T HE R O U T L E D G E COMP ANI ON TO

P H I L O S O P H Y O F L ANGUAGE

Philosophy of language is the branch of philosophy that examines the nature of mean-
ing, the relationship of language to reality, and the ways in which we use, learn, and
understand language.
The Routledge Companion to Philosophy of Language provides a comprehensive and
up-to-date survey of the field, charting its key ideas and movements, and addressing
contemporary research and enduring questions in the philosophy of language. Unique
to this Companion is clear coverage of research from the related disciplines of formal
logic and linguistics, and discussion of the applications in metaphysics, epistemology,
ethics, and philosophy of mind.
Organized thematically, the Companion is divided into seven sections: Core Topics;
Foundations of Semantics; Parts of Speech; Methodology; Logic for Philosophers of Lan-
guage; Philosophy of Language for the Rest of Philosophy; and Historical Perspectives.
Comprised of 70 essays from leading scholars—including Sally Haslanger, Jeffrey
King, Sally McConnell-Ginet, Rae Langton, Kit Fine, John MacFarlane, Jeff Pelletier,
Scott Soames, Jason Stanley, Stephen Stich and Zoltán Gendler Szabó—The Routledge
Companion to Philosophy of Language promises to be the most comprehensive and author-
itative resource for students and scholars alike.

Gillian Russell is Associate Professor of Philosophy at Washington University in

St. Louis.
Delia Graff Fara is Associate Professor of Philosophy at Princeton University.
Routledge Philosophy Companions
Routledge Philosophy Companions offer thorough, high quality surveys and assessments
of the major topics and periods in philosophy. Covering key problems, themes and
thinkers, all entries are specially commissioned for each volume and written by lead-
ing scholars in the field. Clear, accessible and carefully edited and organised, Routledge
Philosophy Companions are indispensable for anyone coming to a major topic or period
in philosophy, as well as for the more advanced reader.

The Routledge Companion to Aesthetics, Second Edition

Edited by Berys Gaut and Dominic Lopes
The Routledge Companion to Philosophy of Religion
Edited by Chad Meister and Paul Copan
The Routledge Companion to the Philosophy of Science
Edited by Stathis Psillos and Martin Curd
The Routledge Companion to Twentieth Century Philosophy
Edited by Dermot Moran
The Routledge Companion to Philosophy and Film
Edited by Paisley Livingston and Carl Plantinga
The Routledge Companion to Philosophy of Psychology
Edited by John Symons and Paco Calvo
The Routledge Companion to Metaphysics
Edited by Robin Le Poidevin, Peter Simons, Andrew McGonigal, and Ross Cameron
The Routledge Companion to Nineteenth Century Philosophy
Edited by Dean Moyar
The Routledge Companion to Ethics
Edited by John Skorupski
The Routledge Companion to Epistemology
Edited by Sven Bernecker and Duncan Pritchard
The Routledge Companion to Philosophy and Music
Edited by Theodore Gracyk and Andrew Kania
The Routledge Companion to Phenomenology
Edited by Søren Overgaard and Sebastian Luft
The Routledge Companion to Philosophy of Language
Edited by Gillian Russell and Delia Graff Fara
Forthcoming:
The Routledge Companion to Seventeenth Century Philosophy
Edited by Dan Kaufman
The Routledge Companion to Eighteenth Century Philosophy
Edited by Aaron Garrett
The Routledge Companion to Philosophy of Mental Disorder
Edited by Jakob Hohwy and Philip Gerrans
The Routledge Companion to Social and Political Philosophy
Edited by Gerald Gaus and Fred D’Agostino
The Routledge Companion to Theism
Edited by Charles Taliaferro, Victoria Harrison, and Stewart Goetz
The Routledge Companion to Philosophy of Law
Edited by Andrei Marmor
The Routledge Companion to Islamic Philosophy
Edited by Richard C. Taylor and Luis Xavier López-Farjeat
The Routledge Companion to Philosophy of Literature
Edited by Noël Carroll and John Gibson
The Routledge Companion to Philosophy of Religion, Second Edition
Edited by Chad Meister and Paul Copan
The Routledge Companion to Aesthetics, Third Edition
Edited by Berys Gaut and Dominic Lopes
The Routledge Companion to Philosophy of Science, Second Edition
Edited by Stathis Psillos and Martin Curd
The Routledge Companion to Ancient Philosophy
Edited by Frisbee Sheffield and James Warren
 PRAIS E FOR THE SER IES
The Routledge Companion to Aesthetics

‘This is an immensely useful book that belongs in every college library and on the book-
shelves of all serious students of aesthetics.’—Journal of Aesthetics and Art Criticism

‘The succinctness and clarity of the essays will make this a source that individuals not
familiar with aesthetics will find extremely helpful.’—The Philosophical Quarterly

‘An outstanding resource in aesthetics . . . this text will not only serve as a handy refer-
ence source for students and faculty alike, but it could also be used as a text for a course
in the philosophy of art.—Australasian Journal of Philosophy

‘Attests to the richness of modern aesthetics . . . the essays in central topics – many of
which are written by well-known figures – succeed in being informative, balanced and
intelligent without being too difficult.’—British Journal of Aesthetics

‘This handsome reference volume . . .belongs in every library.’—Choice

‘The Routledge Companions to Philosophy have proved to be a useful series of high qual-
ity surveys of major philosophical topics and this volume is worthy enough to sit with
the others on a reference library shelf.’—Philosophy and Religion

The Routledge Companion to Philosophy of Religion

‘. . . a very valuable resource for libraries and serious scholars.’—Choice

‘The work is sure to be an academic standard for years to come . . . I shall heartily recom-
mend The Routledge Companion to Philosophy of Religion to my students and colleagues
and hope that libraries around the country add it to their collections.’—Philosophia
Christi

The Routledge Companion to Philosophy of Science

A Choice Outstanding Academic Title 2008

‘With a distinguished list of internationally renowned contributors, an excellent choice
of topics in the field, and well-written, well-edited essays throughout, this compendium
is an excellent resource. Highly recommended.’—Choice

‘Highly recommended for history of science and philosophy collections.’—Library

Journal
This well conceived companion, which brings together an impressive collection of
distinguished authors, will be invaluable to novices and experience readers alike.’
—Metascience

The Routledge Companion to Twentieth Century Philosophy

‘To describe this volume as ambitious would be a serious understatement. . . . full of

scholarly rigor, including detailed notes and bibliographies of interest to professional
philosophers. . . . Summing up: Essential.’—Choice

The Routledge Companion to Philosophy and Film

‘A fascinating, rich volume offering dazzling insights and incisive commentary on every
page . . . Every serious student of film will want this book . . . Summing Up: Highly
recommended.’—Choice

The Routledge Companion to Philosophy of Psychology

‘This work should serve as the standard reference for those interested in gaining a
reliable overview of the burgeoning field of philosophical psychology. Summing Up:
Essential.’—Choice

The Routledge Companion to Metaphysics

‘The Routledge Philosophy Companions series has a deserved reputation for impressive
scope and scholarly value. This volume is no exception . . . Summing Up: Highly
recommended.’—Choice

The Routledge Companion to Nineteenth Century Philosophy

A Choice Outstanding Academic Title 2010

‘This is a crucial resource for advanced undergraduates and faculty of any discipline who
are interested in the 19th-century roots of contemporary philosophical problems. Sum-
ming Up: Essential.’—Choice

The Routledge Companion to Ethics

‘This fine collection merits a place in every university, college, and high school library
for its invaluable articles covering a very broad range of topics in ethics[.] . . . With its
remarkable clarity of writing and its very highly qualified contributors, this volume is
must reading for anyone interested in the latest developments in these important areas
of thought and practice. Summing Up: Highly recommended.’—Choice
 TH E ROU TL E DGE
CO M PAN ION T O
P H IL OSOPHY O F L ANGUAGE

Edited by
Gillian Russell and Delia Graff Fara
 First published 2012
by Routledge
711 Third Avenue, New York, NY 10017
Simultaneously published in the UK
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2012 Taylor & Francis
The right of the editors to be identified as the authors of the editorial material, and
of the authors for their individual chapters, has been asserted in accordance with
sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any
electronic, mechanical, or other means, now known or hereafter invented, including photocopying and
recording, or in any information storage or retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Library of Congress Cataloging in Publication Data
The Routledge companion to philosophy of language/edited by Delia Graff Fara and Gillian Russell.
p. cm.—(Routledge Philosophy Companions)
Includes bibliographical references and index.
1. Language and languages—Philosophy. I. Fara, Delia Graff. II. Russell, Gillian.
P106.R75 2011
121'.68—dc23
2011023340
ISBN: 978–0–415–99310–4 (hbk)
ISBN: 978–0–203–20696–6 (ebk)
Typeset in Goudy by Swales & Willis Ltd, Exeter, Devon
 C ONT ENT S

Notes on Contributors xv
Acknowledgments xxv

Introduction 1
GILLIAN RUSSELL

SECTION I
Core topics 7

1.1 Extension, Intension, Character, and Beyond 9

DAVID BRAUN

1.2 Semantics and Pragmatics 18

CHRISTOPHER GAUKER

1.3 Logical Form 29

KIRK LUDWIG

1.4 Presupposition 42
PAUL DEKKER

1.5 Implicature 53
LAURENCE HORN

1.6 Pragmatic Enrichment 67

FRANÇOIS RECANATI

1.7 Meaning and Communication 79

KENT BACH

1.8 Compositionality 91
JOSH DEVER

1.9 Focus and Intonation 103

DANIEL BÜRING

1.10 Context-Sensitivity 116

TOM DONALDSON AND ERNIE LEPORE
 CO N TENT S

1.11 Relativism 132

JOHN MACFARLANE

1.12 Vagueness 143

J. ROBERT G. WILLIAMS

1.13 Empty Names 153

SARAH SAWYER

1.14 Relevance Theory 163

ROBYN CARSTON

1.15 Truth and Reference in Fiction 177

STAVROULA GLEZAKOS

SECTION II
Foundations of Semantics 187

2.1 Reference 189

TERESA ROBERTSON

2.2 Theories of Truth 199

MATTI EKLUND

2.3 Propositions 209

SCOTT SOAMES

2.4 Concepts 221

CHRISTOPHER PEACOCKE

2.5 Analytic Truth 231

CORY JUHL AND ERIC LOOMIS

2.6 Possible Worlds Semantics 242

DANIEL NOLAN

2.7 Dynamic Semantics 253

SETH YALCIN

2.8 Event Semantics 280

BARRY SCHEIN

2.9 Skepticism about Meaning 295

MICHAEL MCDERMOTT

x
 CO N TENT S

SECTION III
Parts of Speech 305

3.1 Names 307

BARBARA ABBOTT

3.2 Verbs 318

ZOLTÁN GENDLER SZABÓ

3.3 Adjectives 328

CHRIS KENNEDY

3.4 Quantifiers and Determiners 342

G. ALDO ANTONELLI AND ROBERT C. MAY

3.5 Generics 355

SARAH-JANE LESLIE

3.6 Anaphora 367

JEFFREY C. KING

3.7 Descriptions 380

PETER LUDLOW

3.8 Plurals 392

BERNHARD NICKEL

3.9 Adverbs 409

DELIA GRAFF FARA

3.10 Mass Terms 424

FRANCIS JEFFREY PELLETIER

3.11 Indexicals and Demonstratives 438

ALLYSON MOUNT

3.12 Indicative Conditionals 449

ANTHONY S. GILLIES

3.13 Subjunctive Conditionals 466

KAI VON FINTEL

3.14 Questions 478

PAUL HAGSTROM

xi
 CO N TENT S

SECTION IV
Methodology 493

4.1 The Role of Experiment 495

EDOUARD MACHERY AND STEPHEN STICH

4.2 The Role of Linguistics 513

SARAH MOSS

4.3 The Role of Psychology 525

ROBERT J. STAINTON

4.4 The Role of Mathematical Methods 533

LAWRENCE S. MOSS

4.5 The Role of Artificial Languages 544

MARTIN STOKHOF

4.6 The Role of Intuitions 554

MICHAEL DEVITT

SECTION V
Logic for Philosophers of Language 567

5.1 Model Theory: What it Is and What it Isn’t 569

JOHN P. BURGESS

5.2 Logical Quantifiers 579

GILA SHER

5.3 The Logic of Time and Tense 596

ANTONY GALTON

5.4 Modal Logic and its Applications to the Philosophy of Language 609
KIT FINE

5.5 Two-Dimensional Logics and Two-Dimensionalism in Philosophy 624

STEVEN T. KUHN

5.6 Many-Valued Logics 636

NICHOLAS J. J. SMITH

5.7 Dynamic Logic in Natural Language 652

JOHAN VAN BENTHEM

5.8 Intuitionism 667

ALLEN HAZEN

xii
 CO N TENT S

5.9 Richard Montague’s Approach to the Semantics of Natural Languages 678

RICHMOND H. THOMASON

SECTION VI
Philosophy of Language for the rest of Philosophy 691

6.1 Philosophy of Language for Epistemology 693

RAM NETA

6.2 Philosophy of Language for Metaethics 705

MARK SCHROEDER

6.3 Philosophy of Language for Metaphysics 716

6.3.1 The Language of Causation 716
ERIC SWANSON

6.3.2 Dispositional Expressions 729

ALEXANDER BIRD

6.4 Philosophy of Language for Normative Ethics 741

6.4.1 Language, Gender, and Sexuality 741
SALLY MCCONNELL-GINET

6.4.2 Language and Race 753

LUVELL ANDERSON, SALLY HASLANGER,
AND RAE LANGTON

6.5 Apriority 768

SINAN DOGRAMACI

6.6 Necessity and Meaning 782

GILLIAN RUSSELL

6.7 Propositional Attitude Reports 795

DAVID SHIER

SECTION VII
Historical Perspectives 809

7.1 Ancient Philosophy of Language 811

LUCA CASTAGNOLI AND ERMELINDA VALENTINA
DI LASCIO

7.2 Medieval Philosophy of Language 827

GYULA KLIMA

xiii
 CO N TENT S

7.3 Modern Philosophy of Language 841

MICHAEL LOSONSKY

7.4 Frege, Russell, and Wittgenstein 852

MICHAEL POTTER

7.5 Logical Positivism and Quine 860

SANFORD SHIEH

7.6 Ordinary Language Philosophy 873

MICHAEL BEANEY

7.7 Pragmatics and Context: The Development of Intensional Semantics 885

JASON STANLEY

7.8 A Brief History of Generative Grammar 895

ROBERT FREIDIN

Index 917

xiv
 NOTE S ON
CONT R IB UT O R S

Barbara Abbott is a Michigan State University Professor Emerita of Linguistics and

Philosophy of Language. She has worked on such topics as definiteness, proper names,
presuppositions and conditionals, and is the author of Reference (Oxford University
Press, 2010).

Luvell Anderson is the Alain Locke Postdoctoral Fellow in the Department of

Philosophy at Rutgers University. He works on issues in the philosophy of language
and race.

Aldo Antonelli is professor of philosophy at the University of California, Davis.

His research interests include topics in the philosophy of logic and mathematics,
defeasible logics and knowledge representation, philosophical logic, and the theory
of generalized quantifiers.

Kent Bach, Professor Emeritus of Philosophy at San Francisco State University, was
educated at Harvard College and University of California, Berkeley. He has written
extensively in philosophy of language, theory of knowledge, and philosophy of mind.
His books include Thought and Reference (Oxford University Press, 1987, expanded
edition 1994) and, with Michael Harnish, Linguistic Communication and Speech Acts
(MIT Press, 1979).

Michael Beaney is Professor of Philosophy at the University of York, United Kingdom.

He works on the history of analytic philosophy and on conceptions of analysis in the
history of philosophy. He is the author of Frege: Making Sense (Duckworth, 1996),
and editor of The Frege Reader (Blackwell, 1997), Gottlob Frege: Critical Assessments
of Leading Philosophers (with Erich Reck; 4 vols., Routledge, 2005), The Analytic Turn
(Routledge, 2007), and The Oxford Handbook of the History of Analytic Philosophy
(Oxford University Press, forthcoming).

Johan van Benthem is University Professor of Logic at the University of Amsterdam,

and Professor of Philosophy at Stanford University. His current interests are dynamic
logics of information, agency, and games. Relevant works include Exploring Logical
Dynamics (CSLI Publications, 1996) and Logical Dynamics of Information and Interaction
(Cambridge University Press 2010), and he was a co-editor of the Handbook of Logic
and Language (Elsevier Amsterdam, 1997, 2010).
 CO N TRI BUT O R S

Alexander Bird is Professor of Philosophy at the University of Bristol. His research

concerns the metaphysics of science, in particular dispositions, natural kinds, and laws
of nature. He is the author of Nature’s Metaphysics (Oxford University Press, 2007)
and is now working on two projects, one concerning natural kinds and and the other
concerning scientific knowledge.

David Braun is a Professor of Philosophy at University at Buffalo. His main research

interests are in philosophy of language and philosophy of mind.

John P. Burgess has taught since 1975 at Princeton University, where he is now John N.
Woodhull Professor of Philosophy. He has worked mainly in logic and allied areas, and
is author or co-author of A Subject with No Object (with Gideon Rosen); Computability
and Logic (with George Boolos and Richard Jeffrey); Fixing Frege; Mathematics, Models,
and Modality; Philosophical Logic; and Truth (with Alexis G. Burgess).

Daniel Büring is Professor of Linguistics at the University of Vienna, Austria. His main
research areas are semantics, pragmatics, and intonation; his publications include The
Meaning of Topic and Focus (1997, Routledge), Binding Theory (Cambridge University
Press, 2005) and the forthcoming Intonation and Meaning.

Robyn Carston is Professor of Linguistics at University College London. She works

on the semantics/pragmatics distinction, the explicit/implicit communication
distinction, and the interpretation of metaphor. She has published Thoughts and
Utterances (Blackwell, 2002) and a collection of her papers Pragmatics and Semantic
Content is forthcoming (Oxford University Press).

Luca Castagnoli is Lecturer in Ancient Philosophy at the Department of Classics and

Ancient History Durham University. His research interests include ancient logic and
epistemology and he is the author of Ancient Self-Refutation (Cambridge University
Press, 2010).

Paul Dekker is Associate Professor at the ILLC/Department of Philosophy, of the

Universiteit van Amsterdam. His work originates in the philosophy of language
and formal semantics. More specific interests include the dynamic semantics, the
semantics/pragmatic interface, quantification and modality. He has edited a book on
the semantics and pragmatics of questions, and is currently completing a monograph
on an original interpretation of formal dynamic semantics.

Josh Dever is Associate Professor of Philosophy at the University of Texas at Austin.

He works in the philosophy of language, philosophical logic, and metaphysics.

Michael Devitt is Distinguished Professor of Philosophy at the Graduate Center at

CUNY. His research interests include the philosophy of linguistics and the philosophy
of language. He is the author of Designation (Columbia, 1981), Coming to Our Senses
(Cambridge University Press, 1996), Language and Reality (with Kim Sterelny, MIT
Press, 1999), Ignorance of Language (Oxford University Press 2006), and the editor (with
Richard Hanley) of the Blackwell Guide to the Philosophy of Language (Blackwell, 2006).
His most recent book is Putting Metaphysics First (Oxford University, Press, 2010).

xvi
 CO N TRI BUT O R S

Sinan Dogramaci is Assistant Professor of Philosophy at the University of Texas at

Austin. Most of his recent research concerns questions about how we can rationally
infer conclusions that are logical consequences of things we already know.

Tom Donaldson is a PhD student at Rutgers University. He is interested in meaning

and reasoning, and is working on a dissertation on pragmatism.

Matti Eklund is Associate Professor of Philosophy at Cornell University. His published

work is primarily on topics in metaphysics, philosophy of language, and philosophy
of logic.

Kit Fine is University Professor and Silver Professor of Philosophy and Mathematics
at New York University. His principal interests are metaphysics, logic, and the
philosophy of language.

Kai von Fintel is Professor of Linguistics in the Department of Linguistics and

Philosophy at the Massachusetts Institute of Technology and Associate Dean of
MIT’s School of Humanities, Arts, and Social Sciences. He is editor of the journal
Semantics & Pragmatics. His research is primarily in semantics, pragmatics, philosophy
of language, and the intersections thereof. He has worked on modality, conditionals,
presuppositions, quantification, context-dependency, and negative polarity.

Robert Freidin is Professor of Linguistics in the Council of the Humanities at Princeton

University. His research concerns syntax and semantics, focusing on the foundations
of syntactic theory (the central concepts of syntactic analysis and their evolution)
and their role in the study of language and mind. Some of this work is collected
in Generative Grammar: Theory and its History (Routledge, 2007). His most recent
publications include “The roots of minimalism” (with Howard Lasnik) in The Oxford
Handbook of Linguistic Minimalism, “Noam Chomsky’s contribution to linguistics: a
preliminary sketch” in The Oxford Handbook of the History of Linguistics and Syntax:
basic concepts and applications (forthcoming from Cambridge University Press).

Antony Galton is Reader in Knowledge Representation in the College of Engineering,

Mathematics, and Physical Sciences at the University of Exeter. He has published
widely on the topic of spatial, temporal, and spatio-temporal knowledge representation,
with particular emphasis on the conceptual foundations of geographical information
science. Earlier work explored the logical and philosophical underpinning of aspectual
phenomena in natural language.

Christopher Gauker is Professor of Philosophy at the University of Cincinnati. He

is the author of Thinking Out Loud: An Essay on the Relation between Thought and
Language (Princeton, 1994), Words without Meaning (MIT Press, 2003), Conditionals
in Context (MIT Press 2005) and Words and Images: An Essay on the Origin of Ideas
(Oxford University Press, 2011).

Stavroula Glezakos is Associate Professor of Philosophy at Wake Forest University.

Her research focuses on questions concerning the meaning and reference of names
and kind terms.

xvii
 CO N TRI BUT O R S

Delia Graff Fara is an Associate Professor in the Philosophy Department at Princeton

University. Her research and teaching interests fall mainly in the areas of philosophy
of language, metaphysics, philosophical logic, and philosophical linguistics.

Anthony S. Gillies is an Associate Professor of Philosophy at Rutgers University. His

research focuses on issues about language, logic, and information. He has worked on
conditionals, modality, and the dynamics of both meaning and belief.

Paul Hagstrom is Associate Professor of Linguistics in the Department of Romance

Studies at Boston University. His research centers around the syntax, morphology,
and semantics of interrogative sentences and indefinites, and he also works on first
language acquisition of morphosyntax and second language acquisition of definiteness
restrictions. He has published overviews of interrogative semantics (2006, Glot
International, 2003) and of A-not-A questions (Blackwell Companion to Syntax,
2006).

Sally Haslanger is Professor of Philosophy and Director of Women’s and Gender Studies
at MIT. She specializes in feminist theory and critical race theory with an emphasis
on questions concerning metaphysics, epistemology, and philosophy of language.

Allen Hazen is a Fellow of the University of Melbourne’s philosophy department and

Professor of Philosophy at the University of Alberta. His research interests include
metaphysics, philosophy of mathmatics, logic, philosophy of language, philosophy of
mind, epistemology, and history of philosophy.

Laurence Horn is Professor of Linguistics and Philosophy at Yale University. He is

the author of A Natural History of Negation and numerous papers on negation and
pragmatics and the (co-)editor of Negation and Polarity (Oxford University Press,
2000), The Handbook of Pragmatics (Blackwell), Explorations in Pragmatics (de Gruyter,
2001) and The Expression of Negation (de Gruyter, 2010).

Cory Juhl is Professor of Philosophy at the University of Texas at Austin. His research
areas include the philosophy of mathematics, metaphysics, and metasemantics. His
recent papers include works on the fine-tuning argument, on the inverse gambler’s
fallacy, and another paper on the distinction between theoretical entities and
‘stipulata’. He is co-author of Analyticity (2010, Routledge).

Chris Kennedy is Professor and Chair of the Department of Linguistics at the

University of Chicago. His research addresses questions in semantics, pragmatics,
and philosophy of language, primarily through an exploration of the grammar of
comparison, amount, degree, and vagueness. He is the general editor (with Chris
Barker) of Oxford University Press’s Studies in Semantics and Pragmatics and Surveys in
Semantics and Pragmatics.

Jeffrey C. King Professor in the Philosophy Department at Rutgers University. His

research focuses on the philosophy of language, formal semantics, philosophical logic
and metaphysics. He is the author of two books: The Nature and Structure of Content
(Oxford University Press, 2007) and Complex Demonstratives: A Quantificational
Account. (MIT Press, 2001).

xviii
 CO N TRI BUT O R S

Gyula Klima is Professor of Philosophy at Fordham University, New York, and Director
of the Society for Medieval Logic and Metaphysics. His books include John Buridan
(Oxford University Press, 2009), Readings in Medieval Philosophy (Blackwell, 2007),
John Buridan: Summulae de Dialectica, an annotated translation with a philosophical
introduction (Yale University Press, 2001), and ARS ARTIUM: Essays in Philosophical
Semantics, Medieval and Modern (Hungarian Academy of Sciences, 1988).

Stephen T. Kuhn is a professor in the Department of Philosophy and an affiliated

faculty member in the Department of Linguistics at Georgetown University. His
interests include modal and tense logic, philosophy of logic, and the evolution of
moral norms.

Rae Langton is Professor of Philosophy at MIT. She works on moral and political
philosophy, feminist philosophy, speech act theory, metaphysics, and the history of
philosophy.

Ermelinda Valentina Di Lascio is Leverhulme Early Career Fellow at the Department

of Classics and Ancient History of Durham University. Her research interests
include ancient logic and dialectic and she has published on Aristotle’s Sophistical
Refutations, logical papyri, and Wittgenstein.

Ernie Lepore is a professor of philosophy at Rutgers University. He works in philosophy

of language.

Sarah-Jane Leslie is an Assistant Professor in the Philosophy Department at Princeton

University. She is also associated faculty in the Psychology Department, and serves
on the executive committee for the Program in Linguistics. Her work is primarily
in the intersection between philosophy and psychology. She is currently writing a
book on generics for Oxford University Press, which will bring issues in philosophy of
language into contact with recent empirical results from psychology.

Eric Loomis is Associate Professor of Philosophy at the University of South Alabama.

His research areas include metaphysics and early analytic philosophy. He is co-author,
with Cory Juhl, of Analyticity (Routledge, 2010).

Michael Losonsky is Professor of Philosophy at Colorado State University. He is

author of Linguistic Turns in Modern Philosophy (Cambridge University Press, 2006),
Enlightenment and Action from Descartes to Kant (Cambridge University Press, 2001)
and co-editor with H. Geirsson of Readings in Language and Mind (Blackwell, 1996).
He works on the philosophies of Locke and Leibniz, the history of the philosophy of
language, and the history of semantic internalism.

Peter Ludlow (PhD Columbia U, 1985) taught at the state University of New York
at Stony Brook, the University of Michigan, and the University of Toronto before
joining Northwestern University. He is the editor of the Readings in the Philosophy of
Language (Bradford Books, 1997). Lately he has been working on a number of topics
including the semantics of quantified noun-phrases, microlanguages and the dynamic

xix
 CO N TRI BUT O R S

lexicon and natural logic. His most recent book is The Philosophy of Generative
Linguistics (Oxford University Press, 2011).

Kirk Ludwig is Professor of Philosophy at Indiana University, Bloomington. He publishes

in the philosophy of language, mind, action (esp. collective action), metaphysics and
epistemology. He is editor of Donald Davidson (Cambridge University Press, 2003)
and co-author with Ernie Lepore of Donald Davidson: Meaning, Truth, Language
and Reality (Oxford University Press, 2005) and Donald Davidson’s Truth-theoretic
Semantics (Oxford University Press, 2007).

John MacFarlane is Professor of Philosophy at the University of California, Berkeley.

He has written extensively on the philosophy of logic and the philosophy of language,
and is currently working on a book on relative truth and its applications.

Edouard Machery is Associate Professor of History and Philosophy of Science in the

Department of History and Philosophy of Science at the University of Pittsburgh, a
resident fellow of the Center for Philosophy of Science (University of Pittsburgh), a
member of the Center for the Neural Basis of Cognition (Carnegie Mellon University
and University of Pittsburgh), and an associate editor of the European Journal for
Philosophy of Science. He is the author of Doing without Concepts (Oxford University
Press, 2009).

Robert May is Professor of Philosophy and Linguistics at the University of California,

Davis. He is well-known for his research in theoretical linguistics, especially on the
syntax and semantics of quantification, and is currently working in philosophy of
language and linguistics, philosophy of logic, and history of early analytic philosophy.
His books include Logical Form: Its Structure and Derivation (MIT Press, 1985), Indices
and Identity (MIT Press, 1994) and De Lingua Belief (MIT Press, 2006), the last two
co-authored with Robert Fiengo.

Sally McConnell-Ginet, who writes on various topics in semantics, pragmatics, and

the philosophy of language, is Professor Emerita of Linguistics, Cornell. Gender,
Sexuality, and Meaning (Oxford University Press, 2011) and her other work on the
interaction of language with gender and sexuality draw on linguistic and philosophical
perspectives.

Michael McDermott studied philosophy at the University of Sydney under David

Armstrong, and has taught there ever since.

Lawrence S. Moss is Director of the Program in Pure and Applied Logic at Indiana
University, Bloomington, and is Professor of Mathematics and Adjunct Professor
of Computer Science, Informatics, Linguistics, and Philosophy. He is an editor of
the Journal of Logic, Language, and Information, and of Logical Methods in Computer
Science. He is Steering Committee Chair of the North American Summer School in
Logic, Language, and Information.

xx
 CO N TRI BUT O R S

Sarah Moss is Assistant Professor of Philosophy at the University of Michigan. Her

current research applies tools from the philosophy of language to problems in
metaphysics and formal epistemology. She has also published in formal semantics
and pragmatics.

Allyson Mount is Assistant Professor of Philosophy at Keene State College in Keene,

New Hampshire. Her research interests include context sensitivity and the role of
speaker intentions, and she has published work on indexicals and demonstratives.

Ram Neta is Associate Professor of Philosophy at the University of North Carolina at

Chapel Hill. He writes on various topics in epistemology. He is editor (with Duncan
Pritchard) of Arguing About Knowledge (Routledge, 2009).

Bernhard Nickel is Assistant Professor of Philosophy at Harvard University. His

research interests are in the philosophy of language, mind, and science, focusing
especially on their interrelations.

Daniel Nolan is Professor of Philosophy at the Australian National University. He is

the author of Topics in the Philosophy of Possible Worlds (Routledge, 2002) and David
Lewis (Acumen, 2005).

Christopher Peacocke is Professor of Philosophy at Columbia University, and holds

the Wollheim Chair of Philosophy at University College, London. He works on
intentionality, understanding, metaphysics, epistemology, the philosophy of mind,
and the relations between all of the preceding. His books include Sense and Content
(Oxford University Press, 1983), A Study of Concepts (MIT Press, 1992), Being
Known (Oxford, 1999), The Realm of Reason (Oxford University Press, 2004) and
Truly Understood (Oxford University Press, 2008). He is currently working on a book
on the self and the first person. He is a Fellow of the British Academy and of the
American Academy of Arts and Sciences.

Francis Jeffry Pelletier is a Visiting Professor of Philosophy at the University of Alberta.

He is an Emeritus Canada Research Chair in Cognitive Science and Emeritus
Professor of Philosophy and of Linguistics at Simon Fraser University. Within the
areas of philosophy of language and logic he has published work on generics, mass
terms, and semantic compositionality. Additionally, he has published in the areas of
automated theorem proving, artificial intelligence, cognitive science, ancient Greek
philosophy, and history of logic.
Michael Potter is Professor of Logic at Cambridge University. His books include Reason’s
Nearest Kin (Oxford University Press, 2000), Set Theory and its Philosophy (Oxford
University Press, 2004), and Wittgenstein’s Notes on Logic (Oxford University Press,
2009). He has jointly edited Mathematical Knowledge (Oxford University Press, 2007)
and The Cambridge Companion to Frege (Cambridge University Press, 2010).

François Recanati is a research fellow at CNRS and a professorial fellow at EHESS and
the University of St Andrews, he has taught in several major universities around the
world. His recent publications include Literal Meaning (Cambridge University Press,
2004), Perspectival Thought (Oxford University Press, 2007) and Truth-Conditional

xxi
 CO N TRI BUT O R S

Pragmatics (Oxford University Press, 2010). He is the Director of Institut Jean-Nicod

(Ecole Normale Supérieure, Paris).

Teresa Robertson is Associate Professor of Philosophy at the University of California,

Santa Barbara. Her primary interests are in topics that fall into both the areas of
metaphysics and philosophy of language. She has published articles on origin
essentialism, the sorites paradox, and reference to kinds.

Gillian Russell is an Associate Professor in the Philosophy Department at Washington

University in St. Louis. She is the author of Truth in Virtue of Meaning: A defence of
the analytic/synthetic distinction (Oxford University Press, 2008).

Sarah Sawyer is Senior Lecturer at the University of Sussex. Her research interests
are based around the nature of and the connections between thought, language, and
knowledge. Her published work primarily concerns content externalism, justification,
fiction, and singular thought. She is also the editor of New Waves in Philosophy of
Language (2010).

Barry Schein is Associate Professor of Linguistics in the Department of Linguistics at the

University of Southern California. He has published on plurals and event semantics,
and his current work is on the syntax and semantics of conjunction.

Mark Schroeder is Associate Professor of Philosophy at the University of Southern

California. He works in ethics, practical reason, epistemology, and the philosophy of
language, and is author of Slaves of the Passions (Oxford University Press, 2007), Being
For: Evaluating the Semantic Program of Expressivism (Oxford University Press, 2008),
and Noncognitivism in Ethics (Routledge, 2010).

Gila Sher is a Professor of Philosophy at the University of California, San Diego. She
is the author of The Bounds of Logic: A Generalized Viewpoint (MIT Press, 1991). Her
current research centers on logic, truth, and knowledge.

Sanford Shieh is Associate Professor of Philosophy at Wesleyan University. His research

interests include philosophy of logic, history of analytic philosophy, philosophy of
mathematics, and philosophy of language.

David Shier is Associate Professor of Philosophy at Washington State University. His

published work is primarily on propositional attitudes, rationality, and the history of
analytic philosophy.

Nicholas Smith is Senior Lecturer in Philosophy at the University of Sydney. He is the

author of Vagueness and Degrees of Truth (Oxford University Press, 2008) and Logic:
The Laws of Truth (Princeton University Press, 2012).

Scott Soames is Director of the School of Philosophy at the University of Southern

California. His books include What is Meaning?, Philosophy of Language, Reference
and Description, Philosophical Analysis in the Twentieth Century, Beyond Rigidity,
Understanding Truth, and Philosophical Essays.

xxii
 CO N TRI BUT O R S

Robert J. Stainton is Professor of Philosophy at The University of Western Ontario.

Recently, his research has focused on philosophical foundations of linguistics, history
of philosophy of language, and pragmatic impairments.

Jason Stanley is Professor of Philosophy at Rutgers University, New Brunswick. He is the

author of Knowledge and Practical Interests (Oxford University Press, 2005), Language
in Context (Oxford University Press, 2007), and Know How (Oxford University Press,
2011).

Stephen Stich is Board of Governors Professor of Philosophy and Cognitive Science

at Rutgers University and Honorary Professor of Philosophy at the University of
Sheffield. He is a member of the American Academy of Arts and Sciences, a recipient
of the Jean Nicod Prize, and the first recipient of the Gittler Award for Outstanding
Scholarly Contribution in the Philosophy of the Social Sciences. His most recent
book is Collected Essays, vol. 1, Mind and Language (Oxford University Press, 2010).

Martin Stokhof is Professor of Philosophy of Language at the Institute for Logic,

Language and Computation (ILLC), Universiteit van Amsterdam. He has published
on various topics in semantics and pragmatics (such as questions and dynamic
semantics), and on the philosophy of Wittgenstein. His current research interests
focus on issues in the philosophy of linguistics.

Eric Swanson is Assistant Professor of Philosophy at University of Michigan, Ann

Arbor. Much of his recent work is on the language and logic of causation, the language
of subjective uncertainty, and extensions of supervaluationism.

Zoltán Gendler Szabó is A Professor of Philosophy at Yale University. His primary

research focuses on the philosophy of language. He has written on compositionality,
descriptions, context-sensitivity, tense, aspect, modality, and propositional attitudes.

Richmond Thomason is Professor of Philosophy, Linguistics, and Computer Science

at the University of Michigan. He has edited Richard Montague’s philosophical and
linguistic works after Montague’s untimely death in 1971, and his interest in formal
semantics has persisted since that time.

J. Robert G. Williams is Professor in Theoretical Philosophy at the University of Leeds.

His published work is primarily on topics in the philosophy of logic and language and
metaphysics.

Seth Yalcin is an Assistant Professor of Philosophy at the University of California,

Berkeley, where he is also a member of the Group in Logic and the Methodology of
Science. His work usually has something to do with semantics, its foundations, and
its interactions with metaphysics and epistemology.

xxiii
 A CK NOWL ED GM ENT S

A volume of this size doesn’t happen without substantial help, over a long period of time,
from a large number of talented and committed people. So thank you, first of all, to our
contributors. Our list of authors ended up looking like a Who’s Who of contemporary
philosophy of language and we are grateful to everyone who agreed to be a part of this
project. But thanks also to the small army of philosophers who helped to referee papers
and provide feedback for the authors. This volume could not have happened without
their professionalism and selflessness. We are grateful to Ashley Atkins, a graduate stu-
dent in the philosophy department at Princeton, for complinig the index.
We would also like to thank our editors Kate Ahl, and later Andy Beck and Mike
Andrews at Routledge for their patience and for giving us the time to produce the book
that we wanted.
Delia Graff Fara in particular acknowledges an enormous debt to Gillian Russell for
taking on many lions’ shares worth of work on this project. Her limitless enthusiasm,
industriousness, diligence, as well as kindness are inestimable.
Gillian Russell would like to thank her partner, Robert Miller, for all his support.
And also Meshuggah’s Coffee on the Loop, where much of her work on the Companion
was done. It’s still her favorite.
 INT R ODUC T ION
Gillian Russell

1
Do we ever make progress in answering philosophical questions? If we narrow our focus
to the philosophy of language it is apparent that we do. Even acknowledging the dif-
ficulties in measuring such things, it’s clear that more progress has been made in the last
century or so than was made in any previous century, and moreover that there is reason
to expect—if the present culture of training and research can endure—that our students
will be able to do a better job than we.
We are thriving at least in part because of the tools and data we have received from
two other thriving disciplines—both logic and linguistics came into their own in the
20th century. Thanks to logic we have been able to draw and consistently maintain
distinctions that were previously beyond us and consider theories that we used to be
unable to formulate. Thanks to linguistics we have learned about new phenomena and
generalisations about those phenomena, and been exposed to data that has helped us to
rule out or refine some of our own theories.
Readers who are new to the subject will want to ask what the philosophy of language
is exactly and how it differs from logic or linguistics. Such a reader will probably be
frustrated to discover that even researchers who have devoted their lives to the area are
often unsure how to answer such questions and perhaps aren’t too concerned by this.
In practice, hesitation in defining the boundaries of one’s discipline may present little
obstacle to being a successful practitioner. Yet one way to introduce the philosophy of
language to someone to whom it is entirely novel is to give examples of the questions
with which it is concerned: what is meaning? what is truth? what is reference? How are
meaning and reference related? How does linguistic communication work? All these
questions are much more general than something that would be studied in a foreign
language class, or even a class on the anthropology of language. But there is some danger
that these questions sound so abstract and general as to give little idea what a philoso-
pher of language actually does. Moreover, anyone who has actually studied the subject
knows that it also involves questions of much greater specificity, such as: what is the
meaning of the word the? What are the truth-conditions of subjunctive conditionals?
And: how is anaphoric reference determined? And if we put it this way the subject is in
danger of sounding so concrete as to be dull, pedantic, and perhaps worst of all, like we
are attempting to do linguistics without getting proper training in the subject first. That
might have been acceptable before there was a developed science of linguistics, but it
would be jejune now.
 G I LLI A N RU S S E L L

We can perhaps do better by exploring an example. Take our first question: what is
meaning? We normally think that sentences have meanings—they say something—and
that sentences in two different languages, and even two different sentences in the same
language, may have the same meaning. But just what kind of thing is a meaning? Some
observations can help to restrict the set of possible answers: what sentences say—their
meanings—are often the kinds of thing that can be true or false. The meaning of a
whole sentence is a function of the meanings of its parts. Even sentences that no-one
has ever said or thought of may have meanings. Some sentences seem to be useable with
different meanings on different occasions.
That might not seem like much to go on—and I don’t wish to suggest that any of
these claims is unassailable if we discover good reasons to reject them—but this is some-
where to start. It tells us, for example, that meanings must be the kinds of things that
can have parts, that such parts must be composable into something that can be either
true or false, and that there is no simple 1–1 relationship between sentences and their
meanings. There is much information that could be directly or indirectly useful in tak-
ing such a miniature project further. Some of it could come from psychology: meanings
are the kinds of things that speakers of a language are sensitive to, so they must some-
how be the kinds of things that human beings can be sensitive to. Psychology—includ-
ing things like studies of language learning, or linguistic processing by speakers—can
tell us more about how that can happen. Other information comes from linguistics,
especially from the study of specific features of languages. Meanings can be complex,
and the ways in which meanings compose into more complex meanings is quite various.
We would not have the conceptual tools necessary for formulating appropriate theories
without the help of mathematicians and logicians. Recent research into the psychology
of very basic mathematical abilities in humans and animals has suggested that language
was an important tool in the development of human mathematical capacities, but, it is
also quite clear from looking at contemporary philosophy of language that our theories
of language would be miserably poor without mathematics. Taking all this scientific
information and all these mathematical tools, and coming up with theories about what
meanings are and reasons to accept some of those theories over others, is one of the
things that philosophers of language do.

2
There are lots of features of natural languages of which their native speakers are igno-
rant. That is perhaps a contentious way to put it, but the situation is roughly this: you
can be a competent speaker of English without being able to produce an explicit list of
the syntactic markers for context-sensitivity, the rules for turning active constructions
into passives, or for determining the referents of anaphora. (It gets worse. If Tim Wil-
liamson’s Philosophy of Philosophy is to be believed, you can be a competent speaker of
English and not know that all vixens are vixens.) Competent English speakers often are
nonetheless able—as they wander about the world, having conversations and reading
each others postcards—to judge accurately the truth-conditions of sentences contain-
ing context-sensitive expressions, reliably turn actives into passives, and determine the
referents of anaphora. It is just that when asked to come up with a general theory of how
such things are done, if they give an answer at all it will usually be wrong. We might
say that for humans speaking their native languages, linguistic processing—much like
visual object processing—is mostly transparent and automatic. We don’t so much see

2
 I N TRO DU C T I O N

it happening, or intentionally make it happen, as simply find ourselves presented with

the results. Once we acknowledge this it becomes unsurprising that there is much to be
discovered about how language works, including things that no speaker knows or has
ever known.
Still, the myth that competent speakers know all there is to know about their lan-
guages has historically been a barrier to progress, because it has suggested that there is
nothing about a living language that some living person doesn’t already know. One
way in which linguistics has greatly aided the philosophy of language is by making it
increasingly obvious that there is a large gap between what every fluent speaker knows,
and what the experts know. In doing so they have made it clear that more substantive
research into language is possible.
Yet even today, the misconception of full speaker knowledge survives where it has
not been wiped out, and unfortunately that includes many areas of philosophy. Where
it survives it has a tendency to underwrite bad work on language, and undermine the
reception of the good. It underwrites the bad by suggesting that if one is a native speaker
(and perhaps in addition a philosopher of science, or an ethicist, or a metaphysician)
and one proposes a theory of how some part of the language works and can see no
problems with it, then there are no problems with it. A naive philosopher might simply
conclude that there is nothing substantive or serious to be said against his claims about
meanings because he has a hard time envisioning what serious or substantive work there
could be. It undermines the reception of good work because having a favourite answer to
a problem and thinking that there can be no serious or substantive work that contradicts
it tends to undermine one’s motivation to read the literature.
For example, there are areas in philosophy—including value theory, vagueness, the
philosophy of arithmetic, and the philosophy of ordinary objects—where we need both
an account of what sentences about the domain of study are claiming, and an account
of what exists and is happening within the domain itself. We need some philosophy of
language and we need some metaphysics. Sometimes one’s theory of what exists and
is happening in the domain itself can be made more satisfying by a novel account of
the semantics of the sentences about the domain. Ethicists might suggest that value-
ascribing sentences don’t really make assertions, and thereby free themselves from the
requirement of positing certain odd kinds of object. Philosophers of mathematics might
claim that numerals are not really referring expressions, and that sentences employing
them don’t require the existence of numbers to be true, and metaphysicians may find
it simplifies their metaphysical theories if they maintain that some locution in Eng-
lish is systematically ambiguous, or context-sensitive, or creates an opaque context. In
the most egregious cases, this is treated as if it can be done “for free.” The fact that it
results in an improved metaphysical theory, and the author (who is, after all, a compe-
tent speaker) sees no problems with the linguistic part of the theory is taken as grounds
enough—as if there could be no independent, substantial linguistic research that bore
on the question of whether the linguistic part of the theory is true.
That’s a mistake, and to ignore the research that has been done is to ignore significant
evidence that is relevant to whether or not one’s overall picture is correct. It is possible
that non-philosophers of language will find the extra tests that their combined theories
have to pass onerous. But from the disinterested perspective of knowledge gathering,
a rich new source of evidence is to be welcomed—it gives us a new tool for telling the
good theories from the bad.

3
 G I LLI A N RU S S E L L

3
The philosophy of language can itself be of great use to the rest of philosophy, but there
is much more to philosophy than just the philosophy of language. There are genuinely
hard questions in ethics and metaphysics, in epistemology and the philosophy of sci-
ence, that would remain unsolved even if all questions in the philosophy of language
were answered. It is as much a mistake to think that the only substantive philosophy
is philosophy of language as it is to think that nothing concerning language is all that
substantive (and woe betide those philosophers who have done both). Still, the claim
that I want to make here is stronger than many readers will like: I claim that if you
want to be a good philosopher today then you ought to study a lot of philosophy of
language.
The argument for this is both simple and a bit dull. Arguments are the primary tools
and targets of philosophers. When we assess an argument we consider whether the
premises are true, and whether the premises support the conclusion. But the answer
to such questions depends on the meaning of the sentences in the arguments, and so it
is understandably common for opposing philosophical parties to dispute claims about
meaning. Hence philosophers must often evaluate claims about meaning. Meaning con-
troversies include whether knows is a context-sensitive expression, whether value attri-
butions assert propositions and whether numerals are singular terms. Such questions are
not trivial ones, which any fluent speaker is in prime position to answer accurately with
relatively little thought. Nor can the answers be looked up in the index of the aver-
age linguistics textbook. Facility in the philosophy of language makes you better able
to imagine, distinguish and evaluate the different possible answers, and as a result less
likely to make a mistake which will hobble your account in another areas.
It is just possible that philosophers in other areas might find this situation trying.
Perhaps they became epistemologists, ethicists or philosophers of mind etc . . . because
they wanted to do epistemology, ethics or philosophy of mind, where they did not
expect that this involved ploughing through a densely written literature on quantifiers
or gradable adjectives. These other parts of philosophy are no doubt abstract and techni-
cal disciplines themselves, but to make the situation worse, all the input from logic and
linguistics has made some philosophy of language unusually technical and specialised
too, and thus somewhat inaccessible.
So what is a conscientious philosopher to do? Suppose you are already embroiled in
the debate in some area, and you begin to suspect that you ought to know more about
the semantics of questions, or mass terms or subjunctive conditionals. Perhaps this is
because other people in the debate seem to know a lot about it already. Perhaps it is
because when an opponent makes a certain claim you have an inchoate hunch that their
theory clashes with the true semantics and you wish you knew what that was, so that
you could formulate your objection neatly and persuasively. Or perhaps—and maybe
this is the really exciting possibility—perhaps you are suddenly struck with a totally
original idea and it involves claiming that such and such a claim is really just equivalent
to another, less confusing claim, and you want to check and see whether such a thing
would be semantically feasible. You pop next door to see your friendly, neighbourhood
philosopher of language, but she is too busy with the metaphysicians. What should you
do? Fortunately, this is just the kind of emergency that this Routledge Companion to the
Philosophy of Language is designed to handle.

4
 I N TRO DU C T I O N

Contents
This Companion is intended to make the literature on many topics in the philosophy of
language accessible to a variety of audiences: those who are studying the subject for the
first time, those who are already philosophers working in other areas, and even profes-
sional philosophers of language who might want a survey of contemporary work in some
unfamiliar sub-area. For each chapter, we picked a top researcher and asked them to
write a short, interesting introduction to their topic. Many of these introductions are
quite opinionated, but where they take a side on a controversy you should find that they
also give you a good sense of where to look for the alternative approaches, and what you
might put on your reading list if you wanted to know more.
Section I presents work on some of the core ideas in the philosophy of language,
including extensions, intensions, character, presupposition, implicature, communica-
tion, logical form, and context-sensitivity.
Section II then focuses on approaches to some of the key topics that fall under the
heading of ‘semantics’: reference, propositions, concepts, and some rival approaches to
giving a theory of meaning, including possible world semantics, event-based semantics,
and one of the more high-profile and intriguing newcomers, dynamic semantics.
Section III groups together entries focusing on particular kinds of expression, rang-
ing from words like names and adjectives, to complex expressions like a conditionals,
descriptions and questions. Some of these topics—such as descriptions—have already
played an important role in philosophy, and in fact we might say that it was philoso-
phers who kickstarted the linguistics literature. Others—such as questions—have come
to play a role in philosophy only more recently.
The methodology of philosophy—a topic in the philosophy of philosophy—has
attracted significant interest in recent years, and Section IV of this book could be
regarded as a new contribution to that field. Indeed, it could almost have been published
as a stand alone book containing six essays on the methodology of the philosophy of
language. With other parts of the Companion we were usually asking researchers to write
introductions to literatures that we knew already existed, but in this one we wondered
what people would have to say, and commissioned the articles to find out: we wanted a
new literature on the topic, one that would contribute to the “methodological aware-
ness” that Timothy Williamson recommends in his Philosophy of Philosophy and which
could serve as a series of case studies for people writing about the methodology of phi-
losophy more generally. Hence we asked Stephen Stich and Edouard Machery to write
about the use of experiment in the philosophy of language, Robert Stainton to cover
the role of psychology, Larry Moss to talk about mathematical methods, Sarah Moss to
introduce the uses of linguistics, Michael Devitt to talk about intuitions, and finally
Martin Stokhof to discuss the role of artificial languages.
Section V turns the focus to logic. The role that logic plays in the study of language
is itself a controversial, and the section opens with John P. Burgess’ cautionary sermon
“Model Theory: what it is and isn’t,” which warns of the confusions that stem from
confusing model theory with semantics. Subsequent articles introduce and summarise
research both on topics that have traditionally been important to the philosophy of
language, such as quantification and many-valued logics, and on work in logic that has
risen to prominence recently, including two dimensional and dynamic logics.
Section VI focuses on the uses that philosophy of language has been put to in philoso-
phy more broadly. It includes essays on the use of philosophy of language in epistemology

5
 G I LLI A N RU S S E L L

and metaethics, and we asked philosophers in metaphysics and normative ethics to

write essays on language and some particular topic in their area, giving us, for example,
the team of Luvell Anderson, Rae Langton, and Sally Haslanger on language and race,
and Eric Swanson on language and causation.
Contemporary philosophy of language is not an especially historically-oriented part
of philosophy. The historians I know bemoan this, and with some grounds. A standard
introductory course in our area starts in the late 19th century with Frege—not with
the pre-Socratics—and just last week I was talking to a philosopher of language who
suggested that it was a mistake to make students read original texts (such as Kripke and
Burge) and that we would be better off using contemporary textbooks, much as a logi-
cian or a linguist would. The argument appealed to the thought that the big ideas are
rarely presented in their clearest form when they are first discovered, and it is much
easier to learn about them from cleaned-up, contemporary versions. Be that as it may,
the history of philosophy of language is a serious area of study in its own right and igno-
rance is not something to be proud of. Historical ignorance, in particular, may lead to
the repetition of old mistakes and this means that even the most contemporary of phi-
losophers of language has good reason to study her subject’s history. Our final section,
Section VII attempts to provide some. We have gone beyond the familiar story that
starts with Frege, including sections on ancient, medieval and modern philosophy as
well as a chapter on the history of generative grammar. The latter is likely to make fas-
cinating reading for younger philosophers, even if older readers are likely to be horrified
to find events they lived through appearing in the history section. Despite our efforts,
the philosophers who wrote these sections have stressed how much more there is to be
said (especially when we attempted to keep them to word limits). Nonetheless, much of
what is here will be new to the majority of readers and each chapter includes pointers
about where to go if you would like to learn more.

On our cover image

Finally, the cover image for this book is a photograph of five surgical tools, and we
promised the ever-patient Andrew Beck—our editor at Routledge—that we would
explain the choice. Typical jackets for books on language feature famous paintings
of people talking or moderately abstract images containing two heads. Possibly in
pastel colours. Our image is less fluffy. We picked it because the philosophy of lan-
guage is much less fluffy than outsiders normally suppose. We pride ourselves on
clear, precise, both empirically and mathematically informed analysis of subjects
that at first glance appear to be messy and organic. Surgical tools that are designed
for use on the human body seemed like an apt metaphor. In addition, we want
to emphasise the extent to which philosophy of language can itself be a precision
tool—something that philosophers in other areas can exploit in their own work.
Philosophical questions can be difficult and confusing, muddled and, at first, rather
messy. We think of this Companion as offering you tools designed for such work.
Achtung!—they’re sharp.

6
 Section I

CO R E T O P I C S
 1.1
EXT E NS I ON, INT ENSION,
CHAR ACT ER , AND
BE YO ND
David Braun

This article explains some of the technical terms commonly used in semantic theory.
One such term is extension. The notion of an extension is a generalization of the notion
of reference. Proper names refer to objects, and a proper name’s extension is the object
to which it refers. For instance, the extension of ‘Barack Obama’ is Barack Obama, and
the extension of ‘Los Angeles’ is Los Angeles. The extension of a predicate is the set of
objects to which it correctly applies. For instance, the extension of the common noun
‘dog’ is the set of all dogs, the extension of the verb ‘run’ is the set of all things that run,
and the extension of the adjective ‘happy’ is the set of all happy things. Binary predi-
cates correctly apply to pairs of things, and so the extension of a binary predicate is a
set of ordered pairs. For example, the extension of the binary predicate ‘kiss’ is the set of
pairs of objects <x, y> such that x kisses y. The extension of a ternary predicate (such as
‘give’) is a set of ordered triples. Generally, the extension of an n-place predicate (where
n is greater than 1) is a set of n-tuples.
Complex expressions also have extensions. For instance, the extension of ‘kisses
Barack Obama’ is the set of objects to which this phrase correctly applies, namely the
set of objects that kiss Barack Obama. The extension of this complex expression is
completely determined by the extensions of the simple expressions in it: for an indi-
vidual z is a member of the extension of ‘kiss Barack Obama’ if and only if there is a
pair of objects <x, y> such that (i) <x, y> is a member of the extension of ‘kiss’, (ii) y
is identical with the extension of ‘Barack Obama’ (that is, y is identical with Barack
Obama), and (iii) z is identical with x. Declarative sentences are also complex expres-
sions. If we assume that they have extensions, and we also assume that the extension
of every complex expression is completely determined by the extensions of the simple
expressions in it, then we are naturally led to the conclusion that the extension of a
declarative sentence is its truth-value (either truth, if the sentence is true, or falsehood, if
the sentence is false). For example, the sentence ‘Barack Obama runs’ is true if and only
if Barack Obama runs. But the extension of ‘Barack Obama’ is Barack Obama and the
extension of ‘runs’ is the set of things that run. So the extension of the sentence ‘Barack
Obama runs’ is truth if and only if the extension of ‘Barack Obama’ is a member of the
extension of ‘runs’. Therefore, the truth-value of the sentence ‘Barack Obama runs’ is
 DAVI D BR AU N

completely determined by the extensions of the words in that sentence. Similar points
hold for many other simple sentences.
Extensions are important to semantics, but they are not plausible candidates for
the meanings of expressions, for expressions that have the same extension (co-extensive
expressions) can differ in meaning. For example, the predicate ‘renate’ is co-extensive
with the predicate ‘cordate’, and yet the terms differ in meaning, for ‘renate’ means
animal with a kidney whereas cordate means animal with a heart. Every true sentence has
the same extension (truth), and every false sentence has the same extension (falsehood),
but obviously there are pairs of true sentences, and pairs of false sentences, that differ in
meaning: consider, for example, the true sentences ‘Some dogs live in Pittsburgh’ and
‘All cats are mammals’ and the false sentences ‘Some cow is purple’ and ‘Every frog is
a mammal’. Some pairs of co-extensive (co-referring) proper names, such as ‘George
Orwell’ and ‘Eric Blair’, also seem to differ in meaning, for the sentence ‘George Orwell
is George Orwell’ seems to differ in meaning from ‘George Orwell is Eric Blair’.
We can sum up the problems with identifying meanings with extensions by saying
that meanings are finer-grained semantic entities than extensions. Generally, one sort of
semantic entity X is finer-grained than another semantic entity Y if two expressions that
have the same Y can have different X’s.
If meanings are not extensions, then what are they? Many traditional theories say
that the meanings of expressions are individuals and attributes, and complex entities
made up of these. Typically, these views say that every predicate expresses one or more
attributes, and the meaning of a predicate is either the attribute it expresses or the set of
all attributes it expresses. For instance, on some versions of this theory, ‘red’ expresses
the property of being red, while ‘human’ expresses the properties of being rational and
being an animal. Therefore, the meaning of ‘red’ is the property of being red, and the
meaning of ‘human’ is the set consisting of the properties of being rational and being an
animal. The meaning of ‘renate’ is the property of being an animal with a kidney and
the meaning of ‘cordate’ is the property of being an animal with a heart. The latter two
properties are distinct, and so the predicates ‘renate’ and ‘cordate’ differ in meaning.
The meanings of these predicates determine their extensions. For instance, the exten-
sion of ‘renate’ is the set of all things that have the property of being a renate, and so the
latter property determines the extension. Similarly for other predicates: the meaning of
‘kiss’ is the binary relation of kissing, and this relation determines the term’s extension:
namely the set of all pairs <x, y> such that x bears the relation of kissing to y.
On some versions of this traditional theory, the meaning of a proper name is the
object to which it refers, and so the names ‘George Orwell’ and ‘Eric Blair’ have the
same meaning. If the meaning of a proper name is its referent, then the meaning of a
name on this view obviously determines its extension, for they are one and the same
thing. On other versions of the traditional theory, the meaning of a proper name is the
same as that of some definite description: for instance, the meaning of ‘George Orwell’
is the same as that of ‘the author of 1984’. The meaning of a definite description is,
roughly speaking, a property that at most one object can bear. Thus the meaning of
‘George Orwell’ might be the property of being the (sole) author of 1984, whereas the
meaning of ‘Eric Blair’ might be some other property that only Orwell/Blair has. If the
meaning of a proper name is a property that at most one object can satisfy, then the
meaning determines the referent by determining a unique object.
Complex expressions, such as ‘kisses Barack Obama’, also have meanings. On many
traditional views, the meaning of a complex expression E is a structured entity, whose

10
 E XTEN SI O N , I N TEN SI O N , CH A R A C T E R , A N D B E Y O N D

structure reflects the grammar of E and whose constituents are the meanings of the
words in E. For instance, the meaning of ‘kisses Barack Obama’ is something like a
sequence of the meanings of the words in the phrase (‘kiss’ and ‘Barack Obama’), in the
order in which the words appear in the phrase, or a tree-like structure whose ultimate
constituents are the meanings of the words in the phrase. On such views, the meaning of
a complex phrase determines its extension. For instance, the meaning of ‘kisses Barack
Obama’ contains as constituents the meanings of ‘kiss’ and ‘Barack Obama’. The mean-
ing of ‘kiss’ is the relation of kissing, and (as we saw above) this determines the exten-
sion of ‘kiss’. The meaning of ‘Barack Obama’ is either Barack Obama himself or some
property that he has uniquely; either way, the meaning of ‘Barack Obama’ determines
the extension of the name. Finally, as we saw before, the extensions of ‘kiss’ and ‘Barack
Obama’ determine the extension of the phrase ‘kisses Barack Obama’.
The meaning of a declarative sentence is traditionally called a proposition. Declara-
tive sentences are said to express propositions. On many traditional views, propositions
are complex structured entities: if sentence S expresses proposition P, then P has a struc-
ture that reflects the grammar of S, and the ultimate constituents of P are the meanings
of the words in S. For instance, the proposition expressed by ‘Barack Obama runs’ has
the meanings of ‘Barack Obama’ and ‘runs’ as constituents. Propositions that have this
sort of structure are often called structured propositions or Russellian propositions, after
Bertrand Russell (1903), who uses them extensively in his semantic theories. (See also
Kaplan 1989 and Salmon 1986.) A proposition that has an individual as a constituent is
often called a singular proposition. The proposition that a sentence expresses determines
the sentence’s extension (its truth-value). For instance, the meaning of ‘Barack Obama
runs’ has the meanings of ‘Barack Obama’ and ‘runs’ as constituents. These meanings
determine the extensions of ‘Barack Obama’ and ‘runs’, and as we saw earlier, the exten-
sions of these terms determine the truth-value of the sentence.
Many traditional theories of meaning say that people can bear various cognitive
and linguistic relations to propositions. For instance, people can assert, deny, believe,
or doubt the proposition that Barack Obama runs. Furthermore, propositions are the
referents, and extensions, of ‘that’-clauses: for example, the extension of ‘that Barack
Obama runs’ is the proposition that Barack Obama runs. Therefore, the sentence
‘Clinton believes that Barack Obama runs’ attributes to the referent of ‘Clinton’ belief
in the proposition that Barack Obama runs.
Some theorists use alternative terminology for similar notions. Some use ‘denotation’
for (roughly) the extension of a term, and ‘denote’ for the relation between an expres-
sion and its denotation (Mill 1872, Book 1, Chapter 2, Section 5; Russell 1903). Others
use ‘designate’ and ‘designation’ for similar extensional notions. Some use ‘connotation’
for meaning, and ‘connote’ for the relation between an expression and its connotation
(Mill 1872). ‘Comprehension’ is an older term (rarely used now) for the meaning of a
predicate (Arnauld 1662/1964, 51). Intension is a more common term for meaning or
connotation (Hamilton 1860, Volume 2, 101; Carnap 1956). Semantic theories that
describe only the extensions of expressions are often called extensional semantic theories.
Semantic theories that attempt to describe the meanings of expressions are often called
intensional semantic theories.
A more recent, twentieth-century proposal for describing the meanings of expres-
sions relies on the idea that the extension of a term often depends crucially on the way
the world is. For instance, the predicates ‘renate’ and ‘cordate’ are in fact co-extensive,
but there could have been animals with kidneys but no hearts, or animals with hearts

11
 DAVI D BR AU N

but no kidneys, and if either had been the case, then the extension of ‘renate’ would
have been different from the extension of ‘cordate’. We can use this idea to try to distin-
guish between the meanings of ‘renate’ and ‘cordate’, in the following way. Consider all
of the different ways the world might have been, and call these ways possible worlds. The
extension of ‘renate’ at a possible world is the set of things that are renates in that world.
Thus the extension of ‘renate’ varies from possible world to possible world. Now we can
define a function from possible worlds to extensions as follows: Given any possible world
as an input (or argument), this function yields as its output (or value) the set of things
that are renates in that possible world. This function is the possible-worlds intension of
‘renate’, and according to this theory, the meaning of ‘renate’ is identical with this func-
tion. (Most theorists who accept this theory call these functions simply ‘intensions’,
leaving out ‘possible-worlds’. I will say more about this terminology later.) Similarly,
the extension of ‘cordate’ varies from possible world to possible world, and the possible-
worlds intension of ‘cordate’ is a function whose value at any possible world is the set
of things that are cordates in that world. Since it is possible that some cordates are not
renates, the extensions of ‘renate’ and ‘cordate’ differ at some possible world, and so the
possible-worlds intensions of these terms are different functions. If the only meanings of
these terms are their possible-worlds intensions, then they differ in meaning.
We can extend this idea to other expressions. The predicate ‘kiss’ has an extension at
each possible world (the set of ordered pairs in which the first member kisses the second
in that world), and the possible-worlds intension of ‘kiss’ is the function that yields the
extension of ‘kiss’ at each possible world. Similarly for other predicates. The possible-
worlds intension of a proper name is a function which, at every possible world, yields
the extension of that name at that world. Suppose that the extension of a given proper
name can vary from possible world to possible world. Then there may be worlds at which
the extension of ‘Eric Blair’ is different from the extension of ‘George Orwell’. If so, then
the possible-worlds intensions of the names are different, and if meanings are possible-
worlds intensions, then these names differ in meaning. The extension of a declarative
sentence at a possible world is a truth-value, and its possible-worlds intension is a func-
tion that, at every world, yields the sentence’s truth-value at that world. There are some
worlds in which no dogs live in Pittsburgh, but all cats are mammals. In these worlds, the
extension of ‘Some dogs live in Pittsburgh’ is falsehood and the extension of ‘All cats are
mammals’ is truth. Therefore, these sentences have different possible-worlds intensions,
and if their meanings are their possible-worlds intensions, then they differ in meaning.
The view that meanings are identical with possible-worlds intensions is sometimes
called possible-worlds semantics. Many proponents of possible-worlds semantics identify
the proposition expressed by a sentence S with its possible-worlds intension, but some
instead identify the proposition expressed by S with the set of possible worlds in which S
is true. These two ways of speaking about propositions are equivalent, on possible-worlds
semantics. The possible-worlds intension of S divides up the set of possible worlds into
two sets, the set of worlds in which S is true and the set of worlds in which S is false.
Therefore, S’s possible-worlds intension determines the set of worlds in which S is true.
But the set of worlds in which S is true also determines S’s intension, for if W is the set
of worlds at which S is true, then the intension of S is the function that yields truth for
every world in W and yields falsehood for every world not in W. Therefore the two ways
of speaking of propositions are equivalent, according to possible-worlds semantics.
Traditionally, the term ‘intension’ has been used as a virtual synonym for ‘meaning’.
But most contemporary philosophers and linguists use the term ‘intension’ specifically

12
 E XTEN SI O N , I N TEN SI O N , CH A R A C T E R , A N D B E Y O N D

for functions from possible worlds to extensions; these are the functions that I called
‘possible-worlds intensions’ above. Some philosophers and linguists who use the term
‘intension’ in this modern way think that meanings really are identical with such func-
tions (Carnap 1956, Lewis 1972, Montague 1974, Stalnaker 1984). Others who use
‘intension’ in this modern way accept that there are such functions but deny that mean-
ings are identical with them (Kaplan 1989, Salmon 1986, Soames 1987). To avoid
taking a stand on this issue, I shall continue to use the term ‘possible-world intension’
when I speak of functions from possible worlds to extensions, and I shall continue to use
plain old ‘intension’ (without the modifier ‘possible-world’) in the traditional way, as a
term that is, more or less, synonymous with ‘meaning’. Readers should keep in mind that
I am, in this respect, departing from standard contemporary usage.
Along with ‘extension’ and ‘intension’ come other related technical terms. An exten-
sional context is a part of a complex expression in which substitution of co-extensive
expressions does not change the extension of the complex expression. For instance,
the underlined portion of the verb phrase ‘kissed Eric Blair’ is an extensional context
in that verb phrase, because substituting a co-extensive expression, such as ‘George
Orwell’, for ‘Eric Blair’ in this context does not change the extension of the verb phrase
(namely, the set of things that kissed Blair/Orwell). Similarly, the underlined portion of
‘Eric Blair kissed Mary’ is an extensional context in this sentence, for if one substitutes
a co-extensive expression for ‘Eric Blair’ (e.g., ‘George Orwell’), then the extension
(the truth-value) of the entire sentence remains the same. A non-extensional context is
a part of a complex expression in which substitution of co-referring expressions may
change the extension of the complex expression. For instance, the underlined portion
of ‘Necessarily, all renates have kidneys’ is a non-extensional context in the sentence,
for substituting the co-extensive expression ‘cordate’ for ‘renate’ yields a false sen-
tence. Similarly, substituting ‘cordate’ for ‘renate’ in the sentence ‘John believes that
all renates have kidneys’ may result in a change in the truth-value of this sentence, and
so the underlined portion is a non-extensional context in the sentence. Sentences that
contain non-extensional contexts are exceptions to the general rule that the extension
of a complex expression is completely determined by the extensions of its parts. An
intensional context is a part of a complex expression in which substitution of expressions
with the same possible-worlds intension does not change the extension of the complex
expression. For instance, the underlined portion of ‘Necessarily, all renates have kid-
neys’ is an intensional context in this sentence, for we can substitute any expression
with the same intension as ‘renate’, such as ‘animal with kidneys’, and the truth-value of
the whole sentence will remain the same. (Different theorists use the term ‘intensional
context’ differently. By my definition, the underlined portion of ‘Eric Blair kissed Mary’
is both an extensional and an intensional context. But some theorists use the term
‘intensional context’ for all and only contexts that are non-extensional; they would
not count this context as intensional.) A hyperintensional context is a part of a complex
expression in which substitution of expressions with the same possible-worlds intension
may change the extension of the complex expression. For example, the underlined por-
tion of ‘John believes that all animals with kidneys have kidneys’ is a hyperintensional
context in the sentence, for substituting ‘renate’ for ‘animals with kidneys’ may change
the truth-value of the sentence.
Possible-worlds intensions resemble meanings in at least two respects. First, meanings
determine extensions, and so do possible-worlds intensions. Second, two expressions
with the same extension can differ in meaning, and likewise two expressions with the

13
 DAVI D BR AU N

same extension can differ in their possible-worlds intensions. But identifying mean-
ings with possible-worlds intensions is problematic (Soames 1987). The sentences ‘All
dogs are dogs’ and ‘2 + 2 = 4’ are both true at all possible worlds, and so have the same
possible-worlds intension, but seem to differ in meaning. The sentences ‘Barack Obama
exists and he is not an electron’ and ‘Barack Obama exists and 2 + 2 = 4’ are both true at
all possible worlds in which Obama exists, and both false at all worlds in which Obama
does not exist, and so have the same truth-value at all worlds, and therefore have the
same possible-worlds intension, but they differ in meaning. The predicates ‘triangular’
and ‘trilateral’ have the same extension at any possible world, and so they have the same
possible-worlds intension, but they differ in meaning. Many philosophers think that the
extension of a given proper name does not vary from world to world (see Kripke 1980
for arguments for this claim). On their view, proper names are rigid designators, that is,
expressions whose extensions, or designations, are the same at all possible worlds. If
proper names are rigid designators, then two proper names that have the same extension
at the actual world, such as ‘Eric Blair’ and ‘George Orwell’, have the same extension at
all possible worlds, and so have the same possible-worlds intension. Yet some such pairs
of names seem to differ in meaning.
To deal with some of these difficulties in identifying meanings with possible-worlds
intensions, some philosophers propose to identify meanings with structured intensions
(Lewis 1972, Creswell 1985). A structured intension is a structured entity that contains
possible-worlds intensions as constituents. The structured intension of a sentence S is
an entity that contains the possible-worlds intensions of the words in S, within a struc-
ture that reflects the grammatical structure of S. The structured intension of ‘All dogs
are dogs’ contains the possible-worlds intension of ‘dog’, while the structured intension
of ‘2 + 2 = 4’ contains the possible-worlds intension of ‘2’ and does not contain the pos-
sible-worlds intension of ‘dog’. Therefore, these two sentences have different structured
intensions, even though they have the same (unstructured, possible-worlds) intension.
So if meanings are structured intensions, then these sentences differ in meaning. Simi-
larly, ‘Barack Obama exists and is not an electron’ has a different structured intension
from ‘Barack Obama exists and 2 + 2 = 4’ because (for instance) the first contains the
possible-worlds intension of ‘electron’ whereas the second does not. Complex predicate
phrases, such as ‘exists and is not an electron’ and ‘exists and is such that 2 + 2 = 4’ also
have different structured intensions. The expressions ‘trilateral’ and ‘triangular’ also
have different structured intensions, if they are complex predicate phrases made up of
‘tri’ and ‘angular’, and ‘tri’ and ‘lateral’, respectively.
Structured intensions resemble the meanings that the earlier traditional theories
attribute to complex expressions. On the traditional theory, the meaning of ‘Barack
Obama runs’ is a complex entity that has as its constituents the meaning of ‘Barack
Obama’ (which is either the man or a certain sort of attribute) and the meaning of
‘runs’ (the property of running). The structured intension of the sentence is similar,
but has the possible-worlds intensions of ‘Barack Obama’ and ‘runs’ as constituents.
However, some philosophers and linguists hold that attributes just are possible-worlds
intensions. This identification is controversial, but if it is correct, then structured inten-
sions come close to the meanings of the traditional theory. There remains one differ-
ence, however, for some traditional propositions contain individuals (such as Barack
Obama) as constituents, whereas no structured intension does (though a structured
intension can contain a possible-worlds intension whose value at every world is Barack
Obama).

14
 E XTEN SI O N , I N TEN SI O N , CH A R A C T E R , A N D B E Y O N D

Co-referring proper names still present a serious apparent problem for both theories.
Suppose that proper names are rigid designators. Then ‘George Orwell’ and ‘Eric Blair’
have the same possible-worlds intension. Therefore, the sentences ‘George Orwell is
George Orwell’ and ‘George Orwell is Eric Blair’ have the same structured intension.
Yet the sentences seem to differ in meaning. Traditional theories that say that the mean-
ing of a proper name is its referent have a parallel problem, for these theories entail that
the two names have the same meaning, and so entail that the two sentences have the
same meaning. An advocate of structured intensions might try to evade this problem by
denying that proper names are rigid designators; if ‘George Orwell’ and ‘Eric Blair’ differ
in extension at some world, then they have different possible-worlds intensions, and so
differ in meaning. A traditional theorist might deny that the meaning of a name is its
referent, and instead hold that the meaning of a name is an attribute, and claim that the
meanings of these names are distinct attributes. But most theorists reject these options,
for reasons given by Kripke 1980 (see also Chapter 3.1 on Names in this volume).
Gottlob Frege (1892) distinguishes between an expression’s sense and its reference.
(He uses the German terms ‘Sinn’ and ‘Bedeutung’, respectively.) Frege’s notion of
reference is roughly the same as that of extension; his notion of sense corresponds with
the notion of meaning we have discussed so far. But Frege’s theory of sense seemingly
entails that senses are finer-grained than both structured intensions and traditional
meanings. Frege held that if a rational speaker can understand two sentences S and S',
and think that S is true and S' is false, then S and S' differ in sense. A rational speaker
can understand ‘George Orwell is George Orwell’ and ‘George Orwell is Eric Blair’, and
yet think that the first is true and the second is false. So Frege’s theory entails that these
sentences differ in sense. Therefore the Fregean senses of proper names are finer-grained
than possible-worlds intensions, and the Fregean senses of proper names are also finer-
grained than the meanings that certain traditional views attribute to names: namely
those views that identify the meaning of a proper name with its referent. Parallel con-
siderations seemingly show that the Fregean senses of predicates are finer-grained than
possible-worlds intensions and attributes. Consider the sentences ‘All pails are pails’
and ‘All pails are buckets’. The property of being a pail is the same as the property of
being a bucket, and the possible-worlds intension of ‘pail’ is the same as that of ‘bucket’.
Yet it seems that a rational person could understand both of these sentences and think
that the first is true and the second is false. If this is right, then Fregean senses cannot
be identified with possible-worlds intensions or the attributes of traditional theories of
meaning. Philosophers are divided on the plausibility and viability of a theory of fine-
grained Fregean senses. See Kripke (1980) and Salmon (1986) for discussion.
We need even further distinctions, further sorts of semantic entities, and further
semantic vocabulary, to describe a common phenomenon that we have ignored so far:
namely that a single sentence can “say” different things in different contexts. Imagine
that John and Mary both utter the sentence ‘I run’. This sentence is not ambiguous (or
at least it is not ambiguous in the way that ‘bank’ and ‘John is standing next to a bank’
are). Nevertheless, John and Mary assert different propositions, for John says that he
runs while Mary says that she runs (and John does not say that Mary runs and Mary does
not say that John runs). What they say may even differ in truth-value, for it may be true
that John runs, while false that Mary does. In view of this, it is natural to think that the
sentence ‘I run’ has a single meaning, but that it expresses different propositions on dif-
ferent occasions, depending on who utters it. Something similar happens with sentences
containing ‘here’, ‘today’, ‘yesterday’, ‘now’ ‘he’, ‘she’, ‘this’, and ‘that’. Two people who

15
 DAVI D BR AU N

utter a single sentence containing one of these words may very well end up asserting
different propositions.
We can systematize these observations in the following way (as Kaplan 1989 does).
Let us suppose that words and sentences have two kinds of meaning, character and con-
tent. The sentence ‘I run’ has a single character. The character of this sentence is the
meaning that all utterances of it have in common. But this sentence has different con-
tents in different contexts. The content of the sentence in a context is the proposition
that it expresses in that context. The content of ‘I run’ in John’s context is the proposi-
tion that John runs. The content of that same sentence in Mary’s context is the proposi-
tion that Mary runs. The character of the sentence determines what its content will be
in a context. Similar points can be made about single words. The word ‘I’ has a single
character, but varies in content from context to context, depending on who the agent
of the context is. When an expression varies in its content from context to context, we
call it ‘context-sensitive’. Some expressions, such as ‘runs’, have the same context in all
contexts, and so are context-insensitive.
The meanings that we discussed earlier in this chapter should be identified with the
contents of this new view that distinguishes between character and content. The claim
that there are two kinds of meaning, character and content, is compatible with various
views about the nature of contents. If contents are like the meanings of some traditional
theories, then the content of ‘I run’ in John’s context is a structured proposition that
contains John himself and the property of running as constituents. If contents are pos-
sible-worlds intensions, then the content of ‘I run’ in John’s context is a function from
possible worlds to truth-values whose value at any world is truth iff John runs at that
world. Similarly, the content of ‘I’ in John’s context is John himself, if contents are
like the meanings of certain traditional theories; if contents are possible-worlds inten-
sions, then the content of ‘I’ in John’s context is a function whose value at every world
is John. From here on, I will (for the sake of simplicity) assume a traditional theory of
meaning on which contents are individuals, attributes, propositions, and complexes of
such entities.
The character of an expression is usually identified with a function from contexts to
contents, whose value at any context is the content of that expression in that context.
For instance, the character of ‘I’ is a function on contexts whose value at any context is
the content of ‘I’ in that context: namely the agent of the context. Given John’s con-
text, this character yields John himself, and given Mary’s context, it yields Mary herself.
The character of ‘runs’ is a constant function whose value, at every context, is the
property of running. The character of ‘I run’ is a function whose value at every context
is the singular proposition that A runs, where A is the agent of the context.
Typical theories of character (such as that of Kaplan 1989) assume that every context
has its own agent, time, location, and possible world, and also assume that the agent of
a context exists in the world of that context, and is located at the time and place of that
context in that world. A sentence is true at a context C iff its content in C is true in the
world of C. For instance, the sentence ‘I exist’ is true in a context C iff the content of
that sentence in C is true in the world of C. Therefore ‘I exist’ is true in every context,
for the agent of a context always exists in the world of the context. Similarly, ‘I am here
now’ is true in every context. However, the content of ‘I exist’ at a context can be false
at worlds other than the world of the context. For example, the content of ‘I exist’ in a
context C in which Mary is the agent is the proposition that Mary exists; this proposi-
tion is true in the world of C, but is false in some other worlds. Therefore, ‘Necessarily,

16
 E XTEN SI O N , I N TEN SI O N , CH A R A C T E R , A N D B E Y O N D

I exist’ is false in context C, even though ‘I exist’ is true in all contexts. Similarly, ‘I am
here now’ is true at every context, but ‘Necessarily, I am here now’ is false at all contexts
whose agent is a thing that could have been at a different location at the time of the
context.
Some theorists argue that the above theory of character is not sufficiently fine-grained.
Suppose that the content, in a context, of any expression of the form ‘that F’ is the dem-
onstrated object of the context, if that object is F. For instance, the content, in a con-
text, of ‘that dog’ is the demonstrated object of that context, if it is a dog. (This assump-
tion is controversial; see Braun 1994 and King 2001 for discussion.) Now consider the
expressions ‘that dog’ and ‘that dog which is black if it is black’. Anything that is a dog
is also a dog which is black if it is black. So on the preceding theory, ‘that dog’ has the
same content in every context as ‘that dog which is black if it is black’. Therefore, ‘that
dog’ and ‘that dog which is black if it is black’ have the same character, on the above
theory. Yet they seem to differ in meaning in some respect. Some theorists (Braun 1994)
propose to use structured characters to deal with this apparent problem. The structured
character of a complex expression is a structured entity that contains the (functional)
characters of the words in the expression as constituents. For instance, the structured
character of ‘I run’ is a structured entity whose constituents include the character of ‘I’
and the character of ‘run’. ‘That dog’ and ‘that dog which is black if it is black’ differ in
structured character, for the structured character of the latter has the character of ‘black’
as a constituent, whereas the structured character of the former does not.

References
Arnauld, Antoine. 1662/1964. The Art of Thinking: Port-Royal Logic. Translated by James Dickoff and Patri-
cia James. New York: Bobbs-Merrill.
Braun, David. 1994. “Structured Characters and Complex Demonstratives.” Philosophical Studies 74,
193–219.
Carnap, Rudolf. 1956. Meaning and Necessity, second edition. Chicago: University of Chicago Press.
Cresswell, Max. 1985. Structured Meanings. Cambridge, MA: MIT Press.
Frege, Gottlob. 1892. “Über Sinn und Bedeutung,” Zeitschrift für Philosophie und Philosophische Kritik, 100,
pp. 25–50. Translated as “On Sense and Reference.” In Peter Geach and Max Black (eds.), Translations
from the Philosophical Writings of Gottlob Frege. 1952. London: Blackwell.
Hamilton, William. 1860. Lectures on Logic. Boston: Gould and Lincoln.
King, Jeffrey. 2001. Complex Demonstratives: A Quantificational Account. Cambridge, MA: MIT Press.
Kaplan, David. 1989. “Demonstratives.” In Joseph Almog, John Perry, and Howard Wettstein (eds.), Themes
from Kaplan, pp. 481–563. New York: Oxford University Press.
Kripke, Saul. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press.
Lewis, David. 1972. “General Semantics.” In Donald Davidson and Gilbert Harman (eds), Semantics of
Natural Language, pp. 169–218. Dordrecht: Reidel. Reprinted in David Lewis, Philosophical Papers Volume
1, pp. 189–232, 1983. New York: Oxford University Press.
Mill, John Stuart. 1872. System of Logic. London: Longman, Green, Reader, and Dyer.
Montague, Richard. 1974. “Pragmatics.” In Richmond Thomason (ed.), Formal Philosophy: Selected Papers of
Richard Montague, pp. 95–118. New Haven, CN: Yale University Press.
Russell, Bertrand. 1996. Principles of Mathematics. New York: Norton. Originally published in 1903, London:
Allen and Unwin.
Salmon, Nathan. 1986. Frege’s Puzzle. Cambridge, MA: MIT Press.
Soames, Scott. 1987. “Direct Reference, Propositional Attitudes, and Semantic Content.” Philosophical
Topics 15, 47–87. Reprinted in Nathan Salmon and Scott Soames (eds.), Propositions and Attitudes,
pp. 197–239, Oxford: Oxford University Press.
Stalnaker, Robert. 1984. Inquiry. Cambridge, MA: MIT Press.

17
 1.2
SE MANT IC S AND
P RAGMA T IC S
Christopher Gauker

Semantics deals with the literal meaning of sentences. Pragmatics deals with what speak-
ers mean by their utterances of sentences over and above what those sentences literally
mean. However, it is not always clear where to draw the line. Natural languages contain
many expressions that may be thought of both as contributing to literal meaning and
as devices by which speakers signal what they mean. After characterizing the aims of
semantics and pragmatics, this chapter will set out the issues concerning such devices
and will propose a way of dividing the labor between semantics and pragmatics.
Disagreements about the purview of semantics and pragmatics often concern expres-
sions of which we may say that their interpretation somehow depends on the context in
which they are used. Thus:

• The interpretation of a sentence containing a demonstrative, as in “This is nice”,

depends on a contextually determined reference of the demonstrative.
• The interpretation of a quantified sentence, such as “Everyone is present”, depends
on a contextually determined domain of discourse.
• The interpretation of a sentence containing a gradable adjective, as in “Dumbo is
small”, depends on a contextually determined standard (Kennedy 2007).
• The interpretation of a sentence containing an incomplete predicate, as in “Tipper is
ready”, may depend on a contextually determined completion.
• The interpretation of a sentence containing a discourse particle such as “too”, as in
“Dennis is having dinner in London tonight too”, may depend on a contextually
determined set of background propositions (Gauker 2008a).
• The interpretation of a sentence employing metonymy, such as “The ham sand-
wich wants his check”, depends on a contextually determined relation of reference
shifting.
• The interpretation of a conditional sentence, such as “If you turn left at the next
corner, you will see a blue house at the end of the street”, depends on a contextually
determined range of prospects.

I will call devices such as these the in-between phenomena. Insofar as the truth values of
sentences involving in-between phenomena depend on the contextually determined
values of certain parameters, explication of the in-between phenomena seems to belong
to semantics. Insofar as the evaluation of utterances of such sentences depends on the
 SEM A N TI CS A N D P R A G MAT I C S

context of utterance, explication of the in-between phenomena seems to belong to

pragmatics.
Toward dividing the labor between semantics and pragmatics, I will urge that we take
care to distinguish between two questions. One question concerns the conditions under
which a sentence is true relative to a context. That question belongs to semantics. The
other question concerns which context pertains to a given act of speech. That ques-
tion belongs to pragmatics. So characterized, the issue concerns not just how to define
“semantics” and “pragmatics” but how the truth conditions of sentences and the settings
in which sentences are uttered jointly contribute to a successful linguistic exchange.

1 Semantics, Briefly
A semantic theory for a language will specify the literal, conventional meaning of
expressions of the language. Typically, it is assumed that the core of such a specification
of meanings is a specification of extensions. The extension of a proper name might be
the thing named. The extension of a verb phrase might be a set of things that the verb
phrase describes. The extension of a whole sentence is a truth value. The extension of a
quantifier might be a function from sets (sets of sets) to truth values. In simple cases, the
extension of a compound expression will be determined by the extensions of its compo-
nents and its grammatical structure, and, where this is so, a semantic theory is expected
to specify the way in which this is determined. Thus, the truth value of “Socrates is
snub-nosed” will be truth if and only if the extension of “Socrates”, namely, Socrates, is
a member of the set of objects that is the extension of “is snub-nosed”, namely, the set
of snub-nosed things.
Beyond identifying the extensions of expressions, a semantic theory might be
expected to specify the intensions of expressions. Commonly, the intension of an expres-
sion is identified with a function from possible worlds to extensions. Moreover, some
of the expressions of a language may express operations on intensions. Thus, “possibly”
may be thought of as expressing an operation on intensions that takes an intension and
generates the value truth if and only if that intension is a function from worlds to truth
values that generates truth at some world. Propositions are often identified with the
intensions of sentences and are thought of as being, or as determining, functions from
possible worlds to truth values.
Many of the controversies in semantics concern the treatment of specific linguistic
devices within this basic framework. Other controversies have to do with the basic con-
cepts that I have employed in characterizing this framework. For instance, one contro-
versy concerns the use of the notion of possible worlds. Furthermore, one may question
the nature of the relation between expressions and their extensions, which we may call
the reference relation. What makes it the case that “Socrates” refers to Socrates? What
makes it the case that “is snub-nosed” describes the objects in the set of snub-nosed
things? Some philosophers have attempted naturalistic reductions of the reference rela-
tion. Others despair of obtaining a naturalistic reduction and consequently present
ideas about how to formulate a semantic theory without our needing a naturalistically
reducible reference relation.
In general, a semantic theory will relativize the truth conditions of sentences to eval-
uations of various parameters, such as interpretation of the nonlogical lexical items,
model domain, possible world or world-and-time, or, as we will see, context. Various
notions of validity may be defined by picking out a set of these parameters and, while

19
 CH RI STO PH ER G AU K E R

holding the values of the other parameters fixed, saying that an argument (a set of
premises and a conclusion) is valid if and only if for each parameter in that set, if the
premises are all true relative to that parameter, then so is the conclusion. For instance,
we may define one very narrow kind of validity by taking the pertinent set to be the
set of all evaluations of all of the parameters (holding fixed only the interpretation of
the logical vocabulary). We may obtain a much broader kind of validity by keeping the
interpretation, model domain, and world-time fixed and taking the pertinent set to be
only the set of all contexts.
Even theorists who disagree about the basic framework for semantics might nonethe-
less agree that a semantic theory has as one of its aims the characterization of various
kinds of validity. That this is so will be important in what follows, because I will appeal
to such logical properties in trying to decide between different approaches to the in-
between phenomena.

2 Pragmatics, Even More Briefly

Suppose, for the moment, that declarative sentences always literally express complete
propositions relative to a context of utterance. Still, what a speaker means by uttering a
sentence may be more than that the proposition it expresses is true. A speaker may utter
the sentence, “There’s a gas station around the corner”, and if that sentence expresses in
context the proposition that there is a gas station around the corner from place d, where
d is the location at which the speaker is speaking, then one thing the speaker may mean
by uttering that sentence is that there is a gas station around the corner from d. But in
addition the speaker may mean that the gas station is open and has gas to sell. Pragmat-
ics may be conceived as, in part, the study of the way in which speakers manage to mean
more by uttering sentences than what the sentences they utter literally mean relative
to the context in which they are uttered. That, at least, is the conception of pragmatics
that we have inherited from Grice (1989).
Inasmuch as meaning something by an utterance of a sentence is a state of a speaker’s
mind, this way of defining the subject matter of pragmatics assumes that it concerns the
speaker’s state of mind. A different way of approaching pragmatics would be to think of
it as the study of the way in which hearers are able to draw conclusions on the basis of
their understanding of the literal meaning of a speaker’s words and their own knowledge
of the state of the environment in which the speaker’s utterance takes place (Gauker
2001). A hearer who hears a speaker say, “There’s a gas station around the corner”, may
infer from the fact that there is a gas station around the corner and the fact that he or
she is standing at a busy intersection in the middle of the day that probably the gas sta-
tion is open and has gas to sell. The speaker may usually intend the hearer to draw such
inferences, but it is not obvious that the hearer has to think about what the speaker had
in mind in order to do so.

3 The Logic of the In-between Phenomena

Part of the challenge posed by the in-between phenomena is that they introduce logical
data that need to be accounted for. We find arguments that are valid in the sense that
for every context in which the premises are true, the conclusion is true, and other argu-
ments that are invalid in the sense that there is a single context in which the premise is
true and the conclusion is not. So our account of the truth conditions of sentences must

20
 SEM A N TI CS A N D P R A G MAT I C S

explain how the truth value of a sentence relative to a context depends on the content
of the context in such a way that in terms of that explanation we can explain why some
arguments are valid in this sense and others are not.
Imagine a store that sells nothing but unpainted wooden furniture. In conversing
with a customer, the salesman gestures toward the showroom and declares, “Everything
is made of wood”. In the context in which he speaks, we should regard what he says as
true. We would not want to deem it not true, as it would be, for instance, if a plastic
chair were for sale. And yet, the telephone on his desk is not made of wood. The sen-
tence “The telephone is made of wood” is not true in that context (which is not to say
that it is false). So the argument from “Everything is made of wood” to “The telephone
is made of wood” does not preserve truth-in-a-context. But if there were any context in
which “The telephone is made of wood” were true, then in that context “Something is
made of wood” would be true in that context as well.
For another example, in any context in which “Tipper is ready” is true, “Tipper is
ready for something” will surely be true. And yet, there might be a context in which
“Tipper is ready for something” is true, though “Tipper is ready” is not true. Tipper
might be ready to take her next breath but not be ready for the thing that matters in the
context, such as going to the convention center.

4 Approaches to the Middle Ground

4.1 Minimalists
One approach to the in-between phenomena is just to say that, while their contribution
to the proposition that a sentence expresses is independent of the context of utterance,
the propositions expressed are only minimal propositions of a certain sort.
One kind of minimalism is that of Emma Borg (2004, 2007). Borg’s motivation is the
idea that a semantic theory should characterize those aspects of a speaker’s knowledge
of meaning that can be attributed to a cognitive module that deals with language, while
our understanding of contexts cannot be located in any particular cognitive module.
Borg will explain the truth condition of a sentence such as “This is red” as follows
(2004: 192):

(R) If a speaker utters a token of “This is red” and refers with the utterance of
“this” therein to an object x, then that utterance is true if and only if x is
red.

Borg’s theory does not assign to the sentence “This is red” a proposition that it expresses,
but on her theory one can still say that an utterance of “This is red” expresses a proposi-
tion, regarding the object referred to by the utterance of “this” therein, that it is red.
Borg is prepared to allow that in a sentence such as “Tipper is ready” there is a hidden
argument place and a hidden quantification. So she says that “Tipper is ready” has the
same “underlying form” as “Tipper is ready for something” (2007: 350–351).
The minimalism of Cappelen and Lepore (2005) is motivated by the desire to draw a
line. Cappelen and Lepore allow that the proposition that a sentence expresses may be
relative to a context in some ways. A context may be called upon to provide an interpre-
tation for a limited number of items, such as explicit demonstratives, indexicals, words
and inflections indicating tense, and a few others (2005: 1). But they expect that the

21
 CH RI STO PH ER G AU K E R

arguments that might persuade us to countenance varieties of context relativity beyond

those on their list would prove too much, leaving us with no account of what we under-
stand in understanding the meanings of words. They hold that a speaker who utters the
sentence “There are no French girls” (their example, 2005: 41) might thereby state that
there are no French girls in room 401, but what the speaker’s utterance expresses is pre-
cisely the proposition that there are no French girls. A speaker who utters “The students
are ready” might be interpretable as saying that the students are ready for the exam, but
what the speaker’s utterance expresses is precisely the proposition that the students are
ready (2005: 168).
A problem for Borg’s theory is that it is not evident that her truth conditions for
sentences containing demonstratives can be derived from a recursive definition of truth
for a language. For example, the truth condition of “This is not red” ought to be a con-
sequence of (R) and a general fact about the truth conditions of negations. But what is
this general fact about the truth conditions of negations? We could try:

(N) If u is an utterance of a sentence s and v is an utterance of the negation of

s, then v is true if and only if u is not true.

But this would be wrong for two reasons. First, (N) tells us the truth conditions of a
negation only on the condition that the sentence negated has been uttered, though it
is not in general true that if a negation is uttered then the sentence negated has been
uttered. And second, we cannot assume, as (N) does, that the demonstratives in the
utterance of the sentence negated refer to the same objects as the demonstratives in the
utterance of the negation. In place of (N), we might put:

(Nc) If r is the negation of sentence s, then r is true in a context c if and only

if s is not true in c.

But this belongs to a definition of truth-in-a-context, not a definition of truth simplic-

iter, whereas it was Borg’s ambition to avoid the relativization of truth to context.
Further, neither variety of minimalism accounts for all of the logical facts that the
in-between phenomena point to. Neither theory identifies any sense in which the argu-
ment from “Everything is made of wood” to “The telephone is made of wood” might fail
to be valid. Borg’s theory will not explain in what sense “Tipper is ready for something”
fails to imply “Tipper is ready” because she holds that these two sentences have precisely
the same truth conditions. On Cappelen and Lepore’s theory, what we say will depend
on what we say about the proposition that Tipper is ready, but it seems that for anything
that we might say about that, we will end up either affirming that “Tipper is ready for
something” implies “Tipper is ready” or denying that “Tipper is ready” implies “Tipper
is ready for something”, which seem to be wrong results.

4.2 Partialists
A second point of view on the in-between phenomena is that what they show is that
sentences frequently do not express complete propositions at all and that semantic the-
ory cannot aspire to anything more than ascribing to sentences a kind of “gappy” mean-
ing that a speaker can utilize to convey complete propositions to a hearer, the speaker’s
meaning. This is the sort of view advanced by Recanati (2001, 2004), Carston (1998,
2008) and Bach (1994, 2005).

22
 SEM A N TI CS A N D P R A G MAT I C S

The partialists disagree among themselves on several questions. Recanati disagrees

with Carston on certain questions concerning the psychology of interpretation (2004:
39). Carston (2008) does not understand why Bach thinks that the interpretation of
indexicals, such as “I” and “now”, belongs to semantics. And Recanati, in defining
a kind of minimal propositional meaning, sometimes takes into account the hearer’s
point of view (2004: 19); whereas Bach finds no place for such a conception of meaning.
Nonetheless, the partialists all agree in thinking that linguistic convention alone will
usually not get as far as determining for a sentence, or even for a sentence in a context,
a full proposition expressed. What semantics assigns to many sentences is only a proposi-
tion radical (Bach) or linguistically encoded meaning (Carston).
Not only do the partialists not accommodate the special logic of the in-between
phenomena, they are altogether silent on questions of logical relations. Inasmuch
as sentences do not express propositions, not even relative to contexts, we cannot
define logical relations between sentences in terms of relations between the proposi-
tions expressed. Carston, for one, holds that only mental representations, not spoken
sentences, are the proper object of truth-conditional semantics and that they are free
of the in-between phenomena; but to say that is not yet to address the question of
the logic of natural language. One can think of various accounts that one might offer
on the partialists’ behalf. If sentences express some kind of “gappy” meaning, short of
a full proposition, then sentence truth could be relativized to ways of filling the gaps,
and logical relations could be defined in terms of truth relative to ways of filling the
gaps. But, as far as I know, none of the partialists has actually addressed the issue in
print.

4.3 Indexicalists
An influential model for the treatment of the in-between phenomena has been Kap-
lan’s treatment of indexicals such as “I” and “now” (1989). According to Kaplan, the
truth of a sentence must be relativized to both a possible world and a context. A con-
text, for Kaplan, is an object that assigns to each indexical an appropriate referent. For
example, to “now” it assigns a time. Thus, to each expression we can assign a function
from contexts to intensions, which in turn may be thought of as functions from worlds
to extensions. Kaplan called these functions from contexts to intensions characters.
The indexicalist strategy for explicating the in-between phenomena, exhibited in a
series of papers by Stanley (e.g., 2002, 2005), is to treat all of them along the same lines
as indexicals. The in-between phenomena are to be thought of as involving additional,
unspoken lexical items that appear in the deep grammatical structure of a sentence and
receive an interpretation from the context. For example, in Stanley and Szabó 2000,
domains of discourse for quantifiers are supposed to be sets that contexts assign to vari-
ables associated in deep structure with nominal phrases. (I use the term “deep structure”
to refer to mental representations of the grammatical structure of a sentence. Contem-
porary conceptions of this deep structure vary.) Similarly, gradable adjectives, such as
“small”, could be handled by letting context assign to them a standard, such as an aver-
age height (Kennedy 2007). King and Stanley even suggest that the way to handle the
context-relativity of conditionals is to let context assign a similarity relation (between
possible worlds) to the words “if” and “then” (2005: 152).
The indexicalist strategy suggests a clear division of labor between semantics
and pragmatics and also accommodates at least some of the logical facts at issue. A

23
 CH RI STO PH ER G AU K E R

semantic theory for a language will take the form of a recursively specified assignment of
a character, in Kaplan’s sense, to each expression (with contexts now extended as neces-
sary). It will fall to pragmatics to explain what it takes for a given context to be the one
that pertains to a given utterance of a sentence, that is, to be the context such that we
should evaluate the utterance as true (simpliciter) if and only if the sentence uttered
is true relative to that context in the world in which the utterance occurs. The indexi-
calist can even account for some of the special logical facts due to context-relativity,
such as the fact that the argument having “Everything is made of wood” as premise
and “The telephone is made of wood” as conclusion is in a sense not valid, by defining
the pertinent kind of validity thus: an argument is valid if and only if for each context
if the premises are true in that context then the conclusion is true in that same
context.
However, it is doubtful whether the indexicalist strategy can explain all of the logi-
cal facts pertaining to the in-between phenomena. To see this, consider first a sentence
containing two demonstratives, such as:

(1) This is delicious and this is not delicious.

There is nothing contradictory about this sentence, and if it seems odd on the printed
page, that is only because we cannot observe the gestures that would tell us what differ-
ent foodstuffs the two occurrences of “this” denote. To acknowledge that (1) expresses
a truth, we do not have to envision that the context somehow shifts between the evalu-
ation of the first conjunct and the evaluation of the second. Sentence (1) may be true
in a single context because the two occurrences of “this” may denote different objects
in a single context.
By contrast, each of the following sentences is odd in an additional way:

(2) Every student is happy and some student is not happy.

(3) Tipper is ready and Tipper is not ready.
(4) Dumbo is small, and Mickey is large, and Dumbo is larger than Mickey.

There is an oddity in each of these sentences that our semantic theory should enable
us to characterize (Gauker 2010). The oddity cannot readily be characterized as merely
pragmatic. We can perhaps imagine situations in which one of these sentences would
be uttered, but even then the situation is either one in which the speaker has spoken
poorly though we can see what he or she has in mind or one in which the oddity is
deliberately exploited for special effect (for example, to pose a riddle). The contradic-
tory character of (2)–(4) is a problem for the indexicalist because, for the indexicalist,
(2)–(4) will be noncontradictory, just as (1) is, since the context-relative elements in
these sentences should be interpretable independently, just as they are in (1).
Incidentally, it has sometimes been supposed that pure indexicals, such as “I” and
“now”, differ from demonstratives such as “this” in that the former but not the latter
are governed by strict rules such as that “now” always denotes the time of utterance,
and Kaplan himself may have presumed as much. In light of answering-machine cases
and post-it note cases (Predelli 2003), that doctrine has been thrown in doubt. Conse-
quently, it is now widely doubted that there is a special logic of indexicals that would
ensure that the sentence “I am here now” has a special kind of validity.

24
 SEM A N TI CS A N D P R A G MAT I C S

5 A Proposal: Contexts Versus Situations

So far I have followed the common practice of using the term “context” equivocally. I
have treated contexts as parameters relative to which we define the truth of sentences
of a language, and I have spoken of contexts as contexts of utterance, which we may think
of as situations in which an utterance takes place. The first step toward dividing the
labor between semantics and pragmatics in accounting for the in-between phenomena
will be to strictly distinguish between contexts proper and situations. A context will be
a certain formal structure relative to which the truth of sentences will be defined. A
situation will be an arrangement of objects and events such as may include an utterance
of a sentence. Semantics will deal with the definition of truth for sentences relative to
contexts. Pragmatics will deal with the conditions under which a given context is the
one that pertains to a given utterance situation.

5.1 Semantics in Light of this Distinction

Although situations are endlessly complex arrangements of objects and events, contexts
are precisely definable structures. So we have the possibility of precisely defining the
whole set of contexts for a given language. And we have the possibility of defining in a
precise way the conditions under which an arbitrary sentence is true relative to an arbi-
trary context. The logical phenomena pertaining to the in-between phenomena can
then be handled by defining an argument to be in a sense valid if and only if it preserves
truth-in-a-context and defining a sentence to be in a sense contradictory if and only if
it is true in no context.
To accommodate context-relative domains of discourse, we can let a context include
a set of objects, the domain (Gauker 1997b). To accommodate the context-relativity
of gradable adjectives, we can let contexts include, for each gradable adjective, a stand-
ard that an object must surpass in order for the adjective to apply. To accommodate
incomplete predicates, we may include a set of propositions and say that, for example,
“Tipper is ready” is true in a context if and only if there is some activity such that a
proposition to the effect that Tipper is ready for that activity belongs to the contextu-
ally-determined set of propositions and there is no activity such that the proposition
that Tipper is not ready for that activity belongs (Gauker forthcoming). To accom-
modate metonymy we can let contexts include a deferment function that might, for
example, take a ham sandwich as input and yield as output the patron who ordered a
ham sandwich. To accommodate the context-relativity of conditionals we might let a
context include (or even be) a set of prospects (themselves conceived as contexts), so
that we can say that an indicative conditional is true in a context if and only if for each
prospect in (or identical to) the context, if the antecedent is true in that prospect, then
so is the consequent (Gauker 2005).
In light of these characterizations of context and the definitions of logical properties
in terms of truth-in-a-context, we can hope to capture the logical data introduced by
the in-between phenomena. We can say that the argument from “Everything is made
of wood” to “The telephone is made of wood” is invalid, because there is a context c
such that every object in the domain of discourse for c is made of wood but such that the
telephone is not in that domain. The sense in which “Tipper is ready for something”
fails to imply “Tipper is ready” is that there is a context in which the contextually-
determined set of propositions contains the proposition that Tipper is ready for a certain

25
 CH RI STO PH ER G AU K E R

activity, so that “Tipper is ready for something” is true in that context, but also contains
the proposition that Tipper is not ready for a certain other activity, so that “Tipper is
ready” is not true in that context.
To account for the contradictory aspect of (2) we may suppose that the context speci-
fies a global domain not assigned to any particular indexical. In (2), the truth of “Every
student is happy” in a context demands that the domain for the context include only
happy students, while the truth of “Some student is not happy” in context demands that
the domain include at least one student who is not happy. (More work has to be done
to accommodate the kinds of sentences that motivate Stanley, such as “Every student
answered every question” and “Every sailor waved to every sailor”. See Gauker 2010.)
Insofar as a context contains a single contextually determined set of propositions, we
may find that the conditions under which “Tipper is not ready” is true relative to a
context are dual to the conditions under which “Tipper is ready” is true relative to
a context, so that (3) is true in no context. As for (4), a context may specify a single
standard relative to which both “small” and “big” are evaluated, so that if Dumbo is
larger than Mickey, then no matter our contextually determined standard for smallness,
if Dumbo satisfies it, then Mickey will not satisfy our contextually determined standard
for bigness.

5.2 Pragmatics in Light of this Distinction

We may assume that quite generally an utterance of a declarative sentence is true (sim-
pliciter) if and only if the sentence uttered is true relative to the context that pertains
to the utterance. In this light, the question for pragmatics becomes: what is it about the
situation in which an utterance takes place that determines that a particular context is
the context that pertains to that utterance?
Many authors writing about the in-between phenomena, including all of those men-
tioned in Section 4, assume that the shortfall between the semantic properties of the
sentence uttered and what a given utterance of that sentence expresses is made up by
the content of the speaker’s communicative intentions. So long as interpretation is con-
ceived simply as filling the gaps left open by literal meaning, it may be hard to conceive
of any alternative. But once we conceive of the issue for pragmatics as that of identify-
ing the context that pertains to a given utterance, alternatives become conceivable.
One could still insist on the primacy of speakers’ intentions, by holding that nothing
but the speaker’s intention picks out the context pertinent to a speaker’s utterance.
But an alternative that comes into view is that the pertinent context might be fixed by
other features of a situation, such as these: what is perceptually salient in the immediate
environment, the facts that are relevant to the interlocutors in light of their goals, what
it is true or reasonable to say, and anaphoric connections between context-relative ele-
ments in a sentence and prior discourse (Gauker 2008b).
One problem facing the assumption that speaker intention fixes context is that it is
just not clear what the contents of the pertinent speaker’s intentions are supposed to
be. In the case of demonstratives, the answer may be clear enough: the intention that
determines that the pertinent context assigns object o to “that” is the intention to draw
the hearer’s attention to object o. But it is far less clear what the content of the inten-
tion would be that determines the membership of the domain of discourse for the perti-
nent context. It cannot be, for instance, the intention, for some set D, that the sentence
uttered be evaluated relative to a context for which the domain of discourse is D. That

26
 SEM A N TI CS A N D P R A G MAT I C S

account of it would be question begging in that it presumes that we have a prior under-
standing of what it is for a certain set to be the domain of discourse for an utterance.
Another basic question for the assumption that speaker intention fixes context is
how a hearer is supposed to recognize the speaker’s intentions in speaking. Normally,
hearers have little access to what a person has in mind apart from an understanding of
what the person says. But, as the in-between phenomena indicate, hearers normally
have to identify the context pertinent to a speaker’s act of speech in order to figure out
what the speaker has said. So normally, hearers will have to have some means of iden-
tifying the pertinent context without independent access to the speaker’s intention.
The question then becomes whether speakers have sufficient independent access to the
speaker’s intention or whether the speaker’s intention can determine the pertinent con-
text although the hearers normally identify the pertinent context on some other basis.
Finally, one has to give some thought to the mental representations that are the
speaker’s underlying intentions. It is not plausible that every aspect of the meaning of
an utterance that is not written out explicitly in the sentence uttered is somehow writ-
ten out explicitly in the brain. The meanings of mental representations will depend on
contextually determined referents and contextually determined domains of discourse,
much as the meanings of spoken sentences do (Gauker 1997a). So to assume that the
speaker’s intention is what determines the context pertinent to a speaker’s utterance
just puts off explaining how something other than speaker’s intention can determine
the context pertinent to a particular token representation.

6 Beyond the In-between

There are many linguistic phenomena that challenge the core conception of semantics
described in Section 1 although they do not appear to belong to the same family as the
in-between phenomena listed in the introduction. These include anaphoric pronouns,
words that indicate a rhetorical relation between what is being said and prior discourse
(such as “but” and “anyway”), and adjectives whose meanings seem to depend on the
nouns they modify (such “good” in “good actor”). The proposal of Section 5.1 does not
immediately suggest a means of accommodating these phenomena in a compositional, or
recursive, semantic theory. But perhaps it gives us reason to hesitate before giving up on
the idea that that a compositional or recursive theory of suitable semantic values of some
kind, relativized to parameters of some kind, might encompass such phenomena as well.

Related Topics
1.1 Extension, Intension, Character, and Beyond
1.4 Presupposition
1.5 Implicature
1.6 Pragmatic Enrichment and Conversational Implicature
1.10 Context-Sensitivity
1.14 Relevance Theory
3.3 Adjectives
3.6 Anaphora
3.11 Indexicals and Demonstratives
6.1 Philosophy of Language for Epistemology
7.7 Pragmatics and Context: The Development of Intensional Semantics

27
 CH RI STO PH ER G AU K E R

References
Bach, Kent (1994) “Conversational Impliciture”, Mind and Language 9: 124–162.
Bach, Kent (2005) “Context ex Machina”, in Z. Gendler Szabó (ed.) Semantics vs. Pragmatics, Oxford: Oxford
University Press, pp. 15–44.
Borg, Emma (2004) Minimal Semantics, Oxford: Oxford University Press.
Borg, Emma (2007) “Minimalism versus Contextualism in Semantics”, in G. Preyer and G. Peter (eds.)
Context-Sensitivity and Semantic Minimalism: New Essays on Semantics and Pragmatics, Oxford: Oxford Uni-
versity Press, pp. 339–359.
Cappelen, Herman and Ernie Lepore (2005) Insensitive Semantics: A Defense of Semantic Minimalism and
Speech Act Pluralism, Oxford: Blackwell Publishing.
Carston, Robyn (1998) “The Semantics/Pragmatics Distinction: A View from Relevance Theory”, UCL
Working Papers in Linguistics 10: 1–30.
Carston, Robyn (2008) “Linguistic Communication and the Semantics/Pragmatics Distinction”, Synthese
165: 321–345.
Gauker, Christopher (1997a) “Domain of Discourse”, Mind 106: 1–32.
Gauker, Christopher (1997b) “Universal Instantiation: A Study of the Role of Context in Logic”, Erkenntnis
46: 185–214.
Gauker, Christopher (2001) “Situated Inference versus Conversational Implicature”, Noûs 35: 163–189.
Gauker, Christopher (2005) Conditionals in Context, Cambridge, MA: MIT Press.
Gauker, Christopher (2008a) “Against Accommodation: Heim, van der Sandt, and the Presupposition
Projection Problem”, in J. Hawthorne (ed.) Philosophical Perspectives, 22, Philosophy of Language, Oxford:
Blackwell Publishing, pp. 171–205.
Gauker, Christopher (2008b) “Zero Tolerance for Pragmatics”, Synthese 165: 359–371.
Gauker, Christopher (2010a) “Global Domains vs. Hidden Variables”, The Journal of Semantics 27:
298–328.
Gauker, Christopher (forthcoming) “What Tipper is Ready for: A Semantics for Incomplete Predicates”,
Noûs (forthcoming).
Grice, Paul (1989) Studies in the Way of Words, Cambridge, MA: Harvard University Press.
Kaplan, David (1989) “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology
of Demonstratives and Other Indexicals”, in J. Almog, J. Perry, and H. Wettstein (eds.) Themes from
Kaplan, Oxford, Oxford University Press, pp. 481–564.
Kennedy, Christopher (2007) “Vagueness and Grammar: The Semantics of Relative and Absolute Gradable
Adjectives”, Linguistics and Philosophy 30: 1–45.
King, Jeffrey C. and Jason Stanley (2005) “Semantics, Pragmatics, Semantic Content”, in Z. Gendler Szabó
(ed.) Semantics versus Pragmatics, Oxford: Oxford University Press, pp. 111–164.
Predelli, Stefano (2003) Contexts: Meaning, Truth and the Use of Language, Oxford: Oxford University
Press.
Recanati, François (2001) “What is Said”, Synthese 128: 75–91.
Recanati, François (2004) Literal Meaning, Cambridge: Cambridge University Press.
Stanley, Jason (2002) “Making it Articulated”, Mind and Language 17: 149–168.
Stanley, Jason (2005) “Semantics in Context”, In G. Preyer and G. Peter (eds.) Contextualism in Philosophy,
Oxford: Oxford University Press, pp. 221–253.
Stanley, Jason, and Zoltán Szabó (2000) “On Quantifier Domain Restriction”, Mind and Language 15:
219–261.

28
 1.3
L OGICAL F OR M
Kirk Ludwig

Interest in logical form has been driven historically by interest in valid argument forms,
that is, semantic entailment relations between the premises and conclusion of an argu-
ment underwritten by their semantic forms. The pressure to provide an ever deeper and
more general account of valid argument forms has led to a generalization of the notion
of logical form to semantic structure relevant to the evaluation generally of the truth or
falsity of a sentence.
It is a staple of the tradition since Frege that logical form differs, sometimes dra-
matically, from the surface or superficial grammatical form of a sentence. (We return
below to whether logical form may be associated with some further level of syntactical
analysis.) For example, some sentences, like 2 and 3, have multiple readings (“someone
is loved by everyone” and “everyone is such that there is someone he loves”; “relatives
who are visiting can be boring” and “it can be boring visiting relatives”) which should
be assigned different logical forms because the same words contribute differently to
truth conditions on each. Some sentences with the same superficial form, such as 1 and
2, 4–6, 9 and 10, 11 and 12, and 13 and 14, are assigned different logical forms because
they enter into different systematic entailment relations. And some sentences with
superficially different forms, such as 6–8, are assigned the same logical form because
they appear to enter into the same systematic entailments.

1. Mary loves John

2. Everyone loves someone
3. Visiting relatives can be boring
4. The hearth is warm
5. The weather is fine
6. The whale is a mammal
7. Whales are mammals
8. Every whale is a mammal
9. Felix does not bark
10. Pegasus does not exist
11. The president is 6' 1" inches tall
12. The average man is 5' 9" inches tall
13. Jack is a wolf
14. Jack ate a wolf
 KI RK LU D W I G

In the early analytic tradition, the divergence of surface and logical form was asso-
ciated with a divergence between the form of thoughts and the sentences expressing
them. Wittgenstein puts this vividly in his Tractatus Logico-Philosophicus ([1921] 1961:
4.002).

Language disguises thought. So much so, that from the outward form of the
clothing it is impossible to infer the form of the thought beneath it, because the
outward form of the clothing is not designed to reveal the form of the body, but
for entirely different purposes.

Bertrand Russell, in the second of his 1914 Lowell lectures, Our Knowledge of the
External World, defined “philosophical logic” as that portion of logic that concerned the
study of forms of propositions—or, as he called them, “logical forms.” He claimed that

some kind of knowledge of logical forms, though with most people it is not
explicit, is involved in all understanding of discourse. It is the business of philo-
sophical logic to extract this knowledge from its concrete integuments, and to
render it explicit and pure.
([1914] 1993: 53)

The most famous exemplar of this activity is Russell’s Theory of Descriptions (1905),
according to which the logical form of “The King of France is bald” is not that of a sub-
ject–predicate sentence (or even a noun phrase + verb phrase construction—we return
to this divergence later) but a quantificational sentence with internal truth-functional
structure, “There is a king of France and whatever is a king of France is identical with
it and it is bald,” in which the grammatical subject of the original is not treated as con-
tributing any entity to the meaning of the sentence, in contrast to, say, “Louis XIV was
bald” (5, 6, and 12 introduce additional complexities).
In the early twentieth century, especially in the short-lived but influential theory of
Logical Atomism, logical form was directly connected with metaphysics and the philos-
ophy of mind (Russell [1918–19] 1985; Wittgenstein [1921] 1961). The logical form of
sentences was taken to reveal both the forms of possible facts or reality and the thoughts
expressed with them. This interest in logical form is clearly broader than an interest
in inference patterns among sentences induced by the meanings of logical terms. For
Russell, for example, a crucial question about logical form was whether belief sentences
express relations to propositions, a view he held at one point but later rejected (Rus-
sell [1918–19] 1985: 87–8). (Importantly, Russell’s and Wittgenstein’s interest in the
structure of thought blurs the distinction Davidson drew later between investiga-
tion of logical form and conceptual analysis (Davidson [1980] 2001b: 105–6, [1984]
2001c: 31).)
Even apart from Logical Atomism, investigation of logical form is an important com-
ponent of what Strawson (1959) called “descriptive metaphysics,” the project of uncov-
ering the ontology embedded in natural languages, what must exist if the sentences
we endorse are true. Given Quine’s (1948: 33) criterion of ontological commitment,
according to which “a theory is committed to those and only those entities to which the
bound variables of the theory must be capable of referring in order that the affirmations
made in the theory be true,” it follows that one goal of the recovery of logical form is to
reveal the quantificational commitments of natural language sentences (see Davidson

30
 LO G I CA L F O R M

1977; Higginbotham 1993). We will consider a dramatic example below in the event
analysis of action sentences.
One response to the divergence of surface and logical form is to treat propositions or
possible facts as the primary bearers of logical form. Sentences then have the same logi-
cal forms if and only if the propositions or possible facts they express do. This gives us,
though, only a temporary illusion of progress. As Ayer put it,

This would be all very well, if we had any means of determining the logical
forms . . . other than through the grammatical forms of the sentences which
are used to state them. What happens, in practice, is that we decide, on other
grounds, which forms of sentences convey this information most perspicuously,
and that these are not always the grammatical forms in which the sentences are
originally cast.
(1984: 29–30)

Ayer’s last remark points to a popular way of talking about logical form, according to
which the logical form of a natural language sentence is the form of a sentence in a
specially regimented, “ideal,” typically formal language, that translates it—or the forms
of the sentences that translate it if it is structurally ambiguous (Kalish 1952; Harman
1972; Audi 1995: 511–12; Sainsbury 2001). These sentences are sometimes said to be,
and sometimes to give, the logical forms of the originals. The ideal or regimented lan-
guage is to have no ambiguities and to encode in its syntax all differences in the logical
or semantic roles of terms in them. An interpreted formal language for a first-order logic,
designed to make shared validity-relevant semantic features correspond to the construc-
tion of a sentence out of its parts, would be an example. Natural language sentences
have the same form if and only if they are translated by sentences of the regimented
language of the same form.
This is not, however, a satisfactory account, for it leaves us with the question of what
the form of the sentence in the ideal language is (Grandy 1974). If we think of it in
terms of the pattern of terms in it, perhaps the pattern of logical terms, then this makes
the notion of logical form relative to the choice the favored language, and there are
many that would do equally well. Russell’s analysis of “The King of France is bald” can
be rendered into a standard infix logical notation or into Polish or prefix notation (with
“f(x)” for “x is a king of France,” “b(x)” for “x is bald,” and “i(x,y)” in prefix notation for
“x = y”), as in 15 and 16.

15. (x)((f(x) & (y)(f(y) x = y)) & b(x))

16. xKKf(x)yCf(y)i(x,y)b(x)

The logical form of “the king of France is bald” cannot be identified with either, or with
the patterns of the terms in them, since they are distinct and the choice arbitrary. We
can only say what the logical form is relative to a system of regimentation. Yet, the origi-
nal and its translations (if they are) share logical form. Sensitivity to this point explains
why philosophers often talk of such renderings as giving the logical form of the original.
But this leaves us with the question of what it gives and how it is doing it.
The idea that logical form attaches first to propositions points to a resolution, for
propositions (in the Frege–Russell tradition) were introduced to be simultaneously
the objects of thought and the meanings of (declarative) sentences. This suggests

31
 KI RK LU D W I G

looking midway between the sentence and the proposition it expresses to a comprehen-
sive semantic theory that details how words and structures contribute to fixing truth
conditions for sentences.
An influential suggestion of Donald Davidson’s along these lines is that the notion
of logical form is best placed in the context of an interpretive truth theory for a natural
language.

What should we ask of an adequate account of the logical form of a sentence?

Above all . . . such an account must lead us to see the semantic character of
the sentence—its truth or falsity—as owed to how it is composed, by a finite
number of applications of some of a finite number of devices that suffice for the
language as a whole, out of elements drawn from a finite stock (the vocabulary)
that suffices for the language as a whole. To see a sentence in this light is to
see it in the light of a theory for its language. A way to provide such a theory
is by recursively characterizing a truth predicate, along the lines suggested by
Tarski.
(Davidson 1968, [1984] 2001c: 94)

The suggestion in full is that the logical form of a sentence is revealed in the context
of a compositional meaning theory for the language that takes the form of an interpre-
tive Tarski-style truth theory (Tarski [1934] 1983, 1944; Davidson 1967b, 1970, 1973).
Restricting attention to a context-insensitive language, an interpretive truth theory for
an object language L is a finitely axiomatized theory whose axioms provide satisfaction
conditions for the semantically primitive expressions of the language by using metalan-
guage expressions which translate them. With an adequate logic, the theory will entail
all sentences of the form (T), where “s” is replaced by a description of a sentence of L as
constructed out of its semantically primitive components (the need to regiment natural
language sentences introduces a complication which we return to below) and “p” is
replaced by a metalanguage sentence that translates it.

(T) s is trueL iff p

In (T), “is trueL” is a metalanguage truth predicate for sentences of L. Given that what
replaces “p” translates s, we can replace “is trueL iff” with “meansL that” and preserve truth.
A sentence of the form (T) in which this replacement yields a true sentence we call a T-
sentence and we say that it provides interpretive truth conditions for its object language
sentence. (The morals developed extend to a language with context-sensitive elements,
though we cannot take the extension up here. See Lepore and Ludwig 2005: chs. 4–5 for
more detail and the relation of this way of putting the project to Davidson’s own.)
A canonical proof issues in a T-sentence for s by drawing intuitively only on the con-
tent of the axioms for words in s. The canonical proof of a T-sentence shows the seman-
tic role of each primitive expression in the sentence in fixing its interpretive truth con-
ditions, and in this sense shows what its semantic structure is. According to Davidson’s
suggestion, this is to know the logical form of the sentence: “To know the logical form
of a sentence is to know, in the context of a comprehensive theory, the semantical roles
of the significant features of the sentence” ([1980] 2001a: 146).
A simple informal theory, A1–A8, stipulated to be interpretive, will illustrate. The
vocabulary of L consists of the variables, “x,” “x1,” “x2,” . . ., names “Ned” and “Sally,”

32
 LO G I CA L F O R M

the one-place predicate “is honest,” the two-place predicate “loves,” the connectives
“~” and “&” (for negation and conjunction), and an existential quantifier formed
with parentheses around “” followed by a variable, e.g., “(x).” Square brackets will
form a structural description of a sentence. Thus, “[Ned loves Sally]” is equivalent to
“‘Ned’{ ‘loves’{ ‘Sally’”—the concatenation of “Ned,” “loves,” and “Sally.” Similarly,
where “v” and “v'”are metalinguistic variables for variables and names (terms) of L,
“[v loves v']” is equivalent to “v{ ‘loves’{ v',” etc. ‘iff’ abbreviates “if and only if.” We
use functions from terms to objects as satisfiers of formulas. Axioms A1–2 specify assign-
ments to names for the functions we quantify over. For any D, [f(D)] refers to the object
f assigns to D. “I” and “\” take object language formulas as values.

A1. For any function f, f(“Ned”) = Ned.

A2. For any function f, f(“Sally”) = Sally.
A3. For any function f, term v, f satisfiesL [v is honest] iff f(v) is honest.
A4. For any function f, terms v, v', f satisfiesL [v loves v'] iff f(v) loves f(v').
A5. For any function f, for any formula I, f satisfiesL [~I] iff it is not the case that
f satisfiesLI.
A6. For any function f, for any formulas I, \, f satisfiesL [I& \] iff f satisfiesLI and f
satisfiesL \.
A7. For any function f, for any formula I, variable v, f satisfiesL [(v) I] iff for some f'
such that f' is a v-variant of f, f' satisfiesL I.
A8. For any sentence I, I is trueL iff every function f satisfiesL I.

The satisfaction relation generalizes the ‘true of’ relation. If ‘Ned loves Sally’ is true,
then ‘loves’ is true of Ned and Sally taken in that order. Thus, a function f that assigns
Ned to “x” and Sally to “x1” satisfies “x loves x1.” A function f' is a v-variant of f iff f' is
like f except at most in what it assigns to v.
A canonical proof is a sequence of metalanguage sentences such that (i) the last
member is a sentence of the form (T) with no semantic vocabulary on the right hand
side and (ii) each member of it is an axiom or derived from axioms or previously
derived sentences by rules applied so as to permit drawing only on the content of the
axioms.
Axioms A1–4 are base axioms; axioms A5–7 are recursive axioms. A1–2 are refer-
ence axioms for “Ned” and “Sally.” A3–4 are predicate satisfaction axioms, for one- and
two-place predicates. A5–6 provide recursive satisfaction conditions for formulas con-
structed with truth-functional connectives. A7 provides recursive satisfaction condi-
tions for quantified formulas. A8 connects satisfaction with truth of closed sentences.
The form of the axioms may be said to give the role of the object language term in deter-
mining truth conditions of sentences in which it appears and the logico-semantic form
of the vocabulary item for which it is an axiom. Then the forms of the axioms which
are used in the proof of a T-sentence for s and how they are employed reveal the logical
form of s.
A parameter in characterizing semantic form is the classification of axioms. The
framework allows for various classifications. A natural classification for tracing truth-
relevant semantic structure is to treat the singular reference axioms as having the same
form, axioms for predicates with the same number of argument places as having the
same form, and axioms for distinct truth–functional connectives and distinct quantifiers
as having the distinct forms.

33
 KI RK LU D W I G

A proof of a T-sentence for [(x)(Ned loves x) & (x1)(x1 loves Sally)] would instan-
tiate A6 for “&” to this sentence. Then we would instantiate A7 for the existential
quantifier to each of the conjuncts on the right hand side. With A4 for “loves” and
A1 for “Ned” we can then deduce “For any function f, f satisfies [(x)(Ned loves x)] iff
some x is such that Ned loves x” and similarly for [(x1)(x1 loves Sally)] with A4 and
A2. Then instantiating A8 to the sentence, a series of valid substitutions yields the T-
sentence “[(x)(Ned loves x) & (x1)(x1 loves Sally)] is true iff some x is such that Ned
loves x and some y is such that y loves Sally.”
What is the point of the exercise? The proof displays the semantic structure of the
object language sentence in what axioms of what forms are applied at what points in
arriving at a T-sentence for it. The axioms give the type of contribution of the term for
which they are axioms, and how they contribute is given by the place in the derivation
at which it is applied.
The notion of form we arrive at by this method, as Davidson noted, is relative to
both the metalanguage and its logic. We seem to face the same problem as we did ear-
lier in identifying logical form with sentences in a particular ideal language. We want
to abstract away from incidental features of canonical proofs. A suggestion extending
Davidson’s proposal is to fix sameness of logical form of two sentences in their respec-
tive languages in terms of their admitting of corresponding canonical proofs (Lepore
and Ludwig 2002b: 67).

A proof P1 of a T-sentence for s1 in T1 corresponds to a proof P2 for a T-sentence

for s2 in T2

iffdf

(a) P1 and P2 are sentence sequences identical in length;

(b) at each stage of each proof identical rules are used;
(c) the base axioms employed at each stage are of the same semantic type, and the
recursive axioms employed at each stage interpret identically object language
terms for which they specify satisfaction conditions (with respect to contribu-
tions to truth conditions).

In terms of this notion, the suggestion is that (loc. cit.)

For any sentences s1, s2, languages L1, L2, s1 in L1 has the same logical form as
s2 in L2

iffdf

there are interpretive truth theories T1 for L1 and T2 for L2 such that

(a) they share the same logic;

(b) there is a canonical proof P1 of the T-sentence for s1 in T1;
(c) there is a canonical proof P2 of the T-sentence for s2 in T2, such that:
(d) P1 corresponds to P2.

This yields an unrelativized characterization of sameness of logical form. It does not tell
us what thing the logical form of a sentence is but rather when any two sentences in any

34
 LO G I CA L F O R M

languages are the same in logical form. An entity can be introduced using this equiva-
lence relation—the logical form of s1 in L1 = the logical form of s2 in L2 iff s1 in L1 has
the same logical form as s2 in L2—but this provides no additional insight.
This approach can be extended to imperatives and interrogatives by extending truth-
theoretic semantics to fulfillment-theoretic semantics. Sentences have fulfillment con-
ditions of different sorts: truth conditions for declaratives, and compliance conditions
for imperatives and interrogatives. Compliance conditions are spelled out recursively
using the machinery of the truth theory. This is required even for non-declaratives since
they can appear in molecular sentences, and we can quantify into mood markers: e.g.,
“if you are going to the store, buy some milk,” “Invest every penny you earn.” Then the
definition of a canonical proof can be generalized to a fulfillment theory, and likewise
that of a corresponding proof, to generalize the account to nondeclaratives (Lepore and
Ludwig 2002b: 74–6; Ludwig 2003).
Our illustration elides an important stage in the application to natural languages.
Minimally, structural ambiguity and syntactic elision that leaves inexplicit aspects of
how words contribute to truth conditions in natural language sentences require a trans-
lation first into a regimented notation to which the truth theory can be applied (Lycan
1984: ch.1). Much of the work of uncovering the logical form of natural language
sentences is expressed in appropriate regimentation. In practice, regimentations draw
heavily on structures we already know how to incorporate into a truth theory, with the
attendant danger that familiar structures will prove a Procrustean bed for the original
(we review an example below).
Thus, giving a sentence’s logical form with a regimented sentence comes down to
producing a sentence whose semantic structure is presumed (a) to be well understood
and (b) to be the same as that of the original. The property indicated, however, is exhib-
ited only in the light of the semantic theory for the regimented sentence.
What are the constraints on regimentation? An interpretive truth theory requires
that axioms use terms and structures that translate or interpret object language terms
and structures. The meaning of a term or a structure is a matter of the rules for its
use, realized in speakers’ linguistic competencies. These involve both the grammar of
sentences and their interpretation, and are expressed in judgments about grammatical-
ity and entailments and in patterns of usage. Recovering these rules from reflection
on judgments and observations of usage is constrained (i) by the need to incorporate
expressions into a comprehensive semantic and syntactic theory for the language, and
(ii) by the need to take into account the role of pragmatic factors in communica-
tive contexts—for the theory must distinguish responses to sentences based solely on
meaning and responses based in part on what is pragmatically implied by a speaker
(Grice 1989: chs. 1–4). There is no simple general way to describe how to do this,
but two examples will give the flavor of the enterprise and show the power of the
method.
Return first to Russell’s analysis of ‘The king of France is bald’ in 15, repeated here.

15. (x)((f(x) & (y)(f(y)  x = y)) & b(x))

Evidence for this is that the target sentence appears to be true as a matter of meaning if
and only if there is a king of France, there is at most one king of France, and whoever
is king of France is bald. Additional evidence comes from the power of the analysis to
solve certain puzzles. For example, if 17 is a logical truth, an instance of the law of the

35
 KI RK LU D W I G

excluded middle, how do we avoid the conclusion that it is a logical truth that there is
a king of France?

17. The king of France is bald or the king of France is not bald.

The answer on Russell’s theory is that the second disjunct has two readings, as ‘not’ is
treated as modifying the predicate (taking narrow scope) or the whole sentence (taking
wide scope), as shown in 18a and 18b.

18a. (x)((f(x) & (y)(f(y)  x = y)) & ~b(x))

18b. ~(x)((f(x) & (y)(f(y)  x = y)) & b(x))

If we give it the construal in 18a, then 17, though it is committed to there being a king
of France, is not a logical truth, while if we give it the construal in 18b, it is a logical
truth but not committed to there being a king of France. “The king of France is not
bald” is assimilated to “All that glitters is not gold,” which has a true and a false read-
ing as we take “not” to modify “gold” or the whole sentence. “The king of France is not
bald” then contrasts with “Louis XIV is not bald,” where whether “not” modifies the
predicate or the whole sentence makes no difference to its evaluation.
Strawson (1950) objected famously that when there is no king of France, an assertion
of “The king of France is bald” is not false but lacks a truth value because that there is a
king of France is a mere presupposition rather than an entailment of it. For when there is
no king of France, we respond to assertions of “The king of France is bald” not by denying
it but by saying “There is no king of France.” Against this, we feel no hesitation in saying
some assertions of sentences containing a nondenoting definite description are false: for
example, “My father is the present king of France” (Neale 1990: 26–7). The identified
pattern of response, moreover, can be explained on Russell’s theory together with a theory
of conversational pragmatics (Grice 1989: chs. 1–4). If one denies that the king of France
is bald, one will typically be taken to be denying it because one thinks the king of France is
hirsute, since the point of the sentence is to pick out something to ascribe a property to it,
and, thus, attention is focused on the predicate. If one denies it because there is no king of
France or there are two or more, one must be more specific to avoid misunderstanding.
There are other grounds to modify Russell’s account, however. Russell offers a con-
strual in terms of unrestricted quantifiers. The practice goes back to Frege, who rendered
“All philosophers are rich” as “For all x, if x is a philosopher, x is rich” and “Some phi-
losophers are rich” as “Some x is such that x is a philosopher and x is rich.” This intro-
duction of logical connectives can seem unmotivated from the standpoint of syntax.
Reflection on related constructions gives substance to this concern and shows the way
to a resolution. Prima facie, “All philosophers are rich” and “Most philosophers are rich”
should have the same semantic structure, differing only in the quantificational deter-
miner (Neale 1990: 38–44, 2002). However, “Most philosophers are rich” is not logically
equivalent to any first-order construal using unrestricted quantifiers (Rescher 1962). It
is not equivalent to “Most x are such that if x is a philosopher, then x is rich” because
the original is false, while this is true because most things are not philosophers. It is not
equivalent to “Most x are such that x is a philosopher and x is rich” because most things
are neither philosophers nor rich—nor to any other representation using only unre-
stricted quantifiers. “Most philosophers” must be treated a distinct semantic unit, “[Most
x: x is a philosopher],” which restricts the domain to philosophers and requires most to

36
 LO G I CA L F O R M

satisfy the following predicate. But we should then construe “All philosophers are rich” as
“[All x: x is a philosophers](x is rich)” and “Some philosophers are rich” as “[Some x: x is
a philosopher](x is rich)” and “The king of France is bald” as “[The x: x is king of France]
(x is bald).” We can thus retain Russell’s quantificational account of definite descrip-
tions without reading any truth functional structure into our understanding of them.
We see here an example of how regimentation in a familiar notation can distort logico-
semantic form, and how placing the project in the context of a comprehensive semantic
theory for the language can provide a correction. For further discussion, see Neale 1990,
Reimer and Bezuidenhout 2004 and Ludlow (Chapter 3.7) in this volume.
Another celebrated example of the discovery of logical form is the event analysis of
action sentences (Davidson 1967a). The event analysis was introduced to explain the
semantic function of adverbs of action, and in particular to explain modifier drop entail-
ment. In 20, the action verb is modified by four adverbials; 20 entails each sentence
obtained by removing one or more of them, and together with 21 entails 22.

20. Brutus stabbed Caesar [violently] [with a knife] [at the Curia of Pompey]
[on the ides of March]
21. Brutus stabbed Caesar only once.
22. Brutus’s stabbing of Caesar was violent, was done with a knife, at the Curia
of Pompey, on the ides of March.

That these entailments are independent of the particular adverbials and verb shows that
they are a matter of form (this holds of event verbs generally, and of state verbs that
admit of adverbial modification). That 20 together with 21 entails 22 is of particular
interest because 22 contains a description of an event: namely, the event of Brutus’s
stabbing Caesar, and “was violent,” “was done with a knife,” etc., are predicates of the
event described. It can scarcely be an accident that variants of these appear in the adver-
bials in 20. Davidson suggested that these entailments fall into a familiar pattern, which
explains in a uniform way the function of the recurring words in 21–22, if we take 20 to
involve an implicit existential quantifier over events introduced by the action verb, and
the adverbials as contributing predicates of events, as in 23 (we set aside tense for this
abbreviated discussion; see Lepore and Ludwig 2002a for a quantificational treatment).

(e)(stabbing(e, Brutus, Caesar) & violent(e) & with(e, a knife) & at(e, the
Curia of Pompey) & on(e, the ides of March)).

Modifier drop entailment is then an instance of conjunction elimination in the scope of

an existential quantifier; 22 is analyzed as in 24, and so “violent,” “with a knife,” etc., are
exhibited as playing the same semantic role in each of these sentences, and the formal
entailment of 22 by 20 and 21 is made transparent.

24. [the e: stabbing(e, Brutus, Caesar)](violent(e) & with(e, a knife) & at(e,
the Curia of Pompey) & on(e, the ides of March))

In a comment on Davidson’s suggestion, Casteñeda suggested separating out the role

of agent and object (or patient) of the action as separate conjuncts to accommodate
entailments like those between 25 and 26 on the one hand, and 25 and 27 on the other
(1967).

37
 KI RK LU D W I G

25. He flew the spaceship (e)(agent(e, he) & object(e, the spaceship) &
flying(e))
26. He did something (e)(agent(e, he))
27. The spaceship flew (e)(object(e, the spaceship) & flying(e))

This has become a standard feature of the event analysis, now the dominate view of
the semantics of adverbial modification. (See Ludwig 2010 and Graff Fara and Schein’s
chapters (2.8 and 3.9) in this volume for further discussion and refinements.)
The event analysis shows vividly the relation between logical form and descriptive
metaphysics. If the event analysis is correct, then, as we are committed to the truth of
action sentences, we are committed to the existence of events, though there are no
overt quantifiers or terms for events in them.
Let us return now to relate this conception of logical form to formally valid argu-
ments. The truth-theoretic (or fulfillment-theoretic) account characterizes logical form
in terms of semantic structure. In virtue of the meaning of a sentence with a certain
semantic structure (perhaps on a reading), its truth may require the truth of another, as
in the case of modifier drop entailment above. Similarly for a set of sentences and one
or more sentences. The truth theory does not state these relations. However, the logic
of the truth theory, which is needed to carry out proofs of T-sentences, if fully adequate
to the meanings of the terms and structures over which it operates, will determine what
entailments there are based on the forms of sentences. Thus, the truth theory together
with an adequate logic will provide an account of the formal entailment relations among
object language sentences.
What is the relation of this to logical validity? An argument is logically valid on
the standard account if no uniform reinterpretation of its nonlogical elements, holding
fixed the interpretations of its logical terms, makes its premises true without making its
conclusion true. Logical validity is therefore relative to a division of terms into logical
and nonlogical. However, there is no settled view on the correct extension of “logical
constant.” While there is general agreement on the requirement that logic be topic
neutral, and on the classification of certain terms, there are various ways of develop-
ing the idea that lead to different extensions. (See Gomez-Torrente 2002 and MacFar-
lane 2010 for an overview.) Consequently, there is no settled view on the extension of
‘logically valid’.
However, no matter what the criterion, logical validity, if understood in terms of the
notion of a logical constant, is narrower than that of formal validity. For example, 28
entails 29. This is intuitively a formally valid argument, but neither 28 nor 29 contain
any logical terms.

28. Brutus is an honorable man

29. Brutus is a man
30. Brutus is honorable and Brutus is a man

Of course, the representation of the logical form of 28 in 30 contains a logical constant.

But this is to say that 28 and 30 share logical form on the criterion introduced, not that
30 contains any logical terms. On this ground, Evans (1976) distinguished between log-
ical and structural consequences. Logical consequences hinge on the presence of logical
terms, structural consequences on the patterns of types of terms in them. Entailments
like that from 28 to 29 are structural but not logical. Still, this distinction does not seem

38
 LO G I CA L F O R M

significant from the standpoint of semantic theory. That the pattern is created in the
one case by the use of a term and in the other by arrangement of categories of terms
seems a matter of what device is used to subserve a purpose. Thus, it seems a matter of
terminology whether we speak of logical syntax as a semantic structure characterized
indifferently by a pattern formed partly without or entirely without particular terms, or
of formally valid arguments, and distinguish them, as Evans does, into those valid in vir-
tue of a pattern created by certain terms, or a pattern created by a structure in the types
of expressions used (Lepore and Ludwig 2002b: sec. IV and appendix C).
Returning to logical constants, there seems little point in insisting that one of the
competing criteria in the literature identifies the true coin of logic. Each refines the
initial observation that there are terms that form salient patterns relevant to valid-
ity. When we are clear about what distinctions each draws, nothing further is revealed
about the machinery of language by the choice of what terms to bring under the heading
“logical.” In a comprehensive truth or fulfillment-theoretic account of the language,
the role of each expression and structure, whether classified as logical or not, is fully
revealed. The choice lies with the purposes for which the distinction is deployed.
A final topic is the relation of nonovert levels of syntactic description to logical form.
The first suggestion of an alignment between the theory of syntax and logical form
was made by Gilbert Harman (1975), who suggested that deep structure in transforma-
tional grammar could be identified with logical form. More recently, with changes in
the Chomskian program, it has been suggested that syntactic representations at a level
called “Logical Form” or “LF,” distinct from surface structure and deep structure, might
be identified with the logical forms of sentences (Neale 1993). LF is a level of represen-
tation of syntax in Chomskian grammar that provides an interface with a semantics for
the language in the sense that the LF representation makes explicit whatever is relevant
to semantic interpretation, such as quantificational structure (May 1985, 1987, 1989).
This motivates calling it “Logical Form.”
That there are levels of syntactic representation, how many, and how realized, is an
empirical issue (many linguists, following Chomsky (1995), now posit only Phonetic
Form and LF). As usually understood, a level of syntactic representation, distinct from
surface form, is a psychologically real level of representation by speakers over which
rules of syntax are defined. It is conceivable that the hypothesis of LF be shown to be
empirically inadequate. This would not, however, be a reason to say that sentences did
not have logical form. The notion of the logical form of a sentence, then, is not to be
analyzed in terms of LF. We can also note that our earlier argument against identifying
logical form with a sentence in a formal language which makes explicit its semantic
structure, relative to a semantic theory, applies here as well. Still, LF is clearly relevant
to logical form. If LF is real, then a psychologically real level of syntactical description
encodes structural semantic features of sentences. This provides a representation that
can guide regimentations that serve as input to an interpretive truth theory and an
important empirical constraint on them (see e.g. Hornstein 2002). Work on LF, then,
if LF has the relevant properties, will interact in a straightforward way with a theory of
logical form for natural languages, and vice versa.

Related Topics
1.2 Semantics and Pragmatics
1.8 Compositionality

39
 KI RK LU D W I G

1.10 Context-Sensitivity
2.2 Theories of Truth
2.8 Event Semantics
3.4 Quantifiers and Determiners
3.7 Descriptions
3.9 Adverbs
3.11 Indexicals and Demonstratives
4.3 The Role of Psychology
4.6 The Role of Intuitions
4.2 The Role of Linguistics
6.3 The Philosophy of Language for Metaphysics
7.4 Frege, Russell, and Wittgenstein.

References
Audi, R. (ed.) (1995) The Cambridge Dictionary of Philosophy, Cambridge: Cambridge University Press.
Ayer, A. J. (1984) Philosophy in the Twentieth Century, New York: Vintage Books.
Casteñeda, H.-N. (1967) “Comments on Donald Davidson’s ‘The Logical Form of Action Sentences’,” in N.
Rescher (ed.) The Logic of Decision and Action, Pittsburgh: University of Pittsburgh Press.
Chomsky, N. (1995) The Minimalist Program, Cambridge: MIT Press.
Davidson, D. (1967a) “The Logical Form of Action Sentences,” in N. Rescher (ed.) The Logic of Decision and
Action, Pittsburgh: University of Pittsburgh. Reprinted in (Davidson 2001b).
——. (1967b) “Truth and Meaning,” Synthese, 17: 304–23. Reprinted in (Davidson 2001c).
——. (1968) “On Saying That,” Synthese, 19: 130–46. Reprinted in (Davidson 2001c).
——. (1970) “Semantics for Natural Languages,” in Linguaggi nella Societa e nella Tecnica, Milan: Comunita.
Reprinted in (Davidson 2001c).
——. (1973) “In Defense of Convention T,” in Truth, Syntax and Modalit, Amsterdam: North-Holland
Publishing Company. Reprinted in (Davidson 2001c).
——. (1977) “The Method of Truth in Metaphysics,” Midwest Studies in Philosophy, 2: 244–54. Reprinted in
(Davidson 2001c).
——. (2001a) “Comments and Replies,” in Essays on Actions and Events, New York: Clarendon Press.
Original publication, 1980.
——. (2001b) Essays on Actions and Events, 2nd ed., Oxford: Clarendon. Original publication, 1980.
——. (2001c) Inquiries into Truth and Interpretation, 2nd ed., New York: Clarendon Press. Original publica-
tion, 1984.
Evans, G. (1976) “Semantic Structure and Logical form,” in J. McDowell and G. Evans (eds.) Truth and
Meaning, London: Oxford University Press.
Gomez-Torrente, M. (2002) “The Problem of Logical Constants,” The Bulletin of Symbolic Logic, 8 (1):
1–37.
Grandy, R. (1974) “Some Remarks about Logical Form,” Noûs, 8 (2): 157–64.
Grice, P. (1989) Studies in the Way of Words, Cambridge: Harvard University Press.
Harman, G. (1972) “Logical Form,” Foundations of Language, 9: 38–65.
——. (1975) “Deep Structure as Logical Form,” in D. Davidson and G. Harman (eds.) Semantics of Natural
Language, Dordretch: Reidel.
Hornstein, N. (2002) “A Grammatical Argument for a Neo-Davidsonian Semantics In Logical Form and
Language,” in G. Preyer and G. Peter (eds.) Logical Form and Language, Oxford: Clarendon, 345–64.
Higginbotham, J. (1993) “Grammatical Form and Logical Form,” in J. Tomberlin (ed.) Philosophical Perspec-
tives: Language and Logic, Atascadero: Ridgeview.
Kalish, D. (1952) “Logical Form,” Mind, 61: 57–71.
Lepore, E., and K. Ludwig (2002a) “Outline of a Truth Conditional Semantics for Tense,” in Q. Smith (ed.)
Tense, Time and Reference, Cambridge: MIT Press.
——. (2002b) “What Is Logical Form?,” in G. Preyer (ed.) Logical Form and Language, Oxford: Clarendon
Press.
——. (2005) Donald Davidson: Meaning, Truth, Language, and Reality, Oxford: Oxford University Press.

40
 LO G I CA L F O R M

Ludwig, K. (2003) “The Truth about Moods,” in G. Preyer, G. Peter and M. Ulkan (eds.) Concepts of Mean-
ing: Framing an Integrated Theory of Linguistic Behavior, Dordrecht: Kluwer Academic Publishers.
——. (2010) “Adverbs of Action,” in T. O’Connor and C. Sandis (eds.) Blackwell Companion to the Philoso-
phy of Action, Oxford: Wiley-Blackwell.
Lycan, W. (1984) Logical Form in Natural Language, Cambridge: MIT Press.
MacFarlane, J. (2010) “Logical Constants,” in E. Zalta (ed.) The Stanford Encyclopedia of Philosophy
May, R. (1985) Logical Form: Its Structure and Derivation, Cambridge: MIT Press.
——. (1987) “Logical form as a Level of Linguistic Representation,” in E. LePore (ed.) New Directions in
Semantics, London: Academic Press.
——. (1989) “Interpreting Logical Form,” Linguistics and Philosophy, 12: 387–435.
Neale, S. (1990) Descriptions, Cambridge: MIT Press.
——. (1993) “Logical Form and LF,” in C. Otero (ed.) Noam Chomsky: Critical Assessments, London:
Routledge.
——. (2002) “Abbreviation, Scope, Ontology,” in G. Preyer and G. Peter (eds.) Logical Form and Language,
Oxford: Clarendon.
Quine, W. V. O. (1948) “On What There Is,” Review of Metaphysics, 2: 21–38.
Reimer, M., and A. Bezuidenhout. (2004) Descriptions and Beyond, Oxford: Oxford University Press.
Rescher, N. (1962) “Plurality Quantification,” Journal of Symbolic Logic, 27: 373–4
Russell, B. (1905) “On Denoting,” Mind, 14: 479–93.
——. (1985) The Philosophy of Logical Atomism, La Salle: Open Court. Originally published 1918–19 in four
consecutive issues of The Monist, from a transcript of lectures delived Jan.–Mar. 1918.
——. (1993) Our Knowledge of the External World: As a Field for Scientific Method in Philosophy, New York:
Routledge. Original publication, 1914.
Sainsbury, M. (2001) Logical Forms, 2nd ed, Oxford: Blackwell.
Strawson, P. F. (1950) “On Referring,” Mind, 59: 320–44.
——. (1959) Individuals: An Essay in Descriptive Metaphysics, London: Methuen.
Tarski, A. (1944) “The Semantic Conception of Truth and the Foundations of Semantics,” Philosophy and
Phenomenological Research, 4: 341–76.
——. (1983) “The Concept of Truth in Formalized Languages,” in Logic, Semantics, Metamathematics, India-
napolis: Hackett Publishing Company. Original publication, 1934.
Wittgenstein, L. (1961) Tractatus Logico-Philosophicus. Edited by D. F. Pears and B. F. McGuinness, London:
Routledge & Kegan Paul Ltd. Original publication, 1921.

41
 1.4
P R E S UP P O SIT ION
Paul Dekker

1 Introduction
A presupposition is normally a proposition that is or has to be taken for granted, or ‘sup-
posed’, before (‘pre-’) a certain statement or other linguistic act can be made sense of.
A presupposition is not so much the ground you stand on, or the thinking that you rely
on, but more like what comes before the ground you stand on, or what comes before the
thinking that you rely upon. The notion of presupposition is therefore in essence quite
a philosophical subject. By the same token, presuppositions are things which one can
try to communicate, without actually stating them. “Has Alfred stopped smoking pot?”
Even if you do not know the answer to a question like this, it can be used to communi-
cate, not question, that Alfred used to smoke pot.
Notwithstanding, the term has gotten its main import as a decent technical term, in
philosophy and linguistics, and that is how we will mainly deal with it in this chapter.
We will not go into the rhetorical uses, and abuses, of the concept. In this introduc-
tory section we will speak of the prehistory of the subject and in the remainder we fol-
low, roughly, the historical development. In section 2, the original logico-philosophical
interests in and rejections of presupposition are discussed. In section 3, we give a short
overview of the major systematic treatments of presupposition in logic and linguistics in
the 1970s. In section 4, we focus on more pragmatic and dynamic theoretical treatments
of presuppositions that evolved from that, and on some of the most debated issues in this
area. In section 5 we very briefly discuss some recent developments.
While the subject of presupposition is as old as the phenomenon of speaking and
thinking, it has not become known as such until rather recently. Some idea of presup-
position has been commonly attributed to the philosopher Aristotle, although it seems
he has never identified it as such. One of the standard (Aristotelian) logical insights is
that both of the following statements:
(1) All eudeontologists are herbivors.
(2) No eudeontologists are herbivors.
carry the entailment that there are eudeontologists. And so do the following
statements:
(3) Not all eudeontologists are herbivors.
(4) Some eudeontologists are herbivors.
While the examples (1) and (3) seem to be contradictory, as do the examples (2)
and (4), they all four seem to entail that there are eudeontologists. If there were no
 PRESU PPO S I T I O N

eudeontologists, all four examples would be deemed unfortunate. The conclusion then
seems to be that a sentence can fail to be true, and that also its negation fails to be true,
so that there is a gap which has to be bridged for the sentence to be true or false. The gap
that has to be filled is the sentence’s presupposition. Only if the sentence’s presupposi-
tions are satisfied, will we be able to say that the sentence is true or false.
Besides the above-mentioned existential presupposition, to Aristotle (384–322 bc)
may be attributed the concept of categorial presupposition as well. Certain predicates
cannot be predicated of all kinds of substances, because they are of the wrong type
(Aristotle 1975, 2b8–2b22). Nevertheless, the concept of presupposition was not rec-
ognized as such by Aristotle. He has been rather explicit that there is this principle of
the excluded third, which means that a sentence is either true or false, or at least that its
affirmation or its negation must be true. (Aristotle 1975, 17a26ff.)
It has been said that Aristotelian logic has remained unchallenged for 2.5 millen-
nia, and that the concept of presupposition had not been identified over the very
same period in the Western tradition. Some possible and notable exceptions are phi-
losophers and theologicians like Eubulides, Thomas Aquinas, and Petrus Hispanus.
Eubulides (fourth century bc) was one of Aristotle’s contemporaries and antagonists.
He proposed the subject of presupposition as a challenge to Aristostelian logic in a
form which has become famous as the horned man. “Another was, if you have not lost
a thing, you have it; but you have not lost horns; therefore, you have horns” (Laertius
1853, Book VII, chapter “Life of Chrysippus”, section XI); see also Laertius 1853,
Book II, chapter “Life of Euclides”, section IV). This reads like a prehistorical practi-
cal joke, but it was a severe problem for Aristotelian logic. If we, as Aristotle seemed
to do, endorse the principle of bivalence, at least one of the following sentences must
be true:

(5) You have lost your horns.

(6) You have not lost your horns.

Both sentences, however, seem to entail that you have or had horns. If the principle
of bivalence holds, one of the sentences must be true, and, therefore, logic tells us
that you have or had horns. Since we could have embedded any kind of presupposi-
tion beside you having horns, this means that logic would entail any proposition, a
pretty disastrous result. (Seuren 2005, following Kneale & Kneale 1962, claims that,
with six other challenges, Eubulides has defined the agenda of twentieth century
semantics.)
Thomas Aquinas, a thirteenth century theologist and philosopher, apparently
adopted a metaphysical or epistemological concept of presupposition, though clearly
not a linguistic one: “As a result, every science has presuppositions which the learner
must believe” (Aquinas 2000–2009, Faith, Reason, and Theology, Questions I–IV of
the Commentary on Boethius’ De Trinitate) “For in every science, the nature of its
object is presupposed” (Aquinas 2000–2009, “Summa Theologiae, Treatise on Sacred
Doctrine”). Larry Horn (Horn 1985; Horn 1996), and many others following him, have
credited Pedro Julião (a.k.a. Petrus Hispanus, or Pope John XXI, also thirteenth cen-
tury) with the acknowledgment of some notion of presupposition as opposed to a notion
of denotation. However, Pieter Seuren (Seuren 2005) has pointed out that not only the
reference to Peter of Spain is mistaken but also that there is no mention in the text of a
notion of presupposition as we understand it nowadays.

43
 PAU L DEK K E R

2 Logico-Philosophical Treatments
The concept of presupposition as we understand it nowadays was first identified, for-
mally, by the end of the nineteenth century, in the work of Gottlob Frege (1848–1925),
in an analytical, and rather positivist, frame of mind. It appears he first spotted the prob-
lem in his Foundations of Arithmetic. “Der Ausdruck ‘der grösste achte Bruch’ hat z. B.
keinen Inhalt, weil der bestimmte Artikel den Anspruch erhebt, auf einen bestimmten
Gegenstand hinzuweisen” (Frege 1884, p.87–8). (“The expression ‘the largest real frac-
tion’ has, for instance, no content, because the definite article raises the claim to refer to
a definite object.”) (Note, the translations from German are mine, PD.) In his ground-
breaking “Über Sinn und Bedeutung” Frege literally employed the terms “Vorausset-
zung”, [i.e., presupposition]: “. . . wir setzen eine Bedeutung voraus. . . . Nun können wir
uns in jener Voraussetzung freilich irren, und solche Irrtümer sind auch vorgekommen”
(Frege 1892, pp. 31–2) (We presuppose a reference. . . . We can easily be mistaken about
such presuppositions, and such mistakes have occurred.) “Wenn man also behauptet,
‘Kepler starb im Elend’, so ist dabei vorausgesetzt, daß der Name “Kepler” etwas bez-
eichne; aber darum ist doch im Sinne des Satzes “Kepler starb im Elend” der Gedanke,
daß der Name “Kepler” etwas bezeichne, nicht enthalten. (. . .) Daß der Name “Kepler”
etwas bezeichne, ist vielmehr Voraussetzung ebenso für die Behauptung “Kepler starb
im Elend” wie für die entgegengesetzte.” (Frege 1892, p. 40) (If one asserts, “Kepler
died in misery,” it is presupposed that the name “Kepler” denotes something; but the
thought that the name “Kepler” denotes something, is therefore not contained in the
sense of the sentence “Kepler died in misery”. . . . It is more a presupposition for the
assertion “Kepler died in misery” that the name “Kepler” denotes something, as like-
wise for its opposite.) With the logicistic ideal, Frege was not particularly fond of the
possibility of presupposition failure, and called it a “Mangel”, a deficiency, of natural
language: “Dies liegt also an einer Unvollkommenheit der Sprache, von der übrigens
auch die Zeichensprache der Analysis nicht ganz frei ist (. . .). (Frege 1892, p. 41) (This
is to blame a deficiency of language, from which also the analytic language is not free.)
In mathematics, logic, and science, propositions or sentences ought to be true or false
and presupposition failure prohibits them from being so. In this, Frege was both ideal-
istic and realistic about the inherent deficiencies of natural language. In his legacy, in
a note titled “Meine grundlegenden logischen Einsichten”, dated 1915, we read: “Die
logische Arbeit ist gerade zu einem grossen Teil ein Kampf mit den logischen Mängeln
der Sprache, die uns doch wieder ein unentbehrliches Werkzeug ist.” (Frege 1969,
p. 272) (Most of the logical work consists in a struggle with the logical defects of lan-
guage, which, again, is an indispensable tool.)
Bertrand Russell, and Ludwig Wittgenstein, also seem to have realistically acknowl-
edged the problem of presupposition, but idealistically sought to exclude it, either in a
reformation of natural language (Russell), or even in the logical analysis of language
itself (Wittgenstein). Bertrand Russell (1872–1970) is famous, for his theory of descrip-
tions (1905), which seems to have been the dominant theory until Peter Strawson (on
whom more in section 4) attacked it in 1950. A couple of years later, in 1957, Russell,
quite irritated, replied:
“This brings me to a fundamental divergence between myself and many phi-
losophers with whom Mr. Strawson appears to be in general agreement. They
are persuaded that common speech is good enough not only for daily life, but
also for philosophy. (. . .) He [Strawson] admits that the sentence [“the King

44
 PRESU PPO S I T I O N

of France is wise”] is significant and not true, but not that it is false. This is a
mere question of verbal convenience. He considers that the word “false” has an
unalterable meaning which it would be sinful to regard as adjustable, though
he prudently avoids telling us what this meaning is. For my part, I find it more
convenient to define the word “false” so that every significant sentence is either
true or false.” (Russell 1957, pp. 387–9)

Despite these logical and logical positivist reservations, logicians like Jan Łukasiewicz
(in 1920), Dmitri A. Bochvar (in 1937), and Stephen Cole Kleene (in 1938, 1952),
among many others, have studied many-valued logics, which explicitly recognize the
systematic possibility that a sentence be neither true nor false, or be assigned another
value than those two. Bochvar proposed the third value to be understood as “unin-
terpretable,” or “nonsense.” The consequence of this being that if part of a sentence
is rendered nonsense, then the interpretation of the whole crashes and it is rendered
nonsense as well. Bochvar’s interpretation has a very intuitive understanding, and it has
interesting mathematical and computational applications, but the basic idea, under-
stood as an account of presupposition (which it was not intended to be), falls short when
we start considering more involved, quantified, or modal structures. It seems to gener-
ate the prediction that an existentially quantified sentence will fail to be true or false if
the sentence is uninterpretable under one of the assignments to the variable quantified
over. Thus, “Someone’s bike is stolen” would not need to be true if someone’s bike is
stolen, but it would be rendered uninterpretable if someone ‘failed’ a bike.
Bochvar’s system has also become known as a weak Kleene logic, in contradistinction
to a strong Kleene logic, presented by Kleene himself. In Kleene’s (strong) system, the
third value is interpreted as “undecided” or “to be determined” and this makes a con-
ceptually different interpretation of the connectives. If, in this system the values of one
subordinate formula are not yet determined, we may still be able to assign a determinate
value to the whole. So, for instance, if one conjunct of a conjunction is known to be
false, we can establish that the whole conjunction will come out false, no matter what
the other conjunct has as a value. Likewise, if one member of a disjunction is true, the
whole can be known to be true as well. Obviously, a strong Kleene logic, like a weak
Kleene one, agrees with a classical propositional logic, as long as the embedded formulas
are all defined, or decided. But it also allows for determinate truth values for formulas,
one of whose constituent formulas is not (yet) determined. Conceptually, and also com-
putationally, this is an attractive feature. However, it is not obvious how to generalize
this intuition when we start considering more involved, quantified or modal structures.
It seems to generate the prediction that an existentially quantified sentence may come
out not to be false, even if it is undefined for many, if not for all, of the assignments to
the variable quantified over. Thus, “Someone’s bike is stolen” would not be false if none
of those who are ‘known’ to have a bike, have their bike stolen.
Sure enough, multi-valued approaches need not be implemented and generalized as
suggested above. Recent work, by, for instance, Danny Fox (Fox 2008), builds in a
sophisticated and empirically very interesting method on that tradition. Some of the
work goes back to (Soames 1989), which convincingly shows that presuppositions are
not symmetrically projected in coordinations. The presuppositions of a second conjunct
or disjunct are not automatically those of a whole conjunction or disjunction. Asym-
metric truth tables to that effect have already been proposed by Stanley Peters (Peters
1977). The presuppositions of a second conjunct may get satisfied by those of the first,

45
 PAU L DEK K E R

in an asymmetric way; the presuppositions of a conditional consequent clause may get

satisfied by the material in the antecedent clause of the conditional; and a second dis-
junct presupposition may get satisfied by the denial or negation of the first. As a matter
of fact, Peters’ and Soames’s truth tables predate some of the dynamic semantic insights
which will be discussed in section 4.

3 The Linguistic Decade

Before such a dynamic or pragmatic turn could take over, however, the phenomenon
had to be studied in much more detail, and much more extensively. Here is a little
piece of discourse, filled with a couple of standard presupposition triggers, most of them
identified in the 1970s.

(7) Pete found the lost pack of cigarettes. He managed to hide it from his parents.
Ron realized what a good boy Pete was. All the kids thought they were alone.
harry smoked. It was Bert who lit the cigarette. Mary smoked, too. No, only
boys smoked. Amelia had stopped smoking.

The first sentence presupposes that there was some special pack of cigarettes. The sec-
ond presupposes that there is a referent for the pronoun “it,” apparently the pack of
cigarettes; the pronoun “he”, apparently refers to Peter (who has parents) and who
first tried, and then succeeded in hiding the cigarettes from them. The third sentence
(counterfactually) presupposes that Pete was a good boy. The fourth presupposes that
there was a definite group of kids being talked about. The fifth presupposes that at least
someone smoked. The sixth is that someone lit the fatal cigarette. The seventh presup-
poses that somebody else beside Mary smoked, and the eighth that boys smoked. The
last sentence presupposes that Amelia has been smoking earlier.
The presuppositions above are triggered in quite a different number of ways, and peo-
ple disagree as to whether we can speak of a unitary phenomenon. See (Seuren 1985;
Zeevat 1992) for critical discussion. Another problem, however, has been how triggered
presuppositions project from embedded positions. A ground-breaking paper on this was
produced by Lauri Karttunen (Karttunen 1973), and an excellent overview of this prob-
lem is given in (Soames 1989). Karttunen introduced the terminology of ‘plugs’, ‘holes’,
and ‘filters’. So-called holes are sentential operators which inherit the presuppositions
of the sentences they are applied to. Thus, all of the following constructions seem to
come with the presupposition that Alfred has a niece:

(8) Alfred didn’t meet his niece.

(9) Maybe he managed to make an appointment with his niece.
(10) But if he made an appointment with his niece, she didn’t show up.

The general idea is that all three sentences entail that Alfred has a niece, even though it
is not asserted as such. The conclusion is that “not,” “maybe,” and “if” are holes, which
let presuppositions ‘speak up’. This is quite different from so-called plugs which are
claimed to literally block presuppositions.

(11) Alfred said he did meet his niece.

(12) He really believes he ate ice with his niece.

46
 PRESU PPO S I T I O N

Even though these sentences strongly suggest that Alfred has a niece, one cannot be blamed
for speaking falsely if Alfred fails a niece. This means that a presupposition of a sentence S
need not be a presupposition of a sentence say that S, or believe that S, and it has therefore
been claimed that ‘saying that’ and ‘believing that’ are presupposition plugs. Presupposi-
tion ‘filters’ are the most difficult and disputed terms. If . . .-clauses are typical examples
of filters, which sometimes do and sometimes do not inherit the presuppositions of the
consequent clauses. Compare the following two sentences.

(13) If Ben marries Lisa, his sons will be in good hands.

(14) If Ben has children with Lisa, his sons will be in good hands.

Most people agree that the first, but not the second, presupposes that Ben has sons.
The conclusion seems to be that an if-clause may help to filter out a presupposition of
a superordinate clause. The data are very much disputed, as will probably appear from
the next section.
As a next step to the next section, we may have to note that linguists in the 1970s
however already agreed that presupposition is not a purely logical phenomenon. Ruth
Kempson (Kempson 1975) argued that presupposition is basically a pragmatic phenom-
enon, and the conclusions of the leading theorists like Gerald Gazdar, Lauri Karttunen,
and Stanley Peters are equivocal. These authors treat presupposition in conjunction
with conventional implicatures, which cannot be done without reference to a theory of
pragmatics. See, for example, Gazdar 1979. Karttunen and Peters eventually presented
an account of presupposition by separating presupposition (or ‘implicature’) and asser-
tion (‘extension’) along different dimensions (Karttunen & Peters 1979). See Dekker
2008 for a recent dynamic implementation of their views.

4 The Dynamic Semantic Turn

Taking some steps back in history, it may said to have been Peter F. Strawson (1919–
2006) who first formally, identified the concept of presupposition under this name, after
Frege did so under the name of “Voraussetzung” (Strawson 1952, p. 175). Before this,
in 1950, Strawson had supplied his famous response to Russell’s theory of descriptions,
noting that it is not sentences that are true or false and terms that refer but that it
is the uses of sentences which are true or false, or neither, and people’s uses of terms
that refer, or fail to refer—and uses which may yield truth-value gaps. (Strawson 1950,
pp. 327–8) Above we have already seen Russell’s touchy reaction to these insights.
Strawson’s emphasis on the use-based nature of truth, reference, and presupposition
has dominated large parts of the philosophical/linguistic literature ever since. Straw-
son’s core ideas became more fashionable in linguistics following the work of Keith
Donnellan (Donnellan 1966; Donnellan 1978) who once more emphasized the referen-
tial uses, as opposed to their Russellian, attributive uses, of definite descriptions. In his
seminal paper on presupposition and assertion, Robert Stalnaker continued this line of
thought in the early 1970s. Stalnaker emphasizes that “People, rather than sentences or
propositions are said to have, or make, presuppositions in this [pragmatic] sense. . . . To
presuppose a proposition in the pragmatic sense is to take its truth for granted, and to
assume that others involved in the context do the same” (Stalnaker 1970, p. 279). The
understanding of presuppositions has gained more emphasis in (Stalnaker 1978), where
the notion has been more or less formally implemented in terms of updates of common

47
 PAU L DEK K E R

grounds, or in terms of a dynamic notion of meaning. In the latter paper, assertions are
seen to be, basically, proposals to add propositions to a propositional ground assumed to
be common to all participants in a conversation.
The 1980s and 1990s have witnessed several attempts to formalize these pragmatic
conceptions of presupposition further but in different, and rival directions, all of them
with their own intuitively appealing inclinations. Lauri Karttunen (Karttunen 1974),
Irene Heim (Heim 1983), David Beaver (Beaver 1995), and Robert van Rooij (van
Rooij 2005), among many others have elaborated and refined analyses based on the idea
that presupposition relates to information presumed to be given in a discourse situation.
The idea is very simple and intuitively appealing. Consider the following examples.

(15) Julio’s son will be a star.

(16) Julio has a son, and Julio’s son will be a star.
(17) If Julio has a son, Julio’s son will be a star.
(18) Julio’s son will be a star, and Julio has a son.

Example (15), but not example (16), presupposes that Julio has a son; and ecample (17)
does not, and (18) does, presuppose that Julio has a son. A dynamic semantic treat-
ment of 5 assertion and presupposition neatly accounts for these facts, because such
treatments build on the assumption that when the second constituent of a coordinated
(conjoined, or implicational) structure is evaluated, including its presuppositions, this
is against the background of what the first constituent has implied. And not the other
way around. This intuitively explains the asymmetry in the examples above. The main
tenets of such dynamic treatments consist in the idea that participants in a conversa-
tion maintain some kind of a common ground of what they tend to be commonly given
as information. This gets updated any time a contribution to the discourse has been
made and has remained unchallenged. Notice that this notion of a ‘common ground’
is a technical point, and can be suppositional, for instance, if the consequent clause of
a conditional sentence is evaluated relative to a suppositional update of the common
ground with the antecedent clause.
Within the framework of Discourse Representation Theory (DRT) a similar, dynamic
treatment of presupposition and of presupposition projection has been given by Rob
van der Sandt, Bart Geurts, and Hans Kamp (van der Sandt 1988; van der Sandt 1992;
Geurts 1999; Kamp 2001). In these, and similar frameworks, the current states of dis-
courses and dialogues are presented as representations of the information conveyed at
the current state of a discourse. As in the dynamic theories of interpretation, presup-
positions are supposed to be satisfied, but they are not directly identified as propositions,
but, again, as representations of them. This enables them to be handled more as inde-
pendent structures. Within DRT, presuppositions are not interpreted as such, but they
are handled as representational items that need to be ‘resolved’ in the context. Consider
a complicated structure, like the one induced by the following sentence:

(19) Lucy thinks it is not likely that if she fails her exam, she will not be able to
feed her cat.

The crucial, and triggering, part here is “feed her cat” which comes with the presupposi-
tion that some y is Lucy’s cat. Maybe we all know Lucy has a cat, or maybe we all know
Lucy thinks so, or maybe we all know that if what Lucy believes is true, and she also fails

48
 PRESU PPO S I T I O N

her exam, she has a cat. Or, even, maybe we all know, that Lucy thinks that if she fails
her exam than most likely she will not have a cat and be able to feed it. These are all
interpretative options made available in DRT, surviving from the presupposition that
some y is Lucy’s cat, on a most suitable place in the Discourse Representation Structure,
where it renders “Lucy’s cat” as a known entity right there where the term has to be
interpreted. In DRT the different resolutions are achieved by moving the representation
of y being a cat of Lucy’s, around in the Discourse Representation Structure until it ends in
the rationally most comfortable place.
The difference between information being given in an update semantics, or being
represented in a representational framework, like that of DRT, may seem marginal, but it
has given rise to harsh disputes about the actual implementation of these ideas. A most
typical example of the difference is the phenomenon of conditional presuppositions.
The strength and beauty of the update approach is that it comes with its own logic,
which, almost automatically it seems, explains the facts about presupposition projec-
tion. It has been argued, however, that the predictions are too soft to be realistic. Using
the dynamic approach, an assertion of one of the following sentences

(20) Lucy is taking a shower now, and she forgot to feed her cats.
(21) If Lucy is taking a shower now, she forgot to feed her cats.

is not predicted to presuppose that Lucy has cats, but only that if she is taking a shower
now, she has cats. The predictions are rather counterintuitive. It seems the representa-
tional format of DRT is quite capable of handling the problem. Upon their most favora-
ble readings both sentences get resolved by unconditionally getting the information that
Lucy has cats by resolution or by accommodation. Of course, several good arguments
have been made to rescue the dynamic approach. Geurts 1996 gives a critical overview
of the problems, and van Rooij 2007 a constructive reply. Despite the apparent con-
tradictory predictions of the two frameworks, which might be overcome, it seems that
only (Zeevat 1992) has been able to avoid both horns of the dilemma, and combine the
benefits of both types of dynamic approach.
One problem, which the dynamic semantic approach shares with the previously men-
tioned, many valued approach. consists in the treatment of presuppositions in modal
and quantified structures. A most straightforward implemenation of the dynamic ideas
seems to entail that, again,

(22) A fat man was pushing his bicycle.

presupposes that all fat men do actually have a bicycle, an unwanted result for sure. Sev-
eral attempts have been made within DRT, and dynamic semantic alternatives, to deal
with these problems, but no consensus has been reached. It seems that the intuitive and
empirical data available do not point at a direction for solving the issue.

5 Recent Developments
Philosophers and linguists have equally impressively emphasized the phenomenon of
presupposition as a pragmatic one, but it has not been their task or aim to give an
analytic, syntactic, or semantic treatment of it. Not unsurprisingly, the phenomenon
has recently been revived in more pragmatic and cognitively oriented approaches to

49
 PAU L DEK K E R

discourse semantics and in empirical studies. In this last section we briefly summarize
some recent been results and findings.
Reinhard Blutner has developed a DRT-style account of presupposition in the frame-
work of bi-directional optimality theory. Optimality theory has been a very successful
paradigm in phonology and syntax (Prince & Smolensky 1997), and the bi-directional
version of it has been very fruitful in semantics as well. In a bi-directional framework,
optimalization strategies of both speakers and hearers are taken into account. When
it comes to presupposition, put very bluntly, speakers ought not to say things that the
hearers can figure it out themselves, but, also, speakers should say the things which they
presuppose and which the hearers cannot be assumed to figure out for themselves.
Similar in spirit is a recent series of papers in which Philippe Schlenker has argued
for a “transparant” pragmatic theory of presupposition, which is not built on a dynamic
notion of meaning but on a bi-valent truth value assignment. Instead, Gricean-style
pragmatics is invoked to deliver most of the kinds of presuppositions that Karttunen,
Heim and Beaver have argued for (Schlenker 2008). A special issue of Theoretical Lin-
guistics was devoted to proponents and opponents of this revolutionary or, as some
claim, reactionary approach. Recently (Thomason et al.) proposed another computa-
tional perspective on presuppositions, in terms of “enlightened updates”, dealing with
commonsense reasoning about speakers’ and hearers’ intentions, acts, and plans.
Definitely theoretically oriented but much more empirically motivated, are recent
studies conducted by Jennifer Spenader, Bart Geurts and Nausicaa Pouscoulous, and
Emmanuel Chemla. Many discussions about presupposition, like many discussions
about the semantics of expressions in natural language, get hampered by almost unde-
cidable questions about natural language intuitions concerning complicated examples
brought up from the philosopher’s armchair, or the linguist’s keyboard. Until recently,
hardly any empirical tests had been made regarding these intuitions, but fortunately the
aforementioned authors have started to fill this embarrassing gap. In the present state
of affairs, however, it seems difficult to draw any specific conclusions, as the findings are
not yet very decisive.
Other open questions, for which we are not going to claim an answer, and which
are very much related, remain pretty open. There is still a debate about whether or not
presupposition can be conceived of as a property of sentences, utterances, or speakers.
Is there, after all, one notion of presupposition, or do the various so-called triggers get
interpreted differently? Can we make do with a (partial) logic to interpret presupposi-
tions, or do we have to implement a module in a (classical) system of interpretation
that is especially designed to deal with presuppositions? An age-old question, first fully
addressed by Gerald Gazdar, is: how do presupositions and implicatures relate? A final,
old, consideration, the subject of current empirical research, is how presuppositions
really project when they are embedded in modal and quantified structures. Obviously,
it is also not so easy to gather empirical data, just as it is not so easy to gather analytical
armchair data.

References
Aquinas, Thomas 2000–2009. Aristotle on Interpretation. In: The Collected Works of St. Thomas Aquinas,
http://www.corpusthomisticum.org/: Corpus Thomisticum.
Aristotle 1975. Categories and De Interpretation. Oxford: Oxford University Press.
Bäuerle, Rainer, Uwe Reyle & Thomas Ede Zimmermann 2010. Presuppositions and Discourse: Essays Offered
to Hans Kamp, volume 21 of CRISPI. Bingley, UK: Emerald Group.

50
 PRESU PPO S I T I O N

Beaver, David 1995. Presupposition and Assertion in Dynamic Semantics. Ph.D. thesis, CCS, Edinburgh.
Published in 2001 by CSLI Publications, Stanford.
Beaver, David 1997. Presupposition. In: Johan van Benthem & Alice ter Meulen (eds.) Handbook of Logic
and Language, Amsterdam: Elsevier.
Bochvar, Dmitry Anatol’evich 1939. Ob odnom trehznachom iscislenii i ego primeneii k analizu para-
doksov klassicskogo rassirennogo funkcional nogo iscislenija. Matematiciskijs bornik 4. English translation,
1981, “On a Three-valued Logical Calculus and its Applications to the Analysis of the Paradoxes of the
Classical Extended Functional Calculus” in: History and Philosophy of Logic 2, pp. 87–112.
Dekker, Paul 2008. A Multi-Dimensional Treatment of Quantification in Extraordinary English. Linguistics
and Philosophy 31(1), pp. 101–127.
Donnellan, Keith 1966. Reference and Definite Descriptions. Philosophical Review 75, pp. 281– 304.
Donnellan, Keith 1978. Speaker Reference, Descriptions and Anaphora. In: Peter Cole (ed.) Syntax and
Semantics, Vol 9. Pragmatics, New York: Academic Press. pp. 47–68.
Fox, Danny 2008. Two Short Notes on Schlenker’s Theory of Presupposition Projection Theoretical Linguis-
tics 34(3), pp. 237–252.
Frege, Gottlob (ed.) 1884. Die Grundlagen der Arithmetik: eine logisch-mathematische Untersuchung über den
Begriff der Zahl. Breslau: Wilhelm Koebner Verlag.
Frege, Gottlob 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik NF 100,
pp. 25–50.
Frege, Gottlob 1969. Nachgelassene Schriften und Wissenschaftlicher Briefwechsel. Hamburg: Meiner Verlag.
Edited by Hans Hermes and Friedrich Kambartel and Friedrich Kaulbach.
Gazdar, Gerald 1979. Pragmatics. Implicature, Presupposition, and Logical Form. New York: Academic
Press.
Geurts, Bart 1996. Local Satisfaction Guaranteed: A Presupposition Theory and its Problems. Linguistics and
Philosophy 19(3), pp. 259–294.
Geurts, Bart 1999. Presuppositions and Pronouns, volume 3 of CRISPI. London: Elsevier.
Heim, Irene 1983. On the Projection Problem for Presuppositions. In: M. Barlow, D. Flickinger & M.
Wescoat (eds.) Proceedings of WCCFL II. Stanford, California. Reprinted in Steven Davis (ed.) 1991,
Pragmatics: A Reader, Oxford University Press.
Horn, Laurence R. 1985. Metalinguistic Negation and Pragmatic Ambiguity. Language 61, pp. 121–174.
Horn, Laurence R. 1996. Presupposition and Implicature. In: Shalom Lappin (ed.) The Handbook of Contem-
porary Semantic Theory, Oxford: Blackwell, pp. 299–319.
Kamp, Hans 2001. Presupposition Computation and Presupposition Justification: One Aspect of the Inter-
pretation of Multi-Sentence Discourse. In: Myriam Bras & Laure Vieu (eds.) Semantics and Pragmatic
Issues in Discourse and Dialogue, Amsterdam: Elsevier, pp. 57–84.
Karttunen, Lauri 1973. Presuppositions of Compound Sentences. Linguistic Inquiry 4, pp. 169–193. Kart-
tunen, Lauri 1974. Presupposition and Linguistic Context. Theoretical Linguistics 1(1), pp. 181–193.
Karttunen, Lauri & Stanley Peters 1979. Conventional Implicature. In: Choon-Kyu Oh & David A. Din-
neen (eds.) Syntax and Semantics 11—Presupposition, New York: Academic Press. pp. 1–56.
Kempson, Ruth M. 1975. Presupposition and the Delimitation of Semantics. Cambridge: Cambridge University
Press.
Kleene, Stephen Cole 1938. “On Notation for Ordinal Numbers”.The Journal of Symbolic Logic 3(4),
pp. 150–155.
Kleene, Stephen Cole 1952. Introduction to Metamathematics. Amsterdam: Notrh-Holland.
Kneale, William & Martha Kneale 1962. The Development of Logic. Oxford: Clarendon Press.
Laertius, Diogenes 1853. The Lifes and Opinions of Eminent Philosophers. http://classicpersuasion.org/pw/
diogenes/
Łukasiewicz, Jan 1920. O logicetrójwartościowej. Ruchfilozoficzny 5, pp. 170–171. English translation, 1970,
“On three-valued logic”, in Ł. Borkowski (ed.), Selected works by Jan Łukasiewicz, NorthHolland, Amster-
dam, 1970, pp. 8788.
Peters, Stanley 1977. A Truth-Functional Formulation of Karttunen’s Account of Presupposition. In: Susan
F. Schmerling & Carlota S. Smith (eds.) Texas Linguistic Forum, 6, University of Texas at Austin.
Prince, Alan & Paul Smolensky 1997. Optimality: From Neural Networks to Universal Grammar. Science
275, pp. 1604–1610.
van Rooij, Robert 2005. A Modal Analysis of Presupposition and Modal Subordination. Journal of Semantics
22(3), pp. 281–305.
van Rooij, Robert 2007. Strengthening conditional presuppositions. Journal of Semantics 24(3), pp. 289–304.

51
 PAU L DEK K E R

Russell, Bertrand 1957. Mr. Strawson on Referring. Mind 66, pp. 385–389.
van der Sandt, Rob A. 1988. Context and Presupposition. London: Croom Helm.
van der Sandt, Rob A. 1992. Presupposition Projection as Anaphora Resolution. Journal of Semantics 9(4),
pp. 333–377.
Schlenker, Philippe 2008. Be Articulate: A Pragmatic Theory of Presupposition Projection. Theoretical Lin-
guistics 34(3), pp. 157–212.
Seuren, Pieter A.M. 1985. Discourse Semantics. Oxford: Blackwell.
Seuren, Pieter A.M. 2005. Eubulides as a 20th-Century Semanticist. Language Sciences 27, pp. 75–95.
Soames, Scott 1989. Presupposition. In: Dov Gabbay & Franz Guenthner (eds.) Handbook of Philosophical
Logic, Dordrecht: Kluwer, volume IV, Topics in the Philosophy of Language. pp. 553–616.
Stalnaker, Robert 1970. Pragmatics. Synthese 22, pp. 272–289. Reprinted in Donald Davidson and Gilbert
Harman, 1972, Semantics of Natural Language, Reidel, Dordrecht, pp. 272–289.
Stalnaker, Robert 1978. Assertion. In: Peter Cole (ed.) Syntax and Semantics 9—Pragmatics, New York:
Academic Press. pp. 315–332.
Strawson, Peter F. 1950. On referring. Mind 59, pp. 320–344.
Strawson, Peter F. 1952. Introduction to Logical Theory. London: Methuen & Co.
Thomason, Richmond H., Matthew Stone & David DeVault for Theoris. Enlightened Update: A computa-
tional architecture for presupposition and other pragmatic phenomena. In: Donna Byron, Craige Roberts
& Scott Schwenter (eds.) Presupposition Accommodation, Ohio: Ohio State Pragmatic Initiative.
Zeevat, Henk 1992. Presupposition and Accommodation in Update Semantics. Journal of Semantics 9(4),
pp. 379–412.

Further Reading
The reader may get a good sense of the wealth of discussion and examples relating to the topic of presupposi-
tion from the following overview papers and collections: (Soames 1989; Horn 1996; Beaver 1997; Bäuerle
et al. 2010). A special issue of Theoretical Linguistics, 2008, Volume 34/3, has been devoted to the subject, as
well as an ESSLLI workshop on “New Directions in the Theory of Presupposition” in Bordeaux, 2009, organ-
ized by Nathan Klinedinst and Daniel Rothschild (http://essllipresupposition2009.blogspot.com/

52
 1.5
IMP L I C AT UR E
Laurence Horn

A year after unveiling implicature in his William James lectures at Harvard, Paul Grice
(1968) published one of the lectures, a study of the relation of word meaning to utterer’s
meaning situated within his overall framework for speaker meaning.

The wider programme . . . arises out of a distinction I wish to make within

the total signification of a remark, a distinction between what the speaker has
said (in a certain favored and maybe in some degree artificial, sense of ‘said’),
and what he has ‘implicated’ (e.g., implied, indicated, suggested, etc.), taking
into account the fact that what he has implicated may be either conventionally
implicated (implicated by virtue of the meaning of some word or phrase which
he has used) or non-conventionally implicated (in which case the specification
of implicature falls outside the specification of the conventional meaning of
the words used).
(Grice 1968: 225; cf. also Grice 1989: 118)

Grice’s footnote on the first page of this paper—“I hope that the material in this paper,
revised and re-arranged, will form part of a book to be published by Harvard University
Press”—is poignant to read, given that 21 years would elapse before this book would
appear in print, and posthumously at that. Actually, the notions of conventional
and non-conventional (specifically conversational) implicature, as defined here and
as mapped by the familiar diagnostics of cancelability and detachability, had already
been introduced several years earlier without the labels introduced above but already
depicted as distinct varieties of (non-logical) implication (cf. Grice 1961: §3 and discus-
sion in Horn 2004).

Conversational Implicature
Conversational implicature arises from the shared presumption that S and H are inter-
acting to reach a shared goal. A speaker S saying p and implicating q counts on her
interlocutor’s ability to compute what was meant (by S’s uttering p at a given point
in the interaction) from what was said, based on the assumption that both S and H
are rational agents. Speakers implicate, hearers infer; a hearer may recover an implica-
ture, but to implicate is not to infer. Nevertheless, S’s assumption that H will draw the
appropriate inference is what makes implicature a rational expectation. The governing
dictum is the Cooperative Principle: “Make your conversational contribution such as
 LAU REN CE H O R N

is required, at the stage at which it occurs, by the accepted purpose or direction of the
talk exchange” (Grice [1967]1989: 26). This principle is instantiated by a set of general
maxims of conversation whose exploitation potentially yields implicatures:

(1) quality: Try to make your contribution one that is true.

1. Do not say what you believe to be false.
2. Do not say that for which you lack evidence.
quantity:
1. Make your contribution as informative as is required (for the current
purposes of the exchange).
2. Do not make your contribution more informative than is required.
relation: Be relevant.
manner: Be perspicuous.
1. Avoid obscurity of expression.
2. Avoid ambiguity.
3. Be brief. (Avoid unnecessary prolixity.)
4. Be orderly.

Grice launched his characterization of implicature as an aspect of speaker meaning—

what is meant without being said—at a moment when similar notions were in the air,
especially the air of Oxford and its ordinary-language sphere of influence. Thus, a rela-
tion of “contextual implication” was invoked by Nowell-Smith (1954: 80–82) and
evaluated by Hungerland (1960) in her eponymous paper: “A statement p contextually
implies q if anyone who knew the normal conventions of the language would be entitled
to infer q from p in the context in which they occur.” But despite its name and context-
relativity, such a relation differs substantively from implicature. The locus classicus for
contextual implication is the context familiar from G.E. Moore: “When a speaker uses
a sentence to make a statement, it is contextually implied that he believes it to be true”
(Nowell-Smith 1954: 81). But the relation between (my stating) He has gone out and
my believing that he has gone out cannot be assimilated to conversational implicature,
for reasons given later by Grice himself:

On my account, it will not be true that when I say that p, I conversationally

implicate that I believe that p; for to suppose that I believe that p (or rather
think of myself as believing that p) is just to suppose that I am observing the
first maxim of Quality on this occasion. I think this consequence is intuitively
acceptable; it is not a natural use of language to describe one who has said that
p as having, for example, “implied”, “indicated”, or “suggested” that he believes
that p. The natural thing to say is that he has expressed (or at least purported
to express) the belief that p.
(Grice 1989: 42)

Treating the Nowell-Smith/Hungerland relation as a proto-conversational implicature

also instantiates three of Bach (2006)’s “Top ten misconceptions about implicature”,
#1 (“Sentences have implicatures”), #4 (“Gricean maxims apply only to implicatures”),
and #6 (“All pragmatic implications are implicatures”).

54
 I M PLI CAT U R E

The difficulty of canceling such a (putative) implicature without epistemic or doxas-

tic anomaly argues against such an analysis. The same applies to other cases of sincerity
conditions Hungerland cites: e.g. the relation between I promise to p and I intend to p.
Another of Nowell-Smith’s examples of contextual implication does qualify as prefig-
uring Grice on the maxims, specifically Relation, although not implicature as such:
“What a speaker says may be assumed to be relevant to the interests of the audience.”
This maxim may indeed be overridden, as Nowell-Smith and Hungerland both observe.
But Hungerland (1960: 212) points out the heterogeneous nature of any construct
that extends from this principle to the sincerity condition for assertions and promises;
Nowell-Smith himself admits that the violation of the latter leads to “logical oddity,” as
in “It’s raining, but I don’t believe it is,” while overriding the relevance rule chiefly runs
the risk of boredom. As an anonymous reviewer notes, relevance violations may also
yield confusion or signal a disparity of conversational goals; what they do not produce is
the “logical oddity” of Moore’s paradox.
In addition, conversational implicature differs from contextual implication (or non-
demonstrative implication more generally) in being defined as a relation between a
speaker (not a sentence) and a proposition and in arising typically from the exploitation
of the maxims (Grice 1989: 26ff.; cf. Horn 2004); in the case of scalar implicature in
particular, what is implicated depends on what isn’t (but could have been) said. Ceteris
paribus, a speaker uttering Some F are G implicates that (for all she knows) not all F are
G because she would have been expected to express the stronger proposition if she had
been in a position to do so. A speaker may opt for a weaker utterance from a belief that
to express the latter might violate considerations of relevance, brevity, or politeness (cf.
Horn 2004) but especially from a lack of certainty that the stronger counterpart holds.
Exactly a century before the William James lectures, John Stuart Mills offered these
grounds for rejecting Sir William Hamilton’s (1860) semantic account of some as “some
only, some but not all”:

No shadow of justification is shown . . . for adopting into logic a mere sous-

entendu of common conversation in its most unprecise form. If I say to any one,
“I saw some of your children today”, he might be justified in inferring that I
did not see them all, not because the words mean it, but because, if I had seen
them all, it is most likely that I should have said so: though even this cannot be
presumed unless it is presupposed that I must have known whether the children
I saw were all or not.
(Mill 1867: 501)

The last point Mill makes in this passage, invoking the two-stage (epistemically gov-
erned) process involved in the computation of implicature recovery, is built into the
rationality- and cooperation-driven Gricean model but is not directly captured in some
current alternative theories; see Geurts (2010) for elaboration.
Mill’s allusion to a tacit principle that requires the speaker to use the stronger all
in place of the weaker some when possible, and invites the hearer to draw the corre-
sponding inference when the stronger term is eschewed, is echoed by others in his own
time—

Whenever we think of the class as a whole, we should employ the term All;
and therefore when we employ the term Some, it is implied that we are not

55
 LAU REN CE H O R N

thinking of the whole, but of a part as distinguished from the whole—that is,
of a part only.
(Monck 1881: 156)

– and in Grice’s (e.g. Doyle 1951, Fogelin 1967: 20–22; see Horn 1990 and Chapman
2005: Chap. 5 for more on the chronology). Grice’s contribution, beyond securing the
naming rights to the relation in question, was to ground the operation of Mill’s “sous-
entendu[s] of common conversation” within an overall account of speaker meaning and
the exploitation of tacit principles based on assumptions of the rationality and mutual
goals of the interlocutors.
Mill’s reasoning, codified by Grice in his first quantity maxim, is systematically
exploited to yield upper-bounding scalar implicatures associated with relatively
weak scalar operators, those configurable on a scale defined by unilateral entailment
as in <all, most, many, some>. What is said in the use of a weak scalar value like those
boldfaced in (2) is the lower bound (. . . at least . . .); what is implicated, in the absence
of contextual or linguistic cancellation, is the upper bound (. . . at most . . .). What is
communicated, ceteris paribus, is the ‘two-sided reading’ that combines what is said with
what is implicated. Thus in (2d), “If we assert that a man who has acted in a particular
way must be either a knave or a fool, we by no means assert . . . that he cannot be both”
(Mill 1867: 512), but this unasserted ‘exclusive’ understanding of the disjunction may
well be communicated.

(2) one-sided reading o two-sided reading

a. Pat has 3 children. ‘at least 3’ ‘exactly 3’
b. You ate some of the cake. ‘some if not all’ ‘some but not all’
c. It’s possible she’ll win. ‘at least possible’ ‘possible but not
certain’
d. He’s a knave or a fool. ‘. . . and perhaps both’ ‘. . . but not both’
e. It’s warm. ‘at least warm’ ‘warm but not hot’

The alternative view on which each scalar predication in (2) is lexically ambiguous
between one-sided and two-sided readings falls afoul of Grice’s (1989: 47) Modified
Occam’s Razor: “Senses are not to be multiplied beyond necessity.” Scalar implicature
was introduced and formalized in work by Horn (1972, 1989), Gazdar (1979), Hirsch-
berg (1991), and Levinson (2000); cf. also Katzir (2007) and Geurts (2010) for insight-
ful recent discussions of certain problems arising in the implementation of the central
notions involved and Bontly (2005) for a defense of Modified Occam’s Razor as a heu-
ristic in acquisition.
One reflex of the centrality of scalar implicature for natural language is the systematic
pattern of lexical gaps and asymmetries involving the negative incorporation in logical
complexes. Thus, alongside the standard determiners all, some, no, we never find (in
any language) a determiner *nall (‘not all’); alongside connectives corresponding to
and, or, and sometimes nor (= ‘and not’), we have no *nand (= ‘or not’, ‘not . . . and’).
Although some does not contribute the same semantic content as some not (not all), the
use of either of the two values typically results in communicating the same information
in a given context, ‘some but not all’. The mutual quantity implicature holding between
some and some not and the markedness of negation results in the superfluity of the latter
for lexical realization (cf. Horn 1972, 1989 for details).

56
 I M PLI CAT U R E

The implicature-based approach to scalar predications has been vigorously challenged

by relevance theorists (see Carston 1988, 2002, 2004 and work reviewed therein), who
take such sentences to involve not lexical but propositional ambiguity, with the prag-
matically enriched two-sided meanings constituting not implicatures but explicatures,
pragmatically derived components of propositional content. It has been suggested (Saul
2002) that some of the divergences between relevance theory (RT) and Gricean prag-
matics are attributable to the focus of the former on the cognitive task of utterance
interpretation as distinguished from the philosophical tradition in which Grice situated
his own approach to implicature as a component of speaker meaning.
While the standard neo-Gricean view treats the cases in (2) homogeneously, many
have questioned the tenability of this line for number words, as in (2a), arguing that
such predications are semantically underspecified and contextually resolved, rather
than assigned the weak, ‘at least n’ values by linguistic means. The underspecification
line on cardinals, originally proposed by Carston (1988) and Koenig (1991), is extended
in later work including Horn (1992), Geurts (1998), and Bultinck (2005). But while
Ariel (2004) similarly disputes an implicature-based account of the upper bound of
most-statements (the move from most F are G to ‘not all F are G’) based on a putative
parallel between most and the cardinals regarding the status of the upper bound, there is
evidence that most is distinct in behavior from the cardinals and that its meaning, like
that of some, should be assigned the standard neo-Gricean account (unilateral seman-
tics and upper-bounding scalar implicature) (cf. Horn 2006: §4). The strongest support
for this hybrid view is the fact that a simple negative answer to a general scalar question,
as opposed to one involving a cardinal value, always returns a ‘less than’ meaning, since
this context selects descriptive and not metalinguistic negation. If you ask me whether
most of the students passed, my negative response commits me to the proposition that
50 percent or fewer passed, not to the disjunction that either 50 percent-or-fewer passed
or else all of them did. Yet it is just that disjunction that I must be asserting if my
reply negates the proposition that ‘50–99 percent of the students passed’ as it does on
a semantic upper-bound account. On the other hand, if you asked me whether 10 stu-
dents passed and I knew that 15 did, I must first determine whether you were asking me
if at least 10 passed or exactly 10 passed before knowing whether to answer positively
or negatively. A simple ‘No’ in response to ‘Did 10 of the students pass?’ might commit
me to either ‘Fewer than 10 passed’ or ‘Either fewer or more than 10 passed’, depending
on the context. Similarly, I would report that ‘I am surprised that most of the students
failed’ only if I had expected at least half of them to pass, while my report that ‘I am
surprised that five students failed’ is consistent with either a more pessimistic or more
optimistic expectation.
In addition to such linguistic evidence, a now considerable body of empirical work
indicates that the acquisition and processing of cardinals differs along a variety of
parameters from that of other scalar values; note especially the studies in Papafragou
and Musolino (2003), Hurewitz et al. (2006), and Breheny (2008). Thus, while acknowl-
edging the important work on underspecification by Carston (and earlier related obser-
vations of Jay Atlas (1979)), we may need to distinguish the behavior of cardinals
from that of their inexact scalar cousins. It is not obvious how the unitary explicature-
based program for all scalar operators as in Carston (2002) can draw the necessary dis-
tinctions here, any more than the approach in Levinson (2000: 87–90) that retains
the original (Horn 1972) implicature-based line for both cardinal and general scalar
predications.

57
 LAU REN CE H O R N

Two other challenges to the Gricean picture of implicatures involve the number
and status of the maxims and the relationship between implicature and propositional
content. To begin with the former issue, Grice himself later acknowledged (1989:
371ff.) that the four macroprinciples (inspired by Kant) and nine actual maxims in
his inventory were somewhat overlapping and non-coordinate. The number of maxims
has been revised both upward (Leech 1983) and downward. The dualistic program of
Horn (1984, 1989, 2007a) begins by following Grice’s lead (1989: 371) in ascribing a
privileged status to Quality, on the grounds that without the observation of Quality, or
Lewis’s (1969) convention of truthfulness, no question of the observance of the other
maxims can be considered (though see Sperber and Wilson 1986 for a dissenting view).
The remaining maxims are subsumed under two countervailing functional principles
whose role in communication and language change traces back to Paul (1889) and
Zipf (1949). In this Manichaean model, implicatures may be generated by either the Q
Principle (essentially ‘Say enough’, generalizing Grice’s first submaxim of Quantity and
collecting the first two ‘clarity’ submaxims of Manner) or the R Principle (‘Don’t say too
much’, subsuming Relation, the second Quantity submaxim, Relation, and Brevity).
The hearer-oriented Q Principle is a lower-bounding guarantee of the sufficiency of
informative content, exploited to generate upper-bounding (typically scalar) implicata.
The R Principle is a correlate of Zipf’s principle of least effort dictating minimization of
form, exploited to induce strengthening implicata; it is responsible for euphemism, indi-
rect speech acts, neg-raising, and meaning change (Horn 2007a). In accord with the
division of pragmatic labor (Horn 1984), a relatively unmarked form—briefer and/or
more lexicalized—will tend to become R-associated with a particular unmarked, stereo-
typical meaning, use, or situation, while its periphrastic or less lexicalized counterpart,
typically more complex or prolix, will tend to be Q-restricted by implicature to those
situations outside the stereotype, for which the unmarked expression could not have
been used appropriately (as in kill vs. cause to die, or mother vs. father’s wife). Formaliza-
tions of the division of pragmatic labor have been undertaken within bidirectional opti-
mality theory and game-theoretic pragmatics, although the results of such work are a
matter of debate (cf. e.g. Blutner 2004, van Rooij 2009, and references cited therein).
Levinson’s (2000) framework posits an interaction of Q, I (| Horn’s R), and M (Man-
ner) heuristics. Levinson’s reconstruction of the division of pragmatic labor involves not
Q but the M heuristic, given that some differs from all in informative content whereas
kill differs from cause to die in complexity of production or processing. As Levinson
concedes, however, the Q and M patterns are closely related, since each is negatively
defined and linguistically motivated: S tacitly knows that H will infer from S’s failure to
use a more informative and/or briefer form that S was not in a position to have used that
form. Unlike Q implicature, R/I-based implicature is not negative in character and is
socially rather than linguistically motivated, typically yielding a culturally salient stere-
otype (Levinson 2000; cf. also Huang 2006 for a useful overview).
Relevance theorists (e.g. Sperber and Wilson 1986, Carston 2002) invoke one basic
pragmatic principle, that of Relevance as defined in non-Gricean terms. It may be argued,
however, that the RT program is itself covertly Manichaean, given that Relevance itself
is calibrated as a minimax of effort and effect. As Carston (1995: 231) puts it, “Human
cognitive activity is driven by the goal of maximizing relevance: that is . . . to derive as
great a range of contextual effects as possible for the least expenditure of effort.”
A clear departure from earlier Gricean models is in the move to reconceptualize
implicature, and not just within RT. Over the last couple of decades, it has become a

58
 I M PLI CAT U R E

truism within pragmatic theory that linguistically delivered aspects of meaning consti-
tute a relative slender component of both the speaker-intended and the hearer-recovered
message. The gap between what is literally expressed and what is communicated cor-
responds to aspects of meaning variously identified as conversational implicatures, con-
ventional implicatures, presuppositions, implicitures, and explicatures. These are not
simply alternate labels for a given phenomenon but different ways of conceptualizing
the area between the said and the meant (or, for the hearer, between the said and the
interpreted). Situating ‘what is said’ along this spectrum is itself controversial; what
is said for Recanati (2001), Ariel (2008), and the relevance theorists is enriched by
pragmatically derived material (hence constituting an explicature). Levinson, on the
other hand, responds to the apparent need to accept “pragmatic intrusion” into what
is said by allowing conversational implicatures to have truth-conditional consequences
for the propositions in question, contra Grice; in cases like Deirdre Wilson’s It’s better to
meet the love of your life and get married than to get married and meet the love of your life, an
implicature (“P precedes Q”) can feed into (rather than just being read off) what is said.
(See Carston 2002 and Russell 2006 for illuminating discussions of the and case.)
For more orthodox Griceans, the pragmatically enriched proposition in such cases
is distinct from what is said (see Horn 2009 for a defense of what Jenny Saul calls an
“austere” conception of what is said). Bach (2001) observes that once we give up the
“OSOP” (One Sentence, One Proposition) assumption, we can recognize that a given
sentence may express not only more than one proposition but fewer. What is said in
cases like Dana is ready may constitute not a truth-evaluable proposition but a proposi-
tional radical. Filling in such a radical or completing it within a given context to yield
e.g. Dana is ready to write a dissertation yields not what is said (which is tightly con-
strained by the actual syntax) or an explicature (since there is nothing explicit about
it), but rather an impliciture, a proposition implicit in what is said in a given context
as opposed to a true implicature, a proposition read off what is said (or the way it is said).
Similarly, scalar implicatures are reborn as scalar implic-i-tures. What Grice failed to
recognize, argues Bach, is the non-exhaustive nature of the opposition between what is
said and what is implicated.
Bach (2006) goes on to include “Scalar ‘implicatures’ are implicatures” as #9 in his
hit parade of misconceptions: since a speaker uttering “Some of the boys went to the
party” means not two separate things but just one thing, i.e. that some but not all of
them went, this enriched proposition is an impliciture (built up from what is said),
not an implicature (read off what is said). But it should be borne in mind that on the
Gricean account (e.g. in Horn 1972, 1989), the scalar implicature here is “Not all of
the boys went to the party”; this combines with what is said (“Some . . .”) to yield what
is communicated (“Some but not all . . .”). Thus the impliciture incorporates the scalar
implicature rather than supplanting it.
Another of Bach’s misconceptions, #2, is “Implicatures are inferences.” While Lev-
inson (2000) defines generalized conversational implicatures as default inferences,
others argue that they are neither inferences—by definition an implicature is an aspect
of speaker’s meaning, not hearer’s interpretation (cf. Bach 2001, Saul 2002)—nor true
defaults. This last point is especially worth stressing in the light of much recent work
in experimental pragmatics (see e.g. Noveck and Posada 2003, Bott and Noveck 2004,
Breheny et al. 2006, Katsos 2008) suggesting that children and adults do not first auto-
matically construct implicature-based enriched meanings for scalar predications and
then, when the ‘default’ interpretation is seen to be inconsistent with the local context,

59
 LAU REN CE H O R N

undo such meanings and revert to the minimal implicature-free meaning. To the extent
that this empirical work on the processing of implicature recovery can be substantiated
and extended, this is a very interesting result, but not (contrary to some claims) one
that threatens the actual Gricean tradition, which predicts no automatic enrichment
or default interpretation. This is clear from the passage distinguishing generalized and
particularized implicature (Grice 1989: 37, italics added):

I have so far considered only cases of what I might call ‘particularized conver-
sational implicature’ . . . in which an implicature is carried by saying that p on
a particular occasion in virtue of special features of the context, cases in which there
is no room for the idea that an implicature of this sort is normally carried by
saying that p. But there are cases of generalized conversational implicature.
Sometimes one can say that the use of a certain form of words in an utterance
would normally (in the absence of special circumstances) carry such-and-such an
implicature or type of implicature.

The classic contrast at issue here dates back to Grice 1961: §3—the particularized impli-
cature in the “Gricean letter of recommendation” for a philosophy job candidate (“Jones
has beautiful handwriting and his English is grammatical”) vs. the generalized implicature
in the logical disjunction (“My wife is in Oxford or in London”, implicating I don’t know
which). Crucially, an implicature may arise in an unmarked or default context without
thereby constituting a default jor automatic inference. (See Bezuidenhout 2002, Jaszc-
zolt 2005, and Geurts 2009 for more on defaults and their relation to implicature.)
Despite their substantial differences (from each other and from Grice) as to the role of
implicature and the relation between what is implicated and what is said, the proponents
of the approaches touched on above share Grice’s commitment to situating implica-
ture within a rationality-based pragmatics. On a competing view that has recently been
elaborated by Chierchia (2004) and his colleagues, scalar implicatures in particular are
generated locally as part of the grammar and/or the conventional lexical semantics of
weak scalar operators. Support for this variety of defaultism involves an appeal to cases in
which the Gricean model appears to yield the wrong results, thus arguing for local com-
putation of “embedded implicatures.” Others (e.g. Sauerland 2004, Russell 2006, Horn
2006) have challenged these conclusions and defended a global account of implicature
along Gricean lines as both empirically and theoretically desirable. In particular, Geurts
(2009, 2010) provides a broad survey of the landscape. Drawing a distinction between
marked L[evinson]-type cases and unmarked C[herchia]-type cases of putative local-
ity effects, Geurts (2009) argues that unlike the latter type, the Levinsonian contrast-
induced narrowings represent true problems for a classical Gricean (or neo-Gricean)
theory of implicature but shows that these can be handled by allowing upper-bounding
to enter into the reinterpretation of what scalar operators express, a reinterpretation that
is itself pragmatic in nature. In his treatise on Q-implicatures, Geurts (2010) argues that
the conventionalist alternative to a Gricean approach is not only stipulative but also
empirically flawed in failing to predict the full range of implicature-related results.

Conventional Implicature
In addition to introducing the very successful product line of conversational implica-
tures, Grice assembled an apparently inchoate class of phenomena under the rubric

60
 I M PLI CAT U R E

of conventional implicature, a non-cancelable but truth-conditionally transparent

component of content whose success in the scholarly marketplace has been far more
mixed. The coherence of this category has evoked much skepticism—Bach (1999) con-
signs it to the dustbin of mythology, Carston (2002: 134) remarks that “there simply is
no such thing as ‘conventional’ implicature in relevance theory (or, we would argue,
in reality),” while Potts (2005) undertakes a rehabilitation that redraws the descrip-
tive map—but Grice’s sketchy account of conventional content that does not affect
the truth conditions of the asserted proposition has a rich lineage. Frege (1892, 1897,
1918) delineates a somewhat heterogeneous class of meanings that, while of linguistic
interest, do not “affect the thought” or “touch what is true or false.” While most recent
scholarship follows Dummett (1973) in dismissing Frege’s positive proposals in this area
as representing a confused and subjective notion of “tone,” this mischaracterizes Frege’s
actual account of the relevant phenomena. For a range of constructions including dis-
course particles (but, even, Ger. ja, doch), subject-oriented adverbs, epithets, and other
‘loaded’ words, the approach favored by Frege and Grice remains eminently plausible
(Barker 2003; Horn 2007b, 2008; Gutzmann 2008; Williamson 2009).
Further, such an approach can be fruitfully extended to a range of other natural lan-
guage phenomena, including the familiar vs. formal second person singular (‘T/V’)
pronouns of many modern European languages, evidential markers, and, arguably, the
uniqueness/maximality condition on definite descriptions. In addition, certain syntac-
tic constructions can be profitably analyzed along these lines such as the southern U.S.
English “personal dative,” a non-argument pronominal appearing in transitive clauses
that obligatorily coindexes the subject as exemplified in I love me some squid, truth-con-
ditionally equivalent to, but not fully synonymous with, ‘I love squid’ (Horn 2008). In
each case, we find aspects of conventional content that are not entailed and do not fall
inside the scope of logical operators.
The category of conventional implicature poses a complication for the distinction
between the categories of what is said and what is meant. It could be argued that expres-
sions falling under this analysis represent a recalcitrant residue for Grice (who was con-
cerned with delineating what is said and what is conversationally, and hence calculably,
implicated) as they did for Frege (who was concerned with the thought, i.e. with sense
and potential reference); for both, detecting a conventional implicature facilitates the
real work by clearing away the brush. But Grice did undertake to situate this relation
within what we refer to (though he did not) as the “semantics/pragmatics divide.” His
contributions in this area, if not always accepted, are widely recognized, as in this pas-
sage from Davidson (1986: 161–62):

It does not seem plausible that there is a strict rule fixing the occasions on
which we should attach significance to the order in which conjoined sentences
appear in a conjunction: the difference between ‘They got married and had a
child’ and ‘They had a child and got married.’ Interpreters certainly can make
these distinctions. But part of the burden of this paper is that much that they
can do should not count as part of their linguistic competence. The contrast
in which is meant of implied by the use of ‘but’ instead of ‘and’ seems to me
another matter, since no amount of common sense unaccompanied by linguis-
tic lore would enable an interpreter to figure it out. Paul Grice has done more
than anyone else to bring these problems to our attention and help to sort them
out.

61
 LAU REN CE H O R N

But how does this sorting work? If descriptive content, reflecting what is said, is clearly
semantic and if what is conversationally implicated (e.g. the ‘for all I know, not both
p and q’ upper-bounding implicatum associated with the utterance of the disjunction p
or q or the negative effect of the Gricean letter of recommendation) is pragmatic (pace
Chierchia et al.), where is conventional implicature located? One standard view is that
by falling outside what is said, it must be pragmatic (see e.g. Gutzmann 2008: 59). In
Kaplan’s words (1999: 20–21):

According to Grice’s quite plausible analysis of such logical particles as “but”,

“nevertheless”, “although”, and “in spite of the fact”, they all have the same
descriptive content as “and” and differ only in expressive content . . . The argu-
ments I will present are meant to show that even accepting Grice’s analysis,
the logic is affected by the choice of particle . . . If this is correct, then genera-
tions of logic teachers, including myself, have been misleading the youth. Grice
sides with the logic teachers, and though he regards the expressive content as
conventional and hence (I would say) semantic (as opposed to being a conse-
quence of his conversational maxims), he categorizes it with the maxim-gener-
ated implicatures.

To be sure, conventional implicatures are implicatures. But then again, they are con-
ventional; we are indeed dealing here, unlike in the maxim-based cases, with aspects
of content.
Two decades after the William James lectures, Grice revisited these categories in his
Retrospective Epilogue (1989: 359–65), where he distinguished central and non-cen-
tral modes of meaning through the criteria of formality (“whether or not the relevant
signification is part of the conventional meaning of the signifying expression”) and
dictiveness (“whether or not the relevant signification is part of what the signifying
expression says”). Thus, when a speaker says “p; on the other hand, q” in the absence of
any intended contrast of any kind between p and q, “one would be inclined to say that
a condition conventionally signified by the presence of the phrase ‘on the other hand’
was in fact not realized and so that the speaker had done violence to the conventional
meaning of, indeed had misused, the phrase ‘on the other hand’.” Crucially, however,
“the nonrealization of this condition would also be regarded as insufficient to falsify
the speaker’s statement” (Grice 1989: 361). Thus, formality without dictiveness yields
conventional implicature. (As for dictiveness without formality, a plausible candidate
is the pragmatically enriched content of relevance theorists or the truth-conditional
pragmatics of Recanati 2001.)
In uttering a given sentence in a given context, the speaker may intend to communi-
cate, and may succeed in communicating, more than one truth-evaluable proposition,
but these communicated propositions do not necessarily have equal status; in They are
poor but happy, the conjunctive content is truth-conditionally relevant, but the contras-
tive content is not. Yet but is not a synonym of and. Conventional implicatures consti-
tute part of encoded content but not part of truth-conditional content per se; their falsity
does not project as falsity of the expression to which they contribute (cf. Barker 2003).
What they contribute is use-conditional meaning (Kaplan 1999, Gutzmann 2008).
Besides the defined properties of detachability and non-cancelability for such aspects
of content (Grice 1961, 1989), additional diagnostics for conventional implicatures
include their tendency to project out of embedded contexts and their immunity to

62
 I M PLI CAT U R E

certain kinds of objection. In addition, conventional implicatures exhibit a contextual

variability or descriptive ineffability (Potts 2007). Illustrations include the difficulty
of pinning down the precise content of:

(3) • the nature of the contrast/unexpectedness implicated by but (Dummett

1973)
• the characterization of the scalar conventional implicature associated
with even (relative or absolute? unlikelihood or noteworthiness?)
• the nature of the expressive attitude embodied in racial and ethnic slurs
and other epithets (Williamson 2003, 2009, Potts 2007)
• the precise notion of uniqueness or individuability associated with the
use of definite descriptions (cf. Horn 2007b)
• the appropriateness implicatures for tu vs. vous or other T vs. V 2nd
singular pronouns within a given context in a community of practice (T
can be affectionate, presumptuous, comradely, or condescending; V can
be polite, aloof, diplomatic, or hostile; cf. Brown and Gilman 1960)
• the implicatures of intentionality or positive affect for personal datives
in Appalachian English (He found/#lost him a laptop) and related affect-
edness implicatures (benefactive or malefactive) for non-argument
datives in a variety of languages (cf. Horn 2008)

Whence this property of ineffability? Modulo the recognized problems of vagueness, it

is plausible that the edges of truth-conditional meaning should remain largely discrete,
while inconsistency in the mental representation of non-truth-conditionally relevant
content is less pernicious. If you know generally that my use of vous rather than tu signals
some aspect of formal respect, distancing, and/or lack of intimacy, my precise motives
can be left underdetermined, but if you don’t know whether I’m using a second-person
or third-person pronoun, the indeterminacy would be more costly. Similarly, it’s crucial
to know whether I bought the car for myself or for my son, and hence the identity of
the referent for a true indirect object pronoun, but whether or not you can figure out
precisely why I affirm “I bought me a truck for my son” rather than simply “I bought a truck
for my son,” no differences in argument structure or truth conditions emerge.
For Frege and Grice, identifying the class of conventional implicature-licensing con-
structions—including a motley collection of scalar particles, speaker-oriented sentence
adverbs, epithets and slurs, prosodic features, evidential markers, “affected” pronomi-
nals, and word order effects—serves to characterize them in terms of what they are
not: they do not affect the thought or the truth-conditionally relevant meaning of a
given expression, and at the same time they are not derivable from general principles
of rationality. While they share the former property with conversational implicatures,
they differ crucially from them in the latter respect.
Since classical rhetoricians first recognized figures in which we say less and mean
more (“minus dicimus et plus significamus”—Servius and Donatus, cited in Hoffmann
1987: 21), semanticists and pragmaticists have explored the domain of the said and
the meant. In delineating and differentiating conversational and conventional impli-
catures, Grice provided a map for all further explorations of these boundary disputes,
explorations that continue to inform our understanding of the nature of meaning in
natural language.

63
 LAU REN CE H O R N

Acknowledgments
I am grateful to Barbara Abbott, Kent Bach, Betty Birner, Robyn Carston, Gregory Ward,
and the anonymous reviewer for Routledge for their helpful comments, pointers, and
objections; the usual disclaimers apply. Thanks also to the editors for their patience.

Related Topics
1.2 Semantics and Pragmatics
1.6 Pragmatic Enrichment
1.7 Meaning and Communication
1.14 Relevance Theory.

References
Aikhenvald, A. (2004) Evidentiality. Oxford: Oxford University Press.
Ariel, M. (2004) “Most,” Language 80: 658–706.
Ariel, M. (2008) Pragmatics and Grammar. Cambridge: Cambridge University Press.
Atlas, J. D. (1979) “How Linguistics Matters to Philosophy: Presupposition, Truth, and Meaning,” in C.-K.
Oh and D. Dinneen (eds.), Syntax and Semantics 11: Presupposition, 265–81. New York: Academic Press.
Bach, K. (1994) “Conversational Impliciture,” Mind and Language 9: 124–62.
Bach, K. (1999) “The Myth of Conventional Implicature,” Linguistics and Philosophy 22: 327–66.
Bach, K. (2001) “You Don’t Say?” Synthese 127: 11–31.
Bach, K. (2006) “The Top 10 Misconceptions about Implicature. In B. Birner and G. Ward (eds.) Drawing
the Boundaries of Meaning, 21–30. Amsterdam: John Benjamins.
Barker, S. (2003) “Truth and Conventional Implicature,” Mind 112: 1–33.
Beaney, M. (ed., 1997) The Frege Reader. Oxford: Blackwell.
Bezuidenhout, A. (2002) “Generalized Conversational Implicatures and Default Pragmatic Inferences,” in
J. Campbell et al. (eds) Meaning and Truth: Investigations in Philosophical Semantics, 257–83. New York:
Seven Bridges Press.
Blutner, R. (2004) “Pragmatics and the Lexicon,” in Horn and Ward (eds.), 488–515.
Bontly, T. (2005) “Modified Occam’s Razor: Parsimony Arguments and Pragmatic Explanations,” Mind &
Language 20: 288–312.
Bott, L. and I. Noveck (2004) “Some Utterances Are Underinformative: The Onset and Time Course of
Scalar Inferences,” Journal of Memory and Language 51: 437–57.
Breheny, R. (2008) “A New Look at the Semantics and Pragmatics of Numerically Quantified Noun
Phrases,” Journal of Semantics 25: 93–140.
Breheny, R., N. Katsos, and J. Williams (2006) “Are Generalized Scalar Implicatures Generated By Default?
An On-Line Investigation Into The Role of Context in Generating Pragmatic Inferences,” Cognition 100:
434–63.
Brown, R. and A. Gilman (1960) “The Pronouns of Power and Solidarity,” in T. Sebeok (ed.) Style in Lan-
guage, 253–77. Cambridge, MA: MIT Press.
Bultinck, B. (2005) Numerous Meanings: The Meaning of English Cardinals and the Legacy of Paul Grice.
Oxford: Elsevier.
Carston, R. (1988) “Implicature, Explicature, and Truth-Conditional Semantics,” in R. Kempson (ed.)
Mental Representations: The Interface Between Language and Reality, 155–81. Cambridge: Cambridge Uni-
versity Press.
Carston, R. (1995) “Quantity Maxims and Generalized Implicatures,” Lingua 96: 213–44.
Carston, R. (2002) Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell.
Carston, R. (2004) “Relevance Theory and the Saying-Implicating Distinction,” in Horn and Ward (eds.),
633–56.
Chapman, S. (2005) Paul Grice: Philosopher and Linguist. Basingstoke: Palgrave Macmillan.
Chierchia, G. (2004) “Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interface,” in
A. Belletti (ed.) Structures and Beyond, 39–103. Oxford: Oxford University Press.
Chierchia, G., D. Fox, and B. Spector (forthcoming) “The Grammatical View of Scalar Implicatures and the

64
 I M PLI CAT U R E

Relationship Between Semantics and Pragmatics”, in C. Maienborn et al. (eds.) Handbook of Semantics:
An International Handbook. Berlin: Mouton de Gruyter.
Davidson, D. (1986) “A Nice Derangement of Epitaphs,” in R. Grandy and R. Warner (eds.) Philosophical
Grounds of Rationality, 157–74.
Doyle, J. (1951) “In Defense of the Square of Opposition,” The New Scholasticism 25: 367–96.
Dummett, M. (1973) Frege: Philosophy of Language. Oxford: Duckworth.
Fogelin, R. (1967) Evidence and Meaning. New York: Humanities Press.
Frege, G. (1892) “On Sinn and Bedeutung,” Translated by M. Black, in Beaney (ed.) 151–71.
Frege, G. (1897) “Logic,” unpublished paper, appearing in H. Hermes et al. (eds.) Nachgelassene Schriften,
139–61. Hamburg: Felix Meinen, 1969. Extract translated by P. Long and R. White, reprinted in Beaney
(ed.) 227–50.
Frege, G. (1918–19) “Thought,” translated by P. Geach and R. H. Stoothoff, in Beaney (ed.) 325–45.
Gazdar, G. (1979) Pragmatics. New York: Academic Press.
Geurts, B. (1998) “Scalars,” in P. Ludewig and B. Geurts (eds.) Lexikalische Semantik aus kognitiver Sicht,
95–117. Tübingen: Gunter Narr Verlag.
Geurts, B. (2009) “Scalar Implicature and Local Pragmatics,” Mind and Language 24: 51–79.
Geurts, B. (2010) Quantity Implicatures. Cambridge: Cambridge University Press.
Grice, H. P. (1961) “The Causal Theory of Perception,” Proceedings of the Aristotelian Society, Supplementary
Volume 35: 121–52.
Grice, H. P. (1967) “Logic and Conversation,” William James lectures delivered at Harvard University.
Grice, H. P. (1968) “Utterer’s Meaning, Sentence-Meaning, and Word-Meaning,” Foundations of Language
4: 225–42.
Grice, H. P. (1989) Studies in the Way of Words. Cambridge, MA: Harvard University Press.
Gutzmann, D. (2008) On the Interaction Between Modal Particles and Sentence Mood in German. Mainz:
German Institute, Johannes Gutenberg University.
Hamilton, Sir W., of Edinburgh (1860) Lectures on Logic, Volume I. Edinburgh: Blackwood.
Hirschberg, J. (1991) A Theory of Scalar Implicature. New York: Garland.
Hoffmann, M. (1987) Negatio Contrarii: A Study of Latin Litotes. Assen: Van Gorcum.
Horn, L. (1972) On the Semantic Properties of Logical Operators in English. PhD dissertation, UCLA. Distrib-
uted by Indiana University Linguistics Club, 1976.
Horn, L. (1984) “Toward a New Taxonomy for Pragmatic Inference: Q-based and R-based Implicature,” in
D. Schiffrin (ed.) Meaning, Form, and Use in Context (GURT ‘84), 11–42. Washington, D.C.: Georget-
own University Press.
Horn, L. (1989) A Natural History of Negation. Chicago: University of Chicago Press. (Reissued with a new
introduction, Stanford: CSLI, 2001.)
Horn, L. (1990) “Hamburgers and Truth: Why Gricean Inference is Gricean,” BLS 16, 454–71.
Horn, L. (1992) “The Said and the Unsaid,” SALT II, 163–92. Columbus: Ohio State University, Depart-
ment of Linguistics.
Horn, L. (2004) “Implicature,” in Horn and Ward (eds.) 3–28.
Horn, L. (2006) “The Border Wars: a Neo-Gricean Perspective,” in K. Turner and K. von Heusinger (eds.)
Where Semantics Meets Pragmatics, 21–48. Oxford: Elsevier.
Horn, L. (2007a) “Neo-Gricean Pragmatics: a Manichaean Manifesto,” in N. Burton-Roberts (ed.) Pragmat-
ics, 158–83. Basingstoke: Palgrave Macmillan.
Horn, L. (2007b) “Toward a Fregean Pragmatics: Voraussetzung, Nebengedanke, Andeutung,” in I. Kecskes
and L. Horn (eds.) Explorations in Pragmatics: Linguistic, Cognitive, and Intercultural Aspects, 39–69. Berlin:
Mouton de Gruyter.
Horn, L. (2008) “‘I love me some him’: The Landscape of Non-Argument Datives,” in O. Bonami and P.
Cabredo Hofherr (eds.) Empirical Issues in Syntax and Semantics 7: 169–92. Downloadable at http://www.
cssp.cnrs.fr/eiss7
Horn, L. (2009) “WJ-40: Implicature, Truth, and Meaning,” International Review of Pragmatics 1: 3–34.
Horn, L. and G. Ward (eds., 2004) The Handbook of Pragmatics. Oxford: Blackwell.
Huang, Y. (2006) Pragmatics. Oxford: Oxford University Press.
Hungerland, I. (1960) “Contextual Implication,” Inquiry 3: 211–58.
Hurewitz, F., A. Papafragou, L. Gleitman and R. Gelman (2006) “Asymmetries in the Acquisition of
Numbers and Quantifiers,” Language Learning and Development 2: 77–96.
Jaszczolt, K. (2005) Default Semantics: Foundations of a Compositional Theory of Acts of Communication.
Oxford: Oxford University Press.

65
 LAU REN CE H O R N

Kaplan, D. (1999) “The Meaning of Ouch and Oops: Explorations in the Theory of Meaning as Use”, unpub-
lished ms., UCLA.
Katsos, N. (2008) “The Semantics/Pragmatics Interface from an Experimental Perspective: The Case of
Scalar Implicature,” Synthese 165: 385–401.
Katzir, R. (2007) “Structurally Defined Alternatives,” Linguistics and Philosophy 30: 669–690.
Koenig, J.-P. (1991) “Scalar Predicates and Negation: Punctual Semantics and Interval Interpretations,”
CLS 27, 140–55.
Leech, G. (1983) Principles of Pragmatics. London: Longman.
Levinson, S. C. (2000) Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cam-
bridge, MA: MIT Press.
Lewis, D. (1969) Convention. Cambridge, MA: Harvard University Press.
Mill, J. S. (1867) An Examination of Sir William Hamilton’s Philosophy (3d edn). London: Longman.
Monck, W. H. S. (1881) Sir William Hamilton. London: Sampson, Low.
Noveck, I. and A. Posada (2003) “Characterizing the Time Course of an Implicature: An Evoked Potentials
Study,” Brain and Language 85: 203–10.
Nowell-Smith, P. H. (1954) Ethics. Harmondsworth: Pelican Books.
Papafragou, A. and J. Musolino (2003) “Scalar Implicatures: Experiments at the Semantics-Pragmatics
Interface,” Cognition 86: 253–82.
Paul, H. (1889) Principles of the History of Language, translated by H.A. Strong. London: Macmillan.
Potts, C. (2005) The Logic of Conventional Implicatures. Oxford: Oxford University Press.
Potts, C. (2007) “The Expressive Dimension,” Theoretical Linguistics 33: 165–98.
Recanati, F. (2001) “What is Said,” Synthese 128: 75–91.
van Rooij, R. (2009) “Optimality-theoretic and Game-theoretic Approaches to Implicature,” in E. Zalta
(ed.) Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/archives/spr2009/entries/implicature-
optimality-games/
Russell, B. (2006) “Against Grammatical Computation of Scalar Implicatures,” Journal of Semantics 23:
361–82.
Sauerland, U. (2004) “Scalar Implicatures in Complex Sentences,” Linguistics and Philosophy 27: 367–91.
Saul, J. (2002) “What is Said and Psychological Reality: Grice’s Project and Relevance Theorists’ Criti-
cisms,” Linguistics and Philosophy 25: 347–72.
Sperber, D. and D. Wilson (1986) Relevance: Communication and Cognition. Oxford: Blackwell.
Williamson, T. (2003) “Understanding and Inference,” Aristotelian Society Supp. 77: 249–93.
Williamson, T. (2009) “Reference, Inference, and the Semantics of Pejoratives,” in J. Almog and
P. Leonardi (eds.) Philosophy of David Kaplan, 137–59. Oxford: Oxford University Press.
Wilson, D. and D. Sperber (2004) “Relevance Theory,” in Horn and Ward (eds.) 607–32.
Zipf, G. K. (1949) Human Behavior and the Principle of Least Effort. Cambridge, MA: Addison-Wesley.

66
 1.6
P RAGM AT IC
E NRI C HM ENT
François Recanati

1 Pragmatic Enrichment and Conversational Implicature

The phrase “pragmatic enrichment” is sometimes used in a broad sense to refer to the
process by virtue of which the content conveyed by an utterance comes to include
all sorts of elements which are contextually implied without being part of what the
utterance literally means. In his class notes “Pragmatic Enrichment: Introduction by
Examples,” Chris Potts (2007) gives examples like the following:

(1) John and Mary have recently started going together. Valentino is Mary’s
ex-boyfriend. One evening, John asks Mary, “Have you seen Valentino
this week?” Mary answers, “Valentino’s been sick with mononucleosis for
the past two weeks.” Valentino has in fact been sick with mononucleosis
for the past two weeks, but it is also the case that Mary had a date with
Valentino the night before.

Mary’s utterance clearly suggests a negative (and false) answer to the question: “Have
you seen Valentino this week?” Literally, however, she only says that Valentino has
been sick with mononucleosis for the past two weeks, and that is true.
In his William James Lectures delivered at Harvard in the late 1960s, Grice described
the mechanism through which one can mean one thing by saying another (Grice 1989).
The term coined by Grice for that mechanism is “conversational implicature.” Thus,
in (1), Mary ‘says’ that Valentino has been sick with mononucleosis for the past two
weeks, and ‘conversationally implicates’ that she has not seen him this week.
On Grice’s account, implicatures are derived through an inference which enables the
interpreter to grasp the speaker’s communicative intention despite the fact that it’s not
articulated in words. The inference makes use of two crucial premises (in addition to
background knowledge): (1) the fact that the speaker has said what she has said (here,
that Valentino has been sick etc.), and (2) the fact that the speaker, qua communica-
tor, obeys (or is supposed to obey) the rules of communication or, as Grice puts it, the
“maxims of conversation.” Because of the role played by the first premise in the infer-
ence through which they are derived, conversational implicatures belong to the ‘post-
semantic’ layer of interpretation. This means that, in order to derive the implicatures
of an utterance, an interpreter must start by identifying what the speaker literally says.
 F RA N ÇO I S RE C A N AT I

(The properly semantic layer of interpretation corresponds to the identification of what

is said—the proposition literally expressed.)
It is customary in the literature to equate ‘pragmatic’ and ‘post-semantic,’ but this is
not quite right, for two reasons. First, there is a ‘pre-semantic’ layer of interpretation
which, like the post-semantic layer, involves a good deal of pragmatic reasoning aimed
at figuring out the speaker’s communicative intention. Semantic interpretation takes
disambiguated expressions as inputs, and computes their contents as a function of the
contents of their constituents and the way they are syntactically combined (plus the
context, when the sentence contains indexicals or context-sensitive expressions). No
such computation can take place unless the sentence has been properly disambiguated.
In this pre-semantic task, background knowledge and the presumption that the speaker
obeys the maxims of conversation play an obvious role. As David Kaplan points out, if
a haberdasher says “I am out of checks” to a banker, “whether the utterance takes place
in the store or at the bank will help the banker to determine what the haberdasher has
said” (Kaplan 1978: 229), and it will do so by pre-semantically selecting the relevant
reading for the word “check.” Pragmatics is therefore involved at the pre-semantic level,
just as it is involved at the post-semantic level.
Pragmatics is also involved at the semantic level, and that is the second reason why
one should reject the equation of ‘pragmatic’ and ‘post-semantic.’ To compute the
proposition expressed by an utterance, it is necessary to assign contextual values to
indexicals, (unbound) pronouns etc. For example, if the speaker uses a demonstrative
pronoun and says, ‘“She is cute,” the hearer must determine who the speaker means by
“she” in order to fix the utterance’s truth-conditional content. This contextual assign-
ment of values to indexicals and pronouns is a pragmatic affair, just like disambiguation
or the derivation of implicatures. In all three cases one has to use contextual cues (and
the presumption that the speaker obeys the maxims of conversation) to figure out what
the speaker’s intentions are.
To sum up, utterance interpretation involves pragmatic reasoning at three levels:
(i) at the pre-semantic level one must determine which (disambiguated) sentence is
intended among various candidates compatible with the proferred signal; (ii) at the
semantic level one must assign contextual values to the context-sensitive elements in
the sentence in order to determine the proposition expressed; (iii) at the post-semantic
level one must derive the conversational implicatures. Only in the last type of case (the
implicatures) does the inference presuppose that the proposition expressed (what is
said) has been identified.
Now, pace Potts, the phrase ‘pragmatic enrichment’ on its standard use is meant to
contrast with ‘conversational implicature,’ instead of being a non-technical synonym for
it (as it is for Potts). On the standard use, while conversational implicatures belong to
the post-semantic layer of interpretation, pragmatic enrichment belongs to the semantic
layer: It affects the proposition expressed by an utterance. This puts pragmatic enrich-
ment in the same ballpark as the assignment of contextual values to indexicals and free
pronouns.

2 Pragmatic Enrichment and Semantic Under-specification

What, then, is the difference between pragmatic enrichment and the contextual
assignment of values to indexicals and pronouns if both belong to the semantic layer of
interpretation? Following Kaplan, let us call ‘content’ the contribution an expression

68
 PRA G M ATI C ENR I C H ME N T

makes to the truth-conditions of the utterance in which it occurs. Context-sensitive

expressions are such that their linguistic meaning does not fix their content: their con-
tent depends upon the context. In the case of standard indexicals, pronouns etc., the
expression is like a free variable to which a content must be assigned in context, and its
linguistic meaning merely constrains the contents which may be contextually assigned
to it. For example, the linguistic meaning of the pronoun “she” in the above example
carries gender and number features which (presuppositionally) constrain the reference
of the pronoun: the reference must be a female individual. The content of the pro-
noun is whatever is assigned to it in the context of utterance (in virtue of the speaker’s
intentions etc.), provided the assigned value satisfies the presuppositions carried by the
pronoun.
In other cases, a contextual value is not directly assigned to the expression, but the
content of the expression still depends upon the context. For example, ‘ready’ and ‘tall,’
without an accompanying prepositional phrase, are context-sensitive because they can
be used to ascribe different properties in different contexts—the property of being ready
for the walk vs ready for dinner, or the property of being tall for a sixth-grader vs. tall for
a basketball player. Some authors talk of pragmatic enrichment in that sort of case: the
meaning of ‘ready’ or ‘tall,’ they say, is under-specified and has to be pragmatically
enriched through the contextual provision of a meaning constituent expressible by
means of a prepositional phrase (‘for dinner,’ ‘for a sixth-grader’). Since the enrichment
in question is mandatory, the meaning of the semantically under-specified item can be
represented as involving a free variable to which a value must be contextually assigned:
thus ‘ready’ arguably means something like ready for x and ‘tall’ means something like
tall for an X. In the syntax, however, there arguably is no variable—only a semantically
under-specified lexical item whose meaning carries what Kratzer (2009) has called a
“conceptual” variable. The variable is introduced in the semantics, via the lexical entry
for the item, while in the case of pronouns and indexicals, the lexical item occurring in
the syntactic structure is itself (like) a variable.
The treatment of context-sensitive expressions like ‘ready’ and ‘tall’ I have just
described is actually controversial. Some authors think that, associated with lexical
items like ‘ready’ and ‘tall,’ there is a (covert) variable/pronominal expression in the
syntax (Stanley 2000, Marti 2006). In context, a value must be assigned to that covert
variable, just as a contextual value must be assigned to overt pronouns. If this is true,
then there is no need to talk of semantic under-specification or pragmatic enrichment
in that sort of case—it is just an ordinary case of contextual assignment of value to
pronominal elements in syntactic structure.
In other cases, however, the idea of semantic under-specification triggering pragmatic
enrichment may be harder to resist. Thus Kratzer (2010) says that “pragmatic enrich-
ment is not just allowed, but required, for the interpretation of noun-noun compounds
like swan boat.” Neither ‘swan’ nor ‘boat’ is context-sensitive (let us assume)—both lexi-
cal items have a content independent of context—but the compound ‘swan boat’ is: its
content must be pragmatically fixed through the contextual provision of a relevant rela-
tion R between swans and boats, such that only boats bearing that relation to swans end
up in the extension of ‘swan boat.’ What we seem to have here is constructional context-
sensitivity: the context-dependence of the content of ‘swan boat’ is not inherited from
the context-dependence of the contents of its parts, but comes from the construction
itself. The content of the noun–noun construction is semantically under-specified: it has
to be fleshed out through the contextual provision of the relevant relation R.

69
 F RA N ÇO I S RE C A N AT I

To be sure, there have been attempts to provide ‘indexical’ analyses of compound

nouns (CNs). Thus, according to Weiskopf (2007: 175),

The linguistic meaning of CNs contains a phonologically unrealized open vari-

able or otherwise indexical expression that picks out the relation among their
constituents, and this relation is filled in by pragmatic mechanisms that draw
on features of the context of utterance.

But the variable R* which Weiskopf posits is introduced via the semantic rules associated
with the noun–noun construction, which he calls the “rules of CN Interpretation,” e.g.:

[[N1 N2]] =Ox[N’2(x)&R*(x,N’1)]

So it is, as he puts it, “the meaning of CNs” which carries the variable. The variable in
question is therefore a conceptual variable and is not supposed to be present in the syn-
tax. (In this respect it is somewhat misleading to say that the variable is “phonologically
unrealized,” for that notion only applies to elements present in syntactic structure.)

3 Free Pragmatic Enrichment

I have just illustrated the use of the phrase ‘pragmatic enrichment’ in connection with
semantic under-specification and conceptual variables. On that use it is possible to say,
as Kratzer does, that pragmatic enrichment is required for the interpretation of a given
type of expression. But that is not the standard use of the phrase ‘pragmatic enrichment.’
On the standard use, pragmatic enrichment is never required. It is the standard use of
the phrase (and the phenomenon it designates) that I will focus on in what follows.
On the standard use, just as it contrasts with conversational implicatures, pragmatic
enrichment contrasts with the assignment of values to free variables, whether the varia-
bles in question are construed as conceptual (introduced via lexical entries) or syntactic. What
characterizes the contextual assignment of values to free variables of all categories is
that it is mandatory: no proposition is expressed unless a value is assigned to the variable.
Thus, if I say “John is ready,” but the context does not answer the question “Ready for
what?”, I haven’t said anything definite; I haven’t expressed a complete proposition.
Similarly, if I say “she is cute,” but no individual can be contextually singled out as
the reference of the demonstrative pronoun ‘she’, no proposition has been expressed.
Likewise, if I use the phrase ‘swan boat’, but the context does not specify any relevant
relation R between swans and boats, my utterance fails to express a complete content.
In all these cases, where the provision of some contextual element is mandatory, I talk
of ‘saturation’: a slot has to be contextually filled, leaving the utterance semantically
incomplete if it remains unfilled. In contrast, pragmatic enrichment—or, more explic-
itly, free pragmatic enrichment—is not mandatory but optional. There is enrichment
in that sense only when the following three conditions are jointly satisfied: (i) the
context adds some element to the interpretation of the utterance; (ii) that element is
truth-conditionally relevant—it affects the proposition expressed (so this is unlike con-
versational implicatures); yet (iii) its contextual provision is not necessary, in the sense
that if that element was left aside, the utterance would still express a complete proposi-
tion (this is in contrast to indexicals, pronouns, and the other type of context-sensitive
expressions mentioned above).

70
 PRA G M ATI C ENR I C H ME N T

Consider the following example from Robyn Carston:

(2) Jean went to Austria last week and she ran into John.

One naturally interprets the second conjunct (“she ran into John”) as expressing the
proposition that Jean ran into John in Austria. The location of the event of running
into John is contextually assumed to be the same as the location mentioned in the
first conjunct, but that assumption is defeasible, as contextual assumptions typically
are. What is important, however, is that the contextual provision of a location for the
event mentioned in the second conjunct is not mandatory: it is perfectly correct to
interpret a sentence like “she ran into John” as saying that Jean ran into John in some
place or other, without the place in question being contextually specified or specifiable.
This is in contrast with standard indexicals or pronouns: For the sentence does not
express a complete proposition unless the indexicals or unbound pronouns it contains
are assigned definite values.

4 Pragmatic Modulation in General

Free enrichment is only one type of case in which a pragmatic process that is optional
(in contrast to the contextual assignment of values to indexicals and free variables of all
sorts) nevertheless affects the proposition expressed. In Literal Meaning and subsequent
work, I used the term ‘modulation’ (as opposed to ‘saturation’) as a generic term for the
entire class. The general idea is this:

In context the meaning of words is adjusted or ‘modulated’ so as to fit what is

being talked about. Sense modulation is essential to speech, because we use a
(more or less) fixed stock of lexemes to talk about an indefinite variety of things,
situations and experiences. Through the interaction between the context-inde-
pendent meanings of our words and the particulars of the situation talked about,
contextualised, modulated senses emerge, appropriate to the situation at hand.
(Recanati 2004: 131)

Saturation is a pragmatic process of contextual value-assignment that is triggered (and

made obligatory) by something in the sentence itself, namely, the linguistic expression
to which a value is contextually assigned. The expression itself acts as a variable in
need of contextual instantiation. So pragmatics comes into play (in order to determine
which value to assign in context to the variable), but it does so under the guidance of
the linguistic material: the pragmatic process of saturation is a ‘bottom-up’ process in
the sense that it is signal-driven, not context-driven. In contrast, modulation is a ‘top-
down’ or ‘free’ pragmatic process, i.e. a process which is not triggered by an expression
in the sentence but is context-driven and takes place for purely pragmatic reasons—in
order to make sense of what the speaker is saying.
Modulation operates on the meaning of an expression (simple or complex) and
returns a meaning of the same type. In free enrichment, the output sense is more specific
than the input sense. Free enrichment typically works by narrowing down the exten-
sion of an expression through the contextual addition of a component to its meaning.
For example, through free enrichment, the (intransitive) verb ‘to smoke’ can come to
mean to smoke marijuana, if the context makes it clear that that is what we are talking

71
 F RA N ÇO I S RE C A N AT I

about. What is added to the meaning of the verb is a specification of its internal accusa-
tive—the type of thing smoked—just as in (2) what is added is a specification of the
location of the event of running into John.
Sense extension (also called ‘loosening’ or ‘broadening’) is another modulation proc-
ess which works in the opposite direction: The meaning of an expression is made less
specific, and its extension typically wider, through the contextual deletion of a compo-
nent of its meaning (Cohen 2002: chapters 3 and 5; Franks 1995). Thus consider:

(3) The ATM swallowed my credit card

This can be given a literal reading, if we imagine a context à la Putnam in which ATMs
turn out to be living organisms (for only living organisms can swallow). But the sen-
tence can also and typically will be interpreted non-literally. In an ordinary context,
‘swallow’ will be given an extended sense, corresponding to what ATMs sometimes do
with credit cards (something which, indeed, resembles swallowing). The sentence may
be true, on such a reading, even though no real swallowing takes place.
Another key modulation process is predicate transfer (Nunberg 1995). In predicate
transfer the input sense is a property F, and the output sense is another property F’,
definable in terms of F, along the following lines:

F’ =def Ox [(Qy: Rx,y) Fy]

In this schema representing the transferred property F’, “Q” and “R” are placeholders
for, respectively, a quantificational determiner and a relation that is left free (within
certain limits) and will be determined in context. So F’ is the property one has when-
ever, for Q object y one bears relation R to, y is F. In Nunberg’s famous example, “I am
parked out back,” F is the property of being parked out back (a property only a vehicle
can have), R is the ownership relation between cars and car owners, and Q may be the
existential quantifier. F’, then, is the property of having a car that is parked out back. It
is that property which the words ‘parked out back’ arguably ascribe to the speaker in “I
am parked out back.”

5 The ‘Syntactic’ Approach

Even though free enrichment is only one type of modulation among others, the phrase
‘pragmatic enrichment’ is often used as a generic term, synonymous with my own term
‘pragmatic modulation’. One reason for this is that modulation can be thought of as
operating on expressions (‘syntactic’ construal), rather than directly on their meanings
(‘semantic’ construal). When so construed, modulation amounts to supplementing the
overt expression with implicit elements, and so it counts as a form of ‘enrichment’,
whatever its semantic effect.
On the semantic construal of pragmatic modulation, the meaning of an expression
is mapped to a distinct meaning. On the syntactic construal, it is the expression (or a
corresponding mental representation) that is mapped to a more complex expression/
representation through the addition of extra elements. For example, a sentence like (4)
has several readings:

(4) There is a lion in the middle of the piazza

72
 PRA G M ATI C ENR I C H ME N T

On one reading ‘lion’ is given a non-literal interpretation through predicate transfer and
(4) may be true even if, literally, there is no lion in the middle of the piazza (but only a
statue of a lion). An advocate of the ‘syntactic’ approach to pragmatic modulation will
argue that ‘lion’ is mapped to ‘representation of a lion’ through the addition of an opera-
tor representation of. Likewise, sense extension will be said to proceed through the
addition of some kind of similarity operator. On this view pragmatic modulation always
is a form of enrichment or supplementation of a (linguistic or mental) representation,
making it into a more complex representation.
The syntactic approach to pragmatic modulation has one putative advantage over the
semantic approach:It can provide a simple response to an argument against pragmatic
enrichment known as the ‘binding argument’. According to the argument, implicit ele-
ments of content (e.g. the implicit reference to Austria in the second conjunct of (2))
cannot, in general, be due to free pragmatic enrichment, because the elements in ques-
tion can be made to vary with the cases introduced by a quantifier, as in

(5) Whenever I light a cigarette, it starts raining.

On the most salient reading, this means that whenever I light a cigarette, it starts raining
at the location where I light the cigarette. Here the location of the raining event is an implicit
element of content, as is the location of the event of running into John in example (2)
above, but it varies with the cases introduced by “whenever I light a cigarette.” Implicit
bound readings like this raise a problem for pragmatic enrichment accounts, Stanley
(2000) argued, because binding requires a variable that can be bound (here, a location
variable), and variables are syntactic objects. According to free enrichment accounts
of implicit content, however, the location in (2), (5) and similar examples is supposed
to be inarticulated in the syntax, which means that there is no location variable in the
syntax which could be bound; nor could the variable be generated through pragmatic
enrichment, since pragmatic enrichment (on the semantic construal) only maps mean-
ings to meanings. However, assuming that this is correct and that implicit bound read-
ings cannot be generated through pragmatic enrichment semantically construed, they
can still be generated through pragmatic enrichment syntactically construed. Nothing
prevents the syntactic-enrichment process from contextually supplementing the overt
sentence with implicit variables or pronouns directly or indirectly bound by the overt
quantifier (Recanati 2002: 338–42).

6 Two Variants of the ‘Syntactic’ Approach

There are two main variants of the ‘syntactic’ approach to pragmatic modulation: the
natural-language (NL) variant and the language-of-thought (LOT) variant.
On the NL variant, pragmatic enrichment is the process whereby a natural language
expression (that which is actually uttered, e.g. the word ‘lion’ in the above example)
is mapped to another natural language expression (‘lion statue’), which is not uttered
but might have been, and which can be contextually recovered from the sentence that
is actually uttered. This is the ‘shorthand’ view put forward (inter alia) by Kent Bach
(1994, 2000) and Stephen Neale (2000, 2004). On this view modulation works by
enriching or supplementing the sentence with contextually recoverable extra materials,
as in the following examples, from Kent Bach:

73
 F RA N ÇO I S RE C A N AT I

Jack and Jill went up the hill [together].

Jack and Jill are married [to each other].
Jill got married and [then] got pregnant.
You’ll succeed if [and only if] you work hard.
There are [approximately] 30 students per class.
France is hexagonal [roughly speaking].
Otto has [exactly] three cars.
Felix has always been an honest judge [since he’s been a judge].
Adele hasn’t had lunch [today].
You’re not going to die [from this cut].

As Bach (2000: 262–63) puts it, “we generally speak loosely, omitting words that could
have made what we meant more explicit and letting our audience fill in the gaps.” In
such cases “what the speaker means is a qualified version of what he is saying. He would
have to utter the words in brackets (or roughly equivalent words—the exact words don’t
matter) to make what he meant more explicit.” The words in question “are not part of
the sentence (these are not cases of syntactic ellipsis) but the speaker could have uttered
them if he deemed that necessary.”
On the LOT variant, put forward by relevance theorists (Sperber and Wilson 1986,
Carston 2002) and by Jerry Fodor (2001), natural language expressions are assumed to
translate into language of thought strings (sometimes called ‘semantic’ or ‘conceptual’
representations). Pragmatic enrichment is the process whereby the language of thought
string into which the uttered expression translates is contextually enriched with extra
symbols, to which nothing corresponds in the uttered expression. It is the richer lan-
guage of thought string resulting from the enrichment process which determines the
(intuitive) truth-conditions of the utterance: the truth-conditions of the utterance
are the truth-conditions one gets by subjecting the contextually enriched LOT string
(rather than the natural language sentence or its mentalese translation) to model-
theoretic interpretation.
As Paul Elbourne points out, the NL variant is more constrained than the LOT vari-
ant. “Enrichment in [the first kind of] theory can only yield meanings that could be
obtained by syntactically building on the material present in the syntax [of the natu-
ral language expression]. No such constraint is present in the [other kind of] theory”
(Elbourne 2008: 95). “Theories of [the second] kind, then, are potentially vulnerable
in the following way: if a reading turns out to be unavailable for a given sentence, they
cannot avail themselves of syntactic constraints to explain its absence, . . . and they will
face real difficulty if the reading in question can be argued to be pragmatically plausible”
(Elbourne 2010). Still, it is easy to introduce an analogous constraint into the LOT
variant, and it is that sort of hybrid view which Stephen Neale seems to be defending.
On such a hybrid view a LOT string is considered as an admissible enrichment of (the
mentalese translation of) the uttered sentence only if it is the mentalese translation of a
sentence one can get by supplementing the uttered sentence with extra materials.
Even if we take the syntactic constraint on board, there is an obvious overgenera-
tion objection looming. Many readings one could get by freely enriching the uttered
sentence with extra words are simply not available (Stanley 2002, 2005, Elbourne
2010). Extra constraints will therefore be needed to filter out the absent readings.
(See Elbourne 2008 and Hall 2008 for attempts to address some of the overgeneration
worries.) Overgeneration is also a problem for semantic approaches to pragmatic

74
 PRA G M ATI C ENR I C H ME N T

modulation. In general, pragmatic modulation/enrichment is such a powerful mech-

anism that it needs to be suitably constrained. Thus predicate transfer as a general
mechanism ought to make it possible to interpret the sentence “I was once driven by
Jean Gabin” to mean that the speaker owns a car that was once driven by Jean Gabin.
That reading is very unlikely, however, for reasons that are essentially pragmatic and
are well-worth investigating (see Nunberg 1979, 1995 for a pragmatic approach to the
constraints on predicate transfer).

7 Is Pragmatic Enrichment Relevant to Semantics?

It is tempting to appeal to pragmatic enrichment/modulation whenever an element of
meaning, e.g. the location of the event of running into John in (2), remains implicit yet
seems to be truth-conditionally relevant. The standard strategy in linguistics consists in
dealing with such examples by positing unpronounced elements in the syntax. But in
many cases, such covert elements can only be detected by pragmatic means, and there
is only a short step from the idea that they are detected on a pragmatic basis to the idea
that they are generated on a pragmatic basis. As Ariel Cohen (2009) puts it, if an ele-
ment is phonologically null, the hearer can’t hear it; if the hearer nevertheless proceeds
as if that element were present, s/he must apply some reinterpretation of the input so
as to introduce that element. Modulation is the case in which such reinterpretation
proceeds on a purely pragmatic basis.
Despite the appeal of the idea, and the fact that it promises to simplify both syntax
and semantics, there is deep resistance to it and it has been rather underutilized in formal
semantics. In the final sections of this entry, I discuss what I take to be the main sources
of resistance, besides the overgeneration worry which has already been mentioned.
First and foremost, there is resistance to the very idea that free pragmatic processes
might affect ‘what is said’—the proposition expressed by an utterance—as suggested by
the idea that modulation belongs to the semantic layer of interpretation. Free pragmatic
processes are pragmatic processes that are not mandated by the linguistic material but
take place for wholly pragmatic reasons, in order to make sense of the speaker. They are
‘top down’ processes, rather than ‘bottom up’ pragmatic processes triggered by some-
thing in the linguistic signal. That such processes could affect what is said seems to be
a contradiction in terms. ‘What is said’, the truth-conditional content of an utterance,
is what is literally said, and that—by definition—has to be determined by the conven-
tions of the language. Pragmatics can enter the picture, provided its role is to assign a
contextual value to a lexical item in a bottom-up manner, i.e. in accord with (and under
the guidance of) the conventional meaning of that context-sensitive item. In contrast,
what King and Stanley (2005: 118–19) call “strong pragmatic effects,” i.e. pragmatic
effects achieved in order to make sense of the speech act without being linguistically
mandated, take us into the realm of speaker’s meaning, away from literal meaning.
Insofar as this argument is based upon a certain understanding of the phrase ‘what is
said’ (or ‘what is literally said’), it is not substantive, but verbal. There is no doubt that
one can define ‘what is said’ in such a way that only weak pragmatic effects can affect
what is said. But what the advocate of pragmatic modulation means by ‘what is said’
corresponds to the intuitive truth-conditional content of the utterance. Now, the intuitive
truth-conditions of an utterance of (3) or (4) above are affected by free pragmatic proc-
esses of predicate transfer or sense extension: (4) can be intuitively true even though
no real lion stands in the middle of the piazza, and (3) can be true even though no

75
 F RA N ÇO I S RE C A N AT I

real swallowing takes place. So there is a sense of ‘what is said’ in which what is said is
affected by pragmatic modulation. Assuming this is true, this does not prevent us from
defining another notion of what is said, conforming to literalist standards.
The second objection to pragmatic modulation (or rather, to the idea that it affects
semantic content) is the following. If free pragmatic processes are allowed to affect
semantic content, semantic content leaps out of control—it is no longer determined by
the rules of the language but varies freely, à la Humpty Dumpty. But then, how can we
account for the success of communication? Communication (content sharing) becomes
a miracle since there is nothing to ensure that communicators and their addressees
will converge on the same content. Now communication is possible (it takes place all
the time), and there is no miracle. It follows that we should give up the view that free
pragmatic processes play a role in the determination of semantic content (Cappelen and
Lepore 2005: chapter 8).
This argument fails, I believe, because the problem it raises is a problem for every-
body. Whenever the semantic value of a linguistic expression (e.g. the reference of a
demonstrative) must be pragmatically inferred, the question arises: What guarantees
that the hearer will be able to latch on to the exact same semantic value as the speaker?
Whether the pragmatic process at stake is saturation or modulation is irrelevant as far
as this issue is concerned, so the argument fails as an argument specifically intended to
cast doubt on pragmatic modulation.

8 The Systematicity Objection

The most important source of resistance to the idea of pragmatic modulation is the fear
that if something like pragmatic modulation is allowed as a determinant of semantic
content, the project of constructing a systematic, truth-conditional semantics for natu-
ral language will be doomed to failure. Here is a first sketch of an argument for that
conclusion:

In contrast to the contextual assignment of values to indexicals, modulation is

not driven by the linguistic meaning of words. Nothing in the linguistic mean-
ing of the words whose sense is modulated tells us that modulation ought to
take place. Modulation takes place purely as a matter of context, of ‘pragmat-
ics’; what drives it is the urge to make sense of what the speaker is saying. So
modulation is unsystematic. If we allow it as a determinant of semantic content,
we make it impossible to construct a systematic theory of semantic content.

Thus formulated, the argument is not convincing. Let us grant the objector that modu-
lation is unsystematic. Still, it is easy to make room for it within a systematic seman-
tics (Pagin and Pelletier 2007, Westerståhl forthcoming). In general, nothing prevents
unsystematic factors from being handled systematically, by being assigned their proper
place in the theory. In the case at hand, assuming the semantic approach to pragmatic
modulation, we can define a function mod taking as arguments an expression e and the
context c in which it occurs: the value of mod is the particular modulation function g
that is contextually salient/relevant/appropriate for the interpretation of that expres-
sion in that context. If no modulation is contextually appropriate and the expression
receives its literal interpretation, the value of mod will be the identity function. In this
framework, we can distinguish between the literal sense of a simple expression e, namely

76
 PRA G M ATI C ENR I C H ME N T

its semantic interpretation I(e), and the modulated sense M(e)c carried by an occurrence
of e in context c. The modulated sense of an expression e (in context c) results from
applying the contextually appropriate modulation function mod (e, c) = g to its seman-
tic interpretation I(e):

M(e)c = mod (e, c) (I(e)) = g (I(e))

So far, this is very standard: in distinguishing I(e) from M(e)c we are just appealing to
the traditional semantics/pragmatics distinction. What is not standard is the claim that
the semantic interpretation of a complex expression (e.g. a sentence) is a function of
the modulated senses of its parts and the way they are put together (Recanati 2010:
chapter 1). This is what examples like (3) and (4) suggest if we take at face value the
effects of modulation on truth-conditional content which they seem to display. On
the resulting view the semantic process of composition and the pragmatic process of
sense modulation are intertwined. For simple expressions their semantic interpretation
is their literal sense, but for complex expressions pragmatic modulation is allowed to enter
into the determination of semantic content. This is non-standard, certainly, but there is
nothing unsystematic about this view.
The systematicity objection can be understood differently, however. In introducing
modulation (in contrast to saturation), I said that in saturation “pragmatics comes into
play, but does so under the guidance of the linguistic material,” whereas modulation “is
not triggered by an expression in the sentence but takes place for purely pragmatic rea-
sons—in order to make sense of what the speaker is saying.” This suggests that, insofar
as it involves modulation, utterance interpretation does not significantly differ from
“the kind [of interpretation] involved in interpreting kicks under the table and taps
on the shoulder” (Stanley 2000: 396). Stanley objects that “we do not interpret these
latter sorts of acts by applying highly specific rules to structured representations,” as for-
mal semanticists say we do in the linguistic case. Accepting the semantic relevance of
pragmatic modulation thus seems incompatible with the program of formal semantics,
according to which utterance interpretation is a systematic affair and differs, in this
respect, from the (pragmatic) interpretation of gestures.
Thus understood, however, I think the objection is confused. Even though free prag-
matic processes, i.e. pragmatic processes that are not mandated by the standing meaning
of any expression in the sentence, are allowed to enter into the determination of truth-
conditional content, still, in the framework I have sketched, they come into the picture
as part of the compositional machinery. Semantic interpretation remains grammar-driven
even if, in the course of semantic interpretation, pragmatics is appealed to, not only
to assign contextual values to indexicals and free variables but also to freely modulate
the senses of the constituents in a top-down manner. Semantic interpretation is still a
matter of determining the sense of the whole as a function of the (possibly modulated)
senses of the parts and the way they are put together.

Acknowledgment
The research leading to this paper has received funding from the European Research
Council under the European Community’s Seventh Framework Programme (FP7/2007–
2013) / ERC grant agreement no 229 441—CCC.

77
 F RA N ÇO I S RE C A N AT I

References
Bach, K. (1994) “Conversational Impliciture,” Mind and Language 9: 124–62.
Bach, K. (2000) “Quantification, Qualification and Context,” Mind and Language 15: 262–83.
Cappelen, H. and Lepore, E. (2005) Insensitive Semantics, Oxford: Blackwell.
Carston, R. (2002) Thoughts and Utterances: The Pragmatics of Explicit Communication, Oxford: Blackwell.
Cohen, A. (2009) “No Quantification without Reinterpretation,” Workshop Genericity: Interpretation and
Uses, Paris, May 11–13, 2009.
Cohen, L. J. (2002) Knowledge and Language: Selected Essays, Dordrecht: Kluwer.
Elbourne, P. (2008) “The Argument from Binding,” Philosophical Perspectives 22: 89–110.
Elbourne, P. (2010) “Incomplete Descriptions and Sloppy Identity,” Contextualism and Compositionality
workshop, Paris, May 17–19 2010.
Fodor, J. (2001) “Language, Thought and Compositionality,” Mind and Language 16: 1–15.
Franks, B. (1995) “Sense Generation: A ‘Quasi-Classical’ Approach to Concepts and Concept Combina-
tion,” Cognitive Science 19: 441–505.
Grice, P. (1989) Studies in the Way of Words, Cambridge, MA.: Harvard University Press.
Hall, A. (2008) “Free Enrichment or Hidden Indexicals?” Mind and Language 23: 426–56.
Kaplan, D. (1978) “DTHAT,” Syntax and Semantics 9: 221–43.
King, J. and Stanley, J. (2005) “Semantics, Pragmatics, and the Role of Semantic Content,” in Z. Szabo (ed.)
Semantics versus Pragmatics, Oxford: Clarendon Press, 111–64.
Kratzer, A. (2009) Context and Modality, The Context and Content Lectures, Paris, October–December
2009.
Kratzer, A. (2010) Abstract for “Syntactic Constraints on Meaning Composition: Deriving the Meanings of
Conditionals,” Contextualism and Compositionality workshop, Paris, May 17–19, 2010.
Marti, L. (2006) “Unarticulated Constituents Revisited,” Linguistics and Philosophy 29: 135–66.
Neale, S. (2000) “On being Explicit: Comments on Stanley and Szabo, and on Bach,” Mind and Language
15: 284–94.
Neale, S. (2004) “This, That and the Other,” in M. Reimer and A. Bezuidenhout (eds.), Descriptions and
Beyond, Oxford: Clarendon Press, 68–182.
Nunberg, G. (1979) “The Non-Uniqueness of Semantic Solutions: Polysemy,” Linguistics and Philosophy 3:
143–84.
Nunberg, G. (1995) “Transfers of Meaning,” Journal of Semantics 12: 109–32.
Pagin, P. and Pelletier, J. (2007) “Context, Content and Communication,” in G. Preyer and G. Peter (eds.)
Context-Sensitivity and Semantic Minimalism, Oxford: Clarendon Press, 25–62.
Potts, C. (2007) “Pragmatic Enrichment: Introduction by Examples,” class notes, University of Massachu-
setts, Amherst, Fall 2007.
Recanati, F. (2002) “Unarticulated Constituents,” Linguistics and Philosophy 25: 299–345.
Recanati, F. (2004) Literal Meaning, Cambridge: Cambridge University Press.
Recanati, F. (2010) Truth-Conditional Pragmatics, Oxford: Oxford University Press.
Sperber, D. and Wilson, D. (1986) Relevance: Communication and Cognition, Oxford: Blackwell.
Stanley, J. (2000) “Context and Logical Form,” Linguistics and Philosophy 23: 391–434.
Stanley, J. (2002) “Making It Articulated,” Mind and Language 17: 149–68.
Stanley, J. (2005) “Semantics in Context,” in G. Preyer and G.Peter (eds.), Contextualism in Philosophy:
Knowledge, Meaning, and Truth, Oxford: Clarendon Press, 221–53.
Weiskopf, D. (2007) “Compound Nominals, Context, and Compositionality,” Synthese 156: 161–204.
Westerståhl, D. (forthcoming) “Compositionality in Kaplan Style Semantics,” in W. Hinzen, E. Machery,
and M. Werning (eds.) The Handbook of Compositionality, Oxford: Oxford University Press.

78
 1.7
ME ANI NG AND
COMMUNIC AT ION
Kent Bach

“If I didn’t know any words, you wouldn’t know what I mean.” This astute observation,
made by my granddaughter Sophia when she was four, might suggest that knowing what
a speaker’s words mean is all it takes to know what she means in using them and, that
communicating is just a matter of putting one’s thoughts into words. Sophia didn’t sug-
gest that and, indeed, the theme of this chapter is that communication is more compli-
cated than that. For even if you know what my words mean, you might not know what
I mean.

Introduction
Words mean things, speakers mean things in using words, and these need not be the
same. For example, if you say to someone who has just finished eating a super giant
burrito at the Taqueria Guadalajara, “You are what you eat,” you probably do not mean
that the person is a super giant burrito. So we need to distinguish the meaning of a
linguistic expression—a word, phrase, or sentence—from what a person means in using
it. To simplify matters, let us pretend that an utterance is always of a sentence (and, for
mnemonic purposes, let our imagined speaker be a she and hearer be a he).
This chapter is concerned with the relationship between linguistic meaning and
what speakers mean in using language. It will not take a stand on the nature of linguis-
tic meaning itself, a difficult question on which there are many views, some of which
are discussed elsewhere in this volume. However, we will assume that it is one thing
for a sentence to have a certain meaning (or meanings, if it ambiguous) and another
for a speaker to mean something, whether the same thing or something else, in using
it. This leaves open whether what words mean in a language ultimately comes down to
what speakers mean by them, as argued by Grice (1968) and by Schiffer (1972). We
will assume also that speakers ordinarily, as members of the same linguistic community,
share knowledge of the meanings of the expressions they use. This is part of their lin-
guistic knowledge, which also includes knowledge of phonological (or orthographic)
form and syntactic structure. What matters for us is that linguistic knowledge is only
part of the knowledge that people bring to bear when they communicate with one
another. Our examples will only begin to illustrate how, mostly for efficiency’s sake but
sometimes for other reasons, people commonly try to communicate more than what
their sentences mean, and often even manage to make themselves understood.
 KEN T BAC H

We will explore one particularly ingenious idea about this, due to Grice (1957). He
thought that communication involves a special sort of intention on the part of the
speaker and that successful communication involves a special sort of inference on the
part of the intended audience. In using a sentence to try to communicate something,
a speaker has an audience-directed intention that is in a certain way self-referential.
Specifically, the speaker intends the listener to figure out what the speaker means partly
on the supposition that the speaker intends him to do so. The hearer’s job is to figure out
what the speaker means, partly on the basis that he is intended to do so. This is possible
because unlike intentions in general, a communicative intention is one whose fulfill-
ment consists simply in its recognition. In zeroing in on what this involves, we will need
to keep in mind that people generally do not use sentences merely to communicate but
primarily to affect one another in various ways.

Linguistic Meaning and Speaker Meaning

What we mean is generally connected, though sometimes only remotely, to what our
words mean. To appreciate this, consider a case where there is no such connection at
all. Suppose to gain entrance into a private club you must utter a three-word sentence
whose words begin, respectively, with “a,” “b,” and “c.” You say, “Always be cool,” and
you are let in. Clearly the meanings of your words are irrelevant to what you mean (“I’m
a member—let me in”). You could just as well have said, “Antibodies battle chlamydia.”
But this is an exceptional case. Ordinarily the meanings of the words you use do matter.
Nevertheless, their meanings do not determine what you mean in using them. There are
various ways in which this can be.
First of all, this can be because of ambiguity. A sentence can have more than one
meaning because it contains an ambiguous expression, like ‘bar’ in (1),

(1) Because of his excessive drinking, Judge Jones was banned from the bar.

or because it is structurally ambiguous, like (2),

(2) The chicken is ready to eat.

Because of the lexical ambiguity in (1), a speaker, though being literal, could mean
either that Judge Jones was banned from legal practice or that he was banned from a
certain drinking establishment. Similarly, a literal speaker of the structurally ambiguous
(2) could be talking about either a hungry chicken or a broiled chicken. In each case
what the speaker means corresponds to only one of the things the sentence means.
Another way linguistic meaning can fail to determine what the speaker means is via
nonliterality. For instance, although sentence (4) means something analogous to what
(3) means,

(3) Farmer Frank was up to his ears in mud.

(4) Farmer Frank was up to his ears in debt.

a speaker is likely to mean something quite different. What he means is related to but
distinct from the linguistic meaning, since he means that Farmer Frank was only figura-
tively up to his ears in debt. In other cases, involving indirection, a speaker means what

80
 M EA N I N G A N D CO M MU N I C AT I O N

the sentence means (or one of the things it means, if it is ambiguous) but means some-
thing else as well. If a friend asks you for something to drink and you utter (5),

(5) There’s some beer in the fridge.

presumably you mean not only that there is some beer there but also that your friend
may help himself to some of it. Finally, an utterance can be both nonliteral and indirect,
as in a likely case of a mother saying (6) to her slightly sadistic son,

(6) I’m sure Felix likes having his tail pulled.

She means not only that the cat doesn’t like having his tail pulled but also that her son
should stop pulling it. These and similar examples (see Bach and Harnish 1979: chapter
4) illustrate different ways in which what the speaker means can depart from what the
sentence means. The speaker may mean one of the things the sentence means, as with
(1) or (2), something quite distinct from anything it means, as with (4), or both, as
with (5).
There is a minor complication here. We need to distinguish what a speaker means by
an expression and what she means in using it. This distinction is evident from (7), for
example,

(7) Dr. Frankenstein’s lavatory blew up.

where the speaker probably means laboratory by “lavatory.” This distinction is also
needed to handle utterances of ambiguous sentences like (1) and (2), where what a
speaker means by her sentence does not include each of the things it means. In these
cases what the speaker means by the sentence determines which meaning is operative
in her utterance of them. With (4), however, there is no linguistic ambiguity. Despite
what a speaker would mean in using the words “up to his ears,” she does not really mean
anything different by them than she would in uttering (3). The phrase seems not to be
ambiguous but rather to have two uses, one literal and one figurative, one corresponding
to its single meaning and the other a derivative one. The speaker is exploiting the single
(literal) meaning of her words in order to mean something else in using them. With (5)
the speaker means both what the sentence means and something else as well. When she
utters “There’s some beer in the fridge,” she means by those words just what they mean
(not quite, actually, since they do not specify the fridge in question). Yet in uttering (5)
she means more than just that, namely that the hearer may help himself to some beer.

Communicative Intentions
Intuitively, to mean something in uttering a sentence is to intend to communicate
something to one’s audience. But what is it for an intention to be communicative? In
his groundbreaking article “Meaning,” Grice (1957) offered an original answer to this
question. He observed that meaning something (trying to communicate it) is not sim-
ply a matter of thinking something and acting with the intention of somehow causing
one’s audience to entertain that thought. After all, one’s intention could be covert. You
might, for example, make self-deprecating remarks intending to get people to think you
are modest. They might think that but certainly not if they recognize your intention.

81
 KEN T BAC H

Nor is it enough that one’s intention be overt. Say you point to a cracked window with
the intention of getting someone to believe that the window is broken. Seeing that it is,
they will come to believe that but not by virtue of recognizing your intention.
Grice’s idea was that communicative intentions are intentionally overt and that this
feature plays a special role in their fulfillment. That is, in trying to communicate some-
thing to others by saying something, a speaker intends the audience to recognize that
intention partly by taking into account that they are so intended. As Grice character-
ized the distinctively self-referential or “reflexive” character of communicative inten-
tions, a speaker means something by his utterance only if he intends his utterance “to
produce some effect in an audience by means of the recognition of this intention” (Grice
1957/1989: 220). Not just any sort of effect will do, and later we will consider just what
sort of “effect” this is. But first we need to appreciate Grice’s basic idea.
To get a feel for it, consider what goes on in the following games, which have some-
thing in common with linguistic communication. Take the game of charades, in which
one player uses gestures and other bodily movements to help the second player identify
what she has in mind. The first player has a self-referential intention, for part of what
she intends is for the second player to take into account the very fact that she intends
her gestures etc. to enable him to figure out what she has in mind. Nothing like this
goes on in the game of 20 questions, where the second player uses answers to yes-or-no
questions to narrow down the possibilities of what the first player has in mind. Here
the only cooperation required is honest answers on the part of the first player. Like
charades, simple games of tacit coordination, first discussed by Schelling (1960: 54–58),
also involve self-referential intentions. The first player selects and records an item in a
certain specified category, such as a letter of the alphabet, a liquid, a mode of transpor-
tation, a city, or a U.S. president; the second player has one chance to guess it. In this
game either both win or both lose. Both win if and only if the second player guesses right
without any help. But what counts as guessing right? That depends entirely on what the
first player has in mind, and that in turn depends entirely on what she thinks the sec-
ond player, taking into account that she wants him to guess right, will think she wants
him to guess. The second player guesses whatever he thinks she wants him to guess. To
appreciate how this cooperative guessing game works, play this game with a friend. Try
additional categories too, and consider why some work better than others.
When players use the above categories, they usually both pick the letter ‘A’, water,
cars, the city in which they are located, and the current president. Each ‘correct’ choice
stands out in some salient way from other members of the category. Grice’s idea was in
effect that successful communication involves something of the same sort. In uttering a
sentence, a speaker has something in mind that she is trying to convey and intends the
hearer to figure out what that is; in hearing the speaker utter the sentence, the hearer
tries to figure out what the speaker intends to convey, partly on the basis of being so
intended. That is, the hearer is to take into account that he is intended to figure out what
the speaker intends to convey. It is the meaning of the words uttered, of course, that pro-
vides the primary input (along with the fact that the speaker uttered them, presumably
with the intention to communicate something), but what they mean does not determine
what the speaker means. Even if what she means is precisely what her words mean, the
fact that she is speaking literally is not determined by what they mean—she could have
meant something else. What is loosely called “context” plays a key role here, not in
determining what the speaker means—that is a matter of the speaker’s communicative
intention—but of enabling the hearer to ascertain (a different sense of “determine”)

82
 M EA N I N G A N D CO M MU N I C AT I O N

what the speaker means. Context comprises whatever other considerations the hearer is
to take into account in so doing. It is information that is mutually available to the speaker
and the hearer, information that the speaker intends the hearer to take into account in
figuring out what the speaker means (see Bach 2005).

Communication and Speech Acts

If Grice was right, there is something distinctively self-referential about a communica-
tive intention. But what is it that the speaker intends? What sort of effect does she intend
to produce on her audience, partly “by means of the recognition of this intention”? We
cannot take up this question until we reckon with the fact that utterances, though gen-
erally communicative, are not made just with communicative intentions—they are not
merely acts of communication. As first investigated by Austin (1962) in his aptly titled
How to Do Things with Words, a speech act is a multi-layered affair. In this respect they
are no different from most intentional actions. Moving one’s arm in a certain way can,
given the right intentions and circumstances, also be a case of pushing away a second
plate of pasta, of sticking to one’s diet, and of trying to impress one’s spouse. Notice that
this is not a series of actions but, rather, a single bodily movement comprising a multi-
plicity of nested actions. The same occurs when one utters a sentence.

Utterances as Three-level Speech Acts

Austin distinguished three distinct levels beyond the mere act of uttering a sentence.
There is the act of saying something, what one does in saying it, and what one does by
saying it. He dubs these, respectively, locutionary, illocutionary, and perlocutionary acts.
Saying isn’t just a matter of meaning what one’s words mean. The locutionary act is, as
Austin defined it (1962: 95), the act of using a sequence of words “as belonging to a cer-
tain vocabulary . . . and as conforming to a certain grammar, . . . with a certain more or
less definite sense and reference” (this qualification is needed to allow for resolving any
ambiguity and fixing any indexical reference). Importantly, Austin did not mean the
act of uttering particular words, reported with direct quotation as for example in (8):

(8) Bernanke says, “Inflation is not a problem.”

The locutionary act is reported, rather, with indirect quotation:

(9) Bernanke says that inflation is not a problem.

Bernanke did not have to use the words, “Inflation is not a problem,” to say that infla-
tion is not a problem. He did not even have to be speaking in English. Regardless of the
means by which he said that, or even if he had uttered something else, say “Inflation
is not on the horizon,” he could have performed the illocutionary act of assuring the
public that prices and interest rates won’t go up significantly, thereby performing the
perlocutionary act of assuaging at least one of their economic fears.
The sentence uttered does not in general determine the type of illocutionary act
being performed. Just as we can do one of several different things in shaking hands—
introduce ourselves, greet each other, seal a deal, or bid farewell—so we can use a given
sentence in various ways. For example, (10) can be used, depending on who is speaking

83
 KEN T BAC H

to whom and with what intention, as a prediction, a warning, a promise, a threat, or

even an order.

(10) The riot squad will break up the demonstration.

It is partly, but only partly, because of what the sentence means that it has its various
(literal) uses. After all, the sentence expresses, at least relative to a given context of
utterance, a proposition about what a certain riot squad will do regarding a certain
demonstration. However, the meaning of the sentence does not determine whether
it is being used to predict, warn, promise, threaten, or order. And it certainly does not
determine the perlocutionary effect which, depending on the illocutionary act being
performed, could be anticipation, dispersal, assurance, fear, or action.

Communicative vs. Perlocutionary Intentions

Now we can spell out the difference, corresponding to the distinction between illocu-
tionary and perlocutionary acts, between a speaker’s communicative intention and her
further intention in uttering a sentence. Intuitively, an act of communication, linguistic
or otherwise, is an act of expressing oneself. This rather vague idea can be made more
precise if we get more specific about what is being expressed. Take the case of an apol-
ogy. If you utter, “[I’m] sorry I smashed your car,” and intend this as an apology, you
are expressing regret, in this case for smashing the hearer’s car. Indeed, it seems that
an apology just is the act of (verbally) expressing regret for something one did that
adversely affected the hearer. It is communicative because it is intended to be taken as
expressing a certain attitude, in this case regret (for smashing the hearer’s car). That is
only the communicative aspect of an apology. When you apologize, you may intend not
just to express regret but also, hoping the hearer thinks you’re sincere, to seek forgive-
ness. Seeking forgiveness is distinct from expressing regret, even though in saying that
you’re sorry for smashing the hearer’s car you are doing both.
In general, speech acts are not merely acts of saying something and not merely com-
municative, illocutionary acts of expressing an attitude. They are also perlocutionary
acts, performed to produce some effect on the audience. However, since the intended
perlocutionary effect of a given type of illocutionary act can vary, it makes sense to
distinguish different types of speech acts primarily by their illocutionary type, such as
asserting, requesting, promising, and apologizing, which in turn may be distinguished by
the type of attitude expressed. The perlocutionary act is generally, though not always,
an attempt to get the hearer to form some correlative attitude, as this table illustrates.

Illocutionary Act Attitude Expressed Intended Hearer Attitude

statement belief that p belief that p
request desire for H to D intention to D
promise firm intention to D belief that S will D
apology regret for D-ing forgiveness of S for D-ing

These acts exemplify the four main categories of communicative illocutionary acts,
which Bach and Harnish (1979: chapter 3), borrowing partly from the terminology
of both Austin (1962) and Searle (1969), call constatives, directives, commissives, and
acknowledgments.

84
 M EA N I N G A N D CO M MU N I C AT I O N

If a type of act can be distinguished by type of expressed attitude, then an act of

that type is communicationally successful if the hearer recognizes the attitude being
expressed, such as a belief in the case of a statement and a desire in the case of a request,
along with its content (what is believed, desired, or whatever). Any further perlocu-
tionary effect the act has on the hearer, such inducing a belief or an action, or even
just being taken as sincere, is not essential to its being a statement or a request and is
not necessary for its success as an act of communication. It need not be sincere—the
speaker might not actually possess the attitude she is expressing—and the hearer might
not take her to be sincere. But there is no question about the speaker being sincere pos-
sessing the communicative intention itself, for this intention must be identified before
the question of her sincerity can even arise. One can be unsuccessful in conveying one’s
communicative intention—by being too vague, ambiguous, or metaphorical, or even by
being wrongly taken literally—but not insincere about it.

Self-Referential Intentions Again

Now we are in a position to fit Grice’s idea of self-referential intentions, the key to his
conception of communication, into the broader framework of speech act theory. This
will enable us to pin down the sort of effect a speaker can intend “to produce in an audi-
ence by means of the recognition of this intention,” and also to see why there is nothing
paradoxical or mysterious about such an intention.

The Intended “Effect”

As Strawson (1964: 459) first pointed out, the relevant effect is understanding, rather
than, as Searle (1969: 47) added, any further, perlocutionary effect on the hearer. And
understanding, or what Austin (1962) called “uptake,” is a matter of identifying what
attitude (and its content) the speaker is expressing. We can think of meaning some-
thing as intending just such an effect: “to express an attitude is reflexively to intend the
hearer to take one’s utterance as reason to think one has that attitude” (Bach and Har-
nish 1979: 15). This formulation respects the difference between expressing an attitude
and actually possessing it, not to mention having a further intention toward the hearer
(Siebel 2003 expresses some worries about this formulation). Whatever else a speaker
may be doing in performing a speech act, her communicative illocutionary act is just the
act of expressing that attitude. Communicating successfully, being understood, is sim-
ply having the expressed attitude recognized. It does not require the hearer to respond
in any further way. The hearer need not even attribute a belief, desire, or some other
attitude to the speaker. Identifying the attitude the speaker is expressing, whether or
not one actually attributes it to her, is just the sort of thing that can be done by way of
recognizing the speaker’s intention for one to do it.

Reflexive Paradox?
Commenting on the notion of an intention “to produce an effect in an audience by
means of the recognition of this intention,” Grice remarked, “This seems to involve a
reflexive paradox, but it does not really do so” (1957/1989: 219). It seems to because the
intention is self-referential. Indeed, the air of paradox may seem to imbue the hearer’s
inference, inasmuch as the hearer is to identify the speaker’s intention partly on the

85
 KEN T BAC H

supposition that he is intended to do so. If meaning is supposed to be what people do

whenever they speak to one another, there had better be nothing paradoxical about it.
It might seem paradoxical if one confuses iterative intentions with reflexive ones, as
indeed Grice himself seems to have done. As Harman explains,

Grice himself originally states his analysis as involving a self-referential inten-

tion . . . but, because of worries about what he calls “self-reflective paradox,” he
goes on to restate the analysis as involving a series of intentions, each about the
preceding one. This turns out to lead to tremendous complexity in the resulting
theory. Much of this complexity is artificial and due to Grice’s refusal to stick
with the original analysis and its appeal to a self-referential intention.
(Harman 1986: 87–88)

As Harman stresses, Grice’s move to iterative intentions led to increasingly complex

formulations beginning with Strawson’s (1964), followed by Grice’s (1969) own, and
culminating with Schiffer’s (1972), each prompted by counterexamples to the previous
formulation. As Harman suggests, sticking with a self-referential intention rather than
invoking iterative ones avoids this complexity.
Resistance to reflexive intentions has been based not just on fear of paradox
but also, it seems, on a misconstrual of what goes into the content of a communica-
tive intention and into the hearer’s inference to it (Bach 1987, in reply to Recanati
1986). Here is the simplest case of this. It might seem that Grice’s formulation, with
its key phrase “by means of the recognition of this intention,” requires that to under-
stand the speaker the hearer must engage in some sort of circular reasoning. It sounds as
though the hearer must already know what the speaker’s communicative intention is in
order to recognize it. However, this misconstrues what the hearer has to take into
account in order to recognize the speaker’s intention. The hearer does not infer that the
speaker means that p (or is expressing, say, the belief that p) from the premise that
the speaker intends to convey that p. Rather, operating on the assumption that the
speaker, or any speaker, in uttering a sentence intends to communicate something or
other (Bach and Harnish (1979: 12) call this the “Communicative Presumption”), the
hearer takes this general fact, not the identity of the specific intention, into account in
identifying that intention. The situation is analogous to that of the player in those coor-
dination games we discussed. In both cases one has to figure out what one is intended
to figure out partly on the basis that one is intended to, but not by knowing in advance
what it is.

Saying One Thing and Meaning Something Else

Grice elaborated on the case in which a speaker says one thing and means something else
instead or something else as well. Although he did not expressly invoke Austin’s notion
of locutionary act, he did rely on a notion of saying whereby what is said is “closely
related to the conventional meaning of the . . . sentence . . . uttered” (1975/1989: 25).
However, it is not identical to conventional meaning because there can be ambiguity
or context-dependent reference. Usually only one conventional (linguistic) meaning is
operative in a given utterance, and linguistic meaning does not determine what such
words as ‘she’, ‘this’, and ‘now’ are used to refer to. But even with all that fixed, the
speaker might not mean just what she says.

86
 M EA N I N G A N D CO M MU N I C AT I O N

Implicating
Grice coined the term implicature for what a speaker means but does not say. Whereas
what is said may imply something, what a speaker implicates is a matter of her intention
in saying it. For example, suppose we’re dining at a greasy spoon and I ask you what you
think of the pasta you’re eating. You reply with (11),

(11) It’s covered with Parmesan.

probably implicating that the pasta is not very good. That’s a matter of your communica-
tive intention. But how do I recognize your intention? Grice’s answer would be roughly
this. Presumably you are giving an informative answer to my question, but saying the
pasta is covered with Parmesan is on the face of it not very informative. So you must
intend me to infer that you mean something more informative. Since I asked you for
your assessment of the pasta and saying that it is covered with Parmesan, good though
that is, does not tell me much about the pasta itself, you are inviting me to read an assess-
ment of it into the fact that you said what you said and no more. I can infer that you have
nothing else good to say about it, hence that you mean that it is not very good. I make
such an inference partly on the basis that you intend me to. Notice that what I infer is
that you mean that it is not very good. To understand you I do not have to infer that it is
not very good or even that you believe that, contrary to Grice (1975/1989: 31).
Uttering something like (11) is not the only way that you could have conveyed with-
out saying that the pasta was not very good. Suppose you had uttered (12), in a sarcastic
tone:

(12) That’s the best pasta I’ve ever tasted.

Here you intend me, taking it to be obvious that you don’t mean that it’s the best pasta
you’ve ever tasted, to infer that you mean that the pasta is not very good. You intend
the fact that you said something relevant but obviously false to be my basis for figuring
out what you mean.
In explaining what goes on in such cases, Grice proposed a general “Cooperative Prin-
ciple” and specific “maxims”—of quality, quantity, relevance, and manner—to account for
how speakers manage to make themselves understood even when they do not spell out
what they mean. Actually, these maxims are better thought of as presumptions, for it is
on the presumption that the speaker is being truthful, relevantly informative, and oth-
erwise appropriate that the hearer figures out what the speaker means. By saying some-
thing obviously false (or unjustified), underinformative (or perhaps overinformative),
irrelevant, or in a strange way (say by being longwinded or pedantic), one calls the hear-
er’s attention to something ostensibly inappropriate about the utterance and, in effect,
invites the hearer to reinterpret it, perhaps by considering what else one could have
said instead, in such a way that it is appropriate after all. Note, however and contrary
to popular opinion (see Bach 2006), that the maxims (or presumptions) do not apply
only to implicature. Even when a speaker is being completely literal, meaning what she
says and nothing else, the hearer reasons in essentially the same sort of way, although
the reasoning is simpler, since he doesn’t have to reinterpret the utterance. Also, the
Cooperative Principle should be not understood to mean that interlocutors are or ought
to be generally cooperative. It specifically concerns the aim of communication, not the
further perlocutionary aims that people have in saying things to one another.

87
 KEN T BAC H

Implicating and Indirect Speech Acts

Implicating is a kind of indirect speech act, a special case of performing one illocution-
ary act by way of performing another. Consider our earlier example (5) and as well as
(13), both uttered by you at a party of yours.

(5) There’s some beer in the fridge.

(13) The gendarmes are coming.

In uttering (5) you could not only tell someone where the beer is but also indirectly offer
them some. With (13) you could not only inform your guests that the gendarmes are
coming but also indirectly warn them to quiet down. Notice that the direct illocution-
ary act need not be a statement. You might directly ask a question with (14),

(14) Do you know it’s after midnight?

to inform someone indirectly that it is after midnight and perhaps also to suggest that it
is time to go home. In these cases you have two communicative intentions (in uttering
(14) perhaps three), one corresponding to each illocutionary act, and you intend your
audience to recognize one by way of recognizing the other.

Between Saying and Implicating

In contrasting saying and implicating, Grice allowed both for cases in which the speaker
means what he says and something else as well (implicating and indirect speech acts
generally) and ones in which the speaker says one thing and means something else
instead (nonliteral utterances, which Grice also counted as cases of implicating). We
also need to allow for the case in which the speaker says something and doesn’t mean
anything (Bach 2001). Now Grice seems to have assumed that saying and implicating
exhaust the cases of speaker meaning. He overlooked the possibility of an intermediate
phenomenon, albeit one that also exploits the maxims. However, as others have since
observed (Sperber and Wilson 1986; Bach 1994; Carston 2002; Recanati 2004), there
are many sentences whose standard uses are not strictly determined by their meanings
(even with ambiguities resolved and references fixed) but are not implicatures or figura-
tive uses either. For example, if your child comes crying to you with a minor cut and you
assure him,

(15) You’re not going to die.

you do not mean that he will never die (that is false but irrelevant) but merely that he
won’t die from that cut. And if someone proposes going out to dinner and you say,

(16) [Sorry, but] I’ve already eaten.

you do not mean that you have eaten at some previous time (that is obviously true but
irrelevant) but more specifically that you’ve had dinner that evening. In both cases you
do not mean precisely what you are saying but something more specific. On the other
hand, what you do mean isn’t an implicature either. You are using each of the words in

88
 M EA N I N G A N D CO M MU N I C AT I O N

(15) and (16) literally, but you are not using the sentences themselves literally, in that
you are leaving part of what you mean implicit.
In other cases, what the speaker says is not merely not as specific as what she means
but is incomplete in the sense of not being fully propositional. If your spouse is honking
the horn at you and you shout back with (17),

(17) I’m not ready.

you mean that you are not ready to leave. And if your spouse yells back,

(18) We’ll be late.

she means that you will be late for the event you are both planning to attend. In nei-
ther case is there anything in the sentence that corresponds to the implicit reference.
Although the speaker means something definite, the sentences themselves, even with
the references of the indexicals ‘I’ and ‘we’ fixed, lack determinate truth conditions.
As with (15) and (16), though for a different reason (propositional incompleteness),
what the speaker means is more specific than what the sentence means. These exam-
ples appear to violate the grammar school dictum that a sentence, unlike a mere word,
phrase, or “sentence fragment,” always expresses a “complete thought” (I say ‘appear’
because some would argue that such sentences contain hidden constituents). We might
say that whereas what a user of (15) or (16) means is an expansion of the sentence mean-
ing, what a user of (17) or (18) means is a completion of the sentence’s content (this is
Bach’s (1994) terminology, which differs from both Sperber and Wilson’s (1986) and
Recanati’s (2004), whose frameworks are somewhat less Gricean).
Now several of Grice’s critics have pointed out that expansions and completions
are not related closely enough to conventional meaning to fall under Grice’s notion of
what is said but are too closely related to count as implicatures. That is because what
the speaker means is built directly from what the sentence means. Recanati (2004) sug-
gests that the notion of what is said should be extended to cover such cases (he goes so
far as to offer a series of progressively more liberal notions of saying), but clearly he is
going beyond Grice’s conception of what is said as corresponding to the constituents of
the sentence and their syntactic arrangement. Sperber and Wilson (1986: 182) coined
the word “explicature” for this in-between category, since part of what is meant expli-
cates what is said. I propose calling these cases of impliciture (Bach 1994), since part of
what is meant is communicated not explicitly but implicitly, by way of expansion or
completion.

Summing Up
We have contrasted speaker meaning with linguistic meaning and examined Grice’s
ingenious conception of speaker meaning, or communication, as involving a distinctive
sort of self-referential intention directed toward an audience. To make this conception
compelling, we needed to distinguish the specifically communicative “effect” of under-
standing from other effects on the audience. This required bringing in broader notions
from speech act theory. With this framework in place, we then considered a variety
of ways speakers can make themselves understood even if when, as is often, what they
mean is not what they say.

89
 KEN T BAC H

Related Topics
1.2 Semantics and Pragmatics
1.5 Implicature
1.6 Pragmatic Enrichment
1.10 Context Sensitivity.

References
Austin, J. L. (1962) How to Do Things with Words, Oxford: Oxford University Press.
Bach, Kent (1987) “On Communicative Intentions: A Reply to Recanati,” Mind & Language 2: 141–154.
—— (1994) “Conversational Impliciture,” Mind & Language 9: 124–162.
—— (2001) “You Don’t Say?” Synthese 128: 15–44.
—— (2005) “Context ex Machina,” in Z. Szabó (ed.), Semantics versus Pragmatics, Oxford: Oxford
University Press.
—— (2006) “The Top 10 Misconceptions about Implicature,” in B. Birner and G. Ward (eds.), Drawing the
Boundaries of Meaning, Amsterdam: John Benjamins.
—— and Robert M. Harnish (1979), Linguistic Communication and Speech Acts, Cambridge, MA: MIT
Press.
Carston, R. (2002) Thoughts and Utterances, Oxford: Blackwell.
Grice, H. P. (1957) “Meaning,” The Philosophical Review 66: 377–388; reprinted as chapter 14 of Grice
1989.
—— (1968) “Utterer’s Meaning, Sentence-meaning, and Word Meaning,” Foundations of Language 4:
225–242; reprinted as chapter 6 of Grice 1989.
—— (1969) “Utterer’s Meaning and Intentions,” The Philosophical Review 78: 147–177; reprinted as chapter
5 of Grice 1989.
—— (1975), “Logic and Conversation,” in P. Cole and J. Morgan (eds.), Syntax and Semantics, Vol. 3, Speech
Acts, New York: Academic Press; reprinted as chapter 2 of Grice 1989.
—— (1989) Studies in the Way of Words, Cambridge, MA: Harvard University Press.
Harman, G. (1986) Change in View, Cambridge, MA: MIT Press.
Recanati, F. (1986) “On Defining Communicative Intentions,” Mind & Language 1: 213–242.
—— (2004) Literal Meaning, Cambridge, UK: Cambridge University Press.
Schelling, T. (1960) The Strategy of Conflict, Cambridge, MA: Harvard University Press.
Schiffer, S. (1972) Meaning, Oxford: Oxford University Press.
Searle, J. (1969) Speech Acts: An Essay in the Philosophy of Language, Cambridge, Cambridge University
Press.
Siebel, M. (2003) “Illocutionary Acts and Attitude Expression,” Linguistics and Philosophy 26: 351–366.
Sperber, D. and D. Wilson (1986) Relevance, Cambridge, MA.: Harvard University Press.
Strawson, P. F. (1964) “Intention and Convention in Speech Acts,” The Philosophical Review 73: 439–460.

90
 1.8
C OMP OS I T IONALIT Y
Josh Dever

Consider a token linguistic utterance, such as:

(1) A screaming comes across the sky.

(1) expresses some proposition—one we could perhaps identify as the proposition that
a screaming comes across the sky but which we’ll for convenience just call Tyrone. In
virtue of what does (1) express Tyrone? Answers will vary depending on one’s metase-
mantic views, but al-most everyone will agree that the underlying stratum of facts deter-
mining the expression relation is large and complicated. The relevant class of facts may
involve:

1. Psychological facts: facts about intricately structured communicative intentions in

me, and perhaps mental states too of you, and even of others not party to this com-
municative exchange.
2. Usage facts: perhaps up to the full history of how these words, and these syntactic
structures, have been used throughout the history of the English language.
3. Contextual facts: when and where (1) was produced and received, what objects
were salient at those times and locations, and what background cultural and worldly
information is common ground for us.
4. Causal and environmental facts: the ways in which we and our linguistic expres-
sions are causally embedded in the world around us (whether, for example, we are
on Earth or Twin Earth).

For you to understand (1), you must know that it expresses Tyrone. But if Tyrone’s iden-
tity depends on facts scattered across space and time, and often exceedingly difficult to
ascertain, how can you have any hope of understanding (1), or any bit of language? The
problem sounds intimidating, but at the same time we all know that the core of the solu-
tion is clear. One understands (1) by understanding the individual words that compose
(1), and knowing how the meanings of individual words determine the meanings of
complex expressions composed of those words.
The thought, then, is that the complexity of the full determination basis for the
expressive features of (1) is screened off at the lexical level. The complex streams of
determining information sort themselves out at the lexical level—from word down to
world, things are an unwieldy mess, but from word up to sentence, things are relatively
simple and epistemically tractable.
 J O SH DEV E R

Why might we expect natural languages to have this screening structure? General
considerations of the supervenience of the features of wholes on the features of their
parts do not suffice here: we are asking for determination not by all properties of the
parts but only by the specifically semantic properties of the parts; and we are asking for
determination of the meanings of the complex expressions, and meanings are extrinsic
features of expressions, and extrinsic features are typically not determined by features of
parts. One reason to expect the screening structure would derive from epistemic/learn-
ability considerations. Competent language users need to know what propositions are
expressed by particular utterances. But language users cannot plausibly be expected to
know the full determination basis for the expression relation, or to be able to calculate
the proposition expressed even if they did know the full determination basis. Given the
screening structure, however, it becomes unnecessary to know the full determination
basis—all that is needed is to know the meanings of the individual lexical items, and the
rules for determining meanings of complex expressions from the meanings of their parts.
Of course, these epistemic/learnability constraints neither mandate nor are automatically
satisfied by the screening structure. The complexities of the full determination basis
could be screened elsewhere than at the lexical level (for example, at the level of some
tractable characterization of utterance context), and the epistemic/learnability con-
siderations would be equally answered. And screening could occur at the lexical level
without answering those considerations—if, for example, the computation of meanings
of complex expressions from the meanings of their component parts were enormously
complex.
We might also expect the screening structure because we think that the complexities
of the determination basis were already screened at the level of (for example) objects,
properties, and relations, and then think that the lexical meanings simply reflect that
prior screening. Perhaps the complexities of usage and individual psychology are all
manifestations of the various complicated ways in which we, individually and collec-
tively, relate to distal objects, but those complexities play no role in differentially deter-
mining expression of propositions, because the propositions are already determined
by the distal objects. If the proposition expression relation is suitably closely tied to
information flow from the world, and thus suitably independent of the vagaries of our
psychologies, again the screening structure might be predicted.

1 What is Compositionality?
There are many forms a screening structure might take, but the one that has been of cen-
tral interest in the philosophy of language is the property of compositionality. A meaning
theory for a language is compositional just in case for any complex expression in the lan-
guage, the meaning of that complex expression is fully determined by the meanings of
its component expressions plus the syntactic mode of organization of those component
expressions. Compositionality is thus a property that constrains how three things come
together: the meanings of complex expressions, the meanings of components of com-
plex expressions, and the syntactic mode of organization of the component expressions
in the complex expression. Equivalently, compositionality is a property that imposes
two constraints on the determination basis for the meanings of complex expressions.
The factors that contribute to the meaning of a complex expression must be syntactic
and semantic. Thus the meaning of a complex expression cannot be determined by the
phonology or morphology of the words that make it up, or by the context in which it

92
 CO M PO SI TI ON A L I T Y

is uttered except, in both cases, insofar as these factors themselves contribute to the
meanings of the component expressions. (For more careful discussion of the interaction
between compositionality and context sensitivity, see Pagin and Pelletier 2007, Reca-
nati 2011, and Westerstahl 2011.) And the factors that contribute to the meaning of a
complex expression must be local. The meanings of the words that compose the complex
expression can contribute to the meaning of the complex; the meanings of other words
(perhaps those in larger expressions of which the complex expression is itself a compo-
nent part) cannot so contribute.
The compositionality constraint can be given equivalent functional and substitutional
formulations. Functionally, compositionality is a property had by a meaning theory for
a language just in case, for any complex expression in the language, its meaning is some
function of the meanings of its component parts and their mode of syntactic organiza-
tion. What does it mean to treat the mode of syntactic organization as an argument to
a function, as required by this formulation of compositionality? We can think about
this in one of two ways, depending on how we think of the constituents of a complex
expression. Consider a complex expression such as “the hungry tiger,” and assume that
its syntactic structure is [DP [D the] [NP[A hungry] [N tiger]]]. The immediate constituents
of the expression are then the determiner “the” and the noun phrase “hungry tiger,”
while the ultimate constituents are “the,” the adjective “hungry,” and the noun “tiger.” If
we impose the constraint called strong compositionality (Larson and Segal 1995: 78–79),
we require that the meaning of a complex expression be a function of the meanings of its
immediate constituents and their mode of syntactic organization. Since the constituents
are immediate, the syntactic organization can be specified by identifying the particular
syntactic rule that allows those immediate constituents to be joined into the complex
expression. In the case of “the hungry tiger,” the rule is perhaps something like:

• DP o D + NP

The arguments for the meaning composition function are then the semantic values of
the immediate constituents, along with the syntactic rule by which they are combined.
Strong compositionality thus encourages a picture on which each syntactic rule is asso-
ciated with a semantic rule—the rule that determines how constituent meanings are to
be combined when the constituents are syntactically combined by the given syntactic
rule. This is typically called the rule-to-rule conception of compositionality, and it is
most paradigmatically associated with the semantic programme of Richard Montague
(Montague 1970, 1973).
If we allow the meaning of a complex expression to be determined by the meanings of
its ultimate, rather than its immediate, constituents, we impose the constraint of weak
compositionality. In this case, the syntactic structure cannot be treated as a single rule,
and instead needs to be taken as a partial ordering of rules, or as a sequence of stages
of syntactic derivation. If a meaning theory is strongly compositional, it is also, as the
names would suggest, weakly compositional. Given strong compositionality, we can
from the meanings of the ultimate constituents work out the meanings of the slightly-
less-ultimate constituents, and then the meanings of the even-less-ultimate constitu-
ents, and on until we determine the meanings of the immediate constituents, and then
the meaning of the complex. Thus the meanings of the ultimate constituents do, given
strong compositionality, determine the meaning of the whole. But the reverse is not
true—weak compositionality does not entail strong compositionality. If, for example,

93
 J O SH DEV E R

the meaning of a sentence is its truth value, then strong compositionality requires that
all sentential operators be truth-functional, while weak compositionality does not.
The substitutional formulation of compositionality is often easier to work with than
the functional formulation. If the meaning of a complex expression is a function of
the meanings of its component expressions (plus syntax), then it can be shown that
expressions (any constituents, if we begin with strong compositionality; only ultimate
constituents, if we begin with weak compositionality) with the same meaning can be
intersubstituted in a complex expression without altering the meaning of the whole. Let
. be some meaning theory. If the doctor’s nephew is a function of the, doctor, ’s,
and nephew, then there is some function f such that the doctor’s nephew = f (the,
doctor, ’s, nephew). But if doctor = physician, then f (the, doctor, ’s,
nephew) = f (the, physician, ’s, nephew). So the doctor’s nephew = the phy-
sician’s nephew. Functional compositionality thus entails substitutional composition-
ality; the reverse entailment holds as well. (See Hodges 2001 for more detailed proofs.)
Determining what function combines meanings of constituents to yield meanings
of complexes can be a challenging task, one that requires production of a substantial
component of a semantic theory. But checking for intersubstitutability of synonyms
requires no difficult semantic theorizing, making the substitutional test typically easier
to deploy. The substitutional test also makes it possible sensibly to raise the question of
compositionality for semantic theories that do not assign reified semantic values but do
have a notion of sameness of meaning (such as, perhaps, Davidsonian truth theories,
which can be taken to assign to two expressions the same meaning if the two expressions
are governed by axioms the same modulo replacement of mention of the one expression
by mention of the other). It is an immediate consequence of the substitutional test that
a language with no synonyms is trivially compositional, since there are no substitution
instances that could possibly fail to preserve meaning.

2 A First Example
At a first glance, compositionality can easily seem a trivial constraint. How could the
meaning of a complex expression not be determined by the meanings of its component
parts? But even a cursory look at the linguistic data shows that natural languages, even
if they are in the end compositional, certainly display features that resist a compositional
treatment. Consider so-called Frege cases, such as:

(2) Lex Luthor believes Superman is vulnerable to kryptonite.

(3) Lex Luthor believes Clark Kent is vulnerable to kryptonite.

Initial appearances suggest the following:

1. (2) and (3) differ in truth value, and hence also differ in meaning.
2. “Superman” and “Clark Kent” are co-referential, and hence have the same
meaning.
3. (2) and (3) have the same syntactic form, and differ only in the substitution of
“Superman” in (2) for “Clark Kent” to form (3).

But these initial appearances then contradict the principle of compositionality. The substi-
tution of synonymous expressions, while holding the syntax fixed, has altered meaning.

94
 CO M PO SI TI ON A L I T Y

Is this a proof that English and other natural languages are not compositional? No—
there are multiple ways to respond to the prima facie noncompositionality. The responses
fall into four categories: three corresponding to established philosophical positions on
Frege cases, and one not.

Fregeanism
Fregean semantic theories reject the claim that “Superman” and “Clark Kent” have
the same meaning, by rejecting the move from coreferentiality to synonymy. Classi-
cal Fregean approaches run a two-tracked semantics, in which expressions are assigned
both a semantic value of reference and a semantic value of sense. When a meaning
theory assigns multiple types of semantic values, the question of compositionality can
be raised for each type. Standard Fregean approaches strive for compositionality both
in reference and in sense.
However, even given the Fregean doctrine that sense determines reference, compo-
sitionality at the two levels is independent. A Fregean meaning theory can be composi-
tional at the level of sense but not compositional at the level of reference—the substitu-
tion of one expression for another coreferential but not cosensual expression need not
preserve reference, if all we know is that substitution of cosensual expressions preserves
sense. And a Fregean meaning theory can be compositional at the level of reference,
but not compositional at the level of sense. If one expression is replaced with another
cosensual expression, reference is preserved, since the two expressions will also be coref-
erential and we are assuming compositionality at the level of reference. But sense need
not, in general, be preserved.

Direct Reference
We can reject the claim that (2) and (3) differ in meaning. In doing so, we must either
(a) deny also that they differ in truth value, or (b) reject the tacit additional assump-
tion that meaning determines truth value. If we are considering the sentences as types,
option (b) is plausible, and may be developed along the lines of the Hidden Constituent
response considered below. If we are considering the sentences as tokens, option (b) is
more challenging, and the direct reference theorist is more likely to develop the posi-
tion along the lines of (a), as in Salmon (1986). But in either case, if (2) and (3) differ
in meaning, the challenge to compositionality is resolved.
The direct referentialist may attempt to soften the blow of claiming that (2) and (3)
have the same meaning by also claiming that the two sentences give rise (perhaps, in
many/typical contexts) to different conversational implicatures. When conversational
implicatures are invoked, another level of meaning has been introduced, and the ques-
tion of compositionality can be raised for it. There are additional complications in this
case, since the meaning level of implicature does not seem to involve assigning any
meanings to subsentential components (utterances implicate, but the component words
making up the utterances do not individually implicate). However, Grice’s implicature
test of nondetachability (Grice 1975: 39) bears certain relations to a compositionality
requirement—it demands that the implicatures of a sentence be fully determined by the
standard semantic values of the component expressions of the sentence, plus context.
The direct referentialist of this stripe is thus committed to a rejection of the quasi-com-
positional feature of nondetachability.

95
 J O SH DEV E R

Hidden Constituents
We can deny that (2) and (3) differ only in the substitution of “Superman” in (2) for
“Clark Kent” in (3). The overt form of the two sentences makes it appear that this is
the only difference, but we could hold that there are covert components of the two sen-
tences. We might, for example, think that belief is a trinary relation between an agent,
a proposition, and a mode of presentation, and then hold that while the complement
clause to the verb “believes” in (2) and (3) in each case provides the propositional rela-
tum there is an additional hidden constituent of the sentence that provides the mode
of presentation, and that the semantic contribution of that hidden constituent differs
between (2) and (3). We might further think that the hidden constituent is indexical,
so that the difference in meaning between (2) and (3) emerges only at the level of token
contextualized utterances.
For a hidden constituent approach to resolve the challenge to compositionality, the
hidden semantic values must be tied to some (presumably covert) syntactic components
of the two sentences. The approach of Crimmins and Perry (1989), on which the hid-
den constituent is unarticulated, and not the semantic value of any syntactic constitu-
ent of the sentence, leaves the semantics noncompositional. The lesson generalizes: In
order for compositionality to be possible, contextual impact on meaning of complex
expressions must be mediated by contextual impact on meaning of individual words.
Since a single word has no component parts, its meaning can vary from context to con-
text without threatening compositionality (compositionality requires only that, trivi-
ally, the meaning of each individual word be a function of its own meaning). But if there
are contextual impacts on complex expressions not mediated by contexutal impacts on
component words, then two tokens of the same complex, deployed in different con-
texts, can have different meanings, despite both consisting of parts with the same mean-
ings. Context-sensitivity is thus a special case of the general screening-off phenomenon
discussed above: Meanings can and do depend on many disparate facts, but we ideally
expect those disparate facts to be screened off at the level of lexical meaning.

Devious Syntax
The argument for non-compositionality is defeated if (2) and (3) have different syn-
tactic forms. However, there does not seem to be any good reason in this example to
expect a difference in syntactic form. Some related compositionality issues, though, are
subject to plausible “devious syntax” approaches. For example, it is plausible that “that
dingos will eat my baby” and “the proposition that dingos will eat my baby” have the
same meaning: Both refer (in context) to the proposition that dingos will eat my baby.
However, the pair of sentences:

(4) I fear that dingos will eat my baby.

(5) I fear the proposition that dingos will eat my baby.

can clearly differ in truth value—one attributes a perhaps rational worry about wildlife,
while the other exhibits a strange phobia of abstracta. (See King 2002 for more detailed
development of this problem.) Is there a problem here for compositionality? Only on
the assumption that (4) and (5) have the same syntactic structure. But there is clearly
reason to think that they do not—in (4), the verb phrase is the result of combining

96
 CO M PO SI TI ON A L I T Y

a verb with a complementizer phrase, while in (5), the verb phrase is the result of
combining a verb with a noun phrase. Given the rule-to-rule picture sketched
above, we can have different semantic rules associated with these two syntactic com-
bination rules, and compositionally account for the difference in meaning between (4)
and (5).

3 The Strength of Compositionality

An important lesson can be extracted from the above discussion: although there are
many phenomena in natural language that at first blush display non-compositional fea-
tures, there are also a variety of methods for modifying a theory of language to allow
compositional treatments of those phenomena. Because compositionality requires coor-
dination among the three features of meanings of complexes, meanings of components,
and syntactic structure of complexes, these methods can be sorted into three categories.
Faced with an apparent violation of compositionality, we can revise our views on the
meanings of the complex expressions. This is perhaps the most difficult of the three
approaches to pursue, since we typically have more direct access to the meanings of
complex expressions—especially whole sentences—than we do to the meanings of indi-
vidual words. This approach thus faces the challenge of making a convincing case that
appearances about sentence meaning are, in the relevant instances, misleading. How-
ever, given enough flexibility in sentence meanings, any phenomenon can be given a
compositional analysis. Some views on sentence meaning will automatically provide
the requisite flexibility. For example, Horwich (1998: 155) holds a view of sentence
meaning such that “once one has worked out how a certain sentence is constructed from
primitive elements, and provided one knows the meanings of those elements, then,
automatically and without further ado, one qualifies as understanding the sentence.”
On such a view, compositionality would appear to be automatic.
Apparent violations of compositionality can also be approached by adjusting the
assignment of meanings to individual words. Words can be given meaning that are, in
effect, deferred instructions for how to combine with a variety of possible embedding
contexts. Suppose we want “the temperature” to behave differently when it combines
on the one hand with “is 42 degrees” and on the other hand with “is rising rapidly”
(picking out, roughly, a number in the first case and a function from times to numbers
in the second case). Compositionality seems to forbid this, since the meaning of “the
temperature” needs to be determined just by the meanings of its component parts, and
not by facts about where (the particular token of) it is embedded. But if the meaning
of “the temperature” is a function that maps the meaning of “is 42 degrees” to the true
just in case the current heat level is some specific amount, and maps the meaning of “is
rising rapidly” to the true just in case the function from times to heat levels has a cer-
tain local derivative, then “the temperature” will, in a compositional way, display the
desired behavior when combined with either predicate.
The general tactic deployed here is that of type lifting, and in limited forms it plays
a common role in semantic theories. For example, since Montague 1973, it has been
common to take the meaning of a proper name to be, rather than an object, the second-
order property of being a property possessed by that object, in order to make semantic
combination with quantified noun phrases proceed more smoothly. When pursued vig-
orously enough, type lifting can in effect allow any given expression’s semantic interpre-
tation to be arbitrarily delayed, allowing its meaning, in effect, to depend on anything

97
 J O SH DEV E R

in the sentence in which it occurs. In fact, if a sentence contains a single word whose
meaning we can set in whatever way we please, we can then always construct a com-
positional theory that assigns to both the whole sentence and to the other component
words of the sentence the pretheoretically desirable meanings. If compositionality is to
be a nontrivial constraint, then, there must be substantive constraints on what mean-
ings can be assigned to words.
Finally, apparent violations of compositionality can be approached by adjusting the
syntactic form of the sentence. If we are sufficiently liberal with the syntactic theory,
then any combination of sentence meanings and word meanings can be fitted together
in a compositional theory (see Janssen 1997 for details). Liberality of this level may
result in a syntactic structure for the sentence that bears no resemblance to what stand-
ard syntactic theory offers us (compositionality can easily be achieved, for example,
if we endorse a syntactic theory according to which none of the words that seem to
appear in the sentence are in fact syntactic constituents of it). But there are more stand-
ard syntactic techniques that are often appealed to in the quest for a compositional
meaning theory. We may posit hidden syntactic constituents of the sentence, such as
traces. These hidden constituents can then be carriers for semantic values needed to
achieve compositionality. Or we may posit movement, and run the semantic analysis on
a syntactic level at which certain constituents are displaced from their surface position,
allowing them to interact with different parts of the sentence.
Compositionality, then, is by itself a constraint only as strong as one’s commitment
to particular triads of meanings of complexes, meanings of constituents, and syntactic
structures. If we have firm commitments on all three of these matters, we may find
ourselves unable to give a compositional meaning theory for the language. If we are
sufficiently flexible on any of the three, compositionality can always be achieved. It is
thus tempting to think that compositionality simpliciter is not the theoretically central
concept but rather some strengthening of the core principle that combines compo-
sitionality with additional constraints. Here are three representative examples: First,
the project of direct compositionality, for example, seeks to combine compositionality
with a minimization of deviousness in the syntax—ideally, developing a version of the
rule-to-rule approach, in which semantic analysis is run on surface syntactic forms, and
each syntactic rule used in the construction of those surface forms is coupled with some
specific method of combining component semantic values. (See Jacobson 2002, and
numerous papers in Barker and Jacobson 2007.) Second, the project of systematic seman-
tics requires that meanings of complex expressions be not just functions of meanings of
their parts, but particularly tractable functions (see Zadrozny 1995). So, for example, a
systematic semantics might require that meanings always combine by functional appli-
cation, or by concatenation. Systematic semantics often seeks to explain the productiv-
ity of our linguistic understanding—why is it that if someone understands the expres-
sion “red apple” and the expression “large table”, they also understand “large apple”
and “red table.” Compositionality alone is not sufficient to explain these facts about
understanding—there could be a function that maps red and apple to red apple,
and also maps red and table to red table but which maps the latter two arguments
via a vastly more complicated computational algorithm than the former two arguments,
so that understanding of how the first two are mapped to their output fails to extend to
understanding of how the latter two are mapped to theirs. But if we insist that the com-
positional function be sufficiently tractable, then understanding of the one case will
extend reliably to understanding of the other. Third, we could pursue a semantic theory

98
 CO M PO SI TI ON A L I T Y

in which the mechanism of type shifting is minimized, or even outlawed, so as to avoid

the sort of easy compositionality via devious lexical meanings described above.

4 A Second Example
We close by working out in some detail a specific example of how the pursuit of compo-
sitionality can shape a meaning theory. Consider the problem of the meaning of quanti-
fied noun phrases, of the form ‘DET NP’, such as “the logician” or “most philosophers.”
Suppose we have already decided to work in a semantic framework that takes on board
four commitments:

1. The meaning of a sentence is its truth value.

2. The meaning of a proper name is the object to which it refers.
3. The syntax of the language is binary branching, so that each complex expression
has two immediate constituents.
4. The meanings of complex expressions are always determined by functional applica-
tion: The meaning of one immediate constituent takes the meaning of the other
immediate constituent as argument, and produces the meaning of the complex
expression as output.

Consider a simple sentence, such as:

(6) Socrates laughs.

Socrates laughs is a truth value, and it must be determined either by functional appli-
cation of Socrates to laughs, or by functional application of laughs to Socrates.
But Socrates is an object, not a function, so the first option is ruled out. Thus laughs
must be a function from objects to truth values. For convenience, we can talk of the
semantic type of any expression.
Using t to pick out the set of truth values, and e to pick out the set of objects, we say
that sentences are of type t (the meaning of any sentence is an element of t) and proper
names are of type e. Given any types a and b, we can define a derived type (a, b), which
is the set of all functions from elements of type a to elements of type b. “Laughs,” in this
notation, is of type (e, t).
Given this framework, what is the semantic type of “some logician”? Given the
sentence:

(7) Some logician laughs.

we know that either laughs (some logician)is of type t, or some logician (laughs)
is of type t. Given that “laughs” is of type (e, t), the former requires that “some logician”
be of type e. But if “some logician” is of type e, it must refer to an object, and there seems
to be no plausible way to pick a specific object for it to refer to. We will thus set aside
this possibility. If some logician takes laughs as argument to produce a truth value,
then it must be of type ((e, t), t)—roughly, a property of properties.
However, we encounter a difficulty when we have quantified noun phrases in object
position. Start with a name-containing sentence with a transitive verb:

99
 J O SH DEV E R

(8) Socrates sees Plato.

“Sees Plato” must be of type (e, t), and hence “sees” must be of type (e, (e, t). But now
replace “Plato” with “some logician”:

(9) Socrates sees some logician.

If “sees” is of type (e, (e, t)), then it can take arguments of type e only. If “some logician”
is of type ((e, t), t), then it can take arguments of type (e, t) only. Thus neither “sees” nor
“some logician” can take the other as argument, and there is no way to assign semantic
value to “sees some logician” in a way consistent with our starting assumptions.
Our semantic framework thus makes it impossible to place quantified noun phrases
both in subject and in object position. The heart of the problem is this: we have seman-
tic reason to place names and quantified noun phrases in different semantic categories.
But the two types of expression have the same syntactic distribution, so, given the
commitment to compositionality and, in particular, the use of functional application as
the method of semantic composition, they must co-occur in a given construction type
by having one serve as argument in that construction type, and the other as argument-
taker. But that particular trick can be used only in one construction type—we exhaust
the trick when we find a typing for quantified noun phrases in subject position, and then
it is no longer available when we reencounter the same difficulty in subject position.
There are various options for responding to this difficulty. We could, for example,
reject the assumption that proper names are of type e, and unify names and quantified
noun phrases in type ((e, t), t). Then intransitive verbs could be of type (((e, t), t), t),
and transitive verbs of type (((e, t), t), (((e, t), t), t)). Here we will instead hold on our
starting four commitments, and pursue instead a syntactic route. The trouble could be
eliminated if we could deny that quantified noun phrases occur in both subject and
object positions. One path to such a denial is to assume that our syntax includes a rule of
quantifier raising, which takes quantified noun phrases and moves them to the position
of adjoin to the sentence. Thus the underlying logical form of “Some logician laughs”
would be [S [DP some logician ]t [S [NP t ][VP laughs ]]] (where t is a trace left by movement,
and is of type e). Similarly, the underlying logical form of “Socrates sees some logician”
would be [S [DP some logician ]t [S [NP Socrates ] [VP [TV sees ][NP t]]]]. Given these logical
forms, quantified noun phrases that appear, at surface level, in either subject or object
position nevertheless occupy the same position (adjoin to S) at the level of logical form.
Compositionality can then be respected by assigning the same semantic type to both
subject and object quantified noun phrases.
But which semantic type would that be? Given our commitment to sentences being
of type t, and given the modified syntactic position of the quantified noun phrases as
joining with a sentence to form a sentence, we have no choice but to assign them type
(t, t). (Even if we weakened our commitment to sentences being of type t, we would
face the difficulty of typing the trace so that in both subject and object position it could
produce the desired typing for the whole sentence.) Type (t, t), however, is the set of
functions from truth values to truth values, and clearly we will not find in this type the
semantic values we need to assign plausible values to quantified noun phrases.
An empirically adequate and compositional semantics for quantified noun phrases
can be obtained from this impasse by altering our conception of the type t. Following
Tarski 1936, we can take the fundamental semantic values for sentences to be not the

100
 CO M PO SI TI ON A L I T Y

monadic features of truth and falsity but rather the relational features of satisfaction by
sequences. Letting t now be the set of values satisfied by sequence V1, satisfied by sequence
V2, and so on, we discover that since there are many sequences, t becomes a much richer
semantic type. That means that the type (t, t) contains many more functions than
before, and indeed is now a domain in which we can find plausible semantic values for
quantified noun phrases.
The particular route to compositionality set out here is only one option among
many (see Heim and Kratzer 1998: 178–208) for much more detailed discussion), but it
illustrates a general point. Careful attention to the demands of compositionality often
allows the theorist of language to uncover tacit commitments to unexpected forms of
semantic value. In the case of quantified noun phrases, a priority of satisfaction to truth
is revealed; the Fregean will see a discovery of a commitment to senses as emerging in
the same manner; the dynamic semanticist may take the compositional treatment of
cross-sentential anaphora given in Groenendijk and Stokhof 1991 as another case in
point in which the necessity of context-change semantic values is revealed. By requir-
ing coordination between meanings of complexes and meanings of their component
parts, compositionality provides a guard against unpaid semantic bills—commitments
regarding the meanings of sentences must eventually be traced down to their conse-
quences for the meanings of individual words.

References
Barker, C. and P. Jacobson (2007) Direct Compositionality, Oxford: Oxford University Press.
Crimmins, M. and J. Perry. (1989) “The Prince and the Phone Booth: Reporting Puzzling Beliefs”, Journal
of Philosophy 86: 685–711.
Grice, H.P. (1975) “Logic and Conversation”, in P. Cole and J. Morgan (ed.), Syntax and Semantics, 3:
Speech Acts, New York: Academic Press. Reprinted in H. P. Grice (eds.), Studies in the Way of Words,
Cambridge: Harvard University Press.
Groenendijk, J. and M. Stokhof. (1991) “Dynamic Predicate Logic”, Linguistics and Philosophy 14: 39–100.
Heim, I. and A. Kratzer (1998) Semantics in Generative Grammar, Oxford: Wiley-Blackwell.
Hodges, W. (2001) “Formal Features of Compositionality”, Journal of Logic, Language, and Information
10: 7–28.
Horwich, P. (1998) Meaning, Oxford: Oxford University Press.
Jacobson, P. (2002) “The (Dis)Organization of the Grammar: 25 Years”, Linguistics and Philosophy,
25: 601–26.
Janssen, T. (1997) “Compositionality”, in J. van Benthem and A ter Meulen (eds.), Handbook of Logic and
Language, Cambridge: The MIT Press.
King, J. (2002) “Designating Propositions”, Philosophical Review, 111: 341–71.
Larson, R. and G. Segal. (1995) Knowledge of Meaning: An Introduction to Semantic Theory. Cambridge:, MA
The MIT Press.
Montague, R. (1970) “Universal Grammar”, Theoria 36: 373–98.
Montague, R (1973) “The Proper Treatment of Quantification in Ordinary English”, in K.J.J. Hintikka,
J.M.E. Moravcsik, and P. Suppes (eds.), Approaches to Natural Language, Dordrecht: Reidel.
Pagin, P. and J. Pelletier (2007) “Content, Context, and Composition”, in G. Peter and G. Preyer (eds.),
Content and Context: Essays on Semantics and Pragmatics, Oxford: Oxford University Press.
Recanati, F. (2011) “Compositionality, Flexibility, and Context-Dependence”, in W. Hinzen, E. Machery,
and M. Werning (eds.), The Oxford Handbook of Compositionality, Oxford: Oxford University Press.
Salmon, N. (1986) Frege’s Puzzle, Cambridge, MA: The MIT Press.
Tarski, A. (1936). “The Concept of Truth in Formalized Languages”, in A. Tarski, Logic, Semantics, Meta-
mathematics, Indianapolis: Hackett.
Westerstahl, D. (2011) “Compositionality in Kaplan-Style Semantics”, in W. Hinzen, E. Machery, and M.
Werning (eds.), The Oxford Handbook of Compositionality, Oxford: Oxford University Press.
Zadrozny, W. (1995) “From Compositional to Systematic Semantics”, Linguistics and Philosophy 17: 329–42.

101
 J O SH DEV E R

Further Reading
Dever, J. (2006) “Compositionality”, in E. Lepore and B. Smith (eds.), The Oxford Handbook of Philosophy of
Language, Oxford: Oxford University Press.
Pagin, P. and D. Westerstahl. (2010) “Comnpositionality I: Definitions and Variants” and “Compositional-
ity II: Arguments and Problems”, Philosophy Compass 5: 250–264 and 265–82.
Szabo, Z. (2007) “Compositionality”, in Stanford Encyclopedia of Philosophy. (Three recent overviews of com-
positionality that deal more carefully with a number of complexities that arise in the formulation and
application of a principle of compositionality.)

102
 1.9
FOC US AND
I NTONA T ION
Daniel Büring

1 Introduction and Terminology

Intonation describes the way the fundamental frequency of the voice, also called its
pitch or F°, changes over the course of an utterance. A slightly broader term is Pros-
ody, which covers not just intonation but also additional aspects of phonetic realization
such as pauses, lengthening of segments, perhaps loudness, and spectral tilt; intonation
in particular, and perhaps prosody in general, roughly corresponds to the colloquial
term ‘inflection’. Certain aspects of prosody (and intonation) are grammatical in nature
and as such represented in a phonological representation, called Prosodic Structure.
At a minimum, prosodic structure will encode prosodic constituent structure, relative
metrical strength or stress of syllables, and location and nature of certain tonal (or ‘into-
national’) events (see Ladd, 1996).
Focus—at least in connection with languages such as English—is the term used to
describe effects of prosody on meaning and vice versa. The basic idea is that prosodic
prominence in an unusual place serves to emphasize the meaning of the element(s)
thusly marked. For example Kim is assumed to be focused in (1b) but not (1a) or (1c)
and Harry is focused in (1c), but not (1a/b) (we will return to the question of what if
anything is focused in (1a) in section 4.1 below):

(1) a. Kim will read Harry’s book.

b. KIM will read Harry’s book.
c. Kim will read HARry’s book.

To come to terms with the relation between focus and intonation, the following
simplified picture may be instructive: each sentence S has a ‘neutral intonation’ (part
of its ‘neutral prosody’), its citation form, the way it is uttered out-of-the-blue. Neutral
prosody is modeled by the prosodic structure grammar assigns to S (where S is a syntactic
tree, not just a string of words) on purely structural grounds. Where the prosody of a sen-
tence S does not correspond to that neutral intonation, this is the reflex of additional,
meaning-related marking, the sentence’s focusing, or more generally, its Information
Structure (where information structure is simply a cover term for whatever semantic
or pragmatic categories such as focus, background, givenness, topic etc. one thinks may
influence prosody in this way).
 DA N I EL BÜ R I N G

Roughly, neutral intonation is characterized by local frequency maxima (or some-

times: minima), so-called pitch accents (PAs), on (virtually) all content words.
Marked prosody most typically omits PAs on content words at the end of the sentence
(e.g. book in (1c)).
This in turn yields the prosodic impression that the last pitch accent (which is not on
the final content word, as in the neutral prosody) is the strongest, most prominent (e.g.
Harry in (1c)). The semantic or pragmatic intuition is that omitting final PAs yields the
effect of emphasizing the last constituent that does bear a pitch accent.
It is common (though perhaps bad) practice to only indicate the last PA in ortho-
graphic renderings, as in (1) above, and not indicate any PAs in the neutral pronuncia-
tion. An alternative more faithful to the actual intonation is as in (2a) and (b), cor-
responding to (1a) and (c) respectively:

(2) a. kim will read harry’s BOOK.

b. kim will read harry’s book.

In (2), the final PA is marked by larger capitals, reflecting speakers’ judgments (men-
tioned above) that the last PA—even in neutral prosody, (2a)—is more prominent
than the previous ones (interestingly, phonetically, the last PA is not always more elab-
orate—higher, longer or louder—than the others; its main distinctive feature being,
it seems, its relative position in the sentence). The last PA is also referred to as the
Nuclear Pitch Accent (NPA). Marked intonation can thus also be seen as a (left-
wards) shifting of the NPA; its interpretation, on this view, is emphasis on the element
bearing the shifted NPA.
Since Jackendoff (1972), a privative feature [F] is used to mark focused constituents
in the syntactic representation. [F] yields the interpretation of emphasis (or something
like it, see section 3.2 below), and a prosodic realization which—in English at least—
has the NPA within the [F]-marked constituent.
Perhaps [F] is just a convenient way to talk about the abstract concept focus, as
opposed to the prosodic features that realize it, and to structurally disambiguate sen-
tences with the same prosodic structure but—by pragmatic criteria—different foci. If so,
ultimately grammar could interpret prosody directly, without the mediation of [F]. The
details of such an elimination are very unclear at present though, so most works assume
that [F] is a bona fide element of grammatical—specifically: syntactic—representation.
Summarizing our terminological usage: When people hear utterances, they perceive
prominence on certain elements; this perception is based on prosodic properties, most
significantly the placement of pitch accents. On the meaning side, people perceive
emphasis, the grammatical counterpart to which is interpretation as focus (more on
which below), or, more generally, in terms of information structure. The syntactic fea-
ture [F] mediates between emphasis and prominence, via effects on interpretation and
prosodic representation.
We haven’t used the term Stress in the discussion of focus so far. Much of the
nonphonological literature uses ‘stress’ and ‘accent’ interchangeably, usually to describe
PAing (or its perceptive counterpart, prominence). In phonology, stress more gener-
ally describes a strength relation between syllables which can be realized phonetically
in a variety of ways, including accented (more stressed)—not accented (less stressed),
nuclear accent (more stressed)–pre-nuclear accent (less stressed), longer (stressed)–
shorter (less stressed), full vowel (more stressed)–reduced vowel (less stressed), or

104
 F O CU S A N D I N T O N AT I O N

perhaps sometimes not at all. Crucially, stress is a relational, and hence graded notion
(though people occasionally talk about ‘word stress’, ‘phrasal stress’, ‘sentential stress’
as a shorthand for ‘strongest stress within the word/phrase/sentence’), while accenting
is absolute (a syllable is accented or not), and accordingly, a syllable may be stressed to
a certain degree and still be accentless. It is plausible to think that, ultimately, focus
is realized by stress, which in turn results in pitch accenting by prosody-internal prin-
ciples. For the purpose of this overview, however, we will stick to the more directly
observable correlate, accenting.

2 Semantic Meaning of Focus

2.1 Associated Focus

Focus may influence truth-conditional meaning, but it doesn’t itself have truth-condi-
tional meaning. If Kim brought Harry’s book and comb, (3a) is false, but (3b) may be
true; if Kim brought Harry’s and Jo’s book, (3b) is false, but (3a) may be true:

(3) a. Kim only brought Harry’s BOOK to the meeting.

b. Kim only brought HARry’s book to the meeting.

Leaving out only in both sentences makes them truth-conditionally equivalent (though
of course they still convey different emphases). That is, focus sensitive elements
(FSEs) such as only change truth conditions depending on the focus they associate
with; focus alone doesn’t.
The contribution of focus to the meaning of FSEs such as only can be modeled using
alternative sets, also called focus semantic values (Rooth, 1985; vonStechav,
1989). Whatever type of meaning a constituent X ordinarily has, its AS, written [[X]]F,
is a set of meanings of that type. If X contains a focus, [[X]]F corresponds to the mean-
ings of expression that are like X but with the focus replaced by an alternative. Thus the
VP in (3a) has an alternative set of properties (i.e. meanings!) such as {brought Harry’s
book, brought Harry’s bike, . . .} that in (3b) like {brought Joe’s book, brought Frida’s
book, . . .}. The FSE then relates the AS of its argument to its ordinary denotation;
roughly, only VP will denote the set of individuals that have the property denoted by VP
but not other properties from VP’s AS.

2.2 Free Focus Has No Truth-Conditional Content

As for free focus (i.e. focus not associated with a FSE), it is sometimes suggested that
it introduces an existential presupposition, e.g. KIM took Harry’s book would presuppose
that someone took Harry’s book.This idea has been refuted in various ways. First, the
sentence can answer the questions in (4), neither of which implies (and hence license
a presupposition to the effect) that someone took Harry’s book:

(4) a. Who, if anyone, took Harry’s book?

b. Did anyone take Harry’s book?

Second, a sentence like (5) would be falsely predicted to presuppose that someone took
Harry’s book (and hence be contradictory) (Jackendoff, 1972):

105
 DA N I EL BÜ R I N G

(5) NOBody took Harry’s book.

Third, there is clear difference between (6a), which—due to the cleft construction—
does presuppose that someone took Harry’s book, and (6b)—plain free focus—which
doesn’t (Rooth, 1999):

(6) a. Was it KIM who took Harry’s book?

b. Did KIM take Harry’s book?

Similarly, it is sometimes suggested that free focus entails exhaustiveness of the answer.
But again, there is notable difference between plain focus, (7a), and a cleft, (7b), with
only the latter truly entailing exhaustivity (and hence creating a contradiction)(Rooth,
1999):

(7) a. (Who attended the meeting?) KIM attended the meeting. Maybe Jo did
too.
b. (Who attended the meeting?) It was KIM who attended the meeting.
# Maybe Jo did, too.

Data like these show that existence and exhaustivity may be implicatures of focus but
not parts of truth conditional meaning. This leaves us to look for the meaning of free
focus elsewhere.

3 Pragmatic Meaning of Free Focus

3.1 Mentalist and Discourse-Related Approaches

An intuitive characterization of focus meaning would be that focus marks what the
speaker intends to highlight, or emphasize, or regards as most important, or most
informative.This is an example of what I will call a mentalist theory of focus meaning.
The obvious problem with the above mentalist characterization is that the attitudes it
ascribes to speakers are difficult to verify independently, or even elicit systematically.
A variant of this problem regards characterizations like “focus presupposes the
existence of closed set of relevant alternatives”: this is either almost trivially true, if
we understand ‘existence’ to be about actual, real-world existence, or again immensely
difficult to track down, if we talk about existence in the speaker’s (or addressee’s)
mind.
Even more formal-looking characterizations like “focus creates/presupposes an open
formula and then provides a unique value for the variable in that formula” or “focus
instructs the hearer to open a new a file card and write something on it” are haunted by
basically the same problem: It presupposes the reality of a certain kind of representation
in the participants’ minds (assuming in this case that formulas wouldn’t be supposed to
be objects in the real world), to which independent access would be needed in order to
make the claim falsifiable.
For this reason, most formal work on the interpretation of free focus has instead taken
felicity in a discourse to be the main data to be accounted for: which focusing of a given
sentence is felicitous as an answer to a particular question, or in response to a particular
statement, or as a narrative continuation of a previous sentence. The meaning of focus

106
 F O CU S A N D I N T O N AT I O N

under that view consists of discourse appropriateness conditions. Call these discourse
related approaches to focus meaning.
(One may argue that on such a view, focus doesn’t have meaning so much as merely a
pragmatic function; this is a valid point, though I will continue to use the term meaning
in what follows.)
To give an example, answers to questions show a very systematic focusing pattern,
describable as in (8):

(8) Question/Answer Congruence (QAC):

In an answer, focus marks that constituent which corresponds to the wh-
phrase in the question.
(9) Q: Who did Jones’s father recommend for the job?
A: He recommended JONES for the job.

(A formal rendering of) (8) can be seen as part of a discourse grammar, providing a
partial analysis of the meaning of focus via the notion “felicitous focusing in a (ques-
tion-answer) discourse.”
At the same time, many if not most approaches will—implicitly or explicitly—assume
that, truly, QAC in (8) doesn’t constrain the relation between a focused declarative
sentence and a preceding interrogative sentence but between a focused declarative sen-
tence and some sort of context representation construed on the basis of the question (and
maybe additional context as well). This allows the analyst to extend the analysis to
cases in which, say, a pertinent question is assumed to be on participants’ minds but
never explicitly uttered:

(10) (A and B find the door to the classroom locked. Says A:) JONES has keys
to this room.

Here we can assume that the question of who has keys to the room is on A and B’s
mind, even though it hasn’t been uttered. A plausible analysis assumes that focus in A’s
answer is licensed by question/answer congruence (QAC), where ‘question’ is under-
stood as the question under discussion, QUD, in a discourse model. Overt QAC as
in example (9) above would then be the special case of a QUD set by an actual question
utterance.
So even discourse related approaches to focus are mentalist in this sense. But in con-
tradistinction to approaches that talk about ‘importance’, ‘highlighting’ or ‘open formu-
lae in speakers’ minds, discourse related theories show a direct way in which to access
the pertinent mental discourse representation independently: When there is an explicit
cotext (e.g. an interrogative utterance), it has a defined effect on context representation
(e.g. setting the QUD); and all relevant aspects of context representation can at least be
manipulated by particular utterances such as explicit questions, statements, etc.

3.2 Alternative Semantics for Free Focus

Let us assume, then, that discourse context, as construed in conversationalists’ minds,
and to be modeled in a formal approach to focus, includes a question under discussion
(which in turn may be set by an explicit utterance or through more indirect means).
Question/answer congruence can then be defined in terms of AS, as introduced in

107
 DA N I EL BÜ R I N G

section 2.1 above, as follows. The AS of a declarative sentence will be a set of sentence
meanings, for concreteness: a set of propositions. A question denotation, too, is mod-
eled as a set of propositions, roughly the set of all literal answers (true or false). Then
QAC can, to a first approximation, be stated as follows:

(11) A declarative sentence S matches the QUD only if every proposition

in the question meaning is an element of the AS of S. formally: QUD
Ž[[S]]
F

Another discourse context that typically determines focus is found in so-called con-
trastive focus:

(12) Contrastive Focus: Focus marks the constituent that distinguishes a sen-
tence (or parts thereof) from a previously uttered one.
T: Last year Jones’ father recommended Smith for the job.
C: This year he recommended JONES for the job.

The relation between T (for target) and C (for contrast) is simply that the denotation of
T must be an element of the AS of C: [[T]]O [[C]]F (the same rule captures corrections
as a special case of contrast).
As Rooth (1992) points out, both (12) and (8)/(11) can be seen as anaphoric proc-
esses, in the sense that T and Q(UD), respectively, are salient antecedents to which the
focusing in A/C relates. However, an essential disjunction remains, as the antecedent is
a subset of the AS in the Q/A case but an element of AS in the contrastive case.
Finally, some more terminology: In (12C), the entire sentence is the domain of the
focus Jones. What that means is that we work with alternatives at the sentence level
(propositions), not the alternatives to the focus alone (individuals). Syntactically, the
domain consists of the focus and its background; semantically, the background cor-
responds to the invariable part of the alternatives (‘Jones’s father recommended . . . for
the job’), the focus to the variable one (though of course, the alternatives, being set-
theoretic objects, do not literally have an internal structure, i.e. there are no semantic
objects that correspond to focus and background in isolation).
In some cases, the domain of the focus is smaller than the whole sentence (and
accordingly the alternatives are not propositions), e.g. VP in (3), or, as in (13), DP
(from Rooth, 1992, note that there is no contrast at the higher nodes, say, ‘met an X
farmer’).

(13) A CaNADian farmer met an AMERican farmer.

Focus domain—and derivatively, background—are essential in understanding focus;

the notion of focus only makes sense as focus within a domain. The focus is that element
within its domain that makes the (denotation of the) domain different from the target
(or the question). As (9) and (12) aptly illustrate, the focus itself needn’t even be new
(Jones is previously mentioned in the immediate context), it is only ‘new’ relative to its
background, in its domain. Focus is thus by its very nature a relational phenomenon; it
would be a hopeless task to try to grapple with the formal modeling of focus independ-
ent of its background.

108
 F O CU S A N D I N T O N AT I O N

3.3 Given and New

As just pointed out, a notion like ‘not previously mentioned’, or plainly ‘new’, is not
sufficient to model focus. On the other hand, focus alone is not sufficient to explain
(marked) accent patterns generally:

(14) (Context: We got a toy car for Veronica. What else might she like?)
a. She’d probably also like a gaRAGE for her car.
b. A gaRAGE for her car.
c. She’d probably like a bed for her DOLLs.
(15) (What did you buy Max for his birthday?)
I bought him a WATCH at the anTIQUE store.

The focus in (14a) must be the entire object a garage for her car (cf. also the term answer
in (14b)), yet the accent falls on the head noun garage, not the phrase final PP, as usual
(e.g. (14c)). The intuitive reason is that (her) car in (14a/b) (but not (her) dolls in (14c))
is anaphoric, or given.
Conversely, while the answer-focus in (15) must be a watch—which, to be sure, is
accented—there is another pitch accent on the final PP at the antique store, ostensibly
because it is not given.
These examples show that givenness is not just the inverse of focus. Given elements
can occur within a focus, and then get deaccented as in (14a/b), and nongiven elements
can be outside of a focus and be accented, see (15). (‘Deaccenting’ is the usual term for this
phenomenon: An element in a syntactic position which would, in default prosody, receive
an accent, goes accentless. One shouldn’t literally assume, though, that there once was an
accent on that very element which then is removed by some kind of transformation.)
There is also a crucial difference in formal properties: Roughly, an element is given
if its meaning is salient in the discourse, prototypically if some synonymous or semanti-
cally close related expression (e.g. a hyperonym) has been mentioned before. As such,
givenness is privative: One can say of an expression in context, but regardless of its syn-
tactic environment, whether it is given or not. As emphasized in section 3.2 just above,
this is different for focus, which is by its very nature relational.
That being said, it is worth noting that for a constituent C to be given is a degenerate
case of contrastive focusing: namely a focus domain without a focus in it. To sketch the
reasoning: If C doesn’t contain a focus, its AS is the singleton set containing its ordinary
meaning. Thus, any ‘target’ for that focus domain would, in order to be an element of
C’s AS, have to be synonymous with C. Building on this, Schwarzschild (1999) pro-
vides a unification of contrastive focus and givenness, which he calls Givenness (with a
capital G). The remarks about the difference between focus and givenness made above
do not apply to Schwarzschildian Givenness (see op. cit. for further details).

4 Grammatical Properties of Focus

4.1 Size: Broad, Narrow, Sentence-Wide

Using questions and contrasts, we can pragmatically diagnose focus and hence become
more clear about the relation between [F]-marking and its prosodic realization. For

109
 DA N I EL BÜ R I N G

example, (16) can answer any of the questions in (16a), or—preceded by ‘No!’—cor-
rect any of (16b). (17) answers/corrects (17a/b). Crucially, swapping the intonations
between (16) and (17) yields infelicitous discourses:

(16) Excess heat damaged the FUSilage.

a. (i) What happened?/What caused the accident?
(ii) What effects did the excess heat have?
(iii) What did the excess heat damage?
b. (i) Someone shot at the plane.
(ii) Excess heat melted the windows.
(iii) Excess heat damaged the internal wiring.
(17) Excess HEAT damaged the fusilage.
a. What damaged the Fusilage?
b. A saboteur damaged the fusilage.

Such a pattern indicates that the NPA on fusilage in (16) (which is where the NPA falls
in neutral intonation) is compatible with focus (and hence [F]) on the entire sentence,
the verb phrase, or the object nominal, whereas the ‘shifted’ NPA on heat in (17) can
only signal subject focus.
Sentence (16), as a reply to (16ai)/(16bi), is called whole-sentence focus all other
cases in (16) and (17) narrow focus. Note that narrow foci can be larger than a single
word, e.g. VP in the context of (16aii)/(16bii). The term broad focus is sometimes
used synonymously with whole-sentence focus, sometimes somewhat vaguely to mean
foci larger than a single word or phrase.
The size of the focus is occasionally referred to as the scope of the focus, or its domain.
We discourage both usages: Since foci can be scope-bearing elements (e.g. quantifiers),
‘scope of a focus’ to mean its size invites confusion with the logical scope of the focused
element. On the other hand, we use domain of the focus to mean the larger constituent
which includes the focus and its background (see section 3.2 above).
It turns out that what we referred to as the normal intonation of a sentence system-
atically coincides with the realization of a whole-sentence focus. From this we may
conclude that either all sentences have a focus (and the apparently focus-less ones are
simply whole-sentence-foci), or that focusing only changes normal prosody in case there
is a contrast between focus and nonfocus within a sentence.
We can now make precise some more terminology: Sentences may have normal into-
nation, or marked intonation. The former may correspond to sentence-wide focus (or per-
haps the absence of focus altogether), the latter to non-sentence-wide (=narrow) foci (or
some other IS-related property). The inverse is not true: Some narrow focus patterns are
realized by neutral intonation (e.g. object focus in (16aiii)/(16biii)). This is why we speak
of marked intonation and narrow focus but not focus intonation and marked focus.
There is an independent question whether, apart from the placement of PAs, there
aren’t any other cues to the presence of (narrow) focus, prime candidates being the
shape (rising, falling, sharp rising . . .) and scaling (higher than previous, lower than pre-
vious . . .) of PAs. The answer to that question is not clear, but it seems that while there
are unambiguous phonetic cues to narrow focus, these are not obligatorily present in all
narrow focus sentences. For example (16), rendered with four pitch accents of roughly

110
 F O CU S A N D I N T O N AT I O N

descending height can answer any of the questions in (16a) (or correct any of (16b)),
but with only one PA, on fuselage, or the PA on fuselage significantly higher than the
previous ones, may introduce a bias to understand it as a narrow focus answer to (16aiii)
(or a correction to (16biii)).
In closing we note that a single sentence may contain more than one focus:

(18) a. (Who wants tea, or coffee?) KIM wants TEA.

b. (The floor was scratched when John moved the dresser.—No!
It was scratched when) SUE moved the FRIDGE.

Moreover, there may be sentences with discontinuous single foci:

(19) a. (What did you do to John?) I told him about the GAME.
b. (What happened to the book?) michela sent it back to the LIBrary.

Alternative semantics for focus as sketched in section 3.2 provides a way of handling
such cases as well by interpreting both foci in situ.

4.2 Types: Corrective, Presentational, New-Information

Apart from differences in focus size, it is regularly suggested that there are different types
of focus, such as informational, contrastive, corrective, etc.
Informally speaking, informational focus, sometimes also called presentational focus
or new-information focus, would generally mark the nongiven parts of an utterance (see
section 3.3) without necessarily establishing a contrast to a particular alternative (in
many cases, then, answer focus is, or at least can be, presentational). Contrastive focus,
on the other hand, juxtaposes two, perhaps always exclusive, alternatives; the prime
instance of this would be corrections.
The crucial question is what evidence we have that these are distinguished by gram-
matical means, e.g. different pitch accent patterns or types. Prosodic evidence in English
seems inconclusive. In particular, it is hard to distinguish between the claim that a
particular focus is, for example, corrective (and hence realized, say, with a larger pitch
excursion), and the claim that an ordinary), narrow focus is realized in a contrastive
utterance (which therefore has a larger pitch range in general).
On the other hand, Italian and many other Romance languages, have a focus front-
ing construction in which an intonationally distinguished, and pragmatically focused,
element appears in noncanonical left-peripheral position. Focus fronting never happens
with answer foci (these must be realized right-peripherally) but typically occurs in cor-
rections or explicit juxtapositions. This kind of pattern can be taken as evidence that
grammar does distinguish different kinds of foci, although analyses that aim to derive
the differences while maintaining a unified concept of focus have been proposed (Bru-
netti, 2004; L’opez, 2009; Samek-Lodivici, 2006).
It is important to distinguish constructions like Romance focus fronting, which
(ostensibly) marks a particular pragmatic type of focusing, from constructions like the
English cleft (which is also occasionally referred to as a focusing construction). English
clefts have particular semantic properties: namely an existence presupposition and an
exhaustivity claim. As we saw above, focus conversationally implicates the same things.
As such, focusing and clefting often align for pragmatic reasons as in (20a), but it is
straightforward to dissociate them, (20b) (Prince, 1978: 885/898):

111
 DA N I EL BÜ R I N G

(20) a. It was MaGRUDer that leaked it.

b. It was also during these centuries that vast internal migration from
the SOUTH took place.

A more complicated case is presented by Hungarian, which has a preverbal position

that has largely the same semantic properties as English clefts (existence presupposi-
tion and exhaustivity claim). Unlike with English clefts (and Romance focus fronting),
however, answer foci and corrective foci virtually obligatorily appear in that position,
unless they are decidedly incompatible with its semantic properties. Conversely—and
also unlike English clefts, see (20b)—elements in that position are always stressed and
interpreted as pragmatic foci (e.g. in Q/A sequences). É. Kiss (1998) concludes that
foci in the preverbal position in Hungarian are a different type of focus (she calls them
“identificational”) than focus in postverbal position (É. Kiss’s “informational focus”); it
seems equally possible, though, that exhaustivity (and existence) are contributed by the
position (like in English clefts), while the meaning of focus in pre-and postverbal posi-
tion is the same (Brody, 1995; Horvath, 1986, 2010; Szendrői, 2001, 2004).
To conclude, it is hard to determine with certainty whether there are different types
of grammatical focus. It is at least not obvious that intonational focus in English (and
other Germanic languages) differentiates systematically between uses such as contras-
tive, informational, or identificational.

4.3 Focus Realization Rules

We mentioned several times above that one main principle of focus realization in
English (and many other languages) is that foci bear the NPA. Truckenbrodt (1995)
makes the strong claim that this is indeed the only effect focusing has on prosody,
with all other effects following from prosody-internal regularities, and in particular
from what may be called Prosodic Inertia: that those principles that determine
neutral prosody remain active in focused structures as well. As a consequence, neutral
prosody is retained as much as compatible with the requirement that focus bear the
NPA.
Of particular importance is the assumption that the final PA in a prosodic domain
(usually the clause) is grammatically the strongest. This is of course the theoretical
counterpart to our earlier observation that the final PA is perceived as the most promi-
nent. It means that—within a prosodic constituent—one cannot have a stronger PA
followed by a weaker one (though the inverse is possible). Assuming that focus needs
to contain the strongest PA, PAs after the focus must be omitted altogether (since oth-
erwise, they would be strongest). Furthermore and trivially, if the focus doesn’t contain
a PA by virtue of unmarked prosody (say focus on a pronoun, or an auxiliary, or some
other functional element, or a phrase final predicate—see below), one will be added. In
a nutshell: the focus contains a PA; apart from that, everything up to and including the
focus will retain unmarked prosody; everything after the focus until the end of the focus
domain remains accentless.
One particularity of unmarked intonation that is retained under focusing is Predi-
cate Integration, roughly: predicates (in particular: verbs) remain accentless if fol-
lowing an accented argument. This effect has been widely discussed in English intransi-
tives, as well as head final verb phrases in Dutch and German:

112
 F O CU S A N D I N T O N AT I O N

(21) a. Your MOTher called. (not: Your mother CALLED)

b. ein BUCH lesen (not: ein buch LESen)
a book read
‘to read a book’
[German]

Whatever the exact scope and characterization of predicate integration is, its is—under
the view argued for here—not a consequence of some focus realization rule (as it is for
Gussenhoven, 1983; Selkirk, 1984, 1995, cf. the latter’s Focus Projection Rules),
but of unmarked intonation. So if we claimed above that within a focus (and before
it), unmarked intonation occurs, this will include unaccented predicates in integration
configurations. Such predicates will only be accented if they are focused, and their argu-
ment is not. (On integration (under various names) see among many others Schmerling
(1976), Fuchs (1984, 1976), Gussenhoven (1983), Jacobs (1991/2b, 1992, 1999); and,
from a slightly different vantage point, Sasse (1987), Jäger (2001); for a recent prosody-
based analysis Truckenbrodt (2006).)
If, as in (21), an argument–predicate sequence happens to occur in sentence-final
position, integration will lead to a ‘shift’ of the NPA to the left. But crucially, there is no
need for a rule of NPA shift, nor indeed for a rule of NPA placement (such a rules is e.g.
formulated in Cinque, 1993, and widely adopted); it simply follows from two factors:
local PA omission on a predicate due to integration, and the purely prosodic definition
of NPA as the final PA.

5 Summary
This article surveyed some core semantic, pragmatic, and phonological properties of
focus and intonation. Perhaps surprisingly, no definition of focus was offered. Rather, I
suggested a rough demarcation of phenomena potentially amenable to an analysis utiliz-
ing focus (marked accent patterns), introduced a formal tool to formulate focus-related
grammatical conditions (alternative sets) and then gave a semi-formal characterization
of two of the most common and best understood focus phenomena: question/answer
focus and contrastive focus.
A complete theory of focus would offer similar characterizations of all phenomena
within this realm, whatever they may be, and hopefully be able to unify them formally
(unless, of course, they turn out to be properly grammatically distinguished in a way not
attributable to other factors). At present, some generalization of contrastive focus along
the general lines of anaphoric contrast seems most promising in this regard, but many
details remain to be worked out.
Cross linguistic studies of focusing are of utmost importance in this project. Consider-
ing the lack of a general and sufficiently precise definition of focus, the starting point for
the more successful crosslinguistic studies are again uncontroversial instantiations of the
general phenomenon, namely answers and contrastive utterances. Once the grammati-
cal realization of such constructions in a language is well understood, one can proceed
to ask what other pragmatic (or grammatical) configurations in that language receive
the same or a similar realization, thereby widening again the scope of focus theory in
general, and allowing for meaningful crosslinguistic comparison.
Hopefully this article can serve as a guideline to researchers new to this endeavor,
and aid in avoiding common terminological and methodological pitfalls.

113
 DA N I EL BÜ R I N G

Bibliography
Akmajian, Adrian (1973). “The Role of Focus in the Interpretation of Anaphoric Expressions.” In Stephen
Anderson and Paul Kiparsky, eds., A Festschrift for Morris Halle, 215–226. New York: Holt, Rinehart &
Winston.
Beaver, David and Brady Clark (2008). Sense and Sensitivity: How Focus Determines Meaning. Cambridge:
Wiley-Blackwell.
Bosch, Peter and Rob van der Sandt (1999). Focus—Linguistic, Cognitive, and Computational Perspectives.
Cambridge: Cambridge University Press.
Bródy, Michael (1995). “Focus and Checking Theory.” In István Kenesei, ed., Levels and Structures,
Approaches to Hungarian 5, 31–43. Szeged: Jate.
Brunetti, Lisa (2004). A Unification of Focus. Padova: Unipress.
Büring, Daniel (2006). “Focus Projection and Default Prominence.” In Valéria Molnár and Susanne Win-
kler, eds., The Architecture of Focus, 321–346. Berlin/New York: Mouton De Gruyter.
Chomsky, Noam (1971). “Deep Structure, Surface Structure and Semantic Interpretation.” In Danny D.
Steinberg and Leon A. Jakobovits, eds., Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and
Psychology, 183–216. Cambridge: Cambridge University Press.
Cinque, Guglielmo (1993). “A Null Theory of Phrase and Compound Stress.” Linguistic Inquiry 24: 239–267.
Culicover, Peter W. and Louise McNally, eds. (1998). The Limits of Syntax, vol. 29 of Syntax and Semantics.
New York: Academic Press.
van Deemter, Kees (1994). “What’s New? A Semantic Perspective on Sentence Accent.” Journal of Seman-
tics 11:1–31.
van Deemter, Kees (1999). “Contrastive Stress, Contrariety, and Focus.” In Bosch and van der Sandt (1999),
3–17.
Dekker, Paul and Herman Hendriks (1996). “Links without Locations—Information Packaging and Non-
monotone Anaphora.” In Paul Dekker and Martin Stokhof, eds., Proceedings of the Tenth Amsterdam Col-
loquium, 339–358.
Dik, Simon C. (1997). The Theory of Functional Grammar. Part 1: The Structure of the Clause. Berlin, New
York: Mouton de Gruyter. Edited by Kees Hengeveld.
É. Kiss, Katalin (1998). “Identificational Focus and Information Focus.” Language 74: 245–273.
Erteschik-Shir, Nomi (1997). The Dynamics of Focus Structure. No. 84 in Cambridge Studies in Linguistics.
Cambridge: Cambridge University Press.
Fuchs, Anna (1984). “ ‘Deaccenting’ and ‘Default’ Accent.” In Gibbon and Richter (1984), 134–164.
Fuchs, Anna (1976). “Normaler und kontrastiver Akzent.” Lingua 38: 293– 312.
Geurts, Bart and Rob van der Sandt (2004). “Interpreting Focus.” Theoretical Linguistics 30(1): 1–44.
Gibbon, D. and H. Richter (1984). Intonation, Accent and Rhythm: Studies in Discourse Phonology. Berlin.
Gussenhoven, Carlos (1983). “Focus, Mode, and the Nucleus.” Journal of Linguistics 19: 377–417. Reprinted
as chapter 1 in Gussenhoven (1984).
Gussenhoven, Carlos (1984). On the Grammar and Semantics of Sentence Accents. Dordrecht: Foris.
Gussenhoven, Carlos (2007). “Types of Focus in English.” In Chungmin Lee, Matthew Gordon, and Daniel
Büring, eds., Topic and Focus—Cross-linguistic Perspectives on Meaning and Intonation, no. 82 in Studies in
Linguistics and Philosophy, 83–100. Dordrecht: Springer.
Halliday, M.A.K. (1967). “Notes on transitivity and theme in English, part II.” Journal of Linguistics
3: 199–244.
Hendriks, Herman (2002). “Information Packaging: From Cards to Boxes.” In Kees Van Deemter and
Rodger Kibble, eds., Information Sharing: Reference and Presupposition in Language Generation and Interpre-
tation. Stanford: CSLI.
Horvath, Julia (1986). Focus in the Theory of Grammar and the Syntax of Hungarian. Dordrecht: Foris.
Horvath, Julia (2010). “ ‘Discourse Features’, Syntactic Displacement and the Status of Contrast.” Lingua
120(6): 1346–1369.
Jackendoff, Ray (1972). Semantics in Generative Grammar. Cambridge, MA: MIT Press.
Jacobs, Joachim (1991/2a). Informationsstruktur und Grammatik. Opladen: Westdeutscher Verlag (= Linguis-
tische Berichte Sonderheft 4).
Jacobs, Joachim (1991/2b). “Neutral Stress and the Position of Heads.” In Jacobs (1991/2a), 220–244.
Jacobs, Joachim (1992). Integration. Tech. Rep. 14, SFB 282, Düsseldorf, Köln & Wuppertal.
Jacobs, Joachim (1999). “Informational Autonomy.” In Bosch and van der Sandt (1999), 56–81.
Jäger, Gerhard (2001). “Topic-Comment Structure and the Contrast Between Stage-Level and Individual-
Level Predicates.” Journal of Semantics 18: 83–126.

114
 F O CU S A N D I N T O N AT I O N

Ladd, Robert D. (1980). The Structure of Intonational Meaning. Bloomington: Indiana University Press.
Ladd, Robert D. (1996). Intonational Phonology. Cambridge: Cambridge University Press.
Lambrecht, Knud (1994). Information Structure. Cambrige: Cambridge University Press.
López, Luis (2009). A Derivational Syntax for Information Structure. No. 23 in Oxford Studies in Theoretical
Linguistics. Oxford: Oxford University Press.
Miller, Judi (2006). “Focus.” In Keith Brown, ed., The Encyclopedia of Languages and Linguistics, vol. 4,
511–518. Oxford: Elsevier.
Prince, Ellen F. (1978). “A Comparison of Wh-Clefts and it-Clefts in Discourse.” Language 54(4): 883–906.
Reinhart, Tanya (2006). Interface Strategies: Optimal and Costly Computations. Cambridge, MA: MIT Press.
Rizzi, Luigi (1997). “The Fine Structure of the Left Periphery.” In Liliane Haegeman, ed., Elements of Gram-
mar: Handbook of Generative Syntax, 281–337. Dordrecht: Kluwer.
Roberts, Craige (1996). “Information Structure in Discourse: Towards an Integrated Formal Theory of Prag-
matics.” In J.H. Yoon and Andreas Kathol, eds., OSU Working Papers in Linguistics 49: Papers in Semantics,
91–136.
Rooth, Mats (1985). Association with Focus. Ph.D. thesis, University of Massachusetts, Amherst.
Rooth, Mats (1992). “A Theory of Focus Interpretation.” Natural Language Semantics 1: 75–116.
Rooth, Mats (1999). “Association with Focus or Association with Presupposition.” In Bosch and van der
Sandt (1999), 232–244.
Samek Lodivici, Vieri (2006). “When Right Dislocation Meets the Left-Periphery: A Unified Analysis of
Italian Non-Final Focus.” Lingua 116(6): 836–873.
Sasse, Hans-Jürgen (1987). “The Thetic/Categorical Distinction Revisited.” Linguistics 25: 511–580.
Schmerling, Susan F. (1976). Aspects of English Sentence Stress. Ph.D. thesis, Austin.
Schwarzschild, Roger (1997). “Interpreting Accent.” Unpublished ms. Rutgers University.
Schwarzschild, Roger (1999). “GIVENness, AvoidF and Other Constraints on the Placement of Accent.”
Natural Language Semantics 7(2): 141–177.
Selkirk, Elisabeth (1984). Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA:
MIT Press.
Selkirk, Elisabeth (1995). “Sentence Prosody: Intonation, Stress, and Phrasing.” In John A. Goldsmith, ed.,
The Handbook of Phonological Theory, 550–569. London: Blackwell.
von Stechow, Arnim (1989). Focusing and Backgrounding Operators. Tech. Rep. 6, Fachgruppe Sprachwis-
senschaft, Universität Konstanz.
Szendró´i, Kriszta (2001). Focus and the Syntax-Phonology Interface. Ph.D. thesis, University College
London.
Szendró´i, Kriszta (2004). “Focus and the Interaction between Syntax and Pragmatics.” Lingua 114(3):
229–254.
Truckenbrodt, Hubert (1995). Phonological Phrases: Their Relation to Syntax, Focus, and Prominence.
Ph.D. thesis, MIT, published 1999 by MIT Working Papers in Linguistics.
Truckenbrodt, Hubert (2006). “Phrasal Stress.” In Keith Brown, ed., The Encyclopedia of Languages and Lin-
guistics, vol. 9, 572–579. Oxford: Elsevier.
Vallduví, Enric (1990). The Informational Component. Ph.D. thesis, University of Pennsylvania. (published
1992: Garland).
Vallduví, Enric and Maria Vilkuna (1998). “On Rheme and Contrast.” In Culicover and McNally (1998),
79–180.

115
 1 . 10
C O NTE X T -S ENSIT IV IT Y
Tom Donaldson and Ernie Lepore

1 Introduction
We begin with a puzzle. Consider the following two utterances:

(1) I’m Spartacus! [Said by Spartacus]

(2) I’m Spartacus! [Said by Antoninus]

What Spartacus said was true, and what Antoninus said was not. Yet the two slaves
uttered the exact same sentence, so how can this be?
Admittedly, the puzzle is not very hard, and its solution is uncontroversial. The first
person pronoun “I” is—to use a technical term—context-sensitive. When Spartacus
uses it, it refers to Spartacus; when Antoninus uses it, it refers to Antoninus. So when
Spartacus says “I’m Spartacus,” he expresses the true proposition that he, Spartacus,
is Spartacus. And when Antoninus says it, he expresses the false proposition that he,
Antoninus, is Spartacus. The sentence “I’m Spartacus” expresses different propositions
when used by different people.
Another example will help. Contrast these two utterances, made by subjects in a
study carried out by experimental epistemologists:

(3) This is a zebra. [Said by someone while pointing at a zebra]

(4) This is a zebra. [Said by someone while pointing at a cleverly decorated mule]

(3) is true; (4) is false—even though they are tokens of the same sentence. Why? Because
the word “this,” like “I,” is context-sensitive. Roughly, an instance of the word “this”
refers to whichever object is being indicated, or otherwise rendered salient, to the audi-
ence of the utterance. In (3), the object indicated is a zebra, and so, the token of “this”
in (3) refers to that zebra. This animal is the subject of the sentence. So in saying (3),
the speaker truly says, of a certain zebra, that it is a zebra. In the case of (4), on the other
hand, the indicated object is a mule. In saying (4), then, the speaker says of a certain
mule that it is a zebra. Utterance (4) is therefore false.
Enough examples for now—let’s have some definitions. “Context,” as the word is
used in semantics, is a technical term. There is some dispute about exactly what we
should take contexts to be, but for our purposes it suffices to say that the context of an
utterance is comprised of all those features of the circumstance in which the utterance is
made that are relevant to interpretation. A context-sensitive word, phrase, or sentence
 CO N TEXT- SENS I T I V I T Y

is one which can receive different semantic values (interpretations) when used in dif-
ferent contexts. In particular, a declarative sentence is context-sensitive if it can express
different propositions when said in different contexts.
We can now begin to form a list of words which are context-sensitive in an uncontro-
versial and straightforward way:

Pronouns: I, me, my, you, yours, he, him, his . . .

Demonstratives: this, that, those, these . . .
Adjectives: current, present . . .
Adverbs: today, tomorrow, now, here, there . . .

There is little dispute about the context-sensitivity of the words on this list. Most Eng-
lish speakers quickly acknowledge their context-sensitivity when it is pointed out to
them. It is also usually thought that tense morphemes—like the ending “-ed” which
marks the past tense in English—are also context-sensitive (although this has been
disputed).
Other expressions, however, are the subject of controversy. Indeed, a great deal of
dispute revolves around the question of how context-sensitive natural languages are.
What’s more, there is disagreement about how context-sensitivity should be explained
in systematic linguistic theory. We will begin, in sections 2–5, by looking at the ques-
tion, Which natural language expressions are context sensitive? We will go on, in sections
6–9, by investigating how context-sensitivity should be best accommodated within lin-
guistic theory.

2 Further Apparent Examples of Context Sensitivity

2.1 First Example: “On the Left”

“On the left” is not as obviously context-sensitive as “I” and “this,” but a little reflection
suggests forcefully that it is indeed context-sensitive. Two New York tour buses pass one
another on the road, going in opposite directions past the American Museum of Natural
History. On one bus, the guide says, correctly:

(5) The American Museum of Natural History is on the left.

If the guide on the other bus were to say the same thing, all of the passengers would look
the wrong way.
“On the left,” it would seem, is context-sensitive. Roughly, in a given context, some-
thing is “on the left” just in case its direction from the contextually salient location is
about 90 degrees anti-clockwise from the contextually salient “forward” direction.

2.2 Second Example: “It’s Raining”

John is having a three-way phone conversation with his friends Nitika (in Bombay) and
Kurt (in Vienna). The conversation goes like this:

(6) John: What’s the weather like, Nitika?

Nitika: It’s raining.

117
 TO M DO N A LDSO N A N D E R N I E L E P O R E

John: Kurt?
Kurt: It’s not raining.

It seems that Nitika and Kurt could both have uttered the truth—despite the appear-
ance of contradiction. Arguably, when Nitika says “It’s raining,” she says that it is rain-
ing in Bombay, while when Kurt says “It’s not raining,” he says that it is not raining in
Vienna. Does it follow that “It’s raining” is context-sensitive?

2.3 Third Example: “Heavy”

Suppose that Chiara has bought a bike as a present for her niece. She chose this par-
ticular bike because, when compared to the alternatives, it was the most lightweight.
Having bought the bike, Chiara went to the post office, intending to mail the bike to
her niece. She was disappointed to find that she would be charged by the kilogram, and
the postage cost was enormous. Now consider:

(7) This bike is not heavy. [Said by Chiara in the store, admiring its lightweight
construction]
(8) This bike is heavy. [Said by Chiara in the post office, despairing at the cost of
sending the bike by mail]

Arguably, both utterances (7) and (8) are true. In the context of (7), the bike does not
satisfy “heavy,” but in the context of (8), it does. A natural thing to say is that—some-
how—there has been a shift in the “standard” or “threshold” for heaviness.

2.4 Fourth Example: “Green Leaves”

(9) is clearly context-sensitive, because it contains the word “my”:

(9) The leaves on my maple tree are green.

Charles Travis has argued that this sentence is context-sensitive in another, less obvi-
ous, way (Travis 1997). Travis asks us to consider Pia who owns a Japanese Maple tree,
with russet leaves. Thinking that the leaves should be green, Pia paints them, and hav-
ing finished, she says (9)—apparently truly. Shortly afterwards, Pia receives a phone
call from a friend, a botanist looking for green leaves for a study of green-leaf chemistry.
Pia offers him the leaves from her Japanese Maple tree. This time, when she utters (9),
she says something false. The point is not that “my maple tree” has a different referent
on the two occasions—the same tree is referred to both times. Moreover, the tree’s
leaves have not changed color between the first utterance and the second. As Travis
likes to put it, it seems that what “counts as” green is different in the two contexts.

2.5 Fifth Example: Quantifiers

When someone opens his cupboard, peers inside, sighs, and reports:

(10) All the peaches are rotten.

118
 CO N TEXT- SENS I T I V I T Y

He is unlikely to be reporting on a threat to the world fruit supply: rather, his quantifier
“all the peaches” ranges just over those peaches in his cupboard or those in his kitchen,
or some similar contextually salient domain. The very same sentence, used in a different
context, would have different truth-conditions because the domain of quantification
would be different. Arguably, then, sentences involving quantifiers are context-sensi-
tive, because the domain of quantification is contextually determined.

3 Minimalism, and Moderate and Radical Contextualism

Multiplying examples of this sort might quickly lead one to conclude that all terms in
natural language are sensitive to context, and in complicated ways that are hard to
describe systematically. Any given declarative sentence, on this view, can express dif-
ferent propositions in different contexts. One might even begin to think it misleading
to speak of the meaning of a sentence type—on the grounds that the meanings of its vari-
ous tokens have so little in common. This sort of position is sometimes called ‘radical
contextualism’.
Radical contextualists are at one end of a spectrum of different views. At the other
end of the spectrum are the minimalists, who deny there is context-sensitivity in natural
language beyond whatever is introduced by the “easy” examples of context-sensitive
expressions discussed in section [1]. Between the radicals and the minimalists there are
a variety of moderate contextualist positions.

4 Objections to Radical Contextualism

It would be accepted by everyone that (11) is context-sensitive, because it contains the
demonstrative “these”:

(11) These leaves are green.

We could say that (11) has no truth-conditions independently of context. However, most
philosophers would say that an adequate semantic theory for English would provide us
with a generalization of the form:

Relative to a context c, (11) is true iff ____

that is, a generalization which tells us the truth-condition of (11) relative to an arbitrary
context. One’s first guess might be:

(11*) Relative to a context c, “These leaves are green” is true iff the objects
indicated in c are green leaves.

This is perhaps a little crude, but it is usually thought that semanticists should seek
generalizations of roughly this kind. The characteristic claim of radical contextualists is
that this search will fail; motivated by examples like those discussed in section [2], radi-
cal contextualists deny that generalizations like this could capture the subtle and multi-
farious ways in which the truth conditions of English sentences depend on context.
This, then, is the negative claim characteristic of radical contextualism. Problems
arise, however, when radical contextualists get around to defining their views positively.

119
 TO M DO N A LDSO N A N D E R N I E L E P O R E

In particular, it is difficult for radical contextualists to answer the question, What does it
take for an utterance of (11) to be true? Here is Travis struggling with this question:

What could make given words “The leaves are green” true, other than the
presumed “fact that the leaves are green,” is the fact that the leaves counted as
green on the occasion of that speaking. Since what sometimes counts as green
may sometimes not, there may still be something to make other words “The
leaves are green” false, namely that on the occasion of their speaking, those
leaves (at that time) did not count as green.
(Travis 1997: 101–2; see, also, Travis 1996: 457)

This passage suggests an utterance of (11) can be true even when the indicated leaves
aren’t green, because they might still count as green. Apparently, as Travis sees it, the
indicated leaves can count as green even if they aren’t. This should seem strange all by
itself, but the key critical point we want to register is that in saying this Travis has actu-
ally contradicted the central claim of his radical contextualism. It would seem that on
Travis’s view, something like the following is true:

(11Travis) Relative to a context c, “These leaves are green” is true iff the objects
indicated in c count as green leaves in c.

But this is exactly the sort of generalization which radical contextualists claim to be
impossible!
A second inconsistency charge against radical contextualism arises from Travis’s
answer to the question: “What, other than the indicated leaves being green, could ever
suffice for the truth of and utterance of (11)?” To see what’s gone wrong, note that
Travis’s answer itself is provided in a particular context. Let’s call this ‘Travis’s Con-
text’, or TC for short. According to Travis, the sentences in his paper have their truth-
conditions determined in TC. In particular, sentences containing the word “green” in
Travis’s article, must, according to radical contextualism, have the kind of context-
sensitive truth conditions radical contextualism claims all sentences have. So the truth
conditions for (12):

(12) What could make the given words “The leaves are green” true, other than
the presumed “fact that the leaves are green,” is the fact that the leaves counted
as green on the occasion of that speaking.

are determined (in part) by what counts as green in TC, i.e., the semantic contribution
of the expression “green” as it occurs in an utterance of (12) in TC depends on what
counts as green in TC. Let’s say that anything that counts as green in TC is greenTC. But
if this is so, (12) then should be read as (12TC):

(12TC) What could make the given words “The leaves are green” true, other
than the presumed “fact that the leaves are green,” is the fact that the leaves
counted as greenTC on the occasion of that speaking.

We hope it is clear this is not what Travis intends to be saying. He doesn’t mean to
suggest that the leaves counting as greenTC is what would make an utterance of “The

120
 CO N TEXT- SENS I T I V I T Y

leaves are green” true in contexts other than TC (and (12) is clearly about contexts
other than TC). The bottom line is that he seems to be trying to use “green” as it occurs
in (12) in a context-insensitive way. Since, according to him, such context-insensitive
uses are not possible, it follows that (12) either says something false (i.e., that counting
as greenTC is what makes utterances of “the leaves are green” true in contexts other than
TC) or it employs the word “green” in a way inconsistent with radical contextualism.
(The same point applies to our use of “green” on the RHS of (11Travis), and, for that mat-
ter, for our attempt to formulate this objection.)
In short, if these arguments are sound, then the radical contextualists have provided
no coherent alternative to the standard picture. We turn now to more systematic ways
of distinguishing context-sensitive expressions from those which are insensitive to con-
text. In this next section, we will canvass a number of tests for context sensitivity: that
is, procedures for identifying context-sensitive expressions.

5 Tests for Context-Sensitivity

In section [2], we explored a variety of examples which suggest that context-sensitivity
in natural language extends well beyond the easy examples we discussed in section [1].
All of these arguments are more or less controversial. Moreover, taken together they
push us towards radical contextualism—a position we have seen is inherently unten-
able. What we would like is a test with which systematically to distinguish expressions
which are context-sensitive from those which aren’t. In this section we will look at
some of the tests that have been proposed for this purpose.

5.1 The Disquotational Indirect Quotation Test (IDR)

Linguistic practices indifferent to context should expose expressions that are not. We
will begin by relying on (Cappelen and Lepore 2005).
Disquotation is usually a safe policy in insuring accuracy in indirect reporting. When
Alice says “Bob Dole is Italian,” she is usually and most straightforwardly best reported
with “Alice said that Bob Dole is Italian.” Context-sensitive language, however, man-
dates adjustments. If Bob Dole utters “I’m Italian,” the indirect quotation “Bob Dole
said that I am Italian” fails to report him (unless the reporter is Bob Dole). Such obvious
considerations lead to a test for context-sensitivity: the easier it is to indirectly report an
utterance disquotationally, regardless of indifference or ignorance about its context of
use, the less likely it is that its constituents are context-sensitive. In short:

IDR: If reporters can easily and truly indirectly disquotationally report an utter-
ance of a sentence S by an agent A, i.e., with ‘A said that S,’ despite indiffer-
ence about, or ignorance of, its original context of utterance, then it is unlikely
S is context-sensitive.

5.2 The Collectivity Test (Collectivity)

Collective reports of utterances of a sentence often preserve truth. If John and Herman
both utter “China is larger than Russia,” we easily collectively report them with “John
and Herman said that China is larger than Russia.” An exception is when what’s uttered
is context-sensitive. Suppose we know John and Herman each uttered “Bill left today.”

121
 TO M DO N A LDSO N A N D E R N I E L E P O R E

We cannot then accurately collectively report them with “John and Herman both said
that Bill left today,” unless we happen to know that both John and Herman spoke today.
The context-sensitive word “today” blocks collection; and in general, context-sensitive
expressions resist collective accurate reporting. Such obvious considerations lead to a
test for context-sensitivity based on collecting what others say: The easier it is to collect
distinct utterances of a single sentence into a single disquotational report, regardless of
indifference about, or ignorance of, their original contexts of use, the less likely it is that
the sentence uttered is context-sensitive. In short,

Collectivity: Let u and u’ be utterances of S by A and B. If reporters can easily

collect u and u’ into a single true indirect disquotational report, i.e., with ‘A
and B both said that S,’ despite indifference about, or ignorance of, the original
contexts of utterance, it is unlikely S is context—sensitive.

In both tests—and others, cf. (Cappelen and Lepore 2005)—it is presumed that accu-
rate reporting requires reporter and reportees to be expressing the same thought (or
proposition) when the latter’s utterances easily admit of a disquotational or collective
indirect report.

5.3 Agreement Tests (Agree)

When we apply the IDR and Collectivity tests to expressions like “immigrate,” “left,”
“nearby,” “it’s raining,” “local,” “heavy,” and “ready,” the tests suggest that these expres-
sions are not context-sensitive. For example, an utterance of (13) can often easily be
reported with (14), no matter what the environment of the reporter is like, despite
indifference about, or ignorance of, the original context of utterance.

(13) It’s raining.

(14) John said that it’s raining.

Likewise, when John and Herman each utter (13), they can be easily reported collec-
tively with (15) across contexts, despite indifference towards, or ignorance of, John’s
and Herman’s original contexts of utterances.

(15) John and Herman said that it is raining.

“Left” and “nearby” submit to the same collection patterns. How can an expression
behave like this if it is context-sensitive? Several authors have argued that IDR and
Collectivity are poor tests for identifying context-sensitive expressions. Space limita-
tions preempt our going very far into this discussion; however, some of these authors
recommend replacing indirect reporting tests with agreement tests.
Suppose John and Herman both utter (13) indicating different locations. While they
may easily admit of the collective report (15), it is difficult in such circumstances to
characterize them with (16):

(16) John and Herman (both) agree that it is raining.

The same applies to some other supposedly context-sensitive expressions, including

those containing “nearby,” “left,” “local,” “ready,” “enough,” etc. Exploiting such

122
 CO N TEXT- SENS I T I V I T Y

judgments, some authors have recommended introducing Agree-1 and Agree-2 as tests
for context- sensitivity (in place of indirect reporting tests which they see as flawed).

Agree-1: If A utters S, B utters its negation, and they are not easily reported as disagree-
ing, say, with ‘A and B disagree whether S’, then S is semantically context-sensitive
(Cappelen and Hawthorne 2009)

If A says “I’m happy” and B “I’m not happy,” no one would report them with “A and B
disagree”; this is because “I” is context-sensitive.

Agree-2: If A and B both utter S and can be reported as agreeing, say, with ‘A
and B agree that S’, then that is evidence S is semantically invariant across
its distinct utterances. If, on the contrary, distinct utterances cannot be so
reported, this is evidence S is not semantically invariant across its distinct con-
text of utterance.
(Cappelen and Hawthorne 2009: 54–55)

If both A and B utter “John bought a table,” then if they can be correctly reported with
“A and B agree that John bought a table,” this is because “table” is context-insensitive.
In short: if speakers in distinct contexts utter S but can be reported as agreeing, then
S is context-insensitive; and if one utters S while the other utters its negation, then S is
context-sensitive only if they needn’t disagree.
One reason for preferring agreement judgments over indirect reporting ones in ascer-
taining context-sensitivity is that the former do not admit of distributive readings while
the latter may. Agreement and disagreement require coordination on a single proposi-
tion, while indirect reporting does not.

6 Content and Character

So far, we have been discussing the question, “Which expressions are context-sensi-
tive?” We have yet to discuss how to explain context-sensitivity in systematic linguistic
theory. It is to this topic we turn in the next few sections.
We start with the easiest sort of example:

(1) I’m Spartacus! [Said by Spartacus]

(2) I’m Spartacus! [Said by Antoninus]

As noted back in section [1], the first person pronoun “I” refers to Spartacus when Sparta-
cus uses it, and to Antoninus when he uses it. Here’s a possible reaction to the example:

The example shows that distinct tokens of “I” differ in meaning. So the type
word “I” lacks meaning. It only has a meaning when used in context.

This would be an overreaction to the example. There is an important commonality

among the semantic interpretations of the different tokens of the first-person singular
pronoun: the word “I” always refers to the person using it. Anyone who fails to recog-
nize this feature of the word has not understood it, and so is not a competent user of
the word. This suggests that semantic theory should associate a function with the word
“I,” a function which takes a context as its argument and yields as its value the person

123
 TO M DO N A LDSO N A N D E R N I E L E P O R E

speaking in that context. If you like, this function ascribes a meaning to “I” and the
referent of each token of the word is the value of the function when evaluated at the
relevant context.
Let’s look at our second example:

(3) This is a zebra. [Said by someone while pointing at a zebra]

(4) This is a zebra. [Said by someone while pointing at a cleverly decorated mule]

We can associate with the demonstrative type word “this” a function, which takes a
context as argument and yields the object indicated, or otherwise rendered most salient,
in the context. Each token of “this” then refers to the value of the function evaluated
at the relevant context.
Now for some terminology: the content of a token expression is its semantic inter-
pretation in its particular context; the character of a type expression is a function which
maps each context to the appropriate content. Notice that even non-context-sensitive
expressions can be said to have characters—in such cases the character is a constant
function. Notice also that whole sentences have characters. The character of the sen-
tence “I was hungry yesterday,” for example, is a function which maps each context c
to the proposition that the speaker at c was hungry on the day before the time of c. The
terms “content” and “character” come from David Kaplan, who introduced this way of
dealing with context-sensitivity (Kaplan 1989a, 1989b).
This approach to accommodating context-sensitivity is standard for dealing with
all of the uncontroversial cases of context-sensitivity, as discussed in section 1. Other
(putative) sorts of context sensitivity must be treated differently, as we shall see.

7 Unarticulated Constituents
We now return to an example discussed in section 2.2. Consider:

(17) It’s raining. [Said by John Perry’s son, in Palo Alto]

Plausibly, the sentence “It’s raining” is context-sensitive. In some contexts—as in

(17)—it expresses the proposition that it is raining in Palo Alto; in other contexts it
expresses the proposition that it is raining in Paris. Not everyone will agree that “It’s
raining” is context-sensitive in this way, but let’s put that question to the side, and
instead ask how to explain this sort of context-sensitivity, on the assumption that it
exists.
Picking up a line of thought from section 6, we might suggest that the source of the
context-sensitivity in “It’s raining” is that one of the words in the sentence has a non-
constant character. “Rain” is the only good candidate, and it seems that the theory will
have to go something along these lines:

The word “rain,” when used in a context c, refers to the property of raining at
the location contextually salient at c.

When one uses “It’s raining” in a context in which the contextually salient location is
Palo Alto, it will express the proposition that it is raining in Palo Alto. On the other
hand, when one uses the same sentence in a context in which the salient location

124
 CO N TEXT- SENS I T I V I T Y

is Paris, it will express a different proposition. So far, so good—the theory seems to

work well.
One problem for this approach arises when we consider statements like:

(18) It is raining in Palo Alto.

Suppose we use (18) in a context c in which the salient location is Paris. Then, accord-
ing to the proposed theory, the content of “rain” in c will be the property of raining in
Paris. But then, on the proposed theory, the token of (18) would appear to express the
proposition that it is raining in Paris, and it is unclear how “in Palo Alto” could play a
role in determining the proposition expressed. Something has clearly gone wrong!
So it seems that we need an alternative to the standard content/character approach.
When discussing (17), John Perry famously said:

In order to assign a truth-value to my son’s statement [of (17)] . . . I needed

a place. But no component of his statement stood for a place . . . Palo Alto is
a constituent of the content of my son’s remark, which no component of his
statement designated; it is an unarticulated constituent.
(Perry 1986: 206)

Here is Perry’s proposal: the underlying relation here is Rain(t, l)—which holds when
it is raining at time t in location l. (17) specifies one of the arguments for this relation
(the “t” argument), but no syntactic constituent of (17) refers to a location, so the sec-
ond argument is left, as it were, blank. The sentence, therefore, expresses an incomplete
proposition, a proposition with a hole in it (a “propositional radical,” as Kent Bach puts
it (Bach 1994):

(19) Rain (1st September 1996,_)

Different tokens of (17) express different propositions because this propositional radical
gets “filled in” in different ways in different contexts. On this view, (17) expresses the
propositional radical (19), but a particular token of (17) could express the proposition:

(20) Rain(September 1, 1996, Palo Alto)

Some part of what is asserted in such a case does not correspond to a syntactic constitu-
ent of the sentence used to make the assertion. It is, as one says, “unarticulated” (cf.
Perry 1986). It is somehow, in a way never made clear in the literature, “introduced”
into the proposition expressed without any direction from the uttered sentence.

8 Binding Arguments and Hidden Indexicals

A number of critics have challenged the idea of an unarticulated constituent; in par-
ticular, they invoke the so-called binding argument to support the idea that every object
in the proposition expressed is put there by an expression in the sentence uttered. We
begin with an example. Consider:

(21) Wherever John visits, it’s raining.

125
 TO M DO N A LDSO N A N D E R N I E L E P O R E

On its most natural reading, (21) expresses the proposition that for any location x that
John visits, it’s raining at x. On this reading, some sort of quantification is going on.
A natural thought here is that where there is quantification, there must be variable
binding. Accordingly, the adverbial phrase “Whenever John visits” is a quantificational
phrase, binding a variable in the part of (21) that follows the comma. On this view, the
variable does not correspond to any pronounced component of the sentence, but just the
same it must be there. It might seem strange to say that there are syntactic components
of sentences which are not pronounced, but in fact this is routine in modern syntax.
But what has this got to do with context-sensitivity? Hold onto the idea that “rain” is
followed by an unpronounced variable, and look again at (17):

(17) It is raining.

According to the view under discussion, consideration of (21) shows that (17) con-
tains an unpronounced variable. But in the case of (17) there is no quantificational
expression that could bind this variable, and so it occurs free. Since it occurs free, its
interpretation is fixed contextually. Roughly, in any context of use, the variable will
refer to whichever place is most salient in that context of use. On this view, then, the
variable in (17) behaves like an indexical (i.e. like “this” or “that”). Because it is not
pronounced, it is called a “hidden indexical.” Theories of this kind are called “hidden
indexical theories.”
It has been argued, perhaps most notably by (Stanley and Szabo 2000), that unarticu-
lated constituents theories of (17) (as discussed in the last section) cannot be extended
to cope with sentences like (21). If this is right, it provides an argument for hidden
indexical theories over unarticulated constituent theories.
It is perhaps worth pausing to draw attention to the key difference between the unar-
ticulated constituents’ approach and the hidden indexical approach. Both theories
agree that (17), uttered in a certain context, will express the proposition that it is rain-
ing in Palo Alto, and both theories agree that no pronounced component of (17) refers
to Palo Alto. The big difference is that the hidden indexical theory says that there is a
component of (17) that refers to Palo Alto—an indexical expression which is present
in the syntax but not pronounced—while the unarticulated constituents theory denies
that there is a component of (17)—even a hidden component—which refers to Palo
Alto in the context.

9 The Dynamic Lexicon

It is not uncommon for language users to coin new words “on the fly,” or to invent new
meaning for old words. (Clark 1983) gives many examples, including these:

Subjected to the musical equivalent of 72 hours in a dentist’s waiting room,

Bradley was apparently in real danger of being the first tourist ever Muzakked
to death. (From the San Francisco Examiner)
I think that it’s across from a quarry. That’s the only way I can landmark it.
(Said by a person talking about finding a beach).
I stopped in Perry’s for a quick crab. (Herb Caen, meaning a crab that could
be eaten quickly)

126
 CO N TEXT- SENS I T I V I T Y

Examples like these suggest that we should not think that language users just select their
words from a fixed lexicon; rather, the lexicon can be modified and extended in the
course of conversation. As Ludlow (2000) puts it, the lexicon is dynamic.
Clark uses this point to explain some putative instances of context-sensitivity. One
of his examples is that of denominal adjectives—i.e. adjectives derived from nouns, like
“Hegelian” (from “Hegel”) and “metallic” (from “metal”). Many denominal adjectives
have well-established meanings, but others do not. Clark gives this example:

Churchillian, for example, might mean with a face like Churchill, smoking a cigar
like Churchill, with a speaking style like Churchill, or any number of other things.
In principle, the list is unlimited; in practice, it is limited by what the speaker
can assume the addressees know about Churchill, and will be able to see that
he is alluding to.

If Clark is right about this, then sentences containing the word “Churchillian” are con-
text-sensitive, but we do not need to account for this by saying that the word is indexi-
cal, or by postulating a hidden variable, or by saying that such sentences do not express
complete propositions; rather, we should say that the lexicon is dynamic, and the lexical
entry for “Churchillian” is different on different occasions.

10 Three Further Supposed Cases of Context-Sensitivity

A summary paper on the literature on context-sensitivity would not be complete with-
out some discussion of contextualism in epistemology, or mention of contextualist solu-
tions to the liar and sorites paradoxes. However, our treatment of these issues will have
to be cursory, because of limitations of space.

10.1 Contextualism in Epistemology

Well-known arguments from epistemology—How do you know you’re not being tricked
by an evil demon?—might lead an epistemologist to say:

(22) I don’t know that I am forty years old.

However, when filling in his census form later that day, the very same epistemologist
might say:

(23) I know that I am forty years old.

At this point, he is likely to be accused of inconsistency—an accusation that he can

attempt to duck by arguing that “know” is context-sensitive. He might say, for exam-
ple, that there is a shift in which set of epistemic standards are salient between the two
contexts.
Contextualism, in epistemology, is the claim that “know” is context-sensitive.
Contextualists usually employ this doctrine to reconcile skeptical claims (made in the
context of an epistemological discussion) with anti-skeptical claims (made in an eve-
ryday context).

127
 TO M DO N A LDSO N A N D E R N I E L E P O R E

10.2 Contextualist Solutions to the Sorites Paradox

Shortly before leaving for the airport with my two suitcases, my traveling companion asks,
“Are your suitcases heavy?” In fact, the bags differ in weight by only a gram. I reply:

(24) One of my suitcases is heavy; the other isn’t.

It seems that this could not be true: my suitcases hardly differ in weight, so how can I
truly say that one is heavy and the other not? The example might lead one to accept:

First: If two things differ only slightly in weight, then either neither is heavy,
or they both are.

However, it is notorious that claims like First lead to problems. Suppose one has a
sequence of suitcases. The first weighs 1kg, the last weighs 100kg, and adjacent suit-
cases in the sequence differ in weight by only 1g. By repeated application of First, one
can derive a contradiction from the innocuous assumption that the first suitcase is not
heavy and the last is, and the assumption that adjacent bags in the sequence differ only
slightly in weight.
Contextualists about vagueness, recalling that there are independent grounds for
thinking that “heavy” is context-sensitive (see section 2.3), offer a principle like this in
place of First:

Second: In a given context, given any two contextually salient objects that dif-
fer only slightly in weight, either both objects are in the extension of ‘heavy’ in
the context or neither is.

To put it roughly, the line that divides heavy things from the things which are not
heavy moves around so that it never lies between two objects that are contextually sali-
ent and very similar in weight. The reason why it is improper for me to assert (24) is that
both of the suitcases are contextually salient, and they differ only slightly in weight.
The theory is attractive because it rejects First (which leads to contradiction) and
also offers an explanation of why First is so attractive. Vague predicates appear to be
“boundaryless” because the boundary moves in such a way that it never lies between two
contextually salient objects. Wherever you look, there it isn’t.

10.3 Contextualist Solutions to the Liar

Consider sentence (25):

(25) (25) is not true.

What is its semantic status? We have, in answering this question, at least the following
three options:

Option One: (25) is true.

Option Two: (25) is not true.
Option Three: (25) fails to express a proposition; it lacks a truth-condition.

128
 CO N TEXT- SENS I T I V I T Y

Prima facie, Options One and Two are inconsistent, as are Options One and Three.
It turns out that it is difficult to adopt one of these options and then stick with it despite
further thought. Whichever option one chooses, one will find oneself inclined to move
to a different one.
One might start by reasoning as follows. Suppose for contradiction that (25) expresses
a proposition. Then:

“(25) is not true” is true if and only if (25) is not true.

But “(25) is not true” just is (25), so by Leibniz’s law we can infer:

(25) is true if and only if (25) is not true.

This is a contradiction—so by reductio we can infer that (25) does not express a proposi-
tion. This argument gives us a good reason for accepting Option Three.
However, further thought suggests otherwise. We seem to have shown that (25) does
not express a proposition. But it would seem to follow that (25) is not true (since only
a sentence that expresses a proposition can be true). So we should accept Option
Two.
Again, further thought will force us to change our mind. We seem to have given a
sound argument for accepting Option Two—i.e. for accepting:

(25) is not true.

But this just is the sentence (25), so we have given a sound argument for (25). It would
seem to follow that (25) is true after all, and we then conclude that we should accept
Option One. Repeating our initial line of argument will take us back to Option Three—
and on it goes.
So to repeat, whatever semantic status we ascribe to (25), further thought seems to
force us to assign it a different status. The contextualist concludes that (25) must be
changing status; on this view, (25) is context-sensitive; and as we reason, the context
changes in such a way that (25) changes status.

11 Summary
• Some words—including “I,” “this,” “current” and “today”—are uncontroversially
context-sensitive.
• It is controversial how far context-sensitivity extends beyond these ‘easy’
examples.
• At one extreme, minimalists argue that the only context-sensitivity in natural
language is that introduced by the ‘easy’ examples. At the other extreme, radi-
cal contextualists argue that examples of context-sensitivity are so widespread and
multifarious that they cannot be accounted for by a systematic semantic theory.
• Radical contextualist positions are hard to state coherently (see section 4), and are
at any rate at odds with the results of tests for context-sensitivity (see section 5).
• We have discussed several different ways of explaining context-sensitivity: Kaplan’s
character/content approach (section 6); the ‘unarticulated constituents’ approach
(section 7); the ‘hidden indexicals’ approach (section 8) and the ‘dynamic lexicon’

129
 TO M DO N A LDSO N A N D E R N I E L E P O R E

approach (section 9). The character/content approach is standard for the ‘easy’
examples of context-sensitivity discussed in section 1, while the other approaches
are used in more controversial cases.
• It has been argued that postulating context-sensitivity can help us with the liar
paradox and the sorites paradox, and that it can be a useful part of a response to
skepticism in epistemology.

References
Bach (1994) “Conversational Impliciture,” Mind and Language 9: 124–162.
Beall J. and M. Glanzberg (Forthcoming) ‘The Liar Paradox’ in The Stanford Encyclopedia of Philosophy.
Bezuidenhout, A. (2002) “Radical Pragmatics” in J. K. Campbell et al. (eds.) Meaning and Truth: Investiga-
tions in Philosophical Semantics. New York: Seven Bridges Press, pp. 292–302.
Borg, E. (2004) Minimal Semantics. Oxford: Oxford University Press.
Burge, T. (1979) “Semantical Paradox,” Journal of Philosophy 76: 169–198.
Cappelen, H. and J. Hawthorne (2007) “Locations and Binding,” Analysis 67: 95–105.
—— (2009) Relativism and Monadic Truth. Oxford: Oxford University Press.
Cappelen, H. and E. Lepore (2002) “Indexicality, Binding, Anaphora and A Priori Truth,” Analysis 62:
271–281.
—— (2005) Insensitive Semantics. Oxford: Basil Blackwell.
Carston, R. (2002) Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell.
Clark, H. (1983) ‘Making Sense of Nonce Sense’ in Clark (1992) and Flores d’Arcais and Jarvella (1983)
—— (1992) Arenas of Language Use. Chicago: University of Chicago Press.
DeRose, K. (1995) “Solving the Sceptical Problem,” Philosophical Review 104: 1–52.
Devault, D. and M.D. Stone (2006) “Scorekeeping in an Uncertain Language Game,” The 10th Workshop
in the Semantics and Pragmatics Dialogue (brandial 2006), University of Potsdam, Germany.
Fara, D.G. (2000) “Shifting Sands: An Interest-Relative Theory of Vagueness”, Philosophical Topics, 28:
45–81. Published under the name ‘Delia Graff’.
Flores D’Arcais, G.B. and R.J. Jarvella (1983) The Process of Understanding Language. New York: Wiley.
Grice, H.P. (1975) “Logic and Conversation” in P. Cole and J. Morgan (eds.) Syntax and Semantics, 3: Speech
Acts. New York: Academic Press. Reprinted in H. P. Grice (ed.) Studies in the Way of Words, pp. 22–40.
Cambridge, MA: Harvard University Press (1989).
Haegeman, L. (1994) Introduction to Government and Binding Theory, 2nd edition. Cambridge, MA:
Wiley-Blackwell.
Hamm, F., H. Kamp and M. Lambalgen (2006) “There is No Opposition between Formal and Cognitive
Semantics,” Theoretical Linguistics 32: 1–40.
Jaszczolt. K.M. (ed.) (2000) The Pragmatics of Propositional Attitude Reports. Oxford/New York: Elsevier
Science.
Kaplan, David (1989a) “Demonstratives’’ in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan.
Oxford: Oxford University Press, pp. 481–563.
—— (1989b) ‘Afterthoughts’ in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford:
Oxford University Press, pp. 565–614.
Keefe, R. and P. Smith (eds.) (1996) Vagueness: A Reader. Cambridge, MA: MIT Press.
Lewis, D. (1979) “Scorekeeping in a Language Game,” Journal of Philosophical Logic, 8: 339–359.
—— (1996) “Elusive Knowledge,” Australasian Journal of Philosophy 74: 549–567.
Ludlow, P. (1999) Semantics, Tense, and Time: An Essay in the Metaphysics of Natural Language. Cambridge,
MA: MIT Press.
—— (2000) “Interpreted Logical Forms, Belief Attribution, and the Dynamic Lexicon,” in Jaszczolt
(2000).
Parsons, C. (1974) “The Liar Paradox,” Journal of Philosophical Logic 3: 381–412.
Perry, J. (1986) “Thought without Representation,” Proceedings of the Aristotelian Society Supplementary 60:
263–283. Reprinted in Perry (1993).
—— (1993) Problem of the Essential Indexical and Other Essays. Oxford: Oxford University Press.
—— (1998) “Indexicals, Contexts and Unarticulated Constituents” in Proceedings of the 1995 CSLI-Amster-
dam Logic, Language and Computation Conference. Stanford: CSLI Publications.
Raffman, D. (1996) “Vagueness and Context-Sensitivity”, Philosophical Studies 81: 175–192.

130
 CO N TEXT- SENS I T I V I T Y

Recanati, F. (2004) Literal Meaning. Cambridge: Cambridge University Press.

Roberts, C. (2004) “Context in Dynamic Interpretation” in L. Horn and G. Ward. (eds.) The Handbook of
Pragmatics. Oxford: Blackwell, pp. 197–220.
Sainsbury, M. (1996) “Concepts without Boundaries” in Keefe and Smith (1996), pp. 251–264.
Searle, John, (1978) “Literal Meaning,” Erkenntnis, 13: 207–224.
—— (1980) “The Background of Meaning” in J. Searle, J. F. Kiefer and M. Bierwisch (eds.) Speech Act
Theory and Pragmatics. Dordrecht: Reidel, pp. 221–232.
—— (1983) Intentionality. Cambridge: Cambridge University Press.
Shapiro, S. (2006) Vagueness in Context. Oxford: Oxford University Press.
Soames, S. (1999) Understanding Truth. Oxford: Oxford University Press.
Stalnaker, R. (1978) “Assertion” in P. Cole (ed.) Syntax and Semantics Vol. 9. New York: New York Aca-
demic Press, pp. 315–332.
—— (1998) “On the Representation of Context,” Journal of Logic, Language and Information, Special Issue
on Context in Linguistics and Artificial Intelligence, January, 7 (1): 3–19.
Stanley, J. (2002) “Making it Articulated,” Mind and Language 17 (1&2): 149–168. Reprinted in Stanley
(2007).
—— (2005) Knowledge and Practical Interests. Oxford: Oxford University Press.
—— (2007) Language in Context: Selected Essays. Oxford: Oxford University Press.
Stanley, J. and Z. Szabo (2000) “On Quantifier Domain Restriction,” Mind and Language 15 (2): 219–261.
Steedman, M. and M. D. Stone (2006) “Is Semantics Computational?” Theoretical Linguistics 32 (1): 73–89
Travis, Charles (1985) “On What is Strictly Speaking True,” Canadian Journal of Philosophy 15 (2): 187–229.
Reprinted in Travis (2008)
—— (1989) The Uses of Sense. Oxford: Oxford University Press.
—— (1996) “Meaning’s Role in Truth,”Mind 100: 451–466.
—— (1997) “Pragmatics” in B. Hale and C.Wright (eds.) A Companion to the Philosophy of Language. Oxford:
Oxford University Press.
—— (2008) Occasion Sensitivity: Selected Essays. Oxford: Oxford University Press.
Westerstahl, Dag (1985) “Determiners and Contexts Sets” in J. van Benthem and A. ter Meulen (eds.),
Quantifiers in Natural Languages. Dorgrecht: Foris, pp. 45–71.
—— (1989) “Quantifiers in Formal and Natural Languages” in D. Gabbay and F. Guenthner (eds.) Hand-
book of Philosophical Logic, IV (Dordrecht: Kluwer), pp. 1–131.

131
 1 . 11
RE L AT IV ISM
John MacFarlane

Relativism, in the sense at issue here, is the view that some assertions or beliefs can be
characterized as accurate, or objectively correct, only relative to a context of assess-
ment. We will begin by looking at a few cases that make relativist views tempting, then
consider the technical and philosophical issues such views raise.

Motivation

Epistemic Modals
“Joe might be in Boston,” says Maria. In doing so, she expresses a certain state of mind,
making it clear that she does not take her information to rule out Joe’s being in Boston.
But does she assert this? There are reasons to think not. George, who knows that Joe is
just down the hall, might disagree with what Maria has said. And, after Maria learns
that Joe is in fact in Berkeley, she too may think that she was wrong, and retract her
claim: “I said he might be in Boston, but that’s wrong—he can’t be in Boston, since
George just saw him down the hall.”
Noticing these things, one might conclude that Maria hasn’t asserted anything but
has only expressed her uncertainty. But suppose one does want to think of her speech
act as an assertion. What are the options? One could explain the disagreement and
retraction facts by taking her to have made a claim about what is left open by the infor-
mation of a broader group, including her possible interlocutors. But then it becomes
hard to understand how she could have taken herself to be warranted in making this
claim. We need to understand how it is that she can legitimately assert that Joe might
be in Boston on the basis of nothing more than her own ignorance, while others—and
her own later self—can evaluate what she has said in relation to their information.
All of this could be explained if we could make sense of the idea that the accuracy of
Maria’s assertion is not an absolute matter but varies with perspective. We could then
say that it is the assessor’s information, not the speaker’s, that matters for the accuracy of
an epistemic modal claim. That would explain why, when Maria learns that Joe is right
down the hall, she takes her original claim that he might be in Boston to have been
inaccurate, and retracts it. In this way we might capture the subjectivity of epistemic
modal claims—the way in which their accuracy depends on what is known, not on any
objective feature of the world—without construing them (as contextualist views do) as
claims about what some particular person or group knows. (See Egan et al. 2005, Egan
2007, Stephenson 2007, MacFarlane 2011; for critical remarks, see Wright 2007, von
Fintel and Gillies 2008, Dietz 2008.)
 RELATI V I S M

Taste Predicates
“This vegemite sandwich is tasty,” says Joe. In doing so, he expresses his liking for the
sandwich, but does he do anything more? Does he assert something, and if so, on what
does the accuracy of his assertion depend?
Contextualists hold that Joe has asserted something about the relation between the
food and himself, or between the food and a certain body of tastes. But then it is puzzling
why Sasha, whose tastes are very different, should reject Joe’s assertion. On a relativist
view, by contrast, Joe’s assertion can only be classed as accurate or inaccurate relative to
a context of assessment, and it is the assessor’s tastes that matter for its accuracy, rather
than the speaker’s. If Sasha despises the taste of vegemite, while Joe loves it, then Joe’s
assertion will be inaccurate as assessed by Sasha but accurate as assessed by Joe. And, if
Joe later comes to hate the taste of vegemite, his earlier belief will be inaccurate relative
to his current context, and he will rightly consider it to have been wrong. In this way,
the relativist tries to capture the subjectivity of taste claims—the way in which their
accuracy depends on a sense of taste—without reducing them to claims about how vari-
ous foods relate to the tastes of a person or group. (See Kölbel 2002, Lasersohn 2005,
2009, Stephenson 2007, MacFarlane 2007; and for criticism, Glanzberg 2007, Cappelen
and Hawthorne 2009, ch. 4.)

Future Contingents
“There will be a sea battle tomorrow,” says Themistocles, at a time when it is objectively
indeterminate whether there will be a sea battle the next day. Is his assertion accurate or
inaccurate? The question can only be answered, according to a relativist view, relative
to a particular context of assessment. Themistocles’ assertion is inaccurate as assessed
from the moment at which it is made (mo), accurate as assessed from a moment one day
later on a possible future history with a sea battle (m1), and inaccurate as assessed from
a moment one day later on a possible future history without a sea battle (m2). In this
way the relativist splits the difference between the supervaluationist, who holds that
the assertion is (absolutely) inaccurate—and has trouble explaining why it need not be
retracted by the asserter at m1—and the “true futurist,” who holds that the assertion is
either accurate or inaccurate absolutely, even though no objective feature of the situ-
ation breaks the symmetry between the two possible histories. (See MacFarlane 2003,
2008; and for criticism, Heck 2006.)

Such proposals raise two foundational questions in semantics:

The technical question What revisions are needed to standard truth-conditional

semantic frameworks to make room for relativist accounts of this kind?

The philosophical question How can we make philosophical sense of the idea
that the accuracy of an assertion or belief is assessment-relative? Do we really
understand what such proposals say? How do relativist theories differ from non-
relativist theories in their predictions about language use?

In what follows, we will briefly consider how each of these questions might be
answered.

133
 J O H N M A CFA R L A N E

The Technical Question

Index and Context

The task of a semantic theory, according to Lewis (1980), is to define truth at a context
of use for sentences of a language L. Why is this our target notion? Because if we know
the conditions under which sentences of L are true at a context, we know enough to
understand and communicate with speakers of L. One counts as a speaker of L in virtue
of participating in a convention of uttering only true sentences of L and trusting other
L-speakers to do the same. Thus, if we know that a speaker of L utters S in context c, and
we know that in L, S is true at c if and only if p, we can presume that the speaker believes
that p and intends for us to believe this too. And if we want to communicate that q to
another speaker, we have only to select a sentence T that is mutually known to be true
in our context if and only if q.
Because languages contain infinitely many sentences, and we want a finite characteriza-
tion of our semantic knowledge, our definition of ‘true at c’ must be recursive. However,
it is not in general possible to state the condition for a sentence to be true at a context in
terms of the conditions for its constituents to be true at a context. One reason is that the
constituents of some complex sentences are not sentences but open formulas. Another
reason is that a proper treatment of modal and temporal operators requires considering the
truth value of the embedded sentence at times and worlds other than the time and world
of the context. For example, to evaluate ªYesterday: Iº we need to consider the truth
value of I relative to the day before the present day. If we are recursively defining truth at
a context, we will need to evaluate I at a context c' that differs from our current context c
only in the time of the context. However, as Lewis points out (1983, 29), there is no guar-
antee that there will be such a context. Contexts represent possible occasions of use, and
are not arbitrary collections of parameters. Since the agent of the context must exist at the
time and world of the context, if we shift the time back to before the agent of the context
existed, there will be no corresponding context. Moreover, even if there were a context c'
that differed from c only in the time of the context, we would not want to say that ªYester-
day: Iº is true at c iff I is true at c'. For I might contain temporal indexicals like ‘today’ or
‘now’, and these must be evaluated with respect to the original context c. ‘Yesterday it was
warmer than it is now’ can be true, even though the embedded sentence ‘It is warmer than
it is now’ is not true relative to any context (Kamp 1971, Kaplan 1989).
The standard solution to these problems is to recursively define truth at a context and
an index, which is simply a collection of independently shiftable parameters (world, time,
assignment of values to the variables, and so on). The clause for ‘Yesterday’ could be:

Yesterday ªYesterday: Iº is true at c, (w, t, a) iff I is true at c, w, t', a, for some t'
belonging to the day before t.

Although what we really care about at the end of the day is truth at a context, a recur-
sive definition of truth at a context and index can serve our purposes, as long as we can
recover a definition of truth at a context from it. And we can:

Lewisian postsemantics A sentence I is true at c iff I is true at c, wc, tc, a, for any
assignment a. (Here wc and tc are the world and time of the context c.)
(cf. Lewis 1983, 31)

134
 RELATI V I S M

MacFarlane (2003, 2005a) calls this step the postsemantics, to distinguish it from the
recursive definition of truth at a context and index, which we might call the composi-
tional semantics.

Assessment Sensitivity
As Lewis makes clear, the role of the index in such a theory is entirely technical. If there
is no operator that shifts a parameter, then the parameter is not needed in the index.
And, if there is an operator that shifts the parameter, no further justification is needed
for putting that parameter in the index. The index is just part of the machinery for giv-
ing a systematic definition of truth at a context; it plays no further role and is subject to
no further extrasystematic constraints.
So the presence of outlandish parameters of the index (taste, information state, etc.)
does not itself make a semantic theory “relativist” in any philosophically interesting
sense. It does not prevent the theory from assigning truth values to sentences at con-
texts, or making absolute judgements of the accuracy of assertions.
We can see this more clearly if we change our target notion to make room for the pos-
sibility that accuracy is not absolute. Instead of defining truth at a context of use, let us
define truth at a context of use and context of assessment—a possible situation in which
a speech act or mental attitude might be assessed. We can then distinguish two kinds of
contextual sensitivity. An expression is use-sensitive if its extension depends on features
of the context of use, and assessment-sensitive if its extension depends on features of the
context of assessment. Or, making finer discriminations: an expression is F-use-sensitive
if its extension depends on the F of the context of use, and F-assessment-sensitive if its
extension depends on the F of the context of assessment. A “relativist” semantic theory
is one that takes some expressions to be assessment-sensitive. This amounts to taking
accuracy to be relative to contexts of asssement, since

Accuracy An assertion of S at c is accurate, as assessed from c', iff S is true as used

at c and assessed from c'.

A compositional semantics that employs a taste parameter in the indices (perhaps to

make sense of an operator like ‘on any standard of taste’) need not take any expression
to be taste-assessment-sensitive. For such a semantics can be combined with either a
contextualist or a relativist postsemantics:

Contextualist postsemantics S is true as used at c (and assessed from c') iff S is true
at c, wc, tc, where wc is the world of c and tc is the taste of the agent at c.
Relativist postsemantics S is true as used at c and assessed from c' iff S is true at c,
wc, tc', where wc is the world of c and tc' is the taste of the agent at c'
(i.e., the assessor).

According to the contextualist theory, the accuracy of an assertion of ‘This is tasty’ will
depend on the physical disposition of the demonstrated food at the time of the asser-
tion and on the speaker’s tastes at the time of the assertion. According to the relativist
theory, by contrast, it will depend on the physical disposition of the demonstrated food
at the time of the assertion and the assessor’s tastes at the time of assessment. Since the
same assertion can be assessed from indefinitely many perspectives, there are only per-
spective-relative answers to the question whether the assertion was accurate.

135
 J O H N M A CFA R L A N E

Content and Circumstance

Lewis (1980) argued that, for purposes of doing semantics, one need not talk about con-
tents or propositions at all. In that respect he differs from Kaplan (1989), who adopts
a two-stage picture of semantics. In the first stage of such a semantics, we define the
expression relation, specifying for any sentence S and context c the structured proposition
that S expresses at c. In the second stage, we give a recursive truth definition of truth
at a circumstance of evaluation for the structured propositions. In Kaplan’s framework,
circumstances are time/world pairs, but Kaplan acknowledges that “other features,” such
as location, may also be needed (1989, 504). Though it is not clear that Kaplan would
go so far, one could in principle take subjective features like taste or body of information
to be parameters of circumstances.
The truth of a sentence relative to a context is the joint product of these two
stages:

Kaplanian postsemantics A sentence S is true at a context c iff the content

expressed by S at c is true at the circumstance of c, that is, wc, tc .
(cf. Kaplan 1989, 522)

Here, the context plays two distinct roles in determining a truth value for an occurrence
of a sentence. It plays a content-determinative role in fixing which content is expressed,
and a circumstance-determinative role in fixing at which circumstance this content is to
be evaluated. Hence, there will in general be two distinct ways to take some expres-
sion to be F-use-sensitive. One can take the expression’s content to depend on the F of
the context of use, or one can take the F of the context of use to play a role in picking
out the “circumstance of the context”—the circumstance one looks at in determining
whether the sentence is true at the context, and in determining whether an assertion or
belief in the context is accurate.
Contextualist views typically use the first model. Contextualists about ‘tasty’, for
example, often hold that the sentence ‘vegemite is tasty’ expresses different proposi-
tions depending on the tastes of the speaker. But the second model is also an option.
One could take the content of ‘vegemite is tasty’ to have truth values relative to worlds,
times, and tastes, and let the tastes of the speaker (or more generally the taste relevant
at the context of use) help determine which triple is the “circumstance of the context.”
On such a view, which following MacFarlane (2009) we can call nonindexical contextual-
ism, sentences have truth values relative to contexts of use, and accuracy is absolute, not
assessment-relative. An assertion that vegemite is tasty is accurate, simpliciter, just in
case the taste of vegemite is good according to the taste of the speaker (or whatever taste
is relevant at the context of use). On this view, ‘vegemite is tasty’ is taste-use-sensitive,
not taste-assessment-sensitive.
Some writers in the literature use the term “truth relativism” for any view that relativ-
izes propositional truth to “nonstandard” features like tastes or information states. How-
ever, as the possibility of nonindexical contextualist views shows, such relativism at the
level of circumstances of evaluation is compatible with the traditional view that the
accuracy of particular assertions and beliefs is an absolute matter. (Conversely, one can
have assessment sensitivity without any nonstandard parameters of circumstances. For
an example in which circumstances are just worlds, see MacFarlane 2008.) Since most
of the traditional worries about truth relativism are worries about the rejection of this

136
 RELATI V I S M

traditional view, and not about the individuation of propositions, we follow MacFarlane
(2005a) in reserving the label “truth relativism” for views that countenance assessment
sensitivity. (Lasersohn 2005 does not explicitly relativize truth to contexts of assess-
ment, so it might appear that his view is not relativist, in the sense distinguished here.
However, that would be misleading. For Lasersohn, a “context of use” is not a concrete
possible situation in which a sentence might be used but an abstract sequence of param-
eters. How we should set these parameters in interpreting an utterance of ‘This is tasty’
depends, Lasersohn holds, on both features of the concrete speech situation, which help
determine the reference of ‘this’ and the relevant world and time, and on features of the
concrete assessment situation, which determine the “judge” of the context. So both the
concrete use situation and the concrete assessment situation play a role, even though
the distinction between them is not made notationally salient.)
If we want a view on which ‘vegemite is tasty’ is taste-assessment-sensitive, then
again there are two options in a Kaplanian framework. If we give the context of assess-
ment a content-determinative role, then we have a form of content relativism.

Content-relativist postsemantics S is true as used at c and assessed from c' iff the
content expressed by S at c, as assessed from c', is true at the circumstance of c,
that is, wc, tc, where wc and tc are the world and taste of c.

On a content-relativist view, there is no absolute answer to the question which proposi-

tion is expressed by a sentence at a context of use. Different propositions are expressed,
relative to different contexts of assessment. (Content relativism has had few advo-
cates in the literature, as compared with truth relativism. But see Cappelen 2008 and
Weatherson 2009.)
If we give the context of assessment a circumstance-determinative role, then we
get a

Truth-relativist postsemantics S is true as used at c and assessed from c' iff the
content expressed by S at c is true at the circumstance wc, tc', where wc is the
world of c and tc' is the taste of c'.

Here the context of use and context of assessment jointly determine which circum-
stance is relevant to the truth of the sentence relative to these contexts. According
to this view, an assertion or belief that something is tasty will be accurate, as assessed
from another context c', if the way the thing tastes at the time of the assertion or belief
is good according to the tastes of the assessor at c'. So the very same assertion or belief
may be accurate, as assessed by one observer, and inaccurate, as assessed by another.
This is one way of trying to capture the idea that tastiness depends on our evaluative
reactions, without reducing all ascriptions of tastiness to claims about how foods strike
some particular person or group.

Variables
A relativist who adopts a Lewisian or Kaplanian approach can keep the idea that an
adjective like ‘tasty’ is semantically a one-place predicate. There is no extra argument
place for a taste or judge. This may be counted an advantage, but it makes accounting
for binding phenomena more awkward. We must analyze

137
 J O H N M A CFA R L A N E

(1) That’s tasty to Sam

as

(2) To Sam: tasty(that),

where ‘To Sam:’ is an operator that shifts the taste parameter of the index (or, in Laser-
sohn’s 2005 version, a predicate-modifier). And we must analyze

(3) Everyone got something tasty

(on the reading where it means that everyone got something tasty to them) as

(4) [Everybody x][Something y](x got y & to x: tasty(y)).

A tempting alternative is to treat ‘tasty’ as semantically two-place, with an implicit argu-

ment place for a taste. We could then analyze (1) as

(5) Tasty(that, Sam’s taste)

and (3) as

(6) [Everybody x][Something y](x got y & tasty(y, x’s taste))

Following Stephenson (2007), we could take the extra argument place to be filled, when
not bound, by a special unpronounced pronomial element, PROJ. On Stephenson’s
account, PROJ denotes the judge of the index; here, we could take it to denote the taste
of the index, or (if we had no other need for a taste parameter in the index) we could
take its value to be set directly by the taste of the context of assessment.
Thus, although it is sometimes assumed that relativist views require operators to
motivate novel parameters in the index (Cappelen and Hawthorne 2009, ch. 3, Stanley
2005, ch. 7), the issue of assessment sensitivity can be pried apart from questions about
the parameters of indices, and from related questions about whether there are opera-
tors that shift these parameters. This should not be surprising. As we will see below,
assessment sensitivity is a way to explain certain features of the use of nonembedded
occurrences of sentences. Thus it ought to be largely independent of the issues about
embedded occurrences of expressions that motivate indices.

The Philosophical Question

What motivates work on assessment sensitivity is an

Empirical claim There are assessment-sensitive expressions in natural languages.

Some of the things we say and think are assessment-sensitive.

Defending this claim would require detailed examination of the use of particular bits of
language and careful consideration of alternative explanations. However, the empiri-
cal claim presupposes a weaker, conceptual claim that can be established—if it can be
established at all—through philosophical clarification:

138
 RELATI V I S M

Conceptual claim The possibility of assessment-sensitive expressions is coherent

and intelligible. We know what it would be for an expression to be assessment-
sensitive, whether or not this possibility is realized in natural languages.

The ground-clearing of the last section has helped us see what assessment sensitivity is
not, but it has not yet told us what it is. If we are given a definition of truth at a context of
use for a language, then we have a reasonable grip on how its expressions are to be used,
since we know that (to a first approximation) speakers try to assert things that are true at
their current contexts, and expect others to do the same (see Dummett 1981, 302, Lewis
1980, §2, 1983, §III). But what if we are given a definition of truth at a context of use and
context of assessment? How does this notion connect to the use of language? What is the
significance of assessment sensitivity for our linguistic practices? Something must be
said here, or it is not clear that in talking of truth relative to a context of use and context
of assessment, we are saying anything meaningful at all.
It is important to realize that both ‘true as used at a context c’ and ‘true as used at c
and assessed from c'’ are theoretical terms. Ordinary speakers don’t talk of sentences
being true relative to contexts. When they say things like, ‘What Joe said is true’ and
‘None of the claims in Sally’s report are true’, they are using a monadic predicate that
applies to propositions. So, we need to say something about what these relativized
truth predicates mean if we are to use them in theories of meaning. It might seem
tempting to define ‘is true as used at c’ as ‘would express a truth if it were used at
c’, but this doesn’t quite work: It would render the sentence ‘An English sentence is
being used’ true at all contexts, for example. In addition, if the ordinary monadic truth
predicate is nothing more than a logical device for expressing generalizations, as many
philosophers have claimed, then a definition that fixes its extension cannot serve as a
theory of meaning (see Dummett 1959 for the point, Williams 1999 for dissent, and
Patterson 2005 for a recent defense). A more promising approach is to let the meanings
of these technical terms be fixed (as with many technical terms) by their role in a larger
theory—here, a theory of language use. If we understand what is implied about the use
of language by claims made using the relativized truth predicate, then we understand
this predicate well enough.
In the case of ‘true as used at c’, we can get pretty far by saying, with Dummett and
Lewis, that speakers understand each other (in the absence of special reasons) to be try-
ing to assert only truths; that is, they follow a norm

T1 Assert I in c only if I is true as used at c.

Given that knowledge, plus a definition of truth at a context of use for a language, you
will be in a position to understand utterances others make, and to communicate your
own thoughts to others using the language. Simplifying somewhat: When you want to
communicate that you are hungry, you will find a sentence whose truth in your current
context depends on your being hungry; when someone else utters a sentence that your
definition tells you is true in their context just in case they are hungry, you will take
them to be communicating that they are hungry.
But suppose we have a doubly relativized truth predicate: ‘true as used at c and assessed
from c'’? What shall we say about how ascriptions of assessment-relative truth relate to
language use?
We might try quantifying over contexts of assessment in our norm for assertion:

139
 J O H N M A CFA R L A N E

T2 Assert I in c only if I is true as used at c and assessed from c', for some/
most/all contexts of assessment c'.

Or, more plausibly, we might simply privilege a particular context of assessment—the

one the speaker is in when making the assertion:

T3 Assert I in c only if I is true as used at c and assessed from c.

But if this is all we say, assessment sensitivity will be an idle wheel, with no real use in
semantic theorizing. For, given any definition T of ‘true as used at c and assessed from c'’
for a given language, according to which certain expressions are assessment-sensitive,
we can construct another definition T* that yields the same norms for language use as
T does (when combined with T3) but does not make any expressions assessment-sensi-
tive. Here’s the recipe:

T* I is true as used at c and assessed from c' iff according to T, I is true as used
at c and assessed from c.

T3 cannot help us distinguish these two definitions, because T3 only tells us the
significance of the assessment-relative truth predicate for the special case where the
context of assessment is the same as the context of use, and in such cases T and T* will
always agree. If T3 is all we have to say to connect the assessment-relative truth predi-
cate to language use, then, we will never have evidence that favors T over T*, so we
will never have good reason to embrace assessment sensitivity (see MacFarlane 2005a,
which draws on Evans 1985, 348).
However, we can distinguish theories that posit assessment sensitivity from theories
that do not if we consider not just norms for making assertions but norms for retracting,
challenging, and/or justifying them. For, when one is considering a challenge to an asser-
tion, or considering whether to retract an assertion made earlier, there are potentially
two relevant contexts: the context in which the assertion was made and the context in
which it is now being assessed. There is room here, then, for the context of assessment
to play a significant role in the norms for these speech acts.
For example, suppose we combine T3 with the following norm for retraction:

R Retract at c' an assertion of I made at c if I is not true as used at c and

assessed from c'.

Now we can distinguish between relativist and nonindexical contextualist views about
‘tasty’. Both will say that one may assert ‘this is tasty’ at c only if the demonstrated food
is good according to one’s tastes at c. But, given R, the two views will differ about the
condition under which such an assertion must be retracted at some later context c'
where one’s tastes are different. According to the nonindexical contextualist version,
‘this is tasty’ is not assessment-sensitive; so it is true as used at c and assessed from c' just
in case it is true as used at c and assessed from c. If it was okay to make the assertion in
the first place, then, it need not be retracted. According to the relativist view, however,
the truth of ‘this is tasty’ (at a context of use and context of assessment) depends on the
taste of the context of assessment; so, even if ‘this is tasty’ is true as used at c and assessed
from c, it may be false as used at c and assessed from c'. In such a case, the assertion would
have to be retracted, even though one did not violate any norm in making it.

140
 RELATI V I S M

Consider a concrete example on which the relativist and nonindexical contextualist

views differ. Suppose that at age ten, I asserted ‘fish sticks are tasty’. Now my tastes have
changed; I find fish sticks disgusting, and I deny that fish sticks are tasty. Am I obligated
to retract my earlier assertion? The nonindexical contextualist says that I am not. I can
let this earlier assertion stand, since, although it was not not specifically about my tastes,
it concerned my tastes at the time. The relativist says that I must retract; I cannot coher-
ently reject the proposition I asserted as false while holding fast to my earlier assertion.
Of course, claims about what speakers are obligated to do are not themselves claims
about what speakers will do. Still, this is enough to give us an empirical grip on the dis-
tinction. After all, speakers can be presumed to be trying to satisfy the norms, except
when there is good reason not to, and they may have intuitions about their obligations
to retract. If assessment-sensitive and non-assessment-sensitive theories make different
predictions about proprieties for language use, they will also make different predictions
about language use. Arguably, this is enough to vindicate the conceptual claim, and to
tell us what to look for in evaluating the empirical claim. (See MacFarlane 2003, 2005a,
2007 for the general strategy pursued here.)

References
Cappelen, H. (2008) “Content Relativism and Semantic Blindness,” in García-Carpintero and Kölbel
(2008), 265–286.
Cappelen, H. and J. Hawthorne (2009) Relativism and Monadic Truth (Oxford: Oxford University Press).
Dietz, R. (2008) “Epistemic Modals and Correct Disagreement,” in García-Carpintero and Kölbel (2008),
239–263.
Dummett, M. (1959) “Truth,” Proceedings of the Aristotelian Society n.s. 59: 141–62.
Dummett, M. (1981) Frege: Philosophy of Language, 2nd ed. (Cambridge, MA: Harvard University Press).
Egan, A. (2007) “Epistemic Modals, Relativism, and Assertion,” Philosophical Studies, 133: 1–22.
Egan, A., J. Hawthorne, and B. Weatherson (2005) “Epistemic Modals in Context,” in G. Preyer and P.
Peter (eds.) Contextualism in Philosophy (Oxford: Oxford University Press).
Evans, G. (1985) “Does Tense Logic Rest on a Mistake?” in Collected Papers (Oxford: Oxford University
Press), 343–63.
García-Carpintero, M. and M. Kölbel (eds.) (2008) Relative Truth (Oxford: Oxford University Press).
Glanzberg, M. (2007) “Context, Content, and Relativism,” Philosophical Studies 136: 1–29.
Heck, R. (2006) “MacFarlane on Relative Truth,” Philosophical Issues 16: 88–100.
Kamp, H. (1971) “Formal Properties of ‘Now’,” Theoria 37: 227–273.
Kaplan, D. (1989) “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology of
Demonstratives and Other Indexicals,” in J. Almog, J. Perry and H. Wettstein (eds.) Themes from Kaplan
(Oxford: Oxford University Press), 481–566.
Kölbel, M. (2002) Truth Without Objectivity (London: Routledge).
Lasersohn, P. (2005) “Context Dependence, Disagreement, and Predicates of Personal Taste,” Linguistics and
Philosophy 28: 643–86.
Lasersohn, P. (2009) “Relative Truth, Speaker Commitment, and Control of Implicit Arguments,” Synthese
166: 359–74. ¨
Lewis, D. (1980) “Index, Context, and Content,” in Stig Kanger and Sven Öhman (eds.) Philosophy and
Grammar (Dordrecht: Reidel). Reprinted in Lewis (1983).
Lewis, D. (1983) Philosophical Papers, vol. 1 (Oxford: Oxford University Press).
MacFarlane, J. (2003) “Future Contingents and Relative Truth,” Philosophical Quarterly 53: 321–36.
MacFarlane, J. (2005a), “Making Sense of Relative Truth,” Proceedings of the Aristotelian Society 105: 321–39.
MacFarlane, J. (2007) “Relativism and Disagreement,” Philosophical Studies 132: 17–31.
MacFarlane, J. (2008), “Truth in the Garden of Forking Paths,” in García-Carpintero and Kölbel (eds.)
2008, 81–102.
MacFarlane, J. (2009) “Nonindexical Contextualism,” Synthese 166: 231–50.
MacFarlane, J. (2011) “Epistemic Modals are Assessment Sensitive.” in E. Egan and B. Weatherson (eds.)
Epistemic Modality (Oxford: Oxford University Press).

141
 J O H N M A CFA R L A N E

Patterson, D. (2005) “Deflationism and the Truth Conditional Theory of Meaning,” Philosophical Studies
124: 271–94.
Stanley, J. (2005) Knowledge and Practical Interests (Oxford: Oxford University Press).
Stephenson, Tamina (2007) “Judge Dependence, Epistemic Modals, and Predicates of Personal Taste,”
Linguistics and Philosophy 30: 487–525.
von Fintel, K. and A. Gillies (2008) “CIA Leaks,” Philosophical Review 117: 77–98.
Weatherson, B. (2009) “Conditionals and Indexical Relativism,” Synthese 166: 333–57.
Williams, M. (1999) “Meaning and Deflationary Truth,” Journal of Philosophy 96: 545–64.
Wright, C. (2007) “New Age Relativism and Epistemic Possibility: the Question of Evidence,” Philosophical
Issues 17: 262–83.

Further Reading
MacFarlane, J. (2005b) “The Asssessment Sensitivity of Knowledge Attributions,” in T. Szabó Gendler
and J. Hawthorne (eds.) Oxford Studies in Epistemology 1: 197–233. (Argues that a relativist treatment of
knowledge attributions combines the virtues of contextualism and invariantism.)
Recanati, F. (2007) Perspectival Thought: A Plea for (Moderate) Relativism. Oxford University Press. (An
extended argument for what is here called non-indexical contextualism, with applications.)
Richard, M. (2004) “Contextualism and Relativism,” Philosophical Studies 119: 215–42. (Argues for a relativ-
ist treatment of vague terms and knowledge attributions.)
Richard, M. (2008) When Truth Gives Out Oxford University Press. (Argues for a relativist treatment of
vague terms, and against a relativist treatment of taste predicates.)
Zimmerman, A. (2007) “Against Relativism,” Philosophical Studies 133: 313–48. (Argues against a variety of
relativist proposals.)

142
 1 . 12
VAGUE NESS
J. Robert G. Williams

Puzzles of Vagueness
Taking away grains from a heap of rice, at what point is there no longer a heap? It
seems small changes—removing a single grain—can’t make a difference to whether or
not something is a heap; but big changes obviously do. How can this be, since the big
changes are nothing but the small ones chained together? Call this the “little-by-little”
puzzle.
At each stage, while removing grains from the original heap, ask yourself: “Is what
I have at this moment a heap?” At the initial stages, the answer will clearly be “yes.”
At the late stages, the answer will clearly be “no.” But at intermediate stages, the ques-
tion generates perplexity: It’s not clearly right to say “yes,” nor is it clearly right to say
“no.”A hedged response seems better: “It’s a borderline case of a heap”—but that’s not
yet a direct answer. So what is the answer to the original question, when confronted with
what we can all agree to be a borderline case of a heap? Call this the “borderlineness”
puzzle.
The little by little puzzle leads to the sorites paradox (from “sorites”—the Greek word
for “heap”). Here’s one version. Take a line of 10,001 adjacent men, the first with no
hair, the last with 10,000 hairs, with each successive man differing from the previous
by the addition of a single hair (call this a sorites series for “bald”). “Man N” names the
man with N hairs.
Man 0 is bald. Man 10,000 is not bald. Now consider the following collection of
horrible-sounding claims:

(1): Man 0 is bald, and man 1 is not bald.

(2): Man 1 is bald, and man 2 is not bald.
...
(10,000): Man 99,999 is bald, and man 10,000 is not bald.

If we reject the above, surely we must accept their negations:

(1*): it is not the case that: Man 0 is bald, and man 1 is not bald.
(2*): it is not the case that: Man 1 is bald, and man 2 is not bald.
...
(10,000*): it is not the case that: Man 99,999 is bald, and man 10,000 is not
bald.
 J . RO BERT G . W I L L I A MS

But given the various (N*), and the two obvious truths about the extreme cases, a
contradiction follows. Each (N*) is (classically) equivalent to the material conditional
reading of:

(N**) if Man N-1 is bald, then Man N is bald

Since Man 0 is bald, iterated modus ponens reasoning leads to the conclusion that Man
10,000 is bald, contrary to our assumptions. We must either find some way of rejecting
seemingly compelling premises, or find a flaw in the seemingly valid reasoning.
We turn next to the puzzle of borderlineness: If Harry is intermediate between clear
cases and clear noncases of baldness, “Is Harry bald?” seems to have no good, direct,
answer.
There are familiar cases where we cannot answer such questions: Having never seen
Jimmy, I am in no position to say whether he’s bald, simply because I lack relevant
knowledge. Lack of knowledge of the answer might also be appealed to in borderline
cases. But for such borderline cases, appeal to ignorance simply moves the challenge
one stage back. Why would we lack knowledge over whether Harry is bald? After all, it
seems puzzlement persists even if we find out all the relevant underlying facts about the
number and distribution of hairs on Harry’s head. What kind of status is this, where the
question of baldness is not only something we can’t answer, but where we can’t even
conceive of getting in a position to answer?
It’s common to say that it’s indeterminate, or indefinite, or that there’s no fact of the mat-
ter whether borderline-bald Harry is bald. Such terminology connects borderlineness
to a broader class of putative examples of indeterminacy. Examples include partially
defined terms, identities across theory change in science, the problem of the many,
ungrounded sentences that generate semantic paradoxes, and indeterminate condition-
als. It is a vexed question whether a unified analysis can be given of some or all of these
“indeterminacy” phenomena, and what shape it should take if so. (Note too that while
a general theory of indeterminacy might answer some questions about the status of
borderline cases, it’s not at all clear whether it would say anything interesting about the
little-by-little puzzle.)

Sharp Extensions
Sometimes it is suggested that vague language poses obstacles to the very idea of provid-
ing a classical (model-theoretic) semantics for natural language. Is this right?
Focus, for simplicity’s sake, on the basic case of extensional semantics. This assigns
extensions (sets of objects) as the semantic values of predicates. This may be thought
to be problematic when the predicates are vague. Sets of objects, as traditionally con-
ceived, are definite totalities—each object is either definitely a member, or definitely
not a member. Wouldn’t associating one of these totalities with “bald” force us, unjusti-
fiably, to “draw sharp boundaries” for its application? On the other hand, it seems that
we easily say which set should be the extension of “is bald.” It should be {x: x is bald}, i.e.
the set of things which are bald. We can use vague language (“in the theorist’s metalan-
guage”) to say what the extension should be.
Delicate questions about methodology in semantics arise. Can semantic theory say
simply that the extension of “red” (in English, or “rouge” in French) is the set of red
things? Or should we require it give a more illuminating characterization of that set?

144
 VA G U EN E S S

Even on the latter view, so long as the illuminating characterization is given in vague
terms, there’s no obvious obstacle to specifying the extension of a vague term. Thus
Lewis, in discussing his semantic treatment of counterfactuals in terms of the admittedly
vague notion of similarity, says “I . . . seek to rest an unfixed distinction upon a swaying
foundation, claiming that the two sway together rather than independently” (Lewis,
1973, p. 92). It would be problematic if one imposed the requirement that we should be
able to give a semantics for a vague object language within a nonvague metalanguage.
But such a requirement would need to be argued for.
Perhaps, though, the worry is that terms such as {x: x is red} picks out a “vague set”—
an entity not covered by standard set theories. Consider, for example, the set of bald
men B, and the set B+ that is the union of B with {Harry}, where Harry is borderline
bald. Since it’s indefinite whether Harry is bald, it seems it will be indefinite whether
Harry is a member of B; but he’s definitely a member of B+. All other individuals will
be a member of B iff they are a member of B+. It follows that it is indefinite whether B
and B+ contain exactly the same members. By the axiom of extensionality, two sets are
identical iff they have the same members. So it seems as if it will be indefinite whether
B and B+ are identical.
This may be problematic. Indefiniteness in identity is the target of the famous Evans
argument (Evans 1978, Salmon 1981). This purports to show, via an appeal to Leibniz’s
law, that such claims can be reduced to absurdity: Facts about identity are never vague.
So, not only do we require an account of vague sets in general, we also need to find some
way of addressing the Evans–Salmon argument.
However, a supporter of textbook (classical) semantics has responses. One idea is that
terms such as “{x: x is red}” are vague descriptions of precise sets. Suppose the relevant
domain D contains a single object a, and it is indefinite whether a is red. The candidates
to be picked out by “{x in D: x is red}” are the null set and the singleton {a}. It might be
indefinite which of the two it picks out—but a clear mistake to think it picks out some
special entity, a “vague set,” indefinitely identical to both. Indeterminacy in reference
is a well-known escape route from the Evans reductio argument (cf. Lewis 1988). If this
line succeeds, no retreat from classical conception of sharply bounded sets would be
needed in order to appeal to underpin their use in semantics.
However, even if textbook semantics were fine, it wouldn’t explain the two puzzles
of the previous section: borderlineness and the sorites. In the next three sections I
present examples of three kinds of positive accounts of vagueness and how they
tackle the puzzles identified above. I will start with the one that requires least modi-
fication of standard logic and semantics—epistemicism. Each is successively more
revisionary.

Classical Vagueness: Epistemicism

If we stick with classical logic and semantics for vague language, a consequence is that
room for maneuver in explaining the puzzles of borderlineness and the sorites paradox
is drastically reduced. Classicists cannot, for example, say that borderline-bald Harry
is special because it is neither true nor false that he is bald—for this would conflict
with the classical commitment to bivalence (every claim being either true or false). And
unless we want to deny that Jerry Garcia in his prime was bald, and Kojak was not, then
it follows, classically, that in the baldness sorites series, the last nonbald man differs by
only a hair from a man who is bald.

145
 J . RO BERT G . W I L L I A MS

The last claim seems particularly unpalatable—it seems to commit us, incredibly,
to sharp boundaries to baldness. Some endorsing the classical logic/semantics pack-
age argue that no such commitment follows (McGee and McLaughlin 1994), but the
theorists I discuss in this section take the hard-nosed stance that there really are sharp
boundaries for vague predicates. To make a case for this surprising claim, at a minimum
we need some explanation of how the characteristic features of vagueness are consistent
with sharp boundaries.
Epistemicists typically endorse the face-value classicism just described. It’s either true,
or false, that borderline-bald Harry is bald, for example—there is a fact of the matter—
but we do not and (in a certain sense) cannot know which option is realized. Below, I
sketch Williamson’s (1994) elaboration of this idea.
Let us consider borderlineness first. Start from the idea that we are ignorant of whether
Harry is bald, when he is a borderline case. The puzzle was to explain why this was so,
and why the unknowability was of such a strong and ineliminable sort.
Williamson’s proposal makes use of a general constraint on knowledge: the idea that
in order to know that p, it cannot be a matter of luck that one’s belief is that p is true.
Williamson articulates this as a “safety principle,” roughly:

(SAFETY)
For “S knows that p” to be true (in such situation s), “p” must be true in any
marginally different situation s* (where one forms the same beliefs using the
same methods) in which “S believes p” is true.

The idea is that the situations s* represent “easy possibilities”: falsity at an easy possibil-
ity makes a true belief too lucky to count as knowledge.
This first element of Williamson’s view is independently motivated epistemology.
The second element is that the extensions of vague predicates, though sharp, are unstable.
They depend on exact details of the patterns of use of vague predicates, and small shifts
in the latter can induce small shifts in the (sharp) boundaries of vague predicates.
Given these two, we can explain our ignorance in borderline cases. A borderline
case of “bald” is just one where the boundary of “bald” is close enough that a marginally
different pattern of usage could induce a switch from (say) Harry being a member of
the extension of “bald” to his not being in the extension of that predicate. If that’s the
case, then even if one truly believed that Harry was bald, there will be an easy possibility
where one forms the same beliefs for the same reasons, but that sentence is false. Apply-
ing (SAFETY), the belief won’t count as knowledge.
Given that the source of ignorance resides in the sharp but unstable boundaries of
vague predicates, one can see why gathering information about hair distributions won’t
overcome the relevant obstacle to knowledge. This is why the ignorance in borderline
cases seems ineliminable.
What does the epistemicist say about the sorites? Whether we present that argument
via a string of negated conjunctions, or conditionals, or variations and generalizations
thereof, one of the premises will be false: A certain man will be bald, while his neighbor
with just one more hair is not bald. The sorites argument is therefore unsound. But it’s
controversial whether this is enough to resolve our initial little-by-little puzzle. We’d
like to know why we found the idea of a sharp cutoff so incredible (especially since
there is available a very simple, valid argument from obvious premises to this effect).
Williamson can give an account of why we’d never feel able to accept—since we

146
 VA G U EN E S S

cannever know—any one of the individual conjunctions. But that doesn’t explain why
we’re uneasy (to say the least) with the thought that some such conjunction is true. An anal-
ogy: I’ll never know in advance which ticket will win a lottery; but I’m entirely comfort-
able with the thought that one will win. Why don’t we feel the same about the sorites?

Supervaluationism
A common reaction to borderline-bald Harry is that it’s neither true nor false that he
is bald. Assuming that one can only know what is true, this would explain our inevi-
table lack of knowledge in borderline cases. It’s often thought to be a rather plausible
suggestion in itself. Classical semantics builds in the principle that each meaningful
claim is either true or false. So if we’re to pursue the thought that borderline claims are
truth-value gaps, we must revise our semantic framework to some extent. Indeed, we
can know in advance that any semantic theory with truth-value gaps will diverge from
classical semantics even on some of the most intuitively plausible consequences: For it
can be shown under very weak assumptions that truth-value gaps are incompatible with
accepting disquotational principles such as: “Harry is bald” is true if and only if Harry is
bald (see Williamson 1994, ch. 7).
How will the alteration of the classical framework go? One suggestion goes under the
heading “supervaluationism” (see inter alia Fine 1975, Keefe 2000). As we’ll see, the
term is somewhat ambiguous. As an account of the nature of vagueness, supervaluation-
ism is a view on which borderlineness arises from what we might call “semantic indeci-
sion” (cf. Lewis 1993). Think of the sorts of things that might fix the meanings of words:
conventions to apply the word “bald” to clear cases; conventions to apply “not bald” to
clear noncases; various conventions of a more complex sort—for example, that anyone
with less hair than a bald person should count as bald. The idea is that when we list
these and other principles constraining correct interpretation, we’ll be able to narrow
down the space of acceptable (and entirely classical) interpretations of English—but not
to the single intended interpretation hypothesized by classical semantics. At best, what
we’ll get is a cluster of candidates. Let’s call these the sharpenings for English (sometimes
these are called “precisifications” or “delineations”). Each will assign to each vague
predicate a sharp boundary. But very plausibly the location of such a boundary is some-
thing the different sharpenings will disagree about. A sentence is indeterminate (and if
it involves a vague predicate, is a borderline case) just in case there’s a sharpening on
which it comes out true, and another on which it comes out false.
As an account of the semantics of vague language, the core of the supervaluationist
proposal is a generalization of the idea found in classical semantics, that for something
to be true is for it to be true at the intended interpretation. Supervaluationism offers a
replacement. It works with a set of “co-intended interpretations,” and says that for a
sentence to be true, it must be true at all the co-intended interpretations (this is sometimes
called “supertruth”). This dovetails nicely with the semantic indecision picture, since
we can take the “co-intended interpretations” to be what we called above the sharpen-
ings. When a sentence is indeterminate (true on one sharpening and false on another)
neither it nor its negation will be true; and hence we have a truth-value gap.
The slight tweak to the classical picture leaves a lot unchanged. Consider the tautol-
ogies of classical logic, for example. Every classical interpretation will make them true;
and so each sharpening is guaranteed to make them true. Hence, any classical tautol-
ogy is always supertrue, which is enough to make it a supervaluational tautology. (It’s a

147
 J . RO BERT G . W I L L I A MS

matter of dispute whether more subtle departures from classical logic are required, and
whether this matters. (Cf. Fine 1975, Williamson 1994 ch 5, Williams 2008.)
If (super)truth is a constraint on knowledge, supervaluationists can explain why we
can’t know whether borderline bald Harry is bald. On some developments of the posi-
tion, they can go interestingly beyond this. One might argue that insofar as one should
only invest credence in a claim to the extent one believes it true, obvious truth-value-
gaps are cases where we should utterly reject (invest no credence in) both the claim
and its negation (cf. Field 2003). This would mean the information that such-and-such
is borderline gives us a direct fix on what our degree of belief should be in borderline
cases.
What about the sorites? No individual conjunction “Man N is bald and man
N + 1 is not bald” will be true—many of them will instead be truth-value gaps, true on
some sharpenings and false on others. On the other hand, every sharpening can provide
a number n that makes “for some n, man n is bald and man n+1 is not bald” true—differ-
ent ones on different sharpenings—so it’s supertrue overall. This highlights one of the
distinctive (and perhaps disturbing) features of supervaluationism—that disjunctions
and existential generalizations can be true, even if no disjunct or instance is. Either
way, not all the premises of the premises of the argument for paradox will be true, so the
argument is blocked. Just as for the epistemicist, there is a further question about what
accounts for the argument’s original appeal.
Two points to bear in mind about supervaluationism. First, as we’ve seen, the super-
valuationist endorses the claim that “there is a cutoff for bald”—a pair of men differing
by only one hair, with the first bald and the second not. The supervaluationist must
try to persuade you that once you understand the sense in which there is no fact of the
matter where that cutoff is, you will be prepared to accept the existence of cutoffs. Sec-
ond, many want to press the charge that the supervaluationist makes no progress over
the classicist, for reasons of “higher order vagueness.” The thought is that the task of
explaining how a set of sharpenings gets selected by the meaning-fixing-facts is no easier
or harder than explaining how a single classical interpretation gets picked out. However,
(a) the supervaluationist can reasonably argue that if she spells out the notion of “sharp-
ening” in a vague metalanguage, she will regard the boundary between the sharpenings
and nonsharpenings as vague (see Keefe 2000); (b) even if both epistemicist and super-
valuationist were both in some sense “committed to sharp boundaries,” the account
they give of the nature of vagueness is vastly different, and we can evaluate their posi-
tive views on their own merits.

Many-Valued Settings
A distinctive feature of supervaluationism was that while it threw out bivalence (“Harry
is bald” is either true or false) it preserved the corresponding instance of excluded middle
(“Harry is bald or Harry is not bald”). Revising logic and semantics in a more thorough-
going way would allow for a coherent picture where we can finally reject the claim
“there is a single hair that makes the difference between bald and nonbald” without
falling into paradox.
“Many valued” logics can be characterized by increasing the number of truth-values we
work with—perhaps to three, perhaps infinitely many—and offering generalizations of
the familiar stories of how logical constants behave to accommodate this tweak. There
are many ways to develop this (for a survey, see “Many-Valued Logic,” this volume).

148
 VA G U EN E S S

Here is a sample many-valued logic, for a propositional language with conjunctions,

disjunctions, and negations. To characterize the logic, we postulate three values: call
them, neutrally, “1” ,“0.5,” and “0.” For the propositional case, each atomic sentence
will be assigned one of these truth-values. The truth-values then get assigned to complex
sentences recursively. Let us assume a conjunction will be assigned the minimum of the
truth-values of its conjuncts; a disjunction will get the maximum of the truth-values of its
disjuncts; and a negation will be 1 minus the truth-value of the claim negated (you can
easily check that, ignoring the value 0.5 altogether we get back classical truth-tables).
One option at this point is to take this model theory seriously—much as the classicist
and supervaluationist do—and hypothesize that natural language has (or is modeled by)
some many-valued interpretation (or set of interpretations). The most famous version
of this many-valued approach, the “degree theory” of vagueness, works not with the
three values above but with infinitely many, labeled by the real numbers between 1 and
0. The truth-values are thought of as “degrees of truth”—with a sentence getting closer
to “perfect truth” insofar as its truth-value approaches 1. (Cf. inter alia Machina 1976,
Smith 2008.)
Degree theorists will say that a borderline case of a vague predicate has an inter-
mediate truth-value. Harry is bald is neither wholly true, nor wholly false—and this is
the reason why we are uncomfortable giving either answer. Along a sorites series, the
degree of truth of “Man N is bald” falls off smoothly: for N = 0 this may be value 1; for
large N it is a value close to 0. What one says about the sorites argument itself—the
truth-values of the compound claims involved, and the validity of the argument, will
depend on exactly what story is given about the behavior of the connectives and the
characterization of the logic.
Worries about “higher order” vagueness are more pressing when directed against the
degree theorist rather than against the supervaluationist. On the degree-theoretic story,
it appears that Harry will be bald to some particular, precise degree—say 0.5456. But is
it really plausible that there are such sharp facts of the matter about the exact degree to
which Harry is bald? The “higher order indeterminacy” claim is that it is indeterminate
to what degree Harry is bald. But if this higher-order indeterminacy is handled in a
many-valued manner, then the language in which we ascribe degrees of truth to sen-
tences—and hence theorize about degree theory itself—is non-classical. It’s then not
clear that the kind of claims we made earlier to characterize the position, e.g. “every
atomic sentence has a degree of truth,” are fully true. Many-valued model theory is usu-
ally described in a metalanguage tacitly assumed to be classical. It’s far from clear this is
dispensable (cf. Williamson 1994 ch. 4).
Some claim many-valued theories give counterintuitive results. Take a 0.5-valued
sentence A. From the recursive clauses, we can see that “A&~A” will be half-true as
well. Many find this commitment objectionable. (See Williamson 1994 ch. 4 for argu-
ments that all many-valued treatments of the connectives will be similarly problem-
atic.) Degree-theorists don’t agree, and standoff threatens. It’s worth noting that this
objection relies on a particular “degrees of truth” gloss on the many-valued semantics.
But in a three-valued setting, rather than talk of degrees of truth, one could interpret
value 1 as “true,” 0 as “false,” and 0.5 as “neither true nor false” (cf. Parsons 2000). On
this reading, “A&~A” will be described as a truth-value gap, when “A” itself has this
status—which sounds rather different (and more plausible) than describing it as half
true. The opponent of many-valued approaches must check that their objections don’t
depend on a contentious gloss on the targeted view.

149
 J . RO BERT G . W I L L I A MS

We’ve concentrated till now on accounts that take model theory seriously—and which
look for an integrated account of the logic and semantics of vague language. But there’s
another option—to focus attention on the logic, and regard model theory as simply a
descriptive instrument we use to specify the One True Logic.
Logic-first theorists advocate a revisionary take on what arguments are valid and what
combinations of claims or attitudes are consistent. To fix ideas let’s focus on an example,
using the 3-valued assignments described earlier. Recall that a sequent is valid on the
strong Kleene logic if it is “1-preserving” across these assignments, i.e. if it is such that
when all the premises are value 1, the conclusion is value 1 too. In this Kleene logic,
“Av~A” is no longer a tautology, since it need not be value 1. On the other hand,
“A&~A” is still treated as a logical contradiction (every sentence whatsoever follows
from it), since it will never attain value 1, no matter what value A is assigned.
For the Assignment-first theorists we have been considering thus far, the interesting
question was about the various assignments and what they represent. Does the middle
status represent a truth-value gap or a degree of truth? Among the variety of assignments
compatible with the rules, which matches up to the actual distribution of truth statuses?
But for a logic-first theorist, such questions are misguided. The assignments are simply
mappings from sentences to numbers—there’s no need to think that they latch on to
real properties of the sentences, and so no sense in asking about what they represent
or which one among them is “intended”—all they do is collectively characterize a
Kleene-style logic.
A logic-first theorist such as Field (2003a) then uses the logic to handle the dis-
tinctive puzzles of vagueness. Field is committed to a certain normative significance for
logic—that accepting the premise of a valid argument commits one to accepting its
conclusion; and rejecting the conclusion commits one to rejecting its premise. Note
that the Kleene logic does not make excluded middle a tautology, so on Field’s view
there’s no logical reason to accept instances of it. Field advocates rejecting the claim
that “Harry is bald or he isn’t” when Harry is borderline-bald. Since this disjunction
follows, in the logic he endorses, from “Harry is bald,” he rejects this claim too, as well
as its negation. This then is Field’s account of the puzzle of borderlineness: Both direct
answers to “Is Harry bald?” are to be rejected.
Note that this story would collapse if to reject a claim is simply to accept its nega-
tion—we must be able to reject “Harry is bald” without accepting “Harry is not bald.”
That situation can get more dramatic. In the Kleene logic, the negation of excluded
middle “~(Av~A)” is inconsistent (and so must never be accepted). But we are sup-
posed to reject instances of Av~A. So a basic feature of the position is that sometimes
we reject claims whose negations are inconsistent.
One can view the sorites paradox as repeating this pattern on a grander scale.
Instances of “Man N is bald and Man N+1 isn’t” do indeed sound horrible, and we
should reject them. But the sorites paradox does not get started unless we move from
this to endorsing their negations. What the sorites paradox reveals is that the nega-
tions generate a contradiction. But just as with a single instance of excluded middle,
there’s no obligation to move from rejecting the claim to endorsing its (problematic)
negation.
Clearly, the logic-first version of many-valuism is very different from the assignment-
first version. Note, for example, that the higher-order indeterminacy worries mentioned
earlier simply do not get started against the logic-first theorists (at least if they endorse
the instrumental view of model theory). If one’s account has no place for “the intended

150
 VA G U EN E S S

truth-value assignment,” then opponents can hardly ask embarrassing questions about
whether it’s determinate which assignment this is!
On the other hand, the logic-first position is radical. It gives up on the attractive
prospect of a model-theoretic analysis and explanation of validity. And it seems commit-
ted to doing without model theory when directly addressing questions about the seman-
tics and truth status of vague language (Field, for example, endorses a nonsemantic,
disquotational treatment of truth and meaning).

Conclusion
I’ve outlined three sample positive views of vagueness. They are illustrative of broad cat-
egories of theories of vagueness. The categories are not exhaustive, but they do orientate
much contemporary debate.
The most conservative category–represented here by epistemicism–preserves clas-
sical logic, model theory and textbook semantics. The distinctive epistemicist response
is one way of explaining what vagueness could be, even if the apparent commitments
of this setting to “sharp cutoffs” are taken at face value. But this isn’t the only way a
classicist might respond. Fara (2003), for example, agrees with Williamson on the broad
classical setting. But her account of what vagueness is—at least for the cases she focuses
on, gradable adjectives—is very different. Her idea is that (i) the word “red” means
something like significantly redder than is typical; (ii) what counts as “significant” is deeply
interest-relative, and in particular, can vary depending on the focus of our attention;
(iii) as a result of this, we can predict that whenever we focus on a particular potential
“cutoff” for red, the facts about what is now significant to us ensure that the classical
cutoff for “red” is located somewhere other than we are looking. Hence, for Fara, the
distinctive elusiveness of vague predicates. (The idea that context-sensitivity might be
intimately involved with the phenomena of vagueness is something many writers find
appealing, whether or not they work in a classical backdrop (see Raffman 1994, Shapiro
2006).)
The next most conservative response is to design a “special purpose” semantics for
vagueness, allowing semantic features to do significant explanatory work, which nev-
ertheless “saves much of the appearance” of the standard classical setting. Supervalu-
ationism illustrates this possibility. But there are other ways to go, even with the broad
“semantic indecision” framework that gives us our notion of a sharpening. Various authors
have suggested that we might look at the proportion of sharpenings on which a given
claim holds. In terms of this, we can characterize a notion of intermediate degrees of truth:
the degree of truth of “Harry is bald” being the proportion of sharpenings on which it is
true. Thus supertrue sentences (“Kojak is bald”) will be true to degree 1; while if “Harry
is bald” is plumb borderline, true on half and false on half, then it’ll be true to degree
0.5 (cf. Edgington 1997). Such notions may find application within a treatment of com-
parative forms of vague adjectives and modifiers (see Kamp 1975, Lewis 1970).
The final theoretical framework we looked at was thoroughly revisionary, though
the revisionism has more or less radical forms. Preserving realism about model-
theoretic foundations, we find “gap” and “degree” theories. Alternatively, looking at
model theory as a mere descriptive instrument for specifying a logic, we have logic-first
theories. There’s plenty of room for debate about which logic is most appropriate: For
example, Crispin Wright (2003, 2007) argues for intuitionism as the appropriate logic
governing vague discourse. (Wright has been a long-time advocate of drawing lessons

151
 J . RO BERT G . W I L L I A MS

on the nature of semantic theory from careful study of the phenomenon of vagueness–
see Wright (2007).)
Important though these logico-semantic issues are, one should not think that the
whole task of understanding vagueness consists in figuring out the proper seman-
tic treatment of it. As is evident from the case of epistemicism, resources outside the
philosophy of language may be key to diagnosing its characteristic puzzles. Moreover,
logico-semantics may underdetermine the overall theory of vagueness. Getting clear on
the nature of vagueness/indeterminacy is another aspect to theorizing about vagueness
–for example is it a matter of “semantic indecision” as many supervaluationists contend?
Similar formal machinery can be paired with very different responses to this question.
As well as the nature, psychological features demands attention. Should we have zero cre-
dence in Harry being bald, if we know he’s a borderline case (as Field argues)? Or should
have fifty-fifty confidence, as some degree theorists maintain? (Cf. Smith 2008, Schiffer
2003.) Or something else entirely (Wright 2003)?
The attention lavished on vagueness since the mid-1970s shows little signs of con-
verging on a single standard account. But it has given us plenty of options, providing
rich resources for future arguments. To study it is to appreciate the far-reaching impact
of the disarmingly simple puzzles of vagueness.

Bibliography
D. Edgington. (1997) “Vagueness by degrees,” in Vagueness: A Reader, eds. Rosanna Keefe and Peter Smith
(Cambridge, MA: MIT Press), 294–316.
G. Evans. (1978) “Can there be vague objects?” Analysis 38, no. 13: 208.
D. Graff Fara. (2003) “Shifting sands: an interest-relative theory of vagueness,” Philosophical Topics 28, no.
1: 45–81.
H. H. Field. (2003) “Semantic paradoxes and the paradoxes of vagueness,” in Liars and Heaps: New essays on
paradox, ed. J. C. Beall (Oxford: Oxford University Press), 262–311.
H. H. Field. (2003a) “No fact of the matter,” Australasian Journal of Philosophy 81, no. 4: 457–480.
K. Fine. (1975) “Vagueness, truth and logic,” Synthese 30: 265–300.
J. A. W. Kamp. (1975) “Two theories about adjectives,” in Formal Semantics of Natural Language, ed. E.
Keenan (Cambridge: Cambridge University Press), 123–55.
R. Keefe. (2000) Theories of Vagueness (Cambridge: Cambridge University Press).
D. K. Lewis. (1970) “General semantics,” Synthese 22: 18–67.
—— (1973) Counterfactuals (Oxford: Blackwell).
—— (1988) “Vague Identity: Evans misunderstood,” Analysis 48: 128–30.
—— (1993) “Many, but almost one,” in Ontology, Causality and Mind: Essays on the philosophy of D. M.
Armstrong, ed. Keith Campbell, John Bacon, and Lloyd Reinhardt (Cambridge: Cambridge University
Press).
K. F. Machina. (1976) “Truth, belief and vagueness,” Journal of Philosophical Logic 5: 47–78.
Vann McGee and Brian McLaughlin. (1994) “Distinctions without a difference,” Southern Journal of Philoso-
phy supp XXXII: 203–251.
T. Parsons. (2000) Indeterminate Identity: Metaphysics and semantics. New York: Oxford University Press.
D. Raffman. (1994) “Vagueness without Paradox,” Philosophical Review 103.
N. U Salmon. (1981) Reference and Essence (Princeton: Princeton University Press).
S. Schiffer. (2003) The Things We Mean (New York: Oxford University Press).
N. J. J. Smith. (2008) Vagueness and Degrees of Truth (Oxford: Oxford University Press).
J. R. G. Williams. (2008) “Supervaluationism and logical revisionism,” The Journal of Philosophy CV, no. 4.
T. Williamson. (1994) Vagueness (London: Routledge).
C. Wright. (2003) “`Vagueness: A fifth column approach’,” in Liars and Heaps: New essays on paradox, ed. J.
C. Beall (Oxford: Oxford University Press), 84–105.
—— (2007) “Wang’s paradox” in The Philosophy of Michael Dummett, eds. Randall E. Auxier and Lewis
Edwin Hahn (Peterborough, NH: Open Court).

152
 1 . 13
E MP TY NAM ES
Sarah Sawyer

1 Introduction
An empty name is a name that fails to name an individual—a name that fails to refer.
Fiction, myth, and false theory provide plausible examples. Thus, on the face of it,
‘Sherlock Holmes,’ ‘Pegasus,’ and ‘Vulcan’ fail to refer precisely because there is no such
detective as Sherlock Holmes, there is no winged horse Pegasus, and there is no planet
Vulcan whose orbit lies between Mercury and the Sun. Nonetheless, some sentences
containing empty names can be understood, and hence appear to be meaningful—for
example, ‘Pegasus was sired by Poseidon’ and ‘Vulcan was postulated by Le Verrier’;
some empty names appear to name the same individual and hence to be co-referen-
tial—for example, ‘Santa Claus’ and ‘Father Christmas’; and some sentences containing
empty names appear to be straightforwardly true—for example, ‘Sherlock Holmes is a
character from fiction,’ and ‘Pegasus does not exist.’ The expression ‘straightforwardly
true’ is intended to contrast ‘true according to the fiction/myth/failed theory,’ which
require independent treatment. Thus ‘Sherlock Holmes played the violin’ is, on the
face of it, true according to the fiction but not straightforwardly true, whereas ‘Sherlock
Holmes is a character from fiction’ is, on the face of it, straightforwardly true but not
true according to the fiction. A semantic theory of names must make sense of the appar-
ent phenomena of meaning, reference, and truth in the specific context of sentences
containing names that fail to refer (either by accommodating the relevant intuitions or
by explaining them away).
Of course, the question of which names, if any, are empty cannot be answered by
philosophy of language alone. Rather, a semantic theory of names must be provided
without prejudging the answer to that question. A semantic theory of names should
deliver a specification of the conditions under which a name names an individual, and
hence a specification of the conditions under which a name is empty; but the question
of whether those conditions are met on any given occasion of use plausibly lies outside
the scope of a semantic theory. Having said that, there are those who have attempted to
make sense of meaning, reference and truth in the context of prima facie empty names
by postulating referents for those names. This strategy aims to eliminate the problems
surrounding empty names by eradicating (or reducing to an insignificant minimum)
the category of empty names. Terence Parsons has defended the claim that the refer-
ents of apparently empty names are nonexistents (Parsons 1980); Nathan Salmon and
Amie Thomasson have argued that they exist as abstracta (Salmon 1987; Salmon 1998;
Salmon 2002; Thomasson 1998). In what follows I will provide an overview of semantic
 SA RA H SAW Y E R

theories which address the need to make sense of the apparent phenomena of meaning,
reference, and truth in the specific context of sentences containing empty names.

2 Fregean Sense
One may wonder why exactly empty names are thought to be problematic. The problem
is strikingly clear, however, if set against a certain intuitive understanding of how names
and predicates function. Intuitively, the function of a name is to pick out (refer to) an
individual, and the function of a predicate is to pick out (refer to) a property. If this is
right, it is natural to think: first, that a sentence composed of a name and a predicate is true
just in case the object referred to by the name has the property referred to by the predicate
(which presupposes that there is an object named); and, second, that a subject can under-
stand a sentence composed of a name and a predicate only if she knows which object is
referred to by the name and which property is referred to by the predicate (which, again,
presupposes that there is an object named). If this basic understanding of the function of
names and predicates is correct, empty names are essentially defective, and we are bound
to conclude: first, that sentences containing empty names can be neither true nor false
(since there is no object which has or lacks the property referred to by the predicate); and
second, that sentences containing empty names cannot be understood (since one cannot
know which object is referred to by a name that refers to no object).
Frege’s move from an extensional semantic theory akin to the one described above
(a theory that countenances only linguistic expressions and their referents) to an inten-
sional semantic theory that invokes in addition a notion of sense, embodies a rejection
of the intuitive understanding of how names and predicates function. According to
Frege, every linguistic expression expresses a sense, which is objective and is grasped
by anyone who understands the expression. The sense of an expression determines the
referent of that expression, but, crucially, the sense of an expression is available to be
thought of independently of whether the expression in fact refers. Consequently, Frege’s
theory provides an account of how sentences that contain empty names can have mean-
ing and be understood. In order to understand a sentence composed of a name and a
predicate, one need simply grasp the sense of the sentence (the thought expressed),
and this will be available even in the absence of a referent for the name. (It is typically
assumed that the Fregean sense of a name must be descriptive. However, I think there is
good reason to think of the sense of a name as nondescriptive. For one thing, the claim
that names have descriptive senses conflicts with Frege’s claim that sense is transparent:
It is, after all, possible for a subject rationally to assent to a sentence composed of a name
and a predicate, and simultaneously withhold assent from the sentence had by replacing
that name with any description that allegedly expresses the same sense.)
The problems of apparent co-reference and of truth, however, remain. The fact that
‘Santa Claus’ and ‘Father Christmas’ appear to be co-referential cannot be explained by
appeal to the senses of the expressions, since the senses of the expressions will differ.
And sentences such as ‘Pegasus does not exist,’ which contain a name that fails to refer,
will be neither true nor false, since, according to Frege, the truth-value of a sentence is
its referent, and the referent of a complex expression is determined by the referents of its
parts. Thus, if a part of a complex expression lacks a referent, so does the whole, which
in the case of a sentence means that the sentence has no truth-value.
Frege was not primarily concerned with ordinary language. He remarks that it is a
fault of languages that they contain expressions that fail to refer, and maintains that in

154
 EM PTY N A ME S

a logically perfect language no name would be introduced without being provided with
a referent (Frege 1892: 69–70). However, the problems raised by empty names are prob-
lems that must be dealt with satisfactorily if an adequate semantics for ordinary language
is to be found. As such, either Frege’s remarks must be supplemented so as to account for
our intuitions about reference and truth, or we must look to a different semantic theory
for an answer.
Gareth Evans and John McDowell advocate a Fregean theory according to which the
sense of a singular term, such as a name or a demonstrative, is object-dependent, or de
re (Evans 1982; McDowell 1977; McDowell 1984). On this view, expressions that lack
referents thereby lack sense: Consequently, a sentence containing an empty name fails
to express a thought. Rather than helping matters, this appears to leave all three prob-
lems surrounding empty names unresolved—those of meaning, reference, and truth.

3 Russellian Descriptions
Russell became suspicious of intensional notions such as Fregean sense (Russell 1905),
and sought instead to solve the problems raised by empty names within a semantic the-
ory that once again countenanced only linguistic expressions and their referents. Never-
theless, Russell also rejected the intuitive understanding of how names function. Thus,
according to Russell, names, such as ‘Bertrand Russell’ and ‘Vulcan,’ do not express a
Fregean sense, nor do they function to pick out an object directly. Rather, according to
Russell, names are disguised definite descriptions of the form ‘the F.’ Thus, for example,
the name ‘Bertrand Russell’ is taken to be semantically equivalent to a definite descrip-
tion such as ‘the male philosopher who was born on May 18,1872 and wrote “On Denot-
ing” ’; similarly, the name ‘Vulcan’ is taken to be semantically equivalent to a definite
description such as ‘the planet whose orbit lies between Mercury and the Sun.’
Further, and crucially, Russell rejected the intuitive understanding of definite descrip-
tions as singular terms that function to pick out an object directly (which would leave
the problems concerning empty names unresolved). Instead, he argued, definite descrip-
tions such as ‘The F’ are quantificational phrases which fall into the same broad seman-
tic category as expressions such as ‘Some F’ and ‘Every F.’ On this view, the function of
a name, understood as a definite description, in turn understood as a quantificational
phrase, is not to contribute an object to propositions expressed by sentences containing
it: rather, the function of a name is to contribute a complex of properties (Russell 1905;
Russell 1911; Russell 1918). The beauty of this account lies in the fact that a name has
purely extensional semantic content which is nonetheless available whether or not it is
the name of an individual. A name is empty in the relevant sense if and only if there is
no object that uniquely satisfies its descriptive content. Thus ‘Bertrand Russell’ is not
an empty name because there is in fact just one male philosopher who was born on May
18, 1872 and wrote “On Denoting”; whereas ‘Vulcan’ is an empty name because there is
no planet whose orbit lies between Mercury and the Sun.
Russell’s account offers a solution not only to the problem of meaning (and hence
understanding) for sentences containing empty names but also to the problem of truth.
Thus, taking the relevant names to be equivalent to the definite descriptions suggested
above, the sentence ‘Bertrand Russell was a pacifist’ is to be analyzed as ‘There is exactly
one male philosopher who was born on May 18, 1872 and wrote “On Denoting” and he
was a pacifist,’ which is true in virtue of the fact that there is such an individual and he
was a pacifist. In like fashion, the sentence ‘Vulcan was discovered by Le Verrier’ is to

155
 SA RA H SAW Y E R

be analyzed as ‘There is exactly one planet whose orbit lies between Mercury and the
Sun and it was discovered by Le Verrier,’ which is false in virtue of the fact that there
is no such planet. Finally, the sentence ‘Vulcan does not exist’ is to be analyzed as ‘It is
not the case that there is exactly one planet whose orbit lies between Mercury and the
Sun,’ which, in accord with intuition, is straightforwardly true.
The proposed solution to the problem of truth is not unproblematic, however. Rus-
sell’s account implies that all sentences composed of an empty name and a predicate are
false, including: ‘Pegasus is a mythical creature’ and ‘Venus was incorrectly postulated
by Le Verrier.’ Russell was concerned to preserve the law of excluded middle, accord-
ing to which every sentence is either true or false. However, it is not obvious that the
law of excluded middle ought to be preserved, and the particular way in which Russell’s
account preserves it leaves us with unintuitive assignments of truth-values, as indicated.
Moreover, once again the problem of apparent co-reference remains.
Russell’s account of definite descriptions as quantificational phrases has gained a
reasonable degree of support. (Although see Strawson 1950; Donnellan 1966. For a
defence of Russell’s account of definite descriptions see Neale 1990.) His claim that
names are disguised definite descriptions, in contrast, has been largely discredited by
Kripke’s arguments to the effect that names and definite descriptions have different
modal profiles (Kripke 1980). For instance, it is false that Bertrand Russell might not
have been Bertrand Russell (although he might of course have had a different name);
but it is true that Bertrand Russell might not have been the male philosopher who was
born on May 18, 1872 and wrote “On Denoting” (since Russell might not have become
a philosopher). As a concession to the facts about the different modal profiles of names
and definite descriptions, it has been suggested that a Russellian account of names might
nonetheless be advocated specifically for names that are empty. The suggestion here is
that names that name an individual be treated as directly referring expressions that pick
out an individual directly, while empty names be treated as having descriptive content
(e.g. Currie 1990). There is some plausibility to this idea, since (one might think) empty
names are introduced by allegedly reference-fixing descriptions from which they never
manage to break free. The main concern is that such an account would not provide a
unified semantic account of proper names, since the semantics of a name would then
depend on whether or not it refers.

4 Millianism
Qualms about intensional notions, together with the realization that names do not in
general have descriptive content, has led to a striking consensus that names are directly
referring expressions. This is effectively the intuitive account with which we started,
according to which the function of a name is simply to pick out (refer to) an individual.
One influential kind of direct reference theory is Millianism, according to which the
meaning of a name is the object named (Salmon 1986; Soames 1987; Soames 2002).
Millians often maintain that a sentence composed of a name and a predicate expresses
a structured proposition which contains the object referred to by the name and the
property referred to by the predicate; and that a sentence containing an empty name
expresses a ‘gappy’ proposition—a proposition that has a gap where an object would
be if the name were not empty. Consequently, the view runs headlong into the three
problems of empty names with which we started. Different attempts to solve the prob-
lems have emerged consistent with both the claim that a name is a directly referring

156
 EM PTY N A ME S

expression and the claim that a sentence containing an empty name expresses a gappy
proposition. I will look at three primary attempts here. I leave aside the view proposed
by Salmon and Thomasson (Salmon 1998; Salmon 2002; Thomasson 1998). This view
avoids the problems of empty names by treating names from fiction, myth, and failed sci-
entific theories as referring to abstract artefacts, and hence as nonempty. I have argued
against such a view elsewhere (Sawyer 2002). Salmon does maintain that in rare cases a
name can be empty, but such cases are so rare as to be relatively insignificant. The view
proposed by Parsons (Parsons 1980), according to which apparent empty names refer to
nonexistent entities, is discussed below.
The first approach I will discuss here is developed by David Braun (Braun 1993;
Braun 2005). According to Braun, an atomic sentence composed of a name and a predi-
cate expresses a proposition which is true if and only if there is an object referred to by
the name and it has the property referred to by the predicate; and propositions which
are not true are false. Thus every atomic sentence containing an empty name expresses a
gappy proposition which is false: for example, ‘Sherlock Holmes is a detective,’ ‘Pegasus
was sired by Poseidon,’ and ‘Vulcan was postulated by Le Verrier.’ One apparent benefit
of Braun’s view is that sentences of the form ‘a exists’ are also to be treated as atomic
sentences of subject–predicate form. As a result—and in accord with intuition—
sentences such as ‘Sherlock Holmes exists,’ ‘Pegasus exists,’ and ‘Vulcan exists’ express
false propositions, while their negations—‘Sherlock Holmes does not exist,’ ‘Pegasus
does not exist,’ and ‘Vulcan does not exist’—express true propositions.
Less intuitive are the following consequences of the view. First, sentences that differ
only in respect of the empty name they contain express the very same proposition: Thus,
‘If Pegasus exists, then there is a planet between Mercury and the Sun’ and ‘If Vulcan
exists, then there is a planet between Mercury and the Sun’ both express the same, false,
gappy proposition. This goes against an intuition concerning truth—that the former is
false, while the latter is not—and against two related intuitions concerning meaning—
namely, that the sentences differ in meaning, and hence that the names have meaning.
Second, atomic propositions and their predicate-negations express propositions both of
which are false: Thus ‘Pegasus has wings’ and ‘Pegasus does not have wings’ both express
false propositions, which again goes against an intuition concerning truth—this time
that the sentences differ in truth-value (or, perhaps, are truth-valueless).
Braun aims to accommodate the relevant intuitions concerning meaning and truth
by appeal to cognitive facts—the semantic facts alone being unable to accommodate
them. According to Braun, there are different ways in which one can believe or dis-
believe a proposition; and each of these ways corresponds to a different mental state,
which can play a different role in one’s cognitive life. Thus a single proposition, whether
gappy or not, can provide the content of numerous different mental states if that propo-
sition is believed and/or disbelieved in different ways. So how does the appeal to ways
of believing a proposition accommodate the intuitions about meaning and truth?
According to Braun, the (false) intuition that ‘If Vulcan exists, then there is a planet
between Mercury and the Sun’ is true, whereas ‘If Pegasus exists, then there is a planet
between Mercury and the Sun’ is false, is to be explained by the fact that there is one,
false, gappy proposition that we both believe in a ‘Vulcan-ish’ way and disbelieve in a
‘Pegasus-ish’ way. Similarly, the (false) intuition that ‘Pegasus has wings’ and ‘Pegasus
does not have wings’ differ in truth-value is to be explained by the fact that we believe
the false, gappy proposition expressed by the first sentence but disbelieve the false gappy
proposition expressed by the second. Intuitions about differences in meaning are also

157
 SA RA H SAW Y E R

to be explained by appeal to different ways of believing. Since we necessarily believe

a proposition in some particular way, false intuitions about differences in meaning can
be seen as mapped onto—and hence as being sensitive to—real differences in ways of
believing.
The success of this strategy depends upon the viability of the crucial notion of ways of
believing a proposition. Appeal to ways of believing (or something similar) is prevalent
among Millians as a result of the felt need to explain how a subject can take differing
cognitive attitudes toward one and the same proposition. Thus, for example, although
‘Cary Grant is an actor’ and ‘Archie Leach is an actor’ express the same proposition
(since ‘Cary Grant’ and ‘Archie Leach’ are co-referential, directly referring expressions),
it is possible for a rational subject to believe what the first says and not believe what the
second says. For the sake of argument I assume here that this strategy is plausible in the
case of names that refer. However, even if an appeal to ways of believing were legiti-
mate in cases where the relevant names refer, cases where the relevant names are empty
cannot be assumed to be analogous. After all, if the name ‘Vulcan’ lacks meaning, then
plausibly the expression ‘Vulcan-ish’—which on the face of it is a parasitic term—also
lacks meaning: Consequently, there is no clear meaning to the theoretical expression
‘believing in a “Vulcan-ish” way.’ One might think that ‘believing in a “Vulcan-ish”
way’ could be understood metalinguistically, as equivalent to ‘believing by means of
the name “Vulcan’’ ’ , but this won’t do since many different things can share a name.
‘Vulcan’ is the name of a hypothetical planet, a Roman god, a colleague’s pet tortoise,
and more besides. This casts doubt on the general strategy employed, which in turn casts
doubt on the ability of Braun’s account to solve the problems of meaning and truth.
Moreover, even if the notion of ways of believing can be made to work, it is unclear how
it would solve the problem of apparent co-reference.
The second approach has been developed by Fred Adams, Gary Fuller, and Robert
Stecker (Adams et al. 1994; Adams et al. 1997; Adams and Fuller 2007. See also Taylor
2000; Soames 2002). On this view, all gappy propositions expressed by sentences that
contain empty names simply lack truth-value; and intuitions concerning meaning, ref-
erence, and truth are to be explained by appeal to the descriptive information that those
sentences pragmatically convey. More specifically, the introduction of an empty name
into the language (whether intentionally or unintentionally) occurs within the context
of a set of descriptions which are intended to provide information about an individual
bearing that name. These descriptions are then pragmatically conveyed by subsequent
uses of the name, although the descriptions are never part of the meaning of the name.
Fictional names are associated with descriptions from the relevant fiction; mythical
names are associated with descriptions from the relevant myths; and names from failed
scientific theories are associated with descriptions from the relevant scientific theory.
So how does the notion of descriptive information pragmatically conveyed accom-
modate the intuitions about meaning, reference, and truth? According to Adams, Fuller,
and Stecker, since the descriptions pragmatically conveyed by ‘Vulcan’ are different
from those pragmatically conveyed by ‘Pegasus,’ it seems to us that the names differ in
meaning. The names ‘Santa Claus’ and ‘Father Christmas,’ in contrast, plausibly prag-
matically convey the same set of descriptions. They say: “There are two ways that a set
of descriptions may come to be associated with more than one name. An author may
make the associations explicit in the text. . . . Also, an association may be arrived at
as an inference to the best explanation” (Adams et al. 1997: 132). Further, despite the
fact that ‘Pegasus exists’ and ‘Pegasus does not exist’ both lack truth-value, according

158
 EM PTY N A ME S

to Adams, Fuller, and Stecker, we think the former is false and the latter is true because
the former pragmatically conveys the information that a winged horse exists, which is
false, whereas the latter pragmatically conveys the information that a winged horse does
not exist, which is true. Our intuitions concerning meaning, reference, and truth, then,
are explained by our mistaking the information a sentence pragmatically conveys for the
proposition it expresses.
There are at least four problems with the view. (For the first, second, and third see
Everett 2003; for the fourth see Reimer 2001.) First, the difference in modal profile
between names and definite descriptions is equally evident between empty names and
the sets of descriptions they pragmatically convey. Thus ‘Pegasus is Pegasus’ is necessar-
ily true, whereas ‘Pegasus helped Bellerophon in his fight against the Chimera’ is not.
Consequently, the modal profile of a sentence containing an empty name cannot be
explained by appeal to the pragmatic information it conveys; nor can it be explained
by appeal to the proposition expressed, which is gappy and necessarily lacking in truth-
value. Second, if we are prone to mistake the information a sentence pragmatically con-
veys for the proposition it expresses in the case of sentences containing empty names,
then we ought to be prone to the very same mistake in the case of sentences that contain
names that refer, but clearly we are not. Kripke’s arguments to the effect that names are
not equivalent to definite descriptions depends upon us not being subject to this kind of
confusion. Third, intuition favors the thought that the meaning of a sentence contain-
ing an empty name is determinate, whereas the descriptions associated with any given
name will vary across speakers and times. And fourth, the relevant semantic intuitions
about empty names persist even once we are offered a pragmatic explanation of them.
Successful pragmatic explanations of semantic intuitions, however, tend to rid us of the
initial intuitions they are invoked to explain. This indicates that the pragmatic expla-
nation offered by Adams, Fuller, and Stecker is not after all a genuine explanation of the
apparent phenomena of meaning, reference ,and truth surrounding empty names.
The fundamental flaw in the pragmatic approach is that it loses sight of the intuition
that an empty name is a name: that it is a directly referring expression that allows us
to talk about an object independently of the object’s satisfying any descriptions associ-
ated with it. No doubt there are descriptive propositions pragmatically conveyed by a
sentence containing an empty name—just as there are descriptive propositions prag-
matically conveyed by a sentence containing a non-empty name—but appeal to what
is pragmatically conveyed cannot do justice to our intuitions about meaning, reference,
and truth.
The third, rather different, approach is advocated by Marga Reimer (Reimer 2001).
According to Reimer, the intuitions concerning meaning, reference, and truth are all to
be accounted for by the fact that “as speakers of the language, we unreflectively assume
that there are nonexistents and that reference to them is possible” (Reimer 2001: 499).
It is important to realize that Reimer’s approach does not require that the assumptions
we make themselves be true—it may well be incoherent to assume there are non-
existents to which we can refer. But it is, as she points out, clear that the assumptions
are made. This is why we are willing to accept as meaningful and true sentences such
as ‘There is a fictional character called “Sherlock Holmes”, but he doesn’t exist—he’s
only fictional after all’ (Reimer 2001: 499). The acceptance of a philosophical theory
(such as Millianism, for example) may well disrupt our pretheoretic intuitions. But the
problems surrounding empty names concern our pretheoretic intuitions, and these are
to be explained by the pretheoretic assumption of reference to nonexistents.

159
 SA RA H SAW Y E R

Reimer’s approach appears to explain all three apparent phenomena of meaning, ref-
erence, and truth. Essentially this is because the apparent phenomena are based on the
fundamental fact that an empty name is a name. It would be tempting to conclude that
this provides further support for Millianism as a semantic theory. But such a conclusion
would be too hasty. What Reimer has identified is a plausible assumption that speakers
make which explains our inuitions concerning meaning, reference, and truth in the
context of sentences containing empty names. The assumption—together with its abil-
ity to explain the relevant intuitions—is certainly consistent with Milliansim: but the
two are strictly independent. In particular, the assumption is also consistent with a view
proposed by Tyler Burge according to which a name is a predicate true of an object if
and only if the object was given that name in an appropriate way (Burge 1973). Names,
if Burge is right, are predicates potentially true of many different individuals. According
to Burge, a singular use of a name such as ‘Bertrand Russell’ as it occurs in sentences
such as ‘Bertrand Russell was a pacifist’ are to be understood as semantically equivalent
to demonstrative expressions such as ‘that Bertrand Russell’, and hence as falling within
the same semantic category as explicit demonstrative-predicate constructions such as
‘that cat.’ An empty singular use of a name such as ‘Vulcan’ as it occurs in sentences
such as ‘Vulcan is a planet’ are then treated as failed demonstratives. Since this is the
only semantic theory of names that adequately accounts for the fact that different indi-
viduals can share a name (Burge 1973; Sawyer 2010), it is this semantic theory in con-
junction with the assumption Reimer identifies that ought to be accepted.

5 Empty Predicates
The problems of meaning, reference, and truth that surround empty names have been
widely discussed throughout the history of analytic philosophy, and the ability of a
semantic theory to deal with the problems has been taken as a primary consideration
in its assessment. In contrast, it is widely assumed that no such problems arise for predi-
cates: that predicates are semantically unproblematic in virtue of being guaranteed a
referent. This, I think, is misguided, particularly in a naturalistic climate. It is no more
obvious that predicates such as ‘x is a unicorn’ and ‘x is a Greek god’ are guaranteed a
referent, than that names such as ‘Pegasus’ and ‘Poseidon’ are guaranteed a referent.
This is potentially problematic in particular for semantic theories which try to deal
with the problems surrounding empty names by appeal to the guaranteed referents of
predicates. Russell’s theory of descriptions provides an early example of this. Accord-
ing to Russell, a name would be problematic if understood as a singular term, since a
singular term makes no semantic contribution to a proposition expressed by a sentence
containing it if it lacks a referent; but an ordinary name is not problematic if understood
as a definite description, since a definite description always provides a semantic contri-
bution to a proposition expressed by a sentence containing it, even if there is no object
which uniquely satisfies that description. But if it were possible for a definite description
to be empty—not in the sense of there being no object that satisfies it, but in the sense
of there being no set of properties it refers to—the problem of empty names would not
have been solved.
A second example is provided by Parsons’s theory of nonexistent objects (Parsons
1980). According to Parsons, there is an object corresponding to every set of proper-
ties. To some of those sets of properties there corresponds an object that exists, and to
others there corresponds an object that does not exist (a nonexistent object). The realm

160
 EM PTY N A ME S

of nonexistent objects includes objects from fiction, myth, and failed scientific theory,
as well as impossible objects such as round squares. Consequently, Parsons’s ontology
of nonexistents solves the semantic problems raised by empty names because there are
(effectively) no longer any empty names: there are simply names that refer to existent
objects on the one hand, and names that refer to nonexistent objects on the other. (Par-
sons does distinguish referring to a nonexistent object from failing to refer to an object,
but the distinction does not affect the main point here.) But, crucially, the solution to
the semantic problems depends on the claim that every predicate refers to a genuine
property, and in particular that the predicates that define the nonexistent objects refer
to genuine properties. If ‘x is a Greek god’ were an empty predicate, then there could be
no Greek gods, not even nonexistent ones.
A rather different kind of theory that relies on the same questionable presupposition
is the version of Millianism that is committed to the claim that sentences containing
empty names express ‘gappy’ propositions but tries to account for our intuitions con-
cerning meaning, reference and truth by appeal to the descriptive content either of the
sentences themselves, or of the propositions pragmatically conveyed by them. Again,
the presupposition here is that the availability of descriptive content provided by the
predicates is unproblematic (Recanati 1993; Adams et al. 1994; Taylor 2000; Adams
and Fuller 2007). As discussed above, Braun’s appeal to ways of believing also relies on
predicates being guaranteed a referent, since ‘x believes in a “Vulcan-ish” way’ must
refer if his strategy is to work.
If predicates are guaranteed referents, then appealing to the referents of predicates
in an attempt to solve the problems of meaning, reference, and truth for empty names
might be legitimate. If not, then alternative solutions must be found, ones that can in
addition solve the parallel problems of meaning, reference, and truth for empty predi-
cates. Reimer’s claim that our pretheoretic use of empty names involves the presupposi-
tion that we can refer to nonexistents could perhaps be adapted here; in like fashion,
our pretheoretic use of empty predicates involves the presupposition that they refer
to properties. The relevant presuppositions, however, are consistent with more than
one semantic theory, and should not themselves be seen as part of any given semantic
theory.

References
Adams, F., Fuller, G., & Stecker, R. (1994) ‘Vacuous Singular Terms,’ Mind and Language 9: 387–401.
Adams, F., Fuller, G., & Stecker, R. (1997) ‘The Semantics of Fictional Names,’ Pacific Philosophical
Quarterly 78: 128–48.
Adams, F. and Fuller, G. (2007) ‘Empty Names and Pragmatic Implicatures,’ Canadian Journal of Philosophy
37: 449–62.
Braun, D. (1993) ‘Empty Names,’ Noûs 27: 449–69.
Braun, D. (2005) ‘Empty Names, Fictional Names, Mythical Names,’ Noûs 39: 596–631.
Burge, T. (1973) ‘Reference and Proper Names,’ Journal of Philosophy 70: 425–39.
Currie, G. (1990) The Nature of Fiction, Cambridge: Cambridge University Press.
Donnellan, K. (1966) ‘Reference and Definite Descriptions,’ Philosophical Review 75: 281–304.
Evans, G. (1982) The Varieties of Reference, Oxford: Oxford University Press.
Everett, A. (2000) ‘Referentialism and Empty Names,’ in Everett and Hofweber (eds.) Empty Names, Fiction
and the Puzzles of Nonexistence, Stanford: CSLI Press.
Everett, A. (2003) ‘Empty Names and “Gappy” Propositions’ Philosophical Studies 116: 1–36.
Frege, G. (1892) ‘On Sense and Meaning,’ translated in P. Geach and M. Black (eds.) Translations from the
Philosophical Writings of Gottlob Frege, Oxford: Blackwell, 1952.
Kripke, S. (1980) Naming and Necessity, Cambridge, MA: Harvard University Press.

161
 SA RA H SAW Y E R

McDowell, J. (1977) ‘On the Sense and Reference of a Proper Name,’ Mind 86: 159–85.
McDowell, J. (1984) ‘De Re Senses,’ Philosophical Quarterly 34: 283–94.
Neale, S. (1990) Descriptions, Cambridge, MA: Cambridge University Press.
Parsons, T. (1980) Nonexistent Objects, New Haven: Yale University Press.
Recanati, F. (1993) Direct Reference: From Language to World, Oxford: Blackwell.
Reimer, M. (2001) ‘The Problem of Empty Names,’ Australasian Journal of Philosophy 79: 491–506.
Russell, B. (1905) ‘On Denoting,’ Mind 14: 479–93.
Russell, B. (1911) ‘Knowledge by Acquaintance and Knowledge by Description,’ Proceedings of the Aristote-
lian Society 11: 108–28.
Russell, B. (1919) Introduction to Mathematical Philosophy, London: George Allen & Unwin.
Salmon, N. (1986) Frege’s Puzzle, Cambridge, MA: MIT Press.
Salmon, N. (1998) ‘Nonexistence,’ Noûs 32: 277–319.
Salmon, N. (2002) ‘Mythical Objects,’ in J.K.Campbell, M. O’Rourke, and D. Shier (eds.) Meaning and
Truth: Investigations in Philosophical Semantics, New York: Seven Bridges Press.
Sawyer, S. (2002) ‘Abstract Artifacts in Pretence,’ Philosophical Papers 31: 183–98.
Sawyer, S. (2010) ‘The Modified Predicate Theory of Proper Names,’ in S. Sawyer (ed.) New Waves in
Philosophy of Language, London: Palgrave Macmillan Press.
Soames, S. (1987) ‘Direct Reference, Propositional Attitudes, and Semantic Content,’ Philosophical Topics
15: 47–87.
Soames, S. (2002) Beyond Rigidity, Oxford: Oxford University Press.
Strawson, P.F. (1950) ‘On Referring,’ Mind 59: 320–44.
Taylor, K. (2000) ‘Emptiness Without Compromise,’ in A. Everett and T. Hofweber (eds.) Empty Names,
Fictional Names, and the Puzzles of Nonexistence, Stanford: CSLI Press.
Thomasson, A. (1998) Fiction and Metaphysics, Cambridge: Cambridge University Press.

162
 1 . 14
RE L E V ANC E T HEOR Y
Robyn Carston

1 Introduction
Relevance theory (RT) is best known for its account of verbal communication and
comprehension, but it also sets out a general picture of the principles driving the human
cognitive system as a whole and this plays a crucial role in underpinning the particular
claims made about communication and the pragmatic theory that follows from them.
The various post-Gricean accounts of the principles and processes that mediate the
gap between sentence meaning and speaker meaning can be divided broadly into three
classes based on their orientation: linguistic, philosophical, and cognitive-scientific.
Linguistically oriented theories tend to focus on those pragmatic processes which are
the least context-sensitive and most code-like, reflecting default or general patterns of
language use (Levinson 2000; Horn 1984, 2004). Philosophically oriented accounts
tend to follow Grice closely in maintaining his system of conversational norms and
providing rational reconstructions of the “conversational logic” that delivers speakers’
implicated meaning (Neale 1993, chapter 3; Recanati 2001, 2004). Given its cognitive-
scientific orientation, relevance theory pragmatics is concerned with the on-line proc-
esses of utterance interpretation and the nature of the mental system(s) responsible for
them (Sperber and Wilson 1986/1995a; Wilson and Sperber 2004). So it is responsive
to research in evolutionary psychology on the nature of human cognitive architecture,
empirical work on children’s communicative development and experimental meas-
ures of adults’ on-line comprehension, investigations into the relation between prag-
matic competence and theory of mind (the ability to attribute intentions and beliefs to
others), and clinical studies of people with impaired communicative capacities. For a
survey of the ways in which Relevance Theory engages with these issues, see Wilson
and Sperber (2004), Wilson (2005).
Given the philosophical nature of this volume, I will focus less in this chapter on the
cognitive theorizing and experimental work that has built up around relevance theory
than on those issues which have brought it into direct contact with debates in the phi-
losophy of language. These include the meaning and function of singular terms (names,
indexicals, demonstratives) and definite descriptions, the apparent occasion-sensitivity
of word meaning and the extent to which pragmatics may affect the truth-conditional
content of an utterance. These are all issues that bear on the distinction between the
meaning provided by the linguistic system and the meaning that arises through the
pragmatics of human communicative interaction. Broadly speaking, philosophers of
language fall into two camps: semantic minimalists, who maintain that natural language
 RO BYN CA R S T O N

sentences provide a propositional content that is essentially pragmatics-free, and seman-

tic contextualists, who insist that it is only utterances (or speech acts) that express propo-
sitional contents and these are irremediably context-sensitive. Although RT is usually
classified as a contextualist theory, it will be suggested that, on the basis of its cognitive
underpinnings and its emphasis on minds in communication, it occupies a distinct posi-
tion, which I call “pragmaticism.” These philosophically oriented issues are taken up in
section 3, but first, in section 2, I lay out the main tenets of the theory.

2 Relevance Theory—Principles and Processes

2.1 Relevance Theory and Cognition

According to the RT framework, human cognitive systems quite generally are geared
towards achieving as many improvements to their representational contents and to their
organization as possible, while insuring that the cost to their energy resources is kept as
low as reasonably possible. At the center of the theory is a technically defined notion
of relevance, where relevance is a potential property of any input to any perceptual or
cognitive process. An input may deliver a variety of different types of cognitive effects
to the system; it may, for instance, combine inferentially with existing assumptions to
yield new conclusions (known as contextual implications), it may provide evidence that
strengthens existing beliefs, it may contradict and eliminate already held information,
or it may rearrange the way information is stored. Such effects may or may not be benefi-
cial to an individual, that is, they may increase or decrease the accuracy of the cognitive
system’s information about the world and may make useful information easier or harder
to access. An input is relevant to a cognitive system only if it benefits that system, that
is, only if it has positive cognitive effects. The other crucial factor affecting the degree of
relevance of an input (whether an external stimulus or an internal mental representa-
tion) is the processing effort it consumes: Deriving effects from any given input requires a
mobilization of cognitive resources, including attention, memory, and various process-
ing algorithms and heuristics. Thus, the relevance of any input is a trade-off between
the positive cognitive effects it yields and the processing effort it requires: the greater
the ratio of effects to effort the greater the relevance of the input. The basic claim of the
framework is that human cognition is oriented towards maximizing relevance (known
as the Cognitive Principle of Relevance). The evolutionary idea underlying this is that, as
a result of constant selection pressure towards increasing cognitive efficiency, humans
have evolved procedures to pick out potentially relevant inputs and to process them in
the most cost-effective way (Sperber and Wilson 1995).
RT has developed in tandem with two important and closely connected ideas in
cognitive science: (1) that the mind is modular, and (2) that many mental processes are
performed by fast and frugal heuristics. Arguments from evolutionary psychology suggest
that the mind is massively modular, in the sense that a great many distinct dedicated
procedures and processes have evolved to solve specific cognitive problems (Cosmides
and Tooby 1994; Sperber 2002). And it seems that, in solving a wide variety of everyday
problems, we employ rather simple, albeit ecologically rational, heuristics rather than
foolproof algorithms or explicit reasoning processes (Gigerenzer et al. 1999). These
mechanisms making up the mind’s “adaptive toolbox” are “fast and frugal” in that they
carry out limited computations and consult just a small salient subset of all the available
information rather than working through myriad possibilities and comparing candidate

164
 RELEVA N CE T H E O RY

solutions. A process of this sort can only achieve a high degree of accuracy in a particu-
lar domain to whose regularities it is specifically tailored or adapted.
The relevance-theoretic approach to communication situates pragmatics within
this sort of cognitive framework, that is, one which consists of largely domain-specific
capacities, each with the function of solving a specific pressing problem in human
mental life and employing quick, relatively cheap computations to do so (Sperber and
Wilson 2002; Allott 2008). The human pragmatic capacity is such a dedicated system:
Its specific domain is ostensive stimuli (verbal utterances and other acts of ostensive
communication) and the comprehension procedure it employs is a fast and frugal heu-
ristic. This idea is elaborated in the next section.

2.2 Relevance Theory and Linguistic Communication

The starting point for a pragmatic theory is the question of how hearers are able to
bridge the gap between the linguistic meaning encoded in an utterance and the speak-
er’s intended meaning. The most obvious manifestations of this gap are nonliteral uses
of language such as metaphor and irony, and cases where, as well as the proposition she
explicitly expresses, the speaker communicates an additional proposition known as a
conversational implicature, exemplified by speaker Y’s utterance in (1).

(1) X: We need your written report now.

Y: I’ve been very busy recently.
Implicating: I haven’t written the report yet.

There is also a range of pragmatic tasks involved in determining the proposition explic-
itly expressed, including disambiguation, assignment of referents to indexicals, and
filling in missing constituents, as in (2), and various other enrichments or adjustments
of encoded content, as indicated in the examples in (3) (where the particular proposi-
tion expressed given here is, of course, just one of indefinitely many possibilities):

(2) He has taken enough from her.

Expressing: Jim has endured enough abusive treatment from Mary.
(3) a. I’ve eaten.
Expressing: I’ve eaten dinner tonight.
b. Your knee will take time to heal.
Expressing: Your knee will take a substantial amount of time to heal.
c. The water is boiling.
Expressing: The water is very hot [not necessarily strictly at boiling
point].

While the proposition expressed by (3a) contains constituents of content that don’t
appear in the sentence uttered, (3b) involves a narrowing down of “take time” and (3c)
a loosening of the concept encoded by “boiling.”
How, then, is an addressee able to infer the intended meaning from the encoded
linguistic meaning which is just a schematic guide or set of clues? According to RT,
the answer lies with a special property of overtly communicative acts: which is that
they raise certain expectations of relevance in their addressees—that is, expectations
about the cognitive effects they will yield and the mental effort they will cost. Quite

165
 RO BYN CA R S T O N

generally, an utterance comes with a presumption of its own optimal relevance—that is,
there is an implicit guarantee that the utterance is the most relevant one the speaker
could have produced, given her competence and her own current goals, and that it is at
least relevant enough to be worth processing. This is known as the Communicative Prin-
ciple of Relevance and it follows from the Cognitive Principle of Relevance in conjunc-
tion with the overtness of the intention that accompanies an utterance: The speaker
openly requests effort (attention) from her addressee who is thereby entitled to expect a
certain quality of information requiring no gratuitous expenditure of effort. That utter-
ances carry this presumption licenses a particular comprehension procedure, which, in
successful communication, reduces the number of possible interpretations to one:

Relevance-theoretic comprehension procedure:

a. Follow a path of least effort in computing cognitive effects: Test interpretive
hypotheses (disambiguations, reference resolutions, lexical adjustments, implica-
tures, etc.) in order of accessibility.
b. Stop when your expectations of relevance are satisfied.

This procedure (a fast and frugal heuristic) is automatically applied in the on-line
processing of verbal utterances: Taking the schematic decoded linguistic meaning as
input, processes of pragmatic completion and enrichment at the explicit level occur
in parallel with the derivation of the implications of the utterance. Central to the
working of the procedure is a subprocess of mutual adjustment of explicit content and
contextual implications, a process guided and constrained by expectations of relevance.
Here is a brief example involving the adjustment of explicit content in response to
expected implications and where the outcome is a narrowing down of a lexically
encoded meaning:

(4) Bill: I’m doing the 10km circuit run this afternoon. Would you like to
come?
Sue: No thanks, I’m resting today.

The verb “rest” encodes a rather general concept, rest, which covers any degree of
inactivity (physical or mental), from sleeping, to staying awake but not moving much,
to performing a range of not very strenuous tasks (with many more possibilities in
between). Suppose now that Sue is quite an athletic person, who exercises regularly,
then her use of “rest” here is plausibly understood as expressing a concept rest*, which
entails a much lower degree of physical activity than she undertakes on her training
days but is still quite compatible with her pottering about the garden or walking to the
shops. A hearer using the relevance-theoretic comprehension heuristic would narrow
the encoded concept rest just as far as is required to satisfy his expectation of relevance
(e.g. by explaining why Sue is refusing his invitation), and no further. This particular
narrowing would cost Bill little effort, given his knowledge of Sue’s exercise habits, and
provide him with a range of contextual implications (e.g. she won’t come with me today
because she is resting*, she may come another day when she isn’t resting*, etc.). In
different circumstances—for instance, in response to the question, “Would you like to
walk to the corner shop with me?”—rest would have to be narrowed much further,
resulting in a distinct concept rest**.

166
 RELEVA N CE T H E O RY

An interesting RT claim in this context is that metaphoric and hyperbolic uses of

words involve a kind of concept broadening (“loose use”), so fall within this general
process of lexical meaning adjustment, which contributes to explicitly communicated
content. For instance, an utterance of the sentence in (5) could be taken literally, or as
an approximation (if, say, the run referred to was a little less than 26 miles), or as hyper-
bolic (if it was obviously much less than the length of a marathon), or as metaphorical
for a long, arduous, exhausting experience, whether physical or mental. The idea is that
there is a continuum of degrees (and kinds) of concept broadening, with metaphorical
use being the most radical case.

(5) It was a marathon.

For detailed exemplification of the RT-based account of lexical adjustment, resulting

in concept broadening, or narrowing, or a combination of the two, see Carston 2002,
Wilson and Sperber 2002, and Wilson and Carston 2007.
It is a basic assumption of RT pragmatics that the meaning encoded in the linguistic
expression type uttered inevitably underdetermines the content that speakers commu-
nicate, not only their implicatures but also the propositional content they communi-
cate explicitly (the explicature of their utterance). It is claimed that this is not simply
a matter of effort-saving convenience on the part of speakers who could employ fully
explicit (eternal) sentences if they so chose, but rather it is an intrinsic property of
public language systems and no amount of effort to be more explicit by employing ever
longer and more complex sentences will achieve full explicitness (Carston 2002 chapter
1, 2009). Since an act of ostensive communication, verbal or nonverbal, preempts the
addressee’s attention and triggers inferential processes geared toward finding an opti-
mally relevant interpretation of the act, much of the speaker’s intended meaning can be
communicated without being encoded. Assuming that language evolved in the crucible
of an inferential pragmatic system already employed in preverbal communication, the
function of linguistically encoded meaning is to channel the inferential mechanism in
particular directions and to provide just that content that cannot be recovered by infer-
ence alone (Sperber 2000; Origgi and Sperber 2000).
In the next section, implications of this view of the language/pragmatics relation are
considered for some central semantic issues within current philosophy of language.

3 Relevance Theory and Issues in Philosophy of Language

3.1 Reference and Definite Descriptions

Virtually everyone agrees on the inherently context-sensitive nature of indexical and
demonstrative expressions and, therefore, on the fact that understanding utterances
containing them requires a process of pragmatic saturation. However, there is much less
agreement on the role of pragmatics in grasping the content of definite descriptions and
proper names in use. According to recent work in Relevance Theory, although they are
not linguistically marked as context-sensitive and so as requiring pragmatic saturation,
their linguistically encoded meaning does not determine their reference and pragmatics
plays an essential role (see, in particular, Powell 2001, 2003, 2010).
As is well known, the key question on definite descriptions over the last 50 years
has been how to analyze what Donnellan (1966) calls the ‘referential-attributive

167
 RO BYN CA R S T O N

distinction’, that is, the fact that a definite description, “the F,” may be used either to
talk about a particular identifiable individual or to talk about whatever/whoever happens
to be uniquely F. As regards proper names, there are broadly three positions: those who
consider names to be the natural language equivalent of logical individual constants,
those who take them to be descriptive, that is, to contribute properties to truth condi-
tions, and those who see them as closely related to indexicals.
How best to treat the referential-attributive distinction has proved a contentious
issue within the philosophy of language. On the one hand, the truth conditions of
an utterance of a definite description sentence appear to alter according to whether
the description is used referentially or attributively. On the other hand, there seem to
be good reasons to believe that definite descriptions are not linguistically ambiguous:
Although the description ‘the mayor of this town’ may be used either referentially or
attributively, it does not seem to be ambiguous in the way that words like “bank” or
“match” are. But these two observations are hard to reconcile on standard philosophi-
cal assumptions: if (leaving aside indexical expressions) you identify the meaning of
an expression with the contribution that expression makes to truth conditions, then
it follows directly that an expression which is capable of making two different kinds of
contribution to truth conditions has two different meanings and so is ambiguous.
However, Relevance Theory offers a natural way to reconcile these data since it draws
a clear distinction between, on the one hand, the linguistically encoded meaning of a
particular expression and, on the other hand, the contribution that expression makes
to truth-conditional content on an occasion of use. Given this distinction, the fact
that a particular expression may make two (or more) distinct kinds of contribution to
truth-conditional content is not evidence for its ambiguity at the level of linguistically
encoded meaning. According to the relevance-theoretic account, definite descriptions
are linguistically univocal, as Donnellan himself maintained, but are truth-conditionally
ambiguous (Bezuidenhout 1997; Powell 2001). The gap between the encoded meaning
of a definite description and what that description contributes to the propositional con-
tent of a particular utterance is bridged by relevance-guided pragmatic inference.
There has been less work from a relevance-theoretic perspective on proper names
than on definite descriptions, but Powell (1998, 2003, 2010) has addressed questions
about proper names as part of a general relevance-theoretic analysis of the semantics
and pragmatics of singular expressions. According to his analysis, all such expressions
are profoundly context-sensitive: whether they make referential or descriptive contribu-
tions to truth conditions is not a matter of the encoded meanings of these expressions,
but is rather a matter of broad context and pragmatic principles. Powell analyses the
encoded meanings of singular expressions (including here definite descriptions) not in
terms of their contribution to truth conditions, but rather in terms of their contribution
to a hearer’s mental representations. All these expressions, on this view, are marked as
individual concept communicators by virtue of their linguistically encoded meaning. That
is to say, their encoded meaning indicates that what they contribute to explicature is
a concept which is taken to be satisfied by a unique individual, but they are neutral
with regard to whether this concept should be de re (referential) or descriptive (attribu-
tive)—this has to be determined pragmatically. Which constraint a particular singular
expression places on the concepts which may serve as its interpretation varies accord-
ing to the type of singular expression. In the case of a proper name “N,” the constraint
on interpretation is simply that the individual concept should be of a bearer of “N.”
Which concept that is on a particular occasion of use will be determined by pragmatic

168
 RELEVA N CE T H E O RY

inference, constrained by the criterion of optimal relevance. A definite description “the

F,” on the other hand, encodes a more complex condition: It constrains interpretation
to an individual concept of a unique F in a salient context. Again, which is the salient
context and which the intended individual concept (and whether it is referential or
descriptive) on any given occasion is determined pragmatically.
On this account of singular terms, the encoded linguistic meaning provides a set of
indications concerning the conceptual content to be recovered, but, on all occasions of
use, some pragmatic inference is required for the full determination of that content. For
a detailed account of how, following the relevance-based comprehension procedure,
either the referential (de re) or the attributive (descriptive) interpretation is recovered,
see Powell (2001).

3.2 ‘Free’ Pragmatic Processes and Unarticulated Constituents

Ascertaining the proposition that a speaker explicitly communicates (the explicature)
is a matter for pragmatics. A controversial claim in this context is that there are “free”
pragmatic processes that can affect this level of content, where “free” is understood as
not required or directed by any element of the linguistic expression used; see Recanati
(1993: 243) for a clear distinction between linguistically controlled “saturation” proc-
esses and linguistically free pragmatic enrichment processes.
There are, arguably, two kinds of “free” enrichment processes. The first of these is the
modulation or adjustment of lexically encoded meanings, as exemplified above in (4)
and (5). The second, and more contentious, is the pragmatic recovery of components of
content which are not linguistically indicated in any way (and so are known as “unar-
ticulated constituents” of utterance content). The following are some cases for which
relevance-theorists have claimed that the bracketed constituent, a component of the
explicature of a particular utterance of the sentence, is not represented in the linguistic
form:

(6) a. It is raining. {in Oslo}

b. I have eaten. {supper}
c. Jill reported Jack for misconduct and he was fired. {as a result}

It seems that, in comprehending an utterance of (6a), we would very often, if not always,
take there to be a specific place at which the raining is occurring; this would often
be the location of the speaker of the utterance, but need not be. Similarly, for many
utterances of (6b), an object of eating is recovered and, for (6c), a cause-consequence
relation is taken to hold between the two events described in the conjuncts. Accord-
ing to the view that these are linguistically unarticulated constituents of the proposi-
tions expressed by speakers of these utterances, their recovery by addressees is entirely
pragmatically motivated; that is, they are inferred as part of the process of finding an
optimally relevant interpretation.
This kind of “strong” pragmatic effect on the proposition explicitly expressed by an
utterance is resisted by semanticists such as Stanley (2000, 2005), King and Stanley
(2005) and Martí (2006). They see this propositional content as the semantics (the
truth-conditional content) of the sentence, relative to a context of utterance, and insist
that, as such, all its constituents must be provided for in the linguistic logical form (LF)
of the sentence. Pragmatics can only have “weak” effects at this level of content; that is,

169
 RO BYN CA R S T O N

its role is just to supply context-specific values to indexical elements in the logical form
(the saturation process). As well as overt, phonologically realized, indexical elements
such as pronouns and demonstratives, a sentence may contain various covert indexicals
which, although not phonologically manifest, constitute elements of syntactic structure.
So, on this view, the sentences in (6a) and (6b) contain such elements; for instance,
(6a) includes an unpronounced adjunct of location, <at L>, which indicates that a
value is to be pragmatically supplied in just the same way as it is to an occurrence of the
overt demonstratives “here” or “there.” Stanley (2000) finds support for the presence
of this covert location indexical from the fact that, like an overt pronoun, it can enter
into binding relations, as in “Every time John lights a cigarette, it rains,” on the inter-
pretation where the location of each event of raining is bound to the location of each
event of John’s lighting a cigarette. In the case of (6c), on the other hand, no plausible
case can be made for a covert indexical and it is accepted that the cause-consequence
relation is a strong pragmatic effect (the result of a “free” pragmatic process). On that
basis, it is treated as an instance of conversational implicature, hence as falling outside
the truth-conditional content of the utterance (King and Stanley 2005), contrary to
widespread intuitions.
In a critical response to this “indexicalist” semantic account, relevance-theorists
have pointed out that there is no principled limit to the number of covert elements that
such a theory would have to posit (Wilson and Sperber 2002). An utterance of (6b),
for instance, might express a proposition with a range of constituents corresponding to
what is eaten, the time, place, manner of eating, and so on. On Stanley’s analysis, each
of these would have to correspond to a variable or indexical in the linguistic logical
form, a theoretical prediction which Wilson and Sperber take to be a reductio of his posi-
tion. Furthermore, as Carston (2002) points out, although all of these hidden indexical
elements would have to be present at LF, there would be many occasions of utterance
of these sentences on which some of these elements would receive no value. Consider
an utterance of (6b) “I’ve eaten” in response to the question “Would you like some
dinner?” While what was eaten and the time of eating might well be relevant (that the
speaker has eaten a full meal and the eating took place in the recent past), the place
and manner of eating would surely not be and would not receive a contextual value.
However, this seems to be quite different from the way that overt indexicals work:

(7) She put the book on the table.

If the addressee of an utterance of the sentence in (7) is unable to assign a value (a refer-
ent) to “she,” he has not fully understood the speaker’s intended content and something
has gone wrong with the communicative exchange. So it seems that Stanley’s hidden
indexicals are not, after all, simply covert counterparts of the overt indexicals that we
are familiar with but entities of some other, unknown sort. The suspicion, from the
pragmaticist side, is that they are an artefact of the semantic theorizing and don’t actu-
ally exist.
Some within the semanticist camp have taken this objection seriously and looked for
a remedy consonant with the position that any pragmatic processes affecting this level of
propositional content are linguistically controlled. Thus, Martí (2006) has argued that
the problem can be overcome by taking the covert indexical elements to be optional,
so that, if and when they do occur, they must be pragmatically saturated, as is the case
for their overt counterparts. Thus “it is raining” has two underlying logical forms, one

170
 RELEVA N CE T H E O RY

with a covert location indexical, one without. Clearly, this move entails a considerable
increase in linguistic structural ambiguity. For instance, the surface sentence form in
(6b) “I have eaten” would have a variety of underlying logical forms, each with an array
of covert indexicals, differing in number and type (including one with none), indicat-
ing possible contextual completions. In the case of a sentence that has four possible
covert indexicals for different constituents, the optionality position results in 16 logical
forms to cover the range of cases. Thus, the shift from obligatory to optional linguistic
structure comes at a high computational cost. While the semanticist acknowledges this
point, she finds it preferable to positing a “mysterious and computationally intractable”
process of free pragmatic enrichment (Martí 2006: 151).
However, according to Carston (2010), a counter-intuitive proliferation of linguistic
ambiguity is not the only problematic outcome of making covert indexicals optional.
When we spell out the implications of the optionality move for the on-line processes
of comprehension, it seems that the logical forms with their varying numbers of covert
indexicals are redundant. The semanticists’ requirement is that indexical saturation,
rather than a free pragmatic process, is responsible for any unpronounced constituents
that occur in the explicature of utterances of sentences like (6a) and (6b). So a logi-
cally prior task, on any given occasion of utterance, is disambiguation of the surface
phonological form, so that one among the range of logical forms with their varying con-
figurations of indexicals is selected as the intended one. But, according to Carston, this
disambiguation process relies on the recovery of relevant contextual information (e.g.
the specific location of raining or the object of eating) which is identical to the con-
stituents of content which would be supplied directly by a free pragmatic process as part
of the inferential construction of an optimally relevant interpretation. If this is right,
a wholly pragmatically driven process of inferring a constituent of content preempts
any process of indexical saturation and renders superfluous the logical forms containing
covert indexicals.
For further relevance-theoretic arguments against hidden indexicals (whether oblig-
atory or optional) and a response to allegations that free pragmatic enrichment is insuf-
ficiently constrained, see Hall (2008a, 2008b, 2009).

3.3 Relevance Theory and the Minimalism/Contextualism Debate

The linguistic meaning of the sentence a speaker utters is decoded and processed bit
by bit in on-line comprehension, but thinking hypothetically of it as a whole, what it
amounts to, on the RT view, is a template or schema for constructing the proposition
the speaker explicitly communicates. It is seldom, if ever, fully propositional (truth-
conditional) and has no reason to be so, since the propositional content which is com-
municatively intended by the speaker is inevitably recovered by pragmatic inference.
A number of philosophers of language concur with this view of what is delivered by
“semantic” interpretation (characterized by its formal, algorithmic, context-free char-
acter). For instance, Recanati (2001) thinks of linguistic meaning as providing seman-
tic schemata or propositional functions, and Bach (1994, 2006) talks of “propositional
radicals.” Both of them maintain that pragmatic processes (including “free” enrich-
ment) make an essential contribution to retrieving the communicatively intended
propositional content. Thus, on all of these accounts, there is a fairly obvious sense in
which the encoded linguistic meaning is “minimal”: it underdetermines the proposition
explicitly communicated and is standardly less than fully propositional itself.

171
 RO BYN CA R S T O N

However, those philosophers of language who advocate a minimal semantics for

natural language sentences see themselves as being in direct opposition to relevance-
theorists and to “contextualist” philosophers such as Recanati and Bach. The most
prominent current minimalists are Borg (2004, 2007) and Cappelen and Lepore
(2005) and their primary concern appears to be to establish that uttered tokens of
natural language sentences “semantically express” a proposition (a truth-conditional
content) which is essentially context-invariant and pragmatics-free. While Cappelen
and Lepore allow for a limited degree of context-relativity, restricted to the small class
of overt indexicals and demonstratives, Borg (2004) argues that even these receive a
semantic value without recourse to pragmatics or broad context. What is meant here
by “minimalism,” then, is propositional minimalism. It is a stance that follows from the
longstanding assumption that what a semantics for any representational system must
deliver is statements of truth-conditional content. However, it is difficult to see why
we should accept this assumption, or why the minimalists are so attached it, since
they agree with relevance-theorists that this “semantically expressed” proposition is
seldom, if ever, the proposition explicitly communicated by an utterance. They accept
that communicated propositions (both explicatures and implicatures) are pragmatic
entities, derived by nondemonstrative inferential processes geared toward recover-
ing an interpretation that meets expectations of rational communicative behavior,
whether Gricean maxims or the presumption of optimal relevance (see, in particular,
Borg 2004: 110). It seems clear, then, that while what is communicated (by uttering
a declarative sentence) must be propositional (must embody statements or claims
about the world, which an addressee may or may not take on as part of his own
belief system), what is merely semantically expressed or encoded need not be.
(For more detailed RT-oriented discussion of Cappelen and Lepore’s “insensitive”
semantics and of Borg’s semantic minimalism, see Carston 2008a and 2008b,
respectively.)
The two central tenets of contextualist semantics are: (a) that it is not sentences but
utterances (or speech acts) that have truth-conditional content, and (b) that virtually
every word in the language is context-sensitive. See, for instance, Searle (1978, 1980),
Recanati (1993, 2004) and Travis (1985, 1997). Unsurprisingly, then, Relevance
Theory is usually placed squarely within the contextualist camp—by minimalists and
contextualists alike.
However, it’s not clear that this is the best way to characterize Relevance Theory
nor that this distinction between semantic theories is central to what RT is about. It
is first and foremost a theory of communication and interpretation, and its advocacy
of free pragmatic processes (meaning modulation and unarticulated constituents)
is entirely motivated by the aim of providing an account of how it is that hearers
can recover speakers’ intended contents, given that these diverge in a range of ways
from the meaning encoded in the linguistic expressions employed. What contextual-
ists take to be the semantics (the truth-conditional content) of an utterance is, for
relevance theorists, the speaker’s explicature—that is, the proposition she explic-
itly communicates. Since thoughts have truth-conditional content, it follows that
any communicated thought has truth conditions, whether it is explicitly or implicitly
communicated (an explicature or an implicature). So, on this account, which posits
a nonpropositional linguistic semantics and, via pragmatics, a set of communicated
propositions, there appears to be no reason to think of the explicature of an utterance
as its semantics.

172
 RELEVA N CE T H E O RY

RT agrees with contextualist philosophers that the meaning a use of a word con-
tributes to an explicature varies from occasion to occasion. Virtually any expression
can be used by a speaker and understood by a hearer to express a meaning that is dif-
ferent from its linguistic expression-type meaning. For instance, a speaker can use the
word “butterfly” to communicate a concept whose denotation includes human beings
who have certain properties, or the word “bachelor” to communicate a concept whose
denotation includes some married men and excludes some unmarried men. But this is
a rather different phenomenon, involving a different property of the words concerned,
from the context-sensitivity of indexicals and demonstratives, which arguably do not
encode a concept to start with but rather a variable with certain indications about the
kind of value that variable should receive. Cappelen and Lepore (2005) have some
quite effective tests for distinguishing this latter class of linguistic expressions from the
rest and it comes as no surprise that there are such discriminatory tests: There are strong
pretheoretic intuitions that indexicals are special and quite different from words like
“butterfly” or “bachelor.”
Rather than taking the contextualist stance that virtually all words are context-
sensitive, RT makes a distinction between inherent context-sensitivity, on the one hand,
(and agrees that it is confined to pretty much the cases that Cappelen and Lepore cite
and which pass their various tests), and what could be called “pragmatic susceptibility,” on
the other hand. What this means is that virtually every (open class) word can be used
and understood to express a meaning that departs in certain ways from the meaning
that it encodes (its expression-type meaning) and this is because of the human prag-
matic interpretive ability (which includes an acute sensitivity to relevant contextual
factors). Linguistic expressions are tools with certain inherent properties (phonological,
syntactic, and semantic) that we, as normally functioning adult humans, can employ
very flexibly for our communicative purposes by virtue of certain characteristics of our
psychological makeup (specifically, our “theory of mind” capacities, and, in particular,
our attunement to each other’s communicative intentions and our expectations of each
other as rational speakers and hearers). Thus, while there is a limited degree of context-
sensitivity built into linguistic systems, pragmatic susceptibility is a pervasive feature of
language as employed by us in ostensive communication. This is one respect in which
relevance theory is better described as being “radically pragmaticist” rather than “radi-
cally contextualist”: It is us, the users of language, that are sensitive to context, and, as
rational communicating/interpreting agents, we are able, by exploiting this sensitivity
in each other, to make linguistic expressions do a lot more than simply express their
standing linguistic meaning.
With regard to their views on the role of context, there is another way in which
contextualist and pragmaticist orientations differ. In his pioneering work on conversa-
tional logic, Grice set out an inferential schema for deriving conversational implicatures
which employed a large component of theory-of-mind type reasoning. However, he did
not carry this over to the pragmatic processes required for the full identification of what
a speaker has said (explicitly communicated) but spoke here of “context as a criterion.”
It seems that he thought of disambiguation and indexical reference assignment as a
matter of contextual best fit, rather than as involving conversational maxims or proc-
esses of reasoning geared to the recovery of what the speaker intended (Grice 1989: 25,
222). In this regard, certain current contextualists, despite being strong advocates of
free pragmatic processes of enrichment and modulation in recovering what a speaker
has said, remain essentially Gricean. For instance, while Recanati (2004) construes the

173
 RO BYN CA R S T O N

“secondary” pragmatic processes of conversational implicature derivation as maxim-

guided, reflective reasoning, which deploys premises concerning speakers’ mental states
(beliefs and intentions), he takes a different stance on the “primary” pragmatic proc-
esses that contribute to the recovery of explicatures. He sees these as a function of
an automatic, dumb (noninferential) cognitive mechanism responsive to differential
degrees of activation of candidate interpretations, such that the most highly activated
one wins out. It is context (both linguistic and extra-linguistic) that does the work here
and contextual coherence that provides the criterion of correctness. As he puts it, “the
interpretation which eventually emerges . . . results from a blind, mechanical process,
involving no reflection on the interpreter’s part. The dynamics of accessibility does
everything and no ‘inference’ is required. In particular, there is no need to consider the
speaker’s beliefs and intentions” (Recanati 2004: 32).
According to relevance theory, on the other hand, the whole utterance interpretation
process is a matter of (nondemonstrative) inference, and taking account of the speaker’s
competence (including her epistemic states) and preferences (her desires, intentions,
interests) may be required for carrying out any of the pragmatic tasks involved (includ-
ing lexical concept adjustments, disambiguation, fixing of indexical reference). Along
with the propositions communicated (explicatures, implicatures), aspects of the con-
text for the interpretation fall under the speaker’s communicative intention and the
hearer selects it (in the form of a set of conceptual representations) as part of his search
for an interpretation that satisfies his expectations of relevance.
Summing up, the pragmaticism of relevance theory has the following characteris-
tics which distinguish it from contextualism: (a) it is primarily a theory of communi-
cation and interpretation rather than a semantic theory and, although explicatures
have truth-conditional content (as do implicatures, thoughts, and propositional enti-
ties quite generally), there is no useful sense in which they are the semantic content
of anything (a sentence, a sentence token in a context, an utterance); (b) while
only a few words in the language are inherently context-sensitive, the vast major-
ity of words are susceptible to the pragmatics of the speaker–hearer interaction such
that they can be used to communicate an indefinite range of different concepts; (c)
it is not context acting on language that is somehow doing the work of determining
explicature content, but, just as for implicatures, it is the exercise of specific human
mind-reading capacities dedicated to achieving the feat of ostensive communication
and comprehension.

Related Topics
1.2 Semantics and Pragmatics
1.6 Pragmatic Enrichment
1.7 Meaning and Communication
1.10 Context-Sensitivity
2.1 Reference
3.1 Names
3.7 Descriptions
4.3 The Role of Psychology
7.6 Ordinary Language Philosophy.

174
 RELEVA N CE T H E O RY

References
Allott, N. (2008) Pragmatics and Rationality. PhD thesis, University of London.
Bach, K. (1994) “Conversational Impliciture,” Mind and Language 9: 124–162.
Bach, K. (2006) “The Excluded Middle: Semantic Minimalism without Minimal Propositions,” Philosophy
and Phenomenological Research 73 (2): 435–442.
Bezuidenhout, A. (1997) “Pragmatics, Semantic Underdetermination and the Referential/Attributive
Distinction,” Mind 106: 375–409.
Borg, E. (2004) Minimal Semantics. Oxford: Clarendon Press.
Borg, E. (2007) “Minimalism Versus Contextualism in Semantics,” in G. Preyer and G. Peter (eds.) Context-
Sensitivity and Semantic Minimalism. Oxford: Oxford University Press, 339–359.
Cappelen, H. and Lepore, E. (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act
Pluralism. Oxford: Blackwell.
Carston, R. (2002) Thoughts and Utterances: The Pragmatics of Explicit Communication, Oxford: Blackwell.
Carston, R. (2008a) “Linguistic Communication and the Semantics/Pragmatics Distinction,” Synthèse 165
(3): 321–345. (Special Issue on the Semantics/Pragmatics Distinction.)
Carston, R. (2008b) “A Review of E. Borg, 2004. Minimal Semantics. Oxford: Clarendon Press,” Mind and
Language 23: 359–67.
Carston, R. (2009) “The Explicit/Implicit Distinction in Pragmatics and the Limits of Explicit Communica-
tion,” International Review of Pragmatics 1 (1): 35–62.
Carston, R. (2010) “Explicit Communication and ‘Free’ Pragmatic Enrichment,” in B. Soria and E. Romero
(eds.) Explicit Communication: Robyn Carston’s Pragmatics. Basingstoke: Palgrave Macmillan, 217–287.
Cosmides, L. and Tooby, J. (1994) “Origins of Domain-Specificity: the Evolution of Functional Organiza-
tion,” in L. Hirschfeld and S. Gelman (eds.) Mapping the Mind: Domain Specificity in Cognition and Culture.
New York: Cambridge University Press, 85–116.
Donnellan, K. (1966) “Reference and Definite Descriptions,” Philosophical Review 75: 281–304.
Gigerenzer, G., Todd, P.M., and the ABC Research Group. (1999) Simple Heuristics That Make Us Smart.
Oxford: Oxford University Press.
Grice, H.P. (1989) Studies in the Way of Words, Cambridge, MA: Harvard University Press.
Hall, A. (2008a) “Free Enrichment or Hidden Indexicals?” Mind and Language 23(4): 426–456.
Hall, A. (2008b) Free Pragmatic Processes and Explicit Utterance Content. PhD thesis, University of
London.
Hall, A. (2009) “‘Free’ Enrichment and the Nature of Pragmatic Constraints,” UCL Working Papers in Lin-
guistics 21: 93–123.
Horn, L. (1984) “Toward a New Taxonomy for Pragmatic Inference: Q-based and R-based Implicature,” in
D. Schiffrin (ed.) Meaning, Form and Use in Context (GURT ‘84).Washington, DC: Georgetown Univer-
sity Press, 11–42.
Horn, L. (2004) “Implicature,” in L. Horn and G. Ward (eds.) Handbook of Pragmatics, Oxford: Blackwell,
3–28.
King, J. and Stanley, J. (2005) “Semantics, Pragmatics, and the Role of Semantic Content,” in Z. Szabo (ed.)
Semantics vs. Pragmatics, Oxford: Oxford University Press, 111–164.
Levinson, S. (2000) Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge,
MA: MIT Press.
Martí, L. (2006) “Unarticulated Constituents Revisited,” Linguistics and Philosophy 29: 135–166.
Neale, S. (1993) Descriptions. Cambridge, MA: MIT Press.
Origgi, G. and Sperber, D. (2000) “Evolution, Communication and the Proper Function of Language,”
in P. Carruthers and A. Chamberlain (eds.) Evolution and the Human Mind: Modularity, Language and
Meta-Cognition. Cambridge: Cambridge University Press, 140–169.
Powell, G. (1998) “The Deferred Interpretation of Indexicals and Proper Names,” UCL Working Papers in
Linguistics 10: 143–172.
Powell, G. (2001) “The Referential-Attributive Distinction—a Cognitive Account,” Pragmatics and Cogni-
tion 9 (1): 69–98.
Powell, G. (2003) Language, Thought and Reference. PhD thesis, University of London.
Powell, G. (2010) Language, Thought and Reference, Basingstoke: Palgrave Macmillan. (Substantially revised
version of Powell 2003.)
Recanati, F. (1993) Direct Reference: From Language to Thought. Oxford: Blackwell.
Recanati, F. (2001) “What is Said,” Synthèse 128: 75–91.

175
 RO BYN CA R S T O N

Recanati, F. (2004) Literal Meaning.Cambridge: Cambridge University Press.

Searle, J. (1978) “Literal Meaning,” Erkenntnis 13: 207–224.
Searle, J. (1980) “The Background of Meaning,” in J. Searle, F. Keifer ,and M. Bierwisch (eds). Speech Act
Theory and Pragmatics. Dordrecht: Reidel, 221–32.
Sperber, D. (2000) “Metarepresentations in an Evolutionary Perspective,” in D. Sperber (ed.) Metarepresen-
tations: A Multidisciplinary Perspective. Oxford: Oxford University Press, 117–137.
Sperber, D. (2002) “In Defense of Massive Modularity,” in E. Dupoux (ed.) Language, Brain and Cognitive
Development: Essays in Honor of Jacques Mehler, Cambridge MA: MIT Press, 47–57.
Sperber, D. and Wilson, D. (1986/95a) Relevance: Communication and Cognition. Oxford: Blackwell.
Sperber, D. and Wilson, D. (1995) “Postface,” in Relevance: Communication and Cognition, Oxford: Black-
well. Second edition 1995a, 255–279.
Sperber, D. and Wilson, D. (2002) “Pragmatics, Modularity and Mind-Reading,” Mind and Language 17:
3–23. (Special Issue on Pragmatics and Cognitive Science.)
Stanley, J. (2000) “Context and Logical Form,” Linguistics and Philosophy 23: 391–434.
Stanley, J. (2005) “Semantics in Context,” in G. Preyer and G. Peter (eds.) Contextualism in Philosophy:
Knowledge, Meaning, and Truth. Oxford: Oxford University Press, 221–253.
Travis, C. (1985) “On What is Strictly Speaking True,” The Canadian Journal of Philosophy 15 (2): 187–229.
Reprinted in C. Travis 2008, 19–64.
Travis, C. (1997) “Pragmatics,” in B. Hale and C. Wright (eds.) A Companion to the Philosophy of Language.
Oxford: Blackwell, 87–107. Reprinted in C. Travis 2008, 109–129.
Travis, C. (2008) Occasion-Sensitivity: Selected Essays. Oxford: Oxford University Press.
Wilson, D. (2005) “New Directions for Research on Pragmatics and Modularity,” Lingua 115: 1129–1146.
Wilson, D. and Carston, R. (2007) “A Unitary Approach to Lexical Pragmatics: Relevance, Inference and
Ad Hoc Concepts,” in N. Burton-Roberts (ed.) Advances in Pragmatics, Basingstoke: Palgrave Macmillan,
230–260.
Wilson, D. and Sperber, D. (2002) “Truthfulness and Relevance,” Mind 111: 583–632.
Wilson, D. and Sperber, D. (2004) “Relevance Theory,” in L. Horn and G. Ward (eds.) The Handbook of
Pragmatics, Oxford: Blackwell, 607–632.

176
 1 . 15
T R UTH AND
R EF E RE NCE IN F IC T ION
Stavroula Glezakos

Introduction
Fiction is often characterized by way of a contrast with truth, as, for example, in the
familiar couplet “Truth is always strange/Stranger than fiction” (Byron 1824). And
yet, there are those who maintain that we are likely to “learn more about human life and
human personality from novels than from scientific psychology” (Chomsky 1988: 159); on
their view, some truth is best encountered via fiction. The scrupulous novelist points out
that her work depicts no actual person, either living or dead; nonetheless, we use
names from fiction in ways that suggest that we take these names to refer. Philosophers
who investigate fiction aim to reconcile such apparently incompatible phenomena,
and, in general, to account for the myriad ways that we talk, think, and feel about
fiction.

Fictional Characters and Fictional Names

Of great interest to philosophers has been the way that certain statements appear to be
true in virtue of fiction. F. Scott Fitzgerald’s novel The Great Gatsby (1925) contains
descriptions of protagonist Jay Gatsby’s lavish West Egg mansion; it thus seems appro-
priate to classify the following as true and false, respectively:

(1) Jay Gatsby lives in a lavish mansion in West Egg.

(2) Jay Gatsby lives in a polyurethane yurt near Big Sur.

The most straightforward account of truth for a sentence that contains a proper name
and a predicate holds that the sentence is true just in case the referent of the name
stands in the appropriate relation to the property expressed by the predicate. Thus, to
maintain that (1) is true, one must either identify a referent for the name “Jay Gatsby,”
or provide an alternative account of the sentence’s truth.
Many philosophers do, in fact, accept that the name “Jay Gatsby” refers. They disa-
gree, however, about what it refers to, with some maintaining that the referent of “Jay
Gatsby” is not a man, others, that it is a man that is not actual, and yet others, that it is
something that does not exist.
 STAVRO U LA GL E Z A K O S

Fictional Names Refer to Abstract Objects

According to abstract object theorists, fictional names refer to fictional characters,
and critical assessment of a fictional work involves reference to such characters (Searle
1975, van Inwagen 1977, Salmon 1998, and Thomasson 1999, 2003, 2009). Consider,
for example, the following sentence, which one might utter during a discussion of the
literary merits of The Great Gatsby:

(3) Daisy is a less well-developed character than Gatsby.

These theorists maintain that accepting (3) commits one to the truth of (4):

(4) Some fictional character exists.

This is because from (3) we can infer that there exists a fictional character that is less
well developed than Gatsby, and thus, that there exists a fictional character (van Inwa-
gen 1977). Such characters—or “creatures of fiction”—are abstract entities, possessing
(or lacking) the sorts of properties that are at issue within critical discourse, such as
being well developed, being sympathetic, or being modeled on a particular person (van
Inwagen 1977: 302). In uttering (3), one refers to two such characters, and it is their rel-
ative degree of development that determines whether one’s utterance is true or false.
Matters become more complicated when we consider sentences like (1) and (2). If
the referents of fictional names are abstract objects—that is, objects that are neither
physical nor causally efficacious—then descriptions within novels cannot be specify-
ing properties of those objects. And, indeed, abstract object theorists direct us to give
a revised reading to any predicative sentence involving a fictional name and predi-
cate assigned to a character within a novel. An assertive utterance of (1), for example,
should be understood as the claim that in the novel The Great Gatsby, the property of living
in a lavish mansion in West Egg is ascribed to the character of Gatsby—which these theorists
take to be true.
Some abstract object theorists hold that fictional characters come into existence
along with the particular work of fiction in which they figure (Searle 1975, Salmon
1998, and Thomasson 1999). On their view, creation of a work of fiction involves pre-
tended reference to persons, places, and things; by means of such pretence, an author
brings into being not only her novel, but also fictional persons, places, and things. Once
created, a character can appear in a later work by someone other than its creator, and
may be portrayed in ways that depart from the depictions offered in the original work,
so long as the new author’s intentions are appropriately directed at the character from
that earlier story.
In contrast, Wolterstorff (1980) argues that the authorial task of character develop-
ment is best understood as providing specifications for a sort of thing, rather than crea-
tion of a particular thing. On his account, names from fictional works refer to abstract
universals, or “person-kinds,” that exist prior to, and independently of, those works.
One such kind consists of the conjunction of all and only the properties that are, in The
Great Gatsby, assigned to Gatsby; a sentence like (1), if used to assert that the Gatsby-
kind is partly constituted by the property of living in a lavish mansion in West Egg, comes out
not just true but necessarily true. Understanding characters in this way allows us to
suggest, of an actual person who happens to possess a large number of the features that

178
 TRU TH A N D REF ERE N C E I N F I C T I O N

constitute the Gatsby-kind, that he strikes us as an instance of that kind (although,

since no actual human could satisfy all and only the descriptions of Gatsby in the novel,
Gatsby-hood could never actually be instantiated).
Perhaps unsurprisingly, the proposal that fictional characters are existing abstract
entities has met with some objection. One concern is that it does not seem to provide
for the truth of sentences like (5) and (6):

(5) Daisy has a larger wardrobe than Nick Carraway.

(6) Either Daisy has more freckles on her left leg than on her right, or she has
more freckles on her right leg than on her left, or she has exactly the same
number of freckles on both legs.

Since neither the size of Daisy’s wardrobe relative to Nick’s, nor the number of freckles
on Daisy’s legs, are discussed within The Great Gatsby, the sentences are false if given
the reading in The Great Gatsby, . . . And, since abstracta do not have wardrobes or
freckles, the sentences will also be false if they are read as direct predications of the
fictional characters Nick Carraway and Daisy Buchanan. Thus, if we think that it is
possible to make true claims about characters that go beyond the descriptions provided
in the fiction, we may find ourselves unsatisfied by this account.
A deeper worry is that such fictional characters appear to be precariously balanced
between inconsistency and superfluity. The characters that figure in the beliefs, inten-
tions, and emotional responses of authors and readers (see Thomasson 1999; cf. Braun
2005) do not seem to be abstracta. Take, for example, one author’s description of char-
acter development as requiring “the very hard work of feeling a character deeply, of
understanding what happiness or hurt drives that character, so that it becomes impossi-
ble to lay superficial judgment on any behavior” (Silver 2010). Similarly, readers invoke
characters in descriptions of their emotional responses to fiction, as in the teenage
Gatsby enthusiast’s powerful three-paragraph essay about the disgust, attraction, and
exasperation that Daisy inspires in him. If fictional characters are abstracta, they are
in no way appropriate objects of such attitudes; furthermore, being (presumably) caus-
ally inert, they can neither have, nor bring about, reactions of any kind (cf. Friend
2000: 999).
Some abstract object theorists maintain that authors do not write about fictional char-
acters (see, for example, Searle 1975, van Inwagen 1977, and Salmon 1998). Nonethe-
less, on their view, a name that appears in a fictional work can be “imported” into real-
ity, and used to make true assertions about a character—which, according to Salmon, is
“the very same thing that it is the name of according to the story” (Salmon 1998: 80).
If, however, the author’s uses of the name in the novel do not refer to a character, it is
not clear how, within that novel, properties are nonetheless ascribed to that character.
(According to van Inwagen, the relation of ascription of a property to a character in a
work of fiction “must be taken as primitive” (1979: 306).)
Finally, taking fictional names to refer to actually existing abstract objects appears to
commit us to the falsity of sentences that we reflexively assent to, like (7):

(7) Daisy Buchanan does not exist.

To this, proponents of the view respond that (7) has different uses, and that, although
it is true on some of those uses, it is not true on all of them. Someone might utter the

179
 STAVRO U LA GL E Z A K O S

sentence to assert something that these theorists accept—for example, that there is no
person Daisy Buchanan, that Daisy Buchanan is not a real human, or that there is no object
that actually possesses the properties ascribed to Daisy Buchanan in the novel. Only when (7)
is used to assert something like there is no Daisy Buchanan, or the name ‘Daisy Buchanan’
does not refer do these theorists object, since, on their view Daisy Buchanan (the fic-
tional character) does exist. Thus, abstract objects theorists maintain that they can rec-
ognize true readings for (7); all that they deny is that (7) must be given a true reading.

Fictional Names Refer to Non-Existent Objects

Meinong (1904), however, takes (7) to be straightforwardly true. He draws a distinction
between being and existence, and argues that an object need not exist in order to be the
subject of thought and talk. A thing could, for example, be a male human who walks,
talks, loves, and, in general, has the features attributed to Gatsby in the novel, while
lacking the property of existence; that nonexistent object would be the referent of the
name “Jay Gatsby,” and (7) would be true in virtue of the fact that it does not exist.
Van Inwagen (1977: 299) gives a stark statement of one objection to this proposal:
it appears to “necessitate an abandonment of what are commonly called ‘the laws of
logic’.” That is: How can we consistently maintain that there is something that does not
exist? Moreover, if the referent of a fictional name possesses all of the properties attrib-
uted to the character within the fiction, then we should expect the referent of “Gatsby”
to possess the property of existence, since, in the novel, Gatsby is portrayed as an existing
walking and talking man.
Parsons’ (1980) account aims to overcome these challenges (Zalta 1983 and Zemach
1993 for alternative accounts of non-existent objects). He distinguishes between
“nuclear” properties (like red, round, sad, human, etc.—intuitively, features by means
of which an object could be identified), any set of which corresponds to an object, and
“extra-nuclear” properties like existence. A fictional object, Parsons maintains, pos-
sesses only the nuclear properties attributed to it within the story. Thus, sentences (1)
and (7) are true, while (2), (4), (5), and (6) are false; the truth-status of (3) is less clear.
For the majority of philosophers, however, the question of whether we can arrive at a
formally satisfactory account of the truth and falsehood of sentences by way of non-
existent objects is academic, since they remain unreconciled to the supposed distinction
between being and existence: “Is Gatsby a man? Yes. If he is pricked (or shot), does he
not bleed? Yes. Is he worth more than the whole damn bunch put together? Probably.
Just remember, though: he doesn’t exist!”

Fictional Names Refer to Possible Objects

A third account of fictional names (developed in Lewis 1978) takes “Jay Gatsby” to refer
to a man who really does exist, and who walks, talks, and loves—albeit as a denizen of
some other (concrete) possible world. Straightforward readings of (7) come out false
on this view; however, since “Jay Gatsby” doesn’t refer in the actual world, (7) can be
assessed as true if it is understood as the assertion that Jay Gatsby is not actual.
According to the possible objects theorist, at least some of the stories that are fic-
tional in our world—including the one presented in The Great Gatsby—are, in some
other possible worlds, “told as known fact”. In each such possible world, the name
“Jay Gatsby” refers to the individual who “plays the role” of Gatsby in that world.

180
 TRU TH A N D REF ERE N C E I N F I C T I O N

An utterance of (1)—which, according to Lewis, abbreviates “In The Great Gatsby, Jay
Gatsby lives in a lavish mansion in West Egg”—is true just in case in every world w in
which the Gatsby story is told as known fact, the man who plays the role of Gatsby in
w lives in a lavish mansion in West Egg. Thus, it is not really accurate to say that the
possibilist picture provides a referent for the name “Jay Gatsby”; rather, it provides a
(different) referent in each of the possible worlds determined by the story.
One advantage of taking fictional names to refer to possible objects is that it allows
us to recognize the truth of assertions that go beyond what is explicitly predicated of
characters within a fictional work. In all worlds in which the story is told as known fact,
the woman who plays the role of Daisy will have a determinate number of freckles on
her legs, and so (6) is true. (5) will be true, according to Lewis, just in case among the
worlds in which the Gatsby story is told as known fact, a world in which Daisy has a
larger wardrobe than Nick differs less from our actual world than any world in which
Daisy does not have a larger wardrobe than Nick.
Challenges to the possibilist account of fictional reference fall into two categories. In
the first are general concerns about modal realism: for example, that we seem to have
no means by which to verify anything about the concrete possible worlds posited by
Lewis, including their existence. The second variety of objection holds that, even if
we grant that such possible worlds exist, we should nonetheless deny that the referents
of our fictional names are to be found there. Lewis himself recognizes (1978: 263) that
he gives no account of truth for sentences like (3) and (4), and that his proposal is in
that respect incomplete. Furthermore, Donnellan (1970) and Kripke (1972) argue that
only an object that is causally or historically connected to a name’s introduction and
subsequent use can be the referent of that name; if they are correct, then only objects in
the actual world are candidate referents for names from fiction. Thomasson (2009: 16)
finds the possibilist’s identity conditions for fictional characters too restrictive, since
they entail that if the author had ascribed to the character “so much as one different
property (however trivial), she would have written about a different possible person.”
While this does seem like a dramatic consequence, the possibilist will presumably not be
bothered by it, since, on his view, an author who ascribed a different property would be
a different person: any scenario involving what did not actually happen is part of some
other possible world.

Fiction and Make-Believe

As noted in the previous section, some theorists who hold that fictional names refer
acknowledge that, when an author is writing her novel, “there is no point in debating
what sort of thing” she is writing about, since she is not writing about anything (van
Inwagen 1977: 301). But if we accept this, and if our spirits have been dampened by
challenges to the referentialist views discussed in the previous section, we might well
find ourselves reconsidering the enterprise of seeking referents for fictional names.
Bertrand Russell (1919: 169) warns of the folly of any such project; on his view, “it
is of the very essence of fiction that only the thoughts, feelings, etc., [of an author] and
his readers are real, and that there is not, in addition to them, an objective [character].”
Russell takes those who posit referents for fictional names to have lost the “robust sense
of reality” that is required for successful theorizing in this area. Some recent theorists
who join Russell in denying that names from fiction refer do so with a robust sense of a
particular part of reality: our games of make-believe.

181
 STAVRO U LA GL E Z A K O S

These theorists (most prominently, Walton 1990; see also Currie 1990) character-
ize works of fiction as “props” in coordinated exercises of the imagination. Indeed, it is
in virtue of playing a particular role in games of imaginative pretence, rather than any
intentions on the part of an author, that a text is a work of fiction: “the notion of objects
serving as props is independent of and has priority over that of the action of making
objects to be used as props” (Walton 1991: 428). As with all varieties of make-believe,
to consider which reports about fictional content are really true, or to attempt to exter-
nally assess the truth of what is asserted within a fiction, is to make a basic mistake. This
becomes clearer when we consider two children who spend a balmy summer afternoon
inhabiting the roles of Antarctic explorers adrift in iceberg-infested waters. Though the
children may later tell family members lurid tales of their adventures, they would surely
grow impatient with anyone who insisted on being told whether the explorers had really
been growing weak from hunger, and whether they had really been able to eat their fill
of plankton. On the one hand: “Yes, that’s exactly what we were pretending!” And on
the other: “We were just pretending—there aren’t any such explorers!”
Walton’s characterization of fictional ‘truth’ instead involves “there being a prescrip-
tion or mandate in some context to imagine something” (1990: 39). When we follow
the mandate, and imagine that what some sentence expresses is true, and that the names
within it refer, we are not, however, imagining of something that it is a certain way. This
is pretence, after all, and our pretendings take the form there is a man named ‘Jay Gatsby,’
and he lives in a lavish mansion in a place called ‘West Egg.’ Neither The Great Gatsby story,
nor the imaginative uses to which we put it, make it the case that (1) is true, although
(1), unlike (2), is fictionally true, or true in the story—just as, in the story, the name
“Jay Gatsby” is a name of a man, while “Kengo Matsusaka” is not. Pretence theorists
also deny that sentences like (3) and (4) are true, or that assertive utterances of them
commit us to the existence of (real) fictional characters. Instead, they maintain that
those who assert or assent to utterances of these sentences are engaged in a particular
variety of make-believe, in which what is pretended is that there are fictional characters
(Walton 1990, Brock 2002).
Walton’s (1990) view is that, like fictional truth and ‘real’ truth, cognition and emo-
tion within games of pretence should be distinguished from their nonfictional coun-
terparts. Given that I ‘believe’ that Gatsby has just been shot, it is not surprising that I
‘feel’ sad and bitter. However, it is only within the pretence that I believe that Gatsby
has been shot, and that I feel sad and bitter (that is: I am not experiencing real sadness
and bitterness). What is crucial to notice, according to Walton, is that the ‘feelings’ (or
quasi-emotions) that are elicited by fiction are not motivating: a moviegoer may, during
a film about deadly green slime, tremble and clutch his chair with sweaty hands, but
he does not run out of the theater, or believe that he needs to quickly move to higher
ground, as he would if he were really afraid. In contrast, Lamarque (1981) argues that
someone can be genuinely frightened by the thought of green slime, even if he is not
frightened that, for example, the slime will soon engulf the theater.
Walton (1990) maintains that we are sometimes directed by fiction to imagine some-
thing that we actually believe, or know, to be true. According to New (1996), this posi-
tion reveals problematic elements of Walton’s portrayal of imagination and fiction. New
maintains that someone who believes that New York is the most exciting city in the
world, and encounters the sentence “New York is the most exciting city in the world”
in a work of fiction will not imagine, or make-believe, that the sentence is true, or that
the name “New York” refers. Rather, the reader’s thought will involve the actual city

182
 TRU TH A N D REF ERE N C E I N F I C T I O N

and its actual place at the pinnacle of exciting cities; thus, engagement with a work of
fiction cannot, according to New, be understood as an exclusively imaginative exercise.
(For a richly detailed discussion of mental pretence, see Camp 2009; for discussion of
empirical results linking imagination and belief, see Nichols 2004.)
The pretence account also allows for the questionable possibility that a scrupulously
accurate report of events that really did occur could qualify as a piece of fiction if it
came to be used as a prop for certain imaginative exercises. The reverse is true as well: A
work produced as fiction would cease to be such if it no longer figured in our practices of
make-believe. Currie (1990) maintains that even if those who authored the Bible did so
with fictional intentions—that is, even if the Bible was authored by those who intended
it to serve as a prop in games of make-believe—the fact that “sufficiently many” people
do not now read biblical stories as fiction means that the Bible could not, now, be prop-
erly classified as fiction. Thus, the pretence account seems to have the counter-intui-
tive consequence that being a work of fiction amounts to being treated as fiction. But
this cannot be correct: We can, after all, ask about a text that is identified as a piece of
fiction whether it in fact is, and we have a general idea of what sorts of investigations we
might carry out in order to arrive at an answer.

Truth in Fiction
Our focus so far has mainly been on accounts of truth for individual sentences within,
or about, fiction. But philosophers have also considered whether a fictional work, in
its entirety, could express or present a truth, and have attended to the worry that it is
not possible to learn such truths via fiction, since it would seem that even a work that
expressed a truth could not provide evidence in support of it.
Critical assessments of a work of fiction often aim to make explicit the propositions
that it expresses, or that it is committed to. One such analysis of The Great Gatsby, for
example, takes the novel to provide “profound corrective insights into the nature of
American experience,” most notably, that “the American dream, stretched between a
golden past and a golden future, is always betrayed by a desolate present” (Bewley 1954).
Since no such sentence about the American dream appears in the novel, we need some
account of how a text might present content that is not expressed by any particular
sentence(s) within it.
On one view, some fictional works express philosophical theses concerning “man and
his world,” which are revealed to us via examinations of “plot, characters, dialogue,
authorial interpolation, tone and themes” (Weitz 1975: 31). Exactly how, and whether,
the plot and additional elements might be related to such theses is a matter of much
dispute among literary critics. To the extent, however, that we offer arguments in sup-
port of the correctness of our particular interpretation of a novel’s ‘deeper meaning’, or
feel unease if we begin to think that what is ‘said’ by a favorite work of fiction might be
objectionable, it is plausible to suggest that we do take a work to have, or be committed
to, this sort of thesis (see Harcourt 2010).
Other theorists maintain that there is no content that fiction presents assertively,
or ‘as true.’ Beardsley (1981) for example, holds that a work of fiction might well have
a meaning discernable by an interpreter, but he denies that the work could advocate
for (let alone establish) that meaning. In support of this position, Lamarque and Olsen
(1994) emphasize that works of fiction rarely contain anything that could be construed
as arguments for theses that they are purported to advocate.

183
 STAVRO U LA GL E Z A K O S

At least some of the candidate truths from fiction are not, however, of the sort that we
would expect to learn via arguments. Nagel (1974) and Jackson (1982) maintain that
not all truths are propositional, or objectively accessible; some are essentially experien-
tial, or subjective. These truths concern what it is like—for example, to be in a certain
situation, to be part of a certain cultural group, or to be a certain kind of person—and,
according to some philosophers, we can encounter such truths via fiction, and perhaps
even receive confirmation of them.
Nussbaum (1985: 516) argues that we can develop our “moral perception” by actively
engaging with fiction: “certain novels are, irreplaceably, works of moral philosophy.”
Reading such works with sufficient appreciation is a means of becoming sufficiently sen-
sitive, and appropriately responsive, to particular complexities that might be present
in human lives. Echoing Nussbaum, those who advocate for fiction’s place within the
curriculum for the helping professions argue that such practitioners must be capable of
taking the perspective of those that they serve: to successfully counsel or advocate for
the bereaved, for example, “the feelings associated with bereavement . . . must either
be experienced first-hand, or else vicariously through writing or other media which
deliberately attempt to invoke them, that is, which resonate with our own feelings”
(Rolfe 2002: 91).
One might worry, however, about how we are to move from acknowledgment that
fiction is a means of adopting a perspective other than our own, to any claim that such
perspectives from fiction can provide truth, in ways that allow us to recognize it as such.
The main source of difficulty here is that a work’s effectiveness in moving the reader to
take a particular perspective, and ‘see’ things from that perspective, seems in principle
to be independent from the truth of what is viewed from that perspective.
As an example, consider Tolstoy’s powerful short novel The Kreutzer Sonata (1889),
in which the central character, Pozdnyshev, explains how and why he murdered his
wife and the man that he suspected of being her lover. As his narrative unfolds, we
become enlightened regarding the causes of his actions; he himself comes to understand
that the love and hate that he felt for his wife “. . . were one and the same animal feeling
only at opposite poles” (p. 126), and that “sexual passion, no matter how it is arranged,
is evil, a terrifying evil . . .” (p. 110). (In an Afterward, Tolstoy elaborates on the lessons
that his novella is meant to impart, among them, “that men and women [should] be
educated in their families and through social opinion” so that “they would not regard
falling in love and the carnal love connected with it as a poetic and elevated stage as
they look on it now, but rather as a state of bestiality degrading for a human being . . .”
(p. 164).)
Surely, for many readers, the perspective on offer in The Kreutzer Sonata is one that
leaves them cold. Philosophers (Walton 1990, Gendler 2000, Weatherson 2004) have
provided varied accounts of the phenomenon of imaginative resistance—that is, read-
erly refusal, or inability, to follow the imaginative prescriptions within a fiction. But not
all readers will resist the invitation to take the perspective of Pozdnyshev, and to see
things as he does. A skilled storyteller, writing with the fevered passion of true convic-
tion, can make it seem that the world that he is giving us is, in fact, the world. And the
perspectives to which we are provided access via fiction must always be subject to extra-
fictional scrutiny; we cannot do without arguments, proceeding by way of objectively
accessible propositions, if we are to assess purported truth in fiction.

184
 TRU TH A N D REF ERE N C E I N F I C T I O N

Bibliography
Beardsley, M. (1981) Aesthetics: Problems in the Theory of Criticism, 2nd edition, Indianapolis: Hackett.
Bewley, M. (1954) “Scott Fitzgerald’s Criticism of America,” The Sewanee Review LXII (Spring).
Braun, D. (2005) “Empty Names, Fictional Names, Mythical Names,” Noûs 39, pp. 596–631.
Brock, S. (2002) “Fictionalism about Fictional Characters,” Nous 36, no. 1, pp. 1–21.
Byron, G. (1824) Don Juan, Cantos XII.–XIII.–-and XIV. New York: Charles Wiley.
Camp, E. (2009) “Two Varieties of Literary Imagination: Metaphor, Fiction, and Thought Experiments,”
Midwest Studies in Philosophy XXXIII.
Chomsky, N. (1988) Language and Problems of Knowledge: The Managua Lectures, Cambridge, MA: The MIT
Press.
Currie, G. (1990) The Nature of Fiction, Cambridge: Cambridge University Press.
Donnellan, K. (1970) “Proper Names and Identifying Descriptions,” Synthese 21, pp. 335–358.
Fitzgerald, F.S. (1925) The Great Gatsby, New York: Scribner’s.
Friend, S. (2000) Review of Thomasson (1999), Mind 109, pp. 997–1000.
Gendler, T. (2000) “The Puzzle of Imaginative Resistance,” The Journal of Philosophy 97, no. 2, pp. 55–81.
Harcourt, E. (2010) “Truth and the ‘Work’ of Literary Fiction,” British Journal of Aesthetics 50, no. 1,
pp. 93–97
Jackson, F. (1982) “Epiphenomenal Qualia,”Philosophical Quarterly 32: pp. 127–136.
Kripke, S. (1972) Naming and Necessity, Oxford: Blackwell.
Lamarque. P. (1981) “How Can We Fear and Pity Fictions?” British Journal of Aesthetics 21, pp. 291–304.
Lamarque, P. and Olsen, S. H. (1994) Truth, Fiction, and Literature: A Philosophical Perspective, Oxford:
Clarendon.
Lewis, D. (1978) “Truth in Fiction,” American Philosophical Quarterly 15, pp. 37–46. In Philosophical Papers,
vol. 1 (1983), Oxford: Oxford University Press.
Meinong, A. (1904) “On the Theory of Objects,” in R. Chisholm (ed.) Realism and the Background of
Phenomenology (1960) Atascadero: Ridgeview.
Nagel, T. (1974) “What Is It Like to Be a Bat?” Philosophical Review 83, pp. 435–50.
New, C. (1996) “Walton on Imagination, Belief and Fiction,” British Journal of Aesthetics 36 no. 2,
pp. 159–165.
Nichols, S. (2004) “Imagining and Believing: The Promise of a Single Code,” The Journal of Aesthetics and
Art Criticism, Special Issue on Art, Mind, and Cognitive Science.
Nussbaum, M. (1985) “ ‘Finely Aware and Richly Responsible’: Moral Attention and the Moral Task of
Literature,” The Journal of Philosophy 82, no. 10.
Parsons, T. (1980) Non-Existent Objects, New Haven: Yale University Press.
Rolfe, G. (2002) “ ‘A Lie That Helps us See the Truth’: Research, Truth and Fiction in the Helping
Professions,” Reflective Practice 3, pp. 89–102.
Russell, B. (1919) Introduction to Mathematical Philosophy, London: Allen and Unwin.
Salmon, N. (1998) “Nonexistence,” Noûs 32, no. 3, 277–319.
Searle, J. (1975) “The Logical Status of Fictional Discourse,” New Literary History 6, No. 2, pp. 319–332.
Silver, M. (2010) “The Rules of the Game,” posted on The Elegant Variation weblog, http://marksarvas.blogs.
com/elegvar/2010/04/the-rules-of-the-game.html
Thomasson, A. (1999) Fiction and Metaphysics, Cambridge: Cambridge University Press.
—— (2003) “Speaking of Fictional Characters,” Dialectica vol. 57, no. 2, pp. 207–226.
—— (2009) “Fictional Entities,” in J. Kim and E. Sosa (eds.) A Companion to Metaphysics, Blackwell, pp. 10–17.
Tolstoy, L. (1889) The Kreutzer Sonata (1890) “Epilogue to The Kreutzer Sonata,” reprinted in Six Short
Masterpieces by Tolstoy, translated by M.Wettlin, New York: Dell Publishing Co.
van Inwagen, P. (1977) “Creatures of Fiction,” American Philosophical Quarterly 14, no. 4, pp. 299–308.
Walton, K. (1990) Mimesis as Make-Believe, Cambridge, MA: Harvard University Press.
—— (1991) “Reply to Reviewers,” Philosophy and Phenomenological Research 51, no. 2, pp. 413–431.
Weatherson, B. (2004) “Morality, Fiction, and Possibility,” Philosophers’ Imprint 4, no. 3.
Weitz, M. (1975) “Literature without Philosophy: ‘Antony and Cleopatra’,” in K. Muir (ed.), Shakespeare
Survey, Cambridge: Cambridge University Press, pp. 29–36.
Wolterstorff, N. (1980) Works and Worlds of Art, Oxford: Clarendon Press.
Yagisawa, T. (2001) “Against Creationism in Fiction,” Philosophical Perspectives 15, pp. 153–172.
Zalta, E. (1983) Abstract Objects, Dordrecht: Reidel.
Zemach, E. (1993) “Existence and Nonexistents,” Erkenntnis 39, pp. 145–166.
—— (2003) Review of Thomasson (1999), The Philosophical Review 112, no. 3, pp. 427–431.

185
 Section II

FOU N DAT I O N S O F
SEM A N T I C S
 2.1
RE FE R ENC E
Teresa Robertson

1 Introduction: Reference, Semantic Content, and the

View that Content is Reference
Reference may be the most basic of all semantic relations. It is the relation that an
expression (in a particular context) bears to the thing (or things) for which it stands.
(From here on, reference to contexts is suppressed unless suppression is apt to cause con-
fusion.) The clearest examples of this relation in natural language are those involving
the bare demonstratives ‘this’ and ‘that’. When one says, gesturing to the coffee in one’s
cup, “This is good and strong,” the expression ‘this’ refers to the coffee in the cup. Simi-
larly, when I say, indicating Seattle, “That’s my hometown,” the expression ‘that’ refers
to Seattle. Here are some other natural (which is not to say uncontroversial) examples
of the reference relation: The proper name ‘Socrates’ refers to (or in other terminology
designates or denotes) the philosopher Socrates; the definite description ‘the philosopher
executed for corrupting the youth of Athens’ refers to him too; and the pronoun ‘I’ (in
a suitable context) refers to me. The expressions in the five examples just given are all
naturally thought of as singular terms, terms whose fundamental semantic role is to refer
to a single individual. Paradigmatic singular terms are the individual constants and
variables of formal logic. Because the notion of reference finds its natural home in the
relation between expressions for single individuals and the individuals themselves, the
bulk of the philosophical work on reference has been concerned with singular terms or
expressions that are at least arguably singular terms.
Interest in reference has been intimately bound up with interest in meaning, specifi-
cally with interest in semantic content. The philosopher’s notion of semantic content
is a thoroughly intuitive one. (Again, something may be intuitive or natural or naive
but nonetheless controversial.) Consider the English sentence ‘Snow is white’ and the
French sentence ‘La neige est blanche’. Intuitively, these sentences say the same thing.
The single thing that both sentences say (or mean or express) is the proposition that snow
is white. It is this proposition that is said to be the semantic content (or simply the con-
tent) of each of these declarative sentences. It is natural to say that what those sentences
express is true. Taking appearances at face-value, propositions are taken to be bearers of
truth-value (that is, things that are true or false). For similar reasons, they are taken to
be bearers of modal properties (like necessity and contingency) and epistemic proper-
ties (like being knowable a priori or being knowable only empirically). They are also
taken to be objects of other attitudes (that is, in addition to being taken to be objects of
knowledge, they are taken to be objects of assertion, belief, doubt, and the like).
 TERESA RO B E RT S O N

Propositions are commonly thought of as structured entities with component parts.

Consider the propositions expressed by the following sentences: ‘Socrates is wise’, ‘Soc-
rates is Greek’, and ‘Plato is Greek’. Intuitively the proposition expressed by the second
sentence shares a component with the proposition expressed by the first sentence—a
component that is associated with the word ‘Socrates’—in virtue of which each propo-
sition is concerned with the individual Socrates. Similarly, the proposition expressed
by the second sentence intuitively shares a component with the proposition expressed
by the third sentence—a component that is associated with the phrase ‘is Greek’—in
virtue of which each proposition is concerned with the property of being Greek. Con-
sideration of the propositions expressed by slightly more complex sentences like ‘Socra-
tes taught Plato’ and ‘Plato taught Socrates’ reveals that propositions must be structured
entities: if they were not, then the two propositions, which are clearly different since
one is true and the other false, would not be distinguishable since both would simply
consist of the same components. It is natural to think of the structure of the proposition
as at least roughly isomorphic to the structure of the sentence that expresses it.
The idea that the content of a compound expression is a function of the contents of
its parts together with the way those parts are structured is nearly irresistible—at least
as a general rule. There are numerous exceptions to this rule. For example, whatever
the content of ‘Socrates’ is—that is, whatever ‘Socrates’ typically contributes to the
content of sentences in which it occurs—that thing is evidently no part of the follow-
ing sentence:

‘Socrates’ contains eight letters.

Nonetheless it is safe to assume, as most work in semantics does, that this rule holds by
and large. I will take it for granted in this chapter. In fact, for simplicity, I will typically
assume something stronger, namely that by and large the content of a compound expres-
sion is a structured entity composed of the contents of the expression’s contentful parts.
I will say that an expression is compositional when its content is a function of the con-
tents of its parts, together with the way those parts are put together (or, more strongly,
when its content is a structured entity composed of the contents of the parts).
The most straightforward theory about content says simply that content is reference.
On this theory, the proposition that is the content (and referent) of the sentence ‘Soc-
rates is wise’ has as its components the individual Socrates (which is the referent of
‘Socrates’) and the property of being wise (which can, at least for present purposes, be
taken to be the referent of ‘is wise’). This theory unites the semantic relations of refer-
ring and expressing.

2 Trouble over Definite Descriptions

Two of the naive ideas encountered in §1, namely that definite descriptions are singular
terms and that content is reference, come into conflict with one another. Consider
again the definite description ‘the philosopher executed for corrupting the youth of
Athens’. Intuitively, it has contentful parts—parts which themselves have semantic
content. And intuitively descriptions are compositional. (Again, exceptions may arise.
Consider the description ‘the number of letters in “Socrates”’. Whatever the content
of ‘Socrates’ is, it is no part of the content of ‘the number of letters in “Socrates”’. But
here it is the quotation marks and not some feature common to all definite descriptions

190
 REF EREN C E

that is responsible for the exception.) Accepting these intuitions at face-value reveals
that we cannot accept both the straightforward semantic theory that content is refer-
ence and the natural thought that ‘the philosopher executed for corrupting the youth
of Athens’ is a singular term that refers to Socrates: According to modern anatomy at
least, Socrates is not himself a complex entity composed of the semantic contents of
the contentful parts of ‘the philosopher executed for corrupting the youth of Athens’.
There are two obvious ways to deal with this problem, and historically both have had
advocates: Bertrand Russell (Russell 1905) rejected the claim that definite descriptions
are singular terms and John Stuart Mill (Mill 1893) rejected the claim that content is
reference. I will sketch both of these kinds of responses, though I will not aim to discuss
the particular views of Russell and Mill themselves. I will use ‘Russell1*’ and ‘Mill*’ as
labels for the views I discuss.

3 Definite Descriptions: Russell1* and Mill*

Russell1* resolves the trouble over definite descriptions by retaining the thought
that content is reference but giving up the thought that definite descriptions are sin-
gular terms. Russell1* assimilates definite descriptions to other determiner phrases
(phrases of the form ‘every F’, ‘some Fs’, ‘most Fs’, etc.), which are viewed as restricted
quantifier phrases. So far, Russell agrees with Russell1*. Russell himself thought that
such phrases did not have content “in isolation”; however on a slightly more natural
view, a restricted quantifier’s content is a higher-order property—perhaps a property
of properties or a property of classes. For example, the sentence ‘Every philosopher is
Greek’ may be thought of as attributing to the property of being Greek the property
of being (a property that is) instantiated by everything that instantiates the property
of being a philosopher, so that the attributed property (of properties) may be thought
of as the content of ‘every philosopher’. (Roughly this idea is due to Richard Mon-
tague (Montague 1974).) Similarly, a sentence like ‘The philosopher executed for
corrupting the youth of Athens was Greek’ may be thought of as attributing to the
property of being Greek the property of being instantiated by something that uniquely
instantiates the property of being a philosopher executed for corrupting the youth of
Athens.
In contrast, Mill* resolves the trouble over definite descriptions by retaining the
thought that definite descriptions are singular terms but giving up the thought that
content is reference. The idea is that there are (at least) two kinds of semantic value—
content and reference—and that the content of a definite description is not its referent
but instead some complex entity composed of the semantic contents of the description’s
contentful parts. On this view, the description ‘the philosopher executed for corrupt-
ing the youth of Athens’ has as content some complex involving such things as the
property of being a philosopher, the property of being executed, and so on, but the view
retains the natural thought that the description’s referent is Socrates. Deviating as lit-
tle as necessary from the naive theory that content is reference, Mill* allows that some
expressions, and in particular proper names, have as their contents their referents. (It
is worth stressing that one can recognize a thoroughgoing distinction between content
and referent, as Mill* does, while allowing that for some expressions the very same
entity is both. Compare: there is a thoroughgoing distinction between sister and wife,
although for Zeus the same entity is both.)

191
 TERESA RO B E RT S O N

4 Challenges for Russell1* and Mill*

Though the theories of Russell1* and Mill* are very different in their theoretical out-
looks, with the former but not the latter accepting that content is reference, they share
a commitment to the claim that the semantic content of a proper name is its referent.
This claim faces challenges in the form of various puzzles that can be grouped into two
kinds: puzzles involving substitution of coreferential names and puzzles involving empty
names (that is, names that lack referents).
Let’s first look at the substitution puzzles. My father’s given name is ‘Charles’, but for
reasons that are lost to time, people in my family usually call him ‘Pete’. Now consider
the following two sentences.

(1) Charles is Charles.

(2) Pete is Charles.

Assuming that these sentences are compositional, the theories of Russell1* and Mill*
ascribe the same content to both sentences, because they ascribe the same content to
‘Charles’ and ‘Pete’. Or, more fully, because the components of the proposition expressed
by each sentence are the same (my father, the relation of identity, and my father again)
and they are put together in the same way. But, the sentences do not seem to express the
same proposition, since the proposition expressed by (1) is not informative whereas the
proposition expressed by (2) is. This is the puzzle of informativeness presented by Gottlob
Frege (Frege 1892). He presented the puzzle using identity sentences, as I have done,
but that is not crucial to the puzzle, as the following sentences reveal.

(3) Charles is a person, if Charles is a person.

(4) Pete is a person, if Charles is a person.

Again the theories of Russell1* and Mill* ascribe the same content to both sentences,
yet intuition tells us that (3) and (4) express different propositions—one that is not
informative and one that is.
There are a number of similar puzzles that vary from Frege’s only in what particular
property of propositions they highlight. So, for example, one can generate a puzzle con-
cerning epistemic status (rather than informativeness) by noting that intuitively the
proposition expressed by (3) is knowable a priori, whereas the proposition expressed by
(4) is not. Similarly one can generate a puzzle concerning belief by noting that although
Clark Kent is Superman, Lois believes the proposition expressed by (5) but not the one
expressed by (6).

(5) Clark Kent is a reporter.

(6) Superman is a reporter.

(Here I am treating the comic book fiction as fact.) (A family of closely related substi-
tution puzzles involves sentences of the form ªIt is informative that Sº, ªIt is knowable
a priori that Sº, ªSo and so believes that Sº, and so on, where the numbered sentences
above are examples of substituends for ‘S’. Because these sentences involve the operator
‘that’, the full statement of these puzzles would have to address issues that are beyond
the scope of this chapter. In brief, ‘that’ appears to operate somewhat as quotation marks

192
 REF EREN C E

do, so that it takes significantly more than compositionality to underwrite the claims
that according to Russell1* and Mill*, ‘Lois believes that Clark Kent is a reporter’ and
‘Lois believes that Superman is a reporter’ express the same proposition. Nor does com-
positionality provide enough to underwrite the claim that according to Russell1* and
Mill* those two sentences have the same truth-value. Those claims can be adequately
supported, but it takes additional theory to do so.)
It is important to understand that the substitution puzzles that arise for Russell1* and
Mill* do not arise when definite descriptions are involved. These theories are commit-
ted to saying that coreferential names have the same semantic content, but they are not
committed to saying that coreferential descriptions have the same semantic content
or to saying that a name and a coreferential description do. (Here I use “referential”
in a loose way, a way according to which even Russell1*—who strictly speaking thinks
that the referent of a definite description is a higher-order property rather than the
individual that is intuitively its referent—can say that ‘the philosopher executed for
corrupting the youth of Athens’ and ‘Socrates’ are coreferential.) And this allows these
theories to resist the claims that sentences like (7), (8), and (9) below have the same
content, even though Benjamin Franklin was the first postmaster general and also the
inventor of bifocals.

(7) Benjamin Franklin was interested in science.

(8) The first postmaster general was interested in science.
(9) The inventor of bifocals was interested in science.

This means that no analogs of the substitution puzzles involving names arise in the case
of definite descriptions.
Now, let’s look at the most general of the puzzles involving empty names. Intuitively,
the name ‘Santa’ has no referent. Nonetheless, sentences involving it appear to be con-
tentful. Consider the following sentence.

(10) Santa is jolly.

What is the content of this sentence according to Russell1* and Mill*? An entity com-
posed of the referent of ‘Santa’ and the property of being jolly. But ‘Santa’ does not have
a referent. So how can (10) be contentful? One natural response to this question is that
the proposition expressed by (10) must be some kind of “gappy” structured entity, with a
gap where there should be something to which ‘Santa’ refers. But this solution founders
when we consider the following sentence.

(11) Tinkerbell is jolly.

Assuming that ‘Tinkerbell’ has no referent, (11) must also express a gappy proposition;
in fact the same one expressed by (10). But intuitively (10) and (11) do not express the
same proposition. (There are other puzzles involving empty names that are less general
than this one. The most famous of these is the puzzle of true negative existential statements,
which involve sentences like ‘Santa does not exist’. A full discussion of this and the other
more specific puzzles involving empty names is beyond the scope of this chapter.)
Again, it is important to understand that for Russell1* and Mill* there is no analog
to the general puzzle involving empty names that involves instead empty descriptions.

193
 TERESA RO B E RT S O N

For both Russell1* and Mill* a description like ‘the largest prime’ has content: for Rus-
sell1*, its content is something like the property of being instantiated by something that
uniquely instantiates the property of being a prime greater than all others; for Mill*, its
content is some complex involving the semantic contents of the semantically content-
ful constituents of the description.

5 Addressing the Challenges for Russell1* and Mill*:

Russell2* and Frege*
Since the puzzles that arise for Russell1* and Mill* do not arise when descriptions are
involved, the puzzles can be solved by assimilating names to descriptions. This is accom-
plished in a way that retains much of the theoretical outlook of Russell1* (in particular
adherence to the simple theory that content is reference) by Russell2*, who maintains
that names are really just definite descriptions in disguise (Russell 1910–1911). (Again,
my use of a “starred name” indicates that my aim is not to capture the view of a par-
ticular historical figure so much as to tell a purely philosophical story in which figures
a view that bears important similarities to the view of the referent of the “unstarred
name.”) On this view, a name like ‘Santa’ is semantically equivalent to a description
like ‘the fat jolly old elf who delivers presents to good children on Christmas’, which in
turn is thought of as a restricted quantifier phrase. The assimilation of names to descrip-
tions is accomplished in a way that retains much of the theoretical outlook of Mill* by
Frege*, who takes Mill*’s recognition of a distinction between content and referent to
its extreme, holding that for every expression, that which serves as content includes a
mode of presentation of that which serves as referent (Frege 1892). The content is thus
distinct from the referent, at least typically. On this view, names, like descriptions, have
contents that are conceptual in nature. The name ‘Santa’ has as its semantic content
some complex involving such things as being fat, being jolly, and so on, so that ‘Santa’
is semantically equivalent to some definite description like ‘the fat jolly old elf who
delivers presents to good children on Christmas’.

6 Trouble for Russell2* and Frege*

Though the theories of Russell2* and Frege* provide solutions to the puzzles encoun-
tered in §4, there are three powerful arguments against these theories: the modal argu-
ment, the epistemic argument, and the semantic argument. (The first two arguments are
due to Saul Kripke (Kripke 1972/1980). The last is due to Keith Donnellan (Donnellan
1972) and independently to Kripke.) Somewhat surprisingly, two of these arguments
(the modal and epistemic) depend on considerations that are remarkably similar to
those involved in the puzzles of §4.
Let’s suppose with Russell2* and Frege* that the content of ‘Gödel’ is the same as the
content of some description, say ‘the discoverer of the incompleteness of arithmetic’.
Now consider the following two sentences.

(12) Gödel (if he exists) discovered the incompleteness of arithmetic.

(13) The discoverer of the incompleteness of arithmetic (if there is exactly
one) discovered the incompleteness of arithmetic.

The modal argument points out that if ‘Gödel’ and ‘the discoverer of the incompleteness

194
 REF EREN C E

of arithmetic’ have the same content, then (12) and (13) express the same proposition.
(The argument assumes that these sentences are compositional.) But, the argument
continues, the proposition expressed by (12) is not necessary, whereas the proposition
expressed by (13) is. Hence (12) and (13) do not express the same proposition. Hence
‘Gödel’ and ‘the discoverer of the incompleteness of arithmetic’ do not have the same
content.
The epistemic argument is very similar to the modal argument. Both arguments have
the same general form and reach the same conclusion. In place of the modal argu-
ment’s appeal to the fact that the proposition expressed by (12) is not necessary whereas
the proposition expressed by (13) is, the epistemic argument appeals to the fact that
the proposition expressed by (12) is not knowable a priori, whereas the proposition
expressed by (13) is.
The semantic argument, which is sometimes confused with the modal argument,
is a very different sort of argument from the modal and epistemic arguments. Suppose
that in fact the person who we think discovered the incompleteness of arithmetic
(the man we call ‘Gödel’) did not in fact discover the incompleteness of arithmetic
but rather stole the result from his neighbor (a man whom we call ‘Schmidt’). Who
would the name ‘Gödel’ refer to—the thief or the neighbor? Clearly, it would refer
to the thief. (Or, not to put too fine a point on it, ‘Gödel’ would still refer to Gödel
and not to Schmidt.) Who would ‘the discoverer of the incompleteness of arithme-
tic’ refer to—the thief or the neighbor? Clearly, it would refer to the neighbor (that
is, Schmidt) even though we would in the circumstances mistakenly think that it
referred to Gödel. Thus, assuming that content determines referent, the content of the
name ‘Gödel’ is not the same as the content of ‘the discoverer of the incompleteness of
arithmetic’.
Given these problems for Russell2*-Frege* and the problems for Russell1*-Mill* that
were discussed earlier, it is natural to search for some third general view concerning
the semantic content of proper names. But it is evident that there is precious little
room between the view that the semantic content of a proper name is its referent and
the view that the semantic content of a proper name is the kind of thing that could be
expressed by some definite description. Nonetheless philosophers have attempted to
skirt the problems facing the dominant views by claiming that a proper name has as
its content a “nondescriptive sense”. (An early advocate of this approach was Felicia
Ackerman (Ackerman 1979).) Such a sense would have to be some sort of conceptual
entity that is associated with the name and that is such that the thing that falls under
the relevant concept would be the referent of the name, but such a sense would at the
same time have to be something that could not be captured by any definite description.
(A conceptual content that is not but that could be captured by a definite description
would not be a nondescriptive sense.) The exact nature of this line of thought remains
murky and consequently attractive to some. Philosophers have also tried more straight-
forward approaches. There are current theories in the spirit of Russell1*-Mill* that
address the problems of §4. (Prominent defenders of this line include Nathan Salmon
and Scott Soames. Soames (2002) modifies Salmon’s (1986) solution to the problems of
substitution, and endorses Salmon’s (1998) solution to the problems of empty names.)
And also there are numerous theories in the spirit of Russell2*-Frege* that attempt
to solve the problems of §6. (Michael Dummett (Dummett 1973) and Gareth Evans
(Evans 1982) gave early and influential responses to the arguments of Kripke and
Donnellan.) No consensus has emerged.

195
 TERESA RO B E RT S O N

7 Mechanisms of Reference
It has been thought that one advantage of the Russell2*-Frege* view over the
Russell1*-Mill* view is that the former has something to say about the mechanism by
which reference is secured while the latter stands in need of supplementation. To fill
this need, Kripke (1972, 1980) proposed the so-called causal-historical picture of ref-
erence. On this picture, a name may be introduced in a variety of ways. (Here are just
a few examples. Parents of a newborn may introduce a name for their child by decid-
ing on a name and acting on that decision. Similarly, the adoptive parents of an alien
baby with extraordinary superhuman powers may introduce a name for their child
by saying, “Let’s call him ‘Clark’.” A scientist may introduce a name for a particular
individual by saying, “I’ll use ‘Eve’ to name the most recent female ancestor common
to all living humans.” The ritual aspects of these dubbings are presumably brought to
a bare minimum when, for example, someone who lives alone names her pet, but even
workers at a law office may without ceremony simply begin using the name ‘Romeo’
to refer to whoever sent flowers to the managing partner.) Those who introduce the
name use it to refer to the relevant object. In communication, the name gets passed
on to other users, so that the name refers to that same object even in the idiolects of
the newer users. Spelling out the exact details of the nature of the causal chains that
link later uses to initial uses is a difficult empirical task. (For example, why is it that
although there are presumably causal chains linking my use of ‘St. Nicholas’ as well
as my use of ‘Santa Claus’ to early uses of a name of Nicholas of Myra, ‘St. Nicholas’
refers to that saint while ‘Santa Claus’ does not?) What is of philosophical interest is
that on this picture, in sharp contrast to the Russell2*-Frege* view, reference is deter-
mined contextually rather than conceptually.
While the causal-historical picture is just a picture and not a full-fledged theory
(since many empirical questions about the nature of the relevant causal chains remain
unanswered), the picture does provide enough of a supplement to the Russell1*-Mill*
view that it would be unfair to say that it remains a complete mystery on this view how
a name comes to be hooked up to its referent. And in spite of what is often said about
the Russell2*-Frege* view, there is in fact some mystery on this view about this issue.
While it is easy enough for the view to explain how a name’s associated description gets
connected to the name’s referent (since the referent is simply whatever satisfies the
description), the view stands in need of supplementation to explain how the name gets
connected to its associated description.

8 Reference and General Terms

Although the reference relation is most at home in connection with singular terms, it
is not much of a stretch to see the relation as holding between general terms and kinds.
For example, it is natural to think of the general terms ‘water’ and ‘tiger’ as referring to
a particular liquid substance and a particular species of animal respectively. (It is more
of a stretch to see what the analog of the notion of reference is for other expressions
such as predicates and sentences. An ingenious argument, the so-called slingshot, due
to Alonzo Church (Church 1943) and Kurt Gödel (Gödel 1944), gives strong reason
to think that the closest analog to singular term reference for a predicate is the class of
(n-tuples of) individuals to which it applies and that the closest analog for a sentence is
its truth value.) And in fact many of the issues that arose in the discussion of singular

196
 REF EREN C E

terms arise in much the same way for general terms. In particular, the modal, epistemic,
and semantic arguments against the view that a proper name has the same content
as a singular definite description apply equally against the view that a simple general
term, like ‘water’ or ‘tiger’, has the same content as a general definite description, like
‘the colorless, odorless, thirst-quenching liquid that fills the lakes and streams’ or ‘the
species that serves as mascot for Princeton University’. (I leave the details as an exercise
for the reader.)
Though much of the literature on general terms focuses on one-word general terms for
natural kinds, it is far from clear that such terms differ in semantically significant ways
from one-word general terms for nonnatural kinds. In particular, the modal, epistemic,
and semantic arguments can be run using, for example, ‘crayon’ and ‘the favorite writ-
ing instrument of Yoko’. (Hillary Putnam (Putnam 1975) using his Twin-Earth thought
experiment gave a version of the semantic argument in connection with natural-kind
terms. He also argued that artifact terms, like ‘pencil’, are not semantically equivalent
to any definite description.) This is not to deny that a one-word general term may
be synonymous with a compound common noun phrase: surely ‘bachelor’ is synony-
mous with ‘unmarried man’. This highlights a significant point of disanalogy between
terms for kinds and terms for individuals: It is commonplace to refer to kinds not only
with simple terms and definite descriptions but also with compound common noun
phrases like ‘unmarried man’ and ‘guy who won’t take no for an answer’. This fact raises
many questions. Given that ‘bachelor’ and ‘unmarried man’ are synonyms, should the
considerations that favor a Russell1*-Mill* view for simple-kind terms like ‘bachelor’
lead us to favor a Russell1*-Mill* view for compound-kind terms like ‘unmarried man’?
Or rather does the apparent fact that ‘unmarried man’ has descriptive content give us
a consideration in favor a Russell2*-Frege* view for ‘bachelor’? What impact should
answers to these questions have on views about singular terms? Are there singular term
analogs to kind-phrases? Suffice it to say that a more thorough understanding of work-
ings of both singular and general terms is needed to get to the bottom of the notion of
reference.

References
Ackerman, F., “Proper Names, Propositional Attitudes, and Nondescriptive Connotations,” Philosophical
Studies (1979), 55–69.
Church, A., “Review of Carnap’s Introduction to Semantics,” Philosophical Review 52 (1943), 298–304.
Donnellan, K., “Proper Names and Identifying Descriptions,” in D. Davidson and G. Harman (eds.), Seman-
tics of Natural Language (Dordrecht: D. Reidel, 1972), 356–379.
Dummett, M., Frege: Philosophy of Language (New York: Harper and Row, 1973).
Evans, G., Varieties of Reference (Oxford: Oxford University Press, 1982).
Frege, G., “Über Sinn und Bedeutung,” Zeitschrift für Philosophie und Philosophische Kritik 100 (1892), 25–50.
English translation in M. Beaney (ed.), The Frege Reader (Oxford: Blackwell, 1997), 151–171.
Gödel, K., “Russell’s Mathematical Logic,” in P.A. Schilpp (ed.) The Philosophy of Bertrand Russell (New
York: Tudor, 1944), 123–154.
Kripke, S., Naming and Necessity (Harvard University Press and Blackwell, 1972/1980).
Mill, J.S., “Of Names,” in Book I, Chapter II of A System of Logic (New York: Harper and Brothers, 1893),
29–44.
Montague, R., “The Proper Treatment of Quantification in Ordinary English,” in J. Hintikka, J. Moravcsik,
and P. Suppes (eds.), Approaches to Natural Language (Dordrecht: Reidel, 1973), 221–242. Reprinted in
Montague (1974), 247–270.
Montague, R., Formal Philosophy: Selected Papers of Richard Montague, edited and with an introduction by
R. Thomason (New Haven: Yale University Press, 1974).

197
 TERESA RO B E RT S O N

Putnam, H., “The Meaning of ‘Meaning’,” in K. Gunderson (ed.) Minnesota Studies in the Philosophy of Science
VII: Language, Mind, and Knowledge (Minneapolis: University of Minnesota Press, 1975), 130–193.
Russell, B., “On Denoting,” Mind 14 (1905), 479–493.
Russell, B., “Knowledge by Acquaintance and Knowledge by Description,” Proceedings of the Aristotelian
Society 11 (1910–1911), 108–128.
Salmon, N., Frege’s Puzzle, (Cambridge, MA: MIT Press, 1986).
Salmon, N., “Nonexistence,” Noûs 32 (1998), 277–319.
Soames, S., Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity, (Oxford: Oxford
University Press, 2002).

198
 2.2
THE OR IE S OF T R UT H
Matti Eklund

Questions about truth are central to several different parts of philosophy. In what
follows I will summarize central parts of the debates over truth in metaphysics, in phi-
losophy of language, and, more briefly, in philosophy of logic. Throughout I will also be
concerned with the relations between these different debates.

The Metaphysics of Truth

A preliminary question for all discussions of truth is what the primary bearers of the
properties of truth and falsity are: are they propositions or sentences, or something else
altogether, like acts of judgment? For the purposes of the first part of this discussion, I
will assume that propositions are the primary truth-bearers.
The first question I will actually pause on is one which arguably underlies the debate
over the traditional main theories of truth: the correspondence theory, the coherence
theory, the identity theory, and deflationism. By virtue of what are propositions true
and false, respectively? This is a question concerning the relationality of truth. We can
distinguish four main types of answer: (a) propositions are true or false by virtue of how
things stand in extra-propositional reality (except when they are about other proposi-
tions); (b) propositions are true or false by virtue of relations they stand in to other
propositions; (c) the truth or falsity of a proposition is determined by intrinsic features;
(d) the whole question of what a proposition’s truth depends on is misbegotten (Wright
1999).
Alternative (a) should seem the most plausible. Even without getting into details
regarding what a specific theory on which (b) is correct might look like, one might
well think that when a proposition concerns how things stand in nonpropositional
reality, its standing in the right relations to other propositions is not sufficient to make
it true. Explanation (c) obviously runs into trouble with contingency: If the truth of a
proposition is an intrinsic feature, how can a proposition have different truth-values
with respect to different possible scenarios? I will turn to the dismissive suggestion (d)
below.
Alternatives (a) through (d) correspond roughly to the important traditional theo-
ries of truth: (a) spells out the correspondence idea; (b), when supplemented by the claim
that the relation in question is coherence, spells out the idea of coherence; (c) suggests
the identity theory, on which true propositions do not correspond to but simply are facts
(but see below); and (d) should be recognizable as something deflationists about truth
are apt to hold.
 M ATTI EKL U N D

There are ways of conceiving of a correspondence theory of truth, given which it

does not seem as plausible as it does if we take the view to be captured by (a). A
correspondence theory is often formulated as the thesis that truth consists in corre-
spondence with the facts, and to be committed to an ontology of facts. This gives rise to
a number of problematic questions—for example, concerning the existence of nega-
tive and hypothetical facts. However, one can say, consistently with (a), that what
makes a given proposition true: say, that a is not F means simply that a fails to be F.
There is no obvious need to appeal to such entities as facts. Second, a correspondence
theory is sometimes claimed to be bound up with metaphysical realism. But (a) is not so
bound up. One can say that a proposition is true by virtue of how things stand in non-
propositional reality, while holding that nonpropositional reality is mind-dependent
in a way that is incompatible with realism. Third, some are skeptical of a correspond-
ence theory on the grounds that it is hard to elucidate the notion of correspondence
involved. It may be tempting to understand correspondence as having something to
do with similarity. But can propositions really, in any interesting sense, be similar to
facts? And if we are not to elucidate the relation of correspondence by saying that it
is a matter of similarity, just how can correspondence be elucidated? However, it must
be said that (a) can be true even if nothing more informative can be said. Fourth, in
light of the above points, and in light of the fact that (a) doesn’t by itself provide any-
thing like a reductive characterization if truth, it may be thought that (a) by itself is
too modest to count as a correspondence theory. But, importantly, (a) does seem to be
in conflict with natural ways of spelling out other conceptions of truth, and (a) goes
beyond some truly modest understandings of correspondence: for example those in
which a correlation in truth of the proposition that p is true and p is sufficient for cor-
respondence. Answer (a) adds to this that the truth of the proposition depends on what
holds in nonpropositional reality.
Now turn to (b). The coherence theory, according to which truth is coherence with
other propositions, is clearly one way to spell out (b). The coherence theory faces an
immediate objection, known as the Bishop Stubbs objection, from Bertrand Russell’s
example: any proposition—including obviously false ones like the proposition that
Bishop Stubbs was hanged for murder—is part of some coherent collection of proposi-
tions, but coherence with other propositions is not sufficient for truth (Russell 1910).
The coherence theorist’s response is to insist that truth consists in coherence with a
privileged collection of propositions. The question is whether that is a workable way
out. In his study of coherence theories of truth, Ralph Walker (1989) has prominently
argued that there is still a problem: what accounts for the truth of the proposition that
a given collection of propositions is privileged?
Coherence theory is often presented as an historically important competitor to corre-
spondence theory. But the following complication must be stressed. Those traditionally
counted as coherence theorists of truth—prominently the British idealists (for example,
Joachim (1906) and Bradley (1909))—took reality itself to be an organic whole and so to
be characterized by its coherence. Given this view on reality one can emphasize coher-
ence without being a coherence theorist of truth in the sense of (b). For example, one
can use this view on reality to deny that individual propositions are ever (fully) true and
insist instead that only a system of propositions displaying a certain coherence can be
true. But even that wouldn’t be to embrace (b). It is only that proper correspondence
obtains only between larger units than it is otherwise conceived of as obtaining between,
and the units are themselves characterized by coherence. Another characteristic view

200
 TH EO RI ES O F T R U T H

of the idealists was, precisely, idealism, the view that reality is mind-dependent. But it
is possible to insist that reality is somehow mind-dependent, but that a proposition and
the reality which makes it true are two distinct elements of this mind-dependent reality,
and that the proposition is made true by the fact. (There is a debate in the literature
over whether the British idealists really did commit themselves to a coherence theory.
See, for example, Candlish (1989).)
As mentioned, alternative (c) sounds like what is otherwise known as the identity
theory of truth. But there is some reason for caution. The identity theory says that true
propositions are identical to facts. But there are two important ways to interpret this
idea. Either one can take facts to be primary, and use the theory to reduce true proposi-
tions to facts. Or one can take propositions to be primary, and take facts to be no more
than true propositions. In the former case, there may be some reason to think that
truth is intrinsic to the truth-bearer. But there is an obvious problem concerning how
to deal with falsity. False propositions cannot be facts. Hence, an identity theory of the
first kind is committed to an oddly disjunctive treatment of true and false propositions.
suggestion. The second suggestion faces no problem concerning the existence of false
propositions, but it is no longer clear that the identity theory in any way suggests the
idea that truth is intrinsic to the truth-bearers. The identity theory thus conceived is
fully compatible with the idea that the proposition that p is true (the fact that p obtains)
because of how things stand in propositional reality.
If we reject alternatives (b) and (c), the only competitor to (a) is (d). And (d) might,
especially when (a) is divorced from other commitments of the correspondence theory,
sound unmotivated. If we can simply say that the proposition that a is F is made true by
a’s being F, why accept the defeatist suggestion (d)? But here is one possible considera-
tion in favor of (d). One may think that underlying the attractiveness of (a) is the idea
that the proposition that p is true depends on it being the case that p; but it is necessary
that the proposition that p is true iff p, and one might generally be skeptical of the idea
of a dependence relation obtaining between p and q when p and q of necessity obtain
together. Given such skepticism—a kind of skepticism of the hyperintensional—there
is then reason to favor (d) over (a).
Trenton Merricks (2007) holds that while claims of the form that p is true because p
are all true, the dependence expressed by these claims is “trivial.” and truth does not
depend on the world in any “substantive” way. He considers and rejects some more sup-
posedly substantive claims. One is Truthmaker, the claim that for every true claim there
is some entity whose existence makes that claim true. Merricks rejects this, appealing,
for example, to negative existentials. Another is the claim that truth supervenes on
being. This thesis receives several precisifications. Merricks argues that under one pre-
cisification it is problematic in the way Truthmaker is alleged to be and under another
precisification that it fails to spell out the dependence claim. Do Merricks’ arguments
pose problems for (a)? This is unclear; and it is unclear whether Merricks would even
intend them to. Why should the so-called trivial dependence not be sufficient for (a)
to be true? Alternatively, if one does find the so-called trivial dependence troublingly
trivial, one can contend that Merricks’ menu of supposedly more substantive depend-
ence claims is disappointingly short. The point is often made that mere supervenience
claims are not strong enough to capture dependence claims (McLaughlin and Bennett
(2005): section 3.5). As for Truthmaker, why even contemplate that dependence upon
nonpropositional reality should amount to dependence upon the existence of some
particular entity or entities in nonpropositional reality?

201
 M ATTI EKL U N D

Alternative (d), the rejection of the question of what makes a proposition true, is a
deflationary suggestion, one naturally associated with deflationism. But while this view
will be recognized as something a deflationist may hold concerning truth-making, it is,
importantly, not the canonical characterization of deflationism. The canonical charac-
terization is rather that the content of the notion of truth is somehow exhausted by the
equivalence schema,
(ES) The proposition that p is true iff p
or its counterpart for sentences, the disquotation schema,
(DS) “p” is true iff p,
or some more sophisticated variant of one of these schemata. (Sometimes “minimalism”
is used for the proposition-based variant of deflationism and “disquotationalism” for the
sentence-based one.) Let me next focus on (ES), still assuming that propositions are
the primary truth-bearers. In the next section I will discuss what deflationism might
come to.

Deflationism
What might it be for the instances of (ES) to exhaust the concept of truth? One straight-
forward thing it might mean is that these instances jointly, implicitly define ‘true’. But
with the deflationist’s characterization of truth thus understood, it is not obvious that
deflationism should be understood to entail a deflationary view on the question of the
relationality of truth. Why shouldn’t the implicit definition have as a consequence that
‘true’ stands for a property such that whether a proposition has this property depends
on nonpropositional reality? There is the illusion, engendered by some of the rhetoric
surrounding the deflationist’s characterizations of truth, that if the concept of truth is
characterized as indicated, it follows that the property of being true is not a full-blooded
property. But it is generally not the case that if a predicate is introduced by the stipu-
lation that it makes such-and-such sentences true or such-and-such schemata valid,
the predicate does not stand for a (full-blooded) property. Instead, the natural conclu-
sion is simply that the predicate stands for a property which satisfies the conditions
mentioned.
Generally, deflationism tends to be characterized in a number of different ways.
Sometimes it is characterized by reference to specific types of characterizations of the
truth predicate, such as those based around (ES) or (DS). Sometimes it is characterized
metaphysically, as it were, as the claim that truth isn’t a property, or is only a ‘logi-
cal’ property, or does not have an underlying nature. Sometimes—compare the above
discussion of the relationality of truth—it is characterized by its rejecting as mis-
conceived the metaphysical questions to which other traditional theories of truth
seek to provide answers. Sometimes in discussions of the philosophy of language it is
characterized by its denial of the claim that truth plays any sort of theoretical or
explanatory role. It should be clear that the relations between these claims can be
problematized.
Suppose that a would-be deflationist insists that when she says that (ES) exhausts the
notion of truth she does not mean that it exhausts the concept of truth, but rather that
it exhausts the property of truth. What might this amount to? Surely there will be other
facts about the property, not entailed by the instances of (ES) alone, such as that the

202
 TH EO RI ES O F T R U T H

proposition that snow is white is true. So there is a sense in which the instances of (ES)
clearly do not fully exhaust the property. What else might the claim amount to?
One suggestion is that the view is best understood as the claim that truth satisfies a
principle of revelation (Damnjanovic 2010; Wright 2001: 753). The property of truth is,
according to the claim, fully revealed to us when we grasp the concept of truth. In this
it contrasts with water. The property of being water is not fully revealed to us when we
grasp the concept of water: Further investigation is needed to figure out the nature of
this property, namely H2O. But what does the appeal to revelation amount to? Suppose
first that the claim is that the property of truth is such that there is a concept of that
property which when grasped fully reveals the property. There is reason to think that
this cannot be an adequate explication. Is not every property like that? Isn’t it the case
that for every property there is some possible concept—whether one in our actual reper-
toire or not—competence with which fully reveals the property? Compare: water has an
underlying nature; and the concept water is a paradigmatic example of a concept which
doesn’t fully reveal the property it stands for, the property of being H2O. But for all that,
there can be some concept which fully reveals this property—indeed, the concept H2O
may be such a concept.
Suppose that instead we take the revelation claim to be that truth is the kind of prop-
erty such that for every concept of a property of this kind, a grasp of this concept fully
reveals this property. But isn’t this claim implausibly strong? Grasp of the concept of
truth may fully reveal the property this concept stands for. But even so, what might be
the reason for thinking that this applies to every concept of this property?
Someone focusing on the revelation claim can instead urge that the view is that our
ordinary concept of truth fully reveals the property it stands for. But this claim doesn’t by
itself yield a distinctive view on the property of truth, if for all properties there are thus
revelatory concepts. All we are told is that we happen to have a revelatory concept of
this property, while we lack revelatory concepts of other properties.
One way to deal with the problems encountered is to identify deflationism as the
view that the only essential feature the property of being true is that the instances of
(ES) are all true, where ‘essential’ does not mean simply ‘metaphysically necessary’.
(Fine (1994) forcefully presents the case for a notion of essence that cuts more finely
than the notion of metaphysical necessity.) The concept of truth, understood as the
deflationist understands it, is then fully revelatory of the property in the sense that grasp
of the concept provides knowledge of all essential features of the property. But talk of
revelation is then not needed to characterize deflationism: The claim about the essen-
tial features of the property suffices. Notice, however, an irony. Deflationism is often
supposed to make truth metaphysically innocuous and safe. But understood as proposed,
deflationism carries ideological commitment to a metaphysics of essence.

Truth and the Philosophy of Language

Thus far, I have assumed propositions to be truth-bearers and I have been concerned
with a certain metaphysical question: On what does the truth of a proposition depend?
I will now turn to the debate over truth within philosophy of language. Any discussion
of this must take as its point of departure Alfred Tarski’s (1936) theory of truth. For
present purposes, the most important features of a Tarskian account are the following:
Start with a given language. Let axioms governing simple subsentential expressions
state what such expressions refer to, are satisfied by, etc. Then the truth of sentences can

203
 M ATTI EKL U N D

be characterized recursively, using these axioms together with combinatorial axioms.

The canonical theorems of Tarski’s theory are of the form s is true iff p, where s is a name
of an object language sentence and p is a sentence of the metalanguage translating s.
Tarski proved that such a theory can be consistent, despite the semantic paradoxes. A
key idea is that the metalanguage (the language in which the theory is stated) is distinct
from the object language (the language which the theory is a theory of).
A central debate concerns the exact philosophical significance of Tarski’s theory.
An obvious limitation of the theory as it stands is that reference and satisfaction are
characterized only by lists: It is not said what reference and satisfaction are; all that is
provided are lists that tell us, for a given language, what primitive expressions of that
language refer to and are satisfied by. Hartry Field—who early and late has made semi-
nal contributions to the debate—argued (1972) that for Tarski’s theory to provide a
naturalistically acceptable definition of truth, a suitably reductive general account of
reference and satisfaction must be provided, for example along the lines of a causal-
explanatory theory of reference. Field also held that thus elaborated, Tarski’s theory
would be a correspondence theory, albeit one that dispenses with propositions and facts.
Another—deflationist—suggestion is to take the list-like characterizations of reference
and satisfaction to provide as much information about reference and satisfaction as
there is: All there is to reference and satisfaction is captured by these lists. This defla-
tionism would, the idea is, make reference and satisfaction naturalistically acceptable,
not through supplying a reductive account but through denying that they possess an
underlying nature.
Donald Davidson (1967) made the influential but controversial suggestion that a
theory of meaning for a natural language should take the form of a truth theory. The
basic idea—which subsequently has been modified in important ways—was that what a
speaker who understands a language knows is the information stated by a truth theory.
One question concerning Davidson’s proposal is whether this is correct. Another ques-
tion is whether the truth predicate employed in a Davidsonian theory is deflationary or
not. The question of whether the truth predicate employed is deflationary in turn splits
into two. One question is whether the truth predicate (or the reference and satisfaction
predicates in terms of which it is defined) is best conceived of as characterized in the
deflationist’s list-like way; a distinct question is whether the truth predicate has only
a purely expressive role and does not serve a deeper theoretical purpose. While these
two questions for natural reasons tend to get run together, they are separate (Eklund
2010).
It is often suggested that there are two fundamentally different outlooks on questions
of meaning and content: On one view, truth and related representational notions are
to be accorded a central role; on another, such notions have at most an expressive role,
and meaning is instead explained by appeal to use (Field 1994). There certainly is an
important dividing line there. But the picture presented is crude. For one thing, one may
suspect that the opposition between alethic accounts and use-theoretic accounts is illu-
sory: that the accounts provide answers to different questions. A use-theoretic account
can provide the answer to the question of what determines meaning, while a truth-
theoretic account is correct in that what is thereby determined is something best under-
stood in terms of truth-conditions. For another, there are many outlooks on meaning—
for example dynamic semantics, and different versions of inferentialism—such that it is
not straightforward how best to classify them in these terms. On dynamic semantics, the
focus is shifted from the propositions expressed by individual sentences to a sentence’s

204
 TH EO RI ES O F T R U T H

potential to change an information state. But alethic notions are still important, for
the information states are still characterized by what they represent. Inferentialism is
sometimes presented as an alternative to a truth-theoretic semantics. But truth can still
play an important role in an inferentialist story: for example, because what distinguishes
inferences from other associations of thoughts is concern with truth.
Sometimes in discussions about the theoretical role of truth, much of the focus is on
reference, as in Field (1972). But some theorists take the theoretical role of truth to
be importantly different in kind. First, while Davidson early and late has emphasized
the role of a Tarski-style truth theory, he has deemphasized the role of reference. He
has taken what might be called an instrumentalist approach to the axioms governing
subsentential expressions: any such axioms which yield acceptable conclusions regard-
ing the truth of whole sentences are acceptable. Unsurprisingly, given this outlook,
Davidson has been quite cavalier about the possibility of radical referential indetermi-
nacy (Davidson 1979). Truth is central; reference is not. Second, in some parts of the
literature, truth is perceived as important because it is a norm—a norm constitutively
governing the speech act of assertion, or the mental state of believing (Dummett 1981;
Velleman 2000; Wedgwood 2002). One reason it is relevant to distinguish between dif-
ferent reasons for seeing truth as a central notion is that this matters for some debates
over whether truth is properly perceived as central. If the theoretical justification for
centrally employing a notion of truth relies on the idea that reference is an explanatory
notion, arguments that reference is radically indeterminate can be damaging; but if the
role of truth is conceived of differently, such indeterminacy arguments are less relevant
(Eklund 2010).
The relation between these foundational issues in philosophy of language and the
debate over the relationality of truth is not straightforward. For example, it should be
obvious that one can insist on the correspondence claim regarding the relationality of
truth while eschewing the idea that truth has an important role to play in an account
of language. Conversely, one can reject as misguided the question of what the truth of
a proposition depends on while still basing one’s account of language around a notion
of truth.
One reason why the debates as I have described them are not tightly related is that
the correspondence theory as introduced concerns the truth of propositions while in
the debate over the foundations of semantics the issue concerns truth as applied to
sentences. Some theorists—see Field (1986) and McGee and McLaughlin (2000)—
understand the difference between the correspondence theorist and the deflationist
over sentence truth to concern the modal status of instances of (DS): for the cor-
respondence theorist these instances are contingent while for the deflationist they
are necessary. The idea that the instances of (DS) are necessarily true does not seem
true for ordinary language: surely it seems that ‘white’ could have meant green, and if
it had meant ‘green’, then ‘snow is white’ would not have been true, but snow would
still have been white. There are two ways for the deflationist, as Field and McGee &
McLaughlin conceive of her, to deal with this. One is to operate not with (DS) as
stated but with the rather more complex formulation

(DS*) ‘p’ is true in the language I am actually speaking iff p. (McGee 1993: 92)

But an obvious drawback with appeal to (DS*) is that the instances of (DS*) would
appear to be necessary on any acceptable view, due to the presence of the actuality

205
 M ATTI EKL U N D

operator. Another—the strategy of Field (1994)—is to insist that the only truth predi-
cate we need is a deflationary one, such that the instances of (DS) are necessary.
Related to this modal distinction is the idea that for the correspondence theorist
there is something we do with sentences that explains the truth of sentences while the
deflationist denies this (McGee 2004). For a correspondence theorist focusing on the
relation between propositions and reality, questions about what, if anything, explains
the semantic features of sentences are not relevant.

Truth and the Liar

Alongside the discussion in metaphysics about the nature of truth and the discussion
in the philosophy of language about the relation between truth and meaning, there is
the technical discussion about truth, centered on the liar paradox and other semantic
paradoxes. I will not attempt to summarize that debate. Let me just briefly indicate the
relation between some issues that come up in the literature on the paradoxes and some
themes from the discussion of truth in metaphysics and philosophy of language.
The liar paradox seems to show that the schemata (ES) and (DS) are not valid. A liar
sentence, L, saying of itself that it is not true (e.g. “this very sentence is not true”) is
such that Ll™Tr(L). But then if the relevant instance of (DS), LlTr(L), also holds,
we have a contradiction on our hands, Tr(L)l™Tr(L). Corresponding reasoning can
be carried out for propositions, but in the technical literature the focus is often on
sentences.
This apparent invalidity of (ES) and (DS) presents prima facie problems for a number
of different views on truth. Of course one way to deal with this threat posed by the liar is
to argue that the arguments for the invalidity of (ES) and (DS) are unsound, contrary to
appearances. For example, some might deny that there really is a liar proposition; they
would say that a liar sentence fails to express a proposition. But if that fails, what might
then be said? Let me go through a couple of views threatened by the liar and discuss
what routes lie open.
The paradigmatic deflationist takes (ES) or (DS) to somehow fully characterize
truth. Her view is then obviously threatened by the liar. Perhaps the simplest way for the
deflationist to deal with the liar paradox is to retreat from the idea that (DS) or (ES) in
a fully general form is definitional of truth, and instead appeal to a restricted version of
one of those schemata. But what restricted version might be appropriate? (See Horwich
(1998) and, for criticism, McGee (1992).) Alternatively, she might say that (DS) or
(ES) is definitional of truth even if some of the instances are untrue. This general type
of view divides into different subtypes. First, the deflationist can embrace dialetheism,
the view that there are true contradictions. She can then say that even if some instances
of (DS) and (ES) are untrue, still all such instances are (also) true (Priest 2006).
Second, adopting a meaning-inconsistency view, she can say that she is primarily con-
cerned with the concept of truth, and say that her view is that (DS) or (ES) some-
how gives the content of this concept, even though it leads to paradox (Eklund 2002,
2007). Third, she can say that (DS) or (ES) is somehow definitional of the concept,
but properly understood this doesn’t lead to paradox but only to the claim that para-
doxical sentences like the liar sentences cannot stably be assigned a truth-value (Gupta
2005).
Turn next to truth-theoretic approaches to meaning, specifically of the Davidsonian
form, where the canonical theorems are of the form “s is true iff p,” as above. Again

206
 TH EO RI ES O F T R U T H

the instance for L presents a problem. Again the dialetheist view and the meaning-
inconsistency view can be appealed to. For example, some theorists in the Davidsonian
tradition take a meaning theory to be that which competent speakers of a language
cognize, and several such theories have entertained the idea that what a competent
speaker cognizes is an untrue theory, for example due to liar sentences (Patterson 2007;
Badici and Ludwig 2007). A different strategy to which appeal has been made, that of
Davidson himself (1967: 28f), is to distinguish the truth predicate in the metalanguage
that the meaning theory is stated in from the truth predicate of the object language, and
insist that all instances of “s is true iff p,” where the truth predicate is the metalanguage
predicate and p is a translation of the object language sentence, are true. It seems less
plausible to defend a truth-theoretic approach by restrictions on (DS): Surely our com-
petence with the pathological sentences does not differ in kind from our competence
with nonpathological sentences.
Does the liar paradox pose problems for a correspondence theory of truth? From some
presentations of what the correspondence theory involves one might certainly get that
impression. Some friends of the correspondence theory take Aristotle’s remark, that to
say of what is so that it is so or of what is not so that it is not so, is to speak the truth
to characterize a correspondence conception, and Aristotle’s remark can be thought to
receive expression in (ES). But then if (ES) isn’t valid, the correspondence conception is
doomed. But to say with (a) above that proposition-truth depends on nonpropositional
reality is only to say that the truth of the proposition that p depends on is something
non-propositional. However natural it is further to say that it depends on whether p, that
further claim is an added commitment. Second, more importantly, Aristotle’s remark
doesn’t immediately entail the validity of (ES). It entails at most that ‘S’ and ‘S is true’
are always true together and that ‘not-S’ and ‘S is false’ are always true together. It can
then allow for exceptions to (ES) when ‘S’ is neither true nor false. So the correspond-
ence theory is not immediately bound up with (ES). This is not to say, however, that
the liar paradox does not pose any problems for the correspondence theory. Take the
notion of groundedness. A sentence which doesn’t assert anything about the truth-values
of other sentences is grounded, and one which does assert something about truth-values
of other sentences is grounded if the process of tracing through what sentences its truth-
value depends on terminates at grounded sentences (Kripke 1975: 693f). It is natural to
think that if the truth or falsity of a proposition depends on non-propositional reality,
then only grounded sentences can be true or false. But intuitively there are counterex-
amples to this latter claim. Sentences like ~(Tr(L)&~Tr(L)) and Tr(L)›~Tr(L) are
ungrounded but, intuitively, true. There are three possible reactions to this. One is to
take the argument at face value, as refuting the correspondence idea as here spelled out.
A second is to hold on to this correspondence idea and reject the intuitive judgments
about these propositions (Maudlin 2004). A third is to insist that the truth of some
ungrounded propositions is after all fully consistent with the correspondence idea. For
instance, one can perhaps hold that a sentence saying of itself that it is not both true and
untre is made true by the general feature of non-propositional reality that nothing can
both be and fail to be the case.
A different argument against the correspondence view departs from the so-called
no-no paradox (Sorensen 2001). Take two sentences that each says of the other that
it is not true. The only consistent ways of assigning truth-values to these sentences
involves assigning them different truth-values, so, the argument goes, they must have
different truth-values. But since they are perfectly symmetric, there is nothing in

207
 M ATTI EKL U N D

reality that determines one rather than the other to be true; so, given their difference in
truth-value, we have a counterexample to correspondence.

References
Badici, E. & K. Ludwig (2007) “The Concept of Truth and the Semantics of the Truth Predicate,” Inquiry
50: 622–38.
Bradley, F. H. (1909) “On Truth and Coherence,” Mind 18: 329–42.
Candlish, S. (1989) “The Truth about F. H. Bradley,” Mind 98: 331–48.
Damnjanovic, N. (2010) “New Wave Deflationism,” in Pedersen & Wright (2010).
Davidson, D. (1967) “Truth and Meaning,” Synthèse 17: 304–23.
—— (1979) “Inscrutability of Reference,” Southwestern Journal of Philosophy 10: 7–19.
Dummett, M. (1981) Frege: Philosophy of Language, Cambridge, MA: Harvard University Press, 2nd edn.
Eklund, M. (2002) “Inconsistent Languages,” Philosophy and Phenomenological Research 64: 251–75.
—— (2007) “Meaning-Constitutivity,” Inquiry 50: 559–74.
—— (2010) “Rejectionism about Truth,” in Pedersen & Wright (2010).
Field, H. (1972) “Tarski’s Theory of Truth,” Journal of Philosophy 69: 347–75.
—— (1986) “The Deflationary Conception of Truth,” in G. MacDonald & C. Wright (eds.) Fact, Science
and Morality, Oxford: Blackwell.
—— (1994) “Deflationist Views of Meaning and Content,” Mind 103: 249–84.
Fine, K. (1994) “Essence and Modality,” Philosophical Perspectives 8: 1–16.
Gupta, A. (2005) “Do the Paradoxes Pose a Special Problem for Deflationism?” in J.C. Beall & B. Armour-
Garb (eds.) Deflationism and Paradox, Oxford: Oxford University Press.
Horwich, P. (1998) Truth, Oxford: Clarendon Press, 2nd edn.
Joachim, H. H. (1906) The Nature of Truth, Oxford: Clarendon Press.
Maudlin, T. (2004) Truth and Paradox: Solving the Riddles, Oxford: Oxford University Press.
McGee, V. (1992) “Maximal Consistent Sets of Instances of Tarski’s Schema (T),” Journal of Philosophical
Logic 21: 235–41.
—— (1993) “A Semantic Conception of Truth?” Philosophical Topics 21:83–111.
—— (2004) “Ramsey and the Correspondence Theory,” in V. Halbach & L. Horsten (eds.) Principles of
Truth, Frankfurt: Ontos Verlag.
McGee, V. & B. McLaughlin (2000) “The Lesson of the Many,” Philosophical Topics 28: 129–51.
McLaughlin, B. & K. Bennett (2005) “Supervenience,” in E. Zalta (ed.) Stanford Encyclopedia of Philosophy,
http://plato.stanford.edu/entries/supervenience/.
Merricks, T. (2007) Truth and Ontology. Oxford: Clarendon Press.
Patterson, D. (2007) “Understanding the Liar,” in J.C .Beall (ed.) Revenge of the Liar, Oxford: Oxford Uni-
versity Press.
Pedersen, N. & C. Wright (2010) New Waves in Truth, Palgrave Macmillan.
Priest, G. (2006) In Contradiction: A Study of the Transconsistent, Oxford: Oxford University Press, expanded
edition.
Russell, B. (1910) “The Monistic Theory of Truth,” in Philosophical Essays, London and New York: Long-
mans, Green and Co.
Sorensen, R. (2001) Vagueness and Contradiction, Oxford: Clarendon Press.
Tarski, A. (1936) “Der Wahrheitsbegriff in den formalisierten Sprachen,” Studia Philosophica 1: 261–405.
Velleman, D. (2000) “On the Aim of Belief,” in The Possibility of Practical Reason, Oxford: Oxford University
Press.
Walker, R. (1989) The Coherence Theory of Truth, London and New York: Routledge.
Wedgwood, R. (2002) “The Aim of Belief,” Philosophical Perspectives 16: 267–97.
Wright, C. (1999) “Truth: A Traditional Debate Reviewed,” Canadian Journal of Philosophy suppl. vol. 24:
31–74.
—— (2001) “Minimalism, Deflationism, Pragmatism, Pluralism,” in M. Lynch (ed.) The Nature of Truth,
Cambridge, MA: MIT Press.
—— (2003) Saving the Differences: Essays on Themes from ‘Truth and Objectivity’, Cambridge, MA: Harvard
University Press.

208
 2.3
P R OP OS IT IONS
Scott Soames

Propositions are traditionally taken to be entities that satisfy A1–A3.

A1. Some things are asserted, believed, and known. These attitudes relate
those who assert, believe, or know something to that which they assert
believe, or know.
A2. The things asserted, believed, and known are bearers of truth and falsity.
A3. Propositions—the things satisfying A1 and A2—are expressed by sen-
tences. The proposition expressed by S can be designated by the expres-
sions ªthe proposition that Sº, ªthe statement/claim/assertion/belief that
Sº, or simply ªthat Sº.

Since different sentences may be used to assert the same thing, or express the same
belief, and different beliefs or assertions may result from accepting, or assertively utter-
ing, the same sentence, propositions are not identical with sentences that express them.
Intuitively, they are what different sentences, or utterances, that “say the same thing,”
or express the same belief, have in common. The metaphysical challenge posed by prop-
ositions is to identify entities fitting this picture that can play the roles demanded of
them in our theories.
One such theory is semantics. In semantics, propositions are needed as contents of
(certain) sentences, objects of attitudes, entities quantified over (‘Bill questioned sev-
eral propositions Mary asserted’) and referents of certain names (‘Logicism’, ‘Church’s
Thesis’, ‘Goldbach’s Conjecture’) plus demonstratives (‘That’s true’). Propositions are
also needed to state the goals of semantics, which include specifying the contribution
of semantic contents to what is asserted and believed by speakers. In theories of mind
propositions are contents of perceptual and cognitive states that represent the world as
being certain ways, and so have truth conditions. As with sentences, the states and their
propositional contents aren’t identical. Thus, our question becomes ‘What are these
things, propositions, that different, but representationally equivalent, sentences and
cognitive states have in common?’
One natural, though ultimately unsatisfactory, answer is that what they have in com-
mon is simply their truth conditions. On this view, propositions are not things that have
truth conditions; they are the truth conditions that sentences and cognitive states have.
The main problem with this idea is that sameness of truth conditions is not sufficient
for sameness of assertion, belief, or semantic content. Semantic theories that pair a
sentence S with the set of metaphysically possible world-states at which S is true don’t
provide enough information to allow one who knows the theories to understand what S
 SCO TT SO A ME S

means, or to identify the beliefs expressed, or assertions made, by uses of S. This nega-
tive result persists even if we relax the requirement that the world-states—thought of
as maximal properties attributed to the universe—be restricted to those that genuinely
could have been instantiated, and allow, in addition, metaphysically impossible world-
states that can coherently be conceived to be instantiated, and can’t be known a priori
not to be. This result is generalized in Soames (1987, 2008b) to all theories satisfying
(i)–(iii), no matter what they choose as circumstances at which sentences are evaluated
for truth.

(i) Semantic contents of some singular terms, including indexicals and vari-
ables, are their referents relative to contexts and assignments.
(ii) ªP & Qº is true at a circumstance E iff both conjuncts are true at E.
(iii) ªx Fxº is true at E iff Fx is true of some object o at E

Since (i) is well motivated and (ii)–(iii) are central to theories of truth at a circumstance,
we can’t identify the propositions expressed by sentences with sets of circumstances at
which they are true, no matter how fine-grained the circumstances. In short, although
propositions are indispensable, they can’t be extracted from truth-conditional theories.
Rather, they are what the truth conditions of sentences are derived from. Thus, we need
an independent theory of propositions.

The Realist Theories of Frege and the Early Russell

The realist theories of Gottlob Frege (1892a, 1918) and the early Bertrand Russell
(1903) provide a starting point. According to both, propositions are meanings of sen-
tences, bearers of truth, and objects of the attitudes. Since they are meanings, Frege
and Russell took them to be structurally complex entities the constituents of which
are meanings of the constituents of the sentences that express them. This assumption,
though natural, led to what they found to be an unfathomable mystery. Just as sentences
aren’t collections of unrelated expressions but rather have a structural unity that distin-
guishes them from mere lists, and is responsible for their representational character, so
propositions aren’t collections of unrelated expression meanings, but have a unity that
distinguishes them from mere aggregates of their parts, and allows them to represent the
world. This unity is what Frege and Russell found mysterious.
Here is how the problem is put in Russell (1903: 49–50).

Consider, for example, the proposition “A differs from B.” The constituents of
this proposition, if we analyse it, appear to be only A, difference, B. Yet these
constituents, thus placed side by side, do not reconstitute the proposition. The differ-
ence which occurs in the proposition actually relates A and B, whereas the difference
after analysis is a notion which has no connection with A and B. [my emphasis] It
may be said that we ought, in the analysis, to mention the relations which dif-
ference has to A and B, relations which are expressed by is and from when we
say A is different from B. These relations consist in the fact that A is referent
and B relatum with respect to difference. But A, referent, difference, relatum,
B, is still merely a list of terms, not a proposition. A proposition, in fact, is essen-
tially a unity, and when analysis has destroyed the unity, no enumeration of constitu-
ents will restore the proposition. The verb, when used as a verb, embodies the unity of

210
 PRO PO SI T I O N S

the proposition, and is thus distinguishable from the verb considered as a term, though
I do not know how to give a clear account of the precise nature of the distinction.
[my emphasis]

Certainly, there is more to the proposition that A differs from B than the fact that
its constituents are A, B, and difference. There is also the manner in which the con-
stituents occur. Presumably, this has something to do with the fact that the proposition
predicates difference of A and B, and so represents A as being different from B. Since a list
doesn’t predicate anything of anything else, it isn’t representational.
Is this problematic? Consider sentence (1), in which ‘difference’ and ‘identity’ are
nouns, ‘different’ is an adjective that combines with the copula to form the predicate,
and ‘from difference’ is a prepositional phrase modifying the predicate.

(1) [S [N Difference] [VP [V is different [PP [P from] [N identity]]]]

The constituents of the Russellian proposition expressed by (1) are the relations iden-
tity and difference, the latter occurring twice. Understanding (1) involves understand-
ing its words, which provide the proposition’s constituents, plus understanding its syn-
tactic structure, which indicates what is predicated of what. Just as one who understands
the sentence recognizes one expression as predicate and two others as arguments, so,
it might be argued, one who entertains the proposition recognizes from its structural
configuration which constituent is predicated of which. On this view, it is the structural
relations that constituents bear to one another that carry the information about predica-
tion that unifies the proposition, and gives it representational content.
But how, exactly, does the arrangement of its constituents show that the proposi-
tion predicates difference of difference and identity? Consider some candidates for this
proposition.
(2a) <difference, <difference, identity> >
(b) {{difference}, {difference, {{difference}, {difference, identity}}}}
(c) <<identity, difference>, difference >
(d) {{{identity}, {identity, difference}}, {{{identity}, {identity, difference}},
difference}}
(3a) 1

Difference 2

Difference Identity

(3b) 1

2 Difference

Difference Identity

211
 SCO TT SO A ME S

Although any of these structures could serve as a formal model of the proposition
expressed by (1), none could be that proposition. The proposition represents difference as
being different from identity, because it predicates difference of the other two. Since there
is nothing in (2), (3), or any similar formal structure, which, by its very nature, indicates
that anything is predicated of anything, such structures are neither intrinsically repre-
sentational nor capable of being true or false.
We could, if we wished, adopt rules that would allow us to read off information about
predication from the structures, and so interpret them. But this wouldn’t make them
traditional propositions. Such propositions aren’t things we endow with meaning by
providing them with interpretations; they are the meanings, or interpretations, we take
sentences to have. The real problem is Russell’s pair of assumptions: (i) that proposi-
tions are intrinsically representational, independent of us, and (ii) that they are the
sources from which cognitive states and sentences inherit their intentionality. To adopt
these assumptions is to try to explain something relatively clear and obvious by appeal-
ing to something mysterious and incomprehensible. The clear and obvious fact is that
agents do predicate properties of things, and when they predicate being different of things
they—the agents—represent those things as being different. The purported explanation is
that these agents entertain an abstract structure which—intrinsically, and independent
of any relation it bears to us—predicates difference of the things, and so represents them
that way. The problem with this “explanation” is that we have no idea of how a tree-
structure, n-tuple, set-theoretic construction, or any similar abstract structure could,
in and of itself, predicate anything of anything. Because of this, traditional Russellian
propositions are the incomprehensible posits of an explanatory program that doesn’t
explain. As such, there is no reason to believe in them.
The same holds for Fregean propositions, which are also assumed to be representa-
tional independent of us. Frege differs from Russell in postulating “unsaturated” senses
that are intrinsically predicative, and so always occur in a predicative role (Frege,
1892b). Although this may sound attractive, it isn’t, since it leads him to conclude
that neither the sense nor referent of any predicative expression can be designated by a
nonpredicative expression—and, thereby, made the subject of a further predication.
This thesis—that if Pred is a predicate, then the sense of Pred is unsaturated, the referent
of Pred is incomplete, and neither can be designated by any nonpredicative expres-
sion—is self-defeating, as shown by the italicized phrases used to state it (Soames 2010a:
chapter 2).

New Conceptions of Propositions

Since we need propositions in our linguistic and cognitive theories, the failure of tra-
ditional conceptions calls for a new conception that reverses explanatory priorities.
Sentences, utterances, and cognitive states are not representational because of their
relations to inherently representational propositions. Instead, propositions are repre-
sentational because of their relations to inherently representational cognitive states, or
independently representational sentences. On this view, intentionality in mind and
language results from the cognitive activities of agents. There are, in the current litera-
ture, two main ways of fleshing out this idea. The first, presented in King (2007) takes
the existence and representational character of propositions to be dependent on, and
derived from, the prior existence and representational character of sentences that express
them. The second, presented in Soames (2010a), develops a conception of propositions

212
 PRO PO SI T I O N S

as contents of intentional cognitive states generally—including perception and

nonlinguistic belief, which are the basis of more complex, linguistically mediated,
thought.

King’s Conception of Propositions

We begin with sentence (4), and the proposition it expresses.

(4) [S [NP This ] [Pred [Cop is ] [Adj red]]]

The labeled bracketing in (4) displays the syntactic structure of the sentence, indicat-
ing its constituents, syntactic structure, and the order in which the constituents occur.
Abstracting away from details, I use ‘RS’ to name the syntactic relation that the word
‘this’ stands in to the word ‘red’ in (4). Since ‘this’ is an indexical, (4) expresses a propo-
sition only relative to a context of utterance. Imagine a context in which ‘this’ is used
to refer to o. According to King, the proposition—that o is red—expressed by (4), rela-
tive to the context, is a structurally complex entity of which o and the property being
red are constituents, along with RS. He takes this proposition to be the fact designated
by (4F).

(4F) the fact that there is a possible context C and expressions a and b of some lan-
guage L such that (i) some sentence S of L consists of a’s bearing RS to b, (ii) o is
the semantic content of a in C, and the property being red is the semantic content
of b in C, and (iii) in L the relation RS encodes the instantiation function (which
is to say that sentences formed by placing an expression D in the relation RS to an
expression E are true, at a context iff the referent of D at the context instantiates
the referent of E at the context).

Although this may seem like a mouthful, the idea is simple. Propositions are what sen-
tences with the same semantic content have in common—the contents of their con-
stituents, plus a common syntactic structure, the semantic contribution of which is the
same for each sentence. In English, when a term D stands in RS to an adjective E, the
property designated by E is predicated of the referent of D, which means that the sen-
tence is true iff the referent instantiates the property. So the syntax and semantics of
English, plus the existence of (4) and the context C, guarantee the truth of the italicized
clause in (4F). King takes this to mean that (4F) designates a fact, which is the proposi-
tion (4) expresses. Thus, the existence of the proposition that o is red is guaranteed by
the semantic properties of (4) at C. To get the same result for any sentence S and con-
text C* at which S has the same semantic properties that (4) has at C, King quantifies
over all possible contexts, and every sentence of every language, thereby arriving at the
complex general fact that, in his view, is the proposition that o is red.
For King, as for the early Russell, the fact that a is G is a complex entity that consists
of a’s actually instantiating G-hood. As Russell observed, the proposition that a is G can’t
be that fact, for, if it were, its very existence would guarantee its truth (Russell 1912:
chapter 12; Soames 2010a: chapter 4). King avoids this absurdity by taking the proposi-
tion to be the fact that a Rp G-hood—where Rp is the relation expressed by the formula
that results from replacing the underlined occurrences of ‘o’ and ‘the property being red’
in the italicized part of (4F) with a pair of variables. Thus, he solves Russell’s problem

213
 SCO TT SO A ME S

of explaining what “holds together” the constituents of the proposition—they are held
together by the fact that Rp actually relates them—while avoiding the absurd result that
in order to exist a proposition must be true.
Although this result is good, the view faces a number of challenges, including (i–iii).

(i) Positing facts—not as true propositions, but as discrete truthmakers—carries phil-

osophical worries that may themselves be sources of skepticism (Merricks 2007;
Soames 2008a). Possibilia are also a concern. For King, propositional facts involve
quantification over possible contexts, which include merely possible world-states
as constituents. But if the analysis of propositions appeals to a conceptually prior notion
of a possible world-state, then the analysis of such a state can’t appeal to a conceptu-
ally prior notion of a proposition. This is problematic, since analyses of world-states
in terms of propositions are attractive (Adams 1974; Soames 2007, 2010b:
chapter 5). Thus, going down King’s road on propositions precludes taking other
roads one may want to take on related issues.
(ii) King’s belief in actualism—which allows reference to, and quantification over,
only things that actually exist—brings further worries. He brings possible contexts
into his analysis to secure semantic contents (needed for some actual propositions)
that no sentence is ever actually used to express. In order for these propositions
to exist world-states of possible contexts—thought of as maximal properties the
universe fails to instantiate—must also exist. Some of these will, I suspect, involve
merely possible objects (including merely possible agents of some contexts). This
is worrisome. Though the existence of some uninstantiated properties may be
unproblematic, the existence of those involving particular objects (e.g., being iden-
tical with, or distinct from, o) seems to require the objects to exist. If, as I believe,
world-states are properties the existence of which requires the existence of possible
objects—then their actual existence will require the actual existence of possible
objects (Soames 2010b: chapter 5). King takes no definite stand on this, leaving
it open how world-states represent possible objects (King 2007: 42–4, 57, 84).
This leaves his account hostage to the fortunes of contentious metaphysical claims
about which, one might think, the analysis of propositions should be neutral.
(iii) Since English contains both ªthe fact that Sº and ªthe proposition that Sº—which
for King designate different things—his view seems to require ‘that’-clauses to be
ambiguous between the readings they bear in ªPam regrets (the fact) that Sº and
ªPam believes (the proposition) that Sº. But he neither gives any linguistic argu-
ment that this ambiguity exists, nor rebuts seeming evidence to the contrary—for
example., “Pam regrets that she is pregnant. Although her parents don’t realize it
yet, in time they will come to believe it.” Here, the fact regretted is described
as something that will eventually be believed—a proposition. How, given the
supposed difference between the two, can that be?

This difficulty stems from King’s Russellian perspective. Both he and Russell take a
central aspect of the problem of the unity of the proposition to be that of explaining how its
constituents—objects and n-place properties—“hold together.” Both came to think that
the only available explanation is that a relational constituent of the proposition must
actually relate them, which means that the proposition must be a fact in order to exist
at all. When Russell came to this view, he concluded that the only candidate for being
identical with the proposition that Desdemona loved Cassio was the fact that Desdemona

214
 PRO PO SI T I O N S

loved Cassio. Since this ruled out the possibility of falsity, he gave up propositions
(Russell 1912: chapter 12). Although King avoids this conclusion by identifying the
proposition with a different fact, this simply leads to the corresponding difficulty (iii).
The error in both accounts comes from taking the question “What holds the constit-
uents of a proposition together?” too seriously. The misnamed problem of propositional
unity isn’t that of making one object out of many. Sets, sequences, and trees are each
single things with multiple constituents of various sorts. The reason they aren’t proposi-
tions isn’t that their constituents keep falling out. They aren’t propositions because they
don’t represent anything as being any particular way. The real problem for which we
have, as yet, no answer is: “How is it that propositions are able to represent the world,
and so have truth conditions?” Nothing is gained, when answering this question, by
appealing to facts. Even if we suppose that King’s clause (4F) designates a complex fact
F4, simply noting F4’s existence provides no hint that it has truth conditions at all, let
alone that it is true iff o is red. Of course, if it exists, then for some sentence S of some
language and some possible context C, S is true at C iff o is red. But from this we can’t
even conclude that it makes sense to attribute truth conditions to facts, let alone that
F4 has the truth conditions claimed for it.
King eventually realizes this (King 2007: 60). He responds as follows: in and of itself,
F4 isn’t a proposition, and doesn’t have truth conditions. However, it becomes one, and
acquires truth conditions, when we interpret it in a certain way. King sees F4 as a complex
consisting in o’s standing in the two-place relation Rp (defined above) to the property being
red. So understood, it can be seen as acquiring truth conditions, provided that Rp is viewed
as (in his words) “inheriting the semantic significance of RS.” Just as we use the syntactic rela-
tion RS in sentence (4) to predicate the property designated by its second argument of the
referent of its first argument, so we can come to use the propositional relation Rp to predicate
the property which is its second argument of the object which is its first argument. To do
so is to endow the formerly nonintentional F4 with truth conditions. King thinks there was
a time—before English contained attitude verbs, ‘that’-clauses, or modal operators—when
it did contain sentences like (4), plus truth-functional and quantified sentences. Then,
sentences were used in cognition and communication, words had semantic contents, and
sentences had truth conditions. But no propositions yet existed. Although the fact F4
existed, it wasn’t yet a proposition, because no need had arisen to assign semantic significance
to Rp. When the need to report cognitive states was felt, some way of viewing sentences
and cognitive states as sharing representational contents had to be found. Then, speakers
became aware of F4, and implicitly assigned it a significance inherited from sentence (4).
Thus it came to pass that Man created propositions (King 2007: 60–1, 65–7).
In addition to being uncomfortably speculative, this account embodies a troubling
internal tension. On one hand, we are told that propositions are dependent on the
prior existence of language, and the complex cognitive and communicative practices
of those who speak it. Prior to the existence of propositions, speakers are supposed to
have used language to think and communicate—which presumably involved (a) using
certain expressions with the intention of referring to specific objects, (b) using other
expressions with the intention of predicating them of those objects, (c) using sentences
to assert things, and express beliefs, and (d) being confident one’s hearers could recog-
nize one’s intentions, assertions, and beliefs. On the other hand, we are told that belief,
assertion, and intention are attitudes to propositions, which did not then exist. From
this it follows that the agents who used language in King’s prepropositional age must
have done so without intending, believing, or asserting anything. How can that be?

215
 SCO TT SO A ME S

King answers by postulating that our ancestors had what he calls “proto-intentional
states,” sufficient for their “prepropositional language.” But whatever these “proto-
intentional states” are, they can’t be relations to things—call them “proto-propositions”—
that are themselves representational, or have truth conditions, lest they raise the same
problems as genuine propositions do. If the postulated primitive states are not relational in
this way, we need to be told: (i) how, if at all, they are representational; (ii) how—if they
are not—they can give rise to sentences that are representational, and have truth condi-
tions; and (iii) how—if they both represent things as being certain ways and bring it about
that the sentences of the proto-language do too—they can fail to provide propositions at
the same time. King’s discussion (65–7) doesn’t begin to answer these questions. Thus,
it remains mysterious how the sentences that, on his view, must have arisen from them
could be representational, or have the truth conditions on which his account depends.
This point is tied to a final question the relevance of which should have been evident
all along. What is it to entertain one of his linguistically-based propositions? Apart from
points already mentioned, plus an inconclusive paragraph on page 52, King says very
little about this. Essentially, all we are told is that once propositions come into being,
a speaker of a language that contains attitude constructions will be able to entertain
propositions by understanding sentences. No language-independent way of entertain-
ing propositions is specified. But surely, we need an account of propositions as contents
of perceptual experience, perceptually based beliefs, plus imaginative, and other non-
linguistic, intentional states. It is hard to see how this can come from an analysis of
propositions as inherently linguistic facts.

The Cognitive Realist Account of Propositions

This observation leads to a second attempt to ground the intentionality of propositions
in the cognitive states of agents (Soames 2010a). Like King’s approach, it applies to
propositions expressed by sentences. Unlike his approach, it is not centered in language
but applies to cognitive states generally. We begin with the idea that to entertain a
simple proposition is to predicate something of something. To entertain the proposi-
tion that o is red is to predicate redness of o. Although, like negation, predication is a
primitive notion, it is easily illustrated. When we see o as red, we predicate redness of it,
and so entertain the proposition that o is red. We also predicate redness of o, and hence
entertain this proposition when we form the perceptual belief that o is red. We do the
same when we understand an utterance of ‘This is red’, taking the predicate to express
the property redness and the subject to refer to o.
Pretend, for a moment, that the proposition that o is red is the abstract structure (4D).

(4D) [Prop [Arg o] [Pred Redness]]

Prop

Arg Pred

O Redness

216
 PRO PO SI T I O N S

Recognizing structures like these to be theorists’ creations, we may temporarily stipulate

that to entertain (4D) is to predicate redness of o. In so doing, we assign a technical
meaning to the verb ‘entertain’ that explains what we mean by the claim that an agent
entertains this abstract structure. Next, we advance (5a), from which we derive that an
agent entertains the proposition that o is red iff the agent predicates redness of o.

(5a) An agent entertains the proposition that o is red iff the agent entertains
(4D).

Other attitudes are treated similarly. Since to entertain the proposition that o is red is to
predicate redness of o, and since this predication is included in every attitude with that
content, entertaining the proposition is one component of any attitude we bear to it.
To judge that o is red is to predicate redness of o while endorsing that predication. To
believe that o is red is, (very roughly) to be disposed to judge that it is. To know that o is
red is, roughly, for o to be red, to believe that o is red, and to be justified in so believing.
To assert that o is red is to commit oneself, by uttering something, to treat the proposi-
tion that o is red as something one knows. Given these characterizations, we add (5b)
to (5a), generating further empirical claims that bottom out, as before, in claims about
the cognitive acts of agents.

(5b) An agent judges/believes/knows/asserts the proposition that o is red iff the

agent judges/believes/knows/asserts (4D).

More complex propositions are treated similarly. For example, to entertain the proposi-
tion that it is not the case that o is red, represented by (6), is (i), to predicate redness of
o, and thereby to entertain the proposition that o is red, and (ii), to predicate not being
true of that proposition—which, for the moment, we continue to identify with (4D).
One performs this predication—of untruth to the propositional structure—by, in effect,
saying to oneself, “That’s not true,” referring to the result of one’s initial predication.

(6) [Prop[Pred NEG][Arg [Prop [Arg o] [Pred Redness]]]]

Prop1

Pred1 Arg1

NEG Prop2

Arg2 Pred2

O Redness

Entertaining more complex propositions involves more complex sequences of cognitive

acts.

217
 SCO TT SO A ME S

The approach thus far depends on (i) the inclusion of information about what is
predicated of what in the abstract structures we are temporarily allowing to play the role
of propositions; (ii) the identification of what it is to entertain a proposition with spe-
cific acts of predication that occur in perception and both linguistic and nonlinguistic
thought; and (iii) the establishment of propositions as themselves representational bear-
ers of truth conditions, in virtue of what is required to entertain them. What makes the
proposition that o is red represent o as red, is that predicating redness of o is necessary
and sufficient to entertain it. Since one who performs this predication oneself represents
o as red, and since any agent who does this thereby entertains the proposition, we speak
derivatively of the proposition predicating the property of the object, and so represent-
ing it as red—whether or not anyone, in fact, ever entertains it. Since this fact about
the proposition doesn’t change from world-state to world-state, its truth conditions are
invariant. It is true at any world-state w iff o is red at w.
At this point we confront the convenient pretense that identifies propositions with
abstract formal structures like (4D). Having come this far, it is tempting to think that
no further, more realistic, account of propositions is needed. The temptation springs
from the idea that the function of propositions in our theories is nothing more than to
identify and track the cognitive states of agents—which are all we are really interested
in. Just as in physical theory we use numbers, and other abstract objects, to specify rela-
tions that physical magnitudes bear to one another, so, in semantic and cognitive theory
we use propositions to talk about the relations that intentional cognitive states bear to
one another, and to the world. Since what is essential is just that we have a simple and
economical way of doing this, it is tempting to imagine that one abstract conception of
propositions that gets the job done is as good as another. On this instrumentalist view,
there are no genuine questions: “What are propositions really?”, or “What structures do
agents really have before their minds?” Propositions are theoretical fictions.
Although this idea may initially seem attractive, it isn’t. According to the theory,
entertaining a compound structure like (6)—which is supposed to represent the propo-
sition that it is not the case that o is red—requires agents first to refer to, and then
predicate untruth of, its simple propositional constituent, which is the result of their
predicating redness of o. But if agents really do refer to the proposition that o is red, and
predicate untruth of it, then that proposition is no fiction. Since the representational
character of (4D) is due to theorists, and so is merely conventional, neither it nor any
similar structure is a proposition. This suggests a genuinely cognitive realist view. Since
propositions are needed to track cognitive acts, they can be identified with event types,
instances of which involve those very acts. For example, the proposition that o is red is
the (minimal) event type in which an agent predicates redness of o. As such, it is both
intrinsically connected to the cognitive acts it is needed to track, and also something to
which all agents who entertain it bear the same natural relation.
Consider a spoken utterance of the sentence ‘Snow is white’, thought of as an event
that occurs at a particular time and place, and also as a token of the sentence uttered.
So construed, sentences are event-types capable of having multiple occurrences, which
are their tokens. Next imagine an utterance of the sentence followed by an utterance of
“That’s true.” In such a case, the demonstrative may refer either to the utterance, or to
the sentence uttered—illustrating that some event types can be bearers of truth value.
Finally, there are events in which one doesn’t utter anything but simply thinks of snow
as white, thereby predicating whiteness of it. These cognitions are events that occur at
particular times and places, which are instances of a corresponding event type in which

218
 PRO PO SI T I O N S

an agent predicates whiteness of snow. Just as the sentence ‘Snow is white’ can be iden-
tified with an event type of which utterances of it are instances, so the proposition that
snow is white can be identified with an event type of which particular acts of predicat-
ing whiteness of snow are instances. Thus, both event types have truth conditions.
In addition to bearing their truth conditions intrinsically, propositions-as-event-
types are things with which we are acquainted. Since the proposition that o is red is an
event type in which one predicates redness of o, and since every attitude one bears to
this proposition involves this predication, any agent acquainted with his own cognitive
processes—in the sense of being able to make them objects of thought—will be simi-
larly acquainted with the proposition that o is red, by virtue of being acquainted with
(and noting the similarity among) the events in his cognitive life that are instances of
it. Given the means both of thinking of o as red, and of becoming aware of so doing,
one can then make further predications about the content of one’s thought. If, after one
predicates redness of o, one says to oneself, “That’s not true,” one thereby predicates
untruth of the proposition that is the type of cognitive event one has just experienced.
This illustrates how agents are able to entertain compound propositions by predicating
properties of their constituent propositions—which was the bane of the instrumentalist
view.
In this way, the cognitive-realist theory inherits the virtues of that view, without
its deficiencies. Like that view, it provides entities needed as contents of sentences,
bearers of truth, and objects of the attitudes. But while the instrumentalist view sees
nothing beyond the unavoidably arbitrary formal structures—like (4D) and (6)—that
play the role of propositions, the realist account views these structures as merely useful
devices that represent the real propositions to which agents bear natural cognitive relations.
The labeled trees provided by linguistic and cognitive theories encode the structure
and sequence of cognitive acts that are necessary and sufficient for entertaining the real
propositions these structures represent—where entertaining a proposition is performing
the acts needed to token the event-type that it is.
This account addresses the most vexing problems to which traditional propositions
give rise. Unlike the Platonic epistemology traditionally required by theories of proposi-
tions, the cognitive-realist account takes knowledge of propositions to be knowledge of
events that make up one’s cognitive life. It also avoids the pseudo-problem of “the unity
of the proposition,” which—though usually posed as that of explaining how the con-
stituents of propositions “hold together”—serves to mask the real problem of explaining
how propositions can be representational, and so have truth conditions. The traditional
view makes this problem insoluble by taking the representational nature of propositions
to be intrinsic, unanalyzable, and independent of us. By locating the representational
character of propositions in their intrinsic connection to inherently intentional cogni-
tive events, the cognitive-realist account offers a solution.

References
Adams, Robert (1974) “Theories of Actuality,” Noûs 8: 211–31.
Frege, Gottlob (1892a), “Uber Sinn and Bedeutung” Zeitschrift fur Philosophie und Philosophische Kritik 100:
25–50; translated by Peter Geach as “On Sense and Reference,” in Geach and Black (1970).
—— (1892b) “On Concept and Object,” Vierteljahrsschrift fur wissenschaftliche Philosophie 16: 192–205;
translated by Peter Geach, in Geach and Black (1970).
—— (1918), “Der Gedanke,” Beitrage zur Philosophie des deutschen Idealismus 1; translated by A. Quinton and
M. Quinton as “The Thought,” Mind, 1956, 65: 289–311.

219
 SCO TT SO A ME S

Geach, Peter and Black, Max (1970) Translations from the Philosophical Writings of Gottlob Frege, Oxford:
Blackwell.
King, Jeffrey C. (2007) The Nature and Structure of Content, Oxford: Oxford University Press.
Merricks, Trenton (2007) Truth and Ontology, New York: Oxford University Press.
Russell, Bertrand (1903) Principles of Mathematics, New York: Norton.
—— (1912) The Problems of Philosophy, New York: Oxford University Press..
Soames, Scott (1987) “Direct Reference, Propositional Attitudes, and Semantic Content,” Philosophical
Topics 15: 44–87; reprinted in Soames (2009).
—— (2007) “Actually,” in Mark Kalderon, ed., Proceedings of the Aristotelian Society, Supplementary Volume
81: 251–277; reprinted in Soames (2009).
—— (2008a) “Truthmakers?” Philosophical Books 49, 317–27.
—— (2008b) “Why Propositions Can’t be Sets of Truth-Supporting Circumstances,” Journal of Philosophical
Logic 37: 267–76, reprinted in Soames (2009).
—— (2009) Philosophical Essays, Vol. 2, Princeton: Princeton University Press.
—— (2010a) What is Meaning?, Princeton: Princeton University Press.
—— (2010b) Philosophy of Language, Princeton: Princeton University Press.

220
 2.4
CONC EPT S
Christopher Peacocke

Concepts are of interest in their own right; but a good theory of concepts also bears upon
a huge array of further philosophical issues. A good theory of concepts has consequences
for our understanding of justification, of reasons, the a priori, knowledge, psychological
explanation, and the way the mind relates to the world, to name but a few such issues.
‘Concept’ is used very loosely in everyday English, so we need to be more precise
about our target topic. Very roughly, concepts are here understood as meeting the con-
dition that an arbitrary concept C, of whatever category, is distinct from a concept D
if and only if a rational thinker can judge a content A(C) without judging A(D) (cp.
Frege 1993). So the concept I and the concept Descartes are distinct, even as used by
Descartes. During a bout of amnesia, Descartes can rationally judge Descartes is Descartes
without judging I am Descartes. Similarly the concepts circle and shape described by the
equation x2+y2=r2 are distinct, since someone can judge that a seen shape is a circle
without knowing what shape that equation describes. Concepts meeting the given con-
dition are to be distinguished from conceptions, stereotypes, and from whatever subper-
sonal mental representations may underlie or make possible attitudes to contents built
up from concepts. For instance, the stereotype of a C.E.O. of a company may be that of a
sharply dressed middle-aged man. Nevertheless, a thinker can rationally judge, and may
know, that a casually dressed younger woman is a C.E.O of a company.
Just as a theory of meaning and a theory of understanding must be integrated with
each other, so equally a theory of concepts and a theory of the grasp of concepts must be
integrated with one another. If we give an account of what a concept is, as something
that can be true or false of objects, we must also say what it is for concepts so character-
ized to be grasped, to be in the contents of thinkers’ attitudes. Conversely, if we say what
it is for a concept to be grasped, we must also say what it is for concepts so described
to be capable of being true of false of objects. We cannot gain theoretical understand-
ing if we take the capacity of concepts to play both these roles as something primitive
and unexplained (Wilson 2006: Ch 3). Saying that we need a theory that integrates
concepts with grasp of concepts is not yet to say which of the two, if either, is more
fundamental than the other. That has been a disputed issue between those who would
agree that integration is desirable (Peacocke 1992, Fodor 1998).
Dummett draws a distinction between ‘meaning theories’ which specify the mean-
ings of the expressions in a particular language, and the theory of meaning in general.
The theory of meaning which will specify the general form to be instantiated by a par-
ticular meaning theory, presumably together with some rationale for that general form
(Dummett 1991: 22). I similarly distinguish between theories of particular concepts,
 CH RI STO PH ER P E A C O C K E

and a general theory which specifies the general form that should be instantiated by
a theory of any particular concept. As in the case of linguistic meaning, both types of
theory are of philosophical interest.
Theories of intermediate levels of generality, concerned with particular types of con-
cept, can also be philosophically significant. One may be interested in what is com-
mon to all observational concepts, or all concepts of conscious states, or all logical
concepts.
I have been thinking about a theory of concepts for well over a quarter-century, and
could not offer a wholly neutral critical survey even if I tried. So I will not refrain from
advocacy; but I will try to indicate, for each alternative approach, some of the tasks it
faces. It will help us if we divide theories of concepts into three broad types: type I theo-
ries, which hold that there is some constitutive and individuative connection between
concepts on the one hand, and reference and truth on the other; Type II theories, which
hold that there is no such connection; and Type III theories, whose general character
can be indicated by calling them ‘no-theory’ theories.

Type I Theories
Concepts look very much like Fregean senses. In the Grundgesetze, Frege emerges clearly
as a Type I theorist (Frege 1967: §32). He says that the Thought expressed by a sentence
of his formal language is the thought that the conditions for it to denote the True are
fulfilled, where those conditions are determined in a certain canonical fashion by his
semantic rules for the formal language. Since Thoughts are composed of component
senses, this goes hand-in-hand with the view that particular senses are individuated by
the condition for something to be their reference or semantic value, their reference-
rules. This conception can be generalized beyond the formal language Frege was consid-
ering to a vast range of expressions of natural language.
Simply generalizing this aspect of Frege’s conception leaves a lot open, however. All
we have so far in this description of concepts is a specification of what a concept is, and
not an account of what it is to grasp it. Type I theories that agree on this Fregean com-
ponent subdivide and part company on the right way of addressing the issue of grasp.
Some theories aim to give substantive accounts of what it is to grasp a concept with
a given reference-rule. They hold that there is more to be said than is given in a sim-
ple disquotational clause, such as “is in pain” is true of x at t iff x is in pain at t. Under
the approach I myself favor, for each predicative concept, there a nondisquotational
condition tacitly known to one who grasps the concept, and which specifies the condi-
tion for an object to fall under the concept. Grasp of the concept pain consists in tacit
knowledge that an arbitrary subject x is in pain just in case x is in the same subjective
state he himself can recognize when he himself is in pain (Peacocke 2008: Ch 5). Grasp
of the observational concept oval consists in knowledge that an arbitrary thing x is oval
just in case x is of the same shape as things he himself can recognize in perception as
oval (where the representational content of perception captured by as oval is noncon-
ceptual). These richer accounts are in the spirit of what Dummett called a full-blooded
account, though Dummett’s elaboration of that conception contained problematic ele-
ments that need not be present in richness (Dummett 1993). (We can proceed without
his problematic idea that one could use a full-blooded description to explain a concept
to someone who does not already grasp it.) Richer theories of this sort meet the con-
straint noted earlier. The condition that is tacitly known for something to fall under

222
 CO N CEP T S

the concept is also a priori necessary and sufficient for something really to fall under the
concept. Grasp and satisfaction are thereby integrated.
Rejection of such rich theories does not mean that a theorist has nothing to say about
grasp. Under the ‘modest’ conception of McDowell, we may say that someone grasps a
concept if attribution of that concept, with its disquotational rule, is required to make
the person maximally intelligible to us (McDowell 1998). The issue over richness versus
modesty is not whether this condition on maximal intelligibility is correct. It will rather
be over whether the constraints on intelligibility can be elucidated without appealing
to the resources of a rich theory. Certain combinations of experiential state, environ-
ment, and background attitudes will make certain attitudes not intelligibly ascribable to
a subject. There are indefinitely many such constraints. The rich theorist will say that
these constraints have a finite and structured source in the non-disquotational condi-
tions involved in grasp of the concepts in question. The fundamental issue of whether a
theory of concepts should be rich is by no means resolved.
Even if one has sympathy for the rich theorist, the latter has more work to do on
several counts. The rich theorist must elucidate the notion of tacit knowledge on which
he is relying. He must say how it functions in explanation, both of a thinker’s making
particular judgments, and in our assessment of certain combinations of attitude and cir-
cumstance as intelligible or as unintelligible. He must make it plausible that there really
are rich accounts for each of the concepts we employ. He must also say something about
the entirely general notions of reference and truth on which he is relying.
Friends of possible worlds semantics will remark that many aspects of concepts are
captured if, following Chalmers, they are treated as epistemic intensions (Chalmers
2002). This approach uses a notion of an epistemically possible world. The epistemic
intension of an expression is a function from such worlds to extensions. It is the function
that maps any such world w to the extension the expression would have if w were actual.
Such an approach has also to explain grasp of a concept. The advocate of this approach
could say that grasp of the concept consists in tacit knowledge of the principle associat-
ing such a world with the extension the concept would have if the world were actual.
One challenge this approach faces is that of saying why the worlds are needed at all.
The treatment must involve some principle determining, for each expression, an exten-
sion from a world (if it is actual). Whatever that principle is, it seems that it already fixes
a concept or sense, without the need to mention worlds. With such a modification, we
obtain a version of the Fregean view that concepts are individuated by their fundamen-
tal reference rules. All the same, the approach via epistemic intensions is valuable in
explaining the consequences of the Fregean approach when modal operators are in play.
Some accounts of concepts that do invoke reference and truth do so by specifying ref-
erential constraints that apply on top of a level that itself does not immediately involve
reference and truth. Before we can consider such mixed approaches, we have first to
consider treatments of concepts that contain only that reference-free level.

Type II Theories
The most common theories of concepts that aim to eschew reference to reference are
conceptual role theories. These theories hold in common that a concept is individuated
by its role in a thinker’s psychology. This approach treats individuation of a concept
and the account of its grasp simultaneously. Within this very general category, there are
multiple varieties of conceptual role theory. Some aim to individuate a concept by what

223
 CH RI STO PH ER P E A C O C K E

a thinker treats as evidence for certain contents containing it. Others, more pragma-
tist in spirit, mention the consequences in a speaker’s thought and action of accepting
certain contents containing the concept.
Cutting across this distinction are two others. Some theories are intrinsically norma-
tive. They aim to individuate concepts in terms of good reasons for moving to, or mov-
ing from, acceptance of certain contents containing the concept. The other distinction
that cuts across all of the above is the line between those theories (‘individualist’ or
‘internalist’) that attempt to individuate a concept solely in terms of states which, it is
claimed, depend only on conditions within the skin of the thinker, as opposed to those
anti-individualist approaches that take into account environmental and possibly social
relations of the thinker who employs the concept. Influential conceptual role theories
falling within one or another of these categories have been advocated by Sellars (1974),
Field (1977), Block (1986), Brandom (1994), Harman (1999), and many others. Prinz
(2002) offers a theory inspired by classical empiricism, in treating concepts as a certain
kind of perceptually derived representation.
Conceptual role theories have proved tempting in part because it is highly plausible
that there are some constitutive links between the nature of particular concepts and
some aspects of their use by someone who possesses them. We ought to recognize some
special link between observational concepts and their application on the basis of per-
ception; between concepts of sensation and a thinker’s own experience of the sensations
in question; between particular logical and arithmetical concepts, and absolutely mini-
mal competence in the inferential and numerical application of these concepts.
Acknowledgment of these points does not, however, by itself support conceptual
role theories of concepts. For two questions are left unanswered by these acknowledged
links. Are the connections with some part of conceptual role consequential on some-
thing not itself explained in terms of conceptual role? And, are the connections with
conceptual role exhaustive of what individuates the concept?
On the first question, some links with conceptual role are equally explained on refer-
ence-involving theories. Suppose, as above, that a thinker’s grasp of the observational
concept oval is given by his tacit knowledge that an arbitrary thing x is oval just in case
x is of the same shape as things he himself can recognize in perception as oval. This tacit
knowledge will be manifested precisely in a special sensitivity in his application of the
concept to things he perceives to be oval. Similarly, if a thinker has tacit knowledge
that any content of the form A&B is true just in case A is true and B is true, that will
be manifested in his basic inferential practices involving the concept of conjunc-
tion. Similar points can be made for the aspects of conceptual role intrinsic to other
concepts.
On the second question, about the exhaustiveness of conceptual role, it is clear that
at least the particular features of conceptual role cited here fail to individuate fully the
concepts in question. We know what it is for objects too large or too small for us to
perceive to be oval. We know what it is for subjects other than ourselves to be in pain.
It is very implausible that in these cases, there is a conceptual role, not making refer-
ence to reference, that captures these aspects of the concept in question. It is a wholly
empirical matter what would be evidence for a proposition that some object too small
to be perceived by us is in fact oval. It also a wholly empirical matter what would be the
consequences of its being oval. A Quinean point applies here: what such evidence or
consequences might be depends on other empirical information. What the evidence or
consequences are is not something simply written into the identity of the concept oval.

224
 CO N CEP T S

The same applies to concepts of sensations when applied to others; to many past-tense
contents; and to many other contents that we judge.
On Type I approaches to concepts, these phenomena are entirely to be expected. We
appreciate that for an object we cannot perceive to be oval is for it to be the same shape
as the perceptibly oval things. For another subject to be in pain is for her to be in the
same subjective state you are in when you are in pain. There are complex and interest-
ing preconditions for grasping such conditions, such as having a conception of objective
space, and of multiple subjects of experience. These need philosophical elucidation;
but if that can be done, there is nothing illegitimate about such explications of under-
standing and grasp of the concepts involved. These explications do however place us in
the territory of Type I theories. They make essential use of the notion of the property
picked out, either in local or in first-person cases, and they make no sense without some
such commitment. The attractiveness of such explications suggests that reference has a
deeper role to play in the individuation of concepts than is often recognized.
We can distinguish various grades of involvement of reference in an account of con-
cepts. In the conceptual role semantics of Sellars and Brandom, it plays no ineliminable
role. At an intermediate level, we can conceive of mixed theories that claim to individu-
ate concepts by conceptual role but which also require these roles to meet some general
constraints. One such constraint might be that the transitions specified in the individua-
tion of the concept be at least truth preserving as an a priori matter. The mixed accounts
may also try to connect the determination of reference to this truth-preservation. Such
was the approach I investigated in the late 1980s and early 1990s (Peacocke 1992). At a
third level, of more far-reaching involvement of reference with concepts, as illustrated in
the treatments of the preceding paragraph, we cannot even specify what makes something
the concept it is without bringing in referential relations in explaining what determines its
extension. Correspondingly, we cannot elucidate grasp without mentioning tacit knowl-
edge of conditions that involve reference.

Type III Theories

Type III theories hold that it is a mistake about the nature of concepts to seek a substan-
tive theory of concepts, a mistaken goal pursued both by proponents of theories of Type
I and by proponents of theories of Type II. The most developed form of ‘no-theory the-
ory’ of this third type is that offered by Schiffer (2003). Under an approach of this style,
the basic way in which we introduce talk of concepts is by regarding ‘John falls under the
concept is in pain’ as a way of rewriting ‘John is in pain’. According to Schiffer, this is a
something-from-nothing transformation. While it is legitimate to talk of an ontology of
concepts, atomic concepts cannot on his view be used to explain our grasp of complex
contents, nor can they be used in explanation of our understanding of language. On
Schiffer’s view, the meaning of ‘John is in pain’ is given in advance of, and cannot be
explained by, anything having to do with concepts. The idea of an explanatory theory
of concepts is largely a mistake, according to his Type III theory.
Under this Type III treatment, it is illegitimate to explain a person’s understand-
ing of a language compositionally, by associating concepts with atomic expressions. In
his exchanges with me, Schiffer has defended this aspect of his view, and the reader
is referred to that literature (Peacocke 1986, Schiffer 1986). Here I focus on another
challenge to such Type III theories. When we possess a particular concept, is in pain or
any other, we are able to assess, of an indefinite variety of circumstances, whether those

225
 CH RI STO PH ER P E A C O C K E

circumstances make it rational to apply the concept to an object given in a particular

way, or whether they make it rational to apply the negation of the concept to the object
so given, or whether they make it rational to remain neutral on the issue. We have an
open-ended capacity to make these assessments. Since our minds are finite, there must
be some finitary body of information on which we draw, no doubt subpersonally, in
making these assessments of rationality.
Type I theories holding that possession of a concept consists in tacit knowledge of a
substantive, nondisquotational reference can provide precisely such a finitary explana-
tion. The thinker draws on the information given in that substantive, non-disquotational
reference rule, and combines that with his other information and capacities that go
beyond what is constitutive of the concept to reach an assessment of the rationality of
judging the content in the specified circumstances. This gives an answer to the question
that any theory should, in one way or another, answer: ‘how does the concept-possessor
reach that assessment?’
Schiffer’s is not the only conceivable theory of Type III. One can imagine a variety of
minimalism about meaning which, by analogy with minimalist theories of truth, takes
the schema “The sentence ‘A’ means that A” as primitive, and tries to build an account
of meaning from that initial resource. Such a treatment would not take complex mean-
ings as substantive composed from meanings of component expressions, just as mini-
malists about truth do not regard instances of the disquotational schema as genuinely
explained by facts about the references of component expressions. All such views face
challenges in explaining language understanding, and our ability to assess the rational-
ity of attitudes, of the same kind as face Schiffer’s treatment.

Concepts and Rule-Following

We have a certain intuitive conception of applying a concept in a new case. In the new
case, the application of the concept is made for reasons, and the reasons are good (or
bad) in the light of the nature of the concept being applied, together with the facts of
the new case. This intuitive conception of rational application of concepts is challenged
by the sceptical arguments expounded by Kripke (1982) and Wright (1978). These
arguments imply that we cannot cite anything nontrivial that makes one rather than
another judgment involving a given concept correct in a new case. If these skeptical
arguments are sound, then the whole idea of the correctness of a judgment in a new
case being a result of the facts of the new case, together with the identity of the concept
being applied, has to be rejected.
Skeptical treatments commonly involve a communal or anthropocentric replace-
ment for the intuitive conception of correctness and rationality in a new case. Such
skeptical solutions abandon a certain form of objectivity of judgment, in that they hold
that the correctness of a judgment is not constituted independently of one or another
person’s willingness to make the judgment.
Under the conception of concepts as individuated by their fundamental reference
rules, there are straight, nonskeptical answers to these skeptical arguments. A judg-
ment applying a concept in a new case to a given object is correct if the object meets
the condition given in the fundamental reference rule for the concept. The judgment is
rational if it is reached by the thinker’s drawing on her tacit knowledge of the concept’s
fundamental reference rule. That knowledge constitutes her grasp of the concept. So
under this account, it is after all grasp of a concept that contributes to the explanation

226
 CO N CEP T S

of judgments made in new cases. One can expect the explanation to draw also on other
information possessed by the person applying the concept. It is only in a restricted class
of cases that the explanation of concept application draws only on a thinker’s grasp of
a concept.
Those who endorse the intuitive conception of rule-following are sometimes accused
of commitment to an unacceptable Platonism. What is said to be unacceptable is a
conception of correctness that outstrips, even in ordinary cases, anything we could be
supposed to know or to access. But it is a mistake to think that a proper rejection of the
inaccessible, in the description of ordinary cases, forces us into the rejection of the mini-
mal objectivity of judgment. The understanding-condition for oval given above involves
the notion of something’s being of the same shape as the perceptibly oval things. The
holding of such a relation of sameness of shape is something we can on occasion know
to obtain; it is not inaccessible; and it is not something judgment-dependent either.

Concepts and the a Priori

Just as Quine gave compelling arguments that nothing is true purely in virtue of mean-
ings, one can equally argue that nothing is true purely in virtue of the concepts compos-
ing it (Quine 1976). This point is still consistent with the existence of a deep connection
between the nature of concepts and the property of a content of being a priori—that is,
the property of being knowable without justificational reliance on perceptual experi-
ence or other conscious states.
When some state gives a reason for judging a conceptual content, that reason-giving
relation is founded in the nature of the concepts that compose the conceptual content
(Peacocke 2000). A genuine perception of something as oval can give reason for judging
that the object thus perceived is oval. The fundamental reference rule for the concept
oval explains why this should be so. It follows from that reference rule that when such a
perception occurs, the relevant conceptual content That’s oval is true.
Here we need to use the notion of the relatively a priori. It is relatively a priori that
the content That’s oval holds, given the occurrence of a genuine corresponding percep-
tion of the object in question as oval. Similarly, the occurrence to a subject of a pain
justifies him in judging I’m in pain. It is relatively a priori that the content I’m in pain
holds, give the occurrence to the thinker of a pain. Again, the fundamental reference
rules for pain and the first person concept I—that it refers to the agent of the thinking
containing it—explain why these justification relations hold.
This is not at all a conventionalist approach. Indeed, in appealing to the nature of
concepts, the approach is apparently explaining necessary relations (instances of the
relatively a priori) from other certain constitutive truths (the fundamental reference
rules for concepts).
The absolutely a priori contents form a limiting case of the relatively a priori contents.
They are those contents whose acceptance is justified on the basis of certain reasons,
such as proofs, where the justifying reasons do not involve perceptual or other conscious
events or states, and where the reasons can be ratified as sound on the basis of the funda-
mental reference rules for the concepts from which the contents are composed. Logical
and mathematical theorems arguably meet this condition. In developing this approach
to the absolutely a priori, it is important to respect the distinction between what gives
access to a justifying condition, and the justifying condition itself. We perceive proofs
of logical and mathematical theorems. The perception gives access to the proof. What

227
 CH RI STO PH ER P E A C O C K E

justifies is the abstract proof itself. Though proofs are perceived, the justifications to
which these perceptions gives access are not empirical.
It remains a challenge for the theory of concepts to discover how much of the a priori
can be explained by such resources.

Concepts and Relation-Based Thought

In some familiar cases, a subject is able to think of an object under a given concept
only because he stands in a certain relation to that object. The relation makes avail-
able the concept as a way of thinking of the object. This is widely acknowledged for
perceptual-demonstrative concepts of particular objects, such as that man, that tower. It
is also widely acknowledged for recognitional concepts of particular individuals and for
perceptible properties and relations. But in fact the phenomenon is far more widespread
than cases of these sorts.
For every recognizable kind of conscious psychological event or state, there is a con-
cept of it made available by the thinker’s ability to recognize when she is in that state,
or enjoying that event. Such concepts plausibly include our concepts of particular types
of sensation, perceptual experiences, and action-awarenesses. (Recognizability, which
involves memory, is of course less fine-grained than conscious discriminability.) Our
everyday concept of seeing, for instance, is not that of a state that plays a specified
explanatory role. Many people have false beliefs about the role in explanation played
by conscious experience. They think it explains the properties of concurrent action,
when in fact properties of such action are explained by nonconscious states sustained
by the older, dorsal route to representation (Milner and Goodale 1995). But those false
beliefs do not mean that their owners are not really exercising the concept of perceptual
experience. They are exercising that concept because of the recognitional relations in
which they stand to perceptual experience, not because of any beliefs they may or may
not have about the explanatory role of experience.
It is not only spatial, material, and mental states that can be thought about in ways
made available by the relations in which a thinker stands. Concepts (and meanings)
themselves can be thought about in ways made available by the thinker’s grasping those
concepts. A thinker can fully grasp the second-level concept the concept man, the con-
cept parliament only if she already grasps the concepts man and parliament respectively.
Moreover, just as in the example of the concept of perceptual experience, a thinker can
possess these concepts of concepts even if she has radically false beliefs about what it is
to possess or to be a concept.
What makes a thinker’s concept the concept F is not any theory she may have about
concepts, but rather the relation of this second-level concept to her own grasp of the
concept F itself. Suppose that F has the fundamental reference rule R. In grasping F, the
thinker has tacit knowledge of the reference rule R. To possess the concept the concept F
is then to have tacit knowledge that the condition for an arbitrary concept M to be the
reference of the concept F is for M to have the fundamental reference rule R. This tacit
knowledge is, in the nature of its content, available only to someone who grasps the
fundamental reference rule R. A similar point can be made about meanings.
Since philosophers and linguists disagree frequently about what individuates par-
ticular concepts and meanings, it could hardly be the case that correct theories of con-
cepts and meanings are necessary for thinking about a concept F as the concept F. The
generalization of a relation-based treatment to the case of concepts and meanings

228
 CO N CEP T S

explains why here, as elsewhere, false theory is consistent with a uniform notion
the concept F.

Concepts and the Metaphysics of a Domain

How should we conceive of the relation between a theory of concepts of a given domain
(of things, events, and properties) on the one hand, and the metaphysics of that domain
on the other? One Dummett-like thesis about the explanatory priority of a theory of con-
cepts generalizes what Dummett says about the theory of meaning. The thesis states that
metaphysical disputes in general can be resolved within, and should really be regarded
as disputes about, the theory of concepts (Dummett 1991: 13–15). An opposing view
holds that the metaphysics of a domain is explanatorily prior to the nature of concepts
of that domain. I argue here for this opposing view, which I label ‘metaphysics-first’.
The correct metaphysics for a domain constrains accounts of grasp of concepts of
objects and properties in that domain. It does so because it constrains the possible rela-
tions in which a thinker can stand to elements of that domain, and to the instantiation
of its distinctive properties and relations. There is a requirement of consonance of our
theory of grasp of concepts with the metaphysics of what falls under those concepts.
The metaphysics of some domains allows for the intelligibility of the causal operation
of objects, events, and properties in that domain upon thinkers. The primary class of
such cases is the domain of material objects, events, and their properties and relations
in space and time. These mind-independent things can stand in perceptual relations
to thinkers, and thus make available distinctive, perceptually based concepts of them.
Perceptual-demonstratives, recognitional concepts and observational concepts all have
fundamental reference rules that depend on and make reference to these perceptual
relations.
Some conscious mental events stand in causal relations to their owners, without
standing in perceptual relations. A visual experience, or a pain, may rationally cause the
engagement of a thinker’s recognitional capacity for just that kind of conscious event.
Such an account of some mental concepts requires an ontology of events falling under
them, rather than a treatment of such events as some kind of construct from expressive
action.
Unlike physical and mental events and their properties, other domains, such as those
of possible worlds and possible objects, do not stand in causal relations to any mental
states of thinkers, or to anything else. A fortiori, the fundamental reference rules for
such notions cannot mention causal relations to their subject matter. In such cases, it
is much more plausible that the fundamental reference rules must have a more intimate
relation to the very metaphysics of the subject matter. One general model for such cases
is that the thinker has some tacit knowledge of what makes something a possible world,
or a possible object. That knowledge is drawn upon in the thinker’s assessment of the
truth-value of thoughts whose truth conditions involve reference to possible worlds or
possible objects.
In this area, as everywhere else, we need to make use of the distinction between the
order of discovery and the order of philosophical explanation. For example, it may be
that in the first instance we cannot conceive, epistemically, of two distinct possible
objects with all of the same relations to actual objects and to other possible objects.
We may first conceive of this as an epistemic impossibility. But on reflection and fur-
ther philosophical theory, we reach the conclusion that the epistemic impossibility is

229
 CH RI STO PH ER P E A C O C K E

founded in a metaphysical impossibility. Possible objects are individuated in such a

way that there is no genuine metaphysical possibility of the kind in question. Here the
metaphysics remains prior in the order of philosophical explanation, even if it was not
temporally prior in the order of our thought.
There remain many outstanding issues about the relation of epistemology to meta-
physics, issues for which the theory of concepts is pivotal. Epistemic norms in a given
domain are a wide range of cases traceable to the nature of the concepts in the domain.
If the nature of those concepts is given by their fundamental reference rules, and those
rules are in turn constrained by the metaphysics, it follows that the metaphysics contrib-
utes to the explanation of, and constrains, epistemic norms. It remains to be seen how
this dependence plays out in detail across a huge range of domains, from such abstract
domains of thought about the infinite and the possible on the one hand, and on the
other to thought about consciousness, subjects of experience, and norms themselves.

References
Block, N. (1986) “Advertisement for a Semantics for Psychology,” Midwest Studies in Philosophy 10,
615–678
Brandom, R. (1994) Making It Explicit: Reasoning, Representing and Discursive Commitment (Cambridge, MA:
Harvard University Press)
Chalmers, D. (2002) “On Sense and Intension,” Philosophical Perspectives, 16, 135–182
Dummett, M. (1991) The Logical Basis of Metaphysics (Cambridge, MA: Harvard University Press)
Dummett, M. (1993) “What is a Theory of Meaning? (I)” in his collection The Seas of Language (Oxford:
Oxford University Press)
Field, H. (1977) “Logic, Meaning and Conceptual Role,” Journal of Philosophy 74, 379–409.
Fodor, J. (1998) Concepts: Where Cognitive Science Went Wrong (Oxford: Oxford University Press)
Frege, G. (1967) The Basic Laws of Arithmetic tr. and ed. M. Furth (Berkeley and Los Angeles: University of
California Press)
Frege, G. (1993) “On Sense and Reference” in A. Moore, ed., Meaning and Reference (Oxford: Oxford Uni-
versity Press)
Harman, G. (1999) Reasoning, Meaning and Mind (Oxford: Oxford University Press)
Kripke, S. (1982) Wittgenstein on Rules and Private Language (Oxford: Blackwell)
McDowell, J. (1998) “In Defence of Modesty”, in his collection Meaning, Knowledge and Reality (Cambridge,
MA: Harvard University Press)
Milner, D. and Goodale, M. (1995) The Visual Brain in Action (Oxford: Oxford University Press)
Peacocke, C. (1986) “Explanation in Computational Psychology: Language, Perception and Level 1.5,”
Mind and Language 1, 101–123.
Peacocke, C. (1992) A Study of Concepts (Cambridge, MA: MIT Press)
Peacocke, C. (2000) “Explaining the A Priori: The Programme of Moderate Rationalism” in New Essays on
the A Priori eds., P. Boghossian and C. Peacocke (Oxford: Oxford University Press).
Peacocke, C. (2008) Truly Understood (Oxford: Oxford University Press)
Prinz, J. (2002) Furnishing the Mind: Concepts and their Perceptual Basis (Cambridge, MA: MIT Press)
Quine, W. (1976) “Truth by Convention” in his The Ways of Paradox and Other Essays (Cambridge, MA:
Harvard University Press, Second Edition)
Schiffer, S. (1986) “Peacocke on Explanation in Psychology,” Mind and Language 1, 362–371
Schiffer, S. (2003) The Things We Mean (Oxford: Oxford University Press)
Sellars, W. (1974) “Meaning as Functional Classification”, Synthèse 27, 417–437
Wilson, M. (2006) Wandering Significance: An Essay on Conceptual Behavior (Oxford: Oxford University
Press)
Wright, C. (1978) Wittgenstein on the Foundations of Mathematics (London: Duckworth)

230
 2.5
ANAL YT IC T R UT H
Cory Juhl and Eric Loomis

Introduction and Overview

‘Analytic truth’ is ‘truth in virtue of meaning’, broadly construed. Statements that are
not analytically true are typically regarded as ‘synthetic’ statements for which truth or
falsity is determined, in part, by meaning- or language-independent matters of fact.
Other categories of statement, such as analytically false statements, may exist as well.
This chapter will briefly survey some of the historical uses of analytic truth, as well
as objections to it raised by W. V. O. Quine and others. We then turn to some con-
temporary discussions of analyticity, presenting our own account of analytic truth and
examining other recent accounts.

Historical Conceptions of Analytic Truth

Statements such as:

(i) All bodies are extended;

(ii) Everything is either B or not B;
(iii) If A is earlier than B and B is earlier than C, then A is earlier than C;
(iv) Nothing is both red and green all over, and;
(v) 1+1=2

have appeared to philosophers at different times to possess any one of a variety of prop-
erties, and have all been labeled ‘analytic truths’. Immanuel Kant regarded (i) as ana-
lytic on the grounds that the concept expressed by its predicate, ‘extended’, is ‘covertly
contained’ in that expressed by its subject, ‘body’ (Kant 1965: 48–9). This containment
implied, Kant thought, that analytic truths are knowable a priori, since one does not
“go beyond the concept ‘body’” in order to find extension “bound up with it.” Moreover,
such analytic truths are necessary, in the sense that the denial of an analytic truth such
as (i) would lead to a contradiction because, presumably, it would require saying that
extension both was and was not a property of body (1965: 189). For Kant however, a
priority and necessity are not only properties of analytic truths but also features of some
synthetic ones, such as the truths of arithmetic which he believed to be synthetic a
priori (53).
Subsequent philosophers such as Bernhard Bolzano found Kant’s theory of analytic
truth to be both vague and narrow. Bolzano described Kant’s notion of containment
 CO RY J U H L A N D E R I C L O O MI S

as “figurative” and permitting “too many interpretations” (1973: 201). Kant’s notion
was too narrow because it excluded statements such as (ii) from being analytic, for
there is no apparent sense in which being B or not-B is “contained” in the concept of
everything, even “covertly” (201). Bolzano thought a statement like (ii) ought to be
counted as analytic, for no matter what changes we make to the constituent non-logical
concept ‘B’, the statement remains true. Bolzano thus introduced a third feature com-
monly ascribed to one sort of analytic truth—preservation of truth upon the uniform
substitution of nonlogical expressions.
However, Bolzano recognized that even this idea is too narrow to capture a statement
like (iii) within the extension of analytic truth, since (iii)’s truth makes essential use of
the meaning of the nonlogical expression “is earlier than”; if we uniformly replace this
expression by certain others, such as “is the mother of,” it ceases to be true. Nor does
a statement like (iii) appear to be analytic either by Kant’s notion of containment or
by his claim that a statement is analytic just in case its denial leads to a contradiction.
Yet (iii) appeared to later philosophers to possess certain features, such as necessity and
knowability a priori, also supposedly possessed by analytic truths. Moreover, in the early
twentieth century, the emergence of more powerful systems of logic, due to Frege, Rus-
sell, Hilbert, and others, as well as the emergence of axiomatic systems of arithmetic
and related formal sciences, raised the possibility of a further expansion of the scope of
analytic truth to include such statements.
Many philosophers, most famously those of the Vienna Circle of logical empiricism,
thus awakened to the possibility that all necessary a priori statements, including those
of arithmetic, could be given a uniform characterization, thereby removing the need for
Kant’s distinction between analytic and synthetic a priori truths. Vienna Circle mem-
bers such as Rudolf Carnap, and sympathizers such as A. J. Ayer, expanded the notion
of analytic truth to include all statements that are necessary and knowable a priori.
Their doing so involved three important modifications to the notion of analytic truth.
First, they treated analytic truths as expressions of linguistic rules which constitute the
meanings of the expressions of a language. A statement like (iii) could thus be regarded
as partially constitutive of the meaning of “earlier than” (cf. Carnap 1937; Ayer 1952).
Even axiomatic systems for arithmetic or other formal sciences were seen as rules for
the use of the corresponding terms and the concepts thereby expressed. Second, they
considered analytic truths to be empty of “cognitive content” and thus “tautologies” in
the sense advanced by Wittgenstein in the Tractatus (Ayer 1952; Carnap 1956). That
is, while analytic truths may reveal aspects of linguistic usage qua linguistic rules, they
say nothing about how the world is. Third, Vienna Circle members accounted for the
a priori status of analytic truths by regarding them as conventions which were known a
priori in virtue of their status as stipulations (cf. Ayer 1952: 80; Carnap 1967: 178).
The Vienna Circle and those influenced by it thus treated analytic truths as possessing
a variety of characteristics: true by convention, knowable a priori, expressing rules for
the use of words, and “tautologous” or empty of cognitive content.

Objecting to Analytic Truth

In the early 1950s, Willard V. Quine launched a series of criticisms against the notion
of analytic truth that were later developed by others such as Gilbert Harman. Behind
Quine’s criticisms lay his belief that philosophy ought not to be regarded as funda-
mentally distinct from the natural sciences (cf. Quine 1960: 3–4, 24). For many,

232
 A N A LYTI C T R U T H

analytic truths appeared to have a distinctive status, namely to be such that their truth is
independent of matters of language-independent facts, including empirical facts. Ana-
lytic truths thus appeared to reveal a sphere of inquiry that could be conducted inde-
pendently of empirical science. So by criticizing analytic truth, Quine could be seen as
partially undermining any distinctive role or domain of inquiry for philosophy.
Quine’s criticisms of analytic truth took a variety of forms. Some of them rested
on Quine’s view that intensional notions, such as meaning and synonymy, are philo-
sophically suspect. For example, so-called Frege analytic truths, in which a statement is
analytic if it can be converted into a logical truth by the uniform substitution of syn-
onymous expressions, rely on intensional notions and therefore deemed by Quine to be
illicit (Quine 1953: 21–2). Another line of criticism, later developed by Harman, sug-
gested that analytic truths simply don’t exist as a matter of fact; “analytic truth” [is like a]
witch” (Harman 1967: 127). According to this objection, there may be agreement among
philosophers about which statements are called analytic truths, just as there may have
been agreement in some communities about which women were to be called witches. Yet
the extensions of both terms may turn out to be empty.
A third line of Quinean objections to analytic truth involved Quine’s endorsement of
a holistic conception of belief and confirmation that derived from Duhem, Carnap, and
others (cf. Carnap 1937: 317; Quine 1953: 43). According to this holism, a statement
can have empirical content only within a broader system of theory, a system which may
include the entire “fabric” of our beliefs (Quine 1953: 42). An individual hypothesis
H can only be verified given a theoretical background of statements of theory (such
as empirical laws), or other observations, methodological assumptions, mathematical
truths, and logical laws by which H may be inferred from the other statements. If H is
empirically tested and found to be false, any element of the theoretical background in
which it is embedded is, in principle, open to possible rejection. Thus, upon finding that
H is false, we may reject the supporting observations, or the statements of theory, the
methodological assumptions, or even the mathematical or logical laws used in inferring
H. Among these statements are some that the tradition regarded as analytic truths.
Yet we see how they may nonetheless be empirically rejected. As such, no statement
is immune to revision in the way that analytic truths were thought to be. Given any
alleged analytic truth, circumstances can be imagined in which subsequent empirical
discoveries lead to its rejection by leading to modifications of the “web” of belief in
which it is embedded (cf. Quine 1953: 43).
In recent decades, another challenge to analytic truth has emerged from the rise of
“semantic externalism,” deriving from work by Saul Kripke, Hilary Putnam, and others.
Kripke (1972) argued that the reference of proper names, like “Gödel,” and of natural-
kind terms like “gold” or “water,” depends upon factors external to any given language
user, such as causal and historical factors. Such considerations arguably undercut core
elements of the notion of analyticity as traditionally conceived (by the Vienna Circle,
for example). First, many philosophers were persuaded by Kripke and Putnam to cease
regarding the notions of a priori knowledge and necessary truth as coextensive. In par-
ticular, Kripke and Putnam argue that some necessary truths can be known only a pos-
teriori, and that some a priori truths might be contingent. Kripke argued that seemingly
a priori truths, such as that the standard meter stick in Paris is one meter long, might
turn out to be contingent. For while we might claim to know this a priori on the grounds
that it is stipulated, this should not be taken to show that it is necessarily true, Kripke
argued (1972: 55f). That stick S is one meter long is only a contingent feature; it may

233
 CO RY J U H L A N D E R I C L O O MI S

after all have a different length in some counterfactual situations in which it is heated
or stretched. If “necessary truth” is not coextensive with “truth knowable a priori,”
then an analytically true statement cannot be said to imply both necessity and a priori
knowability.
A second consequence of semantic externalism for analytic truth concerns the idea
that an analytic truth is one that is true in virtue of meaning, or one which someone who
understands the sentence must be disposed to assent to. This idea loses its relevance if
the meaning of a statement is not something that a competent user of a language can be
said to know just in virtue of their linguistic competence. Consider again Kripke’s gold
example: the meaning of “gold” is given through an a posteriori discovery by empirical
science, namely that gold has the atomic number 79. This was not known until rela-
tively recently. Nonetheless, English speakers prior to that discovery were competent in
their use of the word “gold” (a fact explained on Kripke’s account by saying that those
speakers nonetheless understood the reference-fixing explanations for “gold”). Seman-
tic externalism thus raises a challenge to the idea that possessing linguistic competence
in the use of certain expressions might be sufficient to account for either the knowledge
of or the truth of those expressions.
Several philosophers have recently explored these and other consequences of seman-
tic externalism for analyticity. Some have argued that semantic externalism renders
the notion of analyticity largely uninteresting (see Salmon 1993; Williamson 2007).
Williamson in particular has developed a series of arguments designed to raise doubts
concerning the view, originating with the Vienna Circle, that analytic truths are non-
cognitive, tautologous, or trivial. Williamson separates the majority of his objections
from those of Quine; unlike Quine he accepts synonymy and allied intensional notions
as legitimate (2007: 52–3). Yet, like Quine, he finds the notion of analytic truth to be
of little philosophical interest. Whether analytic truths are construed “metaphysically”
as truths whose meaning is sufficient for their truth (60–1), or “epistemologically” as
expressing truths knowable in virtue of meaning (77–8), Williamson argues that the
notion of analytic truth achieves few of the explanatory goals required by the philo-
sophical “linguistic turn” deriving from the Vienna Circle and its successors.

A Stipulative Conception of Analyticity

The objections to analytic truth that were raised by Quine, as well as those deriving
from semantic externalism, have led many contemporary philosophers to view the
notion with skepticism. Yet for a variety of reasons, variants of the notion continue
to be defended. Here is one: Let us say that a statement is a sentence-as-understood by
a community of speakers. A statement stands to a sentence roughly as a chess bishop
stands to a mini-sculpture of a bishop. When we describe a mini-sculpture as a chess
bishop we are treating it as an item understood to be governed by rules of use associated
with a particular activity, the game of chess. Similarly, when we describe a sentence as
a statement, we understand it to be governed by rules of use shared by other speakers.
For present purposes we focus on the relatively idealized situation in which all speak-
ers of the language understand the statement in the same way, and take the associated
sentence to be such that everyone takes it to be governed by the same rules.
A statement is analytic in our sense if its associated sentence is understood as
governed by rules of the following form:

234
 A N A LYTI C T R U T H

(Stip) Sentence s expresses a true proposition. Furthermore, that s expresses a

true proposition is empirically indefeasible.

In addition we introduce a stricter notion of analyticity that we will use when we discuss
mathematical and related applications, that of a transcendental stipulation:

(TStip) Sentence s expresses a true proposition. Furthermore, that s expresses a

true proposition is empirically indefeasible. Finally, the proposition p expressed
by s is also empirically indefeasible.

We introduce the notion of a transcendental stipulation in order to distinguish Kripke-

type cases such as the stipulation (J), “Julius is the unique inventor of the zipper (if there
is a unique inventor of the zipper),” from stipulations that “express” propositions that
are understood as empirically indefeasible. The proposition expressed by (J) is a propo-
sition whose truth or falsehood is supportable by empirical evidence, even though the
proposition that (J) expresses a true proposition is understood as empirically indefeasible.
In contrast, in mathematical and many other cases we take the proposition expressed by
the stipulation-sentence s to be empirically indefeasible (for details, cf. Juhl and Loomis
2010: ch. 6).

Responses to Some Objections to Analyticity

As we noted, Quine rejected all “intensional” notions and entities, and so refused to
accept accounts of analytic truth that relied on them. However, since most contem-
porary philosophers accept intensional notions, we will assume them in what follows.
Nonetheless, we wish to consider how the account introduced above fares against two
popular objections to analytic truth.

Objection 1: Synonymy is indeterminate (or there is no relation of synon-

ymy). Since analyticity relies on a notion of synonymy, analyticity is similarly
ill-defined.

The account of analyticity given above does not require appeal to synonymy. The char-
acterization appeals to the notion of a rule, and assumes as a constraint on acceptable
translation schemes that they preserve clear cases of analyticity. Many stipulative defi-
nitions are universally quantified biconditionals (UQBs) of the form “For any x, Fx iff
Gx.” It seems natural to say that F is synonymous with G when everyone agrees on the
stipulation. A focus on this sort of example may be what leads to the impression that
stipulations in general generate synonymies. Yet there is a more general phenomenon
that should be captured, the phenomenon of the adoption of a novel rule for a language,
one which concerns the empirical indefeasibility of the associated sentences. UQB stip-
ulations are relatively easy to understand as empirically indefeasible, since these logical
forms do not entail “categorical” statements that can then potentially conflict with
experience. Nevertheless, we think that the notion of a rule of the form Stip is in the
background in both these typical UQB cases and many other cases that need not have
this logical form. What is required for a coherent linguistic practice is that the rules of
the practice be coherent. In order to maintain coherence, all rules added concerning

235
 CO RY J U H L A N D E R I C L O O MI S

stipulations must be such that the collection of stipulations as a whole does not provide
evidence either for or against any possible experience.
The stipulative account also appeals to the notion of an acceptable translation
scheme. Any good scheme should translate any clearly analytic statement into a clearly
analytic statement, understood as governed by rules of that distinctive sort. This con-
straint insures that no analytic statement will count as given up on the basis of empiri-
cal counterevidence. Quineans appeal to principles of good or better translation, even
though they avoid appeal to synonymy. However, there is no need for a detour through
synonymy in order to defend analyticity.

Objection 2: Saying or believing that sentence s is true does not make s true.
The truth of any sentence is only justifiable on the basis of its playing an essen-
tial role in some best/good empirical theory of the world.

Williamson raises a version of this objection using what he calls “understanding-truth

links.” On one view of stipulations, what it is for stipulations to be true simply is for
people to have a convention of taking them to be true. According to such a view, there
is no gap, for certain sentences, between our believing that they are true and their being
true. Our sharing a belief in their truth, or adopting a convention of taking them to be
true, constitutes their being true. This is analogous to other cases such as whether some
item is “really” a chess bishop or another is “really” a stop sign. There is arguably noth-
ing more to being a chess bishop or a stop sign beyond our having some collection of
intentional states concerning the items. Stipulations and their truth might be thought
to be similar. If we permit stipulations in our language, and our practice is coherent,
then one could claim that there are trivial arguments demonstrating the truth of rel-
evant types of stipulations. In a mathematical proof, for example, if at some stage the
mathematician says, “f(a) = b,” and writes as justification “stipulative definition,” then
given the rules of proof, the truth of the sentence “f(a) = b” might count as proved.
It seems possible to remain agnostic on the question of whether stipulations within
a coherent stipulation practice are true, at least while reflecting on the practice, as
opposed to engaging in it. However, this agnosticism is compatible with recognizing
analytic truths’ philosophical interest. We may show why given our conventions it
seems natural to describe them as true (whether or not they “really are” true), and why
it is not puzzling that we could take them to be true in a distinctively risk-free way, i.e.,
without risk of empirical counterevidence. The practice is designed to permit such risk-
free assertions. Perhaps this is itself a matter to be resolved by stipulations governing
the extension of “true.”
The second part of this objection states that all sentences are to be justified in ways
broadly the same as all empirical statements are. This seems to us to be an optional, and
misleading, picture of how our language works, and a narrow way of understanding how
any possible language practice can be understood.

Mathematics and Other Applications of Analyticity

Thus far we have focused on stipulative definitions. Another possible case for analy-
ticity is mathematics. The current debate about the epistemology of mathematics is
dominated by two main positions: Platonist-type realists on the one hand, and empiri-
cists influenced by Quine and Mill on the other. Many philosophers are uncomfortable

236
 A N A LYTI C T R U T H

with an account that appeals to mathematical intuition, as realism tends to do. Yet
mathematical empiricisms do not seem to plausibly capture an obvious feature of math-
ematical practice: namely that mathematicians ignore empirical evidence of the sort
that empirical scientists require for their research. It is not that empiricists about math-
ematics have not tried. But the accounts given of which we are aware have not proven
satisfying for many philosophers of mathematics. Stipulationism (about mathemat-
ics—we will often omit the qualifier) offers an alternative to these views. Stipulation-
ists can maintain that mathematical axiom-statements share the epistemic, pragmatic,
and semantic profiles of transcendental stipulations, i.e., that they are transcendental
stipulations.
One can motivate stipulationism by noting that understanding mathematics as “ana-
lytic” sheds some light on the epistemology of mathematics and makes some problems
seem less pressing. For example, how do we know that there is a set of concreta {a,b}
whenever the concreta a and b are known to exist? It seems odd to say that we intuit
the existence of such an entity. It is also difficult to imagine setting up an experiment to
find out whether the set really exists over and above the members. It seems instead that
there is something like a stipulative definition of “set” such that a certain conditional
is understood as true or taken to be true in the way that stipulations are, so that given
any concreta x and y, there is a set containing exactly them. Stipulationism might be
extended to other cases of abstracta such as fictional objects, mereological sums, and
propositions (see Schiffer 2003; Juhl 2009). There are three objections to stipulationism
about mathematics that we will consider here.

The “existential” objection: You cannot stipulate things into existence. The
existence or nonexistence of Santa Claus, God, or other objects is a matter of
stipulation-independent fact.

The kinds of objects this objection invokes as examples are the sorts of things on
which empirical evidence is understood to bear. Santa’s existence is something for or
against which we can have empirical evidence; similarly for God, at least as classically
conceived (as a performer of miracles, for example). This is relevant because on a stipu-
lationist view of mathematics, the members of the language community understand
mathematical statements in such a way that the possibility of empirical evidence (the
existence of experiences with some character) bearing on their truth or justification is
ruled out. Speakers are also understood to make other statements for which empirical
evidence is not ruled out. If a given class of statements were understood to be such that
they both permitted experience to bear on them, and simultaneously ruled this possibil-
ity out, the resulting practice would be incoherent. Thus, in any coherent stipulation
practice, that is, any language that permits the introduction of empirically indefeasible
stipulations, stipulations cannot jointly have consequences on which empirical evi-
dence bears.
The upshot of a response to objection 1, then, is that for some kinds of things, or
some concepts, certain statements and/or facts involving them are understood as being
evidentially independent of experience. For such statements and their logical conse-
quences, a coherent stipulation practice can permit the introduction of statements with
arbitrary logical form, even existence claims. In that sense, at least, existence claims can
be stipulated. And the standard Santa Claus, unicorn, and God counterexamples do not
undermine this point (cf. Juhl and Loomis 2010).

237
 CO RY J U H L A N D E R I C L O O MI S

The “mathematics uses empirical evidence” objection: Empirical evidence frequently

counts for or against the truth of mathematical statements. For example, the
fact that a computer prints a certain pattern of ink, that a mathematician utters
a pattern of sound, or that a pattern appears in a textbook can all be empirical
evidence for the truth of various mathematical statements.

Williamson and others think that “experiments” are relevant to mathematical truth
(2007, p. 6). However, it is worth drawing a distinction between the way in which
empirical evidence bears on mathematical statements versus empirical hypotheses.
We can distinguish between first-order and second-order justifications. A first-order or
“canonical” justification for any (nonaxiomatic) mathematical statement consists of a
proof of it. No empirical statement is permitted to be appealed to in a proof. The fact
that a computer produces a physical output in the relevant cases is contingently cor-
related with the existence of a first-order justification, and provides second-order justi-
fication. Similarly for an utterance of a mathematician; we only count the utterance as
evidence if we think that, say, the fact that this person utters that p is correlated with
the existence of a first-order justification. In short, empirical evidence can bear on the
truth of a mathematical statement. But unlike ordinary empirical hypotheses, empirical
evidence bears only in a particular way: namely in this “second-order” way that provides
evidence for the existence of a first-order, purely nonempirical justification.

The “indispensability” objection: Mathematics is as indispensable to physical sci-

ence as electrons and quarks are, and this fact provides empirical evidence that
mathematical entities exist.

This objection sees the support that physical science grants to the existence of objects
like electrons as extending to mathematical “entities.” However, the fact that the enti-
ties referred to by the mathematical-looking terminology applied in physics are distinc-
tively mathematical—outside of space and time, knowable independently of experience,
and the like—plays no role in the explanations and calculations used in physical sci-
ence. For consider that we can take any theory of mathematized physics, and in effect
“Ramseyfy” it (i.e., “quantify out” all predicate terms with second-order quantifiers, and
similarly for constants and objectual quantifiers). The resulting theory would be exactly
as empirically adequate as the original theory. Precisely what parts of the Ramseyfied
theory would be empirically supported in the general case is a complicated matter. But it
is perfectly coherent to say that in such a Ramseyfied theory there is no term referring to
anything distinctively mathematical. Thus, even if every part of the Ramseyfied theory
is taken to be empirically defeasible, the stipulationist practice in which properly math-
ematical statements are treated as empirically indefeasible would remain coherent.
The problem that we are suggesting for the indispensability argument, then, is
this: the entities and predicates being distinctively mathematical does no predictive or
explanatory work. Thus, the claim that mathematical entities per se are indispensable is
not supported by successful empirical applications of mathematical language. All that is
genuinely supported is the existence of some structures or other.
We will take the Quinean view to argue that one must be empiricist about mathemat-
ics. Stipulationists hope to show that one need not be empiricist about mathematics.
Stipulationism yields a description of mathematics that fits better with actual practice
of mathematics and actual scientists than the alternative, empiricist stipulation.

238
 A N A LYTI C T R U T H

This debate about the nature of mathematical statements connects to the general
Quinean picture according to which all statements are empirically defeasible. The
stipulationist thinks that the case of mathematics exposes Quine’s view as an optional
proposal, one that does not fit actual mathematical or scientific practice.

Other Recent Developments

Recently Gillian Russell has produced an important book on analyticity (Russell 2008).
Russell takes as her paradigm case the sentence “I am here now.” There is a clear and
natural sense in which token utterances of that type are (almost) always true, and true
“in virtue of the meanings of the words therein.” Russell takes the notion of analytic
truth to be, at bottom, the notion of a statement or sentence whose truth is explainable
solely by appeal to something akin to meanings of the component words. Her account of
analytic truth is thus one primarily concerned with semantic issues, not epistemological
ones, although she grants the possibility that competent speakers may know the truth
of an analytic claim solely through knowing facts about “reference determiners” of its
component words. As such, Russell’s account, unlike the stipulative account above, is
compatible with the possibility that a broadly Quinean account of mathematical and
other allegedly a priori truths is correct.
According to another approach to analyticity, (cf. Peacocke 1995; Boghossian 1996;
Glock 2003), linguistic or conceptual competence with some term or concept requires
belief in the truth of some statements. In some recent works, both Boghossian and
Peacocke have argued for a priori justification of these meaning-conferring principles
on the basis of their special meaning-conferring status. Against such views, Williamson
argues that linguistic competence may never require belief in any statement or proposi-
tion (2007: 73ff). And even if the competence requirement is correct, a number of crit-
ics (Salmon 1993; Horwich 1998; Williamson 2007) have argued that there is no way
to show that the statements that are meaning-conferring are therefore true or a priori
justified.
One of Williamson’s arguments against the view that linguistic competence entails
believing certain statements considers two philosophers who do not believe that all
vixens are vixens (Williamson 2007: 85ff). One rejects the principle because he thinks
that conditionals whose antecedents and consequents are truth valueless are themselves
truth valueless. This philosopher thinks that some entities x are such that a statement
that x is a vixen is truth valueless. Williamson argues that this case and others can be
constructed in such a way that we will judge that the person in question understands the
notion vixen (as well as implication and universal quantification). Thus linguistic com-
petence does not entail belief, even for this very basic logical truth. Williamson thinks
that this strategy can be generalized, showing that no belief is ever absolutely necessary
for linguistic or conceptual competence.
Closely related to the claim about linguistic competence is the view that various
principles “implicitly define” some terms, and that their role in conferring meaning
entails that the defining principles are true. Horwich gives an argument against the
related idea that meaning-conferring principles must be true (Horwich 1998: 136ff).
Horwich notes that novel theoretical terms can be introduced by “defining” the terms
by appeal to theoretical principles or hypotheses that are taken to hold true for them. It
is plausible to describe such principles as “conferring meaning” on the terms that they
are intended to define. Yet consider the classic case of a now-rejected theoretical type,

239
 CO RY J U H L A N D E R I C L O O MI S

phlogiston. The term “phlogiston” had some theoretical principles that were taken to
hold of it, and which arguably conferred meaning on the term. Yet as it turns out, there
is no phlogiston. To the extent that the theoretical principles entailed that phlogiston
was the cause of some event type, the principles are false. This remains true, even if we
grant that those very principles somehow “define” or confer meaning upon the term
“phlogiston.” Since Horwich thinks that the analytic/synthetic distinction was prima-
rily introduced to resolve some epistemic problems associated with a priori knowledge
or justification, he takes the phlogiston example to show that there is no interesting
analytic/synthetic distinction.
It is possible, Horwich thinks, to grant that the “non-substantive” conditional of
the form “If anything has (theoretical profile) T, then phlogiston has T” is knowable
a priori, in sharp contrast to the truth of the “substantive” defining statements, which
are the consequents of the “trivial” conditional. We think that this tempting move is
misguided. First of all, Horwich does not explain what he means by “substantive” when
he says that logical principles and mathematical principles are substantive, whereas the
conditional above is not. One might think that since the conditional is not logically
true, if logical truths are already substantive, as Horwich thinks they are, then such
a conditional is even more substantive. In any case, given Horwich’s sympathy with
Quine’s confirmational holism and Quine’s rejection of a priori knowledge, it seems
that Horwich (and other Quineans) should think that even the conditional might be
revised under some imaginable empirical circumstances. The conditional has the form,
“If any theoretical concept has feature T, then phlogiston does.” Suppose that physi-
cists were to find that, after all, there is some fluid that satisfies T (its presence causes
temperature to rise, etc.), but that this fluid is not the fluid that physicists would have
been interacting with using their very low-tech instruments of the nineteenth century.
The fluid that does satisfy T requires sophisticated twenty-seventh century superstring-
detection devices to interact with it systematically. In such a situation, physicists argu-
ably should deny the conditional, “If anything has T, then phlogiston does,” since some-
thing satisfies T, as it turns out, whereas phlogiston does not exist. The example is not
much more or less far-fetched than many Quinean thought experiments purportedly
showing that some analytic statement “might be given up.” Furthermore, any attempt
to rule out scenarios by putting more and more information concerning phlogiston into
T makes it less plausible to claim that everything within T is required for conferring
meaning on the term. Thus we think that according to Horwich’s Quinean background
assumptions, even the conditional is “revisable,” and not a priori justified. His move
from the ordinary stipulation to the conditional seems to be a mistake.
That said, the conditional can be coherently stipulated. That is, a conditional of this
form can be held to be true and empirically indefeasible, if it does not have empirical
consequences by itself. Like the UQBs we discussed earlier, such a conditional has a
logical form that is readily immunized from counterevidence. However, mathematical
statements are similarly easily immunized from the empirical. And if it introduces rules
for a novel term into the language, the consequent of a Horwich conditional can be
stipulated as well, as long as it does not bear on even the probability of the truth of any
empirical statement. An important difference between the empirical cases like phlogis-
ton and mathematical cases is that, in practice, scientists actually do give up theoretical
principles taken to be true of a theoretical notion, even if that notion was initially intro-
duced or “defined” via its role in an empirical theory. Mathematical practice contrasts
sharply with empirical science practice in this respect.

240
 A N A LYTI C T R U T H

Examples like phlogiston are interesting but often overgeneralized. One can grant
that for properly theoretical notions, our practices as a matter of fact seem to be of a sort
such that practically any principle originally introduced as “defining” could eventually
be rejected on the basis of empirical counterevidence. In contrast, as a matter of fact
mathematicians practically never justify the rejection of any defining principles on the
basis of empirical evidence. So generalizing from empirical practice to mathematical
practice is overgeneralization. Horwich seems to think that there are substantive and
nonsubstantive statements, and that mathematics and logic fall on the substantive side.
We are instead inclined to draw a distinction between statements understood as empiri-
cally indefeasible and statements understood as defeasible, and to place mathematics in
the former category and most statements of empirical science in the latter.
None of the criticisms against analyticity to date are sufficient to rule it out in every
form, and many of the disputes in ontology and metaontology may benefit from a much
closer and more sympathetic treatment of analyticity, such as the recent accounts can-
vassed above. There remains space for analyticity to have an important and epistemi-
cally illuminating role, even if it cannot be deployed as ambitiously as some philoso-
phers have thought.

References
Ayer, Alfred J. (1952) Language Truth and Logic. New York: Dover Publications Inc.
Boghossian, Paul. (1996) “Analyticity Reconsidered,” Noûs, 30:3, 360–391.
Bolzano, Bernard (1973) Theory of Science. B. Terrell (translator). Dordrecht: D. Reidel.
Carnap, Rudolf (1937) The Logical Syntax of Language. London: Routledge and Kegan Paul.
Carnap, Rudolf (1956) “Empiricism, Semantics, and Ontology.” In Meaning and Necessity, A Study in Seman-
tics and Modal Logic. Enlarged Edition. Chicago: University of Chicago Press, 205–221.
Carnap, Rudolf (1967) The Logical Structure of the World and Pseudoproblems in Philosophy. Second Edition,
R. A. George (translator). Berkeley: University of California Press.
Glock, H. J. (2003) Quine and Davidson on Thought, Language and Reality. Cambridge: Cambridge University
Press.
Harman, G. (1967) “Quine on Meaning and Existence I,” Review of Metaphysics 21, 124–151.
Horwich, P. (1998) Meaning. Oxford: Oxford University Press.
Juhl, C. (2009) “Pure and Impure Stipulata,” Philosophy and Phenomenological Research, LXXIX: 3, 637–652.
Juhl, C., and Loomis, E. (2010) Analyticity. New York: Routledge.
Kant, I. (1965) Critique of Pure Reason. N. K. Smith (translator). New York: St. Martin’s Press.
Kripke, Saul A. (1972) Naming and Necessity. Cambridge, MA: Harvard University Press.
Kripke, Saul (1980) Naming and Necessity. Cambridge, MA: Harvard University Press.
Peacocke, Christopher (1995) A Study of Concepts. Cambridge, MA: MIT Press.
Quine, W. V. O. (1953) “Two Dogmas of Empiricism.” In From a Logical Point of View. Cambridge, MA:
Harvard University Press, 21–46.
Quine, W. V. O. (1960) Word and Object. Cambridge, MA: MIT Press.
Russell, Gillian. (2008) Truth in Virtue of Meaning. Oxford: Oxford University Press.
Salmon, N. (1993) “Analyticity and Apriority,” Philosophical Perspectives 7, Language and Logic: 125–133.
Schiffer, S. (2003) The Things We Mean. Oxford: Oxford University Press.
Williamson, Timothy. (2007) The Philosophy of Philosophy. Oxford: Blackwell.

241
 2.6
POS S IBL E W OR LD S
S E MANT IC S
Daniel Nolan

One approach to specifying the meaning of pieces of languages is to treat those mean-
ings as constructions out of possible worlds and possible objects. This technique is useful
both in logic and in providing the semantics of natural languages. After introducing
possible worlds semantics, this chapter will outline some of the applications that have
convinced many philosophers and linguists of the usefulness of this framework, and will
conclude with a discussion of one of the most conspicuous limitations of the framework,
and some ways this limitation has been dealt with.

Extension and Intension

Pieces of language have extensions: predicates are associated with sets of objects that
satisfy the predicate, sentences have truth-values, and singular referring expressions
designate objects. Some aspects of the meanings of expressions seem to rely on more
than the extensions of those expressions: extensions are often contrasted with “inten-
sions.” In the context of possible worlds semantics, expressions are assigned possible
worlds intensions that are constructed from possible worlds and possible objects using
set theoretic means. For example, the standard possible worlds intension of a predicate
is a function from possible worlds to extensions that yields the predicate’s extension
in each possible world. (These functions are sometimes called the properties associated
with predicates.) One standard way of defining the intensions of declarative sentences
is as a function from possible worlds to truth-values (a function that assigns a sentence
True when it is true according to a world and False otherwise), or, as a simpler construc-
tion that carries the same information, the set of possible worlds where the sentence
is true. These sets of possible worlds are often referred to as possible worlds propositions
expressed by sentences.

Possible Worlds Semantics in Logic and Language

The expression “possible worlds semantics” was first used to describe “semantics” in the
logician’s sense. In this sense, possible worlds semantics is a matter of associating with
a given logic a model that contains worlds; and assignments, relative to those worlds,
of the truth-values of sentences, extensions of predicates, and so on. The best-known
 PO SSI BLE- W O RLD S E MA N T I C S

application of possible worlds semantics is in the semantics of modal logics, usually

attributed to Kripke 1959 and 1963, though see Copeland 2000 for the “prehistory” of
this semantics.
Consider, for example, the propositional modal logic S4, which extends propositional
logic with two operators applied to well-formed formulas: (I) for “it is necessary that
I” and¡(I) for “it is possible that I.” A model of S4 can be given by a ordered triple <W,
R, Q>, where W is a set of worlds, R is a two-place relation on worlds, and Q is an evalua-
tion function from propositional letters and worlds to the truth-values 1 and 0 (for True
and False, respectively). R is an “accessibility relation” between worlds, and to ensure
the model will model S4, we insist that R be both reflexive (so that for all wW, Rww)
and transitive (so that for all w, x, y W, when Rwx and Rxy then Rwy). To specify Q,
we first insist that Q assign exactly one of 1 and 0 to each propositional letter for each
wW. We will represent Q assigning a propositional letter p the truth value 1 at world w
as follows: Qw(p) = 1, and in general assignments of truth-values at worlds will be repre-
sented in the form QD(I) = J for a world D, proposition I and truth-valueJ.
Next, we specify that Q assigns propositional-logical compounds truth-values, given
the truth-value assignments to propositional letters in each world, as those truth-values
are built up in propositional logic. So, for example, when Qw(p) = 1 and Qw(q) = 0, then
Qw(p & q) = 0, Qw(p v q) = 1, Qw(~p) = 0, Qw(q Šp) = 0, and so on. Finally, we specify how
formulas with and ¡ are to be treated by Q. When I is a well-formed formula, (I)is to
be assigned the truth value 1 at a world w by Q (Qw(I) = 1) provided that all the worlds
v such that Rwv are also such that Qv(I) = 1. In other words, a sentence I is specified to
be necessarily true at a world w if and only if in every possible world accessible from w
(or “possible from” w) is a world where I is true. The condition for ¡ is that (Qw(¡I) =
1) provided that some world v such that Rwv is also such that Qv(I) = 1. In other words,
I is possibly true in a world w just in case some world “possible from” w is one where I
is true.
Propositional S4 is provably sound and complete with respect to these models. When-
ever a conclusion * is provable from a set of premises 6 in S4, then every model which
satisfies the above conditions and in which all the premises 6 are true at a world is one
where * is also true at that world. Conversely, whenever it is that case that, in every
model of the above sort, *is true at every world where all of 6are true, then * is prov-
able from 6 in the logic S4. Because of these facts, we can use the model theory to tell
us which arguments are valid in S4 and which are not: if we can provide an S4 counter-
model of an argument from some premises 6 to a conclusion *, then we can tell that *
will not be provable from 6 in our S4 proof theory, for example.
Such models can be complicated to handle predicate calculus versions of S4, as well
as still richer versions of S4. The model given above can also be varied to yield classes
of models for other modal logics: in particular, we get quite a range of different modal
logics just by putting different constraints on the R relation. Possible worlds models like
these, whether for propositional logic or for predicate logic, have enabled modal logi-
cians to see connections between a number of different systems, and in particular seem
to have helped logicians to get a better understanding of the behaviour of iterated modal
operators, such as in the formula ¡p.
Models of this sort have been used in a number of different areas in logic besides the
treatment of “necessarily” and “possibly.” Similar semantics have been offered for tem-
poral logics (for example the logics of “until now,” “from now on” etc.), deontic logics
(the logics of “ought,” “permitted,” and so on), and some epistemic and doxastic logics

243
 DA N I EL N O L A N

(logics of “it is known that . . .” or “. . . is (epistemically) justified”). Note that while

the formal structures are similar, the “worlds” in these models are often better thought
of as times (in the temporal case) or epistemic states (in some epistemic logics). There
is an extended sense of “possible worlds semantics” where the models for a logic are a
possible worlds semantics provided that they are similar, in the right kinds of ways, to
the Kripke models for the logics of necessity and possibility. Compare the way that dif-
ferent mathematical systems are known as “geometries” even if they bear little relation
to the structure of space, provided they bear certain kinds of similarities to systems like
Euclid’s geometry.
Other kinds of relations on worlds have been found useful by philosophical logicians
besides the accessibility relations of normal modal logics. Many nonclassical Relevant
Logics use a ternary relation on worlds to model the implication relations in those log-
ics (see Routley et al. 1983). “Conditional logics” like that of Stalnaker 1968 and Lewis
1973 use a “closeness” relation on worlds to model conditionals, particularly counter-
factual conditionals. Interpretations of the language of probability sometimes employ
measures on sets of worlds: This is one way to apply Kolmogorov’s mathematical theory
of probability.
One thing that it is worth noting about the models used by logicians to investigate
the formal properties of logics is that these models can be very useful even if they have
nothing in particular to do with the project of providing the meanings of the logical
operators and connectives: We could still employ the model theory to help with finding
counterexamples to invalid inferences, for example. On the other hand, in linguistics
and philosophy of language “semantics” is the study of the meaning of expressions (per-
haps particularly that aspect of meaning which is in common across different uses of the
same word). A number of linguists and philosophers have found it useful to investigate
the semantics of natural languages, such as English, employing a framework of possible
worlds. “Possible worlds semantics,” in this sense, involves using theories that postulate
possible worlds in order either to give the meanings of, for example, English expressions,
or at least to illuminate how those meanings work.
The two senses of “possible worlds semantics” are connected. Richard Montague
introduced possible worlds into the systematic study of the semantics of natural lan-
guage, and his approach was to take a simplified fragment of English, and then treat it as
logicians had treated their artificial formal languages. Montague offered a model theory
for his fragment of English that employed possible worlds, possible objects, and various
set-theoretic constructions from them, in order to systematically yield assignments of
truth-conditions to sentences: that is, assignments of truth-values to sentences relative
to each world of the model. His reason for offering this model theory was to exhibit
the way that the meanings of different English words go together to produce truth-
conditions of English sentences. His semantics, in the model-theory sense, was the
core of his semantics, in the sense of a theory about the meaning of English construc-
tions. Montague’s system can be, and has been, applied to many other languages besides
English, and Montague grammar remains a flourishing research program in linguistics.
Montague (1974) is the best place to find Montague’s own contributions.

What Are Possible Worlds?

Possible worlds semantics relies on there being a domain of possible worlds, and usually
requires things in those worlds to be the members of the sets associated with predicates

244
 PO SSI BLE- W O RLD S E MA N T I C S

in each world. So we have possible worlds and possible individuals. Does that mean
that, in order to use possible worlds semantics, we need to think that there is an infi-
nite range of alternate universes, full of merely possible individuals, including strange
individuals like golden mountains and talking donkeys?
Some have argued that this is indeed the best way to understand possible worlds and
possible individuals (Lewis 1986). More common is the view that possible worlds are
some kind of abstract object. Whatever they are, they need to be able to have sentences
or propositions “true according to” them, but there are a number of ways this might
happen. A world literally containing a singing cow would have “A cow sings” true
according to it in one straightforward way. But if worlds are representations, they might
be able to have “A cow sings” true according to them in the same way that a newspaper
that contains the sentence “A cow sings” can. And there are other options: Perhaps a
world composed of states of affairs that do not obtain could have as a component the
state of affairs of a cow singing, for example. Or we could offer a modal analysis of “true
according to”: Perhaps a claim is true according to a world if, and only if, were that world
actualized, the claim would be true. Finally, a theorist might resist analysis of the “true
according to” relation altogether, and just take it to be an unanalyzed fact that “A cow
sings” is true according to one possible world but not according to another.
Some logicians and philosophers of language will be happy to leave the question
of what possible worlds are, and what truth according to a world is, to the metaphysi-
cians. But it is worth noting that the answer given to what possible worlds are, and in
particular what the relation of truth-according-to-a-world is, may put constraints on
what theoretical purposes possible worlds are suitable for. If the modal analysis of “true
according to” is relied upon, then it will be much harder to treat the equivalence of
“necessarily I”and “at all (accessible) possible worlds, I” as explaining necessity: we can
use modality to explain what possible worlds are, or possible worlds to explain modality,
but using each to explain the other risks circularity.
More worryingly, if we treat possible worlds as representations—as sets of sentences,
or sets of propositions, or “maximal” propositions that settle every question—then we
risk undermining their explanatory role in a theory of meaning. If we are explaining
the meaning of sentences by associating them with sets of possible worlds, then on pain
of circularity we should not take possible worlds to be sets of interpreted sentences, for
example.
There are many other questions that could be raised about possible worlds, and
indeed about possible objects. But for some purposes the metaphysics does not matter.
For example, when using possible worlds semantics in the logicians sense, whether the
members of W or the members of the domains associated with each wW are anything
like real possible worlds and the full range of merely possible objects is irrelevant from
the technical point of view.
Indeed, even if we are using possible worlds for the project of semantics in the linguis-
tics sense, we may be able to treat them as only formal modeling devices. Whether or
not we can do so will depend on what sort of explanations they are meant to provide—if
they are invoked only for modeling structural constraints on meanings, it may not mat-
ter what the objects in the models are. On the other hand, if we take ourselves to be
illuminating the meaning of expressions by providing genuine truth-conditions for sen-
tences, then the “worlds” and constructions from them should arguably have something
to do with what can possibly be the case.

245
 DA N I EL N O L A N

Possible Worlds Semantics for Natural Language: A Simple Model

Each kind of linguistic construction is assigned an “intension” or “semantic value” con-
structed from possible worlds, possible objects, and sets ultimately built up from these.
Sentences are assigned, as their intensions, sets of possible worlds: these are the worlds at
which the sentences are true, so in assigning these sets we are specifying the conditions
under which the sentences are true. These possible world truth conditions are identified
as propositions in this framework. Predicates are assigned functions from worlds to sets
of possible objects: the set associated with a predicate and a world is the extension of
that predicate at that world. These functions from worlds to extensions are identified as
properties in this framework. For an object to have a property, in a world, is for the object
to be in the extension of that property at that world.
We can provide intensions for quantifiers: the intension of the universal and existen-
tial quantifiers are certain function from worlds to sets of properties, in the above sense.
The universal quantifier, for example, is assigned, at each world, the set of properties in
common to all the objects at that world, and the existential quantifier is assigned the set
of properties such that at least one thing at that world has that property.
The intensions of larger linguistic structures are functions of the intensions of their
components, so, for example, we can build up the intension of a sentence from the
intensions of its constituents. The rule for a sentence of the form [quantifier];[predicate],
for example, is that a world w is in the intension of the sentence if the intension of the
quantifier is a function from w to a set of properties that includes the property associated
with the predicate. This may become clearer with an example: suppose our sentence is
“something is hungry.” When we evaluate a world, say w5, the intension of “something”
will associate with w5 a set of properties (the set of properties had by at least one thing
in w5). Suppose the property that is the intension of “is hungry” is a member of that set.
Then w5 will be one of the members of the proposition associated with “something is
hungry.” Which is as it should be, since those conditions insure that one of the objects
in w5 is in the extension of the property associated with “is hungry.” If one of the objects
has the property of being hungry in a world, we want “something is hungry” to count as
true there.
If we wanted uniform principles about how different semantic categories went
together, and the two kinds of basic sentences we allowed were [quantifier];[predicate]
and [name];[predicate] sentences, then it would make sense to give names the same
general kind of semantic values as quantifiers. Montague (1973) did this, treating names
as also being associated with sets of properties: “Barack Obama,” for example, would be
associated with a function from worlds to the set of properties that Barack Obama has at
those worlds. “Barack Obama is hungry” would then have a world w in its intension just
in case the set of properties associated with “Barack Obama” at w included the property
associated with “is hungry.” Alternatively, we could have different rules for different
kinds of sentences: For example, we could allow the semantic value of “Barack Obama”
just to be an individual, and the rule for [name];[predicate] sentences to be that a world
w is in the intension associated with such a sentence provided that the individual that
is the intension of the name is in the extension, at w, of the property associated with
the predicate.
Extensional connectives such as “and,” “or,” and “not” can be assigned intensions:
“And,” for example, can take two propositions as inputs and yield a proposition as out-
put (it, in effect, it functions as a set-theoretic intersection on sets of worlds: I and M is

246
 PO SSI BLE- W O RLD S E MA N T I C S

true at the set of worlds that is the intersection of the set of worlds where I is true and
the one where M is true). Intensional operators like the sentential adverbs “necessarily”
and “possibly” can also be treated as functions from propositions to propositions. The
simplest rules are these (where I is a sentence): necessarily I has, as its intension, the
proposition containing every world just in case the intension of Iincludes every world,
and has a null intension otherwise. Possibly I has, as its intension, the proposition con-
taining every world just in case the intension of I includes at least one world, and has
the null set as its intension otherwise. More complicated rules involving accessibility
relations need to be invoked if more complex notions of necessity and possibility are
in play.
Obviously the simple theory presented so far barely scratches the surface of the com-
plexity of natural language: it does not even handle tense or the sorts of sentences we
would represent in predicate calculus with multiple variables. The interaction of the
syntax and semantics has not been addressed either. But hopefully it gives the flavor
of how such a theory might be set up. To see how a possible worlds semantics might
earn its keep, we should discuss some more difficult constructions that possible worlds
semantics has had some success with.

Some Applications of Possible Worlds Semantics for Natural Language

Possible worlds semantics is useful, as we have seen above, for handling sentential
adverbs like “necessarily” and “possibly.” It is useful not just for an “unrestricted” sense
of necessity and possibility, corresponding to truth in every possible world or some pos-
sible world, respectively. It is also useful when dealing with a restricted necessity, such
as saying what is necessary given the laws of nature, or what is feasible given some
other constraints. (This corresponds to a restriction on the set of worlds relevant for the
semantic value.) As was noted in the section above on logic, possible worlds semantics
has also been found very useful in dealing with the so-called deontic modalities such
as “It is obligatory that . . .” or “It is permitted that . . .” and their more natural-sound-
ing English relatives, and also for dealing with so-called epistemic modalities used for
specifying what is known or what is justifiably believed, or what “has to be” or “might
be” in the epistemic sense of those expressions. (See Kratzer 1981 for a possible worlds
treatment of “ought.”)
Possible worlds semantics come into their own for a range of nonextensional uses
of language. For example, consider so-called intensional transitive verbs. Examples
include “seeks,” “owes,” “wants,” “offers,” and a number of others. Whether “John seeks
a city of gold” is true or not does not just depend on the extensions of “John” and “a city
of gold.” Suppose, as is plausible, that there are no cities of gold, and no flying horses.
“A city of gold” and “a flying horse” would thus plausibly have the same extension, but
“John seeks a city of gold” can be true while “John seeks a flying horse” is false, so “seeks”
is intensional. If instead of extensions we use possible world intensions, we can draw
the distinction we need. There are possible cities of gold, and possible flying horses,
so “a city of gold” and “a flying horse” have different intensions—you could see those
intensions as being the property of being a city of gold and the property of being a flying horse,
respectively. Since the intensions are different, if we treat the intension of “seeks” as, in
this case, a function from the two intensions that flank it to a proposition, then we can
get the right answer, in this example, that the first sentence has the truth-value True at
the actual world and the second does not.

247
 DA N I EL N O L A N

One of the applications that caught the attention of the philosophical community
following Stalnaker 1968 and Lewis 1973 was the use of possible worlds semantics to
give a theory of the truth-conditions for conditional sentences. According to Stal-
naker’s proposal, for example, “if A then B” is true at a world w just in case a “selection
function,” which is a function from propositions and worlds to worlds, when it takes as
arguments the proposition associated with A and the world w, yields a world x that is a
member of the proposition associated with B. The effect of this is that the conditional
is true at a world if the “closest” world to w where A is true is also one where B is true.
This formal proposal yields a variety of interesting logics of conditionals depending on
what constraints are put on the selection function, and deals with a number of problems
that faced traditional theories of conditionals. For example, for “if A then B” to be true
seems to require more than that either A is false of B is true (the “material conditional”
account): I am currently sitting down, but it does not seem true that “if I had stood up,
I would have exploded.” On the other hand, requiring that for “if A then B” to be true
B has to be true in every possible world where A is true is too strong a requirement:
“If I had stood up, I would not have exploded” seems true, though there is nothing
absolutely impossible about a situation where I stand up and then explode. A lot of
contemporary work in exploring the semantics of conditionals employs some relative
of this approach.
A third important application of possible worlds semantics is in the semantics of
propositional attitude ascriptions. Sentences like “John believes the number of planets
is 10” or “Mary desires that everyone love someone” seem to require a relationship
between a person and a proposition: The propositions expressed by “the number of
planets is 10” and “everyone love someone,” respectively. Constructions like “believes
that” are not extensional: substituting one true sentence for another in the scope of
“believes that” does not, in general, preserve truth-value. Furthermore, the behavior
of propositional attitude contexts is too complex to just treat “that I” constructions
as, in effect, referring to propositions. “There is someone such that everyone believes
that he is the murderer” cannot be treated as referring to a proposition with “that
he is the murderer,” since “he” is in effect bound by “There is someone.” Moving to
a possible worlds semantics allows phenomena like this to be dealt with better than
existing extensional frameworks.
The above only scratches the surface of work done in the possible worlds tradition.
Using possible worlds semantics arguably illuminates topics in semantics as diverse as
adverbs, progressives, dispositional expressions, the semantics of embedded questions,
and plays a vital role in theorising about effect of context, including in the development
of dynamic semantics. For an introductory discussion to more applications than could
be discussed here, see Partee (1989), especially section III.

The Problem of Necessary Co-Extension

Some sentences are not true at any possible worlds, and some predicates are true of any
possible objects (“is round and not-round” might serve as an example). Furthermore,
some pairs of sentences are true in some possible worlds but are true in all the same pos-
sible worlds, and some pairs of predicates hold of exactly the same possible objects. Since
orthodox possible worlds semantics takes sentences true at the same worlds to have the
same intensions, and predicates true of the same possible objects to share intensions,
this leads to some surprising results.

248
 PO SSI BLE- W O RLD S E MA N T I C S

The possible worlds semantics treatment of intensional transitive verbs, for example,
misfires when impossibilities get involved, or at least the straightforward version of pos-
sible worlds semantics does. Suppose Alan is seeking a proof of the rationality of S. Such
a proof is impossible, so the set of objects, across possible worlds, that meet the condi-
tion of being a proof of the rationality of S is the null set. Now, suppose Ben is seeking
a round square: Again, given the impossibility of round squares, the condition of being
a round square corresponds to the null set. “Proof of the rationality of S” and “round
square” have the same possible worlds intension. So we should be able to conclude, by
substitution of equivalences, that Alan is seeking a round square, and Ben is seeking a
proof of the rationality of S. But that substitution is not valid: Alan seems well within
his rights to deny that he is seeking a round square, his behavior is not that of someone
who is seeking a round square, and our psychological insight has failed if we report that
Alan is doing that. Similar problems arise for verb phrases such as “prays to,” “try to,”
“wish for,” “offer,” and even “owe”—I might owe you a proof of the rationality of S, if I
promised to deliver one.
Propositional attitude ascriptions also are dealt with badly by the basic possible
worlds semantic treatment: anyone who believes one strictly impossible thing (a thing
true in no possible world) believes them all, for example. As well as the problem of
impossible contents, propositional attitude ascriptions have problems with necessar-
ily true contents: Anyone who believes one necessarily true thing believes them all,
according to this model, but even if mathematical truths are necessary, they are not that
easy to discover! There are also problems with beliefs about contingent matters where
what seem to be two different belief contents obtain in the same possible worlds. Sup-
pose Robin Hood and Robin of Locksley are the same person, though few suspect this.
The Sherriff of Nottingham can believe that Robin Hood robbed the abbot, while not
believing that Robin of Locksley robbed the abbot. Or so we would ordinarily think. If
“Robin Hood” and “Robin of Locksley” are rigid designators, and pick out the same indi-
vidual across possible worlds, then the two names have the same possible worlds inten-
sion. So, according to the straightforward possible worlds semantics for names at least,
“The Sherriff believes Robin Hood robbed an abbot” is necessarily equivalent to “The
Sherriff believes Robin of Locksley robbed an abbot”: the Sherriff cannot believe one
without believing the other. A similar problem will arise for natural kind terms, given
the usual theory of how they work—anyone who believes there is water in their glass
also believes there is H2O in their glass, according to a straightforward possible worlds
semantics. But not everyone believes that water is H2O.
Finally, the orthodox possible worlds treatment of conditional statements gives results
that many find odd. Conditionals with necessarily false antecedents that share a conse-
quent must all be treated the same, since those antecedents all share an intension: the
standard line (for example Lewis 1973: 24–26) is to treat all conditionals with necessar-
ily false antecedents as true at all worlds, regardless of the consequent. However, many
people give different verdicts on different “counterpossible” conditionals: “If 8 had been
divisible without remainder only by 1 and itself, it would have been prime” seems good,
but “If 8 had been divisible without remainder only by 1, 2, 5 and itself, it would have
been prime” looks less appealing: in the latter case it is tempting to say it would not have
been prime, but it would have been composite instead. (See Nolan (1997) for a case for
positing counterpossible conditionals with non-trivial truth values.)
There are a number of responses available to advocates of possible worlds semantics
here. The most straightforward is to bite the bullet: Alan is seeking a round square,

249
 DA N I EL N O L A N

though he might not put it that way; the Sherriff does believe Robin of Locksley robbed
an abbot; everyone always knew that water was H2O, and so on. Biting these bullets
often goes with a strategy of explaining the pragmatics of these utterances so that they
are somehow misleading or incorrect to say, even if they are not false (this seems to be
the proposal about some cases in Stalnaker 1978, for example).
Another relatively straightforward response is to say that possible worlds semantics
is a model of the meanings of linguistic expressions that is useful for some purposes but
not others: perhaps it is analogous to Newtonian mechanics, which is a good enough
approximation when it comes to bridge-building but not the right physics to use when
designing particle accelerators. If this approach is taken, it leaves open the question
about what the more fully accurate theory might look like—but for some projects, it
makes sense to put that question off.
Another response is to say that a semantics of natural language should use construc-
tions from possible worlds as one of the components of a “meaning”: perhaps semantic
values should be pairs of possible worlds constructions and something more linguistic,
for example, or that possible worlds intensions should be components in a “structured
meaning” that looks rather like a syntactic structure, as in Creswell (1985). There are
potentially as many different versions of this strategy as there are components we could
add to a semantic value, so it is difficult to generalize about the prospects of this way of
developing possible worlds semantics.
A way of complicating the postulated intensions without bringing in things besides
constructions from possible worlds is to treat intensions as “two-dimensional inten-
sions.” David Kaplan pioneered this move to deal with demonstratives and indexicals:
expressions like “that” or “I” or “tomorrow” that make different contributions to truth-
conditions on different occasions of use (Kaplan 1989). “It will rain tomorrow” can
be true when I say it today but false when I say it tomorrow, for example. Kaplan held
that the meaning of an expression like “that” or “tomorrow” had two aspects: a content,
which was in effect a standard possible worlds intension, and a character, which was a
function from certain n-tuples (with components like the speaker, and the time of utter-
ance) to contents. Such n-tuples can be generalized to worlds of utterance, or worlds of
utterance plus a “center” to indicate which speaker, time, audience etc. in the world is
being picked out—these are sometimes called the ‘context’ of the utterance.
While Kaplan intended his framework for a limited range of context-dependent
expressions, later “two-dimensional” theorists have offered similar treatments for proper
names and natural kind terms. The contents of proper names pick out the same object
in each world, like a standard possible worlds intention, but the character associated
with a name maps the name to different contents depending on features such as which
object is at the source of the reference-chain associated with the indicated speaker,
or the referential intentions of the designated speaker. In this framework, there are
contexts which take “Robin Hood” and “Robin of Locksley” to different contents, and
since the overall functions-from-contexts-to-content intensions are thus different for
the two names, there is scope for those intensions to interact differently with pieces of
language such as “believes that.” (See Garcia-Carpintero and Macia (2006) for a col-
lection containing a number of papers for and against such generalized two-dimensional
treatments.)
A final kind of modification for orthodox possible worlds semantics to deal with the
problem of necessary co-extension is to be more generous with what worlds the theory
allows. We got the problems because some expressions had equivalent applications in

250
 PO SSI BLE- W O RLD S E MA N T I C S

every possible world: “a proof of the rationality of S” and “a round square”, for example,
or “Robin Hood” and “Robin of Locksley”. If we allowed worlds to have the extensions
of these expressions come apart from each other, these problems would not arise.
One way to do this, which is tempting in the case of propositional attitude ascriptions
and intensional transitives involving the activity of agents, is to use epistemically pos-
sible worlds or doxastically possible worlds. These worlds might be more generous than
standard possible worlds: For example, if it is possible to believe I without believing M,
there are “worlds” where Iand Mcome apart, and connectedly that the extensions of
two expressions (like “water” and “H2O” or “proof of the rationality of S” and “round
square”) come apart at some world if it is possible to believe that an object is in one
extension but not in the other. Or one could go even further and drop the restriction
to do with our capacities to believe: If a worlds semantics allowed arbitrary impossible
worlds as well as possible worlds, the problem of necessary co-extension could be guar-
anteed to not arise.
Possible worlds semantics has provided a powerful and impressive framework for sys-
tematically dealing with the semantics of puzzling constructions in language. Whether
the theory of semantics it yields needs to be extended by other devices, and if so what
those devices are, will continue to be one of the important debates in the philosophy of
language in the coming decades.

Related Topics
1.1 Extensions, Intensions, Character, and Beyond
3.1 Names
5.4 Modal Logic
5.9 Montague Grammar
6.6 Necessity and Meaning
6.7 Propositional Attitude Reports.

References
Copeland, B.J. 2000. “The Genesis of Possible Worlds Semantics,” The Journal of Philosophical Logic 31:
99–137
Cresswell, M. 1985. Structured Meanings. MIT Press, Cambridge MA.
Garcia-Carpintero, M. and Macia, J. (eds.) 2006. Two-Dimensional Semantics: Foundations and Applications.
Oxford University Press, Oxford.
Kaplan, D. 1989. “Demonstratives,” in Almog, J., Perry, J., and Wettstein, H. Themes from Kaplan. Oxford
University Press, Oxford. 481–563.
Kratzer, A. 1981. “The Notional Category of Modality,” in Eikmeyer, H.J. and Rieser H. (eds.) Words,
Worlds, and Contexts, de Gruyter, Berlin: 38–74.
Kripke, S. 1959. “A Completeness Theorem in Modal Logic,” Journal of Symbolic Logic 24: 1–14.
Kripke, S. 1963. “Semantical Considerations on Modal Logic,” Acta Philosophica Fennica 16: 83–94
Lewis, D. 1973. Counterfactuals. Blackwell, Oxford.
Lewis, D. 1986. On the Plurality of Worlds. Blackwell, Oxford.
Montague, R. 1973. “The Proper Treatment of Quantification in Ordinary English,” reprinted in Montague
1974, 247–270.
Montague, R. 1974. Formal Philosophy: Selected Papers of Richard Montague. (Thomason, R. ed.) Yale Uni-
versity Press, New Haven.
Nolan, D. 1997. “Impossible Worlds: A Modest Approach,” Notre Dame Journal for Formal Logic 38.4:
535–572
Partee, B. 1989. “Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective,” in S. Allen

251
 DA N I EL N O L A N

(ed.) Possible Worlds in Humanities, Arts and Sciences. Proceedings of Nobel Symposium 65. Walter de Gru-
yter, New York, 93–123.
Routley, R., Meyer, R.K., Plumwood, V. and Brady, R. 1983. Relevant Logics and its Rivals, Volume I. Rid-
geview, Atascardero, CA.
Stalnaker, R. 1968. “A Theory of Conditionals,” in Rescher, N. (ed.) Studies in Logical Theory, American
Philosophical Quarterly Monograph Series, no.2. Basil Blackwell, Oxford., 98–112
Stalnaker, R. 1978. “Assertion,” in Cole, P. (ed). Syntax and Semantics: Pragmatics Vol. 9. Academic Press,
New York, 315–332

252
 2.7
DY NAMIC S EM ANT IC S
Seth Yalcin

1 Introduction
‘Dynamic semantics’ gets used in various, not always compatible ways. On perhaps its most
inclusive use, it applies to a range of semantic systems embracing a thesis we might call:

discourse primacy. It is fundamentally entirely discourses that have truth-

conditions (or more broadly, informational content). Individual sentences
have truth-conditions in at best a derivative sense, insofar as they have some
potential to impact the truth-conditions of a discourse.

This thesis breaks with a long semantic tradition—going back at least to Frege, and
running through Montague and beyond—which revolves around the individual sen-
tence, articulating the meaning of all linguistic expressions in terms of their contribu-
tions to the truth-conditions of a sentence. By shifting the locus of truth-conditions in
systematic theorizing to the discourse, discourse primacy shifts the traditional center
of gravity in semantics. It recommends that we articulate the meaning of a sentence in
terms of its potential to contribute to the truth-conditions, or informational content,
of discourses in which it can occur. This idea was first developed by Kamp [1981] and
(independently) by Heim [1982], in works that formed the starting point for subsequent
theorizing in the dynamic tradition.
Exactly how to model a discourse is itself a theoretical question on which theorists
working in the spirit of discourse primacy may differ. But a common theme in the
dynamic tradition is to model a discourse via aspects of the mental states of the agents
in conversation. For instance, Kamp [1981] models a discourse via a certain kind of
a structured representation (a discourse representation structure, or DRS), which (he
postulates) is an abstract model of the kind of mental representation involved in
linguistic processing and understanding. On his picture, it is fundamentally these
structures that have truth-conditions. And in Heim [1982], the fundamental bearers
of truth-conditions are the coordinated states of presupposition of the participants
of the conversation. It is the content of these states that the meanings of sentences
impact.
Once we take the dynamic turn and accept discourse primacy, we come to a fork
in the road concerning how to articulate the meanings of sentences. One path accepts
a thesis we can call dynamic representation; the other, a thesis we can call dynamic
interpretation.
 SETH YA L C I N

dynamic representation. Sentences encode instructions for updating a cer-

tain kind of representation, and it is this representation, not sentences or other
linguistic expressions, that is the primary object of compositional semantic
interpretation.
dynamic interpretation. The compositional semantic value of a sentence is
a function from a body of information to a body of information.

The first thesis we find in Kamp [1981]; the second in Heim [1982]. The theses are
obviously different. As noted, Kamp’s account centrally involves a certain intermedi-
ate layer of representation, a DRS. He supplies a systematic construction procedure for
mapping a sentence or sequence of sentences to a DRS, and a systematic procedure
for interpreting DRSs. The resulting system is not locally compositional in the usual
sense, for it does not directly associate linguistic expressions with interpretations. (For
relevant discussion see Groenendijk and Stokhof [1991], Muskens [1996], Muskens et
al. [2011], van Eijck and Kamp [2011].)
We don’t directly discuss dynamic representation here. It is better treated against
the larger context of Discourse Representation Theory (DRT), the family of approaches
to semantic theorizing—some straightforwardly compositional, some not—that fol-
lowed in the footsteps of Kamp [1981]. For introductions to DRT, see Kamp and Reyle
[1993], Geurts and Beaver [2007], van Eijck and Kamp [2011].
On a more restrictive use of ‘dynamic semantics’, it applies just to those accounts that
embrace dynamic interpretation. This restricted use is the one in play henceforth. This
article is about the import and upshot of dynamic interpretation—for short, dynamic.
It is about the idea that the compositional semantic value of a sentence is the sort of
thing that updates a body of information—often, though not always, by adding to it more
information. In the jargon, it is about the idea is that the semantic value of a sentence is
a context-change potential, or update function, or program on a body of information.
To get a handle on dynamic, contrast it with its much more traditional cousin,
static:

static. The compositional semantic value of a sentence is a truth-condition—

a function from possible worlds to truth-values, or more generally, a function
from indices (points of evaluation with a model) to truth-values.

We can think of static and dynamic as competing high-level empirical hypotheses

about what kind of thing the semantic value of a sentence is.
We have two aims. First, we should like a basic sense of how, in principle, dynamic
could be incorporated into a theory which performs the sort of explanatory work tradi-
tionally expected from static.
This we tackle in section 2. Second, we should like a sense of some of the specific
linguistic phenomena that have been taken to recommend dynamic over static. This
we broach in section 3.

2 Dynamic Semantic Values

How, in principle, might dynamic semantic values be put to work in explaining the kind
of phenomena we have always expected our semantic values to cover?

254
 DYN A M I C SEMA N T I C S

That prompts the prior question: what is it we expect semantic values to explain?
Theorists of course differ in exactly what work they expect this notion to perform, but
let us focus on a few of the most usual explananda.
First there is the productivity of language use: the empirical fact that competent speakers
can understand and produce complex expressions that they have never before encoun-
tered. This is a typical motivation—perhaps the leading motivation—for embracing
compositionality. [See entry on compositionality.] The assumption of compositional-
ity is a point of common ground between static and dynamic. Typical defenders of
each assume compositionality partly with an eye towards explaining the productivity of
language use
More illuminating differences between the sort of theory one gets out of dynamic
as opposed to static emerge when we look at other explanatory demands we typi-
cally place on semantic values. Besides helping to explain productivity by being com-
positional, the semantic values of sentences are normally expected to do the following
explanatory work:

(i) The semantic values of sentences should play a central role in explaining how
speakers communicate, and in particular, transfer information, by using the sentence.

(ii) The semantic values of sentences should play a central role in explaining
which sentences follow from which, and which sentences are incompatible with
which.

These tasks will be approached in importantly different ways, depending on whether

one embraces static or dynamic.

2.1 Communication
Take first (i). How, assuming static, is meaning leveraged to communicate information?
A familiar and traditional story is the following. By static, sentences have truth-condi-
tions. We take it these truth-conditions constitute a core aspect if not the core aspect,
of the informational content normally communicated in context by an assertion of
the sentence. Building on Grice and Stalnaker, we take it communication takes place
against a background of mutually shared presuppositions—a common ground—and that
the characteristic pragmatic impact of an assertion is to add the informational content
determined by the sentence uttered to the common ground of the conversation. Thus
an assertion normally serves to update the common ground, partly as a function of the
truth-conditional content of the sentence asserted in context. It is a feature of the prag-
matics of assertion that the common ground is normally to be updated in this way. (See
Stalnaker [1975], [1978].)
This story assumes static, but note that on it an asserted sentence is nevertheless
associated with a rule for updating the common ground—in effect, with what we have
called an update function, or context change potential. How then does dynamic differ
from static in respect of (i)? The core difference is that on a dynamic view, the context
change potential of a complex sentence is determined compositionally by the semantic
values of its parts. Ancillary pragmatic assumptions are not required. And correspond-
ingly, the compositional contribution of an embedded sentence—the semantic ingredi-
ent the sentence contributes to the larger linguistic expressions in which it occurs—is

255
 SETH YA L C I N

itself taken to be a context change potential. Thus the context change potential of a
conjunction, for instance, is a literally function of the context change potentials of
its parts. This is quite different than the static version of the story, where the context
change potential of a sentence is only determined with the help of pragmatics, and
applies only in connection with unembedded, asserted sentences.
When it comes to conversational update, the dynamic view rolls into the semantics
what, on a static view, would be parceled to pragmatics. The compositional contribu-
tion of a sentence is just identified with an instruction for updating the informational
context, or common ground, of a conversation.

2.2 Consequence
Turn next to (ii), the demand that the semantic values of sentences play a central role
in explaining which sentences follow from, or are consequences of, which. On a classic
static view, the story is straightforward: sentences have truth-conditions, and conse-
quence is a matter of truth-preservation. An argument is valid, we can say, just in case
if the premises are true, the conclusion must also be true.
Since a dynamic semantics does not generally associate sentences with truth-
conditions, it cannot tell the same story. (It might try to tell the same story indi-
rectly, by associating all declarative sentences with context-change potentials which
determine, in some systematic way, truth-conditions. But this would be a severe
constraint, and one precluding many of the motivating applications for dynamic
semantics in the literature.) To cover the entailments evidenced to us by the produc-
tions of competent speakers, and so perform the explanatory work required by (ii),
dynamic semantic systems are typically equipped with alternative formalizations of
consequence.
Central to most of these formalizations is the notion of a fixed point of a sentence.
The fixed points of a sentence will be those bodies of information which, when
operated on by the update function semantically expressed by the sentence, return
just that body of information back. Intuitively, the fixed points of a sentence will
be those bodies of information that already incorporate the informational update
associated with the sentence. The fixed points of a sentence I are said to support I,
or accept I.
Generally speaking, the update function of a sentence will map bodies of information
to fixed points of that sentence. For example, on the dynamic model, when an asser-
tion gets accepted in conversation, the common ground of the conversation normally
becomes a fixed point of the sentence. (That is why we generally don’t say things twice:
once our conversation is fixed point of a sentence, uttering the sentence will not—at
least as a matter of the semantics of the sentence—alter the information that is com-
mon ground.)
Once we have the idea of a fixed point of a sentence, we can define consequence
in terms of it. Here there are various options, of which we will just mention one
(for additional discussion and further options, see for instance van Benthem [1996],
Veltman [1996], Beaver [2001], Muskens et al. [2011] and references cited therein).
Consequence may be defined as follows: A conclusion \ follows from some premises
I1, . . ., In just in case every body of information which is a fixed point of the premises
is also a fixed point of the conclusion. On this analysis, consequence is the sort of thing
that preserves acceptance. Alternatively put, consequence is an informational update-

256
 DYN A M I C SEMA N T I C S

preserving relation between sentences. The thought is that once a state of information
is updated with the premises of a valid argument, it already incorporates the update
associated with the conclusion.
The idea that the data of semantics largely concerns the truth-conditions of sen-
tences is sometimes treated as a platitude. But in a context in which we are debating
the virtues of static versus dynamic, we must acknowledge that this is really a the-
ory-laden and question-begging way to describe the matter. The dynamic approach
offers a rival characterization of the data of semantics: The data are, at the most
general, data about the potential of a sentence to update or change a conversation. It
is fundamentally these kinds of facts that judgments about consequence and compat-
ibility track.
Lewis has famously written that “Semantics with no treatment of truth-conditions
is not semantics” ([Lewis, 1970, 18]). What he meant to underscore in saying this was
that a semantics should associate expressions with interpretations as opposed to mere
translations. It is worth emphasizing that in this respect, dynamic semantics and truth-
conditional semantics are on the same side of the fence: They are each card-carrying
members of the interpretive (representational, model-theoretic) tradition in seman-
tics. (But there is the related question of whether the role of an intermediate layer of
representation is doing nontrivial explanatory work; we return to this question, in the
context of Heim’s account, briefly in section 3.3.)
Still, dynamic semantics is disruptive to traditional distinctions in other ways. Nota-
bly, talk of sentences expressing propositions, or having propositional content, often
lacks clear sense in a dynamic setting, as does the traditional distinction between a sen-
tence’s informational content and its force. The reason, again, is that the chief locus of
informational content on a dynamic picture is not the sentence but rather the states of
presupposition of the participants in the discourse. It is the content of these states that
fixes the common ground, and sentence meanings are foremost tools for manipulating
this shared body of informational content.

2.3 A Sample Semantics: Negation and Conjunction

The preceding gives an abstract idea of how dynamic might in principle be worked
into a theory which explains the sort of things we expect a compositional semantics
to explain. But a real sense for the dynamic approach can only come through working
with concrete examples. For illustrative purposes, then, let us consider an example of
a toy dynamic semantics, one defined in an ordinary artificial propositional language
L containing just negation and conjunction. We will soon turn to more complicated
systems.

Defs. A model M for L is a triple ¢W, I² where W is a set of possible worlds, and
I is an interpretation function mapping the propositional letters of L to sets of
worlds.
Def. A context set in M is any subset of WM.
Def. For any M, a dynamic valuation [ ] (for M) is a function from wffs of Lto
functions from context sets (in M) to context sets (in M) recursively defined as
follows, where Dis any propositional letter, Iand \ are any wffs, and c is any
context set in M:

257
 SETH YA L C I N

c[D] = c I(D)
c[¬I] = c − c[I]
c[Iš \] = c[I][\]
Def. If Iis a wff, [I] is the context change potential (or update function) of I.

Note we are employing postfix notation: the argument of context change potential is
written to its left, rather than its right. This allows us to use the left–right direction to
reflect the order in which a sequence of sentences updates a body of information.
Central to any dynamic semantics is a stand about how to formally model the
informational context of a conversation—the object that dynamic semantic value act
upon. The above semantics encodes the idea of representing the informational com-
mon ground of a conversation via a context set, the set of possible worlds left open by
what is mutually presupposed by the interlocutors, an idea first developed in detail by
Stalnaker [1975, 1978]. Information growth in conversation is a matter of eliminating
possibilities from the context set. Sentences characteristically act to knock worlds out
of context sets, thereby growing the amount of information taken for granted in the
conversation.
This semantics illustrates the way in which the context change potential of a com-
plex sentence may be defined in terms of the context change potentials of its constitu-
ent parts. Now we can formalize the dynamic notions of acceptance and consequence
informally described above:

Def. c accepts I just in case c [I] = c.

Def. A sentence I is a dynamic consequence of a set of sentences *, * _=i I,
just in case for any c, whenever all of the sentences of *are accepted in c, so
too is I.

And from here, it is not hard to see that this semantics validates the same patterns of
inference as would an ordinary classical semantics defined on the same language. Thus
(Iš \) _= I is validated, ™™I _= Iis validated, and so on. (See van Benthem [1986,
1996], Groenendijk et al. [1995], von Fintel and Gillies [2007], and Muskens et al.
[2011] for more discussion and relevant results.)
This toy dynamic semantics has two notable properties: it is eliminative and distributive.

Def. A dynamic valuation function [·] is eliminative just in case s[I] Ž s, for all
s, I.
Def. An dynamic valuation function [·] is distributive just in case s[I] = is {i} [I],
for all s, I.

Distributivity says that the output context of any update just as well might be arrived at
by updating on singletons in the prior context and aggregating the results. Eliminativity
entails that this aggregate set will always be some subset of the original context.
These two features entail that our toy dynamic valuation function acts much like a
sieve. If we run a bucket of flour through a sieve, the output is a new, smaller quantity of
flour from that bucket. Which particles the sieve lets through is entirely a matter of the
individual particles and their sizes. If we had sieved half of the bucket first and then the
second half, we still would have gotten the same output flour.

258
 DYN A M I C SEMA N T I C S

If your dynamic semantics is eliminative and distributive, it means that it is boring

enough to easily implement statically (van Benthem [1986]). More precisely, it means
that each sentence I corresponds to a static condition, PI, on points, such that:

c[I]= c ˆ PI (cf. the assertion rule of Stalnaker [1978])

for any context c. It means, in other words, that we might as well have an ordinary static
compositional semantics together with the above simple rule for mapping the static
condition a sentence expresses into its context change potential.
What this suggests is that if there is a reason to prefer dynamic to static, it is because
there is reason to believe in more sophisticated updates on contexts, ones that go beyond
the mere sieving we find in eliminative, distributive dynamic systems. And indeed, the
dynamic approaches motivated in the literature typically lack eliminativity, distributiv-
ity, or both.
Is the failure of eliminativity and distributivity necessary and sufficient to
demonstrate the need for dynamicness? We return to that question briefly below
(section 6).

* * *

Now that we have a sense of how dynamic semantic explanations are possible, we can
ask what might make explanations in this style desirable. As competing hypotheses
about a certain class of semantic values, we should favor dynamic over static only
if doing so makes for an overall more explanatory theory. A great many motivations
have been offered for dynamic analyses. We will review just three—our main objec-
tive being only to get acquainted with some early and/or well-known dynamic seman-
tic accounts, and to achieve a rough sense of some of the terrain. The motivations
we review stem from: (i) intersentential and donkey anaphora, (ii) the problem of
presupposition projection, and (iii) the semantics of epistemic modals. We proceed
in that order.

3 Intersentential and Donkey Anaphora

3.1 Indefinites and their Discontents

We are familiar with the idea of using the apparatus of quantification to model certain
kinds of anaphoric reference in natural language. In particular we are familiar with the
idea of giving semantics for

(1) A car drove by and it honked.

in a manner which treats the pronoun it as a variable bound by an existential quantifier

introduced by the indefinite a car, so that we may associate the sentence with truth-
conditions along the lines of:

(2) x(x is a car š x drove by š x honked)

What to say, however, when we break (1) into two sentences?

259
 SETH YA L C I N

(3) A car drove by. It honked.

Naively, it would seem desirable to semantically associate this brief discourse with the
same truth-conditions as (1) (cf. Geach [1962]). Yet this would apparently require the
scope of the indefinite in the first sentence to reach across the sentence boundary so as
to bind it in the second sentence—not an option, given ordinary variable binding is a
strictly intrasentential affair.
Should we instead understand the semantic relation between a car and it in (3) on
the model of coreference? That would call for the availability of a nonquantificational,
referential interpretation of indefinite noun phrases. But this idea quickly runs into dif-
ficulties. For instance, it raises the question why discourse:

(4) The man didn’t buy a car. # It was black.

is defective (cf. Karttunen [1976]). If the indefinite and the pronoun were understood
on the model of coreference, we would be able to hear (4) as saying that a specific car,
which the man didn’t buy, was black. But it is not easy to hear it that way.
So two obvious ways to construe the semantic relation between the indefinite and
the pronoun in (3) face clear obstacles. This puzzle has an intrasentential analogue in
the phenomenon of donkey anaphora (Geach [1962]). Consider:

(5) Every farmer who owns a donkey beats it.

It is obvious that the indefinite in this sentence is not a referring expression. But nor can
we straightforwardly understand the anaphoric pronoun it as bound by the indefinite,
for the pronoun is not within the indefinite’s scope.
Moreover, if we wished to render the truth-conditions of the natural reading of (5) in
the language of predicate logic, we would reach for the following:

(6) xy [(farmer(x) š donkey(y) šowns(x, y)) obeats(x, y)]

Somehow, the deeply embedded indefinite in (5) appears to really take wide scope
over the whole sentence—and moreover express universal, rather than existential,
quantification. It is hard to understand how these truth-conditions could be achieved
compositionally.
It was in large part to address these and related puzzles about intersentential and don-
key anaphora that Kamp [1981] and Heim [1982] each proposed their dynamic semantic
systems. Indeed, the semantics of anaphora is arguably the leading historical motivation
for going dynamic (with presupposition projection coming in at a close second). We
will focus on the file change semantics developed in Heim [1982], but we emphasize that
the overlap between Heim’s and Kamp’s accounts is quite substantial.

3.2 File Change Semantics

In its treatment of anaphora, file change semantics weaves together four distinct inno-
vations. First, it employs a novel representation of context, one which can be seen as
enriching the Stalnakerian context sets used in our toy semantics above. Second, it
treats indefinite noun phrases (a cat) as semantically akin to open sentences (x is a cat),

260
 DYN A M I C SEMA N T I C S

and not as introducing any quantificational force of their own. Third, building on Lewis
[1975], it incorporates the idea of unselective quantification, allowing quantifiers to
bind (not just one but) all variables free in their scope. Fourth and finally, file change
semantics embraces dynamic.
Beginning with the first innovation, Heim recommends that we take the information
presupposed in a conversation to have a structure analogous, in important ways, to that
of a filing cabinet. The metaphor goes back at least to Karttunen [1976: 364]:

Consider a device designed to read a text in some natural language, interpret

it, and store the content in some manner, say, for the purpose of being able
to answer questions about it. To accomplish this task, the machine will have
to fulfill at least the following basic requirement. It has to be able to build
a file that consists of records of all the individuals, that is, events, objects,
etc., mentioned in the text and, for each individual, record whatever is said
about it.

Inspired by this metaphor, Heim models the common ground of a discourse as what we
can call a Heim context. Heim contexts are defined in terms of several other notions. I
will now barrage you with definitions, and then explain the intuitive idea.

Def. An assignment world is a pair of a possible world and a variable

assignment.
Def. A satisfaction set is a set of assignment worlds.
Def. A domain is a set of numerals, each corresponding to possible variable
index.
Def. A Heim context c is any pair of a satisfaction set s and a domain d which
is such that if n  d, then the variable xn is free in c.
Def. A variable x is free in c if for any two assignment functions f and f 'that
differ only in their assignment to x, and any world w if (f,w)  c so is (f ',w).

How is a Heim context anything like a filing cabinet? The elements of a Heim context’s
domain are in one–one correspondence with the files of the cabinet. The cardinality
of this domain gives the number of files, telling us how many distinct individuals are
assumed to be tracked in the conversation at a given point. On Heim’s intended appli-
cation, the domain will tell us which individuals are available for anaphoric reference
in the conversation. In the terminology of Karttunen [1976], the domain tracks the
discourse referents of the conversation at a given point. (It will become clearer how the
domain does this below.) The satisfaction set corresponds to the informational content
of the files taken as a whole, and considered independently of their particular file struc-
ture. It imposes properties and relations on the individuals assumed to be tracked in
conversation, as well as conditions on the world at large.
Think of Heim as beginning with the Stalnakerian context sets of our toy dynamic
semantics, and enriching these context sets in two ways. First, each world in the con-
text set is paired with a variable assignment, yielding an assignment world. Assignment
worlds are the sorts of things we can evaluate open sentences with respect to. Second,
we pair this satisfaction set with a domain. As more individuals are made available for

261
 SETH YA L C I N

anaphoric reference in conversation, the domain grows. (New “files” are opened.) As
more information become presupposed in conversation, the satisfaction set shrinks.
To get a sense of Heim’s system, we can begin with its treatment of donkey sentences
such as (5). Ignoring various subtleties, take it the underlying syntax for our donkey
sentence (5) is as follows (see Muskens et al. [2011] for a brief summary of Heim’s syn-
tactic assumptions):

(5) Every farmer who owns a donkey beats it.

every[x1 is a farmer š x1 owns x2 š x2 is a donkey][x1 beats x2]

Here we observe the second innovation mentioned above: the indefinite noun phrase
a donkey introduces no quantification on its own. It merely contributes what is in effect
an open sentence. (Indeed, Heim takes all noun phrases to express open sentences. We
return to that idea in connection with definiteness below.)
Now to the semantics. Heim’s dynamic semantic values for sentences are functions
on Heim contexts. These semantic values must update both the domain of the context
(which they will do either by leaving it alone, or expanding it) and the satisfaction set
of the context (which they will do either by leaving it alone, or reducing it). It is helpful
to start out by ignoring domain dynamics entirely. This will help us see where exactly
domains are needed.
Focus first, then, on the way in which the meaning of a sentence changes the satisfac-
tion set. Our donkey sentence is built from atomic (open) sentences, conjunction, and
universal quantification, so our objective is to give compositional dynamic semantic
values for these linguistic expressions. We take it that, as in an ordinary static inten-
n
sional semantics, to each n-place predicate F of the language there corresponds an
intension F, taking n individuals and a possible world into a truth value. Then, if sc is the
satisfaction set of c, we can recursively state the dynamical impact of atomic sentences
and conjunctions on satisfaction sets as follows:

sc[F(xi, . . ., xn)] = {(f, w) c : F(f (xi) . . . f (xn), w) = 1}

sc[DšE] = sc[D][E]

The update rule for atomic open sentences is simple: It takes a satisfaction set and
eliminates from it all those assignment worlds not satisfying the open formula. The
update rule for conjunction is also straightforward, and indeed just the same as in our
toy semantics above: It amounts to consecutive update on the input satisfaction set.
Turning to universally quantified sentences, a dynamic rule sufficient to handle (5)
is the following:

sc[every(D, E)]= {(f, w)  c : for every f’ ~ f:

if (f, w)  sc[D], then (f, w) sc[D] [E]

In effect, a quantified statement takes a satisfaction set and eliminates those assign-
ment worlds which (i) would survive update with the restrictor alone but (ii) would not
survive update with the restrictor clause followed by the matrix clause. Observe that
this semantics incorporates the idea of unselective quantification: the object language
quantifier every doesn’t bind just one distinguished variable, but rather binds all vari-
ables free in its scope. This indiscriminate binding enables the single quantifier every to

262
 DYN A M I C SEMA N T I C S

mimic the truth-conditions achieved with the two selective universal quantifiers in (6),
yielding the intuitively correct interpretation for (5). Thus if our initial Heim context
is c, (5) asks us to eliminate those assignment worlds from c which satisfy

x1 is a farmer š x1 owns x2 šx2 is a donkey

but which fail to satisfy

x1 beats x2

Contexts which accept (5), then, are ones that rule out the possibility of ever including
a file on a farmer who is a donkey-owner but not a donkey-beater. Or again, the worlds
left open by the satisfaction sets of such contexts are worlds whose donkey-owning
farmers are all donkey-beaters.
This buys us an account of an important class of donkey sentences. But to pay for it,
we had to give up the idea that indefinites introduce their own quantification. So what
do we say of the undeniable existential readings of indefinites in rather more ordinary
constructions—for instance in our (1) above (repeated below)?

(1) A car drove by and it honked.

Under Heim’s syntactic assumptions, (1) is, at the relevant level of abstraction, equiva-
lent to:

x1 is a car šx1 drove by š x1 honked

We naturally read (1) as an existential claim, however. Why, if it is really an open

sentence?
Heim accounts for the existentially closed reading of indefinites in the context
of defining truth. In characteristically dynamic fashion, she defines truth in the first
instance for contexts, and then derivatively for sentences. We only need her definition
of truth for contexts to explain the intuition that sentences such as (1) are existentially
closed.

Def. A Heim context c is true iff there exists, with respect to the actual world
@, some assignment f of values to variables—equivalently, some sequence of
individuals f in @—such that (f, @)  sc.

This definition tells us how to recover the truth-conditional content of a Heim context
from its satisfaction set. In effect, the satisfaction set represents a complex intensional
relation, and the definition of truth says that a satisfaction set is true just in case this
relation is actually instantiated by some objects. The existential reading of sentences
such as (1) can then be explained by the fact that if this sentence is accepted in a con-
text c, c can be true only if there exists a car that drove by and honked. This definition of
truth in effect enables a form of wide scope existential closure at the end of interpreta-
tion for the variables left free in sentences.
From here, it is easy to handle our case of intersentential anaphora:

263
 SETH YA L C I N

(3) A car drove by. It honked.

On Heim’s analysis, this discourse has exactly the same context change potential as (1).
For recall that on the dynamic account, conjunction merely amounts to consecutive
update. When a conjunction is unembedded, that is equivalent to consecutive asser-
tion. Thus (3) is explained on the same model as (1), with the existential reading again
accounted for by the definition of truth for contexts.
We have, then, a single semantics predicting ordinary existential readings of indef-
inites, donkey anaphoric readings, and intersentential anaphoric readings—a highly
nontrivial achievement.
But we are not yet out of the woods. Although Heim’s definition of truth enables us to
supply the existential readings in cases such as (1) and (3), it cannot supply existential
readings where the indefinite is itself scoped under other operators. Negation provides a
simple example of such a case:

(7) There isn’t a cloud in the sky.

Clearly, we do not want (7) to be equivalent to something like:

™(x1 is a cloud š x1is in the sky)

For when we apply the definition of truth to get the existential reading of the indefinite
the sentence, the resulting interpretation would require that there exists something
which is not a cloud in the sky. But that is wrong. On the natural reading, (7) does not
say, of some thing, that it isn’t a cloud in the sky; rather it simply says that the sky is
cloud-free. What we should like to happen with (7) is that the sentence be existentially
closed before negation is applied:

™x1 (x1 is a cloud š x1 is in the sky)

Heim manages this by allowing negation itself to existentially close free variables not
in already in the domain. This is where domains enter into the story. The central
semantic role of domains is to enable the existential closure of variables in embedded
contexts.
To say how this works, we now must explain how sentences dynamically change
the domain of a Heim context. If c is a Heim context and dc its domain, we can
state the dynamical impact of sentences on domains recursively via the following
schemas:

dc[F (x1, . . ., xn)] = {dci, . . ., n}

d[(DšE] = d[D] d[E]
dc[¬D] = dc
dc[every (D, E)] = dc

Open sentences will expand the domain whenever they include variables which are free
in c; otherwise, they will leave the domain alone. A conjunction expands the domain
whenever one of its conjuncts does. Negated and universally quantified statements such
as (5) will never expand the domain.

264
 DYN A M I C SEMA N T I C S

Observe these rules track intuition about when anaphora is licensed for subsequent
discourse. Atomic open sentences and conjunctions containing indefinites both
generally license subsequent anaphora:

(8) A cloud is in the sky. It is fluffy.

(9) It is breezy and a cloud is in the sky. It is fluffy.

These kinds of sentences have the potential to “open new files” or “make new dis-
course referents available.” By contrast, indefinites under negation and within quanti-
fier restrictors generally do not:

(4) The man didn’t buy a car. # It was black.

(10) Every farmer who owns a donkey beats it. # It is not happy.

Although negated sentences never expand the domain of context, their semantics—in
particular, their impact on satisfaction sets—makes essential reference to the context’s
domain. On Heim’s account, the dynamical impact of negation on satisfaction sets is
the following:

sc[™D]
{(f, w) c : there is no (f ', w) : f ' adc f such that (f ', w)  sc[D]

where

Def. f ' adc f iff for all n d, f(xn) = f '(xn).

A negated sentence changes the satisfaction set by eliminating all those assignment
worlds which could verify the open sentence D. The procedure for doing this effectively
existentially closes the free variables in D which are not in the domain before the nega-
tion. This gets the correct result for (7) above.
Once we add domains, we needn’t change the update rules we have defined on sat-
isfaction sets for atomic open sentences or for conjunction. But we do have to change
our semantics for universal quantification. As the semantics is now stated, it will get
incorrect results for sentences such as:

(11) Every farmer who own a donkey doesn’t beat it.

every [x1 is a farmer š x1 owns x2 š x2 is a donkey] [™(x1 beats x2)]
(12) Every farmer who owns a donkey sells it to a merchant. (Kamp [1981])
every [x1 is a farmer š x1 owns x2 š x2 is a donkey] [x1 sells x2 to x3 š x3 is a
merchant]

The sentence (11) illustrates a case where negation should not existentially close the
variables below it, while (12) illustrates the fact that the universal quantifier should
not bind variables introduced by indefinites which appear in their nuclear scope. The
adjustment to our semantics required to accommodate such examples involves a con-
siderable complication of our update rule for universal quantification. What we need
is for the domain to be locally and “temporarily” updated as interpretation proceeds, so
that the right variables are bound in the right way. We leave it to the reader to verify
whether the following entry will do the job (Heim [1982, 1983a]):

265
 SETH YA L C I N

sc[every(D, E)D)]= {(f,w)  c : for every (f ',w) such that f adc f ' and (f ',w) 
sc[D], there is some (f '',w) such that f ' adc[D] f '' and (f '',w)  sc[D][E]}

We will get a bit further into Heim’s account when we come to the issue of presupposi-
tion projection in the next section. Let me close this section with two questions that
Heim’s dynamic semantics, as so far presented, raises.

3.3 On the File Metaphor

First, there is the question of what exactly to make of Heim’s model of the informational
context. In Heim [1982], she suggests that her representation of context can been seen
as a linguistically motivated extension of Stalnaker’s model of the common ground of
a conversation in terms of a set of possible worlds—his context sets. But on Stalnaker’s
picture, a context set models the content presupposed by a given set of interlocutors. In
what sense can a Heimian context—a pair of a satisfaction set and a domain—be the
content of an attitude like presupposition or belief? Insofar as possible, it would be desir-
able to independently moor this conception of informational content. (Stalnaker’s par-
ticular model of content is motivated by a broadly decision-theoretic style of functional-
ism together with a certain externalist treatment of intentionality. (see e.g. Stalnaker
[1984]). Of course, one may not accept Stalnaker’s particular model of informational
content, but the point is that one has a grip on Stalnaker’s conception of information
quite independently of its application to formal semantics or to the representation of
communication. Not so for Heim’s model, unless more is said.)
Reflection on this issue may push us closer to an interpretation of Heim’s model which
brings it closer to Kamp’s—one on which the role of an intermediate representation really
is, after all, essential. On this reading, Heim contexts would not be contents (as context
sets are for Stalnaker) but rather representations having content. Heim seems to resist this
interpretation (see, e.g., Heim [1990b]), but it suits her preferred metaphor: she calls her
contexts files, and ordinary files are, after all, not contents but bearers of content.
On the other hand, there have been efforts to construe at least the satisfaction set
component of Heim contexts as contents. For work construable in this direction, see
Ninan [2008, 2010], Cumming [2008].

3.4 On the Need for Dynamics

Second, we noted that Heim’s semantics weaves together four distinct ideas. It is clear
enough how the first three ideas cooperate: The novel representation of context (first
idea) allows for the context to be updated by open sentences, and hence by indefinite
noun phrases (second idea). These assumptions in turn facilitate the treatment of don-
key anaphora via unselective quantification (third idea). But we can ask: in what way is
dynamic—the fourth idea—a key ingredient in the story?
Here it is helpful to separate two questions: one about the conceptual necessity of
the dynamic formulation, and another about its comparative elegance and explanatory
power.

I. Can the work done by file change semantics—specifically, its treatment of donkey
and intersentential anaphora—in principle be done within a static semantics, one
defined on exactly the same syntactic structures, and without ad hoc stipulations?

266
 DYN A M I C SEMA N T I C S

II. Does the balance of evidence support the view that a dynamic semantics for donkey
and intersentential anaphora is comparatively more elegant and explanatory than
any static alternative?

The first question is not directly addressed by Heim, and to our knowledge has not
been definitively settled. (Although Heim [1982] itself contains a static semantics
for anaphora covering basically the same data as its file change semantics, the static
account there requires a more complex syntax, and a number of additional stipula-
tions.) However, it is clear that which assignment worlds are eliminated by any given
sentence will depend crucially on whether the variables in that sentence are tracked
by the domain of the relevant file. As a result, no simple static association between
sentences and conditions on assignment worlds will mimic the updates supplied by
file change semantics.
(Might one sidestep this issue by dispensing with domains? Could we just leave out the
complexity introduced by domains, and refer instead everywhere to the set of variables
not free in the relevant satisfaction set? Here the question is whether “free” should be
taken to entail “not-in-the-domain”. Heim thinks not: see [Heim, 1982, 303]. Whether
or not Heim is right, it is clear that such an adjustment would not make it any easier to
associate sentences with fixed conditions on assignment worlds.)
The second question has been, and continues to be, a matter of lively debate within
natural language semantics, and remains open. Perhaps the leading static alternative
to dynamic analyses of donkey and intersentential anaphora is the definite description
(‘d-type’) approach to pronouns, which takes anaphoric pronouns to be semantically
equivalent to certain definite descriptions (see Parsons [1978], Cooper [1979], Heim
[1990b], Elbourne [2005], and references cited therein; cf. Evans [1980], Stalnaker
[1998]). Another important static semantics is the variable-free approach developed by
Jacobson [1999]. For overviews of recent work on pronominal anaphora together with
some cost-benefit analysis, see Elbourne [2008].
Heim herself, in Heim [1982] and Heim [1983b], famously suggested that one reason
for favoring the dynamic character of her file change semantics was that it allowed for a
compelling approach to the projection problem for presuppositions.

4 The Problem of Presupposition Projection

4.1 From Anaphora to Presupposition

We said it would be nice to be able to read (3) along the lines of (2):

(3) A car drove by. It honked.

(2) x(x is a car š x drove by š x honked)

Heim’s semantics enables us to do that. Now suppose we extended this discourse as

follows:

(13) A car drove by. It honked. The car was black.

Intuitively it would seem desirable to analyze (13) along the lines of (14):

267
 SETH YA L C I N

(14) x(x is a car š x drove by š x honked š x was black)

That would require treating definite noun phrases as, like indefinite NPs, open
sentences.
Suppose we did treat them that way. Very well; but what then is the semantic differ-
ence between definites and indefinites? The obviously do differ, notably in respect of
the conditions under which they license anaphora. For instance, the car, unlike a car,
can be read as anaphoric with a previous indefinite ((15)–(16)), and can license subse-
quent pronominal anaphora even when under negation ((17)–(18)):

(15) [A car]i drove by. [The car]i was moving very fast.
(16) [A car]i drove by. # [A car]i was moving very fast.
(17) Bill doesn’t have [the car]i. [It]i is in my possession.
(18) Billy doesn’t have [a car]i. # [It]i is in my possession.

If, following the lead of many generative grammarians, we take it DPs/NPs in logical
form are generally marked rwith a definiteness feature, the question at issue becomes:
What, semantically, does the definiteness feature contribute?
One way to take this question is: what item of truth-conditional content does the defi-
niteness feature contribute? But examples like (15)–(18) suggest that definiteness features
are less content-bearing items than discourse management tools—specifically, devices for
coordinating speaker on referents. Taking this attitude toward the data and building on
earlier work (notably Karttunen [1968]), Heim enriches her system so as to capture the
idea that definites, unlike indefinites, are used to refer to something that is already familiar
at the current state of the conversation. In contrast to indefinites, which generally serve
to introduce new discourse referents, definites pick up on discourse referents already in
play. She does this by defining the notion of appropriateness for atomic sentences:

Def. If c is a Heim context and I an atomic sentence, then I is appropriate with

respect to c iff for every noun phrase NPi contained in I:
(i) if NPi is definite, then NPidc.
(ii) if NPi is indefinite, then NPidc.

and by declaring that the context change potential of a sentence is undefined on con-
texts with respect to which it is inappropriate:

novelty-familiarity condition. If I is not appropriate with respect to c, the

context change potential of I is undefined on c.

Let us walk through how this helps with, e.g., (16). The second sentence of discourse
(16) contains an indefinite coindexed with an earlier indefinite. By the semantics above,
we know that its index must already be in the domain by the time we attempt to update
the context with the sentence. So it is inappropriate. By the novelty-familiarity condi-
tion, its update is undefined on contexts which accept the first sentence. That predicts
infelicity for the discourse, the correct result. By analogous means we can explain the
felicity of (15) and (17), and the infelicity of (18). Observe the novelty-familiarity con-
dition charges the domain feature of context with additional semantic work.

268
 DYN A M I C SEMA N T I C S

(Of course, we needn’t view the novelty-familiarity condition as a special seman-

tic stipulation outside of the compositional semantics proper. We may incorporate the
appropriateness requirement directly into the dynamic semantics given above.)
One gloss on the preceding is this: by being defined on some contexts but not oth-
ers, (in)definite noun phrases take something for granted about the preceding discourse–
specifically, something about the domain of any context to which they are applied.
When what they take for granted is wrong, they are inappropriate and (hence) fail
to update the context. This glosses raises the question: if sentences or NPs can take
something for granted about the domain of the contexts they are applied to, can they
not also take something for granted about the satisfaction sets of the context they are
applied to?
Yes, says Heim [1983b]: in fact, that is just the phenomenon usually called presupposi-
tion. First, define the notion of admittance in terms of the definedness of context change
potential, as follows:

Def. c admits I just in case the context change potential of I is defined on c.

Then define presupposition as a relation between sentences in terms of admittance and

acceptance, as follows:

Def. I presupposes \ just in case every context which admits I accepts \.

Heim conjectured that this definition of presupposition could enable an advance on

the projection problem for presuppositions: roughly, the problem of predicting the presup-
positions of complex sentences in systematic fashion from the presuppositions of their
parts. Presuppositions would be generated wherever we have the relevant undefinedness
of context change potential; they would project whenever the undefinedness of a part
was inherited by the whole.
We can illustrate this approach to presupposition projection with three examples,
the latter two from Heim [1983b]: (i) conjunction, (ii) conditionals, and (iii) univer-
sally quantified sentences.

Conjunctions. The general contours of this account go back to Stalnaker [1974].

Consider:
(19) (a) The king has a son.
(b) The king’s son is bald.

Both sentences presuppose that there is a king, and (19b) moreover presupposes (19a).
Now the following conjunction embeds both sentences:

(20) The king has a son and the king’s son is bald.

but it presupposes only that there is a king. We should like to predict this fact, given
only the meaning of and and the presuppositions of the primitive parts of the sentence.
And we can, with the dynamic semantics for conjunction sketched already in our toy
semantics above. On this view, conjunction is merely consecutive update. Although
the context change potential of the second conjunct is undefined on context sets which

269
 SETH YA L C I N

don’t include the information that the king has a son, the same will not be true for the
context change potential of the whole conjunction. The reason is that when the first
conjunct updates the input context, it yields an intermediate (‘local’) context which
always admits the second conjunct, and which serves as the input to the context change
potential of the second conjunct. In the usual terminology, the second conjunct’s pre-
supposition is locally satisfied by the first conjunct, preventing the undefinedness from
percolating up through the composition, and hence blocking the presupposition from
projecting. By contrast, the presupposition of the first sentence (that there is a king)
does project, for it is never locally satisfied.

Conditionals. The story is similar with conditionals. Consider:

(21) If the king has a son, the king’s son is bald.

This sentence too only presupposes that there is a king, despite the fact that its consequent
presupposes additionally that the king has a son. We should like to predict this fact, given
only the meaning of if and the presuppositions of the primitive parts of the sentence.
We can, with with the dynamic semantics for the indicative conditional offered by
Heim:

sc[Io\]= sc − (sc[I] − sc[I][\]

To update the context with a conditional I o \, we eliminate those possibilities

(assignment worlds) which would survive update with I but not with I followed by
\. According to this rule, a conditional is well-defined on sc only when both sc[I] and
sc[I][\] are. Thus since (19), the antecedent of (21), is only admissible on contexts that
include the information that there is a king, the whole conditional (21) presupposes
that there is a king. By contrast, nothing prevents the conditional from being admis-
sible on contexts that fail to accept that the king has a son. Although the consequent
(20) does presuppose this, and hence is only welldefined on contexts accepting it, the
antecedent serves to locally satisfy this requirement, and thereby prevents the whole
conditional from inheriting this presupposition.
Thus the presuppositions of (21) fall out from the context change potential of the
conditional together with the presuppositions of its parts. Generally speaking, whether
a presupposition projects in a particular setting is a matter of whether it is locally
satisfied. If the presupposition of an expression is not guaranteed to be satisfied by its
local context—by the context it is semantically operating on—the presupposition will
project.

Universally quantified sentences. Consider the following sentence, together with

its Heimian syntactic parse:
(22) Every nation cherishes its king.
every [x1 is a nation] [x1 cherishes x1’s king]

This sentence presupposes

(23) Every nation has a king.

270
 DYN A M I C SEMA N T I C S

Considerations of generality make it very natural to want to trace this presupposition to

‘x1’s king’—specifically, to its presupposing:

(24) x1 has a king.

Of course, that would require making sense of the idea of an open sentence being a pre-
supposition. This is a nontrivial challenge for accounts associating presuppositions only
with proposition-expressing elements (e.g., Stalnaker [1974], Gazdar [1979]), but as
should be obvious, it presents no difficulty for Heim: Her account allows open sentences
to update contexts, and thus enables the definition of presupposition given above to
apply generally. Applied together with her semantics, it yields the prediction that (22)
does indeed presuppose (23), as the reader may confirm. The reader may also wish to
confirm that her semantics correctly predicts that this presupposition of (22) can be
cancelled with an appropriate addition to the restrictor clause:

(25) Every nation with a king cherishes its king.

every [x1 is a nation šx1 has a king] [x1 cherishes x1 ’s king]

See Heim [1983a], Beaver [2001] for further discussion and additional difficulties.
The preceding is intended to give just a flavor of the kind treatment of presupposi-
tion projection possible within a dynamic setting. For a much fuller picture, both of the
historical background and of the complex array data modern theories of presupposition
projection must be tested on, Beaver [2001] is an especially useful source.

4.2 The Question of Explanatory Power

It might be thought that any linguistic connective can be associated with three items:
(1) a truth-conditional contribution (a content property); (2) a specification the presup-
positions it generates, if such there be (a presupposition property); and (3) a specifica-
tion of its permeability for the presuppositions of its arguments (a heritage property).
The heritage property of the indicative conditional, for instance, might specify that an
indicative conditional Io\ (a) presupposes whatever I does, and (b) presupposes I
o\', where \' is what \' presupposes.
Karttunen and Peters [1979] presented a framework in which these three properties
were treated as mutually independent and specified separately for each connective. But
it is natural to conjecture that the heritage property of an expression can somehow be
derived from a specification of its content and presupposition properties. Else we have
a theory which

. . . implies—implausibly—that someone who learns the word “if” has to learn

not only which truth function it denotes and that it contributes no presupposi-
tion, but moreover that it has the heritage property specified . . . It also implies
that there could well be a lexical item—presumably not attested yet—whose
content and presupposition properties are identical to those of “if”, while its
heritage property is different. Heim [1983b]

The worry isn’t that such an expression is logically impossible. There is no contradic-
tion in the idea of such a lexical item. Rather, the worry is that an apparently robust

271
 SETH YA L C I N

empirical generalization about natural languages—roughly, that heritage properties

supervene on content-and-primitive-presupposition properties—is being stipulated
rather than predicted.
Gazdar [1979] argued that Karttunen and Peters’s theory was subject to this kind of
objection, and Heim [1983b] agreed. Her account was intended to satisfy this demand
for explanatory adequacy while being superior to Gazdar’s account in its empirical
coverage. Heim thought she had an explanatory advantage over Karttunen and Peters
partly because, given only the context change potential for a connective, one could
read off both the content and heritage properties.
But as subsequently noted by Rooth (p.c. in 1987 to Heim, quoted in Heim [1990a])
and Soames [1989], differing specifications of the context change potentials of the con-
nectives Heim discusses—differing, in particular, in the heritage conditions they would
impose—could in principle generate the same predictions vis-à-vis the informational
contribution of the expression to context. To illustrate, contrast the standard dynamic
semantics for conjunction (C1) (repeated from our toy semantics) with the deviant
entry (C2), where the conjuncts update the context in reverse order:

(C1) c[I š \= c[I][\]

(C2) c[I š \ = c[\][I]

In the context of Heim’s semantics, these will have the same update impact were both
defined; they differ only in their heritage conditions. So just as Karttunen and Peters need
to stipulate that one among a range of possible heritage properties is the correct one for
(say) the conditional, Heim is compelled to stipulate that one among a range of possible
context change potentials, all equivalent with respect to their informational impact on
context, is the one with the correct heritage condition. Heim conceded this worry, and
agreed that the explanatory power of her account had been overstated (Heim [1990a]).
To be clear: it would be incorrect to say, concerning the limited array of projection
facts Heim’s account was directed at, that the account is completely stipulative, hence
not explanatory or predictive in character. On the contrary: on the basis of a finite set
of stipulations, Heim’s account generates predictions about the presuppositions of an
infinite class of sentences. Rather, the point is simply that (i) a theory which got by with
less—specifically, which built-in fewer assumptions about heritage conditions—would
be more explanatory; and (ii) there is good reason to think that an alternative theory,
better in specifically this respect, can be had.
This is a worry for an account along Heim’s lines, insofar as it shows the account
does less than we might have wanted. Is it an objection to the account? Not really—not,
anyway, until some relevantly more explanatory alternative theory is actually put on
the table.
Recent work by Philippe Schlenker has brought renewed attention to this worry
about dynamic accounts of presupposition projection, and constitutes an important
attempt to deliver the explanatory goods lacking in Heim’s account. See Schlenker
[2008, 2009], and the critical replies to Schlenker by Beaver [2008], Fox [2008], and
Rothschild [2008b].
Another important attempt to fill the explanatory gap in dynamic accounts of pre-
supposition projection, one not discussed by Schlenker, occurs in Beaver [2001]. He
writes:

272
 DYN A M I C SEMA N T I C S

I will borrow from Veltman’s work to show how the context sensitivity of [epis-
temic modal] words like ‘might’ and ‘must’ motivates a dynamic semantics.
None of the alternative CCPs for connectives that have been suggested by
Rooth and Soames would be compatible with this semantics, and it is hard to
imagine how a relevantly different dynamic semantics could still get the facts
right about the meanings of the epistemic modalities.
(146)

Beaver is correct: the dynamic semantics for epistemic modals he considers is not com-
patible with the deviant CCPs offered by Rooth and Soames. Its addition to Heim’s
system would thus appear to increase the system’s overall explanatory power vis-à-vis
the problem of presupposition projection. So let us have a look at this semantics and its
motivation.

5 Epistemic Modals
The dynamic semantics Beaver has in mind is a refinement of the update semantics for
epistemic modals developed by Veltman [1996] (a predecessor of which occurs in Velt-
man [1985]; cf. also Stalnaker [1970], Groenendijk and Stokhof [1975], Groenendijk et
al. [1995]). To illustrate the basic idea of the semantics, we can extend our toy dynamic
semantics from section 2.3 with an epistemic possibility operator  (for epistemic might,
may, possibly, etc.), adding the following clause:

c[I] = {w  c : c[I] z}

(Epistemic must can then be introduced as the dual of .) Given I is not itself modal-
ized, this captures the thought that I is accepted with respect to a context just in case,
roughly, there is a I-world left open by (in) the context. Applied to c, the function [I]
returns either c or the empty set. Veltman uses the metaphor of a test: “Sentences of
the form might I provide an invitation to perform a test on c rather than to incorporate
some new information in it” ([Veltman, 1996, 10]. Sketching roughly the same idea in
earlier work, Stalnaker writes: “A sentence of the form ‘It may be that P’ . . . may be
interpreted as making explicit that the negation of P is not presupposed in the context”
([Stalnaker, 1970, 286–7]).
On this account, I-sentences never, as a matter of their semantics, add information
to the context when tokened unembedded (where “adding information” corresponds
to winnowing down the possibilities, without excluding all possibilities). As a result,
even as one strictly increases the information taken for granted by the context, the
context can go from accepting I to failing to accept it. In the jargon, I is not
persistent:

Def. I is persistent just in case if c accepts I, then I is accepted in all the subsets
of c.

Before we added , our toy dynamic semantics contained only persistent sentences.
It should be evident that this semantics leads to a failure of distributivity. Whether a
set of worlds as a whole contains a I-world is a ‘global’ property of that set. Thus if one
wants to know whether some ™I-world w left open by a context c will survive an update

273
 SETH YA L C I N

with I, the answer will not be settled by some property of w; rather, one needs to know
something about what other worlds there are in c.
All this makes it especially confused to speak of the “content” or “proposition
expressed” by I in a dynamic setting. Such sentences do not semantically serve to
represent the world as being a certain way, or serve to add a proposition to the common
ground; their discourse impact is just different from that of straight factual sentences.
Nor is it helpful to say that  corresponds to a special “force” on this account. Speech
act forces, whatever those are, don’t embed, at least on the usual way of understanding
them; but epistemic modals are freely embeddable on the dynamic account.
The point that epistemic modal claims don’t determine truth-conditions—that they
aren’t the kind of clauses that determine conditions on points, the sets of which con-
stitute contexts—calls into question the assumption, not uncommon in discussions of
presupposition projection, that we can always factor out a truth-conditional contribu-
tion for any dynamic connective.
What is the empirical motivation for this dynamic semantics? Perhaps the most fre-
quently cited data point in its favor concerns an alleged asymmetry in acceptability
between the following two sorts of discourse:

(26) Billy might be at the door (and) . . . it isn’t Billy at the door. (Iš™I)
(27) ? It isn’t Billy at the door (and) . . . Billy might be at the door. (™I š
I)

—Where we are to imagine that what happens between the dots is that the speaker
sees who is at the door. Pairs like this one were first discussed by Veltman, and are
cited approvingly by Groenendijk et al. [1995], Veltman [1996], Beaver [2001], Gillies
[2001], and von Fintel and Gillies [2007], among others. The dynamic account predicts
the supposed difference between these discourses. The update corresponding to (27) is
guaranteed by dynamic lights to “crash” the context, since for all c and I, c[™IšI]
= c[™I][I], which in turn will always be . Once the context comes to accept ™I, it
will fail the test corresponding to I. But not so for (26), the other ordering of the
conjuncts: c[I][™I] will be nonempty as long as c contains some I-worlds and some
™I-worlds. The thought, then, is that pairs like (26)–(27) demonstrate the need for
both a dynamic semantics for epistemic possibility modals and a dynamic semantics for
conjunction, for these together generate the order-sensitivity (noncommutivity) mani-
fested by such pairs.
If correct, this semantics would, as Beaver suggests, help to constrain the space of pos-
sible context change potentials for connectives vis-à-vis the problem of presupposition
projection. For example, on this semantics it is no longer the case that the ordinary con-
junction (C1) and deviant conjunction (C2) have the same update impact where both
defined. On the contrary, they, would come apart exactly on pairs such as (26)–(27). If
ordinary conjunction were defined as in (C2), then (27) would be predicted to be felici-
tous and (26) infelicitous—the incorrect result, says the fan of dynamics for epistemic
modals.
But, stepping back, what is the right thing to say about pairs such as (26)–(27)?
First, it should be duly noted that sentences of the form (Iš™I and (™IšI are
virtually always both marked, and equally so. Generally they sound just as bad as (e.g.)
contradictions. Second: even if, with sufficient rigging of context, (26) can be made
to sound okay, this alone tells us nothing; for with sufficient rigging, (27) will sound

274
 DYN A M I C SEMA N T I C S

okay, as will almost anything else. What then is the point? The point is supposed to be
that pairs such as (26)–(27) manifest an asymmetry in markedness in a certain kind of
situation: namely, one in which monotonic information growth happens in context mid-
sentence. The thought is that if we restrict to such cases, we will see that (26), but not
(27), will sound fine. The is supposed to highlight the need for a nonpersistent update
semantics for I with noncommutative conjunction.
Alas, it does not. First, the data do not support noncommutivity for conjunction. The
sentence (26) is marginal at best with and, even in the scenario described. One senses
some awareness of this fact in the literature, for among the works cited above, only one
(namely Beaver [2001]) actually explicitly includes and in the relevant pairs. Beaver is
of course correct to include it—to draw conclusions about the semantics of an expres-
sion, one presumably should appeal to constructions in which the expression actually
occurs—but judgments do not support non-commutativity.
Moreover, if things are as the dynamic story is telling us, one wonders why there is no
substantial asymmetry in defectiveness between

(28) ?? Suppose Billy might be at the door and he isn’t.

(29) ?? Suppose Billy isn’t at the door and he might be.

According to the dynamic account as so far stated, (29) is predicted to be defective, but
not (28). But the two are both defective, and intuitively, they are defective in the same
way. Perhaps there is some further story the fan of dynamics will tell here, but until
then, we have no data supporting non-commutative conjunction. If anything, we have
data favoring commutativity.
Second, the way that the notion of monotonic information growth is used in setting
up the data is questionable. To see why, consider a view according to which the transi-
tion from believing I to believing ™I normally involves nonmonotonic information
growth. Egan [2007] is perhaps one such example (see also Stephenson [2007]): on his
view, roughly, to believe that Billy might be at the door is to “locate oneself”, in the
sense of Lewis [1979], on a person whose evidence leaves the possibility of Billy’s being
at the door open, where the possibilities left open by one’s evidence are centered worlds.
In the background scenario stipulated for (26)–(27), the speaker’s belief state would, on
Egan’s account, normally nonmonotonically change in respect of the speaker’s views
about his evidence: his belief state would go from excluding centered worlds where
the center’s evidence entails Billy is not at the door to including such worlds. Now we
can ask: Is there reason for Egan to be phased by anything about (26)–(27)? It seems
not: he can simply deny that there really is strictly monotonic information growth in
the (26)-case, and just explain that insofar as the sentences are felicitous, it is because
they express propositions rationally believed at the relevant times of utterance. Could
it be objected against Egan that it is intuitively obvious that the (26)-case involves only
monotonic information growth? No, for it makes little sense to appeal to brute intui-
tions about the technical notion of monotonic information growth.
We conclude pairs such as (26)–(27) provide no evidence for an especially dynamic
account of epistemic possibility modals, or of conjunction.
But what of pairs such as (28)–(29)? Although they do not support a non-
commutative semantics for conjunction, they do support the idea that there is some
semantically significant kind of incompatibility between I and ™I. The defective-
ness of examples such as (28) and (29) is in fact hard to predict on the leading static

275
 SETH YA L C I N

semantics for epistemic modals, namely that of Kratzer [1991]; similiarly with the
accounts of Egan [2007] and Stephenson [2007]. On these accounts I and ™I are
truth-conditionally compatible, and so one expects it to be possible to suppose them
jointly true. But on a dynamic conception the sentences are incompatible, inasmuch
as they are not jointly acceptable by nonempty contexts; as a result, on usual dynamic
ways of defining consequence (such as the definition considered in 2.2 above), it will
come out that ™I _= ™I. Unsurprisingly then, when the dynamic account is suitably
upgraded with semantics for attitude verbs (e.g., Heim [1992]; see also Asher [1987]),
it gets the right results for (28) and (29), and for related constructions. And this is a
indeed a real advantage over its static rivals.
Though not all of them. The static account of Yalcin [2007] can accommodate,
indeed is motivated by, these data (see also Kolodny and MacFarlane [2010], MacFarlane
[2011]). The account of that paper does not assume dynamic, but it does define semantic
notions of acceptance and of consequence entirely analogous to those usual in a dynamic
semantics, and it gets the correct results for sentences like (28) and (29). The abstract
idea behind Veltman’s semantics, that the transition from believing or presupposing I
to ceasing to believe or presuppose I is generally monotonic, certainly has empirical
motivation. (Though see Willer [2010a], Yalcin [2011a] for some further problems and
extensions.) But it can be realized in a static framework.

6 Essentially Dynamic?
Because of the way in which Veltman’s account of epistemic modals introduces vio-
lations of distributivity and persistence, it is sometimes suggested that his modals
are “essentially dynamic operators” (Muskens et al. [2011]; cf. von Fintel and Gil-
lies [2007]). But inlight of our recent observations, this assessment seems somewhat
hasty. It is true, modulo a certain kind of radical context-sensitivity, that one can-
not resolve Veltman’s proposed context change potentials into static conditions on
points—truth-conditions, in the usual sense—together with a simple intersective rule
of context update. But one can effectively resolve it into a static condition on bodies
of information together with a single rule mapping such static conditions and prior
contexts into posterior contexts. And if one can do this, why regard the underlying
idea as essentially dynamic? (Of course, a dynamic formulation might prove to be
more elegant in various respects that its static counterpart, hence to be preferred for
that reason; but that is another story.) For further discussion of this subtle issue, see
Rothschild and Yalcin [unpublished].

7 A Few More References

For more on presupposition and anaphora see, in addition to the citations already given,
Van der Sandt [1992], Geurts [1999], Breheny [2001], Rothschild [2008a], Kripke
[2009]. For a start on dynamics for temporal anaphora, see Ter Meulen [1997], [Steed-
man, 1997, sec. 3.2.3], Stone [1997], Stone and Hardt [1999]. For a start on dynamics
for modal anaphora, see Roberts [1989, 1996], Frank and Kamp [1997], Stone [1997],
Brasoveanu [2007].
Conditionals form an active area of dynamic semantic research. On counterfactuals,
see von Fintel [2001], Gillies [2007], Moss [2010]. See Gillies [2004] for an account of
indicative conditionals in the spirit of Veltman’s update semantics for epistemic modals.

276
 DYN A M I C SEMA N T I C S

See also Willer (2011b) See Yakin for a dynamic account of indicative conditionals
incorporating probabilities.
Interrogatives and focus have formed another active area of dynamic semantic inquiry.
See Ginzburg [1996], Erteschik-Shir [1997], Hulstijn [1997], Groenendijk [1999], Aloni
and Van Rooy [2002], and the papers in Aloni et al. [2007].

References
Maria Aloni and Robert Van Rooy. The dynamics of questions and focus. In Proceedings of SALT, volume
12, pages 20–39. Citeseer, 2002.
Maria Aloni, Alastair Butler, and Paul Dekker, editors. Questions in Dynamic Semantics, volume 17 of Current
Research in the Semantics/Pragmatics Interface. Emerald Group Publishing, 2007.
Nicholas Asher. A typology for attitude verbs and their anaphoric properties. Linguistics and Philosophy,
10(2): 125–197, 1987.
David Beaver. Presupposition and Assertion in Dynamic Semantics. Studies in Logic, Language and Informa-
tion. CSLI Publications, 2001.
David Beaver. As brief as possible (but no briefer). Theoretical Linguistics, 34(3): 213–228, 2008.
Adrian Brasoveanu. Structured Nominal and Modal Reference. PhD thesis, Rutgers, New Brunswick, NJ,
2007.
Richard Breheny. Indefinites and anaphoric dependence: A case for dynamic semantics or pragmatics? In M.
Reimer and A. Bezuidenhout, editors, Descriptions and Beyond. Oxford University Press, Oxford, 2001.
Robin Cooper. The interpretation of pronouns. Syntax and Semantics, 10: 61–92, 1979.
Sam Cumming. Variabilism. The Philosophical Review, 117(4): 525, 2008.
Andy Egan. Epistemic modals, relativism and assertion. Philosophical Studies, 133 (1): 1–22, 2007.
Paul D. Elbourne. Situations and Individuals. MIT Press, Cambridge, MA, 2005.
Paul D. Elbourne. The interpretation of pronouns. Language and Linguistics Compass, 2(1): 119–150, 2008.
Nomi Erteschik-Shir. The Dynamics of Focus Structure. Cambridge University Press, Cambridge, 1997.
Gareth Evans. Pronouns. Linguistic Inquiry, 11(2): 337–362, 1980.
Kai von Fintel. Counterfactuals in a dynamic context. In Ken Kenstowicz, editor, Ken Hale: A Life in Lan-
guage, pages 123–152. MIT Press, Cambridge, MA, 2001.
Kai von Fintel and Anthony Gillies. An opinionated guide to epistemic modality. Oxford Studies in Episte-
mology, 2, 2007.
Danny Fox. Two short notes on Schlenker’s theory of presupposition projection. Theoretical Linguistics,
34(3): 237–252, 2008.
Anette Frank and Hans Kamp. On context dependence in modal constructions. In Proceedings of SALT VII,
pages 21–23, 1997.
Gerald Gazdar. Pragmatics: implicature, presupposition and logical form. Academic Press, 1979.
P.T. Geach. Reference and Generality. Cornell University Press, Ithaca, NY, 1962.
Bart Geurts. Presuppositions and pronouns. Elsevier, 1999.
Bart Geurts and David Beaver. Discourse representation theory. Stanford Encyclopedia of Philosophy, 2007.
Anthony Gillies. A new solution to Moore’s paradox. Philosophical Studies, 105: 237–250, 2001.
Anthony Gillies. Epistemic conditionals and conditional epistemics. Noûs, 38: 585–616, 2004.
Anthony Gillies. Counterfactual scorekeeping. Linguistics and Philosophy, 30(3): 329–360, 2007.
Jonathan Ginzburg. Interrogatives: Questions, facts and dialogue. The Handbook of Contemporary Semantic
Theory, 5:359–423, 1996.
Jeroen Groenendijk. The logic of interrogation. In Proceedings of SALT IX. Institute for Logic, Language and
Computation (ILLC), University of Amsterdam, 1999.
Jeroen Groenendijk and Martin Stokhof. Dynamic predicate logic. Linguistics and Philosophy, 14:39–100,
1991.
Jeroen Groenendijk and Martin Stokhof. Modality and conversational information. Theoretical Lingusitics,
2:61–112, 1975.
Jeroen Groenendijk, Martin Stokhof, and Frank Veltman. Coreference and modality. In S. Lappin, editor,
Handbook of Contemporary Semantic Theory. Blackwell, 1995.
Irene Heim. The Semantics of Definite and Indefinite Noun Phrases. PhD thesis, University of Massachusetts,
1982.
Irene Heim. File change semantics and the familiarity theory of definiteness. In R. Bauerle, C. Schwarze,

277
 SETH YA L C I N

and A. Von Stechow, editors, Meaning, Use and Interpretation of Language, pages 164–90. De Gruyter,
1983a.
Irene Heim. On the projection problem for presuppositions. In M. Barlow, D. Flickinger, and M. Wescoat,
editors, WCCFL 2: Second Annual West Cost Conference on Formal Linguistics, pages 114–25. Stanford
University, 1983b.
Irene Heim. Presupposition projection. In Reader for the Nijmegen Workshop on Presupposition, Lexical Mean-
ing, and Discourse Processes, Univ. of Nijmegen, 1990a.
Irene Heim. E-type pronouns and donkey anaphora. Linguistics and philosophy, 13 (2): 137–177, 1990b.
Irene Heim. Presupposition projection and the semantics of attitude verbs. Journal of Semantics, 9: 183–221,
1992.
Joris Hulstijn. Structured information states: raising and resolving issues. In A. Benz and G. Jager, editors,
Proceedings of MunDial97, University of Munich, 1997.
Pauline Jacobson. Towards a variable-free semantics. Linguistics and Philosophy, 22 (2): 117–185, 1999.
Hans Kamp. A theory of truth and semantic representation. In J. Groenendijk, T. Janssen, and M. Stokhof,
editors, Formal Methods in the Study of Language. Mathematisch Centrum, University of Amsterdam,
Amsterdam, 1981.
Hans Kamp and Uwe Reyle. From Discourse to Logic. Springer, 1993.
Lauri Karttunen. What makes definite noun phrases definite? Technical Report P3871, RAND Corpora-
tion, Santa Monica, CA, 1968.
Lauri Karttunen. Discourse referents. In J. McCawley, editor, Syntax and Semantics 7, pages 363–85, Aca-
demic Press, New York, 1976.
Lauri Karttunen and Stanley Peters. Conventional implicature. In C. K. Oh and D. Dineen, editors, Presup-
position, Syntax and Semantics vol. 11, pages 1–56. Academic Press, 1979.
Niko Kolodny and John MacFarlane. Ifs and oughts. Journal of Philosophy, CVII(3): 115–43, March
2010.
Angelika Kratzer. Modality. In A. von Stechow and D. Wunderlich, editors, Semantics: An International
Handbook of Contemporary Research, pages 639–50. De Gruyter, 1991.
Saul A. Kripke. Presupposition and anaphora: Remarks on the formulation of the projection problem. Lin-
guistic Inquiry, 40(3): 367–386, 2009.
David K. Lewis. General semantics. Synthese, 22:18–67, 1970.
David K. Lewis. Adverbs of quantification. In E. L. Keenan, editor, Formal Semantics of Natural Language,
pages 3–15. Cambridge University Press, Cambridge, 1975.
David K. Lewis. Attitudes de dicto and de se. Philosophical Review, 88(4): 513–43, 1979.
John MacFarlane. Epistemic modals are assessment-sensitive. In Andy Egan and Brian Weatherson, editors,
Epistemic Modality. Oxford University Press, Oxford, 2011.
Sarah Moss. On the pragmatics of counterfactuals. Noûs, 2010.
Reinhard Muskens. Combining Montague semantics and discourse representation. Linguistics and Philosophy,
19(2):143–186, 1996.
Reinhard Muskens, Johan van Benthem, and Albert Visser. Dynamics. In Johan van Benthem and Alice ter
Meulen, editors, Handbook of Logic and Language, pages 607–670. Elsevier, 2nd edition, 2011.
Dilip Ninan. Imagination, Content, and the Self. PhD thesis, MIT, 2008.
Dilip Ninan. De se attitudes: Ascription and communication. Philosophy Compass, 5 (7): 551–567, 2010.
Terry Parsons. Pronouns as paraphrases. Unpublished manuscript, University of Massachusetts, Amherst,
1978.
Craige Roberts. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy,
12(6): 683–721, 1989.
Craige Roberts. Anaphora in intensional contexts. In S. Lappin, editor, Handbook of Contemporary Semantic
Theory. Blackwell, Oxford, 1996.
Daniel Rothschild. Presupposition projection and logical equivalence. Philosophical Perspectives, 22(1):
473–497, 2008a.
Daniel Rothschild. Transparency Theory and its dynamic alternatives: Commentary on “Be Articulate”.
Theoretical Linguistics, 34(3): 261–268, 2008b.
Daniel Rothschild. Explaining presupposition projection with dynamic semantics. Semantics and Pragmat-
ics, 4(3): 1–43, 2011.
Daniel Rothschild and Seth Yakin. Dynamics. Unpublished manuscript, 2011.
Philippe Schlenker. Be articulate: A pragmatic theory of presupposition projection. Theoretical Linguistics,
34(3): 157–212, 2008.

278
 DYN A M I C SEMA N T I C S

Philippe Schlenker. Local contexts. Semantics and Pragmatics, 2(4): 1–78, 2009.
Scott Soames. Presupposition. Handbook of Philosophical Logic, 4: 553–616, 1989.
Robert Stalnaker. Pragmatics. Synthese, 22: 272–89, 1970.
Robert Stalnaker. Pragmatic presuppositions. In M. K. Munitz and P. Unger, editors, Semantics and Philoso-
phy. NYU Press, New York, 1974.
Robert Stalnaker. Indicative conditionals. Philosophia, 5, 1975.
Robert Stalnaker. Assertion. In P. Cole, editor, Syntax and Semantics 9: Pragmatics. Academic Press, 1978.
Robert Stalnaker. Inquiry. MIT Press, Cambridge, MA, 1984.
Robert Stalnaker. On the representation of context. Journal of Logic, Language, and Information, 7: 3–19,
1998.
Mark Steedman. Temporality. In Alice ter Meulen and Johan van Benthem, editors, Handbook of Logic and
Language, pages 895–938. Elsevier, 1997.
Tamina C. Stephenson. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and
Philosophy, 30(4): 487–525, 2007.
Matthew Stone. The anaphoric parallel between modality and tense. Technical Report 06, Institute for
Research in Cognitive Science, University of Pennsylvania, 1997.
Matthew Stone and Daniel Hardt. Dynamic discourse referents for tense and modals. Computing Meaning,
1: 301–320, 1999.
Alice Ter Meulen. Representing Time in Natural Language–The Dynamic Interpretation of Tense and Aspect.
MIT Press, Cambridge, MA, 1997.
Johan van Benthem. Essays in Logical Semantics. Studies in Linguistics and Philosophy. Reidel, Dordrecht,
1986.
Johan van Benthem. Exploring Logical Dynamics. Center for the Study of Language and Information, Stan-
ford, CA, 1996.
Robert A. Van der Sandt. Presupposition projection as anaphora resolution. Journal of Semantics, 9(4): 333,
1992.
Jan van Eijck and Hans Kamp. Discourse representation in context. In Johan van Benthem and Alice ter
Meulen, editors, Handbook of Logic and Language, pages 181–252. Elsevier, 2nd edition, 2011.
Frank Veltman. Logics for Conditionals. PhD thesis, University of Amsterdam, Amsterdam, 1985.
Frank Veltman. Defaults in update semantics. Journal of Philosophical Logic, 25(3): 221–61, 1996.
Malte Willer. Modality in Flux. PhD thesis, University of Texas, 2010a.
Malte Willer. New surprises for the Romsey Test. Synthese, 176(2): 291–309, 2010b.
Seth Yalcin. Epistemic modals. Mind, 116(464): 983–1026, Nov. 2007.
Seth Yalcin. Nonfactualism about epistemic modality. In Andy Egan and Brian Weatherson, editors,
Epistemic Modality. Oxford University Press, Oxford, 2011a.
Seth Yalcin. A counterexample to Modus Tollens. Journal of Philosophical Logic, 2011b.

279
 BA RRY SC H E I N

2.8
EVE NT S E M ANT IC S
Barry Schein

Davidson’s diamonds are a recurrent pattern of inference in natural language, in which

elimination of a modifier phrase or argument, the arrows in (1), is valid; but, introduc-
tion reversing the arrows is not (Davidson 1967, 1985; Castañeda 1967; Parsons 1985,
1990):

(1) Jones (slowly) buttered the toast with a knife in the kitchen.
  OP
Jones (slowly) buttered the toast with a knife. Jones (slowly) buttered the
toast in the kitchen.
  PO
Jones (slowly) buttered the toast.
  OP
Jones (slowly) buttered. The toast was (slowly) buttered.
  PO
There was a (slow) buttering.
(2) e(Agent(e, x1) & slow(e) & butter(e) & Patient(e, x2) &
with (e, x3) & in(e, x4))

A Davidsonian logical form (2), being a conjunction of terms, explains that elimination
of a modifier or argument is elimination of a conjunct. It further represents that all the
conjuncts are about a particular e in a sentence which asserts that an e exists to satisfy
them all. Introduction is thus shown to be a mistaken inference from the existence of
some particulars satisfying given descriptions to the existence of a single one satisfying
a conjunction of them:

(3) e(Agent(e, j) & butter(e) [an x3: knife(x3)] with(e, x3))

Jones buttered with a knife.
e(Agent(e, j) & butter(e) & [the x4: kitchen(x4)] in(e, x4))
Jones buttered in the kitchen.
Ӎ e(Agent(e, j) & butter(e) & [an x3: knife(x3)] with(e, x3) &
[the x4: kitchen(x4)] in(e, x4))
Jones buttered with a knife in the kitchen.

Buttering with a knife in the pantry and again in the kitchen with a spatula does not
imply that Jones ever took knife to butter in the kitchen.
 EVEN T SEM A N T I C S

In the logical forms for (1), the morpheme butter emerges with a single meaning for
all its occurrences in various syntactic and morphological contexts. Variable polyadicity
ranging from zero to four arguments is just an illusion of the monadic butter in construc-
tion with a varying number of conjuncts. Logical form here answers to both inference
in natural language and the univocality of its morphemes. Event semantics, so called,
comprises those approaches to inference and grammar that would, in agreement with
(2), introduce new parameters and decompose the simple sentence into a conjunction
of terms containing them. Events, by the way, are the values for e in true sentences. [See
Schein 2002 for more on the metaphysics of events. Speakers affirm (i)-(iii) without
implying (iv) or (v), a dilemma for Davidsonian event semantics as its uncomplicated
logical forms for (i)–(iii) do imply (iv) and (v) (see Lombard 1985).

(i) The Carnegie Deli faces Carnegie Hall.

(ii) Carnegie Hall faces the Carnegie Deli.
(iii) The Carnegie Deli facing Carnegie Hall is Carnegie Hall facing the
Carnegie Deli.
(iv) F The Carnegie Deli faces the Carnegie Deli.
(v) F Carnegie Hall faces Carnegie Hall.

A necessary complication is to recognize that the material states identified in (iii) are
not the observations under a perspective or scenes of those states that distinguish (i)
and (ii) and undo the unwanted implication. The scenes or perspectives that here res-
cue event semantics from itself are however pervasive wherever there is predication.]
This chapter rummages for more gems of natural language to fix the logical syntax of
talk about events in simple sentences. The existential event quantification with scope
over the entire sentence in (2) turns out to be plural, and the cross-reference to events
from one conjunct to another is by plural pronoun or definite description (plural event
quantifiers and pronouns) rather than bare variable. Instead of relating to the same event
e as in (2), every argument prompts its own events in a more extensive decomposi-
tion (supermonadicity), in which various relations between the several events hold the
sentence together (see (25) below). Emending how quantifier phrases and descriptions
prefix to their scope, adverbialization interposes an adverb derived from the nominal
phrase’s descriptive content so that what is spoken as (4), for example, is always parsed
along the lines of (5):

(4) The butler buttered.

(5) The butler while a butler buttered.

These emendations retain the original explanation for the validity of modifier and argu-
ment elimination and for the invalidity of their introduction; but, the logical form that
replaces (2) paraphrased sounds more like (6):

(6) Jones while Jones participated & it was slow & it caused a buttering &
the toast while toast participated in that, which had assistance which a
knife while a knife participated in, & it was a being-in which the kitchen
while a kitchen participated in.

It gets worse with annotations in logical form referring to perspectival events or states,
scenes or frames of reference for the outside events the sentence is a report of. Event

281
 BA RRY SC H E I N

semantics, its syntax revised and the architecture of simple clauses fortified, is then
found to resolve problems of extensional substitutivity under identity and to clarify the
logical form of identity statements in natural language, deploying event talk to safe-
guard naive reference into an austere ontology.

§1.1. Plural event pronouns make a quick entrance behind adverbs denoting events.
Contrary adverbs (e.g., gracefully vs. clumsily) prove the source of syntactical insight:

(7) Jones gracefully and Godfrey clumsily buttered 613 pastries for brunch.

As what is graceful is not clumsy, Jones buttering 365 brioches and Godfrey, 248
croissants, attend to brunch with their own events. Their graceful and clumsy actions
together butter the pastries. Saying so requires two more existential event quantifiers,
and a plural pronoun dividing its reference between the two events supplants a simple
event variable in relating action to buttering:

(8) e e1 e2((Agent(e1, j) & graceful(e1)) and (Agent(e2, g) & clumsy(e2)) &
[ΆE: Ee1 & Ee2] butter 613 pastries[e, E])

[Atomic formulae, e.g., ‘butter(e)’, ‘Ee’, appear with variables bare or in parentheses.
Brackets indicate a molecular formula with the variables enclosed free, e.g., ‘butter the
toast[e,E]’. N.B. In ‘Agent[E,X]’, ‘cluster[E]’, etc, the brackets flag abbreviation of com-
plex, second-order formulas composed of first-order relations—not primitive, higher-
order relations, v. (15), (18).]
Note that and is univocally the sentential connective. The burden of collective pred-
ication falls on the plural event pronoun here and also in the logical form for (9), identi-
cal to (8) except for omission of the adverbs.

(9) Jones and Godfrey buttered 613 pastries.

Contrary adverbs recur in (10) and (11) with similar mission and further instruction:

(10) Jones gracefully and Godfrey clumsily separated the brioches to the left
and the croissants to the right.
(11) At the loom, Arachne gracefully and her apprentice clumsily criss-crossed
silk threads horizontally and linen threads vertically.

Neither the leftward nor the rightward parade of pastry is a separating on its own; and,
the separating is itself neither to the left nor right, having components of both. So nei-
ther the leftward event nor the rightward event is described by the verb, and the adverbs
modify a separate relation, Theme in (12), describing the brioches’ and the croissants’
participation (pace Kratzer 1996, 2003):

(12) e e1e2e3e4 ((Agent(e1,j) & graceful(e1)) and (Agent(e2,g) &

clumsy(e2)) &
[ΆE12: E12e1 & E12e2] [ΆE34: E34e3 & E34e4] separate[e, E12, E34] &
(Theme(e3,b) & to the left[e3]) and (Theme(e4,c) & to the right[e4]))

Another plural pronoun referring to the leftward and rightward events must, in turn,
relate the brioches’ and croisssants’ participation to the separating, an e that separate(e).

282
 EVEN T SEM A N T I C S

To continue in the spirit of (2) until it is abandoned below for supermonadicity, (10)
could still be held to be about a single separating and the graceful and clumsy on the one
hand and the leftward and rightward on the other could just be two ways to parse it into
mereologically smaller events, into those where Jones and Godfrey are Agents and into
those where the brioches and the croissants are Themes.
Plural event pronouns have been introduced to collect the events in which partici-
pants participate, and they divide their reference accordingly. Verbs also denote plural
events and so then must the quantifier prefixed to the sentence. Plural existential event
quantification (Gillon 1990; Pietroski 2005; Schein 1993, 2006; Schwarzschild 1991,
1996) with a verb denoting plural events is found in (13), given that the same mor-
pheme cluster occurs twice:

(13) The fires will cluster in two clusters (and not in one).
(14) E ([The X: fires(X)] Theme(E,X) & cluster[E] . . .))
(15) cluster[E] ldef eEe & e(Ee o cluster(e))

Divine law and faith that the events scattered in history are distinct find plural event
quantification in (16) too:

(16) In the era chronicled, two Roman bishops against their will and two Angli-
can churchmen at the behest of the throne have joined three English
kings eagerly and ten French princesses reluctantly in holy matrimony.

Here the several joinings the verb denotes are equal in number to the sum of events
executed against the will and events at the behest of the throne and fewer than the
events submitted to eagerly and reluctantly; but, they are all in the plural.

§1.2. A more extensive decomposition of polyadicity is next, supermonadicity. In (16),

the clerics are all joiners and so all Agents. In the conjunction of predicate phrases
in (17), the first describes provocateurs, Agents in actions at the Pentagon and the
second, victims, that is, Patients, acted upon on the Mall:

(17) The Columbia students (noisily) and the Harvard students (quietly) sur-
rounded the Pentagon and were crowded into the Mall.
(18) O[Ei, Ej] ldef e(ei(Eiei & overlap(e,ei)) l ej(Ejej & overlap(e,ej)))
(19) e e1e2e3e4(Participate(e1, C) and Participate(e2, H) &
[ΆE12: E12e1 & E12e2][ΆE34: E34e3 & E34e4] O[E12, E34] &
e3(X Agent(e3, X) & surround the Pentagon[e3,X]) and
e4(X Patient(e4, X) & be crowded into the Mall[e4, X])
The Columbia students participated, and the Harvard students participated; & it all
was a surrounding the Pentagon and a being crowded into the Mall.

Sentence (17) commits a student to neither venue, leaving vague whether she acted at
the Pentagon or was acted upon on the Mall. All that is implied is that the Columbia
students participate in some way as do the Harvard students, and what they do consti-
tutes the events subsequently described, so that some of the Columbia and Harvard stu-
dents must have been at the Pentagon while the rest were on the Mall. The conjuncts
in (19) mentioning the students are necessarily vague about what they are up to in

283
 BA RRY SC H E I N

their noisy and quiet events, asserting only their participation. That vagueness how-
ever threatens translation of (20). In the event that Harvard students mercilessly tease
Columbia students and Columbia students decline to tease themselves, (20) is false:

(20) The Harvard students and the Columbia students teased the Columbia
students.
(21) The Harvard students participate and the Columbia students participate;
& it was a teasing.

But its considered translation paraphrased in (21) is mistakenly true in that both the
Harvard students and the Columbia students have participated, the former as aggressors
and the latter as victims in the same event of teasing. To divorce the vagueness neces-
sary for (17) from its ill effect on (20), the logical form for the latter must resolve two
events, cause and effect:

(22) e1 e2 e3(Participate(e1, C) and Participate(e2, H) &

[ΆE12: E12e1 & E12e2][ΆE3: E3e3] Cause(E12, E34) & tease(e3) &
Participate(e3, C))
The Columbia students participate and the Harvard students participate; & it caused
a teasedness (cf. ‘embarrass’, ‘shame’, ‘humiliate’); & the Columbia students par-
ticipate in that.

The vagueness of the Harvard and Columbia students’ participation notwithstand-

ing, what they do is said to cause the Columbia students’ humiliation. The Columbia
students had no hand in the causal event, and (22) is false, as a translation of (20)
should be.
This argument repeats itself mutatis mutandis wherever nominal phrases can occur
allegedly conjoined, and thus a resolution into smaller events occurs across-the-board at
every argument position. [E.g., 613 diverse hedge funds variously sold and sold to 365 equity
funds and 248 mutual funds in a single day’s trading. The preparation of these Balkan delica-
cies has variously stuffed full and stuffed into other vegetables— thirty eggplants, fifty peppers,
eighty tomatoes, 130 olives and 210 pearl onions.] [Pietroski (1998), arguing from rather
different considerations, also reaches the conclusion that thematic relations themselves
occasion the introduction of an event quantifier. A dyadic relation ‘C(e1, e2)’, if e1 is
meant to be a cause and e2 its effect, leaves no room for reference to the event that is
cause and effect combined, for which purpose one may prefer to substitute a triadic rela-
tion ‘C(e1,e2,e3)’. See Pietroski 1998, 2000 and references cited and Schein 2002.]
In the initial Davidsonian logical form (24) for (23), relations to events relate to the
same event e:

(23) Jones buttered the toast in the kitchen.

(24) Agent(e, x1) & butter(e) & Patient(e, x2) & in(e, x3)
(25) Participate(e1, x1) & Cause(e1,e2) & butter(e2) & Participate(e2, x2) &
in(e2, e3) & Participate(e3, x3)

With supermonadicity, it is small events all the way down, with every nominal phrase,
including each conjoined (see (11)), launching its own. It doesn’t get lonelier than

284
 EVEN T SEM A N T I C S

that, spinning around in a subevent solo, unrelated to anything else except through
the mediation of further relations between events. This revision to clausal architec-
ture is warranted by three elementary observations: i. sentences with multiple conjunc-
tions (17) demand a certain vagueness about the manner of participation in the events
described; ii. this vagueness should not then confuse aggressors and victims (20); and
iii. any participant’s participation may be described by an adverb that is contrary to how
anyone else participates ((7), (10), (11), (17)).

§1.3. Adverbialization of nominal descriptive content is a fact of grammar. First, descrip-

tive anaphora, both implicit and explicit as in (26), co-varying with the events described,
must fall within the scope of the event quantifier adverbialization derives as in (28):

(26) Every fugitive from a federal prison is soon apprehended near it. (after
Enç 1986)
(27) Every fugitive from a federal prisonj as of somei time is soon after thati
time apprehended near itj.
(28) [Every x: e fugitive from a federal prison[e, x]]
[while e: fugitive from a federal prison [e, x]]
e’(soon(e, e’) & apprehended[e’, x] &
[Άy: federal prison(y) from which x is a fugitive in e] near(e’, y))

Second, the adjective respective can only be interpreted as in some way modify-
ing the matrix scope of its host nominal phrase, as if it were the adverb respectively
(Gawron & Kehler 2002, 2004). Suppose that twins rent the same make and model
vehicle and pose for photographs with the two cars in which it is variously true
that:

(29) The twins stood beside their respective rentals.

The twins stood behind their respective rentals.
(30) The twins stood between their rentals.

In contrast to (30), there can be no photograph for which:

(31) *The twins stood between their respective rentals.

For the twins to be in respective possession of their rentals, as the nominal describes
them, puts them in a pair of states in each of which a twin is opposite her own rental.
Such a state frames a standing beside (or a standing behind) if its twin is beside (or
behind) her rental, (29). But, neither of the states that respective possession denotes
frames a standing of a twin between the rental in it, (31). This nonsense that (31) can-
not escape is a consequence of adverbialization demanding that the twins being with
their respective rentals frame their standings between.
The scope of adverbialization tracks the scope of the nominal phrase adverbialized:

(32) At the thriller in Manila, Muhammad Ali’s trainer and Joe Frazier’s
trainer pulled their respective champions apart.
(33) At the thriller in Manila, Muhammad Ali’s trainer and Joe Frazier’s
trainer pulled apart their respective champions.

285
 BA RRY SC H E I N

Given the scope of their respective champions in (32), it is immediate that the states of
respective possession separately frame only the champions being pulled: Ali’s trainer
grabbing Ali and Frazier’s trainer grabbing Frazier pull hard until the champions are
parted. Later noticed is the macabre interpretation in which the states of respective pos-
session each frame a champion being pulled apart by his trainer. The macabre becomes
more salient in (33) to the extent that stylistic movements favor the scope worn on the
sleeve, here the one now including apart within the scope of their respective champions
(slighted is the variation from speakers like myself for whom (33) is unambiguously
macabre to those for whom the contrast between (32) and (33) is rather weak.). Note
that for the only sensible interpretation of (32) and (33), the scope that adverbializa-
tion affords to respective is the minimal one that the host nominal phrase itself requires,
a relation expressing participation in some events—a scope that is too small for the
felicitous use of the proper lexical adverb respectively:
(34) * At the thriller in Manila, Muhammad Ali’s trainer and Joe Frazier’s
trainer respectively pulled their champions apart.
(35) * At the thriller in Manila, Muhammad Ali’s trainer and Joe Frazier’s
trainer pulled apart their champions respectively.
(36) ?? At the thriller in Manila, Muhammad Ali’s trainer and Joe Frazier’s
trainer pulled their champions respectively— apart.
Nominal descriptions are the adverbs dedicated to the smallest relations to events,
reflecting an economy of expresssion and compactness in natural language, which
reserves lexical adverbs for larger phrases containing whole verbs.
[‘While’ appears in the translations above for illustration. Expected is the range of
relations, temporal, causal, causal explanatory, mediating the interpretation of if-clauses
or absolutive clauses including variation therein due to differences between tensed and
reduced clauses.
Adverbialization is grammar sometimes without apparent effect framing the events
reported, The liar was deep asleep—but not while lying. Compare:
(i) The liar fails a polygraph. (ii) The sometimes liar fails a polygraph.
Any liar fails a polygraph. Any sometimes liar fails a polygraph.
Every liar fails a polygraph. Every sometimes liar fails a polygraph.
As expected from adverbialization, the polygraph failures of (i) are an effect of lying, and
any lie that escapes detection is a counterexample to what is asserted. In (ii), it seems that
sometimes lying makes one anxious enough to fail a polygraph even when telling the truth.
To be a sometimes liar is to always be a sometimes liar, for as long as one lives. So, to do
something while a sometimes liar is to do something while oneself. Thus, The liar was deep
asleep is The sometimes liar while a sometimes liar was deep asleep, with sometimes unspoken.]
§2. Extensional Substitutivity under Identity. (Barber 2000; Carlson 1982, Forbes 1997,
1999; Predelli 1999; Saul 1997 and, 1999, 2007; Szabó 2003)
Adverbialization (see Forbes 1999) joins forces with supermonadicity to resolve puz-
zles of extensional substitutivity under identity:

(37) The ponds the temperatures of which were recorded early this morning
were colder than the ponds the temperatures of which were recorded
early this evening.

286
 EVEN T SEM A N T I C S

(38) Every pond the temperature of which was recorded early this morning
was colder than every pond the temperature of which was recorded early
this evening.
(39) Venus Pond early this morning was colder than Venus Pond early this
evening.
(40) * Venus Pond was colder than Venus Pond (was).
From neither (37) nor (38) is there implication that the morning ponds are other than
the evening ponds. If, in fact, Venus Pond (Maui) is among them both, (37) and (38)
imply (39). Insofar as adverbial qualification is necessary to escape contradiction (cf.
(40)), its only source in (37) and (38) are the nominal phrases themselves which,
according to adverbialization, do indeed supply adverbial phrases to qualify participa-
tion in the events described.
As morning is not evening and manhood not boyhood, the adverbs from adverbiali-
zation in (37) and (38) and in (41) cannot apply to the same event, a surpassing in (41),
without contradiction:
(41) a. The man surpassed the boy of his childhood in regrets and not much
else.
b. Many a man surpasses the boy of his childhood in regrets and not
much else.
Only with supermonadicity providing enough events for the derived adverbs to modify
will adverbialization license the substitutions. To substitute the such-and-such for the
so-and-so in a sentence of natural language is always a double substitution in logical
form—the such-and-such while such-and-such-ing for the so-and-so while so-and-so-ing. To
be valid, it is not enough that the such-and-such are the so-and-so. It must also be
that while such-and-such-ing is while so-and-so-ing, that is, that the derived adverbs
describe the same events too.
§3. Frames of reference and scenes. The event that (2) describes and the events super-
monadicity and adverbialization have so far introduced purport to deconstruct the
goings on of the speaker’s conscious report even as she is unaware of the fine-grained
events enveloping her. Further annotations to logical form indicate perspective and
frame of reference, representing the epistemic conditions for observation of the events
reported. If, for example, the frame of reference for (42) is taken to be that of the bal-
lerina’s proximate visual experience, the sentence puts her partner in synchronous orbit
around her (possibly leading from an outstretched arm):
(42) The ice ballerina pirouetted with her partner on the left across the rink.
Alternatively, the frame of reference could be fixed at a moment by the line-of-sight
pointing across the rink and tangent to the ballerina’s curved trajectory, tracking the
partner in a parallel course, on one side or the other depending on whether the speaker
faces the ballerina and the frame of reference is centered on the speaker’s point of view
or on the ballerina’s. The sentence is thus three-ways ambiguous, and it communicates
no determinate thought at all unless the speaker and hearer agree on the frames of refer-
ence intended. A lexical ambiguity—left1, left2 and left3—to sort them out would insult
the concept’s integrity, which must rather be left in f, for different choices for frame of
reference f. In contexts that are not demonstrative for f, as in third-party transmission,

287
 BA RRY SC H E I N

to have understood (42) is to have grasped some descriptive intention as in (43), absent
which, again, no determinate thought is expressed:

(43) In the orbital frames of references f centered on the agent’s proximate

visual experience, the ice ballerina pirouetted with her partner on the left
in f across the rink.

Sentence (44) is also three-ways ambiguous—mercifully, not the nine ways it would
be if the frame of reference for the second token of left were independent and possibly
discontinuous from the first’s:

(44) The ballerina will pirouette with her partner on the left to the middle of
the ice, do a triple somersault, and pirouette with her partner on the left
the rest of the way.

The instruction and knowledge not to switch frames of reference within sentence is
itself a point of grammar. There is no thought here without description of a frame of
reference, and its representation is subject to grammatical constraint.
For purposes of spatial orientation and navigation, analysis of the visual and auditory
scene includes parameters justified by the agent’s accurate navigation even as she is
unaware of them in her experience of coarse-grained objects in location and in motion.
Among the communicative intentions the design of language supports, I include the
intention to convey to a companion witness accurate enough to guide her navigation,
for which she has to acquire a scene from words of mine encoding parameters that I do
not consciously attend to or assert, such as a frame of reference. How the world is at any
given moment, as fixed in true propositions, is not accurate witness for where you are
in it, for which the logical form of utterances should be tricked out to choreograph and
calibrate the speaker’s and hearer’s thoughts and perceptions. (Think of two triangulat-
ing agents formulating and exchanging course corrections based on outputs and inputs
in the language of a visual guidance system.) If narrative is rich enough to support the
communicative intentions of its narrator, who is real or imagined (and always imagined
if not real), it is shortsighted to suppose that the only parameter for narrative continuity
is temporal—a timeline for world history and a succession of points or intervals along
it as the narrative advances its ‘now’. Even a ‘here and now’, the indices for a context
of utterance, will not fix the thoughts expressed in a spatial vocabulary including left.
Here we now are, imagine, at the moon landing site on schedule, and yet we are diso-
riented and disabled for visual navigation if the faulty periscope from our windowless
rover swivels about randomly, no matter how detailed the image transmitted nor acute
my report of it. A sequence of scenes or their narration is orienting and navigational
for observer, speaker, or hearer only if their lines-of-sight are known and the scenes
constitute a survey conforming to some natural conditions on their continuity—condi-
tions that regulate over the course of survey and narrative such parameters as needed to
constitute a natural (stereoscopic, i.e., 3-D) cinematic experience. As with any contex-
tual or discourse parameters, thoughts are more or less dependent on them, and some
narratives may wash out that dependency in adopting an ominiscient narrator or an
arbitrary, unknown or neutral perspective. Nevertheless those parameters are always
there for the asking.

288
 EVEN T SEM A N T I C S

In resolving a celebrated instance of the substitution puzzle, the metaphysics is

spared only when conditions step in on the speaker’s epistemic stance towards the
events under report, granting that the nominal description contains expressions of such
conditions:

(45) Three million passengers crowded National Airlines’ routes last year.
(46) ӎ Three million persons crowded National Airlines’ routes last year.
(after Gupta 1980: 23; Moore 1994)

Frequent flyers recounted under (45)’s protocol keep it from implying (46). If the arith-
metic predicate three million is a simple property ‘3M(_)’ determined by the numeric
identity of what is referred to, three million passengers does not refer to the one million
whole persons who flew frequently. Away we go, denying the identity of passengers and
persons to thwart the inference from (45) to (46) and asserting it in (47):

(47) The three million passengers who crowded National Airlines’ routes last
year were the one million frequent flyers loyal to it.

Worse, the three million passengers in (45) who are fewer persons do not always exist
as such, subject to a condition that Doetjes & Honcoop (1997) aptly call sequencing of
events:

(48) Three million passengers had three million opinions about the food on
National Airlines.
(49) *Three million passengers have three million opinions about the food on
National Airlines.

Sentence (45) can be followed by (48), recounting the passengers’ experience onboard
and continuing to count as many the fewer. In the present tense, however, (49) can-
not count more passengers than persons who fly, despite the fact that these persons
still have three million opinions about flights they remember all too well. The relation
between the count and present tense wants explanation. Instead of the bare ‘3M(_)’,
imagine that three million translates as something like “now counted this way to three
million,” a description of events of measurement (the counter clicks, as it were) under
an explicit protocol. What is counted and what the indefinite description three million
passengers refers to is naively what there is, persons, who are counted and under some
protocols recounted. With the counting now entered into the description, adverbializa-
tion derives (50) and(51):

(50) Three million passengers while counted to three million had three million
opinions about the food on National Airlines.
(51) *Three million passengers while counted to three million have three million
opinions about the food on National Airlines.

Clicking the counter as passengers go by frames or locates opinion recorded contem-

poraneously as in (48); but, there is no counting to three million that frames current
conditions, which, according to (51), is what defeats (49). The speaker is secure in her
constant reference to naive, familiar objects by a sleight of hand that manipulates her
epistemic conditions, in this case, conditions of measurement.

289
 BA RRY SC H E I N

A fundamental relationship exists between the conditions of measurement and the

scenes of what is counted, which under adverbialization frame the events reported.
In the narrow scene or frame of reference in (52), there are exactly two grey regions
if finitely many. There is only one in the wide scene or frame of reference. This is
not a metaphysical claim, for surely the regions counted under the narrow frame of
reference still exist as parts of the grey ring under the wider frame of reference.
They are just not to be counted there, under conditions inappropriate for their
measurement.

(52)

Think of the design of an optical counter, blind to what is counted—passengers or

planets if you say so. To utter two is to take on a perspective from which what is reported
projects a scene the optical counter measures as 2. Counting two commits to the narrow
scene or frame of reference in (52) and to a report of what the events counted two may
frame within it, whether counting passengers or planets, as reflected in number agree-
ment in the following:

(53) The morning star and the evening are playing hide-and-seek in the twi-
light, season after season.
(54) *From the vantage point of Jupiter, (you can see that) the morning star
and the evening star are circling the sun.
(55) From the vantage point of Jupiter, (you can see that) the morning star
and the evening star is circling the sun.

Narrow scenes while the morning star and while the evening star exist only from Earth.
From Jupiter, if anything, what is while the morning star and while the evening star
converges on the same large scene of Venus in orbit, where there is exactly one Venu-
sian region, undermining the plural number agreement.
[Scenes are introduced with different purpose in (Schein 2002) to address puzzles of
event identity, and perspectives in (Schein 1993: chapter 10) to complete the incom-
plete definite descriptions translating pronouns.]
§4. Identity Statements If supermonadicity and adverbialization are everywhere,
they must also touch the most elementary assertions of identity. In any case, parsing
(56)–(58) without them as closures of ªX = Yº is a garden path into a substitution puzzle
without exit, as the sentences could only chase after different values for X and Y:

290
 EVEN T SEM A N T I C S

(56) In a mutual declaration of requited love, the lover and belovèd in the first
love note are not the lover and belovèd in the note returned.
(57) The lover and belovèd in either love note are the lovers exchanging
them.
(58) Ӎ The lover and belovèd in the first love note are not the lovers exchang-
ing them.

The logical forms and lexical items contained in (56)–(58) must admit modification by
Tense at the very least and by the adverbs that, by hypothesis, adverbialization intro-
duces with every nominal phrase. Beside hypothesis, modification in identity statements
finds overt expression in (61), which differs from (60) in truth and differs only in the
presence of a secondary predicate:

(59) Cezanne, Still Life with Vessels, Fruit and Tablecloth and Still Life with
Ginger Jar, Sugar Bowl and Oranges

Photo credit: Scala/Art Resource, NY

g g g

Digital Image © The Museum of

Modern Art/Licensed by SCALA/
Art Resource, NY

(60) T The oranges in the first still life are the oranges in the second.
(61) F The oranges in the first still life are the oranges in the second arranged
the same way.

291
 BA RRY SC H E I N

F They are them arranged the same way.

To be them is not to be them that way. It wants explanation how a classical identity
statement simpliciter comes to be expressed in the same words as conditioned identity
except for the condition itself. There is no room for modification in ªX = Yº. Instead,
an identity relation ‘ͭ[E1, E2]’ is said to hold between events or states only if the partici-
pants in E1 are identical to the participants in E2. ͭdentity holds between the oranges’
participation in the event or state depicted in the first canvas and their participation in
that of the second canvas, verifying (60) (see (62)); but, that is not to say that their state
in the second canvas is also the same arrangement. The further comment describing it
as such results in the falsehood (61) (see (63)):

(62) [the X: E oranges in 1st[E,X]][E1: E (while[E, E1] & oranges in 1st [E,X])]
participate[E1, X] & [ΆE1: that1][ΆE2: that2] ͭ[E1, E2] &
[the X: E oranges in 1st[E,X]][E2: E (while[E, E2] & oranges in 2nd [E,X])]
participate[E2, X]
(63) [the X: E oranges in 1st[E,X]][E1: E (while[E, E1] & oranges in 1st [E,X])]
participate[E1, X] & [ΆE1: that1][ΆE2: that2] ͭ[E1, E2] &
[the X: E oranges in 1st[E,X]][E2: E (while[E, E2] & oranges in 2nd [E,X])]
participate[E2, X] & arranged the same way[E2]

Given the common syntax that (60) and (61) warrant, adverbialization just joins in,
as in (62) and (63), to further condition the events related by ͭdentity. As weak as
ͭdentity is, (56), if it is just a negated identity simpliciter, falsely denies that there are
events with the same participants in one of which the lovers participate while lover and
belovèd and another in which they participate roles reversed. The sentence as intended
is however to be understood as a true denial of a conditioned identity, where the condi-
tion is to be conveyed by an appropriately crafted, tacit secondary predicate:

(64) The lover and belovèd in a love note are not the lover and belovèd in the
note returned oriented the same way.
(65) The lover at the 1st address in a frame of reference1 and belovèd at the 2nd
address there1 are not
the lover at the 1st address in a frame of reference2 and belovèd at the 2nd
address there1, that frame of reference1 oriented the same way as that frame of
reference2.

If all the nominal phrases are addressed to a frame of reference, the secondary predi-
cate is definable as a relation between the two frames of reference described, as the
paraphrase in (65) hints at. Resolving this last class of substitution puzzles takes it all
onboard—adverbialization, frames of reference, and the decomposition of identity
statements to comport with supermonadicity. Yet, it respects that identity statements
are always tensed and sometimes conditioned, while affirming that nominal phrases
refer to what they naively refer to.
With all the event arguments in a supermonadic clause that adverbials, tense, and
secondary predicates might modify, it becomes an empirical question to discover its
grammar, which will lead to a canonical structure for the expression of spatiotemporal
relations in a clause.

292
 EVEN T SEM A N T I C S

(66) For some evenings in 1892, Venus was Hesperus (aligned with a crescent
moon).
For some evenings in 1892, Venus was the evening star (aligned with a
crescent moon).
(67) * For some evenings in 1892, Hesperus was Venus (aligned with a cres-
cent moon).
* For some evenings in 1892, the evening star was Venus (aligned with a
crescent moon).

From the contrast (66)–(67), for example, it appears that the temporal frame adverbial,
Tense, the secondary predicate, and the adverbial derived from the predicate nomi-
nal— in (66), while Hesperus or while the evening star— coincide in the events they
describe, which are to be related by ͭdentity to events described by the subject’s derived
adverbial, while Venus in (66). Both modification and the asymmetry it induces in (66)–
(67) again put the clause structure beyond reach of an analysis based on ªX = Yº.
§5. There is no rest for the Davidsonian who pauses at (2), as a few facts of natu-
ral language lead quickly to plural event quantification, plural event pronouns in lieu
of bare variables, supermonadicity, and adverbialization. Along the way, the plural
event pronouns are the means for and to be the univocal sentential connective in all its
occurences, including (7) and (9), and all nominal phrases survive any context referring
innocently to the familiar and ordinary, despite the puzzles of substitutivity. Metaphysi-
cal arguments to the contrary all rest on errors of syntax, parsing the puzzles without
benefit of the enhanced clausal architecture. Some of these arguments are also deceived
by the smoke and mirrors of scenes and frames of reference by which thoughts are cali-
brated and which logical form records with parameters for yet more events.
[I could not have attempted Event semantics while standing on one foot without
leaning on Cleo Condoravdi, Richard Larson, Paul Pietroski, Roger Schwarzschild and
Achille Varzi. I am grateful to Terje Lohndal for extensive correspondence on an earlier
draft, and special thanks to Alexander Williams for written comments and subsequent
correspondence about the final draft and to Elena Herburger for corrections minutes
before midnight. The article draws on my (2010) ‘And’: Conjunction Reduction Redux,
USC, ms. 750pp.]

References
Alex Barber. 2000. A pragmatic treatment of simple sentences. Analysis 60.4: 300–308.
Greg N. Carlson. 1982. Generic terms and generic sentences. Journal of Philosophical Logic 11.2: 145–181.
Hector-Neri Castañeda. 1967. Comments. In Nicholas Rescher, ed. The Logic of Decision and Action. Pitts-
burgh: University of Pittsburgh Press.
Donald Davidson. 1967. The logical form of action sentences. In Nicholas Rescher, ed. The Logic of Decision
and Action. Pittsburgh: University of Pittsburgh Press.
Donald Davidson. 1985. Adverbs of action. In Bruce Vermaze & Merrill Hintikka, eds., Essays on Davidson:
Actions and Events. Oxford: Clarendon Press.
Jenny Doetjes & Martin Honcoop. 1997. The semantics of event-related readings: a case for pair-quantifica-
tion. In Anna Szabolcsi, ed., Ways of Scope Taking. Dordrecht: Kluwer Academic, 263–310.
Mürvet Enç. 1986. Towards a referential analysis of temporal expressions. Linguistics & Philosophy 9.4:
405–426.
Graeme Forbes. 1997. How much substitutivity? Analysis 57.2: 109–113.
Graeme Forbes. 1999. Enlightened semantics for simple sentences. Analysis 59.2: 86–91.

293
 BA RRY SC H E I N

Jean Mark Gawron & Andrew Kehler. 2002. The semantics of the adjective respective. In L. Mikkelsen and
C. Potts WCCFL 21 Proceedings, eds. Somerville, MA: Cascadilla Press, pp. 85–98.
Jean Mark Gawron & Andrew Kehler. 2004. The semantics of respective readings, conjunction, and filler-
gap dependencies. Linguistics and Philosophy 27: 169–207.
Brendan Gillon 1990. Plural noun phrases and their readings: a reply to Lasersohn. Linguistics & Philosophy
13: 477–485.
Anil Gupta. 1980. The Logic of Common Nouns: An Investigation in Quantified Modal Logic. New Haven: Yale
University Press.
Angelika Kratzer. 1996. Severing the external argument from the verb. In Johan Rooryck & Laurie Zaring,
eds., Phrase Structure and the Lexicon. Dordrecht: Kluwer, pp. 109–137.
Angelika Kratzer. 2003. The Event Argument. Forthcoming from MIT Press.
Ernest Lepore & Brian McLaughlin, eds. 1985. Actions and Events. Oxford: Blackwell.
Lawrence B. Lombard. 1985. How not to flip the prowler: transitive verbs of action and the identity of
actions. In Lepore and McLaughlin (1985), pp. 268–281.
Steven Moore. 1994. Events, objects and eventual objects. In Erin Duncan, Donka Farkas & Philip Spaelti,
eds., The Proceedings of the Twelfth West Coast Conference on Formal Linguistics: Conference Held at UCSC
1993., Stanford: Center for the Study of Language and Information, pp. 577–590.
Terence Parsons. 1985. Underlying events in the logical analysis of English. In Lepore and McLaughlin
(1985), pp. 235–267.
Terence Parsons. 1990. Events in the Semantics of English. Cambridge MA: MIT Press.
Paul M. Pietroski. 1998. Actions, adjuncts and agency. Mind 107: 73–111.
Paul M. Pietroski. 2000. Causing Actions. Oxford: Oxford University Press.
Paul M. Pietroski. 2005. Events and Semantic Architecture. Oxford: Oxford University Press.
Stefano Predelli. 1999. Saul, Salmon, and Superman. Analysis 59.2: 113–116.
Jennifer M. Saul. 1997a. Substitution and simple sentences. Analysis 57.2: 102–108.
Jennifer M. Saul. 1997b. Reply to Forbes. Analysis 57.2: 114–118.
Jennifer M. Saul. 1999. Substitution, simple sentences, and sex scandals. Analysis 59.2: 106–112.
Jennifer M. Saul. 2007. Simple Sentences, Substitution, and Intuitions. Oxford: Oxford University Press.
Barry Schein. 1993. Plurals and Events. Cambridge, MA: MIT Press.
Barry Schein. 2002. Events and the semantic content of thematic relations. In Gerhard Preyer & Georg
Peter, eds., Logical Form and Language. Oxford: Oxford University Press, pp. 263–344.
Barry Schein. 2006. Plurals. In Ernest Lepore & Barry C. Smith, eds. The Oxford Handbook of Philosophy of
Language. Oxford: Oxford University Press, pp. 716–767.
Roger Schwarzschild. 1991. On the Meaning of Definite Plural Noun Phrases, unpublished Ph.D. disserta-
tion, University of Massachusetts, Amherst.
Roger Schwarzschild. 1996. Pluralities. Dordrecht, The Netherlands: Kluwer.
Zoltán Gendler Szabó. 2003. On qualification. Philosophical Perspectives 17(1): 385–414.

294
 2.9
SKE P TI C IS M ABOUT
ME AN ING
Michael McDermott

The several varieties of “meaning scepticism” reject a variety of plausible claims about
meaning (not just about knowledge of meaning). The most important meaning skeptic
is Quine. I shall also consider arguments of “Kripke’s Wittgenstein” which have aroused
some interest.

1 Quine
Quine held, roughly speaking, that all talk of meaning is meaningless. That makes
it sound somewhat paradoxical, however. To understand Quine, we need some
background.
First, Quine was an empiricist. ‘Empiricism’ is the name for a diverse collection of
views united more by family resemblance than by deduction from any central tenets. It
is useful, however, to distinguish two major strands.
(i) The Empiricist Theory of Justification is concerned with the foundations of
knowledge, or of rational belief. It says, roughly speaking, that the foundation is obser-
vation, experience.
What other foundation could anyone suggest? Rejected candidates include divine
revelation, Aristotle, the Church. The classical debate between empiricism and ration-
alism concerned the claims of “reason” to be a source of knowledge or, more properly,
an independent source of knowledge (since it could hardly be denied that reason can pro-
vide knowledge when combined with experience—two persons who have had exactly
the same experience can differ in knowledge). But the current eclipse of empiricism has
been mainly at the hands of “common sense,” particularly our alleged common-sense
knowledge of the mind and modality.
“Observation,” for the most significant empiricists (including Quine), was under-
stood to be always of publicly observable events. There is, on this view, no non-infer-
ential knowledge of the mind, no “introspection”: if knowledge of the mind is possible
at all, it is by inference from observations of behavior. Admittedly, some moderate
empiricists were more liberal. And in much empiricist writing there is also some ves-
tigial feeling that “observation” of the external world is itself really inference from
internal data of some kind. This tension within empiricism was never satisfactorily
resolved. But certainly there is a well established association between empiricism and
“behaviourism”.
 M I CH A EL M CD E R MO T T

The rejection of modal data was, by contrast, completely straightforward, for empiri-
cists: observation is always of actual events.
(ii) The Empiricist Theory of Content says, roughly speaking, that the content of
a statement is to be found in the same kind of thing that would make it reasonable
to believe—experience. It’s not quite that content = evidence, as it is sometimes put:
Evidence of course consists of propositions about past experience, whereas content is a
matter mainly of future experience; but they are the same kind of thing.
This is not the same as the ETJ, and has generally been found less plausible. Many
“empiricists” are content with an uncritically commonsensical conception of con-
tent. They argue from empirical evidence for claims which seem to lack content on
the ETC—for example claims about gods, or an unobserved past, or laws or minds
conceived of in thoroughly “realist” terms. But the most significant and most con-
troversial feature of the empiricism of Quine, and of his Vienna Circle predecessors,
was its acceptance of a version of the ETC. This was verificationism: in its “classical”
form it defined Viennese “logical positivism”; in its “holistic” form it was the center
of Quine’s philosophy.
Verificationism says that the content of a statement (positivism) or theory (Quine)
is determined by what experiences would confirm or disconfirm it. If two statements/
theories would be confirmed and disconfirmed by the same experiences, they have the
same content.
The word ‘content’ is not exactly self-explanatory. It is often used, in empiricist writ-
ings, more or less interchangeably with ‘meaning’. But it would be natural, when preci-
sion matters, to agree that they are not quite the same thing. It is natural to think that
the content of a statement/theory is determined by the meanings of the words (and con-
structions) it uses. But two statements can have the same content even if they contain
different words, with different meanings, e.g. ‘John is taller than Mary’ and ‘Mary is
shorter than John’. They are equivalent by meaning, though not identical in meaning;
they are “synonymous” in a broad (and somewhat artificial) sense only. Similarly it is
natural to say that “analytic” sentences, like ‘Pigs are pigs’, lack content, although of
course they do not lack meaning.
Now how does Quine’s empiricism relate to his views about meaning? The first key to
understanding Quine, I suggest, is to see him as making an even bigger divorce between
content and meaning than that just described. In these terms, Quine’s view is that con-
tent is not even determined by meaning: a theory (not a statement) can have content
without this content being determined by the meanings of its theoretical terms. The
content of a theory is, for Quine, its predictions of experience. His point is that, in
deriving a theory’s implications for experience, recourse to knowledge of meanings will
be unnecessary because the required “truths of meaning” will be explicitly stated (in a
properly regimented theory). For example, deriving ‘x is not green’ from ‘x is red’ will
require no knowledge of the meaning of ‘red’ or ‘green’ because ‘Nothing red is green’
will be available as an extra premise. The theorist can derive ‘x is not green’ from ‘x is
red’ and ‘Nothing red is green’ without caring whether ‘Nothing red is green’ is a “truth
of meaning”.
This, I think, is the most important thesis of “Two Dogmas.” It is, however, a thesis
of the epistemological irrelevance of meaning, rather than the nonexistence of meaning.
Theory construction is thought of by Quine as being done by an individual. Even if
meaning is not relevant in that process, you might think it would be fundamental to the
use of language for communication. On our main topic, the nonexistence of meaning,

296
 SKEPTI CI SM A BO UT ME A N I N G

“Two Dogmas” has much to say; but the argument is rather weak (in my opinion), in
part because the social aspect of language is completely ignored.
The second key to understanding Quine is to see that he does have a good argument
for the nonexistence of meaning—not in “Two Dogmas,” but in Word and Object. And
the argument rests on the ETC. The central question becomes, ‘What is the empirical
content of talk about meaning?’ Quine assumes that it will have something to do with
the behavior of language users—a community of language users. An adequate empirical
theory for this domain, he argues, will not require anything like the common-sense
meaning concepts. Furthermore, the common-sense concepts could contribute no addi-
tional empirical content. Words like ‘meaning’, ‘analytic’, and ‘synonymous’ will have
no place in a properly scientific theory of language. In the end, what’s wrong with talk
of meaning is that it has no content.
So much by way of introduction. I now consider the two main Quine texts in more
detail.

1.1 “Two Dogmas”

To lead up to Quine’s arguments against meaning in TD, let me say something specifi-
cally about analyticity. Quine thinks of ‘analytic’ as being representative of the whole
family of common-sense meaning concepts: It is straightforwardly interdefinable with
‘synonymous’ etc. But it does have a somewhat technical sound, so some preliminary
explanation is in order.
For nonphilosophers, the purpose of language is to express our thoughts. Meaning
is, on this view, a normative relation between sentences and thoughts: The mean-
ings of the words in a statement determine whether or not you should assent to it,
given your beliefs. Sentence content is inherited from belief content. Beliefs stand in
relations of containment: The belief that p&q, for example, contains the belief that
p—there is no such state of mind as believing p&q but not believing p. Derivatively,
there are relations of meaning containment between sentences: For example the con-
tent of the sentence p&q includes that of p—if you assent to p&q while dissenting
from p, this cannot correctly express any genuine state of mind but must involve a
misuse of language. Similarly, there seem to be sentences which cannot be assented to
without misuse of language, for example ‘p&q but not p’. And others which cannot be
dissented from without misuse of language, for example ‘not-(p&q but not p)’. Assent
to such sentences is required merely by the meanings of words, not by constraints
of rationality or whatever pertaining to the underlying nonlinguistic states of mind.
These are the “analytic” sentences. Let us say, to preserve connections, that p&q
“analytically entails” p.
(Common sense is undecided, however, as to just how far these analyticities run,
because it is undecided about the relevant psychology. You can’t believe p&q without
believing p. But can you believe that something is red without believing that it is not
green? Can you believe that all men are mortal and Socrates is a man without believing
that Socrates is mortal? Can you believe that everybody loves a lover and John loves
Mary without believing that everybody loves Mary? (NB lover z lovee.) If we interpret
empiricist talk of analyticity in this natural psychological way, its claim against rational-
ism would be that all these entailments are analytic; all apparent use of deductive reason is
merely verbal; deductive reason provides no new knowledge, it is just the discovery of new
ways of expressing old beliefs; all a priori “knowledge” is analytic. But this seems to go way

297
 M I CH A EL M CD E R MO T T

beyond common sense. The rationalism versus empiricism issue is, however, irrelevant for
our purposes: both sides agree that talk of meaning and analyticity makes sense.)
I have been trying to explain ‘analytic’ on the assumption that people have beliefs
(nonlinguistic). For most of the twentieth century philosophers were pretty much agreed,
however, that common-sense mentalism cannot be taken for granted for purposes of
philosophical clarification. Mental content was held to be more problematic than the
meaning or content of sentences or theories. The most popular method of philosophical
clarification was through the analysis of language, rather than of thought. And the most
popular approach to language involved “truth conditions.” On this approach meaning
is a relation between words and the world (rather than beliefs). Analyticity is vacuity of
the nonlinguistic conditions of truth (rather than assent). The usual formula is: analytic
= true in virtue of meanings alone.
There is a rather obvious difficulty: ‘everything is self-identical’, for example, is ana-
lytic if anything is, but can we deny that part of the reason it is true is that everything is
self-identical? Would it be true if everything was not self-identical?
This truth-conditional definition of ‘analytic’ is Quine’s initial target in TD. He does
not criticize it directly, perhaps relying on the objection above. He considers a variety of
attempts to improve on it. But we might wonder whether this is too quick: could ‘true in
virtue of meanings alone’ actually be made to work? Let me digress to consider a couple
of possible strategies.
(i) The ontologically committed approach. The meaning relation between words
and things generates an association between sentences (some sentences) and non-lin-
guistic “facts”. Whether a sentence is true is normally a matter of whether there is a
corresponding fact. But certain sentences are true without need of a corresponding fact:
These are the analytic sentences. If we had a good theory of facts, and it said that there
is no such fact as the fact that everything is self-identical, that would explain why ‘Eve-
rything is self-identical’ is analytic. Unfortunately, there appears to be no good theory of
facts—not by empiricist standards, anyway. You could try saying that there is a fact that
p iff p is contingently true, or knowable a posteriori, but this would make ‘analytic’ trivi-
ally equivalent to ‘necessary’, or to ‘a priori’, thus undermining the empiricist’s claims
against rationalism.
(ii) The ontologically uncommitted approach (Davidson). The idea here is that ‘true
(in L)’ can do by itself as the central concept of a theory of meaning for L. The theory
will have theorems like:

(1) ‘Schnee ist weiss’ is true (in G) iff snow is white.

Since our total theory will also contain

(2) Snow is white

and

(3) Grass is green,

and since its ‘iff’ will presumably be the material biconditional, it will also contain

(4) ‘Schnee ist weiss’ is true (in G) iff grass is green.

298
 SKEPTI CI SM A BO UT ME A N I N G

But it is plausible to suppose that within our total theory we may distinguish in some
clear and principled way a subtheory, ‘the semantics of G’, which excludes (2), (3), and
(4). Example (1) is thus a “truth of meaning”; (4) is not. (Quine, somewhat similarly,
has no qualms about distinguishing a subtheory called ‘logic’. Or we may distinguish
‘biology’ or ‘genetics’. This is not philosophically contentious. There is no implication
that the truths so singled out are true in some special way.)
Many truths of meaning are biconditionals. But perhaps some unconditional truths
about truth in G, such as (5), will also be theorems of the semantics of G.

(5) ‘Schnee ist weiss oder Schnee ist nicht weiss’ is true (in G).

Would that make ‘Schnee ist weiss oder Schnee ist nicht weiss’ “true in virtue of mean-
ings alone”, i.e. analytic?
The trouble is that any subtheory presumably has as theorems all the logical truths
expressible in it. So this approach too would trivialize the contentious claim that the
truths of logic are all analytic.
Returning now to TD: Quine considers a variety of attempts to improve on ‘true by
meanings alone’.
The first of these is Frege’s definition—S is analytic iff it is either a logical truth or
can be turned into a logical truth by replacing words by synonymous words or phrases.
Once again, the obvious objection is that the definition trivializes the claim that the
truths of logic are all analytic.
Quine, however, has a different (but compatible) objection: the definition is circu-
lar. For ‘synonymous’ is “no less in need of clarification” than ‘analytic’. Quine then
considers certain attempts to define ‘synonymous’. In particular he considers the sug-
gestion that ‘x is synonymous with y (in L)’ means that x and y are intersubstitutable
salva veritate. Quine observes that this suggestion will only work if L contains a specially
tailored notion of “necessity” according to which ‘necessarily p’ is true iff p is analytic.
The circularity is then obvious.
In general, the allegation of circularity is Quine’s main weapon against the various
definitions of ‘analytic’, ‘synonymous’ etc. considered in §§1–4. Since Quine does not
tell us which terms he would not regard as “no less in need of clarification,” this part of
TD is not very impressive.
Section 5 introduces several new ideas: verificationism; a verificationist definition of
‘analytic’; holism.
Verificationism, in its classical form, says that the way the meaning of the words in
a statement determines its content is by determining that certain experiences would
confirm it and certain other experiences would disconfirm it. An analytic statement is
one whose meaning assigns zero content. (As opposed to “nonsense,” where no definite
content, zero or nonzero, is assigned.) Depending on how you understand ‘confirm’, this
seems to imply that an analytic statement is one which (i) no experience would confirm,
or (ii) every experience would confirm. If you understand confirmation as something
like increase in probability, (i) would be the natural choice, since an analytic statement
is presumably one whose meaning gives it maximum probability. A popular alternative
was to take disconfirmation as basic, and defined in terms of analytic entailment: e dis-
confirms h iff h entails not-e; e would then confirm h if it did not disconfirm it. On this
alternative (ii) would be more natural. (Here e is a statement reporting the occurrence
of an experience.) Quine in TD seems to run the two together: an analytic statement

299
 M I CH A EL M CD E R MO T T

would be one which is “vacuously confirmed.” He takes this as the verificationist defini-
tion of ‘analytic’.
A natural objection to this definition would be that ‘confirm’, in the required sense,
is no less in need of clarification than ‘analytic’.
But Quine has a different objection: holism. In the ordinary sense of ‘confirm’, and
presumably in any refined sense needed for verificationism, statements are not generally
capable of confirmation; only theories are.
Holism is certainly a plausible and important thesis in its own right, but for our pur-
poses the question is its relevance to the concept of meaning. Quine seems to think
that it is crucial: since verificationism is basically correct, ‘analytic’ can only make
sense if the verificationist definition can be accepted; but the definition fails on holistic
verificationism.
I don’t think this argument is any good. For holism does not prevent us from saying
that the factual content of theories is determined by the meanings of their words. But if
so much is granted, a notion of synonymy for words and sentences seems to be definable:
x is synonymous with y iff replacement of x by y in any theory formulated in language
L yields a theory with the same “empirical significance.” And then S is analytic iff it is
synonymous with ‘Pigs are pigs’. Quine’s target verificationist definition of ‘analytic’ is
no good, certainly; but there is a holistic verificationist definition which the objection
does not touch.
We should not leave TD without considering the paragraph of §6 which is most
often quoted (including by Quine himself) as a summary of the key argument against
analyticity. “Any sentence can be held true come what may,” says Quine, and “no state-
ment is immune to revision.” Well, certainly we cannot define ‘analytic’ as ‘immune to
revision’—no defender of analyticity denies that you can deny an analytic sentence if
you change its meaning. And certainly we cannot define ‘analytic’ as ‘cannot be revised
without change of meaning’—as a definition, that would be circular. But Quine seems
to be suggesting something stronger: this description is not even true of any non-trivial
class of “analytic” sentences—any sentence can be denied without change of meaning.
This stronger suggestion, however, seems to be quite without support.

1.2 Word and Object

The treatment of linguistic meaning in Chapter 2 of WO is generally regarded as more
radical than that in TD, and inferior. I disagree about “more radical.” The conclusion in
TD was utterly radical—all common-sense talk of meaning is nonsense. (‘Analytic’ was
meant to be representative, not special.) As to “inferior,” this perception has consider-
able justification. The central thesis of Chapter 2 is the “Indeterminacy of Translation,”
and I agree that this is a mess. It is not clear what the thesis is, nor what the arguments
for it are supposed to be. (There is one reasonably clear and plausible argument, but only
for a weaker thesis which came to be known as the Inscrutability of Reference.) Quine
tried to do better later, but without success (on what it now a widely shared view). My
advice is, forget the IT.
But there remains much in Chapter 2 that is not to be ignored. In particular, Quine’s
treatment of ‘analytic’ can easily be separated from the discussion of the IT. The differ-
ence from the treatment of ‘analytic’ in TD is striking. First, Quine’s basic assumption
in WO is behaviorism: he faces directly the question “Is there any empirical content
in talk about meaning?”; and he assumes that, if there is, it consists of predictions of

300
 SKEPTI CI SM A BO UT ME A N I N G

observable behavior: namely the verbal behavior of a community of language users.

Behaviorism didn’t get a look-in in TD. Second, holism takes a back seat. In fact, Quine
even seems to assume, contrary to holism, that the challenge for the defender of mean-
ing is to reduce every individual statement about meaning to a prediction of behavior.
The argument against analyticity is that the best behavioral reduction of ‘S is ana-
lytic’ would be ‘S is socially stimulus analytic’, i.e. ‘S would be assented to, under any
conditions of current stimulation, by any speaker of the language’; but this is not close
enough, since ‘There have been black dogs’ is socially stimulus analytic, but no defender
of the ordinary notion would agree that it is analytic.
Quine realizes that a defender of the intuitive notion would protest that there is a
better behavioral definition. An analytic sentence is, intuitively, one that we all assent
to as a result of shared linguistic habits, not as a result of a shared belief. But Quine’s behav-
iorism prohibits any reference to alleged mental causes of verbal behavior: “All the
objective data that [the linguist] has to go on are the forces that he sees impinging on
the native’s surfaces and the observable behavior, vocal and otherwise, of the native.”
(28) In general, the WO objection to intuitive talk of meaning is that it relies on an
unreduced mentalism.
Unreduced? Could we hope to show that talk of belief etc. is, on the contrary, reduc-
ible to talk of behavior? Quine’s reply is that the only plausible analysis would be along
the lines of Carnap—‘John believes that s’ = ‘John would assent to a sentence of his
language synonymous with s’. It seems that we can only reduce meaning to behavior if
we take mental states for granted; but we can only reduce mental states to behavior if
we take meaning for granted; so neither meaning nor mind can be reduced to behavior.
This is the “intentional circle.” Given his behaviorism, Quine’s conclusion is that there
is no content in talk of meaning or in talk of mind.
This seems to me to be a more interesting argument against meaning than anything
in TD. Of course it falls to the ground if you think the empiricist challenge to mind and
meaning can be met by appeal to “common sense”. Short of that, you could try to get
out of the intentional circle by showing that talk of meaning has behavioral content
without relying on anything mental; many have tried to do this. Or you could try to
establish a behavioral foundation for mind talk and then use that to explain meaning;
this strategy also has had occasional supporters. Or, finally, you could hold that there is
no order of priority (Davidson). As things stand, it is not obvious which of these three
strategies will ultimately be successful—if any. But certainly the prospects of success will
be better if we take holism seriously: ‘meaning’, ‘analytic’, ‘synonymous’ etc., or perhaps
‘belief’, ‘desire’ etc., or perhaps all of these together, may be theoretical terms, in a theory
with real empirical (i.e., behavioral) content—reduction is not necessary.

2 Kripke’s Wittgenstein
KW’s scepticism (Kripke 1982) focuses on the expression ‘. . . means . . . by . . .’: he
argues that it “makes no sense.”
From a common-sense standpoint, this form of words is not unambiguous. Suppose
you call my performance “meretricious”; I say “I presume you meant ‘meritorious’.” I
might be suggesting that you meant (= intended) to say ‘meritorious’ (but your tongue
slipped); this use of ‘meant’ is clearly irrelevant. Or I might be suggesting that you
thought that ‘meretricious’ means ‘meritorious’. This is not KW’s target, either. His
target is a supposed use of ‘. . . means . . . by . . .’ which is not thus parasitic on the use

301
 M I CH A EL M CD E R MO T T

of ‘means’ in relation to a shared public language. His target is a concept of meaning

applicable to a “private language” as well as to a public language.
KW’s argument is that no fact can possibly do what meaning does (according to com-
mon sense). It proceeds in two main stages.

2.1 Stage 1
Common sense holds (says KW) that meaning something by a symbol is a state which
directs the speaker how to use that symbol. For instance meaning addition by ‘+’ is a state
which directs the speaker to assent to ‘68+57=125’; meaning quaddition by ‘+’ is a state
which directs the speaker to assent to ‘68+57=5’. (Quaddition, standard symbol ‘†’, is
the function which differs from addition only in that its value is 5 for arguments greater
than any relevant to the speaker’s previous linguistic behavior.) What might constitute
such a state?
The two leading candidates would appear to be

(i) A rule or intention to conform to certain remembered verbal instructions on how

to use ‘+’;
(ii) A rule or intention to “go on in the same way” as in certain remembered examples
of the use of ‘+’.

But suggestion (i) is unsatisfactory because it just throws the problem back onto the
words in the instructions: Unless there is a determinate fact that the speaker means
such-and-such by them, following the instructions won’t require any determinate ver-
bal behavior. And suggestion (ii) is unsatisfactory because such a rule or intention is
similarly indeterminate: Its content depends on the speaker’s individual sense of
analogy.
The argument so far seems to be very weak. Even without disputing anything substan-
tial in KW’s assumptions, it is easy to suggest a better candidate: meaning addition is a
complex state of mind whose components are intention (ii) and the usual sense of anal-
ogy. Or, better, we can get rid of the implausible appeal to remembered instructions or
examples: Someone means addition iff his rule or intention is to assent to ‘n1 + n2=n3’ iff
n3 is the sum of n1 and n2. Or why not say, even more simply, that the difference between
someone who means addition by ‘+’ and someone who means quaddition is that only
the former means addition? Why is that not an acceptable description of a genuine men-
tal state as it stands? KW emphasizes that “there are no limitations . . . on the facts that
may be cited to answer the sceptic.” (14)

2.2 Stage 2
KW argues, now, that no fact can constitute meaning addition by ‘+’ because mean-
ing stands in a normative relation to verbal behavior. If I mean addition, then I speak
wrongly if I assent to ‘68+57=5’. But it seems that no fact can justify, or make wrong, a
bit of behavior. (No ‘ought’ from an ‘is’.)
KW seems to accept that if we could find a rule or intention with a suitable content, it
would have the required normativity. But this just shows that the real puzzle is how any
fact could constitute a rule or intention.

302
 SKEPTI CI SM A BO UT ME A N I N G

I think this argument may be touching on something interesting and important.

But it is not much of an argument for meaning skepticism as it stands. Not all uses of
‘wrong’ imply genuine normativity. Perhaps the clearest case is simple factual error.
If John says the cat is on the mat and I say he’s right and you say he’s wrong, we just
disagree whether the cat is on the mat. The question is, do rules and intentions have
genuine normativity? I think we need at this point to distinguish between rules and
intentions. Granted, if I intend to give you a dollar, and accidentally give you two, I
give you the wrong amount. But this is not a good argument for the nonfactuality of
intentions. It just shows that, in a suitable context, ‘wrong’ can mean ‘unintended’.
But perhaps rules, and in particular rules of meaning, do have genuine normativity.
However, KW appears to be completely uninterested in the difference between rules and
intentions.
You can intentionally disobey a rule. Perhaps this is the real reason that a “private lan-
guage” (even one used for talking about publicly observable things) is impossible: The
words of a personal shorthand cannot have meaning, in the ordinary sense, because there
is no difference between using a word contrary to its meaning and changing its meaning.
Perhaps, in general, there cannot be a “private rule” about anything. If I “make it a rule”
to always brush my teeth after meals, and one day deliberately leave them unbrushed,
have I broken the rule or abandoned it (perhaps replacing it with a less stringent rule)? If
there is no difference, it would seem that it was not really a rule, in the ordinary sense,
but merely an intention.

2.3 Skeptical Solution?

KW’s meaning skepticism consists in the claim that there is no such fact as meaning
addition—there is nothing which could make ‘Jones means addition by ‘+’ ’ true, while
preserving its normative implications. KW proposes, however, a “skeptical solution”
to this “paradox.” He proposes an alternative general picture of language. We should
understand language by asking, not for truth conditions, but for assertability conditions.
Under what conditions is it justifiable to assert p? What is the utility of having a sentence
with such assertability conditions in our language? And he claims that, on this approach
to language, the target locution does have meaning—as long as the speaker referred to is
regarded as part of a community. ‘Jones means addition by ‘+’ ’ can be justifiably asserted
by Smith if Smith believes that Jones’s use of ‘+’ resembles his own. Similarly ‘Jones
means quaddition by ‘+’ ’ can be justifiably asserted by Smith if Smith believes that
Jones’s use of ‘+’ resembles his own use of ‘†’. The utility of a sentence with such assert-
ability conditions is that if Robinson hears Smith’s remark, he will be able to infer that
Jones’s future use of ‘+’ will resemble his (Robinson’s) use of ‘+’ or ‘†’; the possibility of
this inference depends on community agreement—Robinson must be able to assume
that Smith’s usage resembles his own.
It is hard to see why KW calls this account of ‘. . . means . . . by . . .’ a “skeptical
solution.” For since the replacement of truth conditions by assertability conditions is
offered as a general picture of language, applying equally to such paradigmatically factual
sentences as ‘The cat is on the mat’, there seems to be nothing skeptical about it. And
since it does nothing to show how ‘Jones means addition by ‘+’’ can have normative
implications (when ‘Jones’s future use of ‘+’ will resemble mine’ does not), it does not
seem to be a solution to the paradox.

303
 M I CH A EL M CD E R MO T T

References
Kripke, S. (1982) Wittgenstein on Rules and Private Languages, Cambridge: Harvard University Press.
Quine, W.V. (1960) Word and Object, Cambridge, MA: MIT Press.
Quine, W.V. (1961) “Two Dogmas of Empiricism” in From a logical Point of View, New York and Evanston:
Harper and Row. (First published 1951.)

304
 Section III

FO U N DAT I O N S O F
SEM A N T I C S
 3.1
NAM ES
Barbara Abbott

John Stuart Mill regarded all definite noun phrases (NPs) as names, giving the place
which the wisdom or policy of antiquity had destined for the residence of the Abyssinian princes
as an example of a complex name (1843: 16). However, nowadays most people take
the term “names” as equivalent to “proper names”—just NPs like Madonna and Barack
Obama. Proper names have caused a lot of headaches for semanticians and philosophers
of language, prompting David Kaplan to remark “if it weren’t for the problem of how to
get the kids to come in for dinner, I’d be inclined to just junk them” (1978: 224). This
chapter will try to make clear what the main problem with proper names is, what solu-
tions have been proposed, and why none of those solutions has been universally adopted
as the correct one.

The Problem
The main problem with proper names concerns their contribution to the meaning of
sentences (or utterances) in which they occur. (There are other problems, many of
them related to this one, but we won’t have space to go into them here.) Saying more
about this main problem requires a bit of background. We’ll start by looking at a differ-
ent kind of definite NP—definite descriptions—which we will compare and contrast
with proper names. Definite descriptions are NPs which (in English) begin with the
determiner the. They are linked to their referents by expressing properties of those ref-
erents: the star of Evita refers to Madonna because she has the property of having starred
in the movie Evita, while the mother of Lourdes refers to Madonna because she has the
property of being the mother of Lourdes. Even though those two definite descriptions
(the star of Evita and the mother of Lourdes) refer to the same person (Madonna), they
do it by expressing different properties that Madonna has. These referent-determining
properties are called “senses,” after Gottlob Frege (1892).
The senses of definite descriptions obviously contribute to the meanings of sentences
in which they occur—the propositions those sentences express. They do not, however,
usually matter to the truth value of simple sentences. (1a) and (1b) below differ only in
that one uses the NP the star of Evita and the other uses the mother of Lourdes.

(1) a. The star of Evita will be at the party.

b. The mother of Lourdes will be at the party.

These two sentences obviously mean different things—they express different proposi-
tions. But it is clear that they have the same truth value; if one is true, then the other
 BA RBA RA A B B O T T

must be also. Indeed this is exactly what we would expect, since how we refer to some
entity shouldn’t matter to the truth of what we say about it. This generalization is the
law of substitutivity (also called “Leibniz’s Law”), indicating that we would expect to
be able to substitute one NP for a coreferential one without affecting the truth value of
the containing sentence.
However, there are significant groups of sentences in which one can’t exchange one
definite description for another with the same referent and be sure of maintaining the
same truth value. We focus here on one such group, which consists of sentences about
mental attitudes such as knowledge, belief, desire, hope, anticipation, etc. So, it seems
that (2a) might be true while (2b) is not.

(2) a. Frank knows that the star of Evita will be at the party.
b. Frank knows that the mother of Lourdes will be at the party.

The standard view is that belief, knowledge, desire, etc. are attitudes towards proposi-
tions—hence the term “propositional attitudes” for them. In the case of (2) the proposi-
tions in question are the ones expressed by the embedded sentences following knows that
(these are the same as the sentences in (1)). And given that the properties expressed by
a definite description contribute to the proposition expressed by the sentences it occurs
in, it is not surprising that we can have a change of truth value when we exchange
coreferential definite descriptions. The proposition which Frank is said to know in (2a)
is different from the one he is being said to know in (2b).
We turn now to proper names. Unlike definite descriptions, proper names do not
seem to achieve their reference by expressing properties of the things they denote. In
other words they seem to be nondescriptional, lacking a Fregean sense. This is true even
for names that originally expressed properties. Mill pointed out that, although the Eng-
lish town of Dartmouth originally got its name because it lay at the mouth of the Dart
River, if an earthquake should cause the river to change its course the town would still
be called by that name (Mill 1843: 20).
If proper names do not have any associated descriptional sense, then exchanging
coreferential ones in a sentence should not change the proposition being expressed. The
two sentences in (3) should express the same proposition.

(3) a. Mark Twain was a writer of humor.

b. Samuel Clemens was a writer of humor.

And we expect to be able to substitute coreferential proper names for each other in propo-
sitional attitude sentences without affecting the truth value of those sentences. However,
it’s not clear that this is the case; it seems that (4a) and (4b) could differ in truth value.

(4) a. Mary knows that Mark Twain was a writer of humor.

b. Mary knows that Samuel Clemens was a writer of humor.

So our problem is to explain this apparent difference.

The Solutions
The problem sketched above has been a worry to philosophers of language for a long
time. First, below, we outline some early proposals to solve it. We then review a seminal

308
 NAMES

work of Saul Kripke’s which changed the landscape, and by and large forced people to
look in other directions for a solution. Following that, we take up some more recent pro-
posals, separated into three groups. We’ll pause after the first group to look at another
of Kripke’s contributions, which presented new challenges. We finish with the final two
groups of approaches. We will not be able here to review all the analyses which have
been put forward; important contributions from Mark Richard (1990), Graeme Forbes
(1990), and others have had to be omitted. Nevertheless it is hoped that the reader will
get some idea of the range of possibilities.

Early Solutions
Both Frege and Bertrand Russell concluded from data like those given above that (con-
trary to our intuitions and Mill’s argument about Dartmouth) proper names do express
properties of their referents, similar to those expressed by definite descriptions. Frege
suggested that the name Aristotle might express a property like being the student of Plato
who taught Alexander the Great (1892, 58 n.), and Russell suggested that the name
Bismarck might be equivalent to the first Chancellor of the German Empire (1918: 171).
Given this approach we might propose that Mark Twain means something like ‘the most
famous American humorist’, while Samuel Clemens expresses a different sense (although
it’s a little difficult to say what that one would be).
One problem with such a theory is that it does not explain the difference in impact
between examples such as those in (5).

(5) a. The student of Plato who taught Alexander the Great was a student of
Plato.
b. Aristotle was a student of Plato.

The sentence in (5a) sounds analytic—that is, true just in virtue of its meaning. How-
ever, the sentence in (5b) does not sound analytic. But if the name Aristotle expressed
the description suggested by Frege, then (5a) and (5b) should have exactly the same
meaning, and if one sounds analytic then the other should too.
Others have solved this problem by proposing a looser kind of descriptiveness for
proper names. This approach is sometimes called a “cluster” theory. On this kind of
view, we associate a number of descriptions with a proper name, not all of which actu-
ally need to hold of the denoted entity. John Searle is one follower of the cluster theory.
(Cf. also Wittgenstein 1953: §79.)

Suppose we ask the users of the name ‘Aristotle’ to state what they regard as
certain essential and established facts about him. Their answers would be a set
of uniquely referring descriptive statements. Now what I am arguing is that
the descriptive force of ‘This is Aristotle’ is to assert that a sufficient but so
far unspecified number of these statements are true of this object.
(Searle 1958: 171)

Ultimately Searle concluded that only one of the descriptions need hold true for a name
to refer. That being the case, we no longer have a problem with the examples in (5). On
Searle’s theory, the logical form of (5b) is something like (6).

309
 BA RBA RA A B B O T T

(6) The entity who had at least one of the following properties: X1, X2, . . ., Xn,
had property Xm, where 1d m d n.

And sentences of the form of (6) are not analytic.

Kripke (1972) on Proper Names

In a major series of lectures titled “Naming and necessity” (later published as a book),
Kripke (1972) gave a number of arguments against both single description theories and
cluster theories, and in favor of Mill’s conclusion that proper names are nondescrip-
tional. He pointed out that people can know and use names for individuals when they
don’t know any identifying properties of those individuals. I’ve heard of Anaximander,
for example, and can use his name to ask questions about him, but at present I don’t
know anything more about him than that he was an ancient Greek philosopher. Con-
trary to what the quote from Searle suggests, I could not give any uniquely referring
descriptive statements.
Kripke also pointed out that when we speak of alternative possibilities—different
possible worlds—definite descriptions differ from proper names in potentially referring
to different entities. In the actual world the NP the star of Evita denotes Madonna, but
in different circumstances the movie maker could have picked a different person for the
starring role, and in such a possible world the star of Evita would refer to that person and
not Madonna. Thus (7) is true.

(7) Madonna might not have been the star of Evita.

However, proper names seem to pick the same individual out of any possible world.
Because they don’t express a property of that entity, they have no way to change their
referent under different possible circumstances. Thus (8) does not seem to be true.

(8) Madonna might not have been Madonna.

(In thinking about alternative possible worlds, and how expressions of our language
apply to them, it is important to be clear that we are keeping our language fixed. The
purpose of these exercises is to become clearer on meaning in our language as we speak
it. Thus, it is not relevant that Madonna might not have been named that in a differ-
ent possible world, or that the word star might not have been used for leading movie
actors.)
Kripke called proper names “rigid designators,” meaning that (as we use them) they
refer to the same entity in all possible worlds (1972: 48). He proposed that instead of
being connected to their referents by expressing properties of those referents, proper
names are connected by a historical or causal chain—a chain of uses of the name going
back to the original baptism or dubbing.
These facts present a problem for both the single description theory and the cluster
theory of proper names. No matter which single description we pick, as long as it is a
description of the type suggested by Frege, Russell, or Searle, it will not hold of the same
entity in all possible worlds. (There are what we might call “necessary” descriptions,
like the sum of 5 and 7, which apply to the same entity (12, in this case) in all possi-
ble worlds. However contingent descriptions of the kind we’ve been considering—like

310
 NAMES

the star of Evita—are not of this type.) On the other hand as we’ve just seen, proper
names do pick the same entity out of all possible worlds. The sentences in (7) and (8)
made this contrast clear. Furthermore, like (7), (9) seems also to be true.

(9) Madonna might have had none of the properties commonly attributed to
her.

This is a problem for the cluster theory. If a proper name meant something like ‘the
entity with at least one of following commonly attributed properties: X1, . . ., Xn,’ then
(9) should sound contradictory, but it doesn’t.
Most people who write about this subject have found Kripke’s arguments convincing,
and have looked elsewhere for a solution to the problem presented by proper names in
propositional attitude sentences. Most of the proposed solutions fall into one of three
main categories: metalinguistic approaches, which involve the names themselves in one
way or another; so-called “hidden indexical” approaches, which introduce extra mate-
rial into propositional attitude sentences; and what I’ll call “bite the bullet” approaches,
which try to explain the problem away. We’ll start with a look at the metalinguistic
approaches, and then introduce some more data which is particularly difficult for such
approaches to account for, which will cause us to look at the other two kinds of poten-
tial solution.

Metalinguistic Approaches
There are three subcategories of metalinguistic approach to consider. The first involves
a different analysis of propositional attitude sentences, while the second and third are
new analyses of proper names.

Quotation
Several philosophers (e.g. Quine 1956; Davidson 1969) have suggested that proposi-
tional attitude verbs like know, believe, hope, etc. actually express relations to sentences
rather than propositions. On this quotational view our sentences in (4) above would be
analyzed as in (10).

(10) a. Mary knows-to-be-true “Mark Twain was a writer of humor.”

b. Mary knows-to-be-true “Samuel Clemens was a writer of humor.”

According to this analysis (4a) (= (10a)) attributes to Mary knowledge of the truth of
the sentence Mark Twain was a writer of humor, while (4b) (= (10b)) attributes to Mary
knowledge of the truth of a different sentence—Samuel Clemens was a writer of humor.
Since two different sentences are involved, it is not a problem that one of these could
be true and the other false.
One potential problem with quotational views is that they seem to suggest that any
holder of a propositional attitude must also know a language (although it has been
argued that this is not actually a consequence of such views). However we commonly
attribute knowledge, beliefs, and desires to languageless creatures such as cats, dogs,
chimpanzees, and prelinguistic children. Indeed, animals of all sorts are capable of per-
forming intricate series of tasks in highly organized ways to obtain just the things they

311
 BA RBA RA A B B O T T

need to survive, and it is difficult to imagine that they do this without a thought or a
desire in their head.

Causal description
We saw above that Kripke had proposed that names are linked with their referents via
a historical or causal chain of use (rather than a descriptive sense). A second group of
metalinguistic approaches incorporates this causal chain into a special kind of descrip-
tive meaning for a proper name. (See, for example, Devitt 1981, Searle 1983, Lewis
1997.) One potential problem for this view is that most people do not know anything
about the historical chain leading up to their use of a name. In most cases they will have
completely forgotten when and where they first heard it used. Yet they use these names
confidently, as though nothing were lacking in their knowledge of them. (See Abbott
2010: 117–18, and the works cited there, for further discussion.)

‘The bearer of N’
The third, and currently most popular, kind of metalinguistic approach holds that a
name N expresses a meaning something like ‘the entity named N’ or ‘the bearer of N.’
(See, e.g., Katz 1977, 2001; Bach 1987, 2002; Recanati 1993; Geurts 1997.) Kripke had
responded to an earlier proponent of this kind of approach (Kneale 1962) with several
arguments. For one thing, Kripke thought it was circular: “We ask, ‘To whom does
he refer by “Socrates”?’ And then the answer is given, “‘Well, he refers to the man to
whom he refers’” (1972: 70). And for another, Kripke pointed out that such a theory
would not make sense for other words—we would hardly want to have a dictionary with
entries like “horse: n., those things called ‘horses’” (1972: 69). A later paper of Kripke’s
presented another problem, perhaps even more serious. We turn now to that.

Kripke’s Puzzle
In “A puzzle about belief” (Kripke 1979), Kripke describes the following situation. Pierre
is a monolingual French boy who, based on picture books and postcards, forms a belief
which he expresses by uttering (11).

(11) Londres est jolie.

Unfortunately he is kidnapped, taken abroad, and forced into the slave trade in the very
worst part of London (I’ve filled in a few details here), where he learns English by the
direct method (that is, with no instruction). Now he is inclined to assert (12)

(12) London is not pretty.

although none of his attitudes have changed about the city he still calls Londres. Kripke
asks the simple question: Does Pierre, or does he not, believe that London is pretty?
Kripke gave another version of the puzzle which does not require two languages. Peter
has heard of Paderewski, the famous Polish statesman. He has also heard of someone
with the same name who was a celebrated pianist and composer, but believes this to
be a different person. He also thinks that people in politics are never musically inclined.

312
 NAMES

So the question is, does Peter, or does he not, believe that Paderewski had musical
talent?
One thing shown by these examples is that the problem with proper names in propo-
sitional attitude contexts is not, or not only, one of substitution. In the Pierre case we
have corresponding sentences in two different languages, and in the Paderewski case
there is just one sentence of English involved. But Kripke believes that the examples
show something much deeper, a fundamental problem with our attributions of proposi-
tional attitudes.

When we enter into the area exemplified by . . . Pierre, we enter into an area
where our normal practices of interpretation and attribution of belief are sub-
jected to the greatest possible strain, perhaps to the point of breakdown. So is
the notion of the content of someone’s assertion, the proposition it expresses.
(Kripke 1979: 269, italics in original)

It seems that in Kripke’s view, the simple questions posed about Pierre and Peter are
unanswerable.
What is the consequence of these examples for our examples in (4), repeated here?

(13) a. Mary knows that Mark Twain was a writer of humor.

b. Mary knows that Samuel Clemens was a writer of humor.

Kripke holds that the same considerations apply—that we cannot simply chalk up the
seeming truth of (13a) and falsity of (13b) to a failure of substitutivity. Instead he has
called into question the very possibility of attributing any propositional attitude involv-
ing a proper name.
Kripke’s puzzles seem to present a problem for metalinguistic theories of proper names
and/or propositional attitude contexts. Recall that those theories involve the name
itself—either directly (as in the quotational approaches), or because of the historical
chain that leads to its use (as in the causal description approaches), or in a description
of the form “the bearer of N.” But in Peter’s case we apparently have just a single name,
so it is not clear how any of these approaches can explain the difficulty in attributing to
him a belief about Paderewski. Which is not to say that there have not been responses—
the references cited in the preceding subsection which come after 1979 have all had to
address Kripke’s puzzle, and the interested reader should investigate these resources to
see what their authors have to say.
We turn now to the two remaining categories of response to the problem of proper
names in propositional attitude sentences.

Hidden Indexical Approaches

Generally speaking, hidden indexical approaches such as those of, for example, Crimmins
and Perry (1989), and Schiffer (1992), accept Kripke’s arguments that proper names
are nondescriptional, and hence do not add descriptional content to the propositions
expressed by sentences containing them. In order to solve the problem of proper names
in propositional attitude sentences, they regard such sentences as containing additional
constituents which are not overtly expressed—hence the adjective “hidden.” Further-
more, the exact content of these constituents may depend on the context in which the

313
 BA RBA RA A B B O T T

sentence is uttered—hence the word “indexical.” Let us try to get clear on what this
kind of approach involves.
The basic idea is first to recognize that people have conceptions of the things they
have beliefs and other attitudes about, and then to postulate that one or another of
those conceptions may be implicitly included in a propositional attitude report. These
conceptions are sometimes called “modes of presentation” (using a term which origi-
nated with Frege (1892) but in a slightly different way from the way Frege used it).
When we say, for instance, that Mary knows that Mark Twain was a famous humorist,
we implicitly invoke a mode by which Mark Twain is psychologically presented to Mary
and which would typically include facts about him such as his having written amusing
books and made amusing comments.
Furthermore these theorists believe that these additional constituents, the modes of
presentation, are part of the semantics of the utterance—they play a role in determin-
ing its truth or falsehood. Thus the logical form of a propositional attitude attribution
will involve an additional argument place, not represented in the surface form of the
sentence. We might represent the logical form of our examples about Mary, on this kind
of view, as in (14).

(14) a. Knows (Mary, that Mark Twain was a famous humorist, m1)
b. Knows (Mary, that Samuel Clemens was a famous humorist, m2)

In both (14a) and (14b) Mary is said to be in the knowledge relation to the same propo-
sition, i.e. the proposition that Mark Twain/Samuel Clemens was a famous humorist.
What differentiates (14a) from (14b) are the distinct modes of presentation m1 and m2,
and in order for these sentences to be true, the knowledge Mary is said to have must
be via the correct mode of presentation. Presumably in a context where (14b) strikes
us as false, it will be the case that the unexpressed mode of presentation is one which
does not include the stereotypical facts that we associate with Mark Twain, the famous
humorist.
In arguing for the existence of such unexpressed constituents, Crimmins and Perry
draw an analogy to sentences about the weather (1989: 699–700). One can’t say It’s
snowing and sure of saying something true simply because it’s snowing somewhere or
other. Instead, when we say It’s snowing there is implicit reference to a specific place
where the snow is said to be occurring—usually, but not always, the place where the
speaker is located. In order for the speaker to have said something true, it must be snow-
ing at that location.
Turning to Kripke’s puzzle, hidden indexical theorists will typically hold that a sen-
tence like (15)

(15) Peter believes that Paderewski had musical talent.

will express different things in different contexts, because the mode of presentation
which is implicitly being attributed to Peter will be different. In a context in which
Paderewski’s musical accomplishments are salient, the mode will be one which includes
that kind of thing, and (15) will be true. However, in a context where, perhaps, great
Eastern European political figures are under discussion, a different mode of presentation
will be invoked—one involving Peter’s beliefs about Paderewski as a statesman. In that
kind of context (15) expresses a different proposition, one which is false.

314
 NAMES

Nathan Salmon (1995) has raised a number of objections to the hidden indexical
type of approach. (See also Richard 1993.) One concerns the result just cited—that
(15) will be true on some occasions of its use and false on others. In its surface form,
(15) does not have any relevant indexicals, so to determine that it is on some occasions
true and on others false when none of the relevant facts about the world have changed
is highly counterintuitive. A more fundamental objection is that the modes of presenta-
tion invoked by the hidden indexical theorists should not be considered to be part of the
meaning expressed by propositional attitude sentences but should instead be attributed
to pragmatic considerations. They are so treated in the last kind of approach to be con-
sidered—what I’ve called “bite-the-bullet” approaches.

“Bite-the-Bullet” Approaches
The last kind of theory of proper names in propositional attitude contexts we’ll consider
has been promoted largely by Salmon (e.g. 1986, 1995), and Scott Soames (e.g. 2002).
(Cf. also Braun 1998.) This kind of approach accepts Kripke’s conclusions concern-
ing the nondescriptionality of proper names, and does not posit any additional hid-
den indexical content in propositional attitude ascriptions. What, then, does this kind
of approach have to say about our original sentences about Mary, repeated here once
again?

(16) a. Mary knows that Mark Twain was a writer of humor.

b. Mary knows that Samuel Clemens was a writer of humor.

Salmon and Soames acknowledge that holders of propositional attitudes will conceive
of the propositions involved in various ways—what we’ve been calling “modes of pres-
entation.” However, they hold that specific modes are not represented in any way in
a typical propositional attitude ascription such as those in (16). Instead, it is simply
the case that one holds a particular propositional attitude if there is at least one way in
which one both conceives of the proposition and has the relevant attitude toward it. If
Mary has a way of thinking of Mark Twain according to which she knows that he was a
writer of humor, then (16a) and (16b) are both true.
In order to account for our intuition that (16b) is false, Salmon and Soames point
out that, under the circumstances, the speaker of (16b) has presented its content in a
misleading way. Generally speaking, when one attributes an attitude to another, one
will choose a form of words which reflects the way the attitude holder grasps the propo-
sition in question, and addressees expect speakers to do that. By using the name Samuel
Clemens, the utterer of (16b) suggests that Mary is familiar with Mark Twain under this
name as well, but this suggestion is not an entailment of the utterance. In this way the
bite-the-bullet approaches attribute modes of presentation to pragmatic inference from
an utterance, and not to its semantic content as in the hidden indexical views.
When it comes to Peter and Paderewski, the bite-the-bullet approach says that (15)
(Peter believes that Paderewski had musical talent) is simply true. It is true because there
is at least one way Peter has of conceptualizing Paderewski under which he recognizes
the Pole as an accomplished musician. Under this approach (17) would also be deemed
true.

(17) Peter believes that Paderewski did not have musical talent.

315
 BA RBA RA A B B O T T

That is because Peter has another way of conceptualizing Paderewski under which the
Pole seems definitely not to have been a musician. This is not to saddle Peter with irra-
tionality, however, since two different guises are involved.

Concluding Remarks
It is hoped that the preceding survey has given the reader some idea of the range of
approaches to solving the problem of proper names in propositional attitude contexts.
The very existence of so many different analyses of proper names, and of propositional
attitude sentences, indicates the unfinished state of research in this area. In the back-
ground lies Kripke’s challenge concerning Peter and Paderewski: Is it possible to come
up with a clearly satisfactory analysis of proper names and propositional attitude sen-
tences that will give a definite response to his question? Or is he right, that such attri-
butions lie in an area “where our normal practices of interpretation and attribution of
belief are subjected to the greatest possible strain, perhaps to the point of breakdown”?
Speaking for myself, I incline toward the bite-the-bullet type of theory, but not every-
body would agree.

Related Topics
1.1 Extensions, Intensions, Character, and Beyond
1.13 Empty Names
2.3 Propositions
3.7 Descriptions
6.7 Propositional Attitudes.

References
Abbott, Barbara (2010) Reference, Oxford: Oxford University Press.
Bach, Kent (1987) Thought and Reference, Oxford: Oxford University Press.
—— (2002) “Giorgione was So-called Because of his Name,” Philosophical Perspectives 16: 73–103.
Braun, David (1998) Understanding Belief Reports. Philosophical Review 107: 555–595.
Crimmins, Mark and John Perry (1989) “The Prince and the Phone Booth,” Journal of Philosophy 86:
685–711.
Davidson, Donald (1969) “On Saying That,” in Donald Davidson and Jaakko Hintikka (eds.) Words and
Objections: Essays on the Work of W.V. Quine, Dordrecht: Reidel, 158–174.
Devitt, Michael (1981) Designation, New York: Columbia University Press.
Forbes, Graeme (1990) “The Indispensability of Sinn,” Philosophical Review 99: 535–563.
Frege, Gottlob (1892) “Über Sinn und Bedeutung,” Zeitschrift für Philosophie und Philosophische Kritik, 25–50.
Translated as “On Sense and Reference”, in Peter Geach and Max Black (eds.) Translations from the Philo-
sophical Writings of Gottlob Frege, Oxford: Blackwell, 56–78.
Geurts, Bart (1997) “Good News about the Description Theory of Names,” Journal of Semantics 14:
319–348.
Kaplan, David (1978) “Dthat,” in Peter Cole (ed.) Syntax and Semantics, vol. 9: Pragmatics, New York: Aca-
demic Press, 221–243.
Katz, Jerrold J. (1977) “A Proper Theory of Names,” Philosophical Studies 31: 1–80.
——(2001) “The End of Millianism: Multiple Bearers, Improper Names, and Compositional Meaning,”
Journal of Philosophy 98: 137–166.
Kneale, William (1962) “Modality De Dicto and De Re,” In Ernest Nagel, Patrick Suppes and Alfred Tarski
(eds.) Logic, Methodology and Philosophy of Science. Proceedings of the 1960 International Congress, Stanford:
Stanford University Press, 622–633.
Kripke, Saul (1972) “Naming and Necessity,” in Donald Davidson and Gilbert Harman (eds.) Semantics of

316
 NAMES

Natural Language, Dordrecht: Reidel, 253–355 and 763–769. Page references are to the 1980 monograph
reissue, Cambridge, MA: Harvard University Press.
——(1979) “A Puzzle about Belief,” in Avishai Margalit (ed.) Meaning and Use, Dordrecht: D. Reidel,
139–183.
Lewis, David (1997) “Naming the Colours,” Australasian Journal of Philosophy 75: 325–342.
Mill, John Stuart (1843) A System of Logic, Ratiocinative and Inductive, Being a Connected View of the Principles
of Evidence, and the Methods of Scientific Investigation, London: John W. Parker. Page references are to the
1891 edition, reissued 2002, Honolulu, HI: University Press of the Pacific.
Quine, W.V. (1956) “Quantifiers and Propositional Attitudes,” Journal of Philosophy 53, 177–187.
Recanati, François (1993) Direct Reference: From Language to Thought, Oxford: Blackwell.
Richard, Mark (1990) Propositional Attitudes, Cambridge: Cambridge University Press.
—— (1993) “Attitudes in Context,” Linguistics and Philosophy 16: 123–148.
Russell, Bertrand (1918) “Knowledge by Acquaintance and Knowledge by Description,” in Mysticism and
Logic, London: Longmans, Green. Page references are to the 2004 republication, Mineola, NY: Dover,
165–183.
Salmon, Nathan (1986) Frege’s Puzzle, Cambridge, MA: MIT Press.
——(1995) “Being of Two Minds: Belief with Doubt” Noûs 29: 1–20.
Schiffer, Stephen (1992) “Belief Ascription,” Journal of Philosophy 89: 499–521.
Searle, John R. (1958) “Proper Names,” Mind 67: 166–173.
——(1983) “Proper Names and Intentionality,” in Intentionality: An Essay in the Philosophy of Mind,
Cambridge: Cambridge University Press, 231–261.
Soames, Scott (2002) Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity, Oxford:
Oxford University Press.
Wittgenstein, Ludwig (1953) Philosophical Investigations, translated by G.E.M. Anscombe, Oxford: Basil
Blackwell.

317
 3.2
V E RBS
Zoltán Gendler Szabó

The earliest known Greek grammar by Dionysius Thrax distinguishes eight parts of
speech: noun, verb, participle, interjection, pronoun, preposition, adverb, and con-
junction. The list looks largely familiar to the modern eye, though there are certain
surprises. A number of categories that play a role in modern taxonomies—article,
auxiliary, classifier, complementizer, determiner, etc.—are missing. Some of the tra-
ditional categories appear too narrow: participles, prepositions, and pronouns are
typically viewed by contemporary linguists as subcategories of adjectives, adpositions,
and nouns, respectively. When it comes to interjections, there is a serious question
whether they should count as genuine words at all: greetings, expletives, fillers, and
the like are at best loosely integrated into sentences. But in the larger scheme, these
are quibbles: we continue to classify words roughly the same way as we did around
100 b.c. Given that, it is striking that there is no agreement about the underlying
principles of classification.
Among the usual categories, it’s verbs that probably have the best chance of being
instantiated in all human languages (Baker 2003: sections 2.20, 3.9, 4.6). Dionysius
identifies verbs by a morphological and a semantic criterion: They are words inflected
for person, number, and tense that signify actions or passions. This definition has been
reprinted thousands of times with slight variations. It works remarkably well in prac-
tice—as generations of schoolchildren can attest. But if we are interested in what verbs
are—as opposed to how to spot most of them in Ancient Greek—the traditional criteria
won’t do. Vietnamese, Thai, Mandarin Chinese, and a number of other languages have
no system of verbal inflections. And no matter how far one stretches the meanings of
“action” and “passion,” there are plenty of counterexamples to the claim that all verbs
signify one or the other: The Latin verbs existit [exist], quiescit [rest], friget [show cold],
albet [show white], or claret [show bright] are but a few mentioned already in the Port-
Royal Logic (Arnauld and Nicole 1662: 80). English lacks a morpheme like -et but such
morphemes do exist in other languages besides Latin; cf. the German grün+en and the
Hungarian zöld+ell [show green].
The problems with the traditional definition can perhaps be fixed. We can move
from the purely morphological criterion to a more syntactic one and say that verbs are
the only predicates in the lexicon. We can also sharpen the semantic criterion and
say that verbs are true of events, processes, and states but never of other sorts of enti-
ties. A defense of these characterizations is presented in sections 1 and 2 below. The
proposed definition—verb is a lexical category whose members can be predicated of
events, broadly construed—remains silent on the two features that have been widely
regarded as constituting the very essence of the verb. Aristotle says that “a verb is what
 VERB S

additionally signifies time” (De Interpretatione 16b6) and that “every statement-mak-
ing sentence must contain a verb” (De Interpretatione 17a9). Though neither of these
claims is strictly true, they contain important insights; these will be examined in sec-
tions 3 and 4.

1 Verbs Are Predicates

Officially, expressions belong to the same syntactic category just in case they can
be interchanged within other expressions salva beneformatione (without loss of well-
formedness) and they belong to the same semantic category just in case they can be
interchanged within other expressions salva significatione (without loss of meaningful-
ness); cf. Bar-Hillel 1953. But the official line is routinely ignored because it leads to
excessively fine-grained categories. Consider the bound morpheme -ly. It can be added
to hour, day, month, and year but not to second, minute, decade, and century. But few
would be willing to make a grammatical distinction on such feeble grounds. (It seems
that -ly can combine with nouns denoting medium-length intervals only.) Linguistics
is committed to the unity of the category of nouns, not only in the face of minor differ-
ences in what morphemes they combine with but despite massive differences in syntac-
tic behavior between count nouns and mass nouns, common nouns and proper nouns,
relational and nonrelational nouns, etc. The consensus is that words do different kinds
of things in language and we regard distribution as a mere clue for discovering what
these job-descriptions really are.
The prime division among words is between lexical and functional categories. The
latter were traditionally seen as contributing to the form, rather than the content of
sentences. Paradigmatic examples of functional words are the determiner every, the
conjunction or, and the complementizer that. Functional words are short and tend not
to be loaned; functional categories are closed and tend to be small. In addition, lexical
categories do and functional categories don’t participate in derivational morphology.
For example, nouns verbs, adjectives can be transformed into each other in a variety
of ways but not to adpositions. And there are no morphemes that can attach to adposi-
tions that would turn them into a noun, a verb, or an adjective. This arguably shows
that adposition is a functional category despite the clear spatial meaning many of its
members have. (This is a contested claim: both Chomsky 1970 and Jackendoff 1977
count adposition as a lexical category. The Appendix of Baker 2003 gives a detailed
argument against their view.)
There are apparent exceptions to the claim that functional categories fail to partici-
pate in derivational morphology. For example, in English many prepositions combine
with -ness: cf. betweenness, aboveness and some auxiliaries combine with un-: cf, undo,
unhave. But between and above have adjectival variants and have and do have verbal
ones. Thus it seems reasonable to postulate ambiguity in these cases: perhaps there are
disitinct words pronounced as /bı'twi:n/, /ԥ'b^v/, /du/, /hæv/.
Verb is clearly a lexical category; the question is what distinguishes it from other
lexical categories. My suggestion is simple: Verbs, unlike other lexical categories, are
predicates—expressions used to say something of something. Their predicative nature is
reflected in the fact that they are associated with a set of argument positions. Nouns and
adjectives can have argument positions too: mother and proud both have one marked by
the preposition of. What sets verbs apart for these nouns and adjectives is that they yield
a clause when their arguments are saturated.

319
 ZO LTÁ N G EN DL E R S Z A B Ó

Being a predicate is widely accepted as a necessary condition for being a verb, but it
is equally widely rejected as a sufficient condition. The near-consensus is that a great
many other expressions—in particular, common nouns and intersective adjectives—are
also predicates. Such expressions can certainly be used predicatively, but they clearly
have other uses as well. Common nouns can be used to refer (cf. dog in Dogs are animals)
and intersective adjectives can be used to modify (cf. large in Large dogs bark). In prin-
ciple, we could see either the predicative or the nonpredicative use as primary. By con-
trast, verbs have only predicative uses. A verb can come to refer only if it is transformed
into a noun (for example a gerund) and it can come to modify only if it is transformed
into an adjective (for example a participle). Thus, verbs must be counted as predicates,
but nouns and adjectives needn’t be. We get the tidiest picture if we assume that nouns
and adjectives get to be predicates only as a result of a syntactic transformation.
There is evidence that predicative occurrences of nouns and adjectives contain extra
structure. The first thing that comes to mind in this connection is that simple declara-
tive sentences involving nominal and adjectival predicates require (at least in most
languages) a verbal copula. But this is not conclusive: in small clauses (for example the
embedded clauses of I regard the senator a crook and You consider the senator lucky) there
is no overt verb linking the subject with a nominal or adjectival predicate. A better
reason to believe that nonverbal predicates are more complex than verbal ones comes
from coordination:

(1) We consider the senator lucky and a crook.

(1’) We consider the senator [PREDP [PR‡][AP lucky]] and [PREDP [PR‡][NP a crook]]

Since the adjectival and nominal predicates can be conjoined they must be of matching
categories. This is guaranteed if we assume that when adjectival or nominal phrases are
used predicatively they are embedded into a predicate phrase. The idea can be imple-
mented by positing an empty predication head PR taking an AP or NP as complement.
The subject of predication may start in the specifier position and be subsequently raised
out of the PREDP (Bowers 1993, 2001). The fact that verbs do not coordinate with
nonverbal predicates suggests that they do not need such an empty element to become
predicates (Baker 2003: 78):

(2) *His advisors made the senator win and lucky.

(2’) *His advisors made the senator [VP win]] and [PREDP [PR‡][AP lucky]]
(3) *His advisors made the senator win and a crook.
(3’) *His advisors made the senator [VP win]] and [PREDP [PR‡][NP a crook]]

It has been suggested that PR is lexically realized in English by the as particle in We

regard the senator as lucky and We regard the senator as a crook (Bowers 1993). Many other
languages have particles that arguably realize PR: 3a in Russian, som in Norwegian,
als in German, yn in Welsh, ina in Irish, -kye in Korean, etc. Some of these particles
take only APs or NPs. Edo apparently has distinct particles for constructing nominal
and adverbial predicates: yé performs the former function, rè the latter (Baker 2003:
40). Overt PR is the norm for languages throughout Africa and is also present in many
languages in North and South America, Asia, and Oceania. Many of these are pro-
nounced even when nominal or adjectival predication occurs unembedded (Stassen

320
 VERB S

1997: 77–91). If nominal and adjectival phrases are of different semantic type, this is
certainly not unexpected.
The evidence for the claim that unlike nouns and adjectives, verbs are lexical predi-
cates is substantial. But it is important to keep in mind that this thesis does not auto-
matically yield a semantic characterization of verbs. Predicate is the syntactic category
of expressions that have a number of arguments whose joint saturation yields a clause.
If we assume that all predicates have the same sort of semantic value—for example that
they all denote functions to truth-values—we have something that all verbs have in
common semantically. Bowers 1993 does make this assumption: he follows Chierchia
and Turner 1988 in interpreting PR as a type-shifter mapping properties (conceived of
as individuals) to propositional functions.
But the assumption is far from innocent. Raising verbs (for example seem) and
weather verbs (e.g. rain) take expletive subjects as arguments. If we commit ourselves
to the idea that these denote functions we have to say what their domains are, which
in turn requires assignment of semantic value to the expletive. Perhaps when it seems
to Jill that Jack left then that he left seems to Jill and perhaps when it rains then the rain
rains, but these proposals are rather fanciful. The typical assumption is that expletives
are semantically vacuous. Based on these sorts of considerations one may be tempted
to draw the conclusion that “being a verb is fundamentally a syntactic matter” (Baker
2003: 27)—seem and rain require a syntactic argument but not a semantic one. In the
next section we will see whether the conclusion can be resisted.

2 Verbs Denote Properties of Events

Since the seminal work of Davidson 1967 many have argued that at least some verbs
are predicates of events. Motivated primarily by the inferential behavior of manner
adverbs, for (4) Davison proposed the logical form (4’):

(4) Jill greeted Jack.

(4’) e. greeted (Jill, Jack, e)

Followers of Davidson have extended the proposal in two ways. First, they have argued
that it would be implausible to think that action sentences differ fundamentally from
the rest. A sentence like (5) can also been seen as expressing quantification over under-
lying entities, except that those entities are states, rather than events.

(5) Jill slept at home.

(5’) s (slept(Jill, s)š at home(s))

Bach 1986 introduced the term eventuality for events, processes, and states. To avoid
the slightly awkward term semanticists tend to use the term event in such extended
sense. I will follow this practice. Second, neo-Davidsoneans have emphasized that the
arguments of the three-place predicate in (6) are not on a par. Jill and Jack are assigned
appropriate thematic roles by the verb but the event-argument is non-thematic. We can
think of the thematic roles as specifying ways in which Jill and Jack participate in the
underlying event—the former as its Agent, the latter as its Theme.

(4’’) e (greet(e)š Agent(e, Jill) šTheme(e, Jack)š Past(e))

321
 ZO LTÁ N G EN DL E R S Z A B Ó

The extensions of the original Davidsonean proposal have real advantages. In Mon-
tague semantics verbs are scattered all over the type-theoretic hierarchy: play would
be of type <e, t>, write of type <e,<e,t>>, give of type <e,<e<e,<e, t>>>, etc. In a neo-
Davidsonean semantics all verbs have the same type: They are one-place predicates of
events. Of course, verbs can be subcategorized by the number of arguments they take.
But the linguistically natural way to subcategorize them is more fine-grained than that.
For example, among intransitive verbs one may distinguish between unergative verbs
(for example run, speak, laugh) whose subject is external to the VP and unaccusative
verbs (e.g. arrive, collapse, rust) whose subject is not. This syntactic difference corre-
sponds to a difference in the sorts of thematic roles assigned. For example the Agent
role is always assigned to an external argument while the Theme role is always internal
(Gruber 1965, Chomsky 1981, Grimshaw 1990). The distinction is captured nicely in a
theory that incorporates thematic relations into logical form.
The neo-Davidsonean approach draws clear distinctions among different aspects of verb
meaning. First, there is the core content of the verb, which characterizes an underlying
event. In the case of (4’’), this is the predicate greet, true of all greeting events and nothing
else. Second, there is the argument structure of the verb, which determines the number
and type of participants in the event. In (4’’), this is represented by the second and third
conjuncts. Finally, there is the content contributed by the inflection on the verb. Here this
is the past tense morpheme represented in (4’’) by the last conjunct. Separating the three
components has been the chief reason Neo-Davidsonean theories have been successful
in dealing with difficult phenomena, such as perception reports, aspect, plurals, causa-
tives, focus, secondary predication, and qualification (Higginbotham 1983, Parsons 1990,
Schein 1993, Pietroski 2000, Herburger 2000, Rothstein 2002, Szabó 2003, respectively).
Once the different ways verbs contribute to logical form are distinguished, it is possi-
ble to capture cases where one or another of these three aspects of verb meaning is vacu-
ous or missing. Thus, representing nonfinite occurrences of verbs is unproblematic:

(6) Jill saw Jack run.

(6’) ee'(see(e)šExperiencer(e, Jill)šTheme(e, e')š run(e)š
Agent(e', Jack)š Past (e))
(7) Jack went to run.
(7’) ee'(go(e)š Agent(e, Jack)šGoal(e, e')šrun(e')šAgent(e', Jack)š
Past (e))
(I am ignoring the complication that (7) does not entail that Jack actually ran. Note
that the position occupied by run is not intensional in the usual sense: one can substi-
tute an extensionally equivalent predicate in this position salva veritate.) In (6) and (7)
the verbs in the main clauses bear tense marking but the verbs in the embedded clauses
do not, and so the seeing and going events are in the past, but the time of the running
events is unspecified. This is how the sentences manage to leave it open whether the
events they mention are simultaneous. Light verbs pose no problem either:

(8) Jill took a nap.

(8’) e (Agent(e, Jill)šTheme(e, a nap)šPast(e))
(9) Jill had a nap.
(9’) e (Experiencer(e, Jill)šTheme(e, a nap)šPast(e))

322
 VERB S

Take and have do not contribute a core component to the logical form of (8) and (9):
All these sentences say is that there is a past event whose Agent or Experiencer is Jill
and whose Theme is a nap. There is no need to think that the event is some sort of
taking or having. Presumably, it is a nap. (The semantics leaves it open whether (8) and
(9) describe the same events.)
In the same vein, if—as seems plausible—rain or seem fail to assign thematic roles to
their subjects, the logical form of (10) and (11) can simply be as follows:

(10) It rained.
(10’) e (raining (e)š Past(e))
(11) It seems to Jack that Jill left.
(11’) e (seeming (e)š Experiencer(e, Jack)šTheme(e, that Jill left)šPresent(e))

This removes the obstacle for a purely semantic characterization of verbs mentioned at
the end of the previous section. All verbs—including weather verbs and raising verbs—
are one-place predicates of events. What verbs denote then depends on the semantics
of predication one adopts. The simplest view is that verbs plurally denote events, for
example that run denotes each running event and sleep denotes each state of sleeping.
Light verbs would denote each event and would be semantically distinguished from one
another only in the thematic roles they assign to their arguments. Many light verbs have
homonyms that are regular correlates, which would of course denote events within a
restricted class. So, take in Jack took the letter from the desk denotes each taking event and
have in Jill has a coat denotes each state of having. Those who are less concerned about
positing entities in the semantics can assign a single type of denotation to all verbs. This
could be a function from events to truth-values, a set of events, a kind of events, or a
property of events. For convenience, I will adopt the last alternative.
Just because verbs describe events we cannot conclude that other words don’t do the
same. We certainly seem to be able to talk about events using nouns like kick, freeze, or
war. The suggestion that certain entities can only be denoted through certain types of
expressions isn’t well-motivated. Frege thought predicate denotations (he called them
concepts) cannot be denoted by nonpredicates, which led him to the conclusion that the
concept horse is not a concept (Frege 1892: 182–5). It is best to avoid such paradoxes by
acknowledging that nominal expressions are referentially unconstrained: anything can
(in principle) be referred to by some suitable noun in some suitable language.
What then distinguishes semantically a verb from the corresponding gerund? I suggest
that the distinction lies not in what they denote but in how they denote (Furth 1968,
Burge 2007). Verbs, being predicates, are predicated of things; nouns, being referring
expressions, refer to them. What greeting refers to are greetings and they are precisely
what greet is predicated of. This does not quite make greeting and greet synonyms—there
is more to meaning than denotation. But greet and [PREDP [PR‡] [NP greeting]] probably do
mean the same. Thus the sentence Jill greeted Jack can be correctly paraphrased as some
past greeting of Jack is by Jill.
Functionalist approaches to lexical categories postulate questionable ambiguities
(Baker 2003: 15–16). The sentences Chris hungers, Chris has hunger, and Chris is hungry
are near synonyms, which is not what one would expect if verbs, nouns, and adjectives
denote different sorts of things. The present proposal maintains the functionalist idea
that verbs have a distinctive sort of denotation but denies that this is true of nouns and

323
 ZO LTÁ N G EN DL E R S Z A B Ó

adjectives. Assuming has and is are auxiliaries without a core lexical meaning, all three
of Baker’s sentences describe the very same states—states of hunger Chris is in.

3 Verbs and Times

The definition of a verb advocated here establishes no link between verbs and time.
On the one hand, this is a good thing. There are languages where tense can associate
with nouns and adjectives (e.g. Turkish, Halkomelem) and languages where it can-
not associate with verbs (e.g. Yoruba, Nupe). On the other hand, these languages are
the exception (Stassen 1997). Moreover, the auxiliaries often used to express temporal
reference (such as the English will) typically modify a verb. So while temporality is not
essential to being a verb, the correlation is clearly more than accidental.
Givón 1984 argues that verbs typically denote transitory states of affairs, whereas
nouns and adjectives pick out more permanent ones. How one locates transitory states
of affairs in the flow of time makes a difference, which in turn explains the association
of verbs and times. The obvious problem with this proposal is that verbs correlate with
transience much less than they do with temporality in general. In addition, the explana-
tion presupposes that all lexical categories denote states of affairs—a rather tendentious
claim. Prima facie, neither blue nor dog picks out a state of affairs.
According to the event-based view advocated here, the association of verbs with
times is less direct than on Givón’s account. Verbs denote properties of events and
events are temporally structured. Thus the lexical meaning of a verb already contains
some temporal information. This is what makes it the natural locus for temporal refer-
ence in general. Nouns (and perhaps other words as well) can denote events and their
properties, but they don’t have to.
Lexical aspect (or Aktionsart) is a property of verbs whereby they represent the way
the eventualities they describe unfold in time. Lexical aspect gives rise to a natural and
cross-linguistically valid classification of verbs; the classic proposal about this goes back
to Vendler 1957. He distinguishes four categories: states, activities, achievements and
accomplishments. States (for example knowing) are extended eventualities whose tem-
poral parts could be (and are typically seen as) homogeneous. Activities (for example
walking) are also temporally extended, but they must have temporal parts distinguish-
able from one another. Unlike states and activities, achievements (for example finding)
can be instantaneous and are seen as culminations of processes. Accomplishments (for
example building) are complex events made up of an activity leading to some result-
ant state. Vendler presented a number of linguistic tests that classify verbs depending
on which of these event types they are true of; Rothstein 2004 has updated the tests
and proposed to organize verbs according to the distribution of two lexical features.
Some eventualities are telic (in the sense of having a natural culmination), others are
not; some eventualities have stages (in the sense of having discernible temporal parts),
others do not:

Table 3.2.1

Eventualities +Stage –Stage

+Telic accomplishment achievement
–Telic activity state

324
 VERB S

While there is no consensus about the types of lexical aspect (for a different, equally
influential classification see Steedman 1997), it does seem that verbs and only verbs are
subject to this sort of classification. This fact is explained by the view that verbs obliga-
torily denote events, and that we naturally see events as temporally structured. This
tendency also explains why verbs are natural loci for temporal reference.

4 Verbs and Statements

The idea that verbs signify affirmation has a long history. Here is how the Port Royal
Logic sums up the view and its prime motivation (Arnauld and Nicole 1662: 81):

Why is Petrus vivit [Peter lives] a proposition, and Petrus vivens [Peter living]
not one if you do not add “is” to it as in Petrus est vivens [Peter is living]? Only
because the affirmation contained in vivit was removed from it to make it the
participle vivens. From this it appears that the affirmation that does or does not
exist in a word is what makes it a verb or not.

The obvious objection is that embedded clauses contain verbs but are often not affirmed.
The point did not entirely escape the attention of the authors of the Port-Royal Logic;
they say that in All philosophers assure us that heavy things fall to earth of their own accord,
the speaker’s affirmation is signified only by the verb in the main clause (Arnauld and
Nicole 1662: 95). One might hope that discussion of disjunction and conditionals clears
up what Arnauld and Nicole think about the signification of embedded verbs. Alas, it
does not. Here one could say that the verb of the subordinate clause signifies the affir-
mation of all philosophers, but this won’t help the view in general. The embedded verb
fly in Nobody assures us that heavy objects fly away of their own accord certainly does not
signify anybody’s affirmation.
Since Frege, we have learned to distinguish between the force of an utterance from
the meaning of the expression uttered. The suggestion that some element within a sen-
tence encodes the speech act the sentence is used to perform ignores this distinction. In
principle we could perform all sorts of speech-acts uttering the very same sentence. And
one could even perform illocutionary acts uttering nonsentences (Stainton 2005). Just
imagine a case when it is in dispute whether Peter is alive and suddenly he enters the
room and one utters Peter living pointing at him. It is a bit quaint, but it does seem like
a bona fide statement. So, perhaps we should not say that verbs signify affirmation, but
rather that they signify what makes a linguistic expression capable of being true or false.
The idea is that without a verb, nothing can express a proposition. This was Russell’s
view in the Principles of Mathematics (Russell 1903: 50):

A proposition, in fact, is essentially a unity, and when analysis has destroyed

the unity, no enumeration of constituents will restore the proposition.
The verb, when used as a verb, embodies the unity of the proposition, and is
thus distinguishable from the verb considered as a term, though I do not know
how to give a clear account of the precise nature of the distinction.

The claim is that difference is a “verb considered as a term,” not a “verb used as a verb,”
and this is why the constituents of the proposition A differs from B—which would be A,
difference, and B—placed side by side do not reconstitute the proposition. But the real

325
 ZO LTÁ N G EN DL E R S Z A B Ó

issue is not unity—after all the difference of A from B is not a mere list, but an expres-
sion exhibiting genuine syntactic unity. It’s the kind of unity that matters: without a
verb, Russell contends, we cannot get anything that says something. Alas, some of the
examples mentioned before create a problem for this claim. The embedded clause in I
consider Jill happy manages to be truth-evaluable without containing a verb. (If you think
that without time determination no proposition can be expressed, take the embedded
clause of I consider Jill happy now.) The embedded clause contains the predicate [PREDP‡
[AP happy]], and this appears to be enough for being true or false. (If you think that I
consider Jill happy is but a truncated form of I consider Jill to be happy, consider instead I
made Jill happy, which is exceedingly unlikely to be the truncated form of I made Jill to
be happy.)
What distinguishes the verb live from the participle living and the verb differ from
the noun difference is that verbs are predicates while participles and nouns are not. The
syntactic difference presumably corresponds to a semantic one. As Russell correctly sug-
gests, the difference should not be thought of as a difference in denotation, but rather
as a difference in the manner in which these words denote. Predicates say something of
what they denote, other expressions don’t.
It is not clear why expression of a proposition requires a predicate—to say that only
predicates can say something is not much more than a restatement of the problem. It
does seem that there is a deep connection between truth and predication, one that any
serious philosophy of language must ultimately grapple with. But the connection of
verbs to this perplexing issue is at best indirect—verbs are the only lexical predicates in
natural languages, so the syntactically simplest sentences in these languages can contain
predicates only if they contain verbs.

References
Arnauld, A. and P. Nicole (1662) Logic, or the Art of Thinking, Cambridge: Cambridge University Press,
1996.
Bach, E. (1986) “The Algebra of Events,” Linguistics and Philosophy 9: 5–16.
Baker, M. (2003) Lexical Categories, Cambridge: Cambridge University Press.
Bar-Hillel, Y. (1953) “A Quasi-Arithmetical Notation for Syntactic Description,” Language 29: 47–58.
Bowers, J. (1993) “The Syntax of Predication,” Linguistic Inquiry 24: 591–656.
Bowers, J. (2001) “Predication,” in M. Baltin and C. Collins eds., The Handbook of Contemporary Syntactic
Theory, Oxford: Blackwell.
Burge, T. (2007) “Predication and Truth,” Journal of Philosophy 104: 580–608.
Chierchia, G. and R. Turner (1988) “Semantics and Property Theory,” Linguistics and Philosophy 11:
261–302.
Chomsky (1970) “Remarks on Nominalization”, in R. Jacobs and P. Rosenbaum eds., Readings in English
Transformational Grammar, Waltham, MA: Ginn.
Chomsky, N. (1981) Lectures on Government and Binding, Dordrecht: Foris.
Davidson, D. (1967) “The Logical Form of Action Sentences,” reprinted in Essays on Actions and Events,
Oxford: Oxford University Press, 2001.
Frege, G. (1892) “On Concept and Object,” reprinted in B. McGuinness ed., Collected Papers on Mathemat-
ics, Logic, and Philosophy, Oxford: Basil Blackwell, 1984.
Furth, M. (1968) “Two Types of Denotation,” in Studies in Logical Theory, Cambridge: Blackwell.
Givón, T. (1984) Syntax: A Functional-Typological Introduction, Amsterdam: John Benjamins.
Grimshaw, J. (1990) Argument Structure, Cambridge, MA: MIT Press.
Gruber, J. (1965) Studies in Lexical Relations, PhD Dissertation, MIT.
Herburger, E. (2000) What Counts: Focus and Quantification, Cambridge, MA: MIT Press.
Higginbotham, J. (1983) “The Logic of Perceptual Reports: An Extensional Alternative to Situation Seman-
tics,” Linguistics and Philosophy 80: 100–27.

326
 VERB S

Jackendoff, R. (1977) X-bar Syntax, Cambridge, MA: MIT Press.

Parsons, T. (1990) Events in the Semantics of English, Cambridge, MA: MIT Press.
Pietroski, P. (2000) Causing Actions, Oxford: Oxford University Press.
Rothstein, S. (2001) Predicates and their Subjects, Kluwer: Dordrecht.
Rothstein, S. (2004) Structuring Events: A Study in the Semantics of Lexical Aspect, Blackwell: Oxford.
Russell, B. (1903) The Principles of Mathematics, London: Northon.
Schein, B. (1993) Plurals and Events, Cambridge, MA: MIT Press.
Stainton, R. (2005) “In Defense of Non-Sentential Assertion,” in Z.G. Szabó ed., Semantics vs. Pragmatics,
Oxford: Oxford University Press.
Stassen, L. (1997) Intransitive Predication, Oxford: Oxford University Press.
Steedman, M. (1997) “Temporality,” in J. van Benthem and A. ter Meulen, eds., Handbook of Logic and
Language, Elsevier: North Holland.
Szabó, Z.G. (2003) “On Qualification,” Philosophical Perspectives 17: 385–414.
Vendler, Z. (1957) “Verbs and Times,” Philosophical Review 66: 143–60.

327
 3.3
AD JE C T IV ES
Chris Kennedy

1 Introduction
In The Cambridge Grammar of the English Language, adjectives are characterized as expres-
sions “that alter, clarify, or adjust the meaning contributions of nouns,” in order to allow
for the expression of “finer gradations of meaning” than are possible through the use
of nouns alone (Huddleston and Pullum 2002, p. 526). At a general level, adjectives
gain this capability in virtue of two main characteristics, one of which is semantic and
one of which is syntactic. On the semantic side, they introduce properties. (Whether
they actually denote properties is a question we will address in detail below.) On the
syntactic side, they are able to function as modifiers, and so may (with some restric-
tions) combine recursively with nouns. The result of this combination is a new property
which is typically (though not always) true of a subset of the entities that the original
properties are true of, thereby providing a “finer gradation of meaning” than is possible
using the noun alone. This simple picture hides many important and interesting com-
plexities, however, which provide insights on several topics of central interest to both
linguists and philosophers, including: vagueness, contextualism, relativism, composi-
tionality, and the semantic analysis of significant phenomena such as modality. I begin
with an examination of the distributional properties of adjectives, then summarize the
most prominent analyses of their meanings, and finally conclude with a look at some
of the roles that adjectives have played in reasoning about the issues and phenomena
mentioned above.

2 Distribution
As it turns out, determining exactly what is constitutive of the grammatical category
‘adjective’ is not entirely straightforward. There are a number of distributional tests that
distinguish adjectives from other categories, as we will see below, but it is not the case
that all terms that are traditionally classified as adjectives in a particular language sat-
isfy all of these tests, and it is likewise not the case that the tests apply uniformly across
languages to terms that otherwise share the semantic properties that are traditionally
thought to be associated with adjectives. To keep things simple, I will focus primarily
in this section on adjectives in English, with a few comments here and there about
the behavior of adjectives in other languages. The reader should keep keep in mind,
however, that although all languages have terms that share the semantic properties of
English adjectives, the distributional patterns of these terms can vary. (See Dixon and
Aikenvald 2004 for a detailed discussion of the cross-linguisic properties of adjectives.)
 A DJ ECTI V E S

The first identifying feature of adjectives involves their use as predicate terms. Like verbs,
adjectives may supply the main predicate term in a sentence, and may even introduce
their own arguments, as shown by examples such as (1) and (2). (I’ll assume here that
the verb be in (1a) and (2a) is just providing a host for tense and agreement information,
and is not playing a central role in the meaning of the predicate. Many languages do not
require expression of this element in sentences like these.)

(1) a. That stone is weighty.

b. That stone weighs a lot.
(2) a. The country is dependent on foreign oil.
b. The country depends on foreign oil.

However, only adjectives can serve as the complements of the epistemic verbs seem and
appear, as shown by the following contrasts (* denotes syntactic illformedness):

(3) a. That stone seems/appears weighty.

b. * That stone seems/appears weigh a lot.
(4) a. The country seems/appears dependent on foreign oil.
b. * The country seems/appears depend on foreign oil.

This test doesn’t uniquely pick out adjectives, however: nouns (or rather noun phrases)
can sometimes appear as the complement of seem and appear, especially when their
meanings are in some sense scalar or evaluative. This is illustrated nicely in the follow-
ing lines from The Ship of Fools by Sebastian Brandt (which appear on p. 294 of the 1962
edition of Edwin Zeydel’s 1944 translation, published by Dover):

(5) He seems a burden, seems a pest

To all his brood, a hateful guest,
And yet it almost serves him right,
For he’s a dull and witless wight.

A second diagnostic, which distinguishes adjectives from both nouns and verbs, is
the possibility of direct composition with degree words like rather, very, too, so, enough,
how. For example, of the related terms dependent, depend and dependence, only the first
can directly combine with the excessive degree marker too:

(6) a. The country is too dependent on foreign oil.

b. * The country too depends on foreign oil.
c. * The country has too dependence on foreign oil.

Examples (6b–c) can be repaired by first combining too with much (and in the case of
(6b), moving the whole thing to the left of the verb, deriving depends too much on foreign
oil or depends on foreign oil too much), but this only serves to illustrate the point that it
is only the adjectival form dependent that can directly combine with the degree word.
It should be emphasized, though, that adjectives accept composition with degree words
only to the extent that they are associated with concepts that are, or can be, thought of
as scalar, in a sense to be discussed below.

329
 CH RI S KENN E D Y

Perhaps the most central diagnostic for the class of adjectives is the one that is
implicit in Huddleston and Pullum’s functional/semantic characterization of adjectives
as expressions that “alter, clarify, or adjust the meaning contributions of nouns”: adjec-
tives can directly compose recursively with nouns, forming more complex constituents,
which may then combine with other elements (e.g., a determiner or possessive nomi-
nal) to form a noun phrase, as in (7a–c).

(7) a. a blue ball

b. a round blue ball
c. a large round blue ball

Such uses of adjectives are referred to as instances of ATTRIBUTIVE MODIFICA-

TION. In some languages, adjectives may only be used attributively. For example, in
the Yanaria language of New Guinea, adjectives may directly combine with nouns, as
in (8a), but they may provide the main predicate of a sentence only if they compose
first with a nominal element meaning ‘thing, matter’, as shown in (8b); omission of this
element results in ungrammaticality.

(8) a. haga’ dote’na

tasty food
‘tasty food’
b. ma’i egemo haga-na-e’
this banana tasty-thing-PRED
‘This banana is tasty.’ (Lit. This banana is a tasty thing.)

Even English includes a number of adjectives that have only attributive uses, such as
former, mere, principal and main:

(9) a. This is our former/principal/main objective.

b. * This objective is former/principal/main.

The existence of expressions like these has led some researchers to hypothesize that
the attributive use of adjectives is in some important sense basic, a point to which we
will return in detail below. However, like the other tests, this one also has exceptions,
though they are few and appear to be systematic. For example, there is a class of adjec-
tives which includes asleep, awake, alone (sometimes called a-adjectives, for obvious
reasons) which can appear as complements of seem and appear, but are barred from
attributive position:

(10) a. * Kim photographed two asleep/alive polar bears.

b. Kim photographed two sleeping/living polar bears.

There are, in addition, languages which require noun-modifying adjectives to first com-
bine with a predicative element, effectively turning them into relative clauses (and
calling into question their status as adjectives to begin with; see Baker 2003).
Cases like these show that the possibility of attributive modification is not a nec-
essary condition for adjective status, but it is generally agreed that it is a sufficient
one. Nevetheless, some care must still be taken in applying this test. Nouns may also

330
 A DJ ECTI V E S

combine directly with nouns, as in eyeball, tennis ball, home run ball, or medicine ball, but
in a way that is different from adjectives in two respects. First, the interpretation of
such structures (referred to as NOUN-NOUN COMPOUNDS) is variable and often
context dependent: an eyeball is a part of the body that has the shape of a ball; a tennis
ball is a ball used for playing tennis; a home run ball is a ball that was hit for a home
run (e.g., Barry Bonds’ 756th home run ball was auctioned for $752,467); a medicine ball
could be a ball of medicine, a ball used to deliver medicine, or a piece of gym equipment.
Attributive adjective modification, in contrast, gives rise to much more systematic and
restricted interpretations, as we will see in detail below.
Second, attributive adjectives are different from nouns in compounding structures in
that the former cannot occur outside the latter:

(11) a. a majestic towering home run ball

b. * a majestic home run towering ball
c. * a home run majestic towering ball

In contrast, attributive adjectives can often be reordered without compromising syntac-

tic well-formedness:

(12) a. a majestic towering home run

b. a towering majestic home run

Interestingly, it is not the case that attributive adjective ordering is fully unrestricted.
For example, the default order of the adjectives numerous, inefficient, and American as
attributive modifiers is as in (13a); orders in which numerous is non–initial are ungram-
maticsal (13b–c); and an order in which American precedes inefficient is acceptable just
in case American is understood contrastively or in focus. For example, (13d), with stress
on American (indicated by capitalization), would be acceptable as an answer to the ques-
tion Are there a lot of inefficient cars on the road?

(13) a. There are numerous inefficient American cars on the road.

b. * There are inefficient numerous American cars on the road.
c. * There are inefficient American numerous cars on the road.
d. There are numerous AMERICAN inefficient cars on the road
(but not so many JAPANESE ones).

These ordering restrictions are robust cross-linguistically, holding both in languages like
English, where adjectives precede nouns, and in a mirror-image fashion in languages in
which nouns precede adjectives, though the underlying reasons for the distribution are
not well understood (see Demonte 2008; Svenonius 2008; and Cinque 2010 for recent
discussion).
Sometimes multiple orders are possible but result in significant differences of inter-
pretation. For example, wild Minnesotan rice denotes quantities of uncultivated or unruly
rice, which stands in some relation to Minnesota (most likely it was grown there, though
other interpretations are possible), while Minnesotan wild rice denotes quantities of ziza-
nia palustris (which is in fact not a species of rice). The relative order of the adjective
and the noun, when two orders are possible, can also affect meaning. Consider, for

331
 CH RI S KENN E D Y

example, (14), in which the adjective can either be interpreted nonrestrictively, as in

(14a), or restrictively, as in (14b) (Bolinger 1967; Larson and Marušič 2004).

(14) All of his unsuitable remarks will be eliminated from the final text.
a. All of his remarks will be eliminated; they are unsuitable.
b. All (and by implication, only) those of his remarks that are unsuitable
will be eliminated.

When the adjective occurs postnominally, however, only the restrictive interpretation
is available:

(15) All remarks unsuitable (for publication) will be eliminated from the final
text.

The examples in (16), discussed originally by Bolinger (1967) (see also Larson 1998;
Cinque 1993, 2010; Demonte 2008; and Morzycki 2008), show a similar sensitivity to
the relative order of the noun and the adjective.

(16) a. The visible stars include Capella, Betelgeuse and Sirius.

b. The stars visible include Capella, Betelgeuse and Sirius.

(16a) is truth-conditionally ambiguous: it can be understood as a claim about which

stars are visible at the time of utterance, or as a claim about which stars are intrisically
visible (e.g., capable of being seen by the naked eye). At noon on a sunny day, (16a)
would (normally) be false on the first reading and true on the second. Example (16b), in
contrast, is unambiguous: It has only the ‘currently visible’ reading, and would be false
in the sunny day context. This suggests that the ambiguity in (16a) does not reflect an
ambiguity in visible, but rather has something to do with composition. This conclusion
is further strengthened by the contrast between (17a) and (17b).

(17) a. The invisible visible stars include Betelgeuse.

b. ?? The visible invisible stars include Betelgeuse.

(17a) means that Betelgeuse is among the stars which can generally be seen but are cur-
rently invisible, which is a perfectly coherent thing to say. Example (17b), on the other
hand, sounds a bit odd (indicated by the ‘??’), because it involves definite reference to a
set of stars that are currently visible and intrinsically invisible, which is a combination
of properties that is difficult to have, and possibly even contradictory. As pointed out by
Larson (1998), these facts show that the ‘current’ vs. ‘intrinsic’ distinction is not (or not
only) a function of the relative ordering of the adjective and the noun but (also) reflects
more subtle facts about adjective–noun composition.
Taken together, examples like these show that subtle differences in structure can
affect the truth conditions of sentences with attributive adjectives, a fact that must be
kept in mind when constructing arguments—either linguistic or philosophical—based
on the interpretations of such constructions. Superficially simple structures sometimes
hide an underlying complexity, which must be taken into account by reasoning based
on the meanings of those constructions and the elements they contain.

332
 A DJ ECTI V E S

3 Meaning
I said above that adjectives introduce properties. Two kinds of facts suggest that adjec-
tives also denote properties. First, as we have already seen, adjectives may provide the
main predicate in a sentence. Second, we often see entailments from the attributive
form to the predicative form, as in (18).

(18) a. Cosmo is a hairy brown dog.

b. Cosmo is hairy.
c. Cosmo is brown.

We do not need to know anything about dogs to know that (18a) entails (18b–c). If we
later learn that Cosmo is not merely a dog but also a Westminster Kennel Club cham-
pion, we may also justifiably conclude from (18a) that he is a hairy brown Westminster
Kennel Club champion. If (18a) involves the ascription of three properties to Cosmo
(being a dog, being brown and being hairy), these patterns of reasoning follow.
Adjectives that give rise to such reasoning patterns are often referred to as INTER-
SECTIVE. Not all adjectives are intersective, however, a fact that introduces challenges
for the idea that adjectives as a class denote properties. Consider the following examples,
from Partee 1995. Knowing that (19a) is true does not justify the conclusion in (19b),
because it could be the case that the only respect in which Francis is skillful is in his role
as a surgeon, in which case we would accept the former but most likely deny the latter.

(19) a. Francis is a skillful surgeon.

b. Francis is skillful.

Similarly, the combined truth of (19a) and (20a) do not license the conclusion in (20b):
Francis could be a very skillful surgeon and still have only limited facility with the
violin.

(20) a. Francis is a violinist.

b. Francis is a skillful violinist

Intuitively, skillful in (19a) and (20b) picks out just the subset of surgeons and violinists
who are skillful as surgeons and as violinists respectively. As a result, we cannot conclude
from the truth of e.g. (19a) that Francis is skillful in any other way. Partee (1995) labels
adjectives like skillful SUBSECTIVE, since composition of the adjective with a nominal
constituent returns a subset of the denotation of the nominal, but in a way that does not
support the same inferences as with intersective adjectives: All intersective adjectives
are subsective, but not all subsective adjectives are intersective.
In addition to intersective and subsective adjectives, there is a third group of what
Partee calls NONSUBSECTIVE adjectives including former, alleged, fake, possible, ersatz
and so forth, which are neither intersective nor subsective. The set of objects that satisfy
the description former president of the United States is neither the intersection of the set
of former things (if that even makes sense) with the set of presidents, nor is it a subset
of the set of presidents. Similarly, a fake identification is arguably not an identification at
all. (Nonsubsective adjectives like fake which imply exclusion from the noun meaning
are sometimes called PRIVATIVE.)

333
 CH RI S KENN E D Y

A number of researchers, including Lewis (1970), Wheeler (1972), Cresswell (1973)

and Montague (1974), have taken the existence of non-intersective interpretations of
adjectives as evidence that adjectives do not denote properties, but rather must be ana-
lyzed as expressions that map properties into new properties. (Others have adopted a
more nuanced view whereby attributive uses involve such a meaning, while predicative
uses denote properties; see Siegel 1976 for a sophisticated implementation of this kind
of account, and the kind of linguistic data that can be brought to bear to support it). In
some cases (the intersective adjectives), the output is just the conjunction of the input
with a property introduced by the adjective; in others (the non-intersective ones), the
adjective determines the output property in a more complex way, as we have seen.
Furthermore, apparent predicative uses of adjectives are analyzed as deriving from an
underlying attributive source, so that what is predicated of the subject in e.g. (19b) is
not the property of being skillful, but rather the property of being a skillful one, where
the value of the anaphor one is filled in contextually. English is thus analyzed on a par
with languages like Yagaria (see (8b) above), the only difference being that the surface
syntax obscures rather than reflects the the underlying form.
The attributive analysis of adjectives represents a kind of “generalization to the worst
case” strategy, which can be found elsewhere in compositional analyses of English (cf.
the analysis of proper names as generalized quantifiers in Montague 1974). The advan-
tage of such an approach is that it allows for a general theory of lexical types and com-
positional operations, and if the general goal is to show that the semantic properties of
natural language can be accounted for within a compositional framework, the strategy
is a reasonable one. (Though see Kamp 1975 for a critical assessement of the explana-
tory power of the attributive analysis.) The disadvantage of such an approach is that in
effectively building noun-dependency into the meaning of the adjective, the uniformity
hypothesis doesn’t leave much space for complex structural effects on meaning of the
sort we observed in the previous section for adjectives like visible. At the same time, it
can lead to an over-simplistic assessment of the data, when a more sophisticated analysis
of both noun and adjective meaning can provide us with ways of explaining patterns
like those above without adopting the attributive analysis of adjective meaning.
Consider, for example, (21a), which is ambiguous between the reading in (21b),
in which the adjective is subsective, and the one in (21c), in which the adjective is
intersective.

(21) a. Lee is a beautiful singer.

b. Lee sings beautifully.
c. Lee is a singer who is beautiful.

Larson (1998) shows that this ambiguity can be captured straightforwardly without pos-
iting an attributive semantics for adjectives—i.e., by maintaining the hypothesis that
adjectives denote properties—by extending Davidson’s (1967) well-established analysis
of adverbial modification in action sentences to adjectival modification. Specifically,
nouns like singer are analyazed as relations between events and individuals, and adjectives
are analyzed as properties of either events or individuals. In some cases, the syntax of the
surface form determines exactly what kind of argument the adjective has: in (21b) it is
an event, and in (21c) it is an individual. (The addition of the suffix -ly is also syntacti-
cally conditioned.) In others, such as (21a), the syntax is compatible with either option,
resulting in ambiguity. Depending on which option we choose for the adjective, we

334
 A DJ ECTI V E S

derive the truth conditions in (22a–b) for the sentence which correspond to the read-
ings in (21b–c), respectively.

(22) a. e[beautiful(e) ^ singer(lee,e)]

b. e[beautuful(lee) ^ singer(lee,e)]

Larson’s analysis shows how one kind of subsectivity can be handled by appealing to a
more sophisticated theory of noun meaning; a different kind of subsectivity, manifested
by adjectives that encode scalar concepts, referred to as GRADABLE ADJECTIVES,
can be handled by appealing to a more complex view of adjective meaning. Consider,
for example, the adjective tall. The truth of (23a) does not guarantee the truth of (23b);
likewise, knowing that (23a) is true and knowing that Julian is a basketball player does
not allow us to conclude (23c).

(23) a. Julian is a tall jockey.

b. Julian is tall.
c. Julian is a tall basketball player.

Facts like these lead to the conclusion that tall does not denote a property on its own;
instead, tall comes to denote a property only after determing a “threshold” or STAND-
ARD of height that an object must reach in order to count as tall, which is itself com-
puted on the basis of a relevant set of objects, or COMPARISON CLASS (Kamp 1975;
Klein 1980; Kennedy 2007). To say that Julian is a tall jockey, in other words, is to say
that he is a jockey who is tall relative to the standards for jockeys, which does not entail
that he is tall relative to some other standard or comparison class. In particular, this
does not entail that he is tall relative to whatever standard is approprirate for basketball
players, so we are not licensed to conclude (23c), even if we know that Julian is a
basketball player.
One way of accounting for facts like those in (23) is to adopt an attributive analysis
of adjectives, whereby the noun provides the comparison class for the adjective. On this
view, tall denotes a function from properties to properties of the form in (24), where
stnd picks out an appropriate value from the set of heights we get by applying the height
function to the objects in the denotation of the noun (cf. Klein 1980; Heim and Kratzer
1998; a proper version of this analysis would need to intensionalize the comparison
class, of course).

(24) OPOx.height(x) t stnd{height(y) | P (y)}

However, a closer look at the distributional properties of adjectives like tall shows that
this is not the only possible analysis of their meanings, or of facts like those in (23).
As the following examples show, a central characteristic of gradable adjectives is that
they can appear in a variety of constructions and are linked semantically by encod-
ing different notions of degree: relations to measures, comparison relations, relations of
sufficiency and excess, and so forth.

(25) a. Julian is four feet tall.

b. Julian is taller than Sterling.
c. Julian is as tall as we expected him to be.

335
 CH RI S KENN E D Y

d. Julian is too tall to fit in the box.

e. Julian is tall enough to reach the ceiling.
f. Julian is so tall that he has to buy special clothes.

These kinds of facts have led many researchers to hypothesize that gradable adjectives
do not express relations between properties (or whatever the proper semantic concep-
tion of noun meanings is) and properties, but rather relations between more abstract
representations of measurement, or DEGREES, and properties (see e.g. Bartsch and
Vennemann 1973; Seuren 1973; Cresswell 1976; von Stechow 1984; see Klein 1991
and Bale 2009 for detailed discussions of how degrees can be related to equivalence
classes of individuals). The most common implementation of this view posits the deno-
tation in (26) for tall, where d is a degree of height.

(26) OdOx.height(x) t d

On this view, the function of the complex constituents that combine with the adjec-
tive in the examples in (25) is to fix the value of the degree argument, thereby providing a
standard of comparison, and turning the adjective into a property that holds of an object if
its height exceeds the relevant standard. The predicate in (25a), for example, denotes the
property of having a height that exceeds the degree denoted by four feet; the comparative
construction in (25b) denotes the property of having a height that exceeds the degree of
Sterling’s height; and so on. Importantly, the complex constructions denote properties,
and indeed behave intersectively in attributive position: (27a) entails (27b), and if Ster-
ling and Julian are basketball players as well as jockeys, we may also draw the conclusion
in (27c). (Prenominal comparatives also presuppose that the nominal predicate applies to
both the target and standard of comparison, but this is an independent fact which follows
from the syntax and compositional semantics of comparatives; see Bresnan 1973.)

(27) a. Julian is a taller jockey than Sterling.

b. Julian is taller than Sterling.
c. Julian is a taller basketball player than Sterling.

Paradoxically, in this kind of analysis, it is unmodified occurrences of gradable adjec-

tives of the sort seen in (23a–c) that present the trickiest analytical challenge, since
there is no constituent in the surface form to saturate their degree arguments and turn
them into properties. The usual approach is to hypothesize a phonologically null, “posi-
tive degree” morpheme which does this job, by existentially binding the degree argu-
ment and imposing the restriction that it come from a degree on the scale above a
certain threshold. How exactly this threshold is identified, and the extent to which it
is determined based on discourse context and linguistic context (e.g., by the fact that
an adjective is used attributively vs. predicatively, or by lexical semantic properties
related to the kind of scale the adjective uses), are issues that are resolved differently in
different analyses (see Kennedy 2007 for a comparison of approaches). The end result is
that nonintersectivity in examples like (23a–c) is accounted for not by hypothesizing a
function–argument relation between the adjective and the noun, but by hypothesizing
a more complex semantic analysis of the adjective (phrase).
That said, it should be acknowledged that, like the attributive analysis, this approach
also gives up on the idea that adjectives (the gradable ones, at least) denote properties.

336
 A DJ ECTI V E S

Instead, gradable adjectives denote relations between individuals and scalar values
(degrees), and come to denote properties only through composition with something
that saturates their degree arguments. This has the analytical advantage of providing a
ready account of complex constructions like those in (27) (though such constructions
have also been analyzed in nondegree analyses, which begin from semantic assumptions
about adjective meaning on a par with (24); see below for details). It also has certain
theoretical advantages in the analysis of phenomena of interest to philosophers, such
as vagueness and context-dependence, a point I return to below. However, it has a
significant disadvantage from a purely linguistic perspective: if gradable adjectives do
not directly denote properties but come to do so only through composition with special
degree-saturating morphology, then why is it the case that in all the languages of the
world that we know of, the linguistic form that we think of as introducing the “core”
property associated with the word—the property of being tall, large, rich, happy, and
so forth—is never marked by overt morphology? This is a question which so far has not
been given a satisfactory answer by proponents of degree-based semantic analyses of
gradable adjectives.

4 Philosophical Interest
Adjectives have played a prominent role in a number of philsophical discussions of aspects
of human language, as I will document in this section, but perhaps the most prominent is
their role in the characterization and analysis of vagueness (Chapter 4.13). The problem
of vagueness is essentially the problem of being unable or unwilling to say of any single
point along an ordering generated by the meaning of a particular term whether that point
separates the things that the term is true of from the things that it is false of. Vagueness
is not a feature of adjectives alone, but adjectives provide a particularly rich empirical
ground for investigating it, because so many of them fall into the class of gradable adjec-
tives discussed in the previous section, and so (in their basic, unmodified forms) intro-
duce properties that are true or false of objects depending on their position on a scale.
A central question in work on gradable adjectives and vagueness is whether vague-
ness is the defining characteristic of the class, with their other significant properties,
such as the possibility of forming comparative constructions like those in (25) arising
as a result of this feature, or whether vagueness is derived. The first view is seen in the
work of Wheeler (1972), Kamp (1975), Klein (1980), van Benthem (1982), and most
recently by van Rooij (2011), who provide compositional semantic analyses of various
kinds of comparative constructions in terms of an initial analysis of gradable adjectives
as vague property terms. This approach has the advantage of explaining the apparent
morphological universal mentioned at the end of the previous section: if there is a dif-
ference in morphological complexity between the positive and comparative form of an
adjective, it is always the latter that is complex. (Though it should be noted that many
languages—probably the majority—do not make a morphological distinction between
the forms; see Ultan 1972.)
The second view is associated with degree-based analyses of gradable adjectives of
the sort discussed in the previous section: Since adjectives do not denote properties at
all, but rather relations between individuals and degrees, there is no sense in which the
basic meanings of the terms are vague. Instead, vagueness is introduced composition-
ally through the mapping of such relations to properties. In particular, if this mapping
is achieved through composition with a phonologically null “positive” morpheme, as

337
 CH RI S KENN E D Y

described above, this opens up the analytical possibility of associating vagueness with
the particular semantic features of this morpheme, a move advocated and justified by
Fara (2000) and Kennedy (2007, 2011).
Adjectives have also played an important role in discussions of the implications of
variable judgments about truth for theories of meaning. Recent work on semantic rela-
tivism (see Chapter 4.15) has focused extensively on differences in truth judgments of
sentences containing adjectives of personal taste like tasty and fun (see e.g. Richard
2004; Lasersohn 2005; MacFarlane 2005; Stephenson 2007; Cappelen and Hawthorne
2009), and researchers interested in motivating contextualist semantic analyses have
often used facts involving gradable adjectives (recall the judgments in (23) which show
that the threshold for what “counts as” tall can change depending on whether we are
talking about jockeys or basketball players) to develop arguments about the presence
(or absence) of contextual paramters in other types of constructions, such as knowledge
statements (see e.g. Unger 1975; Lewis 1979; Cohen 1999; Stanley 2004, and Chapters
3.7 and 4.14). Other researchers have attempted to account for the apparent context
sensitivity of these examples without importing context dependence into the semantics
(see e.g. Cappelen and Lepore 2005).
More radically, Charles Travis (1997, 1985, 1994) has used judgments about the truth
of sentences containing color adjectives to argue against the view that sentences deter-
mine truth conditions. Instead, according to him, the semantic value of a sentence at
most imposes some necessary conditions under which it may be true (as well as conditions
under which it may be used), but those conditions need not be sufficient, and the content
of the sentence does not define a function from contexts to truth. However, Travis’s argu-
ment goes through only if it is the case if the truth conditional variability introduced by
color adjectives cannot be linked to a context-dependent element in its logical form or
to an underlying ambiguity. And indeed, there are responses to his work which argue for
each of these positions based on careful and sophisticated linguistic analysis of color adjec-
tives (see e.g. Szabó 2001; Rothschild and Segal 2009; Kennedy and McNally 2010).
The significance of adjectives for general questions about compositionality in lan-
guage goes well beyond cases like Travis’s. Two additional kinds of phenomena are of
particular interest. The first involves sentences like the following, in which the adjec-
tive slow seems to be contributing a different shade of meaning depending on the sort of
thing it is predicated of: a slow quarterback is one who runs (or maybe executes plays)
slowly; a slow road is one on which traffic moves slowly; a slow song is one with a slow
tempo; and a slow book is one that takes a long time to read.

(28) a. Tom Brady is a slow quarterback.

b. Lake Shore Drive is a slow road during rush hour.
c. “Venus in Furs” is a slow song.
d. Remembrance of Things Past is a slow book.

In each case, the meaning contributed by the adjective appears to be systematically

related to a kind of activity that is conventionally associated with the meaning of the
noun. This has led some researchers to hypothesize that the compositional relation
between adjectives and nouns can be even more complex than what we saw above for
beautiful singer, and in particular that it requires a highly articulated lexical semantic
structure for nouns (see e.g. Pustejovsky 1991, 1995; see Fodor and Lepore 1998 for an
opposing view).

338
 A DJ ECTI V E S

The second kind of case involves examples in which prenominal adjectives appear
to have interpretations outside of the noun phrases in which they appear. For example,
adjectives like occasional, sporadic, and rare can syntactically compose with a noun but
have a sentence-level interpretation equivalent to an adverb of quantification, as in
the following passage from a 1989 article in the Chicago Tribune (where the adverbial
occurrence in the third sentence highlights the fact that the adjectival use in the second
sentence has a sentence-level meaning):

(29) “I used to be a pretty good Scotch drinker,” [Tower] said. “I haven’t tasted
Scotch in 12 years. After that I had only wine and perhaps an occasional
martini, occasionally a little vodka with smoked salmon or caviar or
something like that. But that was just occasionally.”

Prenominal average provides an even more striking example of this kind of phenom-
enon: (30a) (from a post on answers.bloglines.com) means that the average number of
people in an American family is 3.14, and so does not give rise to the bizarre inference
that there are actual families which contain 3.14 people, in contrast to (30b), which
does give rise to this inference.

(30) a. The average American family consists of 3.14 people.

b. ?? The typical/normal/usual American family consists of 3.14 people.

The fact that (30a) has the meaning that it does indicates that average can somehow
compose at the sentence level with the numeral and a measure function (which returns
the number of whole people in an American family), rather than directly with the prop-
erty denoted by American family. The analytical challenge presented by these cases is to
show that the actual meanings can be compositionally derived from the surface forms
without resorting to ad hoc stipulations and construction-specific rules; this project is
undertaken by Stump (1981); Larson (1998); Gehrke and McNally (2009) for occa-
sional and related terms, and by Carlson and Pelletier (2002) and Kennedy and Stanley
(2009) for average. While the analyses advocated in these different papers are distinct,
they collectively illustrate a theme that is present in much of the work on adjectives
described in this chapter: that the semantic properties of adjectival constructions are
often more complex than superficial appearances indicate, and must be interpreted
against the backdrop of a sophisticated linguistic analysis.

References
Baker, M. C. (2003) Lexical Categories: Verbs, Nouns, and Adjectives. Cambridge: Cambridge University Press.
Bale, A. (2009) “A universal scale of comparison”. Linguistics and Philosophy, 31: 1–55.
Bartsch, R. and Vennemann, T. (1973) Semantic Structures: A Study in the Relation between Syntax and Seman-
tics. Frankfurt: Athäenum Verlag.
Bolinger, D. (1967) “Adjectives in English: Attribution and predication”. Lingua, 18: 1–34.
Bresnan, J. (1973) “Syntax of the comparative clause construction in English”. Linguistic Inquiry,
4: 275–343.
Cappelen, H. and Hawthorne, J. (2009) Relativism and Monadic Truth. Oxford: Oxford University Press.
Cappelen, H. and Lepore, E. (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act
Pluralism. Oxford: Blackwell.
Carlson, G. and Pelletier, F. J. (2002) “The average American has 2.3 children”. Journal of Semantics, 19:
73–104.

339
 CH RI S KENN E D Y

Cinque, G. (1993) “A null theory of phrase and compound stress”. Linguistic Inquiry, 24: 239–297.
——(2010) The Syntax of Adjectives. Cambridge, MA: MIT Press.
Cohen, S. (1999) “Contextualism, skepticism and the structure of reasons”. Philosophical Perspectives, 13:
57–89.
Cresswell, M. (1973) Logics and Languages. London: Methuen.
Cresswell, M. J. (1976) “The semantics of degree”. In B. Partee (ed.), Montague Grammar. New York: Aca-
demic Press, 261–292.
Davidson, D. (1967) “The logical form of action sentences”. In N. Rescher (ed.), The Logic of Decision and
Action. Pittsburgh: University of Pittsburgh Press.
Demonte, V. (2008) “Meaning-form correlations and adjective position in Spanish”. In L. McNally and C.
Kennedy (eds.), Adjectives and Adverbs: Syntax, Semantics and Discourse. Oxford: Oxford University Press,
71–100.
Dixon, R. M. W. and Aikenvald, A. Y. (2004) Adjective Classes: A Cross-Linguistic Typology. Oxford: Oxford
University Press.
Fara, D. G. (2000) “Shifting sands: An interest-relative theory of vagueness”. Philosophical Topics, 20: 45–81.
Fodor, J. and Lepore, E. (1998) “The emptiness of the lexicon: Reflections on James Pustejovsky’s the genera-
tive lexicon”. Linguistic Inquiry, 29: 269–288.
Gehrke, B. and McNally, L. (2009) “Frequency adjectives and assertions about event types”. In Proceedings
of Semantics and Linguistic Theory 19. Ithaca, NY: CLC Publications.
Heim, I. and Kratzer, A. (1998) Semantics in Generative Grammar. Oxford: Blackwell.
Huddleston, R. and Pullum, G. K. (2002) The Cambridge Grammar of the English Language. Cambridge,
Cambridge University Press.
Kamp, H. (1975) “Two theories of adjectives”. In E. Keenan (ed.), Formal Semantics of Natural Language.
Cambridge: Cambridge University Press, 123–155. Reprinted in 1984.
Kennedy, C. (2007) “Vagueness and grammar: The semantics of relative and absolute gradable predicates”.
Linguistics and Philosophy, 30: 1–45.
——(2011) “Vagueness and comparison”. In P. Egré and N. Klinedinst (eds.), Vagueness and Language Use.
Houndmills, UK: Palgrave, 73–97.
Kennedy, C. and McNally, L. (2010) “Color, context and compositionality”. Synthese, 174: 79–98.
Kennedy, C. and Stanley, J. (2009) “On average”. Mind, 118.
Klein, E. (1980) “A semantics for positive and comparative adjectives”. Linguistics and Philosophy, 4: 1–45.
——(1991) “Comparatives”. In A. von Stechow and D. Wunderlich (eds.), Semantik: Ein internationales
Handuch der zeitgenössischen Forschung (Semantics: an international handbook of contemporary research),
Berlin: de Gruyter, 673–691.
Larson, R. (1998) “Events and modification in nominals”. In D. Strolovitch and A. Lawson (eds.), Proceed-
ings from SALT VIII. Ithaca, NY: CLC Publications, 145–168.
Larson, R. and Marušič, F. (2004) “On indefinite pronoun structures with APs: Reply to Kishimoto”.
Linguistic Inquiry, 35: 268–287.
Lasersohn, P. (2005) “Context dependence, disagreement, and predicates of personal taste”. Linguistics and
Philosophy, 28: 643–686.
Lewis, D. K. (1970) “General semantics”. Synthese, 22: 18–67.
——(1979) “Scorekeeping in a language game”. Journal of Philosophical Logic, 8: 339–359.
MacFarlane, J. (2005) “Making sense of relative truth”. Proceedings of the Aristotelian Society: 321–339.
Montague, R. (1974) “English as a formal language”. In R. Thomason (ed.), Formal Philosophy: Selected
Papers. New Haven: Yale University Press, 188–221.
Morzycki, M. (2008) “Nonrestrictive modifiers in nonparenthetical positions”. In L. McNally and C. Kennedy
(eds.), Adjectives and Adverbs: Syntax, Semantics and Discourse. Oxford University Press, 101–122.
Partee, B. H. (1995) “Lexical semantics and compositionality”. In L. Gleitman and
M. Liberman (eds.), Language. An Invitation to Cognitive Science. Cambridge, MA: MIT Press, 311–360.
Pustejovsky, J. (1991) “The generative lexicon”. Computational Linguistics, 17.
——(1995) The Generative Lexicon. Cambridge: MIT Press. Richard, M. (2004) “Contextualism and relativ-
ism”. Philosophical Studies, 119: 215–242.
Rothschild, D. and Segal, G. (2009) “Indexical predicates”. Mind and Language, 24: 467–493.
Seuren, P. A. (1973) “The comparative”. In F. Kiefer and N. Ruwet (eds.), Generative Grammar in Europe.
Dordrecht: Riedel, 528–564.
Siegel, M. E. (1976) “Capturing the Adjective”. Ph.D. thesis, University of Massachusetts at Amherst.
Stanley, J. (2004) “On the linguistic basis for contextualism”. Philosophical Studies, 119: 119–146.

340
 A DJ ECTI V E S

Stephenson, T. (2007) Towards a Theory of Subjective Meaning. Ph.D. thesis, Massachusetts Institute of
Technology.
Stump, G. (1981) “The interpretation of frequency adjectives”. Linguistics and Philosophy, 4: 221–257.
Svenonius, P. (2008) “The position of adjectives and other phrasal modifiers in the decomposition of DP”.
In L. McNally and C. Kennedy (eds.), Adjectives and Adverbs: Syntax, Semantics and Discourse. Oxford:
Oxford University Press, 16–42.
Szabó, Z. (2001) “Adjectives in context”. In I. Kenesei and R. M. Harnish (eds.), Perspectives on Semantics,
Pragmatics and Discourse: A Festschrift for Ferenc Kiefer. Amsterdam: John Benjamins, 119–146.
Travis, C. (1985) “On what is strictly speaking true”. Canadian Journal of Philosophy, 15: 187–229.
——(1994) “On constraints of generality”. Proceedings of the Aristotelian Society, 94: 165–188.
——(1997) “Pragmatics”. In B. Hale and C. Wright (eds.), A Companion to the Philosophy of Language.
Oxford: Blackwell, 87–106.
Ultan, R. (1972) “Some features of basic comparative constructions”. In Working Papers on Language Univer-
sals, No. 9. Stanford, CA: Language Universals Project, Committee on Linguistics, 117–162.
Unger, P. (1975) Ignorance. Oxford: Clarendon Press.
van Benthem, J. (1982) “Later than late: On the logical origin of the temporal order”. Pacific Philosophical
Quarterly, 63: 193–203.
van Rooij, R. (2011) “Implicit versus explicit comparatives”. In P. Egré and N. Klinedinst (eds.), Vagueness
and Language Use. Houndmills, UK: Palgrave Macmillan, 51–72.
von Stechow, A. (1984) “Comparing semantic theories of comparison”. Journal of Semantics, 3: 1–77.
Wheeler, S. (1972) “Attributives and their modifiers”. Noûs, 6: 310–334.

341
 3.4
Q UANT IFIER S AND
DE TE R M INER S
G. Aldo Antonelli and Robert C. May*

1 Introductory Remarks
When we think about language, and, in particular, about how we use it to speak of
things, there is an intuition that permeates virtually all of our thinking and analysis.
The intuition is that there is in language an inherent cleavage in the class of terms we
use to carry out this task, between those used to speak about specific particulars, and
those used to speak in general about those particulars. From the semantic point of view,
this is the distinction between reference and quantification; it is what is paradigmati-
cally expressed in natural languages by the distinction within the noun phrases between
proper names like “Max” or “Maria” and quantifiers, determiner–noun complexes like
“everybody” or “three women.” It is not hard to read much, perhaps even most, of the
ongoing discussion in the philosophy of language as trying to give content to this intui-
tion; indeed, the field itself dates its birth in part to Russell’s reflections on just this
point in “On Denoting,” when he considers whether definite descriptions are to be ana-
lyzed as akin to names or quantifiers, opting for the latter. Russell’s arguments, assuming
as they do the underlying intuitive validity of the distinction between reference and
quantification, however, raise a meta-question: What is one opting for when adopting a
quantificational analysis? Or to put it a little differently: What does it mean to say that
a term is a quantifier, as opposed to a name? What constitutes being a quantifier? Giving
answers to questions of this sort of course was not Russell’s primary concern,1 but it is
ours, and so our goal in this chapter is to acquaint the reader with the answers that have
emerged, as well as present some novel reflections on these matters.
In searching for answers, a natural first-place to alight is with something elementary.
So consider the two first-order quantifiers “there is” and “for all” that are the staples of
introductory logic courses, and are employed with insouciance in mathematics, linguis-
tics, and computer science courses throughout the undergraduate curriculum. Typically,
these quantifiers are introduced as expressions that can be used in conjunction with a
variable and an open formula to produce a sentence. Students are then told that of these
quantifiers there are at most two: the existential quantifier  and the universal quantifier
 that combine with variables such as x and y and formulas M (x) or M (y) to produce
sentences  xM (x) or yM(y), intuitively saying that something (in some intended
domain or context) is a M, or that everything in that same domain is M. Students are
cursorily told that, grammatically, quantifiers are used to bind variables, and after being
 Q U A N TI F I ERS A N D D E T E R MI N E R S

shown a few examples of the notation’s intended meaning, they are sent on their merry
way. In fact, students are seldom, if ever, told what the quantifiers are, or for that mat-
ter, why there are two (or maybe one, given interdefinability with negation, or perhaps
more), or how they affect the meaning of the expressions in which they occur.
This scenario is perhaps a bit fanciful, but nevertheless it frames questions that are not
only substantive, but also foundational; questions which if left unanswered would leave
us hard-pressed to say that the intuitive difference between reference and quantification
is of any analytic significance. What the answers are, it turns out, are already implicit
in our fanciful scenario, at least in broad form. Thus, the very terms in which the sce-
nario is described imply an answer to the first, and most fundamental, question of what
quantifiers are. They are logical terms, integral aspects of the purely logical machinery;
indeed, that it was a logical analysis was a central selling point for Russell’s theory
of descriptions. Making good on this answer, however, and how it founds answers to
the latter two questions, directs us to attend to the manner in which sentences con-
taining quantifiers are represented. Thus, the sentences xM(x) and yM (y) represent
three characteristics essential to understanding quantification. The first is scope. This
is indicated by the use of bracketing conventions, and is essential, as Frege was the first
to observe, for properly describing the meaning of sentences in which more than one
quantifier occurs, such as x(y(\ (x,y))). The second is the meaning of the quantifier.
This is indicated by the symbols “” and “,” whose meanings are standardly glossed
as the universal and the existential. The third is the order of the quantifier, and this is
indicated by the letter following the quantifier symbol. xM(x) contains a first-order
existential quantifier, but we can also have quantifiers of higher-orders; thus, P P(a)
contains a second-order existential quantifier. What we have learned since Frege and
Russell’s time is that it is through attention to the interaction of these three factors—
scope, meaning and order—that there emerges a notion of quantifiers as logical terms
that is suffciently articulated to distinguish them from referential terms, and thus ground
the ur-intuition of the distinction between reference and quantification.
Now, the factors just mentioned conform in an obvious way to a distinction between
syntax and semantics. Thus scope is inherently syntactic, a matter of the logical form
of sentences containing quantifiers, and, as Tarski showed, if these sentences are com-
posed in the proper manner—that is with the sort of recursive syntax initiated by Frege’s
“conceptual notation”—then it is possible to construct in parallel a definition of truth
suffcient for characterizing logical consequence (Tarski 1956). If this is to be our dis-
tinction between syntax and semantics, then an answer to the second of our questions
is provided, since for the task at hand we need only one quantifier namely, universal
generalization.2 All we need is to be able to state that matters hold generally, for all
things. To be sure, this is a profound result, but it is also limiting in its pragmatism; since
just one is needed for the job at hand, the question of how many quantifiers there are is
moot. This sort of functional answer will thus leave us short, if our goal is to answer this
question in any robust way. Such an answer will require something more than just an
account of the logical form of generalization, and the semantics that depends upon it. It
will also require an account of the logical content of quantification, of the quantities that
are the content of quantifiers, and not just merely an account of generalization. Giving
this sort of robust account is what falls under the purview of the theory of generalized
quantifiers.
The theory of genralized quantifiers has proven to be of ongoing interest largely
because of its application in the semantics of natural language, where its utility has been

343
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

central to the analysis of the broad class of determiner expressions, including not only
those that correspond to the classical logical expressions, like “every” and “some,” but
also to words such as “three” and “most” and “only” that do not.3 The theory has been
remarkably frutiful in isolating those properties that unify these words into a class, and
distinguishing them from those that differentiate them into natural subgroups. Our goal
in this chapter will be to outline answers the theory of generalized quantifiers provides
to the questions we posed above, and this will draw our focus largely to how the theory
characterizes the meanings of quantifiers. But in doing so, we will also see how the
theory articulates the third factor that those answers demand, the order of quantifiers.
This matter has had a vexed history, primarily because of a confluence of meta-logical
and ontological worries that has given rise to an embedded skepticism about second-
order logic.4 Thus, fundamental results obtainable for first-order logic, for example com-
pleteness, do not obtain for second-order logic. Moreover, while it is perhaps clear what
meets the call of first-order variables, it is by no means as clear at the second-order: Are
they to be concepts, properties, attributes, sets? First-order logic, while perhaps accept-
able along these grounds, is however too weak to express substantive mathematical
results, as opposed to second-order systems that contain terms that generalize over the
meanings of predicates. Our novel observations (in section 4) will come on this point;
what we will see is that in the context of a robust system of quantification, (not just
of generalization)—that is, in generalized quantifier theory—a notion of first-orderiz-
ability can be articulated that overcomes the expressive limitations of first-order logic,
but without affecting the semantical features associated with the first order. That is,
the theory of quantification that analytically grounds our initial intuition regarding
the bifurcation of terms, and which affords substantial explanations of the logical and
linguistic properties of quantifiers, also founds the notions needed to derive the core
propositions of cardinal arithmetic.
In what follows, with our framing questions in mind, we will survey aspects of gen-
eralized quantifier theory. There are three parts. In the first, we outline the salient
roots of the theory, noting the emergence of the notion of quantifiers as higher-level
predicates. In the second, we turn to an overview of contemporary developments
directed at articulating set-theoretic foundations for quantifiers, and the relation of
first-order definability to first-order semantics as it emerges in this context. We discuss
here quantifiers that while not first-order definable do have a first-order semantics, and
the importance of these quantifiers for systems that can have significant mathematical
content. In the final section, we consider the extension of the theory from standard to
nonstandard semantics for first-order quantifiers, and note some of the ramifications of
this generalization.

2 The Traditional View

The theory of generalized quantifiers traces its origins all the way back to Aristotle’s
investigation in the Prior Analytics of the logic of the binary determiners Every, Some,
No, and Not all. From a grammatical point of view, these quantifiers are expressions
taking predicates S and P as arguments and returning a sentence as a value (sentences
obtained in this way are traditionally referred to as “categorical propositions”). The
logical relations connecting categorical propositions are codified in the traditional
doctrine of the square of opposition (see Parsons 2006) and theory of the syllogism
(see Figure 3.4.1).

344
 Q U A N TI F I ERS A N D D E T E R MI N E R S

Every S is P No S is P Every S is P No S is P
A contraries E A inner negation E
. .
@ @
@ . @ .
@ . @ .
@ @
@ @
@@ @@
subalterns contradictories subalterns dual outer negation dual
@ @
@ @
@ @
@ @
@ @
? @ ?
@ @@
I subcontraries O I inner negation O
Some S is P Some S is not P Some S is P Some S is not P

Figure 3.4.1 The traditional square of opposition (left) and the modern one (right).

In the traditional square, if two propositions are contradictories, then one of them
must be true and the other false; in other words, they can be obtained from each other
by applying (outer) negation. So, “Every S is P” and “Some S is not P” are contradicto-
ries. If a proposition is subaltern to another one, then it is implied by it. So “Every S is P”
implies “Some S is P,” and “No S is P” implies “Some S is not P.” Thus, the two universal
forms at the top of the traditional square have existential import. If two propositions are
contraries then they cannot both be true (though they can both be false); this is the case
for “Every S is P” and “No S is P” (again, because of the existential import of the posi-
tive universal). And finally, if two propositions are subcontraries then they cannot both
be false (though they can both be true), as is the case for “Some S is P” and “Some S is
not P” (since each is implied by the contradictory of the other one). It should be noted,
however, that a case can be made that only the positive universal was regarded as hav-
ing existential import in ancient and medieval times, a view that is perfectly coherent
(again, see Parsons 2006 for details).
When the four categorical propositions are translated into modern logical notation,
the two universal forms “Every S is P” and “No S is P” lose their existential import,
with the result that only the contradictory connections along the diagonal survive the
transition from the traditional square to the modern one. As mentioned, the contradic-
tory connection can be thought of as “outer negation”: “No S is P” is obtained as the
propositional denial of “Some S is P” (and conversely), and likewise “Some S is not P”
is the propositional denial of “Every S is P” (and conversely). However, the modern
square incorporates also a form of “inner negation,” which applies to the predicate term
P. So the inner negation of “Every S is P” is “Every S is not P” (i.e., “No S is P”) and
conversely the inner negation of “No S is P” is “No S is not P” (i.e.,“Every S is P”). And
in a similar manner “Some S is P” and “Some S is not P” are each the inner negation of
the other one.
The two kinds of negation, inner and outer, can be combined to obtain the dual
transformation: The dual of “Every S is P” is obtained by first taking the inner negation,
yielding “No S is P,” and then the outer negation, yielding “Some S is P.” Equivalently,
the dual of “Every S is P” can also be obtained by taking the outer negation first, fol-
lowed by the inner negation. The dual of “No S is P” can be obtained in a similar man-
ner by applying the two kinds of negation (in either order). As pointed out by Peters

345
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

and Westerståhl (2006, p. 26) any number of applications of the three transformations
to any quantifier in the modern square always leads to another quantifier in the square;
in other words, it’s impossible to land outside of the square by applying the transforma-
tions to any of the quantifiers. Moreover, this is not peculiar to the four Aristotelian
determiners: any quantifier gives rise to its own square of opposition by application of
inner negation, outer negation, and dual.5
The scientific study of the quantifiers as higher-order operators can be found again at
the very beginning of modern logic. In §21 of Frege’s Grundgesetze der Arithmetik we are
asked to consider the forms:
2 a
  a = 4 and   a> 0.
a

These forms of the “conceptual notation” correspond to the modern formulas (a2 = 4)
and a(a> 0) (there exists a square root of 4, and there exists a positive number).
These forms can be obtained from  a  M(a) by replacing the function-name place-
holder M([) by names for the functions [ 2 = 4 and [ > 0 (and the placeholder cannot
be replaced by names of objects or of functions of 2 or more arguments). These two
functions take numbers as arguments and return the value true if those numbers are
square roots of 2 or (respectively) positive, and false otherwise. The kind of arguments
of the functions determines their level; since these arguments are first-level objects (in
this case, numbers) they are referred to as first-level functions. First-level functions that
return the values true and false are called concepts. So the forms displayed above can be
regarded as values of the same function for different arguments. Now, these arguments,
as we just noted, are themselves functions. It follows that the form

  M(a)
a

is a second-level function, because its arguments are first-level functions. Since the argu-
ments of the form  a  M(a) are functions that return truth values (i.e., concepts) the
same holds for the form  a  M(a). Such a form is therefore a second-level concept. But
such a function is also, as one can easily see, a quantifier, for it combines with a variable a
and a formula Mto produce a sentence with a determinate truth value, true or false. This
is not a mere coincidence, but a general fact: from a Fregean point of view, quantifiers
are second-level concepts.

3 The Modern View

The modern view of quantifiers originating with Frege was further developed under the
heading of generalized quantifiers by Mostowski (1957) and Montague (1974). This work
on generalized quantifiers spans linguistics and mathematical logic, the linguists focus-
ing on quantifiers within natural language semantics, and the logicians focusing on the
expressive power and properties such as axiomatizability, decidability etc. The best cur-
rently available systematic treatment of these issues is Peters and Westerståhl (2006).
The modern approach takes a characteristically general stance on the matter. By a
domain (of discourse) we just mean a nonempty set D. A quantifier (viewed as a linguis-
tic—syntactic—entity) can be used in conjunction with a formula to yield a sentence.
In turn, a formula is understood semantically as a subset of D, i.e., the collection of
those d  D that satisfy the formula. Accordingly, a quantifier can be identified with a

346
 Q U A N TI F I ERS A N D D E T E R MI N E R S

collection of subsets of D: for instance, the existential quantifier can be identified with
the collection of all nonempty subsets of D, and the universal quantifier with the col-
lection containing D itself as its only member. To see how this works, consider that a
sentence of the form xM (x) is true over D precisely when every d  D satisfies M, i.e.,
when the extension of M (x) over D is D itself. Hence,  can be identified, semantically,
with {D}. Similarly, for the existential quantifier. In general, a quantifier Q over D, then,
is just a collection of subsets of D: Q Ž ᏼ(D):

1.  = {D};
2.  = {X Ž D: X zÀ};
3. !k = {X Ž D: |X| = k};
4. John = {X Ž D: John  X}.

The quantifier !k (“there are exactly k objects such that . . .”) denotes the collection of
all k-membered subsets of D. The last item above is referred to as a “Montagovian indi-
vidual”: It was Montague who proposed to identify proper names such as “John” with par-
ticular quantifiers, i.e., with those subsets of the domain of which John is a member. Then
“John runs” is true if and only if the predicate “runs” falls under the quantifier John i.e., if
and only if “runs” is among the subsets containing John, i.e., if and only if John runs.
Some quantifiers are best viewed as n-ary relations over ᏼ(D). In fact, binary rela-
tions over ᏼ(D) play a particularly important role, in that they supply denotations for
ubiquitous natural language determiners. A determiner is an expression such as “every,”
“no,” “most,” “between 3 and 5,” etc., which combines with a predicate to form a noun
phrase. Here are some examples beginning with the Aristotelian determiners:6

• Every A is B: Every = {(A,B): A Ž B}

• Some A is B: Some = {(A,B): A B z‡}
• Most A are B: Most = {(A,B): |A B| > |A − B|};
• Twice as many A as B are C: Twice = {(A,B,C): |A C| = 2 ·|B C|}.

It is well known that some binary quantifiers can be represented by means of the cor-
responding unary quantifier applied to a Boolean combination of their arguments:

Every[A,B]= [−A  B],

but for instance Most cannot be so represented. In fact, Most is not even definable in
first-order logic, i.e., there is no formula M of first-order logic using 1-place predicates A
and B as the only nonlogical constants that is true (over some domain D) if and only
if (in D) most A’s are B’s. More generally, we say that a (binary) quantifier Q(A,B) is
first-order definable over D in a language ᏸ if and only if there is a formula M ᏸ (P,Q)
such that

(D, A,B)|= M,

where ᏸ(P, Q) is the language obtained from ᏸ by adding two new predicate symbols
P and Q, and (D, A,B) is the ᏸ-structure with domain D assigning A and B to P and
Q (respectively) as their denotations. For instance, At least two, as students learn in
introductory logic courses, is first-order definable in the pure language of identity.

347
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

A quantifier Q(A,B) is proportional (over finite D) if and only if there are integers n
and m such that Q(A,B) holds precisely when at least of the A’s are B’s:

|AB|  m
Q(A,B) Ÿ t n
|A|

The quantifiers Most, At least half, More than 10%, . . . are all proportional. Propor-
tional quantifiers are not first-order definable.
Quantifiers are distinguished by their arity as well as by the number of their dimen-
sion: The former is the number of arguments (formulas) they take, and the latter is
the number of variables of such formulas. Every(A,B) is a binary monadic quantifier,
whereas Q(R) = 1 œxyRxy is a unary dyadic quantifier. In general, the type of a
quantifier Q can be represented by a k-tuple <n , . . ., nk>, where k represents the arity,
1
and ni the dimension of the i-th argument. Formally:
k
Q (n1, . . ., nk) ŸQ Ž  ᏼ(D i).
n

i = 1
A special case that will be of interest to us is that of monadic quantifiers, i.e., a quantifier
of type <1, . . ., 1>; if the type specification contains n occurrences of “1”, then we have
an n-nary monadic quantifier.
The quantifiers , , !k are all of type <1>; the Aristotelian determiners are of type
<1,1>, the quantifier Q defined above is of type <2>, etc. Since the last one is often
called a “dyadic” quantifier, the dimension of a quantifier is sometimes referred to, in
analogy with its “arity,” as the “adicity.”
The distinction between first-and second-order quantifiers can also be made explicit
by the assignment of types. Consider the sentence PI(P). If we identify the open for-
mula I(P) (with P as a free variable) with the class of predicates that satisfy it, then the
sentence is true if some P satisfies I, i.e., if the class of predicates satisfying I(P) is not
empty. It follows that the second-order quantifier  denotes the collection {X  ᏼ2(D):
X z À} of all nonempty collections of predicates (here ᏼ2(D) is the power-set of the
power-set of D). Since the class of predicates satisfying M(P) is a first-order quantifier
(of type <1> if P is a 1-place predicate), then the second-order quantifier  has type
<<1>>. It’s important to stress that the distinction between first- and second-order quantifiers
is semantical, not merely notational. The distinction also does not coincide with whether a
given quantifier is representable in first-order logic, as in the case of the quantifier Most.
We take the time to review here several important properties that have been singled
out for binary quantifiers Q(A,B):

• Conservativity: Q(A,B)= Q(A,A  B);

• Right monotony: Q(A,B) and B Ž C implies Q(A,C)(All, Most);
• Left monotony: Q(A,B) and A Ž C implies Q(C,B);
• Right anti-monotony: Q(A,B) and C Ž B implies Q(A,C)(No, Few);
• Left anti-monotony: Q(A,B) and C Ž A implies Q(C,B).

Virtually all quantifiers that provide denotations for natural-language determiners are
conservative. For instance, “Some A’s are B’s” is equivalent to “Some A’s are A’s that
are B’s.” The one notable exception is Only, a fact on which much has been written
and that has led some to question whether “only” really denotes a quantifier. Some

348
 Q U A N TI F I ERS A N D D E T E R MI N E R S

quantifiers such as All and Most, are right-monotonic: If “All A’s are B’s” and A Ž C,
then also “All A’s are C’s” (and similarly for “most”).
We are all familiar with the all-important phenomenon of quantifier-dependence,
in which quantifiers can be applied in sequence to a particular predicate as in, e.g.,
xyRxy. We have not yet explained how this can be represented within the framework
of the theory of generalized quantifiers, but it’s intuitively clear how such an account
needs to go. Given R and an object a from the domain, we need to form the collection
Ra of all those objects b such that Rba; then xyRxy holds precisely when the set of a’s
for which Ra falls under the quantifier , falls under . In other words, the combination
of quantifiers  generates a dyadic quantifier that holds of a binary relation R precisely
when for every a there is a b such that Rab.
The notion of iteration provides a generalization of this construction, by allowing us
to combine quantifiers in order to obtain new ones (see Higginbotham and May 1981).
Suppose Q is a quantifier of type <m> (i.e., a quantifier taking m-place relations as
1
arguments) and Q of type <n>. The iteration of Q and Q , denoted by the juxtaposi-
2 1 2
tion Q1 Q2 is a quantifier of type <n + m> defined as follows (where R is a relation of
n + m places and the first two items are convenient abbreviations):

Ra1 . . . am(b1, . . ., bn) Ÿ R(a1, . . ., am, b1, . . ., bn)

R 2(a1, . . ., am) Ÿ Q2(Ra1 . . . am)
Q

Q Q (R) Ÿ Q (R 2)
Q
1 2 1

Notice that composition is associative (but not commutative), so we can write Q Q . . .

1 2
Qn without worrying about parentheses.
Remarkably, the notion of iteration can be used also in the analysis of noun phrases,
when these are interpreted as denoting Montagovian quantifiers. Recall that on such
an account noun phrases such as “John” denote the collection of all subsets X Ž D such
that John is a member of X. According to the above account, the iteration of noun
phrases now gives arity-reducing (projection) operators taking n-place relations to m-
place relations, where m dn (identifying the special case of 1-place relations with the
corresponding sets of elements and the special case of 0-place relations with the truth
values true and false). Then the value of

John(Mary (kissed))

is true if and only if John is a member of kissedMary, i.e., if and only if John kissed Mary,
and false otherwise.
We now turn to a particularly important property of quantifiers, permutation invari-
ance. Invariance under permutations was first identified by Tarski (1986) as charac-
terizing logical notions, on the idea that such notions are independent of the subject
matter and hence insensitive to permutations of the first-order domain. There is wide-
spread consensus that this kind of invariance provides at least a necessary condition for a
notion’s logical character. Invariance under permutations can be applied to the modern
notion of quantifier in order to single out those quantifiers that are logical. By a permuta-
tion of the domain D of objects we mean a function S that maps D onto itself in a one-to-
one fashion. Such a permutation can be “lifted” to subsets of D by taking, for each set A,
its point-wise image S[A] under S, defined as {S(x): x  A}. A quantifier Q is invariant

349
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

under permutations precisely when, for subsets A and B of D, Q(A,B) holds iff Q(S[A],
S[B]) holds (similarly for type <1> quantifiers). For instance, the existential quantifier
 is permutation invariant since S[A] is nonempty whenever A is nonempty. On the
contrary, the Montagovian quantifier John is not, as expected, permutation invariant,
since Scan map a set containing John to one that does not contain it.
Finally we mention the case of cardinality quantifiers, i.e., quantifiers that explicitly
deal with cardinality constraints between the sets they relate. Historically, the first such
quantifiers were the Rescher quantifier R and the Härtig quantifier I (introduced by
Rescher (1962) and Härtig (1965), respectively). Their definitions are as follows:

• R(A,B) œ|A| > |B|;

• I(A,B) œ|A| = |B|.

These quantifiers have been extensively studied from a mathematical point of view
(see Herre et al. (1991) for a survey). Härtig’s quantifier is definable from Rescher’s by
putting I(A,B) œ ™R(B,A) š™R(A,B) (which requires the axiom of choice) but not
vice versa. While they are both obviously permutation invariant, neither quantifier is
conservative, as the following both fail:

• I(A,B) œ I(A,A  B);

• R(A,B) œ R(A,A  B).

In a similar vein, one can introduce a quantifier F, the “Frege quantifier” of Antonelli
(2010a), with the semantics

• F(A,B) œ|A|d|B|.

(The marginal advantage of F over R is that the Härtig quantifier can be directly repre-
sented without assuming the axiom of choice.)
One deceivingly natural way to read F(A,B) is “For every A there is a B” (see Boolos
1981). Such a reading is potentially misleading because it is tantalizingly close to some-
thing that one might want to express by means of alternating universal and existential
quantifiers, while in fact F cannot be so interpreted (it is not even formalizable at the
first order, as it interprets the Härtig quantifier).
At this juncture, it is important to bear in mind that, since all three cardinality
quantifiers have type <1,1>, they are, from a semantical point of view, first order. This
fact might come across as somewhat surprising, since, for instance, I(A,B) would seem
to require the existence of a one-to-one function mapping A onto B, a notion that is
usually regarded as second-order. In this respect it is important to introduce a distinc-
tion between what a statement of the form Q(A,B) asserts and what it expresses. The
statement I(A,B) does express the existence of such-and-such a function, but it does not
assert it (asserting the existence of such a function is properly the purview of higher-
order notions). From a purely semantical point of view, the Härtig quantifier is not any
different from an uncontroversially first-order quantifier, such as Every. The statement
Every(A,B) also expresses the existence of a purportedly higher-order entity, namely, a
function f from A to B such that f (a) = a for every a  A, but it certainly does not assert
the existence of such a function. The fact that both the Härtig quantifier I and the
Aristotelian determiner Every (as well as the Rescher and Frege quantifiers) are first

350
 Q U A N TI F I ERS A N D D E T E R MI N E R S

order should not, of course, lead us to overlook the fact that they are vastly different in
expressive power (see Herre et al. 1991 for details). Whereas Every does not outstrip the
expressive capabilities of standard first-order logic, the Härtig quantifier allows us, for
instance, to provide a complete characterization of the natural numbers, a feat which
lies far beyond the bounds of first-order logic.

4 The Non-Standard View

The upshot of the last section is that both first-and second-order quantifiers depend for
their meaning on having access to the collection of subsets of D. The difference lies in
the fact that first-order quantifiers express the existence of particular subsets, whereas
second-order quantifiers assert it. For instance, the sentence xM(x) expresses the fact
that the extension of M(x) is among the nonempty subsets of D; but in order to assert
that the extension of M(x) is among the subsets of D, one needs a second-order quanti-
fier, as in the following instance of the comprehension principle:

P x(Px l M(x)).

The fact remains that the specification of the meaning of both first-and second-order
quantifiers is equally dependent on the specification of a second-order domain com-
prised of subsets of the first-order domain.
It has been known since the work of Henkin (1950) that the second-order quan-
tifiers, in addition to the standard semantics, also admit of a general, or nonstandard
semantics, in which the second-order domain is not given by the full power set ᏼ(D),
but it is rather constituted by a collection D´ of subsets—as long as such a collection
satisfies certain natural closure conditions (see Enderton 2009). On such a nonstand-
ard semantics, the second-order language is then interpretable into a two-sorted first-
order language comprising (first-order) variables for objects and predicates along with a
membership relation , thereby regaining many of the desirable properties of first-order
logic. It went however apparently unnoticed until the work of Antonelli (2007) that
first-order quantifiers also admit of such a nonstandard semantics.
Consider for instance the ordinary first-order quantifier . On the standard interpre-
tation,  denotes the collection of all nonempty subsets of D. It is possible, however,
to provide also a nonstandard interpretation according to which denotes an arbitrary
collection of nonempty subsets (the collection itself may be empty, although its mem-
bers—if any—may not). Similarly, in order to preserve the duality of  and , on the
nonstandard interpretation  denotes an arbitrary collection of subsets subject only to
the proviso that D itself must be a member of that collection.7 A similar account can
be given also for the Frege quantifier F (which is also first order), where a statement
F(A,B) holds on a nonstandard interpretation not just if |A|d|B|, but if in addition
this fact is witnessed by an injective function f: A o B to be found in a given collection
of functions also supplied by the interpretation (such a collection is also assumed to
satisfy a number of specific constraints—see Antonelli 2010b for details).
One could naturally regard these alternatives not as specifying a “nonstandard”
interpretation for, say,  or F, but as providing the semantics for altogether different
quantifiers with little, if any, connection to the standard  and F. But it is perhaps
conceptually more interesting to recast the situation in terms that make it clear that
while the interpretation of the quantifier remains fixed, it is the underlying semantics

351
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

that is allowed to vary in nonstandard ways. In other words, the same quantifier or F
is being provided either a standard or non-standard semantics. For simplicity, let us
restrict our attention to monadic quantifiers, i.e., quantifiers of type <1, . . ., 1>. Now a
(general) interpretation is a pair (D, D´), where D is a nonempty first-order domain and
D´ Ž ᏼ(D) is the second-order domain. The interpretation is standard if D´ = ᏼ(D).
It is natural in certain contexts to assume that D is closed in some relevant sense, e.g.,
under first-order definability. On such an interpretation (D, D´), a type <1> quantifier
Q is a collection of members of D´, a type <1,1> quantifier is a relation over D´, etc.
Then we can take , say, to select all and only the non-empty members of D´. On such
an account,  receives the standard interpretation on standard domains, and the general
interpretation on nonstandard ones. Thus we can carry over wholesale the classifica-
tion of section 3, automatically providing a nonstandard interpretation for each of the
quantifiers mentioned therein. It should be evident that what this does is to provide an
alternative interpretation for the same first-order quantifiers , , Every, Some, etc.,
rather than introducing new quantifiers altogether.
More importantly, the setup just outlined, in which an interpretation provides both
a first-and a second-order domain, allows us to clarify the sense in which first-order
quantifiers express, while second-order ones assert, the existence of certain subsets of the
domain. In both cases it is clear that the interpretation of the quantifier depends on the
specification of the domain D´ of subsets of D; but while first-order quantifiers range
directly over D´ (they denote collections of members of D´), second-order quantifiers
range over the power-set of D´ (they denote collections of collections of members of
D´). Notice that we thus have also a form of general Henkin semantics for the second-
order quantifiers, where a second-order quantifier is now taken to denote a subset of
the true power-set ᏼ(D´) of D. Essentially, this amounts to building a standard inter-
pretation of the second-order quantifier on top of a nonstandard interpretation for the
first-order ones.8

5 Concluding Remark
We have thus come to a very general understanding of first-order quantifiers and the
role they play in providing denotations for natural-language determiners. The great
advance of the modern view was to identify quantifiers with collections of, or rela-
tions over the second-order domain. This account allowed a great many of such con-
structions to be properly recognized as quantifiers, from the Aristotelian determiners to
cardinality quantifiers, to Montagovian individuals; it also allows us to draw a range of
semantic distinctions among the quantifiers, including a specification (via permutation
invariance) of the logical terms as a subclass. But this view turns out to be even further
generalizable, by allowing the second-order domain to fall short of the full power-set of
the first-order domain—a possibility that has been long acknowledged for second-order
quantifiers, but not for first-order ones. First-order quantifiers, we have emphasized, are
just as dependent on a prior notion of second-order domain as the second-order quanti-
fiers, but in a different way.
Whereas second-order quantifiers assert the existence of subsets of the first-order
domain having certain set-theoretic properties, first-order quantifiers merely express
their existence. And it is this difference, we are proposing, that is at the heart of the
difference in the order of quantifiers.

352
 Q U A N TI F I ERS A N D D E T E R MI N E R S

Notes
* We would like to thank Robert Hirsch for very helpful comments.
1 That concern was in finding a logical way of denoting numbers without engaging the paradoxes, that is,
the project ultimately of Principia Mathematica. But Russell, driven by his epistemological conception,
also sought to efface the distinction. By insisting that all names must be “logical”– they can only name
that with which there is acquaintance—Russell was forced to analyze terms that are intuitively names as
abbreviations for definite descriptions, and hence ultimately by the logical machinery of quantifcation
(see Russell 1919).
2 We assume the interdefinability of existential generalization with respect to negation. Accordingly, we
could have assumed that that the existential is the only quantifier, and defined the universal.
3 This tradition in linguistics stems from the seminal papers by Higginbotham and May (1981), Barwise
and Cooper (1981) and Keenan and Stavi (1986). For the classic discussion of scope and logical form in
natural language, see May (1985). For an insightful overview, see Glanzberg (2006).
4 The chief skeptic being Quine (1970).
5 In actuality, this result is a somewhat narrower, as it obtains only for quantifiers of type <1,1>; see below
for discussion of quantifier types.
6 van Benthem (1983) introduces the notion of a “tree of determiners” as an intuitive way of character-
izing the logical space of set-theoretic relations that demarcates the class of possible determiners. For an
extended discussion, see Peters and Westerståhl (2006), pp. 160ff.
7 One could ask—although the question is somewhat tangential to the overall direction of this contribu-
tion—about the set of sentences that turn out to be valid on the nonstandard interpretation of  and .
The somewhat surprising answer of Antonelli (2007) is that this is the same as the set of validities of
positive free logic (the reason being that the non-standard semantics is equivalent to the outer-domain
semantics of positive free logic).
8 This kind of general interpretation is somewhat narrower than that provided by Henkin semantics for
the second-order quantifiers (thus yielding a stronger logic), for while every second-order interpretation
obtained in this way is a general model in Henkin’s sense, the converse is not necessarily true (although
it’s easy to see how to accommodate a non-standard interpretation for both the first-and second-order
quantifiers).

References
G. A. Antonelli, Free quantification and logical invariance, Rivista di estetica, 33(1): 61–73, 2007 (special
issue “Il significato eluso. Saggi in onore di Diego Marconi,” M. Andronico, A. Paternoster, A. Voltolini
eds).
G. A. Antonelli, The nature and purpose of numbers, The Journal of Philosophy, 107(4): 192–212, 2010a.
G. A. Antonelli, Numerical abstraction via the Frege quantifier, Notre Dame Journal of Formal Logic, 51(2):
161–79, 2010b.
J. Barwise and R. Cooper, Generalized quantifiers and natural language, Linguistics and Philosophy, 4(2):
159–219, 1981.
G. Boolos, For every A there is a B, Linguistic Inquiry, 12: 465–67, 1981.
H. Enderton, Second-order and higher-order logic, in E. Zalta, editor, Stanford Encyclopedia of Philosophy,
2009, URL: http://plato.stanford.edu.
M. Glanzberg, Quantifiers, in E. Lepore and B. Smith, editors, The Oxford Handbook of Philosophy of Lan-
guage, pages 794–821, Oxford University Press, 2006.
H. Härtig, Über einen Quantifikator mit zwei Wirkungsbereichen, in L. Kalmár, editor, Colloquium on the
foundations of mathematics, mathematical machines and their applications, pp. 31–36, Akadémiai Kiad—,
Budapest, 1965.
L. Henkin, Completeness in the theory of types, Journal of Symbolic Logic, 15(2): 81–91, 1950.
H. Herre, M. Krynicki, A. Pinus, and J. Väänänen, The Härtig quantifier: a survey, Journal of Symbolic Logic,
56(4): 1153–83, 1991.
J. Higginbotham and R. May, Questions, quantifiers and crossing, Linguistic Review, 1(1): 41–80, 1981.
E. Keenan and J. Stavi, A semantic characterization of natural language determiners, Linguistics and Philoso-
phy, 9: 253–326, 1986.
R. May, Logical Form: Its Structure and Derivation, MIT Press, Cambridge, MA, 1985.
R. Montague, English as a formal language, in R.H. Thomason, editor, Formal Philosophy, Yale University
Press, 1974, originally published 1969.

353
 G . A LDO A N TO N ELLI A N D R O B E RT C . MAY

A. Mostowski, On a generalization of quantifiers, Fundamenta Mathematicæ, 44: 12–36, 1957.

T. Parsons, The traditional square of opposition, in E. Zalta, editor, Stanford Encyclopedia of Philosophy, 2006,
URL: http://plato.stanford.edu.
S. Peters and D. Westerståhl, Quantifiers in Logic and Language, Oxford University Press, 2006.
W. V. O. Quine, Philosophy of Logic, Prentice-Hall, 1970.
N. Rescher, Plurality quantification, Journal of Symolic Logic, 27: 373–47, 1962.
B. Russell, Knowledge by acquaintance and knowledge by description, in Mysticism and Logic and other Essays,
Longmans, Green and Co., 1919.
A. Tarski, On the concept of logical consequence, in J. H. Woodger, editor, Logic, Semantics, metamathema-
tis. Papers from 1923 to 1938 by Alfred Tarski, pages 409–20, Oxford University Press, 1956, Reprinted in
1983 by Hackett, J. Corcoran, ed.
A. Tarski, What are logical notions?, History and Philosophy of Logic, 7: 143–154, 1986, original text 1966;
edited (with an introduction) by John Corcoran.
J. van Benthem, Determiners and logic, Linguistics and Philosophy, 6(4): 447–78, 1983.

354
 3.5
GE NE R IC S
Sarah-Jane Leslie

Introduction
Generics are statements such as “doctors heal people,” “a tiger is striped,” “the dodo is
extinct,” “a duck lays eggs,”,“kettles boil water,” and “mosquitoes carry the West Nile
virus.” Generic statements express general claims about kinds and categories, rather
than claims about particular individuals. In English, generics can be expressed using a
variety of syntactic forms: bare plurals (e.g. “ducks lay eggs”), indefinite singulars (e.g. “a
tiger is striped”), and definite singulars (“the dog is a mammal”).
Quantified statements, such as “all dogs are mammals” or “most tigers are striped,”
can also be used to express general claims about kinds. However, quantified statements,
unlike generics, can be used to answer the question how much or how many. For exam-
ple, if asked “How many tigers are striped?”, one could felicitously reply by saying “all/
most tigers are striped.” One could not felicitously reply by using the generic “tigers are
striped” (Carlson 1977).
The truth conditions of generics have proved quite elusive for semanticists. For
example, “dogs are mammals” seems to require for its truth that all (possible) dogs be
mammals. “A tiger is striped” or “ravens are black,” however, are somewhat more forgiv-
ing, since they are compatible with the existence of a few stripeless tigers (as Siegfried
and Roy’s performances attest), and white albino ravens. “Ducks lay eggs” and “a lion
has a mane” are more forgiving still; these generics are true even though it is only the
mature members of one gender which possess the relevant properties. This truth condi-
tional laxity is limited in scope, however: We do not accept “ducks are female” or “lions
are male,” even though every egg laying duck is a female duck, and similarly mutatis
mutandis for maned lions. Finally, we accept “mosquitoes carry the West Nile virus,”
even though fewer than 1 percent of mosquitoes carry the virus, while also rejecting
“books are paperbacks,” when over 80 percent of books are paperbacks. The correct
analysis of the truth conditions for generics is a matter of great controversy among theo-
rists working on the problem.

Generics vs. Existentials

The interpretation of sentences containing bare plurals, indefinite singulars, or definite
singulars can be either generic as in (1) existential/specific as in (2):
 SA RA H - J A N E L E S L I E

(1) Tigers are striped

A tiger is striped
The tiger is striped.

(2) Tigers are on the front lawn

A tiger is on the front lawn
The tiger is on the front lawn.

The subjects in (1) are prima facie the same as in (2), yet their interpretations in (1) are
intuitively quite different from those in (2). In (2) we are talking about some particular
tigers, while in (1) we are saying something about tigers in general.
There are some tests that are helpful in distinguishing these two readings. For exam-
ple, the existential interpretation is upward entailing, meaning that the statement will
always remain true if we replace the subject term with a more inclusive term. For exam-
ple, if it is true that tigers are on the lawn, then it will also be true that animals are on
the lawn. This is not so if the sentence is interpreted generically. For example, it is true
that tigers are striped, but it does not follow that animals are striped (Lawler 1973; Laca
1990; Krifka et al. 1995). Another test concerns whether we can insert an adverb of
quantification (in the sense of Lewis 1975,) with minimal change of meaning (Krifka et
al. 1995). For example, inserting “usually” in the sentences in (1) (for example “tigers
are usually striped”) produces only a small change in meaning, while inserting “usually”
in (2) dramatically alters the meaning of the sentence (for example “tigers are usually
on the front lawn). (For generics such as “mosquitoes carry malaria,” the adverb “some-
times” is perhaps better used than “usually.”)

Stage-Level and Individual-Level Predicates

We might wonder why the interpretations of (1) are so different from the interpretations
of (2). The most prominent explanation is that, at least for bare plural (BP) generics, the
interpretations depend on whether the predicates in question are stage-level or individual-
level. In particular, stage-level predicates are thought to give rise to existential interpreta-
tions, while individual-level predicates give rise to generic ones (Carlson 1977).
Intuitively, the distinction between these two types of predicates has to do with
whether the predicate denotes a property that may well be had fleetingly (making for a
stage-level predicate) versus a property that is more stable and longlasting (making for
an individual-level predicate). Examples of stage-level predicates include the predicates
‘is drunk’, ‘is barking’, and ‘is on the lawn’—these properties are normally had only tem-
porarily, or at least intermittently. Individual-level predicates express more stable and
persistent properties, e.g. ‘is tall’, ‘is a mammal’, and ‘is female’.
The distinction between stage-level and individual-level predicates can also be drawn
on the basis of their differential acceptability in a variety of constructions. For example,
only stage-level predicates can be used to fill in “_” in “A saw B _”. That is, we can say
“John saw Mary drunk/barking/on the lawn,” but not “John saw Mary tall/a mammal/
female.” Similarly, only stage-level predicates can be used in the construction “there
are Ks _”: “there are firemen available/outside the window,” but not “there are firemen
male/tall” (Milsark 1974; Carlson 1977; Stump 1985).
The details of exactly how these different types of predicate may give rise to generic
vs. nongeneric interpretations of these sentences is beyond the scope of this entry, as

356
 G EN ERI C S

the proposals tend to involve technical issues in syntax and linguistic semantics. The
original proposal is due to Greg Carlson (1977), however Carlson himself later came
to reject the basics of his account (Carlson 1989). More recent discussion can be found
in Diesing (1992), Chierchia (1995), Kratzer (1995), Cohen and Erteschik-Shir (1997,
2002), Kiss (1998), and others.

Generics and Logical Form

There is some controversy in semantics as to the details of the logical form of generic
sentences, however there is also general (though not universal) agreement about the
broad features of these logical forms. This section highlights the points of general agree-
ment (see Chapter 1.3 Logical Form).

D-Generics and I-Generics

It is helpful to separate out two categories of generic statements. Our first category
includes statements such as “tigers are striped,” “ravens are black,” “a lion has a mane,”
and “the dog is carnivorous,” These statements are naturally thought of as expressing
generalizations about individual members of the kind. For example, we might suppose
that “tigers are striped” is made true by enough individual tigers possessing stripes. The
exact nature of these generalizations is highly controversial.
In our second category, we have statements like “dinosaurs are extinct,” “the dodo
is extinct,” “tigers are widespread,” and “the domestic cat is common,” which are often
thought to predicate a property directly of the kind in question, rather than expressing
generalizations concerning its members. For example, in saying “dinosaurs are extinct,”
one says something about the kind dinosaur: namely that that kind of thing is extinct.
We may notice that it is not possible to say of an individual dinosaur Dino that Dino is
extinct, since only a kind can be extinct.
Examples of this second category are often referred to as D-generics (‘D’ for ‘definite’),
while examples of the first are known as I-generics (‘I’ for ‘indefinite) (Krifka 1987). It
should be noted that the indefinite singular form cannot be used to express a D-generic:
Statements such as “a dinosaur is extinct” and “a tiger is widespread” are not felicitous.
Much of the work on the semantics of generics has been focused on I-generics, since
they have proved the most elusive. It is widely accepted that D-generics are singular
statements which predicate properties directly of kinds. For example, “tigers are extinct”
predicates the property of being extinct directly of the kind Panthera tigris, and would be
true just in case Panthera tigris had the property of being extinct (Krifka et al. 1995). The
semantics of I-generics have proved much less tractable.

The Logical Form of I-Generics

For the remainder of this chapter, we will be concerned only with I-generics, since they
have received the most attention, and been the subject of most controversy. Hence-
forth, I will use the term ‘generics’ to mean specifically I-generics. This section discusses
the dominant theory of the logical form of I-generics, and of necessity involves some
technical material. For readers unfamiliar with the general framework, it will be helpful
to first consult the chapters on Logical Form (1.3), Quantifiers and Determiners (3.4),
and Adverbs (3.9). It is generally agreed that that there is a two-place operator, usually

357
 SA RA H - J A N E L E S L I E

termed Gen, which functions as an adverb of quantification (Lewis 1975; see Chapter
3.9 on Adverbs). Examples of adverbs of quantification include ‘usually’, ‘generally’,
‘sometimes’, and so on. They relate one set of conditions containing at least one free
variable to another set. The variables may range over individuals, or over cases or situ-
ations, as in (3), whose Lewisian analysis is given by (4):

(3) Usually, when John comes home, he sleeps

(4) Usually s [John comes home in s] [John sleeps in s]

This yields a tripartite structure, consisting of a quantifier (“Usually s”), a restrictor (“John
comes home in s”), and a scope (“John sleeps in s”). In this example, the material from
the when-clause is in the restrictor, and the material from the main clause is in the
scope.
Generics are standardly assimilated to this model. A common assumption (Heim
1982; Kamp 1981; Kamp and Reyle 1993; Diesing 1992; Kratzer 1995; and many others)
is that indefinites such as bare plurals and indefinite singulars contribute predicates with
unbound variables to the logical forms containing them. During syntactic processing,
the material in the sentence is divided up into the restrictor and scope (e.g. Diesing
1992). If the sentence contains a quantificational adverb (e.g. ‘usually’, ‘always’), then
any unbound variables in the restrictor are bound by that quantificational adverb, as in
example (4) above. However, if there is no quantificational adverb, the generic operator
Gen is introduced to bind those variables. Thus, to take a simple example, the logical
form of a sentence like “ravens are black” may be given as follows:

(5) Gen x [Ravens(x)] [Black(x)]

Even theorists who do not agree with the assumption that indefinites contribute just
predicates and unbound variables to their logical forms (see, for example, Chierchia
1998) tend to agree that a tripartite structure nonetheless is the correct analysis for
generics. (However, Liebesman (2011) disputes this claim, arguing that I-generics
should have the same logical forms as D-generics.)
A significant virtue of the tripartite structure is that it is readily able to accommodate
intuitions of ambiguity, such as those associated with sentences such as “typhoons arise
in this part of the Pacific,” which can be understood as either “typhoons in general have
a common origin in this part of the Pacific,” or as “there arise typhoons in this part of
the Pacific” (Carlson 1989). These different readings correspond to two different logical
forms, given schematically by (6) and (7) respectively:

(6) Gen x [x is a typhoon] [x arises in this part of the Pacific]

(7) Gen e [e is a (contextually relevant) event involving this part of the Pacific]
[e involves the arising of typhoons]

The term ‘generic’ is sometimes extended to sentences such as “Mary smokes after work,”
since these habitual statements share various features with generics, although they do
not express generalizations concerning kinds. It is often believed that these statements
should be analyzed with Gen (for example, Lawler 1972; Schubert and Pelletier 1989;
Krifka et al. 1995; and many others):

358
 G EN ERI C S

(8) Gen e [Relevant-event-involving-Mary(e) & Occurs-after-work(e)]

[Event-of-Mary-smoking(e)]
For the remainder of this chapter, the term ‘habitual’ rather than ‘generic’ will be used
for these statements.

The Semantics of Generics

A great deal of work has been done on the semantics of (I-)generics, particularly on bare
plural (I-)generics. It is easy to see why this is so: “Ducks lay eggs” is a true generic, while
“ducks are female” is false, yet it is only the female ducks who ever lay eggs. “Mosquitoes
carry the West Nile virus” is true, and “books are paperbacks” is false, yet less than one
percent of mosquitoes carry the virus, while over 80 percent of books are paperbacks.
How are we to account for these puzzling facts?
It is clear that generics are not equivalent to universal statements, but rather permit
exceptions—that is, generics can be true even if some (or sometimes many) members of
the kind lack the property in question. Generics also do not mean ‘most’; it is false that
most mosquitoes carry the West Nile virus and true that most books are paperbacks, but
our intuitions about the truth/falsity of the corresponding generics are reversed.
As these examples suggest, Gen cannot be analyzed as sharing a meaning with any
of the standard quantifiers. One question that comes up immediately is whether Gen
can be considered a quantifier in any sense. Carlson (1977) argued informally that it
could not, since generics do not tell us how much or how many. He notes that, if asked
“how many tigers are striped,” one could reply “most/all/many/some tigers are striped,”
but not simply “tigers are striped.” Leslie (2007) offers a more formal argument to the
effect that Gen cannot be considered a quantifier in any of the standard senses, since it
is not permutation invariant (see Chapter 3.4 Quantifiers and Determiners), nor does it
partition the power set of the domain of discourse.
A significant number of theories concerning the meaning of generics have been offered
over the years. Most theories have focused primarily on bare plural generics, though
some theories are intended to cover indefinite and definite singulars too. I will conclude
with some remarks concerning issues specific to indefinite and definite singulars.

Possible Worlds and Normalcy-Based Approaches

Many accounts of generics are framed in terms of some or other type of quantification
over possible worlds (Dahl 1975; Delgrande 1987; Schubert and Pelletier 1989; Asher
and Morreau 1995; Krifka et al. 1995; Pelletier and Asher 1997; Greenberg 1998; and
others; also see Chapter 2.6 Possible Worlds Semantics). Often, these possible worlds are
employed to capture the intuition that generics tell us something about what is normal
for members of a kind (see Nickel 2008). For example, it is natural to think that a generic
such as “tigers are striped” tells us something about normal tigers; the only exceptions to
it are those tigers who are albino, and so in some respect out-of-step with the norm for
the kind. Similarly, “dogs have four legs” may strike us as true because the only dogs who
do not have four legs either have birth defects or have met with misfortune.
Possible worlds are helpful here, because they allow us to consider, say, worlds in
which things go as normally as possible for a given tiger, even if life is actually quite
abnormal for that tiger. For example, Asher, Morreau, and Pelletier argue that “Ks I”
is true iff for each individual K, the most normal worlds for that K (according to a

359
 SA RA H - J A N E L E S L I E

contextually determined ordering base, see Chapter 2.6 Possible Worlds Semantics), are
such that that K Is (Asher and Morreau 1995; Pelletier and Asher 1997). Glossed in
more intuitive terms, the account states that a generic “Ks I” is true iff each individual
K would have the property I if all was to go as normally as possible for that K. Thus,
while in actuality some dogs are three-legged, one might suppose that, had things gone
more normally for each of those dogs, they would have had four legs.
Pelletier, Asher, and Morreau connect their semantic account of generics to the liter-
ature on defeasible validity (McCarthy 1986; Reiter 1987, and others). A set of premises
defeasibly or nonmonotonically entails a conclusion if the conclusion is likely or reason-
able given the premises, yet it is nonetheless possible for the premises to be true and the
conclusion false. They argue that there are many defeasibly valid patterns of entailment
that involve generics, and that their semantics explains why this is so. For example, if
Fido is a dog and dogs have four legs, then the inference that Fido has four legs is held to
be defeasibly valid. It is possible that Fido is a three-legged dog, but it is still reasonable to
suppose that Fido is four-legged on the basis of the premises. The reasonableness of this
inference is explained on Pelletier, Asher, and Morreau’s semantics—the generic “dogs
have four legs” guarantees that, if things go normally for a given dog, then that dog will
have four legs. Since we have been given no reason to suppose that Fido is abnormal,
one may defeasibly conclude that Fido is four-legged.
Indisputably, a large number of true generics of the form “Ks I” are such that, if things
go normally for a given K, then it has the property I. However, there would seem to also
be a number of true generics which do not fit this profile: “ducks lay eggs,” “lions have
manes,” “mosquitoes carry the West Nile virus,” and “sharks attack swimmers.” The lat-
ter two generics present a significant difficulty for normalcy-based accounts, since it is
surely not normal for a mosquito to carry the virus, nor is it normal for a shark to attack
a bather. For generics such as “ducks lay eggs” and “lions have manes,” several theorists
have argued that the domain of discourse is restricted so that we are talking about only
the female ducks and the male lions respectively, and thus aim to rescue normalcy-based
approaches from this criticism (for example Pelletier and Asher, 1997), since perhaps it
is abnormal for a given female duck to fail to lay eggs. Leslie (2008) argues that if such
domain restriction were available, then a number of false generics would be predicted to
be true. Khemlani, Leslie, and Glucksberg (2009) report empirical evidence that further
suggests that people do not understand these generics to involve domain restriction. If
one cannot appeal to domain restriction to handle “ducks lay eggs,” then such generics
would appear to constitute counterexamples to this sort of approach.

Domain Restriction: Relevant Quantification and Situation Semantics

Appeals to domain restriction are not limited to normalcy-based approaches. In light
of generics such as “ducks lay eggs,” some theorists have argued that generics involve
quantification over relevant individuals (actual or possible), where context deter-
mines which individuals are relevant (e.g. Schubert and Pelletier 1987; Declerk 1991;
Chierchia 1995). On such views, when we consider a generic such as “ducks lay eggs,”
only the mature, fertile female ducks enter into our evaluation of the sentence, because,
for example, they are the only potential egg-layers. The question that arises is how
exactly to determine which individuals are relevant. Schubert and Pelletier (1987) offer
a detailed discussion of how some of these restrictions arise—for example, via presup-
position, focus, linguistic context, or explicit restriction as appropriate.

360
 G EN ERI C S

Ter Meulen (1986), Gerstner-Link (1988), and Cavedon and Glasbey (1994) offer
treatments of generics in the framework of situation semantics (Barwise and Perry
1983). On such views, generics express constraints on situations—for example, “tigers
are striped” expresses the constraint that every situation involving a tiger involves a
striped tiger. Constraints are not reducible to the properties of individuals (Cavedon
and Glasbey 1994). The interpretation of a generic on these views is relative to a given
context (channel), so as on the above views, generics such as “ducks lay eggs” are evalu-
ated only with respect to the female ducks (Cavedon and Glasbey 1994).
As noted above, there are philosophical and empirical difficulties (e.g. Leslie 2008;
Khemlani et al. 2009; see also Cimpian, Gelman, and Brandone 2010) facing any
accounts that rely on domain restriction to handle generics such as “ducks lay eggs.” For
these accounts to succeed, these difficulties would need to be addressed in more detail.

Stereotypes and Prototypes

A somewhat different approach to the semantics of generics is taken by theorists who
suppose that generics express stereotypes or prototypes. On such views, “tigers are
striped” would express that the stereotypical or prototypical tiger is striped, and like-
wise, “sharks attack bathers” would express a belief about the stereotypical or prototypi-
cal shark. Geurts (1985) and Declerk (1986) suggest that generics may be interpreted
as expressing culturally accepted stereotypes, and Platteau (1980), Nunberg and Pan
(1975), and Heyer (1985, 1990) claim that generics express prototypes, in the sense of
Rosch (1978). Thus on such views, “tigers are striped” would express the fact that the
stereotypical or prototypical tiger is striped.
A general concern with views of this type is that they would seem to make the hold-
ing of the relevant sort of false belief sufficient to render the generic true. For example,
suppose that people falsely associate sliminess with snakes in the relevant way (for
example, as a culturally held stereotype, or part of the Roschean prototype, etc.). This
does not make the generic “snakes are slimy” true (Krifka et al. 1995). While it is plau-
sible that people often assert generics on the basis of their prototypical/stereotypical
beliefs, it is implausible to think that these beliefs enter into the truth-conditions of
generics. Stereotypical beliefs can be, and often are, false (see Chapter 6.4.2 Language
and Race).

Probabilitistic Approaches
Cohen (1996, 1999, 2004) argues that generics can be understood in terms of com-
parative probabilities. There are two different ways for a generic to be true on Cohen’s
view. The first way (Absolute) can be illustrated by “tigers are striped.” This is a true
generic because (roughly speaking) a randomly chosen tiger is more likely than not to
be striped. The second way (Relative) a generic can be true involves comparison with
other kinds. For example, on Cohen’s account, “mosquitoes carry the West Nile virus”
is true because (again roughly speaking) if we pick a mosquito and another insect at
random, the mosquito is more likely than the other insect to carry the West Nile virus.
Thus on Cohen’s view, generics are made true (or false) by such probabilistic considera-
tions. More precisely, Cohen’s categories are as follows:
Absolute generics:

361
 SA RA H - J A N E L E S L I E

‘Ks are F’ is true iff the probability that an arbitrary K that satisfies some predi-
cate in Alt(F) satisfies ‘is F’ is greater than .5.

Relative Generics:

‘Ks are F’ is true iff the probability that an arbitrary K that satisfies some predi-
cate in Alt(F) satisfies ‘is F’ is greater than the probability that an arbitrary
member of Alt(K) that satisfies some predicate in Alt(F) satisfies ‘is F’.

As it stands, however, Cohen’s account incorrectly predicts that generics such as “bees
are sterile” will be true, since the probability that a given bee is sterile is greater than .5.
To deal with such cases, Cohen introduces the homogeneity constraint:
Homogeneity Constraint:

The above probability conditions (exceeding .5, or exceeding that of the arbi-
trary alternative to the kind) should hold in all salient partitions of the kind.

Generics such as “bees are sterile” are now predicted to be false, because there is a sali-
ent partition of bees into workers, queens, and drones—and queen bees have a very low
probability of being sterile.
Leslie (2007, 2008) argues that Cohen’s account faces some counterexamples. For
example, humans are more likely to suffer from autism than other mammals, and so
“humans are autistic” would seem to be falsely predicted to be a true (relative) generic
(and it is hard to see how homogeneity could undermine this prediction). This example
illustrates that unique possession of a property by a kind is not generally sufficient for
the truth of a generic, yet the category of relative generic predicts that, modulo homo-
geneity, this should suffice. Leslie (2007, 2008) offers additional counterexamples to
Cohen’s conditions.
Recent empirical work also suggests that the homogeneity constraint does not guide
people’s understanding of generics. For example, Cimpian, Gelman, and Brandone
(2010) found that adults are not at all reluctant to accept a generic that involves a
property found only in one salient partition of a kind, contra the predictions of the
homogeneity constraint (though this was not the intent behind the experiment).

Generics and Psychology

Recent work in psychology by Susan Gelman and her collaborators shows that gener-
ics are very easily acquired by young children. In particular, generics are understood by
young children more easily than explicit quantifiers such as “most”, and even “all” and
“some,” at least when used to make kind-wide generalizations (Gelman 2003, 2004; Gel-
man , Goetz, Samecka, and Flukes 2008, Gelman and Tardiff 1998; Hollander, Gelman,
and Star 2002; Leslie and Gelman forthcoming; Papafragou and Schwarz 2005/2006
and other studies in progress). This is prima facie puzzling since, as the above discussion
indicates, the semantics of generics look to be very complex.
Leslie (2008) argues from these considerations and others that generics give voice to
cognitively default generalizations, and that quantified statements give voice to cogni-
tively more sophisticated, nondefault ones. This hypothesis has gained some empirical
support from the finding that both adults and young children will interpret and recall

362
 G EN ERI C S

quantified statements as generics (Hollander et al. 2002; Khemlani, Leslie, Glucksberg,

and Rubio-Fernandes 2007; Leslie, Khemlani, and Glucksberg 2011; Leslie and Gelman
forthcoming).
Leslie further argues that once we understand the role that generics play in our psy-
chology, we can develop an account of when generics are true and false. Leslie divides
generics into three categories: characteristic (including items such as “ducks lay eggs”),
majority (for example “cars have radios”) and striking/dangerous (for example “mosqui-
toes carry the West Nile virus”). These different classes have different requirements that
the world must meet for the corresponding generic to be true. Cimpian, Brandone, and
Gelmen (2010) report empirical results that support Leslie’s thesis that these particu-
lar factors differentially impact on adults’ willingness to accept generic statements (for
example adults accept generics at lower prevalence levels if the property in question is
strikingly dangerous).

Definite and Indefinite Singular Generics

The above discussion was primarily centered on accounts of bare plural generics, which
have received the most discussion in the literature. Singular generics introduce their
own sets of complications. Unlike bare plurals, singular generics can easily be infelici-
tous. For example, it is perfectly fine to say (9) or (10):

(9) A madrigal is polyphonic

(10) A football hero is popular

But not (11):

(11) *A madrigal is popular

(Notice, however, that the plural version of (11) “madrigals are popular” is perfectly felici-
tous (though perhaps, unfortunately, false).) Lawler (1973) notes that indefinite singulars
are only felicitous when they express properties that are somehow “necessary”, “essential”
or “inherent” to the kind. Burton-Roberts (1977) argues that indefinite singulars carry a
special normative force, while Krifka et al. (1995) take a different tact, arguing that the
facts can be accounted for in terms of logical form. Greenberg (1998) and Cohen (2001)
argue that indefinite singulars can express only “rules and regulations,” in the sense of Carl-
son (1995). These analyses are often implemented by way of possible worlds semantics.
Definite singular generics appear to invoke constraints similar to those of indefinite
singulars, but also some more besides. For example, definite singular subjects are often
infelicitous if they do not refer to well-established kinds (Krifka et al. 1995; example
from Carlson 1977, attributed to Barbara Partee):

(12) The coke bottle has a narrow neck

(13) *The green bottle has a narrow neck

Other constraints on the acceptability of definite singulars have been noted by Nunberg
and Pan (1975), Carlson (1977), Dahl (1985), and others. However, definite singular
generics have received less attention overall than indefinite singulars, while bare plurals
have received by far the most discussion in the literature.

363
 SA RA H - J A N E L E S L I E

New Directions
Recently, several philosophers have begun to consider the role generics, and the gen-
eralizations they express, may play in various forms of social prejudice (for example
Haslanger forthcoming; Leslie forthcoming; see Chapter 6.4.2 Language and Race for
more details). More generally, generics continue to be a topic of considerable interest
to linguists, philosophers, and more recently, psychologists. There has recently been
an explosion of interest among psychologists concerning generics (e.g. Gelman 2003;
Prasada and Dillingham 2006, 2009). Recent and ongoing work in psychology has been
examining issues such as how children acquire generics, how children and adults process
generics, the role of generics in reasoning, and the influence of generics in various forms
of prejudice. This new range of empirical findings will surely be a significant influence
on the topic going forward.

Related Topics
2.6 Possible-Worlds Semantics
3.4 Quantifiers and Determiners
3.8 Plurals
3.9 Adverbs
3.10 Mass Terms
6.3.1 The Language of Causation
6.4.1 Language, Gender, and Sexuality
6.4.2 Language and Race.

References
Asher, N. and M. Morreau. (1995) “What some generic sentences mean”. In G. Carlson and F. J. Pelletier
(eds.) The Generic Book. Chicago: Chicago University Press, pp. 300–339.
Barwise, J., and Perry, J. (1983). Situations and Attitudes. Cambridge, MA: MIT Press.
Burton-Roberts, N. (1977). “Generic sentences and analyticity.” Studies in Language, 1, 155–196.
Carlson, G. (1977). Reference to Kinds in English. Ph.D. dissertation, University of Massachusetts,
Amherst.
Carlson, G. (1989). ‘On the semantic composition of English generic sentences’. In G. Chierchia et al.
(eds.), Properties, Types and Meanings, Vol. II: Semantic Issues, Dordrecht: Kluwer, pp. 167–192.
Carlson, G. (1995). “Truth-conditions of generic sentences: Two contrasting views.” In G. Carlson and F. J.
Pelletier (eds.) The Generic Book. Chicago: Chicago University Press, pp. 224–237.
Cavedon, L., and Glasbey, S. (1994). “Outline of an information-flow model of generics.” Acta Linguistica
Hungarica, 42.
Chierchia, G. (1995). “Individual level predicates as inherent generics”. In G. Carlson and F. J. Pelletier
(eds.) The Generic Book. Chicago: Chicago University Press, pp. 176–224.
Chierchia, G. (1998). “Reference to kinds across languages.” Natural Language Semantics, vol. 6.
Cimpian, A., Brandone, A. C., and Gelman, S. A. (2010). “Generic statements require little evidence for
acceptance but have powerful implications.” Cognitive Science, 34(8), 1452–1482.
Cimpian, A., Gelman, S. A., and Brandone, A. C. (2010). “Theory-based considerations influence the
interpretation of generic sentences.” Language and Cognitive Processes, 25(2), 261–276.
Cohen, A. (1996). Think Generic: The Meaning and Use of Generic Sentences. Ph.D. dissertation, Carn-
egie Mellon University.
Cohen, A. (1999). “Generics, frequency adverbs and probability.” Linguistics and Philosophy, 22, 221–253.
Cohen, A. (2001) “On the generic use of indefinite singulars.” Journal of Semantics, 18(3), 183–209.
Cohen, A. (2004). “Generics and mental representation.” Linguistics and Philosophy, 27(5), 529–556.
Cohen, A., and Erteschik-Shir, N. (1997). “Topic, focus and the interpretation of bare plurals.” In P. Dekker,
M. Stokhof, and Y. Venema (eds.) Proceedings of the 11th Amsterdam Colloquium, pp. 31–36.

364
 G EN ERI C S

Cohen, A., and Erteschik-Shir, N. (2002). “Topic, focus, and the interpretation of bare plurals.” Natural
Language Semantics, 10(2), 125–165.
Dahl, O. (1975). “On generics.” In E. L. Keenan (ed,) Formal Semantics of Natural Language,. Cambridge:
Cambridge University Press, pp. 99–111.
Dahl, O. (1985). Tense and Aspect Systems. Oxford: Blackwell.
Declerk, R. (1986). “The manifold interpretations of generic sentences.” Lingua, 68, 149–188.
Declerk, R. (1991). “The origins of genericity.” Linguistics, 29, 79–101.
Delgrande, J. P. (1987). “A first-order conditional logic for prototypical properties.” Artificial Intelligence,
33, 105–130.
Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press.
Gelman, S. A. (2003). The Essential Child: Origins of Essentialism in Everyday Thought. New York: Oxford
University Press.
Gelman, S. A. (2004). “Learning words for kinds: Generic noun phrases in acquisition.” In D. G. Hall and
S. R. Waxman (eds.) Weaving a Lexicon. Cambridge, MA: MIT Press.
Gelman, S. A., Goetz, P. J., Sarnecka, B. S., and Flukes, J. (2008). “Generic language in parent-child con-
versations.” Language Learning and Development, 4, 1–31.
Gelman, S. A., & Tardif, T. Z. (1998). “A cross-linguistic comparison of generic noun phrases in English
and Mandarin.” Cognition, 66, 215–248.
Gerstner-Link, C. (1988). Uber Generizitat. Generische Nominalphrasen in Singularen Aussagen und
Generischen Aussagen. Ph.D. dissertation, University of Munich.
Geurts, B. (1985). “Generics.” Journal of Semantics, 4.
Greenberg, Y. (1998), Manifestations of Genericity. Routledge: New York.
Haslanger, S. (Forthcoming), “Ideology, generics, and common ground”, in Charlotte Witt (ed.) Feminism
and Metaphysics.
Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of
Massachusetts, Amherst.
Heyer, G. (1985). “Generic descriptions, default reasoning, and typicality.” Thoretical Linguistics, 12,
33–72.
Heyer, G. (1990). “Semantics and knowledge representation in the analysis of generic descriptions.” Journal
of Semantics, 7, 93–110.
Hollander, M. A., Gelman, S. A., and Star, J. (2002). “Children’s interpretation of generic noun phrases.”
Developmental Psychology, 38, 883–894.
Kamp, H. (1981), “A theory of truth and semantic interpretation”. In J. Groenendijk, J. Janssen and M.
Stokhof (eds.) Formal Methods in the Study of Language. Forts Publications. Dordrecht, pp. 277–322.
Kamp, H. and U. Reyle. (1993). From Discourse to Logic. Dordrecht: Kluwer Academic Publishers.
Khemlani, S., Leslie, S. J., Glucksberg, S., & Rubio Fernandez, P. (2007). “Do Ducks Lay Eggs? How Humans
Interpret Generic Assertions.” Proceedings of the 29th Annual Cognitive Science Society. Nashville, TN:
Cognitive Science Society.
Khemlani, S., Leslie, S. J., and Glucksberg, S. (2009). “Generics, prevalence, and default inferences”. Pro-
ceedings of the 31st Annual Cognitive Science Society. (2009) Amsterdam: Cognitive Science Society.
Kiss, K. E. (1998). “On generic and existential bare plurals and the classification of predicates.” In S. Roth-
stein, (ed.) 1998. Events and Grammar. Dordrecht: Kluwer, pp. 145–162.
Kratzer, A. (1995). “Stage-level and individual-level predicates”. In G. Carlson and F. J. Pelletier (eds.) The
Generic Book. Chicago: Chicago University Press, pp. 125–176.
Krifka, M. (1987). An Outline of Genericity, partly in collaboration with Claudia Gerstner. SNS-Bericht 87-
23, University of Tubingen.
Krifka, M., F. Pelletier, G. Carlson, A. Ter Meulen, G. Chierchia and G. Link. (1995). “Genericity: An
introduction”. In G. Carlson and F. J. Pelletier (eds.) The Generic Book. Chicago: Chicago University
Press, pp. 1–125.
Laca, B. (1990). “Generics objects: Some more pieces of the puzzle.” Lingua, 81, 25–46.
Lawler, J. (1972). “Generic to a fault.” CLS 8, Chicago: Chicago Linguistic Society, 247–258.
Lawler, J. (1973). “Studies in English generics.” University of Michigan Papers in Linguistics 1(1).
Leslie, S. J. (2007). “Generics and the structure of the mind.” Philosophical Perspectives, 21(1), 375–403.
Leslie, S. J. (2008). “Generics: Cognition and acquisition.” Philosophical Review, 117(1), 1-47.
Leslie, S. J. (forthcoming). “The original sin of cognition: Fear, prejudice, and generalization.” The Journal
of Philosophy.
Leslie, S. J., and Gelman, S. A. (forthcoming). “Quantified statements are recalled as generics: Evidence
from preschool children and adults”. Cognitive Psychology.

365
 SA RA H - J A N E L E S L I E

Leslie, S. J., Khemlani, S., and Glucksberg, S. (2011). “All ducks lay eggs: The generic overgeneralization
effect”. Journal of Memory and Language, 65, 15–31.
Lewis, D. (1975). “Adverbs of quantification”. In E. L. Keenan (ed.), Formal Semantics of Natural Language.
Cambridge: Cambridge University Press, pp. 3–15.
Liebesman, D. (2011), “Simple generics”, Noûs, 45(3), 409–442.
McCarthy, J. (1986). “Applications of circumscription to formalizing commonsense knowledge.” Artificial
Intelligence, 28, 89–116.
ter Meulen, A. (1986). “Generic information, conditional contexts and constraints.” In On Conditionals,
E. C. Traugott, A. ter Meulen, J. S. Reilly, and C. A. Ferguson (eds.) Cambridge: Cambridge University
Press.
Milsark, G. (1974). Existential Sentences in English. Ph.D. dissertation, MIT.
Nickel, B. (2008), “Generics and the way of normality.” Linguistics and Philosophy, 31 (6), 629–648.
Nunberg, G. and Pan, C. (1975). “Inferring quantification in generic sentences.” In Papers from the Eleventh
Regional Meeting of the Chicago Linguistic Society.
Papafragou, A. and Schwarz, N. (2005/6). “Most wanted.” Language Acquisition 13 (Special Issue: On the
Acquisition of Quantification) 207–251.
Pelletier, F. and Asher, N. (1997). “Generics and defaults”. In J. van Benthem and A. ter Meulen (eds.)
Handbook of Logic and Language. Cambridge, MA: MIT Press, pp. 1125–1179.
Platteau, F. (1980). “Definite and indefinite generics.” In J. van der Auwera (ed.) The Semantics of Determin-
ers. London: Croom Helm, pp. 112–123
Prasada, S. and Dillingham, E.M. (2006). “Principled and statistical connections in common sense concep-
tion.” Cognition, 99, 73–112.
Prasada, S. and Dillingham, E.M. (2009). “Representation of principled connections: A window onto the
formal aspect of common sense conception.” Cognitive Science 33, (3), 401–448.
Reiter, R. (1987). “Nonmonotonic reasoning.” In Annual Review of Computer Science, J. F. Traub, N. J. Nils-
son and B. J. Grosz (eds.) Palo Alto: Annual Reviews Inc, pp. 147–186.
Rosch, E. H. (1978). “Principles of categorization”. In E. Rosch and B. Lloyd (eds.), Cognition and Categoriza-
tion. Hillsdale, NJ: Erlbaum Associates, pp. 27–48.
Schubert, L.K. and Pelletier, F.J. (1987). “Problems in representing the logical form of generics, bare plurals,
and mass terms”. In E. Lepore (ed.), New Directions in Semantics. London: Academic Press, pp. 387–453.
Schubert, L. K. and Pelletier, F. J. (1989). “Generically speaking, or using discourse representation theory to
interpret generics.” In G. Chierchia, , B.H. Partee, and R. Turner, (eds.) Properties, Types and Meaning.
Dordrecht: Kluwer, pp. 193–268.
Stump, G. (1985). The Semantic Variability of Absolute Constructions. Dordrecht: Reidel.

366
 3.6
ANAP H OR A
Jeffrey King

To say precisely what anaphora is in a way that is theory neutral is nontrivial. Very
roughly, cases of anaphora involve one expression, the anaphoric one, being inter-
preted in light of another expression, its antecedent. A paradigmatic case would be the
following:

1. Joanna1 came to talk to me. She1 was upset.

(I’ll use numerical coindexing to indicate intended anaphoric relations.)

Here ‘She’ is the anaphor (anaphoric on ‘Joanna’) and ‘Joanna’ is its antecedent. It is
widely held that expressions of many different syntactic categories produce anaphora.
However, for purposes of the present chapter, we shall concentrate on pronominal ana-
phora since it is the most widely studied form of anaphora and has generated the most
controversy.
The reason for interest in anaphoric pronouns is that it is sometimes not clear how
they are functioning semantically. Before turning to such cases, it is worth saying a few
words about cases in which it is relatively clear how anaphoric pronouns are functioning
semantically. It is generally agreed that in sentences like 1, the pronoun simply picks up
its antecedent’s semantic value. For example, those who think that the name ‘Joanna’
is a referring expression whose semantic value is Joanna will hold that the semantic
value of the anaphoric pronoun is Joanna as well. Hence the pronoun is simply a refer-
ring expression. No one has lost any sleep attempting to understand cases of anaphora
illustrated by examples like 1. Similarly, there is wide agreement that in certain cases,
anaphoric pronouns are functioning semantically as bound variables do in first order
logic. In cases like:

2. [Every politician]1 believes she1 should have more power.

it would appear that ‘she’ is simply a variable bound by ‘Every politician’, so that 2 could
be represented as:

2a. [Every politician: x] believes that x should have more power.

Again here, it would seem that the functioning of the anaphoric expression is well
understood.
However, as indicated above, there are other cases in which it is unclear how ana-
phoric pronouns are functioning semantically and for this reason such cases have

367
 J EF F REY K I N G

attracted lots of attention from linguists and philosophers. One such case is one in
which a pronoun is anaphoric on an indefinite description in another sentence:

3. [A man]1 showed up at Sue’s office yesterday. He1 was with the IRS.

Many philosophers and linguists think that indefinite descriptions are existential quan-
tifiers. Further, it is generally thought that the scope of a quantifier cannot extend
beyond the sentence in which it occurs. That this is so is confirmed by trying to bind
pronouns in other sentences using other quantifiers:

4. *[Every/No student]1 took the final. He1 didn’t do well.

Further, as noted by Evans (1977), if we consider examples like 3 but using different
quantifiers, we see that construing the pronoun as a variable bound by its quantifier
antecedent would yield the wrong truth conditions. It would make

5a. [Exactly one man]1 drank champagne. He1 was ill.

equivalent to

5b. Exactly one man drank champagne and was ill.

But of course this is incorrect. So in 3, we appear to have a pronoun anaphoric on a

quantifier that cannot be construed as a variable bound by the quantifier. It also does
not seem as though the pronoun in 3 is a referring expression. For one thing, holding
that the pronoun is a referring expression would assign the discourse the wrong modal
profile. Suppose exactly one man, Alan, did appear at Sue’s office yesterday and that the
pronoun in the second sentence refers to him. Then, consider a world in which a differ-
ent man showed up at Sue’s office yesterday and in which Alan works for the IRS. Both
sentences of 3 would be true at such a world if the pronoun in the second sentence refers
to Alan (and ‘A man’ in the first sentence is an existential quantifier). But this is incor-
rect. Further, in more complex cases it is much harder to see how pronouns anaphoric
on indefinites whose scope they do not occur in can be referring expressions:

6. John believes [a man]1 broke into Sue’s apartment. Tom believes he1 got in
through the window.

Both sentences in 6 can be true even if no man did break into Sue’s apartment. But
then it seems as though the pronoun here cannot be a referring expression. And surely
it would be desirable to have a theory on which the pronouns in 3 and 6 work in the
same way semantically. So it appears that the pronoun in 3 is not a referring expression,
nor is it a variable bound by its quantifier antecedent. But then how is it functioning
semantically? Because the pronoun in 3 has an antecedent prior to it in the discourse,
examples like this are often called instances of discourse anaphora. So, there is a theoreti-
cal question as to how anaphoric pronouns function semantically in cases of discourse
anaphora.
A second sort of case in which it is unclear how an anaphoric pronoun is functioning
semantically is a case in which in a single sentence a pronoun appears to be anaphoric on

368
 A N A PH O R A

a quantifier but not a variable bound by it. The primary examples of interest here are
so-called donkey sentences. Examples include:

7a. Every man who owns [a donkey]1 beats it1.

7b. If [a man]1 owns [a donkey]2, he1 beats it2.

It is generally agreed that both sentences are true just in case every male donkey owner
beats every donkey he owns. For obvious reasons, sentences like 7a are sometimes called
relative clause donkey sentences and those like 7b are called conditional donkey sentences.
The term donkey anaphora generally applies to both kinds of example. Given the truth
conditions of these sentences, it is clear that the anaphoric pronouns here cannot be
referring expressions. Further, there is reason to think they cannot be variables bound
by their quantifier antecedents either. First, if we try using other quantifiers in 7a/7b we
cannot bind the pronouns:

7’a. *Every man who owns [every/no donkey]1 beats it1.

7’b. *If [a man]1 owns [every/no donkey]2, he1 beats it2.

This is thought to be due to the fact that the scope of a quantifier in a relative clause is
confined to the relative clause; and similarly for the scope of a quantifier in the anteced-
ent of a conditional. Further, in 7b, even if the existential quantifier could take scope
over the entire conditional, it doesn’t seem like the resulting truth conditions could
require every male donkey owner to beat every donkey he owns. So here again, we find
anaphoric pronouns with quantifier antecedents that cannot be understood as referring
or as being bound by their quantifier antecedents. Again, this raises the theoretical
questions of how donkey pronouns are functioning semantically.
I now turn to the main theoretical approaches that give accounts of the semantic
functioning of pronouns in discourse anaphora and donkey anaphora. Due to limita-
tions of space, I shall be content to informally sketch the main ideas of each approach.
The first sort of approach to donkey and discourse anaphora I’ll discuss, which I’ll call
the descriptive approach, claims that the anaphoric pronouns in question function seman-
tically as definite descriptions. Descriptive approaches have been defended by Cooper
(1979), Davies (1981), Neale (1990), Heim (1990), and Elbourne (2005) among oth-
ers. In cases of discourse anaphora, it is easy to see why one might think that pronouns
function semantically as definite descriptions. The second sentences of the following
two discourses seem to mean the same thing:

8a. [A female student]1 took my class and flunked it. She1 was upset.
8b. [A female student]1 took my class and flunked it. [The female student who
took my class and flunked it]1 was upset.

(8b may sound a bit awkward, but that seems to be because the descriptive material in
the description isn’t needed and so is redundant. When such material is needed to figure
out who is being talked about, the awkwardness disappears: ‘A female student took my
class and flunked it but another female student took my class and did well. The female
student who took my class and flunked it was upset.’) Descriptive theories, particular
versions of the descriptive approach, can differ along the following dimensions. First, it
might be claimed that at the relevant level of syntax, the anaphoric pronouns in ques-

369
 J EF F REY K I N G

tion are definite descriptions (Cooper 1979, Heim 1990, Elbourne 2005) or it might
be claimed that the pronouns are merely semantically interpreted as definite descriptions
(Neale 1990). Second, such views can differ as to how the descriptive content of the
description that provides the interpretation of the pronoun is generated. For example,
one could hold that it is generated by being salient in context or one might claim that
it is recovered in some rule-governed way from the linguistic environment of its ante-
cedent. Finally, descriptive theories can differ in terms of the semantics they assign to
definite descriptions and hence to the anaphoric pronouns to which they are semanti-
cally equivalent. For example, Neale (1990) endorses a Russellian treatment of definite
descriptions on which they are quantifiers and on which the truth conditions of sen-
tences of the form ‘The F is G’ require there to be exactly one F, which is also G. On
this treatment such sentences entail the existence and uniqueness of an F. By contrast,
Heim (1990) and, following her, Elbourne (2005) take such sentences to presuppose that
there is a unique F and to assert that it is G. For Heim and Elbourne, if this presupposi-
tion fails we have no truth value for the sentence, whereas for Neale the failure for there
to be a unique F results in falsity. For Heim and Elbourne, these anaphoric pronouns
have as their semantic values functions from situations (or situations and individu-
als—see example 11) to individuals.
While it is easy to see in outline how descriptive theories will treat instances of dis-
course anaphora such as 8a, it should be not at all obvious how they will treat instances
of donkey anaphora such as 7a and 7b above. For Neale (1990) and Heim (1990), 7a
would essentially get regimented as follows:

7aR. ([Every x: man x & [a y: donkey y & x owns y]][the z: donkey z and x
owns z] (x beats z)).

The underlined definite description corresponds to the pronoun ‘it’ in 7a. Note that it
contains a variable bound by the quantifier ‘Every man who owns a donkey.’ But the
problem here is that 7aR either entails (Neale 1990) or presupposes (Heim 1990) that
every man who owns a donkey owns exactly one donkey. And yet native speakers tend
to judge 7as true in a case in which many male donkey owners own more than one
donkey, so long as all male donkey owners beat all donkeys they own. Similar problems
arise with 7b.
I know of two attempts to deal with this problem in the literature. Neale [1990],
following Parsons (1978) and Davies (1981), claims that there are “numberless descrip-
tions,” which he represents as ‘whe F’ (instead of ‘the F’), such that ‘Whe F is G’ is true
iff every F is G (and there is an F). Hence so-called numberless descriptions are essen-
tially universal quantifiers. Interpreting the pronoun in 7a as a numberless description,
we would get an LF as follows:

7aR1. ([Every x: man x & [a y: donkey y & x owns y]][whe z: donkey z and x
owns z] (x beats z))

This is true iff every man who owns a donkey beats every donkey he owns. Though
Neale gets the right truth conditions, positing these numberless descriptions seems a
bit ad hoc. Further, if singular pronouns anaphoric on singular indefinites can have
numberless or Russellian readings, there needs to be an explanation of why many such
pronouns do not have numberless readings including the following:

370
 A N A PH O R A

9. A student came to see me today. She was upset.

There is no reading of the second sentence on which it entails that every student who
came to see me today was upset. Neale suggests that in a case like this, one likely has a
student in mind who provides the grounds for the one’s utterance and so the pronoun
is interpreted as a Russellian description. But his explanation doesn’t work. Suppose I
am telling you that there are some things I know on general grounds. You ask for an
example. I say:

10. Well, a man is eating dinner in a restaurant right now in New York. He is
drinking an expensive glass of wine. He has just finished a long day of work
on Wall Street.

Surely, though I have no particular man in mind, what I’ve said is true iff there is a man
having dinner at a restaurant in New York right now who is drinking a glass of expensive
wine and has just finished a long day of work on Wall Street. Just as surely for example,
the second sentence does not have a reading on which its entails that every man eating
dinner right now in a restaurant in New York is drinking a glass of expensive wine. But
if the pronoun qua description in the second sentence were given a numberless read-
ing, this would be entailed. Here I have no man in mind in making my utterance, so by
Neale’s own lights the numberless readings should be available.
Heim (1990) suggested dealing with the uniqueness presupposition, generated by
treating the pronouns in 7b as definite descriptions, by taking conditionals to quantify
over situations. Very roughly, she claims 7b is true iff every minimal situation s in which
the antecedent is true (which is a situation containing one man who owns one donkey),
can be extended to a situation s’ in which the consequent is true. However, when the
consequent is evaluated in s’, the pronouns qua definite descriptions are evaluated in s.
But there is a unique man owning a donkey and a unique donkey owned by a man in
s. So we just make sure that the man who owns a donkey (in s) beats in s’ the donkey
owned by a man (in s). The uniqueness presuppositions of the pronouns qua definite
descriptions are rendered harmless due to the minimality of s.
However Heim’s treatment of 7b leads to what has been dubbed the proportion prob-
lem. Specifically, Heim’s account predicts that the following sentence is true if there
is one man who owns 100 donkeys and beats them all, while 99 other men own one
donkey each and don’t beat them:

11. If a man owns a donkey, he usually beats it.

The reason Heim’s theory predicts this is that most (“usually”) minimal situations con-
sisting of a single man owning a single donkey are extendable to a situation in which the
man beats the donkey (the one donkey beater is “part” of 100 such situations). However,
the prediction that the sentence is true in this circumstance seems incorrect. Though
Heim (1990) considers ways to deal with the problem, none ultimately succeeds; nor is
the matter dealt with satisfactorily in Elbourne (2005). Further, Elbourne (2005) tries
to extend Heim’s [1990] treatment of 7b, as quantifying over minimal situations in an
attempt to render uniqueness presuppositions of the pronouns harmless, to 7a. Elbourne
construes quantification generally (‘Every man who owns a donkey’) as quantification
over minimal situations. Roughly, when we evaluate ‘it’ in 7a we are doing so with

371
 J EF F REY K I N G

respect to a minimal situation in which a man owns a donkey. But such a situation will
again contain a unique donkey, rendering the uniqueness presupposition of the pronoun
qua definite description harmless. Unfortunately, Elbourne thereby runs into a version
of the proportion problem. His theory predicts that the following sentence is true if,
again, there is one man who owns 100 donkeys and beats them all while 99 other men
own one donkey each and don’t beat them:

12. Most men who own a donkey beat it.

Again, this prediction seems clearly incorrect.

There is a second approach to discourse and donkey anaphora that is in certain ways
similar to certain versions of the descriptive approach (for example that of Davies 1981
and Neale 1990), though it avoids some of its difficulties mentioned above. On these
versions, the anaphoric pronouns in questions are semantically functioning as definite
descriptions, which are in turn understood as quantifiers. On the Context Dependent
Quantifier (CDQ) approach to discourse and donkey anaphora, suggested by Wilson
[1984] and elaborated and defended in King (1987, 1994, 2004), pronouns in discourse
anaphora are understood as quantifiers as well. However, they are quantifiers whose
force (universal, existential, etc.), restriction (what they quantify over), and scope (rela-
tive to other scoped elements) are determined by features of their linguistic environ-
ments. In the case of discourse anaphora:

9. A student came to see me today. She was upset.

CDQ claims that the pronoun is a quantifier that has existential force (inherited from
its antecedent), and is restricted to students who came to see me today. Hence the second
sentence is true iff a student who came to see me today was upset. This seems intuitively
correct.
As for donkey anaphora, King (2004) proposes an account of conditional donkey
anaphora on which, as on Heim’s (1990) account, conditionals quantify over situations.
The rough idea is that in a case like

13. If Sarah owns a donkey, she beats it.

though the pronoun ‘it’ is an existential quantifier ranging over donkeys owned by
Sarah, and hence the consequent is equivalent to Sarah beats a donkey she owns, the
whole conditional is true iff every minimal situation s1 in which Sarah owns a donkey
(i.e. in which the antecedent is true) is part of situation s2 in which Sarah beats a don-
key she owns in s1 (i.e. the consequent is true in s2 in virtue of the donkey in s1, which
of course must be in s2 too). Because King’s (2004) account of donkey conditionals is
so similar to Heim’s (1990), he too faces the proportion problem here. See King (2004)
for an inconclusive discussion of the relevant issues and attempted solutions. As to the
CDQ account of relative clause donkey sentences, it straightforwardly predicts one of
the readings such sentences have. See King (2004) for details.
Let me now turn to the third kind of approach to discourse and donkey anaphora:
discourse representation (DR) approaches. Such approaches were pioneered by Heim
(1982) and Kamp (1981). My exposition here will follow Kamp’s implementation of
the basic ideas. In processing sentences of a natural language discourse, speakers produce

372
 A N A PH O R A

representations of the discourse that Kamp calls discourse representation structures (DRS).
As the discourse proceeds and a new sentence is uttered, it is interpreted in the context
of the DRS the hearer has produced to that point. The DRS is “updated” in the process-
ing of the new sentence yielding a new DRS. As indicated above, it is not possible here
to lay out Kamp’s formal theory, but we can sketch the outlines of the theory and how it
handles discourse and donkey anaphora. The formal theory is comprised by the follow-
ing things. First, there is a formal language in which a discourse, understood as a finite
string of sentences of the language, can be formulated. Second, there are rules of DRS
construction. Given the DRS at a given point corresponding to the processing of the
first n sentences in the discourse, these rules tell you how to update that DRS depend-
ing on what type of sentence the n+1st sentence of the discourse is. Finally, there is a
definition of what it is for a discourse D to be true in a model M relative to a complete
DRS K for D.
Which construction rule applies at a given point in processing a sentence depends on
which expression in the sentence “has widest scope” (i.e. was added last in constructing
the sentence) among the expressions in the sentence that have not already been treated.
In order to explain how DRS approaches handle discourse and donkey anaphora, we
need to informally explain the construction rules for indefinite descriptions, pronouns,
conditionals, and universal quantifiers. We then need to explain how the semantics
applies to the DRS resulting from the application of these construction rules.
Here are some notions from Kamp’s formal theory. A discourse representation (DR) is
a pair <U,Con> , where U is a set of discourse referents (to be discussed below) and Con
is a set of conditions, which are simply sentences of a language that extends the language
that is used to formulate a discourse. A discourse representation structure (DRS) is a set
of DRs. The simplest rules of DRS construction tell us what to add to U (if anything)
or to Con in applying the rules to sentences in the DR in question. The more complex
rules tell us how to add DR s to our DRS. Examples should make this clear.
Let’s begin by considering an example of discourse anaphora.

14. Pedro owns [a donkey]1. It1 thrives.

The beginning DRS for this discourse is a set containing one DR: {<Ø, {Pedro owns a
donkey, It thrives} }. Assuming that ‘Pedro’ has widest scope in the first sentence, the
rule for proper names tells us to add a discourse referent, say ‘u’, to U (=Ø) and add the
conditions ‘u =Pedro’ and ‘u owns a donkey’ to Con (={Pedro owns a donkey, It thrives}),
yielding the following DRS: {<{u},{Pedro owns a donkey, It thrives, u=Pedro, u owns
a donkey}>}. Now assume that ‘a donkey’ in ‘u owns a donkey’ has widest scope of the
expressions not yet treated in the sentence. Then we apply the indefinite rule, which
tells us to add another discourse referent, say v, to U(={u}) and to add the conditions
‘donkey (v)’ and ‘u owns v’ to Con (= {Pedro owns a donkey, It thrives, u=Pedro, u owns
a donkey}) yielding the DRS {<{u,v}, {Pedro owns a donkey, It thrives, u=Pedro, u owns
a donkey, donkey (v), u owns v}>}. At this point, nothing further can be done to the first
sentence. In the second sentence, the pronoun in subject position has widest scope, and
so we apply the pronoun rule to it. Roughly, this rule tells us to find a “suitable” accessible
discourse referent, in this case that is going to be ‘v’, and add to Con ‘v thrives’, yielding
the following complete DRS for our original discourse: {<{u,v}, {Pedro owns a donkey, It
thrives, u=Pedro, u owns a donkey, donkey (v), u owns v, v thrives}>}. Now put very
roughly, Kamp’s (1981) semantics tells us that this DRS is true relative to model M iff

373
 J EF F REY K I N G

there is a function f from U (= {u,v}) into M such that f(u) = referent of ‘Pedro’ in M,
<f(u), f(v)> H extM(‘owns’) and f(v) H extM(‘donkey’) and extM(‘thrives’) (extM (‘owns’)
is, of course, the extension of ‘owns’ in M and similarly for other predicates). In other
words, the discourse is true iff Pedro owns a donkey that thrives. This seems intuitively
correct.
Turning now to donkey anaphora, because Heim’s and Kamp’s original theories treated
relative clause and conditional donkey sentences in virtually the same way, we need only
consider one of these. Let’s try ‘Every man who owns a donkey beats it.’ Here I’ll be a bit
less explicit about the construction. In the previous examples, all DRS were sets contain-
ing one DR. Here that will not be the case. The beginning DRS is: {<Ø, {Every man who
owns a donkey beats it}>}. The ‘Every man’ quantifier has widest scope here, so we apply
the universal quantifier rule first. This rule tells us to add to the DRS two new DRs, in
effect corresponding to the restriction of the quantifier (‘man who owns a donkey’) and the
quantifier’s nuclear scope (‘beats it’). These DRs are respectively <{u}, {man(u), u owns a
donkey}> and <Ø, {u beats it}>. The entire resulting DRS is the following set of three DR’s:
{<Ø, {Every man who owns a donkey beats it}>, <{u}, {man(u), u owns a donkey}>, <Ø,
{u beats it}>}. After applying the indefinite rule to the condition ‘u owns a donkey’ in the
second DR in this DRS and applying the pronoun rule to ‘it’ in the third DR, the resulting
DRS is: {<Ø, {Every man who owns a donkey beats it}>, <{u,v}, {man(u), u owns a donkey,
donkey(v), u owns v }>, <Ø, {u beats it, u beats v}>}. Roughly, this DRS is true in M iff
every function f from {u,v} into M such that f(u) HextM (‘man’), f(v) H extM(‘donkey’)
and <f(u),f(v)> H extM(‘owns’) can be extended to a function f’ such that <f’(u), f’(v)> H
extM(‘beats’). (In the current case f’ won’t extend f, because f’ need only be defined on u
and v, which f is. But in the general case f’ will have to extend f in some cases.)
Thus, the semantics predicts that the original sentence is true iff every man who
owns a donkey beats every donkey he owns. As I indicated above, the conditional ‘If
a man owns a donkey, he beats it’ is treated in virtually the same manner with the
conditional understood as universally quantifying over men and donkeys they own in
virtue of the presence of the indefinites ‘a man’ and ‘a donkey’ in the antecedent of
the conditional.
Before turning to other matters, a couple of remarks are in order. First, note how indef-
inites are treated on the present proposal. Thinking of discourse referents for a moment
as free variables, an indefinite contributes to a DRS a free variable while putting a con-
dition on it (donkey(u)). In our example of discourse anaphora, the indefinite has exis-
tential force only because the semantics includes default existential generalization of
free variables (“a DRS is true relative to M iff there is a function that maps the discourse
referents to elements of M . . .”). By contrast, in the relative clause donkey sentence the
discourse referent/free variable introduced by the indefinite gets bound by the universal
quantifier just as much as does the discourse referent introduced in the processing of the
universal quantifier (corresponding to men who own donkeys). Hence the universal
force of the indefinite in the relative clause donkey (male donkey owners must beat
every donkey they own for the sentence to be true). Second, note how the treatment of
anaphora is implemented. In applying the rule for indefinites, an indefinite introduces
a discourse referent; and the pronoun rule substitutes a “suitable” discourse referent
for the pronoun. Often, of course, this suitable discourse referent will have been intro-
duced by an indefinite. In such cases, the pronoun and indefinite end up being the same
discourse referent/free variable; and the default existential generalization of free vari-
ables results in their being “semantically bound.”

374
 A N A PH O R A

A problem with this early version of a DR approach is that it runs squarely into the
proportion problem, discussed earlier. When we consider sentences like

15a. If a man owns a donkey, he usually beats it.

15b. Most men who own a donkey beat it.

straightforward extensions of the central ideas of the DR approach we are considering

make both sentences true in a case in which one man owns 100 donkeys and beats them
all, while 99 other men each own 1 donkey and don’t beat them. The problem is that on
the DR approach, ‘usually’ in 15a quantifies over farmer/donkey pairs; and ‘Most’ in 15b
does as well (recall how in the relative clause donkey sentence considered above, ‘Every’
bound both the free variables/discourse referents for farmers and donkeys). Though this
problem was corrected in Kamp and Reyle (1993), the proportion problem undermines
what was thought to be an insight of DR approaches: namely the idea that the quantifica-
tional force of indefinites is inherited from other quantifiers that bind multiple variables.
A second issue with the simple DR approach discussed to this point is that anaphora on
indefinites is understood as essentially binding: As we saw, an indefinite introduces a
discourse referent that is then “picked up” by a pronoun with the result that default exis-
tential generalization of free variables/discourse referents treats them both as effectively
bound variables. But this means that DR approaches will have to invoke some other
mechanism to explain many cases of anaphora on indefinites, such as the following:

16. A woman is following Glenn. John thinks she is a spy.

In 16 the second sentence can be read as attributing a general belief to John according to
which he does not have a belief about any particular woman, but believes (roughly) that
the woman who is following Glenn is a spy. But on DR approaches, given the mecha-
nisms we have discussed to this point, such sentences only have readings on which it
is as though the anaphoric pronoun is bound by the indefinite, and on this reading
the second sentence of 16 asserts that John has a belief about a particular woman. See
Asher (1987) for an attempt to extend DR approaches to capture this sort of data which
results in significant additions to and complications of DR approaches. Note that other
accounts (some descriptive approaches and the CDQ account) can get the relevant
readings here straightforwardly, which would seem to be advantage of such approaches
with respect to this data.
Let me turn now to dynamic approaches. Here, there are many versions and so I have
opted to introduce the reader to the general features of such approaches with one of the
simplest and most influential versions: Groenendijk and Stokhof (1991). Henceforth, I
shall refer to their account as GSDL.
To begin with, let’s look at how simple discourse anaphora is handled on GSDL. So
consider:

17. A man loves Joanna. He is rich.

Now in GSDL, indefinites such as ‘a man’ are treated as existential quantifiers. Further,
GSDL idealizes a bit and treats consecutive sentences in discourses as being conjoined.
So we can think of 17 as follows:

375
 J EF F REY K I N G

17a. (x)( man x & x loves Joanna) & x is rich

Here we have rendered the anaphoric pronoun ‘He’ as the variable ‘x’, the same variable
that is the variable of its quantifier antecedent. This represents the anaphoric connec-
tion. The important point to notice is that the anaphoric pronoun/variable in 17a is not
within the syntactic scope of its quantifier antecedent. This corresponds to the fact that
in GSDL, the syntactic scopes of quantifiers are confined to the sentences in which they
occur, as current syntactic theory tells us they should be.
The key to understanding the GSDL account of discourse anaphora lies in under-
standing the semantic accounts it offers of the existential quantifier and conjunction.
Let’s begin with existential quantification. In standard first-order logic, we can interpret
formulas (in a model) by assigning them sets of sequences of individuals (those that
satisfy the formulas). GSDL just takes this a step further by assigning to formulas (in a
model) sets of pairs of sequences, thought of as the input and output sequences. In the
case of an existentially quantified formula ‘(x))’, the idea is that a pair of sequences
<g,h> is in its interpretation just in case there is sequence k differing from g at most on
x such that <k,h> is in the interpretation of ‘)’. Note how “interpreting” the existential
quantifier results in shifting from the “input” sequence g to k, where k is now the “input”
sequence to ‘)’. This makes the existential quantifier “internally dynamic,” capable of
affecting the interpretation of expressions within its syntactic scope. Further, the fact
that the output sequence of interpreting the whole existentially quantified sentence,
here h, is allowed to be a sequence different from the input to the interpretation, here
g, means that the processing of the existentially quantified formula may affect the inter-
pretation of expressions after the existentially quantified formula, and hence outside the
scope of the existential quantifier. This is to say that the existential quantifier is “exter-
nally dynamic,” capable of affecting the interpretation of expressions outside its syntac-
tic scope. As we will see, an expression can be internally dynamic and externally static
(as well as internally and externally static). At any rate, putting things very roughly, the
idea here is that once the existential quantifier “resets” the value of ‘x’ in a sequence
so that it satisfies the formula that the quantifier embeds, that value stays reset and can
affect the interpretation of subsequent formulas.
Turning now to conjunction, the idea here is similar. Again, the fundamental idea is
that the interpretation of the left conjunct can affect the interpretation of the right con-
junct. A bit more formally, a pair of sequences <g,h> satisfies a conjunction just in case
there is a sequence k such that <g,k> satisfies the left conjunct and <k,h> satisfies the
right conjunct. Note how interpreting the left conjunct changes the input sequence for
the interpretation of the right conjunct. Again, this means that conjunction is “inter-
nally dynamic,” possibly affecting the interpretation of expressions in its scope. And
again, that the output of interpreting a conjunction, here h, can differ from the input,
here g, means that a conjunction is capable of affecting things outside of it and hence
outside of the scope of that conjunction sign. Again, this is to say that conjunction is
“externally dynamic.”
Given these treatments of existential quantification and conjunction, it is easy to
show that on GSDL the following two formulas are equivalent even when ‘\’contains
free occurrences of ‘x’:

(x)(I& \ and (x)(I & \)

376
 A N A PH O R A

So if we consider again our example of discourse anaphora 17 and its “representation”

in GSDL 17a:

17a. (x)( man x & x loves Joanna) & x is rich

this ends up being equivalent to

17b. (x)( man x & x loves Joanna & x is rich)

and so the sentences of the discourse are true iff some rich man loves Joanna. Since con-
junction is externally dynamic, we can keep adding sentences with anaphoric pronouns
to similar affect. Thus in a discourse such as

17c. A man loves Joanna. He is rich. He is famous.

the sentences are all true iff some rich famous man loves Joanna. These truth conditions
seem intuitively correct.
Because the treatment of donkey anaphora is a bit more complicated technically, and
because some of the main ideas of GSDL are now on the table, I will be more suggestive
here. I urge the interested reader to consult Groenendijk and Stokhof [1991] directly.
Further, since the treatments of conditional and relative clause donkey sentences are
similar, I’ll only consider one of them.
So consider conditional donkey anaphora:

13. If Sarah owns a donkey, she beats it.

This gets regimented in GSDL as follows:

13a. (x)(Io \

where ‘\’ contains an occurrence of the variable ‘x’ (‘it’) not in the scope of the existen-
tial quantifier in the antecedent.
There are three crucial points to the GSDL treatment here: (1) the existential quan-
tifier is externally dynamic and hence may affect the interpretation of variables outside
its scope, and in particular ‘x’ in the consequent of 13a; ( 2) ‘o’ is internally dynamic
and allows the interpretation of its antecedent to affect the interpretation of its con-
sequent (just as is conjunction). Points 1 and 2 together mean that the quantifier in
the antecedent of 13/13a can “semantically” bind the variable in the consequent, even
though it is not in the syntactic scope of the quantifier. But without doing anything
further, we would be left with 13a having the truth conditions of

13b. (x)(Io \)

where ‘o’ is the standard material conditional. This doesn’t give the intuitive truth
conditions of 13 on the reading that concerns us, since 13b would be true if something
failed to be a donkey Sarah owns. The third and final element we need to get the truth
conditions to come out right is to say that a pair of sequences <h,h> is in the interpre-
tation of a conditional iff for all k such that <h,k> satisfies the antecedent, there is a j
such that <k,j> satisfies the consequent. (That we only look at pairs <h,h> means that

377
 J EF F REY K I N G

conditionals are “externally static,” and so cannot affect the interpretation of expres-
sions outside of them, unlike conjunctions. This explains the following contrast: ‘A
man came in and he was happy. He was rich.’; *‘If a man came in, he was happy. He was
rich.’) This says, roughly, that for any output sequence k of a pair of sequences satisfy-
ing the antecedent of the conditional (assigning a donkey Sarah owns to x in the case
of 13/13a), k is the input of a pair <k,j> that satisfies the consequent, for some j. In the
case of a simple example like 13/13a, j=k. So that the account claims that any output of
a pair of sequences that satisfies the existentially quantified antecedent (which means
that the sequence assigns to ‘x’ a donkey Sarah owns), satisfies ‘Sarah beats x’, and so
also assigns to ‘x’ something Sarah beats. That is, the truth of 13/13a requires Sarah to
beat every donkey she owns.
On the other hand, GSDL and dynamic approaches generally face an issue that arose
in our discussion of DR approaches. Put crudely, GSDL solves the problems of discourse
and donkey anaphora by formulating semantics for quantifiers that allows quantifiers to
semantically bind variables that aren’t in their syntactic scopes. In this they (self-con-
sciously) resemble DR approaches. Consider again our discourse 16, repeated here:

16. A woman is following Glenn. John thinks she is a spy.

As mentioned above, 16 has a reading on which the second sentence of the discourse
attributes a general belief to John. On a dynamic approach to 16, the quantifier in the
first sentence semantically binds the variable in the second sentence. But then this
semantically amounts to quantification into the verb of attitude, and so will not result
in a reading of the second sentence on which it attributes a general belief to Scott.
Hence, dynamic approaches need to invoke some other mechanism to get the reading of
the second sentence in question. There are many examples of this sort that will require
dynamic approaches to invoke additional mechanisms.

References
Asher, Nicholas, 1987, ‘A Typology for Attitude Verbs and Their Anaphoric Properties, Linguistics and
Philosophy, volume 10, number 2.
Barker, Chris and Chung-chieh Shan, 2008, ‘Donkey Anaphora is In-Scope Binding,’ Semantics and Prag-
matics, volume 1, Article 1: 1–46.
Brasoveanu, Adrian, 2008, ‘Donkey Pluralities: Plural Information States versus Non-Atomic Individuals,’
Linguistics and Philosophy, volume 31, number 2.
Chierchia, Gennaro, 1995, Dynamics of Meaning, The University of Chicago Press, Chicago and London.
Cooper, Robin, 1979, ‘The Interpretation of Pronouns’ in Frank Heny and Helmut Schnelle, eds., Syntax and
semantics 10: Selections from the Third Groningen Round Table, 61–92. New York: Academic Press.
Davies, Martin, 1981, Meaning, Quantification and Necessity, London: Routledge and Kegan Paul.
Edelberg, Walter, 1986, ‘A New Puzzle About Intentional Identity,’ Journal of Philosophical Logic, volume
15, 1–25.
Elbourne, Paul, 2005, Situations and Individuals, Cambridge, MA:The MIT Press.
Evans, Gareth, 1977, ‘Pronouns, Quantifiers and Relative Clauses (I),’ Canadian Journal of Philosophy, vol-
ume 7, number 3, 467–536.
Geach, Peter, 1962, Reference and Generality, Ithaca, NY: Cornell University Press.
Geach, Peter, 1967, ‘Intentional Identity,’ Journal of Philosophy, volume LXIV, number 20.
Groenendijk, J. and M. Stokhof, 1991, ‘Dynamic Predicate Logic,’ Linguistics and Philosophy volume 14,
39–100.
Heim, Irene, 1982, The Semantics of Definite and Indefinite Noun Phrases, Ph.D. dissertation, University
of Massachusetts.

378
 A N A PH O R A

Heim, Irene, 1990, ‘E-type Pronouns and Donkey Anaphora,’ Linguistics and Philosophy, volume 13,
137–177.
Kamp, Hans, 1981, ‘A Theory of Truth and Semantic Representation,’ reprinted in Formal Semantics The
Essential Readings, B.H. Partee and P. Portner, eds., Oxford:Blackwell Publishers Ltd.
Kamp, H. and U. Reyle, 1993, From Discourse to Logic: Introduction to Model Theoretic Semantics of Natural
Language, Formal Logic and Discourse Representation Theory, Dordrecht: Kluwer Academic Publishers,
King, Jeffrey C., 1987, ‘Pronouns, Descriptions and the Semantics of Discourse,’ Philosophical Studies, volume
51, 341–363.
King, Jeffrey C., 1994, ‘Anaphora and Operators,’ Philosophical Perspectives, volume 8, 221–250.
King, Jeffrey C., 2004, ‘Context Dependent Quantifiers and Donkey Anaphora’, in New Essays in the Phi-
losophy of Language, Supplement to Canadian Journal of Philosophy, vol. 30, M. Ezcurdia, R. Stainton and C.
Viger, eds., Calgary: University of Calgary Press.
King, Jeffrey C., 2005, ‘Anaphora,’ Stanford Encyclopedia of Philosophy, Stanford: Stanford University.
Kripke, Saul, 2009, ‘Presupposition and Anaphora: Remarks on the Formulation of the Projection Problem,’
Linguistic Inquiry, volume 40, number 3, 367–386.
Lewis, David, 1975, ‘Adverbs of Quantification’ in Formal Semantics of Natural Language, E.L. Keenan, ed.,
Cambridge: Cambridge University Press.
Neale, Stephen, 1990, Descriptions, Cambridge, MA: MIT Press.
Parsons, T, 1978, Pronouns as Paraphrases, Amherst: University of Massachusetts, ms.
Partee, Barbara, 1973, ‘Some Structural Analogies between Tenses and Pronouns in English,’ The Journal of
Philosophy, volume 70, number 18, 601–609.
Van der Sandt, Rob, 1992, ‘Presupposition Projection as Anaphora Resolution,’ Journal of Semantics, volume
9, 333–377.
Wilson, George, 1984, ‘Pronouns and Pronominal Descriptions: A New Semantical ‘Category,’ Philosophical
Studies, volume 45, 1–30.

379
 3.7
DE S C R IP T IONS
Peter Ludlow

1 What are Descriptions?

Ordinarily, when philosophers talk about descriptions, they have two kinds of expres-
sions in mind: definite descriptions—understood to be phrases of the form ‘the F’ (and
their equivalents in other languages), and indefinite descriptions—understood to be
phrases of the form ‘an F’ (and their equivalents in other languages). As we will see, this
way of carving up the kinds of descriptions is far too blunt. First, there are many kinds
of expressions that appear to have this form but that are often argued not to be descrip-
tions. For example, in the expression ‘John is a lawyer’ it is often claimed that ‘a lawyer’
is not a genuine description, but is rather something different—a predicate for example.
Furthermore, it is arguable that there are many kinds of expressions having the surface
form ‘the F’ but that nevertheless have semantical properties unlike “typical” descrip-
tions. For example, there are generics (as in ‘The whale is a mammal’), and plurals (as
in ‘The dogs are barking’).
Second, it is arguable that there are many expressions having surface forms quite
different from ‘the F’ or ‘an F’ that could count as being descriptions. For example, it
seems quite plausible that possessives like ‘my father’ are descriptions (as in ‘the father
of me’). Russell (1905) also proposed that some ordinary proper names could be con-
strued as definite descriptions in disguise. Thus a name like ‘Aristotle’ might be taken
as shorthand for ‘the student of Plato who taught Alexander, wrote The Nichomachean
Ethics etc.’ Furthermore, as we will see, it has been argued that pronouns like ‘it’ might
“stand proxy” for descriptions. So, for example, the pronoun in (1) might be taken to
stand proxy for the corresponding definite description in (1’).

(1) A man came in the room. He turned on the TV.

(1') A man came in the room. [The man who came in the room] turned on
the TV.

Correlatively, it has been suggested that temporal anaphors like ‘then’ and modal ana-
phors like ‘that’ (in ‘that would have been unfortunate’) are kinds of descriptions.
If matters weren’t complicated enough, recent work in philosophy and linguistics
has called into question the very idea of whether the determiners ‘the’ and ‘a’ make the
semantical contribution that philosophers often suppose. The problem is that only a
small subset of natural languages have elements like the articles ‘the’ and ‘a’, and even
for many languages that do have these elements (like English), it may well be that they
are grammatical “particles” with no semantic contribution of their own. Indeed, some
 DESCRI PT I O N S

recent analyses of descriptions have advanced the view that what contributions the
determiners ‘the’ and ‘a’ make are not semantical, but rather pragmatic in nature. All
of these possibilities will be discussed in due course, but for now we will begin with the
analysis of ‘the F’ and ‘an F’ first taken up in Russell (1905, 1919).

2 Russell’s Theory of Descriptions

The key idea of Russell’s proposal is that a sentence like (2) containing an indefinite
description, is understood to have the logical form in (2'),

(2) An F is G
(2') x(F(x) & G(x))

and a sentence like (3) containing a definite description is understood to have the logi-
cal form in (3').

(3) The F is G
(3') x(F(x) &y(F(y)Ÿ x=y) & G(x))

Boiled down to its simplest nontechnical form, the idea is that an expression of the form
in (3) is shorthand for three claims:

i) There is an F.
ii) At most one thing is F.
iii) Something that is F is G.

Following Neale (1990) we will find it useful to substitute (iii') for (iii), which retains
Russell’s truth conditions and allows us to extend the theory to plural descriptions in a
natural way.

iii') Everything that is F is G.

Thus tweaked, Russell’s analysis is that the use of a definite description in a sentence
involves an existence claim, a uniqueness claim, and a maximality claim.

3 Motivations for the Theory of Descriptions

There are three main motivations for the theory of descriptions; the first is metaphysi-
cal, the second involves semantical concerns in the philosophy of language, and the
third is epistemological. Let’s start with the metaphysical motivation.
Consider a negative existential sentence like (4).

(4) The Present King of France does not exist

Because definite descriptions are devices of quantification on Russell’s view they can enter
into scope relations with other operators—in this case, for example, negation. Accord-
ingly, there is a kind of ambiguity in (4), between the following two logical forms.

381
 PETER LU D L OW

(5a) not ([the x: x is the present King of France] exists(x))

(5b) [the x: x is the present King of France] not (exists(x))

If one wants to avoid the ontological entanglements of nonexistent objects, then one is
free to say that (5b) is false (since it involves quantifying over things that don’t exist)
but that (5a) is true (since it is not the case that there is a present King of France). What
is negated in (5a) is not a claim about some particular individual, but rather a general
claim about the world—in effect a claim that the world contains exactly one individual
that is presently the King of France and that whoever is presently the King of France
exists.
Russell also had a number of concerns that today we might call “semantical.” Con-
sider the expressions ‘The Morning Star’ and ‘The Evening Star’. Both refer to (or at
least denote) the planet Venus, but there are contexts in which it seems incorrect to
say that they have the same meaning. For example, it was as an astronomical discovery
that The Morning Star was identical to The Evening Star, so it would be odd to treat an
expression like ‘The Morning Star is The Evening Star’ as merely asserting some object
to be self-identical. In a similar vein, if we utter (6),

(6) George wondered whether The Morning Star is The Evening Star.

we are most likely not saying that George was curious about whether Venus was identi-
cal to itself.
Frege proposed that that the solution to this puzzle involved the introduction of
senses—abstract objects that fix the referents of these expressions, each having a differ-
ent cognitive significance. In the case of (6), the Fregean solution would be to say that
there are different senses attached to ‘The Morning Star and ‘The Evening Star’. Russell
held that descriptions could do the work that senses are supposed to do. If we take the
names in (6) as standing proxy for definite descriptions, then (6) can be unpacked as
in (6').

(6') George wondered whether the star that appears in the morning is identical
to the star that appears in the evening.

Russell saw that scope relations are relevant here as well. So, for example, sentences
like (6) evince what are sometimes called de dicto/de re ambiguities. There are circum-
stances under which George has some object in particular in mind and is wondering, of
that object, whether it is the star that appears in the evening. We can use (6) to report
this fact as well; in this case we may think of the description as taking wide scope rela-
tive to the propositional attitude verb ‘wondered’, yielding a logical form like (7).

(7) the star x that appears in the morning is such that George wondered
whether x is identical to the star that appears in the evening

And of course if George has gone mad and is in fact wondering about the object’s self-
identity, this may be represented as in (8), where both descriptions have wide scope.

(8) the star x that appears in the morning is such that the star y that appears in
the evening is such that George wondered whether x is identical to y.

382
 DESCRI PT I O N S

Metaphysical and semantical concerns were important to Russell in his (1905) paper,
but epistemological concerns were no less significant. This became particularly clear
when he authored his (1910–11) paper “Knowledge by Acquaintance and Knowledge
by Description.” In that paper, Russell distinguished between objects that we are directly
acquainted with and objects that we only know under a description. So, for example, I
might know myself by acquaintance, but I know the tallest man in Iowa only under a
description.

4 Objections to the Theory of Descriptions

The theory of descriptions has encountered its fair share of criticism. This criticism has
ranged from contentions that Russell simply got the truth conditions wrong in impor-
tant cases to nagging worries about the details of the proposal—in particular worries
relating to the nature of the descriptive content. As we will see, none of these concerns
have been completely ameliorated.

4.1 The Challenge to Russell’s Truth Conditions

Strawson (1950) objected that Russell’s theory is simply incorrect about the truth con-
ditions of sentences like ‘The present King of France is bald’. According to Russell’s
analysis, this sentence is false (since it contains an existence claim to the effect that
there is a present King of France), but, according to Strawson, this does not conform
to our intuitions about the truth of an utterance of that sentence. In Strawson’s view,
an utterance of the sentence in a world where there is no present King of France is
neither true nor false; perhaps the sentence has a truth value gap, or perhaps it fails to
express a determinate proposition (Strawson vacillated on this), but either way it does
not appear to be false. Strawson held that this fact supported a referential interpretation
of expressions like ‘The present King of France’—in effect, an interpretation on which
‘The present King of France’ is a logically proper name that only has its referent as its
content.
If there is no present King of France, then an utterance containing such an expression
is somehow defective on Strawson’s view. It is as if I looked into my desk drawer, not
allowing you to see what I was looking at, and said ‘that is a fine green one.’ Strawson
held that utterances like these do not entail the existence of a fine green one or the
present King of France, but rather presuppose their existence. If the expressions fail to
refer, then there is a presupposition failure and the utterance fails to have a determinate
truth value. (Notice that this sort of failure is not supposed to undermine the meaning-
fulness of the sentences that we utter; for Strawson, sentences are meaningful in and
of themselves, independently of the utterance situation. Utterances of meaningful sen-
tences may be true or false, or, if here is a presupposition failure, they may be neither.)
Does this whole debate come down to a case of intuition swapping? Thomason (1990:
327) and Soames (1976: 169) seemed to think so, and Strawson himself (1964) also
came to doubt whether the entailment vs. presupposition debate could be settled by
“brisk little formal argument[s].” Neale (1990) maintained that the matter could be
settled in Russell’s favor, and supported the claim by collecting a number of previously
observed cases in which intuitions about truth conditions clearly do not support Straw-
son’s view. For example, ‘My mother is dating the present King of France’ seems clearly
false, as does ‘The present King of France cleans my swimming pool’, and he concluded

383
 PETER LU D L OW

that these are clearly cases where the Strawsonian truth conditions have gone awry. But
see von Fintel (2004) for a Strawsonian account of these intutions.

4.2 Donnellan’s Distinction and the Argument from Misdescription

Donnellan (1966) observed that there is a sense in which Strawson and Russell are both
right (and both wrong) about the proper analysis of descriptions. He argued that definite
descriptions can be used in (at least) two different ways. On a so-called attributive use,
a sentence of the form ‘The F is G’ is used to express a proposition equivalent to ‘What-
ever is uniquely F is G’. For example, on seeing murder victim Smith’s badly mutilated
corpse, Detective Brown might say “The murderer of Smith is insane” thereby commu-
nicating the thought that some unique individual murdered Smith and that whoever
that individual is, he/she is insane. Alternatively, on a referential use, a sentence of the
form ‘The F is G’ is used to pick out a specific individual, x, and say of x that x is G. For
example, suppose Jones is on trial for Smith’s murder and is behaving quite strangely at
the defense table. I point at Jones and say, “The murderer of Smith is insane,” thereby
communicating the thought that Jones is insane (whether or not Jones is the actual
murderer).
Kripke (1977) responded to Donnellan by arguing that the Russellian account of
definite descriptions could, by itself, account for both referential and attributive uses;
the difference between the two cases could be entirely a matter of pragmatics. Here is
the idea: Grice showed us that there is an important distinction to be made between
what one literally says by an utterance and what one intends to communicate (what
one means) by that utterance. To take a famous example of Grice’s, I might write a let-
ter of recommendation for a student saying that he is very punctual and has excellent
handwriting. Now what I have said is something about the student’s punctuality and
handwriting, but what I meant was that this is a very weak student.
In a similar vein, we could say that when I use a description referentially—say in
Donnellan’s courtroom case—I am literally making a general claim to the effect that
there is a murderer of Smith and that he is insane, but what I mean by that utterance is
that Jones is insane. That is, when I say ‘The murderer of Smith insane’ what I literally
say is that exactly one person, x, is such that x murdered Smith and x is insane, but in
that context I would succeed in communicating the singular proposition (about Jones)
that Jones is insane. Kripke gave several reasons for thinking that this Gricean solution
was preferable to an ambiguity thesis. One reason was a general methodological point
that one should not introduce ambiguities blithely—doing so is a kind of philosophical
cheat.
One of the advantages of employing the Gricean distinction between the proposi-
tion literally communicated and the proposition meant is that it offers an account for
our being ambivalent about Donnellan’s (1966) misdescription cases. In the courtroom
case discussed above, I might say “Smith’s murderer is insane,” and still say something
true even if the crazy man at the defense table is entirely innocent of the charges and
the actual murderer, who is miles from the courtroom, is quite sane. At the same time
there is some pull to say that in such a case one is saying something false too. We can
say that this is a case where what we literally said was false but that what we intended to
communicate—the proposition meant—was true. The two-level theory thus accounts
for our conflicting intuitions.

384
 DESCRI PT I O N S

Similarly, Hornsby (1977) gave the case of my observing the man ranting at the
defense table and (me) saying, “The murderer of Smith is insane” not realizing that the
man at the table is both innocent and quite sane, while the actual murderer is at large
and quite insane. Again we are ambivalent about the truth of what I say, and as Neale
(1990: 91–93) observed, the distinction between the proposition literally expressed and
the proposition meant allows us to understand why. In this case, the proposition literally
expressed is true, but what I intend to communicate is mistaken.
Unfortunately, there are cases where the two-stage theory doesn’t appear to be suf-
ficient. For example, consider a case where we are at the crime scene, and unbeknownst
to Detective Brown there is not one murderer but several—suppose there were several
perpetrators and they were all mad members of an evil cult. When Brown utters the sen-
tence ‘The murderer of Smith is insane’ has he said something true or false? Again we
are in two minds about the matter, but this time the distinction between what is literally
said and what is meant is no help. Let’s call this the “residue of the problem of misde-
scription” (although, strictly speaking, it is not in and of itself a misdescription case).

4.3 The Argument from Incompleteness

Another persistent problem for the classical Russellian theory of descriptions has been
the charge that it fails to account for the problem of “incomplete descriptions” (for discus-
sion see Donnellan 1968, Hornsby 1977, Devitt 1981: chapter 2, 2004, Wettstein 1981,
Recanati 1986, Salmon 1982, Soames 1986, Neale 1990: chapter 2, and Reimer 1992).
In addition, Kripke (1977), while defending Russell’s theory of description against the
problem of misdescription, allows that the argument from incomplete descriptions might
be enough of a problem to force us to accept referential interpretations of descriptions.
The worry, initially raised in Strawson (1950), is that if I say ‘the table is covered with
books’, I do not mean to be suggesting that there is only one table in the world. Unfortu-
nately, that seems to be precisely what the Russellian theory of descriptions is committed
to. (Recall that on the Russellian analysis my utterance is shorthand for ‘there is a table
and only one table and every table is covered with books’.)
One strategy for dealing with this problem is that the context may provide us the
means to flesh out the description. For example, perhaps descriptions can be fleshed
out appropriately if we allow referential devices to be inserted into the description. The
suggestion is that when we speak of the table we are implicitly specifying a spatial coor-
dinate—in effect, we are saying ‘the table over there’.
Neale (1990) has argued that whatever we may want to say about the problem of
incompleteness, it is not very effective as an argument for the referential analysis of
descriptions. For example, it appears that there are numerous examples involving quan-
tified expressions that suffer the same fate as incomplete descriptions. I can say ‘Every-
one came to the party’, not intending to mean everyone in the world. Or I might say, as
Yogi Bera once did, ‘Nobody goes there anymore, it’s too crowded’. It certainly appears
that what is going on in these cases is similar to the cases of incomplete definite descrip-
tions and that there should be a single strategy for accounting for these different cases
of “incompleteness.” (But see Devitt 2004, who argues that these cases are genuinely
different in kind, since descriptions are far more apt to be used referentially.)
Recent work has argued that the problem of incomplete definite descriptions can
be accounted for if we pursue an appropriate theory of quantifier domain restriction.
Stanley and Szabo (2000), for example, take this approach, suggesting that context can

385
 PETER LU D L OW

restrict the domain of quantification. On their proposal, context can even shift within
a sentence, allowing us to make sense of an utterance like ‘The dogs barked at the
dogs’, where I mean to say that one group of dogs barked while the other group perhaps
suffered in silence. (As their proposal is applicable to all quantified expressions and not
just the theory of descriptions, I cannot focus on it here, except to note that it has spawned
a great deal of discussion, including Bach 2000, Neale 2000, and Lepore 2004.)
It is important to note (following Szabo 2000 and Ludlow and Segal 2004) that even
with a fully functional account of quantifier domain restriction there is a lingering prob-
lem here too. Let’s call it the residue of the incompleteness problem. Consider cases
like (9)

(9) Put the book on the book

Now clearly, the first use of ‘the book’ cannot have the same domain of quantification
as the second use, since that would put two books in the domain of quantification and
it would mean that both descriptions in the sentence are incomplete. But one wonders
how valid a domain-shift analysis is here. Is there really a shift in the domain of quanti-
fication between the first utterance of ‘the book’ and the second utterance of that noun
phrase?

5 Dissolving Descriptions

5.1 Unified Theories of Definite and Indefinite Descriptions

Given the Gricean resources discussed above, one might speculate that the distinction
between definite and indefinite descriptions can be collapsed—that is, perhaps ‘the’
and ‘a’ have the same literal meaning and the only relevant distinction between them
is pragmatic.
The motivation for this idea would be as follows. Very few natural languages
have what we would recognize as definite and indefinite descriptions. In most Slavic
languages, for example, ‘the man’ and ‘a man’ would both be expressed in the same
way—in the determiner-free Russian equivalent of ‘man’. Perhaps it is just our infatua-
tion with surface grammatical form that leads us to think that English or Italian really
has two different logical elements corresponding to surface forms ‘the’ and ‘a’. Per-
haps there is a single logical element with different pragmatic application conditions.
That is, perhaps ‘the’ and ‘a’ have the same literal meaning—i.e., for both ‘An F is G’
and ‘The F is G’ the literal semantics is exhausted by the interpretation that [x:
Fx](Gx) receives in a standard truth-conditional semantics. A number of linguists and
philosophers have entertained this idea in recent years, including Kempson (1975),
Breheney (1999), Szabo (2000), Zvolensky (1997), Ludlow and Segal (2004) and
a version of the idea is at least considered in Heim (1982) and Kamp and Reyle
(1993).
Obviously there are important differences between our application of the terms ‘the’ and
‘a’, but it does not follow that these differences in application are part of their semantical
content. Indeed, many synonyms customarily are put to different uses. Take, for example,
Grice (1961, 1975) on the distinction between ‘but’ and ‘and’. On Grice's account ‘but’
and ‘and’ literally mean the same thing, but different “conventional implicatures” are
associated with them; ‘but’ implicates a sense of contrast between the conjuncts.

386
 DESCRI PT I O N S

Ludlow and Segal (2004) offer a similar story about ‘a’ and ‘the’. Following a standard
assumption in traditional grammar, they argue that ‘the’ signals that the object under
discussion is given or familiar in the conversational context. Noun phrases fronted by
the determiner ‘a’ signal that they involve new information. The idea advanced by
Ludlow and Segal, however, is that this slender bit of information, combined with
Gricean principles, is sufficient to generate the uniqueness implication that is carried
by a definite description. That is, an existential claim that there is an F that is G, plus a
signal that this is given (familiar) information, is often enough to allow us to implicate
that there is unique a F that is G.
Armed with this idea, let’s return to the some of the loose threads that we left hang-
ing in section 4 above—the residue of the misdescription problem and the residue of the
incompleteness problem. As we will see, the unified treatment of definite and indefinite
descriptions may provide us an entering wedge for cracking open these puzzles. Let’s
look at the misdescription case first.

The residue of the problem of misdescription

Recall the case in which Detective Brown says ‘the murderer of Smith is insane’, incor-
rectly believing that there was one murderer when in fact the crime was committed by
a ghoulish cult of insane individuals. There is a sense in which Brown spoke falsely, but
there is also clearly some pull for us to say that what he said was true. As we saw in sec-
tion 4, the distinction between the proposition literally expressed and the proposition
meant was not sufficient to account for this ambivalence on our part. But according to
Ludlow and Segal (2004), if we combine this pragmatic distinction with the unified
analysis of definite and indefinite descriptions, there is something we can say about this
last bit of residue.
According to the unified analysis of descriptions, what Detective Brown literally
expresses is not the idea that there was a unique murderer of Smith who is insane. To
the contrary, he literally expresses the proposition that there is at least one murderer of
Smith who is insane. By applying Gricean principles in this context we have made out
that Brown intends to say that there is a unique murderer of Smith and that he is insane.
We are pulled in two directions by this case because what Brown has said is literally true
but what he intended to communicate was, strictly speaking, false.

The residue of the problem of uniqueness

In section 4.3 we considered cases like (9), which did not seem to yield in a natural way
to the device of quantifier domain restriction.

(9) Put the book on the book

But as Szabo (2000) and Ludlow and Segal (2004) have argued, if we combine quanti-
fier domain restriction with the unified analysis of descriptions, the problem seems more
amenable to solution. The idea is the following: What one literally expresses in (9) is
that the hearer should put a book on a book. Pragmatics helps us to make out that one
book in particular is being spoken of (for example the book being pointed at or looked
at), and where it is to be moved.

387
 PETER LU D L OW

5.2 Descriptions as Predicates

At the beginning of this chapter I noted that some uses of the expressions ‘an F’ and ‘the
F’ are argued to be predicational, rather than quantificational. So, for example, consider
the following cases.

(10) John is a lawyer

(11) John is the mayor of Pittsburgh

It is not unusual to think of ‘is a lawyer’ and ‘is the mayor of Pittsburgh’ as predicates.
In the case of (10), we intuitively are not saying that there is a lawyer such that John is
identical to that lawyer. Indeed, as Williams (1983) observed, this appearance is even
more pronounced if we consider cases like (12).

(12) John is not a lawyer

Example (12) does not seem to have the meaning that there is a lawyer such that John
is not identical to that lawyer. It is more natural to take an utterance of (12) as denying
that John has a certain property.
If this is right, then the copula ‘is’ is really just the ‘is’ of predication and semantically
dispensable (significantly, it routinely does not appear in many other languages unless it
is needed to carry tense or other inflectional information). Likewise, we might say that
the indefinite article ‘a’ is semantically inert as well—a mere “grace note” in Higgin-
botham’s (1987) terminology. In this vein, one widely held view, due to Kamp (1984)
and Heim (1982), is that we could regard indefinite descriptions as expressions contain-
ing free variables which are bound by adverbs of quantification (in the sense of Lewis
1975) interpreted as existentially quantified by the model theory, or perhaps bound by
implicit existential closure operators (Diesing 1992).
For linguists it is now standard to think of indefinite descriptions following the copula
as always being predicational, and it is a widespread belief that definite descriptions fol-
lowing the copula are often predicational. Philosophers have also been attracted to this
view. See, for example, Geach (1962: section 39), Wiggins (1965: 42 ff.), Kim (1970:
211 ff.), Wilson (1978), Smiley (1981) and Higginbotham (1987) among others. Does
it follow that we must admit two different kinds of descriptions (quantificational and
predicational)? We saw that the analysis of singular and plural descriptions can be uni-
fied. Can we perform the same unification trick here?
Fara (2001) compellingly made the case that all uses of definite descriptions are predi-
cational—even descriptions in subject position. The idea is that if I say ‘The present
King of France is bald’, I am saying something like the following: Some x is such that x
is the unique present King of France and x is bald. The interesting riff here is the notion
that the uniqueness condition is built into the predicate. Extending the Sharvy (1980)
analysis of plural descriptions, we get the following, where F is the set of all Fs, or the
mass of all substance that is F:

(13)
an F = F;
the F = {x: x  F & y (y FŸ y = x}

388
 DESCRI PT I O N S

But again we might consider collapsing the distinction between definite and indefinite
descriptions here, and say that there is a single rule that interprets all descriptions—
definite, indefinite, singular, plural, (possibly even mass and generic descriptions)—as
predicates. When we supplement this with the rejection of the uniqueness/maximality
clause we get this result:

(13') the F/the Fs/an F/some Fs = F

If this is right then the determiners ‘a’ and ‘the’ are both “grace notes” not just in special
positions but across the board

5.3 Descriptions as Particles

What then are we to say about the determiners ‘the’ and ‘a’? As I noted earlier, they are
rare in the world’s languages and are not even employed in many Indo-European lan-
guages (for example Slavic languages, with the exception of southern Slavic languages
like Bulgarian). But even in languages that deploy these articles it is clear that they are
not behaving as operators. Goldsmith (2009) observes that in both ancient Greek and
contemporary German the definite article can appear in several places within the same
noun phrase (so that, for example, saying ‘the sad clown is crying would involve the
equivalent of ‘the clown the sad is crying’ in ancient Greek). In effect, the determiner
is working as a particle that is carrying syntactic information (for example case) that
is helping to glue the adjective to the noun it modifies. In other instances from con-
temporary German, Goldsmith introduces example constructions with two determiners
that can only be understood as having one quantifier, and closely related constructions
with one determiner that can only be understood as having two operators. Finally, many
of the apparent wide scope readings of descriptions can be accounted for as pragmatic
phenomena (for example, if there is only one familiar or given satisfier of a narrow scope
description there may be the illusion of wide scope).

6 Conclusion
A close study of the syntax and semantics of natural language suggests that constructions
of the form ‘the F’ and ‘an F’ are not only rare in natural languages but misleading in
languages like English. These constructions really don’t carry the logical significance
that Russell and subsequent authors have thought. However, Russell’s core insight
remains intact: Many apparently referential constructions are in fact quantificational,
and many terms that appear to be referential are in fact predicational. What Russell
didn’t see was that surface grammar is even more deceptive than he realized. Elements
like ‘the’ and ‘a’ do not encode quantifiers or uniqueness clauses. The task for philoso-
phers of language now is figuring out what properties of natural language do encode that
information.

Note
1 Thanks to Gary Ostertag for helpful comments.

389
 PETER LU D L OW

References
Bach, K., (2000) “Quantification, Qualification, and Context: A Reply to Stanley and Szabo,” Mind and
Language 15, 262–283.
Breheny, R, (1999) Context Dependence and Procedural Meaning: The Semantics of Definites. Ph.D. the-
sis, Department of Linguistics, University College, London.
Devitt, M., (1981) Designation, New York: Columbia University Press.
—— (2004) “The Case for Referential Descriptions,” in A. Bezuidenhout and M. Reimer (eds.), Descriptions
and Beyond: An Interdisciplinary Collection of Essays on Definite and Indefinite Descriptions. Oxford: Claren-
don Press, 280–305.
Devitt, M., and K. Sterelny, (1999) Language and Reality: An Introduction to the Philosophy of Language, 2nd
edition, Cambridge, MA: MIT Press.
Diesing, M., (1992) Indefinites, Cambridge, MA: MIT Press.
Donnellan, K. S., (1966) “Reference and Definite Descriptions,” Philosophical Review 75, 281–304.
Donnellan, K. S., (1968). “Putting Humpty Dumpty Back Together Again,” Philosophical Review 77,
203–215.
Donnellan, K. S., (1978) “Speaker Reference, Descriptions, and Anaphora,” in P. Cole (ed.), Syntax and
Semantics 9: Pragmatics. New York: Academic Press, 47–68.
Fara, D.G., (2001) “Descriptions as Predicates,” Philosophical Studies 102, 1–42. Published under the name
“Delia Graff.” von Fintel, K., (in 2004) “Would You Believe It? The King of France is Back! (Presupposi-
tion and Truth-Value Intuitions),” in A. Bezuidenhout and M. Reimer (eds.), Descriptions and Beyond:
An Interdisciplinary Collection of Essays on Definite and Indefinite Descriptions. Oxford: Oxford University
Press.
Geach, P., (1962) Reference and Generality, Ithaca: Cornell University Press.
Goldsmith, L., (2009) “Articles, Particles, Syntax,” Manuscript, Northwestern University, Dept. of
Philosophy.
Grice, H. P., (1961) “The Causal Theory of Perception,” Proceedings of the Aristotelian Society, Suppl. Vol.
35, 121–152.
Grice, H. P., (1975) “Logic and Conversation,” in J. Cole and J. Morgan (eds.), Syntax and Semantics, Vol 3:
Speech Acts, New York: Academic Press.
Heim, I., (1982) The Semantics of Definite and Indefinite Noun Phrases, Ph.D. Thesis, University of Mas-
sachusetts, Amherst.
—— (1990) “E-Type pronouns and Donkey Anaphora,” Linguistics and Philosophy 13, 137–178.
Higginbotham, J., (1987) “Indefiniteness and Predication,” in E. Reuland and A. ter Meulen (eds.), The
Representation of (In)definiteness, Cambridge, MA: MIT Press, 43–70.
Hornsby, J., (1977) “Singular Terms in Contexts of Propositional Attitude,” Mind 86, 31–48.
Kamp, H., (1984) “A Theory of Truth and Semantic Interpretation,” in J. Groenendijk et al., (eds.), Formal
Methods in the Study of Natural Language, Amsterdam Centre. Reprinted in J. Groenendijk et al. (eds.),
Truth, Interpretation, and Information, Dordrecht: Foris, 1–43.
Kamp, H., and U. Reyle, (1993) From Discourse to Logic, Dordrecht: Kluwer Academic Publishers.
Kempson, R., (1975). Presupposition and The Delimitation of Semantics. Cambridge: Cambridge University
Press.
Kim, J., (1970) “Events and Their Descriptions: Some Considerations,” in N. Rescher (ed.), Essays in Honor
of Carl G. Hempel. Dordrecht: D. Reidel, 199–215.
Kripke, S., (1977) “Speaker Reference and Semantic Reference,” in P.E. French, T.E. Uehling, and
H.K.Wettstein (eds.),Contemporary Perspectives in the Philosophy of Language, Minneapolis: University of
Minnestoa Press, 6–27.
Lepore, E., (2004) “An Abuse of Context in Semantics: The Case of Incomplete Definite Descriptions,” in
A. Bezuidenhout and M. Reimer (eds.), Descriptions and Beyond: An Interdisciplinary Collection of Essays on
Definite and Indefinite Descriptions. Oxford: Oxford University Press.
Lewis, D., (1975) “Adverbs of Quantification,” in E. Keenan (ed.), Formal Semantics of Natural Language.
Cambridge: Cambridge University Press, 3–15.
Ludlow, P. and G. Segal, (2004) “On a Unitary Semantical Analysis for Definite and Indefinite Descrip-
tions,” in A. Bezuidenhout and M. Reimer (eds.) Descriptions and Beyond: An Interdisciplinary Collection of
Essays on Definite and Indefinite Descriptions. Oxford: Oxford University Press.
Neale, S., (1990) Descriptions, Cambridge, MA: MIT Press.
—— (2000) “On Being Explicit: Comments on Stanley and Szabo, and on Bach,” Mind and Language 15,
284–294.

390
 DESCRI PT I O N S

Recanati, F., (1986) “Contextual Dependence and Definite Descriptions,” Proceedings of the Aristotelean
Society 87, 57–73.
Reimer, M., (1992) “Incomplete Descriptions,” Erkenntnis 37, 347–363.
Russell, B., (1905) “On Denoting,” Mind 14, 479–493.
—— (1910–11) “Knowledge by Acquaintance and Knowledge by Description,” Proceedings of the Aristo-
telean Society. Reprinted in Mysticism and Logic, London: George Allen and Unwin, 1917, and New York:
Doubleday, 1957.
—— (1919) Introduction to Mathematical Philosophy, London: George Allen and Unwin.
—— (1957) “Mr. Strawson on Referring,” Mind 66, 385–389.
Salmon, N., (1982) “Assertion and Incomplete Descriptions,” Philosophical Studies 42, 37–45.
Sharvy, Richard (1980) “A More General Theory of Definite Descriptions,” The Philosophical Review 89,
607–623.
Smiley, T.J., (1981) “The Theory of Descriptions,” Proceedings of the British Academy 67, 331–337.
Soames, S., (1976) A Critical Examination of Frege’s Theory of Presupposition and Contemporary Alterna-
tives, Ph.D. thesis, MIT Dept. of Linguistics and Philosophy.
—— (1986) “Incomplete Definite Descriptions,” Notre Dame Journal of Formal Logic 27, 349–375.
Stanley, J. and Z. Szabo, (2000) “On Quantifier Domain Restriction,” Mind and Language 15, 219–261.
Strawson, P. F., (1950) “On Referring,” Mind, 59, 320–334.
—— (1964) “Identifying Reference and Truth-Values,” Theoria 3, 96–118.
Szabo, Z., (2000) “Descriptions and Uniqueness,” Philosophical Studies 101, 29–57.
Thomason, R., (1990) “Accomodation, Meaning, and Implicature: Interdisciplinary Foundations for Prag-
matics,” in P. Cohen, J. Morgan, and M. Pollack (eds.), Intentions in Communication. Cambridge, MA:
MIT Press, 325–363.
Wettstein, H., (1981) “Demonstrative Reference and Definite Descriptions,” Philosophical Studies 40,
241–257.
Wiggins, D., (1965) “Identity Statements,” in R.J.Butler (ed.), Analytical Philosophy 2nd edition. Oxford:
Basil Blackwell.
Williams, E., (1983) “Semantic vs. Syntactic Categories,” Linguistics and Philosophy 6, 423–446.
Wilson, G., (1978) “On Definite and Indefinite Descriptions,” The Philosophical Review 87, 48–76.
Zvolensky, Z. (1997) “Definite Descriptions: What Frege got Right and Russell Didn’t,” Aporia 7, 1–16.

391
 BERN H A RD N I C K E L

3.8
P L URALS
Bernhard Nickel

The study of plurality has implications for our understanding of the semantics of natu-
ral language and broader issues in philosophy, largely by bearing on matters logical.
Perforce, I’ll emphasize some issues at the expense of others, and I’ll indicate directions
for further study along the way. Other overviews, with different points of emphasis, are
Lønning (1997), Schein (2006), and Rayo (2007).
In the first instance, plurality is a morphological notion and at least in English, it dis-
tinguishes between count and mass nouns. While the former can exhibit it (dogs, trees,
tables, chairs), the latter cannot (*informations, *advices, *broccolis). The semantics of
plurality are far less straightforward.

1 More Than One?

The simplest hypothesis is that plural morphology means more than one. That cannot be
the whole story, however.

(1) a. The President met with bankers on Friday.

b. The President met with more than one banker on Friday.
(2) a. The President didn’t meet with bankers on Friday.
b. The President didn’t meet with more than one banker on Friday.
(3) a. Did you see whales on your trip?
i. Yes, I saw (just) one.
ii. # No, I saw (just) one.
b. Did you see more than one whale on your trip?
i. # Yes, I saw (just) one.
ii. No, I saw (just) one.

While (1a) and (1b) convey exactly the same information, (2a) and (2b) do not. In the
scope of negation, the plural seems to mean one or more. The same is true of questions,
as (3a) and (3b) illustrate. Since (3(a)i) but not (3(a)ii) is a proper response to (3a),
the plural seems to mean one or more. Example (3b) shows that if it meant more than
one, we should expect the opposite pattern of felicitous answers. For this reason, various
theorists have suggested that the plural always means one or more, and that the “more
than one” interpretation is an implicature of some sort. Krifka (2004), Sauerland et al.
(2005), and Spector (2007) implement this strategy in the context of truth-conditional
semantics, Kamp and Reyle (1993) in Discourse Representation Theory.
 PLU RAL S

A similar variability, with an explicit concern for compositionality, was pointed out
by Chomsky (1975).

(4) That unicycle has wheels.

(5) Unicycles have wheels.

It seem as if the bare plural wheels means more than one in (4) but not in (5). The latter
is what has become known as a dependent plural (de Mey, 1981). Chomsky argued that
this presents a problem for compositionality. The plural makes different contributions
to the meaning of the two sentences, but we cannot predict whether the plural means
more than one without knowing what kind of sentence we’re interpreting.
Here, too, several theorists (e.g. Spector, 2007, Zweig, 2008) have suggested that the
plural is semantically number-neutral and that the appearance of a plural meaning in
(4) is the result of an implicature. Working out the details of this approach may have
far-reaching implications for our theory of implicature. Unlike ordinary scalar implica-
tures which can be canceled, we cannot cancel the “more than one” implicature here,
as shown by

(6) # That unicycle has wheels, though I don’t mean to suggest that it has
more than one wheel.

Dependent plurals thus seem to require some other account of implicature, perhaps one
on which implicatures are computed alongside (or as part of) semantic composition
(Chierchia, 2002, 2006).

2 Collectivity, Distributivity, and Cumulativity

There is yet more variability to account for.

(7) All of the children slept.

(8) All of the children gathered in the yard.
(9) Three children ate four pizzas.

The predicate in (7) is what is usually called distributive: if it can be truly applied to a
plurality, it can also be truly applied to each of its members. This connection fails in (8),
making it collective: even if the children gathered in the yard and Jane is one of them,
Jane didn’t gather in the yard. Example (9) exhibits a cumulative reading: there were
three children eating pizzas and between them four pizzas were eaten.
Collectively interpreted sentences such as (8) cannot be analyzed in terms of dis-
tributively interpreted sentences. We thus need to take at least collectivity as basic. But
perhaps collectivity is the only basic phenomenon, and what looks like further inter-
pretations are simply the result of indeterminacy. On this view, (7) and (8) predicate a
property of a plurality, and one way for the property to truly apply to it is for it to apply
to each of its members. Compare: there are many ways for the sentence there are potatoes
in the pantry to be true—the potatoes might be in bags or in a box, but these are just ways
for a single reading to be true (see Harnish, 1991, Higginbotham, 1981, Katz, 1977).
The currently most widespread view holds that the distributive interpretation is a
genuine reading (see Gillon, 1987, for a range of arguments). But what determines

393
 BERN H A RD N I C K E L

whether a sentence is interpreted collectively or distributively? It could be a matter

of the lexical meaning of the verb, so that the meaning of sleep includes the meaning-
postulate that whenever it is true of a collection, it’s also true of each member of the
collection (Scha, 1981). But we see readings that are inexplicable on this approach (see
Winter, 2000).

(10) The girls wore a dress.

On the meaning-postulate approach, we can only predict the reading of (10) on which
there is a single dress for all of the girls because there is no other operator that the exis-
tential quantifier a dress can enter into scope relations with.
Once we posit such an operator, (10) can be interpreted as the girls are such that, for
each of them, there is a dress that she wore. This is a distributive operator: It distributes the
predicated property to each thing in the plurality picked out by the subject.
One could think that the operator is part of the plural noun phrase (NP), so that
collectivity or distributivity is a feature of the NP (see Lakoff, 1972 and Gillon, 1987,
1990, 1992). It is more likely, however, that this distributive operator is part of the verb
phrase (VP), based on examples combining distributivity and collectivity (see, e.g. Beck
and Sauerland, 2000, Landmann, 2000, Lasersohn, 1995, McKay, 2006, Pietroski, 2005,
Schein, 1993, Schwarzschild, 1996, Winter, 2000).

(11) The children woke up and gathered in the yard.

If the NP had to be read either collectively or distributively, one of the predicates

wouldn’t apply. But on the VP-analysis, (11) is essentially equivalent to (12).

(12) The children are such that each of them woke up and they gathered in
the yard.

Gillon (1987, 1990, 1992) and Schwarzschild (1994, 1996) have argued that in order to
capture all the semantic possibilities, the analysis needs to be more complex still. Some
sentences aren’t collective, but the predicate cannot be distributed all the way down
to the individuals involved, either. Suppose we’re buying apples. Each apple costs fifty
cents, and we buy twelve. (13a) and (13b) are true in that case.

(13) a. The apples cost fifty cents.

b. The apples cost six dollars.

In (13a), the VP contains a distributive operator, while it is absent in (13b). In the right
context, perhaps once we’re told that the apples come prewrapped in six-packs, we can
also describe the situation with (14).

(14) The apples cost three dollars.

We’ve already exhausted the possibilities with respect to the distributive operator.
When it’s present, (13a) is true while (13b) and (14) are false. When it’s absent, (13b)
is true and the other two false. Either way, we cannot predict the true reading of (14).
We’d like a formal way of roughly capturing the paraphrase (15).

394
 PLU RAL S

(15) The apples are such that each six-pack among them costs three dollars.

Here, the predicate is distributed to collections that in turn make up the plurality, rather
than to the individuals in that plurality.
Formally, we accomplish this by allowing the distributive operator to distribute the
predicate to contextually salient subclasses of the plurality. One particularly striking
feature of this context-dependence is that certain subclasses are completely unavail-
able to ordinary speakers as targets for distribution (first pointed out by Scha, 1981). In
Figure 3.8.1, we can easily group the lines together in such a way as to make (16)
true.

R1

R2

Figure 3.8.1 Rectangles

(16) The sides of R1 run parallel to the sides of R2.

But we cannot impose a contextual grouping on the lines in Figure 3.8.2 that
verifies (17).

Figure 3.8.2 Lines

(17) The single lines run parallel to the double lines.

Investigating why there are such constraints promises to shed light on how context
can and cannot influence interpretation. The availability of contextual restrictions on
distributivity may also be relevant to other issues, such as the interpretation of definite
descriptions.

395
 BERN H A RD N I C K E L

2.1 Descriptions
Definite descriptions in the singular, such as the king, convey uniqueness, i.e., that there
is only one king. Definite descriptions in the plural, such as the children, convey maxi-
mality, i.e., that the predicate applies to all of the children—see (18) (Sharvy, 1980,
presents a theory that captures the pattern).

(18) The children slept until eight.

The uniqueness and maximality of descriptions seems to be at odds with actual usage,
however. In the case of singular descriptions, this is the phenomenon of “improper
descriptions” (e.g. Strawson, 1950). The child slept can be perfectly appropriate, even
knowing that there’s more than one child in the world. How to account for this is a
hotly debated topic (see Cappelen and Lepore, 2005, Stanley and Szabo, 2000, for some
recent debate).
The corresponding phenomenon for plural definite descriptions is that, even though
descriptions seem to have a maximal flavor, they aren’t perfectly equivalent to universal
claims about the members of the plurality (see, e.g., Brisson, 2003, Brogaard, 2007a,b,
Dowty, 1987, Scha, 1981). (18) can be true, even if some children woke up before
eight. For in the case of plural definite descriptions, we have two sources of quantifi-
cational force—the description and the distributive operator—and the apparent non-
maximality of the plural description may in fact be due to a restricted distributive opera-
tor (discussed in detail by Brisson, 2003). Thus (18) might just mean (19).

(19) The children are such that each one in a (contextually determined)
proper subset drawn from them slept until eight.

3 The Interpretation of “And”

Problems of interpretation also arise with a way of forming plural NPs: conjunction.

(20) a. Mary ran and Sue ran.

b. Mary and Sue ran.
c. Mary stretched and ran.

And is clearly univocal in these examples. They do not exhibit three words with distinct
meanings that just happen to sound the same. But this raises a problem. When and
connects sentences, it has a basically intersective meaning: in terms of possible worlds,
the set of worlds determined by a conjunction is the intersection of the set of worlds
determined by each of the conjuncts. Likewise, and has a basically intersective meaning
when it connects predicates. The set of objects that satisfies the conjunctive predicate
tall, dark, and handsome is the intersection of the sets that satisfy tall and dark and hand-
some, respectively. Extending this intersective idea to noun phrases works far less well.
Indeed, this fact about conjunction and collectivity is one of the earliest observed and
discussed facts in the philosophy of language. Thus Aristotle:

There are two styles of [sophistical] refutation; for some depend on the lan-
guage used, while some are independent of language. Those ways of producing

396
 PLU RAL S

the illusion which depend on language are six in number: they are homonymy,
ambiguity, combination, division, accent, form of expression. [. . .] Upon divi-
sion depend the proposition that 5 is 2 and 3, and even and odd, and that the
greater is equal (for it is that amount and more besides).
(Aristotle, 1984, 166a23–166a35)

That is, we don’t want to conclude from five is two and three that five is both even
and odd by illegitimately “dividing” the predicate into separate identity statements.
The problem of explaining the logical force of this example also exercised Medieval
logicians, for example Peter of Spain (1990, §75) and William of Ockham (1980,
ch. 37).
In contemporary semantic theorizing, finding a unified meaning for conjunction
that accounts for its basically intersective behavior in some contexts and its basically
union-forming behavior in others is of interest because it ramifies into issues about how
semantic composition is implemented in natural language. To say that a language (or
some other system of representation) is compositional is to say that the meaning of its
nonatomic parts is determined by the meaning of its atomic parts and their mode of
combination. Hence, a theory that wants to describe the compositional structure of, say,
English needs to say something about the semantic import of putting words together in
the way that we do. One very popular idea is the following, essentially following Frege
and articulated fully in Heim and Kratzer (1998). (Almost) all composition is a matter
of applying functions to arguments. This basic form of function application can be used
to explain the behavior of all manners of expression, including predicates of arbitrary
arity, quantifiers, and adverbs. But if function application is the only mode of produc-
ing meaningful complex expressions and the surface structure of these sentences cor-
responds closely to their logical form, it’s just about impossible to see how we could give
a single meaning for and.
One historically popular idea for solving this problem made use of a mechanism of
“conjunction reduction.” We know that the surface structure of sentences is the result
of transformations of an initial structure (as is shown by constraints on wh-movement,
for example). Given this fact, it’s at least possible that the different surface configura-
tions in which and can appear are the result of a single underlying structure. Specifically,
(20a) might be the basic case, and (20b) and (20c) are derived from it by “reducing
the conjunction.” And while this operation has syntactic and phonological effects, it
doesn’t have any semantic implications because the level of interpretation relevant to
interpretation—the logical form—is the shared deep structure. In that case, we can
make do with a single meaning for and: phrasal conjunction (see, e.g., Lakoff and Peters,
1969, Massey, 1976, McCawley, 1972, 1988, Smith, 1969).
Unfortunately, the conjunction reduction approach fails because it cannot account
for collectively interpreted noun phrases formed using conjunctions, such as Mary and
Sue met—it’s not equivalent to Mary met and Sue met. For this reason, it may be required
to make use of further principles of composition. The seminal paper in this area is Par-
tee and Rooth (2002), who argue that we should allow the syntax to generate syntactic
configurations that aren’t interpretable by the application of function application. If we
combine this with relatively tightly constrained rules about how the semantic system
should deal with such failures of interpretability—for example, by applying an alterna-
tive rule in such cases—we can give a unified analysis of all occurrences of conjunction.
Winter (2001) argues for a liberalized (and hence more powerful) version of this pro-

397
 BERN H A RD N I C K E L

posal (an alternative view of the mechanisms of composition that focuses on conjunc-
tion rather than function application is presented in Pietroski, 2005).

4 Collectivity and Collections

So far, I’ve assumed that, in some sense, plural NPs pick out a plurality of things (one or
more of them). I end this overview by addressing directly how best to understand this
pretheoretic talk of pluralities, the issue that has been at the center of philosophers’
attention. One way into the problem is to focus not on the interpretation of plurality,
but on a related issue in logic and metaphysics—whether we as theorists can quantify
over absolutely everything.
It is a commonplace that in ordinary speech, quantifiers are restricted. In the exam-
ples in (21), the quantifiers are not interpreted as ranging over every student that has
ever existed anywhere in the world.

(21) a. Every student passed.

b. Some students didn’t turn in their homework.

But for some purposes, it may be of interest to quantify over absolutely everything,
for example, in making ontological claims about what sorts of things there are not. A
metaphysician who says that there are no ghosts would not be happy if we took her to
mean that there aren’t any ghosts around here. She wants to assert something about the
whole universe.
However, on a certain way of understanding our natural language, we cannot pos-
sibly quantify over everything. This issue turns on how to interpret NPs in collectively
read sentences. There are basically three formal strategies. The first makes use only of
standard, singular first order logic (it usually isn’t emphasized that the logic is singular,
but this will be important in a minute). This approach makes use of sets or some other
kind of collectivizing object to capture the truth-conditions of the relevant sentences.
The second retains first-order logic but adds another kind of variable, plural variables.
The third makes use of the resources of second order logic. We can illustrate the three
strategies by considering (22).

(22) The rocks rained down on the village.

For present purposes, we can take the whole verb phrase rained down on the village as an
unanalyzed primitive, and I’ll pretend that the only rocks in the domain of discourse
are the ones that rained down on the village. The first strategy takes the expression the
rocks to denote an object that collects the relevant rocks—perhaps the set that contains
all and only those rocks, perhaps a mereological sum (Simons, 1987), or what have you.
The set-based representation of (22) might then be as in (23).

(23) s(x(Rock(x) l x  s)) š (Rained.Down(s))

This paraphrase has the benefit of retaining simple first order logic, a logical sys-
tem that is formally and intuitively incredibly well understood. But it also has its
drawbacks. A first point to make is that in a way, (23) gets the interpretation of the
predicate wrong. A single object cannot rain down on anything, it can only fall, and

398
 PLU RAL S

the situation doesn’t change even if that object has parts or elements. So in order to
take (23) as an interpretation of (22), we need to interpret the predicate as expressing
a derived notion of “raining down on the village,” call it rain*, so that rain* is true
of an object just in case it has elements or parts and these parts rained down on the
village.
Two further concerns have inclined many researchers in the area to reject the simple
first order paraphrase. At least prima facie, it looks as if it has counterintuitive conse-
quences about the ontological commitments we incur by speaking in a certain way.
Specifically, it looks as if we as speakers of English are, when uttering (22), ontologically
committed to the existence of sets in exactly the same way that we are committed to the
existence of sets when we say things like (24).

(24) There is a set containing just the number one.

But intuitively, there is a real difference in the ontological commitment of a speaker

who accepts only (22) and one who accepts (24). The first order logical paraphrase
doesn’t capture the difference since it analyzes both as quantifying over sets.
We can of course capture the difference by saying something about the theoretical
machinery we use in semantics. In general, not everything we make use of in a semantic
theory is something that the speaker of a natural language we interpret using that theory
is committed to. For example, if we interpret quantifiers as second order relations (as in
generalized quantifier theory), we don’t want to say that this shows that speakers who
use quantifiers in English are thereby committing themselves to the existence of sets.
The use of sets is simply an artifact of efficient theorizing. We can take a similar line
about the difference between (22) and (24): Take the apparent ontological commit-
ment to sets seriously when we interpret sentences explicitly about sets, treat it as an
artifact of efficient theorizing when we don’t. The concern is that we wanted to use the
logical paraphrase in part to lay bare when we’re talking about sets in a way that leads to
ontological commitments, and that isn’t possible on the present line.
Finally, and this has probably been the most influential line of argument, it looks as
if the first order paraphrase in terms of sets or collections can be shown to be straight-
forwardly mistaken in predicting that certain intuitively true sentences are incoherent.
This line of argument was first presented in Boolos (1999b,c), focusing on sentences
like (25).

(25) There are some things which aren’t members of themselves.

Intuitively, (25) is true—every human being serves as a witness to the existential claim.
But if we generally paraphrase plural noun phrases in terms of sets, (25) says that there
is a set, and it contains all and only the things (including sets) that aren’t members of
themselves. Unfortunately, that leads directly to Russell’s paradox.
A proponent of the singular first order paraphrase could block Russell’s paradox by
restricting the domain of quantification in a suitable way. Example (25) leads to para-
dox because at least in principle, anything could be something we refer to with a plural
locution (especially considering the data mentioned in §1 suggesting that the plural
morphology doesn’t mean more than one). This principle is an instance of a comprehen-
sion scheme, which we might put in natural language as (26).

399
 BERN H A RD N I C K E L

(26) Whenever there is one or more thing or things that have some property,
then there are the things that have that property.

Now focus specifically on the consequent of (26). If plural expressions are always inter-
preted as denoting (possibly singleton) sets, then (26) is equivalent to (27). To make
later comparisons easier, we also add the formal counterpart (28).

(27) Whenever there is one or more thing or things that have some property,
then there is a set that contains all and only the things that have that
property.

(28) w(I(w)) o ((s)(y)(Set(s) š (I(y) l y  s))).

Now, Russell’s paradox arises only if the quantifier in the consequent all and only the
things that have that property ranges over all of the sets. To see this, imagine that we’re
implicitly restricting the domain of that quantificational expression. The crudest way of
doing so is to exclude all sets from the domain of quantification in the object language.
In that case, the original troublesome existential claim (25) can be paraphrased in the
metalanguage as (29).

(29) There are some things, and they are all and only the things that aren’t sets
and aren’t members of themselves.

There’s obviously no commitment to the existence of a set that contains all and only
the things (including all of the sets) that aren’t members of themselves. Example (29)
is only committed to a set containing all and only the nonsets that are also nonself-
elemental. No problem there.
That we should resolve Russell’s paradox by appeal to such implicit restrictions has
been argued in detail by Dummett (1981, ch. 15) Dummett (1991) and Parsons (1983a,
b). For concerns about the coherence of the idea that we cannot quantify over every-
thing, see Lewis (1991) Boolos (1999a) and Williamson (2003), as well as many of the
essays collected in Rayo and Uzquiano (2006).
I’ve just presented the problem in terms of sets and a very crude restriction on
quantifiers. What is essential to the dialectic are just two parts. Plural NPs are inter-
preted in terms of an object that somehow collects the objects intuitively denoted
by the NP and there is some kind of membership relation that relates each of these
objects to this collective object—this is what Rayo (2002) calls the surrogate-strategy.
Paradox is avoided by some restriction of the quantifiers of the object language. In
other words, the surrogate strategy is incompatible with quantification over absolutely
everything. And the surrogate strategy is the only way to capture in a singular first
order language the truth-conditions of sentences containing plural NPs. Thus, if
natural language “needs” to be analyzed in terms of a singular first order language, then
it’s impossible to quantify in natural language over everything. I’ll say more about the
notion of “needing to be analyzed” shortly, once we have the other options on the
table.
Another way to block the derivation of Russell’s paradox is to alter the formal system
used to represent the semantics of the language we speak, and this comes in two flavors.
The first retains a first order logic, i.e., one that only quantifies over argument positions

400
 PLU RAL S

but allows plural variables in addition to singular ones. Thus, we allow the logical rep-
resentations in (30).

(30) a. xFx
There is an F.
b. xxFxx
There is one or more Fs.

The logical representation (30b) does not invoke any collectivizing entity. It allows sev-
eral objects to be simultaneous arguments of a predicate. The comprehension principle
(26) is formalized as in (31), where  should be read as is among.

(31) (w)(I(w)) o ((xx)(y)(I(y) l y  xx)).

When we plurally refer to one or more objects, the xxs in this case, we can only do so
when there is at least one object among the xxs. Intuitively, the reason for this restriction
is precisely the motivating thought for plural variables: We want to be able to refer to
various objects without making use of a collectivizing entity. But because there is no such
entity, it’s impossible to refer to any plurality of objects without referring to some objects.
In formal terms, (32), is analytically true, while (33) for example, is analytically false.

(32) (xx)(y)(y  xx).

For any things, there is some thing that’s among them.
(33) (xx)(y)(y  xx l y z y)
There are some things, namely, the nonself-identical ones.

Given such a plural first order language, Russell’s paradox is no longer derivable from
the simple sentence (25), rendered as (34).

(34) (xx)(y)(y  xx l y  y)

The xxs whose extension is defined by (34) aren’t themselves a set, so that (34) does not
entail that there is a set that contains all and only the non-self-elemental things.
Essentially the same effect can be achieved by altering the standard singular first
order logic in another way. Rather than extending it by introducing a special kind of
first-order variable—plural variables—we can move to a second-order logic. Such a
logic is characterized by allowing quantification over predicate positions. Formally, for
example, sentences such as (35) are well-formed, where j is an individual constant such
as a name.

(35) (X)X(j).

It’s hard to state the meaning of second-order logic informally without incurring
unwanted commitments. One might say, for example, that (35) means that there is a
property that John has. But that suggests that the second-order formalism is commit-
ted to the existence of properties we can quantify over. Perhaps the most famous worry
in this direction is Quine’s, who suggests that second-order logic is just set-theory in
“sheep’s clothing” [Quine, 1986]: (35) says neither more nor less than that there is a set
that John is a member of. Using his criterion of ontological commitment (Quine, 1980),

401
 BERN H A RD N I C K E L

it follows that second order logic is ontologically committed to sets. Moreover, this
substantive ontological commitment disqualifies second order logic from being logic
at all, rather than a substantive theory about the world, which in turn raises concerns
about the viability of logicism, the project of showing that arithmetic can be reduced
to logic.
Why exactly Quine’s criterion is supposed to have this consequence is less than
completely clear. If it turns out that an existential quantifier that binds predicate posi-
tions quantifies over sets, and ordinary (unquantified) sentences with predicates can
be the true instantiations of such existentially quantified sentences, then the ordinary
sentences should be committed to sets—after all, the predicate must pick out one of
the things quantified over. For further discussion, see Rayo and Yablo (2001), Wright
(1983).
Indeed, this discussion just recapitulates Frege’s concept horse problem (Frege, 1997).
Quantificational idioms in natural language tend to be object oriented, so that even
when we try to explain quantification over non-object positions, we end up introducing
objects that play the role of non-objects, such as sets, classes, or properties. The fact that
we run into the concept horse problem in trying to elucidate the semantics of second
order logic has made the following observation, due to Boolos, particularly important.
If we consider a specific family of second order logical theories, those that only allow
quantification over one-place predicate positions, we are restricting ourselves to the
so-called monadic second order logics. And Boolos showed that we can translate mon-
adic second-order logic completely mechanically into first order plural logic. The key
idea is to think of the one-place predicates quantified over in the second order logic as
denoting all of the things that fall under them, not by denoting a set of these things, but
simply denoting them plurally. Thus, we simply rephrase what looks like predication
in (36a) as a partitive construction in (36b) (see Higginbotham, 2000, Hossack, 2000,
Linnebo, 2004, for further discussion).

(36) a. John runs.

b. John is one of the runners.

Nonetheless, we want to keep second order logic separate from plural first order logic.
As Rayo and Yablo (2001) and Williamson (2003) point out, the positions quantified
over are formally distinct, one being predicative, the other objectual.
Now that the options are on the table, we can return to the issue of what it means to
say that natural language “needs to be analyzed” using one or another formal tool, in this
case singular first order logic, plural first order logic, or second order logic. Formal lan-
guages are artificial objects. We can use them to represent logical properties of sentences
of a natural language, such as entailment, equivalence, or incompatibility because these
logical relations are precisely defined in the formal language. But we also need to make
sure that we understand the formal language, not by learning to speak it but by having
an interpretation of the language. Such an interpretation must itself be couched in some
meta-language which is primitively understood (i.e., not by being interpreted in some
further meta-language). This is particularly important for the basic building blocks of
the logic, its primitive vocabulary and its logical operators.
A good example concerns modality: we can have a substantive debate about whether
talk about possibility or necessity is clear enough for certain philosophical purposes to
simply be accepted, or whether it itself needs to be explained in terms of some other

402
 PLU RAL S

notion. Goodman (1983) and Quine (1966) are famous skeptics about the intelligibil-
ity of our modal talk and want to reconstruct (some aspects of) that talk in other terms.
In Quine (1973, 1982) are classic expressions of similar skepticism regarding second
order logic. First order logic with quantification is the least controversial such logic,
and the one whose basic building blocks of terms, functional expressions, relations,
sentential connectives, and quantifiers have the easiest correspondence in natural
language.
But as we’ve seen, first order logic also has its limitations. If it captures the semantics
of the language we primitively understand—aka natural language—we cannot quantify
over absolutely everything. So there is some incentive to conduct one’s metaphysical
discussions in a logic that goes beyond the strictures of singular first order logic. And the
best way to show that we understand the formal tools employed in such a discussion is
to show that we use basically these very same expressive resources in natural language.
That is to say, if our best semantics for natural language is one that eschews the surro-
gate strategy in favor of plural first order or monadic second order logic, then the use of
these formalisms is much less problematic. In a nice turn of phrase, Williamson suggests
that the issue is what language we use as our home language, the language in which we
are happy to work, at least for the time being, without seeing it through the lens of a
meta-language (Williamson, 2003, 459).

4.1 Cumulativity and Events

Let’s look at some data that bear on this issue. The simplest example of cumulativity is
(9).

(9) Three children ate four pizzas.

The crucial observation about cumulativity is that the two quantifiers three children and
four pizzas are scopeless with respect to each other. That is to say, neither of the follow-
ing two scope relations captures the cumulative reading of (9).

(37) a. For each of three children x, there are four pizzas that x ate.
b. For each of four pizzas x, there are three children that ate x.

Both (37a) and (37b) are too weak, since they’re true in the situation in which the
cumulative reading is true, but they’re also compatible with situations in which the
cumulative reading is false. Example (37a) is compatible with a total of twelve pizzas
being eaten by the three children, (37b) with a total of twelve children doing the eating
of four pizzas. One option, proposed by Scha (1981) in a paper that shaped the debate
on these issues, is that there is a special dyadic quantifier. That is, at the level of logical
form, the sentence has a single element three-four that applies to two NPs and a rela-
tional predicate to yield the cumulative reading. If that theory is correct, we would see
that natural language furnishes us with very powerful quantificational resources (formal
details of binary quantifiers can be found in Keenan, 1992, van Benthem, 1989).
An alternative to the dyadic quantifier approach involves embracing a Davidsonian
event-analysis for the semantics of English. The core idea of this analysis is that predi-
cates introduce event variables into the logical forms of sentences in which they appear,
and that all the other parts of the sentence are analyzed as predicating something of the

403
 BERN H A RD N I C K E L

event thus introduced (see Davidson, 1980, Parsons, 1990). On this strategy, (38a) is
analyzed as (38b).

(38) a. John buttered the toast.

b. e(Buttering(e) š Agent(e, j) š Theme(e, t))

We can use such an event-based analysis to capture cumulative readings, as well. Infor-
mally, we analyze (9), three children ate four pizzas as (39).

(39) There were some eating events in which three children did the eating
and four pizzas were consumed.

One very powerful reason to prefer such an event-based analysis of cumulativity to the
polyadic quantificational analysis inspired by Scha comes from ditransitive verbs such
as give. Consider (40).

(40) Three ATMs gave two customers two passwords (each).

The reading of interest is one in which we take two customers to not be distributive
with respect to the three ATMs, i.e., we take there to be a total of only two custom-
ers, but where each of the customers got two passwords. This is probably the favored
reading, since we can’t take two passwords to be cumulative, as well—that would mean
that somehow, two ATMs must have collaborated in the giving of at least one of the
passwords.
Schein (1993) argues in great detail that this kind of mixed cumulative/distribu-
tive quantification cannot be captured with any analysis other than an event analysis
along the lines of (39). On the polyadic approach, for example, we cannot scope two
passwords “inside” the three-two complex that takes ATMs and customers as arguments,
since there is no such spot to scope to. As such, these readings strongly speak in favor
of a Davidsonian event-based analysis, though McKay (2006) has tried for an analysis
that does without them.

4.2 The Unique Role Requirement

Landmann (2000), following Parsons (1990), argues that it is part of the semantics of
such an event-analysis that events have a unique-role requirement. For each event,
there can be only one thing that fills its agent role, one thing that fills its patient role,
and so on. The kind of argument at issue concerns, for example, the analysis of (41a)
as (41b).

(41) a. John and Mary met.

b. e(Meeting(e) š Agent(e, j) š Agent(e, m))
c. e(Meeting(e) š Agent(e, j))
d. John met.

(41b) isn’t viable as an analysis, since (41b) entails (41c), which we would think
is the logical form of (41d). But clearly, (41d) isn’t entailed by (41a), it isn’t even
grammatical.

404
 PLU RAL S

The unique role requirement provides an argument for what I earlier called the sur-
rogate strategy for representing collectivity. If the unique role requirement is correct, so
is the surrogate strategy, since it is the only way to make it so that a single thing fills the
agent role of collective predicates such as meet, surround, or rain down.
A natural response (see Pietroski, 2005, Schein, 1993) is to generalize the Davidso-
nian analysis to bind not a singular event variable, but a plural event variable. Using the
example of cumulative readings to illustrate the idea, we can paraphrase (40) as (42).

(42) There were some events of giving something out, three ATMs did the giv-
ing, two customers were their recipients, and two passwords were given.

That is to say, all we want or can say about the ATM situation is that there were some
events with the ATMs as agents that crucially do not allow us to draw any inferences
regarding each of the individual events. By parallel reasoning, we analyze John and Mary
met as (43).

(43) There were some events, they were a meeting, and John and Mary were
their agents.

And though the events all together constitute a meeting, we cannot infer anything
about whether the individual events that make up the meeting are themselves
meetings.
This is not the end of the dialectic. Landmann cites examples such as (44a).

(44) a. The children in the choir sang a requiem.

b. There were some events, they were a singing, the children were their
agents, and their theme was a requiem.

(44a) can be true if some kids were singing, some kids were just pretending, and some
hid at the back of the choir and thought about comic books. And he suggests that at
this point, it’s not clear in what sense all of these children were agents of singing events,
except insofar as they were part of a collection that, as a whole, engaged in some sing-
ing—Schein (2005, n. 62) contains a response.

5 Conclusion
Plurals provide challenges and resources for semantic theorizing. The challenge is to
give a semantics that can account for distinctively plural phenomena, such as collectiv-
ity and cumulativity, while also interacting with other semantic features of a sentence,
such as definite descriptions and conjunction to predict the full range of available read-
ings compositionally. But they also allow us to motivate powerful semantic resources,
such as plural quantification and additional sources of quantificational force, such as
distributive quantifiers that are introduced by the verb phrase. Because plurality is such
a systematic phenomenon, we can parlay empirically well-supported theorizing, such as
debates about the event-based analysis of predicates, into fundamental issues about the
correct logic for natural language and thence the proper logic with which to conduct our
metaphysical and logical inquiry.

405
 BERN H A RD N I C K E L

Related Topics
1.3 Logical Form
1.5 Implicature
1.6 Pragmatic Enrichment
2.6 Possible Worlds Semantics
2.7 Dynamic Semantics
3.4 Quantifiers and Determiners
3.5 Generics
3.7 Descriptions
3.10 Mass Terms.

References
Aristotle. Sophistical refutations. In Jonathan Barnes, editor, The Complete Works of Aristotle, pages 278–
314. Princeton UP, Princeton, NJ, 1984.
Sigrid Beck and Uli Sauerland. Cumulation is needed: A reply to Winter (2000). Natural Language Seman-
tics, 8(4): 349–371, 2000.
George Boolos. Reply to Charles Parsons’ “Sets and Classes”. In Richard Jeffrey, editor, Logic, Logic, and
Logic, pages 30–6. Harvard UP, Cambridge, MA, 1999a.
George Boolos. Nominalist platonism. In Richard Jeffrey, editor, Logic, Logic, and Logic, pages 73–87.
Harvard UP, Cambridge, MA, 1999b.
George Boolos. To be is to be a value of a variable (or to be some values of some variables). In Richard Jeffrey,
editor, Logic, Logic, and Logic, pages 54–72. Harvard UP, Cambridge, MA, 1999c.
Christine Brisson. Plurals, all, and the nonuniformity of collective predication. Linguistics and Philosophy,
26(2): 129–184, 2003.
Berit Brogaard. Sharvy’s theory of definite descriptions revisited. Pacific Philosophical Quarterly, 88: 160–180,
2007a.
Berit Brogaard. The but not all: A partitive account of plural definite descriptions. Mind and Language,
22(4): 402–426, 2007b.
Herman Cappelen and Ernie Lepore. Insensitive Semantics. Blackwell Publishing, Malden, MA, 2005.
Gennaro Chierchia. Scalar implicature, polarity phenomena, and the syntax/pragmatics interface. In A.
Belletti, editor, Structures and Beyond, pages 39–103. Oxford UP, Oxford, 2002.
Gennaro Chierchia. Broaden your views: Implicatures of domain widening and the “logicality” of language.
Linguistic Inquiry, 37(4): 535–590, 2006.
Noam Chomsky. Questions of form and interpretation. In Robert Austerlitz, editor, The Scope of American
Linguistics, pages 159–196. Peter de Ridder Press, Lisse, 1975.
Donald Davidson. The logical form of action sentences. In Essays on Actions and Events, pages 105–148.
Oxford UP, Oxford, 1980.
Sjaak de Mey. The dependent plural and the analysis of tense. In Victoria A. Burke and James Pustejowski,
editors, Proceedings of NELS 11. GLSA, University of Massachusetts, Amherst, MA, 1981.
David Dowty. Collective predicates, distributive predicates, and all. In F. Marshall, editor, Proceedings of the
Third ESCOL. Ohio State University, Columbus, OH, 1987.
Michael Dummett. Frege: Philosophy of Language (Second Edition). Harvard UP, Cambridge, MA, 1981.
Michael Dummett. Frege: Philosophy of Mathematics. Harvard UP, Cambridge, MA, 1991.
Gottlob Frege. On concept and object. In Michael Beaney, editor, The Frege Reader, pages 181–93. Black-
well, Oxford, 1997.
Brendan Gillon. The readings of plural noun phrases. Linguistics and Philosophy, 10(2): 199–220, 1987.
Brendan Gillon. Plural noun phrases and their readings: A reply to Lasersohn. Linguistics and Philosophy,
13(4): 477–485, 1990.
Brendan Gillon. Towards a common semantics for English count and mass nouns. Linguistics and Philosophy,
15(6): 597–640, 1992.
Nelson Goodman. Fact, Fiction, and Forecast. Harvard UP, Cambridge, MA, 1983.
Robert M. Harnish. Logical form and implicature. In Steven Davis, editor, Pragmatics: A Reader, pages
316–364. Oxford UP, New York, 1991.

406
 PLU RAL S

Irene Heim and Angelika Kratzer. Semantics in Generative Grammar. Blackwell, Malden, MA, 1998.
James P. Higginbotham. On second-order logic and natural language. In Gila Sher and Richard Tieszen,
editors, Between Logic and Intuition: Essays in Honor of Charles Parsons, pages 79–99. Cambridge UP,
Cambridge, UK, 2000.
James P. Higginbotham. Reciprocal interpretation. Linguistic Research, 3: 97–117, 1981.
Keith Hossack. Plurals and complexes. British Journal for the Philosophy of Science, 51: 411–43, 2000.
Hans Kamp and Uwe Reyle. From Discourse to Logic. Kluwer, Dordrecht, 1993.
Jerrold J. Katz. Propositional Structure and Illocutionary Force. Harvard UP, Cambridge, MA, 1977.
Edward Keenan. Beyond the Frege boundary. Linguistics and Philosophy, 15 (2): 199–221, 1992.
Manfred Krifka. Bare nps: Kind-referring, indefinites, both, or neither? In R. B. Young and Y. Zhou, editors,
Proceedings of SALT 13. CLC Publications, Ithaca, NY, 2004.
George Lakoff. Linguistics and natural logic. In Donald Davidson and Gilbert Harman, editors, Semantics of
Natural Language. D. Reidel, Dordrecht, 1972.
George Lakoff and Stanley Peters. Phrasal conjunction and symmetric predicates. In D. Reibel and S.
Schane, editors, Modern Studies in English: Readings in Transformational Grammar. Prentice Hall, Engle-
wood Cliffs, NJ, 1969.
Fred Landmann. Events and Plurality. Kluwer Academic, Dordrecht, 2000. Peter Lasersohn. Plurality, Con-
junction, and Events. Kluwer, Dordrecht, 1995. David K. Lewis. Parts of Classes. Blackwell, Oxford, 1991.
Øystein Linnebo. Plural quantification. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy.
Winter 2004. URL http://plato.stanford.edu/archives/win2004/entries/plural-quant/
Jan Tore Lønning. Plurals and collectivity. In Johan van Bentham and Alice ter Meulen, editors, Handbook
of Logic and Language, pages 1009–1053. Elsevier and MIT Press, Amsterdam, 1997.
Gerald J. Massey. Tom, dick, and harry, and all the king’s men. American Philosophical Quarterly, 13(2):
89–107, 1976.
James D. McCawley. A program for logic. In Donald Davidson and Gilbert Harman, editors, Semantics of
Natural Language. D. Reidel, Dordrecht, 1972.
James D. McCawley. The Syntactic Phenomena of English, Vol. I. University of Chicago Press, Chicago,
1988.
Thomas J. McKay. Plural Predication. Oxford UP, Oxford, 2006.
Charles Parsons. The liar paradox. In Mathematics in Philosophy, pages 221–67. Cornell UP, Ithaca, NY,
1983a.
Charles Parsons. Sets and classes. In Mathematics in Philosophy, pages 209–20. Cornell UP, Ithaca, NY,
1983b.
Terence Parsons. Events in the Semantics of English: A Study in Subatomic Semantics. MIT Press, Cambridge,
MA, 1990.
Barbara Partee and Mats Rooth. Generalized conjunction and type amiguity. In Paul Portner and Barbara H.
Partee, editors, Formal Semantics: The Essential Readings, pages 334–356. Blackwell, Oxford, 2002.
Peter of Spain. Summulae logicales. J. Benjamins Co, Amsterdam, 1990. Translated by Frances P. Dinneen.
Paul M. Pietroski. Events and Semantic Architecture. Oxford UP, Oxford, 2005.
W.V.O. Quine. Three grades of modal involvement. In The Ways of Paradox and Other Essays, pages 158–
176. Harvard UP, Cambridge, MA, 1966.
W.V.O. Quine. The Roots of Reference. Open Court, La Salle, IL, 1973.
W.V.O. Quine. On what there is. In From a Logical Point of View, Second Edition, pages 1–19. Harvard UP,
Cambridge, MA, 1980.
W.V.O. Quine. Methods of Logic, Fourth Edition. Harvard UP, Cambridge, MA, 1982.
W.V.O. Quine. Philosophy of Logic, Second Edition. Harvard UP, Cambridge, MA, 1986.
Agustin Rayo. Word and objects. Noûs, 36(3): 436–464, 2002.
Agustin Rayo. Plurals. Philosophy Compass, 2(3): 411–427, 2007.
Agustin Rayo and Gabriel Uzquiano. Absolute Generality. Oxford UP, Oxford, 2006.
Agustin Rayo and Stephen Yablo. Nominalism through denominalization. Noûs, 35(1): 74–92, 2001.
Uli Sauerland, Jan Anderssen, and Kazuko Yatsushiro. The plural is semantically unmarked. In S. Kesper and
M. Reis, editors, Linguistic Evidence. Mouton de Gruyter, Berlin, 2005.
Remko Scha. Distributive, collective, and cumulative quantification. In J. Groenendijk, T. Janssen, and M.
Stokhof, editors, Truth, Interpretation, and Information (GRASS 2). Foris, Dordrecht, 1981.
Barry Schein. Plurals and Events. MIT Press, Cambridge, MA, 1993.
Barry Schein. Plurals (unabridged). ms., USC, 2005. URL http://www-rcf.usc.edu/~schein/Plurals2005.pdf
Barry Schein. Plurals. In Ernest Lepore and Barry C. Smith, editors, The Oxford Handbook of Philosophy of

407
 BERN H A RD N I C K E L

Language, pages 716–767. Oxford UP, Oxford, 2006.

Roger Schwarzschild. Plurals, presuppositions, and the sources of distributivity. Natural Language Semantics,
2: 201–248, 1994.
Roger Schwarzschild. Pluralities. Kluwer, Dordrecht, 1996.
Richard Sharvy. A more general theory of definite descriptions. The Philosophical Review, 89(4): 607–624,
1980.
Peter Simons. Parts. A Study in Ontology. Oxford UP, Oxford, 1987.
Carlotta Smith. Ambiguous sentences with and. In D. Reibel and S. Schane, editors, Modern Studies in Eng-
lish: Readings in Transformational Grammar. Prentice Hall, Englewood Cliffs, NJ, 1969.
Benjamin Spector. Aspects of the pragmatics of plural morphology: On higher-order implicatures. In Uli
Sauerland and Penka Stateva, editors, Presupposition and Implicature in Compositional Semantics, pages
243–281. Palgrave Macmillan, Houndsmills, UK, 2007.
Jason C. Stanley and Zoltan Gendler Szabo. On quantifier domain restriction. Mind and Language,
15(2–3): 219–261, 2000.
Peter F. Strawson. On referring. Mind, 59(235): 320–344, 1950.
Johan van Benthem. Polyadic quantifiers. Linguistics and Philosophy, 12(4): 437–464, 1989.
William of Ockham. Summa Logicae, Part 2. Notre Dame UP, Notre Dame, 1980. Translated by A. J.
Freddoso and H. Schuurman.
Timothy Williamson. Everything. Philosophical Perspectives, 17(1): 415–465, 2003.
Yoad Winter. Distributivity and dependency. Natural Language Semantics, 8 (4): 27–69, 2000.
Yoad Winter. Flexibility Principles in Boolean Semantics. MIT Press, Cambridge, MA, 2001.
Crispin Wright. Frege’s Conception of Numbers as Objects. Aberdeen UP, Aberdeen, 1983.
Eytan Zweig. Dependent Plurals and Plural Meaning. PhD thesis, New York University, 2008.

408
 3.9
ADV ER BS
Delia Graff Fara

Adverbs are a motley crew. They include words or phrases that modify (a) verbs, (b)
adjectives, or (c) other adverbs:

(1) a. He fainted surreptitiously;

b. Their friendship was faintly surreptitious;
c. He fell to the ground shockingly quickly.

According to one contemporary grammarian, adverbs include any word or phrase that
answers one of the following questions: “In what way?”, “When?”, “Where?”, “Why?”,
“How?”, “How much?”, “Under what condition?”, “To what degree?”, or “How often?”
(http://www.esldesk.com/grammar/adverbs (January 2011)). These diagnostic questions
qualify the following words and phrases as adverbs:

(2) a. He wept ruefully;

b. Yesterday, he wept ruefully;
c. Yesterday, he wept ruefully in his bed;
d. Yesterday, he wept ruefully in his bed because she lied;
e. She managed to juggle by using sticky tape;
f. She managed to juggle by using a lot of sticky tape;
g. She managed to juggle while drunk by using a lot of sticky tape;
h. She managed to juggle while very drunk by using a lot of sticky tape;
i. She often managed to juggle while very drunk by using a lot of sticky
tape.

(We might be skeptical about counting ‘a lot’ as an adverb. But otherwise the diagnostic
questions seem to give no false positives.)
Words like ‘therefore’, which are used to relate independent clauses or states of
affairs, are traditionally considered to be “conjunctive” adverbs. Here are a few more
examples.

(3) a. She thought that the coffee tasted foul; accordingly, she spit it out.
b. She had to spit out the coffee; otherwise she would have retched.
c. Between the coffee’s tasting foul and her retching after drinking it, the
day got off to a very bad start.

Once we admit conjunctive adverbs to the motley crew just discussed, we can add to
 DELI A G RA F F FA R A

our initial list of diagnostic questions the following, more cumbersome one: “What’s the
connection between these state of affairs?”
Adverbs can also be used to express an attitude, on the part of a speaker, toward states
of affairs expressed in a complete sentence:

(4) a. Fortunately, Ian didn’t fall on his head;

b. Hopefully, he’ll go see a doctor.

Notice that the sentence adverbs in (4) would not normally be understood as falling
into any of the categories of adverb described so far. For example, neither would
normally be understood as modifying a verb. First, ‘fortunately’ in (4a) would not
normally be understood as modifying ‘falling’. For if I were to claim that Ian didn’t
fall fortunately on his head, I would be saying something compatible with his having
fallen on his head in some other way; for example, accidentally or clumsily. But (4a)
is false if Ian fell on his head at all—no matter how he did it. Second, ‘hopefully’ in
(4b) would not normally be interpreted as modifying either ‘go’ or ‘see’. If I were to
claim that he will go hopefully to see a doctor or that he will go to hopefully see a
doctor, I would be making a claim about the future. I would be committing myself to
his going to see the doctor at some future point in time. What I would be saying, in
either of those cases, would be false if he were never to go see the doctor. But if (4b)
is truth evaluable at all, then its truth does not depend on whether he does in fact go
see the doctor in the future, but only on whether I hope (now) that he will go see the
doctor in the future.
“Sentence adverbs” need not occur at the beginning of a sentence. The following are
easily interpreted as equivalent to the sentences in (4):

(5) a. Ian fortunately didn’t fall on his head;

b. He’ll hopefully go see a doctor.

The following sentences show, however, that although the positions of the “evaluative
adverb” ‘luckily’ and the “epistemic adverb” ‘probably’ are not fixed within a sentence
(6–7), their ordering relative to each other is fixed (8).

(6) a. Luckily, he’s going to win.

b. He’s luckily going to win.
(7) a. Probably, he’s going to win.
b. He’s probably going to win.
(8) a. Luckily, he’s probably going to win.
b. # Probably, he’s luckily going to win.

The syntactician Guglielmo Cinque has argued that there is a hierarchy of finely indi-
viduated categories of adverbs which constrains their ordering relative to one another,
and that this holds across many languages (Cinque 1999). Cinque used such facts in
turn to argue that each type of adverb had its own functional projection within a sen-
tence, where this entails that each adverb contributes a binary-branching tree having
at least three terminal nodes to the phrase structure of the sentence in which it appears.
The important implication of this is that it would require syntactic structures to be sig-
nificantly more complex than had been previously thought.

410
 A DVER B S

Let us turn to the question of how adverbial modifiers are to be represented logically.
Philosophers have been concerned to capture entailments concerning adverbial modi-
fiers as logical entailments. Return to some of our earlier sentences. Each sentence in (9)
entails any below it.

(9) a. Alexander wept ruefully in his bed;

b. Alexander wept ruefully;
c. Alexander wept.

As an aside, note that the following sentences, in contrast to those above, will not in all
contexts entail their successors:

(10) a. Alexander wept ruefully in his bed because she lied;

b. Alexander wept ruefully because she lied;
c. Alexander wept because she lied.

These sentences do not represent entailments since her having lied may not have been
the cause of his weeping; her having lied may have been the cause of the weeping’s hav-
ing taken place in his bed as opposed, say, to her bed (in which case (9a) could be true
while false). Or her having lied may have been the cause of the weeping’s being rueful as
opposed, say, to remorseful (in which case (9a) (9b)or (9c) were true while (9c) or (9b)
were false). (See David Beaver and Brady Clark (2008) for a brief discussion of this feature
of ‘because’—in particular as it relates to focus—which compares the effect of focus on
‘because’-clauses with the effect of focus on counterfactuals. The latter is discussed by, for
example, Fred Dretske (1972), Kai von Fintel (1994), and Mats Rooth (1999).)
Leaving that issue aside, though, we now question how to logically represent the
sentences in (9) so as to reflect the entailment relations that hold among them. In
first-order logic, we would paraphrase ‘Alexander wept’ as an atomic formula, Wa, con-
taining a monadic predicate and a name, where W stands for ‘wept’ and a stands for
‘Alexander’. It looks, at face value, like adverbial modification can be represented in
the canonical language for first-order logic in either of two ways: (i) by introducing
new predicates with the same number of argument places as the unmodified original, to
stand in for the modified predicate; or (ii) by adding extra argument places to accom-
modate the added adverbial modifiers.
To pursue the first strategy, we would paraphrase ‘Alexander wept ruefully in his bed’
by admitting different monadic predicates for the different ways of modifying our verb
‘wept’.

(i) a. Wx: ‘x wept’;

b. Wcx: ‘x wept ruefully’;
c. Wccx: ‘x wept ruefully in x’s bed’;

‘Alexander wept ruefully’ would be paraphrased as Wca.

Problem: This approach leaves the entailment relations between ‘x wept’, ‘x wept rue-
fully’, and ‘x wept ruefully in x’s bed’ unaccounted for. (For discussion of this issue, see
Richmond Thomason and Robert Stalnaker (1973).) The different monadic predicates
W, Wc, et cetera, bear no logical relations to one another: No atomic monadic predica-
tion logically entails any other.

411
 DELI A G RA F F FA R A

To pursue the second strategy, we would paraphrase ‘Alexander wept ruefully’ by

admitting a two-place predicate W2xy:

(ii) a. Wx: ‘x wept’;

b. W2xy: ‘x wept in way y’

‘Alexander wept ruefully’ would now be paraphrased as W 2ar, where r is treated as stand-
ing for the rueful way.
But this method of accommodating adverbial modification does not yield the desired
logical entailment from ‘Alexander wept ruefully’ (W 2ar) to ‘Alexander wept’ (Wa): no
atomic dyadic predication logically entails any atomic monadic one.
To fix this problem, while only slightly varying the theme, one could propose that
instead of having the two different predicates for paraphrasing ‘wept’—a monadic one,
and a dyadic one—we could have just the dyadic one and represent the superficially
monadic English predication ‘x wept’ as an existential quantification: ‘x wept in some
way or other’ (yWxy). This latter sentence would count as the logical paraphrase of
both ‘Alexander wept’ and ‘Alexander wept in some way’. The entailment of ‘Alexander
wept’ by ‘Alexander wept ruefully’ would then be a special case of existential generaliza-
tion, since yW 2ay is entailed by W 2ar.

(iic) a. yW2ay: ‘Alexander wept’;

b. W2ar: ‘Alexander wept ruefully’;

Problem: indefinitely many modifiers may be added to the monadic predicate ‘wept’:

(icc) a. W3arb: ‘Alexander wept ruefully in his bed’;

b. W4arbt: ‘Alexander wept ruefully in his bed at two o’clock’;
c. W5arbtd: ‘Alexander wept ruefully in his bed at two o’clock while
dressing’.

There is no limit to the number of argument places that a single ‘wept’ predicate needs
in order to capture, as cases of existential generalization, the entailment relations that
result from adverbial modification. (See Donald Davidson 1967, 106–108.)
So the add-a-predicate approach obliterates the syntactic and semantic complexity of
adverbial modification. And while the second version of the add-a-place approach does
represent adverbial modification as yielding a difference in both syntactic and semantic
complexity, it doesn’t seem to adequately capture the hierarchical structures of those com-
plexities. On the syntactic side, the sentence ‘Alexander wept ruefully’ gets paraphrased
as W2ar. But then the superficially simpler ‘Alexander wept’ gets paraphrased as under-
lyingly more complex: yW2ay. On the semantic side, when we say that Alexander wept
ruefully, we don’t think of the name and the adverb here as having the same syntactic or
semantic status. When we say that Alexander wept ruefully, we don’t mean that a certain
relation holds, now represented as W2, between Alexander and the rueful way. Rather, we
mean that the property of being rueful applies to the property of weeping to yield a property
in turn—namely, the property of weeping ruefully—that applies to Alexander.
There is a third way of giving logical expression to adverbial modification that
both keeps within the bounds of a standard first-order language and yields the entail-
ments under discussion as logical entailments. The approach is associated with Donald

412
 A DVER B S

Davidson (1967, 1985, e.g.) and Terence Parsons (1990, 1995, e.g.). They propose
that all action sentences are to be logically represented as existential quantifica-
tions over events, with various properties being attributed to those events. (For more
detailed presentation of event-based semantics, see Barry Schein “Event Semantics,”
(Chapter 2.8, this volume).)
For example, the past-tense sentence ‘Alexander wept’ is on this view to be under-
stood as having a more complex logical form than appears on the surface. Parsons
understands it as saying that there is an event that has Alexander as its agent, that is a
weeping event, and that has “culminated” (since it is in the past perfect):

(11) a. Alexander wept;

b. e(Agent(a, e) šWeeping(e) šCulminated(e)).

When we further modify ‘wept’ by describing the manner of the weeping, this is con-
strued as an additional property of the event: This event was not only a weeping but
also a rueful event:

(12) a. Alexander wept ruefully;

b. e(Agent(a, e) šWeeping(e) šCul(e) šRueful(e)).

Further modifications are likewise represented as attributions of properties to an event:

(13 a. Yesterday, Alexander wept ruefully in his bed;

b. e(Yesterday(e) šAgent(a, e) šWeeping(e) š Cul(e) šRueful(e)
šLocated(e, a’s bed).

Given these “logical forms” (as Davidson and Parsons call them), the entailment of (11)
by (12), and of (12) in turn by (13), becomes a straightforward logical entailment: an
instance of conjunction elimination within the scope of an existential quantifier. Many
points deserve mention. I will single out a few.
One might question whether the logical forms in (11–13) contain the correct predi-
cations of the weeping event in question. For example, one might question whether
Alexander was really an agent of the weeping. To say that he was the agent is to say that
the weeping was among his actions—that weeping was something he did. So shouldn’t
we say that he was not the agent of the weeping (unless he did it on purpose) but rather
the experiencer of the weeping? Questions such as these don’t constitute a criticism of
the theory itself, just an internal criticism of one particular way of applying the theory
to one particular sentence.
A similar question arises about the appropriateness of a certain other attribution to
the event which does constitute an objection to the very heart of the event analysis.
One might question whether the thing that was rueful was the event itself, as the analy-
sis requires. The following illustrate some appropriate ways of attributing properties to
events. Events can last for a long or short time. Events can involve knives or icepicks.
But are they the kind of things that can themselves be rueful? Being a weeping is a
property of the event, just as taking a long time is. But being rueful is not like that. It is
not a property of the event. To say that an event itself was rueful is to make a category
mistake. Doesn’t it seem rather that being rueful is a property of weeping rather than a
property of the event itself?

413
 DELI A G RA F F FA R A

The event analysis incurs some ontological commitments. First, and most obviously,
it requires that events exist—that they are in the domain of our quantifiers. Jonathan
Bennett, in true purist style, points out that the theory itself is not committed to
the existence of events, only to speakers often committing themselves to the existence of
events. Given the plausible principle that speakers in the relevant cases often speak truly,
the stronger commitment follows (Bennett 1988, p. 173). Second, it requires (given
plausible principles) that single events may have many of their properties contingently
and may correctly be described in a number of ways. Brutus’s stabbing of Caesar violently,
with a knife is the very same event as some stabbing of Caesar by Brutus. Why? To be
a stabbing of Caesar by Brutus is to be an event that is a stabbing that has Brutus as its
agent and Caesar as its object. To be a violent stabbing of Caesar by Brutus, with a knife,
is to be a stabbing that has Caesar as its object, Brutus as its agent, a knife its instrument,
and violence as a manner in which it was done. Conjunction elimination yields that the
violent stabbing of Caesar, with a knife, by Brutus, is itself a stabbing of Caesar. Since
there was only one stabbing of Caesar, these must be the very same event.
But why think, it might be countered, that there was only one stabbing of Caesar?
After all, there was the stabbing, the violent stabbing, the violent stabbing with a knife
. . . . The thought continues: These are not the same events because Brutus might have
stabbed Caesar gingerly, or he might have stabbed him with an ice pick. But the violent
stabbing couldn’t have been done gingerly, and the stabbing with a knife couldn’t have
been done with an ice pick.
As one would expect, the metaphysical problems of how to individuate events mani-
fest themselves as linguistic problems about the semantics of certain expressions that
express properties of events—which includes adverbs on the views of Davidson and Par-
sons. One concern of Davidson’s is the adverb ‘slowly’; he thought it had to be excluded
from his analysis. His worry was that there could be circumstances in which it would be
true to say that Daisy crossed the Channel slowly and others in which it would be false
to say, speaking of the very same event, that Daisy crossed the Channel slowly. Upon
learning that it took Daisy ten hours to cross, one could truly say that she crossed slowly.
(It normally takes just three hours to cross.) Application of the event analysis to this
case would yield the following formalization:

(14) e(Crossing(e) šCul(e) šAgent(d, e) šObject(c, e) šSlow(e))

But if upon subsequently learning that Daisy crossed by swimming, one said that Daisy
crossed quickly, one would again speak truly. This would express a truth:

(15) e(Crossing(e) šCul(e) šAgent(d, e) šObject(c, e) šQuick(e))

But since there was only one crossing event and since a single event cannot be both slow
and quick, Davidson concluded that that ‘slowly’ and ‘quickly’ don’t really express proper-
ties of events at all—since whether they could be truly predicated of an action depended
on how that action was described or conceived—and that his analysis would therefore
not apply to them. Many other adverbs would presumably fall into this class: for example,
‘gracefully’, ‘beautifully’, and ‘skillfully’. But if the application of the event analysis of
adverbs is really that narrow, then one wonders if it should be accepted at all.
Parsons saves the analysis by proposing—surely correctly—that just as many adjec-
tives must be interpreted relative to a comparison class (e.g., ‘tall’ and ‘strong’), so too

414
 A DVER B S

must their corresponding adverbs. This had been proposed by Gilbert Harman some
years earlier (Harman 1972).
Another class of adverbs that Davidson excludes from his analysis could be those that
yield opaque contexts. One adverb Davidson focuses on is ‘intentionally’. The problem
is that Mary might have intentionally emptied her bath water into the river without
having intentionally contaminated the river. But the water-emptying and the contami-
nation were one and the same event. So whether the adverb ‘intentionally’ applies to
an event depends on how the event is conceived of or described. It therefore cannot
express a property of an event per se. (The example is a variant of one of Elizabeth
Anscombe’s (Anscombe 1957).)
A further, significant, worry about the underlying motivation of the event analysis of
adverbial modification is that its “logical forms” leave no room for a direct connection
between semantic composition and syntactic composition.
Richmond Thomason and Robert Stalnaker presented an alternative theory of
adverbial modification on which adverbial modifiers can be treated as predicate modi-
fiers—ones which therefore do give logical representations of English sentences that
better represent their syntactic structure as well as our intuitive understanding of their
semantic structure. Their account differs in another way from the event-based analy-
sis: It does not count ‘John wept’ as a structural or logical consequence of ‘John wept
ruefully’. The inference from the latter to the former is valid because of the particular
meaning that the adverb ‘ruefully’ has. This seems like the right view to take since ‘John
allegedly wept’ does not entail ‘John wept’. One who makes the corresponding infer-
ence does not understand the word ‘allegedly’.
Their theory involves the following strategy: first develop a logical language that allows
for the existence of predicate-modifying operators and which is therefore unlike standard
languages for first-order logic; give that language a well-defined semantics; and then give a
semantic analysis of English sentences by correlating them with the sentences of the logi-
cal language (in cases of ambiguity, with more than one) by an intuitive rather than algo-
rithmic procedure, but directly, without further English paraphrases as intermediaries.
The crux of their extension of first-order logic is the inclusion of predicate abstracts.
A predicate abstract is formed from an open formula by prefixing it with a variable-bind-
ing abstraction operator, such as x̂. When F is a monadic predicate, and x̂ an abstraction
operator, x̂Fx will be a monadic predicate. Predicate modifiers can then be prefixed to
these new predicates (as well as the original ones) to form predicates. Let’s look at an
example. If F stands for ‘fainted’, then these are both monadic predicates: F and x̂Fx.
These can be read respectively as ‘fainted’ and ‘is an x such that x fainted. These transla-
tions suggest that the semantics for the language should yield the semantic equivalence
of the two predications. Given Thomason and Stalnaker’s intensional semantics, they
do: the property that is true of something in a possible world just in case F is true of that
thing in that possible world.
Predicate modifiers can then be prefixed to either of these monadic predicates to
yield a monadic predicate. For example, let the letter Vbe the adverbial modifier ‘sur-
reptitiously’. Then VF and Vx̂Fx are equivalent monadic predicates that could be read,
respectively, as ‘surreptitiously fainted’ and ‘is an x such that x surreptitiously fainted’.
These predicates would both be true of a thing in a world just in case that thing surrepti-
tiously fainted in that world.
The main feature of including predicate abstracts in a formal language with predicate
modification is that it allows for the formation of complex predicates by abstraction

415
 DELI A G RA F F FA R A

from non-atomic open formulas. Consequently, abstraction makes room for scope dis-
tinctions, for example, with quantifiers. For example, in addition to the atomic formula
K(a, y) (‘a kissed y’), there is a non-atomic formula y(Kay) (‘a kissed everyone’) that
embeds it. Now let Gstand for the adverb ‘deliberately’. We then have the following as
distinct claims:

(16) yGK(a, y)
(‘everyone is a y such that a deliberately kissed y’);
(17) Gx̂(yK(x, y))(a)
(‘deliberately kiss everyone was what a did’).

These two clearly come apart. You can kiss each person deliberately without its being
kissing-everyone that you did deliberately—for one, because you might know that you’ve
kissed everyone even though you have in fact kissed everyone.
How would this distinction be treated on the event-based approach?

(18) xe(Agent(a, e) šKissing(e) šPatient(x, e) šManner(e, deliberate));

(19) e(Kissing(e) š Agent(a, e) š Patient(everyone, e) š Manner
(e, deliberate)).

The first logical form here would represent the case involving universal quantification
over individuals; the second one would represent the deliberateness of the all-kissing,
which which must now be treated as a temporally extended event with a collective as
patient. Moreover, if we construe the entire second statement as attributing a prop-
erty to the collective everyone, then we must say that this is not a distributive property;
otherwise

(19) would entail (18).

Predicate modification must be intensional in view of cases like the following. Suppose
that all eatings are french-fry eatings: ‘ate something’ and ‘ate french-fries’ have the
same extension. Still, ‘surreptitiously ate something’ and ‘surreptitiously ate french-
fries’ need not have the same extension: People could be surreptitious about its being
fries that they eat without being surreptitious about eating tout court. They could hide
the fries but not the chewing. Accordingly, on Thomason and Stalnaker’s semantics for
their formal language, predicates are assigned propositional functions as their semantic
values—these are functions from an individual to the set of possible worlds in which
that individual satisfies that predicate. Adverbial modifiers are assigned functions from
propositional functions to propositional functions, with no further semantic require-
ment on which functions those are. For example, some modifiers will be restrictive,
while others will not be. ‘Deliberately did )’ is restrictive in the sense that something
cannot have deliberately done ) without having done ). In contrast, ‘allegedly’ is not
restrictive. Someone might be alleged to have done something that they didn’t do at
all. Likewise, some modifiers may be extensional (‘while wearing shorts’), while others
are not (‘deliberately’).
We now turn to a different class of adverbs, falling under a phrase coined by David
Lewis—”adverbs of quantification” (1975). Lewis demonstrates that adverbs can be
used not only for predicate modification but also for quantification. Like ‘more than

416
 A DVER B S

three’, they can answer the question “How many?” And like ‘most’, they can answer the
question “What proportion?” The following pairs illustrate the equivalence between
adverbs of quantification and standard quantificational expressions.

(20) How many? (A small number.)

a. Ballerinas rarely smoke.
b. Few ballerinas smoke.
(21) What proportion? (More than half.)
a. Ballerinas usually smoke.
b. Most ballerinas smoke.

The adverbs in these sentences can each be understood as quantifying over ballerinas.
There are interpretations of (20) and (21), however, in which the ‘ly’ words clearly
do have adverbial uses—at least, that is, if phrases answering the question “How often?”
are all adverbs. For these sentences have frequency readings—distinct from their quan-
tificational readings—that can be equated with the following:

(20) c. Smoking is something that ballerinas rarely do.

 Ÿ (True, for example, if: ballerinas (in general) are not the agents of
smoking events more than three times a year.)
(21) c. Smoking is something that ballerinas usually do.
 Ÿ (True if: ballerinas (in general) are the agents of smoking events more
than thirty times a day.)

Lewis begins his discussion of adverbs of quantification with the claim that even on
what seem to be their frequency interpretations, some adverbs really need to be inter-
preted as answering a “How many?” or a “What proportion?” question.

(22) Caesar always awoke before dawn.

(Lewis’s example involves ‘seldom’ rather than ‘always’ (Lewis 1975, 4).)

One might conceive of this as a answer to the question, “How often did Caesar awake
before dawn?” But, as Lewis notes, we must not interpret (22) as meaning that all times
were times at which Caesar awoke before dawn—even if we were to restrict those to
times at which Caesar was alive. Most times were times at which Caesar was not waking
at all. Rather, we should interpret (22) as meaning that all of Caesar’s awakenings were
before-dawn awakenings. This is an answer not to a “How often?” question, but rather
to a “What proportion?” question. It answers the question “What proportion of Caesar’s
awakenings were before-dawn awakenings?”
But still, we might respond to Lewis, the adverb in (22) is not straightforwardly an
adverb of quantification: first, because it does answer a “How often?” question even if
that answer can be equivalently paraphrased as an answer to a “What proportion?” ques-
tion; second, because it does not quantify over any particular domain explicitly singled
out in the sentence in the way that (20a) and (21a) above do.
Lewis also argued that when adverbs are used as adverbs of quantification, they need
not be understood as quantifying over anything like times or even events at all.

417
 DELI A G RA F F FA R A

(23) A quadratic equation usually has two solutions.

This sentence means that most quadratic equations have two solutions (in an appropri-
ately defined sense of ‘most’). The adverb does not quantify over times: the sentence
does not mean that most times are ones at which quadratic equations have two solu-
tions; any quadratic equation with two solutions has those very same two solutions for
ever and for always. Nor, he argued, can the adverb in (23) be understood as quantifying
over events—it does not mean, for example, that most solvings of quadratic equations
result in the finding of two solutions. Most quadratic equations have never yet been and
never will be solved at all.
If adverbs of quantification do not always range over times or even events, then what
do they range over? Lewis’s answer was that they range over what he called “cases”: all
cases of Caesar’s awakenings were cases of before-dawn awakenings; most cases of quad-
ratic equations are cases of equations with two solutions.
Consideration of adverb-involving variants of the “donkey sentence” (‘Every farmer
who owns a donkey beats it’) made famous by Peter Geach (1962) led Lewis to the view
that adverbs of quantification are “unselective quantifiers.” They bind all free variables
in their scope. Let’s consider Lewis’s first donkey sentence.

(24) A man who owns a donkey always beats it now and then.

Note first that there are two adverbial phrases here, ‘always’ and ‘now and then’. The
latter adverb phrase is interpreted as a frequency adverb, as providing an answer to
the question “How often do the donkeys get beaten?” Its presence prevents ‘always’
from getting a frequency interpretation. ‘Always’ here must be interpreted as a universal
quantifier that ranges over individuals rather than times or events. The sentence means
this:

(25) Every man who owns a donkey is such that every donkey he owns gets
beaten by him now and then.

It’s not merely that for each man who owns a donkey, beatings by him of donkeys he
owns happen now and then. Rather, it’s that every donkey that is owned by a man gets
beaten now and then by that man. The double occurrence of the quantifier word ‘every’
in (25), which represents the truth conditions of (24), leads Lewis to think that the
adverb of quantification ‘always’ unselectively binds those variables that correspond to
men as well as those that correspond to donkeys in the logical analysis of the sentence.
(He does not put things this way. I am reconstructing his argument.) Lewis would offer
the following paraphrase: every case of a pair x, y!where x is a man and y is a donkey
that x owns, is a case in which x beats y now and then. A “case” of a pair x, y!meet-
ing the required conditions (of having x as a man and y as a donkey that x owns) is an
assignment of values to the variables x and y, where those values meet the condition
that x beats y now and then.
Here is another of his regimented examples.

(26) a. Sometimes it happens that x sells stolen goods to y, who sells them to
z who sells them back to x. (Lewis 1975, 6)

418
 A DVER B S

The adverb of quantification ‘sometimes’ here ranges over all assignments to the quad-
ruple of variables x, s, y, z!, and the sentence says that some such assignment meets
the required condition of s’s being stolen goods, x’s selling s to y, y’s selling s to z, and z’s
selling s back to x. In standard logical notation this would be written as follows:

(27) sxyz(s is stolen goods š x sells s to y š y sells s to z š z sells s to x).

But in order to reflect the fact that the English word ‘sometimes’ by itself does the duty
of this string of multiple selective quantifiers, ‘sxyz’, Lewis preferred to use a nota-
tion with an unselective existential quantifier, , that binds every free variable within
its scope:

(28) (s is stolen goods šx sells s to y šy sells s to z šz sells s to x).

An important one of Lewis’s examples was an ‘if’ version of the donkey sentence men-
tioned above. That sentence, (24), contained an indefinite description with an embed-
ded relative clause, ‘a man who owns a donkey’. The relative clause can be dispensed
with by re-phrasing the sentence with an ‘if’-clause:

(29) If a man owns a donkey, he always beats it now and then.

Lewis observed that adverbs of quantification could be “restricted”, as he put it, by an

‘if’ clause which would lead to what he called a “three-part construction” consisting of
(i) an adverb of quantification, (ii) a restriction on that adverb, and (iii) the remaining
material of the sentence. The quantifier words in these three-part constructions differ
from the standard ones in two ways. They differ by being unselective—by binding all
free variables in their scope. But they also differ from the standard quantifiers in being
binary rather than unary. The standard quantifiers, v, and v, are unary in the sense
that they combine with a single sentence to form a sentence, binding a selected one of
the free variables in that formula.
For example, to form the first-order paraphrase,

(29c) xy(x is a man and y is a donkey and x owns y o x beats y now and
then),

of (29), one first prefixes the selective unary quantifier y to a single (open) sentence,
then prefixes the selective unary quantifier x to the resulting (open) sentence to yield
(29c).
Using Lewis’s unselective, binary quantifier, one forms a truth-functional equivalent
of (29c) by prefixing the unselective binary quantifier to the open sentences: ‘x is a
man and y is a donkey and x owns y’ and ‘x beats y and now and then’ (he uses ‘+’ as a
separator):

(30)  + (x is a man š y is a donkey šx owns y) + (x beats y now and then).

In this case the initial clause, ‘a man owns a donkey’—logically paraphrased as the open
sentence ‘x is a man and y is a donkey and x owns y’—forms the second part of the three-
part construction. The remaining material of the sentence—logically paraphrased as
the open sentence ‘x beats y and now and then’—forms the third part of the three-part

419
 DELI A G RA F F FA R A

construction. The truth conditions for the universally quantified three-part construc-
tion are given by this clause:

+ ) + < is true just in case all variable assignments that satisfy ) also
satisfy <.

Once we associate each adverb of quantification with the explicitly quantificational

expression, Q, we can generalize that truth clause:

Q + ) + < is true just in case Q variable assignments that satisfy ) also

satisfy <.

For example, the follow triples of sentences display the logical forms and truth condi-
tions that are assigned by this procedure to the displayed English ‘if’-sentences.

(31) ‘Usually’ Ÿ‘Most’

a. If a man owns a donkey, it usually kicks him now and then;
b. most + (x is a man šy is a donkey šx owns y) + (y kicks x now and
then);
c. Most assignments g for which g(x) is a man and g(y) is a donkey and g(x)
owns g(y) are such that g(y) kicks g(x) now and then.
(32) ‘Rarely’ Ÿ‘Hardly any’
a. If a man owns a donkey, it rarely kicks him all the time;
b. Hardly any + (x is a man šy is a donkey šx owns y) + (y kicks x all the
time);
c. Hardly any assignments g for which g(x) is a man and g(y) is a donkey
and g(x) owns g(y) are such that g(y) kicks g(x) all the time.

One could ask why the English a. sentences are assigned the logical forms displayed
in the b. sentences. Irene Heim would, relatively soon after the publication of Lewis’s
paper, develop a systematic and elegant theory of indefinite descriptions and anaphoric
pronouns that provided answers to just that question (Heim 1982). She also devel-
oped an analysis of definite descriptions within her system which allowed for a unified
semantics for definites and indefinites—one which accounted for the difference between
them as a difference with respect to a small number of easily statable features. In later
developments by linguists, adverbs of quantification were treated as selective quantifiers
that bind situation variables; the corresponding situations are subsequently claimed to
have certain properties. (See, for example, Stephen Berman (1987), Heim (1990), Kai von
Fintel (1994, 2004), and Paul Elbourne (2005). Angelika Kratzer (1989) gave an explicit
account of the structure of the situations needed for situation-based approaches.)
There are two problems with Lewis’s analysis that deserve mention before we con-
clude. One of these has come to be known as the proportion problem. As we already
noted, adverbs of quantification can be understood as answering questions frameable as
“How many?” or “What proportion?” questions. To say that children rarely like vegeta-
bles is to say the the number of children who like vegetables is small. (Or perhaps it is
the proportional claim that the percentage of children who like vegetables is small.) To

420
 A DVER B S

say that children usually like candy is to say that the proportion of children who like
candy is greater than 50 percent. Analogously, to say (33a), that a man who owns a don-
key usually beats it now and then is to say (33b), that the proportion of donkey-owning
men who beat their donkeys is greater than 50 percent:

(33) a. A man who owns a donkey usually beats it now and then.
b. The proportion of donkey-owning men who beat their donkeys now
and then is greater than 50 percent.

But that is not quite the truth condition that Lewis’s analysis assigns to (33a). Let a
man-o-donkey pair be a pair consisting of a man and a donkey he owns; let a man-b-
donkey pair be a pair consisting of a man and a donkey he beats now and then. Then
according to Lewis’s view, to say that a man who owns a donkey usually beats it now and
then is to say that the proportion of man-o-donkey pairs that are man-b-donkey pairs is
greater than 50 percent.

(33) a. A man who owns a donkey usually beats it now and then
c. The proportion of man-o-donkey pairs that are man-b-donkey pairs is
greater than 50 percent.

The problem is, (33b) and (33c) do not have the same truth conditions. Think of it this
way. We can rephrase (33b) as follows:

(33) bc. The proportion of men that are in a man-o-donkey pair that are in a
man-b-donkey pair is greater than 50 percent

The difference between (33b) and (33c) shows itself when there is a single donkey-
beating man who owns a disproportionate number of the world’s owned donkeys. If that
one man owns a thousand donkeys that he beats now and then, while the remaining
five-hundred donkey-owning men, who each own one donkey, never beat their donkey
ever, then (33c) is true while (33b)—and (33a) along with it—is false.
A second problem is that the prosody of an assertion with adverbs of quantification
can affect its truth conditions. Consider a variant of (22):

(34) Caesar usually befriended his enemies.

Normally, this would be understood to mean that most of Caesar’s enemies were such
that he befriended them.
But the truth conditions of this sentence can also be affected by stress. Let’s use
capitals to indicate stress, which for our purposes is any way of making a phrase promi-
nent. (For more detailed discussion of focal stress and its effect on meanings, see Daniel
Büring “Focus and Intonation” (Chapter 1.9, this volume). Fred Dretske’s “Contrastive
Statements” (Dretske 1972) has been seminal in this area.)

(34c ) Caesar usually befriended his enemies.

With stress of ‘enemies’, the sentence means that most of Caesar’s befriendings were
befriendings of his enemies. But contrast this:

421
 DELI A G RA F F FA R A

(34cc ) Caesar usually befriended his enemies.

With stress on ‘his’, the sentence means that most of Caesar’s enemy-befriendings were
befriendings of his own enemies. Further, contrast again this:

(34c) Caesar usually befriended his enemies.

With stress on ‘befriended’, the sentence means that most of Caesar’s actions towards
his enemies were befriendings. (We would want to limit this to actions of some relevant
type.) Here is another example, taken from Mats Rooth (1985, 164):

(35) In Saint Petersburg, officers always escorted ballerinas;

(36) In Saint Petersburg, officers always escorted ballerinas.

The first sentence means that all escortings by officers in Saint Petersburg were escort-
ings of ballerinas. The second sentence means that all escortings of ballerinas in Saint
Petersburg were escortings by officers. If Piotr, an officer, escorts Nastenka, who is not a
ballerina, then (35) is false, but (36) might still be true. If Sergei, who is not an officer,
escorts Larissa the ballerina, then (36) is false, but (35) might still be true.
Here we have only skimmed the surface of the issues raised by our motley crew of
adverbs. I have focused on two issues discussed by philosophers: event-based semantics
for adverbial modification and the semantics of adverbs of quantification. The latter
topic has been taken up almost exclusively by linguists, and I have not tried to system-
atically set out their dialectic. There are also many strands that could be followed which
we never picked up at all: for instance, how to interpret frequency adverbs, and whether
we really do have an adequate characterization of the category ‘adverb’ on which to base
our theories.

References
Anscombe, Elizabeth (1957), Intention, Blackwell, Oxford.
Beaver, David I. & Clark, Brady Z. (2008), Sense and Sensitivity: How Focus Determines Meaning, Wiley-
Blackwell, Oxford.
Bennett, Jonathan (1988), Events and Their Names, Oxford University Press, Oxford.
Berman, Stephen R. (1987), ‘Situation-based Semantics for Adverbs of Quantification’, in J. Blevins & A.
Vainikka, eds., University of Massachusetts Occasional Papers, Vol. 12, GLSA Publications, pp. 47–68.
Büring, Daniel (2011), ‘Focus and Intonation’, in G. Russell & D. G. Fara, eds., The Routledge Companion to
Philosophy of Language, Routledge, London.
Cinque, Guglielmo (1999), Adverbs and Functional Heads: A Cross-Iinguistics Perspective, Oxford University
Press, Oxford.
Davidson, Donald (1967), ‘Causal Relations’, Journal of Philosophy 64: 691–703.
Davidson, Donald (1985), ‘Adverbs of Action’, in B. Vermazen & M. Hintikka, eds., Essays on Davidson:
Actions and Events, Oxford University Press, Oxford. Page references are to reprint in Davidson (2001).
Davidson, Donald (2001), Essays on Actions and Events, 2nd edition, Clarendon Press, Oxford.
Dretske, Fred (1972), ‘Contrastive Statements’, Philosophical Review 81(4): 411–437.
Elbourne, Paul D. (2005), Situations and Individuals, MIT Press, Cambridge, MA.
Geach, Peter (1962), Reference and Generality, Cornell University Press, Ithaca. Page references are to the
3rd edition (1980).
Harman, Gilbert (1972), ‘Logical Form’, Foundations of Language 9: 38–65.
Heim, Irene (1982), The Semantics of Definite and Indefinite Noun Phrases, PhD thesis, University of
Massachusetts, Amherst.

422
 A DVER B S

Heim, Irene (1990), ‘E-Type Pronouns and Donkey Anaphora’, Linguistics and Philosophy 13(2): 137–177.
Kratzer, Angelika (1989), ‘An Investigation of the Lumps of Thought’, Linguistics and Philosophy, 12(5):
607–653.
Lewis, David (1975), ‘Adverbs of Quantification’, in E. Keenan, ed., Formal Semantics of Natural Language,
Cambridge University Press, Cambridge, pp. 3–15.
Parsons, Terence (1990), Events in the Semantics of English: A Study in Subatomic Sentences, MIT Press, Cam-
bridge, MA.
Parsons, Terence (1995), ‘Thematic Relations and Arguments’, Linguistic Inquiry 26(4): 635– 662.
Rooth, Mats (1985), Association with Focus, PhD thesis, University Massachusett, Amherst.
Rooth, Mats (1999), ‘Association with Focus or Association with Presupposition?’, in P. Bosch & R. van
der Sandt, eds., Focus: Linguistic, Cognitive, and Computational Perspectives, Cambridge University Press,
Cambridge, pp. 232–244.
Schein, Barry (2011), ‘Event Semantics’, in G. Russell & D. G. Fara, eds., The Routledge Companion to Phi-
losophy of Language, Routledge, London.
Thomason, Richmond H. & Stalnaker, Robert C. (1973), ‘A Semantic Theory of Adverbs’, Linguistic Inquiry
4(2): 195–220.
von Fintel, Kai (1994), Restriction on Quantifier Domains, PhD thesis, University of Massachusetts,
Amherst.
von Fintel, Kai (2004), ‘A Minimal Theory of Adverbial Quantification’, in H. Kamp & B. Partee, eds.,
Context-Dependence in the Analysis of Linguistic Meaning, Elsevier Ltd, Oxford, pp. 137–175.

423
 3 . 10
MAS S TER M S
Francis Jeffry Pelletier

1 Introduction
Mass terms are those such as ‘water’, ‘computer software’, and ‘knowledge’. They are
contrasted with count terms such as ‘person’, ‘computer program’, and ‘belief’. Intui-
tively, mass terms refer to “stuff” while count terms refer to “objects.” Since mass terms
refer to stuff, they (but not count terms) allow for measurement: ‘a liter of water’, ‘three
CDs worth of computer software’, ‘many books worth of deep knowledge’. Since count
terms refer to objects, they (but not mass terms) allow for counting, quantifying, and
individuating: ‘a person’, ‘three computer programs’, ‘each belief of his’. Philosophers
from many areas have found this distinction to be of interest—the metaphysical ques-
tion of the primary existence of gunk vs. things is one obvious area, but also issues in
the notions of identification and reidentification have been thought to be related to the
distinction: Is it the same building when all the concrete has been replaced with new
concrete? Been replaced with stone? A statue can cease to exist without its constituent
matter ceasing to exist. Does this mean there are two entities here: the statue and the
parcel of matter? Further afield, but still relevant, are questions about the referents of
“abstract” mass terms, such as ‘Curiosity is an admirable quality to have’. The fact that
‘knowledge’ is a mass term while ‘beliefs’ is a count term have led some to question the
account of knowledge as a justified true belief.
The examples just given were from English. Not all languages follow English in their
characterization of this distinction—indeed, perhaps no other language is quite like
English. Even within the Indo-European language group, where this distinction most
closely follows that of English, there are relevant differences. And in the wider realm
of the world’s languages, there are those that do not allow plural/singular marking on
individual nouns but only on larger phrases. There are languages that do not have a
plural/singular marking for noun phrases at all (nor agreement with verb phrases); there
are languages that do not have quantifiers that operate on nouns or noun phrases; there
are languages that do not have determiners like ‘a(n)’ and ‘the’, even though they mark
singular/plural. Thus, the examples given in the previous paragraph—which make it
seem that there are clear syntactic ways to distinguish count from mass nouns—do not
have the same purchase (or perhaps no purchase at all) in these languages. And this can
be seen as raising questions concerning the philosophical relevance of, or interest in,
the distinction.
Within the study of mass terms, the philosophical problems traditionally encountered
include distinguishing mass from count terms (is it a syntactic or a semantic distinction,
or something else?), deciding the extent of the classification (does it include more than
 M A SS TE R MS

noun phrases?), describing the semantic underpinnings of mass terms (since they are
not true of individuals, how can a model theory be developed?), and explaining the
ontology presupposed by mass terms vs. count terms. Alongside these concerns, there
is the meta-philosophical question of the extent to which the linguistic practices of the
speakers of a language can be used as evidence for how those speakers view reality or,
indeed, as evidence for what reality is like.
The examples above suggest that the distinction is designed to subdivide noun phrases,
but there have also been attempts to extend the distinction to (some) adjectives (Quine
1960: p. 104), and especially to verb phrases. Verb phrases designating processes like
to eat would be +mass while those designating achievements like to prove would be
+count. See Mourelatos (1978) and Bach (1982a, b), as well as later works inspired by
them, for this extension of the notion +mass/+count.

2 +MASS and +COUNT as Syntax

Many descriptive grammars of English, e.g., Quirk et al. (1985), give a syntactic charac-
terization of the +mass/+count distinction within the category of noun (N). That is,
they view the fact that some noun (e.g., water ) is a mass term as giving an explanation
for why some combinations with other words are ungrammatical. For example, they
might say

(1) a. Mass nouns, unlike count nouns, do not have plural forms and thus all
verb agreement is singular.
b. Mass nouns, unlike count nouns, do not admit of numeral modifiers.
c. Mass nouns, unlike count nouns, do not allow “individuative” quanti-
fiers such as each, every, some (stressed), few, several, many, . . .
d. Mass nouns, unlike singular count nouns, employ measurement terms
such as much, a lot of, (a) little.

(There are further problematic cases for the distinction: “collectives” such as team, com-
mittee, army, herd, alphabet, and “pluralia tantum” which has a large number of subcate-
gories but includes words like pliers, binoculars, suds, intestines, bleachers, ruins, groceries,
spoils, valuables, contents, etc.)
The syntactic characterization in (1) is supposed to account for the following
classifications:

(2) Mass Nouns: water, blood, cutlery, knowledge, carpeting, advice, . . .

(3) Count Nouns: person, dog, spoon, belief, carpet, suggestion, . . .

These are all Ns viewed as being in the lexicon . . . lexical nouns, to give them a name.
The +mass/+count features are viewed by Quirk et al. and others of this syntactic per-
suasion to be a part of the lexical characterization of the nouns. These features are to be
inherited from the lexical items into the larger and larger syntactic units that are pres-
ent in extended phrases: blood as lexical N contains the syntactic feature +mass, which
is inherited by the common noun phrases (CNP) bright red blood and bright red blood that
is on the floor and the full noun phrase (NP) the bright red blood that is on the floor. The fact
that this longer phrase is also +mass is what ultimately explains why

425
 F RA N CI S J EF F RY P E L L E T I E R

(4) *The bright red blood that is on the floor are slippery

is ungrammatical. (Because the fact that the phrase is +mass prohibits it from being
plural, as (1a) says, and hence the agreement with the verb phrase does not happen.)
Violations of the constraints involving +mass and +count yield ungrammatical results
that have the same status as other syntactic violations; (4) is no more a part of English
than is

(5) *Dog the quickly

It might be noted that both of the mass and count categories contain terms that are
“abstract”: knowledge and advice are +mass, while belief and suggestion are +count.

3 +MASS and +COUNT as Semantics

Some descriptive grammars of English, e.g., Huddleston and Pullum (2002), think of
the +mass/+count distinction as a description of the semantic properties of the deno-
tation of the terms. In this type of view, mass meanings contrast with count meanings:

(6) a. Mass meanings are true of stuff; count meanings are true of things
b. Mass meanings are divisive in their reference; count meanings are true of a
unit as a whole
c. Mass meanings are cumulative in their reference; (singular) count mean-
ings are not true of groups of that which they are true
d. Stuff that mass meanings are true of cannot be counted; count meanings
are true of individuated items that can be counted
e. Stuff that mass meanings are true of can be measured; (singular) count
meanings are not measurable

Some theorists take the divisiveness and the cumulativity conditions together to be
called the homogeneous in reference condition.
In the semantic view, the fundamental difference between mass and count terms
is that count terms are true of objects—entities that are distinct from each other even
while being of the same type, and thus one can distinguish and count them—while
mass terms are true of stuff that is undifferentiated with respect to the term being used
to describe it. This in turn explains why mass terms, unlike count terms, are divisive in
their reference: they permit something that the mass term is true of to be arbitrarily sub-
divided and the term to be true of these parts as well. Taking the water in the glass to be
something that is water is true of, it can be divided into parts and is water will be true of
both parts. And again, mass terms, unlike count terms, are also cumulative in their refer-
ence: Putting the water contained in two glasses into a bowl yields something of which
is water is true. But the same is not the case with a count term like dog. Chopping up a
dog does not yield more things of which is a dog is true, nor do two dogs make a thing of
which is a dog is true. (Other than in a Frankenstein-like scenario.)
In a semantic approach, the features +mass/+count are descriptions of the semantic
value of lexical nouns and the larger common noun phrases, etc. Thus, they do not
figure in the syntactic well-formedness constraints of a grammar, but would emerge as
a description of what the semantic values of the embedded nouns are, and how these

426
 M A SS TE R MS

semantic values get altered by the syntactic combination of those nouns with other
words. These features do not syntactically rule anything out; the most that can be
said is that certain combinations are “semantically anomalous,” and hence can’t be
interpreted.
As with the syntactic version of +mass/+count, the lexicon supplies individual
words with a set of syntactic features and also a set of semantic values. Larger and larger
phrases that contain the noun also contain the semantic information mentioned in the
lexical items, modified in accordance with rules that describe the semantic effect of
being syntactically combined in the manner that is employed.
The difference between +mass/ +count as syntax and +mass/ +count as semantics
thus is whether these features are seen as syntactic well-formedness constraints that
yield ungrammaticality when violated or as semantic interpretability constraints upon
syntactically correct sentences that yield semantic anomaly when violated.

4 Some Problems for the Syntactic Approach

In the Syntactic Approach, lexical Ns are assigned either a +mass or +count feature,
and this feature controls the syntactic (in)admissibility of larger phrases. But there are
many words that have both mass and count meanings, for instance

(7) a. Concrete terms

(i) a lot of chocolate / many more chocolates
(ii) more discipline / an academic discipline
(iii) too much paper / write a paper
b. Abstract terms
(i) much discussion / three different discussions
(ii) a lot of difference / two differences
(iii) much more data / many more data

The examples in (7) are just the tip of the iceberg, There are many more of these “dual
life” terms, sometimes forming regular patterns, but sometimes not:

(8) Mass terms used “countily”:

a. Pinot Noir is wine / Pinot Noir is a wine
b. Kim produces sculpture / Kim is producing a sculpture
c. Beer on the table / Three beers on the table / Eight beers on tap
(9) Count terms used “massily”
a. Leslie has more car than garage
b. Chris Pronger, 6c6s worth of defenseman . . .
c. Some people like data better than theory

As (Huddleston and Pullum 2002: p. 335) remark “. . . the dual use of chocolate is not
remotely exceptional but is representative of an extremely widespread phenomenon,”
and they follow this with a list of 25 examples chosen over a wide variety of types
of nouns that illustrate just how widespread the phenomenon of a noun having two
equally salient meanings where one is +mass but the other +count.
There’s also the “universal grinder” of Pelletier (1975), which is like a meat grinder
except that it can accommodate any object, no matter how large, and its teeth are so

427
 F RA N CI S J EF F RY P E L L E T I E R

powerful and fine that it can grind anything, no matter how strong. Insert an object that
falls under any (concrete) count noun into one side . . . for example, a hat. Push the
button, and the result is that there is hat all over the floor. (True despite the fact that
we might have some other term, e.g., felt, also describing what is on the floor.) Another
push of the button and we can have book all over the floor. An unfortunate accident
might generate curious cat all over the floor.
One might also think of “universal packagers” in this regard, that take any item of
which a mass term is true and convert it into an object. Any time there is a use for
a particular type of some mass then there can be a count term that describes it—for
example, a finely silted mud, which can be a name for a type of mud and also a predicate
that is true of all individual exemplars of this type. And if there is a standardized amount
of M that is employed in some use, then there will be a count term that describes this
amount, such as a beer or an ice cream. Furthermore, there seems always to be a count use
for any alleged mass term M, meaning (roughly) a kind of M. Putting all these together,
a term like a scotch could be true of individual servings (thus being independently
true of each piece of the actual matter in the various glasses), or true of the different
standardized amounts (so that two instances of the same standard one-ounce serving
count as only one such standardized amount), or true of the different kinds of scotch on
the table or available at the bar. Thus any of ‘four’, ‘three’, or ‘five’ could be true answers
to the question “How many scotches are on the table?”
These considerations show that the appropriate theory needs to talk about meanings
of terms, or uses of the terms, or maybe occurrences thereof (some occurrences are +mass,
others of the same word are +count). But then this is no longer a syntactic account!
And the syntactic approach just doesn’t work. For, it will turn out that since any noun
can be either mass or count, a +mass/+count syntactic distinction does no work—
nothing is ruled out by the syntactic rules.

5 Some Problems for the Semantic Approach

Many Ns have both a natural mass and a natural count sense. So the basic lexical item
that gets entered into a phrase structure description of a sentence will be one of these
senses. It is never very clear how this is supposed to be effected in a grammar, but we
will simply assume that there is some way that this can be done. But even if we assume
that the N and the correct sense is entered there nonetheless seem to be some serious
difficulties that are semantic mirrors of the difficulties found in the syntactic approach.
Many formal semanticists (e.g., Link 1983; Chierchia 1998a, b; Pelletier and Schubert
1989/2003; Landman 1991) take the characteristics in (6) to be best accounted for in
terms of a semi-lattice theory. A semi-lattice has no lowest elements and is atomless.
The idea is that anything that water, for example, might be true of has subparts—things
in the lattice that are its parts—of which water is true; and any two elements in the
water-lattice find a joined element also in the lattice that represents the merge of those
two elements.
But it should be noted that many mass terms obviously are not “atomless” in the sense
required by this theory. Consider

(10) furniture, cutlery, clothing, equipment, jewelry, crockery, silverware, foot-

ware, bedding, toast, stemware, gravel . . .

428
 M A SS TE R MS

Clearly there are atomic parts of these; yet they are considered mass terms by any of
the traditional grammars. So it cannot be an atomless mereology that accounts for the
mass nature of these words; and since it doesn’t account for the mass nature of these
particular words, there seems to be no reason to think it accounts for the mass nature
of any words.
Some theorists, e.g., Huddleston and Pullum (2002), take this as evidence that terms
like those in (10) are of a different nature than what we have been calling ‘mass terms’,
and are to be treated differently. Huddleston and Pullum call them ‘aggregate terms’
and semantically distinguish them from other mass terms by their being true of “very
different sorts of things.” The idea is that furniture, for example, is true of sofas, chairs,
tables, carpets, and so on, and that these are “very different” from one another. But a
true mass term, for example ‘blood’, is really true only of one kind of thing.
But one might still wonder: Are any words at all that obey the condition on
divisiveness? Are there really any words that are atomless—whose referent has no
smallest parts? Doesn’t water have smallest parts: H2O molecules perhaps? A standard
defense of the divisiveness condition in the face of these facts is to distinguish between
“empirical facts” and “facts of language.” It is an empirical fact that water has smallest
parts, it is said, but English does not recognize this in its semantics: the word water
presupposes infinite divisibility.
It is not clear that this is true, but if it is, the viewpoint suggests interesting questions
about the notion of semantics. If water is divisive but water isn’t, then water can’t be
the semantic value of water (can it?). This in turn suggests a notion of semantics that is
divorced from “the world,” where semantics is not be a theory of the relation between
language and the world. But it also would seem not to be a relation between language
and what a speaker’s mental understanding is, since pretty much everyone nowadays
believes that water has smallest parts. Thus, the mental construct that in some way
corresponds to the word water can’t be the meaning of water either. This illustrates a
kind of tension within “natural language metaphysics” (Bach 1986a, b).
Another problem with the semantic approach to the distinction comes from pairs of
words where one is mass and the other is count and yet the items in the world that they
describe seem to have no obvious difference that would account for this. On the intuitive
level, it seems that postulating a semantic difference should have some reflection in the
items of reality that the terms designate. But this is just not true. There seems to be
nothing in the referent of the following mass vs. count terms that would explain how
they should be distinguished—as they intuitively are. (See McCawley 1975 for further
examples.)

(11) a. Concrete terms

(i) baklava vs. brownies
(ii) garlic vs. onions
b. Abstract terms
(i) success vs. failures
(ii) flu vs. colds

To many, these examples and their surrounding facts have seemed to prove that the
linguistic features of +count and +mass do not have any backing in reality. Nor any
backing in people’s intuitive understanding of when a word will be +mass or +count or
what it is for a word to be +mass or +count.

429
 F RA N CI S J EF F RY P E L L E T I E R

6 +COUNT and +MASS Semantic Theories

A rather heterogeneous group of topics has been included within the subject matter of
the semantics of +mass vs. +count terms. Given that the general topic concerns the
content of lexical items, we can expect that much of the discussion will differ from that
of the more traditional compositional semantic theories.
The older philosophical literature on mass terms (and many other terms) tended to
find different meanings for the terms under consideration, depending on what role they
had in a sentence. Theorists would consider one of the following types of uses of a mass
term:

• As names, as in Water covers most of the globe

• As predicates true of quantities/portions of matter, as in John drank some water
• As predicates true of objects, as in This ring is gold
• As predicates true of kinds/substances, as in Claret is wine
• As predicate modifiers, as in She is wearing a gold bracelet

To some this suggested that +mass terms were ambiguous.

In general, a mass term in predicative position may be viewed as a general term

which is true of each portion of the stuff in question, excluding only the parts
too small to count . . . A mass term used in subject position differs none from
such singular terms as ‘mama’ . . . , unless the scattered stuff that it names be
denied the status of a single sprawling object.
(Quine 1960: pp. 97–98)

To others it suggested that a theory should pick one of these ways that mass terms are
used in sentences as basic and either ignore the others or try to generate them by some
“semantico-syntactic trick.” Parsons (1970) took them always to be names of substances
(“in the chemistry sense”), but when a mass noun M was “in predicate position” such an
occurrence was transformed into is a quantity of M. (This same transformation occurred
in some other positions also.) One of the earliest attempts to employ mereology as a
tool for representing mass terms was Moravcsik (1973). In such a framework nouns (and
adjectives) designated mereological wholes and the copular connection becomes is a part
of. Pure mereological approaches have been found wanting, because of the problem of
“minimal parts”—classical mereology has no minimal parts other than the empty part,
yet it is generally assumed that the designations of ordinary language +mass nouns do
have minimal parts. So Moravcsik invoked a notion of is a part having relevant structural
properties. This has also been found wanting (see Montague 1973; Pelletier 1974), but
more clever ways of operating do seem to avoid these problems (e.g., Bunt 1979, 1985;
see also Burge 1972; Ojeda 1993; Moltmann 1998). A parallel—and related—develop-
ment has been the notion of a (join) semi-lattice to support part–whole structures. This
was first urged onto the mass term semantics world by Link (1983) and picked up by
many others.
A related question is whether it is possible to show a unity within the +mass/+count
realm. Gillon (1992, 1999), who adopts a syntactic view of the +mass/+count distinc-
tion, argues for a “common semantics” for +mass and +count nouns. Count nouns are
assigned a set of atomic individuals, and a mass noun is assigned the mereological join of

430
 M A SS TE R MS

the individuals of which it is true. Chierchia (1998a, b) instead assigns mass nouns to be
“inherently plural.” And in Chierchia (2010), +mass and +count are given the same
type of semantic value as one another, with the exception that +mass terms, unlike
+count terms, are said to have “vague minimal parts.”
One understudied aspect of this concerns “abstract” mass nouns (advice, freedom,
information, . . .). Many theorists wish to invoke a mereology or a semi-lattice as the
relevant semantic structure for concrete mass nouns such as water, blood, steel, . . . , but
mereology (in particular) seems completely out of place for abstract nouns. Intuitively,
+mass is independent of whether the noun is ±abst, and therefore any semantic
technique that is correct for concrete mass terms should be applicable to abstract
mass terms. Differences between abstract and concrete mass terms should be due to
the ±abst, and not due to +mass. But mereology just doesn’t apply to +abst,+mass
terms; hence, it should not be a defining semantic feature of –abst (concrete) +mass
terms either. And so, mereology would not be a part of the semantics of +mass, but at
most a consequence of the interaction of a more general semantic account of +abst and
+mass. Some attempts towards an account of the semantics for +abst, +mass terms
using the semantics of comparatives can be found in Nicolas (2010); an account using
the methodology of Natural Semantic Metalanguage (Wierzbicka 1996) can be found
in Goddard and Wierzbicka (2010). I think all these authors would admit that their
account needs to be expanded before it can truly be said to describe +abst, +mass
nouns in general.
Accounts of the meaning of items that are lexically +mass can be divided into two
sorts: those that are viewed as part of formal semantics and those that think of give a
more “ordinary language” account of the meaning. The former will try either to employ
existing constructs from formal semantics or else will recommend the use of some novel
logical or mathematical machinery to be incorporated into formal semantics. The latter
will offer more of a “dictionary sense” to these lexical entries, often phrased in terms of
some set of cross-cultural semantic primitives.
This latter methodology has been most deeply developed within the Natural
Semantic Metalanguage (NSM) framework of Wierzbicka (1996). Indeed, many of
the works within this framework have concerned themselves directly with the issue of
the meaning of +mass terms, and they have outlined a quite rich structure within the
category of +mass nouns. (See in particular Wierzbicka 1988; Goddard and Wierzbicka
2002; Goddard 2009.) Although there are many who find the NSM framework to be
wanting (e.g., Barker 2003; Matthewson 2003; Riemer 2006; Jackendoff 2007), the work
has produced the most detailed descriptions of different types of +mass (and +count)
nouns and ought to be studied for that reason alone.
Within the formal semantic group of theories, it has long been recognized that
standard first-order logic seems unsuited to representing sentences involving mass terms.
For example, even if one grants that (12a) is to be represented as (12b), nonetheless
Tarski’s classic (13a) can’t reasonably be represented as (13b), because there are no
plausible values for x:

(12) a. Men are mortal.

b. x(Man(x) Š Mortal(x))
(13) a. Snow is white.
b. x(Snow(x) Š White(x))

431
 F RA N CI S J EF F RY P E L L E T I E R

For, what could be the value of x in “For all x, if x is snow, then x is white”? Intuitively,
we want it to be “snowy stuff,” but the idiom of classical logic is committed to the val-
ues being objects/things/entities—elements of the domain. However, those are what is
designated by +count nouns, not +mass nouns.
Earlier attempts invoked relational constants into the first-order language to accom-
modate +mass terms. Parsons (1970), for example, used names for substances and the
relations quantity-of and consitituted-by as well as a “substance-forming operator.” Burge
(1975) considers two different theories along these lines: a “relational” account that
analyzes sentences like (14a) as (14b).

(14) a. This ring is now gold

b. Gold(r, now)

He rejects this account because it takes the basic individuals to be stages of objects. His
preferred account employs the basic Three-place relation of ‘x constitutes y at (time) t’,
analyzing (14a) as

(15) (x)(Gold(x) ^ C(x, r, now) )

One or the other of these two general approaches have been adopted by a wide range of
theorists who want to employ as much of ordinary first-order logic as possible.
An important topic concerns the “dual life” that many (most? all?) nouns lead in
English and similar languages. What should the semantic value of the lexical item be
for such nouns? Consider chocolate, for example. Should its semantic value be the set
of chocolates? But of course not only do we have the chocolates that are delivered to
one’s lover on Valentine’s Day but there is the chocolate that they are made from. Here
is where a first-choice point arises. Should we take the set of chocolates to be basic
and somehow “derive” the chocolate from them? Or should we take the chocolate stuff
as basic and derive the set of chocolates from that? Or should there be two separate
meanings that are each basic? Or should there be one meaning that is “unspecified for
+mass/+count” thereby implicitly including both meanings?
The literature on mass nouns seems to be committed to taking one of the meanings as
basic and deriving the other. The process is called “coercion” and is said to be triggered
by features of the surrounding linguistic context (or, sometimes, by the nonlinguistic
context). For example, if the mass meaning is taken as primitive, then a sentence like
Abelard gave Heloise seven chocolates is said to have ‘chocolate’ coerced into a +count
meaning by the presence of the plural and the number modifier. If the sentence were
Each chocolate was made by hand, then the presence of ‘each’ will coerce the basic mass
meaning into a count meaning. Alternatively, if the count meaning is taken as primitive,
then a sentence like The box contained two kilos of chocolate would be seen as coercing
the basic count meaning into the mass meaning by means of the measure phrase ‘two
kilos of’.
It is never clear why one meaning vs. the other is taken as primitive. It is as if the
theorists have some special insight into what the true, real, and underlying meaning
is. It seems unlikely to be due merely to frequency; maybe it is some combination of
frequency and saliency. When examples are given, they do not use nouns like chocolate
but rather house, child, car, . . . . It then seems more plausible to say that Sally owns too
much house! employs a coerced mass meaning of house—which is “really” a +count

432
 M A SS TE R MS

term. But this seems to become a more suspect attitude when one views the full range of
“dual life” terms. The list in (Huddleston and Pullum 2002: p. 335) shows such a truly
wide variety of these terms that one would hesitate to employ the notion of ‘coercion’
so rapidly.
Some theorists have thought that there should be a distinction made within +mass
nouns for those that are homogeneous vs. those that are “atomic.” The sort of distinction
these theorists have in mind is between words like blood and furniture. We have already
seen in conjunction with (10) above that Huddleston and Pullum (2002) wish to make
these form separate categories within −count, on the basis that these “aggregate terms”
are true of “very different types of things.” This issue of how a mass term can be a cover
term for a variety of different subtypes, each of which is a count term, is discussed in
the psychology literature under the heading of “mass nouns as superordinate terms” (see
Markman 1985; Wisniewski et al. 1996, 2003; Takatori and Schwanenflugel 2008), with
an emphasis on whether there is some basis “in reality” for the notion of an ‘individual’
vs. ‘stuff’. This is often called “natural atomicity.” We should notice that, while the
atomic parts of furniture are rather large—making it clear that the natural atomicity has
been fulfilled—the atomic parts of cutlery are smaller and those of gravel are very small
indeed. And as I mentioned above, pretty much every English speaker believes that
there are atomic parts of any purported mass term, even such prototypical ones as ‘water’
or ‘blood’. What does seem true, however, is that for some mass nouns, ‘gravel’ perhaps
or maybe ‘coffee’, just what counts as atomic parts is “vague.” The idea is that while
there are clear cases of the atomic parts of (most?) +count nouns, and of some +mass
nouns (the clear-cut “aggregates”), and while some words perhaps designate “complete
homogeneity” (maybe ‘space’?), these endpoints merge into one another in the same
way that any of the traditional vague predicates do. And then the issue of what nouns
should be +count and which should be +mass is of the same nature as vagueness: here,
the vagueness of what counts as an atomic part. This line of research is pursued by
Chierchia (2010).

7 Some Diachronic and Cross-Linguistic Data

Chierchia (2010) gives a very helpful three-way division of how various languages deal
with the +mass/+count distinction. Without insisting on the exhaustivity of its clas-
sification, or even on the ultimate “truth” of its vision, we can nonetheless use the labels
to give general characterizations. According to this division, the world’s languages fall
into one of the following three groups with regard to +mass/+count.

1. Number marking languages, which have overt number features that obligatorily
appear on nouns. Here the +mass/+count distinction applies to the nouns directly.
(Most?) Indo-European languages, e.g., current English, are such languages.
2. Classifier languages, which do not have obligatory number marking on nouns (and
arguably do not have a singular/plural contrast at all on nouns). Lexical nouns in
such languages could be viewed as +ma ss, although there is a +ma ss/+cou nt dis-
tinction that is active more generally. (For this reason it might be better to view the
lexical nouns as unspecified for +ma ss/+cou nt .) The classifiers in these languages
enforce the +ma ss/+cou nt distinction, but at the level of an entire “classified noun
phrase.” (Most?) Asian languages, e.g., Mandarin, Japanese, and Korean, are such
languages.

433
 F RA N CI S J EF F RY P E L L E T I E R

3. Languages lacking both obligatory number marking and obligatory classifier sys-
tems. Various Amerindian languages, various South American languages, and
various Austronesian languages are such languages. It is not clear whether these
languages can be said to have a +mass/+count distinction, although some scholars
have argued for it.

Toyota (2009) reports that a close study of the use of mass and count terms in the five
recognized phases of the development of the English language shows that there have
been substantial changes in the counting system of English, especially in the use of
classifiers. Apparently, the earlier English did not make a distinction between mass
and count nouns, using classifiers exclusively and therefore having the nouns all be
+mass. The +mass/+count distinction eventually emerged around around 1500. Toy-
ota concludes that in fact English changed from a classifier language to the current
number-marking, non-classifier language, and as part of this change came to mark a
+mass/+count distinction within lexical nouns.
The Chinese languages and the Korean-Japanese languages are often argued as
not making a +mass/+count distinction within the lexical noun, because in these
languages no noun can directly combine with numerals. Instead, a classifier—a word
that indicates a way to “individuate” what is being discussed—is always needed. The
classifier might designate a measure, or some container, or some shape (etc.) that the
referent of the noun is to have. Many writers, e.g., Hansen (1976), Sharvy (1978),
Krifka (1995), Chierchia (1998a, b), have concluded that the referent of the noun
is therefore to be understood as “mass stuff,” waiting to be “classified” into an object
or a portion or some shape, etc. This is true for nouns such as ‘man’ as well as ones
like ‘water’. And hence, the meaning of all lexical nouns is +mass. However, the
work of Cheng and Sybesma (1999) has convinced many that the appropriate place
to look for the +mass/+count distinction in these languages is the classifier system
itself. And thus Chierchia (2010) can now write that although it is possible to view
the lexical nouns as +mass, “there is a +mass/+count distinction that is active more
generally.”
Other languages have fewer syntactic constructions that would give clues as to
whether a noun is +mass or +count. There are languages that have number marking
(e.g., singular/plural), but it is not obligatory to use it. And there are classifier-like
languages where the use of the classifiers is not obligatory. More notably, there are
languages where there in fact is no number marking or classifiers; and languages
that have no determiners (like ‘a(n)’ or ‘sm’) at all. There are languages where all
quantification is adverbial, and not over nouns: One says ‘Bears are always dangerous’
rather than ‘All bears are dangerous’. And there are even languages where one uses
the same construction to describe the result of adding two bowls of water together to
form one bowl as would be used to consider putting two monkeys in the same vicinity
to each other.
An ever-increasing literature on cross-linguistic comparisons of mass/count are
available. Good starting points are Massam (2012) and Doetjes (2012).

8 Concluding Remarks
We have seen that the +mass/+count distinction has been thought to have philosoph-
ical ramifications within metaphysics, both as to the ultimate nature of reality and to

434
 M A SS TE R MS

issues of identity and reidentification. The extension of the +mass/+count distinction

to the realm of verb phrases also yields a philosophically interesting area of investigation
into differentiation among processes, events, achievements, and so on. We have also
seen that the existence of mass nouns in natural language has challenged traditional
first-order logic as a representation language, and sparked development of mereological
and latticetheoretic approaches as representational media.
There are two standard approaches to the +mass/+count distinction: syntactic and
semantic. But there appear to be severe difficulties with making either of them into a
general theory. Furthermore, the facts surrounding how +mass/+count is manifested
in languages other than the Indo-European ones might seem to show that the basis and
rationale for making the distinction—and perhaps any philosophical consequences that
might seem to follow from the distinction—are not really valid as claims about reality
in general or about how people might conceptualize reality.
One topic not broached in this survey is the psycholinguistic work done on the
+mass/+count distinction. Most of this work has been done in English, and even
that done in other languages has mostly concentrated on numbermarking languages
with only a small amount done with classifier languages and none at all done in other
languages. Nonetheless, it seems that the results are important for a universal theory
of +mass/+count. Further philosophical studies (and further linguistic studies) of the
+mass/+count distinction in English will need to be cognizant of this work, for it
provides a touchstone of empirical reality in the way speakers employ the distinction,
and it cannot be ignored in favor of “a more beautiful theory.” There is a vast amount
of such work, only some of which has been mentioned here. See Barner and Snedecker
(2005), Markman (1985), Wisniewski et al (1996, 2003), Takatori and Schwanenflugel
(2008) and all the works mentioned in those works.

References
Bach, E. (1986a). The algebra of events, Linguistics and Philosophy 9: 5–16.
Bach, E. (1986b). Natural language metaphysics, in R. B. Marcus, G. Dorn, and P. Weingartner (eds.),
Logic, Methodology, and Philosophy of Science, VII, North Holland, Amsterdam, pp. 573–595.
Barker, C. (2003). Paraphrase is not enough, Theoretical Linguistics 29: 201–209.
Barner, D. and Snedeker, J. (2005). Quantity judgements and individuation: Evidence that mass
nouns count, Cognition 97: 41–65.
Bunt, H. (1979). Ensembles and the formal semantic properties of mass terms, in F. J. Pelletier (ed.),
Mass Terms: Some Philosophical Problems, Reidel, Dordrecht, pp. 249–277.
Bunt, H. (1985). Mass Terms and Model Theoretic Semantics, Cambridge UP, Cambridge.
Burge, T. (1972). Truth and mass terms, Journal of Philosophy 69: 263–282.
Burge, T. (1975). Mass terms, count nouns and change, Synthèse 31: 459–478. Reprinted in (Pelletier
1979, pp. 199–218).
Cheng, L. and Sybesma, R. (1999). Bare and not-so-bare nouns and the structure of NP, Linguistic
Inquiry 30: 509–542.
Chierchia, G. (1998a). Plurality of mass nouns and the notion of ‘semantic parameter’ , in S. Rothstein
(ed.), Events and Grammar, Kluwer, Dordrecht, pp. 53–103.
Chierchia, G. (1998b). Reference to kinds across languages, Natural Language Semantics 6: 339–405.
Chierchia, G. (2010). Mass nouns, vagueness and semantic variation, Synthèse 174: 99–149.
Doetjies, J. (2012) Count-mass distinctions across languages, in C. Maienborn, K. von Heusinger, and
P. Portner (eds.), Semantics, de Gruyter, Berlin.
Gillon, B. (1992). Towards a common semantics for English count and mass nouns, Linguistics and
Philosophy 15: 597–640.

435
 F RA N CI S J EF F RY P E L L E T I E R

Gillon, B. (1999). The lexical semantics of English count and mass nouns, in E. Viegas (ed.), The
Breadth and Depth of Semantic Lexicons, Kluwer, Dordrecht, pp. 19–37.
Goddard, C. (2009). A piece of cheese, a grain of sand: The semantics of mass nouns and unitizers, in
F. J. Pelletier (ed.), Kinds, Things and Stuff, Oxford UP, New York, pp. 132–165.
Goddard, C. and Wierzbicka, A. (2002). Meaning and Universal Grammar: Theory and Empirical
Findings, John Benjamins, Philadelphia.
Goddard, C. and Wierzbicka, A. (2010). The semantics of ‘abstract nouns’: Modern NSM analyses
and the thinking of John Locke and Jeremy Bentham. Submitted.
Hansen, C. (1976). Mass nouns and ‘a white horse is not a horse’, Philosophy East and West 26:
189–209.
Huddleston, R. and Pullum, G. (2002). The Cambridge Grammar of the English Language, Cambridge
UP, Cambridge.
Jackendoff, R. (2007). Conceptual semantics and natural semantic metalanguage theory have different
goals, Intercultural Pragmatics 4: 411–418.
Krifka, M. (1995). Common nouns: A contrastive analysis of English and Chinese, in G. Carlson and
F. J. Pelletier (eds.), The Generic Book, University of Chicago Press, Chicago, pp. 398–411.
Landman, F. (1991). Semantic Structures, Kluwer, Dordrecht.
Link, G. (1983). The logical analysis of plurals and mass terms: A lattice theoretical approach, in R.
Baüerle, C. Schwarze, and A. von Stechow (eds.), Meaning, Use and Interpretation, de Gruyter,
Berlin, pp. 303–323.
Markman, E. (1985). Why supordinate category terms can be mass nouns, Cognition 19: 31–53.
Massam, D. (2012). Count and Mass across Languages, Oxford UP, Oxford.
Matthewson, L. (2003). Is the meta-language really natural?, Theoretical Linguistics 29: 263–274.
McCawley, J. (1975). Lexicography and the count-mass distinction, Berkeley Linguistic Society,
Vol. 1, pp. 314–321. Reprinted in J. McCawley (ed.), Adverbs, Vowels, and Other Objects of Wonder,
University of Chicago Press, Chicago, 1979, pp. 165–173.
Moltmann, F. (1998). Part structures, integrity, and the mass–count distinction, Synthèse 116: 75–111.
Montague, R. (1973). Reply to Moravcsik, in K. Hintikka, J. Moravcsik, and P. Suppes (eds), Approaches
to Natural Language, Reidel, Dordrecht, pp. 289–294. Reprinted as the proper treatment of mass
terms in English in (Pelletier 1979, pp. 173–178).
Mourelatos, A. (1978). Events, processes and states. Linguistics and Philosophy 2: 415–434.
Moravcsik, J. (1973). Mass terms in English, in K. Hintikka, J. Moravcsik and P. Suppes (eds.),
Approaches to Natural Language, Reidel, Dordrecht, pp. 263–285.
Nicolas, D. (2010). Towards a semantics for mass expressions derived from gradable expressions,
Recherches Linguistiques de Vineennes 39: 163–198.
Ojeda, A. (1993). Linguistic Individuals, CSLI Press, Stanford.
Parsons, T. (1970). An analysis of mass and amount terms, Foundations of Language 6: 363–388.
Reprinted in (Pelletier 1979, pp. 167–171).
Pelletier, F. J. (1974). On some proposals for the semantics of mass terms, Journal of Philosophical Logic
3: 87–108.
Pelletier, F. J. (1975). Non-singular reference: Some preliminaries, Philosophia 5: 451–465. Reprinted
in (Pelletier 1979, pp. 1–14).
Pelletier, F. J. (ed.) (1979). Mass Terms: Some Philosophical Problems, Kluwer Academic Pub.,
Dordrecht.
Pelletier, F. J. and Schubert, L. (1989/2003). Mass expressions, in F. Guenthner and D. Gabbay (eds.),
Handbook of Philosophical Logic, 2nd edition. Vol. 10, Kluwer, Dordrecht, pp. 249–336. Updated
version of the 1989 version in the 1st edition of Handbook of Philosophical Logic.
Quine, W. V. (1960). Word and Object, MIT Press, Cambridge, MA.
Quirk, R., Greenbaum, S., Leech, G. and Svartvik, J. (1985). A Comprehensive Grammar of the English
Language, Longman, New York.
Riemer, N. (2006). Reductive paraphrase and meaning: A critique of Wierzbickian semantics,
Linguistics and Philosophy 29: 347–379.

436
 M A SS TE R MS

Sharvy, R. (1978). Maybe English has no count nouns: Notes on Chinese semantics, Studies in Language
2: 345–365.
Takatori, Y. and Schwanenflugel, P. (2008). Superordinate category terms and mass-count noun
status, Linguistic Anthropology 2: 199–209.
Toyota, J. (2009). When the mass was counted: English as classifier and nonclassifier language, Skase
Journal of Theoretical Linguistics 6: 118–130.
Wierzbicka, A. (1988). Oats and wheat: Mass nouns, iconicity, and human categorization, in A.
Wierzbicka (ed.), The Semantics of Grammar, John Benjamins, Amsterdam, pp. 499–560.
Wierzbicka, A. (1996). Semantics: Primes and Universals, Oxford UP, Oxford.
Wisniewski, E., Imai, M. and Casey, L. (1996). On the equivalence of superordinate concepts,
Cognition 60: 269–298.
Wisniewski, E., Lamb, C. and Middleton, E. (2003). On the conceptual basis for the count and mass
noun distinction, Language and Cognitive Processes 18: 583–624.

437
 3 . 11
IND E XI C ALS AND
DE MONS T R AT IV ES
Allyson Mount

Indexicals are as puzzling as they are common. We use them effortlessly all the time,
yet upon closer inspection their very nature is the subject of vigorous philosophical
debate. Indexicals are expressions that pick out different people, places, or things in
different contexts. They include pronouns like ‘I’, ‘she’, and ‘it’; temporal expressions
such as ‘now’ and ‘today’; locational terms like ‘here’; and adjectives such as ‘actual’ and
‘present’. Expressions like ‘nearby’ and ‘local’ are sometimes considered indexicals as
well, though it is contentious whether indexicality should be construed broadly enough
to include such expressions. Demonstratives, such as ‘this’ and ‘that’, are a subset of
indexicals, those typically accompanied by pointing or other gestures to identify the
referent. Both indexicals in general and demonstratives in particular give rise to puzzles
about linguistic meaning and reference.

Puzzles and Problems

Suppose Nadia and Leon have arranged to meet at a coffee shop at noon. Knowing
that she has a tendency to be late, Leon calls Nadia on her cell phone ten min-
utes before the appointed time and asks whether she is close to the coffee shop yet.
Nadia, who is early for once, answers, “I am here now.” In this context, ‘I’ designates
Nadia, ‘here’ designates the coffee shop, and ‘now’ designates the time of her utter-
ance (ten minutes before noon). Her utterance seems clear, unambiguous, and
straightforward.
Now suppose Nadia knows that Leon planned to get an oil change on his way to
their meeting. After chatting for a minute, she asks how things went at the garage.
Leon replies, “I am still here now. There was a long line.” In this context, ‘I’ designates
Leon, ‘here’ designates the garage, and ‘now’ designates the time of his utterance (nine
minutes before noon). Like Nadia’s statement, Leon’s reply seems clear, unambiguous,
and straightforward.
This exchange is entirely ordinary and commonplace. Yet even this simple conversa-
tion raises several important questions about indexical meaning and reference. First,
what do the two utterances of ‘I’, ‘here’, and ‘now’ have in common, given that they
designate different people, places, and times? How can a philosophical theory capture
the invariant meaning of indexicals while still respecting the fact that indexicals do, by
their very nature, pick out different things in different circumstances? I will call this the
 I N DEXI CA LS A N D DE MO N S T R AT I V E S

Puzzle of Invariant Meaning. Unlike some of the other puzzles described below, this one
has a solution that is widely (though not unanimously) accepted.
The most obvious candidates for the invariant meaning of ‘I’, ‘here’, and ‘now’ are
that ‘I’ always refers to the speaker or writer of the utterance, ‘here’ always refers to
the place of utterance, and ‘now’ always refers to the time of utterance. That seems to
explain what is going on in the exchange between Nadia and Leon, and in most other
ordinary cases. Indeed, it seems like it must be impossible to utter ‘I am here now’ falsely,
or to utter ‘I am not here now’ truly. But there is at least one situation in which it is
apparently possible to utter ‘I am not here now’ truly: on answering machine greetings.
If the straightforward rules above accurately capture the invariant meaning of these
indexicals, how can we explain this phenomenon? This puzzle is commonly called the
Answering Machine Paradox, although similar problems arise with handwritten notes,
even when no mechanical technology is involved.
With demonstratives, things become even trickier. Suppose that Susan is at an ani-
mal shelter, choosing a cat to adopt. The cats are all tangled together in a playful mass.
Susan, who has a strong preference for cats that are a certain shade of gray, says “I want
that [pointing to the light gray head of a cat] instead of that [pointing to the darker gray
hindquarters of a cat].” Unbeknownst to her, she has pointed at different parts of the
same cat, instead of at separate cats. How can we explain the fact that her two utter-
ances of ‘that’ seem to differ in significance, even though her gestures pick out the same
cat? This is the Problem of Multiple Utterances for demonstratives, and it has several
variations.
Other puzzles arise when there is a discrepancy between the speaker’s intentions and
the object her gesture actually identifies, and when complex demonstrative phrases are
used instead of bare demonstratives. If the gray cat runs away and an orange cat slips
into its place just as Susan says “I want that cat,” does ‘that cat’ refer to the gray cat as
she intended, or to the orange cat she pointed at? What if she inadvertently points to
a ferret instead of a cat? The Problem of Mistaken Gestures involves some kind of mis-
match between the gesture and the speaker’s intentions. The Puzzle of Quantificational
Uses arises when attempting to explain both cases where complex demonstratives are
apparently directly referential (as in the previous example) and cases where they seem
to function as quantifiers.
This is certainly not an exhaustive list of puzzles involving indexicals and demonstra-
tives. Notably, puzzles that are primarily about the nature of beliefs with indexical con-
tent are omitted, though beliefs and their linguistic expression are often intertwined.
(See Lewis 1979 and Perry 1979.) However, the puzzles above do highlight some of the
main challenges that philosophers of language face in providing a compelling account
of indexicality. The main strains of thought in response to these puzzles will be traced
in what follows.

Kaplan’s Theory
On one hand, all utterances of ‘I’ (or ‘today’ or ‘here’, etc.) clearly have some aspect of
meaning in common. On the other hand, different utterances of these expressions are
used to refer to different things. The Puzzle of Invariant Meaning raises fundamental
questions about the nature of indexicals precisely because these two facts are indisput-
able yet difficult to reconcile. While meaning and reference are different phenomena,
with nonindexical terms meaning generally affects reference in fairly straightforward

439
 A LLYSO N M O U N T

ways. The relationship between the meaning and reference of indexicals is more
complex, but sorting it out is a necessary first step in any theory of indexicality.
A great deal of headway has been made on this problem. Recognition that there is
a problem to sort out goes back at least to the early 20th century. In “The Thought”
(1918), Gottlob Frege comments that “If someone wants to say the same today as he
expressed yesterday using the word ‘today’, he must replace this word with ‘yesterday’.
Although the thought is the same, its verbal expression must be different so that the
sense, which would otherwise be affected by the differing times of utterance, is read-
justed” (Frege 1918: 40). Subsequent scholarship has questioned whether some of Frege’s
claims about the expressions later known as indexicals are consistent with his overall
theory of language (Perry 1977). Whether or not they are, Frege’s explicit recognition
that such expressions are more complex than they first appear certainly contributed to
a rising awareness of the issue.
The Puzzle of Invariant Meaning calls attention to the fact that a multi-faceted
approach is needed to capture how indexicals work. Like many other linguistic puzzles,
this one was initially addressed by logicians who sought to represent language systemati-
cally in order to clarify what contribution different expressions make to truth condi-
tions. Building on the work of Rudolf Carnap and others, Richard Montague (1974)
developed the idea of using an index to represent linguistic context. At the time of his
writing, much philosophical energy was being devoted to possible worlds semantics,
where a formal distinction was made between the world in which an utterance is actu-
ally made and other possible worlds in which that utterance could be evaluated. The
way context was represented varied from theory to theory, but the essential idea for
present purposes is that context could be formally represented as an index that con-
tained places for times, places, people, worlds, and whatever other factors were needed
to correspond to aspects of the natural language in question. Since English contains
tense operators, the index needs a place for time; since it contains modal operators, the
index needs a place for world; since it contains the word ‘you’, the index needs a place
for addressee, and so on.
Montague’s work paved the way for David Kaplan to expand and refine the idea of
using an index to represent linguistic context. In his highly influential paper “Demon-
stratives” (1989a, though first circulated in 1977), Kaplan develops a system of double-
indexing to capture both the invariant meaning of indexicals and the way in which their
reference shifts from context to context. The importance of double-indexing for ‘now’
was recognized earlier by Hans Kamp (1971). However, a full appreciation that Kamp’s
insights applied to indexicals more generally did not immediately follow, and it took
Kaplan’s wide-ranging and systematic development to clarify and broaden the idea.
Kaplan conceives of context as an n-tuple with a place for agent, time, location,
and world. He distinguishes between contexts of use and circumstances of evaluation.
Contexts of use are occasions on which an utterance is made. The referent of “pure”
indexicals depends on the context of use, he claims, and the referent of demonstratives
depends on the associated demonstration (Kaplan 1989a: 492). Once the referent has
been fixed, however, the proposition can then be evaluated with respect to other possi-
ble circumstances of evaluation. Kaplan emphasizes that demonstratives are directly ref-
erential. Once the referent of an indexical is established in the actual context of use, it is
the very same individual or object that is relevant to the evaluation of the proposition in
other circumstances of evaluation, rather than the individual the indexical would have
picked out if the utterance had been made in those other circumstances. As we shall see,

440
 I N DEXI CA LS A N D DE MO N S T R AT I V E S

this feature of Kaplan’s account has recently been challenged on various grounds, most
notably by Jeffrey King’s quantificational theory of complex demonstratives.
The clarification that context of use differs from circumstances of evaluation led
Kaplan to distinguish between character and content. The character of an indexical is
a rule for determining the content in each context of use. This can be represented by
a function from contexts to contents. In essence, the character captures the meaning
that remains constant. Content, on the other hand, is “what is said” (the proposition, for
assertions) by the utterance. It varies from context to context and can be represented by
a function from circumstances to extensions.
The character rule for ‘I’, for instance, states that ‘I’ is a directly referential expres-
sion that refers, in each of its utterances, to the person who utters it (Kaplan 1989a:
520). In a particular context of use, the content of ‘I’ is the individual determined by
the character rule—i.e., the speaker or writer. What distinguishes indexicals from non-
indexical expressions is that indexicals have a context-sensitive character instead of a
fixed character. Since character is determined by linguistic convention and remains
constant from use to use, Kaplan suggests that character is more aptly described as the
meaning of indexicals.
This brings us to a potential solution to the Puzzle of Invariant Meaning: Character
rules capture the aspect of meaning that remains the same from use to use, and Kap-
lan’s theory explains the relationship between the meaning and referents of indexicals.
While Kaplan’s focus is on providing a theoretical framework for indexicals and demon-
stratives rather than on specifying what the character rules are for each indexical, one
could work within his framework to do so.
The basic tenets of Kaplan’s theory have been widely accepted, but it is not without
its detractors. Most critiques are not a wholesale rejection of the view but disputes about
various details. For instance, a division between “pure” indexicals and true demonstra-
tives is central for Kaplan. Pure indexicals always refer to the relevant feature of the
context, as represented by corresponding part of the n-tuple. For these indexicals, Kap-
lan says, “[N]o associated demonstration is required, and any demonstration supplied
is either for emphasis or is irrelevant . . . The linguistic rules which govern their use
fully determine the referent for each context” (Kaplan 1989a: 491). True demonstra-
tives, in contrast, require an associated demonstration, which Kaplan says is “typically,
though not invariably, a (visual) presentation of a local object discriminated by a point-
ing” (Kaplan 1989a: 490). Other theorists have discussed Kaplan’s contrast between
pure indexicals and true demonstratives in terms of “automatic” versus “discretionary”
indexicals. Some version of this distinction has been accepted by many philosophers of
language, though the details vary and the distinction is not uncontentious (for discus-
sion, see Wettstein 1984; Recanati 2001; Perry 2001; Bach 2005; and Mount 2008).
This brings us to the next puzzle: the Answering Machine Paradox. It is a challenge
for proponents of the pure indexical view because they must either insist that utter-
ances of ‘I am not here now’ are always false despite a common intuition that it can be
uttered truly on answering machine recordings, or they must explain why some feature
of the answering machine scenario justifies a different view of indexical reference for
this particular type of case.
Kaplan explicitly states that ‘I am here now’ cannot be uttered falsely (Kaplan 1989a:
509), and he appears equally committed to the claim that ‘I am not here now’ cannot
be uttered truly. In a footnote, he acknowledges that recorded messages may introduce
additional difficulties, but he does not see this as a challenge to his distinction between

441
 A LLYSO N M O U N T

pure indexicals and demonstratives (Kaplan 1989a: 491 fn 12). Others have seen it as a
real puzzle in need of resolution, if not necessarily a true paradox. Alan Sidelle (1991)
suggests that answering machine recordings should be treated as deferred utterances,
where the utterance actually occurs at the time the message is played, rather than at the
time it is recorded. This allows him to say that ‘now’ does refer to the time of utterance
(preserving Kaplan’s character rule), yet it explains how ‘I am not here now’ can be
uttered truly, since ‘now’ refers to the time of playback.
Stefano Predelli (1998) proposes another alternative: utterances involving ‘I’,
‘here’, and ‘now’ should be interpreted with respect to the context that is intended by
the speaker, rather than with respect to the context of utterance. In many cases the
intended context of interpretation will just be the context of utterance, but the two do
not necessarily coincide. Corazza et al. (2002) explore the consequences of Predelli’s
view, arguing that it is implausible because it allows ‘I’ to refer to virtually anyone, sim-
ply because the speaker intends so. Instead, they propose that the relevant context is the
one determined by convention. Since answering machines are devices conventionally
understood to broadcast messages to be interpreted at the time of playback, it is the time
of playback that is relevant—not simply because of the speaker’s intention, but because
social conventions dictate such an interpretation.
Others have joined in the debate, responding to the Answering Machine Paradox
either by modifying Kaplan’s character rules to account for recorded messages or by
replacing some other element of his theory. Regardless of which view one favors, some
response to the puzzle is needed.

Simple Demonstratives
The criteria for what makes something a demonstration, and what happens when the
demonstration is ambiguous or various other contingencies arise, have been a matter of
continuing disagreement. The Problem of Mistaken Gestures arises when the speaker
uses a gesture to aid hearers in identifying the intended referent, but for some reason
things go astray. This could happen when the intended demonstratum suddenly moves,
or when the speaker hallucinates and there is no object in the indicated spot at all, or
when the gesture is thwarted in another way. One strain of thought is that speaker inten-
tions are private and internal, so intentions cannot play a constitutive role in what a
particular utterance of a demonstrative refers to, though they certainly are important in
what the speaker means to communicate. This type of view usually gives more weight to
publicly observable gestures whenever gestures and intentions conflict. Another strain
of thought is that speaker intentions—or at least a certain sort of intention—must figure
prominently in any theory of demonstratives. This type of view leads to questions about
whether intentions are relevant to the semantics of demonstratives or just to pragmatics
(a topic taken up in a later section). Kaplan addresses some concerns about the role of
demonstrations and intentions in “Dthat” (1978a) and “Afterthoughts” (1989b). More
recently, Kent Bach and Marga Reimer rekindled interest in the topic, discussing vari-
ous cases where speaker intentions and gestures diverge (Bach 1987 and 1992; Reimer
1991a and 1991b).
Like the Problem of Mistaken Gestures, the Problem of Multiples Utterances comes
in several varieties. One version involves co-reference, as when Susan points and utters
‘that’ twice, unaware that her gesture picks out different parts of the same cat. A similar
and more widely discussed problem of co-reference arises for proper names and is known

442
 I N DEXI CA LS A N D DE MO N S T R AT I V E S

as Frege’s Puzzle. The solution Frege (1892) advocates for proper names is to distinguish
between Sinn and Bedeutung, or sense and reference. Proper names can have different
senses even when they refer to the same person, he argues, and this explains how some-
one could have differing attitudes about a proposition depending on how it is expressed.
One might try to explain the related version of the Problem of Multiple Utterances
similarly. When Susan unknowingly points to different parts of the same cat and says “I
want that [pointing to the light gray head of a cat] instead of that [pointing to the darker
gray hindquarters of a cat],” the cognitive value of each utterance of ‘that’ (paired with
a demonstration) differs, even though they refer to the same object. However, it is far
from clear that this strategy alone can solve the problem.
A second version of the Problem of Multiple Utterances is highlighted by David
Braun (1996). Braun argues that Kaplan’s theory of demonstratives cannot adequately
account for utterances in which the same demonstrative is used multiple times to refer
to different objects. The trouble arises because Kaplan identifies the linguistic meaning
of demonstratives with their character, and character is a function from contexts to con-
tents. But if character is the linguistic meaning that remains constant from use to use, it
seems that any two occurrences of the same demonstrative in the same context should
have the same content (refer to the same object). Yet this flies in the face of the observa-
tion that sentences like “That is bigger than that” or “That [pointing to an aged book] is
older than that [pointing to an antique vase]” can be uttered truly even though nothing
is bigger or older than itself. While Kaplan does propose a formal language that treats
different occurrences of ‘that’ as associated with different demonstrations, thus having
different referents, Braun argues that Kaplan’s formal treatment avoids the problem but
implies that character alone does not capture the linguistic meaning of demonstratives.
Braun considers and rejects a modification of Kaplan’s theory that allows context
shifts in midsentence. This modification seeks to solve the problem by saying that the
context in which the first ‘that’ is uttered is relevantly different from the context in
which the second ‘that’ is uttered. In other words, the context shifts within the short
time it takes to utter the sentence, allowing each occurrence of the demonstrative to
refer a different object while still maintaining that ‘that’ has a single character that is
its linguistic meaning. Braun rejects this alternative and instead argues that demonstra-
tives have three distinct types of meaning, not two. First, there is content, as in Kap-
lan’s theory. Second, there is linguistic meaning that remains constant from context
to context. Third, there is character that a demonstrative acquires when associated
with a demonstration. The linguistic meaning of the expression, when combined with a
demonstration, yields a character. That character then works with the context to yield
a certain content or referent. This avoids Kaplan’s problem by allowing different dem-
onstrations accompanying each utterance of ‘that’ to affect its character. So while the
linguistic meaning of ‘that’ remains constant, character (and content) differs.
The Problem of Mistaken Gestures, the Problem of Multiple Utterances, and cri-
tiques such as Braun’s all make it clear that even simple demonstratives raise challeng-
ing questions. Not surprisingly, complex demonstratives—noun phrases like ‘that cat’
and ‘this green crayon’—raise additional questions about how demonstratives work.

Complex Demonstratives
A version of the Problem of Mistaken Gestures also arises for complex demonstratives.
This happens when the object demonstrated is not of the sort described by the phrase, as

443
 A LLYSO N M O U N T

when Susan utters ‘that cat’ while pointing at a ferret. In one sense, this version of
the problem is clearer because the use of a noun adds linguistic constraints, making it
extremely unappealing to think that the speaker’s intentions alone could ever carry
more weight than the observable features of the utterance. But things are complicated
by the fact that the speaker’s intentions and gestures are even more misaligned than
with simple demonstratives, raising additional questions about reference failure. Pro-
viding a plausible theory of what is going on in cases of this sort is especially difficult
because so many variations exist.
The Problem of Mistaken Gestures is certainly challenging, but the discussion is gener-
ally framed within the familiar Kaplan-style framework, though there may be considerable
disagreement about the details. The Puzzle of Quantificational Uses is based on a more
sweeping critique which, if substantiated, would undermine Kaplan’s whole approach.
It is essential to Kaplan’s view that demonstratives (and indexicals in general) are
directly referential. Failure to recognize this, Kaplan believes, is a primary reason why
previous theories were misguided. King (2001) provides data that is difficult to reconcile
with the direct reference view. He argues that the contributions that complex ‘that’
phrases make to propositions are akin to those made by quantificational noun phrases
like ‘every skier’ and ‘most swimmers’. This contrasts with the direct reference approach,
which treats such phrases as contributing individuals as the content. Hence, the core of
the dispute is whether complex ‘that’ phrases are referring expressions or quantifiers.
King finds direct reference accounts problematic in many respects. He argues that
examples involving what he calls NDNS uses (no demonstration no speaker reference),
quantification in, and narrow scope readings all favor a quantificational view. Quantifica-
tion in occurs in “Every father dreads that moment when his oldest child leaves home.”
The complex demonstrative phrase is bound by the quantifier ‘Every father’ and does not
refer to any particular individual. Sentences such as “That professor who brought in the
biggest grant in each division will be honored” are ambiguous but can be continued in a
way that makes it clear that a narrow scope reading is intended (King 2001: 10). Since
no particular individual is identified by the complex demonstrative phrase in either of
these types of examples, direct reference accounts seem ill-equipped to explain them.
NDNS uses require a bit more explanation. King notes that ‘that’ phrases can be
used without an accompanying demonstration, to talk about something that is not in
the speaker’s immediate environment, when the speaker has no particular individual in
mind. This is an NDNS use. He consider the sentence “That student who scored one
hundred on the exam is a genius,” uttered by Greg in a situation where Greg has no idea
who scored one hundred, but has been assured by his trustworthy teaching assistant that
exactly one student received a perfect score. While a direct reference theorist could
simply say ‘that student who scored one hundred on the exam’ contributes a particular
object to the proposition, such a response is less plausible when the above sentence is
embedded in a belief report. If someone else reports, “Greg believes that that student
who scored one hundred on the exam is a genius,” the reporter’s use of the ‘that’ phrase is
itself an NDNS use. Her claim is true, and King argues that the direct reference theorist
lacks the resources to explain its truth. If the complex demonstrative phrase contributes
an individual to the singular proposition, then the truth of the reporter’s claim would be
the result of Greg standing in the belief relation to that proposition. But Greg had only
general grounds for his assertion, so King argues that this analysis is strained.
On the whole, King’s quantificational view does not entail that speaker intentions
are totally irrelevant to the semantics of ‘that’ phrases. Intentions are relevant, he says,

444
 I N DEXI CA LS A N D DE MO N S T R AT I V E S

in the sense that they can determine properties that further restrict the quantification,
beyond the restriction by the descriptive material in the phrase itself (King 2001: 34).
He also acknowledges that some uses of ‘that’ phrases are rigid, while others are not. The
difference rests on whether it is the speaker’s perceptual intentions or descriptive intentions
that determine the property further restricting the quantification. However, he argues
against ambiguity approaches, which treat rigid uses as directly referential while accept-
ing a quantificational approach only for nonrigid uses. King claims that a unified quan-
tificational approach is preferable, since his account does give a principled explanation
of why some uses are rigid and other are not.
King’s view has spawned extensive discussion in recent years, leading to various mod-
ifications and clarifications by proponents of both direct reference and quantificational
theories. (See, for example, Braun 2008a and 2008b; King 2008a and 2008b; see also
Lepore and Ludwig 2000.) The Puzzle of Quantificational Uses is a puzzle of how to
reconcile examples that apparently support a direct reference view with examples that
apparently support a quantificational view, in a theoretically satisfying way. The debate
remains alive and well.

Indexicals and the Semantics-Pragmatics Distinction

No discussion of indexicals and demonstratives would be complete without mentioning
the debate over the semantics-pragmatics divide. The debate has implications for many
aspects of language, not just indexicals, and the issue is far too complex to do more than
just scratch the surface here. (See Chapter 1.2 on Semantics and Pragmatics in this
volume.) In brief, the question is whether indexicals themselves are referring expres-
sions in the semantic sense, or whether they are just used by speakers to refer in a prag-
matic sense.
In the mid-20th century, ordinary language philosophers saw reference as a property
of uses of expressions rather than of expressions themselves. Kaplan’s theory under-
mines this ordinary language view, since it implies that it does make sense to speak of
indexicals themselves referring—just relative to a context, not simpliciter. He shows
that it is possible to give a logic of demonstratives (1978b). We have already seen some
of the challenges to Kaplan’s theory raised by Braun, King, and others. Behind the
scenes in all of this lurk questions about what role context plays. Does it simply provide
parameters that can be incorporated into meaning rules of indexicals directly? Should
we conceive of context as including demonstrations, and if so, what is the best way to
represent them? Are speaker intentions part of context in the semantic sense, or are
things like beliefs and intentions extra-linguistic factors that are certainly relevant to
communication but not to semantics? Should we abandon talk of indexicals referring in
favor of saying that indexical reference is just speaker’s reference?
John Perry (2001) provides a theory of indexicals and proper names that addresses
some of these foundational issues. He compares referentialist and descriptivist views of
indexicals, ultimately drawing insights from both to support his own referential-reflex-
ive view.
The referentialist view of indexicals holds that indexicals designate objects, and
statements containing indexicals express propositions about those objects (Perry 2001:
7). A statement containing ‘I’, for instance, expresses the same proposition as a similar
statement where ‘I’ is replaced with the name of the speaker, since both designate the
same person. This view seems to get counterfactual truth conditions right; in evaluating

445
 A LLYSO N M O U N T

whether the proposition is true at other possible worlds where the descriptions associ-
ated with the person vary, it still seems like that person (the speaker who uttered ‘I’ in
the actual world) is the one relevant to evaluation of the proposition, even if she is not
speaking in the other world. However, referentialists have difficulty explaining how
co-referring terms can differ in cognitive significance, as Frege recognized long ago. If
the listener does not know he is speaking with Michelle Obama, he might believe the
proposition expressed by “Michelle Obama is Barak Obama’s wife” without believing
the proposition expressed by “I am Barak Obama’s wife” [uttered by Michelle Obama].
Descriptivist views of indexicals come in many variations, but the basic idea is that
“statements involving indexicals express propositions that incorporate identifying con-
ditions . . . that derive from the rules of the language” (Perry 2001: 9). Hans Reichen-
bach’s theory of token-reflexives took this approach (Reichenbach 1947). Reichenbach
suggests that what indexicals contribute to a proposition is a reflexive content about
that token of the indexical itself. For instance, “I prefer tangerines to oranges” really
means something like “The speaker of this token of ‘I’ prefers tangerines to oranges.”
Supporting this view is the intuition that competent speakers of English understand
the reflexive content of ‘I’ even when they do not know who the speaker is. The same
approach can be used for other indexicals and demonstratives, with ‘here’ meaning the
place where this token of ‘here’ is spoken, and so on.
Building on earlier work by Arthur Burks (1949), Perry takes insights from both refer-
entialist and descriptivist views and develops what he calls the referential-reflexive the-
ory. According to Perry, both reflexive contents and referential contents are associated
with indexicals. Reflexive content explains what is going on in co-reference cases. “I am
Barak Obama’s wife” and “Michelle Obama is Barak Obama’s wife” [uttered by Michelle
Obama] differ in their reflexive contents even though ‘I’ and ‘Michelle Obama’ refer to
the same person. Thus, a token-reflexive view of indexicals seems promising. However,
as Perry acknowledges, it is odd to think that when we use indexicals we are making
statements about the words themselves. In most ordinary cases, we aren’t—we are talk-
ing about the actual subject matter, not about words. That is where referential content
comes in, contributing to what is officially said by the utterance (Perry 2001: 12–13).
Perry distinguishes between automatic and discretionary indexicals in a now-familiar
way, and also between narrow and wide context. He says, “narrow context consists of
facts about which things occupy the essential roles involved in the utterance, which
I will take to be the agent, time, and position. These roles are filled with every utter-
ance . . . The wide context consists of those facts, plus anything else that might be rel-
evant, according to the workings of a particular indexical” (Perry 2001: 59). According
to Perry, some indexicals (such as ‘I’) rely only on narrow context, while others rely
on wide context. This type of distinction between narrow and wide context has been
used by others as well, often in conjunction with a distinction between pure/automatic
indexicals and discretionary indexicals/true demonstratives. It allows one to say that
narrow context is the type of context relevant to semantics, while still acknowledging
the common practice of using the word ‘context’ in a broader sense to include speaker
intentions, beliefs, and other factors.
Space does not permit a full discussion of Perry’s theory here. (See O’Rourke and Wash-
ington 2007 for commentary.) The key point for present purposes is that the issue of how
to conceive of context is often tied to the issue of how to draw the line between seman-
tics and pragmatics, and this in turn affects one’s view of indexicals. Bach (2007) has
questioned Perry’s conception, focusing on two of Perry’s assumptions: that utterances

446
 I N DEXI CA LS A N D DE MO N S T R AT I V E S

(rather than sentences) have semantic contents, and that indexicals and demonstra-
tives are inherently referential. Pointing out apparent inconsistencies in Perry’s view, he
charges Perry with blurring the line between semantic “official content” and speaker
meaning. Perry accepts Kaplan’s view that speakers’ directing intentions play a part in
the reference of discretionary indexicals. But if that is true, Bach argues, then intentions
do so pragmatically. Kaplan’s view only works for automatic indexicals, which depend
only on the contextual parameters included in his n-tuple. Speaker intentions are not
part of context. Bach concludes that “the meanings of discretionary indexicals are not
utterance-reflexive” (Bach 2007: 408), since discretionary intentions are communica-
tive intentions, and communicative intentions only affect pragmatics, not semantics.
The issue of where to draw the line between semantics and pragmatics is directly
related to the Problem of Mistaken Gestures as well. Appealing to a distinction between
narrow and wide context solves some problems while exacerbating others. If neither ges-
tures nor intentions are included in narrow context, that may make it more difficult to
explain the semantic role of demonstratives. If gestures but not intentions are included
in narrow context, that paves the way for treating gestures as taking precedence when
they conflict with intentions in simple demonstrative utterances. However, it further
confounds attempts to clarify the relationship between complex demonstratives and
gestures, since a speaker could use a complex demonstrative phrase while pointing at an
object that is not of the sort indicated (for example, pointing to a ferret while uttering
‘that cat’).

Conclusion
The puzzles we began with remain unresolved, but over the last few decades progress has
been made on many fronts. Kaplan-style character rules are leading candidates to address
the Puzzle of Invariant Meaning, although modifications may be needed to explain the
Answering Machine Paradox and other unusual cases. Quantificational accounts of
complex demonstratives provide another sort of explanation altogether. The jury is still
out on which approach better accounts for the data. However, at least it is now clear
that the examples involved in the Puzzle of Quantificational Uses require explanation,
in addition to the more familiar cases where a direct reference account seems promising.
This insight sharpens the focus on exactly what remains to be explained.
The Problem of Multiple Utterances highlights the fact that the relationship
between the meaning and reference of indexicals remains unclear. How this debate
is eventually resolved has implications for other areas of philosophy as well, espe-
cially for philosophy of mind and epistemology. Likewise, the Problem of Mistaken
Gestures raises difficult questions about the role of gestures and speaker intentions.
Aside from being interesting in its own right, this issue is directly related to the ongo-
ing debate about how best to understand linguistic context, and where to draw the
line between semantics and pragmatics. With all of these puzzles remaining, it is no
surprise that philosophers of language continue to explore many facets of indexicals
and demonstratives.

Related Topics
1.1 Extensions, Intensions, Characters, and Beyond
1.2 Semantics and Pragmatics

447
 A LLYSO N M O U N T

1.10 Context Sensitivity

2.1 Reference
5.9 Montague Grammar.

References
Bach, K. (1987) Thought and Reference, New York: Oxford University Press.
——(1992) “Intentions and Demonstrations.” Analysis, 52(3): 140–146.
——(2005) “Context ex machine,” in Zoltán Gendler Szabó (ed.), Semantics versus Pragmatics, New York:
Oxford University Press.
——(2007) “Reflections on Reference and Reflexivity,” in M. O’Rourke and C. Washington (eds.), Situating
Semantics, Cambridge, MA: MIT Press.
Braun, D. (1996) “Demonstratives and Their Linguistic Meanings.” Noûs, 30: 145–173.
——(2008a) “Problems for a Quantificational Theory of Complex Demonstratives.” Philosophical Studies,
140 (3): 335–358.
——(2008b) “Persisting Problems for a Quantificational Theory of Complex Demonstratives.” Philosophical
Studies, 141 (2): 243–262.
Burks, A. (1949) “Icon, Index, and Symbol.” Philosophical and Phenomenological Research, IX: 673–689.
Corazza, E., Fish, W. and Gorvett, J. (2002) “Who is I.” Philosophical Studies, 107 (1): 1–21.
Frege, G. 1892. “On Sense and Nominatum,” in A.P. Martinich (ed.), The Philosophy of Language, 5th edn,
New York: Oxford University Press.
——(1918) “The Thought,” in A.P. Martinich (ed.), The Philosophy of Language, 5th edn, New York: Oxford
University Press.
Kamp, H. (1971) “Formal Properties of ‘Now’.” Theoria, 37 (3): 227–273.
Kaplan, D. (1978a) “Dthat,” in A.P. Martinich (ed.), The Philosophy of Language, 5th edn, New York: Oxford
University Press.
——(1978b) “On the Logic of Demonstratives,” in A.P. Martinich (ed.), The Philosophy of Language, 5th
edn, New York: Oxford University Press.
——(1989a) “Demonstratives,” in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan, Oxford:
Oxford University Press.
——(1989b) “Afterthoughts,” in Matthew Davidson (ed.), On Sense and Direct Reference: Readings in the
Philosophy of Language, Boston: McGraw Hill.
King, J.C. (2001) Complex Demonstratives: A Quantificational Account, Cambridge, MA: MIT Press.
——(2008a) “Complex Demonstratives, QI Uses, and Direct Reference.” Philosophical Review, 117 (1):
99–117.
——(2008b) “Complex Demonstratives as Quantifiers: Objections and Replies.” Philosophical Studies, 141
(2): 209–242.
Lepore, E. and Ludwig, K. (2000) “The Sementics and Pragmatics of Complex Demonstratives.” Mind, 109
(434): 199–240.
Lewis, D. (1979) “Attitudes De Dicto and De Se.” Philosophical Review, 88 (4): 513–543.
Montague, R. (1974) Formal Philosophy, New Haven, CT: Yale University Press.
Mount, A. (2008) “The Impurity of ‘Pure’ Indexicals.” Philosophical Studies, 138 (2): 193–209.
O’Rourke, M. and Washington, C. (eds.) (2007) Situating Semantics: Essays on the Philosophy of John Perry,
Cambridge, MA: MIT Press.
Perry, J. (1977) “Frege on Demonstratives.” Philosophical Review, 86 (4): 474–497.
——(1979) “The Problem of the Essential Indexical.” Noûs, 13: 3–20.
——(2001) Reference and Reflexivity, Stanford: CSLI Publications.
Predelli, S. (1998) “I Am Not Here Now.” Analysis, 58(2): 107–115.
Recanati, F. (2001) “Are ‘Here’ and ‘Now’ Indexicals?” Texte, 127: 115–127.
Reimer, M. (1991a) “Do Demonstrations Have Semantic Significance?” Analysis, 51(4): 177–183.
——(1991b) “Demonstratives, Demonstrations, and Demonstrata.” Philosophical Studies, 63: 187–202.
Reichenbach, H. (1947) Elements of Symbolic Logic, New York: The Free Press.
Sidelle, A. (1991) “The Answering Machine Paradox.” Canadian Journal of Philosophy, 21: 525–539.
Wettstein, H. (1984) “How to Bridge the Gap Between Meaning and Reference.” Synthese, 58: 63–84.

448
 3 . 12
I NDICAT IV E
C OND ITI ONALS
Anthony S. Gillies

1 The Target
Sometimes philosophy gets started by pointing. Setting out the target for theories of
indicative conditionals is a case in point. So some examples to point at:

(1) a. If the gardener didn’t do it, (then) it was the butler.

b. If your blue marble is in the box, then your red one might be under the couch.

Our target is the class of ordinary if . . . then . . . constructions like these. They express
conditional information: information about what is or might or must be, if such-and-
such is or turns out to be the case.
Indicatives (the name is not great but is entrenched) stand apart from two other sorts
of conditional constructions. First, from subjunctive or counter factual conditionals.
An Adams-pair:

(2) a. If Oswald didn’t kill Kennedy, then someone else did.

b. If Oswald hadn’t killed Kennedy, someone else would have.

The conditional information these traffic in is different: (2a) is an indicative and true;
it’s counterfactual cousin (2b) is false. Subjunctive conditionals—that is, those with a
distinctive tense/aspect marking like (2b)—say something about would or might have
been if such-and-such had been.
(The “subjunctive” marking doesn’t go exactly with counterfactuality. Indeed, it’s
neither necessary (as Anderson (1951)-type examples show) nor sufficient (as sport-
scasterese-type conditionals show). See von Fintel 1998 (and the references therein) for
the status of the connection between “subjunctive” marking and counterfactuality.)
Second, indicatives stand apart from biscuit conditionals. Austin’s (1956) example:

(3) There are biscuits on the sideboard if you want them.

A fine thing to say, but not normally a thing that expresses conditional informatinon.
While it is open to connect a story about indicatives and counterfactuals, and while it
is open to connect a story about indicatives and biscuit conditionals, we will be setting
such outreach aspirations aside. Having pointed a few times, let’s get going.
 A N TH O N Y S. G I L L I E S

2 Landscape
Saying that indicatives carry conditional information isn’t yet to say very much. We a
want a better grip on:
i. Just what that conditional information is. This is a story about what indicatives in
natural language mean (what semantic values they have) and how they are used in
well-run conversation (what their pragmatic profile looks like).
ii. Just how sentences of natural language express those meanings. This is a story about
how if interacts with hte rest of out language (that holds good both for composition-
ally deriving how ifs express conditional information and for seeing how conditional
sentences contribute their conditional meaning to embedding environments).
Suppose that semantic values (whatever they are) determine truth values at points of
evaluation (worlds,situations, whatever). This holds, let’s assume, for borth iffy sen-
tences. Not wholly innocent—as we’ll see—but a place to start.
Let’s distinguish between (i) conditional sentences of natural laguage and (ii) condi-
tional connectives of some formal languages that serves to represent the logical forms of
conditional sentences. The aim is then to associate an if-in-English (via well-behaved
mapping that we won’t bother with) with an if-in-the formal-language that then is
associated (via a well-behaved mapping we will bother with) with its semantic value.
(There are ways to express conditional information in natural language without resort-
ing to ifs (in some languages, it’s the only way). The issues in this particular ballpark
won’t matter too much for our purposes so we can (pretty) safely focus on the if . . .
then of English.) For simplicity we’ll stick with a simple propositional language (with
sentential connectives ™, š,›,Š) together with a connective (if ·)(·) for the indica-
tive. (Whether indicatives can be represented by a binary conditional connective in a
regimented intermediate language is also, as we’ll see, uo for grabs.) This indirect route
to assigning meanings to indicatives isn’t required, is in principle dispensable (assuming
our mappings are well-behaved), but makes for an unobstructed view.
One way to make sense out of the idea of conditional information is to let some ver-
sion of the Ramsey Test guide us. Here’s what Ramsey (1929/1990: 155) says: “If two
people are arguing ‘If p, will q?’ and are both in doubt as to p, they are adding p hypo-
thetically to their stock of knowledge and arguing on that basis about q. . . .” The adding
this involves can’t be feigned or hypothetical belief. (Thomason is credited with this
observation in van Fraassen 1980.) If it were, the Ramsey Test would go wrong:
(4) If my students are cheating in class, then I will not discover it (because they’re so
clever).
The state I get into by feigning belief in the antecedent won’t be one in which I believe
the consequent. The right way to understand argumenting for the Ramsey test is
restricting a body of information by the information carried by the antecedent. A more
neutral version of the Ramsey Test Skirts this: the conditional information carried by
(if P) (Q) is true/accurate/acceptable in a situation iff Q is true/accurate/acceptable in
that situatuion-plus-the-information-that-P.
Theories of indicatives are constrained by the patterns of intuitive entailment they
participate in. So what-intuitively-entails-what is important. We want to explain those
patterns as best we can. Consider whether indicatives generally go in for a deduction
theorem (a.k.a. conditional proof):

450
 I N DI CATI VE CO ND I T I O N A L S

(5) *Š{P}
= Q iff *
= (if P)(Q)

If so, then we have an especially tight connection between what (if ˜)(˜) means and
what we say about entailment. the point is that ‘entailment’ is as much part of the theo-
retical machinery as is anything. Take it as given that what we may say about if’s can
interact in non-trivial ways with what we say about entailment. Then the best route to
the best theory may be one in which what we say about entailment bends to the will of
the if’s as much as the other way around.
A concrete example: what Stalnaker (1975) calls the direct argument.

(6) a. Either the butler did it or the gardener did it.

P›Q
b. So: if the butler didn’t do it, then the gardener did.
(if ™P)(Q)

Seems like (6a) entails (6b). The trouble is that treating such or-to-if arguments as
entailments seems to get us quickly to the conclusion that indicatives mean just what
their corresponding material conditionals do. Suppose semantic values are something
that determine truth values and that entailment is preservation of truth. And assume
the direct argument is an entailment (that P › Q
= (if ™P)(Q)). Then P Š Q entails
the indicative (if P)(Q).

Argument. Suppose P Š Q is true (at a world w). Then so is ™P › Q. Thus by the direct
argument so is (if P) (Q) (at w). (The entailment in the other direction isn’t usually
disputed.) …

We can follow where the argument leads or we can look for places to get off the boat.
Another example: a boundedness argument (Gibbard 1981; Veltman 1985). Indica-
tives seem to fall somewhere on the spectrum of logical strength between strict implica-
tion and material implication: when P entails Q the indicative (if P)(Q) must be true
and the falsity of P Š Q entails the falsity of the indicative (if P)(Q). That last bit—that
indicatives are bounded from below by material conditionals—is equivalent to saying
that they go in for modus ponens. And then there is import/export:

(7) a. If the gardener is away and the driver is away, then the mansion is empty.
(if (P š Q))(R)
b. If the gardener is away, then if the driver is away then the mansion is empty.
(if P) ((if Q)(R))

Pairs like these seem to be mutual entailers. Again it seems that indicative conditionals
and material conditionals say the same thing, that the truth of P Š Q is sufficient for the
truth of the indicative (if P)(Q).

Argument. Suppose ™P is true. Since (™P š P) entails Q and indicatives are true when
their antecedents entail their consequents, we have that

(8) (if (™P š P))(Q)

is true. By import/export

451
 A N TH O N Y S. G I L L I E S

(9) (if ™ P)((if P)(Q))

is true. And since ™P is true and material conditionals are the lower bound, the mate-
rial conditional (if P)(Q) is true, too. Now suppose Q is true. The argument goes just as
before, except now we rely on the fact that Q š P entails Q. So if ™P › Q is true so is (if
P)(Q). …
As before, we either follow the argument to the material conditional or we look for
escape routes. That is a tidy way of organizing our choices.
One escape route—flat-footedly denying modus ponens—isn’t very popular. Though,
as things go in philosophy, that’s not because no one has pushed for it. McGee (1985)
sees boundedness arguments as pitting modus ponens against import/export. He thinks
there are counterexamples to modus ponens and not to import/export, so that forces his
choice. (The counterexamples all involve a right-nested indicative we think true (and
whose antecedent we think true) even though we do not especially think the embedded
conditional on its own is true. There is a not-small literature on the status of the coun-
terexamples.) Others deny some mixture of modus ponens and import/export, but not
obviously for reasons connected to boundedness-style arguments (Lycan (2001) denies
both import/export and (for good measure) modus ponens, but not without cost).

3 Horseshoe
The horseshoe theory is both the simplest response to arguments like the boundedness
argument and the simplest story about indicatives: (if ·)(·) has just the truth conditions
that Š does.

Horseshoe Theory Indicatives are material conditionals:

(if P)(Q)c,w = 1 iff either P is false or Q is true (at w, in c).

This obviously treats indicatives as truth-functional.

Assuming truth-functionality, is there any other way of assigning truth conditions?
No.

Argument. This

(10) If Jimbo is taller than 6 feet, then Jimbo is taller than 5 feet. (if P)(Q)

should be true (at w) no matter what. Possible heights for Jimbo: (i) over 6 feet (ante-
cedent true), (ii) between 5 and 6 feet (antecedent false, consequent true), and (iii) less
than 5 feet (antecedent and consequent false). These are the conditions under which
P ŠQ is true. So there’s one indicative true when its corresponding horseshoe is. Assum-
ing truth-functionality, this has to be true for any indicative. Moreover, indicatives
must be false when their corresponding horseshoes are false: otherwise, all indicatives
would always be true. So the horseshoe is the only choice. …
The horseshoe theory is, gently put, not the most widely held view these days. But
it’s not as though it has nothing going for it. Indicatives certainly behave like material
conditionals in mathematical contexts. And the horseshoe theory says that instances
of the direct argument, import/export, and modus ponens—an impressive who’s who in
properties conditionals seem to have—strike us as entailments because they are.

452
 I N DI CATI VE CO ND I T I O N A L S

It’s easy for the horseshoe theory to claim entailments (to conditionals). That is
because the material conditional is so weak. The problem is—for just that reason—we
have more entailments (to conditionals) than we want. (In)famously among them: the
paradoxes of material implication.

(11) a. Carl came alone.

??So: if Carl came with Lenny, neither came.
b. Billy got here first.
??So: if Alex got here before Billy, Billy got here first.

These don’t seem like entailments even though the truth of either ¬P or Q at w secures
the truth of P ŠQ at w. The horseshoe champion owes us some answers.
What we need is an extra-semantic explanation saying that these are entailments, all
right, but ones we can live with because the pragmatic facts about conditionals—how
they are reasonably and appropriately used—explains their weirdness. That is Grice’s
(1975) strategy. Conditionals like the one in (11a) are true but ruled out on general
grounds. Once we know Carl came alone, it is pointless (though true) to say that either
he didn’t come with Lenny or neither came. Since it is pointless, as Lewis (1976: p.142)
says, “also it is worse than pointless: it is misleading.” Ditto for (11b): if you are in a
position to say Q then you shouldn’t be in the business of saying the weaker condition-
alized thing. This kind of explaining away is principled. It doesn’t rely on anything
special about conditionals: once we take on board the Gricean picture of conversa-
tional implicature, we can say something about why the examples in (11) are odd even
though the underlying entailments are real. (Lewis (1976) initially goes for this defense
of the horseshoe theory for indicatives, using (part of) it to explain why assertability of
indicatives goes with their conditional probability and not with the probability of their
truth.)
Another kind of explanation: say that indicatives conventionally implicate something
extra. Jackson (1991) develops this defense, saying that they require “robustness”: when
you issue an indicative (if P)(Q) it conventionally implicates that your credence in? P Š
Q conditional on P—which is to say that the probability of Q conditional on P—is high
enough. (Lewis eventually drops the conversational story and instead goes for a variant
of Jackson’s conventional implicature story.) So if is supposed to be like but: truth-
functional in meaning with an extra, conventionally encoded signal. The problem
with, for instance, (11a) is that our reason for thinking the conditional is true is that we
think the antecedent false—that’s a paradigm case where robustness fails (we drop the
conditional upon learning that—to our surprise—Carl came with Lenny). Similarly for
(11b). There is little independent evidence of this conventional implicature however
(see Bennett 2003: §16 and Edgington 2009: §4.2).
The implicature explanations differ (in mechanism and commitment) but have a lot
in common. Both say that it is some extra-semantic fact about using indicatives that
explains away the oddness of the entailments in (11). Both say that when it comes to
indicatives it’s assertability preservation (not some notion of entailment tied to the
semantic values of indicatives) that matters and that judgments about entailment may
well be clouded by judgments about assertability preservation.
But both also have difficulty explaining facts about indicatives occurring unasserted
in embedded environments. The mechanisms for the assert-the-stronger conversational
implicature explanation don’t apply. And we have no story about how a special-purpose

453
 A N TH O N Y S. G I L L I E S

robustness conventional implicature projects out from under embedding constructions.

That is too bad because conditionals do occur in larger environments and the horseshoe
theory inherits some pretty bad commitments when they do. Everyone knows it was
either the gardener, the driver, or the butler. So when the novice on the beat declares
that if the gardener’s alibi checks out then you can arrest the butler, you rightly object:

(12) a. It’s not so that if the gardener didn’t do it then the butler did.
b. Just because the gardener didn’t do it that doesn’t mean the butler did.

In (12a) we have an ordinary indicative under a wide scope sentence negation. (This
may be inelegant English, but it’s still English.) What you say doesn’t entail that the
gardener didn’t do it. (And it doesn’t entail that the butler didn’t do it.) This line of
reasoning is not good:

(13) Since the driver is still a suspect, it’s not so that if the gardener didn’t do it then
the butler did.
??So: it wasn’t the gardener.

The point is that you can be all signed-up for ™ (if P)(Q) without being signed-up for
the truth of P. But if ™(P Š Q) is true then so is P. This is the price material conditionals
pay for being so weak: their negations are strong. So we need to explain away this entail-
ment. But the implicature gymnastics invoked thus far don’t seem up to it. Maybe some
other pragmatic mechanism is at work—perhaps the apparent widescope negation isn’t
a regular negation at all but is really a denial or metalinguistic negation operator (Horn
2001): when you say (13a) you are not asserting the negation of a conditional but you
are denying that you will assert the embedded conditional. (This is Grice’s reply. Horn
seems happy enough to follow him in part because the negation is so awkward—a test,
he says, for metalinguistic negation. Merely being resistant to embedding under nega-
tion lumps things wrong, though: some modals don’t like it (may and must) even though
others don’t mind at all (can and have to). But may and can (and must/have to) express
the same thing and that thing (whatever it is) is a thing we can sign-up for negating.)
This seems dodgy: we’re now wheeling in another explaining-away to save our original
explaining-away. And the new explaining-away won’t help once we embed the (appar-
ent) negated conditional in an even larger environment:

(14) If there is no God, then it’s not so that if I pray my prayers will be answered.

So the implicature defense needs work, including saying how the speech act of denial com-
positionally mixes with the assignment of semantic values for the rest of the language.

4 Variably Strict Conditionals

Assume that indicatives do not have the truth conditions of material conditionals.
They say more. What can that more be? Stalnaker (1975) argues that indicatives share
a core semantics with counterfactuals. Both kinds of conditionals, he says, are variably
strict conditionals, but indicatives are governed by additional pragmatic mechanisms
that counterfactuals aren’t subject to. The variably strict semantics allows indicatives
to say more than their horseshoes, and the pragmatic mechanisms explain why certain
patterns that aren’t entailments nevertheless seem so compelling.

454
 I N DI CATI VE CO ND I T I O N A L S

Start with orderings over the set of possibilities: for each w, assume there is a (con-
nected, transitive) relation dw recording relative similarity or closeness between worlds.
There are two points of interaction between contexts and the ordering: (i) what order-
ing is relevant is contextually determined and (ii) Stalnaker posits a substantive prag-
matic constraint linking contextually relevant possibilities and the ordering. Let’s sim-
ply model contextually relevant information determined by contexts as functions from
worlds to sets of compatible worlds.

(15) a. If there is sugar in the coffee, it tastes sweet.

(if P)(Q)
b. #So: If there is sugar and diesel oil in the coffee, it tastes sweet.
(if (P^ R))(Q)
(16) a. If it rains, it won’t pour.
(if P)(™Q)
b. #So: If it pours, it won’t rain.
(if Q)(™P)

The nearest P-worlds need not include the nearest (P^R)-worlds, so (15a) won’t entail
(15b). Similarly, the nearest rainy worlds can all be drizzly worlds,in which case (16a) is
true. Still, the nearest pouring-worlds will be rainyworlds and so (16b) will be false.
Since indicatives are stronger than horseshoes, the direct argument isalso classified
as a non-entailment:

Counterexample. Suppose w is a (™P^Q)-world and v is a (P^ ™Q)-world and that v is

the closest P-world to w. …
Why are instances of it—like (6)—so compelling? Stalnaker’s (1975)answer is that it
is a reasonable inference. Put it this way:

Reasonable Inference Suppose P is successfully asserted (at w in c) and c’ is the resulting

posterior context. P, so : Q is a reasonable inference iff Q is accepted in c’.
This is not a semantic property. It depends on what speakers are up to in asserting
various things. In order to be successfully asserted, a premise has to first be felicitously
asserted. In order for a conclusion to be accepted in a context, it has to get a clean bill
of pragmatic health. Since or-to-if begins with a disjunctive premise and ends with a
conditional conclusion, we need two additional pieces of information. Begin with the
disjunctive premise: in order for a disjunction to be felicitously asserted in a context
either disjunct might be true without the other. That seems plausible.
The constraint on indicatives is more substantive. In order to use an indicative condi-
tional (owing perhaps to their epistemic connection) there is an additional requirement
that the selected antecedent world(s) must be compatible with the context, assuming
the antecedent is:

Variably Strict Semantics Indicatives are variably strict conditionals:

Assume for any worlds w, v and proposition X, that dw is such that:

i. (Centering) w is minimal in dw: if v dw, w then w = v

ii. (Limit) X has at least one dw-minimal world
iii. (Uniqueness) X has no more than one dw-minimal world
Then (if P)(Q)c,w = 1 iff Q is true at the dw-minimal P-world.

455
 A N TH O N Y S. G I L L I E S

This is Stalnaker’s (1968) set-up. (Lewis’s (1973) version differs by making neither the
Limit Assumption nor the Uniqueness Assumption. The pragmatic mechanisms for
indicatives that Stalnaker develops are independent of whether the core semantics is
Stalnakerian or Lewisian. I’ll make the Limit Assumption and be willfully sloppy about
whether what’s required is the closest antecedent world or the set of closest antecedent
worlds.)
This predicts that indicatives say more than their corresponding horseshoes. The
paradoxes of material implication won’t be reproduced: the inferences in (11) are
non-entailments for a variably strict conditional. And while the material conditional
validates antecedent strengthening and contraposition, variably strict conditionals
don’t:

Pragmatic Constraint If P is compatible with the context c, then the dw-minimal P-

world(s) is compatible with c.
This is motivated this way. First: indicatives are happiest being asserted when their
antecedents might, in view of the context, be. If you want to say something condi-
tional on P in a context in which it is settled that ™P, then the counterfactual is
what you need to reach for. Second: in asserting indicatives, you are trying to say
something about the possibilities compatible with the context. So the worlds relevant
to whether an indicative is true had better be compatible with the context. This plus
the plausible requirement for disjunctions guarantees that the direct argument is a
reasonable inference:

Argument. Suppose an assertion of (6a) is successful (felicitous and accepted) at w in a

context c. So c(w) contains some (P š ™Q)-worlds and some (™P š Q)-worlds. Since
the disjunction is then accepted, we eliminate worlds from c(w) where it isn’t true. In
the posterior cc(w), there are only (P › Q)-worlds. Some are (P š ™ Q)-worlds and
some are (™P š Q)-worlds. Now consider the indicative (6b): (if ™P)(Q). Its anteced-
ent is compatible with cc (there are ™P-worlds in cc(w)). Thus the pragmatic constraint
ensures that the (dw-)closest ™P-world(s) are also in cc (w). But there aren’t any (™P š
™Q)-worlds in cc (w). So the closest ™P-worlds will be Q-worlds. So the indicative is
accepted in cc. …

That’s why, according to Stalnaker, or-to-if instances—though invalid—are so

compelling.
A variably strict semantics also fails to classify pairs of conditionals like (7) as mutual
entailments: whatever the merits of import/export, given the variably strict semantics,
it ain’t a validity. That’s because the nearest Q-world to the nearest P-world to w need
not be the same as the nearest (P š Q)-world to w, even when both P and Q are com-
patible with the context. In the case of the direct argument, there is an independently
plausible felicity-requirement on the disjunctive premise—that in the prior context
either disjunct may be true without the other—that does real work in making sure the
posterior context will be one in which the indicative is accepted. For import/export it’s
not obvious what such an extra, independently plausible felicity-requirement on the
conditional premises (in either direction) would look like. And so it’s not obvious how
to use the concept of reasonable inference to explain why import/export instances—
though invalid—are so compelling.

456
 I N DI CATI VE CO ND I T I O N A L S

5 NTV and Conditional Assertion

The boundedness argument can be seen as an argument that there is no good way of
assigning truth conditions to indicatives: the only options come up short either by giving
up something important or by saddling us with the horseshoe theory. Gibbard (1981)
sees it that way and sees it as reason conclude, with Adams (1975), that indicative con-
ditionals don’t have truth conditions at all: they do not report conditional information
but are a means of expressing conditional belief. Edgington (1995, 2009) and Bennett
(2003) (among others) follow them to this “N(o)T(ruth)V(alue)” conclusion. (This
view sounds like it denies that indicatives have truth values. In that it does not disap-
point. But the spirit is to deny something more. We’ve assumed that conditionals have
the same sort of semantic value that other bits of declarative language have. NTVers
deny that indicative conditionals have or traffic in the same kind of semantic unit of
exchange that other declarative parts of the language do: it’s not truth and it’s not some
richer-than-than-truth thing either.)
Another argument against truth conditions, this one due to Edgington (2009: §2):
truth conditions come in two flavors (truth-functional and non-truth-functional) and
neither seems right for indicatives. Truth-functional truth conditions are out because
the horseshoe is out. That leaves non-truth-functional ones. Saying that the truth
conditions for (if P)(Q) are non-truth-functional means that the truth values of P and Q
at w don’t fix the truth value of (if P)(Q) at w. So if P is false at w and Q is true at w then
sometimes (if P)(Q) is true at w and sometimes it isn’t. That’s variability. But we have
the or-to-if inference: if all the information you have is that ™P › Q then that is always
sufficient for (if P)(Q). This is uniformity. Edgington says variability and uniformity are
on a collision course and no non-truth-functional theory can be squared with this. If
all that is right, we are out of options truth conditions-wise. (I myself think this isn’t all
right, since for all that’s been said it’s possible that the variability be variability in truth
value at a given world between contexts and the uniformity be uniformity across worlds
compatible with a given context (Gillies 2009: §4).)
A third argument is based on hedging. Lewis (1976, 1986) showed that on pain
of triviality there are no conditional propositions such that their probability of truth
always equals the conditional probabilities of their consequents given their anteced-
ents. (The basic result has been fortified and extended: see, e.g., Hájek (1994). Bennett
(2003: §25–31) surveys some of the main territory.) There is a connection that needs
explaining:

(17) a. Probably, if my team doesn’t sign some big players they won’t win.
b. If my team doesn’t sign some big players they won’t win.

The hedged (17a) seems to say that the probability that my team will continue their
losing ways, conditional on them continuing their frugal/mismanaging ways, is high
enough. I don’t think that they definitely won’t win if they don’t make some big
moves, but I think that’s likely (enough). So the degree to which I am signed up for
the bare (17b) seems to track what (17a) says. Suppose B is a probability function rep-
resenting degrees of belief or assertability or whatever. That’s some reason to sign-up
for this:

The Equation B((if P)(Q)) = B-plus-P(Q)

457
 A N TH O N Y S. G I L L I E S

We haven’t yet said anything about how we arrive at this posterior probability B-plus-
P(Q). Maybe it’s through conditionalization, maybe not. The Equation implies:

Linearity If B is a mixture of B1 and B2 then B-plus-P is a mixture of B1-plus-P and

B2-plus-P.

Proof. Assume B is a mixture of B1 and B2. Consider any P and Q. So B((if P)(Q)) is a
mixture of B1((if P)(Q)) and B2((if P)(Q)). But since we’re assuming The Equation that
means B-plus-P(Q) is a mixture of B1-plus-P(Q) and B2-plus-P(Q). …

Gärdenfors (1982) showed that Linearity is at odds with conditionalization. So The

Equation implies that the plus in B-plus-P can’t be conditionalization:

Observation The Equation is not compatible with conditionalization.

One way of seeing things: Gärdenfors showed that Linearity exactly characterizes
changing the B’s by imaging (imagining and conditionalization can agree only at the
boundaries). Another way: conditionalization implies conservativism: whenever B(Q)=1
and B(P) > 0, then BP(Q) = 1. Linearity rules that out (modulo triviality). Lewis’s
(1976) main triviality results then follow as corollaries.
(All of this is replicated if we think of the B’s as sets of full beliefs. The qualitative
counterpart of The Equation:

(RT) (if P)(Q)  B iff Q  B P

Again, no restrictions on how we arrive at the posterior B P. (RT) implies the qualita-
tive counterpart to Linearity:

(Mon) B1 Ž B2 implies B1 P Ž B2 P.

(Suppose Q  B1 P. Then (if P)(Q)  B1 and so (if P)(Q)  B2. So Q  B2 P.) Just as
Linearity is at odds with conservativity, (Mon) is at odds with it’s counterpart “preserva-
tion” (™P  B implies B Ž B P) (Gärdenfors 1988; Segerberg 1989).)
There are nearby problems, too. I’m dead certain that the Cubs didn’t win it all this
year, but think it’s possible that they won at least 70 games (at a certain point, you have
to just quit keeping track).

(18) a. ??Probably, if the Cubs won at least 70 games, then they won it all this year.
b. If the Cubs won at least 70 games, then they won it all this year.

Given what I know, there’s no hedge low enough for (18a) to be OK and no degree of
belief small enough for me to sign up for the bare (18b) even to that minimal degree.
Bradley (2000) argues that treating indicatives as things expressing propositions—sub-
ject to the same probability hedges that non-conditional stuff is—means having to say
that it is sometimes OK to give positive weight to a conditional and its antecedent even
though you give no weight at all to its consequent.
So if indicatives express propositions of the normal sort and scope under probability
operators, we have trouble. NTVers say so much the worse for the idea that indicatives

458
 I N DI CATI VE CO ND I T I O N A L S

express propositions, have truth conditions, and determine truth-values at all. The rea-
son you sign-up for an indicative just to the degree that you conditionally sign-up for
its consequent given its antecedent is the straightforward reason that indicatives are
vehicles expressing (not representing) such conditional signings-up.
NTVers don’t have to worry about triviality under hedges, but this is an open route
only if they can explain why inferences involving indicatives seem so compelling—it
can’t be because they are entailments, the preservation of the main semantic value
from premises to conclusion or something like that. Here they have something to
say: our intuitive judgments are tracking Adams’s (1975) notion of probabilistic valid-
ity—an argument is probabilistically valid iff the uncertainty of its conclusion is no
more than the sum of the uncertainties of its premises (for an introduction see Adams
(1998)). Modus ponens and modus tollens are probabilistically valid, the paradoxes
of material implication (11) aren’t. But neither is conditional proof. And without
saying something extra and special-purpose about embedded conditionals, neither is
import/export.
Embedding is another worry for NTVers. That is because they have to worry about what
we’re doing when we issue ordinary indicatives. Conditionals aren’t, they say, for assert-
ing. They are vehicles of conditional assertion. So in uttering an indicative (if P)(Q) (at
w, in c), we are conditionally asserting Q-on-P in c at w. That affects c(w) in the same way
that flat-out asserting Q affects c(w)-plus-P. But for this to fly, Q has to be the sort of thing
that can be flat-out asserted in the first place. That means NTVers have revisionary work
to do to explain away apparently negated conditionals like (12) and revisionary work
to do to explain away the apparently right-nested indicatives featured by import/export.
They also face hard choices when it comes to epistemic modals. Consider:

(19) a. If Red isn’t in the box, Blue must be.

b. If Red isn’t in the box, Blue might be.

Either we extend the NTV thesis to the modals or not. If so, then embeddings like
(19) are prima facie ruled out. If not, we cut ties between the modals and conditionals.
Sometimes NTVers say that embedding facts go the other way: since indicatives
don’t always freely embed (they are tough to negate and it’s tough to left-nest them) this
suggests that they don’t express propositions. This is too fast. Some lexical items have
syntactic restrictions on embedding that have nothing to do with their semantics. We’ve
seen this already: might and must resist embedding under negation (and resist embedding
under deontic modal operators) even though can and have to aren’t so fickle.

6 Epistemic Operators
Set aside NTV views. Interaction between indicative conditionals and epistemic opera-
tors (hedges and modals alike) is still tricky. Examples:

(20) a. If the gardener isn’t guilty, the butler must be.

b. If he didn’t tell Harry, he probably told Tom.
c. If Carl is here, then presumably Lenny is here.
d. If the Cubs get good pitching and timely hitting, they might win.

These operators seem to occur embedded, giving us (apparently) instances of this:

459
 A N TH O N Y S. G I L L I E S

(21) (if P)(operator Q)

This takes if-of-English to contribute a conditional relation as its meaning, and further
says that that conditional relation holds between P-worlds and worlds where opera-
tor Q is true. The trouble is there seems to be no such conditional relation. Grant
some minimal assumptions about what conditional relation between (relevant) ante-
cedent-worlds and consequent worlds an (if ·)(·) picks out (at w in c). (For instance,
the relation has to be idempotent and right upward monotonic and it has to care about
consequents.) Then:

Observation The only conditional relation that (if P)(Q) can express between the rel-
evant antecedent worlds and consequent worlds is Ž.

For proofs (of slightly different versions of this), see van Benthem 1986; Veltman 1985;
Gillies 2010.
Now grant some minimal assumptions about the if-relevant worlds at w: (i) w is
always relevant to an indicative (if P)(Q) at w; (ii) the if-relevant worlds (in c at w) are
compatible with c. And grant that modals are quantifiers over worlds compatible with
the context: a might at w in c is an existential quantifier over c(w) and must is the dual
universal quantifier.
To highlight the trouble: I have lost my marbles and know that one and only one—
either Red or Yellow—is in the box.

(22) a. Red might be in the box and Yellow might be in the box.
might P š might Q
b. If Yellow isn’t in the box, then Red must be.
(if ™ Q)(must P)
c. If Red isn’t in the box, then Yellow must be.
(if ™ P)(must Q)

These are all true. But it seems they can’t all be true together if (if ·)(·) means Ž.
No matter what semantics for indicatives we pick, the sentences in (22) are incom-
patible. While it’s a lamentable fact that my marbles are lost, it’s still a fact not an
impossibility.

Proof Sketch. Suppose otherwise and that we have just two worlds compatible with the
context, w and v. Look at w. it’s either a ™Q-world or a ™P-world. Suppose it’s a ™Q-
world. Since (if ™Q)(must P) is true at w in c all the if-relevant ™Q-worlds are worlds
where must P is true. But w must be one of the relevant if-relevant worlds and it’s a ™Q-
world, too. So it’s a world where must P is true. So c(w) has only P-worlds compatible
with it. But might Q is true at w! Mutatis mutandis if w is a ™P-world. (This is equally
an apparent argument against modus ponens for indicatives, assuming that
entailment is simple preservation of truth(-at-a-point). …

Ah, the problem is that we got the scope relations wrong! Epistemic operators, you say,
scope over the indicatives in environments like (20). Replace (21) with

(23) operator (if P)(Q)

460
 I N DI CATI VE CO ND I T I O N A L S

This is no better, really. For one thing if operator is a probability hedge, triviality looms.
(There are other reasons, but they involve questions about whether we decide to be
egalitarian (all worlds compatible with the context are created equal when it comes to say-
ing which are if-relevant) or chauvinistic (not egalitarian). Those questions are left open
given what we’ve said, but when paired with the widescoping strategy in (23), each leads
to different trouble (Gillies 2010: §6).) We appear to be out of choices, scope-wise.

7 Restrictor View
This trouble is a lot like some trouble Lewis (1975) saw for ifs in quantificational envi-
ronments like these:

{ }
Always
(24) Sometimes if a farmer owns a donkey, he beats it.
Seldom

What single connective could if contribute in each of these? Maybe something iffy
would work if the adverb is the universal always, but conjunction would be better if
its an existential like sometimes and neither looks good for seldom. The if has the job of
restricting the quantifiers. Lewis’s (1975) conclusion: the restricting job isn’t one that
a conditional operator can do. So sentences like those in (24) are not instances of a
conditional operator plus an adverb of quantification. Instead, he said, the ifs are a non-
connective whose only job is to mark an argument slot for the adverb of quantification.
Kratzer’s idea (developed in, e.g., Kratzer 1981, 1986, 1991, 2010) is that this holds
not just for ifs under adverbs of quantification but for ifs across the board. The thing all
ifs do is restrict operators. So they aren’t iffy. The choice between (21) and (23) is not
what we are after. Instead what we have is something like this:

Restrictor Analysis Conditionals are restricted operators with logical forms like this:
Quantifier/Operator + if-clause + consequent clause
operator(P)(Q)

When the operator is a modal like must or probably or might, the job of the if-clause is to
restrict the domain over which the modal quantifies. If there is no overt operator, then
since ifs restrict operators, there must be a covert (necessity) one.
This picture is general and powerful. (For more on the restrictor view, see von Fintel
(2011) and the references therein.) It is, for instance, easy to see why indicatives go in
for import/export. It is also easy to see how and why conditionals (which is to say “con-
ditionals”) and operators—like the unrestricted mights in (22)—interact. This is true for
fancier modals and hedges, too:

(25) a. If he’s a Quaker, he’s presumably a pacifist.

b. If the bet is on odd, it’s probably a loser.
c. If the coin is fair, then the probability of heads is ½.

Each of these expresses some restricted hedge: that he’s presumably a pacifist (given he’s
a quaker), that the bet won’t be won (given it’s on odd), that the probability of heads is

461
 A N TH O N Y S. G I L L I E S

 (given the coin is fair). The hedges—presumably, probably, and x-probably—are modal
operators. Treating ifs as restrictors means that when we look at environments like these
we shift our attention to the operators involved. Getting straight about them is how
we get straight about the conditionals in which they occur. Though our current topic
is indicatives and not hedges like these, I’ll just briefly mention two routes to modeling
their contribution that fit hand/glove-wise with the restrictor view of ifs.
Route one: introduce a qualitative ordering between possibilities and eek truth-
conditions for a hedge at w in c out of this. This is the route that Kratzer takes. (There
are a lot of choices for the qualitative structure used and choices for how to connect
it to the hedges. Yalcin (2010) surveys some of the choices.) Route two: introduce a
more fine-grained body of information at hand determined by a context c. In particular:
allow contexts to provide probability measures over the possibilities compatible with
them. (Here, too, there are lots of choices.) Each route ties hedges to a certain body
of information and that body of information can be restricted in pretty straightforward
ways: all the choices allow for a well-understood way of “updating” information that
bears a systematic and well-understood connection to conditionalization. (For more on
the choices for representing the needed uncertainty the hedges trade on, and how those
ways of modeling it go in for updating, see Halpern 2003.)

8 Re-Inventing Conditional Connectives

There are still reasons to explore properly iffy operators that carry conditional mean-
ings. One reason: the charge is that the restricting job that ifs unquestionably do is a job
that can’t be done by any conditional operator. Does the charge stick? It’s worth know-
ing. (Another reason: the way hedged ifs (and their partnering iffed hedges) behave
in conversation seems to require a conditional connective to figure in the story. (The
history of the problem is a little involved, but see von Fintel & Gillies (2011).) That
would be surprising not least because Lewis-style triviality results seem to push exactly
the other way. But the jury is currently still out on this particular bundle of puzzles.) The
re-inventions sketched here also have an answer on offer for what the extra bit is that
indicatives say over and above their corresponding horseshoes (naturally enough the bit
the different re-inventions offer do differ).
One way that if can express a uniform conditional meaning while still doing its
restricting job is for the conditional meaning to be a gappy one—gaps for when things
aren’t as the if-clause says—and for our story of embedding operators to be sensitive to
those gaps. The idea is Belnap’s (1970) and it has been revived and pushed by Huitink
(2009). (Lewis knew this was a way out, but dismissed it.)
Here is the bare bones Belnap view: a conditional (if P)(Q) (in c at w) says pretty
much the same as Q does (in c at w) provided P is true. Otherwise, it says nothing.
Two ways to try to make this go: “boring” and “interesting”. The boring way says
sameness is sameness of truth-value, interesting that it is sameness of semantic value.
Interestingly, it is the boring way that seems more promising. This is what Huitink
does.

Belnap Conditional Indicatives express gappy propositions:

{
[[(if P)(Q)]] c,w = [[Q]] c,w
undefined
if [[P]] c,w = 1
otherwise

462
 I N DI CATI VE CO ND I T I O N A L S

This treats (if P)(Q) like the horseshoe when P is true but diverges from it in just the
cases that the horseshoe theory goes wild. But it doesn’t by itself meet the challenge of
getting the restricting behavior right. If the operator that needs restricting is out-scoped
by the if we will get the wrong and unrestricted truth-conditions and if the operator does
the out-scoping we will still get the wrong and truth-conditions. The fix is easy. First:
widescope the operators. Second: reign in their quantificational domains to survey only
worlds/cases/situations/whatever at which the embedded sentence gets a truth-value.
That’s it.
There is another way to re-invent conditional connectives based on Ramsey’s
(1929/1990) suggestion. The Ramsey test is a recipe for when to accept a conditional: you
accept (if P)(Q) in a state B iff Q is accepted in the subordinate state got by taking B and
adding the information that P to it. But we want a story about what ifs mean, not so much
about when they are/ought to be accepted. Again the fix is easy: (if P)(Q) in a context
c says that all (relevant) P-possibilities are possibilities at which Q is true. What is the
context relevant for checking at those P-possibilities whether Q is true? The subordinate
context got by taking c and adding the information that P to it. Making the straightfor-
ward choice for how the adding goes (zoom in to the possibilities where P is true):

Shifty Conditional Indicatives express shifty propositions:

[[(if P)(Q)]] c,w = 1 iff c(w)  [[P]] c Ž [[Q]] c+P
where c + P = Ov.c(v)  [[P]]c

If we narrowscope the relevant operators, leaving their plain vanilla semantics intact,
this is enough to get the restricting behavior of if-clauses to mesh with them expressing
a genuine conditional connective.
This re-invention is equivalent to dynamic semantic accounts that types all sen-
tences in the language as programs or instructions for changing the context. (The clas-
sic references: Groenendijk Stokhof (1991); Veltman (1996).) Intuition: a program
means what it does, and what it does depends on what things are like—the state you are
in—when you execute it. So, a program’s content is a relation between prior states and
corresponding posterior states.
Dynamic semantics treats all sentences that way: their contents are the characteristic
changes that successful assertions of them induce. A simple example:

Dynamic Semantics for Propositional Logic Take a state (context) s to be a set of worlds.
i. s[P] = {w  s: P is true in w} for atomic P’s
ii. s[™P] = s \ s[P]
iii. s[P š Q]=s[P][Q]

Here the relation is in fact a function [.] from states to states (read it postfix). So (suc-
cessfully asserting) atomic sentences tell us to throw away worlds where they aren’t true,
negations tell us to throw out what would survive an update with the things negated,
and conjunctions tell us to process things in order. The point of each these instructions
is to have a non-null upshot.
Not all instructions have that aim. You say to me: Check whether the game is on. You’re
not instructing me to change anything game-wise or TV-wise but instructing me to see
whether the state we are in has a certain property. It’s a test program. The dynamic idea
is that indicatives are Ramsey-inspired test instructions:

463
 A N TH O N Y S. G I L L I E S

Dynamic Conditional Indicatives express test programs:

s[(if P)(Q)] = {w  s: s[P][Q] = s[P]}

Let’s say that a sentence P is true in state s iff the information that P is already present
in s. That is: P is true in s iff s[P] = s. Then we can put things this way: an indicative
(if P) (Q) is true in a state s iff Q is true in the subordinate state s[P] got by taking s and
adding the information P to it. That is very Ramseylike.

Acknowledgments
I’ve learned a lot about conditionals from a lot of people. Some of my personal intel-
lectual debts are reflected in the references, some not (or not enough). So my biggest
debtholders: Nicholas Asher, Josh Dever, Frank Veltman, and, of course, my frequent
collaborator Kai von Fintel. Finally, I owe a tremendous debt to John L. Pollock. I’d like
to dedicate this to his memory. He will be missed (unconditionally).

References
Adams, Ernest W. 1975. The Logic of Conditionals. D. Reidel.
Adams, Ernest W. 1998. A Primer of Probability Logic. Center for the Study of Language and Information.
Anderson, Alan Ross. 1951. A note on subjunctive and counterfactual conditionals. Analysis 12(2). 35.
Austin, John L. 1956. Ifs and cans. In Proceedings of the British Academy, 109–132.
Belnap, Nuel D. 1970. Conditional assertion and restricted quantification. Noûs 4(1). 1–12.
Bennett, Jonathon. 2003. A Philosophical Guide to Conditionals. Oxford University Press.
van Benthem, Johan. 1986. Essays in Logical Semantics. D. Reidel.
Bradley, Richard. 2000. A preservation condition for conditionals. Analysis 60(3). 219–222.
Edgington, Dorothy. 1995. On conditionals. Mind 104(414). 235–329.
Edgington, Dorothy. 2009. Conditionals. In Ed Zalta (ed.), Stanford Encyclopedia of Philosophy, vol. Spring
2009 Edition. URL http://plato.Stanford.cdu/archives/ win2003/entries/conditionals/.
von Fintel, Kai. 1998. The presupposition of subjunctive conditionals. In U. Sauerland & O. Percus (eds.),
MIT Working Papers in Linguistics: The Interpretive Tract, vol. 25, 29–44.
von Fintel, Kai & Anthony S. Gillies. 2011. Hedging your Ifs and vice versa. Manuscript, MIT/Rutgers.
van Fraassen, Bas C. 1980. Review of Brian Ellis, Rational Belief Systems. Canadian Journal of Philosophy 10.
497–511.
Gärdenfors, Peter. 1982. Imaging and conditionalization. The Journal of Philosophy 79. 747–760.
Gärdenfors, Peter. 1988. Knowledge in Flux. MIT Press.
Gibbard, Allan. 1981. Two recent theories of conditionals. In W. L. Harper, R. Stalnaker & G. Pearce
(eds.), Ifs, 211–248. D. Reidel.
Gillies, Anthony S. 2009. On truth-conditions for If (but not quite only If). The Philosophical Review 118(3).
325–348. doi:10.1215/00318108–2009–002.
Gillies, Anthony S. 2010, Iffiness. Semantics and Pragmatics 3(4). 1–42. doi:10.3765/sp.3.4.
Grice, H. Paul. 1975. Logic and conversation. In P. Cole & J. L. Morgan (eds.), Syntax and Semantics. Aca-
demic Press.
Groenendijk, Jeroen & Martin Stokhof. 1991. Dynamic predicate logic. Linguistics and Philosophy 14.
39–100.
Hájek, Alan. 1994. Triviality on the cheap? In Ellery Eells & Brian Skyrms (eds.), Probability and Condition-
als: Belief Revision and Rational Decision, 113–140. Cambridge University Press.
Halpern, Joseph Y. 2003. Reasoning about Uncertainty. MIT Press.
Heim, Irene. 1983. On the projection problem for presuppositions. WCCFL 2. 114–125.
Huitink, Janneke. 2009. Domain Restriction by Conditional Connectives. Goethe-University Frankfurt.
—— (2010c) “What ‘Intuitions’ are Linguistic Evidence?” Erkenntnis 73: 251–64.
—— (2010d) “Linguistic Intuitions Revisited,” British Journal for the Philosophy of Science 61: 833–65.
—— (2011a) “Experimental Semantics,” Philosophy and Phenomenological Research LXXXII: 418–35.
—— (2011b) “Methodology and the Nature of Knowing How,” Journal of Philosophy CVIII: 205–18.

464
 I N DI CATI VE CO ND I T I O N A L S

Horn, Laurence R. 2001. A Natural History of Negation. CSLI Publications, reissue edition edn.
Jackson, Frank. 1991. Conditionals. Oxford University Press, USA.
Kratzer, Angelika. 1981. The notional category of modality. In H. J. Eikmeyer & H. Rieser (eds.), Words,
worlds, and contexts, new approaches in word semantics, 38–74. Berlin: de Gruyter.
Kratzer, Angelika. 1986. Conditionals. Chicago Linguistics Society 22(2). 1–15.
Kratzer, Angelika. 1991. Modality. In Arnim von Stechow & Dieter Wunderlich (eds.), Semantics: An inter-
national handbook of contemporary research, 639–650. Berlin: de Gruyter.
Kratzer, Angelika. 2010. Papers on Modals and Conditionals. Oxford University Press.
Lewis, David. 1973. Counterfactuals. Blackwell.
Lewis, David. 1975. Adverbs of quantification. In E. Keenan (ed.), Formal Semantics of Natural Language.
Cambridge University Press.
Lewis, David. 1976. Probabilities of conditionals and conditional probabilities. The Philosophical Review 85.
297–315.
Lewis, David. 1986. Probabilities of conditionals and conditional probabilities II. The Philosophical Review
95. 581–589.
Lycan, W. G. 2001. Real Conditionals. Oxford University Press.
McGee, Vann. 1985. A counterexample to modus ponens. Journal of Philosophy 82(9). 462–471.
Ramsey, Frank P. 1929/1990. General propositions and causality. In H. A. Mellor (ed.), F. Ramsey,
Philosophical Papers. Cambridge University Press.
Segerberg, Krister. 1989. A note on an impossibility theorem of Gärdenfors. Noûs 23. 51–354.
Stalnaker, Robert. 1968. A theory of conditionals. In Nicholas Reseller (ed.), Studies in Logical Theory,
98–112. Basil Blackwell.
Stalnaker, Robert. 1975. Indicative conditionals. Philosophia 5. 269–286.
Veltman, Frank. 1985. Logics for Conditionals. Ph.D. thesis, University of Amsterdam.
Veltman, Frank. 1996. Defaults in update semantics. Journal of Philosophical Logic 25. 221–261.
doi:10.1007/BF00248150.
Yalcin, Seth. 2010. Probability operators. URL http://semanticsarchive.net/Archive/Tg3OTQyZ/yalcin.
po.2010.draft.pdf.

465
 KA I VO N F I N T E L

3 . 13
S UBJUNC T IV E
C OND ITI ONALS
Kai von Fintel

1 Introduction
Conditional sentences, canonically of the form “if p, q”, whisk us away to a scenario
described by their antecedent and then make a claim about it in their consequent.
There are two main kinds of conditionals, as illustrated in the well-known Oswald/
Kennedy minimal pair (due to Adams 1970):

(1) If Oswald didn’t kill Kennedy, someone else did.

(2) If Oswald hadn’t killed Kennedy, someone else would have.

Clearly, the two conditionals differ in meaning. The conditional in (1) signals that it is
an open possibility that Oswald didn’t kill Kennedy and will be judged true by anyone
who knows that Kennedy was in fact assassinated. The conditional in (2), in contrast,
signals that it is taken for granted that Oswald did in fact kill Kennedy and makes the
somewhat dubious claim that Kennedy’s assassination was inevitable, perhaps based on
a vast conspiracy. The same difference in meaning can be illustrated with a similar pair
(due to Bennett or Stalnaker?), where it is even harder to hear the second conditional
as making a plausible claim:

(3) If Shakespeare didn’t write Hamlet, someone else did.

(4) If Shakespeare hadn’t written Hamlet, someone else would have.

This chapter will be concerned with the meaning of conditionals of the second kind.
Conditionals of the first kind are usually called “indicative” conditionals, while con-
ditionals of the second kind are called “subjunctive” or “counterfactual” conditionals.
The “indicative” vs. “subjunctive” terminology suggests that the distinction is based in
grammatical mood, while the term “counterfactual” suggests that the second kind deals
with a contrary-to-fact assumption. Neither terminology is entirely accurate.
It is clear that the outward difference between conditionals of the two kinds lies
in their tense/mood/aspect syntax, but it is not reliably a difference in indicative vs.
subjunctive mood properly speaking. In languages that have a past subjunctive (such
as German), antecedents of conditionals of the second kind do indeed appear in the
subjunctive mood but if a language does not have a past subjunctive, some other form is

466
 SU BJ U N CTI VE CO N D I T I O N A L S

used. English uses an indicative antecedent with an additional layer of past tense mor-
phology (“if Oswald didn’t kill Kennedy” becomes “if Oswald hadn’t killed Kennedy”).
It is that additional layer of past tense morphology (which doesn’t obviously contribute
an actual past meaning) that quite reliably signals conditionals of the second kind. So,
instead of calling them subjunctive conditionals, one might call them “additional past”
conditionals.
While our initial examples did appear to signal that their antecedent was false, condi-
tionals of the second kind do not always carry the suggestion of counterfactuality:

(5) If Jones had taken arsenic, he would have shown just exactly those symptoms which
he does in fact show. (Anderson 1951)

A doctor who utters (5) might be prepared to use it as part of an argument that the
antecedent is in fact true, so the conditional could not be conveying counter-factuality
as part of its meaning.
Conversely, there are some formally indicative conditionals that express
counterfactuality:

(6) If he has solved this problem, I’m the Queen of England.

(7) If Messi waits just a second longer, he scores on that play.

[Indicative counterfactuals of the kind in (7) are common in sportscast play-by-play

commentary. They have not (yet) been studied in the semantic literature.]
Even though inaccurate, the terms indicative/subjunctive/counterfactual are so
entrenched that it is presumably futile to try to reform usage. The association of condi-
tionals of the second kind with counterfactuality is somewhat tenuous. In comparison,
the connection of the distinction to tense/mood/aspect morphology is undeniable even
if the morphology in conditionals of the second kind isn’t strictly speaking subjunctive
in many languages. So, the indicative vs. subjunctive terminology is the slightly less
inappropriate one and will be the one we adopt here.

2 The Meaning of Subjunctives

Conditionals whisk us away to a scenario where the antecedent holds. The most influ-
ential semantic theories about subjunctive conditionals, in particular the pioneering
works of Stalnaker and Lewis, treat them as making claims about “possible worlds” that
may be quite different from the actual world. Nevertheless, when one looks at examples
like the Oswald/Kennedy conditionals, one sees that even the counterfactual version
makes palpable claims about reality: “if Oswald hadn’t killed Kennedy, someone else
would have” makes the claim that in actuality, Kennedy’s assassination was inevita-
ble, perhaps because there was a vast conspiracy in place or because one thinks that
Kennedy’s actions and policies inexorably provoked an assassination. How does talk
about other possible worlds connect to the actual world?
All of the possible worlds semantic approaches answer this concern by closely tying
the identification of the relevant set of possible worlds to facts in the actual world.
Stalnaker 1968 grounds the basic idea in the same “Ramsey Test” that has inspired
accounts of indicative conditionals. For Stalnaker, the test describes how we evaluate
conditionals: we hypothetically add the antecedent to our information state and assess

467
 KA I VO N F I N T E L

the consequent in the resulting state. While canonical indicatives have antecedents
that are compatible with the relevant information state, canonical subjunctives involve
antecedents that contradict some prior information. Thus, the adding of the anteced-
ent will necessitate some revisions of the information state. (It should be noted that
the proper analysis of Anderson-type examples, as in (5), in this picture is not entirely
obvious. See von Fintel 1998 for some discussion.)
Correlated with this pyschological process of assessing a conditional, Stalnaker pro-
poses a truth-conditional semantics that starts from the actual world, considers the ante-
cedent, and looks for worlds that differ minimally from the actual world while making
the antecedent true. It is in those worlds that the consequent is then evaluated. So,
in the Oswald/Kennedy case, we keep fixed all actual facts that are not strictly tied to
Oswald killing Kennedy. The subjunctive conditional claiming that if Oswald hadn’t
killed Kennedy, someone else would have, thus amounts to saying that there were facts in
the actual world that would have led to Kennedy’s assassination one way or the other.
Tying the semantics of conditionals to the notion of comparative similarity between
possible worlds may seem like it doesn’t help us nail down the meaning of conditionals
all that much, since similarity obviously is quite a vague and context-dependent notion.
In a brilliant and influential move, Lewis (1973: 91ff) defended the use of this notion by
pointing out that conditionals themselves are inherently vague and context dependent.
Therefore a proper analysis of conditionals needs to correctly reflect their vagueness
and context dependence. He argued that the comparative similarity relation between
possible worlds is just the right tool to do so.
As an illustration of the vagaries of judging similarity and difference between worlds,
consider Quine’s famous pair of conditionals (1960: 221; NB: he actually calls the pair
“Goodman’s, nearly enough”):

(8) If Caesar were in command, he would use the atom bomb.

(9) If Caesar were in command, he would use catapults.

Quine wrote: “we feign belief in the antecedent and see how convincing we then find
the consequent. What traits of the real world to suppose preserved in the feigned world
of the contrary-to-fact antecedent can be guessed only from a sympathetic sense of
the fabulist’s likely purpose in spinning his fable.” His “preserving of traits of the real
world in the feigned world of the antecedent” corresponds quite directly with the deci-
sion about which properties of worlds carry more weight in the judgment of similarity
between worlds. In (8), we seem to keep constant Caesar’s ruthlessness, while in (9) we
(also) care about the technologies that were actually at his disposal in his time.
The similarity/difference-based semantics remains quite a schematic framework, to
be filled in with contextually relevant considerations for assessing similarities and differ-
ences. In later work, in response to several worries about the account, Lewis and others
suggested constraints on the kind of similarity relation underlying conditionals. This
debate continues rather unabated.
A variant of the semantics based on a similarity ordering is given by “premise seman-
tics”, inspired by Goodman (1947) and Rescher (1964) and developed in rival forms by
Kratzer (1977, 1979, 1981) and Veltman (1976). Here, we start with a description of the
actual world (a set of sentences or propositions) and revise it to make it accommodate
the addition of the antecedent. Then, we check whether the consequent follows from
the resulting set of premises. The revision process in this account intuitively parallels

468
 SU BJ U N CTI VE CO N D I T I O N A L S

the determination of the antecedent worlds most similar to the actual world in the
Stalnaker/Lewis framesvork. And in fact, Lewis (1981) showed that technically the two
approaches are intertranslateable. More recent work in the premise semantic tradition
includes Kratzer 1989, Veltman 2005, Kanazawa et al. 2005, Kratzer 2005.

3 Nonmonotonicity
There is an austere way of spelling out a possible worlds analysis of conditionals accord-
ing to which the context supplies us with a set of relevant possible worlds, those that
in all relevant respects are similar enough to the actual world, and that the conditional
then makes a claim about all of the antecedent worlds in that set. This is known as
“strict implication”. In contrast, the analysis proposed by Stalnaker and Lewis is a “vari-
ably strict” analysis (terminology due to Lewis), in which we start from the antecedent
and identify among the antecedent worlds those worlds that are as similar as possible to
the actual world. The variably strict semantics in distinction to the strict implication
analysis predicts a variety of nonmonotonic behaviors, and Stalnaker and Lewis argued
that those predictions are borne out.
Under a Strict implication analysis, the pattern known as Strengthening the Ante-
cedent, is predicted to be valid:

(10) Strengthening the Antecedent

if p, q Ÿ if p&r, q

If all (contextually relevant) p-worlds are q-worlds, then a fortiori all p&r-worlds, a
subset of the p-worlds, have to be q-worlds. This pattern becomes invalid in the Stal-
naker/Lewis analyses. If the p-worlds that are most similar to the evaluation world are
all q-worlds, that does not necessitate that the most similar p&r-worlds are also all
q-worlds. Lewis (1973) gives a humorous example:

(11) Failure of Strengthening the Antecedent

If kangaroos had no tails, they would topple over. Ÿ
e
If kangaroos had no tails but used crutches, they would topple over.

The variably strict analyses explain why the inference in (11) fails: the worlds where
kangaroos have no tails but that are otherwise as similar as possible to the evaluation
world are not worlds where kangaroos use crutches, so the first conditional does not
connect logically to the second conditional.
Other patterns that are expected to be valid under the strict implication analysis but
arguably aren’t are Hypothetical Syllogism and Contraposition:

(12) Failure of the Hypothetical Syllogism (Transitivity)

If Hoover had been a Communist, he would have been a traitor.
If Hoover had been born in Russia, he would have been a Communist.
 Ÿ e If Hoover had been born in Russia, he would have been a traitor.
(13) Failure of Contraposition
If it had rained, there wouldn’t have been a terrific cloudburst. Ÿ
e
If there had been a terrific cloudburst, it wouldn’t have rained.

469
 KA I VO N F I N T E L

[The example in (12) is from Stalnaker 1968, and the example in (13) is a subjunctive
version of an example from Adams 1975.]
The non-monotonic analyses predict both of these invalidities correctly.

4 The Dynamic Strict Analysis

Let us return to the apparent failure of Strengthening the Antecedent. Lewis tried to
forestall the idea that what is treated as semantic non-monotonicity in his account
could actually be explained in a strict implication account by saying that the contex-
tually relevant set of worlds that the conditional quantifies over is easily shifted in a
sequence of sentences. He argued that this move would not be able to explain the well-
formedness of what became known as Sobel Sequences:

(14) If the USA threw its weapons into the sea tomorrow, there would be war; but if
all the nuclear powers threw their weapons into the sea tomorrow, there would be
peace.

Lewis deliberately put this example “in the form of a single run-on sentence, with the
counterfactuals of different stages conjoined by semicolons and but”, suggesting that it
would be a “defeatist” move to say that in such a tight sequence the context could shift
in response to the introduction of a new antecedent clause.
Defeatist or not, based on an observation by Heim (MIT class handout), von Fintel
(2001) develops such an account. Heim had noted that Lewis’ Sobel Sequence cannot
be reversed:

(15) ??If all the nuclear powers threw their weapons into the sea tomorrow, there
would be peace; but if the USA threw its weapons into the sea tomorrow, there
would be war.

This is unexpected from the point of view of a semantically non-monotonic analysis.

In von Fintel’s paper, a dynamic strict analysis is developed in which the antecedent
has the potential to expand the “modal horizon”, the set of contextually relevant pos-
sible worlds which the conditional then ranges over. It is shown that if the expansion
of the modal horizon is governed by the same similarity ordering used in the Stalnaker/
Lewis systems, the analysis replicates the truth-conditions of those systems for isolated
or discourse-initial conditionals. The context shifts become only relevant in sequences
of conditionals and then create the appearance of semantic non-monotonicity. One
crucial argument von Fintel gives for his account is that negative polarity items are
licensed in the antecedent of conditionals and that therefore we would prefer a monot-
onic analysis. It turns out, however, that only a very special notion of monotonicity
(dubbed Strawson Downward Entailingness) holds for von Fintel’s conditionals: these
conditionals are downward monotone in their antecedent only under the assumption
that the initial context is such that the modal horizon is already large enough to be
unaffected by any of the conditionals in the sequence. This idea is explored for other
puzzles for NPI-licensing in von Fintel 1999. The dynamic strict analysis is developed
further by Gillies (2007) and critically compared to a pragmatically supplemented non-
monotonic analysis by Moss (2010).

470
 SU BJ U N CTI VE CO N D I T I O N A L S

5 The Restrictor View

The dominant approach to the semantics of conditionals in linguistics is not so much
an alternative to the accounts we have discussed so far, and in particular not to the
Stalnaker/Lewis analysis, but a radical rethinking of the compositional structure of con-
ditional sentences. It began with Lewis’s 1975 paper on adverbial quantification, which
dealt with sentences like

(16) If it is sunny, we always/usually/mostly/rarely/sometimes/never play soccer.

Lewis argued that there was no plausible semantics for the conditional connective that
would interact compositionally with the adverbs of quantification to give correct truth-
conditions for these sentences. Instead, he argued that the if-clause added no conditional
meaning of its own to the construction. The idea is that the only “conditional” operator
in the structure is the adverb and that if merely serves to introduce a restriction to that
operator. In other words, where naïvely one would have thought that (16) involved the
combination of an adverbial quantificational operator with the conditional expressed
by if, Lewis argued that there was just one operator and that if didn’t express any kind of
conditional operator of its own.
Lewis himself did not generalize this idea; nowhere else in his writings does he give
any indication that if’s found elsewhere are to be treated on a par with the if in adverbi-
ally quantified sentences. (It should be noted that in the adverbial quantification paper,
Lewis does suggest that the if found in construction with probability operators is also
not a conditional operator of its own, although he doesn’t say whether it is to be seen
as a restrictor in those cases. It is a shame that Lewis did not connect his insights in the
adverbial quantification paper to the problems surrounding conditional probability, as
discussed for example in Lewis 1976; cf. also Hájek 1993. Kratzer (1986) does make the
connection; for some recent discussion see Rothschild 2010 and Egré & Cozic 2011.)
Kratzer took the logical step and argued that Lewis’ idea should be applied to all condi-
tional constructions. She put the point very concisely in Kratzer 1986: “The history of the
conditional is the story of a syntactic mistake. There is no two-place if... then connective
in the logical forms for natural languages. If-clauses are devices for restricting the domains
of various operators. Whenever there is no explicit operator, we have to posit one.”
The central idea is that if itself does not carry any distinctive conditional meaning,
rather it is, so to speak, a helper expression that modifies various quantificational/modal
operators. This doesn’t just apply when an overt operator combines with an if-clause but
also when an if-clause occurs on its own with no overt operator in sight. In that case,
Kratzer suggest, there must a covert, or at least not obviously visible, operator. What
one might call bare indicative conditionals either contain a covert epistemic necessity
modal or a covert generic frequency operator (| usually/always):

(17) If he left at noon, he’s home by now. [epistemic necessity]

(18) If he leaves work on time, he has dinner with his family. [generic frequency]

In bare subjunctive conditionals, one should consider the possibility that the modal
form would is the operator restricted by the if-clause, an idea bolstered by the fact that
there are if-less would-sentences (see Kasper 1992 and Schueler 2008):

(19) I would have beaten Kasparov.

471
 KA I VO N F I N T E L

It should be pointed out that while one may have a desire to have a uniform(ish) analy-
sis of indicative and subjunctive conditionals, partially because both are if ... then con-
structions, the restrictor analysis opens up a potentially large gap between them. The
uniform presence of if would be almost entirely beside the point: how big the difference
between the two kinds is depends on what, if any, difference there is between the modal
operators present in them.

6 The Limit and Uniqueness Assumptions, Conditional

Excluded Middle, and Might
Lewis and Stalnaker differ in their assumptions about the similarity ordering. Stalnaker
assumes that for any (non-contradictory) antecedent and any evaluation world, there
will be a unique most similar antecedent world. Lewis neither makes this Uniqueness
Assumption (he calls it “Stalnaker’s Assumption”) nor the weaker Limit Assumption
(that for any antecedent and evaluation world, there is a set of most similar anteced-
ent worlds). Informally, here, we have been using the Limit Assumption but not the
Uniqueness Assumption when we talk about the most similar or closest antecedent
worlds. For discussion of this difference, see Lewis (1973: 19–21) and Stalnaker (1984:
Chapter 7, esp. 140–142); Pollock (1976), Herzberger (1979), and Warmbrod (1982)
argue for the Limit Assumption as well. More recently, Williams (2010b) and Swanson
(2010) have revisited the topic.
In his argument against the Limit Assumption, Lewis adduces cases that seem to show
that the closeness to the actual world can get asymptotically closer:

(20) If this one inch line were more than an inch long, ...

For any world in which the line is 1+x inches long, there will be a world where the line
is just a little bit shorter but still more than an inch long. So, there will be no world
where the line is as close to its actual length as possible while still more than an inch
long. Stalnaker argues that in actual use, natural language would not make the fine-
grained distinctions needed to threaten the Limit Assumption in such cases.
Cases that throw doubt on the Uniqueness Assumption are less recherché. Quine’s
example (1950: 14) will do:

(21) If Bizet and Verdi had been compatriots, Bizet would have been Italian.
(22) If Bizet and Verdi had been compatriots, Verdi would have been French.

If the Uniqueness Assumption were correct, exactly one of (21) and (22) would be true.
But they both seem dubious.
A virtue of the Uniqueness Assumption is that it validates the principle of Condi-
tional Excluded Middle: either if p, q or if p, not q. Surprisingly, this validity seems to
persist even in the problematic Bizet–Verdi case:

(23) Either if Bizet and Verdi had been compatriots, Bizet would have been Italian, or
(if Bizet and Verdi had been compatriots) Verdi would have been French.

Stalnaker suggests that (21) and (22) are semantically indeterminate (because it is
indeterminate which resolution of the similarity ordering is contextually salient), but

472
 SU BJ U N CTI VE CO N D I T I O N A L S

that (23) is true nevertheless because it would be true under any reasonable resolu-
tion of the indeterminacy. He proposes implementing this suggestion in a supervalu-
ation framework. Klinedinst (2011) explores this further and shows how it can make
good on the suggestion by von Fintel and Iatridou (2002) and Higginbotham (2003)
that a CEM-validating semantics for conditionals is behind the intuitive equivalence
of pairs of conditionals under quantifiers (contra Leslie 2009; see also Huitink
2010):

(24) Every student would have failed if he had goofed off.

(25) No student would have passed if he had goofed off.

The validity of CEM also is involved in the intuitive relation between if-conditionals
and only if-conditionals:

(26) Only if the Queen had been home, would the flag have flown.
(27) If the flag had flown, the Queen would have been home.

See Barker 1993 and von Fintel 1997 for discussion.

Another consequence of making the Uniqueness Assumption, is that Stalnaker can-
not treat would and might as duals (as proposed by Lewis). Instead, he suggests that might
is a higher (usually epistemic) operator that takes a whole counterfactual in its scope.
And the embedded would-counterfactual is of course analyzed as usual. He gives the
following paraphrase relation:

(28) If John had been invited, he might have come to the party.
(29) It might be the case that if John had been invited, he would have come to the
party.

Consider the following contrast:

(30) It’s not the case that John must come to the party. But he might.
(31) #It’s not the case that John would have come to the party if he had been invited.
But he might have.

(31) should be fine according to Lewis, it’s like saying not all but some. Stalnaker explains
its anomaly this way: in the first sentence the speaker says that he knows that John
wouldn’t have come, in the second she says that it’s compatible with her knowledge
that he would have.
Lewis has a purported counterexample to this analysis. Suppose there is in fact no
penny in my pocket, although I do not know it since I did not look. “Then ‘If I had
looked, I might have found a penny’ is plainly false.” But it is true that it might be, for
all I know, that I would have found a penny if I had looked.
Stalnaker contends that under the epistemic reading of might, the might-conditional
is not in fact plainly false. He concedes a non-epistemic reading where the statement
is false, but proposes to capture it by relativizing might to a situation where the speaker
knows all the relevant facts. “This will yield a kind of quasi-epistemic possibility—
possibility relative to an idealized state of knowledge.” This reading, as he points out,
comes out almost identical to Lewis.

473
 KA I VO N F I N T E L

The stakes in this debate are increased by the observation that might-conditionals
are very easy to read as true. If might and would are duals, then would-conditionals are
predicted to be very hard to read as true. In fact, Hájek (2009) claims that (almost) all
counterfactuals are false because of this. See DeRose (1999) for an earlier discussion and
Hawthorne (2005), Williams (2008, 2010a) for more recent relevant work. One might
hope that thinking about the contextual dynamics of might-conditionals (see Gillies
2007 and Moss 2010) would defuse the argument.

7 Tense and Aspect

We will not be able to discuss the syntax of conditionals in this chapter (cf. Bhatt &
Pancheva 2006) but we should take a look at the morphological fine structure of condi-
tionals. It is quite apparent that in English at least, the indicative/subjunctive classifica-
tion of conditionals is marked by tense and aspect morphology:

(32) If Grijpstra played his drums, de Gier played his flute.

(33) If Grijpstra had played his drums, de Gier would have played his flute.

The earliest works taking the role of tense and aspect in the semantics of conditionals
seriously came from Dudman (1983, 1984, 1988, 1989). Work on the interaction of
tense and conditionals in philosophical logic includes Nute (1982, 1991); Slote (1978);
Thomason and Gupta (1980), and Thomason (1985). A more recent seminal contribu-
tion is Iatridou (2000). Since then there has been a proliferation of work on this topic;
see Arregui (2005, 2007, 2009); Copley (2006); Ippolito (2007); Kaufmann (2005);
Schulz (2008); von Stechow (2007). Here, we can only introduce some basic facts and
generalizations.
The central observation is that what is commonly called subjunctive in “subjunc-
tive conditionals” is an additional layer (or two) of past tense morphology, no matter
whether the referred to state of affairs is temporally located in the past, present, or
future:

(34) a. If Roman comes to the party tomorrow, it will be a grand success.

b. If Roman came to the party tomorrow, it would be a grand success.
c. If Roman had come to the party tomorrow, it would have been a grand
success.
(35) a. If Roman is at the post office now, he is missing the meeting, meeting.
b. If Roman were at the post office now, he would be missing the meeting.
c. If Roman had been at the post office now, he would have been missing the
meeting.
(36) a. If Roman left before noon, he arrived in time.
b. If Roman had left before noon, he would have arrived in time.

Iatridou (2000) discusses this basic pattern (although she doesn’t discuss the two layer
pasts in future or present conditionals) and proposes that the additional past does not
serve a temporal function. Instead, she argues that the past tense has a schematic seman-
tics that can be applied both temporally and modally: past is an “exclusion feature”.
When past is used temporally it marks the times talked about as distinct from file now

474
 SU BJ U N CTI VE CO N D I T I O N A L S

of the speaker (an additional wrinkle is needed to explain why past means past rather
than non-present = past or future). When past is used modally it marks the worlds talked
about as distinct from the actual world of the speaker (this does not mean that modally
used past is a counterfactuality marker; rather, the intent is to derive something very
much like the Stalnaker-analysis of the import of subjunctive marking).
The alternative to Iatridou’s account is to try to maintain that the additional pasts
in subjunctive conditionals do after all retain their usual temporal meaning. This idea
goes back to Dudman (1983, 1984, 1988, 1989) and has been pursued by Ippolito (2003,
2007) and Arregui (2005, 2009), among others. We do not have the space to survey the
details of these accounts. Let’s rather look at a simplified sketch. Suppose that the extra
layer of past tense marks that what the conditional quantifies over is a set of worlds that
were accessible from the evaluation world at a past time but may not be anymore. This
is typically embedded in a branching futures version of possible worlds semantics. As
the time index progresses, more and more open futures are precluded. Imagine that at
some point in time, it was an open possibility that Roman would leave before noon, but
by the present time it is settled that he did not. Then, assuming that the conditional
employs a “historical necessity”-type of accessibility relation, the time index needs to
be moved to the past to make sure that the domain of accessible worlds includes at least
some worlds where he did leave before noon. Hence, the need for past tense marking on
the modal (would = will + PAST) in (36b); the past tense in the antecedent may be a
mere agreement phenomenon.
What then about the indicative conditional in (36a)? Clearly, if we assume a histori-
cal necessity modal, at the time of utterance it is already settled whether Roman did
or did not leave before noon. So, if there need to be at least some antecedent worlds
in the domain of the modal, the covert modal in (36a) cannot be a historical necessity
modal. Thus, it is not mysterious why (36a) is naturally analyzed as involving a (covert)
epistemic necessity modal.
In this story, then, the difference between indicative and subjunctive is twofold: (i)
type of accessibility relation/type of modal (epistemic vs. historical), (ii) time index on
the modal (present vs. past). An obvious question is whether these differences cross-cut:
are there past epistemic conditionals? are there present historic necessity conditionals?
The answer to the second question is possibly yes: If Roman comes to the party tomorrow,
it will be a grand success might arguably be a non-epistemic conditional. The answer to
the first question might be expected to be no, since it is well-known that epistemic
modals resist embedding under past tense.
One possibly problematic fact for the view just sketched comes from hindsight coun-
terfactuals (Barker 1998, Edgington 2003):

(37) [A randomly tossed coin comes up heads.]

a. If you had bet on heads, you would have won.
b. If you bet on heads, you will win.

While (37a) seems acceptable and true after the coin has come up heads, there is no
time in the past at which (37b) would have been rational to assert. While that doesn’t
mean that there wasn’t a time at which the indicative conditional was true, it does
throw some doubt on the simple idea that the only difference between (37a) and (37b)
is the temporal perspective.

475
 KA I VO N F I N T E L

Related Topics
3.12 Indicative Conditionals.

References
Adams, Ernest W. 1970. Subjunctive and indicative conditionals. Foundations of Language 6(1). 89–94.
Adams, Ernest W. 1975. The logic of conditional: an application of probability to deductive logic, Dordrecht:
Reidel.
Anderson, Alan Ross. 1951. A note on subjunctive and counterfactual conditionals. Analysis 12(2). 35–38.
Arregui, Ana. 2005. On the accessibility of possible worlds: The role of tense and aspect. Amherst, MA: Univer-
sity of Massachusetts at Amherst dissertation.
Arregui, Ana. 2007. When aspect matters: The case of would-conditionals. Natural Language Semantics
15(3). 221–264.
Arregui, Ana. 2009. On similarity in counterfactuals. Linguistics and Philosophy 32(3). 245–278.
Barker, Stephen. 1993. Conditional excluded middle, conditional assertion, and only if. Analysis 53(4).
254–261.
Barker, Stephen J. 1998. Predetermination and tense probabilism. Analysis 58(4). 290–290.
Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In The Blackwell companion to syntax, vol. 1,
638–687. Oxford: Blackwell.
Copley, Bridget. 2006. Temporal orientation in conditionals (or, how I learned to stop worrying and love
UFOs). Ms, CNRS/Université Paris 8.
DeRose, Keith. 1999. Can it be that it would have been even though it might not have been? Noûs 33(sl3).
385–413.
Dudman, V.H. 1983. Tense and time in English verb clusters of the primary pattern. Australian Journal of
Linguistics 3(1). 25–44.
Dudman, V.H. 1984. Parsing if-sentences. Analysis 44(4). 145–153.
Dudman, V.H. 1988. Indicative and subjunctive. Analysis 48(3). 113–122.
Dudman, V.H. 1989. Vive la révolution! Mind 93(392). 591–603.
Edgington, Dorothy. 2003. Counterfactuals and the benefit of hindsight. In Phil Dowe & Paul Noordhof
(eds.), Cause and chance: Causation in an indeterministic world, 12–27. London: Routledge.
Egré, Paul & Mikaël Cozic. 2011. If-clauses and probability operators. Topoi 30(1). 17–29.
von Fintel, Kai. 1997. Bare plurals: bare conditionals, and Only. Journal of Semantics 14(1). 1–56.
von Fintel, Kai. 1998. The presupposition of subjunctive conditionals. In Uli Sauerland & Orin Percus (eds.),
The interpretive tract, (MIT Working Papers in Linguistics 25), 29–44. Cambridge, MA: MIT Press.
von Fintel, Kai. 1999. NPI licensing, Strawson entailment, and context dependency. Journal of Semantics
16(2). 97–148.
von Fintel, Kai. 2001. Counterfactuals in a dynamic context. In Michael Kenstowicz (ed.), Ken Hale: A life
in language, 123–152. Cambridge, MA: MIT Press.
von Fintel, Kai & Sabine Iatridou. 2002. If and when If-clauses can restrict quantifiers. Ms, MIT.
Gillies, Anthony S. 2007. Counterfactual scorekeeping. Linguistics and Philosophy 30(3). 329–360.
Goodman, Nelson. 1947. The problem of counterfactual conditionals. Journal of Philosophy 44(5). 113–128.
Hájek, Alan. 1993. Conditional probability. Princeton, NJ: Princeton University dissertation.
Hájek, Alan. 2009. Most counterfactuals are false. Draft of monograph.
Hawthorne, John. 2005. Chance and counterfactuals. Philosophy and Phenomenological Research 70(2).
396–405.
Herzberger, Hans. 1979. Counterfactuals and consistency. The Journal of Philosophy 76(2). 83–88.
Higginbotham, James. 2003. Conditionals and compositionality. Philosophical Perspectives 17(1). 181–194.
Huitink, Janneke. 2010. Quantified conditionals and compositionality. Language and Linguistics Compass
4(1). 42–53.
Iatridou, Sabine. 2000. The grammatical ingredients of counterfactuality. Linguistic Inquiry 31(2). 231–270.
Ippolito, Michela. 2003. Presuppositions and implicatures in counterfactuals. Natural Language Semantics
11(2). 145–186.
Ippolito, Michela. 2007. Semantic composition and presupposition projection in subjunctive conditionals.
Linguistics and Philosophy 26(6). 631–672.
Kanazawa, Makoto, Stefan Kaufmann & Stanley Peters. 2005. On the lumping semantics of counterfactuals.
Journal of Semantics 22(2). 129–151.

476
 SU BJ U N CTI VE CO N D I T I O N A L S

Kasper, Walter. 1992. Presuppositions, composition, and simple subjunctives. Journal of Semantics 9(4).
307–331.
Kaufmann, Stefan. 2005. Conditional truth and future reference. Journal of Semantics 22(3). 231–280.
Klinedinst, Nathan. 2011. Quantified conditionals and conditional excluded middle. Journal of Semantics
28(1). 149–170.
Kratzer, Angelika. 1977. What must and can must and can mean, Linguistics and Philosophy 1(3). 337–355.
Kratzer, Angelika. 1979. Conditional necessity and possibility. In Rainer Bäuerle, Ura Egli & Arnim von
Stechow (eds.), Semantics from different points of view, 117–147. Berlin: Springer.
Kratzer, Angelika. 1981. Partition and revision: The semantics of counterfactuals. Journal of Philosophical
Logic 10(2). 201–216.
Kratzer, Angelika. 1986. Conditionals. Chicago Linguistics Society 22(2). 1–15.
Kratzer, Angelika. 1989. An investigation of the lumps of thought. Linguistics and Philosophy 12(5).
607–653.
Kratzer, Angelika. 2005. Constraining premise sets for counterfactuals. Journal of Semantics 22(2). 153–158.
Leslie, Sarah-Jane. 2009. If, unless, and quantification. In Robert J. Stainton & Christopher Viger (eds.),
Compositionality, context and semantic values: Essays in honour of Ernie Lepore, 3–30. Berlin: Springer.
Lewis, David. 1973. Counterfactuals. Oxford: Blackwell.
Lewis, David. 1975. Adverbs of quantification. In Edward Keenan (ed.), Formal semantics of natural language,
3–15. Cambridge: Cambridge University Press.
Lewis, David. 1976. Probabilities of conditionals and conditional probabilities. The Philosophical Review
85(3). 297–315.
Lewis, David. 1981. Ordering semantics and premise semantics for counterfactuals. Journal of Philosophical
Logic 10(2). 217–234.
Moss, Sarah. 2010. On the pragmatics of counterfactuals. Noûs. Published online at http:dx.doi.org/10.1111/
j.1468-0068.2010.00795.x/
Nute, Donald. 1982. Tense and conditionals. Tech. rep. Deutsche Forschungsgemeinschaft.
Nute, Donald. 1991. Historical necessity and conditionals. Noûs 25(2). 161–175.
Pollock, John. 1976. Subjunctive reasoning. Dordrecht: Reidel.
Quine, Willard Van Orman. 1950. Methods of logic. New York: Henry Holt and Company.
Quine, Willard Van Orman. 1960. Word and object. Cambridge, MA: MIT Press.
Rescher, Nicholas. 1964. Hypothetical reasoning. Amsterdam: North-Holland.
Rothschild, Daniel. 2010. Do indicative conditionals express propositions? Forthcoming in Noûs.
Schueler, David. 2008. The syntax and semantics of implicit conditionals: Filling in the antecedent. Los
Angeles, GA: UCLA dissertation.
Schulz, Katrin. 2008. Non-deictic tenses in conditionals. Semantics and Linguistic Theory (SALT) 18.
Slote, Michael A. 1978. Time in counterfactuals. The Philosophical Review 87(1). 3–27.
Stalnaker, Robert. 1968. A theory of conditionals. In Nicholas Rescher (ed.), Studies in logical theory,
98–112. Oxford: Blackwell.
Stalnaker, Robert. 1984. Inquiry. Cambridge, MA: MIT Press.
von Stechow, Arnim. 2007. Tense and presuppositions in counterfactuals. Handout from a talk in Oslo,
October 10, 2007.
Swanson, Eric. 2010. Conditional excluded middle without the limit assumption. Ms, forthcoming in
Philosophy and Phenomenological Research.
Thomason, Richmond & Anil Gupta. 1980. A theory of conditionals in the context of branching time. The
Philosophical Review 89(1). 65–90.
Thomason, Richmond H. 1985. Note on tense and subjunctive conditionals. Philosophy of Science 52(1).
151–153.
Veltman, Frank. 1976. Prejudices, presuppositions and the theory of conditionals. In Jeroen Groenendijk
& Martin Stokhof (eds.), Amsterdam papers informal grammar, volume I, 248–281. Amsterdam: Centrale
Interfaculteit, University of Amsterdam.
Veltman, Frank. 2005. Making counterfactual assumptions. Journal of Semantics 22(2). 159–180.
Warmbrod, Ken. 1982. A defense of the limit assumption. Philosophical Studies 42(1). 53–66.
Williams, J. Robert G. 2008. Chances, counterfactuals, and similarity. Philosophy and Phenomenological
Research 77(2). 385–420.
Williams, J. Robert G. 2010a. Chancy counterfactuals, redux: Response to Dodd. Ms.
Williams, J. Robert G. 2010b. Defending conditional excluded middle. Noûs 44(4). 650–688.

477
 PAU L H A G S T R O M

3 . 14
QUE S T IONS
Paul Hagstrom

1 Questions, Interrogatives, and Interrogative Acts

What we make of questions depends on our goals. Coming from semantics and logic,
we aim to clarify the relationships among questions and between questions and their
(actual or potential, complete or partial) answers. Coming from pragmatics, we want to
understand the role of questions in structuring discourse, and how they respond to and
change the context. Questions also have a role in the process of rational inquiry and
theory development.
First, to disentangle the term “question.” One can ask a question. In so doing, one
performs a speech act, an interrogative act, of the sort performed when making an asser-
tion or uttering an imperative. An interrogative act is a kind of request, a directive to
answer.
Interrogative acts are often performed using a sentence in a particular syntactic form,
an interrogative form. The interrogative forms in English often involve a change in word
order, use of interrogative words such as if, whether, and who, or a particular intonation
contour (1). Interrogatives can stand alone as simple sentences (matrix interrogatives),
or be embedded as an indirect question (2), just as an indicative sentence can (3).

(1) What did John buy?

(2) I asked Mary [what John bought].
(3) I told Mary [that John bought a laptop].

Here we will concentrate on a more abstract notion of question: the object an inter-
rogative denotes. A question in this sense is something you can have, discuss, wonder
about, dismiss, or revisit.
These conceptual divisions parallel those we make for propositions: Interrogative
acts and assertions are speech acts, indicatives and interrogatives are sentence forms,
and propositions and questions are abstract referents. These notions are dissociable:
One can perform an interrogative act with an indicative form, use an interrogative form
to make an assertion, or express a question without an interrogative form.
In what follows, we will be surveying some of the basic phenomena, issues, argu-
ments, and conclusions pertaining to questions. However: Our time together is brief,
and the field is vast. Other recent overviews are worth consulting, including Ginzburg
(1996), Higginbotham (1996), Groenendijk and Stokhof (1997), Harrah (2002), Hag-
strom (2003), and Dekker et al. (2007).

478
 Q U ESTI O N S

2 Relating Questions to Logic and Semantics

The issue of how to relate questions to logic has been a running theme in the litera-
ture on questions. Classically, the study of logic deals with inference, deductions, valid
argumentation, relating propositions in terms of truth values. A common goal in the
investigation of questions is to describe their relationships to propositions and to other
questions with a precision and predictive power comparable to that achieved for propo-
sitions—but the centrality of truth values in propositional logic (and its extensions)
leaves questions without an obvious niche to occupy. Whatever else we know about
questions, the intuition is strong that a question differs from a proposition insofar as a
question itself can be neither true nor false. How then can we define notions like entail-
ment or equivalence?
Prior and Prior (1955) make some progress toward an “erotetic logic” (a logic of ques-
tions), drawing on early proposals from Whately (1854, 1877), but their conclusion is
that questions will not submit to a proper treatment within symbolic logic. However,
the developments over the subsequent half century entitle us to consider their pes-
simism premature. As described below, many current approaches succeed in bridging
questions and propositions by viewing questions in terms of their answers.

3 Background on Propositions and Possible Worlds

Let us begin by settling on an characterization of a proposition (informally—for discus-
sion of the nuances and issues that arise here, see Chapters 4.1, 5.1, 5.4, 5.5, 5.10). The
universe (or “world”) is in some state. We may know some aspects of this state, but we
can imagine ways in which they might have been different. We will call the set of all
states a universe could potentially be in the set of possible worlds. Somewhere in there
is the actual world, but knowing which it is would require being able to fully specify the
state of the actual world.
A proposition could be true or false: It is true in some possible worlds, false in others.
Knowing a proposition to be true narrows the possible locations of the actual world—
the actual world must be in that subset of the possible worlds where the proposition is
true. Gaining information (coming to know that some proposition is true) shrinks the
set of possible worlds containing viable candidates for being the actual world.
We can thus take a proposition to characterize a set of possible worlds (those in which
it is true). From the other direction, we can take any set of possible worlds to define
a proposition (although perhaps in only an abstract sense—not every set of possible
worlds corresponds to a proposition expressible in natural language).

4 Questions, Answers, and Partitions (Hamblin 1958)

Hamblin (1958) proposed three Postulates (4), relating questions, statements, and
answers.

(4) Postulate 1. An answer to a question is a statement.

Postulate 2. Knowing what counts as an answer is equivalent to knowing
the question.
Postulate 3. The possible answers to a question are an exhaustive set of
mutually exclusive possibilities.

479
 PAU L H A G S T R O M

Hamblin’s first Postulate combines the ideas that (a) an answer is a vehicle for informa-
tion, and (b) the “unit of information” is the proposition. Specifically, the first Postu-
late excludes sentence fragments from the domain of answers. While it is felicitous in
discourse to respond to “Who left?” by saying “Adam,” “Adam” is not an answer to the
question—it is an elliptical form of the answer “Adam left” and conveys the proposition
that Adam left.
The second Postulate encodes the idea that, just as propositions are distinguished
by their truth conditions, questions are distinguished by their answerhood conditions.
In response to “Who left?” it seems clear that “Adam left” would count as an answer,
whereas “Someday the Sun will go out” would not (even if true). A question divides
propositions into those that are possible answers and those that are not. Postulate 2
illustrates a specific case, but (even without confronting the issue of what Hamblin
meant by “knowing the question”) we can take it more generally as setting up an equiva-
lence between a question and its answers.
The third Postulate characterizes the set of answers. A question presents a partition of
the possible worlds into distinct cells, such that any given possible world is compatible
with exactly one of the answers. To ask a canonical question is to present the hearer
with this set of cells and request help locating the actual world in one of them: Which,
of these ways that the world could be, is it?

5 The Possible, the True, and the Exhaustive

Before turning to broader empirical questions, we will review a few other analyses for
comparison, each of which builds on Hamblin’s (1958) insight. One issue of particular
concern to these subsequent approaches is the ability to derive appropriate meanings
compositionally from the syntactic form (see also Chapters 2.9, 4.6).

5.1 The Tale of Adam, Betty, Christy, and David

Listen:
Once upon a time, there were four students. Their names were Adam, Betty, Christy,
and David. Adam and Christy were spies, working for Mr. X. One evening, all four went
to a party. But before it was over, Christy and David left.
The End.

5.2 Sets of Possible Answers (Hamblin 1973)

Hamblin (1973) proposed an extension of Montague’s (1970) English as a Formal Lan-
guage to capture matrix questions. In so doing, he reinterprets the denotation of a ques-
tion as being the set of its possible answers, abandoning his earlier Postulate 3 (and
the partition arising therefrom). The set of possible answers does not partition possible
worlds because they can overlap—more than one might be true in a given possible
world. The advantage of this view is the ability to derive the set of possible answers from
the syntactic form of the question. The question Who left early? in the scenario given
above has the denotation AnsPoss (5).

(5) AnsPoss: {Adam left early, Betty left early, Christy left early, David left
early}

480
 Q U ESTI O N S

Hamblin (1973) proposed that (a) question words like who denote sets of individuals
(e.g., who denotes {Adam, Betty, Christy, . . .}), and (b) the mechanism of semantic
composition operates on sets. A predicate (set) (like {left}) is applied to an argument
(set) by “distributing” the application of the predicate(s) “pointwise” over the members
of the argument, collecting the results in a set. The derivations of Adam left and Who
left? are sketched below to illustrate.

(6) Adam left.

a. {LEFT}({ADAM})
b. {LEFT(ADAM)}
(7) Who left?
a. {LEFT}({ADAM, BETTY, CHRISTY, DAVID, . . .})
b. {LEFT(ADAM), LEFT(BETTY), LEFT(CHRISTY), LEFT(DAVID), . . .}

A derivation ending with a singleton set has derived a proposition, and a derivation
ending with a multiple-membered set has derived a question. In Hamblin’s system, all
of the work in question formation is done by the question word: In (7), who introduces
a multi-membered set into the derivation, and this set-hood “propagates” out to the end
of the derivation, yielding a set of propositions.
Note that the AnsPoss set in (5) does not contain the proposition nobody left. Under
this analysis, absent some more elaborate story, the response nobody left must be treated
not as a possible answer, but as a rejection of the question.
Hamblin’s (1973) proposed set-based compositional mechanism was not initially
widely used, but has more recently been repopularized through Rooth’s (1985) influen-
tial work on focus semantics, and has since found its way back to analyses of questions
as well (see, e.g., Rullmann and Beck 1997; Hagstrom 1998; Kratzer and Shimoyama
2002; Beck 2006; Cable 2007). The term “Hamblin-semantics” in recent literature gen-
erally refers to the compositional procedure reviewed above. See also Groenendijk and
Roelofsen’s (2009) development of a similar analysis (“inquisitive semantics”) within a
dynamic semantics framework (Chapter 2.7) and with further elaboration concerning
interrogative force and disjunctions.

5.3 Sets of True Answers (Karttunen 1977)

Karttunen (1977) updated Hamblin’s (1973) proposal for Montague’s (1973) PTQ
framework, aiming to account for embedded questions—which Karttunen took to be
primary (reducing matrix questions to embedded questions under an implicit interroga-
tive speech act “I ask you (to tell me)”).
The mechanics of Karttunen’s proposal need not concern us here, apart from noting
that his derivation proceeds by adding a question-forming operator (which we might
syntactically identify as a complementizer) to a proposition, followed by composi-
tion of any question words. This structure lines up well with the popular hypothesis
within generative syntax (Huang 1982) that all question words move to a structur-
ally high position in the syntax prior to interpretation—Karttunen’s proposal can
be viewed as providing a kind of semantic motivation for this syntactic operation.
This synergy may, in part, explain the widespread adoption of Karttunen’s semantic
analysis—but, whatever the reason, Karttunen’s (1977) proposal quickly became the
“standard view.”

481
 PAU L H A G S T R O M

Karttunen’s final denotation for questions differs from Hamblin’s in one essential
respect: While Hamblin’s denotation contains the set of possible answers, Karttunen’s
contains the true answers. This revision is motivated by the interpretation of questions
embedded under verbs like know and tell. If know is given a distributive analysis when
provided with a set of propositions (that is, to know a set of propositions is to know each
of the propositions in the set), the interpretation of (8) indicates that only the true
propositions should be in the denotation of the embedded question. Accordingly, Kart-
tunen’s denotation for the question Who left early? is AnsTrue (9), “the propositions
that jointly constitute a true and complete answer to the question.”

(8) Mr. X knows who left early

(9) AnsTrue: {Christy left early, David left early}

As in Hamblin’s (1973) approach, nobody left early will never be a member of AnsTrue,
although Karttunen anticipated the possibility that AnsTrue might be empty in his
proposed semantics for know (see section 6.8).

5.4 Complete Answers under Partitions

(Groenendijk & Stokhof 1982, Higginbotham 1993)
To AnsPoss and AnsTrue above, we can add AnsExh (10), reflecting Hamblin’s
(1958) partition proposal, where only one member of AnsExh is true in any given pos-
sible world.

(10) AnsExh: {Nobody left early, Only Adam left early, Only Betty left early,
Only Christy left early, Only David left early, Only Adam and Betty left early,
Only Adam and Christy left early, Only Adam and David left early, Only Betty
and Christy left early, Only Betty and David left early, Only Christy and David
left early, Everyone but Adam left early, Everyone but Betty left early, Every-
one but Christy left early, Everyone but David left early, Everyone left early}

More developed partition proposals are outlined by Groenendijk and Stokhof (1982)
and Higginbotham (1993). Groenendijk and Stokhof propose that a question is rep-
resented as a contingent proposition, or a propositional concept. This is a proposition
whose extension in any possible world is the true and complete answer in that world.
When the propositional concept whether Adam left early is evaluated with respect to a
world in which Adam left early, it amounts to the proposition that Adam left early; when
evaluated with respect to a world in which Adam did not leave early, it amounts to the
proposition that Adam did not leave early. This captures rather neatly the fact that Mr. X
knows whether Adam left early means that Mr. X knows the true answer to the question.
Under the partition view, nobody left early has full citizenship among the possible
answers, since it describes one of the cells of the partition of how the world might be.

5.5 Context Dependence in Questions-as-Answers Approaches

In the illustrations of AnsPoss, AnsTrue, and AnsExh above, the answers were lim-
ited to propositions concerning the protagonists and situations in the Tale. This type
of context-sensitivity is a relatively deep component of questions—were the question

482
 Q U ESTI O N S

“Who left early?” posed in the context of the Tale, the response “Vyacheslav Molotov”
does not seem to constitute an answer (despite being true: in 1945, Vyacheslav Molotov
did leave early from a meeting with U.S. President Harry S. Truman).
One way in which context can manifest itself is as a constraint on the domain of indi-
viduals who quantifies over or refers to—perhaps comparable to contextual constraints
on the interpretation of Everybody left (which can be evaluated as true even where not
all of humanity left; see, e.g., von Fintel 1994). Another way that context can play a
role is in ruling out otherwise viable answers; for example, if the context contains a pre-
supposition that exactly one person left early (perhaps contributed by a singular which
phrase, as in which person left early?), then answers entailing that several people left early
cease to be viable.
Integrating context sensitivity into an understanding of questions raises several issues.
Here are two: If a question is defined in terms of its answers and the answers depend on
context, do we conclude that the same interrogative refers to different questions in dif-
ferent contexts? When an otherwise possible answer conflicts with a presupposition in
the context, does it cease to be a possible answer for the purposes of defining the ques-
tion? (Section 6.5 below contains some further discussion of contextual effects.)

5.6 Other Proposals

The analyses outlined above (resulting in AnsPoss, AnsTrue, and AnsExh) are the
most influential and commonly discussed, but of course other proposals have been
advanced as well. Let me briefly mention two, both opposed to the reduction of ques-
tions to answers: Krifka (2001a) argues for a more complex denotation of questions that
can apply directly to fragment answers (denying Hamblin’s (1958) Postulate 1), and
Ginzburg (1995) argues that questions are a basic type and opts for a formalization of
discourse rather than properties of questions per se. Neither proposal will be reviewed
further here, although we will revisit some of the arguments against reducing questions
to answers. Other well-known analyses not even discussed here include Åqvist (1965),
Belnap and Steel (1976) and Hintikka (1976).

6 Basic Data and Analytical Issues

6.1 Complete Answers, Partial Answers, and Irrelevant Statements

An advantage that AnsExh (“partition theories”) has over both AnsPoss and AnsTrue
is that it can distinguish complete answers, partial answers, and irrelevant statements.
In (11), AnsExh (10) partitions the set of possible worlds into 16 cells, each corre-
sponding to a possible combination of leavers and nonleavers.

(11) Who left the party early?

a. Christy and David left the party early, and nobody else did.
b. Adam did not leave the party early.
c. Adam brought his camera to the party.

A complete answer to the question singles out the cell containing the actual world.
Example (11a) is one such proposition; (11b) is not. However, (11b) is a partial answer,
by virtue of ruling out some of the cells—the eight cells in which Adam left early are no

483
 PAU L H A G S T R O M

longer viable candidates for containing the actual world. Example (11c), on the other
hand, does not appear to rule out any cells, making it an irrelevant statement. (Although
note: context can affect this. If it is known that, as a rule, when Adam brings his camera
to the party, he leaves last—perhaps because his spying duties require him to photo-
graph everyone as they leave—then, we could deduce (11b) from (11c)).
Those who discuss complete answers, partial answers, and irrelevant statements are
generally in agreement about the characterizations just given. Although these distinc-
tions are a natural consequence within a partition analysis, it is generally possible to
derive a partition even within a nonpartition analysis (see section 6.8).

6.2 Differentiating Yes–No Questions and Alternative Questions

Karttunen (1977) observed an issue with AnsTrue even as he formulated it: The rep-
resentations of positive and negative yes–no questions are the same. The issue arises
equally with AnsPoss and AnsExh. Under all these approaches, a yes–no question
divides a proposition p from its negation ¬p. Negation in the question will have no
effect on this division. Yet, the intuition is that Did Adam leave early? and Didn’t Adam
leave early? differ.
Krifka (2001b) adds a further observation: Alternative questions where the alter-
natives are p and ¬ p also share this division. That is, all three questions in (12) are
predicted to have the same denotation. Yet these questions do not behave alike. For
example, only (12a) and (12b) admit the responses “yes” and “no”—(12c) requires a
more specified response.

(12) a. Did Christy leave?

b. Didn’t Christy leave?
c. Did Christy leave or didn’t she?

This kind of equivalence is difficult to escape on a partition view. Büring and Gunlogson
(2000) and Romero and Han (2004) discuss aspects of the problem, calling upon Ladd’s
(1981) work on the pragmatic constraints on negative yes–no questions. Both pairs of
authors ultimately provide pragmatic proposals, but there is an option for distinguish-
ing these questions within the semantics proper: The answer set for yes–no questions
contains only the explicit proposition, while that for alternative questions contains
both alternatives ({Christy left} for (12a), {Christy didn’t leave} for (12b), {Christy left,
Christy didn’t leave} for (12c)). Whether such an analysis will ultimately work out
remains to be seen (Büring & Gunlogson argue that it won’t), but if it does provide a
way of extricating ourselves from the equivalence problem, it also positions AnsPoss
ahead of the other possibilities for the basic question denotation.

6.3 Semantic Types and Classes of Embedding Verbs

Verbs that permit an interrogative to be embedded as a complement come in at least
two varieties. Some, like wonder or ask, operate on the question itself, as a question,
based on the meaning of the verb. Others, like know and tell, relate an individual and
a proposition—even when such verbs take an embedded interrogative, the question
is interpreted as its (propositional) answer. A sentence like (13a) puts Mr. X in a
“wondering” relation to a question, but (13b) doesn’t mean that Mr. X told me the

484
 Q U ESTI O N S

question—what Mr. X told me was the answer to the question. Verbs that view a ques-
tion in terms of its answer are also usually able to embed a indicative (14).

(13) a. Mr. X wondered who left the party early.

b. Mr. X told me who left the party early.
(14) Mr. X told me that Christy and David, and nobody else, left the party
early.

Proposition-embedding verbs themselves come in at least two classes, those that are
factive, and those that are not (Kiparsky & Kiparsky 1970). Factive verbs presuppose
the truth of their complement. Baker (1968) observed that verbs like tell, not normally
factive with an indicative complement, become factive with an interrogative comple-
ment. That is, in (14), what Mr. X told me could easily be false, but in (13b), what Mr.
X told me (the answer to the question) is necessarily true. This effect follows naturally
from both the AnsExh and AnsTrue accounts.
One fundamental issue is what place questions (and their answers) should have in
our ontology—what are their semantic types? Equally pressing are the issues of what
(if anything) differentiates the types of matrix questions and embedded questions, and
what (if anything) differentiates embedded questions under wonder from those under
know. Is there a consistent type for questions at all, in either context?
Here are some answers attempted in the literature: Karttunen (1977) takes questions
to be sets of propositions, augmenting the semantics of embedding verbs like know to
convert questions into propositions. Ginzburg (1992) argues that questions should be
considered to be an irreducible basic type. Krifka (2001a) proposes that the type of
a question is not uniform but is determined by the type of a given fragment answer.
Groenendijk and Stokhof (1982) take questions and propositions to have the same type
under know, while questions under wonder are propositional concepts.
In support of this last view, Groenendijk and Stokhof (1982) observe that it is pos-
sible to conjoin interrogatives and indicatives, as in (15). Given the common assump-
tion that only like categories can be conjoined, this puts embedded interrogatives and
embedded indicatives in the same class.

(15) a. John knows that Peter has left for Paris, and also whether Mary has
followed him.
b. Alex told Susan that someone was waiting for her, but not who it
was.

If questions are taken to have a uniform type, the differentiation between questions under
know and under wonder is handled either by augmenting the lexical semantics of the verbs
(adding an operator that converts from the basic type when needed), or by providing
the semantic interpretation procedure with a “type-shifting” function that converts one
type to another on demand when required. This latter class of analyses includes Lahiri’s
(2002) “Interrogative Raising” rule, a syntactic rule that moves an embedded question
when a type clash necessitates it (much as “Quantifier Raising” moves quantifiers).
A related challenge is the existence and interpretation of concealed questions like
(16), where what appears to be a noun phrase in the syntax is interpreted as an identity
question (17). Concealed questions are discussed by, among others, Baker (1968), Heim
(1979), Romero (2005, 2006), Nathan (2006).

485
 PAU L H A G S T R O M

(16) Leslie has forgotten the capital of Vermont.

(17) Leslie has forgotten what the capital of Vermont is.

6.4 Exhaustivity and Mention-Some Readings

Groenendijk and Stokhof (1982) explore the properties embedded questions have in
logical arguments, and observe that both (18) and (19) are intuitively valid arguments
(at least here, where the domain of people is known).

(18) Mr. X knows who left early

Christy left early
Mr. X knows that Christy left early
(19) Mr. X knows who left early
Adam did not leave early
Mr. X knows that Adam did not leave early

Groenendijk and Stokhof observe that only (18) follows from AnsTrue, while both
follow from AnsExh (which they advance as an argument for AnsExh). In the context
above, AnsTrue is weakly exhaustive, allowing the conclusion that Mr. X knows of all
those who left early that they left early, while AnsExh is strongly exhaustive, allowing
the further conclusion that Mr. X also knows of all those who didn’t leave early that
they didn’t leave early.
It is an empirical (and somewhat tricky) question whether our judgments warrant
a strongly exhaustive notion of questions in all cases. Berman (1991) claims that it is
possible for Mr. X to be surprised about who left early without being surprised about who
didn’t leave early. This judgment pattern is (by design) predicted to be impossible under
a strongly exhaustive interpretation like AnsExh—and so, to the extent that such judg-
ments are possible, weakly exhaustive meanings like AnsTrue are still needed (see also
section 6.8).
Another challenge for strong exhaustivity arises when question interpretations seem
even weaker than weakly exhaustive. Hintikka (1976) observed readings of questions
under know that have an “existential” or “mention-some” reading. For example, all that
is required for (20) to be true is that Adam knows the location of some post office—it is
not necessary that Adam knows the location of all post offices.

(20) Adam knows where he can find a post office.

6.5 Resolving Questions and Sufficiency of Answers

To the extent that the denotation of a question is identified with its answers, an issue
arises concerning the extent to which the “completeness” of an answer varies with con-
text. Ginzburg (1992) observes that the conditions under which a question is “resolved”
are affected both by the asker’s goals in considering the question, and aspects of the
asker’s own knowledge state. Boër (1978) and Hintikka (1981) provide related discus-
sion as well.
Two examples of the effects of context are given below, from Ginzburg (1995). The
first example relates resolvedness to the questioner’s knowledge, and the second relates
it to the contextually relevant grain-size.

486
 Q U ESTI O N S

First: Someone asks a conference organizer “Who has been attending talks?”, and
receives in reply a complete list of names. The asker might later characterize this as
either: (a) having been told who was attending these talks (if the asker knows who the
names refer to), or (b) not having been told who was attending these talks (if the asker
is a reporter, for whom the names are not meaningful). Even a completely exhaustive
and true answer may still fail to resolve the question.
Second: A traveler is asked where she is and she responds “Amsterdam.” This answer
indicates either (a) that she knows where she is (if she’s just stepped off an airplane), or
(b) that she doesn’t know where she is (if she’s driving around deep in the city). In the
first scenario, “Amsterdam” is sufficiently precise to resolve the question; in the second
scenario, it is not.
Aloni (2001) approaches these issues within a partition framework by making use of
context-dependent “conceptual cover,” representing alternative ways of conceptualizing
the individuals in a world. This allows influence from a contextually governed “perspec-
tive” while still maintaining a context-invariant semantics for interrogatives. Beck and
Sharvit (2002) and Williams (2000) also call upon contextually determined covers to
divide questions into subquestions, as part of their analysis of the “quantificational varia-
bility effect” (exemplified in sentences like For the most part, Mr. X knows who left the party
early) discussed by Berman (1991), Lahiri (2002), and Groenendijk and Stokhof (1993).
Regardless of whether these particular applications of covers are exactly right, the
notion of a cover seems to have the right character for bringing context-dependency into
semantics. Covers have been productively put to use in other areas, such as in the inter-
pretation of plurals (Higginbotham 1981; Gillon 1987; Schwarzschild 1992) (see also
Chapter 1.5).
The issues concerning “grain-size” in Ginzburg’s (1992) examples may submit to
treatments within situation semantics (as, in fact, in Ginzburg’s own analysis; see also
Kratzer (1989, 2002) and Chapter 5.10).

6.6 Questions with Quantifiers and Quantifying into Questions

Questions featuring a quantifier, such as (21), are often ambiguous, interpreted as one of
(21a) (the individual reading, answered by, e.g., “Professor Smith”), (21b) (the list read-
ing, answered with a list, e.g., “John likes Professor Smith, Pete likes Professor Jones, and
Bill likes Professor Nelson”), or (21c) (the functional reading, answered with a function,
e.g., “his advisor”). The distinction between the individual and list readings feels like a
scope distinction, parallel to that in someone loves everyone—the intuition being that,
in (21a), the question (or question word) takes scope over the quantifier, while in (21b)
the quantifier takes scope over the question.

(21) Which professor does everyone like?

a. Which professor is the x such that, for every person y, y likes x?
b. For every person y, which professor is the x such that y likes x?
c. What is the function f such that, for every person y, y likes f (y)?

The ambiguity of such questions is limited in several ways. One constraint is syntactic,
a consequence of grammatical function. Whereas (21) is ambiguous, (22) is not—it has
only an individual reading. In (21), the quantifier is the subject, while in (22), the ques-
tion word is the subject.

487
 PAU L H A G S T R O M

(22) Which professor likes everyone?

Another constraint is on the quantifier itself. Few quantifiers lead to a list reading of
this kind. Universal quantifiers like everyone do, but few people and nobody do not.
The questions in (23) have no list reading, only individual readings (“caffeine-free
diet soda”) and functional readings (“their least favorite food”). There is some debate
about whether quantifiers like (at least) two, many, and most can generate a list reading;
Chierchia (1993) claims they cannot, Lahiri (2002) claims that they can. But certainly,
monotone decreasing quantifiers like no or few cannot (Lahiri 2002). However, when
this type of question is embedded (24), the restriction on the quantifier type is relaxed
(Szabolcsi 1997).

(23) a. What did few people buy?

b. What did nobody buy?
(24) a. John listed what everybody bought.
b. John listed what most people bought.
c. John listed what few people bought.
d. John listed what nobody bought (only what people sold).

Several authors have noted an issue these interpretations raise: It is not obvious what
quantifying into a question would mean (see, e.g., Chierchia 1993). Quantification is nor-
mally defined over semantic objects that evaluate to a truth value. The interpretation of
“everyone left early” is something like ‘for every person x, the proposition that x left early
is true’. To the extent that a question has a different semantic type (cf. section 6.3), the
quantification rule must be modified if it is to be allowed to take scope over a question.
Several analytical options are open. If questions are themselves propositions (as pro-
posed by Groenendijk & Stokhof 1982), there is no issue. If not, the quantificational
mechanism can be augmented (Higginbotham 1993), or one can reanalyze the list read-
ing such that quantification into questions is not required. This latter route has been
taken in various ways. Pafel (1999) proposes splitting the semantic components finely
enough that a quantifier can take scope above a question word but still below the ques-
tion-forming operator. Berman (1991) and Chierchia (1991), relying on the fact that
functional readings do not require the quantifier to take scope over the question, derive
list readings from functional readings (taking a list to be essentially an explicit func-
tion). Krifka (2001b) analyzes list readings as conjoined interrogative speech acts, thus
not resulting from quantification in the usual sense.

6.7 Knowledge and Inquiry

Arguments between approaches to questions often turn on intuitions about questions
embedded under know, which raises the possibility that the thorny issues in character-
izing “knowledge” could be muddying the waters. Cursory comments follow concerning
some recent philosophical work that explicitly addresses semantic issues.
Stanley and Williamson (2001) discuss “knowledge-that” and “knowledge-how,” argu-
ing that the latter should be reduced to the former, building on some of the same models
of questions we have seen in this chapter.
Schaffer (2007, 2009) casts a critical eye on semantic proposals that reduce “knowl-
edge-wh” to “knowledge-that.” Schaffer’s particular objection is that two distinct

488
 Q U ESTI O N S

questions might be “convergent” (have same true answer), predicting no difference in

meaning when embedded under know. Schaffer concludes that the representation of
knowledge must include not only propositions but also the questions they answer. How-
ever, Aloni and Egré (2010) and Aloni et al. (2009) demonstrate that understanding
questions in terms of AnsExh (perhaps with the addition of conceptual covers, section
6.5) prevents convergent questions of this sort from arising. With respect to what con-
stitutes knowledge, Kratzer’s (2002) discussion of the distinction between facts and true
propositions provides an accessible and recent starting point.
Another area where the semantics and logic of questions intersect with traditional
philosophical pursuits is in the examination of the role questions and answers play in
scientific inquiry; see discussions by Hintikka (1981), Bromberger (1992), and Colling-
wood (1940). Here, the focus is on the adequacy of answers in advancing theories, what
counts as an explanation, and what types of questions are susceptible to explanation-
type answers. This work raises the issue of whether a reduction of questions to AnsPoss,
AnsTrue, or AnsExh is compatible with situations where the possible answers, or even
the forms they might take, are unknown (cf. Bromberger’s (1965) “p-predicament”).
In general, such considerations indicate that membership in the set of possible answers
must be determined by procedure or rule—that a candidate proposition can be identified
as being or not being a possible answer upon encountering it.

6.8 Intertranslatability
Adjudicating between the approaches to questions we’ve seen is not an easy task; each
has its own motivations and potential shortcomings, and compelling arguments can be
made that reach seemingly conflicting solutions. One might even conclude that more
than one representation is necessary. This section focuses on a proposal outlined by
Heim (1994) (see also Lahiri 2002) for deriving AnsExh from AnsTrue.
Karttunen (1977) noted a special case in his construction of AnsTrue: If AnsTrue
is empty, knowledge of AnsTrue is to be translated as knowledge that AnsTrue is
empty (rather than as, say, no relevant knowledge). Heim (1994) observes that general-
izing this special case, such that knowledge of AnsTrue always translates to knowledge
that AnsTrue is whatever it is, has an effect that is in almost every respect equivalent
to knowledge of AnsExh. Hence, AnsExh need not be the basic question denotation,
as it can be derived as needed in the context of know to capture the strongly exhaustive
intuitions (reviewed in section 6.4). In Heim’s words: “Maybe one never really bears
the know-relation to a question unless one also knows that the answer to this question
is the answer to it.”
Heim proceeds to show that the path between AnsTrue and AnsExh is one-direc-
tional—AnsExh can be derived from AnsTrue, but not vice versa. This means that
AnsTrue is genuinely richer, but is this additional richness needed? Verbs like surprise
might constitute such a case (if one can be surprised at who left but not at who didn’t
leave).

7 Concluding Remark
Primarily, the discussion has outlined the ways in which a question can be viewed in
terms of its answers and the ways in which this relation can be made explicit, with a

489
 PAU L H A G S T R O M

taste of the advantages and disadvantages of each. This conception remains dominant
and common to investigations into questions—though, certainly, arguments have been
advanced against reducing questions to their answers.
A great deal more can be said about the topics touched upon here (and about those
that haven’t been), but this sketch of selected issues relevant to the contemporary study
of questions should provide a foothold for approaching the primary literature.

References
Aloni, M. (2001), Quantification under conceptual covers, PhD thesis, University of Amsterdam. Aloni,
M. and Egré, P. (2010), ‘Alternative questions and knowledge attributions’, The Philosophical Quarterly
60(238), 1–27.
Aloni, M., Égré, P. and de Jager, T. (2009), ‘Knowing whether A or B’, Synthese pp. 1–27. Online, available
http://www.springerlink.com/content/g672737l15732087/fulltext.pdf.
Baker, C. L. (1968), Indirect questions in English, PhD thesis, University of Illinois.
Beck, S. (2006), ‘Intervention effects follow from focus interpretation’, Natural Language Semantics 14(1),
1–56.
Beck, S. and Sharvit, Y. (2002), ‘Pluralities of questions’, Journal of Semantics 19, 105–157.
Belnap, Jr., N. D. and Steel, Jr., T. B. (1976), The logic of questions and answers, Yale University Press, New
Haven.
Berman, S. (1991), On the semantics and logical form of WH-clauses, PhD thesis, University of Massachu-
setts, Amherst.
Boër, S. E. (1978), ‘’Who’ and ’whether’: Towards a theory of indirect question clauses’, Linguistics and Phi-
losophy 2, 307–345.
Bromberger, S. (1965), An approach to explanation, in R. J. Butler, ed., ‘Analytical philosophy’, 2nd series,
Basil Backwell, Oxford.
Bromberger, S. (1992), On what we know we don’t know: Explanation, theory, linguistics, and how questions
shape them, University of Chicago Press; Center for the Study of Language and Information, Chicago;
Stanford.
Büring, D. and Gunlogson, C. (2000), ‘Aren’t positive and negative polar questions the same?’, Ms., UCSC.
Cable, S. (2007), The grammar of Q: Q-particles and the nature of wh-fronting, as revealed by the wh-ques-
tions of Tlingit, PhD thesis, Massachusetts Institute of Technology.
Chierchia, G. (1991), Functional WH and weak crossover, in D. Bates, ed., ‘Proceedings of the Tenth West
Coast Conference on Formal Linguistics’, CSLI, Stanford, pp. 75–90.
Chierchia, G. (1993), ‘Questions with quantifiers’, Natural Language Semantics 1, 181–234.
Collingwood, R. G. (1940), An essay on metaphysics, Clarendon Press, Oxford.
Dekker, P., Aloni, M. and Butler, A. (2007), The semantics and pragmatics of questions, in M. Aloni, A.
Butler and P. Dekker, eds, ‘Questions in dynamic semantics’, Elsevier, Amsterdam, pp. 1–40.
Gillon, B. (1987), ‘The readings of plural noun phrases in English’, Linguistics and Philosophy 10, 199–219.
Ginzburg, J. (1992), Questions, queries, and facts: A semantics and pragmatics for interrogatives, PhD thesis,
Stanford University, Stanford, CA.
Ginzburg, J. (1995), ‘Resolving questions, I’, Linguistics and Philosophy 18, 459–527.
Ginzburg, J. (1996), Interrogatives: Questions, facts and dialogue, in S. Lappin, ed., ‘The handbook of contem-
porary semantic theory’, Blackwell, Oxford, pp. 385–422.
Groenendijk, J. and Roelofsen, F. (2009), Inquisitive semantics and pragmatics, in J. M. Larrazabal and L.
Zubeldia, eds, ‘Meaning, content and argument: Proceedings of the ILCLI International Workshop on Seman-
tics, Pragmatics and Rhetoric’, University of the Basque Country Press.
Groenendijk, J. and Stokhof, M. (1982), ‘Semantic analysis of wh-complements’, Linguistics and Philosophy
5, 175–233.
Groenendijk, J. and Stokhof, M. (1993), Interrogatives and adverbial quantification, in K. Bimbo and A.
Mate, eds, ‘Proceedings of the Fourth Symposium on Logic and Language’, Aron Publishers, Budapest,
pp. 1–29.
Groenendijk, J. and Stokhof, M. (1997), Questions, in J. van Benthem and A. ter Meulen, eds, ‘Handbook of
Logic and Language’, Elsevier/MIT Press, Amsterdam/Cambridge, MA.
Hagstrom, P. (1998), Decomposing questions, PhD thesis, Massachusetts Institute of Technology.

490
 Q U ESTI O N S

Hagstrom, P. (2003), ‘What questions mean’, Glot International 7, 188–201.

Hamblin, C. (1958), ‘Questions’, Australasian Journal of Philosophy 36, 159–168.
Hamblin, C. (1973), ‘Questions in Montague English’, Foundations of Language 10, 41–53.
Harrah, D. (2002), The logic of questions, in D. Gabbay and F. Guenthner, eds, ‘The handbook of philosophical
logic’, 2nd edition, Vol. 8, Kluwer Academic Publishers, Dordrecht, pp. 1–60.
Heim, I. (1979), Concealed questions, in R. Bauerle, U. Egli and A. von Stechow, eds, ‘Semantics from dif-
ferent points of view’, Springer, Berlin.
Heim, I. (1994), Interrogative semantics and Karttunen’s semantics for know, in R. Buchalla and A. Mit-
twoch, eds, ‘Proceedings of IATL 1’, Akademon, Jerusalem, pp. 128–144.
Higginbotham, J. (1981), ‘Reciprocal interpretation’, Journal of Linguistic Research 3, 97–117.
Higginbotham, J. (1993), Interrogatives, in K. L. Hale and S. J. Keyser, eds, ‘The view from building 20: Essays
in honor of Sylvain Bromberger’, MIT Press, Cambridge, MA, pp. 195–227.
Higginbotham, J. (1996), The semantics of questions, in S. Lappin, ed., ‘The handbook of contemporary seman-
tic theory’, Blackwell, Oxford, pp. 361–383.
Hintikka, J. (1976), ‘The semantics of questions and the questions of semantics’, Acta Philosophica Fennica
28(4).
Hintikka, J. (1981), ‘On the logic of an interrogative model of scientific inquiry’, Synthese 47(1), 69–83.
Huang, C. T. J. (1982), Logical relations in Chinese and the theory of grammar, PhD thesis, Massachusetts
Institute of Technology.
Karttunen, L. (1977), ‘Syntax and semantics of questions’, Linguistics and Philosophy 1, 3–44.
Kiparsky, P. and Kiparsky, C. (1970), Fact, in M. Bierwisch and K. Heidolph, eds, ‘Progress in Linguistics’,
Mouton, The Hague, pp. 143–173.
Kratzer, A. (1989), ‘An investigation of the lumps of thought’, Linguistics and Philosophy 12, 607–653.
Kratzer, A. (2002), ‘Facts: Particulars or units of information’, Linguistics and Philosophy 25, 655–670.
Kratzer, A. and Shimoyama, J. (2002), ‘Indeterminate pronouns: The view from Japanese’, Paper presented
at the 3rd Tokyo Conference on Psycholinguistics.
Krifka, M. (2001a), For a structured account of questions and answers, in C. Féry and W. Sternefeld, eds,
‘Audiatur vox sapientiae’, Akademie Verlag, Berlin, pp. 287–319.
Krifka, M. (2001b), ‘Quantifying into question acts’, Natural Language Semantics 9, 1–40.
Ladd, D. R. (1981), A first look at the semantics and pragmatics of negative questions and tag questions, in
‘Proceedings of the Chicago Linguistic Society (CLS)’, Vol. 17, pp. 164–171.
Lahiri, U. (2002), Questions and answers in embedded contexts, Oxford University Press, Oxford.
Montague, R. (1970), English as a formal language, in B. Visentini et al., eds, ‘Linguaggi nella società e nella
tecnica’, Edizioni di Comunità, Milan, pp. 189–224.
Montague, R. (1973), The proper treatment of quantification in ordinary English, in J. Hintikka, J. Moravc-
sik and P. Suppes, eds, ‘Approaches to natural language’, Reidel, Dordrecht, pp. 221–242.
Montague, R. (1974), Formal philosophy: Selected papers of Richard Montague, Yale University Press, New
Haven.
Nathan, L. (2006), On the interpretation of concealed questions, PhD thesis, Massachusetts Institute of
Technology.
Pafel, J. (1999), ‘Interrogative quantifiers within scope’, Linguistics and Philosophy 22, 255–310.
Portner, P. and Partee, B., eds (2002), Formal semantics: The essential readings, Blackwell Publishing,
Oxford.
Prior, M. and Prior, A. (1955), ‘Erotetic logic’, The Philosophical Review 64(1), 43–59.
Romero, M. (2005), ‘Concealed questions and specificational subjects’, Linguistics and Philosophy 28(6),
687–737.
Romero, M. (2006), On concealed questions, in J. Howell and M. Gibson, eds, ‘Proceedings of SALT 16’,
CLC Publications, Ithaca, NY.
Romero, M. and Han, C. (2004), ‘On negative yes/no questions’, Linguistics and Philosophy 27, 609–658.
Rooth, M. (1985), Association with focus, PhD thesis, University of Massachusetts, Amherst.
Rullmann, H. and Beck, S. (1997), Reconstruction and the interpretation of which-phrases, in G. Katz, S.-
S. Kim and H. Winhart, eds, ‘Reconstruction: Proceedings of the 1997 Tübingen Workshop’, Vol. 127,
Arbeitspapiere des Sonderforschungbereichs 340, pp. 223–256.
Schaffer, J. (2007), ‘Knowing the answer’, Philosophy and Phenomenological Research 75(2), 383–403.
Schaffer, J. (2009), ‘Knowing the answer redux: Replies to Brogaard and Kallestrup’, Philosophy and Phenom-
enological Research 78(2), 477–500.
Schwarzschild, R. (1992), ‘Types of plural individuals’, Linguistics and Philosophy 15, 641–675.

491
 PAU L H A G S T R O M

Stanley, J. and Williamson, T. (2001), ‘Knowing how’, Journal of Philosophy 98(8), 411–444.
Szabolcsi, A. (1997), Quantifiers in pair-list readings, in A. Szabolcsi, ed., ‘Ways of scope taking’, Kluwer,
Dordrecht, pp. 311–347.
von Fintel, K. (1994), Restrictions on quantifier domains, PhD thesis, University of Massachusetts,
Amherst.
Whately, R. (1854), Elements of logic, Morton & Griswold, Louisville, KY.
Whately, R. (1877), Elements of rhetoric: Comprising an analysis of the laws of moral evidence and of persuasion
with rules for argumentative composition and elocution, Harper & Brothers, New York.
Williams, A. (2000), Adverbial quantification over (interrogative) complements, in R. Billerey and B. D.
Lillehaugen, eds, ‘Proceedings of the 19th West Coast Conference on Formal Linguistics’, Cascadilla Press,
Somerville, MA, pp. 547–587.
Åqvist, L. (1965), A new approach to the logical theory of interrogatives, Filosofiska Föreningen, Uppsala,
Sweden.

492
 Section IV

M ET H O D O L O G Y
 4.1
THE R OLE OF
E X P E R IM ENT
Edouard Machery and Stephen Stich

In this chapter, we argue that philosophers of language should emulate linguists, who
are increasingly replacing the traditional informal reliance on their own and their
colleagues’ intuitions with systematic experimental surveys of ordinary speakers’ intui-
tions. This experimental turn has enabled linguists to alleviate some significant con-
cerns about their methodology, and a similar experimental turn in the philosophy of
language would, we believe, enable philosophers to alleviate similar concerns about
their methodology.
Here is how we will proceed. In Section 1, we will describe the experimental turn in
generative linguistics, and argue that a similar methodological shift is called for in the
philosophy of language. In Section 2, we review some of the burgeoning experimental
work in the philosophy of language, which illustrates both the need for an experimental
approach in the philosophy of language and the benefits to be gained. In the remainder
of the chapter, we focus on the most well-known experimental work in the philosophy
of language: the cross-cultural studies of intuitions about the reference of proper names.
In Section 3, we address an influential objection against this work. In Section 4, we
argue that an experimental turn in the philosophy of language is bound to have some
significant philosophical implications, illustrating this claim with the implications of
the cross-cultural work on intuitions about reference.

1 Emulating the Experimental Turn in

Generative Linguistics

1.1 The Experimental Turn in Generative

Linguistics
Following Chomsky (1957), most syntacticians in the generative tradition rely on a
familiar methodology: develop theories that accommodate judgments about linguisti-
cally significant properties of sentences and their parts. While various sorts of judgments
have been used, judgments about the acceptability of sentences have played a central
role in generative syntax. Because the judgments that serve as evidence for syntactic
theories are typically made quickly without being consciously inferred (they spring to
mind, as it were), they are often called “intuitions.” Syntactic theories in generative
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

linguistics are evaluated on the basis of their consistency with these intuitions: they
are supported if they entail that an intuitively acceptable construction is grammati-
cal or that an intuitively unacceptable construction is not grammatical, and they are
undermined if they entail that an intuitively unacceptable construction is grammatical
or that an intuitively acceptable construction is not grammatical. Theories are also
challenged if they entail that some sentence can be read in two different ways while it is
intuitively judged to have a single reading, and in a variety of other ways. Naturally, just
as scientists sometimes discard some empirical findings, generative syntacticians also
recognize that inconsistencies between a theory and some judgments can be explained
away by assuming that for various reasons the judgments are erroneous; these discounted
judgments are often called “performance errors.” For instance, one might fail to notice
one of the readings of a genuinely ambiguous sentence or one might find a grammatical
construction (for example, a garden-path sentence) unacceptable because one fails to
parse it in the right way.
Until the 1990s, linguists studying a particular syntactic phenomenon typically relied
on their own intuitions and on the intuitions of other linguists. For instance, in The
Logical Structure of Linguistic Theory, Chomsky argues that in English the position of
verb particles is determined not by the length of the sentential constituents, but by their
complexity. He writes (1975: 477):

While . . . both “the detective brought in the suspect” and “the detective
brought the suspect in” are grammatical, in general the separability of the
preposition is determined by the complexity of the NP object. Thus we could
scarcely have “the detective brought the man who was accused of having stolen
the automobile in.” It is interesting to note that it is apparently not the length
in words of the object that determines the naturalness of the transformation,
but, rather, in some sense, its complexity. Thus “they brought all the leaders
of the riot in” seems more natural than “they brought the man I saw in.” The
latter, though shorter, is more complex.

For present purposes, the important point is that Chomsky supports his hypothesis by
appealing to his own intuitions, assuming that these are representative of the intuitions
of lay competent speakers of English. (For empirical evidence that this assumption is
not entirely correct, see Wasow & Arnold 2005).
In recent years, this feature of the methodology of generative linguistics has been
extensively criticized. Schütze, in his landmark study of the methods of generative
linguistics, The Empirical Base of Linguistics (1996), made a powerful case that rely-
ing on syntacticians’ intuitions is methodologically inappropriate. Although eliciting
the intuitions of ordinary competent speakers can be fraught with difficulties (e.g.,
Schütze 2005; Fanselow 2007), experimental work on linguists’ and ordinary com-
petent speakers’ intuitions has not shown that the former are more reliable than the
latter, provided that ordinary speakers understand the nature of the task (Schütze
1996; Culbertson & Gross 2009). Thus, syntacticians’ reliance on their own intuitions
and those of their colleagues cannot be justified on the grounds that their expertise
make these intuitions more reliable. At best, this long-standing feature of their meth-
odology could be justified on pragmatic grounds since it is easier for syntacticians to
consult their own and their colleagues’ intuitions than to survey the intuitions of ordi-
nary competent speakers. This would be entirely reasonable if syntactians’ reliance on

496
 TH E RO LE O F EX P E R I ME N T

their own and on their colleagues’ intuitions did not have important drawbacks. But
that is not the case. First, and most obviously, syntacticians’ theoretical commitments
risk influencing their intuitions, undermining the evidential role of these intuitions.
Several examples discussed by Schütze show that this worry is not merely speculative
(see also Labov 1975). Lasnik and Saito (1984) assert that sentences such as “Why
do you think that he left?” are ambiguous; Aoun et al. (1987) claim that they are not.
Unsurprisingly, the ambiguity of this type of sentence follows from Lasnik and Saito’s
theory, but not from Aoun et al.’s theory! Second, and equally important, syntacti-
cians’ appeal to their own and to their colleagues’ intuitions can blind them to the
dialectical variation within a given language and biases their understanding of the var-
iation in syntactic properties. Thus, for example, some linguists (for example, Chom-
sky 1976) have held that the contraction “wanna” is unacceptable in sentences such
as (1) because “who” leaves an unpronounced trace after “want” when it is fronted
from its position in (2):

(1) Who did you wanna meet your parents?

(2) You wanted who to meet your parents?

However, some ordinary speakers find sentences such as (1) perfectly acceptable, sug-
gesting that in this respect the grammar of English may vary across individuals (Labov
1996; Wasow & Arnold 2005; but see Phillips 2009 for discussion). Worse, syntactians’
intuitions might be entirely unrepresentative of the language they speak. For instance,
a number of claims made about idioms have been shown to be inaccurate (Nunberg et
al. 1994; Riehemann 2001; for a review of the intricate debate about the “wanna” con-
traction, see Goodall 2006; see Featherston 2007 and Riemer 2009 for other examples).
Though Jackendoff (1997) claimed that the idiom “raise hell” is syntactically inflex-
ible and cannot be turned into a passive construction, Riehemann (2001) showed that
passive versions of “raise hell” are actually often used (discussed in Wasow and Arnold
2005: 1487–8). Finally, Wasow and Arnold (2005: 1485–6; see also Sampson 2005)
suggest that syntacticians’ reliance on their own and their colleagues’ intuitions pre-
vents them from paying attention to easily available counterevidence, perhaps because
of a confirmation bias of some sort.
Linguists and philosophers have not only criticized syntacticians’ reliance on their
own and their colleagues’ intuitions, they also have put forward an alternative meth-
odology: the careful survey of the intuitions of ordinary competent speakers (Bard et al.
1996; Schütze 1996; Cowart 1997; Marantz 2005; Sorace & Keller 2005; Featherson
2007). Such intuitions are not contaminated by linguistic theories. By surveying them
linguistics avoid relying on intuitions which may be unrepresentative and they are more
likely to detect dialectical variation within a linguistic community. Furthermore, as
even critics of this alternative methodology have acknowledged (Phillips 2009), sur-
veying a large number of ordinary speakers has a number of methodological advan-
tages (for discussion of some of these, see Featherston 2007; Myers 2009). For instance,
aggregating a large number of acceptability ratings is likely to be the only reliable way
to identify subtle differences in acceptability (Bard et al. 1996; Sorace & Keller 2005;
Sprouse 2007).
For these and other reasons, Schütze issues the following recommendation
(1996: 212):

497
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

Linguistics has much to gain and nothing to lose by taking data collection, and
particularly judgment collection, much more seriously, both with regard to the
insights that will be gained and the theoretical issues that will be clarified, and
with regard to the standing of the field as a scientific endeavor in the larger
academic setting. The realization seems to be growing that the psychology of
grammaticality judgments can no longer be ignored.

This methodological recommendation has been positively received among genera-

tive syntacticians, who rely increasingly on experimental surveys of people’s intuitions
(Schütze 2005; Myers 2009). Semanticists also have started using experimental meth-
ods (Schütze 2005; for an example, see Pietroski et al. 2009), while the field of experi-
mental pragmatics is burgeoning (Noveck & Reboul 2008).
Work on Picture Noun Phrases (PNPs) provides a good illustration (Keller & Asu-
deh 2001; see also Featherson (2005) and Haider (2007) on that-traces in German).
While binding theory originally predicted that anaphors can be bound when pronouns
can’t, and vice versa (Chomsky 1981), PNPs such as (3) and (4) appear to show that
this is not the case:

(3) Hannai found a picture of heri.

(4) Hannai found a picture of herselfi.

To deal with this type of PNP, various modifications of binding theory have been pro-
posed, including the hypothesis that coreference in some constructions (particularly,
some PNPs) is governed by pragmatic considerations rather than by syntactic or struc-
tural features. Attempting to decide between the proposed hypotheses, Keller and Asu-
deh (2001) found that the acceptability judgments made by lay speakers of English were
at odds with all the approaches to coreference in PNPs. Contrary to the contention of
pragmatic theories, coreference in PNPs turns out to be influenced by structural fea-
tures, but in an entirely unexpected manner. Keller and Asudeh conclude (2001: 488)
that “contrary to all that has been written in the syntactic literature, anaphors can be
bound by the subject even in PNPs with possessors.”

1.2 Moral for Philosophers of Language

The similarities between the typical methods in the philosophy of language and in
generative linguistics are striking. In both fields, intuitions are an important source of
evidence, and philosophers overwhelmingly rely on their own intuitions and those of
their colleagues much as generative syntacticians did until the 1990s. We believe that
the three reasons we have discussed that have led a growing number of syntactians to
replace reliance on their own and their colleagues’ intuitions with carefully designed
experimental surveys of the intuitions of ordinary competent speakers carry over to the
philosophy of language straightforwardly.

(1) Despite pronouncements about the superiority of the intuitions of phi-

losophers of language over those of ordinary people (Ludwig 2007; Devitt
2011), there is in fact no empirical evidence supporting this claim.
(2) Philosophers’ theoretical commitments risk influencing their intuitions.
(3) Philosophers’ reliance on their own intuitions risk blinding them to the

498
 TH E RO LE O F EX P E R I ME N T

diversity of the relevant intuitions (see next section for some robust evi-
dence supporting this claim) or to the fact that their own intuitions are
unrepresentative.
(Stich 2010)

Thus there is a strong case to be made for philosophers of language to follow the lead
of generative syntacticians and to replace the appeal to their own and their colleagues’
intuitions with systematic empirical studies of ordinary speakers’ intuitions. In the next
section, we review some work that has done just that.

2 Empirical Research in the Philosophy of Language

In this section, we review the burgeoning experimental work in the philosophy of lan-
guage. We believe that this work vindicates our concerns about traditionalists’ methods
in the philosophy of language, and illustrates the need for an experimental turn in this
field too.

2.1 Cross-Cultural Variation in Intuitions about Reference

As noted in Section 1, syntacticians are replacing the reliance on their own and on their
colleagues’ intuitions with systematic surveys of the intuitions of competent speakers in
part on the grounds that, by relying on the intuitions of a small segment of humanity,
they might be led to ignore the diversity of people’s intuitions and what that diversity
reveals about the language they speak. As we now show, much the same concern applies
to the philosophy of language.
Philosophers of language have long been interested in explaining how words (proper
names, natural kind terms, etc.) refer, and existing theories of reference can be usefully
classified into two distinct types: descriptivist theories and causal-historical theories.
Focusing on proper names, descriptivist theories propose that a name refers to what-
ever individual best satisfies the description competent speakers associate with this
name (Searle 1958; Jackson 1998). Causal-historical theories propose that a proper
name refers to the individual the proper name was introduced to refer to, provided that
there is an appropriate causal chain connecting the introduction of the term with cur-
rent users of the term. This is the case, causal-historical theories maintain, whether or
not the individual satisfies the description competent speakers associate with the name
(Kripke 1972/1980; Devitt 1981).
To decide between competing theories of reference, philosophers of language have
relied on what Machery et al.(2004) have called the method of cases (for discussion, see
Deutsch 2010 and Devitt 2011). The method of cases appeals to the fact that people—
philosophers, linguists, and ordinary speakers alike—appear to have intuitions about
what words refer to in actual and possible situations. Appealing to these intuitions, the
method of cases proposes that the correct theory of reference for words of type T is the
one which is best supported by competent speakers’ intuitions about the reference of
words of type T in actual and possible situations.
To see more concretely how the method of cases works, we briefly review one of
Kripke’s most widely discussed cases. In this case (“the Gödel case”), a name is widely
associated with a description that is false of the original bearer of that name, a, but true
of some other person, b. Because descriptivist theories of reference hold that a name

499
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

refers to the individual that best satisfies the description competent speakers associate
with it, descriptivist theories entail that the name in Kripke’s example refers to b, the
satisfier of the description. But, Kripke maintains, this is just wrong.

Suppose that Gödel was not in fact the author of [Gödel’s] theorem. A man
called “Schmidt,” whose body was found in Vienna under mysterious circum-
stances many years ago, actually did the work in question. His friend Gödel
somehow got hold of the manuscript and it was thereafter attributed to Gödel.

Now Kripke asks us to consider an “ordinary man” who is well educated in mathematics
and can give an account of the incompleteness theorem. But, like many other people,
practically the only thing he has heard about Gödel is that he discovered the incom-
pleteness of arithmetic.

On the [descriptivist] view . . . when our ordinary man uses the name “Gödel,”
he really means to refer to Schmidt, because Schmidt is the unique person sat-
isfying the description “the man who discovered the incompleteness of arith-
metic.” . . . But it seems we are not. We simply are not.
(Kripke 1972/1980: 83–4)

In contrast, causal-historical theories of reference are consistent with the intuition that
“Gödel” continues to refer to its original bearer, a, because a is the person causally-
historically linked with contemporary uses of the name. It is clear that most philosophers
share Kripke’s intuition about this case (and other cases), and even contemporary descrip-
tivists acknowledge that these intuitions have falsified traditional forms of descriptivism
(Evans 1973; Jackson 1998).
But how widely shared are these intuitions by competent speakers who are not profes-
sional philosophers? On the basis of Nisbett and colleagues’ findings about the varia-
tion of cognitive styles across cultures (Nisbett 2003), Machery and colleagues (2004)
hypothesized that the intuition elicited by the Gödel case might differ between Eastern
and Western cultures. To test this hypothesis, they constructed two vignettes, “the
Gödel vignette” and “the Tsu Ch’ung Chih vignette.” The Gödel vignette closely fol-
lowed Kripke’s own wording of the case:

Suppose that John has learned in college that Gödel is the man who proved
an important mathematical theorem, called the incompleteness of arithmetic.
John is quite good at mathematics and he can give an accurate statement of
the incompleteness theorem, which he attributes to Gödel as the discoverer.
But this is the only thing that he has heard about Gödel. Now suppose that
Gödel was not the author of this theorem. A man called “Schmidt,” whose
body was found in Vienna under mysterious circumstances many years ago,
actually did the work in question. His friend Gödel somehow got hold of the
manuscript and claimed credit for the work, which was thereafter attributed to
Gödel. Thus, he has been known as the man who proved the incompleteness
of arithmetic. Most people who have heard the name “Gödel” are like John;
the claim that Gödel discovered the incompleteness theorem is the only thing
they have ever heard about Gödel. When John uses the name “Gödel,” is he
talking about:

500
 TH E RO LE O F EX P E R I ME N T

(A) the person who really discovered the incompleteness of arithmetic?

or
(B) the person who got hold of the manuscript and claimed credit for the
work?

The Tsu Ch’ung Chih vignette had the same structure, but it used names of Chinese
individuals (Machery et al., 2004: B9–10). Participants were undergraduates at Rutgers
(U.S.A.) and at the University of Hong-Kong, an English-speaking university (for more
detail, see Machery et al. 2004).
As Machery and colleagues had predicted, Chinese participants were significantly
more likely than American participants to have descriptivist intuitions—that is, intui-
tions in line with what descriptivist theories of reference say the relevant proper names
refer to (see Figure 4.1.1). In fact, most Chinese turned out to have descriptivist intui-
tions, while most Americans turned out to have causal-historical intuitions—intuitions
in line with what causal-historical theories of reference say the relevant proper names
refer to. Machery and colleagues also found a surprising amount of within-culture vari-
ation. For instance, while a majority of American participants had causal-historical
intuitions, about 40 percent had descriptivist intuitions. Machery et al. (forthcoming)
have replicated these findings, while Machery et al. (2010) have shown that Chinese
participants have similar intuitions when the vignette is written in English and when it
is written in Cantonese.

2.2 Intuitions about Truth-Value

Following Kripke’s own wording of the Gödel case and the standard practice among
philosophers working on reference (for example., Evans 1973), Machery et al. (2004)
examined people’s intuitions about the reference of a proper name—namely intuitions
expressed by sentences such as “When x uses the name y, he is talking about . . .” or “y
refers to . . .”. Other intuitions, such as intuitions about the truth-value of sentences
containing names, are also relevant for theories of reference (for discussion, see, for

100%
Proportion of Causal-Historical Intuitions

Westerners
90%
Easterners
80%
70%
58%
60% 55%
50%
40%
32%
29%
30%
20%
10%
0%
Gödel Case Tsu Ch'ung Chih Case

Figure 4.1.1 Results from Machery et al. (2004)

501
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

example, Martí 2009 and Machery 2011). Machery et al. (2009) examined whether
intuitions about the truth-value of sentences containing names and intuitions about the
reference of these names diverge or, rather, whether they are congruent. They presented
participants from three countries (Mongolia, India, France) with one of the follow-
ing two vignettes, and Machery and Olivola have recently done the same thing with
American participants (unpublished data):

Ivy is a high school student in Hong Kong. In her astronomy class, she was
taught that Tsu Ch’ung Chih was the man who first determined the precise
time of the summer and winter solstices. But, like all her classmates, this is
the only thing she has heard about Tsu Ch’ung Chih. Now suppose that Tsu
Ch’ung Chih did not really make this discovery. He stole it from an astrono-
mer who died soon after making the discovery. But the theft remained entirely
undetected and Tsu Ch’ung Chih became famous for the discovery of the pre-
cise times of the solstices. Everybody is like Ivy in this respect; the claim that
Tsu Ch’ung Chih determined the solstice times is the only thing people have
heard about him. Having read the above story and accepting that it is true,
when Ivy says, “Tsu Ch’ung Chih was a great astronomer,” do you think that
her claim is: (A) true or (B) false?

The second vignette was identical except for the question, which was:

Having read the above story and accepting that it is true, when Ivy uses the
name “Tsu Ch’ung Chih,” who do you think she is actually talking about:
(A) the person who (unbeknownst to Ivy) really determined the solstice
times?
or
(B) the person who is widely believed to have discovered the solstice times,
but actually stole this discovery and claimed credit for it?

Both vignettes are inspired by Kripke’s Gödel case, and they are similar to the Tsu
Ch’ung Chih vignette used in Machery et al. (2004).
Two findings emerge from this study (see Figure 4.1.2). First, in four different coun-
tries, which vary tremendously in terms of culture, the proportion of causal-historical
intuitions is similar when participants are asked about the truth-value of a sentence
containing a proper name or about the reference of this name, suggesting that people’s
intuitions about the truth-value of sentences containing proper names track their intui-
tions about the reference of proper names. When someone judges that a proper name
x refers to a particular individual y in a fictional case, her intuition about the truth of
a sentence containing x in this case seems to depend on what is said about y. Second,
the study found further cross-cultural variation in the intuitions that are relevant for
studying reference. Consistent with previous work, Americans are more likely to have
causal-historical intuitions about the reference of proper names than people from Mon-
golia, India, and France when asked about what a proper name refers to, and they were
more likely than French people to have causal-historical intuitions when asked about
the truth-value of a sentence containing proper names.

502
 TH E RO LE O F EX P E R I ME N T

100%

Proportion of causal-historical intuitions

90% Speaking about
Truth-value
80%

70% 67% 67% 66%

64%
60% 58% 56% 56%

50% 48%

40%

30%
20%
10%
0%
USA India France Mongolia

Figure 4.1.2 Results from Machery et al. (2009)

2.3 Externalist Intuitions

Empirical work on intuitions relevant to the philosophy of language is not restricted to
intuitions about proper names. Jylkkä et al. (2009) have examined whether lay speakers
have externalist intuitions (see also Genone and Lombrozo forthcoming; for relevant
work in psychology, see Braisby et al.1996; Hampton et al. 2007). Externalism about
natural kind terms is, roughly, the view that the reference of natural kind terms (for
example, “gold” and “cat”) is determined by facts about substances and kinds about
which speakers might have no knowledge (such as the essence of chemical kinds), while
internalism is, roughly, the view that the reference of natural kind terms is determined
by the descriptions speakers associate with these terms (Putnam 1975; for review, see
Wikforss 2008). Furthermore, some externalists hold that externalism is true in virtue
of the way competent speakers use natural kind terms: A natural kind term t refers in an
externalist manner if and only if competent speakers intend t to refer to a class of enti-
ties that share some fundamental properties that may be unknown to the speakers (for
example, a particular genetic structure for the entities in the extension of “cat”).
To examine whether people have externalist or internalist intuitions, Finnish par-
ticipants were presented with vignettes like the following:

A yellowish, bitter-smelling, fragile mineral called zircaum occurs widely in

Mid-Siberian soil. Scientists generally believe that zircaum is the compound
ACB. In Northern Norway, a deposit of a substance just like zircaum is found—
it is yellowish, bitter-smelling, fragile, etc. When scientists examine its deep
structure, they conclude that it is ACB, just like zircaum is believed to be.

At this point, participants were asked whether the novel substance is zircaum on a 7-
point scale (question 1). Then, participants were presented with the following passage:

A few weeks after the discovery in Northern Norway, scientists examine the
Mid-Siberian substance more closely. Using methods and instruments more

503
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

exact than previously available, they find out that they were wrong about the
deep structure of the substance: the substance is KML instead of ACB. How-
ever, the substance found in Northern Norway was indeed ACB, just as the
scientists thought it was.

Participants were then asked whether they considered their earlier judgment justified
(question 2) and correct (question 3):

When in the earlier situation you judged that ‘the substance found in North-
ern Norway is/is not zircaum’, was your answer a) justified b) strictly speaking
correct?

Jylkkä and colleagues found that on average Finnish participants harbored external-
ist intuitions (a second experiment replicated these findings). Concerning question 1,
when a novel substance had the chemical structure of a known element (for example,
zircaum), participants tended to judge that it was a sample of this element, while when
the novel substance did not have the chemical structure of a known element, they
tended to judge that it was not a sample of this element (mutatis mutandis for species).
Concerning question 3, two thirds of participants said that their answer to question 1
was mistaken after having been told that the beliefs about the constitution of the known
substance (for example, zircaum in the story above) were mistaken. Thus, most Finnish
participants reported externalist intuitions. However, consistent with our hypothesis
that philosophers of language might be led to ignore the variation in intuitions when
they appeal only to their own intuitions, Jylkkä and colleagues also found that intuitions
about natural kind terms vary substantially among Finnish people. Indeed, Jylkkä and
colleagues also report that people were rarely consistently externalist or internalist.

2.4 Conclusion
The findings reviewed in Section 2 provide support for our claim that philosophers
of language would be wise to emulate the experimental turn in generative linguistics
since some of the intuitions that are of most interest to philosophers of language appear
indeed to vary across and within cultures. Philosophers of language exclusive reliance
on their own and their colleagues’ intuitions obscures much of this variation and makes
it more likely that they will develop and embrace philosophical theories that are unable
to account for the variation of these intuitions across and within cultures.

3 Responding to Some Criticisms of Recent Empirical Research

in the Philosophy of Language
In the remainder of this chapter, we focus on the most developed body of experimental
evidence bearing on the philosophy of language—namely the cross-cultural research on
intuitions about the reference of proper names. The findings reviewed above have been
criticized on various grounds (Ludwig 2007; Martí 2009; Deutsch 2009, 2010; Jackman
2009; Lam 2010; Devitt 2011; Sytsma & Livengood 2011 Ichikawa et al., forthcoming).
We will focus on the most frequently raised criticism (Ludwig 2007; Deutsch 2009).
Because the question participants were asked in Machery et al. (2004) failed to distin-
guish between speaker’s reference and semantic reference, the critics argue, it is unclear

504
 TH E RO LE O F EX P E R I ME N T

whether the cross-cultural data bear on theories of reference and should be of interest
to philosophers of language. In this section, we first spell out the objection in greater
detail, and then offer two distinct replies.

3.1 Speaker’s Reference and Semantic Reference

The vignettes used in Machery et al. (2004) and in several other articles ask partici-
pants whom a speaker is talking about when he uses a proper name (“Gödel” or “Tsu
Ch’ung Chih”). Ludwig (2007) and Deutsch (2009) have complained that this question
is ambiguous since it could be interpreted as asking either (i) What is the speaker’s ref-
erence of a given token of “Gödel”? or (ii) What is the semantic reference of “Gödel”?
The semantic reference of a proper name is the reference it has in virtue of the conven-
tional rules of the natural language this name belongs to (Kripke 1977). Thus, “Barack
Hussein Obama” refers to Barack Obama in virtue of the rules of English. But in some
contexts a speaker may use a given proper name to refer to an individual who is not the
semantic reference of this name. For instance, suppose that in 2006 John sees on TV an
individual in Baghdad who looks like Dick Cheney and that John says, “Cheney is wear-
ing a nice suit.” In reality, Cheney, who is hiding in his ultra-secret bunker, has sent a
look-alike to Baghdad. Because John intends to refer to the individual he sees on TV,
the speaker’s reference of his utterance of “Cheney” is the look-alike in Baghdad, even
though the semantic reference of “Cheney” is the former vice-president. Furthermore,
Ludwig and Deutsch hold that only intuitions about semantic reference are relevant
in determining how the reference of proper names is fixed. While philosophers of
language are likely to understand that the question in the Gödel vignette is meant to
bear only on semantic reference, lay competent speakers might not be aware of the
distinction between these two ways of referring or, if they are aware of this distinction,
they might not realize that the question is meant to elicit intuitions about semantic
reference. It is thus unclear whether their answers express a judgment about the seman-
tic reference of “Gödel” or “Tsu Ch’ung Chih” in the vignettes used in past research.
Deutsch and Ludwig conclude that the experiments conducted by Machery and col-
leagues fail to provide any evidence that the relevant intuitions about reference vary
across cultures.

3.2 Is This A Plausible Objection?

Let’s begin with a few uncontroversial points. It makes sense to ask about speaker’s ref-
erence only for the occurrences of a proper name, and not for a proper name qua type.
In addition, to identify the speaker’s reference of the occurrence of a proper name, one
needs to know the communicative intention of the speaker. Thus, in the example given
above, we can determine that the speaker’s reference of “Cheney” is the look-alike in
Baghdad because the cover story and the sentence in which the proper name “Cheney”
is embedded provide enough information about John’s communicative intention: He
intends to talk about the suit of the individual he is seeing on TV. If no information
were presented—for instance, if we were just asked who “Cheney” refers to when John
uses it—we would focus on the semantic reference of “Cheney” because the question
would seem to bear on the reference of this proper name qua type and because no con-
textual information would be provided that would allow us to identify the speaker’s
communicative intention.

505
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

Now consider the Gödel vignette in Machery et al. (2004). At the end of the vignette,
participants are asked who John is talking about “when he uses the name ‘Gödel’.”
Since no specific utterance is mentioned and no contextual information is provided
that would enable participants to determine John’s communicative intention, it is hard
to see how participants could understand the question to be about the speaker’s refer-
ence of John’s utterance of “Gödel.” Rather, we submit, on the only plausible interpre-
tation of the question, it is asking about the reference of “Gödel” qua type—that is, it is
asking about the semantic reference of the term in John’s language. If this is right, then,
pace Ludwig and Deutsch, the question in the probes is not ambiguous.

3.3 Some Preliminary Empirical Evidence

Though we doubt that the question posed to participants in the Machery et al. (2004)
study is ambiguous, others may not be convinced. This motivated Machery et al. (forth-
coming) to design a further study in which they attempt to eliminate the ambiguity
between speaker’s reference and semantic reference in various ways. Their findings ten-
tatively confirm that Chinese tend to have genuine descriptivist intuitions about the
reference of proper names, while Americans tend to have causal-historical intuitions.
Here we report only one of their manipulations.
Suppose that Americans and Chinese alike tend to have causal-historical intuitions
about the reference of proper names. Suppose also that the minority of American partici-
pants and the majority of Chinese participants in Machery et al. (2004) who answered
that John is talking about the man who discovered the incompleteness of arithmetic
when he uses “Gödel” did so because they reported an intuition about whom John intends
to be talking about. That is, their answer expressed an intuition about speaker’s reference,
not semantic reference. Then, if Chinese and American participants were given a version
of the Gödel case in which John has the intention to refer to the man who stole the theo-
rem, a very large majority of participants should say that John is talking about the man
who stole the theorem: Participants whose answer expresses an intuition about semantic
reference should answer that John is talking about the theorem stealer since by hypothesis
they have causal-historical intuitions, while participants whose answer expresses an intu-
ition about speaker’s reference should also answer that John is talking about the theorem
stealer because in the vignette John has the intention to talk about the theorem stealer.
To test this prediction, Deutsch and colleagues designed a vignette that began with
the original Gödel story, and then continued,

One night, John is sitting in his room, revising for his mathematics exam by
going over the proof of the incompleteness theorem. After a while, he says to
his roommate, “Gödel probably got a huge number of awards from mathemati-
cal societies!”
When John utters that sentence, he is talking about:
(A) the person who got hold of the manuscript and claimed credit for the
work.
(B) the person who really discovered the incompleteness of arithmetic.

In this vignette, it is clear that John intends to talk about the man who in fact stole
the theorem. Naturally, John does not intend to talk about the man who stole the

506
 TH E RO LE O F EX P E R I ME N T

theorem under the description “the man who stole the theorem,” but it remains true
that he intends to talk about this individual since, given that the discoverer of the
proof died right after having discovered the proof, only the theorem stealer is likely
to have won numerous awards from mathematical societies. So, proponents of the
speaker’s reference / semantic reference objection should expect a very large major-
ity of participants to answer A. However, Deutsch and colleagues found that, when
presented with this vignette, about 75 percent of American participants but only 50
percent of Chinese participants, answered that the speaker is talking about the man
who stole the theorem. Because 50 percent of Chinese participants and 25 percent of
American participants answered B, one can conclude that it is not the case that every-
body has causal-historical intuitions about the semantic reference of proper names
and that Americans or Chinese who give answers in line with descriptivist theories of
reference are not merely reporting their intuitions about speaker’s reference. At the
very least, these findings show that genuine intuitions about semantic reference vary
among Americans and among Chinese. Furthermore, they undermine the criticisms of
the cross-cultural work on intuitions that appeal to the distinction between speaker’s
and semantic reference to defend the idea that across cultures people tend to have
causal-historical intuitions.
We conclude, first, that the findings concerning the intuitions about the reference of
proper names cannot be explained away by appealing to the distinction between speak-
er’s reference and semantic reference. Second, and most important for our purposes in
this chapter, far from undermining our call for an experimental turn in the philosophy
of language, this objection can best be evaluated experimentally: it calls for more and
more subtle experiments.

4 Implications of the Experimental Turn

In this final section, we examine the implications the experimental turn might have
for the philosophy of language. For the sake of simplicity, we focus on the implications
of the findings about variation in intuitions about the reference of proper names,
and we argue that philosophers of language face a dilemma: If they acknowledge that
ordinary speakers have genuine intuitions about semantic reference that are typically
reliable, then in light of the findings discussed in the two previous sections, they
need to revise their theories of reference dramatically. However, if they deny that
ordinary speakers have such intuitions and reject our call for an experimental turn in
the philosophy of language, they then open the door to a challenge about the reality
of semantic reference. We further hypothesize that an experimental turn in the phi-
losophy of language will have equally significant implications for other philosophical
issues.
Do ordinary speakers of (say) English have genuine intuitions about semantic refer-
ence? That is, are they able to identify the semantic reference of proper names in actual
and hypothetical situations without confusing it with the speaker’s reference of these
names in these situations?
Suppose, first, that ordinary speakers of English do have genuine intuitions about
semantic reference that are usually reliable. We have reviewed a growing body of exper-
imental evidence that hypothetical situations elicit different answers both within and
across cultures, and we have argued that participants’ answers in these experiments
do reflect their intuitions about semantic reference. This suggests that the reference

507
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

of proper names might be determined differently for different competent speakers of a

given language (English in the current case) and that descriptivist reference determina-
tion might be more common in some cultures than in others. If this is correct, theories
of reference should then be modified to account for such variation in reference deter-
mination. Reimer (2009) is a pioneering example of a theory of the reference of proper
names that is responsive to the variation in intuitions about the reference of proper
names—both within and across cultures—illustrated in Sections 2 and 3.
Suppose however that philosophers of language dismiss the arguments presented in
Section 3 and maintain that lay people do not have intuitions about semantic reference,
perhaps because they do not distinguish between speaker’s reference and semantic refer-
ence. In this case, ordinary speakers’ intuitions can’t be used to decide among competing
theories of reference proposed by philosophers of language, and surveying them would
be a waste of time. But if this is the case, philosophers of language need to clarify the
grounds for positing that proper names actually have a semantic reference in addition
to the reference that occurrences of names have as a result of speakers’ communicative
intentions, much as scientists do when they introduce a theoretical distinction between
two kinds of properties (for example, mass and weight). Philosophers of language must
tell us what phenomena are best explained by postulating that proper names refer in
virtue of the rules of the language they belong to.
We suspect that many philosophers of language would reply that philosophers of lan-
guage’s own intuitions justify positing this distinction: by virtue of their acquired exper-
tise, they might reply, philosophers of language have acquired new intuitions about the
reference of proper names, and these intuitions suggest that proper names can have two
distinct kinds of reference, speaker’s reference and semantic reference (Ludwig 2007;
Devitt 2011). Philosophers of language might go on to note that experts (e.g., radi-
ologists) typically acquire new, reliable intuitions by virtue of their acquired expertise
(Devitt 2006; Devitt 2011).
We are prepared to concede that many philosophers of language probably do have
intuitions consistent with a distinction between speaker’s reference and semantic refer-
ence. However, for three distinct reasons, we doubt that if ordinary competent speakers
do not have genuine intuitions about semantic reference, philosophers’ intuitions can
justify the hypothesis that proper names have a semantic reference in addition to the
reference their occurrences have as a result of speakers’ communicative intentions. First,
this appeal to the intuitions of experts stands in stark contrast to the growing skepticism
in generative linguistics about intuitions that are shared only by linguists. It is curious to
see philosophers embrace a methodology (relying on theoreticians’ intuitions while dis-
carding lay speakers’ intuitions) that linguists are now increasingly questioning. Second,
in many fields, experts’ intuitions (for example, radiologists’ judgments about x-rays) can
be validated by objective criteria, and our confidence in the reliability of their intuitions
largely depends on such validation (for review, see Weinberg et al. 2010). Since it is (to
put it mildly) less than clear what could validate the intuitions of philosophers of lan-
guage about the reference of proper names, one wonders why these intuitions should be
trusted. Third, the broader the consensus among experts, the more credible their intui-
tions. In this respect, philosophers’ intuitions about reference are also on shaky grounds.
Some influential linguists and philosophers of language have denied that names have
a semantic reference, and it is hardly likely that these experts share the intuitions that
support postulating that proper names have a semantic reference in addition to speak-
er’s reference (for example, Strawson 1950; Chomsky 1995, 2000; Pietroski 2003). It is

508
 TH E RO LE O F EX P E R I ME N T

particularly notable that skepticism about referential semantics has long been a
key theme in Chomsky’s work. In New Horizons in the Study of Language he writes
(2000: 17):

In general, a word, even of the simplest kind, does not pick out an entity of the
world, or of our “belief space”. Conventional assumptions about these matters
seem to me very dubious.

Further on in the book Chomsky elaborates on his skepticism:

A good part of contemporary philosophy of language is concerned with analyz-

ing alleged relations between expressions and things, often exploring intui-
tions about the technical notions “denote,” “refer,” “true of,” etc. said to hold
between expressions and something else. But there can be no intuitions about
these notions, just as there can be none about “angular velocity” or “protein.”
These are technical terms of philosophical discourse with a stipulated sense that
has no counterpart in ordinary language. . . . If we rerun the thought experi-
ments with ordinary terms, judgments seem to collapse or, rather, to become so
interest-relative as to yield no meaningful results. . . . [I]t is not at all clear that
the theory of natural language and its use involves relations of “denotation,”
“true of,” etc. in anything like the sense of the technical theory of meaning.
(Chomsky 2000: 130)

Thus, not only is the reliance of philosophers of language on their own intuitions a
departure from the methodological developments in linguistics, there is also little rea-
son to trust philosophers’ intuitions since they are not externally validated and they are
unlikely to be shared by some of the most prominent linguists.
To recap, if philosophers of language dismiss the type of cross-cultural studies sur-
veyed earlier on the grounds that ordinary speakers don’t have genuine intuitions about
semantic reference, it is then unclear how they can justify positing the distinction
between speaker’s reference and semantic reference. But then why should we believe
that names in English (and, mutatis mutandis, other kinds of words) have a reference in
virtue of the rules of English, in addition to the reference their occurrences have as a
result of speakers’ communicative intentions? On grounds of parsimony, it would seem
that one should follow Chomsky’s lead and construct our theory of natural language
without invoking semantic reference.
Thus, philosophers of language are faced with a dilemma. If they acknowledge that
ordinary speakers have intuitions about semantic reference, then theories of reference
need to be substantially modified to accommodate the variation in reference determi-
nation, as Reimer (2009) has proposed to do. If they deny that ordinary speakers have
intuitions about semantic reference, then since philosophers’ intuitions do not seem
sufficient to justify the assumption that proper names have a semantic reference, it
becomes unclear why we should hold that names, instead of people, refer.

Conclusion
It is time for philosophers of language to emulate linguists in taking an experimen-
tal turn. The objections against linguists’ appeal to their own intuitions carry over to

509
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

the methods currently dominant in the philosophy of language, and the burgeoning
research in experimental philosophy has made the limits of these methods increasingly
clear. Philosophers of language who cling to traditional methods are depriving them-
selves of a rich source of evidence about how language actually works.

Bibliography
Aoun, J., Hornstein, N., Lightfoot, D. & Weinberg, A. (1987). “Two Types of Locality,” Linguistic Inquiry
18: 537–77.
Bard, E., Robertson, D. & Sorace, A. (1996). “Magnitude Estimation of linguistic acceptability,” Language
72: 32–68.
Bauer, L. (1990). “Be-heading the Word,” Journal of Linguistics 26: 1–31.
Braisby, N., Franks, B. & Hampton, J. (1996). “Essentialism, Word Use, and Concepts,” Cognition 59:
247–74.
Chomsky, N. (1957). Syntactic Structures, The Hague: Mouton.
Chomsky, N. (1975). The Logical Structure of Linguistic Theory, Chicago: University of Chicago Press.
Chomsky, N. (1976). “Conditions on Rules of Grammar,” Linguistic Analysis 2: 303–51.
Chomsky, N. (1981). Lectures on Government and Binding, Foris: Dordrecht.
Chomsky, N. (1995). “Language and Nature,” Mind 104: 1–61.
Chomsky, N. (2000). New Horizons in the Study of Language and Mind, Cambridge: Cambridge University
Press.
Cowart, W. (1997). Experimental Syntax: Applying Objective Methods to Sentence Judgments, Thousand Oaks,
CA: SAGE Publications.
Culbertson, J. & Gross, S. (2009). “Are Linguists Better Subjects?”, British Journal for the Philosophy of Science
60: 721–36.
Deutsch, M. (2009). “Experimental Philosophy and the Theory of Reference,” Mind & Language 24: 445–66.
Deutsch, M. (2010). “Intuitions, Counter-Examples, and Experimental Philosophy,” The Review of Philoso-
phy and Psychology 1, 447–460.
Devitt, M. (1981). Designation, New York: Columbia University Press.
Devitt, M. (2006). “Intuitions in Linguistics,” The British Journal for the Philosophy of Science 57: 481–513.
Devitt, M. (2011). “Experimental Semantics,” Philosophy and Phenomenological Research 82, 418–435.
Evans, G. (1973). “The Causal Theory of Names,” Supplementary Proceedings of the Aristotelian Society 47:
187–208.
Fanselow, G. (2007). “Carrots—Perfect as Vegetables, but Please Not as a Main Dish,” Theoretical Linguistics
33: 353–67.
Featherston, S. (2007). “Data in Generative Grammar: The Stick and the Carrot,” Theoretical Linguistics
33: 269–318.
Genone, J. & Lombrozo, T. (forthcoming). “Concept Possession, Experimental Semantics, and Hybrid
Theories of Reference,” Philosophical Psychology.
Goodall, G. (2006). “Contraction,” In M. Everaert and H. van Riemsdijk (eds.) The Blackwell Companion to
Syntax, vol. 1 (pp. 688–703), Malden, MA: Blackwell.
Haider, H. (2007). “As a Matter of Facts—Comments on Featherston’s Sticks and Carrots,” Theoretical
Linguistics 33: 381–94.
Hampton, J. A., Estes, Z. & Simmons, S. (2007). “Metamorphosis: Essence, Appearance, and Behavior in
the Categorization of Natural Kinds,” Memory & Cognition 35: 1785–800.
Ichikawa, J., Maitra, I. & Weatherson, B. (Forthcoming). “In Defense of a Kripkean Dogma,” Philosophy and
Phenomenological Research.
Jackendoff, R. S. (1997). The Architecture of the Language Faculty, Cambridge, MA: MIT Press.
Jackman, H. (2009). “Semantic Intuitions, Conceptual Analysis, and Cross-Cultural Variation,” Philosophi-
cal Studies 146: 159–77.
Jackson, F. (1998). “Reference and Description Revisited,” Philosophical Perspectives 12: 201–18.
Jylkkä, J., Railo, H. & Jussi, H. (2009). “Psychological Essentialism and Semantic Externalism: Evidence for
Externalism in Lay Speakers’ Language Use,” Philosophical Psychology 22: 37–60.
Keller, F. & Asudeh, A. (2001). “Constraints on Linguistic Coreference: Structural vs. Pragmatic Factors,”
In J. D. Moore & K. Stenning (eds.) Proceedings of the 23rd. Annual Conference of the Cognitive Science
Society (pp. 483–88), Mahwah, NJ: Lawrence Erlbaum Associates.

510
 TH E RO LE O F EX P E R I ME N T

Kripke, S. (1972/1980). Naming and Necessity, Cambridge, MA: Harvard University Press.
Kripke, S. (1977). “Speaker’s Reference and Semantic Reference,” Midwest Studies in Philosophy 2: 255–76.
Labov, W. (1975). “Empirical Foundations of Linguistic Theory,” In R. Austerlitz (ed.) The Scope of Ameri-
can Linguistics (pp. 77–133), Lisse: Peter de Ridder.
Labov, W. (1996). “When Intuitions Fail,” Chicago Linguistics Society 32: 77–105.
Lam, B. (2010). “Are Cantonese Speakers Really Descriptivists? Revisiting Cross-Cultural Semantics,” Cog-
nition 115: 320–9.
Lasnik, H. & Saito, M. (1984). “On the Nature of Proper Government,” Linguistic Inquiry 15: 235–89.
Ludwig, K. (2007). “The Epistemology of Thought Experiments: First-Person Approach vs. Third-Person
Approach,” Midwest Studies in Philosophy 31: 128–59.
Machery, E. (2011). “Variation in Intuitions about Reference and Ontological Disagreements,” In S. D.
Hales (ed.) A Companion to Relativism, (pp. 118–36) New York: Blackwell.
Machery, E., Deutsch, M. & Sytsma J. (forthcoming). “Speaker’s Reference and Cross-Cultural Semantics.”
In A. Bianchi (ed.), On Reference, Oxford University Press.
Machery, E., Mallon, R., Nichols, S. & Stich, S. P. (2004). “Semantics, Cross-Cultural Style,” Cognition
92: B1–12.
Machery, E., Mallon, R., Nichols, S., Stich, S. P., Deutsch, M. & Sytsma, J. (2010) “Semantic Intuitions:
Reply to Lam,” Cognition 115: 320–29.
Machery, E., Olivola, C. Y. & de Blanc, M. (2009). “Linguistic and Metalinguistic Intuitions in the Philoso-
phy of Language,” Analysis 69: 689–94.
Marantz, A. (2005). “Generative Linguistics within the Cognitive Neuroscience of Language,” The Linguis-
tic Review 22, 429–445
Martí, G. (2009). “Against Semantic Multi-Culturalism,” Analysis 69: 42–8.
Myers, J. (2009). “Syntactic Judgment Experiments,” Language and Linguistics Compass 3: 406–23.
Nisbett, R. E. (2003). The Geography of Thought: how Asians and Westerners Think Differently . . . And Why,
New York: Free Press.
Novack, I. A. & Reboul, A. (2008). “Experimental Pragmatics: A Gricean Turn in the Study of Language,”
Trends in Cognitive Sciences 12: 425–31.
Nunberg, G., Sag, I. A. & Wasow, T. (1994). “Idioms,” Language 703: 491–538.
Phillips, C. (2009). “Should we Impeach Armchair Linguists?,” In S. Iwasaki (ed.), Japanese/Korean Linguis-
tics 17, Palo Alto, CA: CSLI Publications.
Pietroski, P. (2003). “The Character of Natural Language Semantics,” In A. Barber (ed.) Epistemology of
Language (pp. 217–56), New York: Oxford University Press.
Pietroski, P., Lidz, J., Hunter, T. & Halberda, J. (2009). “The Meaning of ‘Most’: Semantics, Numerosity,
and Psychology,” Mind & Language 24: 554–85.
Putnam, H. (1970). “Is Semantics Possible?” In H. Kiefer and M. Munitz (eds.) Languages, Belief and
Metaphysics (pp. 139–52), Albany: SUNY Press.
Putnam, H. (1975). “The Meaning of ‘Meaning’,” In H. Putnam (ed.), Philosophical Papers, Vol. 2: Mind,
Language, and Reality (pp. 215–71), New York: Cambridge University Press.
Reimer, M. (2009). “Jonah Cases,” In A. Everett (ed.) Empty Names, Oxford: Oxford University
Press.
Riemer, N. (2009). “Grammaticality as Evidence and as Prediction in a Galilean Linguistics,” Language
Sciences 31: 612–33.
Riehemann, S. Z. (2001). A Constructional Approach to Idioms and Word Formation, Stanford University
Dissertation.
Sampson, G. (2005). “Quantifying the Shift Towards Empirical Methods,” International Journal of Corpus
Linguistics 10: 15–36.
Schütze, C. T. (1996). The Empirical Base of Linguistics, Chicago: The University of Chicago Press.
Schütze, C. T. (2005). “Thinking About What We Are Asking Speakers to Do,” In S. Kepser and M.
Reis (eds.) Linguistic Evidence: Empirical, Theoretical, and Computational Perspectives (pp.457–485), Berlin:
Mouton de Gruyter.
Searle, J. (1958). “Proper Names,” Mind 67: 166–73.
Sorace, A. & Keller, F. (2005). “Gradience in Linguistic Data,” Lingua 115: 1497–524.
Sprouse, J. (2007). A Program for Experimental Syntax: Finding the Relationship Between Acceptability
and Grammatical Knowledge, College Park, MD: University of Maryland dissertation.
Stich, S. P. (2010). “Philosophy and WEIRD Intuition,” Behavioral and Brain Sciences 33:110–11.
Strawson, P. F. (1950). “On Referring,” Mind 59: 320–44.

511
 EDO U A RD M A CH ERY A N D S T E P H E N S T I C H

Sytsma, J. M. & Livengood, J. (2011). “A New Perspective Concerning Experiments on Semantic Intui-
tions,” The Australasian Journal of Philosophy 89: 315–32.
Wasow, T. & Arnold, J. (2005). “Intuitions in Linguistic Argumentation,” Lingua 115: 1481–96.
Weinberg, J. M., Gonnerman, C., Buckner, C. & Alexander, J. (2010). “Are Philosophers Expert Intuiters?”,
Philosophical Psychology 23: 331–55.
Wikforss, Å. (2008). “Semantic Externalism and Psychological Externalism,” Philosophy Compass 3: 158–81.

512
 4.2
THE R OLE OF
L I NGUI ST IC S
Sarah Moss

The interaction of linguistics and the philosophy of language regularly inspires florid
description. One philosopher describes his discipline as the “initial central sun, semi-
nal and tumultuous: from time to time it throws off some portion of itself to take sta-
tion as a planet,” and he goes on to suggest that linguistics may be among its satellites
(Austin 1956, 232). In the words of one linguist, philosophy is more like “a rummage
sale that sometimes offers useful items for the home improver” (Nunberg 2002, 679).
Higginbotham 2002 offers the tantalizingly ambiguous observation that “linguistics and
philosophy, like steak and barbecue sauce, have much to give each other” (575). And
contemporary interactions between linguists and philosophers of language are so exten-
sive that these descriptions fail to capture the extent to which it is increasingly difficult
just to delineate the fields.
Following the maxim that it would be “better to show than tell” something about
the role of linguistics in the philosophy of language, I will focus on a few case studies
that illustrate key aspects of this relationship. This entry is necessarily limited in scope.
Partee 2004b and 2005 provide an excellent account of early interactions between lin-
guists and philosophers, so this entry focuses on contemporary literature rather than the
historical role of linguistics in the philosophy of language. Furthermore, I set aside the
extensive influence of linguistics on other subfields of philosophy, such as the influence
of psycholinguistics in the philosophy of mind, the semantics of knowledge construc-
tions in assessing intellectualism in epistemology (cf. Stanley & Williamson 2001), and
the semantics of event-related counting in the metaphysics of persistence (cf. Moss
2010b). I focus on the relationship between philosophy of language and three branches
of linguistics—syntax, semantics, and pragmatics. In each section, I discuss a case study
in detail and then briefly mention other notable instances of the role of linguistics in
the philosophy of language.

1 Syntax
The first case study is addressed most directly in Stanley 2000: the debate over whether
there are in fact unarticulated constituents of the propositions expressed by utterances.
For instance, it is uncontroversial that both (1) and (2) are context-sensitive, and that
the place where they are uttered is relevant to their truth conditions:
 SA RA H MO S S

(1) They are serving drinks at the bar near here.

(2) They are serving drinks at the local bar.

Some have argued that unlike (1), sentences such as (2) do not contain any element
that designates the place of the context of utterance (cf. Pragmatic Enrichment for
references and an assessment of this debate). Instead there is a “post-semantic” level of
interpretation in which such unarticulated constituents of the proposition are identi-
fied. Perry 1986 introduces the notion of an unarticulated constituent, and Perry 1998
clarifies:

In some cases of implicit reference there is a feature, a trace, a sort of phantom

expression, that serves in place of an expression, so the referred to constituent
really isn’t unarticulated . . . I am interested in the theoretical possibility and
coherence of truly unarticulated constituents. (9)

Against Perry, Stanley 2000 argues that sentences such as (3) are evidence against the
unarticulated constituent analysis:

(3) Every newspaper reporter went to a local bar to hear the news.

Stanley argues that (3) contains a covert location variable that is semantically and
hence syntactically bound on the natural reading of (3), and that is a fortiori present as
a syntactically articulated element of the sentence.
Stanley also suggests that covert pronouns associated with expressions such as ‘local’
are responsible for weak crossover effects, restrictions on binding easily observed in sen-
tences containing overt pronouns. For instance, one cannot hear a reading of (4) on
which the overt pronoun ‘her’ is bound by the lower quantifier ‘every reporter’:

(4) #Heri local bar sponsored [every reporter]i.

Stanley argues that one can see the same effect in (5), namely the absence of a reading
that says that every reporter was sponsored by a bar local to that reporter:

(5) #A local bar sponsored every reporter.

In other words, the same syntactic constraints on binding seem responsible for the
absence of the intended bound readings of (4) and (5) (though see Rothschild &
Segal 2009 for dissent regarding such examples). Stanley concludes that the behav-
ior of relational expressions (e.g. ‘local’) in weak crossover constructions such as (5)
“strongly suggests the existence of a covert pronominal element in relational expres-
sions” (423).
This is a classic case of syntactic arguments playing a role in the philosophy of lan-
guage. The arguments in fact originate with linguists: Partee 1989 cites Mitchell 1986
as the first to identify a “much broader class of contentful context-dependent elements
which can also exhibit bound-variable-like behavior, such as local” (260). Partee also

514
 TH E RO LE O F LI N G U I S T I C S

points out weak crossover effects in sentences like (6), on the reading where each pro-
fessor’s local union is in question:

(6) #?The leader of the local union wrote a letter to every untenured
professor.

Condoravdi & Gawron 1996 give similar examples and argue that such examples dem-
onstrate that “the contextual determination of the interpretation of implicit arguments
is not just a matter of pragmatic principles” (6). These linguists explicitly mention
Crimmins 1992 as an advocate of unarticulated constituents, and as such they take him
to be one philosopher targeted by their analysis.
It is perhaps regrettable that philosophers engaged in this dispute seldom mention
that Partee 1989 and Condoravdi & Gawron 1996 ultimately reject the pronominal
analysis of relational expressions defended in Stanley 2000. In particular, Condoravdi
& Gawron 1996 argue that “implicit arguments pattern with definite descriptions rather
than pronouns in allowing for associative anaphoric readings” (10). Partee 2009 notes
that semanticists rarely advocate the covert pronoun analysis argued against in Partee
1989. But she mentions that semanticists do accept the main point of Stanley 2000:

Formal semanticists are quite uniformly convinced that the Pragmatic Enrich-
ment Approach cannot be right, because it does not predict any purely linguistic
constraints on possible enrichment, and there are indeed strong anaphora-like
constraints on what is possible. (8)

It is important for the opponent of pragmatic enrichment to consider a range of

responses to the above syntactic arguments. For instance, one might claim that the
effects of pragmatic enrichment may be semantically controlled by other expressions.
See Cappelen & Lepore 2002, 2005 for responses to the above arguments and Cohen &
Rickless 2007 for a rejoinder, and see Marti 2006 and Elbourne 2008 for further refer-
ences and critical discussion of both camps. Regardless of one’s opinion about unarticu-
lated constituents, it is clear that syntactic arguments have played a major role in this
dispute.
Furthermore, the arguments just considered have consequences for other debates in
the philosophy of language, far removed from discussions of pragmatic enrichment. For
instance, take a topic that has been part of the philosophy of language canon since Frege
1892: the correct theory of attitude ascriptions. Concerns raised by direct reference
theorists (e.g. Kripke 1972, Putnam 1975) motivate many philosophers to claim that
coreferential names have the same semantic value. Recent advocates of this doctrine
give various proposals for reconciling two contrasting claims: first, that coreferential
names have the same semantic value even when embedded in intensional contexts,
and second, that sentences such as (7) and (8) can nevertheless have different truth
conditions:

(7) Lois believes that Superman is strong.

(8) Lois believes that Clark Kent is strong.

Furthermore, (9) can have different truth conditions in different contexts, depending
on which way of thinking of Paderewski is salient (cf. Kripke 1979):

515
 SA RA H MO S S

(9) Peter believes that Paderewski had musical talent.

Crimmins 1992 argues that different guises of Paderewski are unarticulated constituents
of propositions expressed by (9) in different contexts, while Schiffer 1992 discusses the
proposal that modes of presentation are covert implicit arguments of attitude verbs (cf.
Ludlow 1996 and Schiffer 1996 for syntactically driven assessments of hidden-indexical
theories of attitude ascriptions). Richard 1990 takes a third approach: that ‘believes’ in
(9) denotes different belief relations in contexts in which different sentences count as
acceptable translations of sentences Peter accepts.
The same arguments marshalled against the theory that locations are unarticulated
constituents of propositions expressed by ‘local’ sentences can be brought to bear on the
theory that guises are unarticulated constituents of propositions expressed by ‘believes’
sentences. Consider the following case:

You are approached by Cicero, a man you recognize to be a famous spy. Cicero
says, “I see you know me as a notorious spy. But you also know me as one of your
closest friends, and without realizing that I am a spy, you share secrets with me
as with any trusted person. And the same goes for every woman in this room:
each knows me as one of her closest friends, and none yet realizes that I am a
spy.”

In such a case, you may truly say:

(10) No woman in this room yet realizes that Cicero is a spy.

Arguably this utterance has the following reading:

(11) No woman in this room yet realizes that Cicero is a spy when thinking
about Cicero in the one-of-her-closest-friends way.

But this bound reading tells against the unarticulated constituent analysis of (10) and
other attitude ascriptions as much as the bound reading of (3) tells against the unarticu-
lated constituent analysis of ‘local’ sentences. Stanley 2000 generalizes his argument to
refute a narrow indexical analysis of relational expressions. I leave it to the reader to
decide if the analogous variant of the above argument presents an equal challenge for
the narrow indexical semantics for attitude ascriptions defended in Richard 1990. Here
the central point is simply that syntactic arguments are relevant in the context of an
entirely self-sufficient philosophical debate.
There are a range of other notable instances of syntactic arguments bearing on classic
questions in the philosophy of language. Several arguments regarding quantifier scope
have been applied in assessing quantificational analyses of definite descriptions; see
Glanzberg 2007 for critical discussion of some of this literature. Others have applied
syntactic arguments in assessing quantificational accounts of complex demonstratives;
see King 2001 for one example and Lepore & Johnson 2002 and King 2010 for critical
discussion. And it has been suggested that the appropriate syntactic representation of
tense has implications for the nature of the objects of assertion, e.g. for whether such
objects vary in truth value across times; see Ninan 2010 for a critical discussion of the
role that syntactic arguments play in this literature.

516
 TH E RO LE O F LI N G U I S T I C S

2 Semantics
A second case study for the influence of linguistics in the philosophy of language is
presented in the discussion of subjunctive conditionals in von Fintel 2001. Giving an
analysis of subjunctives has been of philosophical interest since Chisholm 1946 and
Goodman 1947. Stalnaker 1968 and Lewis 1973 advocate competing theories of sub-
junctives, and many others pursue details of their dispute (see Subjunctive Condition-
als). But the competing accounts given by these philosophers do share a central feature:
according to both, Antecedent Strengthening is invalid. For instance, consider the fol-
lowing felicitous sequence from Lewis 1973:

(12) a. If the USA threw its weapons into the sea tomorrow, there would be
war.
b. If the USA and the other nuclear powers all threw their weapons into
the sea tomorrow, there would be peace.

Stalnaker and Lewis both advocate variably strict conditional analyses, according to
which ‘if p, would q’ expresses a proposition that is true roughly just in case all the
closest p worlds are q worlds. In particular, Lewis 1973 says that ‘if p, would q’ expresses
a proposition that is non-vacuously true at a world just in case some p-and-q world is
closer to that world than any p-and-not-q world. Stalnaker 1968 says that ‘if p, would q’
expresses a proposition that is true at a world just in case the closest p world is a q world.
Both predict that strengthening the antecedent of (12-a) need not preserve truth, so
(12-a) and (12-b) can both be true.
Recent papers in formal semantics challenge the variably strict conditional analysis.
In von Fintel 1999, 2001, von Fintel argues that Stalnaker and Lewis fail to explain the
infelicity of other sequences such as:

(13) a. If the USA and the other nuclear powers all threw their weapons into
the sea tomorrow, there would be peace.
b. #If the USA threw its weapons into the sea tomorrow, there would be
war.

In order to account for the asymmetry between (12) and (13), von Fintel endorses a
strict conditional analysis (cf. Gillies 2007 for a similar challenge and response). He
argues that ‘if p, would q’ is true just in case all the p worlds in a contextually determined
domain are q worlds, and he adds that ‘if p, would q’ itself demands that the relevant
domain contain some p worlds. So after a speaker utters (13-a), the context shifts so
that the domain of the strict conditional expressed in (13-b) contains worlds where all
nuclear powers disarm, and (13-b) therefore fails to be true. In brief, von Fintel argues
that he can beat Stalnaker and Lewis at their own game, since the variably strict analysis
that does so well at accounting for (12) fails to account for our judgment about (13) (cf.
Moss 2010a and Moss 2010c for responses on behalf of the variably strict conditional
analysis).
Furthermore, von Fintel is not only interested in accounting for judgments about
more creative sequences of subjunctives. He also raises distinctively semantic concerns
that bear on the dispute. The most interesting linguistic motivation for his strict condi-
tional analysis comes from facts about negative polarity items (NPIs), expressions that

517
 SA RA H MO S S

are licensed only in the scope of negation and other “negative” expressions. English
‘ever’ is one example:

(14) I haven’t ever watched that movie.

(15) #I have ever watched that movie.

Ladusaw 1979 argues that NPIs are licensed in downward entailing environments, i.e.
environments that reverse entailment relations (such as the relation between ‘fat cat’
and ‘cat’). For example, ‘every’ is downward entailing in just its first argument and ‘no’
is downward entailing in both, as evidenced by the following entailment relations:

(16) Every cat sat. Ÿ Every fat cat sat.

(17) Every cat sat. Ÿ
e Every cat sat on the mat.
(18) No cat sat. Ÿ No fat cat sat.
(19) No cat sat. Ÿ No cat sat on the mat.

And ‘ever’ is licensed in the environments that are downward entailing (cf. von Fintel
1999 for a catalog of literature disputing and refining this generalization):

(20) Every boy who ever watched that movie liked it.
(21) #Every boy who watched that movie ever liked it.
(22) No boy who ever watched that movie liked it.
(23) No boy who watched that movie ever liked it.

Furthermore, antecedents of conditionals—including subjunctives—license NPIs:

(24) If you had ever watched that movie, you would have liked it.

In 2001, von Fintel argues that this fact is evidence for the strict conditional analysis of sub-
junctives. Recall that on the variably strict conditional account, antecedents of subjunc-
tives are not downward entailing environments, i.e. Antecedent Strengthening is seman-
tically invalid. But the strict conditional acccount holds that antecedents are downward
entailing, as the behavior of (24) suggests. One might respond that superlatives also license
NPIs, as in ‘the oldest boy who ever watched that movie liked it’. Fans of the variably strict
analysis could argue that a superlative element (e.g. ‘closest’) in the logical form of subjunc-
tives licenses NPIs in sentences like (24). See von Fintel 1999 for a critique of this proposal.
Whatever one may make of this debate, it is clear that facts about negative polarity items
present further data for any philosopher developing a theory of subjunctive conditionals.
Besides tests involving negative polarity items, several tools from semantics regularly
appear in the philosophy of language literature. Philosophers use minimal pairs of sen-
tences in constructing and testing hypotheses. Discussions of verb phrase ellipsis and
constraints on binding are scattered throughout the literature. Cross-linguistic studies
have introduced constraints on philosophical theories of language fragments (e.g. con-
nections between research on evidentials and epistemic modals, and discussions of non-
English demonstratives outlined in Wolter 2009). Tests for distinguishing predicational
from specificational copular sentences have played a role in debates about the referents
of ‘that’-clauses (cf. Pryor 2007). These examples suggest that perhaps semantics also is
a source of useful items for the home improver.

518
 TH E RO LE O F LI N G U I S T I C S

In addition to providing philosophers with useful tools, semantics is a source of research

interests common to linguists and philosophers of language. Semanticists have studied
literally every topic in the first section of this volume (cf. Parts of Speech). Obviously
there has been much shared research on the semantics of various referring expressions,
including anaphoric expressions (cf. Anaphora), indefinite and definite descriptions (cf.
Descriptions), and complex demonstratives (cf. Indexicals and Demonstratives). Seman-
tic analyses of other language fragments have implications for foundational questions
about the nature of assertion, e.g. research on epistemic modals (cf. Swanson 2008 for
references) and other language of subjective uncertainty and predicates of personal taste
(cf. Relativism). Philosophers have also actively engaged in linguistics research at the
developmental stage of novel semantic frameworks, e.g. from the development of formal
semantic systems (Lewis 1970, Montague 1970, 1973) to event semantics (Davidson
1967), double-indexing (Kamp 1971, Kaplan 1977), alternative semantics (Hamblin
1973), dynamic semantics (Stalnaker 1974, 1978, Lewis 1975, 1979a, 1979b, Kamp 1981,
Veltman 1996), situation semantics (Barwise & Cooper 1981), and structured proposi-
tions (Cresswell & Stechow 1982). In each instance, philosophers have contributed to
the development of research areas squarely situated in the field of formal semantics.

3 Pragmatics
In a number of cases, philosophers have also been present in the early stages of research
programs in pragmatics. Recently linguists (e.g. Potts 2005) have developed more for-
mal approaches to theories of speech acts first introduced by Grice 1987, and the same
can be said for early discussions of presupposition in Stalnaker 1970, 1974, Lewis 1979b
and Kripke 2009 (originally delivered as a lecture in 1990). The third case study I shall
discuss involves applying some more recent literature on presupposition in giving an
account of the following classic pair of sentences:

(25) Hesperus is Hesperus.

(26) Hesperus is Phosphorus.

Since the introduction of these sentences in Frege 1892, philosophers of language have
confronted the challenge of accounting for the fact that sentences differing only by the
substitution of coreferential names can nevertheless convey different information.
Several direct reference theorists have argued that pragmatic differences between
(25) and (26) are responsible for their difference in informativeness (cf. Salmon 1986,
Soames 2002, Thau 2002). Recent considerations raised by linguists confirm this gen-
eral strategy but suggest a novel pragmatic vehicle for the content conveyed in (26),
namely presupposed contents. Here again the account is informed by empirical consid-
erations, namely a relatively straightforward test battery for presuppositions. For exam-
ple, possessives typically introduce presuppositions that project through negation and
antecedents of conditionals, so that each of the following sentences conveys that John
has a dog (see Ch. 1 of Geurts 1999 for more detailed discussion of projection tests for
presuppositions):

(27) John’s dog is hungry.

(28) John’s dog is not hungry.
(29) If John’s dog is hungry, then John should feed him.

519
 SA RA H MO S S

But presuppositions can also be locally accommodated in the consequent of a condi-

tional. Such conditionals will therefore fail to convey the same presupposition, e.g. the
following does not convey that John has a dog:

(30) If John has a dog, John’s dog is hungry.

Several theorists have argued that names convey contents that exhibit similar behavior.
For instance, the following are only appropriate if the name ‘Fido’ meets some familiar-
ity condition (cf. Sommers 1982 for an early instance of this observation):

(31) Fido is hungry.

(32) Fido is not hungry.
(33) If Fido is hungry, then someone should feed him.

But this condition of familiarity can be met by a local context, so that e.g. the speaker of
(34) does not presuppose that the name ‘Fido’ is familiar to his audience:

(34) If Fido is that pink poodle over there, then Fido is hungry.

Maier 2009 argues that “dynamic semantics and presupposition theory have given rise
to a fruitful analysis of proper names as presupposition inducers” (265).
Details of presuppositional theories of names vary widely. Geurts 1997 argues that
“the meaning of a name N is ‘the individual named N’, where the semantic contribution
of the definite article is to be analysed in presuppositional terms” (340). Swanson 2006
claims that the speaker of (31) presupposes merely that “the thing she associates with
‘Fido’ in the context of utterance is the thing the addressee associates with ‘Fido’ in that
context” (11). Maier 2009 operates in a framework according to which names intro-
duce discourse referents that are themselves objects of presupposition. Roberts 2009
argues that names are special because the discourse referent that satisfies the familiarity
presupposition of a name “carries the information that the individual bearer. . . does so
by virtue of a socially recognized, causal dubbing event by an (authorized) agent, and
the subsequent social propagation of the name-individual association. . . to the current
occasion of use” (7). But these disparate accounts are unified in spirit: they say that
sentences such as ‘Hesperus is Phosphorus’ convey presupposed information that distin-
guishes them from trivial identity sentences.
Direct reference theorists are interested in the difference in the cognitive import of
(25) and (26). In explaining this difference, they also aim to give a theory of our judg-
ments about attitude ascriptions:

(35) John believes that Hesperus is Hesperus.

(36) John believes that Hesperus is Phosphorus.

So we have arrived where we began: Insofar as the original syntactic arguments outlined
in §1 are dispositive, general pragmatic theories of differences in cognitive import are
constrained by syntactic considerations. For instance, they must accommodate bound
readings of ascriptions such as (10):

(10) No woman in this room yet realizes that Cicero is a spy.

520
 TH E RO LE O F LI N G U I S T I C S

Providing for these considerations is compatible with a pragmatic account. Presupposi-

tions introduced by names may be locally accommodated in attitude ascriptions. Stal-
naker 1988 introduces the notion of a “derived context” corresponding to embedded
clauses in ascriptions, and there are other formal variations of this notion (cf. Swanson
2010 for an alternative approach and further comparisons between names and other
presupposition-inducing expressions). In the course of giving any such account, it will
be an important fact that locally accommodated presuppositions can be bound by higher
operators. For instance, roughly speaking, (37) presupposes that every woman in the
room believes that she herself already met John:

(37) No woman in this room believes that she will meet John again.

For pragmatic theorists, similar readings of locally accommodated presuppositions are

apt to figure in explanations of syntactic considerations raised by bound readings of
attitude ascriptions such as (10). Just to take one example, advocates of Discourse Rep-
resentation Theory may claim it as an advantage that their frameworks easily model
such behavior (e.g. see Asher 1986 for a DR-theoretic analysis of puzzles about attitude
ascriptions, such as those in Kripke 1979). In any case, one can see that constraints
delivered by linguistics may be useful guides to further progress here.

4 Reflections
Semanticists are separated from philosophers of language in virtue of being alert to quite
different adjacent fields: morphology, phonology, and syntax as opposed to metaphysics,
epistemology, and the philosophy of mind, for instance. As a result, even linguists and
philosophers engaging in similar research programs sometimes find different questions
interesting and different arguments most useful for theory building. Examples given here
illustrate that the different perspective of the linguist is a valuable asset; it is a reason
why linguistics offers the philosophy of language so many useful tools for constraining
theory development. In this respect, linguistics is like other disciplines discussed in this
section of the volume, while unique in the extent to which some of its practitioners and
some philosophers are engaged in a common enterprise.
Ultimately the case studies canvassed here point to a simple conclusion: linguistics plays
a variety of distinctive roles in the philosophy of language. Linguists have contributed
empirical constraints on theories about classic topics in the philosophy of language, in
some cases thereby reinvigorating or redirecting otherwise settled debates. Linguists have
been in direct dialogue with philosophers as collaborators and opponents of particular the-
ories. And linguists sometimes continue philosophical debates until they may fairly claim
to have appropriated that debate as part of linguistics proper. On some occasions, linguists
and philosophers of language have operated in parallel on problems, and the role of lin-
guistics in philosophical discussions has not always been as great as it could have been.
Thomason & Pelletier 2002 point out one unique facet of the role of linguistics in
philosophy: “the results of linguistics—and especially the results of the branches of
linguistics concerned with semantics—provide direct intuitions of the sort that have
always served as the starting points of philosophical positions” (509). In light of this
observation, it is understandable that the role of linguistics in the philosophy of lan-
guage is not only varied but malleable. The question of what role linguistics should
play is in part a philosophical question about the role of intuitions in philosophical

521
 SA RA H MO S S

inquiry, and answers to this question vary according to the philosopher, the particular
inquiry under discussion, and the times. The roles played by intuitions in each field are
themselves evolving in the twenty-first century as online engines become more refined
guides for linguists’ corpus searches and as experimental philosophy research expands,
inciting metaphilosophical debate. But even as the role of linguistics in the philosophy
of language is constantly changing, one may recognize it as a constant fact that the rela-
tionship between the disciplines is greatly productive for both.

References
Asher, Nicholas. 1986. “Belief in Discourse Representation Theory.” Journal of Philosophical Logic, vol. 15
(2): 127–189.
Austin, J. L. 1956. “Ifs and Cans.” In Philosophical Papers, J. O. Urmson & G. J. Warnock, editors. Oxford
University Press, Oxford, first edn.
Barwise, John & Robin Cooper. 1981. “Generalized Quantifiers and Natural Language.” In Portner & Partee
(2002), 75–126.
Cappelen, Herman & Ernest Lepore. 2002. “Indexicality, Binding, Anaphora, and A Priori Truth.” Analysis,
vol. 62: 271–281.
——. 2005. Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act Pluralism. Blackwell Pub-
lishing Ltd., Malden, MA.
Chisholm, Roderick M. 1946. “The Contrary-to-Fact Conditional.” Mind, vol. 55 (220): 289–307.
Cohen, Jonathan & Samuel C. Rickless. 2007. “Binding Arguments and Hidden Variables.” Analysis, vol.
67 (1): 65–71.
Condoravdi, Cleo & Jean Gawron. 1996. “The Context-Dependency of Implicit Arguments.” In Quanti-
fiers, Deduction, and Context, Makoto Kanazawa, Christopher Piñón & Henriëtte de Swart, editors, 1–32.
CSLI Publications, Stanford.
Cresswell, Maxwell & Arnim Stechow. 1982. “De Re Belief Generalized.” Linguistics and Philosophy, vol. 5 (4):
503–535.
Crimmins, Mark. 1992. Talk About Beliefs. MIT Press, Cambridge.
Davidson, Donald. 1967. “The Logical Form of Action Sentences.” In The Logic of Decision and Action,
Nicholas Rescher, editor, 81–95. University of Pittsburgh Press, Pittsburgh.
Elbourne, Paul. 2008. “The Argument From Binding.” Philosophical Perspectives, vol. 22 (1): 89–110.
von Fintel, Kai. 1999. “NPI Licensing, Strawson Entailment, and Context Dependency.” Journal of Seman-
tics, vol. 16: 97–148.
——. 2001. “Counterfactuals in a Dynamic Context.” In Ken Hale: A Life in Language, Michael Kenstowicz,
editor. MIT Press, Cambridge.
Frege, Gottlob. 1892. “On Sinn and Bedeutung.” In The Frege Reader, Michael Beaney, editor, 151–171.
Blackwell Publishers, Ltd., Oxford. Translated by Max Black.
Geurts, Bart. 1997. “Good News About the Description Theory of Names.” Journal of Semantics, vol. 14:
319–348.
——. 1999. Presuppositions and Pronouns. Elsevier, Amsterdam.
Gillies, Thony. 2007. “Counterfactual Scorekeeping.” Linguistics and Philosophy, vol. 30: 329–360.
Glanzberg, Michael. 2007. “Definite Descriptions and Quantifier Scope: Some Mates Cases Reconsidered.”
European Journal of Analytic Philosophy, vol. 3: 133–158.
Goodman, Nelson. 1947. “The Problem of Counterfactual Conditionals.” Journal of Philosophy, vol. 44 (5):
113–128.
Grice, Paul. 1987. “Logic and Conversation.” In Studies in the Way of Words. Harvard University Press,
Cambridge.
Hamblin, Charles. 1973. “Questions in Montague English.” Foundations of Language, vol. 10: 41–53.
Higginbotham, James. 2002. “On Linguistics in Philosophy, and Philosophy in Linguistics.” Linguistics and
Philosophy, vol. 25 (5–6): 573–584.
Kamp, Hans. 1971. “Formal Properties of ‘Now’.” Theoria, vol. 37: 227–274.
——. 1981. “A Theory of Truth and Semantic Representation.” In Formal Methods in the Study of Language:
Mathematical Centre Tracts 135, J. Groenendijk, T. Janssen & M. Stokhof, editors, 277–322. Stichting
Mathematisch Centrum, Amsterdam.

522
 TH E RO LE O F LI N G U I S T I C S

Kaplan, David. 1977. “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology
of Demonstratives and Other Indexicals.” In Themes from Kaplan, Joseph Almog, John Perry & Howard
Wettstein, editors, 481–563. Oxford University Press, Oxford.
King, Jeffrey C. 2001. Complex Demonstratives: A Quantificational Account. MIT Press, Cambridge.
——. 2010. “Syntactic Evidence for Semantic Claims.” Manuscript, Department of Philosophy, Rutgers
University.
Kripke, Saul. 1972. “Naming and Necessity.” In Semantics of Natural Language, Donald Davidson & Gilbert
Harman, editors, 253–355. Reidel, Dordrecht.
——. 1979. “A Puzzle About Belief.” In Readings in the Philosophy of Language, Peter Ludlow, editor, 875–920.
MIT Press, Cambridge.
——. 2009. “Presupposition and Anaphora: Remarks on the Formulation of the Projection Problem.” Lin-
guistic Inquiry, vol. 40 (3): 367–386.
Ladusaw, William. 1979. Polarity Sensitivity as Inherent Scope Relations. Ph.D. thesis, University of Texas
at Austin.
Lepore, Ernest & Kent Johnson. 2002. “Does Syntax Reveal Semantics?: A Case Study of Complex Demon-
stratives.” Philosophical Perspectives, vol. 16: 17–41.
Lewis, David K. 1970. “General Semantics.” In Lewis (1983), 189–232.
——. 1973. Counterfactuals. Basil Blackwell Ltd., Malden, MA.
——. 1975. “Adverbs of Quantification.” In Formal Semantics of Natural Language, Edward L. Keenan, editor,
3–15. Cambridge University Press, Cambridge U.K.
——. 1979a. “A Problem About Permission.” In Papers in Ethics and Social Philosophy. Cambridge University
Press, Cambridge U.K.
——. 1979b. “Scorekeeping in a Language Game.” In Lewis (1983), 233–249.
——. 1983. Philosophical Papers, vol. 1. Oxford University Press, Oxford.
Ludlow, Peter. 1996. “The Adicity of ‘Believes’ and the Hidden Indexical Theory.” Analysis, vol. 56 (2):
97–101.
Maier, Emar. 2009. “Proper Names and Indexicals Trigger Rigid Presuppositions.” Journal of Semantics, vol.
26 (3): 253–315.
Marti, Luisa. 2006. “Unarticulated Constituents Revisited.” Linguistics and Philosophy, vol. 29: 135–166.
Mitchell, Jonathan. 1986. The Formal Semantics Point of View. PhD. dissertation, Department of Linguis-
tics, University of Massachusetts, Amherst.
Montague, Richard. 1970. “Universal Grammar.” Theoria, vol. 36: 373–398.
——. 1973. “The Proper Treatment of Quantification in Ordinary English.” In Portner & Partee (2002),
17–34.
Moss, Sarah. 2010a. “Subjunctive Credences and Semantic Humility” Forthcoming in Philosophy and Phe-
nomenological Research.
——. 2010b. “Four-Dimensionalist Theories of Persistence.” Forthcoming in Australasian Journal of
Philosophy.
——. 2010c. “On the Pragmatics of Counterfactuals.” Forthcoming in Noûs.
Ninan, Dilip. 2010. “Semantics and the Objects of Assertion.” Linguistics and Philosophy, vol. 33 (5):
335–380.
Nunberg, Geoffrey. 2002. “Do You Know What It Means to Miss New Orleans?” Linguistics and Philosophy,
vol. 25 (5–6): 671–680.
Partee, Barbara H. 1989. “Binding Implicit Variables in Quantified Contexts.” In Partee (2004a), 259–281.
——. 2004a. Compositionality in Formal Semantics. Blackwell Publishers, Ltd., Oxford.
——. 2004b. “Reflections of a Formal Semanticist.” In Partee (2004a), 1–25.
——. 2005. “Reflections of a Formal Semanticist as of Feb 2005.” Manuscript, University of Massachusetts,
Amherst.
——. 2009. “Lecture 12: Implicit Arguments and Point of View.” Manuscript, University of Massachusetts,
Amherst.
Perry, John. 1986. “Thought Without Representation.” Supplementary Proceedings of the Aristotelian Society,
vol. 60: 137–152.
——. 1998. “Indexicals, Contexts, and Unarticulated Constituents.” In Computing Natural Language, A.
Aliseda, R. van Gabeek & D. Westerståhl, editors, 1–11. CSLI Publications, Stanford.
Portner, Paul & Barbara H. Partee, editors. 2002. Formal Semantics: The Essential Readings. Blackwell,
Oxford.
Potts, Christopher. 2005. The Logic of Conventional Implicatures. Oxford University Press, Oxford.

523
 SA RA H MO S S

Pryor, Jim. 2007. “Reasons and That-Clauses.” Philosophical Issues, vol. 17 (1): 217–244.
Putnam, Hilary. 1975. “The Meaning of ‘Meaning’.” In The Twin Earth Chronicles, Andrew Pessin & Sanford
Goldberg, editors. M. E. Sharpe, Armonk, NY.
Richard, Mark. 1990. Propositional Attitudes: An Essay on Thoughts and HowWe Ascribe Them. Cambridge
University Press, Cambridge.
Roberts, Craige. 2009. “Names as Definites.” Ms., Department of Linguistics, Ohio State University.
Rothschild, Daniel & Gabriel Segal. 2009. “Indexical Predicates.” Mind and Language, vol. 24 (4):
467–493.
Salmon, Nathan. 1986. Frege’s Puzzle. MIT Press, Cambridge.
Schiffer, Stephen R. 1992. “Belief Ascription.” Journal of Philosophy, vol. 89 (10): 499–521.
——. 1996. “The Hidden-Indexical Theory’s Logical Form Problem: A Rejoinder.” Analysis, vol. 56 (2):
92–97.
Soames, Scott. 2002. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. Oxford Uni-
versity Press, Oxford.
Sommers, F. 1982. The Logic of Natural Language. Clarendon, Oxford.
Stalnaker, Robert C. 1968. “A Theory of Conditionals.” In Ifs: Conditionals, Belief, Decision, Chance, and
Time, William L. Harper, Robert Stalnaker & Glenn Pearce, editors, 41–55. D. Reidel, Dordrecht.
——. 1970. “Pragmatics.” In Stalnaker (1999), 31–46.
——. 1974. “Pragmatic Presuppositions.” In Stalnaker (1999), 47–62.
——. 1978. “Assertion.” In Stalnaker (1999), 78–95.
——. 1988. “Belief Attribution and Context.” In Stalnaker (1999), 150–166.
——. 1999. Context and Content. Oxford University Press, Oxford.
Stanley, Jason. 2000. “Context and Logical Form.” Linguistics and Philosophy, vol. 23 (4): 391–434.
Stanley, Jason & TimothyWilliamson. 2001. “Knowing How.” Journal of Philosophy, vol. 98 (8): 411–444.
Swanson, Eric. 2006. “Interactions with Context.” PhD. dissertation, Department of Linguistics and Phi-
losophy, MIT.
——. 2008. “Modality in Language.” Philosophy Compass, vol. 3 (6): 1193–1207.
——. 2010. “Propositional Attitudes.” Forthcoming in Semantics: An International Handbook of Natural Lan-
guage Meaning, edited by Claudia Maienborn, Klaus von Heusinger, and Paul Portner.
Thau, Michael. 2002. Consciousness and Cognition. Oxford University Press, Oxford.
Thomason, Richmond & Francis Pelletier. 2002. “Twenty-Five Years of Linguistics and Philosophy.”
Linguistics and Philosophy, vol. 25 (5–6): 507–529.
Veltman, Frank. 1996. “Defaults in Update Semantics.” Journal of Philosophical Logic, vol. 25 (3).
Wolter, L. 2009. “Demonstratives in Philosophy and Linguistics.” Philosophy Compass, vol. 4 (3): 451–468.

524
 4.3
THE R OLE OF
P S YCHO LOGY
Robert J. Stainton

1 Introduction
Are psychological facts relevant to philosophy of language; and, in particular, does
scientific psychology have a legitimate role to play? For example, is it methodologi-
cally permissible for philosophers of language to rely upon evidence from neurological
development, experiments about processing, brain scans, clinical case histories, longi-
tudinal studies, questionnaires, etc.? If so, why? These two questions are the focus of
this survey.
Psychology may seem obviously relevant. I thus begin by introducing arguments
against relevance, to motivate the discussion. I will urge that these ultimately fail, and
that the appearance of relevance should be taken at face value. In this first section, the
focus is on the possible relevance of psychological evidence available to the lay person.
Next, I introduce positive arguments for relevance, from examples—shifting the spot-
light to recherché evidence drawn from scientific psychology. To foreshadow the main
conclusion, psychology, including specifically the methods and results of contemporary
cognitive psychology, are relevant because there are connections, both necessary and
contingent, between language and the human mind.

2 Two Arguments for Irrelevance and Why They Fail

The negative portion of the paper will focus on two arguments with the same conclu-
sion. Here is the first:

P1: If philosophy of language is not beholden to facts about natural languages,

then empirically discovered facts about human psychology are irrelevant
to philosophy of language.
P2: Philosophy of language is not beholden to facts about natural languages.
C: Empirically discovered facts about human psychology are irrelevant to
philosophy of language.

P1 requires little comment. It is hard to see how human psychology could be probative,
if facts overtly about human language are not. What reasons can be given for P2? One
might urge that philosophy in its entirety is a priori. Let us ignore this sweeping, radical
 RO BERT J . STA I N T O N

idea. More promising is the view that philosophy of language is not “about” natural
language. It is about all possible languages, one might say. Or alinguistic propositions.
Or how to construct a more perfect language. A thorough response would require an
entire chapter. I will content myself with making three brief points. First, the issue is not
whether philosophy of language is about actual spoken languages but rather whether it
is beholden to empirical results about them. Second, philosophy of language has long
paid attention to facts about actual, spoken languages. Historically, philosophers from
Plato and Aristotle, through the Cartesians, Hobbes and Locke, were concerned with
actual language, and they adduced evidence (as they saw it) from that domain. Similarly,
for more recent theorizing: think of recent debates about compositionality, inferential-
ism in semantics, and the semantics-pragmatics boundary; the question has not been
whether a language could have, or should have, such-and-such properties. Nowadays,
a great deal of philosophy of language makes use of empirical evidence from theoreti-
cal syntax. Third, and finally, I think the tradition has it right: philosophy of language
should be beholden to facts about human language. Reflection upon why philosophers
care about language makes this clear.
Here are examples of philosophical appeals to language, chosen essentially at random.
Arguably, the most central debate in 20th-century philosophy of language has been
about which expressions are definite descriptions, and what definite descriptions mean.
Are descriptions devices of reference, or of quantification? Do sentences containing
definite descriptions entail existence, semantically presuppose it, or merely pragmati-
cally suggest it? Philosophers have also argued about whether ‘know’ is context sensitive
in a way that would address skeptical puzzles. They’ve asked, in the context of analyz-
ing perception, whether ‘S saw a green apple’ entails ‘It seemed to S that something
was green’, or whether it is even consistent with the latter. Finally, concerned with
the nature and existence of moral facts, they have debated whether ethical sentences
such as ‘Abortion is immoral’ have truth conditions. Though I cannot argue the point
here, what seems to matter to these larger debates in epistemology, philosophy of mind,
and ethics is not merely whether there could be, or should be, sentences (or alinguistic
propositions) with the requisite features.
The first argument fails because the antecedent of the major premise is too sweep-
ing to be plausible. The second tries to get by with something weaker. It also adds an
“ontological twist”:

P3: If philosophy of language is beholden to facts about natural languages, but

not to psychological facts, then empirically discovered facts about human
psychology are irrelevant to philosophy of language.
P4: Natural languages are not psychological entities.
P5: If natural languages are not psychological entities, then philosophy of lan-
guage is not beholden to psychological facts.
C: Empirically discovered facts about human psychology are irrelevant to
philosophy of language.

To understand the argument, a bit of background will be useful. According to a long-

standing philosophical tradition, words are the primary locus of meanings; what they
mean are ideas, i.e., items in the mind; and logic describes the mental rules for manip-
ulating ideas. Gottlob Frege revolutionized logic and philosophy of language in the
late 19th century by insisting, instead, that theorists look first at complete sentences.

526
 TH E RO LE O F PS Y C H O L O G Y

Doing so, we find that meanings are formally structured propositions. For instance,
the content of the sentence ‘It’s not the case that Aristotle is alive’ can be decom-
posed into three quite different semantic parts, none of which has to do with images,
sensations, etc. And logic describes not mental operations but truth-theoretic rela-
tions among such contents. Thus, as per P4, neither lexical meanings nor semantic
relations are ideas or mental processes. Frege also played a key role in the develop-
ment of artificial logical languages. This provided further support, by affording an
alternative model for the ontology of languages: languages are abstract entities, not
psychological ones.
Jerrold Katz offers another argument for P4. Consider a perfectly commonsensical
distinction between an object O and the human cognizing of O. Mathematics pro-
vides paradigm examples: geometry, for instance, is by no means the same as human
cognition of geometry. By the same token, Katz (1984: 193) would have us draw “a
fundamental distinction between the knowledge speakers have of their language and the
languages that speakers have knowledge of . . . . [T]he subject-matter of linguistics is, in
this sense, independent of psychological sciences—just as the subject-matter of logic
and mathematics is independent of the sciences concerned with people’s logical and
mathematical ability”. It is only by running these together that languages appear to be
ontologically psychological.
Having explained the second argument, let us now evaluate it. I purposely couched
P4 and P5 so as to equivocate between two issues. One is whether languages dwell
within the mind in the way that dreams, pains, and hallucinations do. This is the issue
that Frege and Katz’s arguments pertain to. These notwithstanding, many philosophers
and linguists follow Noam Chomsky, and insist that languages are mental things in
just this sense. (If so, it is plain why psychology would be relevant.) However, we may
set that entire debate to one side: The relevance of psychology does not require that
languages be ideas. Let’s focus on another reading of ‘psychological entity’—namely, an
entity with some kind of necessary connection to (human) psychology. For, the argu-
ment’s soundness requires that P4 be sustained on that sense as well.
Are languages “psychological entities” in this weaker sense? Languages have both a
meaning and sounding side, and they are psychological on both counts. Beginning with
the combinatorial aspects of language, the requirements that the rules of syntax be recur-
sive and those of semantics be compositional are driven by psychological considerations,
specifically by the empirical fact about human psychology that our minds are finite. That
is, natural languages are constrained essentially by our psychology in the sense that, what-
ever their nature, our limited minds must be able to grasp them. (A similar lesson can be
drawn from the fact that no truly massive list of sentences could be learned by minds like
ours, in the time available.) Turning from combinatorial rules to particular expressions, as
Barbara Partee (1979) has stressed, some contents are out-and-out mentalistic: proposi-
tional attitude sentences; terms for pain, dreams, mental images, emotions; etc.
What’s more, not just exceptionally but in general, linguistic contents are linked
necessarily with human psychology. As Chomsky (2000) and Jackendoff (1983, 2002)
often insist, many linguistic contents pertain to “objects for us.” The things they
mean are essentially connected to our properly human interests and points of view.
Beginning with two obvious cases, human languages contain words for complex social
entities: mortgages, Bollywood, Tuesday. They also include words for fictional and
mythical entities. (For example, being an atheist, I think it impossible, wholly inde-
pendently of human psychology, to give the meanings of: ‘venal sin’, ‘purgatory’, and

527
 RO BERT J . STA I N T O N

‘transubstantiation’; ‘nirvana’, ‘Navaratri’ and ‘Ganesh’; ‘Qiyamah’ and ‘Ramadan’;

etc.) Other expressions, while picking out ordinary physical objects, nonetheless have
contents which are response-dependent. Consider ‘cloud’, ‘tea’, ‘pet’, ‘green’, ‘weapon’,
and ‘bush’. A collection of water droplets is a cloud only if it is perceptible by humans
using the naked eye; a pet is an animal that humans treat as such; not all green things
share a physically specifiable reflectance property but instead are those items treated
“greenly” by the human visual system; and so on. To be clear: the point is not that
clouds and pets, mortgages and Bollywood, or even sins, religious holidays and gods,
dwell within the human mind. (That is the reading of P4 which we set aside.) None-
theless, the content of all these words is essentially tied to human psychology. Such
are what many, maybe even most, of our quotidian substantives express, and what our
demonstratives are prone to pick out in context. Thus P4 properly construed is not so
much subject to a few scattered exceptions but is root-and-branch false.
Equally, the “sounding” side of language cannot be individuated independently of
human psychology. The philosophical tradition notwithstanding, linguistic expressions
are not pairings of acoustic patterns, or waves, with meanings—they are, rather, pair-
ings of something more like clusters of contrastive phonological features with mean-
ings. And the latter are creatures of human psychology. Given space constraints, the
details must be set aside. But the point can be made by means of commonsense exam-
ples. Think of the various pronunciations of ‘cotton’—by a child, a woman, someone
with a cold, someone whose larynx has been removed, a second-language learner, or a
computer. These are acoustically quite different from one another. What unites such
signals is how human minds process them: we hear them as the same. Or again, to the
ears of an Anglophone ‘van’ and ‘ban’ are different sounds. But for Spanish speakers,
these are the same linguistic sound—because ‘v/b’ is not a meaning-carrying contrast
in that language. And English speakers treat the ‘t’ of ‘table’, ‘tea’, and ‘writer’ as the
same sound, though their articulation and acoustics are different. Conversely, the very
same acoustic wave can realize various linguistic sounds, in different contexts, because
of varying semantic import. Finally, acoustic patterns per se do not have word bounda-
ries, syllables, vowels, an onset-coda contrast, intonation contours, etc. (See Isac &
Reiss 2008: 30–32, 109–114 for introductory discussion.) In short, not just meanings but
“linguistic sounds” are necessarily connected to human psychology.

3 A Positive Case for Relevance: Three Examples

I began with two arguments for irrelevance. These helped motivate the questions of
whether psychology is relevant to philosophy of language and, if so, why. Roughly
speaking, the first very sweeping argument failed because of some of the ends that phi-
losophy of language serves. The second, more circumspect, ontology-based, argument
failed because it equivocated on ‘psychological entity’: reading that phrase such that P4
is prima facie plausible, P5 is not; reading it such that P5 is prima facie plausible, P4 is not.
Those having been rebutted, we may now revert to the null hypothesis, namely that
psychological facts are indeed relevant. This speaks to the matter of “whether.” With
respect to “why,” insofar as there are necessary connections between natural languages
and human psychology, it would be surprising indeed should evidence about the latter
prove irrelevant to the former.
I will now provide a positive case for relevance on the basis of three examples.
Notably, they illustrate not just that ordinary, everyday psychology is relevant to

528
 TH E RO LE O F PS Y C H O L O G Y

philosophy of language, but that recherché evidence from advanced scientific psy-
chology is. This, of course, further buttresses the positive answer to our first question.
The examples also show that, pace P5 even on its weak reading, evidential relevance
does not require a necessary connection among (a) a background philosophical issue;
(b) a corresponding question about actual human language; and (c) evidence from
psychology. This yields a second answer to our “why” question, in terms of merely
contingent connections.

Proper Names
In laying out the examples, it will be useful to recall a thesis which traces, via W.V.O.
Quine, to Pierre Duhem. They note that it is not merely a hypothesis all on its own which
yields predictions but rather a hypothesis taken in conjunction with ancillary com-
mitments. Now, hypotheses of philosophical interest are no exception to this rule. In
particular, a properly philosophical hypothesis, conjoined with the right ancil-
lary hypotheses, can yield a prediction which is verified or falsified by scientific
psychology.
For instance, to introduce our first hypothesis H, it has been suggested that names
share the content of descriptions of persons. There are clear philosophical motivations for
H, both metaphysical and epistemological. Very roughly, if proper names are synony-
mous with descriptions, then nonreferring ones can have meaning without undue onto-
logical commitment; they won’t embed salva veritate in certain modal contexts; and per-
sonal identity will presumably depend upon the retention of the descriptive properties.
What’s more, turning to epistemology, deploying a name can afford knowledge of the
person absent acquaintance with its bearer; and propositional attitudes will not pertain
to “persons themselves,” but rather to descriptive contents. (It is precisely because H
is motivated philosophically in these ways, that it is a hypothesis within philosophy of
language.) Now, H itself is not a claim about human psychology. It may even be, pace
the discussion above, that the truth or falsity of H is metaphysically independent of
human psychology. Nonetheless, the methods and results of scientific psychology are
relevant to its evaluation. That is because H can be (dis)confirmed in conjunction with
ancillary commitments A, including postulated laws.
As Valentine et al. (1996) explain at length, proper names are psychologically
special. Processing them is notoriously slow; they are especially vulnerable to brain
damage; and, as experience of “tip of the tongue” moments makes intuitively clear,
names are hard to remember. (Experiments confirm this: when subjects are presented
with unfamiliar faces, and taught various facts about the people shown, including their
names, the latter prove significantly harder to recall. See Cohen and Faulkner 1986.) It
seems, and this is another element of A, that the best explanation of this “specialness,”
is some kind of psychological law pertaining to the content of names; that, rather than
their form, seems to be why they pattern this way. H and A together yield a prediction
P, namely that a synonymous description of a person will exhibit the same psychologi-
cal profile as the corresponding name. However, as experimental, clinical, and other
work shows, this is not the case. Thus, a hypothesis in philosophy of language is called
into question by evidence from sophisticated scientific psychology. (See Segal 2001 for
detailed and careful discussion.)

529
 RO BERT J . STA I N T O N

Knowledge Attributions
There has been much philosophical debate about the conditions under which knowl-
edge attributions are strictly speaking true. One motivation for the debate is a Moore-
style argument against external world skepticism. The idea goes that, as ordinary speak-
ers use the sentence ‘Rob knows that Hitler is dead’, the sentence is true; and, so used,
the truth of this sentence entails that Rob has knowledge of the external world. So,
such knowledge exists. A natural reply to this argument is to insist that, appearances
notwithstanding, the sentence is false. This is H. Of course, one cannot just stop there:
one must explain away the strong intuition that the sentence is true. The reply thus
continues: ‘Rob knows that Hitler is dead’ strikes people as true because they confuse
what it literally says with things that its usage merely conversationally implicates, for
example, that Rob’s belief is warranted, that one can safely proceed on his say-so, etc.
This explanation of the appearances is A. Again, H is not a claim about the human
mind. Maybe its truth is not even conceptually, logically, or ontologically connected
to cognitive scientific facts. Nonetheless, H conjoined with A yields a prediction: that
people who cannot grasp conversational implicatures will reject everyday knowledge
attributions as false. This prediction, P, is disconfirmed by evidence from speakers with
Autism Spectrum Disorders (ASDs): despite exhibiting good formal language abilities,
people with ASDs suffer from serious pragmatic impairments, including in particular
failing to notice conversational implicatures. Yet, pace P, they do not present as external
world skeptics. (Other deficits should yield related (dis)confirmation, by the way—spe-
cifically Semantic-Pragmatic Disorder and Right Hemisphere Dysfunction. For a useful
survey, see Asp and de Villiers (2010) and Cummings (2009). Incidentally, ASDs afford
another interesting evidential link between clinical psychology and philosophy of lan-
guage. Linguistically able people with ASDs famously have trouble attributing complex
mental states to others. This would seem to call into question the Gricean (1957) idea
that meaning something, for example making a statement, of necessity involves higher-
order intentions. See Andrews 2002 and Gluer and Pagin 2003 for discussion.
The import of the foregoing examples is not, of course, that considerations from
scientific psychology have established that names are directly referential rather than
descriptive, or that external world skepticism is false. The point is that, even barring
necessary connections, Duhem’s Thesis yields links between philosophical hypoth-
eses involving language and evidence from cognitive psychology: Because of diverse
and unimagined contingent (including nomic) connections, “properly philosophical”
hypotheses about language give rise, in unpredictable ways, to empirical predictions
about the mind.

Assertion
A final example illustrates just how far removed the background philosophical issues
can be from the curious psychological findings. In the context of defending metaphysi-
cal antirealism, it has been proposed that truth be explicated in terms of warranted
assertion. Doing so, of course, means that assertion itself cannot be analyzed in terms
of truth: that would be circular. A promising way out is to explain the speech act of asser-
tion socially/formally, specifically in terms of the use of declarative sentences in conventionally
specified conditions (Dummett 1973). This is H. Now, if H holds, then, appearances
notwithstanding, subsentential words and phrases cannot be used to assert. Instead, for

530
 TH E RO LE O F PS Y C H O L O G Y

instance, when someone displays a letter and says of it ‘From Chomsky’, what they pro-
duce must be an elliptical declarative sentence. This attempt to explain away a seeming
counterexample to H yields P: someone who cannot engage in sentential ellipsis cannot
make such an assertion. But P is open to psychological (dis)confirmation. In particular,
people who cannot yet, or can no longer, form or comprehend elliptical sentences, can
nonetheless make and understand assertions with bare words and phrases. (See Stainton
2006 for extended discussion.) Granted, it seems outlandish that antirealism and the
nature of truth could be tied to aphasias and child language. And yet, the former have
been connected, by hypothesis, to the speech act of asserting; and thence to a formative
of a certain kind, the declarative; and this philosophical line of thought necessitates
endorsing a linguistic conjecture about ellipsis; and so, by a long and winding road, the
whole account becomes subject to arcane psychological evidence.

4 A Parting Shot: Irrelevance In Practice and In Principle

My contention in this chapter, with respect to our first question, has been that psychol-
ogy, including advanced scientific psychology, does have a legitimate evidential role
in practice. With respect to the second question, namely why this should be so, the
main lesson is this: the actual nature of human language is relevant to certain properly
philosophical questions; and, because of both contingent and necessary connections,
human psychology is relevant to the actual nature of human language; that is why, by
the transitivity of ‘relevant to’, psychology is relevant to philosophy of language.
I have not restricted myself to the issue of whether such evidence is irrelevant in
principle. Surprisingly, certain philosophers have rejected the latter view as well. So, I
will end with a word about it.
It does remain epistemically possible that philosophy of language is disconnected
from psychology in general, and scientific psychology in particular. Hence it could still
turn out, in the long run, that MRIs, error rates, etc., are of no use. Nonetheless, as
Fodor (1981) has stressed, as long as we do not know what is in fact connected to what,
and do not know in particular what linguistic facts supervene on, we should not rule out
psychological evidence a priori. Consider again in this light the three examples above.
What rebuttals are rationally permissible? It would be natural to insist that the alleged
psychological data are incorrect, or at least oversimplified. Or again, one could grant
that the data are accurate but reply that it can be accommodated to, or even supports,
the philosophical account at issue. For example, one might scrutinize more carefully the
speech of people with ASDs, and counter that they actually do use ‘know’ as the anti-
Moorean predicts. Maybe names and descriptions do indeed share the same meaning
but behave differently with respect to processing for some other reason. And maybe, as
Stanley (2000) has argued, some apparent uses of words and phrases are elliptical after
all, while others simply do not amount to assertions. All these rebuttals are appropri-
ate. However, they involve bringing in more psychological evidence, not less. In con-
trast, what should emphatically not be accepted is simple dogmatic defiance: “That’s all
very fascinating. But I refuse even to offer a response. Those are psychological findings,
and I already know full well, on the basis of metaphilosophy and the metaphysics of
language(s), that such results cannot pertain to the philosophy of language.”
This alone shows that, even if at the idealized end of inquiry there turned out to be
no connection between philosophy of language and psychology, nonetheless the latter
should not now be dismissed out of hand. Put otherwise, only an omniscient creature

531
 RO BERT J . STA I N T O N

could have a handle on what “the considerations in principle relevant to philosophy of

language” are. Not being such creatures, we are well advised to pursue evidence which
seems relevant—even when it comes from psychology.

References
Andrews, Kristin (2002). “Interpreting Autism”. Philosophical Psychology 15(3): 317–332.
Asp, Elisa and J. de Villiers (2010). When Language Breaks Down: Analysing Discourse in Clinical Contexts.
London: Routledge.
Chomsky, Noam (2000). New Horizons in the Study of Language and Mind. Oxford: Cambridge University
Press.
Cohen, G. and D. Faulkner (1986). “Memory for Proper Names.” British Journal of Developmental Psychology
4: 187–197.
Cummings, Louise (2009). Clinical Pragmatics. Cambridge: Cambridge University Press.
De Villiers, Jessica, Robert J. Stainton and Peter Szatmari (2007). “Pragmatic Abilities is Autism Spectrum
Disorder”. In P.A. French and H.K. Wettstein (eds.) Philosophy and the Empirical. Volume XXXI of Mid-
west Studies in Philosophy. Oxford: Blackwell, pp. 292–317.
Dummett, Michael (1973). Frege: Philosophy of Language. London: Duckworth.
Fodor, Jerry (1981). “Some Notes on What Linguistics Is About”. In N. Block (ed.) Readings in Philoso-
phy of Psychology, Volume 2. Cambridge: Harvard University Press, pp. 197–207. Reprinted in J.J. Katz
(ed.)(1985). The Philosophy of Linguistics. Oxford: Oxford University Press, pp. 146–160.
Gluer, Kathrin and Peter Pagin (2003) “Meaning Theory and Autistic Speakers.” Mind and Language 18(1):
23–51.
Grice, H. Paul (1957). “Meaning.” Philosophical Review 66: 377–388.
Isac, Daniela and Charles Reiss (2008). I-Language. Oxford: Oxford University Press.
Jackendoff, Ray (2002). Foundations of Language. Oxford: Oxford University Press.
—— (1983). Semantics and Cognition. Cambridge, MA: The MIT Press.
Katz, Jerrold (1984). “An Outline of Platonist Grammar.” In T.G. Bever, J.M. Carroll, and L.A. Miller (eds.)
Talking Minds. Cambridge: The MIT Press, pp. 1–33. Reprinted in J.J. Katz (ed.)(1985). The Philosophy of
Linguistics. Oxford: Oxford University Press, pp. 173–203
Partee, Barbara (1979). “Semantics—Mathematics or Psychology?” In R. Bauerle, U. Egli, and A. von Ste-
chow (eds.) Semantics from Different Points of View. Berlin: Springer, pp. 1–14.
Segal, Gabriel (2001). “Two Theories of Names.” Mind and Language 16(5): 547–563.
Stainton, Robert J. (2006). Words and Thoughts: Subsentences, Ellipsis and the Philosophy of Language. Oxford:
Oxford University Press.
Stanley, Jason (2000). “Context and Logical Form.” Linguistics and Philosophy 23: 391–434.
Valentine, Tim, Tim Brennen, and Serge Bredart (1996). The Cognitive Psychology of Proper Names. London:
Routledge.

532
 4.4
THE R OLE OF
MATHE MAT IC AL
ME TH OD S
Lawrence S. Moss

1 On the Contribution of Mathematics to the Study of Language

Mathematical models have contributed to our understanding of linguistic phenomena.
This chapter makes the case for this, and at the same time it tries to do so in a way that
highlights problematic points. I would like to start with a few theses about the uses of
mathematics in linguistics:

(1) Applying mathematical ideas has changed the kinds of questions we ask
about language.
(2) Applying mathematical ideas has enabled us to use metaphors from
mathematics.
(3) Applying mathematical ideas has facilitated explorations of structure.
(4) In order to get computationally sensible models of language, it is necessary
to apply a range of mathematical techniques and ideas.

I am going to explore all of these points, in a general way in this section and then in
expanded case studies in the remainder of the chapter. The overall point is that math-
ematical modeling in the study of language should be seen in the same light as mathemati-
cal modeling in other areas, say physics or sociology; it is a useful tool, perhaps not the
only useful tool, but a very useful one nonetheless: At the same time, it must be admit-
ted that not all applications of mathematics are successful. Finally, it is a difficult matter
in the philosophy of mathematics to account for the very usefulness of mathematics
in the first place. Why should it be that a discipline which traffics in abstractions be
so connected to the concrete? How is it that facts whose truth seems necessary come
to be connected to a world so contingent? We shall not take up these issues here; they
are far beyond our scope. But they form a backdrop to the issues that is worth keeping
in mind.
The first point about the questions we ask is partly supported by a quote from Noam
Chomsky’s book Syntactic Structures, arguably among the most influential books in the
history of linguistics:
 LAW REN CE S . MO S S

Precisely constructed models for linguistic structure can play an important role,
both positive and negative, in the process of discovery itself. By pushing a pre-
cise but inadequate formulation to an unacceptable conclusion, we can often
expose the exact source of this inadequacy and, consequently, gain a deeper
understanding of the linguistic data. More positively, a formalized theory may
automatically provide solutions for many problems other than those for which
it was explicitly designed. I think that some of those linguists who have ques-
tioned the value of precise and technical development of linguistic theory have
failed to recognize the productive potential in the method of rigorously stating
a proposed theory and applying it strictly to linguistic material with no attempt
to avoid unacceptable conclusions by ad hoc adjustments or loose formulation.

We take a “precisely constructed model” to be a mathematical model, even if no math-

ematics is used. What Chomsky means here by a “negative” role is that even a theory
which leads to bad predictions might help people construct a better theory. It is more
interesting that the “positive role” that he mentions is not just that one can use a model
to solve a problem, it is that the model might enable one to find “solutions for many
problems other than those for which it [the theory] was explicitly designed.”
My second point, on metaphors, is a point worth making on practically all uses of
mathematics. One rarely uses the deepest results of any topic in mathematics a instead
it is the; transfer of ideas from the abstract world of mathematics to the applied area.
For example, when one sees applications of dynamical systems in cognitive science (see,
e.g., Port and van Gelder (1995)), it is largely the metaphors related to concepts like
attractor connected to the long-term qualitative behavior of dynamical systems. One
rarely sees applications of the best-known theorems in the subject. For linguistics, this
point is obscured when we turn to most uses of mathematics in areas like syntax, for
there the mathematics was developed specifically with the application, or something
like it, in mind. However, it does hold for uses of logic. I cannot think of any specific
results from logic that are used in any area of linguistics. To be sure, one finds uses of
results such as the Cut Elimination Theorem for various logical systems used in connec-
tion with grammar formalisms, to mention just one example. But nothing is used in a
widespread way. On the other hand, some of the underlying ideas of modern logic such
as the syntax/semantics distinction and the notion of formal derivation are as important in
formal semantics as they are in logic. This point on metaphors will be discussed a bit
more when we turn to the examples of phylogenetic reconstruction.
Point (3) is motivated by the fact that language comes to us without any evident
structure. It is up to theoreticians to propose whatever structures they think are useful,
and these include structures related to all areas of language from the phonological level
to the level involving the evolution of human language. Mathematical models are the
primary way that scientists in any field get at structure.
We now take up point (4) on the use of mathematical ideas in computational settings
involving language. I am thinking of natural language processing (NLP) and computational
linguistics (CL). Among the fields of mathematics that have been used in connection with
NLP and CL are: automata theory (many of the tools in actual use come from this area),
linear algebra (in connection with latent semantic analysis, to mention just one illustra-
tion), algorithms and computational complexity theory, and probability theory and statis-
tics. It is hard to imagine much serious work in NLP and CL without these mathematical
tools. (Incidentally, this point might be what Chomsky had in mind in the quote above by

534
 TH E RO LE O F M ATH EM AT I C A L ME T H O D S

using a model to find solutions to problems other than the one in which one was originally
interested. His pioneering results on regular and context-free grammars are not muchused
in syntax; they are practically forgotten there. But they are of great use in NLP and CL.)
Going further than linguistics, let me make a point related to the uses of mathematics
in the broader field of cognitive science. When one thinks of mathematics as an applied
discipline, the leading application areas would probably be physics and engineering.
However, this is not the only way of motivating the subject: one could instead think
about the cognitive sciences, including linguistics, psychology, artificial intelligence,
and the like. A very large number of mathematical disciplines could instead be moti-
vated by their application in these fields. Indeed, one could well imagine that someday
these areas of applied mathematics might be seen as the centerpieces of the subject.

1.1 Objections to Mathematics

Before taking up objections to the theme of this chapter, we should mention a few
points which we are not claiming.

(5) We are not claiming that all of the applications of mathematics in linguis-
tics are correct or even interesting.
(6) We are not claiming that mathematical tools are the only valid ones for
the study of language.

Regarding the first point, it is actually difficult from our vantage point concerning
mathematical modeling to make the claim that one or another model is incorrect; the
strongest assertions we can make are that the end results of the modeling are empiri-
cally false, or that they conflict with what “everyone in the field knows,” or that they
lead to uninteresting questions. Even with this in mind, we do not claim that all uses of
mathematical ideas in any given field are useful or interesting. With regard to contem-
porary linguistics, this is trivial point: on any given debate, probably proponents of all
sides marshall arguments and theories that are on our view applications of mathematical
methods.
It might be good at this point to mention some studies related to language that do not
use mathematical methods. One example could be comparative philology, the study of
texts to infer relations between languages (exemplified by the deciphering of the Rosetta
Stone by Jean-François Champollion in 1922). Another, closer to this volume, could be
areas connected to the philosophy of language which have no trace of mathematics.
Here are a few objections to the use of mathematics in connection with the study of
language:

(7) In mathematics, everything is either true or false, black or white. Thus

mathematics would never be able to accommodate the kinds of degrees
of truth that we see in language: words partly mispronounced, sentences
partly grammatical, ideas that do not reflect anything with perfect clarity.
(8) Language is a psychological and social phenomenon. Therefore, the only
legitimate form of research on language is via psychology and sociology, or
other fields that place human phenomena at the forefront. Using mathe-
matics leads to a sterile science that ignores what is most interesting about
language.

535
 LAW REN CE S . MO S S

(9) There is nothing wrong with the use of mathematics in connection with
language. In principle, it could be a good thing. However, nothing so far
has been done that is of lasting value to linguistics.

One reply to objection (7) is that mathematics does not commit us to the view that the
objects it models bear the same kind of clear-cut distinctions that one does indeed find
inside of mathematics. To take one example, consider probability and its applications.
The standard theory of probability proceeds the same way as any field of mathematics:
There are definitions and axioms, and statements that either follow from the axioms or
which do not. People who work on the subject almost never disagree on claims about
what is true in the subject, and what is false. (Of course, they may well disagree on the
intended applications of the theory.) But all of the uses of probability theory are specifi-
cally about phenomena that are not clear cut, that exhibit randomness, chance, and
unpredictability. Turning specifically to language, we do indeed find many applications
of probability theory and statistics. In fact, there are many uses of probability in connec-
tion with syntax in natural language processing.
Nevertheless, there is an aspect of (7) that does seem apt. Many of the items in this
handbook are applications of logic. (Some of them are actually applications to logic, but
we set these aside for this discussion.) The predominant methodology in logic is indeed
that sentences are true or false. (There is a field called paraconsistent logic which holds
that sentences can be both true and false. The view that sentences might be neither true
nor false is fairly commonplace.) And so logic might be the wrong tool for doing certain
kinds of modeling. However, using mathematics does not commit us to using modern logic
(or any particular branch of mathematics), so criticism (7) seems ill-placed. To state our
reply again, in this chapter the use of mathematical methods is understood very broadly. It
encompasses neural networks for example: they enable one to use mathematical meta-
phors and ideas just as surely as other forms of modeling.
Objection (8) seems valid at first glance, but as soon as one looks into the literature
in the social sciences, one indeed does find applications of mathematical ideas and
results. So this objection would seem to fall flat. For example, serious work on human
memory makes use of mathematical models. (Alternatively, the objection to mathemat-
ics in linguistics would have to be widened to an objection to many other applications
of mathematics. With a wider circle of applications, it would seem to be a harder and
harder case to make.)
I mention the last objection mostly because it is forcefully made by Noam Chomsky
and yet contrasts sharply with his 1957 position. In 1993 he gave the Patten Lectures
at Indiana University. At a lecture attended by linguists, logicians, and philosophers,
he was asked what results of mathematics had the most implications for linguistics.
His answer was essentially that none of the applications had any real value, with one
possible exception (the Chomsky-Schützenberger Theorem). Further, he took a posi-
tion that I have outlined as objection (9). In a sense, this chapter as a whole is a kind of
answer to this objection.

2 Examples and Case Studies

The remainder of this chapter discusses two applications of mathematics to linguistics
which either expand on what we have already mentioned or which present issues that
are worthy of further discussion. If we had more space, we would have gone into details

536
 TH E RO LE O F M ATH EM AT I C A L ME T H O D S

on the application of game theory to pragmatics, since it would have fit the topic of the
handbook. For a collection of papers on this topic, see Benz et al. (2006). We might also
have discussed generalized quantifiers, and also the mathematics of learnability.

2.1 Phylogenetic Estimation in Historical Linguistics

One stunning application of mathematics to the study of language pertains to historical
linguistics and the reconstruction or estimation of the phylogenetic tree of languages. As
the name suggests, there is a parallel with biology. Consider the following trees:
The one on the left is intended as a representation of the evolution of a family of lan-
guages, with the edges representing descent in a very long time period. On the right we
have a similar tree for part of the primate family. Please do not take the content of these
trees seriously; we are only presenting them as examples. The trees in both historical
linguistics and in evolutionary biology share some key features: They represent long-
term change, they are partly derived from historical evidence (texts or bones), but the
evidence is scant, and they are frequently the topics of debate. The same mathematics
is used in both cases; it is mainly probability and algorithms. In fact, the same people
work on both topics.
From the point of view of this Chapter, the interesting question concerns the role of
the technical work itself, and so we return to the four points mentioned at the beginning
of this chapter. Work on phylogentic reconstruction may well change the questions his-
torical linguists ask. In broad terms, the way that one uses linguistic data to reconstruct
trees is to compare features of living languages such as cognate words (e.g., English one
and Spanish uno) or parallel syntactic structures (whether the language mainly has verbs
at the end or not). The actual choice of the features is controversial, and so the results
of work in the area could inform the practice in historical linguists. Further, the way one
uses the data in reconstructing or estimating the trees is also controversial. That is, there
is a host of algorithmic options, and very little in the way of actual records to know what
counts as a correct tree. One can propose a computer simulation of the evolution and then
run algorithms on this. But in the case of language, we again have the problem of justify-
ing the use of the simulations. In any case, the actual practice in this area might well con-
tribute to a clarification of the questions that people in the field ask. I take this to be the
most interesting connection to the topics of this chapter. In addition, we might well find
metaphors used in computational biology finding their way into historical linguistics.

Low
GermanERRRR
}} EE RRR
}} EE RRR ◦ OOO
E OOO
Old Anglo- Low Fran-  OO
Frisian Saxon
O. Saxon
conian   ◦ MMM
  MMM
    M
◦=
Mid. Eng.
Mid.-Low Mid.    ===
German Dutch   
gorilla human chimp bonobo
Platt- Dutch,
Frisian English
deutsch Flemish

Figure 4.4.1

537
 LAW REN CE S . MO S S

Another interesting point for us is that, at least so far, the work in this area is not
universally accepted:

Over the last 10 or more years, there has been a tremendous increase in the
use of computational techniques (many of which come directly from biology)
for estimating evolutionary histories of languages. Despite the enthusiasm with
which the popular press has received these studies, it is probably fair to say that
much of the community in historical linguistics has been skeptical of the claims
made by these studies, and perhaps even dubious about the potential for such
approaches to be of use.

This is from an overview tutorial of current work on this topic; see Nichols and War-
now (2008). The question, then, is what all the work comes to. It is possible to use
phylogenetic reconstruction to address questions of historical linguistics. For example,
Warnow et al. (1996) consider long-standing questions concerning Indo-European lan-
guages, such as whether the Italic and Celtic branches should be siblings. But it is hard
to make a definitive assessment of the meaning of the results. Overall, this example of
mathematical methods in the study of language underscores both the usefulness of those
methods, and also some of their troublesome aspects.

2.2 Infinity and Recursion in Language

Our most extended discussion centers on the relation of language to infinity. It is often
claimed that every human language is in a precise sense infinite, and that this point is
important. At the same time, the matter is not without considerable controversy. The
touchstone of the controversy is not so much infinity per se but its cousin, recursion.
Writing in a popular journal Science, Hauser et al. (2002) state:

(10) All approaches agree that a core property of FLN [the faculty of language
in the narrow sense] is recursion . . . . FLN takes a finite set of elements
and yields a potentially infinite array of discrete expressions. This capac-
ity of FLN yields discrete infinity (a property that also characterizes the
natural numbers) . . . . It has been recognized for thousands of years that
language is, fundamentally, a system of sound-meaning connections; the
potential infiniteness of this system has been explicitly recognized by
Galileo, Descartes, and the 17th-century “philosophical grammarians”
and their successors, notably von Humboldt.

A set is a definite collection of objects. These are usually mathematical objects, but it
is crucial in this discussion to consider also sets consisting of words, sentences, and the
like. As a mathematical preliminary, the set N of natural numbers is the set

{0, 1, 2, . . ., n, . . .}

A set S is finite if there is some natural number n such that the elements of S may be
matched with the set {1, 2, . . ., n} in a manner which is one-to-one (no repeats in the
matching) and onto (every number in {1, 2, . . ., n} is matched with some s S. A set is
infinite if it is not finite.

538
 TH E RO LE O F M ATH EM AT I C A L ME T H O D S

For example, the set {, o} is finite, and the set E of even numbers is infinite. Also,
it is a standard fact of set theory that N is not a finite set. (This is not entirely obvi-
ous, since our definition of finite was made independently.) Further, no set which has a
matching with N is finite, either.
The main claim under discussion here is

(11) The set of sentences in every natural language is an infinite set.

We shall say later why one would want to make this claim (11) and also how one argues
for or against it. But before we turn to those points, it should be noted that the claim that
natural languages are infinite is not to be found in other disciplines besides modern lin-
guistics. It seems to be a by-product of asking mathematical questions about language.
The main reasons for (11) have to do with what are sometimes called recursive con-
structions in language. Among these are boolean compounding, intensification of adjec-
tives using very, iterations of possessives, and iteration of propositional attitudes. To be
specific, here is one concrete argument for (11) in the special case of English.
Let

GP = {my greatn grandparents : n  N}.

In a little more detail, (great)n denotes the result of writing down the word great n times
in a row. Thus great1 = great, great2 = great great, etc. So the first item in GP is my grand-
parents, since with n = 0 we simply drop the word great. The second item in GP is my
great grandparents, etc.
We now have a matching between N and GP:

(12) N GP
0 my grandparents
1 my great grandparents
2 my great great grandparents
... ...
n my (great)n grandparents
... ...

The reason that (12) describes a matching is that if j and k are different numbers, then
my (great)j grandparents and my (great)k grandparents have different numbers of words
and hence are different strings of words. It is clear that each element of GP is a noun
phrase (NP) of English. Thus the collection of English NPs is infinite. Similarly, we can
find a collection of English sentences which is infinite: just take GP and append died at
the end.
However, the subtext for claiming (11) is that (11) provides an argument for the gen-
erative grammars. A generative grammar is a kind of mathematical object which embod-
ies the idea of building linguistic structures from the “bottom up” rather than the “top
down.” That is, by a generative grammar one means one of a large number of alternative
definitions; what exactly counts as a generative grammar is, I think, up to debate, and I
am very sure that there are going to be many borderline cases. What the various flavors
of generative grammars have in common is that they employ or reflect a metaphor of

539
 LAW REN CE S . MO S S

building structure starting with basic, unanalyzed forms, and following this they expand
on these by using rules which build complex forms from simpler ones.
An interesting point here is that made in Tiede and Stout (2010):

Hauser et al. (2002) claim that a core property of the human language faculty is
recursion and that this property “yields discrete infinity” of natural languages.
On the other hand, recursion is often motivated by the observation that there
are infinitely many sentences that should be generated by a finite number of
rules. It should be obvious that one cannot pursue both arguments simultane-
ously, on pain of circularity.

The message of the quote is that it is not always clear whether one is using infinity to
justify recursion or recursion to justify infinity. Mathematically, the latter move is legiti-
mate while the former is not. And one reading of arguments in support of recursion in
language is that they are disguised forms of arguments for the generative approach to
grammar. For this reason, writers such as Pullum and Scholz (2010) who are ill-disposed
toward generative grammars have cast a suspicious eye on the argument for infinity
based on recursion, and even on (11) as it applies to English. However, we maintain
that the argument above for (11) is valid. The only assumption that should be debated
is whether all rows of the chart in (12) are indeed NPs of English. (Of course, this is
tantamount to the infinitude claim concerning English in the first place.)
Now what are the arguments against (11)? One claim is that the collection of sentences
in a natural language like English is not a set at all. That is, it is not a “definite collection
of objects.” Whether a given string S of words is a sentence might not be a clear-cut mat-
ter. One person might hold S to be a sentence, and another to deny this. Even worse, the
same person might hold differing opinions at differing times (or even at the same time).
But if a language is not a definite collection, is there any way to think about it mathe-
matically? There are indeed many such ways, and one would be as a probabilistic language.
This would be a function ᐉ from strings on some alphabet to the unit interval [0, 1]. In
the case at hand, the “alphabet” would be the words of English (not the letters). For
example, one might interpret ᐉ(S) as the probability that a randomly selected speaker of
English assents to the assertion that S is a sentence of the language.
Returning to (11), one could object to our argument on the grounds that for some
large n, my (great)n grandparents is not a noun phrase of English: perhaps when n is larger
than the number of particles in the universe, for example. So the matter of infinity and
recursion then hinges on a deeper matter, whether we model English as an idealized col-
lection in the first place. For more on this point, see Tiede and Stout (2010).
But what about other languages? Up until now, all of our work has been on the special
case of the claim in (11) that the set of sentences of English is infinite. What about
other languages? What about the stronger claim that (11) makes? For the most part,
one senses that there is a consensus that the argument should apply to all languages.
However, there is an argument that this is false. The most recent version of this comes
from work of Daniel L. Everett on the language Pirahã, spoken in northwestern Brazil.
In papers and also in publications such as Everett (2007), he claims that there is no
recursion in the syntax of Pirahã:

So in the case of Pirahã, the language I’ve worked with the longest of the 24
languages I’ve worked with in the Amazon, for about 30 years, Pirahã doesn’t

540
 TH E RO LE O F M ATH EM AT I C A L ME T H O D S

have expressions like “John’s brother’s house”. You can say “John’s house”, you
can say “John’s brother”, but if you want to say “John’s brother’s house”, you
have to say “John has a brother. This brother has a house”. They have to say it
in separate sentences.

Indeed, this casts doubt on (11) in the form that we have stated it, and hence on the
quote in (10). That is, if recursion is not a feature of all human languages, it is then not
a universal of language the way (10) claims, and so what is the point of making it the
centerpiece of grammar?
Our purpose here has been to discuss the matter both for its own sake and also as an
illustration of the main points in the Chapter, mathematical methods in the study of
language. Our sources for this discussion are Pullum and Scholz (2010), Tiede and Stout
(2010), and Keenan and Moss (n.d.).

2.3 Refinement: Are Natural Languages Regular?

A further refinement of the question concerning the infinity of language concerns the
place of a natural language in the Chomsky hierarchy. To state the issue, we need some
definitions from mathematical linguistics. We need the definition of regular languages, and
we also need to apply the Pumping Lemma for them. One can find these topics in any
book on formal language theory or mathematical linguistics. But even without these, we
can discuss the matter informally. From a mathematical point of view, not all infinite
sets are created equal. There are many ways to define scales which measure complexity,
for example. At the very bottom end of one such scale, we have what are called the
regular languages. A regular expression on the words of English is an expression such as

my great+ grandparents
the(boy + girl) swam
(the (friend + enemy) of)+ the President died

Each of these has a denotation, a set of sequences of words of English. The denotation of
the first is listed in (12). The denotation of the second is the set with just two elements,
the sequences the boy swam and the girl swam. In effect, the + means “make a choice.”
The + means “repeat one or more of times.” So the denotation of the third set is an
infinite set, including the following:

the friend of the President died

the enemy of the President died
the friend of the enemy of the President died
the friend of the enemy of the enemy of the President died

A regular language is a set of sequences of words which happens to be the denotation of

a regular expression on the words of English. We want to present an argument that Eng-
lish is not a regular language. That is the set of English sentences cannot be the denota-
tion of any regular expression, no matter how long, complicated, or convoluted.
Here is the proof: suppose toward a contradiction that English were regular. It is a
standard fact that the intersection of two regular languages is regular. This is not at all
obvious from our definition. But granting this, consider the language L defined by

541
 LAW REN CE S . MO S S

L = English {the, boy, girl, saw, ran}+

From the point about intersections of languages, this language L must also be regular.
Here are some sentences in L:

the girl ran

the girl the girl saw ran
the girl the girl the girl saw saw ran
the girl the girl the girl the girl saw saw saw ran

We know that nobody can possibly understand this last sentence, but there are reasons
why linguists want to take it to be part of ideal English. These are the same reasons that
one would want to take all of the expressions in (12) to be English NPs.
The main points about L are that

(13) No sentence in L has two “the”s in a row.

(14) No sentence in L has two nouns in a row.
(15) Every noun phrase in every sentence in L has exactly as many nouns as
verbs.

The first assertion is a general fact about English. The other two are justified by examin-
ing L. If we assume that ran cannot take a direct object, then the sentences above are a
full listing of L. If we use ran transitively, then we get sentences like the girl saw the girl
the girl saw, with the girl the girl saw as an object noun phrase. But these still conform to
the last point. Consider a long sentence Sn in English of the form

(the girl)n+1 sawn ran,

with n to be determined shortly. Our sentence Sn belongs to L, and Sn must have one
more “girl” than “see”.
At this point, we wish to use a famous result about regular languages called the Pump-
ing Lemma. It would take us too far afield to state it formally, but the gist of it is that
for any given regular language, there is a number n so that if we have a sequence that
belongs to the language whose length is at least n, then there must be some repeatable
piece of the first n words. Informally, any regular expression whose denotation is the
given language would have to involve an expression with the superscript +, and then
we get some repetitive structure. More formally, the lemma tells us that we may (a) find
some number n; (b) to decompose the sentence Sn with this n as xwy for some non-empty
subword w of S included in the first n words of Sn; (c) “pump” Sn = xwy by repeating w,
to get xw2y; and (d) do all this in such a way that xw2y is also in the set L. The Pumping
Lemma does it all. Getting back to our sentence Sn and the decomposition xwy, let us
ask what w can possibly be. It must be (the girl)k for some k, or else (the girl)k the. But this
string xw2y would not belong to L: its subject noun phrase would have more nouns than
verbs. Since xw2y is not in L, we have a contradiction. This means that English cannot
possibly be regular.
For more on this topic and related ones, see the collection Savitch et al. (1987).
As we end let us reflect on the interaction of mathematics and linguistics in it, as a
reprise of our original points in Section 1.

542
 TH E RO LE O F M ATH EM AT I C A L ME T H O D S

The argument that we have seen can be a motivation to look at an important gram-
matical construction, center-embedded relative clauses. This hints at the contribution of
mathematical methods in syntax, a point which seems to me to be clear. There are also
many metaphors that come from mathematics and computer science to the study of syn-
tax. The notion of structure is especially interesting and problematic in this discussion.
In the first place, the general feeling is that the formalism of regular languages does not
get at the kind of structure that linguists traditionally favor. So the argument we have
seen may be used to motivate looking beyond the regular languages. This was the origi-
nal purpose in the argument. In contrast, opinions are somewhat mixed about whether
natural language “should” be regular in the first place. My sense is that the majority
feel that constructions like center embedding (and the argument we have seen) indeed
show that English is not regular. For other languages, there are even arguments to the
effect that they should not be context-free. A few people reject the argument and feel
that English is either outright regular, or “fairly close.” Turning to our last general point,
although the regular languages are not so much used in contemporary linguistics, they
form a very useful tool in computational linguistics. All in all, the points which we made
in this section are typical points about the useful and sometimes problematic relation of
mathematics and language.

Acknowledgments
My thanks to the editors of this volume for their patience. I am also grateful to Robert
Rose and Hans-Jörg Tiede for comments and discussion. All the remaining deficiencies
are entirely my own responsibility.

References
Benz, A., Jäger, G. and van Rooij, R., eds. (2006), Game Theory and Pragmatics, Palgrave Macmillan.
Everett, D. L. (2007), Recursion and human thought: why the Pirahã don’t have numbers. See http://www.
edge.org/3rd_culture/everett07/everett07_index.html
Hauser, M. D., Chomsky, N. and Fitch, W. T. (2002), “The faculty of language: What is it, who has it, and
how did it evolve?,” Science 298(5598), 1569–1579.
Keenan, E. L. and Moss, L. S. (n.d.), Mathematical Structure in Language. Unpublished ms., UCLA.
Nichols, J. and Warnow, T. (2008), “Tutorial on computational linguistic phylogeny”, Language and Linguis-
tics Compass 2(5), 760–820.
Port, R. and van Gelder, T. (1995), Mind as Motion: Explorations in the Dynamics of Cognition, Bradford
Books/MIT Press.
Pullum, G. K. and Scholz, B. C. (2010), Recursion and the infinitude claim, in H. van der Hulst, ed., Recur-
sion in Human Language, Vol. 104, Mouton de Gruyter, Berlin, pp. 113–138.
Savitch, W. J., Bach, E., Marsh, W. and Safran-Naveh, G., eds. (1987), The Formal Complexity of Natural
Langugage, D. Reidel, Dordrecht.
Tiede, H.-J. and Stout, L. N. (2010), Recursion, infinity, and modeling, in H. van der Hulst, ed., Recursion
in Human Language, Vol. 104, Mouton de Gruyter, Berlin, pp. 1–12.
Warnow, T., Ringe, D. and Taylor, A. (1996), Reconstructing the evolutionary history of natural languages,
in Proceedings of ACM–SIAM Symposium on Discrete Algorithms (SODA), Association for Computing
Machinery and the Society of Industrial and Applied Mathematics, pp. 314–322.

543
 M A RTI N STO K H O F

4.5
T HE RO LE OF
AR T IFIC IAL
L ANGUAGES
Martin Stokhof

1 Introduction
When one looks into the role of artificial languages in philosophy of language it seems
appropriate to start with making a distinction between philosophy of language proper
and formal semantics of natural language. Although the distinction between the two
disciplines may not always be easy to make since there arguably exist substantial histori-
cal and systematic relationships between the two, it nevertheless pays to keep the two
apart, at least initially, since the motivation commonly given for the use of artificial
languages in philosophy of language is often rather different from the one that drives
the use of such languages in semantics.
Of course, this difference in motivation should not blind us for the commonalities
that exists between the two disciplines. Philosophy of language and formal semantics
have a common history, and arguably also share some of their substance. Philosophy of
language is by and large an outgrowth of work in the analytical tradition in philosophy
in the first half of the twentieth century. Both ordinary language philosophy, with its
emphasis on the description of actual language use, as well as the more logic-oriented
and formally inclined school of logical positivism contributed to the definition of phi-
losophy of language as a separate philosophical discipline, with its own set of problems
and methods to solve them. Another major contributor to the establishment of phi-
losophy of language as a distinct discipline has been modern linguistics, in particular
generative grammar in the tradition of Chomsky that became the dominant paradigm
in linguistics in the fifties and sixties of the previous century. And as it happens, both
the generative tradition of Chomsky and analytic philosophy in its formal and less for-
mal guises have been important factors in the development of formal semantics as well.
Thus, it should come as no surprise that the two have something in common. That the
communalities go beyond a common ancestry, but are reflected in substance and meth-
ods as well, will be argued later on.
However, be that as it may, it is still a good idea to keep philosophy of language and
formal semantics separate, at least initially, since the role that is assigned to artificial
languages and the ways in which these languages are employed in both does differ in a
number of respects that are worth keeping in mind. Accordingly, we will begin with a
 TH E RO LE O F A RTI F I C I A L L A N G U A G E S

brief characterization of the way artificial languages play a role in philosophical inves-
tigations (section 2). Then we will look in some detail at their use in formal semantics,
focusing on the philosophical presuppositions of the way in which they are employed
there (section 3). The differences and resemblances are then the subject of section 4.

2 Philosophy
In philosophical research on language we mostly find artificial languages being used
as notational aides, in much the same way as they are used in, for example, analytic
metaphysics, in traditional epistemology, or in philosophy of science. The main goal
of this use of artificial languages, such as those of first-order logic, modal logic, and the
like, is to clarify a particular argument, or to precisely formulate a conjecture or a thesis.
We could call this use of artificial languages that of “easing the conversation,” as the
purpose of it is to make certain elements in an argument or thesis more readily acces-
sible and thereby more easy to evaluate critically (or more convincing as the case may
be). Getting the relative scope of quantificational expressions (‘everyone’, ‘most of us’,
‘some people’,. . . ‘always’, ‘sometimes’, . . .) and modal operators (‘necessarily’, ‘obliged’,
‘possibly’, ‘allowed’ . . .), or the exact nesting of Boolean connectives and operators
(‘and’, ‘or’, ‘if . . . then . . .’, . . ., ‘not’, . . .), right is often crucial for the understanding,
and subsequent evaluation, of an argument, or for the grasp of the precise content of a
particular thesis. Of course, in many cases the required disambiguation or explication
can be provided without taking recourse to the use of formal notation, simply by explic-
itly stating, for example, the intended relative scope of two quantificational expressions.
But such paraphrases may become somewhat cumbersome and complex, and anyway,
the use of notation, apart from being helpful, also has a certain aesthetic appeal. A
classical case where the use of notation helps to, if not solve, then at least make clear
a particular problem is Aristotle’s treatment of temporal modalities in the famous sea
battle discussion in Peri Hermeneias, which has created a large number of diverging
interpretations that in part revolve around the question what are the exact scopal prop-
erties that should be assigned to the relevant terms in Aristotle’s text. Concretely, the
question is whether Aristotle holds that the truth of a proposition about a future event
depends on causal determinism (in which case we should construct Aristotle’s premise
as ‘It is necessary that p or it is necessary that not-p’), or whether he merely subscribes
in the relevant passages to the law of excluded middle (in which case the premise reads
as ‘It is necessary that (p or not-p)’.
Although this is certainly a proper use of artificial languages, we should also remark
that it is not the most deep use. Be that as it may, what is more troubling is that in many
cases authors that use a piece of notation forego the trouble of specifying what it means.
And this actually may hamper understanding rather than facilitating getting a certain
point across. In many cases authors rely on an implicit understanding on part of their
readers of the notation they are using. Often, this may be excused: for instance, when
one uses the quantifier-notation of first-order logic, it would be somewhat pedantic to
explicitly state interpretation clauses for the quantifiers, given that there is an accepted,
standard way of interpreting them. However, if one uses the box-operator from modal
logic, for example, in a partial formalization of the premise or the conclusion of an argu-
ment, such a use really makes sense only if one also indicates which interpretation one
assigns to it. With such a wide variety of modal systems around, one really can not rely
on a common understanding on the part of all of one’s readers.

545
 M A RTI N STO K H O F

This use of notational devices borrowed from various kinds of formal languages is
quite akin to the way in which the use of mathematical notation has become part of
our general conversational repertoire. We all employ Arabic numerals and simple arith-
metical notation all the time, as if they were part of the natural language we use. This
requires no knowledge or expertise beyond a basic level of practical ability. Certainly no
acquaintance with anything as sophisticated as, for example, the Peano-axioms, or the
Frege–Russell definitions of the natural numbers, is assumed here. This use of arithmetic
is really all about easing the conversation, i.e., about making certain practical dealings
with quantities and their basic properties more manageable.
Such use of notation borrowed from formal languages can very well be regarded as
amounting to an extension of the natural language in which it is incorporated. Cf.,
the characterization that Wittgenstein gives in Philosophical Investigations, section 18
(Wittgenstein 2009):

[. . .] ask yourself whether our language is complete;—whether it was so before the

symbolism of chemistry and the notation of the infinitesimal calculus were incor-
porated in it; for these are, so to speak, suburbs of our language. (And how many
houses or streets does it take before a town begins to be a town?) Our language
can be seen as an ancient city: a maze of little streets and squares, of old and new
houses, and of houses with additions from various periods; and this surrounded by
a multitude of new boroughs with straight regular streets and uniform houses.

For a more restricted set of users, logical notation, such as is borrowed from propositional
or predicate logic, functions in much the same way. They use the notation to facilitate
argumentation and debate, but not much hinges (in general, at least) on their having
specific expertise in the logical systems that these languages are part of.
This is different in the context of the natural sciences, where the use of mathemat-
ics goes substantially beyond a ‘mere’ notational use. Here it is the formal systems as
such, and not just the formal languages that provide the notational tools, that are being
used. For example, the use of Riemann geometry in relativity theory is not a expedient
notation but incorporates substantial claims about the nature of space and time. Hence,
this formal system can be used to formulate explanations of phenomena and results
that have been observed, and to deduce predictions that can be checked by further
observation and experiment. The formal system here is more than a tool; it is part of
the substance of the theory in question. Accordingly, here the use of a particular formal
system is not just motivated by ease of use or similar concerns, it is subject to empirical
validation or falsification, at least in principle. Inasmuch as the formal system is part of
the theory, any experimental result or observed phenomenon that verifies or falsifies the
theory reflects on the formal system by implication.
This seems to mark an important difference between the use of formal languages and
systems in philosophy and in the sciences. That does not mean that in philosophy it is
only the conversational and clarificatory use that we may encounter. Quite another use,
less frequent but arguably more substantial, that is made of artificial languages in the
context of philosophical discussions about language is when it is the properties of artifi-
cial languages themselves that are used as premises in some argumentation about natural
language. Obviously, such a use is not of artificial languages as such (as is the case in the
conversational use noted above) but of systems, i.e., of languages with an interpretation
(which can be model-theoretic or proof-theoretic). An interesting example of such an

546
 TH E RO LE O F A RTI F I C I A L L A N G U A G E S

application is Putnam’s appeal to the Löwenheim–Skolem theorem for first-order logic,

as part of an argument against the possibility of providing natural languages with a ref-
erential semantics. (Cf., Putnam 1983; for a more concrete version of the argument, cf.
Putnam 1981.) The theorem states that every countable first-order theory that has an
infinite model has a model of size k for every infinite cardinal k. This means that such
theories are not able to fix their models up to isomorphism. Putnam uses this result to
argue that meaning does not fix reference, since given a set of true sentences in a given
language we can always, by permutation of the objects in the domain, change the refer-
ence of subsentential terms without this affecting the truth values of the sentences in
question. This shows, according to Putnam, that meaning as such does not fix reference.
It is important to note that an argument like this works on the basis of an assumption:
that natural languages are sufficiently like formal languages in the relevant respects so as
to allow a transfer of properties and results from formal languages (or systems) to natural
languages. In this case, the assumption must be that any natural language ‘contains’
something like a first-order language and its associated model theoretic semantics. In
many respects, this use of formal languages in philosophy of language is akin to the use
that is made of, for example, epistemic logic, or probability theory, in formal epistemol-
ogy. On the assumption that the phenomenon under consideration (natural language
meaning, or epistemic reasoning) has a hidden structure that is sufficiently similar to
what is explicit in some formal system (first-order logic, or epistemic logic), one projects
results obtained with regard to the latter (such as (un)decidability) onto the former.
One of the problems in this type of use of formal languages is that the background
assumptions (such as: ‘natural languages contain first-order logic’) are hardly ever dis-
cussed explicitly, let alone justified. But if we really use formal languages in this way,
i.e., as formal systems that are part of a theory and that are instrumental in the deduc-
tion of certain conclusions, then we rely, not only on the applicability of the system as
such but also on a prior decision to use this system rather than some other one. That
priori decision of course calls for independent motivation. In the natural sciences such
independent motivation comes from explanatory and/or predictive success with regard
to independently acquired experimental and observational results. However, also in this
type of philosophical application that is exactly what seems to be lacking.
Yet another use of artificial languages is when a particular language, or rather system, is
appealed to as an arbitrator in order to decide certain issues that are in and of themselves
not really linguistic in nature. For example, classical first-order logic is often held to encom-
pass an appropriately parsimonious ontology, and thus is used as a standard to determine
whether or not some entities that are assumed by some philosophical argument or theory
actually are ‘proper’ entities. Quine’s work on ontological commitment provides a classic
example. This is a restrictive use, one that intends to single out a limited range of bona fide
entities. On the other end of the spectrum we can locate positions that claim that whatever
can be analyzed in some sufficiently general mathematical system, for example Zermelo–
Fraenkel set theory (with or without the axiom of choice), or category theory, is something
that one might justifiably appeal to. In both kinds of use, however different, it is the formal
system that is used as a standard for what can, or cannot, be done in and through natural
language. However, as in the previous kind of use, this type of argument really depends on
the availability of external justification for the choice of the particular formal system that
one employs as a standard. And, in such a case, such a justification seems to be lacking.
However, one might think it is formal semantics, as the empirical study of natu-
ral language meaning that uses similar formal languages—namely, those of logic and

547
 M A RTI N STO K H O F

mathematics—as its main tools, that actually provides such a justification. Therefore, it
is now time to look at the use of artificial languages in that realm.

3 Semantics
The use of artificial languages in the study of the meaning of natural languages started
taking off in a systematic manner only at the end of the 1960s. At the time, the rise of
generative grammar in the Chomsky tradition had done much to expel philosophical
doubts as to whether natural languages are systematic enough to allow the application
of formal tools, such as those from model theoretic semantics, to them. Tarski had stated
in his famous 1944 paper ‘The Semantic Conception of Truth’ that:

The problem of the definition of truth obtains a precise meaning and can be
solved in a rigorous way only for those languages whose structure has been
exactly specified. For other languages—thus, for all natural, ‘spoken’ lan-
guages—the meaning of the problem is more or less vague, and its solution
can have only an approximate character. Roughly speaking, the approximation
consists in replacing a natural language (or a portion of it in which we are inter-
ested) by one whose structure is exactly specified, and which diverges from the
given language ‘as little as possible.’

For some this meant that in order to become susceptible to any kind of formal treat-
ment natural languages have to be reformed and regimented (beyond recognition, as
some others complained). And yet other philosophers took the same ‘observation’ as
evidence that any attempt to treat natural languages on a par with formal ones was
wrong-headed to begin with, and that one would do better to study natural language in
an informal, and much more descriptive manner. This approach, pioneered by Austin,
Warnock, Ryle, and others, commonly known as ‘ordinary language philosophy’, was
quite aptly also described as ‘linguistic phenomenology.’
Both views, however, start from the assumption that natural languages are indeed
not systematic enough to allow formal treatment, which is, of course, a complaint that
has been leveled against natural languages by philosophers for centuries. The work of
Chomsky in generative linguistics apparently inspired much more confidence in phi-
losophers and logicians to assert that perhaps natural languages weren’t as unsystematic
and misleading as their philosophical predecessors had made them out to be after all.
To be sure, there had already been exceptions, such as Hans Reichenbach, whose
Elements of Symbolic Logic (Reichenbach 1947) contains a large section devoted to the
application of logic in the description of natural language phenomena, parts of which (in
particular his treatment of the natural language tense system) became very influential
much later on. And the potential relevance of Chomsky’s work for philosophical seman-
tics was already noted in 1953 by Yehoshua Bar-Hillel who wrote a programmatic paper
in which he explores the possible connections. This attempt at cooperation met with a
negative reaction from Chomsky himself, who throughout his entire career has remained
hostile to the idea of some form of model theoretic semantics being of any relevance to
linguistics. So, it is rather despite this reaction from the leader of the generative move-
ment that at the end of the 1960s formal semantics began to flourish nonetheless.
A prominent example is Donald Davidson who claimed, in his seminal paper ‘Truth
and Meaning’ from 1967, that:

548
 TH E RO LE O F A RTI F I C I A L L A N G U A G E S

Philosophers of a logical bent have tended to start where the theory was and
work out towards the complications of natural language. Contemporary lin-
guists, with an aim that cannot easily be seen to be different, start with the
ordinary and work toward a general theory. If either party is successful, there
must be a meeting. Recent work by Chomsky and others is doing much to bring
the complexities of natural languages within the scope of serious theory.

This sentiment is echoed in many of the early papers that constituted formal semantics
as a discipline at the crossroads between linguistics, philosophy, and logic: generative
linguistics shows how to capture the syntax of natural languages in a systematic and
formal theory; in order to extend this to semantics, philosophy provides the necessary
conceptual apparatus, consisting of analyses of meaning, reference, and truth that in
many respects go back to the early days of analytic philosophy; and logic contributes
the formal tools with which these concepts can be applied in a systematic fashion to
natural language.
Of course, this cooperation between philosophers, linguists, and logicians took on
different forms, and not everyone applied formal languages in formal semantics in
exactly the same way. One of the most influential approaches turned out to be the one
pioneered by Richard Montague. In his 1970 paper ‘Universal Grammar’, which out-
lines the theoretical machinery behind what became known as ‘Montague grammar’,
he states:

There is in my opinion no important theoretical difference between natural

languages and the artificial languages of logicians; indeed, I consider it possible
to comprehend the syntax and semantics of both kinds of languages within a
single natural and mathematically precise theory. On this point I differ from a
number of philosophers, but agree, I believe, with Chomsky and his associates.

This is a strong statement. No doubt Montague was well aware of the numerous differ-
ences that exist between the formal languages of logic and mathematics and natural
languages, yet apparently he is also of the opinion that these are not ‘important theo-
retical differences’ that would constitute an obstacle to describing both with the same
mathematical means.
Montague’s statements suggest that formal languages and natural languages are on a
par, but that is not how formal semanticists actually construe the relationship between
the two. Their focus of interest is natural language meaning, and hence they use formal
languages as tools in their investigations. The most explicit form this takes is when an
interpreted formal language is used as a model for (some part of) natural language. The
idea behind this methodology is that we can describe and explain the semantic prop-
erties of natural language expressions by setting up a systematic relationship between
these expressions and those of some suitable formal language, and then use the seman-
tics of the latter to act as a proxy for that of the former. This procedure of ‘indirect
interpretation’, as it is often called, relies on the assumption that the formal system that
is being used in this way somehow provides explanatory power with regard to the natural
language. This approach was made popular by Montague’s paper ‘The Proper Treatment
of Quantification in Ordinary English’ (Montague 1973), but quite similar ideas can be
found on early work of Lewis (1972), Cresswell (1973), and others.
This methodology has in essence remained unchanged from the early days of formal

549
 M A RTI N STO K H O F

semantics until pretty much the present-day. As then, formal semanticists model a
particular phenomenon—for example, anaphora, or aspect, or vagueness—by either
taking an existing formal system that has been developed independently and, in most
cases, for different purposes, or by defining one themselves, and then use it to model
the natural language phenomenon by providing a more or less strict and systematic
translation of the relevant part of natural language into the language of the formal
system. In the old days, the tendency was to stick to one particular formal system, that,
of course, then had to have all the expressive power one could imagine one would
need at some point. Montague used a system of intensional higher-order type theory,
Cresswell went all the way and just used ZFC set theory; Davidson chose to restrict
himself, for independent philosophical reasons, to the use of standard first-order logic.
In particular in linguistic circles, Montague’s choice initially won the day, but over the
years subsequent developments have seen a plethora of formal systems being used by
formal semanticists, unfortunately more often than not without much attention being
paid to the overall compatibility of these systems. Domain theory, property theories,
belief revision systems, event calculus, different many-valued logics, various nonmo-
notonic logics, dynamic logic, various forms of game theory, second-order type theory,
Martin-Löf’s type theory, untyped lambda-calculus, Boolean algebras, lattices of vari-
ous kinds, set theory with or without ur-elements—basically everything in the book
has been thrown at natural language phenomena at some point. And then there are
the ‘custom built’ systems, such as various systems of discourse representation theory,
or situation theory.
What is interesting about the use of this wide variety of formal systems is how the
choices are motivated. In many cases, especially those where the semanticists use exist-
ing frameworks, they aren’t motivated, at least not explicitly. In such cases one rather
gets the impression that the formal system that is being employed isn’t so much chosen
for any particular properties it may have, but simply because ‘it gets the job done’. For
that to be the case it must of course have enough, and the right, expressive power, but
in most cases this is, as it were, settled on incidental, almost ad hoc grounds, not on the
basis of a prior investigation of the properties of the system in question. In some cases,
though, the latter do play a role and may actually turn up in the motivation for the use
of a particular system. Relevant considerations here usually center around properties
such as (un)decidability, (in)completeness with respect to some proof system, expres-
sive power (for example, first-order versus higher-order quantification), and similar
concerns. These are, of course, properties of formal systems as such, not necessarily of
the natural languages that are studied with the aid of them. However, the questions and
concerns that are raised almost always relate to the natural languages, not to the formal
systems. Thus, a decision to choose a formal system can be found to be motivated by
concerns about the learnability of the natural language, or about its effective, practical
applicability in, say, common sense reasoning. The assumption behind this is, then, that
the relevant properties of the formal system that is used in the analysis of the formal
language can, in some way, be translated back to the natural language, and in that way
address the kind of concerns just mentioned.
Thus, artificial languages and formal systems are being used in the analysis of the
semantics of natural languages in different ways and with different motivations. How-
ever, what appears to be a common assumption is that the formal system is regarded as a
model of the natural language. In actual analyses and descriptions it is, of course, never
considered to be a model of an entire natural language, since it is always only certain

550
 TH E RO LE O F A RTI F I C I A L L A N G U A G E S

aspects of a language (some class of expressions, a particular type of construction) that

are under discussion. But in each case what is being analyzed and described is taken
to be modeled by the formal system in the sense that all the relevant properties of the
natural language are assumed to be adequately represented by properties of the formal
system. For it is only on such an assumption that it makes sense to think of the formal
system as a representation of the relevant semantic properties of the natural language.
This is a widely held view, most of the time simply assumed, sometimes explicitly
stated. Textbooks are often the best source of statements of such basic points of view.
The following is a quotation from a introduction to natural language semantics by Hen-
riëtte de Swart (De Swart 1998):

Given that direct interpretation of natural language with respect to the outside
world (or some model of it) is not always easy, many semanticists opt for the
indirect approach. We know that a translation can sometimes help us to deter-
mine the meaning of an expression. Suppose I speak French, but you don’t, and
we both speak English. In that case, I can teach you something about the mean-
ing of a French expression by translating it into English [. . .] The same ‘trick’
can be used with the translation of natural language into some formal language.
Suppose I can describe the meaning of an English expression by translating it
into an expression of a formal language. Because there will be a full and explicit
interpretation procedure for the expressions of this formal language, I will imme-
diately have grasped the meaning of the English expression. Of course, I will
only have access to it in an indirect way, namely via a translation procedure, but
as long as the translation is perfect, the exact meaning will be captured.

This clearly illustrates that the idea of formal languages as models of natural languages
is built right into the core methodology of formal semantics: namely, that of studying
natural language by providing translations of relevant fragments into the language of
some formal system. The same idea can be found in other textbooks (Cf., for example,
Chierchia & McConnell-Ginet (2000):

Is it possible to regard logical form [. . .] as providing us with a theory of seman-

tic interpretation, with a theory that characterizes what we grasp in processing
a sentence? [. . .] We think it is possible, as our logical forms do meet the main
requirements that semantic representations are generally expected to meet.

However, it relies on two assumptions with regard to its object of study and the way in
which this can be accessed. It is clear that using formal systems to study natural language
meaning in this way, i.e., by devising translations between the two, works only if we can
assume that both the meanings of the expressions of the formal language as well as those
of the natural language are determinate and available prior to the analysis being carried
out. They are determinate in the sense that they are able to guide the translation, in the
sense of providing the necessary criteria for determining when the translation is actu-
ally correct. And they are available in the sense that as semanticists we have access to
them independent of and prior to the use to which we put them. (For a more systematic
analysis of the determinacy and availability assumptions, including their role in concerns
with language reform in the work of Frege, Russell, early Wittgenstein, and others, see
Stokhof 2007.)

551
 M A RTI N STO K H O F

Notice that we need these determinacy and availability assumptions with regard to
both the meanings of the natural language expressions to be analyzed as well the mean-
ings of the formal language that are used. Both are needed for otherwise there would be
no way to judge the correctness of the translation, and this is essential if the methodol-
ogy of indirect interpretation via translation is to succeed. This means that prior to the
specification of the translation both are assumed to be given: we need to know what the
meanings of the natural language expressions are and what meanings are assigned to the
formal language expressions before we can start defining the translation or judge the
correctness of any attempts. Consequently, what such an indirect specification of mean-
ing does, at best, is to represent them in another way than the original natural language
expressions themselves do. This may very well lead to a more perspicuous representa-
tion, or one that has other, technical advantages, but the one thing this methodology
will not do is to actually provide meanings for natural language expressions, nor will
it allow us to discover their formal properties, because prior to the analysis these are
assumed to be determined and to be already available.
Where does this leave us with regard to how and why formal languages are used in
semantics? For that we need to go back to philosophy.

4 Semantics and philosophy

The way in which interpreted formal languages are used in formal semantics to model
relevant aspects of the semantics of natural language has much in common with the
‘Putnam’-type applications in philosophy of language. Properties of the formal system
are used to explicate and explain certain features of natural language meaning, which
is made viable to begin with by the assumption of sufficient similarity between the two
domains. As we noted above, this calls for independent motivation and justification,
which philosophy appears not quite able to provide. In the sciences, this is different:
there, data from experiments and observation can be used to test whether the assumed
similarity between the formal system and the empirical domain actually obtains. This
works, first of all, because the data used for testing can be obtained independently (at
least in principle), and, secondly, because the formal system is integrated in the theory
about the empirical domain.
The question now is whether in formal semantics independent justification of these
assumptions can be procured as well. If so, then formal semantics is like the sciences and
it may even provide the needed empirical justification for some of the philosophical
analyses that make use of a similar methodology. If not, then formal semantics seems to
be in the same boat as philosophy.
That the very methodology of formal semantics rests on the assumptions of deter-
minacy and availability is ample reason to think that the required independent and
prior justification is lacking here: the method of modeling natural language meaning by
means of translation into expressions of an interpreted formal language assumes rather
than justifies the existence of the required similarities between the former and the lat-
ter. So by itself this use of formal systems does not show that it is adequate, nor do the
predictions and explanations that are based on it.
This point is reinforced by the observation that, unlike in the sciences, where the
formal system that is being employed is built right into the heart of the theory, in formal
semantics no such intrinsic link seems to exist. In view of the wide variety of formal
systems being used, and taking into account that the choice of formal systems does not

552
 TH E RO LE O F A RTI F I C I A L L A N G U A G E S

appear to be limited by any substantial empirical argument, we must conclude that these
systems play a very different role.
What might that role be? It seems we can make the following fundamental distinc-
tion between two ways of looking at what we do when we apply artificial languages in
the study of natural language. First of all, we can look at a formal language as a model of
a natural language. This is the traditional, and still dominant, view on the matter that
we have outlined above, and which encounters the problems of justification that we
also discussed. Second, we may consider a formal language as a tool—one of the many
tools, we should add—in the study of natural language meaning. We can use a particular
formal language to display certain inference patterns involving generalized quantifiers,
say, and use another one to deal with, for example, the division between lexical and
world knowledge. We can employ the compositionality with which we endow our for-
mal languages to test whether a certain fragment of a natural language allows for a simi-
larly compositional description. (But note that much here depends on what we have
taken compositionality of natural languages to consist in to begin with, a question that
arguably is not straightforwardly empirical.) What is crucially different on this second
approach is that the adequacy criteria for our choice and employment of a particular
formal language have to come from elsewhere: They derive from the practical concerns
that we have, which by themselves are completely agnostic with respect to the tools
that we may use or need to use. (And these concerns are even agnostic as to whether
logical languages are the best tools. That, then, also becomes a matter that is up for
discussion: For some purposes stochastic tools may be arguably better.)
On this second approach we employ formal languages in the study of natural lan-
guages, not because ‘there is no important theoretical difference’ between the two but
because they are useful tools. A formal language is not a model of a natural one, but
rather a tool that can be used to provide a ‘perspicuous representation’ of some part
or aspect of it. In quite a similar vein one might look at the use of formal languages in
philosophical analysis of language: There, too, the primary aim that is served by their
employment is perspicuity and clarity. Of course, the content in both is different: For-
mal semantics deals with empirical phenomena, philosophical analysis with conceptual
considerations. But the use both make of formal languages as a means for perspicuous
representation is basically the same.

References
Chierchia, G. & McConnell-Ginet, S. (2000) Meaning and Grammar, Cambridge, MA: MIT Press
Cresswell, M. (1973) Logics and Languages, London: Methuen
Davidson, D. (1967) “Truth and Meaning,” Synthese 17
Lewis, D. (1972) “General Semantics,” Synthese 22:18–67
Montague, R. (1970) “Universal Grammar,” Theoria 36: 373–98
Montague, R. (1973) “The Proper Treatment of Quantification in Ordinary English,” in J. Hintikka, J.
Moravcsik, and P. Suppes (eds.) Approaches to Natural Language, Dordrecht: Reidel
Putnam, H. (1981) “A Problem About Reference” in Reason, Truth and History, Cambridge: Cambridge
University Press
Putnam, H. (1983) “Models and Reality,” in Realism and Reason, Cambridge: Cambridge University Press
Reichenbach, H. (1947) Elements of Symbolic Logic, New York: Macmillan
Stokhof, M. (2007) “Hand or Hammer? On Formal and Natural Languages in Semantics,” The Journal of
Indian Philosophy, 35(5–6): 597–626
Swart, H. de (1998) Introduction to Natural Language Semantics. Stanford: CSLI
Wittgenstein, L. (2009) Philosophische Untersuchungen / Philosophical Investigations, Oxford: Blackwell, 4th
revised edition

553
 M I CH A EL D E V I T T

4.6
T HE RO LE OF
I NTUITIONS
Michael Devitt

1 The Received View

How should we go about finding the truth about language? The received view is that
we should proceed by consulting our intuitive judgments about language, our “intui-
tions.” Indeed, it would be hard to exaggerate both the apparently dominant role
of such intuitions in the philosophy of language and the agreement among philoso-
phers that these intuitions should have this role. This emphasis on intuitions reflects,
of course, a widely held view about the methodology of “armchair philosophy” in
general.
Saul Kripke’s Naming and Necessity, one of the most influential works in the philoso-
phy of language, is often, and rightly, cited as an example of heavy reliance on intui-
tions. And Kripke is explicit that intuitions should have this important role:

Of course, some philosophers think that something’s having intuitive content

is very inconclusive evidence in favor of it. I think it is very heavy evidence in
favor of anything, myself. I really don’t know, in a way, what more conclusive
evidence one can have about anything, ultimately speaking.
(Kripke 1980: 42)

Stephen Neale gives another statement of the received view:

Our intuitive judgments about what A meant, said, and implied, and judgments
about whether what A said was true or false in specified situations constitute the
primary data for a theory of interpretation, the data it is the theory’s business to
explain.
(Neale 2004: 79)

These statements are quite general. Here are some typical claims about particular topics
in the theory of language:

Semantic Content: Herman Cappelan and Ernie Lepore attribute to a range of phi-
losophers versions of the following, which they regard as the “Mistaken Assump-
 TH E RO LE O F I NT U I T I O N S

tion”: “A theory of semantic content is adequate just in case it accounts for all
or most of the intuitions speakers have about speech act content, i.e., intuitions
about what speakers say, assert, claim, and state by uttering sentences” (Cappelan
and Lepore 2005: 53).

Reference: Stephen Stich thinks that the following view is “favored, albeit tac-
itly, by most philosophers”: “the theory of reference is attempting to capture the
details of a commonsense theory about the link between words and the world”
where that theory involves, at least, a generalization of referential intuitions
(Stich 1996: 6).

Truth Conditions: Jason Stanley and Zoltan Szabó endorse the view that “account-
ing for our ordinary judgements about the truth-conditions of various sentences is
the central aim of semantics” (Stanley and Szabó 2000: 240).

What Is Said: Robyn Carston thinks that the various criteria in the pragmatics
literature for placing “pragmatic meanings” into “what is said”, “in the end,
. . . all rest . . . on speaker/hearer intuitions” (Carston 2004: 10). François
Recanati claims that “‘what is said’ must be analysed in conformity to the
intuitions shared by those who fully understand the utterance” (Recanati
2004: 14).

Syntax: Noam Chomsky claims that ‘linguistics . . . is characterized by atten-

tion to certain kinds of evidence . . . largely, the judgments of native speakers’
(Chomsky 1986: 36). Liliane Haegeman, in a popular textbook, says that ‘all
the linguist has to go by . . . is the native speaker’s intuitions’ (Haegeman
1994: 8).

2 The Task?
It is clear from these claims, and other similar ones, that intuitive judgments are
commonly thought to provide the evidence or, at least, the main evidence for theories
of language. Yet, often, such claims seem to suggest a stronger view: the very task of
theories of language is to explain or systematize competent speakers’ intuitions about
language.
This view of the task is very puzzling. For, the obvious way to describe the task of
the theory of language is to explain the nature of language, to explain properties like
meaning, truth, reference, and grammaticality—real properties of linguistic expres-
sions playing some sort of explanatory role. If we start from this view, surely as good
a starting place as one could have, why take the task to be to capture ordinary intui-
tions about such properties, intuitions that must simply reflect folk theory? That
would seem to be appropriate only if we assume that the folk must be right about lan-
guage. But why assume that? We don’t suppose that the folk are authorities on phys-
ics, biology, or economics; why suppose that they are authorities on linguistics and
semantics?
So let us set aside the idea that the study of linguistic intuitions is the task of theories
of language and turn to the more plausible view that these intuitions are evidence for
those theories.

555
 M I CH A EL D E V I T T

3 Experimental Semantics
The evidential role of intuitions in semantics, indeed, the evidential role of intuitions
in philosophy generally, has recently grabbed the attention of a group known as “experi-
mental philosophers.” They have noted that the intuitions that play this evidential
role are usually not those of the folk but rather those of the philosophers themselves. In
response, they have conducted experiments to test whether the folk share these intui-
tions. Thus, in a well-known experiment, the intuitions that Kripke airs about reference
were tested against those of undergraduates in Rutgers and Hong Kong (Machery et
al. 2004). The results raised serious doubts in the experimenters’ minds about Kripke’s
refutation of the description theory of names. There is plenty of room for debate about
whether, and to what extent, the results should raise these doubts (Martí 2009, Devitt
2011a, Ichikawa et al. 2011). But one thing such experiments surely should do is focus
our attention on two important questions: Why is the evidential role given to intuitions
thought to be appropriate? Is it really appropriate? These questions will be the focus of
this chapter.

4 “Cartesianism”
It would, of course, be appropriate to give a person’s intuitions an important evidential
role if we could be confident that they reflected knowledge. And the received view is
that a competent speaker of a language does indeed have knowledge about her language,
propositional knowledge, “tacitly” at least, simply in virtue of being competent in the lan-
guage: “It is an undeniable feature of the notion of meaning . . . that meaning is transpar-
ent in the sense that, if someone attaches a meaning to each of two words, he must know
whether these meanings are the same” (Dummett 1978: 131). “The natural view is that
one has some kind of privileged semantic self-knowledge” (Loar 1987: 97). The idea of
this sort of privileged access—that we are in a special position to know about our own
competence—is an instance of general “Cartesianism”:

Since Descartes, it has seemed undeniable to most philosophers that each of us

has a privileged way of knowing about his or her own mental states. . . . when-
ever we have a thought, belief, intention, or desire, we can in principle come
to know what we think, believe, intend, or desire just by internal examination,
without engaging in an empirical investigation of the external world.
(McKinsey 1994: 308)

The idea that we have a Cartesian access to semantic facts seems to be an almost unques-
tioned part of the semantic traditions of Frege and Russell. Consider, for example, the
assumption that our competence consists in propositional knowledge of truth condi-
tions. Herbert Heidelberger (1980) has shown how widespread this assumption is with
references to Wiggins, Strawson, Davidson, Frege, Wittgenstein, Quine, and Carnap.
He points out that it seems to be regarded as “uncontroversial . . . harmless . . . perhaps
unworthy of serious discussion” (p. 402). Gareth Evans says, “perhaps no one will deny
it” (1982: 106).
It is worth noting that one of the most famous arguments in the philosophy of lan-
guage seems to rest on Cartesiansim. This is the Fregean argument that ‘Hesperus =
Hesperus’ and ‘Hesperus = Phosphorus’ must differ in meaning, hence ‘Hesperus’ and

556
 TH E RO LE O F I NT U I T I O N S

‘Phosphorus’ must differ in meaning, because they differ in cognitive significance or

informativeness. But why would the latter epistemic difference be taken to establish the
former semantic difference? The reason, I suggest (Devitt 1996: 172–3), is that it is taken
for granted that two sentences mean the same only if competent speakers know that
they do; see Dummett above. And speakers could know this only if sentences did not
differ in cognitive significance for them. For, if they differed in significance, speakers
would not think that they were synonymous.

5 A Priori Knowledge?
Why should we suppose that ordinary competent speakers have this knowledge of
semantic facts? Many seem to think that the knowledge is a priori. Thus Jerrold Katz
claims: “We know sense properties and relations of expressions on the basis [of] the
speaker’s a priori linguistic intuitions in clear cases” ( Katz 1997: 21). And Michael
McKinsey thinks that it is “fairly clear” that “the principle that the meanings of words are
knowable a priori . . . is taken for granted by most philosophers of language and by many
linguists” ( McKinsey1987: 1).
Now, of course, the idea that some knowledge is a priori is widespread in philosophy.
Nonetheless, Quine has raised serious doubts about it. The main problem with the
idea, in my view (Devitt 1994, 1996; 1998; 2010a), is that we do not have even the
beginnings of an account of what a priori knowledge is. We are simply told what it isn’t:
namely, empirical knowledge. Still, suppose we set such general doubts aside and accept
that at least our knowledge of mathematics and logic is a priori, what could be the basis
for supposing that our knowledge of meanings is too? The meaning of a word is presumably
constituted by relational properties of some sort: “internal” ones involving inferential
relations among words and/or “external” ones involving certain direct causal relations
to the world referred to. Where the meaning is partly constituted by a certain external
relation—plausible examples are the meanings of ‘Cicero’, ‘elm’, and ‘water’—then it is
hard to see how a priori reflection on what is “inside the head” could establish that such
a relation constituted a meaning. But even the meaning-constituting internal relations
pose serious problems for the a priori view. Let Susan be a competent speaker alleged
to have a priori knowledge that, say, the inferential relation between ‘bachelor’ and
‘unmarried’ is part of the meaning of ‘bachelor’. The first problem for the a priori view
is that it requires that Susan comes to this belief about meaning rather than the belief
that the inferential relation is simply the reflection of familiar knowledge about bachelors.
Why should we suppose that her competence alone leads her to make this apparently large
theoretical step? Suppose, however, that she does make it. The second and more serious
problem for the a priori view is that we seem to have no basis for thinking that, simply in
virtue of her competence, Susan’s belief about meaning is justified. We have no basis for
giving her belief any special epistemic authority, and thus turning it into knowledge. We
need a plausible explanation of these allegedly nonempirical processes of belief forma-
tion and justification and some reasons for believing in them.

6 Embodied Theory?
If the view that competent speakers have a priori knowledge of linguistic facts does
not hold up, what else could justify the ubiquitous reliance on intuitions in the phi-
losophy of language? Philosophers provide little in the way of an answer. As Jaakko

557
 M I CH A EL D E V I T T

Hintikka remarks, talking of philosophy generally: “One searches the literature in vain
for a serious attempt to provide” a justification for the appeal to intuitions (Hintikka
1999: 130). In similar vein, Timothy Williamson remarks: “there is no agreed or even
popular account of how intuition works, no accepted explanation of the hoped-for cor-
relation between our having an intuition that P and its being true that P.” He describes
this as “a methodological scandal” (Williamson 2007: 215). Still, we can look to lin-
guistics for a possible answer for semantic intuitions. (Perhaps we should see this answer
as an explication of the obscure a priori answer.)
The common linguistic view of intuitive judgments is expressed in passages like the
following: “It seems reasonably clear, both in principle and in many specific cases, how
unconscious knowledge issues in conscious knowledge . . . it follows by computations
similar to straight deduction” (Chomsky 1986: 270)
I have described the common view as follows: linguistic competence, all on its
own, “provides information about the linguistic facts. . . . So these judgments are not
arrived at by the sort of empirical investigation that judgments about the world usually
require. Rather, a speaker has a privileged access to facts about the language, facts cap-
tured by the intuitions, simply in virtue of being competent . . .” (Devitt 2006a: 483–4;
2006b: 96)
On this view, intuitive syntactic judgments are, “noise” aside, “the voice of compe-
tence.” Let’s call this thesis “VoC.” We can identify two versions of it, a “standard” one
which requires linguistic rules (and principles) to be represented in the mind and a “non-
standard” one which does not. According to the standard version, suggested by the Chom-
sky quote, speakers derive their intuitive judgments from their representations of linguistic
rules by a causal and rational process like a deduction. According to the nonstandard
version, the intuitions must be provided somehow by embodied but unrepresented rules
(Devitt 2006a: 482–6; 2006b: 96–8).
Stich has suggested that philosophers of language might be guided by linguistics in
seeking a justification for the authoritative role given to semantic intuitions (Stich
1996: 40; see also Hintikka 1999, Williamson 2007). Stich’s particular concern is with
referential intuitions. He suggests that philosophers may think that speakers derive their
referential intuitions from a representation of referential principles. So, just as, accord-
ing to the standard version of VoC, the true grammar that linguists seek to discover is
already represented in the mind of every speaker, so too, according to this suggestion,
are true semantic theories of reference, meaning, and the like. Semantic intuitions,
like syntactic ones, are the result of something like a deduction from a represented
theory. Thus, speakers have Cartesian access to linguistic facts simply in virtue of being
competent.
Stich’s suggestion concerns the standard version of VoC, which posits representations
of theories of language in the minds of ordinary speakers. What about the nonstandard
version which does not posit these representations? Stich does not consider this version
as a model for semantic intuitions, but if the model worked for syntax it should work for
semantics as well.
Someone who took the usual linguistic view of grammatical intuitions might well be
tempted by this analogous view of semantic intuitions. So, it is interesting to note that
Chomsky is not tempted. He expresses skepticism about “contemporary philosophy of
language” and its practice of “exploring intuitions about the technical notions ‘denote’,
‘refer’, ‘true of’, etc.” He claims that “there can be no intuitions about these notions,
just as there can be none about ‘angular velocity’ or ‘protein’. These are technical terms

558
 TH E RO LE O F I NT U I T I O N S

of philosophical discourse with a stipulated sense that has no counterpart in ordinary

language.” (Chomsky 1995: 24)
So Chomsky is skeptical about the use philosophers make of semantic intuitions.
But he is not, of course, similarly skeptical about the use linguists make of syntac-
tic ones. Why the difference? If skepticism about semantic intuitions is appropriate,
then surely just the same skepticism is appropriate about the syntactic ones, and for
just the same reason. All the terms in linguistic theory are, in the relevant sense, tech-
nical and theory-laden. A few like ‘grammatical’ and ‘sentence’ have counterparts in
ordinary language but so too do ‘denote’ and ‘refer’. Semantic and syntactic intuitions
are on a par. Chomsky seems to have given a good reason for rejecting VoC
altogether.
In the last section we looked dimly on the idea that linguistic intuitions are a priori.
We have just raised a preliminary doubt about VoC. Before raising more doubts, we
need to describe an alternative view. This alternative accompanies the view that lin-
guistic competence is simply a skill.

7 Competence as a Skill
It is natural to follow the folk in talking of a person’s linguistic competence as a form
of “knowledge.” We noted in section 4 that it is common to think of this knowledge
as propositional, knowledge-that, leading to the idea that speakers have Cartesian access
to linguistic facts. But there is an alternative. If we must follow the folk in talking of
competence as knowledge, we can think of it as mere knowledge-how. However, I think
it is better to talk of the competence as simply a skill or ability.
What skill or ability is the competence? Accepting, as we should, that “language
expresses thought,” I give the following answer:

the competence is the ability to use a sound of the language to express a thought
with the meaning that the sound has in the language in the context of utter-
ance; and the ability (together with some pragmatic abilities) to assign to a
sound a thought with the meaning that the sound has in the language in the
context of utterance (similarly for inscriptions, etc.).
(Devitt 2006b: 148)

We can move to a more theory-laden view of competence if we adopt the popular, and
in my view correct, representational theory of the mind according to which any thought
involves standing in a certain functional relation to a mental representation. Competence
is then “the ability to translate back and forth between meaningful mental representations
and the sounds of the language.” And if we go further to the controversial language-of-
thought hypothesis according to which the mental representation is language-like, the
translation is “between mental sentences and the sounds of the language” (ibid.). Finally,
linguistic competence is complex, consisting of syntactic competence and lexical compe-
tence. Thus, going along with the language-of-thought hypothesis, syntactic competence
is the ability to translate back and forth between the syntactic structures of the sounds of
the language and the structures of mental sentences. And lexical competence is the ability
to translate back and forth between the words of the language and mental words.
Why think that that linguistic competence is just a skill or ability? Briefly, because
it has all the marks of one: It has limited plasticity; it is extraordinarily fast; the process

559
 M I CH A EL D E V I T T

of exercising it is unavailable to consciousness; once established, it is “automatic” with

the result that it can be performed while attention is elsewhere (Devitt 2006b: 209–10);
it is very likely acquired by “implicit learning” (Devitt 2006b: 219). But shouldn’t we
suppose that in the case of linguistic competence, the skill involves knowledge-that? I
argue not (Devitt 2006b). We should suppose that only if we have some powerful rea-
sons for doing so. Otherwise the supposition seems gratuitous. Why suppose that simply
in virtue of being competent in a language a person must have propositional knowledge
about the language? Why suppose that speakers have this sort of Cartesian access to
linguistic facts? Why not suppose, rather, the modest view that any knowledge of these
facts that a speaker may have comes from ordinary empirical reflection on linguistic
phenomena?
Assumptions that competence involves propositional knowledge are very immodest.
One would expect, therefore, that they would be well supported by arguments. This is not
what we find. Arguments for them are few and far between and remarkably thin. (Stanley
and Williamson 2001 may rate as an exception; Devitt 2011b is a response.) One gets the
impression that these propositional assumptions are thought to be too obvious to need
argument.

8 Intuitions As Empirical Judgments

If linguistic competence is simply a skill, requiring no propositional linguistic knowl-
edge, it is hard to see how it could be the source of our linguistic intuitions as VoC
requires. But what is the alternative? I think that we should see intuitions as ordinary
empirical judgments (Devitt 1994; 1996; 2006a,b,c; 2011a; see also Kornblith 1998).
Like the skill-view of competence, this alternative is modest: It makes do with cognitive
states and processes we were already committed to.
The alternative is based on a view of intuitions in general: They “are empirical the-
ory-laden central-processor responses to phenomena, differing from many other such
responses only in being fairly immediate and unreflective, based on little if any con-
scious reasoning” (Devitt 2006a: 491; 2006b: 103). From this perspective, we should
trust a person’s intuitions to the degree that we have confidence in her empirically based
expertise about the kinds under investigation. Sometimes the folk may be as expert as
anyone: Intuitions laden with “folk theory” are the best we have to go on. Perhaps
this is the case for a range of psychological kinds. For most kinds, it clearly is not: We
should trust intuitions laden with established scientific theories. Consider, for example,
a paleontologist in the field searching for fossils. She sees a bit of white stone sticking
through gray rock, and responds immediately, “a pig’s jawbone.” We trust her intuitive
judgment in a way that we would not trust folk judgments because we know that it is the
result of years of study and experience of old bones; she has become a reliable indicator
of the properties of fossils. Similarly, we trust the intuitions of the physicist over those
of the folk about many aspects of the physical world where the folk have proved notori-
ously unreliable. And recent experiments have shown that we should have a similar
attitude to many psychological intuitions. Thus, the cognitive psychologist, Edward
Wisniewski, points out that:

researchers who study behavior and thought within an experimental frame-

work develop better intuitions about these phenomena than those of the intui-
tion researchers or lay people who do not study these phenomena within such

560
 TH E RO LE O F I NT U I T I O N S

a framework. The intuitions are better in the sense that they are more likely to
be correct when subjected to experimental testing.
(Wisniewski 1998: 45)

Although we may often be right to trust an intuition in the short run, it is crucial to see
that nothing rests on it in the long run. We can look for more direct evidence in scien-
tific tests. In such a scientific test we examine the reality the intuition is about. These
scientific examinations of reality, not intuitions about reality, are the primary source of
evidence. The examinations may lead us to revise some of our initial intuitions. They
will surely show us that the intuitions are far from a complete account of the relevant
bit of reality.
What then should we make of linguistic intuitions? All these intuitions arise from reflec-
tion on linguistic data (understood as pieces of linguistic usage, on the model of “primary
linguistic data”). The competent speaker has ready access to a great deal of linguistic data
just as the competent typist has to a great deal of typing data and the competent thinker
has to a great deal of thinking data: The competent speaker and her competent fellows
produce linguistic data day in and day out. So she is surrounded by tokens that may, as
a matter of fact, refer to so and so, be true in such and such circumstances, be grammati-
cal, be ambiguous, and so on. So she is in a position to have well-based opinions about
language by reflecting on these tokens. This is not to say that she will reflect. Indeed,
a totally uneducated person may reflect very little and hence have few if any intuitive
judgments about her language. Still it is clear that the normal competent speaker with
even a little education does reflect on linguistic reality just as she reflects on many other
striking aspects of the world she lives in. And this education will usually provide her with
the terms and concepts of folk semantics and linguistics, at least. As a result she is likely
to be able to judge in a fairly immediate and unreflective way that a token does refer to so
and so, is true in such and such circumstances, is grammatical, is ambiguous, and so on.
Such intuitive opinions are empirical central-processor responses to linguistic phenom-
ena. They have no special authority: although the speaker’s competence gives her ready
access to data it does not give her Cartesian access to the truth about the data.
Still, are these intuitions likely to be right? I think we need to be cautious in accept-
ing them: Thinking about language is notoriously hard and the folk are a long way
from being experts. Still it does seem to me that their “simplest” intuitions, involving
syntactic and semantic vocabulary that we suppose the folk have mastered well enough,
are quite likely to be right. So we can often be confident about judgments that a name
“refers” to x, that this pronoun must “refer to the same thing” as that name, and that
this expression is “ambiguous.” In sum, we have good reason to suppose that the core
judgments of folk linguistics, reflecting the “linguistic wisdom of the ages,” are good
evidence for linguistic theories.
This having been said, the intuitions that philosophers and linguists should prefer are
the ones that they do, as a matter of fact, mostly prefer: those of philosophers and linguists
themselves. For, they are much more expert. This is particularly so when we get beyond the
simple cases to fanciful ones like Kripke’s Gödel (1980) and Putnam’s Twin Earth (1975).
Just as the intuitions of paleontologists, physicists, and psychologists in their respective
domains are likely to be better than those of the folk, so too the intuitions of philosophers
and linguists (c.f. Machery et al. 2004).
To say that intuitions, whether those of philosophers or of the folk, are evidence is
not to say that they are the only, or even the primary, evidence. Indeed, we can look for

561
 M I CH A EL D E V I T T

more direct, less theory-laden, evidence by studying what the intuitions are about, the
linguistic reality itself. But how do we do that? I shall address this briefly in section 10.

9 Rejecting VoC
I think (Devitt 2006a,b,c) that there are several reasons for preferring this modest
explanation of intuitions to VoC, whether as a proposal for syntactic intuitions or as
one, following Stich’s suggestion, for semantic intuitions. (For discussion, see Textor
2009, Fitzgerald 2010, Devitt 2010c, d.) I shall mention the main ones.
The main consideration against the standard version of VoC is as follows. It is of
course possible that the competent speaker’s intuitions about her language are reliable
because they are derived from her mental representations of theories of the language.
It has been argued, mistakenly in my view, that we need to posit such representations
to explain language acquisition and use. In any case, we surely do not need the posit to
explain the reliability of intuitions about language. Consider the analogous phenom-
ena for typing and thinking. We can explain the reliability of intuitions about those
processes by adverting to cognitive states and processes that we are already commit-
ted to, without positing representations of the rules that govern the processes. These
modest explanations seemed perfectly adequate for the job and, indeed, much more
plausible than their representational rivals. So do the similarly modest explanations in
the linguistics case. Language is an important part of the human environment. It is not
surprising that empirical reflection on linguistic data, aided by some education, should
make people fairly reliable detectors of the most obvious facts about language. We are
surely similarly reliable about other important parts of the environment: for example,
the physical, biological, and psychological parts.
The main consideration against the nonstandard version of VoC is that we do not
have any idea how embodied but unrepresented rules might provide linguistic intui-
tions. Not only do we lack the details needed for a plausible explanation, but attention
to other similar systems gives good reason to suppose that the linguistic system does
not provide these intuitions and so we could never have the details. The explanation
would require a relatively direct cognitive path from the embodied rules of the lan-
guage to beliefs about expressions of that language, a path that does not go via central-
processor reflection on the data. What could that path be? Consider some other
examples. It is very likely that rules that are embodied but not represented govern our
swimming, bicycle riding, catching, typing, and thinking. Yet there does not seem to
be any direct path from these rules to relevant beliefs. Why suppose that there is such a
path for linguistic beliefs? Why suppose that we can have privileged access to linguistic
facts when we cannot to facts about these other activities? We do not have the begin-
nings of a positive answer to these questions and it seems unlikely that the future will
bring answers.

10 Other Evidence
On the view I have presented linguistic intuitions are fallible empirical judgments that
are, at best, only indirect evidence of linguistic reality. Where can we find more direct
evidence? This is a large question, beyond the scope of this chapter. I shall briefly men-
tion some points made elsewhere (Devitt 1994, 1996, 2006a, b, 2008, 2011a; see also
Martí 2008).

562
 TH E RO LE O F I NT U I T I O N S

We should start by identifying the “linguistic reality” in question. This reality is to be

found in the utterances of language users. But what is it about these utterances that is theo-
retically interesting? It is natural to think that our interest is in their meanings. Related to
this, since “language expresses thought,” it is natural to think that we are interested in the
meanings (contents) of thoughts. But this talk of meanings is sadly vague, as many have
noted: It is far from clear what counts as a meaning that needs explaining. We need to be
much more precise in identifying the subject matter of semantics. We should identify the
meanings of utterances and thoughts with certain properties of them that are crucial to
their causal roles. Of particular interest here are the roles of thoughts and utterances
in causing intentional behaviors and informing us about the world. So we can identify
the reality that concerns us with the properties of thoughts and utterances in virtue of
which they play those causal roles.
This reality can provide lots of direct evidence of its nature. The main source is “the
corpus,” the utterances that the folk have produced and are producing as they go about
their lives. We can look at the circumstances and consequences of these productions.
Indeed, the linguistic intuitions that are indirect evidence—for example, that people
ignorant about Einstein nonetheless refer to him by ‘Einstein’ (Kripke 1980)—come
largely from reflecting on these features of linguistic usage. The technique of “elicited
production” could provide further direct evidence: Experimental situations “are designed
to be uniquely felicitous for production of the target structure” (Thornton 1995: 140).
Next, rather than creating situations in which we see what people say or understand, we
can describe such situations and ask people what they would say or understand. Finally,
and relatedly, we can look to ordinary opaque attitude ascriptions. Day in and day out,
folk, and social scientists, use “content clauses,” usually ‘that’ clauses like that in “Frank
said that Madonna is gorgeous,” to ascribe properties to thoughts and utterances for
the very purpose of explaining behavior and learning about the world. Insofar as these
ascriptions are right—and we have good reason to suppose that they mostly are right
because they are mostly successful—they provide evidence of the nature of the reality
that concerns us.
This direct evidence often involves behaviors that are aptly called “intuitive,” but it is
very different from the indirect evidence provided by the intuitions that that have con-
cerned us here: For the behaviors are linguistic performances, whereas those intuitions are
judgments about language.

11 Conclusion
Intuitive metalinguistic judgments seem to be the main source of evidence in the phi-
losophy of language. We should reject the idea that this is justified because those intui-
tions are a priori or “the voice of competence.” Rather, we should see the intuitions as
empirical judgments. As such, they are often good, albeit indirect, evidence. And the
intuitions we should prefer are those of philosophers and linguists, because they are
more expert. However, we can find more direct evidence by looking to the linguistic
reality that these intuitions are about.

Acknowledgements
Thanks to Fiona Cowie and David Sosa for helpful comments on a draft and to Elmar
Geir Unnsteinsson for help with research.

563
 M I CH A EL D E V I T T

Related Topics
4.1 The Role of Experiment
4.2 The Role of Linguistics
6.5 A Priority.

References
Cappelen, H., and E. Lepore (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act
Pluralism, Oxford: Blackwell Publishing.
Carston, R. (2004) “Truth-Conditional Content and Conversational Implicature,” in C. Bianchi (ed.) The
Semantic/Pragmatics Distinction, Standford: CSLI Publications.
Chomsky, N. (1986) Knowledge of Language: Its Nature, Origin, and Use, New York: Praeger Publishers.
—— (1995) “Language and Nature,” Mind 104: 1–61.
DePaul, M. R., and W. Ramsey (eds.) (1998) Rethinking Intuition: the Psychology of Intuition and its Role in
Philosophical Inquiry, London: Rowan and Littlefield Publishers.
Devitt, M. (1994) “The Methodology of Naturalistic Semantics,” Journal of Philosophy 91: 545–72.
—— (1996) Coming to Our Senses:A Naturalistic Program for Semantic Localism, Cambridge: Cambridge
University Press.
—— (1998) “Naturalism and the A Priori,” Philosophical Studies 92: 45–65. Reprinted in Devitt 2010b.
—— (2006a) “Intuitions in Linguistics,” British Journal for the Philosophy of Science 57: 481–513.
—— (2006b) Ignorance of Language, Oxford: Clarendon Press.
—— (2006c) “Intuitions,” in V. G. Pin, J. I. Galparaso, and G. Arrizabalaga (eds.) Ontology Studies Cuad-
ernos de Ontologia: Proceedings of VI International Ontology Congress (San Sebastian, 2004), San Sebastian:
Universidad del Pais Vasco. Reprinted in Devitt 2010b.
—— (2008) “Explanation and Reality in Linguistics,” Croatian Journal of Philosophy 8: 203–31
—— (2010a) “No Place for the A Priori,” in M. J. Shaffer and M. Veber (eds.) What Place for the A Priori,
Rodopi Press, forthcoming. Reprinted in Devitt 2010b.
—— (2010b) Putting Metaphysics First: Essays on Metaphysics and Epistemology, Oxford: Oxford University
Press.
—— (2010c) “What ‘Intuitions’ are Linguistic Evidence?” Erkenntnis 73: 251-64.
—— (2010d) “Linguistic Intuitions Revisited,” British Journal for the Philosophy of Science 61: 833-65.
—— (2011a) “Experimental Semantics,” Philosophy and Phenomenological Research LXXXII: 418-35.
—— (2011b) “Methodology and the Nature of Knowing How,” Journal of Philosophy CVIII: 205-18.
Dummett, M. (1978) Truth and Other Enigmas, Cambridge, MA: Harvard University Press.
Evans, G. (1982) The Varieties of Reference, J. McDowell (ed.), Oxford: Clarendon Press.
Fitzgerald, G. (2010) “Linguistic Intuitions,” British Journal for the Philosophy of Science 61: 123–60.
Haegeman, L. (1994) Introduction to Government and Binding Theory, 2nd edn, Oxford: Blackwell Publishers.
Heidelberger, H. (1980) “Understanding and Truth Conditions,” in P. A. French, T. E. Uehling, Jr., and H.
K. Wettstein (eds.) Midwest Studies in Philosophy Vol. 5, Studies in Epistemology, Minneapolis: University of
Minnesota Press.
Hintikka, J. (1999) “The Emperor’s New Intuitions,” Journal of Philosophy 96: 127–47.
Ichikawa, J., I. Maitra, and B. Weatherson, B. (2011) “In Defense of a Kripkean Dogma,” Philosophy and
Phenomenological Research, forthcoming.
Katz, J. (1997) “Analyticity, Necessity, and the Epistemology of Semantics,” Philosophy and Phenomenological
Research 57: 1–28.
Kornblith, H. (1998) “The Role of Intuition in Philosophical Inquiry: An Account with no Unnatural Ingre-
dients,” in DePaul and Ramsey 1998.
Kripke, S. A. (1980) Naming and Necessity, Cambridge, MA: Harvard University Press.
Loar, B. 1987 “Subjective Intentionality,” Philosophical Topics 15: 89–124.
McKinsey, M. (1987) “Apriorism in the Philosophy of Language,” Philosophical Studies 52: 1–32.
—— (1994) “Individuating Belief,” in J. Tomberlin (ed.) Philosophical Perspectives, 8: Logic and Language,
Atascadero, CA: Ridgeview Publishing Company.
Machery, E., R. Mallon, S. Nichols, and S. P. Stich (2004) “Semantics, Cross-Cultural Style,” Cognition 92:
B1–B12.
Martí, G. (2009) “Against Semantic Multi-Culturalism,” Analysis 69: 42–48.

564
 TH E RO LE O F I NT U I T I O N S

Neale, S. (2004) “This, That, and the Other,” in M. Reimer and A. Bezuidenhout (eds.) Descriptions and Beyond,
Oxford: Clarendon Press.
Putnam, H. (1975) “The Meaning of ‘Meaning,’” in Mind, Language and Reality: Philosophical Papers Vol. 2.
Cambridge: Cambridge University Press.
Recanati, F. (2004) Literal Meaning, Cambridge: Cambridge University Press.
Stanley, J., and Z. G. Szabó (2000) “On Quantifier Domain Restriction,” Mind and Language 15: 219–61.
Stanley, J. and T. Williamson (2001) “Knowing How.” Journal of Philosophy 98: 411–44.
Stich, S. (1996) Deconstructing the Mind, New York: Oxford University Press.
Textor, M. (2009) “Devitt on the Epistemic Authority of Linguistic Intuitions,” Erkenntnis 71: 395–405.
Thornton, R. (1995) “Referentiality and Wh-Movement in Child English: Juvenile D-Linkuency,” Language
Acquisition 4: 139–75.
Williamson, T. (2007) The Philosophy of Philosophy. Oxford: Blackwell Publishing.
Wisniewski, E. J. (1998) “The Psychology of Intuitions,” in DePaul and Ramsey, 1998.

565
 Section V

LO G IC F O R
PH ILO SOP H E R S
O F LA N G U A G E
 5.1
MOD E L THEOR Y:
WHAT IT IS AND
WHAT IT ISN’ T
John P. Burgess

Introduction
Model theory is the branch of logic concerned with relations between formal lan-
guages and extralinguistic structures. The most basic technical notion is that of a for-
mula being true in a given structure, and the subject originated with Tarski and his
famous definition of truth for classical first-order languages. On a common usage fol-
lowed here, a model is just a structure of the kind sentences are true or false in. Model
theory contrasts with proof theory, which is concerned only with formal deducibility
relations among sentences, regardless of any connection with extralinguistic structures.
Important post-Tarski contributors to the model theory of classical first-order languages
have included Abraham Robinson, who gave the subject an orientation towards appli-
cations to abstract algebra, where the pertinent structures are groups and rings and
fields, and Saharon Shelah, who introduced methods of great technical sophistication.
Many mathematical logicians today use “model theory” to refer only to technically
sophisticated, algebra-oriented, first-order model theory, a subject with little bearing
on philosophy. Model theory has been developed also, however, for nonclassical formal
languages (tense, modal, and other), especially by Saul Kripke and his successors. This
branch of the subject claims applications to philosophy of language and theoretical
computer science.
In the usage of computer science and philosophical logic, model-theoretic concepts
are often called “semantic,” and proof-theoretic concepts “syntactic.” This usage can be
traced back to Tarski, who noted that the word “true” was used in several senses, and
to distinguish the one that concerned him drew on vocabulary used in Morris (1938),
where syntax is concerned with relations among words, semantics with relations between
words and things, and pragmatics with relations among words and things and people.
Tarski was concerned with the notion of truth according to which “Snow is white” is
true iff snow is white. William James, by contrast, took “Snow is white” to be true iff it
is useful to believe that snow is white. (Here “iff” abbreviates “if and only if.”) Tarski’s
notion is “semantic” in involving words and a thing (snow), while James’s is “prag-
matic” in involving also people (believers). Tarski called his development of a rigor-
ous definition of truth as he conceived it “the establishment of scientific semantics.”
 J O H N P. BU R G E S S

Linguists, by contrast, apply the term “semantics” to the branch of their subject con-
cerned with meaning in natural language. That there is a tension between this usage
and the Morris–Tarski usage is evident if one considers the contrast between (i) deno-
tation or reference or extension and (ii) connotation or sense or intension. Morris’s
words-and-things conception of semantics is more suggestive of (i), while the linguists’
meaning conception of semantics is more suggestive of (ii).
It is hard to keep the two notions of “semantics” apart in theoretical computer sci-
ence, since there one is concerned with artificial languages, and the specification of
the conditions under which a sentence of such a language is to count as true in a given
structure is about all one gets by way of a specification of what such a sentence is sup-
posed to mean. Elsewhere it is crucial to distinguish, since fallacies of equivocation
between the two senses of “semantics” can be damaging to the understanding of scope,
limits, and status both of model theory and of meaning theory. For instance, confusion
of “formal semantics” or model theory, to which Tarski’s definition of truth is central,
with “linguistic semantics” or meaning theory can lead straight to Davidsonianism, the
contentious claim that specification of truth conditions is central to giving a theory of
meaning for natural language; but to arrive at Davidsonianism that way would be to pre-
judge a controversy that can only be fairly judged after an examination of rival research
programs and empirical linguistic data.
In logic, the nature of any relationship between models and meaning varies with the
formalism under consideration (classical first-order, tense, modal, or other) and must be
considered separately in each case. The methods of model theory are too flexible for it
to be safe to assume that, just because there is a rigorously defined notion of model for a
given formalism, it must have some intuitively intelligible meaning. And they are often
too contrived for it to be safe to assume that, just because some feature is present in the
models used for a given formal language, an analogous feature must be present also in
whatever that formal language may be intended to represent. What follows here is an
elaboration and illustration of such warnings.

Classical first-order logic

On one traditional conception, logic’s central concern is with consequence or implica-
tion, where premises have as a consequence or imply a conclusion iff form alone guaran-
tees that if the premises are all true, then the conclusion is true. The notion of validity is
the special, zero-premise case: A conclusion is valid if its form alone guarantees that it
is true. The general case can be reduced to the special case, since A1, . . ., An imply B iff
the conditional, sometimes called the “leading principle” of the inference from A1, . . .,
An to B, namely the following:

(1) If A1 and . . . and An, then B

is valid.
In modern logic, forms are represented by formulas, the notion of validity is applied
in the first instance to formulas, and validity of a conclusion is identified with being
an instance of a valid formula: the result of substituting specific sentences, predicates,
or whatever for the sentence letters, predicate letters, or whatever in the formula (and
if quantifiers are involved, fixing what they are supposed to range over). Intuitively,
a formula is supposed to be valid if it is guaranteed that all instances must be true, but

570
 M O DEL TH E O RY

logicians could never begin work on classifying formulas as valid or invalid if they had
to wait for philosophers of logic to achieve agreement as to just what the “guarantee”
or “must” here amounts to. For present purposes we need only a definition that gets
the extension right, and may take a formula to be valid if all instances are true, never
mind about must. Such a definition might not get the extension right if our language
were severely impoverished, but fortunately it isn’t. Or at least, it presumably isn’t on
the understanding, here adopted, that instantiation may be by expressions containing
free variables (or analogous expressions of natural language), with truth in all instances
requiring truth of all instantiations for all values of any free variables (or analogous expres-
sions). For example, Px may be instantiated by “x is an element of the set u,” with
validity requiring truth whatever set the u in question is taken to be, or by “x likes it,” with
validity requiring truth whatever item the it in question is taken to be.
One can show that a given conclusion is not a consequence or implication of given
premises by replacing the nonlogical words in premises and conclusions by letters to
give a formula, and then presenting an instantiation in which other nonlogical words
to replace the letters in which the premises become true and the conclusion false. This
is not essentially different from the method traditionally used in mathematics to show
that, for instance, the parallel postulate does not follow from other axioms of Euclidean
geometry, except that traditionally the form was not explicitly exhibited, and one sim-
ply presented a substitution of new nonlogical words for the original ones.
The most basic goals of model theory for pure logic are to define rigorously: first, a
technical notion of model capturing all and only the features of an instance that are relevant
to truth; and second, a technical notion of truth in a model such that truth in the intuitive
sense of an instance will coincide with truth in the technical sense in a corresponding model. It
then follows that the intuitive notion of validity as truth in all instances will coincide
with the rigorously defined notion of validity as truth in all models. For this coincidence
to obtain and be useful, the class of models must be (a) broad enough that there is a
model corresponding to any instance; and (b) narrow enough to be more tractable than
the class of all instances.
At the level of classical sentential logic, instantiation will simply be substitution of sen-
tences S, S’, . . . for sentence letters p, p’, . . ., and since all that matters for the truth of the
result is the truth-values of the sentences, and not, for instance, their meaning, for a model
we may simply take an assignment of truth-values to sentence letters. We can then define
the notion M |= A of the truth in a model M of a formula A built up from sentence letters
using negation and conjunction and disjunction ~ and š and ›, inductively as follows:

(2a) M |= pi iff M(pi) = true

(2b) M |= ~A iff not M |= A
(2c) M |= A š B iff M |= A and M |= B
(2d) M |= A › B iff M |= A or M |= B

(Similarly for the conditional o and biconditional l.) A valid formula is then one
that is true in all such models. The desiderata (a) and (b) are achieved, the former obvi-
ously, the latter outstandingly, since for a formula containing only specific sentence let-
ters (for example, the three p, q, r) there will be only finitely many (in the example, just
eight) models or combinations of truth values, whereas there are an infinity of instances
(triples of sentences that might be put in for p, q, r). This reduction from infinite to
finite is responsible for the decidability of classical sentential logic.

571
 J O H N P. BU R G E S S

At the level of classical predicate logic, the notion of model becomes more compli-
cated. What matters for the truth of an instance is, first of all, which things are being
spoken of, or in jargon, the universe of discourse or quantifier domain; and further, for
each k-place predicate, which k-tuples of elements of that universe it is true of. (If
constants are present, they must be assigned distinguished elements of the domain; and
if function symbols are present, they must be assigned functions on the domain, of the
right number of places; but for simplicity we ignore such complications.) Differences
in meaning between the predicates in different instantiations will not matter, so long
as the predicates are true of the same things, so the notion of model may ignore them.
The official definition of a model M accordingly takes one to consist of a nonempty set
M as quantifier domain, and an assignment to each k-place predicate letter P of a set
PM of k-tuples of elements thereof, called M’s interpretation of P. The notion of truth in
a model also becomes more complicated, involving a detour through the notion of an
open formula’s satisfaction by elements in a model; but Tarski’s definition is given in all
standard logic textbooks today, and need not be repeated here.
As for desideratum (b), the class of models is vast, but basic results of model theory
imply that truth in all models is equivalent to truth in all models of a restricted kind.
To begin with, the Löwenheim-Skolem theorem states that a formula will be true in all
models iff it is true in all models with a countable quantifier domain.
As for desideratum (a), there is a problem. We want to apply classical logic to evalu-
ate arguments in, among other fields, set theory. We need therefore to consider formulas
with a single two-place predicate letter and the instantiation in which the quantifier
domain consists of all sets, and what is put in for the predicate letter is “is an element of.”
But there is no model corresponding to this instantiation, since our official definition of
model requires the quantifier domain to be a set, and there is no set of all sets. In Kreisel
(1967), where the problem was first noted, a solution is provided, based on comparison
of the notions of: (i) formal provability, (ii) validity in the intuitive sense (truth in all
instances, even if the domain is not a set), (iii) validity in the official sense (truth in all
models, where the domain must be a set). Intuitively, (i) is found to imply (ii) on inspect-
ing the axioms and rules of formal proof; and obviously (ii) implies (iii); but the Gödel
completeness theorems tells us (iii) implies (i), so the intuitive notion (ii) is caught in a
scissors between the technical notions (i) and (iii), and all three notions agree. (For sec-
ond-order logic, by contrast, and with it plural logic, there is no completeness theorem
and no scissors, and the assumption that every formula that is valid in the official sense
is valid in the intuitive sense is known to be unprovable from the usual axioms of set
theory, leaving the ultimate significance of the model theory in some doubt.)

Tense logic
The simplest time-distinctions are expressed in English not through overt quantifica-
tion over “times” but by verb inflections or auxiliary verbs. There is nothing comparable
to these verb modifications in classical logic, which was designed for application to pure
mathematics, where nothing ever has been or ever will be other than as it is. Tense logic
as developed in Prior (1967) and elsewhere introduces operators F and P, pronounced
“(sometime) will be” and “(sometime) was.” In terms of these, other operators G =
~F~ and H = ~P~, pronounced “always will be” and “always was” can be defined. For
simplicity we consider only the future pair F, G in examples. The form of such a natural
language sentence as

572
 M O DEL TH E O RY

(3) If he ever goes, she will go later

might be represented using these operators thus

(4) G(p o Fq)

with p/q for “he/she goes.”

Though nothing ever changes in mathematics, mathematics nonetheless can be and
is applied to physics, where everything changes. The application requires some depar-
ture from our usual ways of speaking and thinking, adding a phrase “at time t” to each
verb, and replacing the present tense “is” that contrasts with “was” and “will be” with a
tenseless “|is|” that is short for “was or is or will be.” Taking this approach, the form of
the premise (3) might be represented classically as

(4') t(t0 < t š Pt o t'(t < t' š Qt'))

with t0 standing for the present time, < for the relative-futurity or earlier-later relation,
and Pt or Qt for “he or she |goes| at time t.” Comparison of (4) and (4’) suggests how
in general tense-logical formulas can be transcribed as classical-logical formulas, hence
how any natural language representable by a tense-logical formula is representable by a
classical-logic formula. Each G or F will correspond to an  or , each sentence letter pi
will correspond to a one-place predicate letter Pi, with Pit to be thought of as meaning
something like “pi |is| true at time t.”
Classical first-order logic counts the premises as implying the conclusion in very few
tense-logical arguments as they stand, though the conclusion is often a consequence of
the stated premises plus some further unstated premise about the structure of time, such
as the transitivity of the earlier-later relation:

(5) tt't''(t < t' š t' < t'' o t < t'')

It is the job of the physicist to say which such assumptions are true, and of the logician
to say what follows from each. Tense logic seeks to develop systems of axioms and rules
for deriving tense-logical theorems working entirely within the tense-logical language.
The tense logician would like to be able to tell the physicist which axioms hold on which
assumptions about time, but there is this difficulty, that the physicist’s assumptions about
time are formulated in the mathematical style (5) as principles about a timeless relation of
earlier to later among entities called times, and the kind of language with which tense logi-
cians directly work. It is in order to connect the two languages that tense logicians intro-
duce a notion of model that is a direct adaptation of the classical notion of model for the
first-order language with one two-place predicate < and many one-place predicates Pi.
A model M consists of (i) a set M of elements representing “times,” (ii) a two-place
relation <M on the set representing “earlier-later,” and (iii) a component telling us for
each sentence letter pi and each time t whether pi is true or not at t. Here (i) and (ii) are
just are the classical quantifier domain and interpretation of <. Classically (iii) would
be represented by an assignment to each Pi of an interpretation PiM, to be thought of as
the set of times when pi is true; tense-logically, (iii) is more conveniently represented
by a function assigning each time t a function Mt assigning each sentence letter a truth
value; the two representations are mathematically equivalent.

573
 J O H N P. BU R G E S S

The notion of the truth of a formula A at a time t in a model M, written M, t |= A can

then be defined by recursion on the complexity of the formula, adapting the classical
definition thus:
DISPLAY

(6a) M, t |= pi iff Mt(pi) = true

(6b) M, t |= ~A iff not M, t |= A
(6c) M, t |= A š B iff M, t |= A and M, t |= B
(6d) M, t |= A › B iff M, t |= A or M, t |= B
(6e) M, t |= FA iff M, t' |= A for some t' with t <M t'
(6f) M, t |= GA iff M, t' |= A for all t' with t <M t'

A model may be required to have also (iv) a designated element t0 representing the
“present,” in which case truth in the model as a whole is defined as truth at t0, and valid-
ity as truth in all models; more usually, validity is simply defined as truth at all times in
all models.
A certain minimal axiom system has been identified, which gives as theorems all
formulas that are valid without special assumptions about the structure of time, and
various additional axioms corresponding to various such special assumptions have also
been identified. For instance, corresponding to transitivity (5) we have the axiom

(7) FFp o Fp

intuitively meaning roughly

(8) Whatever will be going to be is going to be.

Prior offered something like (8) as the “cash value” of what is asserted using tense-
less mathematical language by something like (5). The “cash values” of quite sophisti-
cated physical theories of time, as in relativity theory, have been worked out by tense
logicians.
If one confused “semantics” in the sense of model theory with “semantics” in the
sense of meaning theory, one would likely be led to conclude that it is not (8) that
gives the cash value of (5), but (5) that gives the cash value of (8). One would likely
be led, that is to say, to the conclusion that sentences involving tense-operators—and
presumably therewith the sentences of tensed natural language whose surface features
tense-operators were intended to represent—deep down really mean something that
quantifies over such entities as “times” or “instants”: that what appears to be an histori-
cally late, sophisticated, philosophico-scientific posit, the “instant of time,” was really
an implicit assumption of ordinary speech all along.
But any conclusion of this sort about natural language would be an empirical linguistic
hypothesis, requiring empirical linguistic evidence. There may well be good evidence of
the right kind for some such hypothesis; but if so, that only makes it the more important
for the student of these matters to be alert to the distinction between such genuine
empirical linguistic evidence and the spurious support that would be provided by an
equivocating argument sliding fallaciously from the presence of “times” or “instants” in
models to their presence in “semantics” to their presence in meaning.

574
 M O DEL TH E O RY

Modal logic
Modal logic is concerned with the distinction between what actually is and what is in
one sense or another necessary or possible. The simplest modal distinctions are expressed
in English through modal auxiliary verbs such as “must” and “may.” There is nothing
comparable in classical logic, because in pure mathematics, for which that logic was
designed, nothing could have been other than as it is. Modal logic as developed by C.
I. Lewis (1918) and his successors introduces operators … and ‘, pronounced “neces-
sarily” and “possibly,” the two being interdefinable since … = ~‘~ and ‘ = ~…~, and
develops systems of axioms and rules for deriving modal theorems.
Historically, systems quickly proliferated without much regard to intuitive interpre-
tation, differing in their axioms regarding iterated modalities, such as

(9a) …p o……p
(9b) ‘p o…‘p

For instance, the systems known as S4 and S5 both have (9a) as a theorem, while the
former lacks but the latter has (9b) as well. Modal logic was pursued for half a century,
and dozens of systems were developed, before logicians gave much attention even to so
basic a distinction as that between metaphysical possibility (concerning what potentially
might have been if the world had been otherwise than it actually is) and logical possibil-
ity (concerning what can without internal self-contradiction be supposed about how
the world actually is).
The fact that formal development ran ahead of intuitive interpretation was in some
way a good thing, since it left the formalism open to unanticipated interpretations and
applications, which have included the epistemic interpretation of intuitionistic logic, as in
Shapiro (1985), the provability interpretation of modal logic, or provability logic for
short, as in Boolos (1995), as well as what is called dynamic logic in theoretical computer
science. But it does leave the logician with the question which of the many systems
in existence is appropriate for which of the various senses of modality that have been
distinguished.
During the period before logical and metaphysical modality were clearly distinguished,
a model theory was developed for many modal systems, primarily Kripke (1963a), in
order to address such issues as decidability. Successively considering such formulas as

(10a) ~(p š q) š ‘(p š q)

(10b) ~(p š q) š ~‘(p š q) š ‘‘(p š q)
(10c) ‘(p š q š ‘(~p š q)) š ‘(p š ~q š ~‘(~p š q))
(10d) ‘(p š q š ‘(~p š q)) š ‘(p š q š ~‘(~p š q))

will lead to successively concluding that a model will have to (a) represent not only
what combination of truth values the sentences used to instantiate sentence letters
actually have, but also what combinations they could possibly have had, and (b) rep-
resent possibilities “of different orders,” not only actual possibilities but also possible
possibilities, and (c) represent not only possibilities of various orders but also which
higher-order possibilities are possible relative to which lower-order possibilities, (d)
allow multiple possibilities for which the sentences involved have the same combina-
tion of truth values.

575
 J O H N P. BU R G E S S

All these features are present in a Kripke model M, which consists of (i) a set M of
elements representing possibilities of various orders, (ii) a two-place relation <M on the
set representing relative possibility, and (iii) a function assigning each element t of the
set a function Mt assigning each sentence letter a truth value. This is, of course, exactly
what a model for tense logic amounts to, and the same definition of truth in a model is
used in both logics, … and ‘ replacing G and F, so that clauses (6abcd) are retained,
and (6ef) replaced by

(11e) M, t |= ‘A iff M, t' |= A for some t' with t <M t'

(11f) M, t |= …A iff M, t' |= A for all t' with t <M t'

(Historically, the models for modal logic came first, and those for tense logic later.)
Traditionally, the elements of the set M are called “worlds” and the relation <M
“accessibility.”
A certain minimal axiom system K has been identified that gives as theorems all for-
mulas that are valid without special assumptions about the accessibility relation among
worlds; and various additional axioms corresponding to various such special assump-
tions have also been identified. Notably

(12a) …p o p
(12b) …p o ……p
(12c) p o …‘p

correspond respectively to the assumptions of reflexivity and transitivity and symmetry

of accessibility.
But does this tell us anything about which of the many modal axioms systems is
appropriate for which of the various notions of necessity that have been distinguished?
Here the contrast with tense logic is striking. Physicists’ assumptions about the struc-
ture of time do naturally present themselves as assumptions about the earlier-later rela-
tion among times. Different notions of necessity do not naturally present themselves as
assumptions about the “accessibility” relation among “worlds.”
This is not to say that models are useless, or useful only for technical purposes.
About a decade elapsed between the time when the model theory became well known
through Kripke (1963b) and the time when the distinction between metaphysical and
logical modality became well known through Kripke (1972), but during that decade
the model theory already had an important influence even on intuitive as opposed
to technical issues. It motivated the general rejection of the so-called converse Barcan
formula

(13) …xPx o x…Px

by confirming preexisting suspicions that this formula somehow was saying that what-
ever exists necessarily exists, and also by suggesting an axiom system that avoids (13) as
a theorem. (On the other hand, the mere existence of a formal model theory, which in
itself shows nothing more than that the axioms systems are formally consistent, encour-
aged complacency in the face of Quine’s questioning whether “quantifying in” to modal
contexts made any intuitive sense. If metaphysical modality is meant, it does; if logical
modality is meant, it does not, or so it is argued in Burgess (1988).)

576
 M O DEL TH E O RY

Moreover, even in connection with the question of which axiom system is appro-
priate for which notion of modality, the model theory has played an auxiliary role. If
one wishes to argue that S5 is sound—that every theorem of the system is true in all
instances—when the box is read as “it is true by virtue of logical form that,” there is no
reason to consider models. One may simply argue intuitively that whatever sentence
Sis put in for p in (12c), the result will be true, and similarly for every other axiom.9
It is when one wishes to argue that S5 is complete for logical necessity in the indicated
sense—that every formula true in all instances is a theorem, or equivalently, that every
formula whose negation is not a theorem is true in some instance—that the model
theory becomes useful.
To illustrate how, consider a formula whose negation is not a theorem, such as

(14) ‘(p1 š p2) š ‘(p2 š p3) š ‘(p3 š p1) š ~‘(p1 š p2 š p3)

What we would like to show is that it has an instantiation that is true. What model
theory tells is is that it has a model in which it is true. What we need to do is derive an
instantiation from a model.
To show how this may be done, consider the model with three worlds w1, w2, w3, all
accessible from each other, and with each pi true at precisely the wj for j z i. We first
want, corresponding to the three worlds wi of the model, three sentences Zi whose logi-
cal form will guarantee that exactly one of them is true. These can be obtained starting
from any two logically independent sentences V1 and V2, say:

(15a) V1 = “Snow is white”

(15b) V2 = “Grass is green”

as follows:

(15c) Z1 = V1 and V2
(15d) Z2 = V1 and not V2
(15e) Z3 = not V1

(For a k-world model, we would need to start with n independent sentences, where k d
2n.) Then to each pi, associate the disjunction Si of the Zj corresponding to wj where it
is true. The results amount to

(15f) S1 = “Snow is not white or grass is not green”

(15g) S2 = “Snow is not white or grass is green”
(15h) S3 = “Snow is white”

Logical form does not preclude any two of the Si being true but does preclude all three
being true, as required to give a true instance of (14). The method generalizes.
The correct logic for the provability interpretation has also been determined, and
again models play an auxiliary role, though most of the hard work is getting from a
model for a formula to a true instance of that formula. There is not space to enter into
details here, nor to enter into the situation in other nonclassical logics, such as condi-
tional or intuitionistic or relevance/relevant logic. In every case, mathematical results
about model theory can be useful for philosophical purposes when properly employed.

577
 J O H N P. BU R G E S S

In every case, the model theory does not come with directions for its proper employ-
ment, which rather can only be discovered by philosophical analysis and reflection.

References
Boolos, George (1995) The Logic of Provability, Cambridge: Cambridge University Press.
Burgess, John P. (1998) “Quinus ab omne nævo vindicatus,” in A. A. Kazmi, ed., Meaning and Reference,
Canadian Journal of Philosophy Supplement vol. 23, 25–65.
Kreisel, Georg (1967) “Informal rigor and completeness proofs,” in I. Lakatos, ed., Problems in Philosophy of
Mathematics, Amsterdam: North Holland, 138–186.
Kripke, Saul (1963a) “Semantical analysis of modal logic 1, normal propositional calculi,” Zeitschrift für
Mathematische Logik und Grundlagen der Mathematik, vol. 9, 113–116.
(1963b) “Semantical considerations on modal logic,” Acta Philosophical Fennica, vol. 16, 83–94.
(1972) “Naming and necessity,” in D. Davidson & G. Harman, eds., Semantics of Natural Language,
Dordrecht: Reidel, 253–355 and 763–769.
Lepore, Ernest & Kirk Ludwig (2007) Donald Davidson’s Truth-Theoretic Semantics, Oxford: Oxford Univer-
sity Press.
Lewis, Clarence Irving (1918) A Survey of Symbolic Logic, Berkeley: University of California Press.
Morris, Charles (1938) Foundation of the Theory of Signs, International Encyclopedia of Unified Science,
vol. 1, no. 2, Chicago: University of Chicago Press.
Prior, Arthur (1967) Past, Present and Future, Oxford: Clarendon Press.
Shapiro, Stewart (1985) (ed.) Intensional Mathematics, Amsterdam: North Holland.

578
 5.2
LO GICAL QUANT IF IER S
Gila Sher

This chapter offers a logical, linguistic, and philosophical account of modern quan-
tification theory. Contrasting the standard approach to quantifiers (according to
which logical quantifiers are defined by enumeration) with the generalized approach
(according to which quantifiers are defined systematically), the chapter begins with
a brief history of standard quantifier theory and identifies some of its logical, linguis-
tic, and philosophical strengths and weaknesses. It then proceeds to a brief history of
generalized quantifier theory and explains how it overcomes the weaknesses of the
standard theory. One of the main philosophical advantages of the generalized theory
is its philosophically informative criterion of logicality. The chapter describes the
work done so far in this theory, highlights some of its central logical results, offers an
overview of its main linguistic contributions, and discusses its philosophical
significance.

I Standard Modern Quantifier Theory

1 Historical Beginnings
Modern quantifiers were first introduced by Frege (1879). The quantifiers All and Some
were already recognized as logical operators by Frege’s predecessors, going all the way
back to Aristotle. But Frege offered a new analysis of these operators that enabled him
to deal with multiple-quantifier prefixes and, more generally, multiple nesting of quan-
tifiers. This, in turn, enabled him to extend traditional logic from a logic of “terms”
(1-place predicates standing for properties) to a logic of relations. These changes revo-
lutionized logic. Among the many achievements of Frege’s theory, three are especially
relevant for the present discussion:

(a) Frege’s theory generated a considerably larger array of quantifiers than traditional
logic: Starting with All as his basic logical quantifier, Frege construed not just the
traditional Some, No, and Not All as (defined) logical quantifiers but also infi-
nitely many others, for example, Exactly n, At Least n, and At Most n, for every
natural number n.
(b) Frege’s theory offered a systematic interpretation of multiple-quantifier prefixes
and other nesting quantifiers.
(c) Frege’s expansion of logical quantifiers gave rise to languages with enormous
expressive power. Extensive parts of natural language as well as the entire
 G I LA SH E R

language of classical mathematics and many segments of the language of science

are expressible using his quantifiers.

Frege regarded 1st-order quantifiers as 2nd-order functions or concepts. The 1st-order

quantifier Some is the 2nd-order concept of nonemptiness, the 1st-order All is the 2nd-
order concept of universality. Using contemporary terminology, “(x))x” says that the
extension of ) is not empty, and “(x))x” says that the extension of ) is universal.
Frege sanctioned both 1st-order and higher-order quantifiers. The former quantify
over individual variables, i.e., variables ranging over individuals; the latter quantify
over functional and predicative variables, i.e., variables ranging over functions, prop-
erties, and relations. For reasons that go beyond the scope of the present chapter (but
see discussion of completeness in Section II.1 below), most logicians, linguists, and
philosophers today focus on 1st-order quantifiers. In what follows I will limit myself to
such quantifiers.
Note: Throughout this chapter I am using contemporary terminology.

2 Semantic and Proof-Theoretic Definitions of Quantifiers

Tarski (1933) developed a semantic definition of truth for languages formulated within
the framework of modern logic. The recursive entries for the standard quantifiers—All
() and Some ()—are: given an assignment g of objects (in the intended universe) to
the variables of the language,

() “(x))x” is true under g iff (if and only if) at least one object a (in the
universe) satisfies “)x” under g;
() “(x))x” is true under g iff every object a (in the universe) satisfies “)x”
under g;

where a satisfies “)x” under g iff it satisfies “)x” when all the free variables of “)x”
other than x are assigned values by g.
Gentzen (1935) developed a proof-theoretic method for defining logical constants in
terms of introduction and elimination rules. Such rules tell us, for a given logical constant,
under what conditions we can derive a statement governed by it from a statement (or
statements) not governed by it, and under what conditions we can derive a statement
(or statements) not governed by it from one governed by it. The entries for the standard
logical quantifiers are, essentially:

() Introduction: )(a) |— (x) )x

Elimination: (x) )x |— )(a) provided “a” plays a merely auxiliary role
() Introduction: )(a) |— (x) )x provided “a” is arbitrary
Elimination: (x) )x |— )(a)

where “a” is an individual constant and certain syntactic conditions are given for
“ ‘a’ plays a merely auxiliary role” and “ ‘a’ is arbitrary.” Frege’s, Tarski’s, and Gentzen’s
characterizations identify roughly the same quantifiers, though the last generalizes dif-
ferently from the first two.

580
 LO G I CA L Q U A N T I F I E R S

II Strengths and Weaknesses of Standard Quantifier Theory

1 Logical Results
The logic of the standard quantifiers, i.e., standard 1st-order logic, is rich in logical, or
rather meta-logical, results. Three results that are especially relevant for our discussion
are:

(a) The Löwenheim-Skolem Theorem (1915/20): If a sentence V (set of sentences

6) has a model, it has a countable model, i.e., a model whose cardinality
(i.e., the cardinality of its universe) is smaller than or equal to the cardinal-
ity of the natural numbers (0).

(By definition, V/6 has a model iff there is a model M such that V/all the sentences
of 6 is/are true in M.) This result is naturally viewed as a limitation on the expressive
power of standard 1st-order logic: Within the framework of this logic we cannot “seri-
ously” define the notion of “uncountably many” and many other infinitistic notions,
since every consistent statement of the form “There are uncountably many F’s” comes
out true in some countable model, i.e., a model in which there are only countably
many F’s. This result is sometimes called “Skolem’s Paradox,” although it is not a
real paradox. But it does imply that the representational powers of the standard 1st-
order quantificational languages are limited, since they cannot determine uniquely the
intended meanings of formal terms and, as a result, the meaning of claims (or theories)
containing such terms. For example, a consistent theory that says there are uncount-
ably many objects has a model (i.e., is true in a model) that does not have uncountably
many objects.

(b) The Completeness Theorem (Gödel 1930): Standard 1st-order logic has a
complete proof system. I.e., A sentence V follows logically from a set of
sentences 6 in the semantic sense (6|=V) iff it follows logically from it in
the proof-theoretic sense (6|–V).

The completeness result is perhaps the most important theoretical result of stand-
ard quantification theory. It says that in standard 1st-order logic the proof-theoretic
relation of logical consequence is just as powerful as the semantic relation of logical
consequence. This result is one of the main reasons most contemporary philosophers
and logicians prefer standard 1st-order logic to standard (full) 2nd-order logic. (Full)
2nd-order logic is incomplete: Its proof system is less powerful than its semantic
system.
One consequence of the completeness of standard 1st-order logic is the co-extension-
ality of the semantic and proof-theoretic definitions of the standard quantifiers. Since
proofs, unlike models, are finite structures, this might be viewed as an advantage of the
proof-theoretic definition over the semantic definition.

(c) The Compactness Theorem (Gödel 1930). Two significant formulations

are: (i) If V follows logically from an infinite set 6, then it follows logically
from a finite subset of 6; (ii) If every finite subset of 6 has a model, 6 has a
model.

581
 G I LA SH E R

The compactness result is naturally viewed as an advantage: it enables us to deal with

infinite sets by dealing with their finite subsets.
It should be noted, however, that the connection of these theorems to the standard
1st-order quantifiers is not always simple. For example, compactness and completeness
fail for logics like Lf,Z, which has formulas with arbitrarily large conjunctions and dis-
junctions, but finitely many quantifiers.

2 Linguistic Strengths and Weaknesses

Standard quantifier theory made a substantial contribution to linguistic semantics and
syntax through the works of Geach (1962), Montague (1973), Kamp (1981), May
(1985), Partee (1987), and numerous others. Not only did it allow linguists to analyze
complex linguistic structures involving multiple-quantifier prefixes (e.g., “(x)(y),”
“(x)(y)(z)”) and other nesting quantifier structures (e.g.,”(x)[(y) . . . Š (y)—]”),
it enabled them to formulate important questions concerning quantifier scope, indefi-
nite nouns, anaphora, etc. A well-known problem of this kind is the quantificational
structure of “donkey sentences,” i.e., sentences like

(1) Every farmer who owns a donkey, beats it.

(See discussion in, e.g., Neale 1990.) Moreover, the theory of standard quantifiers
played a central role in the development of important linguistic theories (for example,
LF theory), the establishment of new semantic frameworks (for example, Montague
grammar), and so on.
But the expressive power of the standard quantifiers is linguistically very limited.
Even such simple quantitative determiners as “most”, “few”, “half”, “an even number
of”, “all but ten”, “finitely many”, “uncountably many”, and so on, cannot be expressed
by any (well-formed) combination of standard 1st-order quantifiers. As far as quantifier-
structures in natural language are concerned, it is quite clear that standard quantifier
theory offers no more than a preliminary framework for their study.

3 Philosophical Strengths and Weaknesses

Standard quantifier theory has had a privileged status in contemporary analytic philoso-
phy since its birth in the early years of the 20th century. It is interwoven into most ana-
lytic theories, and its position has been sealed by such works as Quine’s influential Phi-
losophy of Logic (1970). Even philosophers who have challenged the standard theory have
rarely viewed its choice of quantifiers as a problem. They have assigned  and  a large
array of alternative interpretations, going all the way from the substitutional interpre-
tation (Marcus 1972) to the intuitionistic (Dummett 1973) and paraconsistent (Priest
& Routley 1989) interpretations, but few have contested its restriction to the standard
quantifiers. Perhaps the most celebrated feature of the standard theory is the completeness
of its proof method, i.e., the fact that it is integrated into a powerful logic with a complete
proof system (see above). This, indeed, is a desirable feature. For many, including Quine,
the incompleteness of stronger quantifier theories justifies their rejection.
But standard quantifier theory is also widely recognized to have many problems. Here
I will focus on six problems of philosophical significance (some mentioned above, and
some interrelated):

582
 LO G I CA L Q U A N T I F I E R S

(a) Inability to Account for Basic Inferences

In his 1962 abstract, Rescher points out that standard “quantificational logic is unequal
to certain childishly simple valid arguments” like “Most things are A’s, Most things are
B’s; therefore: Some A’s are B’s”, and “Most C’s are A’s, Most C’s are B’s; therefore:
Some A’s are B’s” (ibid.: 374). Since these inferences are of the same kind as the infer-
ences logic is designed to account for (for example,” All things are A’s, Some things are
B’s; therefore: Some A’s are B’s”), this state of affairs is problematic.

(b) Limited Logical Expressive Power

As noted in Section II.2 above, many elementary quantificational notions, includ-
ing quantitative determiner notions like “most”, “few”, “half”, “finitely many”, and
“uncountably many” are indefinable in terms of the standard quantifiers (and other
standard logical constants).

(c) Limited Formal Expressive Power

As demonstrated by the Löwenheim-Skolem Theorem (Section II.1(a) above), many
formal (mathematical) notions, and in particular infinitistic notions like “uncount-
able”, are not only indefinable as logical notions within the standard quantificational
framework but are also inadequately definable as nonlogical notions within that frame-
work. That is, it appears that the standard framework does not enable us to define them
in a way that adequately captures their meaning. (In logicians’ terminology, it enables
us to define “uncountable” only as a “relative” notion but not as an “absolute” notion.)

(d) Absence of a Theoretical Criterion for Logical Quantifiers

It is a surprising fact about modern logic that it has a theoretical, precise, systematic,
informative, and philosophically explanatory criterion for logical connectives but not
for logical quantifiers or predicates. Logical connectives (at least in classical logic) have
a precise and philosophically informative criterion, truth-functionality, which is given a
mathematically precise definition by Boolean algebra. Every logical connective is repre-
sented by a Boolean operator in a 2-element Boolean algebra, and every Boolean opera-
tor of that kind determines a logical connective. Philosophically, every truth-functional
connective—i.e., a connective that takes into account only the formal pattern of truth-
values of its arguments—is logical, and every logical connective is truth-functional.
(On the philosophical ramifications of this criterion, see (e) below.)
In contrast, for logical quantifiers and predicates we have nothing more than an
utterly uninformative, unexplanatory, and unsystematic characterization, namely, a
characterization by enumeration. All this characterization says is that  and  are logical
quantifiers and = is a logical predicate, and that any quantifier definable from  and 
(using logical predicates and connectives) is also a logical quantifier, and any predicate
defined from = (using other logical constants) is also a logical predicate. Why these rather
than others are logical, why only these are logical, remains a mystery. We may try to justify
the standard choice on pragmatic grounds, by appealing to custom or convenient fea-
tures of the associated logic (for example, completeness), but unless we add a theoretical
dimension to such a justification it remains philosophically shallow. Furthermore, even

583
 G I LA SH E R

on purely pragmatic grounds, how do we know that a stronger 1st-order system, i.e., a
1st-order system with additional logical quantifiers, will not give us an overall more
desirable logic? The current situation is thus philosophically unacceptable.

(e) Lack of Philosophical Understanding of Logicality

Given the central role that logic plays in all areas of knowledge, including philoso-
phy, the nature of logicality requires a theoretical account. The nature of logicality,
however, has so far resisted such an account. One of the problems facing us is finding
a fruitful standpoint from which to launch the investigation, i.e., a standpoint, or a
perspective, which will enable us to formulate the question in a precise, systematic,
and philosophically enlightening terms. It is quite clear that such a perspective requires
an informative account of the logical constants. We know that logical truth, logical
consequence, logical consistency, etc., are largely due to the logical constants involved;
understanding the nature of the logical constants is therefore a prerequisite to a theo-
retical understanding of logicality.
Our experience with sentential logic shows that a theoretical criterion for logical con-
stants can, indeed, make a significant contribution to our philosophical understanding
of logicality. Philosophically, the truth-functionality of the logical connectives means
that they abstract from everything but the bare truth pattern of their arguments, and
this explains why logical consequences based on these connectives preserve truth rather
than other features of sentences. Furthermore, this, together with the fact that truth-
functional connectives correspond to mathematical operators, i.e., operators governed
by necessary laws, and the fact that the test for logical truth/consequence based on such
connectives is truth/preservation-of-truth under all possible variations in truth patterns,
explain some of the philosophically significant features of sentential logic. In particular,
they explain its great generality, topic neutrality, immunity to (at least most) empirical
discoveries, and necessity. A philosophically informative criterion for logical quantifiers
(and predicates) is needed to extend this understanding to all of logic.

(f) The Shaky Status of the Semantic Definition of Logical Consequence

for Quantificational Logic.
In his 1936 paper, “On the Concept of Logical Consequence,” Tarski constructed a
semantic definition of logical consequence for quantificational logic that established
logical semantics (model theory) as one of the two major fields of contemporary logic,
along with proof theory. Tarski’s semantic definition says that:

(LC) 6|=V iff every model of 6 is a model of V (or: there is no model in which
all the sentences of 6 are true and V is false).

Tarski anchored his semantic definition in the intuitive idea that a logical consequence
is a necessary-and-formal consequence. To determine the adequacy of his definition, he
asked whether it accurately captured this idea. Investigating this question, he realized
that the answer varies according to the choice of logical constants. For some choices of
logical constants—for example, treating Is-a-Property-of-Napoleon as a logical quanti-
fier or treating  as a nonlogical constant—the answer is clearly negative, while for
others, including the standard choice of logical constants, no conflict seems to occur.

584
 LO G I CA L Q U A N T I F I E R S

But even in the latter case, Tarski realized, it is not clear that his definition fully cap-
tures the idea it was supposed to capture. Adding such nonstandard logical quantifiers
as Finitely Many also accords with this idea. The scope and justification of the contem-
porary definition of logical consequence, thus, require a systematic criterion for logical
quantifiers.
A solution to all the above problems is offered by generalized quantifier theory.

III Generalized Quantifier Theory

1 Brief History
The history of generalized quantifiers starts with Mostowski (1957). Mostowski sug-
gested that we might generalize the standard notion of logical quantifier along two
dimensions, syntactic and semantic. Syntactically, a logical quantifier is a variable-
binding operator that generates new formulas from old formulas. Semantically, a logical
quantifier over a universe A is a cardinality function from subsets of A to a truth-value,
satisfying a certain invariance condition. For example, the existential and universal
quantifiers over A, A and A, are functions that, given a subset B of A:

(2) A(B) = T iff the cardinality of B, |B|, is larger than 0; i.e., iff |B| > 0.
(3) A(B) = T iff the cardinality of the complement of B in A is 0; i.e., iff
|A–B| = 0.

It is easy to see that both  and  satisfy the following invariance condition:

(4) A quantifier Q on A is invariant under all permutations of A.

That is: For every A, B, B’, where A is a nonempty set and B, B’ are subsets of A: If B’
is obtained from B by some permutation of A, then: QA(B) = QA(B’), where a permuta-
tion of A is a bijection p: A oA, i.e., a 1–1 function from A onto A. Using “Ž“ for “is
a subset of” and “p*” for “the function on subsets of A induced by p”, we can spell out
(4) by:

(5) If A is a nonempty set, B Ž A, B’ Ž A, and there is a permutation p of A

such that p*(B) = B’, then: QA(B) = QA (B’).

Generalizing, Mostowski arrived at the following criterion for logical quantifiers:

(PER) A Quantifier Q is logical iff for any non-empty set A, QA is invariant

under all permutations of A.

Under this criterion, every 2nd-order 1-place cardinality predicate is a 1st-order logi-
cal quantifier, and every 1st-order logical quantifier is a 2nd-order 1-place cardinality
predicate. Examples of logical quantifiers in addition to  and , are Most (interpreted
as, say, “more than half”), An Even Number of, Finitely Many, Uncountably Many,
and D-Many for every cardinal number D, finite or infinite. Such quantifiers appear in
English sentences such as

585
 G I LA SH E R

(6) Most things are beautiful.

(7) There is a finite number of stars.

They are symbolized by the formulas

(8) (Mx)Bx
(9) (Fx)Bx,

using obvious abbreviations.

Examples of 2nd-order 1-place predicates that are not logical quantifiers are “some
men”, “all women”, “is a property of Napoleon”, and “is a color property”.
Lindström (1966) generalized Mostowski’s criterion further in two ways. First, he
extended it to 1st-order predicates of all types (for example, “is red” and “=”), and
second, he extended it to 1st-order quantifiers (2nd-order predicates) of all types (for
example, “is a relation between humans” and “is symmetric”).
Lindström’s criterion for logical quantifiers and predicates can be formulated as
follows:

(ISO) A Quantifier/predicate Q is logical iff for any Q-structure S, Q is invari-

ant under all isomorphisms of S.

By a Q-structure S we mean a structure <A, E1, . . ., En>, where A is a nonempty set (the
universe of S), and E1, . . ., En are elements or structures of elements of A of the same
types as the corresponding arguments of Q. Let S = <A, E1, . . ., En> and S’ = <A’, E’1, .
. ., E‘n>. We say that structures S and S’ are isomorphic—SS’—iff there is a bijection
(a 1–1 and onto function) f from A to A’ such that for 1didi n, f*(Ei) = E‘i, where f* is
induced by f. Q is invariant under all isomorphisms of Q-structures S iff for any Q-struc-
tures S, S’, if SS’, then QA(E1, . . ., En) = QA’(E’1, . . ., E‘n).
Among the new quantifiers/predicates satisfying ISO are, in addition to all the above
examples of quantifiers satisfying PER, the identity predicate of standard 1st-order logic
as well as nonstandard quantifiers such as the 2-place Most and Only, and the 1-place
Symmetric and Well-Ordered. As the last two examples suggest, logical quantifiers,
according to ISO, are not limited to (although they include all) cardinality quantifiers.
Examples of a predicate and a quantifier which do not satisfy ISO are “is red” and “is a
relation between humans”.
We can indicate the type of a given quantifier by an n-tuple representing the types of
its arguments, using <1> for a property on, or a subset of, the universe; <1,1> for a pair
of subsets of the universe; <2> for a 2-place relation on the universe, and so on. Quanti-
fiers of types involving only 1’s we call “predicative”, those of types involving at least
one number larger than 1 we call “relational” or “polyadic”. Thus, the standard  and 
are predicative quantifiers of type <1>, but the 2-place Most is a predicative quantifier
of type <1,1>, and the 1-place relational (polyadic) Symmetric is of type <2>. The last
two are defined, semantically, as follows:

Given a universe A, subsets B,C of A, and a binary relation R on A,

(M1,1) (M1,1)A(B,C) = T iff |BˆC| > |B–C|.
(S2) (S2)A(R) = T iff R is symmetric.

They appear in English sentences such as:

586
 LO G I CA L Q U A N T I F I E R S

(10) Most presidents are women.

(11) Equality is a symmetric relation.

Syntactically, these sentences have the logical forms:

(12) (M1,1x)(Px,Wx).
(13) (S2xy) Exy.

Literally, (13) say, “Symmetric xy, x = y”. It means something like “For x and y to be
equal is for them to be symmetrically equal.”
Tarski (1966) formulated essentially the same criterion, albeit in terms of permuta-
tions. As demonstrated by McGee (1996), a formulation in terms of isomorphisms is
preferable.
Note: the presentation of this section is based on Sher (1991).

2 Logical Results
Generalized quantifier theory has had a considerable impact on the development of
mathematical logic since the 1960s. It has led to the development of a new field, model-
theoretic logic, and the establishment of numerous results. (For a representative collec-
tion of articles in this field see Barwise & Feferman 1985.) Two results of special interest
are due to Lindström (1969, 1974) and Keisler (1970).

Lindström’s Theorem
By offering a model-theoretic generalization of the notion of logical constant, Mostowski
and Lindström created tools for comparing the strength of different 1st-order logical sys-
tems (with respect to their ability to define model-theoretic structures). These tools ena-
bled Lindström to arrive at an important characterization of standard 1st-order logic:

(FOL) Standard 1st-order logic is the strongest logic that has both the com-
pactness/completeness and the Löwenheim-Skolem properties (see
Section II.1 above).

Keisler’s Result
Keisler showed that completeness is not limited to standard 1st-order logic; some
stronger logics are also complete. In particular 1st-order logic with the generalized
quantifier Uncountably Many has a complete proof system. (In contrast, 1st-order logic
with the quantifier At Least Denumerably Many does not.) Keisler’s result weakens
the case for the standard 1st-order quantifiers based on completeness (see II.1. and II.3.
above).

3 Linguistic Results
Generalized quantifier theory has led to a renaissance in linguistic semantics. We can
divide its linguistic contributions into two: contributions to the study of natural-lan-
guage determiners, and contributions to the study of natural-language polyadic (relational)

587
 G I LA SH E R

quantifiers. In both cases the restriction to logical quantifiers is relaxed, but here we will
limit ourselves to logical quantifiers.

(a) Determiners
Determiners form an important linguistic category of noun modifiers (different from
adjectives). Determiners include quantity modifiers that are naturally construed as quan-
tifiers. Some of these—for example, “all”, “every”, “each”, “any”, “a”, “some”, “at least
(exactly, at most) one”—are naturally construed as the standard quantifiers,  and , or
as quantifiers defined in terms of these. However, many quantitative determiners can-
not be analyzed using the standard quantifiers. A few examples are “most”, “few”, “half”,
“more . . . than—”, “finitely many”, “uncountably many”, “0”, and so on. Generalized
quantifier theory enables linguists to account for the semantics of all these determiners
by treating them as predicative quantifiers. Among the pioneers in this field are Barwise
& Cooper (1981), Higginbotham & May (1981), van Benthem (1983), Westerståhl
(1985), and Keenan & Stavi (1986).
Quantifiers representing determiners are for the most part of type <1,1>, and some-
times of more complex types, such as <<1,1>,1>. We have seen above how Most1,1 is
defined. 1,1 is defined as follows:

(1,1) 1,1A(B,C) = T iff |BˆC|>0

We may say that 1,1 is generated from 1 by relativizing it to a subset of the universe, B.
Most existential quantifications in English have this (or some derivative) form. In the
simplest case, linguists assign

(14) Some B’s are C’s

the logical form

(15) (1,1x)(Bx, Cx)

rather than

(16) (1x)(Bx & Cx).

A determiner-quantifier of type <<1,1>1> appears in

(17) More students than teachers attended the lecture,

whose logical form is:

(18) (O<1,1>,1x)(Sx, Tx, Lx).

O<1,1>,1 is defined by:

(19) O<1,1>,1A(B,C,D) = T iff |BˆD| > |CˆD|.

588
 LO G I CA L Q U A N T I F I E R S

Linguists and logicians have developed a rich theory of the characteristic features of
determiner-quantifiers. For an up-to-date account see Peters & Westerståhl (2006),
Chs. 3–6.

(b) Relational or Polyadic Quantifiers

Generalized quantifier theory enables us to go beyond determiners to more complex
quantificational structures. The study of relational or polyadic quantifiers in natural
language has been pursued by Higginbotham & May (1981), Keenan (1987, 1992), van
Benthem (1989), Sher (1991), and many others. The simplest relational quantifiers are
of type <2>. We have already defined one relational quantifier of this type, Symmetric.
One way to arrive at relational quantifiers is by combining predicative quantifiers
in some order. In the simplest case, relational quantifiers of this kind correspond to
linear quantifier-prefixes (of other quantifiers); in more complex cases they correspond
to non-linear, branching or partially-ordered quantifier-prefixes.
An example of a relational quantifier equivalent to a linear prefix of predicative
quantifiers is 2, where

(20) (2xy) )xy

is logically equivalent to

(21) (1x)(1y) )xy.

While 2 simply says “for every’ x there is a y such that”, a more intricate relational
quantifier that resembles it but cannot be expressed by a linear prefix of ‘s and ‘s, says
“Every x has its own y”. This quantifiers appears, for example, in

(22) For every drop of rain that falls, a flower grows,

(23) Every family has its own misery. (Every family is unhappy in its own way.
Tolstoy)

If we symbolize this quantifier by “o2”, the logical structure of (22) and (23) is:

(24) (o2xy)(Bx, Cy).

Other interesting relational quantifiers appear in:

(25) Different students answered different questions on the exam. (Keenan

1987)
(26) No three students answered the same number of questions. (ibid.)
(27) Most pairs of people are not married to each other.

An up-to-date discussion of relational (polyadic) quantifiers appears in Peters & West-

erståhl (2006), Ch. 10.
A special type of relational quantifier corresponds to a branching quantifier-prefix.
But the branching prefix is more interesting when thought of as a complex prefix. Two
examples of irreducibly branching quantifier-prefixes are:

589
 G I LA SH E R

(x)(y)
(28) )xyzw. (Henkin 1959)
(z)(w)

(M1x)
(29) )xy. (Barwise 1979)
(M1y)

English sentences exemplifying these forms, or in the case of (29), the more complex
form:

(M1,1x)\1x
(30) )xy (Barwise 1979)
(M1,1y)\2y

are:

(31) Some relative of each villager and some relative of each townsman hate
each other. (Hintikka 1973)
(32) Most boys in my class and most girls in your class have all dated each
other. (Variation on Barwise 1979)

The branching form raises interesting questions of interpretation. In particular, there is

no consensus on the interpretation of non-monotone-increasing branching quantifiers
(Q1 is monotone increasing iff Q1(B) = T, BŽC Ÿ Q1(C) = T) and on how to formulate
a general semantic definition of branching quantifications that will deal with branching
prefixes of any complexity. Most importantly, the branching form raises the question
of compositionality, i.e., the question of whether the interpretation of logically complex
formulas is always determined by that of their simpler components. The key issue is the
structure of the branching prefix, which need not be a tree structure (but only a partial
ordering). In addition to the works mentioned above, see Westerståhl (1987) and Sher
(1990, 1997). For a further development (IF languages) and a discussion of composi-
tionality see Hintikka & Sandu (1989, 2001).

4 Philosophical Results
Generalized quantifier theory provides solutions to all the philosophical problems
plaguing the standard theory that were discussed in Section II.3 above. How successful
these solutions are, is an open question. But generalized quantifier theory undoubtedly
makes an important contribution to philosophy by creating a framework for confronting
philosophical questions that were earlier thought to resist any systematic investigation.
This is especially true of the last three questions, (d)–(f), which concern fundamental
issues in the philosophy of logic.
Briefly, the solutions to the six problems raised in Section II.3 are:

(a) Basic (Non-standard) Logical Inferences

The problem of recognizing and accounting for basic inferences based on quantifier-
notions is largely, and perhaps fully, solved by the generalized theory. All inferences

590
 LO G I CA L Q U A N T I F I E R S

based on such quantitative notions as “most”, “few”, “half”, “finitely many”, “as many
as”, etc., can be accounted by this logic.

(b) Logical Expressive Power

This problem is also largely (perhaps fully) solved by generalized quantifier theory. All
common quantitative notions, including all quantitative determiner notions, can be
construed as logical notions in this theory. Indeed, every classical mathematical notion
can be construed as a logical quantifier, predicate, or function according to this theory,
often by transforming it to a 2nd-order notion. As a result, the expressive power of gen-
eralized 1st-order logic as a whole is very large. (See McGee 1966.)

(c) Formal Expressive Power & the Löenheim-Skolem Theorem

In many systems of generalized 1st-order logic the Löenheim-Skolem theorem does not
hold. As a result, formal or mathematical notions that lose their intended meaning
in standard 1st-order logic retain their intended meaning in such systems. There are
two possible ways of achieving this result: by constructing such notions as logical, and
by constructing them as nonlogical within an appropriate generalized logical system.
The first was noted in (b) above. We can, for example, construct the formal notion
“uncountably many” as a logical quantifier, UNC1, defined, for an arbitrary universe A
by:

(UNC1) UNC1A(B) = T iff |B| > 0.

Since the interpretation of logical constants is built into the apparatus of models (see
Sher 1991), UNC1 will preserve its intended meaning in every model. Sometimes,
however, we prefer to define a formal notion as a nonlogical notion (for example, in
order to explain its meaning or establish its properties). In this case we might be able
to construct an appropriate generalized framework that blocks the Löenheim-Skolem
theorem, and define the desired concept within that framework without subjecting it
to the Löenheim-Skolem effect. (The gain in expressive power is not totally free; often,
the price includes loss of compactness and other properties that make standard 1st-order
logic easier to work with.)

(d) A Theoretical Criterion for Logical Quantifiers

Generalized quantifier theory provides a criterion (necessary and sufficient condition)
for logical quantifiers and predicates—ISO—that is as theoretical, systematic, precise,
and comprehensive as is the truth-functional, Boolean criterion for logical connectives.
Both criteria are essentially invariance criteria. The truth-functional criterion says that
a sentential connective is logical iff it does not distinguish anything about its argu-
ments besides their bare truth value. ISO says that a quantifier/predicate is logical iff it
does not distinguish anything about its arguments besides their bare formal structure.
That is, logical connectives are invariant under all truth-preserving variations in their
arguments; logical quantifiers are invariant under all structure-preserving variations in
their arguments. Both criteria are also precise mathematical criteria: The sentential cri-
terion is based on Boolean algebra, the quantificational criterion is based on set theory

591
 G I LA SH E R

(most commonly ZFC). Finally, both criteria determine a maximal collection of logical
constants, connectives, and quantifiers, respectively. In so doing, both sanction the
existence of a far larger number of logical constants than one would anticipate based
on (mere) intuition, indeed an infinite number of logical connectives and quantifiers.
(The fact that we cannot reduce the collection of logical quantifiers to a small finite
collection as in the case of the logical connectives is to be expected. The power of quan-
tificational logic is so much greater than that of sentential logic that it cannot have as
simple a logical apparatus as the latter’s.)

(e) Philosophical Understanding of Logicality

For reasons indicated in the discussion of this topic in Section II.3 above, a theoreti-
cal criterion for logical constants can be instrumental in providing a philosophically
informative characterization of logicality. What is the philosophical content of ISO?
The idea of invariance under isomorphisms is a systematic rendering of the intuitive idea
of being indifferent to, or not distinguishing, the individual characteristics of objects in
any given universe of discourse. There is a very long tradition of viewing this idea as
characterizing logic. Two examples, taken from Kant and Frege, are:

[General logic] treats of understanding without any regard to difference in the

objects to which the understanding may be directed.
(Kant 1981/7: A52/B76)
Pure logic . . . disregard[s] the particular characteristics of objects.
(Frege 1879: 5)

ISO offers a precise characterization of this condition and places it at the center of our
understanding of logic. To be logical is to abstract from, or disregard, everything about
objects and their relations besides their formal features or structure. That is, to be logical
is to be formal or structural in a very strong sense. (It is, in a certain sense, to be maxi-
mally formal or structural.)
This characterization explains some of the intuitive features of logic: its high degree
of necessity, extreme generality, topic neutrality, strong normativity, and even apriority
(or quasi-apriority). Very briefly, since logical constants are formal, the laws governing
them are necessary. Since logical constants apply regardless of what objects exist in the
domain (universe of discourse), their laws are highly general and topic neutral. Since
they are not affected by the empirical features of objects (or the laws governing these
features), their laws are (largely) nonempirical. And since their invariance is greater
than that of physical and other constants (which are not invariant under all isomor-
phisms), the laws associated with them are normatively stronger than those associated
with other constants (logic does not have to obey the laws of physics, but physics has to
obey the laws of logic). For sources and further discussion see Tarski (1966), Sher (1991,
1996, 2008), McGee (1996), and others.

(f) Foundation of Logical Consequence

By providing a general criterion for logical constants (combining ISO for predi-
cates and quantifiers and the Boolean criterion for logical connectives), generalized

592
 LO G I CA L Q U A N T I F I E R S

quantifier theory enables us to evaluate the adequacy of Tarski’s semantic definition of

logical consequence. The question is whether Tarski’s model-theoretic test yields intui-
tively necessary and formal consequence in languages with logical constants satisfying
the above criterion. The answer appears to be positive. Since (i) Tarskian consequences
depend on the logical constants of the given language and on the apparatus of models;
since (ii) according to the above criterion, logical constants are formal (in the sense
of taking into account only the formal or structural features of their arguments); and
since (iii) Tarskian models represent all formally possible situations relative to the given
language, consequences based on formal constants satisfying Tarski’s test are necessary
and formal.
For a view claiming that the problem of logical constants is a “red herring”, see Etch-
emendy (1990). For a discussion of this and the other issues raised in (d)–(f), see Sher
(1991, 1996, 2008).

5 Criticisms and Alternatives

To date, generalized quantifier theory has been criticized mostly on two grounds.
Some of the criticisms (for example, Hanson 1997 and Gómez-Torrente 2002) con-
cern the applicability of ISO to natural language quantifiers. The critics present exam-
ples of natural-language quantifier expressions that purportedly satisfy ISO yet are intui-
tively nonlogical. They conclude that the characterization of logical constants should
be pragmatic rather than theoretical.
Other criticisms (for example, Feferman 1999) focus on more theoretical issues, like
the relation between logic and mathematics and certain meta-mathematical features of
quantifiers satisfying ISO (although here too, “gut feeling” plays a central role). Two
logicians (Feferman 1999 and Bonnay 2008) offer alternatives to ISO, for example,
invariance under homomorphisms (Feferman). For response to some of these criticisms
see Sher (2001, 2003, 2008).

Related Topics
1.3 Logical Form
1.8 Compositionality
3.4 Quantifiers and Determiners
3.6 Anaphora
4.4 The Role of Mathematical Methods
5.1 Model Theory
5.9 Montague Grammar
7.4 Frege, Russell, and Wittgenstein

References
Barwise, J. (1979) “On Branching Quantifiers in English,” Journal of Philosophical Logic 8: 47–80.
Barwise J. & R. Cooper (1981) “Generalized Quantifiers and Natural Language,” Linguistics & Philosophy 4:
159–219.
Barwise, J. & S. Feferman (eds.) (1985) Model-Theoretic Logics, New York: Springer-Verlag.
Bonnay, D. (2008) “Logicality & Invariance,” Bulletin of Symbolic Logic 14: 29–68.
Dummett, M. (1973) “The Philosophical Significance of Intuitionistic Logic,” in Truth and Other Enigma,
Cambridge, MA: Harvard University Press, 1978, pp. 215–47.

593
 G I LA SH E R

Etchemendy, J. (1990) The Concept of Logical Consequence, Cambridge, MA: Harvard University Press.
Feferman, S. (1999) “Logic, Logics, and Logicism,” Notre-Dame Journal of Formal Logic 40: 31–54.
Frege, G. (1879) “Begriffsschrift, a Formula Language, Modeled upon That of Arithmetic, for Pure Thought,”
in J.v. Heijenoort (ed.) From Frege to Gödel: A Source Book in Mathematical Logic, 1879–1931, Cambridge,
MA: Harvard University Press, 1967, pp. 1–82.
Gärdenfors, P (ed.) (1987) Generalized Quantifiers: Linguistic & Logical Approaches, Dordrecht: D. Reidel.
Geach, P.T. (1962) Reference & Generality: An Examination of Some Medieval & Modern Theories, Ithaca:
Cornell University Press.
Gentzen, G. (1935) “Investigations into Logical Deduction,” in M.E. Szabo (ed.) The Collected Papers of
Gerhard Gentzen, Amsterdam: North-Holland, 1969, pp. 68–131.
Gödel, K. 1930. “The Completeness of the Axioms of the Functional Calculus of Logic”, in Collected Works,
Vol. I, Feferman et al. (eds.), New York: Oxford University Press, 1986, pp. 103–123.
Gómez-Torrente, M. (2002) “The Problem of Logical Constants,” The Bulletin of Symbolic Logic 8: 1–37.
Hanson, W.H. (1997) “The Concept of Logical Consequence,” Philosophical Review 106: 365–409.
Henkin, L. (1959) “Some Remarks on Infinitely Long Formulas”, in Finitistic Methods: Proceedings of the
Symposium on Foundations of Mathematics, Warsaw: P.W.N. and Pergamon Press, 1961, pp. 167–83.
Higginbotham, J. & R. May (1981) “Questions, Quantifiers and Crossing,” Linguistic Review 1: 41–79.
Hintikka, J. (1973) “Quantifiers vs. Quantification Theory,” Dialectica 27: 329–58.
Hintikka, J. & G. Sandu (1989) “Informational Independence as a Semantic Phenomenon,” in J.E. Fenstad
et al. (eds.), Logic, Methodology, and Philosophy of Science, Amsterdam: Elsevier, pp. 571–89.
—— (2001) “Aspects of Compositionality,” Journal of Logic, Language, and Information 10: 49–61.
Kamp, H. (1981) “A Theory of Truth & Semantic Representation,”in J. Groenendijk et al. (eds.), Truth,
Interpretation, and Information, Dordrecht: Foris, 1984, pp. 1–41.
Kant, I. (1981/7) Critique of Pure Reason, Tr. N. Kemp Smith, London: Macmillan, 1929.
Keenan, E.L. (1987) “Unreducible n-ary Quantifiers in Natural Language,” in Gärdenfors 1987: 109–50.
—— (1992) “Beyond the Frege Boundary,” Linguistics and Philosophy 15: 199–221.
Keenan, E.L. & J. Stavi (1986) “A Semantic Characterization of Natural Language Determiners,” Linguistics
& Philosophy 9: 253–329.
Keisler, H.J. (1970) “Logic with the Quantifier ‘There Exist Uncountably Many’,” Annals of Mathematical
Logic 1: 1–93.
Lindström, P. (1966) “First Order Predicate Logic with Generalized Quantifiers,” Theoria 32: 186–95.
—— (1969) “On Extensions of Elementary Logic,” Theoria 35: 1–11.
—— (1974) “On Characterizing Elementary Logic,” in S. Stenlund (ed.), Logical Theory and Semantic Analy-
sis, Dordrecht: D. Reidel, pp. 129–46.
Marcus, R.B. (1972) “Quantification and Ontology,” Noûs 6: 240–50.
May, R. (1985) Logical Form: Its Structure and Derivation, Cambridge, MA: MIT Press.
McGee, V. (1996) “Logical Operations,” Journal of Philosophical Logic 25: 567–80.
Montague, R. (1973) “The Proper Treatment of Quantification in Ordinary English,” in Formal Philosophy:
Selected Papers, R.H. Thomason (ed.), New Haven: Yale University Press.
Mostowski, A. (1957) “On a Generalization of Quantifiers,” Fundamenta Mathematicae 44: 12–36.
Neale, S. (1990) Descriptions, Cambridge, MA: MIT Press.
Partee, B.H. (1987) “Noun Phrase Interpretation and Type-Shifting Principles,” in J. Groenendijk et al.
(eds.), Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, Dordrecht: Foris,
pp. 115–143.
Peters, S. & D. Westerståhl (2006) Quantifiers in Languag eand Logic, Oxford: Oxford University Press.
Priest, G. & R. Routley (1989) “The Philosophical Significance and Inevitability of Paraconsistency,”
in G. Priest et al. (eds.) Paraconsistent Logic Essays on the Inconsistent, Munich: Philosophia Verlag, pp.
483–539.
Quine, W.V. (1970) Philosophy of Logic, 2nd edition, Cambridge, MA: Harvard University Press 1986.
Rescher, N. (1962) “Plurality-Quantification,” Abstract, Journal of Symbolic Logic 27: 373–4.
Sher, G. (1990) “Ways of Branching Quantifiers,” Linguistics and Philosophy 13: 393–422.
—— (1991) The Bounds of Logic: A Generalized Viewpoint, Cambridge, MA: MIT Press.
—— (1996) “Did Tarski Commit ‘Tarski’s Fallacy’?,” Journal of Symbolic Logic 61: 653–86.
—— (1997) “Partially-Ordered (Branching) Generalized Quantifiers: A General Definition,” Journal of
Philosophical Logic 26: 1–43.
—— (2001) “The Formal-Structural View of Logical Consequence,” Philosophical Review 110: 241–61.
—— (2003) “A Characterization of Logical Constants Is Possible,” Theoria 18: 189–97.

594
 LO G I CA L Q U A N T I F I E R S

—— (2008) “Tarski’s Thesis,” in D. Patterson (ed.), New Essays on Tarski and Philosophy, Oxford: Oxford
University Press, pp. 300–39.
Tarski, A. (1933) “The Concept of Truth in Formalized Languages”. In J. Corcoran (ed.) Logic, Semantics,
Metamathematics, 2nd ed. Indianapolis: Hackett, 1983, pp. 152–278.
—— (1966) “What Are Logical Notions?” History and Philosophy of Logic 7 (1986): 143–54.
Van Benthem, J. (1983) “Determiners and Logic,” Linguistics and Philosophy 6: 447–78.
—— (1989) “Polyadic Quantifiers”, Linguistic sand Philosophy 12: 437–64.
Westerståhl, D. (1985) “Logical Constants in Quantifier Languages,” Linguistics and Philosophy 8: 387–413.
—— (1987) “Branching Generalized Quantifiers and Natural Languages,” in Gärdenfors, pp. 269–98.

595
 5.3
THE L OGIC OF T IM E
AND TENSE
Antony Galton

1 Introduction
Classical logic makes no special provision for handling the expression of time and tempo-
ral phenomena; yet time undeniably plays an important part in language, surfacing both in
the grammar (as tense and aspect) and in the lexicon (notably in the form of adverbs such
as “now”, “then”, “always”, and prepositions and connectives such as “before”, “after”,
“since”, “until”). For logic to be of service to the analysis of language, ways must be found
of expressing the formal content of such linguistic items within a logical framework.
We may distinguish two broadly different approaches to increasing the expressive
range of a logical system; the difference between them relates to the distinction between
the logical and nonlogical elements of a logical language. In standard first-order predi-
cate logic, the logical vocabulary comprises the Boolean connectives š›™and ,
the quantifiers and , and, for first-order logic with identity, the symbol =. These sym-
bols have fixed interpretations: that is, their meanings are determined by the definition
of the logic, not by the user. By contrast, the interpretation of individual constants,
function symbols, and predicate symbols is not determined in the logic itself; rather,
the user is free to specify their interpretations, subject only to the constraint that they
adhere to the correct form laid down by the logic: An individual constant must be inter-
preted as denoting an element of the domain of discourse , an n-ary function symbol as
denoting a function of type n  , and an n-ary predicate symbol as denoting a set of
n-tuples of domain elements. Thus the formula x(P(x)  y(Q(y)šR(x, y))) might
be understood as saying that every positive integer has a prime factor, that every human
being has a mother, or that every country has a capital city (among innumerable other
possibilities), depending on how the nonlogical symbols P, Q, and R are interpreted;
but throughout all the possible interpretations, the logical symbols , , š, and  are
interpreted in exactly the same way.
If we wish to use a logic to handle some new range of expressive possibilities, then
there are, in general, two approaches we may adopt:

1. To adhere to the original logic but capture the new range of meanings by extending
the interpretations of the nonlogical vocabulary.
2. To modify the logic itself by introducing new logical constants whose fixed mean-
ings relate to the desired additional expressive power.
 TH E LO G I C O F TI ME A N D T E N S E

The latter approach is appropriate if the new meanings to be covered are sufficiently
universal in their applicability, featuring in a wide range of specific subject domains.
This accords with the traditional view of logic as transcendental, that is, transcending
any specific subject matter. The first approach, by contrast, suggests itself when the
new meanings clearly belong to some specific domain of discourse: If, for example, we
wished to investigate the logic of kinship relations, it would be natural to introduce
predicates or functions whose interpretations correspond to natural language terms such
as “mother”, “brother”, or “cousin”, without seeking to modify the inherent expressive
and inferential apparatus of the formal logical system employed.
In the case of time and temporal phenomena, the choice between the two approaches
is finely balanced. On the one hand, temporality is pervasive, permeating much of our
language, reflecting the “universal” character of time as the medium in which events
and processes occur. On the other hand, much of the historical development of logic,
especially in the last two centuries, is inseparable from its use as a medium for formal-
izing mathematics, a domain of discourse from which the temporal element is conspicu-
ously lacking. At least from a Platonistic perspective, mathematical truths are eternal
and unchanging: our knowledge and understanding of those truths may change, but for
a Platonist these are in no way constitutive of the truths themselves. From this point
of view, time and temporality may seem to lack the degree of universality required for a
topic to be enshrined in the fixed logical vocabulary of a logic.
As a result of this finely-balanced choice, the historical development of temporal logic
has followed two different approaches, each with its own adherents, and a considerable
body of work devoted to exploring their relative merits and the relations between them.
Under the first approach, the expression of temporality is accomplished through the use
of a first-order language whose nonlogical vocabulary includes elements which are to
be given a temporal interpretation; this approach encompasses a range of rather differ-
ent possibilities, which are discussed in section 2. The second, more radical, approach
goes beyond first-order logic by introducing modal-type operators as the primary vehicle
for the expression of temporality. This approach originated with tense logic, which is
described in section 3.

2 First-order Approaches to the Logic of Time

The task here is to express the temporal content of propositions using a logic which
does not contain resources specifically designed for that purpose—in particular, stand-
ard first-order predicate calculus. The most obvious way of doing this, the method of
temporal arguments (MTA), is described in the first subsection below; subsequent subsec-
tions cover alternatives to MTA, in which content normally thought of as propositional
in nature is reified.

2.1 The Method of Temporal Arguments

In the method of temporal arguments, any atomic proposition whose truth value can vary
with time is equipped with an additional argument-place to hold a reference to the time
at which, in any particular occurrence, it is to be evaluated. To represent the sentence “It
is raining,” for example, instead of a simple 0-ary predicate Raining, we could use a 1-ary
predicate Raining(–), where the argument place is to be filled by a term denoting a time,
as e.g., Raining(t1), where t1 refers to, say, 11.45 p.m. (GMT) on March 30 2010.

597
 A N TO N Y G A LT O N

Using this style of analysis, one can express a wide range of temporal propositions,
for example:

“It is raining whenever we are t(In(we, scotland, t)  Raining(t))

in Scotland”
“John was in London all last t(During(t, lastWeek)  In(john, london, t))
week”
“John kissed Mary last Tuesday” t(During(t, lastTuesday) š Kiss (john,
mary, t))
“We are going to Scotland t(During(t, july) š GoTo(we, scotland, t))
in July”
“John has never kissed Mary” ™t(Before(t, now) š Kiss(john, mary, t))

These are “naive” renderings, with no attempt at capturing delicate semantic nuances,
and the reader will doubtless find cause to question them: The point, however, is to give
an initial indication of the general style of analysis to be expected from MTA. Some of
the nuances are discussed below.
The chief elements of this kind of representation are:

1. Explicit references to specific times, represented by individual constants (lastWeek,

lastTuesday, july, now). As presented, these appear to be indexical (since, e.g.,
the interval referred to by “last week” depends on when it is uttered), but in prin-
ciple they could be replaced by non-indexical expressions such as “the week begin-
ning April 12, 2010”—although it is a question of some philosophical moment
whether temporal discourse can be completely purged of indexicality without loss
of content.
2. Generalized temporal reference via the quantified variable t.
3. Relations between times, as expressed by the predicates During and Before.
4. “Thematic” content, describing specific states of affairs or events, e..g., Kiss(john,
mary, –).

Formally, we have a first-order language whose nonlogical vocabulary comprises sets

T of individual constants denoting times, R of temporal relations, and P of monadic
predicates over times. The predicates in P may have additional nontemporal structure
(e.g., Kiss(john, mary, –) is further decomposed into a ternary predicate Kiss(–,–,–) and
individual constants john and mary). Reasoning with such a language is supported by
axioms specifying properties of time. For example, the temporal precedence relation
might be stipulated to be transitive:

tt't"(Before(t, t') š Before(t', t")  Before(t, t"))

and irreflexive:

t™Before(t, t).

Should these individual times be instants or intervals? Technically, instants are easier
to handle. If the flow of time is linear, distinct instants t and tc must stand in one of two
relations: either t precedes tc or tc precedes t. By contrast, two distinct intervals can stand

598
 TH E LO G I C O F TI ME A N D T E N S E

in any of twelve different relations, depending on how the endpoints of one are related
to the endpoints of the other. Since an interval is determined by the two instants mark-
ing its endpoints, any representation that uses intervals can, in principle, be converted
into one which only uses instants. For conformity with everyday language, however, this
may not be the most sensible choice. A common expression of temporality is the loca-
tion of an event within an interval: “I went to London yesterday”, “His father died in
1998”, and so on. Any formalism lacking direct representation of intervals must resort
to paraphrase to express such facts. Indeed, it has been suggested that the notion of an
instant does not correspond to any empirically observable phenomenon, and that all
references to individual times should be to intervals, which derive their empirical reality
from being the times at which events happen (Allen 1984).
How is the thematic content of a temporal proposition related to the times to which it
is ascribed? If t is an instant, then In(john, london, t) says that John is in London at that
instant. Since being in London necessarily persists over an interval, that instant must
form part of some interval throughout which John is in London. If, on the other hand,
t is an interval, then In(john, london, t) could be understood as saying either that John
is in London throughout t, or that he is there at some time during t. Ordinary language
does not come down decisively one way or the other. John’s being in London is a state;
further complications arise when we consider events, as in Kiss(john, mary, t). Com-
pare “John kissed Mary yesterday” and “John kissed Mary at midnight”. In the former,
the kiss takes place at a time falling within the interval designated yesterday, without
being coextensive with it. In the latter, the kiss occupies a—presumably short—interval
which contains, or starts at, midnight.
To use the formal apparatus of MTA to represent the meanings of ordinary language
sentences, we must be sensitive to issues such as these. The flexibility of natural lan-
guage makes this enterprise difficult: there is seldom any tight one-to-one correspond-
ence between the meanings of natural language sentences and the meanings expressible
within a precisely defined logical system. Hence it may not be appropriate to try to
express all linguistic meanings in logical form; the virtue of logical analysis is in provid-
ing a means for expressing precisely, within a mathematically watertight framework, a
range of possible meanings, in order to provide a set of controlled exemplars with refer-
ence to which the vagaries of natural language can be discussed.

2.2 Reified Temporal Logics I: Type-reification

For the purposes of reasoning about temporal information, one drawback of MTA is
that it does not allow one to express in first-order form general statements such as “An
effect cannot precede its cause” or “A state of affairs that obtains throughout the day
must obtain throughout the morning”, which play a part, at least implicitly, in everyday
temporal reasoning. The need to accommodate such statements has been keenly felt in
artificial intelligence (AI), where the task of automating reasoning forces one to tease out
many general assumptions underlying ordinary reasoning but not explicitly expressed.
A widely-advocated solution within AI is the use of a “reified” notation for expressing
temporal facts. Where MTA writes, e.g., Kiss(john, mary, t), the reified system of (Allen
1984), has Occurs(kiss(john, mary), t). Here the thematic content expressed in MTA
by the predicate Kiss(john, mary, _) is expressed by means of a term kiss(john, mary)
referring to an event-type, here conceived as an entity in its own right (hence “reified”),

599
 A N TO N Y G A LT O N

to which is ascribed the property of occurring at various times. To express a causal rela-
tion ship between two events, Allen writes Ecause(e, t, e', t'), meaning that an occur-
rence of event-type e at time t caused an occurrence of event-type e' at t'. If t and t' are
instants, then the rule that an effect cannot precede its cause could be expressed by

ee'tt'(Ecause(e, t, e', t') ™Before(t', t)).

The reified approach allows explicit distinctions between states, processes, and
events. For Allen, an event forms a unitary whole, no proper part of which can be an
event of the same type. This is consistent with the categories of “accomplishment” and
“achievement” in the well-known classification of (Vendler 1967). In reified logic, this
may be expressed by the axiom:

ett'(Event(e) šOccurs(e, t) šProperPart(t', t)  ™Occurs(e, t')).

For this to make sense, Occurs(e, t) must mean that an occurrence of event-type e is
exactly coextensive in time with the interval t. This in turn means that our example
“John kissed Mary last Tuesday” must be expressed as

t(ProperPart(t, lastTuesday) šOccurs(kiss(john, mary), t)).

This could be abbreviated to

OccursIn(kiss(john, mary), lastTuesday),

as advocated in (Galton 1990).

In contrast to events, states do not form discrete individual occurrences, instead
exhibiting the property of dissectivity:

stt'(State(s) šHolds(s, t) šProperPart(t', t)  Holds(s, t')).

For example, if it was raining all day yesterday, then it was raining at lunchtime yester-
day. Note the use of “Holds” rather than “Occurs”: for Allen, the ways in which states
and events are realized at times are sufficiently distinct to warrant the use of different
predicates to express them.

2.3 Reified Temporal Logics II: Token-reification

Although type-reification provides useful expressive power, it may be objected to on
philosophical grounds, since it seems to accord event and state types first-class status
as elements of one’s ontology. An alternative form of reification, which countenances
individual event occurrences but not general event types, is sometimes proposed instead.
This is known as token-reification.
Token-reification can be traced back to Davidson’s well-known paper on the logical
form of “action sentences” (Davidson 1967). This addressed the problem of the apparent
“variable polyadicity” of verbs. To give a simple example, the English verb “sing” can
be used intransitively, as in “John sang”, or transitively, as in “John sang Rule Britannia”.

600
 TH E LO G I C O F TI ME A N D T E N S E

Naïvely, one might try representing these by Sing(john) and Sing(john, ruleBritannia).
In logic, however, a predicate has a fixed number of argument places (its polyadicity),
so the two occurrences of Sing here must represent different predicates—Sing1 (john)
and Sing2 (john, ruleBritannia), say. The problem then arises of accounting for the logi-
cal relation between the two predicates: specifically, that Sing2 (john, ruleBritannia)
implies Sing1 (john).
In this case there is a simple solution: Any singing must be a singing of something
(even if only a random sequence of notes), so Sing1 (x) could be defined to mean ySing2
(x, y). But now consider a sequence such as

John sang.
John sang Rule Britannia.
John sang Rule Britannia in the bathroom.
John sang Rule Britannia in the bathroom at midnight.

where each sentence is implied by the ones following it. One might suggest using a four-
place predicate Sing4, representing the sentences by the formulae

xyzSing4(john, x, y, z)
yzSing4 (john, ruleBritannia, y, z)
zSing4 (john, ruleBritannia, bathroom, z)
Sing4 (john, ruleBritannia, bathroom, midnight)

but this is unsatisfactory since there may be no limit to the number of extra arguments
needed (e.g., consider “John sang Rule Britannia loudly in the key of E in the bathroom
at midnight”), making it impossible to fix the polyadicity of predicates in advance as
logic requires.
Davidson’s suggestion was to treat these sentences as referring to a particular indi-
vidual event, to which are ascribed various properties: that it was a singing by John of
Rule Britannia, its location was the bathroom, its time was midnight, it was loud, its key
was E, etc. Using an existentially quantified variable “e” to denote this event, we can
then express the last example using the formula

e(Sing(john, ruleBritannia, e) šLocation(e, bathroom) šTime(e, midnight) š

Loud(e) šKey(e, eFlat)).

Formulae corresponding to the shorter sentences above then follow as logical conse-
quences from this by standard first-order logic. Since “e” here represents an individual
occurrence, or event-token, this is an example of token-reification as defined earlier.
In computer science, a similar method was proposed by Kowalski & Sergot (1986),
who introduced the “event calculus” as a system for reasoning about events in a logic-
based framework, with intended application to database updates and discourse represen-
tation. Vila & Reichgelt (1996) also advocated a form of token-reification for temporal
reasoning in AI.
To round off the discussion of first-order approaches to time, consider how the sentence
“Napoleon invaded Russia in 1812” might be represented in the three approaches:

• Method of Temporal Arguments: Invade(napoleon, russia, 1812).

601
 A N TO N Y G A LT O N

• Type-reification: OccursIn(invade(napoleon, russia), 1812).

• Token-reification: e(Invade(napoleon, russia, e) šOccursIn(e, 1812)).

For further discussion and comparison of these methods and their variants, see (Galton
2006).

3 Modal Approaches to the Logic of Time

From a linguistic point of view, one drawback of the first-order approaches is that the
relationship between the expression of propositions in natural language and their logi-
cal representations can be very indirect. The sentence “When John arrived, Mary had
already left”, for example, might be represented using MTA as

tt'(Arrive(john, t) šLeave(mary, t') šPast(t) šBefore(t', t)),

or using token-reification as

ee' tt'(Arrive(john, e) š Leave(mary, e') š Time(e, t) š Time(e, e') š

Past(t) šBefore(t', t)).

To offer either of these formulae as an analysis of the English sentence is to claim that
it refers, implicitly, to unspecified times t and t', and, in the reified analysis, to indi-
vidual events e and e'. Thus these analyses impute to the English sentence a significant
ontological load. Conversely, some elements expressed directly in the English sentence,
notably the simple past tense in “arrived”, the pluperfect tense in “had left”, and the
temporal relation indicated by “when”, receive no direct expression in the logic, their
senses being distributed across different elements of the representation. For some pur-
poses, a logical representation which more directly mirrors the forms of ordinary lan-
guage, such as tense, might be preferred. This is one of the motivations underlying the
modal approach to time.

3.1 Tense Logic

Tense logic was the creation of Arthur Prior and arose directly from modal logic via
a proposal to reduce modal notions to temporal ones, defining the possible as “what
is or will be”, and the necessary as “what is and always will be”, which Prior attributed
to the third-century bce logician Diodorus Cronus. If modality is defined in this
way, what existing system of modal logic is obtained? To determine this, we need a
logic of futurity, and this led directly to the construction of a “calculus of tenses” (Prior
1967).
In tense logic, the standard logical apparatus is supplemented by tense operators, which
are applied to propositional formulae to produce new formulae. Prior’s original logic has
four operators, with intended meanings as follows:

PI: It has been the case that I

HI: It has always been the case that I
FI: It will be the case that I
GI: It will always be the case that I

602
 TH E LO G I C O F TI ME A N D T E N S E

where I stands for any proposition. The operators P and F correspond roughly to the
simple past and future tenses of natural language: if I stands for “It is raining”, PI would
stand for “It was raining”, and FI for “It will be raining”. Because these operators form
propositions out of propositions, they can be iterated, leading to “compound tenses”
such as PFI (“It was going to rain”), FPI (“It will have rained”), and PPI (“It had
rained”). Using this formalism, our example “When John arrived, Mary had already
left” might be represented as P(Arrive(john) š PLeave(mary)), where the simple past
tense of “John arrived” corresponds to the first P operator, while the pluperfect tense of
“Mary had left” corresponds to the nesting of the second P operator within the scope
of the first.
The relationship between the “weak” P and F operators and their “strong” counter-
parts H and G is exactly analogous to that between the weak and strong modal operators
‘ and …. Thus PI may be defined as ™H™I, since to say that it has been true that I
is precisely to deny that it has always been false that I, and likewise FI is equivalent to
™G™I. Conversely, GI and HI are equivalent to ™P™I and ™F™I respectively. Thus
in principle we can dispense with P and F and express everything in terms of H and G,
or vice versa.
Prior’s approach to formalizing the logic of tenses was axiomatic: the meanings of the
operators were not given explicitly but implicitly through the use of axioms, that is,
tense-logical formulae postulated as necessarily true. A minimal set of axioms, making
no presuppositions about the structure of time, is:

1. I GP I
2. I HF I
3. H(I )  (HI H)
4. G(I )  (GI  G)

Axiom 1 asserts that whatever is true now will always have been true (the past is irrevo-
cable), and axiom 2 asserts that whatever is true now has always been going to be true.
The latter may smack of fatalism, but arguably, in accepting it, we are not committing
ourselves to asserting that present facts were inevitable, merely that, as things turned
out, they became true. Philosophically, this is already highly problematic, bearing as it
does on the problem of future contingents, which has exercized philosophers at least
since Aristotle. The remaining axioms are less contentious: Axiom 3 states that if I has
always implied , then if I has always been true, so has ; and Axiom 4 the analogous
statement for the future.
To use these axioms in reasoning, one requires also some rules of inference. In addi-
tion to a standard set of rules for classical logic, one requires rules (RH and RG) which
say that if I can be derived from the axioms, then so can HI and GI: in other words,
having proved some formula, one can assert that it always has been and will be true.
The system thereby obtained is called minimal tense logic, designated Kt. An example of
a theorem of Kt is

I šP  P( šFI),

e.g., if it is raining and was sunny, then it has been the case that it was sunny and would
be raining. (For a proof of this formula, see (Galton 1984, p.43).)
Many plausible tense-logical formulae cannot be derived as theorems of Kt, and it is

603
 A N TO N Y G A LT O N

natural to ask which additional formulae should be postulated as axioms. The formula
FFI  FI, for example, says that whatever will be future is already future. This seems
sensible since what will count as future from the perspective of some future time must
surely form part of what counts as future now. Although this seems obvious, it makes a
substantive assumption about the organization of time: that the relation “is in the future
of” is transitive. The converse formula, FI  FFI, says that anything currently in the
future will still be in the future at some future time. This also makes a substantive assump-
tion about time. For suppose time t is immediately followed by tc, with no other times
between them; if I is true at tc but at no other times, then at t, FI is true but FFI is false,
falsifying FI FFI. Hence to posit this formula as an axiom is, implicitly, to rule out a
model of time in which one time can immediately follow another—and thus to assume a
dense model, in which between any distinct times there is another (and hence, by itera-
tion, infinitely many). Both these formulae, then, make substantive assumptions about
time, and as such go beyond Kt.
Quite early in the development of tense logic, serious expressive limitations became
apparent. For example, Prior’s tense operators cannot be used to express sentences such
as “Mary will be unhappy until John arrives” or “John has been unhappy since Mary
left.” For this reason, Kamp (1968) introduced two new binary operators S and U, whose
intended meanings are as follows:

S I : It has been the case that I , and between then and now it has always been
the case that .
UI : It will be the case that I , and between now and then it will always be
the case that 

Informally, these can be read as “Since I , ” and “Until I , ”, and the two sentences
above can be expressed as UArrive(john)Unhappy(mary) and SLeave(mary)Unhappy
(john). Note, however, that whereas the former formula implies FArrive(john), the Eng-
lish sentence does not imply that John will arrive; a more satisfactory rendering of that
sentence might be UArrive(john)Unhappy(mary) › GUnhappy(mary). Having intro-
duced S and U, we can dispense with Prior’s operators, since PI and FI can be defined
as SI (I ›™I ) and UI (I ›™I ) respectively, with G and H defined in terms of P and
F in the usual way.

4 Relationships Between the Approaches

4.1 Model Theory for Tense Logic
Although the two approaches to the logic of time may appear very different, there are
important links between them. In presenting the operators used in the modal approach,
reference was made to their “intended meanings”, without any indication of how opera-
tor and meaning are linked. In general terms, how are meanings assigned to logical for-
mulae? Under the axiomatic approach, this is done implicitly, by specifying the logical
relationships between the formulae (i.e., what can be deduced from what, and what is
true unconditionally). The alternative is to assign meanings to formulae explicitly by
specifying how they are to be interpreted, as was done informally above. This may be
made formal through the discipline of formal semantics, typically in the form of a model
theory mapping elements of the logic to abstract structures which serve as models for

604
 TH E LO G I C O F TI ME A N D T E N S E

real-world situations represented by the formulae.

We earlier observed that the formula FFI  FI , if asserted as an axiom, implicitly
presupposes that the relation “is in the future of” is transitive. To make such statements
explicit, we interpret formulae with respect to a temporal frame T comprising a set T of
individual times (assumed to be instants) ordered by a precedence relation : thus t  tc
means that time t precedes time tc. (Note that we use “t” to refer to a time in a temporal
frame, as distinct from “t” used earlier as a temporal variable in a first-order language.)
Temporal frames are directly analogous to the frames used in modal logic. To evaluate
formulae as true or false at different times, one must specify, for each atomic formula I ,
its truth value V(I ,t) at each time t in the temporal frame under consideration. This
is called a valuation over the frame T. An interpretation I=(T,V) of the tense-logical
language consists of a temporal frame together with a valuation. The meaning of a tense
operator is specified by stating the conditions under which a formula containing that
operator is satisfied (i.e., evaluated as true) at a time by any given interpretation. For the
two future-tense operators, this is done as follows (and analogously for the past):

• I satisfies FI at t if and only if, for some tc T, t  tc and I satisfies I at tc.
• I satisfies GI at t if and only if, for every tc T, if t  tc then I satisfies I at tc.

We may also speak of a temporal frame satisfying a formula, meaning that every inter-
pretation over that frame satisfies the formula at every time. Given these definitions,
the theorems of minimal tense logic Kt are precisely those formulae satisfied in every
temporal frame, thus justifying the claim that Kt makes no specific assumptions about
the temporal ordering.
By contrast, the formula FFI  FI , which is not a theorem of Kt, is satisfied by a
temporal frame T if and only if its precedence relation is transitive, that is:

ttct"(t  tc štc  t"  t  t").

To prove this, suppose first that  is transitive, and suppose interpretation I satisfies FFI
at t. This means there are times tc, t"  T such that t  tc, tc  t", and I satisfies I at t".
Since  is transitive, tc  t˝, so I satisfies FI , and therefore also FFI  FI , at t. Hence
every interpretation over the frame satisfies this formula at all times, so the frame itself
satisfies it. Conversely, suppose  is not transitive: then there are times tc, t" such that
t  tc, tc,  t", but t  t". Define a valuation V over T such that I is true at t" and at no
other times. Then the interpretation (T,V ) satisfies FFI at t (since it satisfies I at a
time in the future of a time in the future of t), but it does not satisfy FI at t (since the
only time I is satisfied is not in the future of t). Hence it does not satisfy FFI  FI at
t, so the frame does not satisfy this formula either.
There are many such cases where a tense-logical formula I is satisfied by a temporal
frame T if and only if the frame satisfies some specified first-order condition. Another
example is the formula FI  FFI , which is satisfied by dense temporal frames, satisfying
the first-order condition

ttc(t  tc t"(t  t" št"  tc)).

The systematic investigation of these correspondences between tense-logical (and

605
 A N TO N Y G A LT O N

more generally modal) formulae and the first-order conditions on frames which satisfy
them is called correspondence theory; for details, see (Van Benthem 1991). Not all tense-
logical formulae correspond to first-order conditions in this way, and not all first-order
conditions on temporal frames correspond to tense-logical formulae. No tense-logical
formula, for example, is valid on all and only the irreflexive frames, which satisfy the
condition t™(t  t).
Using the model theory, we can systematically translate tense-logical formulae into
first-order formulae, in the style of either MTA or a reified system. Thus PRaining (i.e.,
“It has been raining”) translates as t(Before(t, now) šRaining(t)) or t(Before(t, now)
šHolds(raining, t)).

4.2 Hybrid Tense Logic

The correspondences described in the preceding section suggest one way in which
the two approaches to the logic of time might be unified: the first-order approach
provides a way of specifying the semantics of formulae produced under the modal
approach, enabling tense-logical formulae to be translated into first-order formulae.
This does not in itself overcome certain apparent expressive limitations of the modal
approach, and in this section we examine how a closer rapprochement between the
two approaches might be achieved, enabling at least some first-order formulae to be
expressed in modal form.
A deficiency of tense logic, in either Prior’s original form or Kamp’s extended ver-
sion, is that it does not lend itself readily to expressing sentences containing explicit
reference to times, such as “John kissed Mary at midnight”. One possibility is to treat
“midnight” not as an individual term, but as a statement, expressing the proposition “It
is midnight.” Then the formula P(Midnight šKiss(john, mary)) states that at some past
time it was midnight and John kissed Mary. If Midnight is understood in a generic sense
as true whenever it is midnight, this does not locate the kiss at a specific midnight occa-
sion; but if instead we introduce a proposition Midnight20100412, which is only true
at midnight on April 12 ,2010, then we can write P(Midnight20100412 š Kiss(john,
mary)) to say that John kissed Mary at that time.
This idea lies at the heart of hybrid tense logic (HTL), which marries the spirit of tense
logic with the explicit temporal references of the first-order approach. In addition to
“ordinary” propositions, HTL includes special propositions called nominals which can
only be evaluated as true at exactly one time. A nominal such as Midnight20100412
may be thought of as “naming” the unique time at which it is true. In what follows we
use t, tc t" etc. for nominals, and I, , etc. for general propositions.
To assert that I holds at the time designated by nominal t, with no assumption as to
whether t is past, present, or future, one can put P(t šI) › (t šI) ›F(t šI), which is
abbreviated to @t I. We can regard @t as a kind of tense operator, although unlike the
usual tense operators it commutes with negation, i.e., ™@tI is equivalent to @t ™I. To
specify the semantics of @ in the model-theoretic style, it is convenient to write g(t) to
denote the unique time at which t is true. Then we can put

• I satisfies @tI at t if and only if I satisfies I at g(t).

Superficially, @tI looks like a syntactic variant of a type-reified formula of the form
Holds(s, t) (see section 2.2). The parallel is not exact, however, since whereas I is a

606
 TH E LO G I C O F TI ME A N D T E N S E

propositional formula, s is a term denoting a state. Hence whereas operators of the form
@t can be iterated to yield formulae such as @t @t'I, Holds cannot: Holds(Holds(s, t'), t)
is ill-formed. This difference need not be serious, however, since from the semantics of
@, the formula @t @t'I is equivalent to @t'I, so all but the last of a string of @ operators
can be dropped without change of meaning.
The introduction of nominals changes the complexion of tense logics considerably.
For example, in ordinary tense logic the formula F(I š) šF(I šF)  F( šF) is not
valid, but if I is an HTL nominal then it is. Again, in the model theory, the class of all
irreflexive frames is exactly characterized by the hybrid formula t  ™Ft, but not by any
ordinary tense-logical formula.
There is no space here to discuss HTL further. It is the temporal form of hybrid logic,
which is related to modal logic just as HTL is related to tense logic. For further details,
see (Areces, Blackburn & Marx 2001).

5 Concluding Remarks
In the space available it has not been possible to do more than outline the main ways
formal logic has been used for representing time and tense. The logic of time is of inter-
est to researchers in a number of different areas, notably philosophy, linguistics, compu-
ter science, and mathematics, and therefore it has been approached from many different
points of view. See the references and further reading for some pointers.

Related Topics
2.8 Event Semantics
5.4. Modal Logic.

References
Allen, J. (1984), ‘Towards a general theory of action and time,’ Artificial Intelligence 23, 123–54.
Areces, C., Blackburn, P. & Marx, M. (2001), ‘Hybrid logics: Characterization, interpolation and complex-
ity’, Journal of Symbolic Logic 66(3), 977–1010.
Davidson, D. (1967), The logical form of action sentences, in N. Rescher, ed., The Logic of Decision and
Action, University of Pittsburgh Press, Pittsburgh, Reprinted in D. Davidson, Essays on Actions and Events
(Oxford: Clarendon Press, 1980).
Galton, A. P. (1984), The Logic of Aspect: an Axiomatic Approach, Clarendon Press, Oxford.
Galton, A. P. (1990), ‘A critical examination of Allen’s theory of action and time,’ Artificial Intelligence 42,
159–88.
Galton, A. P. (2006), ‘Operators vs arguments: The ins and outs of reification’, Synthese 150, 415–441.
Kamp, J. A. W. (1968), Tense Logic and the Theory of Linear Order, PhD thesis, University of California,
Los Angeles.
Kowalski, R. A. & Sergot, M. J. (1986), ‘A logic-based calculus of events,’ New Generation Computing 4,
67–95.
Prior, A. N. (1967), Past, Present, and Future, Clarendon Press, Oxford.
Van Benthem, J. F. A. K. (1991), The Logic of Time: A Model-Theoretic Investigation into the Varieties of
Temporal Ontology and Temporal Discourse, 2nd. edn, Kluwer Academic Publishers, Dordrecht.
Vendler, Z. (1967), Verbs and times, in Linguistics and Philosophy, Cornell University Press, Ithaca, chapter
4. Reprinted in Mani et al. (2005).
Vila, L. & Reichgelt, H. (1996), ‘The token reification approach to temporal reasoning’, Artificial Intelligence
83, 59–74.

607
 A N TO N Y G A LT O N

Further Reading
Fisher, M., Gabbay, D. & Vila, L., eds, (2005), Handbook of Temporal Reasoning in Artificial Intelligence, Else-
vier, Amsterdam. (An emphasis on formal treatments directed towards computational implementation.)
Mani, I., Pustejovsky, J. & Gaizauskas, R., eds, (2005), The Language of Time: A Reader, Oxford University
Press, Oxford. (A useful compendium of articles covering different approaches to handling time in natural
language.)
Øhrstrøm, P. & Hasle, P. F. V. (1995), Temporal Logic: From Ancient Ideas to Artificial Intelligence, Kluwer
Academic Publishers, Dordrecht. (Offers a useful historical perspective.)

608
 5.4
MODAL L OGIC AND
ITS AP P L ICAT ION T O
THE P HI L OSOPHY OF
L ANGUAGE
Kit Fine

Modal logic is the logic of possibility and necessity and of other such notions. It began,
as did logic in general, with Aristotle, in his theory of the ‘modal syllogism’; and vari-
ous notions and principles of modal logic were extensively discussed in the middle ages.
But the subject only came into its own at the beginning of the twentieth century (see
Goldblatt 2005 for an account of its recent history).
I begin by presenting some basic material on the possible worlds’ approach to modal
logic and then show how it relates to certain key topics in the philosophy of language.
For reasons of space, I have had to be very selective and, inevitably, a great deal of inter-
esting material has not been covered.

1 Language and logic

We begin by describing the language of sentential modal logic and the various systems
that can can be formulated by its means.

1.1 Language
The language L of classical (non-modal) sentential logic may be taken to consist of the
following symbols:

(i) the sentence letters p1, p2, p3 . . .;

(ii) the truth-functional connectives ~, ›, š and Š;
(iii) the parentheses (and).

1.2 Formulas
The formulas of L are generated by means of the following rules:
 KI T F I N E

(i) each sentence letter is a formula;

(ii) if A is a formula then so is ~A;
(iii) if A and B are formulas, then so are (A › B), (A › B), (A š B) and (A Š B).

The language L(…)of modal sentential logic is obtained by adding a modal operator …
for ‘necessity’ to the language L. There is one additional rule for generating formulas:

(iv) If A is a formula, then so is …A.

We adopt the following abbreviations:

¡A (‘possibly A’) for ~ …~A;

A o B (‘A strictly implies B’) for …(A Š B).

We follow standard notation in using ‘Š’ for material implication and in using ‘o’
for various stricter forms of implication. It should be noted that embeddings of … are
allowed. Thus ……A or …A Š ……A or …(…A Š A) Š …A are all formulas. A large
part of modal logic is concerned with the behavior of embedded modality.

1.3 Systems
A modal system is characterized by a set of axioms and rules. Say that a formula of L(…)
is truth-functionally valid if it is a substitution-instance of a tautology of L. The formula
(…A › ~…A), for example, is truth-functionally valid since it is a substitution-instance
of the tautology (p1 › ~p1). The axiom and rules for the minimal modal system K are
then as follows (where A and B are any formulas of L(…)):

Truth-functionality. All truth-functionally valid formulas;

Distribution. …(A Š B) Š (…A Š …B);
Modus Ponens. A, A Š B/ B;
Necessitation. A/…A.

The theorems of the system K are all those formulas that can be obtained from the axioms
by means of the rules. To be more exact, the set of K-theorems is the smallest set to con-
tain the axioms of Truth-functionality and Distribution and to be closed under the rules
of Modus Ponens and Necessitation (thus if A and A e B are in the set then so is B and if
is A is in the set then so is …A). We may write ‘Š A’ to indicate that A is a theorem of
K (and similarly for other systems of modal logic). What motivates the choice of K is that
it is the weakest system to which the semantic techniques described below will apply.

2 Semantics
We now describe the possible worlds semantics for modal logic.

2.1 Models
A K-model M is an ordered triple (W, R, M), where W is a non-empty set, R is a binary
relation on W (i.e. R Ž W × W), and M is a function taking each sentence-letter p1, p2,

610
 M O DA L LOG I C

p3 . . . into a subset of W. Intuitively, W is the set of ‘worlds’, relative to which formulas

are evaluated, R is an accessibility-relation on worlds, holding between worlds w and v
when v is relevant to the evaluation of a necessity formula at w, and M is a ‘valuation’
that specifies, for each sentence-letter p, the set of worlds M(p) at which p is true.
An example of a K-model is pictured below:

p0 p1 p2 p3 . . .
•o•o•o• . . .
0 1 2 3 ...

And it may be formally defined by letting:

W = {0, 1, 2, . . .};
R = {(m, n): n = m + 1}; and
M = {(pn, {n}): n = 0, 1, 2, . . .}.

Thus in this model, each sentence-letter pn is true at exactly the one world n and the
only world accessible from a given world m is its ‘successor’ n + 1. (We might think of
the worlds as ‘days’, starting with day 0 and continuing into the infinite future; acces-
sibility is the ‘tomorrow’ relation; and pn is sentence stating that it is day n.)

2.2 Truth
Given a K-model M =(W, R, M), we may define when a formula is true at a world of the
model. We use the notation ‘w = A’ to indicate that A is true at w (or ‘w = MA’ if we
wish to be explicit about the underlying model M). The definition is as follows:

M(i) w = p iff w M(p), for any sentence-letter p;

M(ii) w = ~B iff it not the case that w = B;
M(iii) w = B › C iff w = B or w = C (and similarly for š and Š);
M(iv) w = …B iff v = B whenever wRv.

It is readily determined on the basis of these clauses and the abbreviations that have
been laid down that:

M(v) w = ¡B iff v = B for some v for which wRv; and

M(vi) w =B oC iff v = C whenever wRv and v = B.

A possibility formula ¡B is true at a world just in case the embedded formula B is true at
some accessible world; and a strict implication B oC is true at a world just in case the
consequent C is true in any accessible world in which the antecedent B is true.
Let us illustrate with the example-model above (in which we might think of ‘…’ as
meaning tomorrow). The sentence-letter pn will be true at the world n (i.e. n = pn) given
that n M(pn). Since n–1Rn for n > 0, it follows that ¡pn is true at n-1 (i.e. n-1 = ¡pn).
Use ¡m for ¡¡ . . . ¡ (m times). It then follows, for n > m, that ¡m pn is true at n–m (i.e.
n–m = ¡m pn). The reader may readily establish the converse, namely that k = ¡m pn only
if k = n–m; and so ¡m pn Š ¡l pn will be false at the world n–m whenever l zm.

611
 KI T F I N E

2.3 Validity
We say that a formula A is true in a model M = (W, R, M) if it is true in every world
w W of the model; and we say that a formula A is K-valid if it is true in every model.
We write = KA to indicate that the formula A is K-valid. We may show, for example,
that any instance …(A Š B) Š (…A Š …B) of the Distribution axiom is K-valid. For
take any model M = (W, R, M) and world w W and suppose that w = …(A ŠB) and
w = …A. We must then show that w = …B. But if w = …(A Š B), then v = B whenever
v = A and wRv; and if w = …A, then v = A whenever wRv. So v = B whenever wRv;
and consequently w = …B, as required. On the other hand, the formula ¡mpn Š ¡1pn,
for n > m and l zm, is not K-valid since it is not true at the world n–m in the example-
model above.

3 Soundness and completeness of K

It is a central result concerning K that its theorems and validities coincide. It is useful
to divide this result and its proof into two parts.

Theorem 1 (Soundness of K) Any K-theorem is K-valid.

Proof The proof is by induction on the generation of the K-theorems from the
K-axioms by means of the K-rules. We must first show that the K-axioms are
valid and that the K-rules preserve validity. We have already taken care of the
Distribution axioms; and the other cases are equally straightforward.
Theorem 2 (Completeness for K) Any K-valid formula is a K-theorem.

The proof of this result is more difficult than for Soundness. Let us briefly sketch one
method of proof, using a so-called ‘Henkin’ model.
Say that a set ' of formulas is K-consistent if  (defined as ~) is not K-deducible
from '; and say that a set ' of formulas is maximally K-consistent if it is K-consistent
but not properly contained in any K-consistent set. It is the maximally K-consistent sets
that will become the worlds of the canonical model. The following result is familiar from
the standard completeness results for classical sentential and quantificational logic:

Lemma 3 (Lindenbaum’s Lemma) Every K-consistent set of formulas ' is

included in a maximal K-consistent set of formulas *.
For ' set of formulas, let …(') = {A: …A  '}. We define the canonical K-model
MK = (WK, RK, MK) for K by:
(i) WK = {': ' is a maximally K-consistent set of formulas};
(ii) RK = {(', *): ', * WK and …(') Ž*};
(iii) MK = {(p, V): p is a sentence-letter and V = {' WK: p '}}.

We think of the worlds in the model MK as containing exactly those formulas that they
make true. This motivates clause (iii), since we want p to be true in ' just in case it is a
member of '. It can also be seen to motivate clause (ii), for if …A is true in the world '
then we certainly want A to be true at any accessible world *and it is hard to see what
further condition might reasonably be imposed on an accessible world.
It may be shown that the canonical model is faithful to our intentions:

612
 M O DA L LOG I C

Lemma 4 (Truth-lemma). Let ' be any maximally K-consistent set. Then:

A  ' iff ' = A (in the model MK).

Membership of a maximally consistent set coincides with truth in the canonical

model.
Completeness for K (theorem 2) can now be established. For suppose that A is not a
theorem of K. We may then find a maximally K-consistent set *containing ~A. By the
consistency of *, it follows that A * and so, by lemma 4, A is not true at world * in
the canonical model MK and hence A is not valid.
The proof of theorem 2 establishes two further results. Say that a set of formulas ' is
satisfiable if some model M and world w of the model, each formula A of ' is true at the
world in the model. Then:

Strong Completeness Every K-consistent set of formulas ' is satisfiable.

Compactness ' is satisfiable if each finite subset of ' is satisfiable.
For given that ' is K-consistent, we may obtain a maximally K-consistent
set containing ' and then proceed as before; and given that each finite subset
of ' is satisfiable, it will follow that ' is K-consistent and hence satisfiable by
Strong Completeness.

4 Some other systems

A similar result is readily established by similar means for various other systems of modal
logic. Let us give an illustrative example.

4.1 Completeness for T

Let the system T be the result of adding all formulas …A Š A as axioms to K (each such
axiom says that A is the case if A is necessarily the case). Say that a model M = (W, R,
M) is a T-model if its accessibility relation R is reflexive, i.e. if wRw whenever w  W; and
say that a formula is T-valid if it is true in every T-model. We then have the following
soundness result for T:

Theorem 5 (Soundness for T) Any T-theorem is T-valid.

Proof As for the soundness result for K. The one new case that needs to be considered
is the T-validity of …A ŠA. So take any model M = (W, R, M) and any world w W,
and suppose that …A is true at w. Then A is true at v whenever wRv. We need to show
that A is true at w. But R is reflexive; so wRw; and so A is true at w.
The proof of the results leading up to the completeness of K go through as before,
but using T-relative notions in place of K-relative notions. This is because the proofs
of these results depend only upon the presence of certain axioms and rules in K and not
upon their absence. Thus they will go through for any other system that contains those
axioms and rules. However, in order to establish the completeness of T, we also need to
show:

Lemma 6 The canonical model MT = (WT, RT, MT) for T is a T-model.

613
 KI T F I N E

Proof We need to show that RT is reflexive, i.e. that …(') Ž ' for any maximally T-
consistent set of formulas '. So suppose that ' T …A. Since …A Š A is a T-theorem,
it follows that ' T A; and so A ', as required.
We therefore obtain:

Theorem 7 (Completeness for T) Any T-valid formula is a T-theorem.

And, of course, Strong Completeness and Compactness can also be obtained in the
same way as before.

4.2 Other Extensions

A similar strategy can be employed for other systems. Suppose, for example, that we add
the axiom:

4 …A Š……A

to K (or to T). It can then be shown that a formula will be a theorem of the resulting
system K4 (or S4) just in case it is true at all worlds of all models whose accessibility
relation is transitive (or reflexive and transitive).
A good number of more general completeness results can be established but the reader
should be warned that not all modal logics are complete for some class of frames (see any
standard text, such as Hughes & Cresswell (2009), for details).

5 Modal Predicate Logic

The addition of modality to first-order quantificational logic raises serious philosophical
and technical problems. On the syntactic side, matters are relatively straightforward.
To the rules for the generation of the formulas of first-order logic, we add clause 1.2(iv)
above:

If A is a formula, then so is …A.

This allows …xPx, x…Px and …x…Px, for example, to be formulas.

5.1 The Problem of Quantifying In

There is a difficulty, first pointed out by Quine (1953), over the interpretation of formu-
las in which we ‘quantify into’ a modal context. Consider his example, ‘for some number
x it is necessary that x > 7' (of the form x…Px). For this to be true, the open sentence
‘necessarily x > 7' must be satisfied by some object. But whether this open sentence is
satisfied by a given object depends upon how the object is described. Describe it as’9'
and the open sentence is satisfied (since necessarily 9 > 7); describe it as ‘the number of
planets’ and the open sentence is not satisfied (since it is not necessary that the number
of planets > 7). Thus purely objectual satisfaction or quantification appear to make no
sense in this case.
There is a corresponding problem about the satisfaction of a formula at a world. The
formula ‘x > 7' is satisfied by the number 9 in the actual world. But suppose I ask: is it

614
 M O DA L LOG I C

satisfied by the number in some other world? Whether it is would appear to depend
upon how the number is to be identified in that world. Identify it through its position
in the number series and the answer is ‘yes’; identify it through its role in numbering the
planets and the answer may be ‘no.’
There have been three main semantical responses to this problem–the objectualist,
the conceptualist, and the counterpart-theoretic. We consider each in turn and show
how it leads to a distinctive style of semantics for quantified modal logic.

5.2 Objectualism
The objectualist disputes the data upon which the statement of the problem depends.
He claims that it does make sense to say that the sentence ‘necessarily x > 7' is satisfied
by a particular number, independently of how it is described, and, similarly, he claims
that there is no problem of identifying an object from one world to another; the object
is simply given in each world as the object that it is. (Such an approach is championed
in Kripke (1980).)
A simple objectualist semantics might go as follows. A quantificational K-model M is
an ordered quadruple M = (W, D, R, M), with (W, R) a K-frame as before, D (domain)
a non-empty set, and M (valuation) a function taking each world w  W into a classical
first-order model with domain D (thus Mw will assign a subset of Dn to each n-ary predi-
cate and similarly for any other non-logical constants of the language.
Given a model M = (W, D, R, M), a world w W and an assignment D (taking each
variable into an individual from D), we may now define what it is for a formula A to be
true at w relative to the assignment D, which we symbolize as w =D A:

(i) w =D Px1 . . . xn iff <D(x1), . . ., D(xn)> Mw(P);

(ii) w =D ~B iff not w =D B, and similarly for the other truth-functional
connectives;
(iii) w =D xB iff w =D' B for any assignment D' that differs at most on x from
D;
(iv) w =D …B iff v =D B whenever wRv.

The various forms of validity may then be defined in the usual way.

5.2 Conceptualism
The conceptualist grants that purely objectual quantification into modal contexts makes
no sense but, rather than abandoning quantification into such contexts altogether, he
replaces it by a corresponding form of quantification over individual concepts. Thus the
quantifier ‘for some number x’ in ‘for some number x, x > 7' is taken to range over number
concepts, such as the concept of being ninth in the number series or the concept of being the
number of planets; and the quantified statement will therefore be true since the open sen-
tence ‘necessarily, x > 7' is true when ‘x’ takes the concept of being ninth in the number series
as its value. (Such an approach, in one of its forms, is championed by Carnap (1947).)
A quantificational K-model M is now an ordered quadruple M = (W, D, C, R, M), with
(W, D, R, M) as before and C (concepts) a non-empty set of functions from W into D.
Thus each individual concept is identified with the function which takes each world
into the individual that is picked out by the concept in that world. An assignment

615
 KI T F I N E

D should now be taken to be a function from the variables into the members of C.
Truth-relative-to-an-assignment (w =D A) is then defined much as before, the principal
difference being to the first clause:

(i)c w =D Px1 . . . xn iff <D(x1)(w), . . ., D(xn)(w)> Mw(P).

In each world, the concepts behave in the manner of the individuals which they pick
out.
A plausible condition on C is that, for every individual i  D and every world w  W,
there should be a concept c in C for which c(w) = i. The maximal set of concepts will be
obtained when every function from W into D is required to belong to C. On the other
hand, a minimal set of concepts may be obtained by subjecting C to the condition that
c(w) z d(w) whenever c zd; for each individual in each world there will be a ‘privileged’
concept which picks it out.
Partial concepts, i.e. partial functions from W into D, might also be allowed. The con-
cept of the sun, for example, will not pick out any individual in a world without heav-
enly objects. The admission of partial concepts considerably complicates the semantics,
since we need to decide how to modify (i)c above when one of the arguments D(x1)(w),
. . ., D(xn)(w) is not defined. Should Px1 . . . xn be taken to be false at w (relative to the
assignment D) or ‘gappy’ (neither true nor false)? And if gappy, then how should the
other clauses in the truth-definition be modified to deal with the gaps?

5.3 Counterpart Theory

Like conceptualism, counterpart theory rejects the intelligibility of objectual quantifi-
cation into modal contexts. But rather than reconstructing quantification as quantifi-
cation over concepts, it reconstructs it in terms of the tracking of individuals from one
world to another. Suppose we are wondering whether a given individual satisfies ‘it is
necessary that x > 7'. Then according to the counterpart-theorist, we should determine
whether each acceptable way of tracking the individual across different worlds preserves
the truth of ‘x > 7'. (Such an approach, in one of its forms is championed by Lewis
(1968).).
A model M is now an ordered quadruple M = (W, D, T, R, M), where (W, D, , R, M)
is as before and T (tracking) is a relation on (n+1)-tples of the form (w, i1, . . ., in), with
n > 0, w  W and i1, . . ., in  D. T is subject to the conditions that it only relates (n+1)-
tples to (n+1)-tples, that (w, i1, . . ., in)T(v, j1, . . ., jn) implies wRv, and that (w, i1, . . .,
in)T(v, j1, . . ., jn) implies (w, ip(1), . . ., ip(n))T (v, jp(1), . . ., jp(n)) for any permutation p on
{1, 2, . . ., n}. Intuitively, T relates (w, i1, . . ., in) to (v, j1, . . ., jn) if one may acceptably
reidentify i1, . . ., in in w as j1, . . ., jn in v. Note that ‘simultaneous’ tracking is allowed; and
one might also allow current tracking to depend upon past tracking, though with some
complication to the semantics. A special case of tracking is by concepts. Thus, given a
set of concepts C, we may take (w, i1, . . ., in)T(v, j1, . . ., jn) to hold if, for some concepts
c1, . . ., cn in C, c1(w) = i1, . . ., cn(w) = in and c1(v) = j1, . . ., cn(v) = jn.
The truth-definition is as for the objectual semantics except that the clause for ‘…’
requires appeal to tracking. Where the free variables of B are x1, . . ., xn:

w =D …B iff v =E B whenever wRv and (w, D(x1), . . ., D(xn))T(v, E(x1), . . .,
E(xn)).

616
 M O DA L LOG I C

Thus as we go to the different accessible worlds v, we track the individuals D(x1), . . .,

D(xn) that are in play in the given world w.

5.4 Variable Domains

In our presentation of the previous semantical schemes, it was presupposed that the
domain D of quantification was constant from one world to another. But one may wish
it to vary. If, for example, the quantifiers are to range over all actual objects or over all
actual objects of a certain sort, then the domain should vary with what is actual.
A model under the objectualist approach (similar modifications may be made under
the other approaches) will now be a quadruple (W, D, R, M), with (W, R, M) as before
and D (domains) a function that takes each world w  W into a set Dw, at least one of
which is non-empty. The clauses in the truth-definition are as before, but the clause for
the quantifier takes the form:

w =D xB iff w =D B for any assignment Dcthat differs at most on x from D and
for which Dc(x) Dw.

The quantifier ranges, in each world, over the actual objects of the world.
When the domain is constant, the formula x…B { …xB is valid, since the effect
of either prefix, ‘x…’ or ‘…x’ is simultaneously to quantify over all worlds w  W and
all individuals i  D. But when the domain is allowed to vary, the validity of the formula
may fail in either direction. Suppose first that wRv, that Dv is not included in Dw, and
that P is a one-place predicate true in each world of the members of Dw. Then x…Px
is true at w while …xPx is false at w. Suppose next that wRv, that Dw is not included
in Dv and that P is a one-place predicate that is true in each world u of the members of
Du. Then …xPx is true at w while x…Px is false at w.
Suppose we have a variable-domain model (W,D, R, M) as specified above. We may
then interpret the quantifiers as ranging over all possible objects D+ = wWDw
, i.e. over
all objects that are actual in one or another possible world. The clause for quantified
formulas is then:

w =D 3xB iff w =D' B for any assignment Dcthat differs at most on x from D and
for which Dc(x) Dw.

We might introduce an existence or actuality predicate E to single out the objects which
are existent (or actual) in a given world:

w =D Ex iff D(x) Dw.

The actualist quantification xB may then be seen as a relativized form of possibilist
quantification, 3x(Ex Š B).

5.6 Identity
We may add an identity predicate ‘=’ to the language of quantified modal language.
Under the objectual semantics, this will be governed by the following clause:

w =D x = y iff D(x) = D(y).

617
 KI T F I N E

It should be evident that the formula x = y Š …(x = y) (the ‘necessity of identity’) will
then be valid and that Ex might be defined by y(x = y).
Under the conceptualist semantics, we might take ‘=’ to be governed by the clause:

w =D x = y iff D(x) = D(y)

requiring that the concepts be the same or by the clause:

w =D x = y iff D(x)(w) = D(y)(w)

requiring that the objects picked out by the concepts be the same. In the latter case, the
formula x = y Š …(x = y) will no longer be valid since x and y may be assigned concepts
(such as the morning star and the evening star) which only contingently pick out the same
object.
The necessity of identity may also fail under the counterpart-theoretic semantics if it
is allowed that the same object assigned to x and y can subsequently be tracked to dif-
ferent objects within a given possible world.

5.5 Completeness
The question of completeness is much more problematic for quantified modal logic than
sentential modal, although many of the standard systems of quantified modal logic can
be proved to be complete (see Brauner & Gilardi (2007) for a recent survey).

6 Applications to Specific Constructions

The modalities are open to a number of different interpretations. Let me briefly
indicate what some of them are and how they may be of interest to the philosophy of
language.

6.1 Alethic
Under the alethic interpretation, ‘…’ is interpreted as necessary truth. However, differ-
ent forms of necessity might be distinguished—the metaphysical, the natural, the logi-
cal, for example—and it is plausible that each of them should be subject to somewhat
different axioms. Many philosophers have supposed that S5 (S4 plus the axiom A Š…
¡A) is the correct logic for the notion of metaphysical necessity that has played such a
prominent role in recent philosophical discussion; and in providing a semantics for this
system, we may take the ‘worlds’ of a model to be metaphysical possible worlds and take
the accessibility relation to be ‘universal’:

w = …A iff v = A for every world v in W.

6.2 Epistemic/Doxastic
There are two principal epistemic interpretations—‘…A’ (or ‘KA’) is read either as
‘P knows that A’ or as ‘it is a priori that A’; while under the doxastic interpretation,

618
 M O DA L LOG I C

‘…A’ (or ‘BA’) is read as ‘P believes that A’ . We may take the worlds to be epistemically
or doxastically possible worlds, worlds that are possible are far as it is known or believed;
and we may take one world to be accessible from a given world if it is compatible with
what is known or believed in the given world. It is often supposed that S4 is the cor-
rect logic of knowledge for the ideal knower and that K4 (possibly with the addition of
…(…A ŠA)) is the correct logic for the ideal believer.

6.3 Tense-logical
There are two principal tense-logical interpretations—‘…A’ (or ‘GA’) is read as ‘it will
always be that A’ and ‘…A’ (or ‘HA’) is read as ‘it always was that A’. In this case,
the worlds should be taken to be instants of time and the accessibility-relation should
be taken to be the earlier–later relation. Different views on the structure of the ear-
lier–later relation correspond to different logics. When time is assumed to be linear, for
example, the corresponding logic is K4.3 (K4 plus the axiom …(A & …A ŠB) ›…(A
& …A ŠB)) (Prior (1967)).

6.4 Deontic
Under this interpretation, ‘…A’ (or ‘OA’) is read as ‘it is obligatory that A’ and its dual
‘¡A’ (or ‘PA’) is read as ‘it is permissible that A’. This interpretation is somewhat less
natural than the others since obligation and permission are more naturally treated as a
feature of acts rather than of propositions. Worlds are, perhaps, epistemically possible
states and one world is accessible from a given world when it is compatible with what
is obligatory in the given world. The system D (K plus the axiom …A Š¡A, signifying
that the obligatory is permissible) is a natural choice of logic under this interpretation
(Hilpinen (1970)).

6.5 Vagueness-theoretic
Suppose that Herbert is a borderline case of a bald man. Then we might express that this
is so as the claim that it is not definite that he is bald and not definite that he is not bald.
This suggests that we might interpret ‘…’ as a definiteness operator D (‘it is definitely
the case that’). A world w may now be understood as a precification of the language, one
under which each vague expression is made completely precise, and R may be under-
stood as the relation that holds between v and w when v is an admissible precisification
of the language relative to w. An interpretation of this sort has been proposed by Fine
(1975) and similar interpretations have been pursued by Williamson (1994) and others.
Different assumptions about the existence of ‘higher order’ vagueness will then lead to
different logics for the operator D.

6.6 Conditionals
We have so far given different interpretations for the modalities ‘…’ and ‘¡’, but the
possible worlds semantics can be applied to other kinds of operator as well. A prime
example is that of the counterfactual conditional ‘if A were to be the case, then C
would be the case’ (A > C). Given a formula A, let |A| be the set of worlds in which
A is true (relative to a given model); and given a set of worlds V and a particular world

619
 KI T F I N E

w, let C(V, w) be the set of those worlds of V that are closest to w. Along the general
lines of the semantics proposed by Stalnaker (1968) and Lewis (1973), we may then
take:

w = A > C iff v = A for every world v in C(|A|, w).

The counterfactual A > C is to be true if its consequent is true in all of the closest worlds
in which the antecedent is true.
The formal development of these various interpretations has been put to a large
number of different uses, both inside and outside of philosophy. Of special interest to
the philosophy of language is their role in understanding the corresponding locutions of
ordinary language—of tense, modal auxiliaries, etc. But the applications have to made
with a great deal of care since many aspects of ordinary use will be distorted or ignored.
The sentence ‘he does not know Obama is President’, for example, embodies the pre-
supposition that Obama is President though this is no part of the formal semantics for
the knowledge operator or, again, ‘Yesterday, John was ill’ means that John’s illness was
yesterday not at some time prior to yesterday. But application of the symbolism and
semantics of modal logic does provide an important initial guide to how the ordinary
locutions are to be understood.

7 Application to More General Topics

Let me briefly discuss three more general areas of application within the philosophy of
language.

7.1 Meaning
One naturally takes the aim of semantics to be the systematic assignment of a mean-
ing to each meaningful expression of a given language. But what are meanings? The
semantic value assigned to an expression by classical logic is merely its extension—the
denotation in the case of a singular term, a set of individuals in the case of a one-place
predicate, and a truthvalue in the case of a sentence. But expressions with the same
extension may differ in their meaning. The meanings of ‘the morning star’ and ‘the
evening star’, for example, are quite different even though their denotations are the
same.
A more adequate counterpart to meaning may be obtained by letting the semantic
value of an expression be its possible worlds ‘intension’, rather than its extension.
The intension of an expression, intuitively speaking, is a function which takes each
possible world into the object that would be the extension of the expression in that
world. Thus the intension of a singular term will be an individual concept, a function
taking each world into an individual, the intension of a one-place predicate will be a
‘property’, taking each world into a set of individuals, and the intension of a sentence
will be a ‘proposition’, a function taking each world into a truth-value or what is, in
effect, a set of possible worlds. Using intensions in place of extensions, we may dis-
criminate more finely between the meanings of different expressions. The intensions
of ‘the morning star’ and ‘the evening star’, for example, will not be the same since
they will diverge on any world in which different ‘stars’ appear in the morning and in
the evening.

620
 M O DA L LOG I C

A framework of this sort, in which extensions and intensions can be assigned to

expressions of increasingly complex type, has been proposed as a general model for our
understanding of ordinary language (Montague (1970)).

7.2 Rigid Intensions

Of special interest are the rigid or constant intensions. These pick out the same object
in each possible world; and an expression may be said to be a rigid designator when its
intension is rigid, i.e. when its designation is the same in each world. Each necessary
truth, for example, will be a rigid designator since it will designate the Truth in each
possible world.
The notion of rigid designation has become important in recent philosophy of lan-
guage through work of Kripke (1980), in which he argued that proper names (such as
‘Aristotle’ or ‘London’) and certain common nouns (such as ‘water’, ‘heat’) were rigid
designators while definite descriptions (such as ‘the famous teacher of Alexander’, ‘the
capital of London’) were not.

7.3 Intensional Objects

Natural language appears to contain reference to intentional objects which do not
appear to exist or, at least, do not exist in the usual way. It is correct to say, for example,
that Newton was looking for the philosopher’s stone, even though the philosopher’s
stone does not exist. This then raises the question of how such apparent reference to
non-existents is to be handled. A natural suggestion within the possible framework is
to treat such intentional objects as intensions. Thus the philosopher’s stone would be
a function taking each possible world into the object that is the philosopher’s stone in
that world (and perhaps into nothing if there is no such object).

7.4 Indexicality
The sentences of natural language are often context-dependent; their truth-value
depends upon the context in which they uttered. So, for example, the truth-value of the
sentence ‘I am sitting’ will vary with the speaker and the time of utterance.
The dependence of truth-value upon context is akin to the dependence of truth-
value upon the world; and so we may use a general notion of index-relative evaluation
to encompass these two forms of dependence. Thus a sentence such as ‘I am sitting’ may
be taken to have a truth-value that is relative to a triple index (w, t, s), consisting of the
world of evaluation w, the time of utterance t, and the speaker s. Different ‘operators’
may then pick up on these different indices. ‘I might be sitting’, for example, will be true
at an index (w, t, s) if ‘I am sitting’ is true at an index (v, t, s) for some accessible world
v; and ‘I will be sitting’ will be true at an index (w, t, s) if ‘I am sitting’ is true at (w, u,
s) for some later time u.
A more interesting form of multiple indexing arises when we require two or more
component indices of the same type. Consider ‘you will regret what you are now doing’.
In evaluating this sentence, ‘will’ will take us from the time of utterance t to a subse-
quent time u. But in evaluating the implicitly embedded ‘you regret what you are now
doing’ at u, ‘now’ must be taken to refer, not to u, but to t. Thus the evaluation of the
sentence requires that two temporal indices be kept in play—the ‘fixed’ time of utter-
ance t and the ‘floating’ time of evaluation u.

621
 KI T F I N E

7.5 Content/Character
The possible worlds framework has been helpful in distinguishing between different
aspects of meaning. Suppose that I now say ‘I am sitting’ and that you now say ‘I am
sitting’. Then in one way, we say the same thing and, in another way, we do not. Use
‘character’ for the former and ‘content’ for the latter. The content of our utterances may
then be represented by a set of possible worlds, the set of worlds in which I (in the one
case) and you (in the other case) are now sitting, while the character of our utterances
may be represented by the function which takes each context (as specified by a person
and a time) into the corresponding content, i.e. that set of possible worlds in which the
specified person is sitting at the specified time (Kaplan (1989)). And each of these is
distinct again from the previous notion of content, as given by the set of triples (w, t, s)
for which the speaker s is sitting at t in w.

7.6 Limitations
The possible worlds approach to meaning is subject to some well-known limitations. It
cannot distinguish, for example, between knowing one necessary truth from knowing
another. Or again, it may be permitted that I post the letter but not permitted that I post
the letter or post the letter and burn down the post office, even though the two embed-
ded clauses are true in the same possible worlds.
There is a question of how seriously to take these difficulties. My own view is that
they cannot properly be overcome or ignored and that the possible worlds approach, for
all of its success, can only be regarded as the first step towards a more adequate account
of meaning.

References
Almog J., Perry J., Wettstein H. W. (1989) Themes from Kaplan, Oxford: Oxford University Press.
Blackburn P., Benthem J. van, & Wolter F. (eds.) (2007) Handbook of Modal Logic, Amsterdam: Elsevier.
Brauner T., Ghilardi S. (2007) ‘First Order Modal Logic’ in P. Blackburn et al. (2007), pp.549–620.
Carnap R. (1947) Meaning and Necessity, Chicago: University of Chicago Press.
Fine K. (1975) ‘Vagueness, Truth and Logic’, Synthese 30, 265–300; reprinted in Vagueness: A Reader (eds.
R. Keefe and P. Smith), Cambridge: MIT Press (1996).
Goldblatt R. (2005) ‘Mathematical Modal Logic: A View of its Evolution’ in Handbook of the History of
Logic: VII (eds. D. M. Gabbay and J. Woods), Amsterdam: Elsevier.
Hilpinen R. (ed.), (1970) Deontic Logic: Introductory and Systematic Readings’, Dordrecht: D. Reidel.
Hughes G. F., Cresswell M. (2009) A New Introduction to Modal Logic, London: Routledge.
Kaplan D. (1989) ‘Demonstratives’ and ‘Afterthoughts’ in Almog et al. (1989).
Kripke S., (1980) Naming and Necessity, Cambridge: Harvard University Press.
Lewis D., (1973) Counterfactuals, Cambridge: Harvard University Press.
Lewis D. (1968) ‘Counterpart Theory and Quantified Modal Logic’, Journal of Philosophy, 65 no. 5,
113–26.
Montague, R. (1960) ‘On the Nature of Certain Philosophical Entities’, Monist, 53, 159–194; reprinted in
Thomason (1974), 148–187.
Montague, R. (1970). Pragmatics and Intensional Logic. Synthèse, 22, 68–94. Reprinted in Thomason
(1974), 119–147.
Prior A. N. (1967) Past, Present and Future, Oxford: Oxford University Press.
Quine W. V. O. (1953) ‘Reference and Modality’ in From A Logical Point of View, Cambridge: Harvard
University Press.
Stalnaker R. (1968) ‘A Theory of Conditionals’ in N. Rescher (ed.) Studies in Logical Theory, American
Philosophical Quarterly Monograph Series , No. 2', Oxford: Blackwell, 98–112.

622
 M O DA L LOG I C

Thomason, R. H. (ed.) (1974) Formal Philosophy, Selected Papers of Richard Montague. New Haven and
London: Yale University Press.
Williamson T. (1994) Vagueness, London: Routledge.

Further Readings
Benthem. J. van (2010) Modal Logic for Open Mind Stanford: CSLI publications (overview, with applications
to philosophy and computer science).
Chagrov, A., Zakyaryascher, M. (1997) Modal Logic Oxford: Clarendon Press (a more advanced text).
Fitting, M., Mendelsohn, R. (1998) First-order Modal Logic, Dordrecht: Kluwer (touches on many issues of
interest to the philosopher of language).

623
 5.5
TWO- D IME NSIONAL
L OGI C AND T W O-
DIME NS I ON ALISM IN
P HI L OSOPHY
Steven T. Kuhn

Introduction
The label “two-dimensional” has been applied to disparate logical systems. The first sec-
tion of Humberstone 2004 nicely conveys the diversity. For present purposes, we may
understand an n-dimensional logic to be one in which sentences acquire truth values
relative to n parameters. Thus classical propositional logic, which represents sentences as
being true or false simpliciter, is zero-dimensional. Common modal or tense logics, which
represent sentences as being true or false relative to a world or a time, are one-dimen-
sional. Logics aiming to represent both tense and modality assign truth values relative to
both time and world. (Thomason 1984 provides an overview.) They are two-dimensional.
For other linguistic phenomena, other parameters have been employed: for personal pro-
nouns, a speaker, an addressee and a salience order among some class of individuals; for
here and there, a pair of locations; for more faithful representations of tense (in a remark-
able anticipation of ideas discussed here, dating to Reichenbach 1947), a time of utter-
ance and a time referred to. It may sometimes be appropriate to view some parameters as
constructed from others. To represent a notion of historical necessity according to which
the past is determined and the future is open, worlds may be seen as constructed out of
instants and their temporal order. Proper understanding of actions, events, states, and the
like may require that we distinguish among sentences evaluated at instants and intervals
and to view one as constructed from the other. For other applications it may be appropri-
ate to view the parameters as primitive and perhaps independent. To represent tense and
metaphysical necessity, we may take time and world as independent parameters. Alterna-
tively, we may take time to be something internal to each world, allowing for the possibil-
ity that it has different structures in different worlds. All such generalizations of standard
modal and tense logics may be regarded as many-dimensional, the dimension correspond-
ing to the number of parameters to which the truth of a formula is relativized.
Recent interest in “two-dimensionalism” in philosophy stems from the observa-
tion that parameters of the kind mentioned above may play two distinct roles in the
 TW O - DI M EN SI O NA L L O G I C S

determination of a sentence’s truth value. The nature of these roles and the parameters
that are supposed to play them is delineated differently by different authors, but it may
be convenient to use the label 2D for two-dimensional logics whose parameters are both
possible worlds. We thereby encompass the frameworks of Davies and Humberstone
1980 and Stalnaker (1981, 2004), an approximation to that of Chalmers (2006, 2006a,
forthcoming), and illustrative fragments of those of Lewis 1981 and Kaplan (1978). We
label the dimensions utterance world and evaluation world, or simply (to avoid prejudic-
ing later interpretations) u-world and e-world. From a technical perspective, much of
the interest in logics of this kind stems from the fact that the homogeneity of the two
parameters allows each to take the value of the other. From a philosophical perspec-
tive, the dual role of possible worlds has been said to illuminate some of the most deep
and central topics in contemporary philosophy: the existence and nature of contingent
apriori and necessary empirical truths, the role of conceptual analysis in the acquisition
of knowledge, the link between conceivability and necessity and the nature of Fregean
sense, secondary qualities, narrow content, meaning, and communication. Whether
these claims of illumination are warranted depends on which, if any, of the various inter-
pretations of the 2D framework are coherent. This is a contested matter. In this survey
we will describe the logical machinery and outline its deployment by a few philosophers,
emphasizing possible connections to the contingent apriori and necessary empirical.

Logical Machinery
However contested its various interpretations, the technical machinery required for a
2D logic of the kind described above is relatively straightforward. We may take a 2D
model to be a pair M = (W,V), where W is a nonempty set (the possible worlds) and
V is a function (the valuation) that assigns a subset of W × W to each sentence letter.
A sentence letter p is true in M at a pair (u,v) of possible worlds (u-world and e-world,
respectively) if (u,v)6V(p). We may consider languages with a variety of connectives
and constants, including those with the following clauses defining A is true at (u,v) in
M (written (u,v) =MA ). (The quantifiers here are all taken to range over W.)

(u,v) =M …A iff ucvc (uc,vc) =MA

(u,v) =M …1 A iff vc (u,vc) =MA
(u,v) =M …2 A iff uc (uc,v) =MA
(u,v) =M c1 A iff (u,u) =MA
(u,v) =M c 2 A iff (v,v) =MA
(u,v) =M |
×A iff (v,u) =MA
(u,v) =M …
{A iff w (w,w) =MA
(u,v) =M ….. A iff w (v,w) =MA
(u,v) =M … . A iff w (w,u) =MA
(u,v) =M I iff u=v

We may say that a formula A is valid if, for any model M and any world u of M (u,u)=MA,
and that A is strongly valid if, for any M and any worlds u,v of M (u,v)=MA. (Motivation
for the definitions of validity will be suggested in subsequent sections.)
These connectives (or their informal counterparts) have appeared in the logic and phi-
losophy literatures on two-dimensionalism under a variety of names and notations. The
nomenclature above follows a convention once suggested by Brian Chellas. Connectives

625
 STEVEN T. K U H N

roughly analogous to the standard necessity operator are represented by something

whose outer boundary is a square. Duals of these connectives are named by rotating the
boundary 45 degrees to to obtain a diamond. For example,‘ {, the dual of …{, is a con-
nective having the truth conditions: (u,v)=M ‘ { A iff w (w,w)= A. If a connective C
M
is self dual (so that CA always takes the same truth value as ™C™A) then it is repre-
sented by something whose boundary (in the cases above, a circle) remains the same
when so rotated. Beyond these conventions, numbers appearing within the connective
indicate the coordinate on which the truth of a formula made by applying the connec-
tive depends. … .. and …. are intended to evoke the image of dominos. If their truth
clauses are written in terms of accessibility relations those relations, (t,u)R(v,w) iff u = v,
and (t,u)R(v,w), iff t = w are the relations that permit dominos to be placed to the right
and left of others in play. It is common to view the pairs of worlds as forming an array in
which (for reasons unclear) the vertical distance from the top determines the u-world and
the horizontal distance from the left indicates the v-world. On this understanding, the
identity pairs (u,u) form a diagonal from upper left to lower right. …1 A, …2 A and A are true
at a pair if A is true throughout the corresponding row, the corresponding column, and
the diagonal, and the other connectives can likewise be understood in geometric terms.
A logical system based on the connectives …, … 1, …2, c
1, c2, and {¯ (and the notion
of strong validity), referred to as the basic two-dimensional system B, is axiomatized in
Segerberg 1973. Systems based on … .. and …
. are investigated in Kuhn 1989 and Venema
1992.
These connectives are interdefinable in various ways. For example, each square and
diamond connective is interdefinable with its dual.

…A may defined as … 1 …2 A, …
2… .…
1 A, … .. …
. A, or … .. A;
… 2c
{ A, as … 1 A or …2c 1 A;
c ×c
2 A, as | 1 A or ‘ .. (IšA);
c ¯c
1 A, as { 2 A or ‘ . (IšA);
2 A, as {
… ¯…1|× A or c 2… . A;
1 A, as {
… ¯…2| × A or c1… .. A;
.
. A, as c
… 2…1 A or {¯… 1 A;
. A, as c
… 1…2 A or {
¯… 2 A.

Thus all of the above connectives except I are definable in B and in fact they can all be
defined in a language that contains just … 1, c
1 , and | 2 and c
× as primitive. Adding … 2 to
these three connectives allows us to obtain a useful normal form result. To see this, note
that the following biconditionals are valid (where each i and j inside a circle or square
is either 1 or 2, ı̃ is the “opposite” number as i and { may be c 1,c 2 or {
¯:

{i{ j A l{ i A
{i |× A l{ i A
{i …i A l…i A
{i …ı̃ A l…
i| ×A
×{
| i A l{ ı̃ A
¯|
{ × A lA
{
¯…i A l…
ı̃ {
¯A
{(A›B) l{ A›{B
{™A l™{A

626
 TW O - DI M EN SI O NA L L O G I C S

By means of these biconditionals, we can push every circular connective occurrence

inward, past occurrences of squares and boolean connectives, shortening all strings of
circular connectives to length at most one, until each occurrence of c 1, c
2 and |× gov-
erns only a sentence letter. The observation that all formulas can be written in this
“inner circle” form facilitates a proof that the basic two-dimensional logic B is equiva-
lent to the fragment, PL(x,y), of ordinary predicate logic that contains only binary
predicate letters and the two variables x and y. Note that the 2D models M = (W,V)
can be viewed as models for dyadic predicate logic. W becomes the domain of the model
and the set of pairs that V assigns to the i’th sentence letter pi becomes the extension of
the i’th predicate, Pi. A translation t that maps the inner circle formulas one–one onto
PL(x,y) can be defined as follows.

t(pi) = Pixy
t(c1 p ) = P xx
i i
t(cpi) = Piyy
2
¯pi) = Piyx
t({
t(A›B) = t(A)›t(B)
t(™A) = ™t(A)
1 A) = y t(A)
t(…
2 A) = x t(A)
t(…

We can now establish by formula induction that truth of an inner-circle formula A at

the pair (u,v) corresponds to truth of the predicate logic formula t(A) when u and v are
assigned to x and y. In symbols, (u,v) =MA iff M=t(A)[u,v] (and consequently, for all M
in PL(x,y), M=[u,v] iff (u,v)=Mt–1(M).) Thus, the inner circle formulas and the formulas
of PL(x,y) are essentially the same and, since any formula with connectives above other
than I is equivalent to an inner-circle formula, the full two-dimensional logic without I
is essentially PL(x,y). It is perhaps not as obvious as it may seem that every two-dimen-
sional formula can be expressed with two variables. The formula ‘ .. (p š‘
.. ‘ .. (p š‘.. p )),
1 2 3
for example, might naturally be associated with the four-variable formula xyzw(P1xy
š P2yz š P3zw). By the construction above, however, it can be seen to be equivalent to
x(y(P1xy š x(P2yx š y(P3xy))). Our translation can be extended to include I by
setting t(I) to (x = y) (and t(c 1 I), t(c
2 I), t(|
× I) to (x = x),(y = y) and (y = x)), thereby
establishing the equivalence of the full system to the version of PL(x,y) with identity,
which we might write PL=(x,y). Since identity is not definable in PL(x,y), it follows that
I is not definable in B. None of the various flavors of two-dimensionalism make use of
all the connectives described above, but most can be understood as built upon some of
them, and hence as built upon some fragment of PL=(x,y).

Davies and Humberstone

The application in which the technical apparatus sketched above lies closest to the
surface is contained in the influential Davies and Humberstone 1980,where it is used to
explicate a distinction drawn in Evans 1979 between “deep” and “superficial” necessity.
(u,v)=MA is read “A is true at v from the perspective of u as the actual world.” … 1 and
c1 (written there as … and A ) are read necessarily and actually. In a language with just
these two connectives, the truth value of A at (u,v) depends only on the truth values of
its subformulas at pairs (u,vc). u “stores” the world regarded as actual so that the actually

627
 STEVEN T. K U H N

operator can refer to it even when that operator lies within the scope of necessity opera-
tors. While this may be a plausible characterization of the semantic roles of actually and
necessarily, it has the peculiar consequence that c 1 A l… 1c 1 A is strongly valid, i.e., that
actually A is equivalent to necessarily actually A.
If, at v, A is true at u from the perspective of u as actual then, at any w, A is true at u
from the perspective of u as actual. Davies and Humberstone diagnose the apparent pecu-
liarity as stemming from the thought that “another world might have been actual,” and
suggest … 2 (with the reading fixedly) as a way of expressing that a formula is true no matter
what world is taken as actual. The reader puzzled by the previous formula is confusing it
with c 1 A l… 2c 1 A, which is not valid (even in the weaker sense). Furthermore, we can
understand the right side of this biconditional, or its equivalent … { A (read here as fixedly
actually A), to represent something close to Evans’ deep necessity, and … 1 A to represent his
superficial necessity. Evans’s examples of these notions employed the device (which may
or may not be exemplified in natural language) of descriptive names. A descriptive name
for x is a name that behaves semantically like the definite description the object that is actu-
ally P, where P is a predicate that holds only of x and is actually interpreted as c 1 . Thus
descriptive names are rigid (i.e., they have fixed denotations) with respect to e-worlds,
but not u-worlds, whereas ordinary proper names are said to be rigid with respect to both
worlds and ordinary definite descriptions (for example, “the tallest woman in the room,”)
may be rigid with respect to neither. We may call a description formed with the actually
operator in this way a rigidified description. If O is the descriptive name associated with
the actual length of the standard meter bar, then the sentence if there is a unique length of the
standard meter bar, then O is the length of the standard meter bar is deeply but not superficially
necessary. The sentence if there is a unique length of the standard meter bar, then the length
of the standard meter bar is the length of the standard meter bar, however, is both superficially
and deeply necessary. Within Davies’ and Humberstone’s framework, we can understand
Evans’s distinction as saying that the first sentence is true along the diagonal, i.e., “which-
ever world had been actual, [the sentence] would have been true at that world considered
as actual,” while the second is true along the horizontal, i.e., from the perspective of the
actual world as actual, the sentence is true at every world. In logical notation, A is deeply
necessary with respect to pair (u,v) and model M if (u,v)=M… { A and superficially neces-
sary with respect to (u,v) and M if (u,v)=M… 1 A. Since …
{ A is true at one pair iff it is true
at all, can we take deep necessity to be a property of sentences relative to models. We can
similarly take superficial necessity to be a property of sentences relative to models and u-
worlds. In applying these notions to natural language, we understand superficial necessity
simpliciter as superficial necessity with respect to the actual world.
The informed reader will have already noted the similarity between the example here
of a deeply, but not superficially, necessary sentence and the example in Kripke 1972 of
a sentence expressing a contingent truth knowable apriori. Indeed, once we see how the
example is constructed, the framework of Davies and Humberstone allows the construc-
tion of even simpler “toy”examples of the phenomenon. … { (P lc 1 P) is strongly valid,
but … 1 (P l c 1 P) is not even weakly valid, so sentences of the form (P l c 1 P) are
all deeply, but not superficially, necessary sentences. Indeed, Davies and Humberstone
tell us, “one can know apriori that grass is actually green iff grass is green.” Kripke’s and
Evans’s examples can be replaced by ones requiring no special views about the semantics
of proper names. Davies and Humberstone are cautious about the degree to which deep
necessity and apriority coincide. If true proper names are rigid with respect to u-worlds
and e-worlds, then true identity and difference statements employing them must be

628
 TW O - DI M EN SI O NA L L O G I C S

deeply necessary, yet a plausible case can be made that some of these are not knowable
without observation. In the other direction, other interpretations of the 2D framework
allow arguments that sentences true along the diagonal are knowable apriori, but Davies
and Humberstone are content to remark that they have yet to find an example repre-
sentable in their formal language that is not. Subsequent work has shown less restraint.
In a similar way, we can construct toy examples of the necessary empirical. Since
Grass is green is true at (u0,u0) where u0 represents the actual world, grass is actually green
is true at (u0,v) for every possible world v, and so it is superficially, but not deeply, nec-
essary. Since there is no way to know that this sentence is true without knowing that
grass is green, we have a simple example of a necessary truth, not knowable apriori. Dav-
ies and Humberstone consider, without endorsement, suggestions that philosophically
more interesting examples might have similar structure. Water is H2O (for chemically
naive speakers). Red objects reflect light of wavelength about 650nm. Actions possessing a
property that actually arouses in me a feeling of disapproval are wrong. The subject of the
sentence in each case can be regarded as containing an explicit or implicit actuality
operator. So understood, each may be seen as superficially, but not deeply necessary,
and as necessary but not knowable apriori.
When a sentence is asserted, we may take it as asserting the truth of the sentence at
the actual world from the perspective of that world considered as actual. If the connec-
tives of 2D logic represent logical form, the two notions of validity correspond to two
common characterizations of logical truth. A sentence is valid if it is true (i.e., it can be
truly asserted) in virtue of its logical form. It is strongly valid if it is necessarily true in
virtue of its logical form. Thus, this interpretation of the 2D framework, if correct, would
show that the two characterizations come apart. Another useful distinction, somewhat
related, is made by Davies and Humberstone (with attribution to Michael Dummett)
and emphasized in Lewis 1981. Considered as assertions, our two sentences about the
meter bar and, more generally, two sentences of the form P and c 1 P, apparently say
the same thing. Yet each contributes differently the truth of sentences containing them.
For … 1 (grass is green) is false, while …1c 1 (grass is green) is true. We may say that A
and B have the same assertive content (relative to M) if they agree on the diagonal, i.e.,
if =M …{(A lB). They have the same ingredient sense, if they agree at all pairs, i.e.,
If =M…1 …2 (A lB).
Although Davies and Humberstone are relatively cautious in their claims on behalf
of two-dimensionalism, it may be useful to list a few questions about their interpretation
of the logical machinery. Analogous questions often apply to interpretations discussed
in subsequent sections.

1. Is the interpretation of (u,v)=MA (and the role of the u-world in particular) coher-
ent? How should we understand locutions like true at v from the perspective of u
considered as actual, and if the actual world were u. It is not obvious that these come
to the same thing or that the second describes an entertainable condition. Perhaps
the motivating “intuition” that another world might have been actual is confused
and ought to be discouraged rather than represented.
2. Is the set of u-worlds and the set of e-worlds really the same? In answering question
one, we may determine that not every possible world might have been actual or
that the objects we might “consider” as possible worlds are not possible worlds at all.
(See discussion of Chalmers, below.) We then lose the homogeneity characteristic
of 2D logic that made talk of the diagonal sensible.

629
 STEVEN T. K U H N

3. Are the explanations of the necessary-empirical and the contingent-apriori

plausible? In the Davies–Humberstone framework, the idea underpinning these
explanations is that these sentences contain, explicitly or implicitly, something
like an actuality operator. Notice also that the puzzling pairs of properties are attrib-
uted to sentences,whereas Kripke spoke of statements.

Kaplan
David Kaplan is routinely cited as a father of two-dimensionalism, but the logical
machinery employed in Kaplan 1978 and 1989, and its motivation and interpretation,
are quite distinct from those described in other sections of this survey (and he seems
anxious to distance himself from the enterprise –1989, p 512). The main concern in
the Kaplan papers is the logical and semantic properties of indexical expressions like
I, here, now and the lessons that these might have for nature of meaning generally.
In particular, Kaplan observes that the outlook outlined in the first paragraph here,
whereby traditional modal logic is generalized by simply replacing possible worlds with
n-tuples of parameters, will fail to capture the sense in which a sentence like I am here
now, is deeply and universally true. If we restrict the n-tuples to the “proper” ones, say
the 4-tuples (w, a, p, t) where agent a in world w is producing a linguistic expression at
place p and time t, we will falsely represent the sentence as necessarily true. If, on the
other hand, we allow the parameters to vary independently, we will improperly suggest
that it can be uttered falsely. The solution is to recognize that the parameters here can
play two roles. They can be features of context in which an expression is produced or
the circumstances in which the content expressed by an expression in such a context is
evaluated. The content of an utterance of I am here by David Kaplan in Portland (in
the actual world) on March 26, 1977, for example, is the proposition that Kaplan was
in Portland then. Since this proposition is false when evaluated in circumstances where
he was elsewhere on that date, it is not necessary. Contents are supposed to be “what is
said” by an expression in a context. They can be conveniently represented by functions
from circumstances to extensions, though we should remember that such functions are
not identical to contents they represent. We can similarly represent another variety of
meaning, which Kaplan calls character by functions from contexts to contents.
If we take possible worlds as the sole features of context and circumstance, the logic of
demonstratives in Kaplan 1978 is built on the 2D logic described here with connectives
c1 (actually) and … 1 (necessarily) and validity understood in the weaker sense. Because
Kaplan’s version is formulated within predicate, rather than propositional, logic, how-
ever, he is able to introduce a formal device by which the rigidified descriptions of
Davies and Humberstone (and therefore the descriptive names of Evans) can be repre-
sented. For any term t, dthat t, is a term whose extension at (u,v) is the extension of t at
(u,u). Thus, if s represents the length of the standard meter bar, then dthat s represents the
actual length of the standard meter bar. What Kaplan’s logic obviously omits is anything
like the connectives … 2 (fixedly) and … { (fixedly actually), that feature so prominently in
the discussion of Davis and Humberstone. If the interpretation of {A at a pair (u,v)
depends on the interpretation of A at pairs (uc,w) for u zuc, then { is operating on
the character of A rather than merely its content. Kaplan 1989 labels such operators,
and the expressions that they might represent, monsters, and maintains that English
(unless being used metalinguistically) has no such devices. Although Davies and Hum-
berstone’s fixedly is not an expression of ordinary English and Kaplan’s I, here and now

630
 TW O - DI M EN SI O NA L L O G I C S

are not monsters, the basis for the general claim is somewhat unclear and it has been
questioned by a number of authors. (See, for example, Schlenker 2003.)
Kaplan does not emphasize the connections between this framework and the con-
tingent apriori or the necessary empirical. Kaplan 1978 (p. 85) alludes to a “structural”
similarity between his character/content distinction and Kripke’s apriori/necessary one.
Kaplan 1989 (p.550) mentions examples reminiscent of those in Davies and Humber-
stone. If s and t are definite, nonrigid descriptions for which s = t is a contingent, empir-
ical truth, then dthat s = dthat t is both necessary (because all identity statements with
rigid designators are necessary) and empirical (because it is true iff s = t). On the other
hand, s = dthat s is both contingent (because s is nonrigid) and knowable apriori (because
it is valid in the logic of demonstratives). The argument for this last claim would seem
to indicate that the contingent apriori extends beyond the cases mentioned by Kripke to
include examples like Kaplan’s I am here now. Kaplan nowhere makes such an observa-
tion, perhaps because the considerations raised in question three of the previous section
are so salient for these examples. I am here now is valid in Kaplan’s logic, and so we can
know apriori that it is true. Its content in a given u-world, however, is something empiri-
cal, like the proposition that Kaplan was in Portland in 1977. While we might know
apriori that the sentence I am here now is true, it requires empirical knowledge to know
what proposition the sentence expresses in a particular context. Soames 2005 maintains
that all of Kaplan’s examples of puzzling pairs (but not Kripke’s) are defective because
the proposition alleged to be apriori is not the same as that alleged to be empirical (and
that the citations above represent a misleading and unfortunate digression in Kaplan’s
otherwise insightful observations). Lewis 1994, and other of what Soames calls “strong”
two-dimensionalists, however, maintain that all alleged examples of puzzling pairs have
this feature.

Lewis
In an earlier paper (1981), Lewis suggested that the idea that there is some intuitive,
univocal notion of “what is said” captured by Kaplan’s content is illusory. On the Lewis
1981 version of two-dimensionalism, the appropriate distinction is not between features
of context and features of circumstance but rather between context as a whole and a
small package (or index) of contextual features that are independently shiftable. Every
sentence gets a truth value at a context, but, for a compound sentence, that truth value
depends on the truth values of its subsentences at indices in which certain contextual
features have been shifted (and of the mode of compounding in question). I have been
here is a compound of the sentence I am here using the it has been that mode. The truth
value of an utterance of I have been here in a context with Kaplan as speaker in Port-
land in 1977 is determined by the truth values of I am here at indices in which the time
coordinate of that context is shifted toward the past. After the coordinate has shifted,
the index may correspond to no context at all. (Kaplan may not be there then.) Since
any sentence may occur either alone or as part of a compound, we must evaluate all
sentences relative to both contexts and indices. These are the two dimensions. From
this perspective, the absence of monsters is a matter of stipulation. If there is any opera-
tor { for which the truth of {A in a context c depends on the truth of A when some
feature of c has shifted then that feature is, by definition, part of the index. This does
not rule out the possibility that two different operators might make use of the same
contextual feature. In that case indices may require two coordinates to store different

631
 STEVEN T. K U H N

values of that feature. Perhaps this situation would be as monstrous as the presence of
the context-altering creatures that frightened Kaplan.

Stalnaker
For Stalnaker, the need for 2D logic arises in connection with an explanation of asser-
tion. We start with the familiar idea that the content of an assertion, a proposition, can
be roughly identified with a set of possible worlds (the worlds in which the proposition
is true). Each participant brings to a conversation a set of presuppositions, i.e., a set of
propositions p such that he is disposed to act as if he takes p to be true, takes the others
to take it to be true as well, takes them to take him to take it to be true, and so on. The
intersection of these propositions, the speaker’s context set, will comprise all worlds that
the speaker takes to be “live options” in the conversation. The participants in a conver-
sation value communication. Since communication is impeded when presuppositions
diverge, context sets will tend to become non-defective, i.e., identical for all. Given a
nondefective context, conversants have a further common interest in narrowing it, that
is, in more fully specifying the way the world is taken to be, though if they disagree
about the facts their interests in how this should be done may conflict. An assertion of
the proposition p is, in essence, a proposal to narrow the context set by eliminating all
worlds where p is false (or, equivalently, by selecting just those making p true).
One difficulty with this very plausible picture is that necessary propositions are
nowhere false, and so cannot narrow any context set. Yet utterances like Hesperus is
identical to Phosphorus, it is now three o’clock, and an ophthamologist is an eye doctor appear
to be respectable assertions of just such propositions. These appearances, Stalnaker con-
tends, are deceiving. Hesperus is Phosphorus does standardly express a necessary proposi-
tion. Since asserting a necessary proposition fails to narrow the context set, however,
it would violate a fundamental principle of conversation. We are therefore justified in
interpreting one who appears to be doing so as asserting another instead. (Stalnaker is
here invoking ideas of Paul Grice. See Chapter 7.7 of this volume.) That Hesperus is
Phosphorus expresses a necessary proposition follows from the fact that Hesperus and
Phosphorus rigidly denote the planets visible in the evening and morning sky and that
these are one and the same. The speaker who appears to assert that Hesperus is Phos-
phorus, it is reasonable to suppose, fails to presuppose this last condition. His context
set includes a world w in which the morning and evening appearances are of a single
planet and a world wcin which they are of different planets. In w his utterance expresses
a necessary truth; in wc, a necessary falsehood. The situation can be modeled in the
now familiar way by a 2D model M, where (u,v)=MA indicates that what A expresses
in u is true in v. If the truth values of A are arranged geometrically as described above,
the row of truth values at vertical position u represents the proposition expressed by
A in u. The speaker is proposing to eliminate wc, but not w, from the context set. He
cannot do this by asserting propositions that would be expressed by A or ™A in w or
wc, for each of those would eliminate either nothing or everything. He can do it by
asserting the proposition expressed in either world by c 1 A (†A in Stalnaker’s nota-
tion). c 1 A can be regarded as expressing, at every world, that the proposition expressed
by Hesperus is Phosphorus is true, or that the denotations of Hesperus and Phosphorus are
identical. In general, it is often reasonable to interpret what appears to be an assertion
of a necessarily true proposition that A as instead an assertion of a contingent proposi-
tion that c 1 A.

632
 TW O - DI M EN SI O NA L L O G I C S

Stalnaker 2004 emphasizes that he is not suggesting a semantical theory under which
the proper name Hesperus, like Kaplan’s indexicals I and here, gets different interpreta-
tions in different contexts. Semantical considerations still determine that Hesperus is
Phosphorus standardly expresses either a necessarily true proposition or a necessarily
false one, and empirical considerations help us determine that it is the former. Prag-
matic considerations that presuppose and utilize the standard interpretation, however,
may lead us to conclude that an utterance of that sentence is an assertion of an entirely
different, contingent proposition. Utterances of Hesperus do denote different individu-
als in different u-worlds, but this should not be thought of as reflecting the meaning of
Hesperus. Under appropriate conditions the context set might contain a world where
such an utterance denotes a horse or a real number.
Similar considerations suggest that Stalnaker’s two-dimensionalism, though perhaps
explaining how sentences that standardly expresses necessary propositions can convey
information about the world, has little to tell us about the existence and nature of
the contingent apriori or the necessary empirical. Stalnaker 1981 remarks that … { can
be understood as an apriori truth operator, but Stalnaker 2004 rightly expresses strong
misgivings. If it is sometimes appropriate to interpret Hesperus is Phosphorus as express-
ing something necessarily false, it may be similarly appropriate to do so for 7+5 = 12,
Tuesday follows Monday, or any other standardly apriori sentence. If so, then these will
be false somewhere on the diagonal and an application of … { will produce sentences
everywhere false.

Chalmers
The currency of the term two-dimensionalism is largely due to David Chalmers, the sub-
ject’s most voluble and energetic expositor. Yet, on Chalmers’ preferred interpretation,
2D logic, as described here, is largely hidden from view. Having climbed the 2D ladder,
Chalmers is happy to put it aside. (Some qualification is warranted regarding this and
other remarks in this section. A large proportion of Chalmers’ writings on two-dimen-
sionalsm is devoted to interpretations other than what I call his preferred interpreta-
tion. “Semantic pluralism” allows him to regard some of these as viable alternatives,
capturing notions of meaning distinct from his own primary concern. Others are pre-
sented as reformulations of the preferred framework, taking different notions as primi-
tive than the preferred framework but, under appropriate philosophical assumptions,
arguably equivalent to it. The intention is to avoid losing the reader who might have
scruples about his choice of primitives. My remarks concern the preferred interpreta-
tion. A further warning: in this section more than others, qualifications and refinements
are omitted to convey basic ideas in reasonable space.)
Chalmers sees two-dimensionalism as clarifying connections between cognitive con-
tent and necessity revealed to us by Frege and Carnap’s discussions of sense and inten-
sion, but subsequently obscured by Kripke’s persuasive examples of necessary empirical
statements. For this reason, he deliberately seeks an interpretation satisfying a condi-
tion like the one Stalnaker repudiated. He takes u-worlds to be scenarios, rather than
possible worlds, corresponding to “ways that the world might turn out to be, for all
we know apriori.” On the preferred analysis, scenarios are identified with equivalence
classes of complete sentences in some idealized (infinitary) language capable of repre-
senting every hypothesis about the world. A complete sentence is one that is epistemi-
cally possible (this notion is primitive) and implies S or ™S for every sentence S in the

633
 STEVEN T. K U H N

language. Sentences within equivalence classes all imply each other. Some scenarios
imply (“verify”) utterances of Hesperus is Phosphorus, others verify utterances of Hespe-
rus is not Phosphorus, but none verifies the conjunction of these and none verifies Venus
is not Venus. A scenario verifying Hesperus is not Phosphorus might have that sentence
as a conjunct, but it might instead have the bright celestial objects visible in the morning and
evening are Mars and Mercury, respectively, or perhaps no celestial object appears in both
the morning and evening. In general, Chalmers argues, for any scenario u, there will be a
“relatively limited” vocabulary V such that if u verifies an utterance S, then u verifies
some sentence D containing only expressions of V that implies S. There is no require-
ment that expressions of V be qualitative and no requirement that objects mentioned
in s be definable in V.
Since scenarios and worlds are distinct, this conception affords no notion of a diago-
nal. Instead, Chalmers defines the “1-intension” of a sentence S to be the function that
assigns to each scenario u the value T if u implies S and F if it implies ™S. It now becomes
plausible to argue that S is apriori if and only if its 1-intension always takes the value T.
Though it is unclear that his preferred interpretation requires it, Chalmers argues that
there is a correspondence between scenarios and worlds (or centered worlds, i.e., worlds
with time, place, and speaker marked) so that the 1-intension corresponds to a diagonal
after all, but he insists that it should not be seen as deriving from it. The “2-intension”
is just the familiar function from possible worlds to truth values. Thus, sentences are not
(or not fundamentally) evaluated at scenario-world pairs, but at worlds and scenarios
alone. The intuition is that an utterance of S is verified by scenario u if it is true if the
actual world turns out to really be as described by u. It is satisfied by possible world v if
it would be true if the actual world were as described by v. The scenario according to
which no body is visible morning and evening verifies Hesperus is not Phosphorus, but
the world with no such body still satisfies Hesperus is Phosphorus. If the interpretation
has the properties it was designed to, then all the puzzling pair examples have similar
structures. An utterance is true apriori if and only if its 1-intension takes the value T
at all scenarios and it is necessary if and only if its 2-intension takes the value T at all
worlds. It should not be surprising that each of these conditions can obtain without the
other. Whether that implies that there are necessary empirical and contingent apriori
propositions depends on whether one takes the 1-intension and 2-intension of an utter-
ance to correspond to two propositions, both expressed by that utterance, or whether
one takes them to be parts of a single proposition expressed by it. Chalmers leans toward
the latter. To say that puzzling-pair examples have epistemological and metaphysical
properties in virtue of different aspects of their meanings does seem more reasonable
than to insist they express two propositions.

References
Chalmers, D. (2006) “The foundations of two-dimensional semantics,” in M. Garcia-Carpintero & J. Macia,
eds., Two-Dimensional Semantics: Foundations and Applications, New York: Oxford University Press, 2006.
—— (2006a) “Two-dimensional semantics,” in Oxford Companion to the Philosophy of Language, E. Lepore
and B. Smith, eds., New York: Oxford University Press.
—— (forthcoming) “The nature of epistemic space,” in A. Egan and B. Weatherson, eds., Epistemic Modality.
Oxford: Oxford University Press.
Davies, M. and L. Humberstone (1980) “Two notions of necessity,” Philosophical Studies, 38:1–30.
Evans, M. (1979) “Reference and contingency,” The Monist 62: 161–189.
Humberstone, L. (2004) “Two-dimensional adventures,” Philosophical Studies 118: 17–65.

634
 TW O - DI M EN SI O NA L L O G I C S

Jackson, F. (1998) From Metaphysics to Ethics: A Defense of Conceptual Analysis. Oxford: Oxford University
Press.
Kaplan, D. (1978) “On the logic of demonstratives,” Journal of Philosophical Logic, 8:81–98.
—— (1989) “Demonstratives,” in J. Almog, J. Perry, and H. Wettstein, eds., Themes from Kaplan, 481–563,
Oxford: Oxford University Press.
Kripke, S. (1972) “Naming and necessity,” in D. Davidson and G. Harman, eds., Semantics of Natural Lan-
guage, 253–355, Dordrecht: D. Reidel and Company.
Kuhn, S. (1989) “The domino relation: Flattening a two-dimensional logic,” Journal of Philosophical Logic,
18(2):173–195.
Lewis, D. (1981) “Index, context, and content,” in S. Kanger and S. Ohman, eds., Philosophy and Grammar:
Papers on the Occasion of the Quincentennial of Uppsala University, 79–100, Dordrecht: D. Reidel Publishing
Co.
—— (1994) “Reduction of mind,” in S.Guttenplan, ed., Companion to the Philosophy of Mind, Oxford:
Blackwell.
Reichenbach, H. (1947) Elements of Symbolic Logic, New York: Random House.
Schlenker, P. (2003) “A Plea for Monsters,’ Linguistics and Philosophy, 26(1): 29–120.
Segerberg, K. (1973) “Two-dimensional modal logic,” Journal of Philosophical Logic, 2(1):77–96.
Stalnaker, R. (1981) “Assertion,” in P. Cole, ed., Radical Pragmatics, New York: Academic Press.
—— (2004) “Assertion Revised: On the interpretation of two-dimensional modal semantics,” Philosophical
Studies, 118(1–2) 299–322.
Thomason, R. (1984) “Combinations of tense and modality,” in D. Gabbay and F. Guenthner, eds., Hand-
book of Philosophical Logic, Volume II: Extensions of Classical Logic, 135–165. Dordrecht: D. Reidel Publish-
ing Co.
Venema, Y. (1992), “A Note on the Tense Logic of Dominoes,’ Journal of Philosophical Logic
21(2):173–182.

Further Reading
Garcia-Carpintero, M. and J. Macia, eds. (2006) Two-Dimensional Semantics, Oxford: Oxford University
Press. (An anthology including interpretations, applications, and criticism of the two-dimensional frame-
work. The contributions of Davies, Evans, and Stalnaker to Stoljar and Davies (2004), are reprinted, and
it also contains Chalmers 2006 and eleven other papers.)
Marx, M. and Y. Venema (1997) Multi-Dimensional Modal Logic, Kluwer, Dordrecht Academic Publishers.
(The only book-length treatment of many-dimensional logics. Contains a chapter on two-dimensional
logic with a wealth of technical results on logical systems with connectives like those described here—but
not the normal form and translation results sketched above.)
Soames, S. (2005) Reference and Description: The Case Against Two-Dimensionalism, Princeton: Princeton
University Press. (A detailed and sustained attack on two-dimensionalism, which the author sees as a mis-
guided attempt to revive Fregean descriptivism in the face of compelling arguments by Saul Kripke against
it. If Chalmers is the subject’s most energetic expositor, Soames is its most energetic critic.)
Spencer, C. (2010) “Two-dimensional semantics,” Oxford Bibliographies Online—Philosophy. (A richly anno-
tated, extensive and current bibliography.)
Stoljar, D. and M. Davies, eds. (2004), Philosophical Studies 118(1–2). A double issue of the journal contain-
ing material from a 2002 conference. In addition to Humberstone 2004 and Stalnaker 2004, it contains a
useful introduction by the editors, a shortened version of Chalmers 2006, papers by David Braddon-Mitch-
ell, Martin Davies, Frank Jackson, Frederick Kroon, Michaelis Michael, Philip Pettit, Laura Schroeter,
and a posthumous publication of Gareth Evans’s comments to Davies and Humberstone concerning their
1980 paper.

635
 5.6
MANY-VALUED
L OGIC S
Nicholas J.J. Smith

1 Introduction
A many-valued (aka multiple- or multi-valued) semantics, in the strict sense, is one
which employs more than two truth values; in the loose sense it is one which counte-
nances more than two truth statuses. So if, for example, we say that there are only two
truth values—True and False—but allow that as well as possessing the value True and
possessing the value False, propositions may also have a third truth status—possessing nei-
ther truth value—then we have a many-valued semantics in the loose but not the strict
sense. A many-valued logic is one which arises from a many-valued semantics and does
not also arise from any two-valued semantics [Malinowski, 1993, 30]. By a ‘logic’ here
we mean either a set of tautologies, or a consequence relation. We can best explain these
ideas by considering the case of classical propositional logic. The language contains the
usual basic symbols (propositional constants p, q, r, . . .; connectives ™, š, ›ol;
and parentheses) and well-formed formulas are defined in the standard way. With the
language thus specified—as a set of well-formed formulas—its semantics is then given
in three parts. (i) A model of a logical language consists in a free assignment of semantic
values to basic items of the non-logical vocabulary. Here the basic items of the non-logi-
cal vocabulary are the propositional constants. The appropriate kind of semantic value
for a proposition is a truth value, and so a model of the language consists in a free assign-
ment of truth values to basic propositions. Two truth values are countenanced: 1 (rep-
resenting truth) and 0 (representing falsity). (ii) Rules are presented which determine
a truth value for every proposition of the language, given a model. The most common
way of presenting these rules is via truth tables (see Figure 5.6.1). Another way of stating
such rules—which will be useful below—is first to introduce functions on the truth val-

α β ¬α α∧β α∨β α→β α↔β

1 1 0 1 1 1 1
1 0 0 1 0 0
0 1 1 0 1 1 0
0 0 0 0 1 1

Figure 5.6.1 Classical truth tables

 M A N Y- VA LU ED L O G I C S

ues themselves: a unary function ™ * and four binary functions š *, ›

*o
* and l * (see Figure
5.6.2). Representing the truth value of D(on a given model) as [D], we then specify the
truth values of compound formulas as in Figure 5.6.3. Once one becomes familiar with
the distinction between connectives and truth functions, it is customary to use the same
symbols for both and to let context disambiguate. As it generally increases readability, I
shall mostly follow this practice below (i.e. omit the *’s on truth functions). (iii) Deni-
tions of tautology and logical consequence are introduced. In this case, a tautology is a
proposition which gets the value 1 on every model (e.g. p ›™p, p o p), and a proposition
Dis a logical consequence of the set of propositions *(written *_ D) if, on every model
on which every proposition in * has the value 1, Dhas the value 1 (e.g. {p, p o q} _ q, {p}
_ p ›q). Classical logic is then the language just introduced together with either the set
of tautologies, or the consequence relation, just dened. The definition of a logic in terms
of a consequence relation is more powerful, in that once we have the consequence rela-
tion, we can reconstruct the set of tautologies as the set of propositions D such that À _ D.
However sometimes we are interested only in tautologies—hence we allow that a logic
may be specified just by giving a set of tautologies, without a consequence relation.
As we have just seen, a logic can be specified as the one which arises from a certain
semantics. Logics can also be specified in terms of proofs. Proofs come in many different
forms. Consider, for example, axiomatic proof systems. Finitely many propositions are
taken as axioms. One or more rules of inference are specified: they take one or more prop-
ositions as input, and give a proposition as output. A formal proof (in a given axiomatic
system) is then defined as a finite list of propositions, each of which is either an axiom, or
follows from earlier propositions in the list by a rule of inference. A proposition D is said to
be a theorem of the system if there is a proof whose last line is D. There are many known
axiom systems which have the property that their theorems are exactly the tautologies
of classical logic. Thus, classical logic is said to be finitely axiomatizable. We can also
define consequence relations using axiomatic proof procedures, by allowing assumptions
in proofs. So axiomatic proof systems give us an alternative way of characterizing logics.

    
¬ ∧ ∨ → ↔
1 1 0 1 1 1 1
1 0 0 1 0 0
0 1 1 0 1 1 0
0 0 0 0 1 1

Figure 5.6.2 Classical truth functions


[¬α] = ¬ [α]

[α ∧ β] = [α] ∧ [β]

[α ∨ β] = [α] ∨ [β]

[α → β] = [α] → [β]

[α ↔ β] = [α] ↔ [β]

Figure 5.6.3 Rules for assigning truth values to compound propositions

637
 N I CH O LA S J . J . S MI T H

In §2 we look at systems of many-valued semantics and their associated logics and in

§3 we mention some of the uses to which these systems have been put. Overall space
constraints, together with the judgement that readers coming to the existing literature
from a philosophy of language background will find it harder to gain an overview of the
different kinds of many-valued systems than to find information on applications of one
or other of these systems to particular topics of interest, led to the decision to devote the
bulk of the available space to §2.

2 Systems of Many-Valued Logic

2.1 Three Values
Suppose we take the language of classical propositional logic, and give it a semantics
which countenances a third truth value—which we shall write as *—as well as the clas-
sical 1 and 0. Taking a (three-valued) model to be a free assignment of one of these three
values to each basic proposition, the classical truth tables will no longer determine a
truth value for every proposition of the language. For example, if p has the value * on a
given model, then the tables do not specify a value for compound propositions (e.g. ™p,
p o q) which have p as a component. One way to remedy this situation is to replace the
classical tables with three-valued truth tables. There are many such tables (see Bolc and
Borowik [1992, ch.3] for a survey); we begin with the three that have played the most
prominent role in the literature. Figure 5.6.4 shows the Bochvar (aka Kleene weak) tables
[Rescher, 1969, 29–30] [Kleene, 1952, 334]. The idea here is that a compound whose
components all take classical values takes the value that the classical tables assign, while
if any of its components takes the value *, the compound takes the value *. Figure 5.6.5
shows the Kleene (strong) tables [Kleene, 1952, 334]. As before, a compound whose
components all take classical values takes the value that the classical tables assign. As
for the case where a component has the value *, there are two useful ways of thinking
about what the tables dictate. First, we can think of the * as an unknown classical value.
So we suppose the * is a 1 and calculate the value that the classical table would give,
and we suppose the * is a 0 and calculate the value that the classical table would give: if
we get 1 both times, then that is the value in the new table; if we get 0 both times, then
that is the value in the new table; and if we get 1 once and 0 once, then the value in

α β ¬α α∧β α∨β α→β α↔β

1 1 0 1 1 1 1
1 * * * * *
1 0 0 1 0 0
* 1 * * * * *
* * * * * *
* 0 * * * *
0 1 1 0 1 1 0
0 * * * * *
0 0 0 0 1 1

Figure 5.6.4 Bochvar (aka Kleene weak) tables

638
 M A N Y- VA LU ED L O G I C S

α β ¬α α∧β α∨β α→β α↔β

1 1 0 1 1 1 1
1 * * 1 * *
1 0 0 1 0 0
* 1 * * 1 1 *
* * * * * *
* 0 0 * * *
0 1 1 0 1 1 0
0 * 0 * 1 *
0 0 0 0 1 1
Figure 5.6.5 Kleene (strong) tables

the new table is *. (When we are trying to determine what value should be in the new
table where two components both have the value *, we calculate all four possibilities
given by replacing each * with a 1 or a 0.) Second, we can think of * as lying between 1
and 0 on a scale from more true to less true: * is more true than 0 but less true than 1.
Then we can see the conjunction as taking the least true of the values of its conjuncts,
the disjunction as taking the most true of the values of its conjuncts, and the conditional
and biconditional as defined in the standard classical way from the other connectives:
DoEdf ™D›E(or™Dš™E))andDlEdfDoEš EoD. Figure 5.6.6 shows
the Łukasiewicz tables [Łukasiewicz and Tarski, 1930]. These are exactly like the Kleene
tables except for the values of DoE and DlE when Dand Eboth have the value *: in
the Łukasiewicz tables these values are 1, whereas in the Kleene tables they are *.
We come now to the third stage of presenting a semantics: dening notions of tautol-
ogy and/or consequence. The standard way of defining these notions is to specify a subset
of the truth values as designated. A tautology is then a proposition which takes a des-
ignated value on every model, and Dis a logical consequence of * iff, on every model
on which every proposition in * has a designated value, Dhas a designated value. (The

α β ¬α α∧β α∨β α→β α↔β

1 1 0 1 1 1 1
1 * * 1 * *
1 0 0 1 0 0
* 1 * * 1 1 *
* * * * 1 1
* 0 0 * * *
0 1 1 0 1 1 0
0 * 0 * 1 *
0 0 0 0 1 1

Figure 5.6.6 Łukasiewicz tables

639
 N I CH O LA S J . J . S MI T H

earlier definitions in the classical case emerge from this template by setting 1 as the only
designated value among the two classical values 1 and 0. Marking designated values
by underlining, we represent this choice as {0, 1}.) In the three-valued case, the most
obvious choice is to set 1 as the only designated value: {0, *, 1}. However it is also not
unreasonable to set both 1 and * as designated values: {0, * , 1}. On the former choice, a
tautology is a proposition which is always true (and consequence is a matter of preserva-
tion of truth); on the latter choice, a tautology is a proposition which is never false (and
consequence is a matter of preservation of non-falsity). By combining different choices
of designated values with different truth tables, we get different logics:

B3: {0, *, 1} and Bochvar tables

Bc3: {0, * , 1} and Bochvar tables
K3: {0, *, 1} and Kleene tables
Kc3, aka LP [Priest, 2008, 124]: {0, * , 1} and Kleene tables
Ł3: {0, * 1} and Łukasiewicz tables
Łc3: {0, * , 1} and Łukasiewicz tables

These logics differ more or less from classical logic. For a start, it is not hard to see
that any semantics in which 0 is not a designated value and whose tables agree with
the classical tables where only 1’s and 0’s are involved—this includes all the systems
just introduced—will be such that all its tautologies are classical tautologies. Going
the other way, all classical tautologies come out as tautologies of Bc3 and Kc3 (given the
original stipulation that a many-valued logic is one which arises from a many-valued
semantics and does not also arise from any two-valued semantics, this means that these
logics—considered as sets of tautologies (the situation is different when we consider
consequence relations)—are not many-valued logics: for while they do arise from many-
valued semantics, they also arise from the classical two-valued semantics); B3 and K3,
however, have no tautologies at all; Ł3 has some of the classical tautologies (e.g. p o p)
but not all (e.g. p›™p has the value * when p does); Łc3 has all the tautologies of Ł3 and
some more besides (e.g. p›™p), but it still does not have all the classical tautologies (e.g.
™p o ™p) ›™(™p o p) has the value 0 when p has the value *). The story is different
again when it comes to consequence relations; for example, while Bc3 and Kc3 have the
same tautologies as classical logic, they do not have classical consequence relations (e.g.
q is a consequence of {p š ™p} in classical logic, but not in Bc3 or Kc3: consider a model
on which p has the value *—hence so does p š ™p—and q has the value 0).

2.2 Finitely Many Values

The three-valued tables of the previous section can all be generalized to the case where
we have any finite number n of truth values. It is convenient to represent these n values
as fractions:

0 1 n–2 n–1
...,
’
n–1 n–1 ’ n – 1’ n – 1

So where n = 5, for example, we have the following values: 0,  ,  ,  , 1. To generalize the

Bochvar tables, we say that [D] is whatever the classical tables dictate when all D ’s com-
ponents have the values 0 or 1; and otherwise it is the ‘middle value’. (Where the number

640
 M A N Y- VA LU ED L O G I C S

n of values is odd, the middle value is simply  , i.e. half of the top value 1; where n is
even, the ‘middle’ value is taken to be n – 2 , i.e. half of the second-top value [Rescher,
2(n – 1)
1969, 43–4].) To generalize the Kleene tables, we use the rules in Figure 5.6.7. To gen-
eralize the Łukasiewicz tables, we use rules exactly like those for the Kleene systems,
except that we replace the rule for the conditional with the following:

[DoE ] = min(1, 1 – [D] + [E])

(In the rule for the biconditional in the Łukasiewicz systems, the conditional is then
this Łukasiewicz conditional, not the Kleene conditional.) The idea here is that if the
consequent is at least as true as the antecedent, then the conditional is completely true,
while if the antecedent is truer than the consequent—and the difference between their
truth values is k—then the conditional has the value 1 – k, that is, it is k less than fully
true. It is routine to verify that if we set n = 3, then the three sets of rules just given
specify the three sets of tables given in the previous section (with * written as ) and if
we set n = 2, then these rules all specify the classical truth tables.
If we now choose a set of designated values, logics emerge. For example, taking the
Łukasiewicz rules together with 1 as the only designated value yields the n-valued logics
Łn. Lindenbaum showed that the set of tautologies of Łm is a subset of the set of tautolo-
gies of Łn (m, n t2) just in case n – 1 is a divisor of m – 1 [Łukasiewicz and Tarski, 1930,
48]. So, for example, the tautologies of Ł9 are a subset of those of Ł5 (and likewise of
Ł3), because 4 (and 2) is a divisor of 8. It follows that none of these many-valued log-
ics has more tautologies than classical (two-valued) logic. The logics Łn are all finitely
axiomatizable (see e.g. Malinowski [1993, 39]), but not every finitely many-valued logic
is; for example, Rescher [1969, 157–9] presents a three-valued logic which is not finitely
axiomatizable (given substitution and modus ponens as rules of inference).
We turn now to a different strategy for generating n-valued systems, due to Post
[1921]. Note that in the Kleene and Łukasiewicz (but not Bochvar) systems, the truth
value of ™D is as far below 1 as the truth value of D is above 0; that is, the distance
between 0 and [D] is the same as the distance between [™D] and 1. This treatment of
negation requires that there be a meaningful notion of distance between the truth values
of the system. In Post’s systems, by contrast, the truth values are merely ordered: given
any two truth values, we can say which of them is the truer; but we cannot compare the
distances between different pairs of values. We represent the truth values of the n-valued
Post system as follows:

t1, t2, . . ., tn

[¬α] = 1 − [α]
[α ∧ β] = min([α], [β])
[α ∨ β] = max([α], [β])
[α → β] = [¬α ∨ β]
[α ↔ β] = [(α → β) ∧ (β → α)]
Figure 5.6.7 Rules for n-valued Kleene systems

641
 N I CH O LA S J . J . S MI T H

The ordering of the values is this: ti is less true than tj just in case i < j. Figure 5.6.8 shows
the truth functions of this system. So the value of ™D is the value immediately after the
value of D(in the ordering of the truth values from least true to most true), except in the
case where the value of Dis the top value tn, in which case the value of ™D is the bottom
value t1. The rule for disjunction is familiar from Kleene and Łukasiewicz: the value of
the disjunction is the truer of the values of the disjuncts. The remaining connectives are
defined in the standard ways from ™ and ›. Taking Post’s rules and setting n = 3 yields
a three-valued system different from any of the systems examined in §2.1; setting n = 2
yields classical logic.
It is a familiar fact that in classical logic, some connectives can be defined in terms
of others. Similar kinds of results hold in many-valued logics. For example, Łukasiewicz
took ™ and o as primitive connectives, and defined the others in terms of them (see
Figure 5.6.9). Similarly, Bochvar took ™ and š as primitive, and Post took ™ and › as
primitive. It is also sometimes possible to define the connectives of one system within
another system. For example, Kleene conditional is definable in terms of Kleene nega-
tion and disjunction (Figure 5.6.7); but Kleene negation and disjunction coincide with
Łukasiewicz negation and disjunction; so Kleene conditional is definable in terms of
Łukasiewicz negation and disjunction. In fact, all the Kleene connectives are definable
in terms of Łukasiewicz connectives but not vice versa; and all the Bochvar connectives
are definable in terms of Kleene (and hence Łukasiewicz) connectives but not vice versa
[Bergmann, 2008, 91–2]. A set of connectives is said to be functionally complete if we can
define all possible connectives from the connectives in that set. In an n-valued system,
an m-place connective has a truth table with nm rows (one for each possible assignment
of the n truth values to the m component propositions). We specify an m-place con-
m
nective by putting a truth value in each row of the table. Thus there are n(n ) possible
m-place connectives. It is well known that in the classical (two-valued) case, all these
connectives can be defined using only the five standard connectives introduced at the
outset. (Indeed, we do not need all of them: just ™ and š will do, or ™ and o, etc.) This
property of functional completeness of (some subset of) the set of five standard con-
nectives carries over to some, but not all, many-valued logics. In every n-valued Post


ti+1 if i = n
¬ti =
t1 if i = n
t i ∨ tj = tmax(i,j)
t i ∧ tj = ¬(¬ti ∨ ¬tj )
ti → tj = ¬ti ∨ tj
ti ↔ tj = (ti → tj ) ∧ (tj → ti )

Figure 5.6.8 Truth functions for n-valued Post systems

α∨β =df (α → β) → β
α∧β =df ¬(¬α ∨ ¬β)
α↔β =df (α → β) ∧ (β → α)
Figure 5.6.9 Defined connectives in Łukasiewicz systems

642
 M A N Y- VA LU ED L O G I C S

system, the set containing ™ and › is functionally complete. In no n-valued Łukasiewicz

system is the set of five standard connectives functionally complete—although for some
(but not all) n this set is precomplete, which means that it is not functionally complete
but becomes so with the addition of any connective which is not already definable in
terms of the standard five [Urquhart, 2001, 266–8].
In all the n-valued systems that we have considered so far, the set of truth values is
linearly ordered—but it need not be so. For example, Belnap [1977] considers a system
with four truth values—T, F, B and N—ordered as in Figure 5.6.10 (with x < y iff one
can get from x to y by following arrows). N and B are distinct truth values, with neither
greater than the other—so we cannot take conjunction as min and disjunction as max.
However, every pair of values has a supremum (a least value that is greater than or equal
to both values in the pair) and an inmum (a greatest value that is less than or equal to
both values in the pair). A partially ordered structure with this property is called a lat-
tice. In any lattice of truth values, the strategy of defining conjunction and disjunction
as min and max generalizes to defining them as inf and sup. So, for example, we can take
N š B = F and T š B = B, and N › B = T and N › F = N.
We turn now to a new strategy for generating finitely many-valued systems, due to
Jaśkowski [1936]. Suppose we have an m-valued system X and an n-valued system Y. We
can then form a new, (m un)-valued system X uY by taking the product of these two sys-
tems. The truth values of the product system will be elements of the Cartesian product
XQ uYQ , where XQ (YQ ) is the set of values of X (Y); that is, they will be pairs whose first
element is a value of X and whose second element is a value of Y. Truth functions of the
product system are then specified coordinatewise (Figure 5.6.11: superscripts on truth
functions indicate to which system they belong). One system of particular interest is the
product of the classical system with itself. This system has four values:

¢0, 0², ¢0, 1²¢1, 0²¢1, 1²

> T _@@
~~~ @@
~~~ @@
~~ @
N @
`@ B
@@ ~~?
@@ ~~
@ ~~~
F
Figure 5.6.10 A lattice of four truth values
X Y
¬x, y = ¬ x, ¬ y
X Y
x1 , y1 ∧ x2 , y2 = x1 ∧ x2 , y1 ∧ y2
X Y
x1 , y1 ∨ x2 , y2 = x1 ∨ x2 , y1 ∨ y2
X Y
x1 , y1 → x2 , y2 = x1 → x2 , y1 → y2
X Y
x1 , y1 ↔ x2 , y2 = x1 ↔ x2 , y1 ↔ y2
Figure 5.6.11 Truth functions for product systems

643
 N I CH O LA S J . J . S MI T H

Its truth table for negation (for example) is shown in Figure 5.6.12. Note that its truth
tables for conjunction and disjunction could alternatively be arrived at by setting ¢x1, y1²
d ¢x2, y2²iff(x1 dx2 and y1 dy2), and then following the inf/sup strategy of the previous
paragraph. When it comes to choosing designated values for a product system, there are
two obvious choices: ¢x, y² is designated iff x is designated in system X and/or (choose one)
y is designated in system Y. In a self -product system X uX, both choices yield the same set
of tautologies—which is simply the set of tautologies of X [Rescher, 1969, 101].

2.3 Infinitely Many Values

If we wish to countenance more than finitely many truth values, there are two particu-
larly natural options: we can take as truth values all the rational numbers between 0 and
1 inclusive, or the real interval [0, 1] (containing all the real numbers between 0 and
1 inclusive). Łukasiewicz considered both these options. The rules for assigning truth
values to compound propositions stated in §2.2 carry over unchanged. If we take 1 as
the only designated values, the two resulting logics are named Ł and Łc respectively.
0
Interestingly, they both have the same set of tautologies [Rescher, 1969, 38–9].
A semantics which takes [0, 1] as its set of truth values (and treats the connectives
truth-functionally) is called a fuzzy semantics; a logic that arises from a fuzzy semantics is a
fuzzy logic. Among philosophers, the best-known rules for assigning truth values to com-
pound propositions in a fuzzy semantics are the Zadeh rules; the statement of these rules
is exactly the same as that of the Kleene rules in Figure 5.6.7. Among logicians, the fuzzy
systems of most interest are the t-norm fuzzy logics. Where š * is a binary function on [0, 1],
which we are going to use to define conjunction, it is natural to want š *to satisfy the con-
ditions shown in Figure 5.6.13. A binary function on [0, 1] which satises these conditions

α ¬α
0, 0 1, 1
0, 1 1, 0
1, 0 0, 1
1, 1 0, 0

Figure 5.6.12 Negation in the self-product of classical logic

 
x∧y = y∧x
   
(x ∧ y) ∧ z = x ∧ (y ∧ z)
 
x1 ≤ x2 implies x1 ∧ y ≤ x2 ∧ y
 
y1 ≤ y2 implies x ∧ y1 ≤ x ∧ y2

1∧x = x

0∧x = 0
Figure 5.6.13 Conditions on t-norms

644
 M A N Y- VA LU ED L O G I C S

is called a t-norm. If it is also continuous—in the usual sense—then it is a continuous t-

norm. A t-norm fuzzy logic is determined by a choice of a continuous t-norm as the truth
function for conjunction: truth functions for the other connectives are then defined in
terms of this t-norm. The truth function o * for conditional is the residuum of the t-norm:

xo
* y = max{z: x š
* z dy}

and the truth function ™

*for negation is the precomplement of this conditional:

™
*x = xo
* 0

Biconditional is then defined in the standard way using conditional and conjunction.
Disjunction will be discussed below. The most important examples of continuous t-
norms are shown in Figure 5.6.14. These three are fundamental in the sense that every
continuous t-norm is a combination of them [Hájek, 1998, 32]. Note that the Gödel t-
norm is the min operation used to define conjunction in Zadeh/Kleene logic; it is the only
idempotent t-norm (i.e. one which satisfies the condition x š
* x = x) [Klir and Yuan, 1995,
63]. The conditionals and negations derived from these three t-norms (as residuum and
precomplement) are shown in Figure 5.6.15. The Łukasiewicz operations are the familiar
ones—which is why the t-norm that gives rise to them is called the Łukasiewicz t-norm.
Note however that this t-norm is not the min operation used to define conjunction in
Łukasiewicz’s many-valued logics (§2.2): that min operation is the Gödel t-norm, and
the conditional and negation to which it gives rise are not Łukasiewicz’s. The Gödel
operations are named for their discussion in Gödel [1986]. Note that while the Gödel
conjunction is the same as the conjunction in Zadeh/Kleene logic, the conditionals and
negations in these logics are different. The conditional arising from the product t-norm
was discussed in Goguen [1968–69] and is often referred to as the Goguen conditional.


L
 ukasiewicz: x ∧ y = max(0, x + y − 1)

Gödel: x ∧ y = min(x, y)

Product: x∧y =x·y
Figure 5.6.14 Important t-norms


1 if x ≤ y 
 ukasiewicz: x → y =
L ¬x=1−x
1 − x + y if x > y
 
 1 if x ≤ y  1 if x = 0
Gödel: x→y= ¬x=
y if x > y 0 otherwise
 
 1 if x ≤ y  1 if x = 0
Product: x→y= ¬x=
y/x if x > y 0 otherwise

Figure 5.6.15 Residua and precomplements

645
 N I CH O LA S J . J . S MI T H

It is common to add an additional ‘weak’ conjunction to each of these fuzzy log-

ics—with the t-norm conjunction then being termed ‘strong’—defined as follows (with
subscripts w and s indicating strong and weak):

(DšwE  df
Dšs(Do E)

Whatever continuous t-norm is used for šs here (and where ois its residuum), it turns
out that weak conjunction is always the min operation [Hájek, 1998, 36]. Disjunction
also comes in strong and weak forms. Weak disjunction is defined in terms of weak
conjunction and conditional:

(D›wE  df
((Do E) o E) šw ((Eo D) o D)

It always turns out to be the max operation [ibid.]. Strong disjunction is defined as the
dual of strong conjunction:

(x›sy = 1 – ((1 – x) šs (1 – y))

(Where the strong conjunction is a t-norm, this dual operation will always be a t-conorm.
A t-conorm is a binary operation › * on [0, 1] which satisfies conditions exactly like those
for a t-norm—i.e. put › *  for š
* throughout Figure 5.6.13—except that we replace the
final two conditions with 1 › * x = 1 and 0 › * x = x.) In our three systems, this definition
yields the operations shown in Figure 5.6.16. The duality of strong conjunction and
strong disjunction means that where negation is defined as ™x = 1 – x—for example, in
Łukasiewicz t-norm logic—the de Morgan laws will hold (when stated in terms of nega-
tion, strong conjunction and strong disjunction).
Hájek [1998] introduces an axiom system for a logic BL (basic logic) which proves all
and only those formulas which come out as tautologies (taking 1 as the only designated
value) no matter what continuous t-norm is taken as the truth function for (strong)
conjunction. Each of the three specific t-norm logics mentioned above is then axioma-
tized by the addition of an axiom or axioms—different in each case—to BL.
Thus far we have mentioned only propositional logics. Most many-valued seman-
tics for propositional logic can be generalized to predicate logic by following through
the following fundamental idea. We identify a subset S of a background set U with its
characteristic function fS: a function from U to the set of truth values. Where x is an
object in U and y is a truth value, fS(x)= y means that x is a member of S to degree y. So
the truth values now function as values or degrees of membership (of objects in sets), as
well as values or degrees of truth (of propositions). An n-place predicate is assigned an
extension as its semantic value: a subset of the set of n-tuples of members of the domain;
that is, a function from this set of n-tuples to the set of truth values. The truth value of

 ukasiewicz: x ∨s y = min(1, x + y)
L
Gödel: x ∨s y = max(x, y)
Product: x ∨s y = x + y − x · y
Figure 5.6.16 t-conorms/strong disjunctions

646
 M A N Y- VA LU ED L O G I C S

the atomic formula Rab, for example—comprising a two-place predicate R followed by

two names a and b—is then whatever value the extension of R assigns to the ordered
pair comprising the referent of a followed by the referent of b. In order to give rules for
assigning truth values to universally and existentially quantied formulas, we need to
generalize the conjunction and disjunction operations (respectively) so that they assign
values to sets—not just pairs—of values. For details, see Smith [2008, §1.2, §2.2].

2.4 Non-Truth-Functional Systems

All the systems we have looked at so far are truth-functional: the truth value of a com-
pound proposition depends only on the truth values of its components; that is, the
semantics of the connectives are given by associating them with truth functions. One
notable method for assigning multiple values to propositions in a non-truth-functional
way is that of probability logic, where values—probabilities—are assigned in accordance
with the probability calculus. The value of p › q, for example, does not then depend
only on the values of p and q: it also depends on the content of p and q, and in particular
on whether these propositions are mutually exclusive. Interestingly, the set of tautolo-
gies of probability logic (with 1 as the only designated value) is the same as the set of
tautologies of classical logic [Rescher, 1969, 186–7]. Another notable method is that
of supervaluations [van Fraassen, 1966]. We begin with a three-valued model (i.e. a
mapping from basic propositions to the values 0, 1 and *). Say that a classical model M2
extends a three-valued model M3 iff M2 is exactly like M3 except that where M3 assigns
* to a basic proposition, M2 assigns 1 or 0. The rule for assigning truth values to com-
pound propositions is then this: [D] = 1 (0) on M3 iff [D] = 1 (0) on every classical model
which extends M3. So, for example, where p and q both have the value *, p š q will also
have the value *; yet even though ™p will have the same value as q, p š™p will have
the value 0. So the system is not truth-functional. A variant of supervaluationism is
subvaluationism, which differs by having ‘some’ in place of ‘every’ in the rule for assign-
ing truth values to compound propositions: [D] = 1 (0) on M3 iff [D] = 1 (0) on some
classical model which extends M3. If we take 1 as the only designated value, then the
(single-conclusion) consequence relation arising from the supervaluationist semantics
is identical to the classical consequence relation (see e.g. Smith [2008, 82]), while that
arising from the subvaluationist semantics is not (e.g. adjunction fails: {DE } z D š E ).
However, if we move to a multiple-conclusion consequence relation, then the duality
between supervaluationism and subvaluationism is restored: neither yields the classical
consequence relation [Hyde, 1997]. Another variant is the degree-theoretic form of
supervaluationism, which assigns values in the set [0, 1]. We start with a fuzzy model
(i.e. a mapping from basic propositions to [0, 1]), introduce a measure on the set of clas-
sical extensions of this fuzzy model, and then say that the degree of truth of a compound
sentence on the fuzzy model is equal to the measure of the set of classical extensions on
which it is true. For further details see Smith [2008, §2.4.1].

2.5 Many-Valued Systems in the Loose Sense

Some of the motivations for many-valued systems to be discussed in §3 lead most natu-
rally not to the idea of additional truth values, alongside the classical 0 and 1, but to the
idea of propositions possessing neither (or both) of the classical values. This idea leads to
what we termed at the outset many-valued systems in the loose sense. Instead of adding

647
 N I CH O LA S J . J . S MI T H

truth values, we stick with the classical set {0, 1}, but we alter the definition of a model.
Instead of consisting in a function which assigns a truth value to each basic proposition,
it now consists in a relation between the set of basic propositions and the set of truth
values. Where a proposition is related to just one truth value we have the analogues of
classical truth and falsity; where a proposition is related to neither truth value we have a
truth value gap; where a proposition is related to both truth values we have a truth value
glut. Allowing gaps or gluts but not both yields a three-status system; allowing both yields
a four-status system. The question of how to assign truth values to compound proposi-
tions when some of their components have gaps or gluts may be addressed in ways analo-
gous to any of those already considered in relation to three- or four-valued systems: no
essentially new ideas are involved—simply the translation of existing ideas into the new
setting (e.g. we might employ truth tables—with empty and/or multiply-filled cells in
place of third and/or fourth values—or we might employ a version of supervaluationism;
for further details see e.g. Smith [2008, §2.3, §2.4]). Some writers (e.g. Kripke [1975,
700, n.18]; cf. also van Fraassen [1974, 231]) are adamant that there is a deep conceptual
difference between positing a third truth value and allowing truth value gaps, but in
practice it is often more straightforward to formulate systems with additional values
and assignment functions rather than two values and assignment relations, and so even
when the guiding idea is many-valued in the loose sense, the implementation is often
many-valued in the strong sense (e.g. Blamey [1986], Langholm [1988]).

3 Uses of Many-Valued Logics

One of the earliest motivations for many-valued logics—going back to Aristotle
[Łukasiewicz, 1930, 63–4]—was the consideration of modal propositions (e.g. ‘It is pos-
sible that p’) and propositions concerning contingent future events (e.g. ‘I shall be in
Warsaw at noon tomorrow’). In relation to the former, Łukasiewicz [1930, 51] saw that
the semantics of the operator ‘It is possible that’ cannot plausibly be given by any of the
four possible one-place two-valued truth functions; in relation to the latter, he argued
that assigning ‘I shall be in Warsaw at noon tomorrow’ either 1 or 0 as truth value now
conflicts with the fact that it is possible, but not necessary, that I shall be in Warsaw at
noon tomorrow, and he writes that his “three-valued system of propositional logic owes
its origin to this line of thought” [53]. This was also an important motivation for Peirce
and MacColl (see Rescher [1969, 4–5] for references).
The consideration of probabilistic propositions, and in some cases propositions aris-
ing from quantum mechanics, was a motivation for MacColl, Zawirski, Reichenbach,
and Carnap (see Rescher [1969, 14–5, 184–8, 210–1] and Malinowski [1993, 66] for
references).
Consideration of the paradoxes—set-theoretic (e.g. Russell’s) and/or semantic (e.g.
the Liar, where it seems impossible to assign either truth value 1 or 0 to ‘This sentence is
false’)—was a motivation for Bochvar, Moh Shaw- Kwei and others (see Rescher [1969,
13, 29, 207] for references).
Kleene was motivated by a consideration of partial recursive predicates. Think of such
a predicate as coming with an algorithm. Given some objects as input, the algorithm
terminates with the answer ‘Yes’: the predicate is true of these objects. Given some
(other) objects as input, the algorithm terminates with the answer ‘No’: the predicate
is false of these objects. But for some (other) objects, the algorithm does not terminate
with either answer: the predicate is undefined for these objects. Hence Kleene’s three

648
 M A N Y- VA LU ED L O G I C S

truth values: t (true), f (false) and u (undefined). A related application of many-valued

logics is in the analysis of category mistakes arising from predicates which are truly
applicable to some objects, falsely applicable to others, and inapplicable to yet other
objects [Rescher, 1969, 28].
Kleene [1952, 335] also considers a different interpretation of his three values: “t, f, u
must be susceptible of another meaning besides (i) ‘true’, ‘false’, ‘undefined’, namely (ii)
‘true’, ‘false’, ‘unknown (or value immaterial)’. Here ‘unknown’ is a category into which
we can regard any proposition as falling, whose value we either do not know or choose
for the moment to disregard; and it does not then exclude the other two possibilities
‘true’ and ‘false’.” This kind of epistemic application of many-valued logics is also found,
for example, in Belnap [1977].
Many-valued logics have important applications in proof theory. For example, they
can be used to show that certain axioms in a given axiomatization (e.g. of classical
propositional logic) are independent of others. This application originates with Bernays;
for illustrative examples see Malinowski [1993, 105–6] and Mendelson [1997, 43–5]
(and for some examples from modal logic see Lewis and Langford [1932, Appendix
II]). They can also be used for showing the consistency of a set of axioms (i.e. that not
everything can be deduced from it). This application originates with Post [Rescher,
1969, 106].
Like classical logic, many-valued logics have applications in circuit/switching the-
ory; for discussion and references see Malinowski [1993, 109–10] and Hähnle [2001,
§8]. They have also been applied to scheduling problems [Ackermann, 1967, 78].
Many-valued logics have been employed in automated theorem-proving; for a discus-
sion of the many further applications of many-valued theorem provers see Hähnle
[1993, ch.7].
Many-valued logics have been applied to the analysis of presupposition. ‘Bill knows
that Ben is late’ presupposes that Ben is late; ‘The king of France is bald’ presupposes
that there is a king of France; and so on. The many-valued approach to presupposition
holds that if one of a sentence’s presuppositions is false, then the sentence is neither true
nor false. This is known as a semantic approach to presupposition, because it holds that
presupposition failure affects the truth value of a sentence. For further details see Gamut
[1991, 178–90, 212–4].
A closely related use of many-valued logics is in the treatment of singular terms which
lack referents (e.g. names such as ‘Pegasus’ and definite descriptions such as ‘the king of
France’). On this approach, atomic statements involving such terms are neither true nor
false. This idea, which can be found for example in Strawson [1950], was the motivation
for the development of supervaluationism by van Fraassen [1966].
The final use of many-valued logics which we shall mention here is in the treatment
of vagueness, where the basic thought is that statements about borderline cases of vague
predicates (e.g. ‘Bob is tall’, where Bob is neither clearly tall nor clearly not tall) are
neither true nor false. Supervaluationist and fuzzy approaches in particular have played
an important role in the vagueness literature (see Smith [2008] for further discussion
and references).

Acknowledgements
Thanks to Allen Hazen, Libor Běhounek, and Richard Zach for very helpful comments
on an earlier draft. Thanks to the Australian Research Council for research support.

649
 N I CH O LA S J . J . S MI T H

Related Topics
1.4 Presupposition
1.5 Implicature
1.8 Compositionality
1.12 Vagueness
1.13 Empty Names
2.1 Reference
5.8 Intuitionism.

References
Robert Ackermann. An Introduction to Many-Valued Logics. Routledge & Kegan Paul, London, 1967.
Nuel D. Belnap. A useful four-valued logic, 1977. In Dunn and Epstein [1977], pages 8–37.
Merrie Bergmann. An Introduction to Many-Valued and Fuzzy Logic: Semantics, Algebras, and Derivation
Systems. Cambridge University Press, Cambridge, 2008.
Stephen Blamey. Partial logic. In D. Gabbay and F. Guenthner, editors, Handbook of Philosophical Logic,
volume III, pages 1–70. D. Reidel, Dordrecht, 1986.
Leonard Bolc and Piotr Borowik. Many-Valued Logics, volume 1. Springer- Verlag, Berlin, 1992.
J. Michael Dunn and George Epstein, editors. Modern Uses of Multiple- Valued Logic. D. Reidel, Dordrecht,
1977.
D.M. Gabbay and F. Guenthner, editors. Handbook of Philosophical Logic, volume 2. Kluwer, Dordrecht, 2nd
edition, 2001.
L.T.F. Gamut. Logic, Language, and Meaning, volume I. University of Chicago Press, Chicago, 1991.
Kurt Gödel. On the intuitionistic propositional calculus. In Solomon Feferman et al., editor, Kurt Gödel:
Collected Works, volume I, pages 222–5. Oxford University Press, New York, 1986. Originally published
in 1932.
Joseph Goguen. The logic of inexact concepts. Synthese, 19:325–73, 1968–69.
Siegfried Gottwald. A Treatise on Many-Valued Logics. Research Studies Press, Baldock, 2001.
Reiner Hähnle. Automated Deduction in Multiple-valued Logics. Clarendon Press, Oxford, 1993.
Reiner Hähnle. Advanced many-valued logics, 2001. In Gabbay and Guenthner [2001], pages 297–395.
Petr Hájek. Metamathematics of Fuzzy Logic. Kluwer Academic Publishers, Dordrecht, 1998.
Dominic Hyde. From heaps and gaps to heaps of gluts. Mind, 108:641–60, 1997.
Stanisław Jaśkowski. Investigations into the system of intuitionist logic, 1936. In McCall [1967], pages
259–63.
Stephen Cole Kleene. Introduction to Metamathematics. D. Van Nostrand, Princeton, 1952.
George J. Klir and Bo Yuan. Fuzzy Sets and Fuzzy Logic: Theory and Applications. Prentice Hall, Upper Saddle
River, NJ, 1995.
Saul Kripke. Outline of a theory of truth. Journal of Philosophy, 72:690–716, 1975.
Tore Langholm. Partiality, Truth and Persistence. CSLI, Stanford, 1988.
Clarence Irving Lewis and Cooper Harold Langford. Symbolic Logic. Century, New York, 1932.
Jan Łukasiewicz. Philosophical remarks on many-valued systems of propositional logic, 1930. In McCall
[1967], pages 40–65.
Jan Łukasiewicz and Alfred Tarski. Investigations into the sentential calculus, 1930. In Tarski [1956], pages
38–59.
Grzegorz Malinowski. Many-Valued Logics. Clarendon Press, Oxford, 1993.
Storrs McCall, editor. Polish Logic: 1920–1939. Clarendon Press, Oxford, 1967.
Elliott Mendelson. Introduction to Mathematical Logic. Chapman & Hall, London, fourth edition, 1997.
George Metcalfe, Nicola Olivetti, and Dov Gabbay. Proof Theory for Fuzzy Logics. Springer, 2009.
Emil L. Post. Introduction to a general theory of elementary propositions. American Journal of Mathematics,
43:163–85, 1921.
Graham Priest. An Introduction to Non-Classical Logic: From Ifs to Is. Cambridge University Press, Cambridge,
second edition, 2008.
Nicholas Rescher. Many-valued Logic. McGraw-Hill, New York, 1969.
Nicholas J.J. Smith. Vagueness and Degrees of Truth. Oxford University Press, 2008.
P.F. Strawson. On referring. Mind, 59:320–44, 1950.

650
 M A N Y- VA LU ED L O G I C S

Alfred Tarski. Logic, Semantics, Metamathematics: Papers from 1923 to 1938. Clarendon Press, Oxford, 1956.
Translated by J.H. Woodger.
Alasdair Urquhart. Basic many-valued logic, 2001. In Gabbay and Guenthner [2001], pages 249–95.
Bas C. van Fraassen. Singular terms, truth-value gaps and free logic. Journal of Philosophy, 63:481–95, 1966.
Bas C. van Fraassen. The labyrinth of quantum logics. In Robert S. Cohen and Marx W. Wartofsky, editors,
Logical and Epistemological Studies in Contemporary Physics (Boston Studies in the Philosophy of Science,
volume XIII), pages 224–54. D. Reidel, Dordrecht, 1974.
Robert G. Wolf. A survey of many-valued logic (1966–1974), 1977. In Dunn and Epstein [1977], pages
167–323.

Further Reading
For readers whose only background in logic is a typical introductory course or book: Bergmann [2008] pro-
vides a thorough introduction to Bochvar, Kleene and Łukasiewicz three-valued logics, and fuzzy logics.
For readers with greater background in logic: Malinowski [1993], Urquhart [2001] plus Hähnle [2001], and
Gottwald [2001] provide wide-ranging surveys of the field, including important details about consequence
relations, functional completeness, proof theory, decidability and complexity, and other topics of central
importance in logic; for more on fuzzy logics see Hájek [1998] and Metcalfe et al. [2009].
For readers interested in the history of the subject: Rescher [1969] includes a comprehensive bibliography of
work on many-valued logics to 1965; Wolf [1977] extends this to 1974.
For readers wishing to explore applications of many-valued logics with particular relevance to philosophy of
language: Smith [2008] gives an extended argument for the conclusion that the correct account of vague-
ness must involve a many-valued semantics.

651
 5.7
DYNAMIC L OGIC IN
N A T URAL LANGUAGE
Johan van Benthem

1 Dynamic Perspectives in Natural Language

Truth-conditional Semantics
Standard first-order logic defines truth in a three-part scheme: a language, structures D
of objects with relations and operations, and maps from language to structures that drive
semantic evaluation. In particular, “interpretation functions” I map predicate letters to
real predicates, while variable assignments s map individual variables to objects. Logi-
cians often lump D and I together into a “model” M, and then interpret formulas:

formula M is true in model M under assignment s (M, s |= M)

with a recursive definition matching syntactic construction steps with semantic opera-
tions for connectives and quantifiers. This pattern has been applied to natural language
since Montague 1974, stating under which conditions a sentence is true. Compositional
interpretation in tandem with syntactic construction works even beyond logical and
natural languages: it is also a well-known design principle for programs (van Leeuwen
1990). And the paradigm finds an elegant mathematical expression in universal algebra
and category theory.

From Products to Activities

Still, the above semantics merely describes a static relationship between sentences and
the world. But truth is just one aspect of natural language, perhaps not even its crux.
What makes language constitutive of human life seem dynamic acts of assertion, inter-
pretation, or communication. In recent years, such acts have entered logical theory,
from interpretation to speech acts and discourse. This is often considered pragmat-
ics—but often, the natural semantic meaning is the use. In a maxim from mathematics
and computer science: ‘never study representation without transformation’. We can-
not understand the static structure of a language without studying the major processes
it is used for. Our very lexicon suggests a duality between product and process views.
“Dance” is both a verb and a noun, “argument” is an activity one can pursue and its
product that logicians write down.
 DYN A M I C L O G I C

Concrete Dynamic Systems

There are many sources for current dynamic semantics of natural language. One is the
seminal work of Kamp and Heim on anaphoric interpretation of pronouns as creating
discourse representation structures that get modified as speech proceeds, and that can be
matched against reality when the need arises. But the most incisive example has been
“dynamification” of existing logics. First-order logic is a pilot for static truth conditions,
but it can also model essentials of the process of evaluation. The latter involves shifting
relationships between variables and objects (Groenendijk and Stokhof 1991). Consider
the truth condition for an existential quantifier:

M, s |= x M iff there is an object d in M such that M, s[x: = d] |= M.

Intuitively, this calls for a search through available objects d in M for the variable x
until we hit the first d for which M holds. The latter object is then available for fur-
ther reference, as it should, say, in little texts like “xPx. Qx” (“A man came in. He
whistled.”) that support anaphoric reference between the two occurrences of x across
sentence boundaries.
Dynamic ideas work much more widely, with temporal expressions as a major para-
digm (ter Meulen 1995). As for sentence-level processes, sentences can change infor-
mation states of hearers by elimination of all models from a current set that do not
satisfy the formula. This folklore idea of range-with-elimination underlies the account
of conversation in Stalnaker 1978, or the “update semantics” of Veltman 1996, where
meanings are potentials for changing information states. Thus, dynamic linguistic acts
come into the scope of logic, making them amenable to the compositional analysis that
has served truth-conditional views in the past.

Richer Versions: Social Dynamics and Games

Classical semantics has no actors, since it is about bare relationships with reality. Dynamic
semantics is about single agents that compute on discourse structures, or change single
minds. But language is about speakers and hearers that create shared meanings, and over
that channel, engage in meaningful activities, cooperative or competitive. This requires
a study of information flow in multi-agent communication where “social” knowledge
about what others know and mutual expectations are crucial. And beyond single speech
acts, there is a longer-term strategic aspect. I choose my words toward an end, depend-
ing on how I think you will take them, and next, so do you. Such behavior over time is
the realm of game theory. Dynamic logics and games will be discussed below.

Literature
For dynamic semantics, see Dekker 2008, Groenendijk et al. 1996, and for discourse
representation theory, Kamp and Reyle 1993. Other sources include speech act theory
(Searle and Vanderveken 1985) and “score-keeping games” (Lewis 1979). The “Dynam-
ics” chapter in van Benthem and ter Meulen 1997 (updated in 2010 with an appen-
dix on “Information Dynamics”) adds links to dynamic logics, artificial intelligence,
and computational linguistics. Kamp and Stokhof 2008 has extensive philosophical
reflection.

653
 J O H A N VA N B E N T H E M

2 Parallels with Computer Languages and Computation

The Unity of Languages
Philosophy has long known a tension between formal and natural language methods,
with Russell’s “Misleading Form Thesis” claiming that natural language obscures the
logical form of statements. By contrast, Montague proclaimed the unity of formal and
natural languages in their design principles and theoretical properties. Later authors
tested “Montague’s Thesis” on programming languages: formally designed, but driving
a real communicative practice between humans and machines. They found striking
analogies with natural language, from category structure to paradoxes of intensionality.

Semantics of Programs: Computing Change

A major challenge in logics of computation is giving a meaning to imperative programs.
These are not propositions that are true or false but instructions for changing states of
a computer, or indeed any process. As it turns out, first-order assignment semantics is a
great fit here, allowing for a compositional definition of the relation of successful transi-
tion between assignments for program execution:

s1 [[S]]M s2: there is a successful execution of program S starting in s1

and ending in s2.

Here the assignments s, originally an auxiliary device, become important in their own
right, as memory states of a computer. A typical case are atomic programs x:=t, where
the assignment s1 changes to a new s with all values for variables the same, except that
x is now set to [[t]]Ms1, the value of the term t in M under the old assignment s1. In a
picture, we now view meanings in terms of transition arrows between states:

 S


|s
s1 2

Thus, action is identified with pairs <input state, output state>. There are much richer
process views in computer science, but we will use this simple extensional format in
what follows.

Compositional structure
Like propositions, programs have complex syntactic structure, and their interpretation
proceeds inductively: we match them with semantic ones. Here are three basic opera-
tions (the textbook Harel 1987 explains them with kitchen recipes):

Sequential composition S1; S2

Guarded choice IF M THEN S1ELSE S2
Iteration WHILE M DO S

For instance, the semi-colon ; denotes sequential composition of relations. Its transitions
arise from first making a successful transition for S1, and then one for S2. The WHILE
loop is unbounded; we make a computer run for as long as it takes to achieve ¬M.

654
 DYN A M I C L O G I C

Logics of Programs
The oldest computational program logic is Hoare calculus of correctness assertions that
express what a given program does in terms of standard propositions:

{M} S{\} “after every successful execution of program Sstarting from

a state where precondition Mholds, postcondition \holds”:

Note the co-existence of programs S and propositions M: there is no conflict between

statics and dynamics. Also, non-determinism is allowed: a program may have several
executions. Now we can do program logic, much as rules for connectives analyze static
propositions:

Example Rules of the Hoare Calculus.

{M} S1 {\} {\} S2 {F}
{M} S1 ; S2 {F} composition
{M š F} S1 {\} {M š ¬F} S2 {\}
{M} IF F THEN S1 ELSE S2 {\} guarded choice
{M} S {M}
{M} WHILE \ DO S {M š ¬\} iteration

Checking soundness of these inference rules will make you understand a lot about the
logic of change.

From Programs to General Actions

The preceding describes a logic of action for any dynamic event, not just shifts in vari-
able assignments or states of a computer. Logics like this have been applied to natural
language, as we shall see, but also to conversation, strategies in games, and quantum-
mechanical measurements that change a physical system. In this process, a reversal of
perspective has occurred. Thinking of all the dynamic effects of using language, we
might consider the computational stance as primary. Van Benthem 1996 even claims
that natural language is a programming language for cognitive actions.

3 Technical Background: Dynamic Logic of Action

Behind program semantics lies a logic familiar to philosophers studying intensional
notions.

Modal Logics of Process Graphs

Consider process graphs M = (S, {Ra}, V) with a set of states S, a family of binary transi-
tion relations Ra for basic actions (sometimes written o a), and a valuation V interpret-
ing proposition letters p as local properties of states. Over such models, one can interpret
a language with labeled modalities over “action-accessible states”:

M, s |= [a]M iff M, t |= M for all t with s Ra t.

655
 J O H A N VA N B E N T H E M

The dual existential modality <a>M is defined as ¬[a]¬M. Hoare correctness statements
are modal implications M o [S ]\. (A good textbook is Blackburn et al. 2000.) Major
uses of modal logic today are action and knowledge, notions that will return.

Complex Actions: Propositional Dynamic Logic

The same formalism can deal with complex actions. The language now has compo-
nents, one of programs (P) and one of formulas (F):

F := atomic propositions | ¬F | (FšF) | [P]F

P := atomic actions | (P;P) | (P‰P) | P* | F?

For elegance and sweep, program operators are now the regular operations of composi-
tion, Boolean choice, Kleene iteration, and tests for formulas. The semantics matches
the mutual recursion in the syntax. M, s |= I says that I is true at state s, while M, s1,
s2 |= S says the transition from s1 to s2 is a successful execution of program S. Here are a
few key clauses:

• M, s |= [S]I iff for all s’ with M, s, s’ |= S, we have M, s’ |= I

• M, s , s |= a iff (s , s ) Ra
1 2 1 2
M, s , s |= S ;S iff there is an s with M, s , s |= S and M, s , s |= S
1 2 1 2 3 1 3 1 3 2 2
M, s , s |= S ‰S iff M, s , s |= S or M, s , s |= S
1 2 1 2 1 2 1 1 2 2
M, s , s |= S* iff some finite sequence of S–transitions in M
1 2
connects the state s with the state s
1 2
M, s , s |= I? iff s = s and M, s |= I
1 2 1 2 1

Axiomatic System
Propositional dynamic logic has a natural proof system PDL. We give it here to show
how logics of action and change can be designed just like classical logics:

• All principles of the minimal modal logic for all modalities [S]
• Computation rules for decomposing program structure:
[S ;S ]I l [S ][S ]I
1 2 1 2
[S ‰S ]I l [S ]I š [S ]I
1 2 1 2
[I?]\ l (Io \
[S*]I l I š [S][S*]I
• The Induction Axiom (Iš [S*](I o [S]I)) o [S*]I

PDL can derive all Hoare rules, while generalizing modal logic and relational algebra.
All its theorems are valid, and there is a nice completeness proof. And in all this, PDL
is decidable. PDL distils the “essence of computability”: cf. Harel et al. 2000.

Process theories: the larger picture

The above suggests a study of process equivalences, as processes have many natural levels
of detail. Also, the decidability of PDL suggests a balance between the expressive power

656
 DYN A M I C L O G I C

of a logic and the computational complexity of its laws. This balance is also crucial to
understanding natural language. Finally, PDL only studies sequential operations on pro-
grams, while parallel composition is the reality in network computing. This deep subject
is pursued in Lambda Calculus, a system close to Montague’s work, and Process Alge-
bra (Bergstra et al. 2001). Parallel computation is simultaneous action, which can also
be modeled by games (Abramsky 2008). Interestingly, our linguistic performance, too,
is a parallel composition of grasping meanings and engaging in discourse at the same
time.

4 Dynamic Semantics of Natural Language Sentences

Many systems of interpretation highlight actions that deal with anaphora, temporality,
and many other expressions. A major paradigm is Discourse Representation Theory
(Kamp and Reyle 1993), but here we present an approach due to Groenendijk and
Stokhof 1991.

Translation Lore
Dynamic predicate logic (DPL) “dynamifies” first-order logic. Here is why. One is usually
taught some folklore to make first-order formulas fit actual linguistic forms:

1 A man came in. He whistled. The two underlined phrases can co-refer.
2 * No man came in. He whistled. The two underlined phrases cannot co-refer.
3 * He whistled. A man came in. The two underlined phrases cannot co-refer.
4 If a man came in, he whistled. The two underlined phrases can co-refer.

The obvious translation x Cx š Wx for 1 does not give the right scope, and one uses
a bracket trick: x (Cx š Wx). The translation for 2: ¬x Cx š Wx does give the right
scope, the quantifier does not bind the free variable in Wx. The translation for 3: Wx
š ¬x Cx , too, is correct. But the translation for 4: x Cx o Wx has the wrong scope,
and one uses brackets plus a quantifier-change (though the sentence has oas its main
operator): x (Cx o Wx).

Dynamifying Standard First-Order Semantics

DPL assigns dynamic meanings without tricks. It reinterprets first-order formulas I as
evaluation procedures, transition relations between assignments like with programs:

Atoms as tests
M, s1, s2 |= Px iff s1 = s2 and IM(P)(s1(x))
Conjunction as composition
M, s1, s2 |= Iš \ iff there is s3 with M, s1, s3 |= I and M, s3, s2 |= \
Negation as failure test
M, s1, s2 |= ¬I iff s1 = s2 and there exists no s3 with M, s1, s3 |= I
Existential quantification as random reset
M, s1, s2 |= x iff s2 = s1[x:=d] for any object d in the domain.

657
 J O H A N VA N B E N T H E M

Example Dynamic evaluation and bindings explained.

(1) Evaluating x Cx š Wx composes a random reset with two successive test actions.
This moves from states s to states s[x:=d] where both C(d), W(d) hold. (2) Wx š x
Cx composes a test, a random reset, and one more test. This moves from states s where
W(s(x)) holds to states s[x:=d] where C(d) holds: no binding achieved. (3) The non-
binding is explained by the negation test, which leaves no new value for x to co-refer.
(4) To get the implications right, we define Io\ as ¬(I š ¬\). This works out to a
new test: Every successful execution of I can be followed by one for \. This does what it
should for both implications.

Logic as Evaluation Algebra

Conceptually, DPL makes predicate logic a theory of two basic actions: variable resets
and atomic tests. “Standard logic” then becomes a mix of general relation algebra, at the
level of the dynamic logic PDL, plus specific laws for reset actions on first-order models.
Thus we see an intriguing fact. The basic process logic of evaluation is decidable—the
undecidability of first-order logic arises from debatable special mathematical features of
assignments (van Benthem 1996). DPL views also apply to discourse, suggesting notions
of “dynamic inference”: cf. the mentioned sources.

We leave the reader with a thought. If meaning is dynamic, what computational proc-
ess drives natural language? Do typical program structures like WHILE and * iteration
make sense?

5 Logical Dynamics of Conversation

We now move from sentences to discourse, and information flow in communication.

Example Cooperative questions and answers.

I ask you in Amsterdam: “Is this building the Rijksmuseum?” You answer: “Yes.” This is
a simple thing we all do all the time, but subtle information flows. By asking the ques-
tion, I tell you that I do not know the answer, and that I think you may know. And by
answering, you do not just convey a topographical fact—you also make me know that
you know, and as a result, you know that I know that you know, etc.
Common knowledge, at every depth of iteration, mixes factual information and social
information about what others know. The latter is the glue of communication, according to
philosophers, economists, and psychologists. Hence we need logics that treat information
flow with actors on a par. We do this by “dynamifying” the static logic of knowledge:

Epistemic Logic
The epistemic language EL extends propositional logic with modal operators KiI (i knows
that I), for each agent i in a total group I, and CGI: I is common knowledge in the sub-
group G. The inductive syntax rule is as follows:

p | ¬I | I›\| KiI | CGI

This language describes the Question/Answer scenario with formulas like

658
 DYN A M I C L O G I C

(i) ¬KQM š ¬KQ¬M (Q does not know whether M),

(ii) ¬KQ¬(KAM › KA¬M) (Q thinks that A may know the answer).

After communication, we have KAM š KQM, KQKAM š KAKQM, and even C(Q, A}M.

Formally, consider models M = (W, {~i | iG}, V), with worlds W, accessibility rela-
tions ~i for agents iG between worlds, and V a valuation as usual. Pointed models (M,
s) have an actual world s for the true state of affairs (perhaps unknown to the agents).
Here accessibility no longer encodes actions but information ranges: the options agents
see for the actual world. Further conditions on ~i encode special assumptions about
agents’ powers of observation and introspection. Very common are equivalence relations:
reflexive, symmetric, and transitive. Such “information diagrams” interpret the epis-
temic language. Here are the key clauses:

 M, s |= KiI iff for all t with s ~i t: M, t |= I

 M, s |= CGI iff for all t that are reachable from s by some finite
sequence of arbitrary ~i steps (i G): M, t |= I

We draw one model for a simple question answer episode (omitting reflexive arrows).
Agent Q does not know if p, but A is informed about it:

In the actual world (the black dot), the following formulas are true: p, KAp, ¬KQp š
¬KQ¬p, KQ(KAp › KA¬p), C(Q, A} (¬KQp š ¬KQ¬p), C(Q, A} (KAp › KA¬p). This is a good
reason for Q to ask A about p: He knows that she knows the answer.

p Q ¬p

Figure 5.7.1

After the answer “Yes”, intuitively, this changes to the following one-point model:

Figure 5.7.2

Now, common knowledge C{Q, A} p holds at the actual world. Epistemic logic sharpens
intuitive distinctions about information, especially levels of group knowledge. Commu-
nication often turns implicit group knowledge into explicit knowledge.

Axiom Systems for Epistemic Inference

Complete logics capturing epistemic reasoning about oneself and others are known
(Fagin et al. 1995). The base system is a minimal modal logic. Structural restrictions to
equivalence relations add S5 axioms of introspection, while the complete logic of com-
mon knowledge can be axiomatized with PDL-techniques.

659
 J O H A N VA N B E N T H E M

A Dynamic Turn: Public Update by Elimination

Now for the logical dynamics of information flow. A pilot system for exploring this
starts from a folklore view: An event !P yielding the information that P is true shrinks
the current model to just those worlds that satisfy P. This is called public hard information.
More precisely, for any epistemic model M, world s, and P true at s, the model (M|P, s)
(‘M relativized to P at s’) is the submodel of M whose domain is the set of worlds {tM |
M, t |=P}. Drawn in a simple picture, an update step then goes

from M s to M|P s

P ¬P

Figure 5.7.3

This mechanism models communication, but also acts of public observation. It has
been applied to games and other social scenarios. Crucially, truth values of formulas
may change after update: agents who did not know that P now do. We need a logic to
keep things straight.

Dynamic Logic of Public Announcement

The language of public announcement logic PAL adds action expressions to EL, plus
matching dynamic modalities, defined by the syntax rules:

Formulas P: p | ¬I | I›\| KiI | CGI | [A]I

Action expressions A: !P

Here the semantic clause for the dynamic action modality “looks ahead” between dif-
ferent models:

M, s |= [!P]I iff if M, s |= P, then M|P, s |= I

A typical assertion here is [!P]KiI which states what agent i will know after
receiving hard information that P is the case. Reasoning about information flow revolves
around a dynamic recursion equation that relates new knowledge to old knowledge an
agent had before:

The following equivalence is valid for PAL: [!P]KiIl(P oKi(P o [!P]I

The reader may find it helpful to prove this. The complete and decidable logic for
knowledge under public communication is well understood. PAL is axiomatized by any
complete epistemic logic over static models plus recursion axioms

660
 DYN A M I C L O G I C

[!P]q l Poq for atomic facts q

[!P]¬I l Po¬[!P]I
[!P]Iš\ l [!P]Iš [!P]\
[!P]KiI l P oKi(P o [!P]I

There is more here than meets the eye. The logic PAL uncovers many subtleties of
natural language. Suppose that in our question-answer episode, A had not said !P, but
the equally true “You don’t know that P, but it is true” (¬ KQP š P). The latter “Moore
sentence” achieves the same update, but it has become false afterward! Statements
switching their own truth values are essential in understanding conversation, puzzles,
and games (Geanakoplos and Polemarchakis 1982). For much more about PAL and
related systems, including links to epistemology, cf. Baltag et al. 1998, van Ditmarsch et
al. 2007, van Benthem 2011.

Compared to dynamic semantics, logical dynamics of information flow has a dis-

course focus. This is “pragmatics”—but from a logical point of view, the border with
semantics is thin.

6 The Logical Dynamics of Agency

Public announcement is just the start of a dynamics of interactive agency. We mention
a few dimensions of a richer picture of what language users are and do.

From Knowledge to Belief

Language users do not just have knowledge, but also beliefs. What they hear involves
belief revision (Gaerdenfors 1988), and this process can be triggered by information
that is “soft” rather than hard, depending on the reliability of the source. Dynamic
logics for belief revision “dynamify” static doxastic logics, where agents believe what
is true in their “most plausible” worlds. This time, update does not eliminate worlds: It
transforms the relative plausibility that agents assign to worlds. Forming and correcting
beliefs is a learning ability that is more essential to human intelligence than just record-
ing hard information. Rationality is not being correct all the time, but having a talent
for correction.

Private Information
Information flow is driven by a differential: we do not all know the same things.
Dynamic epistemic logics can also model private communication in a group (think of
emails with bcc rather than cc), a phenomenon of high complexity. They even deal
with lying and cheating, a central topic in real language, since most communication is
unreliable to some extent. The usual Gricean focus on helpful truthful communication
seems otherwordly.

Questions and Issue Management

Questions do not just convey information, they also direct discourse by raising and
modifying topics. This is crucial to language, communication and inquiry. A logical

661
 J O H A N VA N B E N T H E M

dynamics of questions must represent issues and actions modifying these. Two recent
flavors are Groenendijk 2009, van Benthem and Minica 2009.

Preference and Evaluation Dynamics

Agency is not just information dynamics. Rational decision and strategic interaction
involve a balance of information and evaluation, encoded in our preferences. Entan-
glement of information and evaluation, and preference change pervade deontic logic
(Gruene Yanoff and Hansen 2010) and games. But again natural language remains close.
We often change our evaluation of situations by speech acts such as suggestions or com-
mands from some moral or esthetic authority. Philosophers have also drawn attention to
the normative character of “discourse obligations” (Brandom 1994).

And Beyond
Further relevant features of agency studied in logic include trust, intentions and commit-
ments, but we the picture should be clear by now. Dynamic analysis of language may at
the same time have to be an analysis of the agents using that language.

7 Games in Logic and Natural Language

Beyond logics of agency lies a world of games. Dynamic logics describe single update
steps, but the next level of language use is strategic behavior over time. What I say now
responds to what you say, but it is usually directed toward a future goal, and part of a
long-term plan. We also saw that language involves iterated social knowledge of agents
about each other and multi-agent equilibrium. Both have received sophisticated treat-
ments in game theory.

Evaluation Games and “Game-Theoretic Semantics”

We illustrate games for natural language with a simple pilot system, again in terms of
first-order logic (Hintikka and Sandu 1997). Evaluation of formulas I can be cast as a
game for two players. Verifier V claims that Iis true in a model M, s, Falsifier F that it is
false. Here are natural moves of defense and attack:

atoms Pd, Rde, . . . V wins if the atom is true, F if it is false

disjunction I›\ V chooses which disjunct to play
conjunction Iš\ F chooses which conjunct to play
negation ¬I role switch between the players, play continues with I

Next, the quantifiers make players look inside M’s domain of objects:

x I(x) V picks an object d, and play continues with I(d)

x I(x) the same move, but now for F

The game schedule is determined by the form of I. Consider a model M with two
objects s, t. Here is the complete game for the formula x y xzy in M, pictured as a tree
of possible moves:

662
 DYN A M I C L O G I C

F
x:= s x:= t
V V
y:= s y:= t y:= s y:= t
loseV winV winV loseV
This is a game of perfect information: players know throughout what has happened.
Branches are the possible plays, with 2 wins for each player. But V is the player with a
winning strategy, she has a rule for always winning. For a more exciting example, look at
the following network with arrows for directed communication links (all self-loops are
present but not drawn):

1e f2
c c
 

e
3e 4
The formula x y (Rxy ›z(Rxz š Rzy)) says that any two nodes can communicate in
two steps. Just analyze the game and see who can win.

Logic and Games

The crucial fact about evaluation games is this equivalence:

A formula is true iff Verifier has a winning strategy,

while it is false iff Falsifier has a winning strategy. This follows from Zermelo’s Theo-
rem on “determined games,”, a stepping stone toward solution procedures for games
with richer preferences than winning and losing (“Backward Induction”). In evalua-
tion games, logical constants change from “control expressions” for procedures to game
actions like making a choice or performing a role switch. Thus, at the heart of natural
language, there is a multi-agent game algebra of its users. Actually, in game theory, the
norm is imperfect information: Players need not know exactly where they are in a game
tree (think of card games). Hintikka and Sandu 1997 claim that “branching patterns”
of independence between quantifiers in natural langage involve imperfect information
about objects chosen by one’s opponent.
Evaluation games are not a realistic account of discourse, which is usually about con-
sistency rather than truth. But modern logic has many further games that are relevant.

“Logic Games”
Games in logic analyze argumentation (Lorenzen 1955), compare models (Ehrenfeucht
1957), construct models (Hodges 1985), etc. In each case, winning strategies for various
agents encode basic notions. A winning argumentation strategy is a proof for the claim,

663
 J O H A N VA N B E N T H E M

or if the opponent wins, a counter-model. Games are at the heart of modern logic, and
quantification is deeply tied to dependent action.

Signaling Games for Meaning

A quite different use of games in language has emerged in Parikh 2000, Jaeger and van
Rooij 2007, and Gaerdenfors and Warglien 2006. These start from the signaling games
in Lewis 1969 that analyze basic lexical meanings. We have situations and linguistic
objects that can represent them. Agents might choose any association, but stable con-
ventions are Nash Equilibria in a game where a Sender chooses a coding scheme and
Receiver a decoding, with some plausible assumptions on their utility functions. Thus
meanings become equilibria in language games. Richer infinite “evolutionary games” can
even explain diachronic phenomena, or emergence of linguistic conventions, using
thought experiments in terms of fitness and stability against invaders.

Integrating Different Games

Signaling games are very different from logic games, where meanings are given. Integra-
tion of these perspectives on natural language is an open problem.

Logic and Game Theory

We have discussed special games for linguistic and logical activities. But there is also an
interface of logic and general game theory, in the study of strategies, information, and
reasoning of agents. This involves epistemic, doxastic, and dynamic logics for analyzing
rational play and game solution. This relates to the sense in which computer scien-
tists have embraced games in multi-agent systems (Shoham and Leighton-Brown 2008),
and philosophers in epistemology (Stalnaker 1999). While this interface is not disjoint
from language and logic games, we will not pursue it here: cf. van der Hoek and Pauly
2006 for this area of investigation.

Coda: Temporal Perspective

Lexical meaning assignment, evaluation, or argumentation are special-purpose short-
term processes. These run against the backdrop of an infinite linguistic process over
time: the “operating system” of natural language. Here language dynamics meets with
temporal logics (Parikh and Ramanujam 2003, Belnap et al. 2001), learning theory
(Kelly 1996), and infinite computational processes (Grädel et al. 2003).

8 Discussion: Putting the Dynamics Together

Dynamic Semantics Versus Dynamic Logic

Dynamics in this chapter has two different strands. “Dynamic semantics” is a new account
of meaning, replacing truth-conditional accounts, and generating “nonstandard logics.”
By contrast, dynamic logics of information update keep the old language with its seman-
tics and logic, but add dynamic events as a new layer. The former approach is implicit:
The dynamics “loads” the meaning of the old language, while the latter approach is

664
 DYN A M I C L O G I C

explicit, the dynamics occurs in operators extending the old language. The implicit/
explicit contrast occurs widely in logic—but what fits natural language best?

Combined Architectures
How can we turn our carousel of dynamic activities and games into one integrated story
of language use? Combining logics can be tricky. Simple decidable logics for knowledge
and time combine into highly undecidable logics for agents with perfect memory (Halp-
ern and Vardi 1989). No integration is known for dynamics, and we may first need an
account of “linguistic agents”, the way Turing analyzed “computing agents”.

Cognitive Realities
Finally, natural language is an interface where logic meets reality—and so dynamic log-
ics meet cognitive science. Van Benthem 2010 proposes studying language in a broad
sense here, including “successful insertions” of new logic-inspired behavior.

9 Conclusion
We have shown how natural language meets with dynamic logics of meaning and
agency, leading to new interfaces between logic, philosophy, linguistics, computer sci-
ence and game theory.

References
Abramsky, S. (2008) ‘Information, Processes and Games’, in P. Adriaans and J. van Benthem (eds) Hand-
book of the Philosophy of Information, Amsterdam: Elsevier, 483–549.
Baltag, A, L. Moss and S. Solecki (1998) “The Logic of Public Announcements, Common Knowledge and
Private Suspicions,” in Proceedings TARK 1998, Los Altos: Morgan Kaufmann, 43–56.
Belnap, N., M. Perloff, and M. Xu (2001) Facing the Future, Oxford: Oxford University Press.
Bergstra, J., A. Ponse and S. Smolka (eds.) (2001) Handbook of Process Algebra, Amsterdam: Elsevier.
Blackburn, P., M. de Rijke, and Y. Venema (2000) Modal Logic, Cambridge: Cambridge University
Press.
Brandom, R. (1994) Making It Explicit: Reasoning, Representing, and Discursive Commitment, Cambridge, MA:
Harvard University Press.
Dekker, P. (2008) “A Guide to Dynamic Semantics,” Amsterdam: Institute for Logic, Language and
Computation.
Ehrenfeucht, A. (1957) “Application of Games to Some Problems of Mathematical Logic,” Bulletin Polish
Academy of Sciences Cl. III, 35–37.
Fagin, R., J. Halpern, M. Vardi, and Y. Moses (1995) Reasoning about Knowledge, Cambridge, MA: MIT
Press.
Gaerdenfors, P. (1988) Knowledge in Flux, Cambridge, MA: MIT Press.
Gaerdenfors, P. and M. Warglien (2006) “Cooperation, Conceptual Spaces, and the Evolution of Seman-
tics,” in: P. Vogt et al. (eds.) Symbol Grounding and Beyond, Heidelberg: Springer, 16–30.
Geanakoplos, J. and H. Polemarchakis (1982) “We Can’t Disagree Forever,” Journal of Economic Theory 28,
192–200.
Grädel, E., W. Thomas and Th. Wilke (eds.) (2003) Automata, Logics, and Infinite Games, Heidelberg:
Springer.
Groenendijk, J. (2009) “Inquisitive Semantics: Two Possibilities for Disjunction,” in P. Bosch, D. Gabelaia
and J. Lang (eds.) Proceedings Seventh Tbilisi Symposium on Language, Logic and Computation, Heidelberg:
Springer, 80–94.
Groenendijk, J. and M. Stokhof (1991) “Dynamic Predicate Logic,” Linguistics and Philosophy 14, 39–100.

665
 J O H A N VA N B E N T H E M

Groenendijk, J., M. Stokhof, and F. Veltman (1996) “Coreference and Modality,” in S. Lappin (ed.) Hand-
book of Contemporary Semantic Theory, Oxford: Blackwell, 179–216.
Grüne Yanoff, T. and S.-O. Hansen (2010) Preference Change, Dordrecht: Springer.
Halpern, J. and M. Vardi (1989) “The Complexity of Reasoning about Knowledge and Time,” Journal of
Computer and System Sciences 38, 195–237.
Harel, D. (1987) Algorithmics: The Spirit of Computing, Reading, MA: Addison-Wesley.
Harel, D., D. Kozen, and J. Tiuryn (2000) Dynamic Logic, Cambridge, MA: MIT Press.
Hintikka, J. and G. Sandu (1997) “Game-Theoretical Semantics,”,in: J. van Benthem and A. ter Meulen
(eds.) Handbook of Logic and Language, 361–410.
Hodges, W. (1985) Building Models by Games, Cambridge: Cambridge University Press.
Jaeger, G and R. van Rooij (2007) “Language Structure: Psychological and Social Constraints,” Synthese
159, 99–130.
Kamp, H. and U. Reyle (1993) From Discourse to Logic, Dordrecht: Kluwer.
Kamp, H. and M. Stokhof (2008) “Information in Natural Language,” in P. Adriaans and J. van Benthem
(eds.) Handbook of the Philosophy of Information, Amsterdam: Elsevier, 49–111.
Kelly, K. (1996) The Logic of Reliable Inquiry, Oxford: Oxford University Press.
Lewis, D. (1969) Convention, Cambridge, MA: Harvard University Press.
Lewis, D. (1979) “Scorekeeping in a Language Game,” Journal of Philosophical Logic 8, 339–359.
Lorenzen, P. (1955) Einführung in die Operative Logik und Mathematik, Berlin: Springer Verlag.
Meulen, A. ter (1995) Representing Time in Natural Language: The Dynamic Interpretation of Tense and Aspect,
Cambridge, MA: MIT Press.
Montague, R. (1974) Formal Philosophy, New Haven CT: Yale University Press (edited by R. H.
Thomason).
Parikh, P. (2000) The Use of Language, Stanford: CSLI Publications.
Parikh R. and R. Ramanujam (2003) “A Knowledge-Based Semantics of Messages,”Journal of Logic, Lan-
guage and Information 12, 453–467.
Searle, J. and D. Vanderveken (1985) Foundations of Illocutionary Logic, Cambridge: Cambridge University
Press.
Shoham Y. and K. Leyton-Brown (2008) Multiagent Systems: Algorithmic, Game Theoretic and Logical Founda-
tions, Cambridge: Cambridge University Press.
Stalnaker, R. (1978) “Assertion,” in P. Cole (ed.) Syntax and Semantics 9, New York: Academic Press,
315–32.
Stalnaker, R. (1999) “Extensive and Strategic Form: Games and Models for Games,”Research in Economics
53, 293–291.
Van Benthem, J. (1996) Exploring Logical Dynamics, Stanford: CSLI Publications.
Van Benthem, J. (1999) Logic in Games, Amsterdam: Institute for Logic, Language and Computation. To
appear in Texts in Logic and Games, Springer.
Van Benthem, J. (2011) Logical Dynamics of Information and Interaction, Cambridge: Cambridge University
Press.
Van Benthem, J. and S. Minica (2009) “Dynamic Logic of Questions,” in J. Horty and E. Pacuit (eds.), Pro-
ceedings LORI-2 Chongqing, Heidelberg: Springer Publishers, 28-42. To appear in Journal of Philosophical
Logic, 2011.
Van Benthem, J. and A. ter Meulen, eds. (1997) Handbook of Logic and Language, Amsterdam: Elsevier Sci-
ence Publishers.
van Ditmarsch, H., W. van der Hoek and B. Kooi (2007) Dynamic Epistemic Logic, Dordrecht: Springer.
Van der Hoek, W. and M. Pauly (2006) “Modal Logic for Games and Information,”, in P. Blackburn, J.
van Benthem, and F. Wolter (eds.) Handbook of Modal Logic, Amsterdam: Elsevier Science Publishers,
1077–1148.
Van Leeuwen, J. (ed.) (1990) Handbook of Theoretical Computer Science, Amsterdam: Elsevier.
Veltman, F. (1996), “Defaults in Update Semantics”, Journal of Philosophical Logic 25, 221–261.

666
 5.8
I NTUITI ONISM
Allen Hazen

1
Intuitionism is a viewpoint in the philosophy of mathematics, developed by the Dutch
mathematician L.E.J. Brouwer, which rejects many standard (“classical”) mathemati-
cal results and methods, and in particular rejects as invalid some principles of standard
(“classical,” two-valued) elementary logic. It is held by a minority of mathematicians
(a minority even of those willing to go public with philosophical statements) but has
inspired much mathematical research: the term ‘intuitionism’ is also used for the body
of mathematics developed in accordance with the strictures of the philosophy, a body
by now of sufficient mathematical richness to be of interest even to researchers not
concerned with philosophical issues. The term ‘constructivism’ is also used for similar
philosophical viewpoints and mathematical methods: when a contrast is made between
them, ‘intuitionism’ refers to a viewpoint or a set of mathematical methods more like
those of Brouwer himself and ‘constructivism’ more generically to these and related
ones. ‘Intuitionistic (or ‘constructive’) logic’ refers to a specific version of propositional
and first-order logic first formalized by Brouwer’s younger colleague Arend Heyting.
There doesn’t seem to be any useful analogy with the views known as intuitionistic in
such areas of philosophy as meta-ethics, nor is constructive mathematics connected to
what is called ‘social constructivism’ about mathematics and mathematics education.
The mathematical notion that is perhaps most central to the motivation of the intu-
itionistic philosophy and constructivist mathematics is that of a constructive proof,
which can be illustrated with elementary geometrical examples. Typical existence proofs
in Euclidean geometry are given by explicit construction. To show that, for example, for
any triangle there is a circle passing through its three vertices, one specifies a sequence of
operations with (idealized) straight-edge and compass which will result in the drawing
of exactly this circle. Generalizing: to prove, constructively, that a certain mathemati-
cal object exists, you present it, or at least present a method (the construction) which
will, provably, find it. For an example of a nonconstructive proof, consider the theorem
that the closed unit interval is sequence compact: for any infinite sequence of points
in [0,1], there exists a subsequence (i.e. an infinite sequence of points all occurring, in
the same order though perhaps with gaps, in the original sequence) that converges to
a point in the interval. Proof: divide the interval into two (overlapping) subintervals,
[0,.5] and [.5,1], each of length a half: at least one will contain infinitely many points of
the sequence. Now divide this subinterval (or, say, the left one if both contain infinitely
many points of the sequence), and divide it into two subintervals of length a quarter.
. . . Thus there is an infinite sequence of successively shorter nested closed intervals
 A LLEN H A Z E N

(the n-th being of length 1/2n): their intersection contains a single point. Now consider
the subsequence of the original sequence of points whose first point is the first from the
original sequence lying in the chosen half, whose second is the next point in the origi-
nal sequence lying in the chosen quarter, and so on: this subsequence converges to the
point in the intersection of the intervals. This is a perfectly acceptable proof in standard
mathematics, but it gives no hint as to where the point of convergence is! Even if the
original sequence of points is specified in some explicit manner (say by presenting an
algorithm computing their location one after another), there may be no equally explicit
way of displaying the converging subsequence or locating the point of convergence. (If
giving an algorithm is our standard for explicit specification, this follows immediately
from the unsolvability of the Halting Problem for Turing Machines.)
There is an obvious sense in which a constructive proof is more informative than a
non-constructive. Quite independent of any philosophical commitments, a mathema-
tician might prefer constructive proofs, might find nonconstructive proofs unsatisfying,
might even feel that a nonconstructive proof was somehow “not cricket” as mathematics:
thus Paul Gordan’s legendary “This is not mathematics! This is theology!” in response
to a nonconstructive existence proof by David Hilbert. (The context and intention of
Gordan’s supposed exclamation aren’t clear; the saying has become legendary, though,
because it is felt to express an attitude that was more than idiosyncratic.)
Brouwer was not interested in formal logic but saw that if nonconstructive proofs
are rejected as invalid, then some traditional principles of logic are “untrustworthy.”
Proof of the negation of an assumption by reductio (showing that from the assumption
absurdity can be derived) was all right: how else are you to prove a negation? The closely
related indirect proof, however—inferring an arbitrary, not necessarily negative, conclu-
sion when absurdity is derived from the assumption of its negation—was not: using
indirect proof to establish an existence theorem is the archetype of the nonconstruc-
tive. Other principles could be rejected because if accepted they would license indirect
proof. Thus double negation elimination (from Not-Not-A to infer A) would give indirect
proof with one extra step: use reductio to establish the negation of the negation of an
existence claim, and then d.n.e. Similarly for the Law of Excluded Middle (either A
or Not-A, for an arbitrary proposition A): it would allow us to argue “Either the exist-
ence claim is true or it isn’t; the latter is absurd, so the former must hold.” (Note how
the L.E.M. is assumed in the nonconstructive proof of sequence compactness above:
to show that, an arbitrary interval containing infinitely many points of the sequence,
either its left half or its right must contain infinitely many, we must assume of each half
that either it contains infinitely many or it doesn’t contain infinitely many.) Addition
of any of these principles to intuitionistic logic yields a formulation of classical, two-
valued, logic.
A subtler, but perhaps more fundamental, point involves quantifiers. Suppose you
have established that for every object x in the domain, either A holds or Fx, where A
is some statement not containing the variable x. (Typical examples of this form arise
when F is a “decidable” predicate, one for which it holds for every x that either Fx or
Not-Fx: let A be the statement that there exists some y for which Not-Fy.) Classically
you can infer that either A holds or it holds that for every x Fx. (This is the converse
of one of the prenexing operations, moving a universal quantifier into one disjunct
of a disjunction.) Intuitionistically you can’t: statements with embedded quantifiers
are not in general equivalent, intuitionistically, to prenex statements. As with the
three principles mentioned above, if this inference were allowed it would license some

668
 I N TU I TI ON I S M

nonconstructive proofs: where F is a decidable predicate, it would allow a nonconstruc-

tive proof of the existence of a non-F object. It can, however, be added to intuitionistic
first-order logic without affecting the purely propositional fragment of the logic; the
resulting logic, intermediate in strength between intuitionistic and classical, is often
called “constant domain” (for short: CD) logic, as it is characterized in Kripke’s seman-
tics by models with a constant domain across “possible worlds”.
There is actually some debate about the status of this last principle among construc-
tivist mathematicians. In my rough and informal initial characterization of constructive
proof I said that presenting an algorithm that will (provably) find an example counts
as a constructive proof of an existence theorem. Now let F be a decidable predicate of
natural numbers, in the sense that there is an algorithm for deciding, with respect to a
given natural number, whether F holds of it or not. (Intuitionists would take any proof
that such an algorithm works as establishing that, for every natural number n, either Fn
or Not-Fn, so any predicate decidable in this sense is also decidable in the sense sug-
gested in the previous paragraph.) Then if it is proven (by reductio) that it is not the
case that for every n, Fn, we could (and a classical recursive function theorist would) say
that there is an algorithm for finding a non-F number: just test the natural numbers, one
after another, for F-ness until the first counterexample appears! Some mathematical
constructivists (in particular the Russian school stemming from the work of A. Markov)
would count this as a constructivistically acceptable proof of the existence of a non-F
number; Brouwerian intuitionists wouldn’t. Presenting an algorithm doesn’t, for the
stricter constructivists like Brouwer, count as a constructive proof unless the proof the
algorithm will work is itself constructive.

2
One could feel that constructive proofs are more interesting than nonconstructive, or
deem them more valuable, or even argue that mathematics journals should not waste
their page space by publishing nonconstructive proofs, without holding that they are
literally invalid, that they fail to establish the truth of their conclusions. It is this claim
that makes intuitionism or constructivism interesting for philosophers not specifically
concerned with the methodology of mathematics, but it is not obvious how it should
be justified, and Brouwer’s own exposition is on the Delphic side. We can approach the
issue by considering disjunctive statements, such as instances of the law of excluded
middle. As a starting point, on the natural interpretation, a disjunction is true just in
case at least one of its disjuncts is. (As the many proposals for the use of Supervaluations
in semantics show, this is a defeasible principle, but I will take it as a starting point in
this chapter.) Similarly, an existential quantification is naturally thought of as true only
if it has some true instance (though here there is the complication that the “witness”
verifying the existential quantification might not have a name). If we can prove that
one of the disjuncts is true, then we may obviously infer that the disjunction is; this
principle is shared by intuitionistic and classical logic. In the other direction, classical
logic holds some disjunctions to be true, or even logically valid, even though neither dis-
junct is provable: the law of excluded middle holds even for propositions which are nei-
ther provable nor refutable. Intuitionistic logic is more circumspect. No disjunction is a
logical theorem of intuitionistic logic unless at least one of its disjuncts is, and, simi-
larly, no existential quantification is unless some specific instance is provable. Nor can
a disjunctive (or existential) conclusion be derived unless one or more premisses are of

669
 A LLEN H A Z E N

disjunctive or existential character (in a sense that is a bit picky to make precise but
intuitively clear). Nor is this a matter of logic alone: natural formal axiomatizations
of intuitionistic mathematical theories (including, centrally, what is called “Heyting
Arithmetic”: essentially first-order Peano Arithmetic with intuitionistic logic) permit
the proof of disjunctive (existential) theorems only when disjuncts (instances) are
provable.
These results about intuitionistic formal systems are striking, and are often cited as
confirming the “constructive” character of the systems, but they suggest something more.
Particular formal axiomatizations cannot be taken as exhaustive of the notion of intuti-
tionistic provablility, any more than a particular axiomatization exhausts that of prov-
ability in classical mathematics. (Gödel’s incompleteness theorems carry over directly.
The consistency of Heyting Arithmetic is unprovable in Heyting Arithmetic, just as the
consistency of classical P.A. is unprovable in P.A., but would be affirmed by intuition-
ists just as classical mathematicians would affirm the consistency of P.A.) Still, it is a
tempting conjecture that “real” intuitionistic provability satisfies the same “constructiv-
ity” criteria: disjunctions provable only when disjuncts are, existential quantifications
only when instances are. What this suggests is that truth and proof are more intimately
connected than they are on the classical picture. For a classical logician, proof is a valu-
able but not always applicable way of coming to know that a proposition is true, but the
truth of the proposition doesn’t depend on the existence of a proof. (Every instance of
the law of excluded middle has at least one true disjunct, but in many cases no proof of
either disjunct is available.) The intuitionist, however, suggests that no proposition is
true unless it has a proof: perhaps, in some sense, it is the proof that makes it true.
One way of interpreting this close relationship between truth and proof is, in a broad
sense, epistemological. A number of philosophical commentators on intuitionism, most
notably Michael Dummett, have argued that what we understand in understanding a
mathematical proposition involves essentially a grasp of how, in princple, it could be
proven. As a result it does not make sense to attribute truth to a mathematical proposi-
tion unless we could, in principle, come to know it: could prove it. Some, Dummmett
again being the most prominent, have attempted to generalize this conception beyond
mathematics, claiming that the understanding of any proposition depends on a grasp of
how it can be verified, and that truth can, therefore, never “transcend” what we could
in principle come to know. This has led to what is known as “Anti-Realistic” or “Assert-
ibility” semantics: probably this is the main reason why philosophers of language have
taken as much interest in intuitionism in recent decades as they have. Inevitably, given
the attempted generalization, much of the literature in the philosophy of language is
not specifically concerned with mathematical intuitionism or the ways in which math-
ematical theorizing and subject matter are special.
Rather than attempting to summarize this literature, I would like to describe a few
features of intutitionistic mathematics and logic that seem to me of particular philo-
sophical interest. The broadly epistemological interpretation just alluded to, focusing
on what it is to understand a mathematical proposition, is attractive in many ways,
particularly in its recognition of the centrality of the notion of (constructive) proof in
intuitionism. A different, more ontological, interpretation is, however, also possible,
and I think gives a simpler way of understanding the peculiarities of intuitionistic math-
ematics and of motivating intuitionistic logic.
Mathematical theorizing, at least in its superficial form, appears to be about abstract
entities having, in general, no physical (or psychological) embodiment. To take the

670
 I N TU I TI ON I S M

simplest example, the natural number series is infinite: after every natural number
there is another. Many, many, philosophers have been sceptical of the existence
of such entities. Whitehead and Russell, for example admitted that their Axiom of
Infinity (needed in Principia Mathematica to guarantee the endlessness of the number
series) was dubious, and Dedekind’s ingenious attempt to prove the existence of an
infinite “system” by citing a series of ideas tends to be dismissed nowadays, often
with sarcastic comments about how even Dedekind couldn’t hold infinitely many
thoughts in mind simultaneously! We have, however, clear enough ideas about how,
for example, the number series should go on that we can have a principled math-
ematical theory about it even if it doesn’t exist. One might, and on the ontological
interpretation intuitionists do, hold that reasoning about “imaginary” entities should
be governed by a different logic from reasoning about things we are confident really
exist. In reasoning about real things, the law of excluded middle is unobjectionable:
one disjunct or the other of an instance of the law will be true in virtue of the facts
about the objects referred to, even if we don’t (or can’t) know which. When the
objects we reason about don’t exist, this justification of the law of excluded middle
is no longer available. It is plausible that the “truth” about nonexistent objects is
simply whatever follows from our conception of them, so that it is appropriate to use
a logic which allows truth and provability to coincide. This point of view, I would
argue, motivates intuitionistic logic as well as the epistemological interpretation, and,
indeed, motivates it for “liberalized” forms of constructivism which are not covered by
the epistemological interpretation.

3
For Brouwer some of the objects of mathematics, namely the “mental constructions”
someone has already consciously effected, are real. If one clearly perceives (with the
mind’s eye) some finite pattern—the solution to a chess problem, for example—then
it is a real thing about which we can reason classically. Over time we can effect more
constructions: create new patterns. (A platonist, holding that forms have an abstract
existence independent of any physical or mental realization, would say we don’t really
create patterns, but intuitionists are not platonists.) The whole infinite set of all possible
finite patterns, however, does not exist. The universe the mathematician reasons about,
then, can be thought of as, not completed, but under construction. So, how should we
reason about something that is still under construction?
A nonmathematical example can illustrate the principle. Suppose you are an ethical
property developer: you want to sell properties in a new suburb your company is devel-
oping, but you also want your sales representatives to be scrupulously honest when they
describe them to potential buyers. Honesty, however, doesn’t involve describing the
current state of the development: potential buyers aren’t interested in the bulldozers
currently at work, but rather in what the neighborhood will look like when they move
in. What the sales representatives must do, then, is to describe things (honestly) in
language appropriate to the description of an already-built suburb, even though much
of it is still unbuilt, and though some details will depend on decisions not yet made.
Obviously, anything that has already been built can be described: the street pattern,
perhaps, and the model houses that have been built on one block, but we will want to
go beyond that. The fundamental constraint is that the sales representative should not
say anything that might be falsified later in the construction process.

671
 A LLEN H A Z E N

On the basis of these considerations we can formulate a recursive definition of what

will count as a true (“honest”) description of an unbuilt neighborhood. As a base
case we take the principle that what has already been completed can be described.
The recursive clauses for conjunction, disjunction, and existential quantification are
completely standard: a statement of one of these forms is true just in case both of its
conjuncts or at least one of its disjuncts is true or there is an object in the domain
verifying its matrix. The remaining logical operators cannot have their standard inter-
pretations if the truth of a statement is to be robust against future development. Thus,
for example, the sales representatives can’t assert a negation, such as “No building on
this block is over three stories,” simply because no tall buildings have been placed there
yet; there must be some guarantee that none will be built (for example. a prohibition
in the building code, or perhaps the fact that two-story houses have already been built
on all the sites on the relevant block). Similarly for conditionals; not only must it not
be the case that construction so far verifies the antecedent but not the consequent,
but there must be a guarantee that if, at any later stage of the development, the anteced-
ent comes to be true, the consequent will come true at the same time (or have already
come true when the antecedent does). As for universal quantifications (for example
“Every building will be compliant, not only with the current fire code but with the
strengthening being considered by the county”), not only must the objects already built
in the domain satisfy the matrix, but there must be a guarantee that any new ones added
will too.
It’s hard to predict all the questions potential home-buyers may ask: we, as develop-
ers, can’t just issue our sales representatives to read from the documentation we provide
in response. They will have to reason, to deduce answers. And the standard of validity
for their reasoning ought to be truth preservation, in the sense of truth just described.
The honest deductive logic for them to use is intuitionistic logic. (As should be obvious
to any expert reading this, the fantasy about ethical real estate development is simply a
way of dressing up the standard Kripke-model semantics for intutionistic logic.)
The ontological interpretation of mathematical constructivism, then, takes it to be
a sort of semi-fictionalist account of mathematics. Some mathematical objects can be
treated as real existents, but in theorizing about them we put them into the context of
a larger system of objects, most of which exist “only in the mind.” The full system of
mathematical objects treated in a given branch of mathematics is “under construction.”
But, whereas a housing development may actually be completed at some stage, a math-
ematical universe will always be under construction.
The description so far has been abstract: the example was non-mathematical, and
not much has been said about how the distinction between real and fictitious entities
applies in actual constructivist mathematics. It turns out that there are several options
available here, and different mathematicians and philosophers of mathematics have
not all found the same option attractive. For Brouwer, and probably for the major-
ity of the minority of mathematicians who consider themselves constructivists, finite
objects—natural numbers, or the sequences of moves in chess games—can be consid-
ered real. In his first publication rejecting classical logic (“On the untrustworthiness
of logical principles,” 1908) he noted that two-valued logic was appropriate in reason-
ing about a finite domain (at least if the atomic predicates used are decidable). This
can be shown formally in intuitionistic arithmetic. The condition on decidable pred-
icates is met—an atomic sentence of arithmetic is just a numerical equation whose
truth value can be calculated by grade-school algorithms—and it is provable in Heyting

672
 I N TU I TI ON I S M

Arithmetic that the instances of the law of excluded middle for statements involving
only bounded quantifiers (for all x less than t/there exists an x less than t) hold. The
infinite series of the natural numbers, however, cannot be completed, and so is not real.
Hence excluded middle ceases to hold when unbounded quantifiers (for all x/there
exists an x) occur.
Others, however, are attracted by a less restrictive conception. Daniel Velleman (not-
ing similarities between his position and those of such early twentieth-century “semi-
intuitionists” as Baire, Borel, and Lebesgue) has argued that a denumerable structure,
such as the natural number series, can be considered real and reasoned about classically,
but that nondenumerable sets, such as that of real numbers, shouldn’t be. (He also notes
that a position like this, called “denumerabilism,” was described, though not endorsed,
by Robert McNaughton.)
Even further from Brouwer’s position is the “liberal intuitionism” proposed by Law-
rence Pozsgay as a philosophy of set theory. Here sets can be treated as real, and classical
logic used when all quantifiers are restricted to range over the members of some set, but
the full set-theoretic universe is treated as “under construction,” and reasoned about in
intuitionistic logic.
Real human beings, of course, are severely limited in what they can understand or
remember or visualize: to use Descartes’s example, we don’t distinguish a thousand-
sided figure from a ten-thousand-sided by forming distinct visual images. The connec-
tion of any sort of constructive mathematics with actual human cognitive capacity is
thus tenuous: an epistemological motivation must appeal to knowability by idealized
human minds. The stricter versions of constructivism require less idealization, but it is
hard to see the kind of mind that could visualize arbitrary denumerable sets, or arbitrary
sets, as even an idealization of our own.

4
Thinking of a formalized theory as describing, not a fully realized system of actually
existent objects but something still under construction with possibilities of being
extended in different ways, puts us in the domain of modal logic. The idea of inter-
preting intuitionistic logic in some sort of modal logic is much older than the model-
theoretic semantics of intuitionistic logic; it seems plausible to conjecture that Kripke
was led to his analysis of intuitionistic logic by putting his own slightly earlier model-
theoretic treatment of modal logic together with the already familiar modal translations
of intuitionistic logic.
Gödel announced the possibility of a translation of intutionistic propositional logic
in a (classical) modal logic in an abstract of 1933. In it he stated that, under any of a
variety of translation schemes, all (and, he correctly conjectured, only) the intuition-
istically valid formulas of propositional logic translate into theorems of the modal logic
S4: subsequent research has shown that (not always with the same translation schemes)
many other modal logics can be substituted for S4. He suggested that the necessity
operator could be thought of as expressing the notion it is provable that, noting that, in
view of the Incompleteness Theorem, this had to be understood in terms of informal
provability and not provability in a particular axiomatic system. (Modal logics whose
necessity operators could be interpreted in terms of provability in a formalized system
were discovered late and Goldblatt showed that such a “provability logic” could be used
to interpret intuitionistic logic.)

673
 A LLEN H A Z E N

The connection with Kripke’s semantics for intuitionistic logic is clearest with a
translation scheme slightly different form the ones Gödel considered. We can define it
(extending it to full first-order logic) recursively as follows:

Basis: an atomic formula is “translated” by prefixing a necessity operator to it.

(In our earlier fantasy about property development, this corresponds to the
assumption that the developer won’t pull down already completed buildings.)
Recursion:
(i) the translation of a conjunction, disjunction or existential quantifica-
tion is simply the conjunction, disjunction or existential quantifica-
tion of the translations of its conjuncts, disjuncts or matrix
(ii) the translation of a negation, conditional or universal quantification
is obtained by prefixing a necessity operator to the negation, condi-
tional, or universal quantification of the translation(s) of the formula
negated, the antecedent and consequent, or the matrix.

Exactly the theorems of intuitionistic first-order logic translate into theorems of a ver-
sion of first-order S4 incorporating the Converse Barcan Formula but not the Barcan
Formula.
This is formally all very nice (and considering modal translations allows easy proofs
of many metamathematical theorems about intuitionistic logic), but what does it mean?
S4 has often been proposed as an epistemic logic, with the necessity operator interpreted
as expressing knowability (Hintikka), or (as in Gödel’s abstract) provability. Around
1980 several authors (notably Goodman and Shapiro) suggested a program of “epis-
temic mathematics”: the use of an explicit modal operator within classical mathemat-
ics to recover, in a classical context, the useful distinctions (between, for example,
constructively proven existence theorems and mere “nonuniversality” theorems) that
intuitionism displays. One formal result came out of this work: the exact translation of
the intuitionistic into modal systems extends beyond logic to first order Arithmetic.
At a conceptual level, epistemic mathematics doesn’t seem to capture the critical side
of constructivism. (To be fair, its proponents didn’t claim it did.) Brouwer and other
constructivists rejected some classical theorems, but “epistemic arithmetic,” for exam-
ple, is a proper extension of classical PA. In the case of arithmetic there is an additional
potentially embarrassing feature. One trivial theorem of classical arithmetic which is
asserted (and asserted to be necessary) in epistemic arithmetic is that every natural
number has a successor. As a consequence, every possible world in a model of epistemic
arithmetic, or of a Kripke model of Heyting Arithmetic, must have infinitely many
objects in its domain, including objects corresponding to every natural number! The
idea that the natural number series is “under construction” seems to be represented in
these models only by allowing some worlds to contain, in addition to all the “genuine”
natural numbers, further non-standard numbers.
This can, however, be avoided. Intuitionistic logic and arithmetic can be interpreted
in modal systems with modal logics weaker than S4, and in particular modal logics
lacking the principle that all “necessary” propositions are true. Such logics are often
interpreted deontically (necessary = it ought to be the case that) or doxastically (neces-
sary = the subject’s beliefs logically commit them to believing that), the latter, at least,
suggesting an interpretation of intuitionism as a sort of fictionalism about mathemati-
cal objects. For a variant of our fantasy about ethical property developers, the necessity

674
 I N TU I TI ON I S M

operator could perhaps be interpreted as something like “it is the case that, or, if not, the
development plan we are committed to following will bring it about that.” A suitable
logic of this sort can be obtained from S4 by replacing the principle

If Nec(A) then A

with the weaker

If NecNec(A) then Nec(A);

an interpretation of intutitionistic logic in this logic will use a modified translation

scheme on which the translations of disjunctions and existential quantifications, as well
as negations, conditionals and universal quantifications, start with a necessity operator
(for more detail see Hazen 1990). This allows an interpretation of intuitionistic arith-
metic in which every “possible world” has a finite domain. (The interpretation stands
to E.W. Beth’s semantics for intutionistic logic roughly as the translation into S4 does
to Kripke’s semantics.)

5
Model-theoretic accounts of intuitionistic systems are typically studied with the
methods of classical mathematics; they are, as it were, a nonintuitionist’s way of explain-
ing intuitionism. They have, however, been studied constructively, and there are
intuitionistic proofs of completeness theorems for parts of intuitionistic logic.
There have also been nonmodel-theoretic semantic accounts of intuitionistic logic
and particular intuitionistic mathematical systems, notably S.C. Kleene’s notion of real-
izability, and the related theory of constructions of N.D. Goodman. It is surprising that
these have not drawn more attention from philosophers of language, for they can be seen
as the most precise and detailed attempt to work out what, in the metaphysical literature,
is called a theory of truthmakers. The idea here is that, for each true proposition, there
is an object whose existence makes the proposition true. In the metaphysical literature,
truthmakers have often been taken to be facts or states of affairs (cf. Olson), or in some
cases ordinary objects. In Kleene’s original version or realizability, the realizer of a true
arithmètic statement is a number, taken as encoding (via a scheme of Gödel-numbering)
some complex of numbers and functions. Thus, for example, the realizer of a true existen-
tial quantification encodes some particular witnessing number together with a realizer
for the statement that this number realizes the matrix. The realizer of a statement of the
form “for every m there exists an n such that . . .” will encode an algorithm computing,
for an input m, an appropriate n; the realizer of a statement of the form “for all n, either
F(n) or Not-F(n)” will encode an algorithm for deciding, with respect to any given n,
whether it is F or not, and so on. The notion (and variations on it) has proven techni-
cally fruitful: for example, since all theorems of Heyting Arithmetic have realizers, one
way of showing that a sentence is not a theorem is to show that it lacks realizers.
As the multiplicity of interpretations suggests, intuitionistic logic is a a very natural
logic with nice properties. Most people familiar with it have a difficult to formulate but
intuitively vivid sense that the proof theory of intuitionistic logic, and in particular its
natural deduction formulation, is “cleaner” than that of classical logic. It has turned
out to have many applications. Some of these are obviously related to its constructiv-
ist motivations: computer scientists have taken an interest in intuitionistic logic and

675
 A LLEN H A Z E N

mathematical systems formulated in it because potentially useful algorithms can often

be extracted from a proof in a constructive system, as suggested by the examples of the
information encoded by realizers in the previous paragraph. On the other hand, intui-
tionistic logic turns out to be the “correct” one for formalizing domains, such as Smooth
Infinitesimal Analysis, which are not obviously inspired by the constructivistic critique
of classical mathematics.
It should be noted, however, that adoption of an intuitionistic or constructivistic
standpoint does not necessarily commit one to this particular formalization of proposi-
tional and first-order logic. Intuitionism is a doctrine about the content of mathematics;
the language in which this content is expressed is of secondary interest. If we define our
logical constants differently, we can express the same content with a different formal
logic. In standard intuitionistic logic, a disjunction can only be proved if one of the
disjuncts is provable, but the negation of a conjunction can be established by a simple
reductio. One could define a “constructive” kind of negation, for which reductio fails
and the negation of a conjunction is provable only if the negation of a conjunct is.
Kleene’s student David Nelson developed a system of this sort: the logic of constructible
falsity. In compensation for abandoning reductio, this logic incorporates many familiar
principles: the double negation equivalence, all the De Morgan equivalences (some
fail in Heyting’s logic), and the analogues of De Morgan’s laws for quantifiers. Nelson
showed that this logic could be used to express the same mathematical content as stand-
ard intutitionistic logic: systems of arithmetic based on Heyting’s and Nelson’s logics
are intertranslatable.

6
Intuitionism is not just about logic, and intuitionistic mathematics is not simply the
result of restricting the logical inferences allowed in classical mathematics. Even when
they assert (verbally) identical theorems, classical and constructive mathematicians
often use different proofs. Because they don’t have the cheat of indirect proof available
to them, constructivists have been forced to develop methods which, though classically
valid, were not discovered by classical mathematicians. There is a deeper kind of differ-
ence, however; Brouwer and his successors defined new concepts that have no role in
classical mathematics.
Brouwer’s theory of real numbers is an example. Real numbers are a generalization
of fractions: not all reals are rational. We appeal to them in giving mathematical treat-
ments of measurable quantities, where measurement is conceived as susceptible of ever
greater precision: a real number, then, is conceived as the limit of a potentially infinite
sequence of rational approximations. With appropriate codings, real numbers can be
thought of as given by infinite sequences of natural numbers, but an infinite sequence
is not something an intuitionist will believe in as a completed whole. One option is to
treat talk of infinite sequences as a façon de parler for algorithms generating them, and
interesting mathematical theories of real numbers can be developed along these lines.
(This is the approach adopted by the school of constructive mathematics stemming
from the worrk of the American mathematician Errett Bishop.) Brouwer chose instead
a less constrained notion, that of a choice sequence: a sequence of numbers determined
by the successive choices of a free agent. (To ensure that the sequence is infinite, we
can imagine some convention for continuing it if the chooser dies or loses interest, but
this is not a major concern. Intuitionism is interested in mathematically possible future

676
 I N TU I TI ON I S M

developments, and there is no mathematical guarantee that the chosing won’t go on

forever.) At any given time, only a finite initial part of the sequence will have been
chosen: it is still “under construction,” and it cannot be said to have or lack properties
that depend on choices that haven’t yet been made. In particular, suppose two people
are choosing numbers (and so, given some recipe for converting infinite sequences of
naturals into real numbers, determining real numbers), and have so far made the same
choices. There is no mathematical guarantee that they will continue to make the same
choices, and so no guarantee that the real numbers they are generating are the same, but
also no guarantee that their choices will diverge. The resulting theory of real numbers
differs dramatically from that in classical mathematics (though it does yield most of the
standard theorems needed for applications in science and engineering). One conse-
quence is of logical interest. In propositional logic, any classically contradictory formula
is also intuitionistically contradictory (and the double negation of any classical tautol-
ogy therefore intuitionistically valid), but this does not extend to first order logic: the
negation of a valid formula of first order logic is not in general intuitionistically con-
tradictory. By appeal to the theory of choice sequences, and the theory of real numbers
based on it, we can show that this is not a mere empty possibility. In fact it is a theorem
of the intuitionistic theory of real numbers that it is not the case that, for every pair of
real numbers x and y, either x = y or it is not the case that x = y.
Another example, suggested by Brouwer later in his career and not accepted by all
intuitionists, is the theory of the Creative Subject. This is an attempt to make use,
within mathematics, of the idea that a proposition is only true if proven. The idea is to
consider a special kind of choice sequence, chosen by the mathematician in accordance
with the rule “choose 1 if, at the time of choice, you have proven proposition A, but
choose 0 if you haven’t.” If truth is the same thing as proof, then it is perhaps plausible
to say that the proposition is true if and only if this sequence contains a 1. Eliminating
the reference to the choosing subject, Saul Kripke formulated this as an axiom schema,
Kripke’s Schema, that could be incorporated in an intutionistic axiom system having
variables both for natural numbers and for choice sequences:

There exists a sequence s such that A if and only if for some n, the n-th
number in s is 1.

This can be used to obtain the power of higher-order logic: let sequences stand for sets,
and use Kripke’s Schema to give the effect of the comprehension schema.

References
Brouwer, L.E.J. (1908). “Over de Onbetrouwbaarheid der logische Principes” (“On the Untrustworthiness of
the Principles of Logic”), Tijdschrift voor Wijsbegeerte, vol. 2.
Hazen, Allen. (1990). “Actuality and Quantification”, Notre Dame Journal of Formal Logic 31(4):498–508

677
 5.9
R ICHARD MO NT AGUE’ S
A P P ROACH T O T HE
S E MANTIC S OF
N A TURAL L A NGUAGES
Richmond H. Thomason

1 Logic and Psychology in Semantics

In the nineteenth century, the foundations of logic were a matter of debate. In [Frege,
1953], Gottlob Frege ably argued that logic should be assigned to mathematics rather
than to psychology. The conception of logic as a branch of mathematics has dominated
the field since that time. Logic is generally considered to be a part of mathematics, and
the research methods that logicians use are mathematical.
Similar issues reappear in the twentieth century, in connection with cognitive sci-
ence. In the 1950s, semantics began to appear as a respectable part of linguistics, but
the foundational questions this raises—is it mathematics or psychology?—are less
clearly and uniformly resolved. Linguistics is heterogeneous, drawing on psychological
and sociological methods, experimentation, data analysis, and applied mathematics.
Probably most linguists working in syntax and semantics follow Chomsky in believ-
ing that their subject matter is psychological. But, although there are alternatives to
formal semantics, such as “cognitive semantics” ([Jackendoff, 1996, Talmy, 2000a]),
the prevailing approach to linguistic semantics, based on Richard Montague’s work,
is thoroughly mathematical. Although some people, notably Barbara Partee ([Partee,
1978, Partee, 1979]) have addressed the discrepancy between the practices and foun-
dational beliefs of semanticists, the tension has not been much addressed. Probably the
main reason for the acceptance of logical techniques in semantics is that there is really
no competitor when it comes to articulating a detailed, explanatory theory.
So the predominant approach to linguistic semantics, deriving from Montague’s
work, is based on mathematical logic. This has several consequences.

2 Formalization and Formalized Languages

The systems of logic developed by the early logicists (and especially, Frege and Russell)
were concerned prinarily with explicating the notion of a correct mathematical proof,
 M O N TA G U E G R A MMA R

and were not equipped with a semantics. Since then the techniques for formalization
have deepened and sharpened, and apply to a much broader spectrum of languages, with
many uses.
Formalizing a language begins with the presentation of an explicit syntax. In the
simplest case, this is simply a definition of the formulas of the language. In more general
cases, Formula or Sentence is just one of many syntactic categories, and the syntax
must define the set of well-formed expressions belonging to each of these categories.
Depending on the purpose of the formalized language, the formalization may include
a characterization of the associated proofs, or it may articulate some associated algo-
rithms. (Proofs are not important for most programming languages, but a compiler is
essential.)
Finally, a formalized language needs a semantic component. Without a semantics,
a formalization is blind, lacking a criterion of correctness for proofs and algorithms. In
the form developed by Alfred Tarski and used by Montague, a semantics has two parts:
(1) a characterization of the models, the structures that serve to interpret the language,
and (2) a definition of the satisfaction relation, according to which every formula of the
language takes on a truth value. Where M is a model and I is a formula, [I]M = T if I is
true in M, and [I]M = F if Iis false in M.

3 Natural Languages and Formalized Languages

Montague opens his article [Montague, 1970a] with the following noteworthy claim.

I reject the contention that an important theoretical difference exists between

formal and natural languages.

As far as I can tell, the rejected contention never existed in the form of a well-developed
and defended position. Certainly it was accepted at the time by many philosophers—
ordinary language philosophers as well as their formally inclined colleagues—that the
artificial languages studied by logicians and natural languages like English were differ-
ent in kind. And logic textbooks by thoughtful logicians, such as [Quine, 1940] and
[Church, 1959], begin with informal accounts of differences between natural languages
and the “well constructed” languages used by logicians, with particular emphasis on the
ambiguities to be found in natural languages. The differences these authors note are
important, but they don’t claim they are in any sense “theoretical.”
The idea that the techniques used by logicians could be applied to natural languages,
and that natural languages are in principle formalizable along similar lines, was not
new. It is mentioned (briefly, but with approval) in [Carnap, 1934] and [Tarski, 1944].
Yehoshua Bar-Hillel took up Carnap’s idea with enthusiasm in [Bar-Hillel, 1963]. Bar-
Hillel was more familiar with linguistics than most philosophers of his time. Citing
work by structural linguists as well as early transformational work by Zelig Harris and
Noam Chomsky, he presses the idea that no differences in principle exist between nat-
ural languages and the constructed languages studied by logicians, and suggests that
logical research could be useful in linguistics, especially in semantics. In a review of
[Bar-Hillel, 1954], an earlier paper along simpler lines, Noam Chomsky rejects the pro-
posal; see [Chomsky, 1955]. Chomsky’s review, which argues in effect that the idealiza-
tions made by logicians render their work largely irrelevant to linguistics, is the most
explicit expression of which I’m aware of the view that natural languages are different in

679
 RI CH M O N D H . T H O MA S O N

principle from logical languages. I don’t know if Montague was familiar with this review.
He certainly was aware in general terms of Chomsky’s work: In “English as a Formal
Language” he says that he that he doesn’t consider the work of Chomsky and his stu-
dents to be helpful in advancing the goal of “serious syntax and semantics.”
The ideological contrast here is rather stark, and despite the efforts of people like
Barbara Partee, I don’t believe it has ever been successfully reconciled. It is therefore
somewhat surprising that the main influence of Montague’s work has been within
linguistics.
In any case, what was dramatically new about Montague’s contribution was not the
opening salvo, but the way he turned his claim into an explicit research project, which
he then proceeded to carry out. After a brief preamble in [Montague, 1970a], he imme-
diately begins to develop an explicit syntax for an extended and ambitious “fragment”
of English. Montague preferred to avoid syntactic complications by restricting himself
to constructions that—although they are definitely grammatical English—sound stilted
and unnatural. For instance, his fragment contains Every man such that that man loves a
woman in Amsterdam walks rapidly, rather than the more natural Every man who loves a
woman in Amsterdam walks rapidly. Nevertheless, the sheer scope of this fragment and
the others that Montague presented at about this time is astonishing.

4 Montague’s Work in Semantics

Montague’s publications bearing directly on natural language semantics began in 1968,
and his work in this area continued until his death early in 1971. The earlier papers,
[Montague, 1968] and [Montague, 1970b], were concerned with developing Intensional
Logic—the logical part of the semantic project. Three later papers, [Montague, 1970c],
[Montague, 1970a], and [Montague, 1973], develop and describe the general frame-
work, and present grammars for “fragments” of English, treating these fragments using
the logical methods described in Section 2, with an explicit syntax and a model-theo-
retic semantics. A sixth paper published during this period, [Montague, 1969], is an
outlier. This paper describes Intensional Logic and advocates it as a framework for doing
rigorous philosophy. These papers, and some others, were published together in the
posthumous collection [Montague, 1974].
As far as I know, no history of Montague’s intellectual development is available, and
materials may not exist for such a work, unless someone interviews those still alive who
knew him during this period. Montague was a logician first and foremost, but unlike
most mathmatical logicians of his time, he believed that the rigorous mathematical
techniques of logic could be used in areas other than mathematics, including philoso-
phy. According to [Partee, 2009]—drawing on a personal communication from Hans
Kamp—Montague was frustrated, in teaching introductory logic, by the informality and
lack of rigor in the part of the course dealing with the relation between logical formulas
and English, and this was a large part of the motivation for his work in the semantics of
natural language.
David K. Lewis taught at UCLA from 1966 through 1969, and according to [Partee,
2005] Montague and Lewis talked often during that period. Lewis’s approach to formal
semantics, published in [Lewis, 1970], is similar to Montague’s in many ways; it differs
in lacking the syntax and semantics for an ambitious “fragment,” but compensates for
this by presenting ideas for dealing with sorts of constructions that Montague never
mentions. Clearly, the two must have influenced each other in many ways.

680
 M O N TA G U E G R A MMA R

5 Words, Phrases, and their Meanings

When we seek a meaning, it is usually the meaning of a word. So it seems that once we
understand the meanings of a sentence’s words, the meaning of a sentence or any other
grammatical phrase comes for free. In the leading linguistic theories of meaning, and
certainly in the theories derived from logical approaches, this is reversed. The theory
has to account for the meanings of phrases, including sentences. This leads theore-
ticians to concentrate on the linguistic analogues of satisfaction definitions in logic.
The task of providing semantic rules, showing how the interpretations of grammatical
phrases depend on the meanings of their parts is sufficiently complicated to attract most
of the attention. Words must have interpretations that are compatible with the seman-
tic rules, but often this means only that their interpretations must have suitable types.
The specific interpretation of a lexical item turns out to be of little importance. This
is fortunate for the theories, because in most cases it would be difficult or impossible
to provide appropriately specific interpretations. Typically, the only natural language
words that receive specific interpretations are those, like every, and, and or, that cor-
respond to “logical constants.”
On the other hand, natural languages contain many more syntactic constructions
and ways of combining them than typical artificial languages. A sentence like

(1) Every decent man who loves a woman who loves him should eventually want to
marry her

is relatively simple and perfectly intelligible, but at least nine different constructions
interact in producing the interpretation. (1) The combination of a transitive verb
with its direct object, as in loves+a woman. (2) The combination of a verbal phrase
with its subject, as in he+loves a woman. (3) The combination of a relative clause
with a nominal phrase, as in woman+who loves him. (4) The combination of an article
or quantifier with a nominal phrase, as in a+woman. (5) The combination of a tense
with a verbal phrase, as in Present+love=loves. (6) The combination of a modal auxil-
iary with a verbal phrase, as in should +eventually want to marry her. (7) The combination
of an adjective with a nominal as in decent+man. (8) The combination of an adverb
with a verbal phrase, as in eventually +want to marry her. (9) The combination of a verb
with an infinitive phrase, as in want +to marry her. (In these examples, ‘combination’
means ‘syntactic combination’, and ’+’ denotes a syntactic operation. For example,
‘nominal phrase’ and ‘relative clause’ are to be understood as syntactic categories,
and the word order that results when they combine would be given by a syntactic
rule. But in these examples, I am being informal and somewhat arbitrary about the
syntax.)
It would be impossible to account for the meanings of even moderately complex sen-
tences without an architecture for classifying possible meanings and their interactions.
Perhaps Montague’s most important new contribution was to provide such an architec-
ture, and to show how it could be systematically used to explain the meanings of natural
language phrases and their modes of combination. It is appropriate that Montague’s
publications on natural language semantics begin with Intensional Logic, because the
materials for producing appropriate meanings lie in the logic.

681
 RI CH M O N D H . T H O MA S O N

6 A Logic Based on Functions

The idea that meanings combine with other meanings to form new meanings creates
a fundamental need for suitable modes of combination. Montague looks to logic to
provide these modes, and logic prizes parsimony: Many of the most impressive results in
logic show how to produce remarkable expressive results from minimal resources. The
best theory would manage with only one mode of combination. The most likely candi-
date for this one mode is functional application. Functions are pervasive in mathemat-
ics, and mathematics provides powerful mechanisms for defining functions.
Moreover, functions are plausible vehicles for combining the meanings of constitu-
ents into larger meanings. For instance, it makes a certain amount of sense to say that
the meaning of an adjective like decent is a function taking the meaning of a nominal
phase, such as man, into the meaning of a new nominal phrase, decent man. If the inter-
pretation of decent is a function f, then [[decent man]], the interpretation of the phrase
decent man, will be f([[man]]): so here we have obtained the meaning of the phrase
X + Y by taking the meaning of X to be a function, and applying this function to the
meaning of X.
If the composition of meanings is a matter of functional application, the logic of
meaning composition will be a logic of functions. At this point, earlier developments in
logic dovetail nicely with the needs of semantics. These developments provide not only
an appropriate logic of functions but the crucial idea of a logical type.
Both Frege and Bertrand Russell undertook the project of producing a logic that
would allow mathematics, and especially the mathematics underlying the calculus, to
be presented by means of a series of definitions. Frege’s combination of set theory and
logic led to a formalization that Russell showed to be inconsistent, replicating the para-
dox of the set consisting of all sets that are not members of themselves. Russell’s for-
malization of the mathematics of real numbers, in [Whitehead and Russell, 1910–1913],
avoided this paradox by enforcing a system of logical types; a set must have a type higher
than the type of any of its members.
After the publication of Principia Mathematica in 1910, a series of developments in
logic produced simplified and improved formalizations of the theory of logical types.
These developments culminated in 1940, with Alonzo Church’s presentation in
[Church, 1940] of a version of type theory based entirely on functions. Functions and
sets are theoretically interchangeable. On the one hand, you can regard a function as a
set of ordered pairs, and on the other, you can identify a set X with the function fX that
outputs the truth-value T for the input x if x X, and otherwise outputs F.
Combining the notion that every object in the ontology of the logic has a type with
a logic based on functions, we arrive at the idea that two types must be associated
every function: (1) the type of the function’s domain, of its inputs, and (2) the type
of the function’s range, of its outputs. A function from individuals to truth-values, for
instance, would be characterized by two components: the type of individuals and the
type of truth-values. These last two types are not functional; they are basic.
The system of all types can then be characterized as follows: (1) there are two basic
types, the type e of individuals and the type t of truth-values; (2) if  and  are types,
then ¢, ²is the type of functions from things of type  to things of type ; (3) there
are no other types. For instance, ¢e, t² is the type of functions from individuals to truth-
values, and ¢e, ¢e, t²²is the type of functions from individuals to functions from individu-
als to truth-values.

682
 M O N TA G U E G R A MMA R

Church’s language for the logic of functional types has a family of equality symbols—
for each type  there is an equality constant of type ¢, ¢, t²²—and relies on only two
constructions: functional application and lambda abstraction. Functional application
involves an expression ] of type ¢, ² and an expression K of type ; ](K) is then an
expression of type , and denotes the result of applying the function denoted by ] to the
object denoted by K.
Lambda abstraction is more complicated, but it proved to be highly useful in Mon-
tague’s linguistic applications of type theory. It is a general method for constructing
explicit names of functions. For instance, in arithmetic, Ox x2 + 1 denotes the function
that inputs a number and outputs the result of squaring that number and adding 1. In
general, an expression ] of type , containing a variable, say x, of type , will denote an
implicit function whose output is the denotation of  when x is assigned an input deno-
tation of type . Ox] explicitly denotes this function, which has type ¢, ². Lambda
abstraction on formulas of type t is a special case: Oxe[x = x], for instance, is the constant
function that inputs an individual and outputs the truth-value t.

7 Typing Phrases of Natural Language

The apparatus of typing, originally motivated by the paradoxes, has become useful in a
remarkably broad range of applications. If Church’s types can be assigned in a plausible
way to the phrases of natural language, this will support a method of semantic interpre-
tation that uses functional application to interpret syntactic combinations. Montague
showed that this can be done, and the methods he developed continue to be used by
formal semanticists.
For instance, take the sentence

(2) Andy is single and unemployed.

The subject, Andy, is a proper name; if we assume that proper names denote individuals,
then Andy will have type e. We can leave the verb is uninterpreted or assign it the iden-
tity function. The phrase single and unemployed is what logicians would call a one-place
predicate, so it would have type ¢e, t². These types allow us to obtain the interpretation
of the entire sentence by functional application, applying the denotation of single and
unemployed to the denotation of Andy. The adjectives single and unemployed have type
¢e, t². Therefore, the and that combines them must have type ¢¢e, t², ¢¢e, t², ¢e, t²²². This
last type indicates a function that inputs (sequentially) two adjective denotations, and
outputs an adjective denotation. These type assignments make good intuitive sense,
and provide a functional interpretation of the entire phrase.
Of course, there will be an and that combines sentences, as in

(3) Andy is single and Andy is unemployed.

This and will have type ¢t, ¢t, t²². Since each lexical item can have only one logical type,
we have to declare (at least) two senses of and: an and1 that combines sentences and an
and2 that combines adjectives.

683
 RI CH M O N D H . T H O MA S O N

8 Interpretation, Translation, and Meaning Postulates

Once the syntax of a fragment has been specified, its semantics is given by a model-
theoretic interpretation. The models of the theory are specified, and for each model M,
the interpretation [], D]M of the phrase ] in M has to be defined for each phrase ]of the
fragment, where D is a disambiguator. (Disambiguators will be explained in Section 9;
for the moment, I’ll ignore this complication.)
The model-theoretic interpretation can be done in either of two ways. The interpre-
tation can be direct, with a satisfaction definition like those that are used for logical lan-
guages. Or it can be indirect, using a logical language that already has a model-theoretic
interpretation and defining a translation into the logic for each disambiguated phrase.
Montague sometimes (for instance, in [Montague, 1970a]) uses the direct method, and
sometimes (in [Montague, 1973]) the indirect method.
The indirect method is preferable, because it provides a logical language (the target
of the translation) in which to state meaning postulates. The need for such postulates
in semantics was somewhat of an embarrassment for Carnap, but they play a different
and important role in Montague’s theory, making it possible to nullify the complicat-
ing effects of the apparatus of types. We can illustrate this point with the two ands of
Section 7.
Since and1 is a logical constant, we can completely specify its meaning by means of a
meaning postulate. In Postulate (4), and1c is the logical translation of and1. (This follows
Montague’s notational practice in [Montague, 1973].) Using lambdas, the postulate
says that this sense of and means the same thing as the logical conjunction š(which is
available in Church’s logic).

(4) and1c = Oxt Oyt [x šy]

Another postulate relates the interpretation of the two ands.

(5) and2c = Ox¢e,t² Oy¢e,t²Oze[x(z) šy(z)]

It follows from this postulate that Examples (2) and (3) are logically equivalent. With
such equivalences, the need to proliferate ands is not so harmful.
Montague’s uniform treatment of noun phrases as generalized quantifiers uses similar
techniques, and was especially influential. Proper names like ‘Andy’ and quantifica-
tional noun phrases are similar in their syntactic distribution and behavior; linguists
classify them together as noun phrases or NPs.
Quantificational phrases have a natural place in Church’s system of types. Since Frege
introduced the idea, logicians have been familiar with the identification of numbers as
sets of sets: five dimes, for instance, would correspond to the set of all sets containing five
dimes. In functional terms, this would mean that five has type ¢¢e, t², ¢¢e, t², t²², so that
five dimes will have the type ¢¢e, t², t². But other quantificational phrases, such as every
dime and a dime can be interpreted in the same way; the former corresponds to the set
of sets that contain all dimes, and the latter to the set of sets that contain at least one
dime. Using meaning postulates, this would mean, for instance, that the translation of
every would satisfy the following constraint.

(6) every = Ox¢e,t² Oy¢e,t²ze [x(z) oy(z)]

684
 M O N TA G U E G R A MMA R

With this new type assignment to NPs, the interpretation of the subject–predicate rela-
tion remains functional but is reversed: in a sentence like

(7) Every dime is round,

round has type ¢e, t², every dime has type ¢¢e, t², t²², and the meaning of the whole sen-
tence is obtained by applying the function that the subject denotes to the set denoted by
the predicate. In effect, the sentence is interpreted as saying “The set of round things is a
member of the set of all sets containing every dime.” This policy towards NPs, together
with the uniform use of function application in translation, makes for logical transla-
tions that look more like English than logical formulas. For instance, the translation of
Sentence (7) is Formula (8).

(8) [everyc (dimec)](roundc)

The textbook translation, xe [dimec(x) oroundc(x)], would be obtained by applying

meaning postulates to (8).
Typing constraints in the logic provide a way to verify formulas such as (8), known
as type checking. The formula everychas type ¢¢e, t², ¢¢e, t², t²², giving everyc(dimec) the
type ¢¢e, t², t². But dimec has type ¢e, t², giving Formula (8) the type t, which is the type
assigned to sentences.
By making the interpretation of NPs uniform, we have given proper names—which
intuitively denote objects—a higher type than we would expect. Again, meaning postu-
lates can prevent the unwanted effects of this assignment of types. In this case, we want
to associate an individual with each proper name, serving as its customary denotation,
and invoke a meaning postulate saying that the interpretation of a proper name is the
set of sets that contain the associated individual. The following meaning postulate will
accomplish this for the name Andy.

(9) xey¢e, t² [Andyc(y) ly(x)]

9 Obstacles to the Formalization of Natural Language

The reasons that led many logicians and philosophers to feel that natural and logical
languages were fundamentally different form a list of challenges to Montague’s project.
These include: ambiguity, vagueness, context-sensitivity, and intensionality. It is inter-
esting to see how Montague addressed them in carrying out his project.
Ambiguity. Montague, along with many subsequent semanticists, assumes that the
only sources of ambiguity are lexical and structural. Lexical ambiguity is familiar; it
arises whenever a single word has two or more different senses. Structural ambiguity
occurs when the syntactic rules can produce the same phrase in more than one way. The
sentence I know what you know, for instance, is structurally ambiguous, since what you
know can either be an embedded question or a relative clause—and it has two different
meanings, corresponding to the syntactic alternatives.
A disambiguator for a phrase, then, is a specification of a sense for each lexical item
in the phrase and of a syntactic derivation for the phrase. The interpretation of a phrase
must be relativized to a disambiguator.
If structural ambiguities can’t be explained without finding appropriate syntactic
derivations, semantic evidence about ambiguity can precipitate decisions about how

685
 RI CH M O N D H . T H O MA S O N

the syntax should be arranged. This happens in Montague’s fragments in the case of
sentences that exhibit quantifier-scope ambiguities, such as Montague’s example, Every
man loves a woman. Conservative syntacticians at the time found the idea of semantic
arguments for syntactic structure unwelcome, but now such treatments of quantifier-
scope ambiguities are widely accepted in linguistics. Montague’s approach to semantic
ambiguity seems, in fact, to have held up well.
Vagueness. Montague offers no solution to the problem of vagueness. This topic has
been studied extensively—mostly in the philosophical literature—since 1973. In my
opinion, we do not have an entirely successful solution to the problem yet. However,
this doesn’t seem to be problem specifically for the semantics of natural languages, but
for any language intended to be used in domains where complete precision is lacking.
Context-sensitivity. Montague considered this problem in his papers on pragmatics.
His solution, which occurred independently to many others, was to make interpretation
in a model sensitive to various contextual factors. An extensive subsequent literature
developed from this starting point. This is an area where the theories have changed
and gradually become more sophisticated. But these developments enhance Montague’s
program of formalizing natural languages, rather than threatening it.
Intensionality. The problem of intensionality was, of course, stressed by Willard V.O.
Quine and other philosophical skeptics about theories of meaning; see, for instance,
[Quine, 1953]. It is a direct and glaring problem for the scheme of interpretation as we
have presented it so far.
In interpreting natural language, the problem of intensionality centers on the logi-
cal type that is associated with sentences. A propositional attitude construction, such
as believe, combines with a sentence to form a verb phrase: The present tense form of
believe, for instance, combines with Los Angeles is the capital of California to form the VP
believes that Los Angeles is the capital of California. This then combines with a subject like
Bertha to form the sentence

(10) Bertha believes that Los Angeles is the capital of California.

So the interpretation [[Bertha believes that Los Angeles is the capital of California]] of the
entire sentence (which is just a truth-value, T or F, in Church’s type theory) will be a
function of the interpretation [[Los Angeles is the capital of California]] of the subordi-
nate clause (which is also a truth-value). Something has gone very wrong, since this
implies that if Example (10) is true, then Bertha believes any false proposition what-
soever. Montague addresses the problem by providing an enhanced interpretation of
sentences, based on possible-worlds semantics: the new interpretation is a set of possible
worlds, or—to put it in functional terms—a function from possible worlds to truth-val-
ues. Possible-worlds semantics—and especially, the theory of how quantifiers interact
with modal operators—was relatively new and undeveloped as a branch of logic when
Montague turned his attention to it. Montague provided one solution to the problem
of how to combine quantification with modality, while retaining a reasonable model-
theoretic interpretation. But he went much further than this, extending modal logic to
include higher-order quantification and generalizing earlier ideas from [Carnap, 1956].
Montague called the logical system that resulted from these ideas “Intensional Logic.”
IL provides an alternative to Frege’s theory of sense and denotation with a well-devel-
oped model-theoretic interpretation.

686
 M O N TA G U E G R A MMA R

10 Intensional Logic
Although Montague didn’t formulate it quite this way, it is easiest to think of Intensional
Logic as the type theory that results from the addition of a new type, w, to Church’s type
theory—the type of possible worlds. Call the extensional types of the new logic those that
do not involve this new type. For each extensional type  (and indeed for any type ),
there is a corresponding intensional type, ¢w,².
Only two changes are needed in the models and satisfaction definition for the
extended theory. (1) Satisfaction is relativized to a world: [[] ]]M,w is the value assigned
to ] in M at world w. (2) Two new constructs, š and ›, need to be interpreted. The
“up” operator, š, is an intensionalizer; if ]is a formula of type , š]is a formula of type
¢w,². In effect, šperforms lambda abstraction over an implicit world variable: where ]
has type , [[š] ]]M,w is the function f from the set of worlds to objects of type  such that
f(w) = [[] ]]M,w. The ›operator is an extensionalizer of formulas of intensional type. If ]
has type ¢w,²then ›]is wellformed and has type , and [[›] ]]M,w = [[] ]]M,w(w).
The intension of the type formerly assigned to sentences is ¢s, t², the type of propo-
sitions or functions from worlds to truth-values, and propositional attitudes are then
treated as usual in possible-worlds semantics. But there are many other sources of inten-
sionality in natural language, and Montague deserves the credit for recognizing that a
type-theoretic modal logic provides the flexibility for dealing quite generally with the
phenomena. In courses on Montague Grammar, the point was illustrated with sen-
tences like John seeks two unicorns.
Of course, the use of possible-worlds semantics inherits the problems of that approach,
especially a commitment to “logical omniscience.”
For an exhaustive treatment of IL as a logic, with applications, see [Gallin, 1975].
Linguists have gradually come to accept possible worlds as a working tool, and have
found many applications of the idea of which philosophers may not be aware. See [Par-
tee, 1989] for an excellent survey of the topic, as well as the more recent general hand-
book articles concerning Montague Grammar.

11 From Montague Grammar to Formal Semantics

Barbara Partee reports in [Partee, 2005] (note 36) that she coined the term “Mon-
tague Grammar” around 1971. But, as Montague’s ideas and later additions have been
absorbed into the background of linguistic semantics, and the field has diversified, the
term has been replaced by “formal semantics.” Linguists have, in general, not taken
Montague’s approach to syntax very seriously, and so Montague’s semantic ideas have
been combined with many more popular syntactic frameworks. At the same time, the
theories have been fruitfully applied to an enormously wide range of constructions, in
English and many other languages. And the logical framework has been enriched, most
notably with various dynamic mechanisms. For more information on these develop-
ments, see [Partee, 1996] and [Partee and Hendriks, 1996].
Philosophers should take note that the term “Montague Grammar” is somewhat out-
dated, and that linguists have for some time been pursuing projects in areas of formal
semantics that are very relevant to philosophy. There is also much of interest to phi-
losophers in formal pragmatics, a younger field which would not have been possible
without relatively mature theories of linguistic semantics. Many papers of interest in
both fields have appeared in the journal Linguistics and Philosophy, and many others are
archived at http://semanticsarchive.net/.

687
 RI CH M O N D H . T H O MA S O N

12 Philosophical Consequences
Montague’s idea that Intensional Logic provided a way of formalizing philosophy,
turning it into a subject much more like mathematics, was unrealistic. And its uncom-
promising presentation would make it difficult even for sympathetic philosophers
to see that there is a genuine and important point here. The philosophical useful-
ness of a slightly less rigorous approach in the same spirit has been demonstrated in
the subsequent work of philosophers like David Lewis and Robert Stalnaker. This
influence is already well entrenched in various areas of philosophy, particularly in
metaphysics.
Time always works in favor of specialization and division of labor. But there is not
yet a happy division between formal semantics/pragmatics and the related areas of
philosophy. There is much overlap but too little communication. I believe that it
would be beneficial for researchers working in philosophy of language, or develop-
ing a point that depends on language, to form a habit of studying the relevant work
of formal semanticists. This work can add a new dimension to philosophical project.
Linguists are trained to deploy sophisticated theory and systematic use of linguistic
evidence in ways that go far beyond what is typical in philosophy. This work is often
directly relevant to philosophical issues, and can add new depth to the treatment of
these issues.
Looking at language systematically can also be a good way to achieve coverage. To
a very large extent, what Emmon Bach calls “natural language metaphysics” in [Bach,
1986]—working out the formal semantics of natural language constructions that seem
to incorporate basic ways we conceptualize the world—overlaps with metaphysi-
cal projects having to do with topics such as causality, time and change, and mereol-
ogy. (See the references in [Bach, 1986]. And [Dowty, 1979] is still a useful source on
this topic.) But because of the need for linguistic coverage, it is driven by system-
atic considerations that seem to be missing in the comparable areas of contemporary
philosophy.
Relationships between linguistics and philosophy have always been stressed by the
journal Linguistics and Philosophy. This journal is perhaps the most useful contemporary
and historical source of interactions between formal semantics and various areas of phi-
losophy. Some of these interactions are discussed in [Pelletier and Thomason, 2002],
a survey editorial that appeared in the twenty-fifth anniversary volume of the journal.
And the papers that appeared in the same issue were specially invited to illustrate con-
nections between the two fields.

References
Emmon Bach. Natural language metaphysics. In Ruth Barcan Marcus, Georg J.W. Dorn, and Paul Wein-
gartner, editors, Logic, Methodology, and Philosophy Of Science, VII: Proceedings of the Seventh International
Congress of Logic, Methodology, and Philosophy of Science, Salzburg, 1983, pages 63–81. North-Holland,
Amsterdam, 1986.
Yehoshua Bar-Hillel. Logical syntax and semantics. Language, 30(2):230–237, 1954.
Yehoshua Bar-Hillel. Remarks on Carnap’s logical syntax of language. In Paul A. Schilpp, editor, The Philoso-
phy of Rudolph Carnap, pages 519–543. Open Court Publishing Company, LaSalle, Ilinois, 1963.
Rudolph Carnap. Logische Syntax der Sprache Schriften zur wissenschaftlichen Weltauffassung. Verlag von Julius
Springer, Vienna, 1934.
Rudolph Carnap. Meaning and Necessity. Chicago University Press, Chicago, 2nd edition, 1956. (First edi-
tion published in 1947.).

688
 M O N TA G U E G R A MMA R

Gennaro Chierchia and Sally McConnell-Ginet. Meaning and Grammar: An Introduction to Semantics. The
MIT Press, Cambridge, Massachusetts, 2nd edition, 2000.
Noam Chomsky. Logical syntax and semantics: Their linguistic relevance. Language, 31(1):36–45, 1955.
Alonzo Church. A formulation of the simple theory of types. Journal of Symbolic Logic, 5(1):56–68, 1940.
Alonzo Church. Introduction to Mathematical Logic, Vol. 1. Princeton University Press, Princeton, 1959.
David R. Dowty, Robert Wall, and Stanley Peters. Introduction to Montague Semantics. D. Reidel Publishing
Co., Dordrecht, 1981.
David R. Dowty. Word Meaning in Montague Grammar. D. Reidel Publishing Co., Dordrecht, Holland,
1979.
Gottlob Frege. The Foundations of Arithmetic. Oxford University Press, Oxford, 2nd edition, 1953. (Origi-
nally published in 1884. Translated by J.L. Austin.).
Daniel Gallin. Intensional and Higher-Order Logic. North-Holland Publishing Company, Amsterdam, 1975.
Leon Henkin. Completeness in the theory of types. Journal of Symbolic logic, 15:81–91, 1950.
Ray Jackendoff. Semantics and cognition. In Shalom Lappin, editor, The Handbook of Contemporary Seman-
tic Theory, pages 539–559. Blackwell Publishers, Oxford, 1996.
David K. Lewis. General semantics. Synthése, 22(1–2):18–67, 1970.
Richard Montague. Pragmatics. In R. Klibansky, editor, Contemporary Philosophy: A Survey. I, pages 102–
122. La Nuova Italia Editrice, Florence, 1968. Reprinted in Formal Philosophy, by Richard Montague, Yale
University Press, New Haven, CT, 1974, pp. 95–118.
Richard Montague. On the nature of certain philosophical entities. The Monist, 53:159–194, 1969.
Richard Montague. English as a formal language. In Bruno Visentini et al., editor, Linguaggi nella Societá,
e nella Tecnica, pages 189–224. Edizioni di Communità, Milan,1970. Reprinted in Formal Philosophy, by
Richard Montague, Yale University Press, New Haven, CT, 1974, pp. 188–221.
Richard Montague. Pragmatics and intensional logic. Synthése, 22:68–94, 1970. Reprinted in Formal Philoso-
phy, by Richard Montague, Yale University Press, New Haven, CT, 1974, pp. 119–147.
Richard Montague. Universal grammar. Theoria, 36:373–398, 1970. Reprinted in Formal Philosophy, by
Richard Montague, Yale University Press, New Haven, CT, 1974, pp. 222–246.
Richard Montague. The proper treatment of quantification in ordinary English. In Jaakko Hintikka, Julius
M.E. Moravcsik, and Patrick Suppes, editors, Approaches to Natural Language: Proceedings of the 1970
Stanford Workshop on Grammar and Semantics, pages 221–242. D. Reidel Publishing Co., Dordrecht, Hol-
land, 1973. Reprinted in Formal Philosophy, by Richard Montague, Yale University Press, New Haven,
CT, 1974, pp. 247–270.
Richard Montague. Formal Philosophy: Selected Papers of Richard Montague. Yale University Press, New
Haven, Connecticut, 1974. Edited, with an introduction, by Richmond H. Thomason.
Barbara Hall Partee. Montague grammar, mental representations, and reality. In Peter A. French, Theodore
E. Uehling, Jr., and Howard K. Wettstein, editors, Contemporary Perspectives in the Philosophy of Language,
pages 195–208. University of Minnesota Press, Minneapolis, 1978.
Barbara Partee. Semantics—mathematics or psychology? In Rainer Bäuerle, Urs Egli, and Arnim von Ste-
chow, editors, Semantics from Different Points of View, pages 1–14. Springer-Verlag, Berlin, 1979.
Barbara H. Partee. Possible worlds in model-theoretic semantics: A linguistic perspective. In Sture Alén,
editor, Possible Worlds in Humanities, Arts and Sciences, pages 93–123. Walter de Gruyter, Berlin, 1989.
Barbara H. Partee. The development of formal semantics in linguistic theory. In Shalom Lappin, editor, The
Handbook of Contemporary Semantic Theory, pages 11–38. Blackwell Publishers, Oxford, 1996.
Barbara H. Partee. Reflections of a formal semanticist as of Feb 2005. http://people.umass.edu/partee/docs/
BHP Essay Feb05.pdf
Barbara H. Partee. Reminiscences on philosophy-linguistics interactions in the beginnings of formal seman-
tics. https://udrive.oit.umass.edu/partee/RichfestPartee.ppt.pdf.
Barbara H. Partee and Herman L.W. Hendriks. Montague grammar. In Johan van Benthem and Alice ter
Meulen, editors, Handbook of Logic and Language, pages 5–91. Elsevier Science Publishers, Amsterdam,
1996.
Francis Jeffry Pelletier and Richmond H. Thomason. Twenty-five years of linguistics and philosophy. Lin-
guistics and Philosophy, 25(5–6):507–529, 2002.
Willard V.O. Quine. Methods of Logic. Holt, Reinhart and Winston, New York, 1940.
Willard V.O. Quine. Three grades of modal involvement. In Proceedings of the XIth International Congress of
Philosophy, Volume 14, pages 65–81, 1953.
Leonard Talmy. Toward a Cognitive Semantics: Concept Structuring Systems, volume 1. The MIT Press, Cam-
bridge, Massachusetts, 2000.

689
 RI CH M O N D H . T H O MA S O N

Leonard Talmy. Toward a Cognitive Semantics: Typology and Process in Concept Structuring, volume 2. The
MIT Press, Cambridge, Massachusetts, 2000.
Alfred Tarski. The semantic conception of truth and the foundations of semantics. Philosophy and Phenom-
enological Research, 4:341–375, 1944.
Richmond H. Thomason. Introduction. In Richmond H. Thomason, editor, Formal Philosophy: Selected
Papers of Richard Montague, pages 1–69. Yale University Press, New Haven, Connecticut, 1974.
Alfred North Whitehead and Bertrand Russell. Principia Mathematica. Cambridge University Press, Cam-
bridge, England, first edition, 1910–1913. Three volumes.

690
 Section VI

PH ILO SO P H Y O F
LANGUAGE FOR THE
REST O F PH I L O S O P H Y
 6.1
P HIL OS OPHY OF
L ANGUAGE F OR
E P IS T E M OLOGY
Ram Neta

The present article surveys some of the ways in which recent epistemology has been
informed by work in the philosophy of language. In particular, we focus on four impor-
tant issues.

1 Is Knowing How to Do Something a Form of Knowing that

Something Is the Case?
Gilbert Ryle (1949) begins chapter 2 as follows: “In this chapter I try to show that when
we describe people as exercising qualities of mind, we are not referring to occult episodes
of which their overt acts and utterances are effects; we are referring to those overt acts
and utterances themselves” (Ryle 1949, 25). In the course of trying to show this, Ryle
occasionally describes our qualities of mind, or the overt acts and utterances in which
we express those qualities, as “knowledge how,” and he also occasionally describes the
occult episodes spuriously posited as the causes of these acts or utterances as “knowledge
that.” Since the chapter in which these occasional descriptions occur is entitled “Know-
ing How and Knowing That,” it is commonly assumed that Ryle means to be claiming
that knowing how to do something is not a matter of knowing that something or other
is the case. (In fact, I find nothing in Ryle’s text that commits Ryle to this view. What
Ryle is committed to is that the know-how that is exhibited in our overt acts and utter-
ances is not a matter of our knowing any rules, and that it is not a matter of our acts and
utterances having any particular antecedents.)
But whether or not Ryle thought that knowing how was distinct from knowing that,
a number of other philosophers (for example, Putnam 1996, Lewis 1990, Devitt 1996)
have assumed that they are distinct. Is this assumption correct? The assumption was
challenged by Vendler 1972, who writes:

It will be objected here that in some cases of knowing how it is impossible to

tell, in words, what one knows. I know how to tie a necktie, but I could not tell
you in words alone. I grant this, but point out that this situation is possible with
nearly all the knowing wh forms. I know what coffee tastes like, what the color
 RA M N E TA

magenta looks like, where it itches on my back, when I should stop drinking,
how the coastline of Angola runs, but I could not tell you in words alone. I must
have, however, some other means to supplement words: pointing, offering a
sample, a sketch, a demonstration; or saying ‘now.’ By these means I can tell you,
or show you, what I know: I know that magenta looks like this (offering a sam-
ple), that it itches here (pointing), that I should stop drinking now. The need to
supplement words with non-linguistic media affects knowing how, and knowing
wh in general, exactly because it affects the corresponding knowing that.
(Vendler 1972, 104–5)

The point that Vendler makes in the passage just quoted is echoed by Ginet 1975 and
then substantially elaborated and defended by Stanley and Williamson 2001. Stanley
and Williamson argue for Vendler’s point by appeal to the syntax and semantics of
knowledge ascriptions involving embedded questions, for example,

John knows where the newspaper is.

James knows why the indicator light is on.
Tanya knows when the next show begins.
Sonia knows how to ride a bicycle.

The syntax of such sentences can be represented as follows:

John knows [where PRO the newspaper is t].

James knows [why PRO the indicator light is on t].
Tanya knows [when PRO the next show begins t].
Sonia knows [how PRO to ride a bicycle t].

PRO is an unpronounced pronoun that occurs in the subject position of the embedded
clauses, and t is a trace left by moving the “wh-” term (“where”, “why”, etc.) into subject
position. Syntactically, ascriptions of knowledge how are of the same form as ascriptions
of knowledge where, knowledge why, and knowledge when. Syntax thus suggests that knowl-
edge how is a kind of knowledge wh-.
But how should we understand knowledge wh- . . .? Since wh- clauses express ques-
tions, it is perhaps best to start by asking: what is the semantic value of a question?
According to the seminal Groenendijk and Stokhof 1984 (which was influenced by
Hamblin 1958), the semantic value of a question is the set of propositions that can
answer that question, consistently with the presuppositions of the question. For instance,
the question “When was Abraham Lincoln born?” has as its semantic value the set of
propositions P such that each member of P is a proposition of the form Abraham Lincoln
was born at t, where t is a time at which, consistently with the presuppositions of the
question (for example, that there was exactly one year in which Abraham Lincoln was
born), Abraham Lincoln could have been born. And the question “Why don’t human
beings have wings?” has as its semantic value the set of propositions Q such that each
member of Q is a proposition of the form Human beings don’t have wings because s, where
s is a proposition that, consistently with the presuppositions of the question, can explain
why human beings don’t have wings. In each case, the semantic value of a question is
a set of propositions that, consistently with the presuppositions of the question, can
constitute an answer the question.

694
 PH I LO SO PH Y O F LA N G U A G E F O R E P I S T E MO L O G Y

Given this semantics for questions, how should we understand ascriptions of knowl-
edge wh- . . .? It is standard to understand such ascriptions as follows: S knows wh- =
there is a such that p is the true answer to the question expressed by the wh- clause, and
S knows that p. Applying this generalization straightforwardly to the case of knowledge
how, we get the somewhat obscure result that S knows how to F = there is a p such that p
is the true answer to the question “how to F?”, and S knows that p. This result is obscure
because it is obscure what sort of question is intended by an interrogative sentence of
the form “How to F?” And that, Stanley and Williamson say, is because clauses of the
form [how PRO to F t] are ambiguous at two different points. First, the infinitive (for
example, “to ride a bicycle”) can be used to express a deontic modality, or a nondeontic
modality. And second, PRO can refer back to the subject of the unembedded clause (for
example, Sonia), or it can be used in a way that is equivalent to the English pronoun
“one”. Thus, we get four possible interpretations of “Sonia knows how to ride a bicycle.”
Roughly:

Sonia knows how she herself is able to ride a bicycle.

Sonia knows how she herself ought to ride a bicycle.
Sonia knows how one is able to ride a bicycle.
Sonia knows how one ought to ride a bicycle.

If we interpret “Sonia knows how to ride a bicycle” in the first of the four ways listed
above, then we can say this: Sonia knows how to ride a bicycle = there is a p such that
p is the answer to the question “How is Sonia able to ride a bicycle” and Sonia knows
that p. If we interpret “Sonia knows how to ride a bicycle” in the second of the four ways
listed above, then: Sonia knows how to ride a bicycle = there is a p such that p is the
answer to the question “How ought Sonia to ride a bicycle?” and Sonia knows that p.
If we interpret “Sonia knows how to ride a bicycle” in the third of the four ways, then:
Sonia knows how to ride a bicycle = there is a p such that p is the answer to the question
“How is one able to ride a bicycle?” and Sonia knows that p. And so on. But notice that,
on any of the four interpretations given above, for Sonia to know how to ride a bicycle
is for there to be some p (some proposition, or perhaps some fact–see the next section)
such that Sonia knows that p. And so, Stanley and Williamson conclude, knowing how
is a kind of knowing that.

2 Is it Propositions that Are Known, or Facts?

Russell 1918 coined the phrase “propositional attitude” to refer to psychological states
that have propositional contents: for example, beliefs, wishes, hopes, and so on. To say
that these psychological states have propositional content is to say that these psychologi-
cal states have a kind of representational content that is individuated more finely than
by its truth-conditions. Most epistemologists assume that knowledge has precisely this
same kind of representational content; they assume that what one knows is just the
same kind of thing that one believes, wishes, or hopes, and it is a kind of content that
is individuated more finely than by its truth-conditions. There seem to be a couple of
very simple and decisive arguments for this popular view. First, suppose I ask you: “Do
you believe that the ship has sailed, or do you know it?” It seems that I am asking whether
your relation to the proposition that the ship has sailed is the relation of knowledge or
of belief. But if it is reasonable for me to ask this question, so understood, then it is just

695
 RA M N E TA

as reasonable for me to suppose that knowledge, like belief, has propositional contents.
And so, it is generally thought, knowledge has propositional content, just as belief does.
Second, suppose that John knows that Superman is in New York and James knows that
Clark Kent is in New York. Superman is identical to Clark Kent, but does it follow that
John and James know the very same thing? It seems that this does not follow, and so it
may seem that the content of knowledge, like the content of belief, is individuated more
finely than by its truth-conditions. Again, it is concluded, knowledge has propositional
content, just as belief does.
Vendler 1972 challenged this orthodoxy, and the arguments for it. Vendler adduced
evidence for the hypothesis that, while belief is a relation to a proposition, knowledge
is a relation to something that is not a proposition. Knowledge, on Vendler’s view, is
a relation to a fact. What is the difference between a proposition and a fact? And why
does Vendler think that beliefs are relations to the former and knowledge to the latter?
Let’s address these questions in order.
The difference between a proposition and a fact is indicated by a couple of things.
First, some propositions are true and others are false, but facts are neither true nor false.
Facts, unlike propositions, do not have truth-values. Second, substitution of co-referring
terms in a clause need not preserve which proposition is denoted by that clause. But if
there is a fact that corresponds to that clause, then the very same fact corresponds to
any clause that results from substitution of co-referring terms in a referential transparent
context. For instance, the proposition that Hesperus is in the sky differs from the proposi-
tion that Phosphorus is in the sky: those who are ignorant of astronomy might reasonably
believe one of these propositions without believing the other. But the fact (if it is a fact)
that Hesperus is in the sky is none other than the fact that Phosphorus is in the sky.
Now that we’ve offered these two indications of the difference between facts and
propositions, let’s consider Vendler’s argument for the claim that we believe proposi-
tions, but we know facts. In support of this claim, Vendler adduces some evidence.
First, consider the following sentences:

(a) John believes that grass is green.

(b) James believes what John believes.
(c) James knows what John believes.
(d) John suspects that grass is green.
(e) James believes what John suspects.
(f) James knows what John suspects.

The conjunction of (a) and (b) implies that James believes that grass is green, and so
does the conjunction of (d) and (e). But the conjunction of (a) and (c) does not imply
this, nor does the conjunction of (d) and (f). But, if the only difference between (b)
and (c), and between (e) and (f), is the substitution of “knows” for “believes,” then why
should there be this difference?
Second, notice the difference between:

(g) James knows where John’s car is parked.

(h) * James believes where John’s car is parked.
(i) James knows why John parked his car there.
(j) * James believes why John parked his car there.

696
 PH I LO SO PH Y O F LA N G U A G E F O R E P I S T E MO L O G Y

(k) James knows how to park a car.

(l) * James believes how to park a car.

The first sentence of each of the above pairs is perfectly grammatical, but not the second
is clearly ungrammatical. Why?
Here’s how Vendler answers these questions. Knowledge, for Vendler, is a relation that
we bear to facts, and facts, unlike propositions, are denoted by wh- clauses, for example,
“whom Onassis married”, “when the party will start”, “why the isotope decayed”, and
so on. (Notice that wh- clauses do not have truth-values, and are referentially transpar-
ent.) That’s why (g), (i), and (k) are grammatical. But belief is a relation that we bear to
propositions, not to facts. That’s why (h), (j), and (l) are ungrammatical.
But then if belief contents cannot be denoted by wh- clauses, why are (b) and (e)
grammatical? (b) and (e) are grammatical because, as it occurs in those sentences, the
phrase “what John believes” is not a wh- clause–in fact, it is not a clause at all. Rather, it
means the same as “that which John believes,” and this is a singular term that refers to
whatever it is (a proposition, according to Vendler) which John believes. Since knowl-
edge cannot have propositional content, the phrase “what John believes” cannot mean
the same thing in (c) and (f) as it means in (b) and (e); in (c) and (f) “what John
believes” is a wh- clause.
But if this is right, then how can we explain away the seemingly plausible arguments
given above, to the effect that what one person can believe another can know, or that
the contents of our knowledge are individuated more finely than by their truth-condi-
tions? When I ask “Do you believe that the ship has sailed, or do you know it?,” the pro-
noun “it” does not refer anaphorically to the proposition that the ship has sailed; in fact,
according to Harman 2002, this pronoun is not being used referentially at all. Rather, it
is what Geach 1975 calls a “pronoun of laziness.” It is to be interpreted as standing in for
“that the ship has sailed.” (Harman offers another example of such a pronoun in “The
man who gives his paycheck to his wife is better off than the man who gives it to his
mistress,” where “it” does not refer back to the first man’s paycheck, but instead stands
in for the phrase “his paycheck.”)
What about the argument that the contents of our knowledge are individuated more
finely than by their truth-conditions? It’s true that, if John knows that Superman is
in New York and James knows that Clark Kent is in New York, then John and James
need not know the same thing. But that’s not because John’s knowledge and James’s
knowledge are of propositions with different truth-conditions. Rather, it’s because
John knows one fact–namely, where Superman is–and James knows another fact–
namely, where Clark Kent is. And recall that facts, unlike propositions, don’t have
truth-conditions.

3 What, Besides the Things Mentioned in 2, Are the Relata of the

Knowledge Relation?
Let’s use the term “knowable” as a count noun, to denote those things (be they proposi-
tions or facts) that are known. We can all, presumably, agree that for one of these know-
ables to be known, it must stand in the knowledge relation to a knower at a time, and
perhaps in a world. So the knowledge relation involves, at least, a knowable, a knower,
a time, and perhaps a world. But does it have any other relata besides these? According
to contrastivists, it does have at least one other relatum: namely, a contrast. But what is a

697
 RA M N E TA

contrast, and on what grounds do contrastivists argue that any such thing is one of the
relata of the knowledge relation?
Let’s begin by noticing that many, if not most, knowledge ascriptions are not of the
form “S knows that p”, but rather of the form “S knows wh- . . .” (for example, “I know
when the movie starts”, “she knows where the plaza is”, “he knows why the Senate
Finance Committee rejected the public option”, “they know how to speak Portuguese”).
But how should we understand knowledge wh- . . .? Recall from section 1 above that it is
standard to understand such ascriptions as follows: S knows wh- = p is the true answer to
the question expressed by the wh- clause, and S knows that p. This view is propounded
by Hintikka 1975, Lewis 1982, Boer and Lycan 1986, Higginbotham 1996, and Stanley
and Williamson 2001, among others. But, according to Schaffer 2007, this popular view
falsely predicts that whenever two questions have the same answer, knowing the answer
to one question implies knowing the answer to the other. Consider, for instance, the
following two questions:

(a) Is the capital of North Carolina Raleigh or Charlotte?

(b) Is the capital of North Carolina Raleigh or Los Angeles?

It’s possible for someone know the answer to easy question (b) without knowing the
answer to hard question (a). But since the questions have the same answer–namely,
that Raleigh is the capital of North Carolina–it follows that knowing the answer to
either question cannot simply consist in knowing that Raleigh is the capital of North
Carolina. More generally, Schaffer claims, knowing wh- cannot consist in knowing that
p, where p is the true answer to the question expressed by the wh- clause.
Schaffer 2005 and Schaffer 2007 argue that the popular attempt to reduce knowing
wh- to knowing that gets things backwards. The reduction goes in the other direction:
To know that p is to know the answer to some question that has p as its true answer.
To know the answer to some question that has p as its answer is to know that, among
the members of the set of propositions that is the semantic value of that question, the
one true member is p. For instance, to know that Abraham Lincoln was born in 1809
is to know the answer to some question that has as its true answer the proposition that
Abraham Lincoln was born in 1809. What question could this be? Well, it could be
the question “When was Abraham Lincoln born?”; or it could be the question “Which
American president was born in 1809?”; or it could be the question “What is the signifi-
cance of the year 1809 in Abraham Lincoln’s life?”; or it could be any number of other
questions. To know that Abraham Lincoln was born in 1809 is to stand in a particular
relation–the knowledge relation–to some such question, as well as to the proposition
Abraham Lincoln was born in 1809.
Defenders of noncontrastivist orthodoxy may challenge Schaffer’s argument for
contrastivism by claiming that what makes question (a) above harder than question
(b) is simply that, when a listener hears question (b), she gets more useful information
than she does when she hears question (a). Upon hearing question (b), the listener
might reason as follows: “I am being asked whether the capital of North Carolina is
Raleigh or Los Angeles, so it is probably one or the other. But it is certainly not Los
Angeles, so it is probably Raleigh.” The same reasoning is not available to someone
who hears question (a). And so, the objection goes, the reason that someone could
know the answer to question (b) without knowing the answer to question (a) is not
that knowing the answer to one question differs from knowing the answer to the other

698
 PH I LO SO PH Y O F LA N G U A G E F O R E P I S T E MO L O G Y

but rather because hearing question (b) puts one in a better epistemic state than does
hearing question (a).
In response to this challenge and several others, Schaffer and Knobe forthcoming
have attempted a defense of contrastivism by appeal to a much wider data set, gathered
from surveys concerning hypothetical cases. It remains to be seen whether contrastivism
provides the best explanation of the large body of survey data that they have gathered.

4 How to Explain the Apparent Context Sensitivity of

Knowledge Ascriptions?
What is meant by the “apparent context sensitivity of knowledge ascriptions”. We can
explain this by focussing on a famous passage from DeRose 1992:

Consider the following cases:

Bank Case A. My wife and I are driving home on a Friday afternoon. We plan
to stop at the bank on the way home to deposit our paychecks. But as we drive
past the bank, we notice that the lines inside are very long, as they often are on
Friday afternoons. Although we generally like to deposit our paychecks as soon
as possible, it is not especially important in this case that they be deposited
right away, so I suggest that we drive straight home and deposit our paychecks
on Saturday morning. My wife says, ‘Maybe the bank won’t be open tomorrow.
Lots of banks are closed on Saturdays.’ I reply, ‘No, I know it’ll be open. I was
just there two weeks ago on Saturday. It’s open until noon.’
Bank Case B. My wife and I drive past the bank on a Friday afternoon, as in Case A,
and notice the long lines. I again suggest that we deposit our paychecks on Saturday
morning, explaining that I was at the bank on Saturday morning only two weeks
ago and discovered that it was open until noon. But in this case, we have just writ-
ten a very large and very important check. If our paychecks are not deposited into
our checking account before Monday morning, the important check we wrote will
bounce, leaving us in a very bad situation. And, of course, the bank is not open on
Sunday. My wife reminds me of these facts. She then says, ‘Banks do change their
hours. Do you know the bank will be open tomorrow?’ Remaining as confident as
I was before that the bank will be open then, still, I reply, ‘Well, no. I’d better go
in and make sure.’
Assume that in both cases the bank will be open on Saturday and there is noth-
ing unusual about either case that has not been included in my description of
it. It seems to me that (1) when I claim to know that the bank will be open
on Saturday in case A, I am saying something true. But it also seems that (2) I
am saying something true in Case B when I concede that I don’t know that the
bank will be open on Saturday. Yet I seem to be in no better position to know
in Case A than in Case B. It is quite natural to say that (3) If I know that the
bank will be open on Saturday in Case A, then I also know that it will be in
Case B. (DeRose 1992, 913)

Now, in question 4, the phrase “the apparent context sensitivity of knowledge ascrip-
tions” is intended to denote a particular phenomenon, and Bank Cases A and B
are intended to serve as examples of that phenomenon. What’s puzzling about this

699
 RA M N E TA

phenomenon is brought out by the apparent conflict between the claims that DeRose
calls (1), (2) and (3) in the last quoted paragraph. If we accept the conditional (3)–
namely, that if I know that the bank will be open on Saturday in Case A, then I also
know that it will be in Case B–then, how can my ascription of this knowledge to myself
in Case A be true, while my denial of knowledge to myself in Case B is also true? To
explain the apparent context sensitivity of knowledge ascriptions, in the sense in which
IV requests such an explanation, is to answer this question.
Many philosophers have, in effect, rejected the question by denying (3). For
instance, Fantl and McGrath 2002, Hawthorne 2004, and Stanley 2005 all argue
that whether one knows that p can depend upon the (actual or perceived) costs of
falsely believing that p. When the costs are higher, the other necessary conditions of
knowledge are more stringent. If this view is correct, then, given that the cost of
the bank’s being closed on Saturday is higher in case B than in case A, it is open to
us to claim that, despite the believer’s being just as confident of the bank’s being open
in case B as in case A, it is nonetheless true that the believer knows in case A
but doesn’t know in case B. That is one prominent basis on which (3) has been
denied.
Another basis on which (3) has been denied is given by Bach 2005, who finds implau-
sible the stipulation that the agent is just as confident of the bank’s being open in case
B as she is in case A. Given the dialogue described in case B, wouldn’t the agent’s con-
fidence in the bank’s being open naturally tend to drop? And if his confidence didn’t
drop, then wouldn’t this indicate that the agent’s confidence in the bank’s being open
is not entirely rational, and therefore not knowledge-constituting? Bach, thinking that
any failure on the agent’s part to lower his confidence in the bank’s being open would
be irrational, and thereby undermine the agent’s knowledge, rejects (3), and so rejects
the demand to explain what we are here calling the apparent context sensitivity of
knowledge ascriptions.
But while the philosophers mentioned above have rejected (3), most philosophers
would not be willing to do so. They are then confronted with the following (nonexclu-
sive) choices: they could either reject (1), or they could reject (2), or they could reject
the conjunction of (1) and (2) without rejecting either conjunct (more on this third
option in a moment), or, finally, they could accept both (1) and (2) and claim that the
truth of both (1) and (2) is somehow, despite appearances, consistent with the truth of
(3). Only sceptics adopt the first of these three options (see, for example, Unger 1975):
there is nothing special about (1) that renders it less worthy of acceptance than most
of our everyday knowledge ascriptions. The only basis for rejecting (1), then, would
equally well serve as a basis for rejecting most of our everyday knowledge ascriptions,
and to do that is nothing short of scepticism.
But what of the other three options? Let me consider them in reverse order, starting
with the fourth. Philosophers who support the fourth option call themselves “contex-
tualists,” and we will follow their usage here. (I should note, however, that the term
“contextualism” has been used by other philosophers to denote views that are com-
pletely different from those being discussed here, and are rather designed to respond to
the Agrippan regress, see, for example, Annis 1978 and Williams 1991.) Contextualists
claim that (1) and (2) are consistent: they can both be true. They can both be true
because the token sentence that I utter in Case A–“I know that it’ll be open”–is not,
according to contextualists, the negation of the token sentence that I utter in Case
B–“Well, no [namely, I don’t know that the bank will be open].” The latter might seem

700
 PH I LO SO PH Y O F LA N G U A G E F O R E P I S T E MO L O G Y

to be the negation of the former, but this appearance is misleading: in fact, the two can
have the same truth value, say contextualists.
How can the two sentences have the same truth value? Contextualists can hold dif-
ferent views on this issue. We can frame the options available to a contextualist by
appeal to the following prominent account of what it is for a sentence, uttered in a par-
ticular context, to be true: “If c is a context, then an occurrence of F in c is true iff the
content expressed by F in this context when evaluated with respect to the circumstance
of the context” (Kaplan 1989).
According to most contextualists (e.g., Cohen 1988, DeRose 1992, Lewis 1996, Hel-
ler 1999, Rieber 1998, Neta 2002, Schaffer 2005) the reason why the token sentence
that I utter in Case A is not the negation of the token sentence that I utter in Case B
is that the content expressed by the sentence depends upon the context in which the
sentence is tokened. Knowledge-ascribing and knowledge-denying sentences do not
express the same content in different contexts, these contextualists say. Precisely how
the content of the sentence depends upon the context – what features of the context it
depends upon, just how it depends upon them – is in dispute among these contextual-
ists. For instance, Cohen 1988 takes the content of knowledge-ascribing and knowl-
edge-denying sentences to depend upon a standard of justification that is relevant in
that context of utterance; DeRose 1995 takes it to depend upon a standard of closeness
of possible worlds that is relevant in that context; and so on. But all of these contextual-
ists agree that knowledge-ascribing and knowledge-denying sentences express different
contents in different contexts of utterance. In contrast to this popular form of contextu-
alism, MacFarlane 2009 sketches what he calls a “nonindexical contextualism,” accord-
ing to which the content of knowledge-ascribing and knowledge-denying sentences
does not depend upon their context of utterance, but their truth-value does so vary,
because different contexts of utterance establish a different circumstance of evaluation
for the expressed content.
That is an extremely brief sketch of some of the many varieties of contextualism.
What unites all of them is that, when confronted with DeRose’s Bank Cases, they will
all claim both that I am saying something true in Case B when I concede that I don’t
know that the bank will be open on Saturday, and that I am saying something true in
Case A when I claim to know that the bank will be open on Saturday, even though they
also grant that if I know in Case A, then I know in Case B.
Now, let’s consider the alternative, noncontextualist responses to the Bank Cases.
In particular, we can sketch those views according to which I speak falsely (and yet
appropriately) when I say, in Case B, that I do not know that the bank is open on
Saturday. According to all of these responses, when I say, in Case B, that I do not know
that the bank will be open tomorrow, even though what I say is false, it is nonetheless
appropriate for me to say it. That’s because, by saying it, I give it to be understood that I
cannot simply rely on the bank’s being open tomorrow – and what I give to be under-
stood is both true and conversationally relevant. Had I instead said that I do know
that the bank is open tomorrow, I would have given it to be understood that I can
simply rely on the bank’s being open tomorrow – and what I would then have given to
be understood is false. This is why, according to all of the noncontextualist responses
that fall into this group, I speak falsely and yet appropriately when I say, in case B, that
I do not know that the bank will be open on Saturday. Where these noncontextual-
ist views differ is in their account of the means by which I give something true to be
understood in case B.

701
 RA M N E TA

According to Stainton 2010 and Harman 2007, when I say, in Case B, that I do
not know that the bank will be open tomorrow, then even though the sentence that I
assertorically utter is false, it is nonetheless appropriate for me to assert it, and that is
because the content of my assertion is not the same as the content of the sentence that I
utter. The content of my assertion – roughly, what it is that I claim to be true – is clearly
true, even though the sentence that I assertorically utter is not true. This can be the
case since, as noted above, the content of my assertion is not always the same as, and
not exclusively determined by, the content of the sentence that I utter in performing
that act.
According to Rysiew 2007 and Brown 2005, when I say, in Case B, that I do not know
that the bank will be open tomorrow, even though what I say is false, it is nonethe-
less appropriate for me to say it. That’s because, by saying it, I conversationally implicate
that I cannot simply rely on the bank’s being open tomorrow – and what I give to be
understood is both true and conversationally relevant. (To say that I conversationally
implicate a certain content is to say that my assertion is conversationally cooperative
only if I aim to give it to be understood that that content is true. See Grice 1975.) Had
I instead said (truthfully) that I do know that the bank is open tomorrow, I would have
conversationally implicated that I can simply rely on the bank’s being open tomorrow
– and what I would then have implicated is false. This, according to Rysiew and Brown,
is why I speak falsely, and yet still appropriately, in Case B.
Finally, according to a position suggested by Turri 2010, when I say, in Case B, that
I do not know that the bank will be open tomorrow, I speak appropriately not because
the content of my assertion or of my implicature is true. Rather, I speak appropriately
because, given the high stakes I face in case B, a serious and sincere utterance of “I know
that the bank will be open tomorrow” in that case would not count as a mere assertion
that I know that the bank will be open tomorrow but as an illocutionary act of a stronger
kind–call it a “guarantee” that I know that the bank will be open tomorrow. But, while
I might know that the bank will be open tomorrow, and while I might even be in a posi-
tion to assert this, I am certainly in no position to guarantee this.
Contextualists’ chief worry about all of these noncontextualist attempts to explain
why I speak falsely, and yet nonetheless appropriately, in case B when I say that I don’t
know whether the bank will be open on Saturday is just this: While all of these non-
contextualist views make it clear why it would be inappropriate for me to claim to have
knowledge in case B, none of them posit a well-precedented mechanism to explain why
it would be appropriate for me to claim not to have knowledge in case B.
There is one more alternative response to the bank cases that has been much dis-
cussed lately, and this is the form of relativism espoused by Richard 2004 and MacFar-
lane 2005. According to their view, Kaplan’s account (stated above) of what it is for
a sentence, uttered in a particular context, to be true is too simple. Sentence truth is
not determined simply by the content of a sentence, along with its context of utterance
(and the circumstance of evaluation determined by that context of utterance). Rather it
is also relative to a context of assessment. The very same content expressed in the very
same context of utterance (fixing one and the same circumstance of evaluation) can be
true relative to one context of assessment and false relative to another. Relativists claim
that, relative to any particular context of assessment, either my knowledge-ascription in
case A will be false and my knowledge-denial in case B will be true, or vice versa. But,
they will say, there is no context of assessment relative to which both my knowledge-
ascription in case A and my knowledge-denial in case B will count as true.

702
 PH I LO SO PH Y O F LA N G U A G E F O R E P I S T E MO L O G Y

References
Annis, D. (1978) “A Contextualist Theory of Epistemic Justification,” American Philosophical Quarterly 15:
213–219.
Bach, K. (2005) “The Emperor’s New ‘Knows’ ” in G. Preyer and G. Peter (eds.) Contextualism in Philosophy,
Oxford: Oxford University Press.
Boer, S. and Lycan, W. (1986) Knowing Who, Cambridge, MA: MIT Press.
Brown, J. (2005) “Comparing Contextualism and Invariantism on the Correctness of Contextualist Intui-
tions,” Grazer Philosophische Studien 69: 71–100.
Cohen, S. (1988) “How to be a Fallibilist,” Philosophical Perspectives 2: 91–123.
DeRose, K. (1992) “Contextualism and Knowledge Attributions,” Philosophy and Phenomenological Research
52: 913–29.
—— (1995) “Solving the Skeptical Problem,” Philosophical Review 104: 1–52.
Devitt, M. (1996) Coming to Our Senses, Cambridge: Cambridge University Press.
Fantl, J. and McGrath, M. (2002) “Evidence, Pragmatics, and Justification,” Philosophical Review 111:
67–94.
Geach, P. (1975) “Back-Reference,” Philosophia 5: 192–206.
Ginet, C. (1975) Knowledge, Perception, and Memory, Ithaca: Cornell University Press.
Grice, P. (1975) “Logic and Conversation” in P. Cole and J. Morgan (eds.) Syntax and Semantics 3: Speech
Acts, New York: Academic Press.
Groenendijk, J. and Stokhof, M. (1984) Studies in the Semantics of Questions and the Pragmatics of
Answers, University of Amsterdam dissertation.
Hamblin, C. (1958) “Questions,” Australasian Journal of Philosophy 36: 159–68.
Harman, G. (2002) “Reflections on Knowledge and its Limits,” Philosophical Review 111: 417–28.
—— (2007) “Epistemic Contextualism as a Theory of Primary Speaker Meaning,” Philosophy and Phenom-
enological Research 75: 173–9.
Hawthorne, J. (2004) Knowledge and Lotteries, Oxford: Oxford University Press.
Heller, M. (1999) “The Proper Role for Contextualism in an Anti-Luck Epistemology,” Philosophical Perspec-
tives 13: 115–29.
Higginbotham, J. (1996) “The Semantics of Questions” in S. Lappin (ed.) The Handbook of Contemporary
Semantic Theory, Oxford: Oxford University Press.
Hintikka, J. (1975) The Intensions of Intentionality and Other New Models for Modality, Dordrecht:
Kluwer.
Kaplan, D. (1989) “Demonstratives” in J. Almog, J. Perry, and H. Wettstein (eds.) Themes from Kaplan, New
York: Oxford University Press.
Lewis, D. (1982) “Whether Report” in T. Pauli (ed.) Philosophical Essays Dedicated to Lennart Aqvist on His
Fiftieth Birthday, Uppsala: University of Uppsala Press.
—— (1990) “What Experience Teaches” in W. Lycan (ed.) Mind and Cognition: A Reader, Blackwell:
Oxford.
—— (1996) “Elusive Knowledge,” Australasian Journal of Philosophy 74: 549–67.
MacFarlane, J. (2005) “The Assessment Sensitivity of Knowledge Attributions,” in T. Gendler and J. Haw-
thorne (eds.) Oxford Studies in Epistemology, Oxford: Oxford University Press.
—— (2009) “Nonindexical Contextualism,” Synthese 166: 231–50.
Neta, R. (2002) “S Knows that P,” Noûs 36: 663–89.
Putnam, H. (1996) “Introduction” in A. Pessin and S. Goldberg (eds.) The Twin Earth Chronicles, Armonk,
NY: M.E. Sharpe.
Richard, M. (2004) “Contextualism and Relativism,” Philosophical Studies 119: 215–42.
Rieber, S. (1998) “Skepticism and Contrastive Explanation,” Noûs 32: 189–204.
Russell, B. (1918) “The Philosophy of Logical Atomism,” Monist
Ryle, G. (1949) The Concept of Mind, London: Methuen.
Rysiew, P. (2007) “Speaking of Knowing,” Noûs 41: 627–62.
Schaffer, J. (2005) “Contrastive Knowledge,” in T. Gendler and J. Hawthorne (eds.) Oxford Studies in Epis-
temology 1, Oxford: Oxford University Press.
—— (2007) “Knowing the Answer,” Philosophy and Phenomenological Research 75: 383–403.
Schaffer, J. and Knobe, J. (forthcoming) “Contrastive Knowledge Surveyed.”
Stainton, R. (2010) “Contextualism in Epistemology and the Context Sensitivity of ‘Knows’” in
J.C. Campbell, M. O’Rourke, and H. Silverstein (eds.) Knowledge and Skepticism, Cambridge, MA: MIT
Press.

703
 RA M N E TA

Stanley, J. (2005) Knowledge and Practical Interests, Oxford: Oxford University Press.
Stanley, J. and Williamson, T. (2001) “Knowing How,” Journal of Philosophy 98: 411–44.
Turri, J. (2010) “Epistemic Invariantism and Speech Act Contextualism,” Philosophical Review 119:
Unger, P. (1975) Ignorance, Oxford: Oxford University Press.
Vendler, Z. (1972) Res Cogitans, Ithaca: Cornell University Press.
Williams, M. (1991) Unnatural Doubts, Princeton: Princeton University Press.

704
 6.2
P HIL OS OPHY OF
L ANGUAGE F OR
ME T AE T HIC S
Mark Schroeder

Metaethics is the study of metaphysics, epistemology, the philosophy of mind, and the
philosophy of language, insofar as they relate to the subject matter of moral or, more
broadly, normative discourse—the subject matter of what is good, bad, right or wrong,
just, reasonable, rational, what we must or ought to do, or otherwise. But out of these
four ‘core’ areas of philosophy, it is plausibly the philosophy of language that is most cen-
tral to metaethics—and not simply because ‘metaethics’ was for a long time construed
more narrowly as a name for the study of moral language. The philosophy of language is
central to metaethics because both the advantages of and the open problems facing dif-
ferent metaethical theories differ sharply over the answers those theories give to central
questions in the philosophy of language. In fact, among the open problems over which
such theories differ are included, particularly, further problems in the philosophy of
language. This chapter briefly surveys a range of broad categories of views in metaeth-
ics and both catalogues some of the principal issues faced by each in the philosophy of
language, as well as how those arise out of their answers to more basic questions in the
philosophy of language. I make no claim to completeness, only to raising a variety of
important issues.

1 Cognitivist-realist Theories
Probably the default view in metaethics is to assume that moral language does not differ
in any deep or important way from nonnormative language, so that sentences like ‘steal-
ing is wrong’ or ‘it’s good that Max is happy’, like sentences like ‘stealing is common’ or
‘it seems that Max is happy’ are about some subject matter in the world—what is wrong
or good. On the simplest version of such a view, ‘it’s good that Max is happy’ expresses
a single, invariant proposition across all contexts of utterance, but much the same issues
face more careful views according to which all context-dependence of ‘it’s good that
Max is happy’, if there is any, is of an ‘unsurprising’ sort—due to facts such as that ‘good’,
like ‘tall’, is a gradable adjective, or due to its tense.
One of the primary issues facing such metaethical theories in the philosophy of lan-
guage concerns how we manage to talk about normative properties or facts. Conse-
 M A RK SCH R O E D E R

quently, metaethical realists are particularly interested in, or at least ought to be particu-
larly interested in, the theory of reference—metasemantics, as it is sometimes called. The
issues of importance to metaethical realists in the theory of reference, however, depend
on the realist’s answer to the metaphysical question of just what normative sentences
are about. According to the nonreductive metaethical realist, normative sentences are
about a sui generis normative domain, separate from any natural fact investigatable by
the rest of the natural or social sciences. Consequently, nonreductive metaethical real-
ists have a large stake in the debate between causal and inferential-role theories of refer-
ence—if sui generis normative facts are causally inert, then a causal theory of reference
will not enable us to think or talk about them. This drives some work by metaethical
realists outside of the philosophy of language—to account for how nonreductive, sui
generis normative facts could in fact play a causal role (see particularly Sturgeon [1985],
[2006] and Thomson [1996])—and it motivates a general interest in noncausal theories
of reference, for those who don’t accept this conclusion (see particularly Wedgwood
[2007]).
Reductive metaethical realists, in contrast, hold that normative properties or facts
can be analyzed in terms of nonnormative or natural properties or facts—so that the
subject matter of normative sentences is part of the world revealed to us by the natural
and social sciences. Reductive realists’ stake in the theory of reference is different: they
must be able to explain why the analyses that they provide are surprising. Reductive
metaethical realists therefore have a deep stake in understanding philosophical analysis
in general and in the philosophy of language underlying surprising property identities,
in particular. Much of metaethical inquiry has been concerned with this set of issues,
and the most extensive work by metaethical realists in the philosophy of language has
been devoted to it (see particularly Moore [1903], Boyd [1988], Brink [1989], Railton
[1986], Smith [1994], and Jackson [1998]).
Reductive realists have been quick to point out that philosophy is full of surpris-
ing identities—including the identity of water with H2O. But things are not quite so
easy. It is familiar how the identity of water and H2O could be surprising—because it is
an empirical finding—and familiar how it could be an empirical finding—because
‘water’ could have referred to something else, if our environment had been differ-
ent, even though people took it to have the same inferential significance. But it is far
from clear that ‘wrong’ could easily have referred to something else, simply because
our environment had been different and even though people took it to have the same
inferential significance—on the contrary, if we imagine cases in which people call dif-
ferent things wrong merely because of a difference in their environment, it is fairly
easy to pump the intuition that we simply disagree with these people about what is
wrong. (See particularly Hare [1952], Dreier [1990], and Horgan and Timmons [1991],
[1992].)
The apparent non-‘twin-earthability’ of ‘wrong’ raises general issues in the philoso-
phy of language about whether surprising analyses like the claim that water is H2O
must be aposteriori (or might be synthetic without being aposteriori) and about whether
making an aposteriori discovery always involves ruling out some genuine metaphysical
possibility—as discovering that water is H2O involves ruling out the possibility that
we are on a twin Earth. This has interested metaethicists in whether other terms—for
example, for functional kinds—exhibit similarly twin-Earth resistant behavior, and in
examples of difficult-to-discover but non-aposteriori analyses outside of metaethics—for
example, the mathematical definition of continuity (see particularly Copp [2000], van

706
 PH I LO SO PH Y O F LA N G U AG E F O R ME TA E T H I C S

Roojen [2006], Schroeder [2007b]). It also raises general questions from the philosophy
of language about how people could succeed at talking about the same thing as one
another, even while having very different ideas about the extension of those terms (see
particularly Horgan and Timmons [2000], Dreier [1993]).

2 Contextualist Theories
As suggested earlier, context-dependence in moral sentences that is due to unsurpris-
ing features such as tense or the presence of gradable adjectives is something for which
ordinary realist theories in metaethics can allow. But some theories—interestingly con-
textualist theories—hold not only that ‘it is good that Max is happy’ expresses different
propositions in different contexts of utterance but that it does so even when controlling
for such unsurprising sorts of context sensitivity. It is possible, of course, for a nonre-
ductive metaethical realist to also endorse this kind of surprising contextualism, but in
practice, surprising contextualist theories are used primarily to make it easier to locate
normative facts in the natural world. On such views, an important part of why it does
not seem to be possible to identify moral facts with any particular natural facts is that
moral claims express different natural facts, in different contexts of utterance (see par-
ticularly Dreier [1990], Timmons [1999], and Finlay [2004], [2009]).
The traditional form taken by metaethical contextualism is speaker subjectivism. The
most naive forms of speaker subjectivism identify the meanings of ‘stealing is wrong’ and
‘I disapprove of stealing’, for some attitude verb like ‘disapprove’, which picks out a
negative desire-like attitude. More cautious forms identify not the meaning of these two
expressions but their truth relative to contexts of utterance and assignments (see Ayer
[1936] for an argument which motivates this move). More cautious forms yet allow that
there may not be any word in ordinary English for the requisite desire-like attitude, so
‘disapprove’ needs to be understood as a stipulative technical term (see Urmson [1968]
for an argument that motivates this move). Contemporary twists on speaker subjec-
tivism include actually rigidified speaker subjectivism, and speaker-group or conversa-
tional subjectivism, which appeals to something more like ‘we disapprove of stealing’
(for both, see Dreier [1990]).
Another contemporary twist on speaker subjectivism is to hold that there is a func-
tion from the speaker’s attitudes to the content of ‘stealing is wrong’ but that these con-
tents are not themselves about the speaker’s attitudes. This view can be thought of as
identifying the contents of ‘stealing is wrong’ and something like ‘stealing instantiates
dthat[the property I disapprove of actions for instantiating]’ (compare Barker [2000],
Schroeder [2009]). Even more flexible views require that the context contribute some
parameter, and allow that in some contexts this parameter is fixed by the speaker’s
attitudes or interests but also allow for ‘disengaged’ contexts in which the parameter
can be fixed by other people’s attitudes or interests, in the way that it is plausible that
uses of ‘local’ refer by default to the neighborhood of the speaker’s location but can also
easily refer to the neighborhood of some other location (see Timmons [1999], Finlay
[2009]).
Contextualists in metaethics face a set of problems familiar from other areas of phi-
losophy. In particular, natural tests for context-dependence, including the behavior
of cross-contextual indirect discourse reports and truth ascriptions, seem to yield the
wrong results for many normative sentences. For example, consider the following bit of
reasoning:

707
 M A RK SCH R O E D E R

Immanuel (in one context): Lying is always wrong.

Jeremy (in a different context): Immanuel just said that lying is always
wrong.
Jeremy: But lying is not always wrong.
Jeremy: So Immanuel just said something false.

On the face of it, Jeremy’s reasoning in this scenario makes perfect sense. But from the
contextualist’s point of view, it is very puzzling. For either Jeremy is correctly report-
ing what Immanuel said or he is making an error. If he is correctly reporting what
Immanuel said, then since Immanuel’s context is different and hence he did not in
fact say anything inconsistent with what Jeremy means by going on to say that lying is
always wrong, it would seem to follow that Jeremy’s inference is invalid. But it seems
both that Jeremy can correctly report what Immanuel said in this way, and that he can
legitimately go on to reason in this way—at one and the same time. Sophisticated con-
textualist views can account for the correctness of the indirect discourse report (in some
contexts), and can account for the validity of the inference (in other contexts), but the
hard thing is accounting for both at the same time (and in the same context—compare
Schroeder [2009]).
Another important and related issue facing contextualists is how to explain how
speakers are able to engage in moral disagreements with one another. This problem
was particularly acute for speaker-subjectivists, for whom even speakers in the same
conversation were in relevantly different contexts; focusing instead on the speaker’s
conversational partners allows people in the same conversation to disagree but still
runs into challenges explaining disagreement across conversations (compare Dreier
[1990]).

3 Relativism
An alternative to metaethical contextualism is genuine relativism. According to
genuine relativism, though ‘stealing is wrong’ expresses the same proposition in
different contexts of utterance (again controlling for possible effects of less controver-
sial context-dependent features), that proposition is not true or false simpliciter but may
be assessed differently as true or false from different perspectives (see MacFarlane
[2005], [2007]). Like metaethical contextualism, relativism is compatible with
nonreductive views in metaethics, but it is hard to see what the motivation for
such a position would be; relativism is more naturally motivated on similar grounds
to those for contextualism—the relativist may claim that the truth of a normative
proposition relative to a given point of view is itself a straightforward natural matter of
fact.
As a matter of the history of the literature, metaethicists have been slow to embrace
and articulate genuinely relativist views; despite the historical importance of moral
relativism, many careful discussions of relativism in metaethics interpret it as a kind of
contextualism (compare Harman [1975], [1978], Wong [1984]). This is in part due to
the fact that the resources for articulating relativism in clearly coherent ways are a more
recent development in the philosophy of language. Given the historical importance of
relativism in metaethics—dating at least to the presocratics—the philosophical issues
surrounding relativism are particularly important for metaethics.

708
 PH I LO SO PH Y O F LA N G U AG E F O R ME TA E T H I C S

4 Noncognitivist Theories: Expressivism

Probably the most interesting issues arising for metaethics in the philosophy of language
arise for metaethical theories which depart quite radically from conventional ideas
about semantic content. These include all of the theories in the ‘noncognitivist’ tradi-
tion exemplified by Hägerström, Ogden and Richards, Schlick and Carnap, Ayer and
Stevenson, Hare, Blackburn, and Gibbard. The most developed contemporary strand
of this tradition is the contemporary theory of metaethical expressivism, according to
which rather than understanding the meaning of normative sentences in terms of what
propositions they express or under what conditions they are true, we should understand
the meaning of normative sentences in terms of what it is to have the corresponding
normative thoughts (see especially Blackburn [1993] and Gibbard [1990] and [2003]).
From the expressivist’s point of view, speaker subjectivists were on the right track to
think that the right way to understand the meaning of ‘stealing is wrong’ is to compare it
to ‘I disapprove of stealing’, but speaker subjectivists went too far to identify the mean-
ing or even the truth conditions of these two sentences.
Compare, for example, what would happen if we identified the meanings of ‘grass
is green’ and ‘I believe that grass is green’. It follows from that view that anyone who
believes that grass is green and says ‘grass is green’ speaks truly. This is analogous to the
prediction of speaker subjectivism that anyone who disapproves of stealing and says
‘stealing is wrong’ speaks truly. Moreover, this view predicts that ‘if I didn’t believe
that grass is green, then grass wouldn’t be green’ expresses a truth in the mouth of
every speaker; this is analogous to the prediction of speaker subjectivism that ‘if I didn’t
disapprove of stealing, then it wouldn’t be wrong’ expresses a truth in the mouth of
every speaker. So expressivists reason that speaker subjectivists were making the very
same mistake as this view about the meaning of ‘grass is green’ does. Since we all know
that the right relationship between ‘grass is green’ and ‘I believe that grass is green’
does not lead to these problems, expressivists propose to postulate that relationship
between ‘stealing is wrong’ and ‘I disapprove of stealing’. We can stipulatively say that
the first member of each pair expresses the attitude that the second member reports (see
Schroeder [2008], Gibbard [2003]).
This view quickly lands expressivists in the midst of needing to answer very general
questions in the philosophy of language about what the relationship is between ‘grass is
green’ and the belief that grass is green. To answer these questions, expressivists need
to take stands about the nature and norms on assertion and on the primacy of mental
versus linguistic content. It also leads expressivists to a very different conception of the
primary goals of semantic inquiry—whereas conventional theories take the goal of a
semantic theory to be to assign truth values or at least truth conditions to sentences,
expressivists are led to the idea that a semantic theory needs to work by associating each
sentence, ‘P’, with what it is to think that P. This novel expressivist approach to the
objectives and structure of semantic theorizing incurs deep commitments for expressiv-
ists, and the projects of justifying it and of carrying it out go to the heart of the semantic
enterprise.
One potential source of solace for metaethical expressivists is the potential for
applications of a similar semantic framework outside of metaethics. For example, con-
ditionals, probability judgments, epistemic modals, and truth have all been motivated
as calling for a distinctively expressivist treatment (see Adams [1975], Gibbard [1981],
Price [1983], Yalcin [2011], Schroeder [2010]). On such a view, conditionals express,

709
 M A RK SCH R O E D E R

rather than report, the conditional confidence of the speaker in the consequent, con-
ditional on the antecedent; probability judgments and epistemic modals express, rather
than report, the speaker’s credences; and truth and falsity ascriptions express, rather
than report, the speaker’s agreement or disagreement with the object of the ascription.
The better motivated such further applications for expressivism are, the more optimism
it makes sense to have that something like the expressivist’s approach to semantic the-
ory can be carried out—so proponents and opponents of metaethical expressivism alike
have a stake in all of these topics.

5 Noncognitivist Theories: Illocutionary Theories

Whereas expressivism seeks to account for the meaning of normative sentences by
accounting for the nature of normative thought, an earlier generation of noncognitivist
theories sought to account for the meaning of normative sentences by accounting for
the nature of normative assertion. These broadly illocutionary theories seek to explain
the meaning of a sentence, ‘P’, by saying what ‘P’ is used to do. Like expressivism, these
theories depart from more conventional semantic approaches by declining to tell us
what normative sentences are about, or what sort of state of the world would make them
true—what is significant about the meaning of normative sentences, on such views, is
that they are used to commend, to cajole, to discourage, or to incite—for example (see
particularly Ayer [1936], Stevenson [1937], [1944], and Hare [1952]).
Illocutionary theories of normative language are obviously deeply involved in need-
ing to resolve a wide range of issues from speech act theory; somewhat less obvious but
equally important is the interest for speech act theories of an adequate understanding
of the semantics of mood. Sentential mood is, after all, the conventionally mediated
aspect of linguistic use that is most transparently connected with illocutionary force—
though declaratives can be used to ask questions and interrogatives can be used to issue
commands and imperatives can be used to make assertions, declarative mood is con-
ventionally associated with assertion, interrogative mood is conventionally associated
with questioning, and imperative mood is conventionally associated with commands
or requests. Illocutionarily based metaethical theories take these facts as grounds for
optimism that a semantic theory must ultimately be able to do the kinds of things that
they ask of it (see Hare [1952], Smart [1984]).
Unfortunately, it isn’t immediately obvious that a general theory of mood must be
able to do the things that metaethical illocutionary theories require of it—for example,
on many conventional approaches to mood, it is a feature only of whole sentences, and
hence doesn’t embed. So, although there are complex imperative sentences like ‘don’t
steal’ and ‘steal everything you can’, these are not cases in which imperatives embed in
more complex constructions, but in which the imperative mood takes scope over a com-
plex sentence—in other words, ‘don’t steal’ has the structure, !!(~(you steal)), rather
than the structure, ~(!!(you steal)). But importantly, normative predicates do embed in
complex constructions. So if mood doesn’t embed, mood is not a very useful comparison
for illocutionary metaethical theorists.
However, there is some evidence that mood can embed in complex constructions.
For example, there is some evidence that conditional interrogatives like ‘if it’s raining,
then why am I not wet?’ are not interrogatives about conditionals, and that conditional
imperatives like ‘if you’re going to the store, then bring me some milk’ do not issue
commands to do conditional things (compare Barker [2004]). Moreover, though it is

710
 PH I LO SO PH Y O F LA N G U AG E F O R ME TA E T H I C S

difficult to form conjunctions of sentences of different moods using ‘and’, sentences like
‘close the window, for it’s freezing in here’ and ‘it’s cold in here, but leave the window
open’ make perfect sense and do seem to combine conjuncts of imperative and declara-
tive moods. Advocates of illocutionary metaethical theories are wise to be particularly
interested in such cases, and in whether they put pressure on an adequate semantic
theory of mood to allow for ways in which conventional markers of illocutionary force
like mood can in fact compose into logically more complex constructions—as norma-
tive predicates need to be able to do (compare especially Boisvert and Ludwig [2006]).

6 General Issues Facing Noncognitivist Theories

In addition to the issues in the philosophy of language of special interest to expressivists,
and those of special interest to illocutionary theories, both strategies for implementing
the general noncognitivist program in metaethics face a shared set of difficult issues in
the philosophy of language, centered around the famous Frege–Geach Problem.
The core of the Frege–Geach Problem is that of how to account for compositional-
ity. The central problem is that since these theories are nonextensional, they can’t rely
on the feature of broadly extensional theories of meaning, that negation corresponds
to complementation, conjunction to intersection, and disjunction to union. This is
because lacking the state of mind expressed by ‘P’ is not the same as having the state of
mind expressed by ‘~P’ and having both the states of mind expressed by ‘P’ and by ‘Q’ is
not the same as having the state of mind expressed by ‘P&Q’. Similarly (for illocution-
ary theories), performing the illocutionary act associated with ‘~P’ is clearly different
from simply not performing the illocutionary act associated with ‘P’, and performing
the illocutionary act associated with ‘P or Q’ is clearly different from either performing
the illocutionary act associated with ‘P’ or performing that associated with ‘Q’. So if a
noncognitivist theory of meaning is to explain what it is to think that ~P on the basis of
what it is to think that P, or to explain what illocutionary act ‘~P’ is suited to perform
on the basis of what illocutionary act ‘P’ is suited to perform, the task is far from straight-
forward (see especially Unwin [1999], [2001] and Schroeder [2008]; see Geach [1965]
for the version of this problem which gave it its name).
Moreover, all noncognitivist metaethical theories need to confront very general
issues in the philosophy of logic. Whatever meaning is associated with ‘not’, it had
better suffice to explain why for any sentence ‘P’, ‘P’ and ‘it is not the case that P’ are
inconsistent. But the conventional way of understanding inconsistency is just truth-
conditional inconsistency. Since noncognitivist theories of meaning do not work by
generating truth values or truth-conditions, they have no direct way of guaranteeing
that ‘P’ and ‘it is not the case that P’ cannot both be true. Consequently, they must
either take a very indirect approach of trying to earn their way to this very simple result,
or they must construct an alternative entire theory of inconsistency—and relatedly, of
validity and logical consequence (see particularly Blackburn [1984], Gibbard [1990],
[2003], van Roojen [1996], and Schroeder [2008]).
Both varieties of noncognitivist theory are also deeply committed to understanding
the nature of truth. According to early proponents of noncognitivist theories, norma-
tive sentences cannot be true or false, but contemporary proponents have sought to
explain how they can be true or false, by appeal to broadly deflationary accounts of
truth (see especially Blackburn [1993]). This is an important move, not only because
we do in ordinary discourse attribute truth or falsity to normative claims but in order to

711
 M A RK SCH R O E D E R

make good on the logical relationships between normative sentences on anything like
ordinary views about what logical relationships involve—even by a very indirect route.
Noncognitivists are also committed to solving problems about truth-aptness that do not
face other theories; for example, a noncognitivist who takes expressives like ‘dammit,
not stealing!’ as a semantic model for the meaning of ‘stealing is wrong’ needs to explain
why the latter is truth-apt even though the former is not (compare Dreier [1996]).
Noncognitivists also face special challenges in accounting for the semantics of attitude
ascriptions and indirect discourse reports—sentences like ‘Max thinks that stealing is
wrong’ and ‘Max said that stealing is wrong’. Indeed, given the special constraints under
which noncognitivist semantic theories operate, there are pressing questions, even if
they can account for the simple compositionality of the ‘truth-conditional’ connectives,
of whether they will be able to account for more general compositional issues—includ-
ing fitting in with generalized quantifiers, generics, modals, tense, and mood. Since
noncognitivist theories really propose a quite different way of thinking about the goals
of a semantic theory, their ultimate debt is to be able to carry such a semantic program
out across the board, which means that they essentially have commitments for under-
standing the semantics of every part of natural languages (see particularly Schroeder
[2008]). Commitments for noncognitivism in the philosophy of language run deep.

7 Hybrid Theories
Some metaethical theories hold that it is not sufficient to understand the meaning of
normative sentences to focus either on what such sentences are about or on what such
sentences are used to do—for the meaning of normative sentences involves both a
‘descriptive’ content and an expressive or illocutionary component. Whereas expressiv-
ists focus on the nature of assertion in general or on expressives and illocutionarily ori-
ented noncognitivists focus on imperatives and performatives, hybrid theorists typically
take as their examples racial slur terms, and sometimes other ‘thick’ evaluatives. Racial
slurs seem to many to be transparently associated both with a particular descriptive
content (that of belonging to a certain race) and to be conventionally associated with
expressing contempt or denigrating (no coincidence of etymology, here). It is the posi-
tion of hybrid views that normative words are at least this much like racial slurs. (See
particularly Copp [2001], Barker [2000], Boisvert [2008], and Ridge [2006], [2007].)
In fact, however, theories about the semantics of racial slurs vary widely, and so
hybrid theorists are well advised to pay attention not only to slurs but also to pejora-
tives and epithets like ‘jerk’, ‘asshole’, and ‘bastard’, each of which also seems to many
to be associated with some descriptive content as well as a derogatory function, and to
paradigmatic ‘thick’ evaluatives like ‘courageous’ and ‘kind’. Another good model for
hybrid theorists to be interested in is the phenomenon of intonational sneer—of the
kind associated with the claim not simply that someone is fat, but that he is ‘FA-at’
(compare Blackburn [1992]). Though intonation is a feature of language that it is easy
to forget about when we spend most of our time as theorists comparing representa-
tions of sentences on the written page, it is very plausible that intonational sneer is
conventionally determined—it works even with unnatural words, for example, in the
claim that someone is ‘SMA-art’, in the same tone of voice. The diversity of features
involved with racial slurs, pejoratives and other epithets, ‘thick’ evaluative terms, and
intonational sneer gives hybrid theorists a variety of models to work with in seeking
to understand normative language, but each of these topics needs to be much better

712
 PH I LO SO PH Y O F LA N G U AG E F O R ME TA E T H I C S

understood in its own right, before any hybrid theory is properly up and running (see
especially Hay [2011]).
One of the most important things hybrid theorists need to understand about
normative language, as well as about each of these examples from other parts of lan-
guage, is how the illocutionary or expressive component of these terms’ meaning
projects through environments like negation, the antecedents of conditionals, attitude
ascriptions, and indirect-discourse reports. A variety of the payoffs for hybrid theories
turn out to turn on the hypothesis that the expressive component of normative language
projects through negation and the antecedents of conditionals, but not through belief
ascriptions—but it’s far from clear that racial slurs work in this way, so much remains for
hybrid theorist to sort out (see Schroeder [2009], Potts [2003], Hay [2011]).

8 General Issues for Metaethics from the Philosophy of Language

In addition to all of the specific issues catalogued above facing metaethicists in virtue
of the specific metaethical theories which they adopt, there are general issues reated
to language of broad interest to all views in metaethics. For example, any theory about
‘good’ or the nature of the property it expresses—whether realist or expressivist—needs
to start from a proper understanding of the argument structure of ‘good’—whether it
expresses a monadic property of states of affairs, for example, or a relation between states
of affairs and agents whom those states of affairs are good for, or a function from prop-
erties to properties, or something else altogether (compare Ziff [1960], Geach [1956],
Thomson [2003], Kraut [2007]). Similarly, any theory about ‘ought’—whether realist or
expressivist—needs to start from a proper understanding of whether it is a context-inde-
pendent propositional operator, a context-dependent propositional operator (and if so
what arguments are provided by context), or whether it expresses a relation between
agents and actions (compare Wedgwood [2007], Schroeder [2011], Cariani [2009]).
And arguably substantive issues in both normative and metaethics can also turn on the
argument structure of claims about reasons (see Schroeder [2007a]). Answering all of
these questions requires addressing general issues from both linguistics and the philoso-
phy of language.
One particular issue of general interest to metaethicists is the relationship between
epistemic and deontic modals. Early treatments of deontic logic were inspired by analo-
gies with epistemic and alethic modalities, and both cross-linguistic data and the fairly
pervasive availability of both epistemic and deontic readings for the very same modals
(‘must’, ‘have to’, ‘may’, ‘ought’, ‘should’) encourages the hypothesis that these terms
make a uniform semantic contribution to both epistemic and deontic readings (see
von Wright [1951], Kratzer [1977], [1981]). Depending on how this uniform semantic
contribution is made out, it potentially constitutes a considerable challenge to meta-
ethicists who think that normative or deontic language is special (see especially Finlay
[2009]).
As I’ve been indicating in this chapter, the issues connected to metaethics from the
philosophy of language are extremely diverse; I’ve had almost nothing to say about any
of them here, only an extreme bird’s-eye glance at the sweep of the territory. But it’s
safe to say that there is much work for philosophers of language to do in metaethics, and
much work for metaethicists to do, in the philosophy of language.1

713
 M A RK SCH R O E D E R

Note
1 Special thanks to Gillian Russell and Delia Graff Fara.

References
Acton, H.B. [1936]. ‘The Expletive Theory of Morals.’ Analysis 4: 42–45.
Adams, Ernest [1975]. The Logic of Conditionals: An Application of Probability to Deductive Logic. Dordrecht:
Reidel.
Ayer, A.J. [1936]. Language, Truth, and Logic. New York: Dover.
Barker, Stephen [2000]. ‘Is Value Content a Component of Conventional Implicature?’ Analysis 60(3):
268–279.
—— [2004]. Renewing Meaning. Oxford: Oxford University Press.
Blackburn, Simon [1984]. Spreading the Word. Oxford: Oxford University Press.
—— [1992]. ‘Morality and Thick Concepts.’ Proceedings of the Aristotelian Society suppl. vol. 66: 285–299.
—— [1993]. Essays in Quasi-Realism. Oxford: Oxford University Press.
—— [1998]. Ruling Passions. Oxford: Oxford University Press.
Boisvert, Daniel [2008]. ‘Expressive-Assertivism.’ Pacific Philosophical Quarterly 89: 169–203.
Boisvert, Daniel, and Kirk Ludwig [2006]. ‘Semantics for Nondeclaratives.’ In B. Smith and E. Lepore, eds.,
The Oxford Handbook of the Philosophy of Language. Oxford: Oxford University Press.
Boyd, Richard [1988]. ‘How to be a Moral Realist.’ Printed in Sayre-McCord [1988], 181–228.
Brink, David [1989]. Moral Realism and the Foundations of Ethics. New York: Cambridge University Press.
Cariani, Fabrizio [2009]. ‘The Semantics of ‘Ought’ and the Unity of Moral Discourse.’ Ph.D. Dissertation,
University of California, Berkeley.
Carnap, Rudolf [1935]. Philosophy and Logical Syntax. Bristol, UK: Thoemmes Press.
Copp, David [2000]. ‘Milk, Honey, and the Good Life on Twin Earth.’ Synthese 124(1): 113–137.
—— [2001]. ‘Realist-Expressivism: A Neglected Option for Moral Realism.’ Social Philosophy and Policy 18:
1–43.
Dreier, James [1990]. ‘Internalism and Speaker Relativism.’ Ethics 101(1): 6–25.
—— [1993]. ‘The Structure of Normative Theories.’ The Monist 76(1): 22–40.
—— [1996]. ‘Expressivist Embeddings and Minimalist Truth.’ Philosophical Studies 83(1): 29–51.
—— [2006]. ‘Negation for Expressivists: A Collection of Problems with a Suggestion for their Solution.’
Oxford Studies in Metaethics, volume 1. Oxford: Oxford University Press, 217–233.
Finlay, Stephen [2004]. ‘The Conversational Practicality of Value Judgment.’ Journal of Ethics 8: 205–223.
—— [2009]. ‘Oughts and Ends.’ Philosophical Studies 143: 315–340.
Geach, Peter [1956]. ‘Good and Evil.’ Analysis 17: 33–42.
—— [1965]. ‘Assertion.’ Philosophical Review 74: 449–465.
Gibbard Allan [1981]. ‘Two Recent Theories of Conditionals.’ In William Harper, Robert Stalnaker, and
Glenn Pearce, eds., Ifs. Dordrecht: Reidel.
—— [1990]. Wise Choices, Apt Feelings. Cambridge, MA: Harvard University Press.
—— [2003]. Thinking How to Live. Cambridge, MA: Harvard University Press.
Hägerström, Axel [1953]. Inquiries into the Nature of Law and Morals. Edited by Karl Olivecrona and trans-
lated by C.D. Broad. Stockholm: Almqvist and Wiksell.
Hare, R.M. [1952]. The Language of Morals. Oxford: Oxford University Press.
—— [1970]. ‘Meaning and Speech Acts.’ The Philosophical Review 79(1): 3–24.
Harman, Gilbert [1975]. ‘Moral Relativism Defended.’ Reprinted in Harman [2000], 3–19.
—— [1978]. ‘Relativistic Ethics: Morality as Politics.’ Reprinted in Harman [2000], 39–57.
—— [2000]. Explaining Value and Other Essays in Moral Philosophy. Oxford: Oxford University Press.
Hay, Ryan [2011]. ‘Hybrid Expressivism and the Analogy Between Pejoratives and Moral Language.’ Euro-
pean Journal of Philosophy, published online in Early View, February 2011
Horgan, Terence, and Mark Timmons [1991]. ‘New Wave Realism Meets Moral Twin Earth.’ Journal of
Philosophical Research 16: 447–465
—— [1992]. ‘Trouble for New-Wave Moral Semantics: The Open Question Argument Revived.’ Philosophi-
cal Papers 21(3): 153–175.
—— [2000]. ‘Copping Out on Moral Twin Earth.’ Synthese 124(1): 139–152.
Jackson, Frank [1998]. From Metaphysics to Ethics. Oxford: Oxford University Press.
Kratzer, Angelika [1977]. ‘What Must and Can Must and Can Mean.’ Linguistics and Philosophy 1: 337–355.

714
 PH I LO SO PH Y O F LA N G U AG E F O R ME TA E T H I C S

—— [1981]. ‘The Notional Category of Modality.’ In H.J. Eikmeyer and H. Rieser, eds., Words, Worlds, and
Contexts. New York: De Gruyter.
Kraut, Richard [2007]. What is Good and Why: The Ethics of Well-Being. Cambridge, MA: Harvard University
Press.
MacFarlane, John [2005]. ‘Making Sense of Relative Truth.’ Proceedings of the Aristotelian Society 105:
321–339.
—— [2007]. ‘Relativism and Disagreement.’ Philosophical Studies 132: 17–31.
Moore, G.E. [1903]. Principia Ethica. Cambridge: Cambridge University Press.
Ogden, C.K., and I.A. Richards [1923]. The Meaning of Meaning. New York: Harcourt Brace.
Perry, R.B. [1926]. General Theory of Value: Its Meaning and Basic Principles Construed in Terms of Interest.
Cambridge, MA: Harvard University Press.
Potts, Christopher [2003]. The Logic of Conventional Implicature. Oxford: Oxford University Press.
Price, Huw [1983]. ‘Does ‘Probably’ Modify Sense?’ Australasian Journal of Philosophy 61: 396–408.
Railton, Peter [1986]. ‘Moral Realism.’ Philosophical Review 95: 163–207.
Ridge, Michael [2006]. ‘Ecumenical Expressivism: Finessing Frege.’ Ethics 116(2): 302–336.
—— [2007]. ‘Ecumenical Expressivism: The Best of Both Worlds.’ Oxford Studies in Metaethics 2: 51–76.
Ross, W.D. [1939]. Foundations of Ethics. Oxford: Clarendon Press, chapter 2.
Sayre-McCord, Geoffrey, ed. [1988]. Essays on Moral Realism. Ithaca: Cornell University Press.
Schroeder, Mark [2007a]. ‘Reasons and Agent-Neutrality.’ Philosophical Studies 135(2): 279–306.
—— [2007b]. Slaves of the Passions. Oxford: Oxford University Press.
—— [2008]. Being For: Evaluating the Semantic Program of Expressivism. Oxford: Oxford University Press.
—— [2009]. ‘Hybrid Expressivism: Virtues and Vices.’ Ethics 119(2): 257–309.
—— [2010]. ‘How to be an Expressivist About Truth.’ In Cory Wright and Nicholas Jang Peterson, eds.,
New Waves in Truth. New York: Palgrave Macmillan, 282–298.
—— [2011]. ‘Ought, Agents, and Actions.’ Philosophical Review 120(1): 1–41.
Searle, John [1962]. ‘Meaning and Speech Acts.’ Philosophical Review 71: 423–432.
Smart, J.J.C. [1984]. Ethics, Persuasion, and Truth. Oxford: Oxford University Press.
Smith, Michael [1994]. The Moral Problem. Oxford: Basil Blackwell.
—— [2003]. ‘Neutral and Relative Value After Moore.’ Ethics 113(3): 576–598.
Stevenson, C.L. [1937]. ‘The Emotive Meaning of Ethical Terms.’ Reprinted in Stevenson [1963], 10–31.
—— [1944]. Ethics and Language. Oxford: Oxford University Press.
Sturgeon, Nicholas [1985]. ‘Moral Explanations.’ In Copp and Zimmerman, eds., Morality, Reason, and
Truth. Totowa: Rowman and Allanheld.
—— [2006]. ‘Moral Explanations Defended.’ In James Dreier, ed., Contemporary Debates in Moral Theory.
Malden, MA: Blackwell.
Thomson, Judith Jarvis [1996]. ‘Moral Objectivity.’ In Gilbert Harman and Judith Jarvis Thomson, Moral
Relativism and Moral Objectivity. Malden, MA: Blackwell.
—— [2003]. Goodness and Advice. Princeton: Princeton University Press.
Timmons, Mark [1999]. Morality Without Foundations. Oxford: Oxford University Press.
Unwin, Nicholas [1999]. ‘Quasi-Realism, Negation and the Frege-Geach Problem.’ The Philosophical Quar-
terly 49(196): 337–352.
—— [2001]. ‘Norms and Negation: A Problem for Gibbard’s Logic.’ The Philosophical Quarterly 51(202):
60–75.
Urmson, J.O. [1968]. The Emotive Theory of Ethics. New York: Oxford University Press.
van Roojen, Mark [1996]. ‘Expressivism and Irrationality.’ The Philosophical Review 105(3): 311–335.
—— [2006]. “Knowing Enough to Disagree: A New Response to the Moral Twin Earth Argument.” In R.
Shafer-Landau, ed., Oxford Studies in Metaethics, volume 1. Oxford: Oxford University Press, 161–193.
von Wright, G.H. [1951]. ‘Deontic Logic.’ Mind 60: 1–15.
Wedgwood, Ralph [2007]. The Nature of Normativity. Oxford: Oxford University Press.
Wong, David [1984]. Moral Relativity. Berkeley: University of California Press.
Yalcin, Seth [2011]. ‘Nonfactualism about Epistemic Modality.’ in Andy Egan and Brian Weatherson, eds.,
Epistemic Modality. Oxford: Oxford University Press.
Ziff, Paul [1960]. Semantic Analysis. Ithaca: Cornell University Press.

715
 6.3
P HI L OS O PHY OF
L ANGUA GE F OR
ME TAP H YSIC S
6.3.1
The Language of Causation
Eric Swanson

What parts of natural language count as parts of the language of causation? Here are four
ascriptions of singular causation:

(1) The fall caused the vase to break.

(2) The fall was the cause of the vase’s breaking.
(3) The fall was a cause of the vase’s breaking.
(4) The fall was causally relevant to the vase’s breaking.

Ascriptions of singular causation are used to make claims about particular instances of
causation between particular causal relata. Section 1 explores some features of claims
like (1)–(4).
There are also ascriptions of general causation, as in (5):

(5) Smoking causes lung cancer.

Sentences like (5) are used to represent a causal connection between smoking ‘in gen-
eral’ and lung cancer ‘in general,’ as opposed to (6), which would be used to represent a
causal connection between Al’s smoking and his lung cancer.

(6) Al’s smoking caused his lung cancer.

Section 2 discusses ascriptions of general causation.

Then there are expressions that might have some sort of covert causal component to
their meaning:
 TH E LA N G U A G E OF C AU S AT I O N

(7) Al boiled the water.

(8) Betty’s insulting Carl disgusted Dawn.

The nature of any such ‘causal component’ is not entirely clear. Some have gone so far
as to say that (7) just means that Al caused the water to boil, and that (8) just means that
Betty caused Carl to feel insulted and that Betty’s insulting Carl caused Dawn’s disgust.
These claims are likely too strong, as we will see in section 3, but this does not show
that there is no causal component to claims like (7) and (8). There also may be a covert
causal component to sentences like

(9) Binge drinking kills brain cells.

The similarities between (5) and (9) suggest that (9) resembles ascriptions of general
causation more closely than it resembles ascriptions of singular causation.
Finally there are passages the proper interpretation of which seems to require appro-
priate beliefs about what causes what. Knowing that the city council’s fear of violence
would likely cause them to refuse the demonstrators a permit helps an addressee inter-
pret the ‘they’ in (10) as referring to the city council.

(10) The city councilmen refused the demonstrators a permit because they
feared violence.

Knowing that advocating revolution would likely cause one’s application for a per-
mit to be refused helps an addressee interpret the ‘they’ in (11) as referring to the
demonstrators.

(11) The city councilmen refused the demonstrators a permit because they
advocated revolution. (Winograd 1972: 295)

Section 4 is devoted to the relationship between causation and the structure of

discourse.

1 The Language of Singular Causation

It is controversial exactly what demands a successful theory of causation must meet.
This is at least in part because it is not always clear to what extent a particular judg-
ment about a causal claim should be explained in terms of facts about the language of
causation and to what extent it should be explained in terms of facts about causation.
Following Anscombe (1971a), Nancy Cartwright has even argued that

There are a variety of different kinds of relations picked out by the abstract
term ‘causes’ and a variety of different—correct—uses of the term for a variety
of different purposes, with little of substantive content in common.
(Cartwright 2006a: 56–57; see also Cartwright 1999a
and Cartwright 2004a, and Machamer 2000a)

717
 ERI C SWA N S O N

At this point no theory of causation has a claim to being the ‘standard theory.’ But it
will be helpful to have a clear, influential analysis in mind as we discuss the language of
singular causation, and David Lewis’s theory in his paper (1973a) “Causation’’ is a good
starting point.
There Lewis analyzes singular causation in terms of patterns of subjunctive
conditionals, sometimes called counterfactuals (see Subjunctive Conditionals,
Chapter 3.13, this volume). To a first approximation, he holds that e depends causally
on c if and only if (12) and (13) are both true. (For important refinements, see Lewis:
1986e.)

(12) If c had occurred, e would have occurred.

(13) If c hadn’t occurred, e wouldn’t have occurred.

On Lewis’s analysis, causation itself is the ancestral of this relation of causal depend-
ence. Put differently: suppose that there is a Lewisian causal chain <e1, e2, e3, . . ., en> such
that for every natural number m that is less than n, em+1 causally depends on em. Then for
all m < n, em is a cause of en.
Lewis’s analysis of causation, like many others, predicts that c was a cause of e in some
very counterintuitive cases. For example, Caesar’s death ‘depends causally’ on his birth.
But in most if not all contexts, assertions of (14) and (15) sound very odd:

(14) Caesar’s birth was the cause of his death.

(15) Caesar’s birth caused his death.

This oddness doesn’t worry Lewis. He acknowledges that we “sometimes single out one
among all the causes of some event and call it ‘the’ cause, as if there were no others’’
(Lewis 1973a). This is not surprising: There are many independent reasons for thinking
that definite descriptions in general—and so ‘the cause’ in particular—are heavily sen-
sitive to conversational context (see Descriptions and Context-Sensitivity, Chapters,
3.7 and 1.10, this volume). Lewisians in particular should expect context sensitivity
since counterfactuals themselves are heavily context sensitive (see again Subjunctive
Conditionals, Chapter 3.13, this volume). And although they convey different presup-
positions—(14) presupposes that Caesar’s death had a cause, and (15) does not—it
doesn’t seem implausible that (14) and (15) have roughly the same truth conditions
(see Presupposition and Implicature, Chapters 1.4 and 1.5, this volume).
Lewis elsewhere discusses the putative context sensitivity of ‘knows’ and ‘knowledge’
in detail (see Lewis 1996a and Philosophy of Language for Epistemology, Chapter 6.1,
this volume; see also Lewis 1979c). In marked contrast he prescinds from the context
sensitivity in ‘the cause’ and ‘to cause’:

I have nothing to say about these principles of invidious discrimination. I am

concerned with the prior question of what it is to be one of the causes (unselec-
tively speaking). My analysis is meant to capture a broad and nondiscrimina-
tory concept of causation.
(558–559)

This position—with the assumption that one can dispel any misleading features of
causal language by talking about ‘a cause’ or ‘one of the causes’—is very common (see,

718
 TH E LA N G U A G E OF C AU S AT I O N

e.g., Hall 2004a: 228). It is reminiscent of John Stuart Mill’s position on causes and
conditions:

Since . . . mankind are accustomed, with acknowledged propriety so far as the

ordinances of language are concerned, to give the name of cause to almost any
one of the conditions of a phenomenon, or any portion of the whole number,
arbitrarily selected . . . it will probably be admitted without longer discussion,
that no one of the conditions has more claim to that title than another, and
that the real cause of the phenomenon is the assemblage of all its conditions.
Mill 1843a: 403

Lewis seems to assume that we can easily talk about all the causes of a phenomenon, and
Mill seems to make the slightly stronger assumption that we can easily talk about all the
causes of a phenomenon at once: That’s just what it is to talk about “the real cause of the
phenomenon’’ in the way that Mill does.
In Swanson (2010c) I argue that these assumptions are wrong. Whether something
can felicitously be cited as a cause of some effect can depend, in part, on what else has
already been cited as a cause of that effect, and on what we might in the future want
to cite as a cause of that effect. A Lewisian might apply my theory roughly as follows:
Citing one representative of a given Lewisian causal chain as a cause of a certain effect
generally crowds out other potential representatives of that causal chain to the effect, so
that they can no longer felicitously be cited as a cause of that effect. Speakers are thus
under pressure to choose a representative of a given causal chain that is likely to serve
evolving conversational needs well.
For an example, consider Hartry Field’s bomb case. Billy plants a bomb in a room.
Suzy comes into the room, notices the bomb, and flees. Suzy later has a checkup and
is found to be in perfect health. According to Field, Stephen Yablo, and many others,
“the bomb is not a cause’’ of Suzy’s health, although the presence of the bomb caused
Suzy to flee, and Suzy’s fleeing is a cause of Suzy’s perfect health the next day (Yablo
2004a: 119). From this it follows that the causation relation is nontransitive. For if e1 is
Billy’s planting the bomb, e2 is Suzy’s fleeing, and e3 is Suzy’s good health the next day,
then unless the causation is nontransitive, it follows from the fact that e1 is a cause of
e2 and the fact that e3 is a cause of e1 that e1 is a cause of e3. A genuine counterexample
to the transitivity of causation would be a counterexample to Lewis’s analysis among
many others. But notice that Billy’s planting the bomb makes Suzy’s good health the
next day quite surprising. Ordinarily, we would think that Suzy’s good health should be
credited to her fleeing, not to Billy’s planting the bomb. These reasons and others make
Billy’s planting the bomb a relatively poor representative of the causal paths through
it to Suzy’s good health the next day. So the Lewisian who appeals to my theory to
explain why some causal claims are infelicitous in certain contexts has a metaphysically
neutral explanation of why it would be unusual to count Billy’s planting the bomb as
a cause of Suzy’s good health. (And in fact it is possible to count the planting of the
bomb as part of a causal chain leading to Suzy’s good health, as long as the speaker
makes it clear that the planting of the bomb does not crowd out other events on that
causal chain.) So with appropriate attention to the context sensitivity that Lewis and
Mill try to ignore, we can actually defuse an important objection to many theories of
causation. Metaphysicians who take their theories to be constrained by our judgments
about causal locutions should thus pay close attention to the respects in which causal

719
 ERI C SWA N S O N

locutions are context sensitive. And this is true whether the context sensitivity of causal
locutions manifests itself semantically or pragmatically (see Semantics and Pragmatics
Chapter 1.2 this volume).
If any of the context sensitivity of singular causal talk does manifest itself in the
semantics, there may be other important upshots for the metaphysics of causation. Note
that Lewis assumes that singular causation is a binary relation, simply relating exactly
one cause to exactly one effect. The burgeoning literature on the role of contrasts in
the language of singular causation puts pressure on this popular assumption. For exam-
ple, Jonathan Schaffer argues that the context sensitivity involved in causal contrasts
is semantic, and as a result that “causation is a quaternary, contrastive relation: c rather
than C* causes e rather than E*, where C* and E* are nonempty sets of contrast events’’
(Schaffer 2005a: 297). He takes this hypothesis to help capture at least some of the
context sensitivity of claims about singular causation. For closely related discussion,
see van Fraassen 1980b; Holland 1986a; Glymour 1986a; Hilton 1986a, 1990a, and
1995a, and the sources cited therein; Hitchcock 1996a; Woodward 2003b; Maslen
2004a; Northcott 2008a; and Schaffer 2010b. For other work on the context sensitivity
of the language of singular causation see Collingwood 1937a; Anderson 1938a; Gask-
ing 1955a; Hart 1985a: 35–37 (originally published in 1959); Gorovitz 1965a; White
1965a; Zwart 1967a; Wright 1971a; Mackie 1980b: 34–38, 119–120 (originally pub-
lished in 1974); Unger 1977a and 1984a: 38, 58–60, 107; Lewis 1986c: 216; Horgan
1989a; Menzies 2004a, 2007a, and 2009a; Hall 2007a; Hitchcock 2007a and 2009a; and
Maslen 2009a.

2 The Language of General Causation

Recall that mere correlation between smoking and lung cancer does not suffice for the
truth of (5):

(5) Smoking causes lung cancer.

Neither is a causal connection between any one person’s smoking and their lung
cancer sufficient for (5) to be true. And neither is any particular instance of singular
causation necessary for its truth. For example, the truth of (16) is compatible with the
truth of (5).

(16) Al’s smoking didn’t cause him to have lung cancer.

On the basis of these differences between ‘singular’ and ‘general’ (or ‘token’ and
‘type’) causation, some philosophers distinguish between them on a metaphysical
level. For example, some hold that “a distinguishing mark of a general causal sentence
is that its causal relata are properties’’ (Menzies 1989a: 59). (See also Good 1961a,
1961b; Cartwright 1979a; Sober 1984a; and Eells 1991a.) Others think that the dif-
ferences between attributions of singular and general causation do not mark any deep
metaphysical differences. For example, when he discusses examples like these Lewis
writes:

Presumably those are quantified statements involving causation among par-

ticular events (or non-events), but it turns out not to be easy to match up the

720
 TH E LA N G U A G E OF C AU S AT I O N

causal generalizations of natural language with the available quantified forms.

A sentence of the form “c-events cause e-events,’’ for instance, can mean
any of

(a) For some c in c and some e in e, c causes e.

(b) For every e in e, there is some c in c such that c causes e.
(c) For every c in c, there is some e in e such that c causes e

not to mention further ambiguities. . . . These problems are not about causa-
tion, but about our idioms of quantification.
Lewis 1973a: 558

But (as Lewis recognizes) none of (a)–(c) provide very good glosses of sentences
like (5).
John Carroll offers a more promising hypothesis: Ascriptions of general causation
are really generic ascriptions of singular causation (Carroll 1988a and 1991a). To get a
sense of what this hypothesis means, consider two uncontroversial examples of generic
sentences (for an extensive discussion of genericity, see Generics, Chapter 3.5, this
volume):

(17) Lions have four legs.

(18) The dodo is extinct.

(17) is true even though some lions have fewer than four legs. And no particular lion’s
having four legs is necessary for its truth. Its exception-permitting character resembles
that of (5). (18) is true even though it would be a category mistake to ascribe the prop-
erty of being extinct to any particular dodo. (And besides, there are no longer any dodos
to whom that property might be ascribed.)
Examples of generic sentences in the literature often include bare plurals (as in (17))
or ‘kind-referring’ definite descriptions (as in (18)), and ascriptions of general causation
use these kinds of expressions less frequently. But consider

(19) Car accidents cause a wide range of personal injuries.

(20) The smoking of cigarettes causes nearly a half a million deaths each
year.

Like (17) and (18), (19) and (20) admit exceptions and do not require anything of any
particular car accident or smoking of cigarettes. And there are also uncontroversial
examples of gerundive and nominalized generics (Krifka 1995a: 102–105):

(21) Smoking tobacco is a bad habit.

(22) The smoking of tobacco is a bad habit.

(For further discussion of such sentences, see Carlson 1977a: 300–301; Conrad 1982a;
Chierchia 1982a, 1984a; and the other sources cited by Krifka et al.) (21) and (22)
obviously resemble the ascriptions of general causation (23) and (24):

721
 ERI C SWA N S O N

(23) Smoking tobacco causes lung cancer.

(24) The smoking of tobacco causes lung cancer.

These sentences resemble standard examples of generics in other ways, too. For
example, adding ‘usually’ or ‘typically’ to (23) and (24) produces “at most a slight change
of meaning’’ (Krifka 1995a: 9): (23), (24), and (25) mean roughly the same thing.

(25) Smoking tobacco typically causes lung cancer.

And “it is very difficult to transform a characterizing sentence into the progressive with-
out its losing its generic character’’ (Krifka 1995a: 12). For example, the meanings of
(26) and (27) are far from the most natural readings of (24) and (25):

(26) Smoking tobacco is causing lung cancer.

(27) The smoking of tobacco is causing lung cancer.

To be sure, none of these observations answer the historically important questions about
ascriptions of general causation and about generic sentences. But it seems likely that
sustained interaction would benefit those primarily interested in the former and those
primarily interested in the latter.

3 The Hidden Language of Causation

George Lakoff (1965a) influentially argues that sentences like (28) and (29) have the
same “deep structure’’:

(28) Floyd caused the glass to melt.

(29) Floyd melted the glass.

The truth of this hypothesis (or a hypothesis close to it) would be profoundly impor-
tant to philosophical thought about causation. Among other things it would open up a
whole new range of causal locutions against which to test our metaphysical theories of
causation. But in response to Lakoff and others Jerry Fodor argues that “one can cause
an event by doing something at a time which is distinct from the time of the event. But
if you melt something, then you melt it when it melts.’’ So although (30) is fine, (31) is
odd (Fodor 1970a: 432–433).

(30) Floyd caused the glass to melt on Sunday by heating it on Saturday.

(31) # Floyd melted the glass on Sunday by heating it on Saturday.

And D. A. Cruse observes that (32) is fine although (33) is odd (Cruse 1972a: 522)

(32) John caused the reflection to move.

(33) # John moved the reflection.

On Lakoff’s side see also McCawley 1968a and 1972a; and Lakoff 1972a; against Lakof-
fians see also Katz 1970a; Smith 1970a; Chomsky 1972a; Kac 1972a and 1972b; Shiba-
tani 1972a; Wierzbicka 1975a; and Morreall 1976a.

722
 TH E LA N G U A G E OF C AU S AT I O N

From out of this debate emerged views that distinguish between “direct’’ or “manip-
ulative’’ causation on the one hand, and “indirect’’ or “directive’’ causation on the
other (see, for example Nedyalkov 1969a and 1973a; Shibatani 1973, 1973a, 1973b,
and 1976a; McCawley 1976a; Talmy 1976a and 2000a; Comrie 1981a and 1985a; and
Wierzbicka 1988a). James McCawley, for example, argues that

. . . “lexical causatives’’ such as kill and transitive open refer to direct causation,
whereas corresponding periphrastic causatives such as cause to die and cause to
open are unspecified as to the kind of causation, since the [Gricean] coopera-
tive principles would dictate the use of the lexical causative where direct causa-
tion is involved.
(McCawley 1978a: 257)

(On Gricean cooperative principles see Presupposition and Implicature, Chapters 1.4
and 1.5 this volume.) Masayoshi Shibatani writes that this distinction is “Perhaps the
single most important semantic distinction linguists make in accounting for different
causative forms’’ (Shibatani 2000b: 11). And yet he observes that “Despite its great
importance, the relevant notion has not been satisfactorily defined, and grammarians
have been using the terms ‘direct causation’ and ‘indirect causation’ and related ones
rather vaguely without a rigorous definition’’ (11).
Shibatani’s observation notwithstanding, it is not clear that any distinction in the
area could do all the work that the ‘direct’/‘indirect’ distinction is put to. For example,
John Morreall observes that

Causing is not an action. Killing, on the other hand, is an action; for example,
it can be done quickly or slowly. . . . [So] if kill were derived from cause to die, we
would expect to see sentences like
(34) *John slowly caused Mary to die.
But the closest we can come to a cause to die sentence corresponding to [‘John
slowly killed Mary’] is [(35)].
(35) John caused Mary to die slowly.
But in [(35)] it is clear that it is Mary’s dying that is slow, and not John’s action
of killing. (Morreall 1976a: 516–517)

Responding to this argument by contending that ‘direct causation’ is an action, and

so can be done quickly or slowly, does not look very promising. But if periphrastic
causatives are “unspecified as to the kind of causation,’’ as McCawley suggests in the
quotation above, then there should be a reading of (34) on which ‘caused’ denotes
‘direct’ causation, and so a reading on which it means the same as ‘John slowly killed
Mary.’ There are various ways to respond to this argument. One might, for example,
reject the putative distinction between ‘direct’ and ‘indirect’ causation. Or one might
reject McCawley’s appeal to conversation implicature, or reject this particular deriva-
tion of ‘kill.’ Unfortunately I cannot assess the possible responses here.
The literature has focused on lexicalizations that seem related to ascriptions of sin-
gular causation, but lexicalizations that seem related to ascriptions of general causation

723
 ERI C SWA N S O N

are interesting as well. For example, a Lakoffian might have thought that (9) and (36)
have the same deep structure:

(9) Binge drinking kills brain cells.

(36) Binge drinking causes brain cells to die.

But paralleling Morreall’s observation, (37) is fine and (38) is odd.

(37) Binge drinking slowly kills brain cells.

(38) # Binge drinking slowly causes brain cells to die.

This is again hard for Lakoffians to explain, and again appealing to a distinction between
direct and indirect causation does not look very promising.
I do not hope to resolve these issues here, but just to make the point that philosophers’
skills would be useful in testing how robust the putative distinctions between direct and
indirect causation really are (for an opening salvo, see Thomson 1987a; but see also
1990a: 207). Philosophers’ skills would also be useful in drawing any important distinc-
tions finely enough so that they have real explanatory power. Whatever the results of
such efforts, a better understanding of how the subtleties of lexical causatives reflect and
influence our thoughts and intuitions about causal notions would likely benefit a wide
range of philosophers interested in causation and related topics. We might even get evi-
dence relevant to the nature of syntax and semantics (Pietroski 2005a) and to cognitive
science (see, for example, Wolff 2003a and 2007a; Song 2004a; and Sloman 2005a).

4 Causation and Discourse Interpretation

There is much to be learned from considering sentences in isolation. But sentences rarely
actually occur in isolation; a sentence is generally part of a discourse. And a discourse
is not a motley collection of unconnected sentences: it is more or less cohesive. A dis-
course “forms a unified whole . . . [that is] best regarded as a semantic unit: a unit not of
form but of meaning’’ (Halliday 1976a: 1–2). The unified nature of naturally occurring
discourses is an important aspect of linguistic meaning. And so for many Frege’s ‘context
principle’—“it is only in the context of a proposition that words have any meaning’’
(Frege 1980: 73, originally published in 1884)—does not go far enough. Discourses have
elements of meaning that do not adhere to any particular sentence (see Anaphora, Rel-
evance Theory, Dynamic Semantics Chapters 3.6, 1.14, and 2.7 this volume).
The extent to which a discourse strikes its interpreters as unified depends to some
extent on the causal connections between the parts of the discourse that interpreters
are able to perceive. Indeed, causal relations play an important enough role in discourse
interpretation that it is routine to categorize discourse relations in part in terms of their
relationship with causality (see, e.g., Halliday 1976a; Mann 1988a; Hobbs 1990a; Sand-
ers 1992a; Lascarides 1993a; Kehler 2002a; and Asher 2003a). Appeal to causal rela-
tions is essential for explaining how we interpret even quite prosaic discourses. Consider
the following simple example:

(39) Max opened the door. The room was pitch dark.
(40) Max switched off the light. The room was pitch dark.

724
 TH E LA N G U A G E OF C AU S AT I O N

Alex Lascarides and Nicholas Asher observe that (39) and (40) have similar syntax
but very different “natural interpretations’’: “The event and state in [39] temporally
overlap, whereas in ([40]) they do not’’ (Lascarides 1993a: 437). How should we explain
this difference between (39) and (40)? Lascarides and Asher postulate a discourse rela-
tion “Result (D, E): The event described in D caused the event or state described in E’’
(440) and hypothesize that we tacitly know that “the room being dark and switching off
the light, if connected, are normally such that the event causes the state’’ (466). The
‘normally’ is important here because the relevant belief is defeasible. This is to respect
the fact that if (40) is embedded in a larger discourse, we can find an interpretation on
which the switching off of the light and the darkness of the room do overlap:

(41) Max and Nell realized that the monkeys were all finally asleep, and
decided to turn off the light in the monkey pen to conserve power. Nell
stumbled around the control room trying to find the switch, but failed.
Max switched off the light. The room was pitch dark. But Max was used
to finding his way around the control room at all hours of the night.

The total theory that Lascarides and Asher develop is complicated enough that I will
not go into its details here. The important point is that our interpretation of discourse
is influenced by our conceptions of the causal relations in the world. The relationship
between causation and language is extensive enough that surprisingly broad ranges of
natural language are, in some sense, parts of the language of causation.

5 Conclusion
Philosophers interested in causation sometimes abstract away from the language of cau-
sation as though such language is likelier to mislead or to confuse than to edify, and as
though we can think about causation in a way that is untainted by facts about language.
To be sure, causal language does not wear its proper analysis on its sleeve, and there is
much more work to be done on causal language. Nevertheless, because it is unlikely that
we can cleanly excise causal thought from causal language, we should pay attention to
both if we are to pay attention to either (see also The Role of Experiment and The Role
of Intuitions, Chapters 4.1 and 4.6, this volume). Philosophers are well-suited to push
work on the language of causation forward. And at the same time, careful attention to
the language of causation could benefit a wide range of philosophical projects.

Note
Thanks to Sarah Moss for helpful discussion, and thanks to an anonymous referee for helpful comments on
an earlier version of this chapter.

References
Anderson, John. 1938. “The Problem of Causality.” Australasian Journal of Philosophy, vol.16 (2): 127–142.
Anscombe, G. E. M. 1971. Causality and Determination: An Inaugural Lecture. Cambridge University Press.,
London.
Asher, Nicholas and Alex Lascarides. 2003. Logics of Conversation. Cambridge University Press,
Cambridge.
Beebee, Helen, Christopher Hitchcock, and Peter Menzies, editors. 2009. The Oxford Handbook of Causa-
tion. Oxford University Press, Oxford.

725
 ERI C SWA N S O N

Carlson, Gregory N. 1977. Reference to Kinds in English. Ph.D. thesis, University of Massachusetts,
Amherst. Published in 1980 by Garland Press, New York.
Carroll, John W. 1988. “General Causation.” PSA: Proceedings of the Biennial Meeting of the Philosophy of Sci-
ence Association, vol. 1: Contributed Papers: 311–317.
—— 1991. “Property-Level Causation?” Philosophical Studies, vol. 63: 245–270.
Cartwright, Nancy. 1979. “Causal Laws and Effective Strategies.” Noûs, vol. 13: 419–437.
—— 1999. The Dappled World. Cambridge University Press., Cambridge.
—— 2004. “Causation: One Word, Many Things.” Philosophy of Science, vol. 71 (5): 805–819.
—— 2006. “Where Is the Theory in our ‘Theories’ of Causality?” Journal of Philosophy, vol. CIII (2): 55–66.
Chierchia, Gennaro. 1982. “Nominalization and Montague Grammar: A Semantics without Types for Nat-
ural Languages.” Linguistics and Philosophy, vol. 5: 303–354.
—— 1984. Topics in the Syntax and Semantics of Infinitives and Gerunds. Ph.D. thesis, University of Mas-
sachusetts, Amherst.
Chomsky, Noam. 1972. “Some Empirical Issues in the Theory of Transformational Grammar.” In Goals of
Linguistic Theory, P. S. Peters, editor, 63–130. Prentice-Hall, Englewood Cliffs.
Collingwood, R. G. 1937–1938. “On the So-Called Idea of Causation.” Proceedings and Addresses of the
American Philosophical Association, vol. 38: 85–112.
Collins, John, Ned Hall, and L. A. Paul, editors. 2004. Causation and Counterfactuals. MIT Press, Cam-
bridge., MA.
Comrie, Bernard. 1981. Language Universals and Linguistic Typology. The University of Chicago Press,
Chicago, second edn.
—— 1985. “Causative Verb Formation and other Verb-Deriving Morphology.” In Language Typology and
Syntactic Description, Volume III: Grammatical Categories and the Lexicon, Timothy Shopen, editor, 309–
348. Cambridge University Press, New York.
Conard, Bent. 1982. Referring and Non-referring Phrases: A Study in the Use of the Gerund and the Infinitive.
Publications in the Department of English, University of Copenhagen.
Cruse, D. A. 1972. “A Note on English Causatives.” Linguistic Inquiry, vol. 3 (4): 520–528.
Eells, Ellery. 1991. Probabilistic Causality. Cambridge University Press, Cambridge.
Fodor, Jerry A. 1970. “Three Reasons for Not Deriving ‘Kill’ from ‘Cause to Die’.” Linguistic Inquiry, vol. 1
(4): 429–438.
van Fraassen, Bas C. 1980. The Scientific Image. Oxford University Press, Oxford.
Frege, Gottlob. 1980. The Foundations of Arithmetic. Northwestern University Press, Evanston, IL, second
edn. Translated by J. L. Austin.
Gasking, Douglas. 1955. “Causation and Recipes.” Mind, vol. 64 (256): 479–487.
Glymour, Clark. 1986. “Statistics and Causal Inference: Comment: Statistics and Metaphysics.” Journal of
the American Statistical Association, vol. 81 (396): 964–966.
Good, I. J. 1961a. “A Causal Calculus (I).” British Journal for the Philosophy of Science, vol. 11 (44):
305–318.
—— 1961b. “A Causal Calculus (II).” British Journal for the Philosophy of Science, vol. 12 (45): 43–51.
Gorovitz, Samuel. 1965. “Causal Judgments and Causal Explanations.” Journal of Philosophy, vol. 62 (23):
695–711.
Hall, Ned. 2004. “Two Concepts of Causation.” In Collins et al. (2004), 225–276.
—— 2007. “Structural Equations and Causation.” Philosophical Studies, vol. 132: 109–136.
Halliday, M. A. K. and Ruqaiya Hasan. 1976. Cohesion in English. Longman, London.
Hart, H. L. A. and Tony Honoré. 1985. Causation in the Law. Oxford University Press, Oxford, second edn.
Hilton, Denis J. 1990. “Conversational Processes and Causal Explanation.” Psychological Bulletin, vol. 107
(1): 65–81.
—— 1995. “Logic and Language in Causal Explanation.” In Causal Cognition: A Multidisciplinary Debate,
David Sperber, David Premack, & Ann James Premack, editors, chap. 16, 495–525. Oxford University
Press, Oxford.
Hilton, Denis J. and Ben R. Slugoski. 1986. “Knowledge-Based Causal Attribution: The Abnormal Condi-
tions Focus Model.” Psychological Review, vol. 93 (1): 75–88.
Hitchcock, Christopher. 1996. “The Role of Contrast in Causal and Explanatory Claims.” Synthese, vol.
107: 395–419.
—— 2007. “Three Concepts of Causation.” Philosophy Compass, vol. 2/3: 508–516.
Hitchcock, Christopher and Joshua Knobe. 2009. “Cause and Norm.” Journal of Philosophy, vol. 106 (11):
587–612.

726
 TH E LA N G U A G E OF C AU S AT I O N

Hobbs, Jerry R. 1990. Literature and Cognition. Center for the Study of Language and Information,
Stanford.
Holland, Paul W. 1986. “Statistics and Causal Inference.” Journal of the American Statistical Association,
vol. 81 (396): 945–960.
Horgan, Terence. 1989. “Mental Quausation.” Philosophical Perspectives, vol. 3: 47–76.
Kac, Michael B. 1972a. “Action and Result: Two Aspects of Predication in English.” In Kimball (1972),
117–124.
—— 1972b. “Reply to McCawley.” In Kimball (1972), 151–156.
Katz, Jerrold. 1970. “Interpretive Semantics vs. Generative Semantics.” Foundations of Language, vol. 6:
220–259.
Kehler, Andrew. 2002. Coherence, Reference, and the Theory of Grammar. CSLI Publications, Stanford.
Kimball, J., editor. 1972. Syntax and Semantics, vol. 1. Academic Press, New York.
Krifka, Manfred, Francis Jeffry Pelletier, Gregory N. Carlson, Alice ter Meulen, Godehard Link, and Gennaro
Chierchia. 1995. “Genericity: An Introduction.” In The Generic Book, Gregory N. Carlson and Francis
Jeffry Pelletier, editors, 1–124. University of Chicago Press, Chicago.
Lakoff, George. 1965. On the Nature of Syntactic Irregularity. Ph.D. thesis, Harvard University.
Lakoff, George and John Robert Ross. 1972. “A Note on Anaphoric Islands and Causatives.” Linguistic
Inquiry, vol. 3 (1): 121–125.
Lascarides, Alex and Nicholas Asher. 1993. “Temporal Interpretation, Discourse Relations and Common-
sense Entailment.” Linguistics and Philosophy, vol. 16: 437–493.
Lewis, David K. 1973. “Causation.” Journal of Philosophy, vol. 70: 556–567.
—— 1979. “Scorekeeping in a Language Game.” In Philosophical Papers, vol. 1, 233–249. Oxford University
Press, Oxford.
—— 1986a. “Causal Explanation.” In Lewis (1986c), 214–240.
—— 1986b. “Counterfactual Dependence and Time’s Arrow.” In Lewis (1986c), 32–66. With postscript.
—— 1986c. Philosophical Papers, vol. 2. Oxford University Press, Oxford.
—— 1996. “Elusive Knowledge.” In Papers in Metaphysics and Epistemology. Cambridge University Press,
Cambridge.
Machamer, Peter, Lindley Darden, and Carl Craver. 2000. “Thinking About Mechanisms.” Philosophy of
Science, vol. 67: 1–25.
Mackie, J. L. 1980. The Cement of the Universe: A Study of Causation. Oxford University Press, Oxford,
second edn.
Mann, William C. and Sandra A. Thompson. 1988. “Rhetorical Structure Theory: Toward a Functional
Theory of Text Organization.” Text, vol. 8 (3): 243–281.
Maslen, Cei. 2004. “Causes, Contrasts, and the Nontransitivity of Causation.” In Collins et al. (2004).
Maslen, Cei, Terry Horgan, and Helen Daly. 2009. “Mental Causation.” In Beebee et al. (2009), chap. 24,
523–553.
McCawley, James D. 1968. “Lexical Insertion in a Transformational Grammar Without Deep Structure.”
In Papers from the Fourth Regional Meeting, Chicago Linguistic Society, C. J. Bailey, B. J. Darden, and A.
Davison, editors, 71–80.
—— 1972. “Kac and Shibatani on the Grammar of Killing.” In Kimball (1972), 139–149.
—— 1976. “Remarks on What Can Cause What.” In Shibatani (1976), 117–129.
—— 1978. “Conversational Implicature and the Lexicon.” In Syntax and Semantics 9: Pragmatics, Peter
Cole, editor, 245–258. Academic Press, New York.
Menzies, Peter. 1989. “A Unified Account of Causal Relata.” Australasian Journal of Philosophy, vol. 67 (1):
59–83.
—— 2004. “Difference-Making in Context.” In Collins et al. (2004), 139–180.
—— 2007. “Causation in Context.” In Causation, Physics, and the Constitution of Reality: Russell’s Republic
Revisited, Huw Price and Richard Corry, editors. Oxford University Press, Oxford.
—— 2009. “Platitudes and Counterexamples.” In Beebee et al. (2009), chap. 17, 341–367.
Mill, John Stuart. 1843. A System of Logic, Ratiocinative and Inductive. John W. Parker, London.
Morreall, John. 1976. “The Nonsynonymy of ‘Kill’ and ‘Cause to Die.’” Linguistic Inquiry, vol. 7.
Nedyalkov, V. P. and G. G Sil’nickij. 1969. “Tipologija Morfologičeskogo i Leksičeskogo Kauzativov”
[Typology of Morphological and Lexical Causatives]. In Tipologija Kauzativnyx Konstrukcyi: Morfologičeskij
Kauzativ, A. A. Xolodovič, editor, 20–50. Nauka, Leningrad.
—— 1973. “The Typology of Morphological and Lexical Causatives.” In Trends in Soviet Theoretical Linguis-
tics, Ferenc Kiefer, editor, 1–32. D. Reidel Publishing Co., Dordreet

727
 ERI C SWA N S O N

Northcott, Robert. 2008. “Causation and Contrast Classes.” Philosophical Studies, vol. 139: 111–123.
Pietroski, Paul M. 2005. Events and Semantic Architecture. Oxford University Press, Oxford.
Sanders, Ted J. M., Wilbert P. M. Spooren, and Leo G. M. Noordman. 1992. “Toward a Taxonomy of
Coherence Relations.” Discourse Processes, vol. 15: 1–35.
Schaffer, Jonathan. 2005. “Contrastive Causation.” Philosophical Review, vol. 114 (3): 297–328.
—— 2010. “Causal Contextualisms.” Ms., Australian National University.
Shibatani, Masayoshi. 1972. “Three Reasons for Not Deriving ‘Kill’ from ‘Cause to Die’ in Japanese.” In
Kimball (1972), 125–137.
—— 1973a. “Lexical Versus Periphrastic Causatives in Korean.” Journal of Linguistics, vol. 9: 281–297.
—— 1973b. A Linguistic Study of Causative Constructions. Ph.D. thesis, University of California,
Berkeley.
—— 1973c. “Semantics of Japanese Causativization.” Foundations of Language, vol. 9: 327–373.
——, editor. 1976. Syntax and Semantics 6: The Grammar of Causative Constructions. Academic Press, New
York.
——. 2000. “Introduction: Some Basic Issues in the Grammar of Causation.” In The Grammar of
Causation and Interpersonal Manipulation, Masayoshi Shibatani, editor, 1–22. John Benjamins
Publishing Company, Amsterdam.
Sloman, Steven. 2005. Causal Models: How People Think about the World and Its Alternatives. Oxford Uni-
versity Press, Oxford.
Smith, Carlota S. 1970. “Jespersen’s ‘Move and Chance’ Class and Causative Verbs in English.” In Linguistic
and Literary Studies in Honor of Archibald A. Hill, M. Ali Jazayery, Edgar Polomé, and Werner Winter, edi-
tors, vol. II, 101–109. Mouton, The Hague.
Sober, Elliot. 1984. “Two Concepts of Cause.” PSA: Proceedings of the Biennial Meeting of the Philosophy of
Science Association, vol. 2: Symposia and Invited Papers: 405–424.
Song, Grace and Philip Wolff. 2004. “Linking Perceptual Properties to Linguistic Expressions of Causation.”
In Language, Culture, and Mind, Michel Achard & Suzanne Kemmer, editors, chap. 17, 237–249. CSLI
Publications, Stanford.
Swanson, Eric. 2010. “Lessons from the Context Sensitivity of Causal Talk.” Journal of Philosophy,
vol. 107 (5): 221–242.
Talmy, Leonard. 1976. “Semantic Causative Types.” In Shiabatani (1976), 43–116.
—— 2000. Toward a Cognitive Semantics, Volume I: Concept Structuring Systems. MIT Press, Cambridge
MA.
Thomson, Judith Jarvis. 1987. “Verbs of Action.” Synthese, vol. 72: 103–122.
—— 1990. The Realm of Rights. Harvard University Press, Cambridge MA.
Under, Peter. 1977. “The Uniqueness in Causation.” American Philosophical Quarterly, vol. 14: 177–188.
—— 1984. Philosophical Relativity. University of Minnesota Press, Minneapolis.
White, Morton. 1975. Foundations of Historical Knowledge. Harper & Row, Publishers, New York.
Wierzbicka, Anna. 1975. “Why ‘Kill’ Does Not Mean ‘Cause to Die’: The Semantics of Action Sentences.”
Foundations of Language, vol. 13: 491–528.
—— 1988. The Semantics of Grammar. John Benjamins, Amsterdam.
Winograd, Terry. 1972. Understanding Natural Language. Academic Press, New York.
Wolff, Philip. 2003. “Direct Causation in the Linguistic Coding and Individuation of Causal Events.” Cogni-
tion, vol. 88: 1–48.
—— 2007. “Representing Causation.” Journal of Experimental Psychology, vol. 136 (1): 82–111.
Woodward, James. 2003. Making Things Happen: A Theory of Causal Explanation. Oxford University Press.
von Wright, Georg Henrik. 1971. Explanation and Understanding. Cornell University Press, Ithaca.
Yablo, Stephen. 2004. “Advertisement for a Sketch of an Outline of a Prototheory of Causation.” In Collins
et al. (2004), 119–137.
Zwart, P. J. 1967. Causaliteit. van Gorcum, Assen.

728
 6.3.2
Dispositional Expressions
Alexander Bird

1 Introduction
The class of dispositional expressions is varied and its extension is debated and poten-
tially very broad. Goodman (1954: 3) says that it “includes not only predicates ending
in ‘ible’ and ‘able’ but almost every objective predicate, such as ‘is red’.” We may distin-
guish covertly dispositional expressions: “soluble” and “fragile” for example, and overtly
dispositional expressions: “disposed to dissolved when immersed in water,” “disposed to
break when struck.” What makes these dispositional is, roughly speaking, that they indi-
cate that the objects of which they are predicated would behave in certain ways (their
manifestations) in certain circumstances (their stimuli).
Dispositional expressions are one kind of thing. Dispositions are another. The former
is a subclass of predicates. The latter is a subclass of properties. The connection between
the two is not at all straightforward. For dispositional and non-dispositional expressions
may co-refer (“avarice” and “the property most conspicuously displayed by Molière’s
Harpagon”). And while the class of dispositional predicates is neither empty nor
exhaustive, it is a matter of debate whether any or all properties are truly dispositional in
nature. Paradigm dispositional expressions (for example “soluble”) indicate the definite
occurrence of a manifestation in suitable circumstances; the corresponding properties
are “sure-fire” dispositions. In addition to these are terms (such as “irascible”) that indi-
cate some degree of liability for the manifestation to occur in suitable circumstances.
Correspondingly, in addition to sure-fire dispositions there is a class of disposition-like
properties that includes capacities, tendencies, and propensities.
While there has been a continuous tradition of generating and refining analyses of
dispositions since the 1930s, the function and nature of such analyses have not be con-
stant. Initially philosophers focused on analyzing dispositional expressions or concepts
(linguistic or mental items), but more recent philosophical concern purports to be more
with the properties themselves (features of the world). At the same time the motivation
for attempts at analysis has shifted. The tradition starts with Carnap’s attempt to recon-
cile the role of dispositional terms in science with his verificationism. In the middle part
of the twentieth century, interest in dispositions centered on Ryle’s dispositional view
of the mind and then later on the functionalist theory of mental states. More recently,
fundamental natural properties have been held to be dispositional in nature and the
analysis of dispositions is invoked in accounting for the laws of nature. This 80-year
history of the analysis of dispositions reveals changing conceptions of the function of
philosophical analysis and its relationship to the philosophy of language.
 A LEXA N DER B I R D

2 Carnap on Testability and Disposition Concepts

Carnap’s interest in dispositional expressions stems from a dilemma presented by his
commitment to a certain brand of empiricism. That empiricism gives a special place to
science. Science is the paradigm of what can be known and what can be said. Indeed
the logical positivists sometimes suggest that science is coextensive with the extent of
possible knowledge. According to Ayer (1936), “There is no field of experience which
cannot, in principle, be brought under some form of scientific law, and no type of specu-
lative knowledge about the world which it is, in principle, beyond the power of science
to give.” Given this emphasis on the primacy of science, it is important for the positivists
to accommodate rather than reject the statements of science—which stands in contrast
to their attitude towards the statements of metaphysics. And dispositional expressions
play an important role in the statements of science. Carnap’s principal example is the
predicate “soluble,” but many central terms in science may be considered dispositional
also (although particular cases may be disputed). “Fitness” in evolutionary biology con-
cerns an organism’s disposition to survive and breed (Mills and Beatty 1979); “electric
charge” denotes a body’s disposition to exert a force on another charged body (Broad
1933: 267); and so forth. Thus the commitment to science implies to a commitment to
provide a satisfactory philosophical account of such predicates.
This empiricism also seemed to the positivists to provide a particular role for philosoph-
ical analysis. For the all-encompassing role of science also demands, it appeared to them,
that philosophy should not be regarded as making contentful statements about the world.
If philosophers claim to be so doing, then their efforts would fail and should be excluded
from philosophy. Hence the rejection of metaphysics. The propositions of philosophy, says
Ayer (1936: 76), “are not factual but linguistic in character . . . they express definitions,
or the formal consequences of definitions.” Such a view requires a demarcation principle
to differentiate those statements that are factual (the statements of science) from those
that purport to be factual but which are not (the statements of metaphysics). Furthermore,
since philosophy is in the business of supplying definitions, it invites us to supply princi-
ples that constrain what counts as an acceptable definition. As is well known, both tasks
are achieved by the verification principle, which Carnap (1936: 420) expresses thus, “the
meaning of a sentence is in a certain sense identical with the way we determine its truth
or falsehood; and a sentence has meaning only if such a determination is possible.” Like
Ayer, Carnap (1935) emphasizes that philosophy is primarily concerned with language:

The function of logical analysis is to analyze all knowledge, all assertions of

science and of everyday life, in order to make clear the sense of each such
assertion and the connections between them. One of the principal tasks of the
logical analysis of a given proposition is to find out the method of verification
for that proposition.

So logical empiricism demands that the role of philosophy is to analyze the propositions
of science in terms of their method of verification, and in the light of the central role
they play in science, it requires that we provide such an analysis of sentences containing
dispositional predicates in particular. However, if we take the method of verification
to require direct observation, then a problem arises, for the satisfaction of dispositional
predicates is not always directly observable. That a particular crystal is soluble is not
directly observable, so long as it is not placed in water. A sample of barium sulphate and

730
 DI SPO SI TI O N A L E X P R E S S I O N S

a sample of sodium chloride may look just the same. If neither is immersed in water, then
there is no apparent difference between the solubility of the former and in the insolubil-
ity of the latter. Likewise a charged particle and a neutral particle will behave the same
so long as there is no electric or magnetic field. Since an important scientific difference
is not always directly verifiable, we have a prima facie challenge to verificationism.
While the immediate context of Carnap’s article ‘Testability and meaning’ is the
analysis of a key class of scientific predicates, logical empiricists also had a broader
concern with dispositions. Phenomenalism has always been attractive to empiricists
as a way of maintaining the exclusive epistemological role of experience without suc-
cumbing to skepticism. But since a simple reduction to actual experiences leads to well-
known problems, unless one resorts to Berkeley’s theological solution, it is natural to
consider reduction to possible experiences, as in Mill’s “permanent possibilities of sensa-
tion” or later Mach’s “functional relations of elements” (where elements are, more or
less, the same as sensations). It is natural to interpret Mach’s relations as dispositions.
Mach himself intended “function” to have its mathematical sense. But, then, Schlick
(1918: 212–14) complains that we would be reducing a material thing to “something
quite shadowy,” and, furthermore, something that is a relation between things that do
not exist (the possible but non-actual sensations).
So, in order to satisfy the verificationist demand that meaningful statements should
be verifiable, the testing that will verify or confirm the presence or otherwise of a dispo-
sition needs to be more than direct observation. It might be tempting to think that “sol-
ubility” and “charge” name unobservable properties of things that cause their behavior
in water or in electro-magnetic fields. But this is precisely the kind of metaphysics that
positivism rejects. Instead we should understand these expressions in terms of the kind
of observable test that would confirm their correct application. Clearly an appropriate
test of the difference between a soluble item and an insoluble item is to observe their
dissolving or not dissolving when the items are placed in water. So, according to Carnap
(1936), a natural first pass at a definition of “x is soluble” is “whenever x is placed in
water, x dissolves.”
The use of “whenever” is a mistake, for as Mellor (1974: 106) notes, this makes solu-
bility an immutable property. It may be that in using “whenever” Carnap was influenced
by a sense of the modal nature of disposition ascriptions, to which we shall come later.
Be that as it may, whether an object is soluble now should not depend on how it is at
previous or later times, for things can gain and lose dispositions. If we ignore this error,
we can express Carnap’s view thus:

(D) Sx iff (Wx o Dx)

where ‘S”’ symbolizes ‘x is soluble’; ‘Wx’ is ‘x is placed in water; ‘Dx’ is ‘x dissolves. Wx

is the test condition (later stimulus condition) and Dx is the response condition (also
manifestation condition).
The objection to (D) that Carnap immediately raises concerns an item that is never
in its history placed in water, for example a match which is never placed in water and
then is burned up. We see that the right-hand side of (D) is satisfied trivially. Thus the
match counts as soluble even though it clearly is not.
Carnap then considers the “bilateral reduction sentence”:

(R) Wx o (Sx l Dx).

731
 A LEXA N DER B I R D

(R) provides a test both for the presence of and for the absence of solubility, and to that
degree satisfies the requirements of the verification principle concerning meaning. On
the other hand, as Carnap recognizes, (R) does not say anything about the solubility
or otherwise of some item that is not in water. And so (R) does not provide any way of
eliminating talk of solubility.
These points against (R) were also made by Storer (1951). Storer’s response was to
note that a soluble item not in water is like, in other respects, items that are in water
and dissolving and is unlike, in that respect, items that are insoluble. So to say that
something is soluble is to say:

either it is in water and dissolves; or it has some property B, such that B is pos-
sessed by some other item that is in water and does dissolve and B is not pos-
sessed by any item that is in water and does not dissolve.

Storer’s proposal is notable in that it is the first to introduce a component that would later
become to be thought of as the causal basis of the disposition. This is moving away from
Carnap’s concern with verifiability, to the extent that the possession of the causal basis is
not itself something that is guaranteed to be verifiable. For example, the sort of property
that Storer had in mind would be “being sugar.” Being sugar is not a directly verifiable
property, and might be held to be one that is verified by the identification of its character-
istic dispositions. Furthermore, without restricting the quantification over properties (for
example to natural properties) there are substitutions for B that make any object soluble
(for example the property of being identical either to this stone [insoluble, not in water] or
to that sugar cube [in water and dissolving], which is a property possessed by the stone).
More immediately problematic is the fact that we can imagine that an object might
have a disposition yet be the only thing of its kind to have that disposition. If it does not
undergo the relevant test procedure (for example being placed in water) then Storer’s
analysis will deny that the object has the disposition. For example, we might imagine
that industrial chemists devise an entirely new material that is soluble. The basis of its
solubility is a novel feature of this material, and so no other substance dissolves in virtue
of possessing this property. If the chemists only ever make one sample of this material
which they burn in due course, without ever placing in water, then this material will not
count as soluble. On the other hand, had the chemists made a second, identical sample,
which they did place in water (and so which does dissolve), the first sample would count
as soluble. In short, the objection is that whether or not an object has a disposition such
as solubility cannot depend on whether other similar objects exist and are subjected to
the relevant test.

3 Introducing Stronger than Material Conditionals

Carnap and his contemporaries sought to provide an analysis of dispositional concepts
employing only second-order, classical, extensional logic. The suggestion that the
second- order quantification should be limited to natural properties is one proposal that
would break away from this restriction. A more important proposal in the same direc-
tion concerns the nature of the appropriate conditional. Ryle (1949: 123) asserts that
“To say that this lump of sugar is soluble is to say that it would dissolve, if submerged
anywhere, at any time and in any parcel of water.” Storer (1951: 134) says concerning
definitions of dispositional concepts (such as colour predicates):

732
 DI SPO SI TI O N A L E X P R E S S I O N S

The peculiarity of all such definitions is the occurrence of sentences of the

type: “If so and so were to happen, then such and such would be the case.” In
a current phrase, all definitions of dispositional predicates involve the use of
contrary to fact conditionals.

So both Ryle and Storer recognize the connection between dispositions and counter-
factuals but retreat from making much of this connection when giving further detail,
primarily because of empiricist concerns at the metaphysical implications of taking
counterfactuals at face value. Counterfactual conditionals (and other subjunctive con-
ditionals, which are understood to be included) have a modal component, seemingly
telling us about non-actual potentialities. Ryle takes it that there can be no fact of the
matter concerning non-actual potentialities. Consequently, the sentence “this lump of
sugar would dissolve if placed in water” does not assert some factual truth, such as the
attribution of a property to a thing. Rather, along with law statements, such assertions
must be understood as inference tickets: one is entitled to infer from “this lump of sugar
is in water” to “this lump of sugar is dissolving.” In effect, the modal feature of disposi-
tions is located in the inference ticket. Ryle does not tell us what features of the world
entitle us to employ such an inference ticket.
Storer points out that the second half of the second disjunct in his analysis (“B is
not possessed by any item that is in water and does not dissolve”) is equivalent to:
everything that is B is such that if it is placed in water then it dissolves. That covers
the conditional component of the counterfactual conditional, but not the modal. The
modal feature is in effect what Storer is trying to capture by the idea of there being a
causal basis that is shared between an item not undergoing the test and other items that
are being tested. Implicitly, Storer is suggesting that the causal basis is a property that
would bring about the response in the object, were it to be tested—but without resort-
ing to modal language.
In the light of the forgoing it is not surprising that philosophers should eventu-
ally conclude that dispositions could not be analyzed using the material conditional.
Rather, a stronger than material conditional needs to be employed, which we may
symbolize by ‘Ÿ’ (without saying too much about its nature). We can also use ‘Sx’
to denote the test/stimulus condition, ‘Mx’ to denote the response/manifestation con-
dition, and ‘D(S,M)’ to denote the disposition to yield manifestation M in response to
stimulus S.
Sellars (1958) asserts that ascription of a disposition is simply to assert a relation of
implication between the stimulus and manifestation. Hence:

(W) D(S,M)x iff Sx Ÿ Mx.

The philosophical task, according to Sellars, is to explain what ‘Ÿ’ means. Sellars him-
self draws on the idea of ‘causal implication’ (which we find also in Pap 1958), and
expresses the idea found in Storer and also in Burks (1955), that we may need to appeal
to the idea that when there is dispositional relationship between Sx and Mx, there is a
kind or property to which x belongs and which plays some kind of causal or nomologi-
cal role in bringing about Mx. These authors were concerned whether their notions of
causal implication are compatible with a Humean regularity view of causation and law
(Malzkorn 2001: 343).

733
 A LEXA N DER B I R D

4 The conditional analysis of dispositions

The position in the 1950s was that philosophers recognized that dispositional and
counterfactual assertions are related and that both of these have connections with
statements concerning laws and causes. Goodman (1954) distinguished the analysis of
counterfactuals from analyzing the meaning of law statements. On the other hand, by
his own admission, Goodman was unable to articulate the details of their relationship.
Furthermore, he remained committed to a Humean view of laws that distinguishes them
from other regularities only in virtue of our propensity to use them in inferences and
predictions (cf. the Rylean inference ticket view of dispositions and laws mentioned
above).
The discussion of the analysis of dispositions was given a major impetus by the devel-
opment of a semantics for counterfactuals by Stalnaker (1968) and Lewis (1973), fol-
lowing earlier work by Kripke on semantics for modal logic. The semantics provided
for counterfactuals made them philosophically respectable, while also articulating their
problematic relationship with laws. Lewis also provided an account of causation in
terms of counterfactuals, allowing a further dissociation of counterfactuals, laws, and
causes. Thus it was possible to see that the causal conditional of Burks, Pap, and Sellars
is a conflation of two related but separate notions, the counterfactual conditional and
causation.
Furthermore, Lewis’s account is based on an objective Humean view of laws rather
than one depending on inference tickets or habits, thus providing a firm metaphysical
basis for understanding dispositions from a Humean/empiricist perspective. At the same
time, concern about the empiricist credentials of any analysis was waning as a result of
a more general retreat from empiricism in the philosophy of science, and the resurgence
of interest in metaphysics, especially modal metaphysics, thanks to the work of Kripke
and others.
In the light of the above, we can replace the ‘Ÿ’ in (W) with a pure counterfactual
conditional: ‘ o’. Thus (W) becomes what would become known as the (simple) con-
ditional analysis of dispositions, the basis of much of the recent discussion of disposi-
tional expressions:

(CA) D(S,M)x iff Sx oMx.

By the 1960s, it was widely accepted that dispositional statements either mean the same
as or at least entail counterfactual or subjunctive conditionals. Armstrong (1969: 23)
tells us, as if it were not much more than a platitude, that “If we consider the attribu-
tion of (say) brittleness to a particular piece of glass then one outstanding feature of the
attribution is that it licenses certain conditionals. If the glass remains unbroken, then
conditionals will be ‘counterfactual’. If the glass had been struck, it would have broken.”
However, until Stalnaker and Lewis, counterfactuals were themselves sufficiently myste-
rious (and even suspect) that authors felt obliged not to rest content with analyzing dis-
positions in terms of counterfactuals but were required to bypass the counterfactuals and
to give an account in yet further terms that reflect wider philosophical (typically meta-
physical) concerns. After Lewis, it was possible to distinguish acceptance of the (sub-
junctive/counterfactual) conditional account of the meaning of disposition statements
from discussion of the metaphysics of dispositions. Thus (CA) was generally accepted,
while debates focused on metaphysical issues that will be mentioned below.

734
 DI SPO SI TI O N A L E X P R E S S I O N S

5 Objections to the conditional analysis and further developments

The conditional analysis, however, suffers from now well-known flaws. Martin’s (1994)
counterexample of a finkish disposition exploits the fact that in a normal case of stimu-
lating a disposition in order to bring about its manifestation, that process takes time.
The object may be stimulated at time t, but the manifestation occurs only at t+G. Dispo-
sitions come and go. Very hot glass is not fragile. So a fragile glass might lose its fragility
by being heated. Let us imagine that were the stimulus to occur at t, that would cause
the disposition to disappear very quickly, certainly well before t+ G. As a consequence,
the process that would normally lead to the manifestation is interrupted, and the mani-
festation does not occur. So, for example, striking a fragile glass causes it to be heated
very rapidly, sufficiently rapidly that the process of breaking is halted, and the glass does
not break. In such a case the glass is fragile at t, but, since it is struck at t but does not
subsequently break, it is false at t that were the glass struck it would break. (See Lewis
(1997) for an attempt to reform (CA) to handle finks.)
Other objections focus on the fact that even if the disposition remains in place, its
presence plus the stimulus may not be causally sufficient to bring about the manifesta-
tion. For environmental conditions may need to be appropriate, and the causal process
may need to take place in a particular way. If such conditions are interfered with, the
manifestation may not occur. Such interferers are masks (Johnston 1992) or antidotes
(Bird 1998).
The 1990s and 2000s saw increased interest in the analysis of dispositions coming
from a number of quarters. In the philosophy of language, Kripke (1982) considered
and rejected dispositional accounts of rule-following, meaning, and understanding. But
Martin and Heil (1998) argued that such a rejection depends on accepting (CA). The
falsity of (CA) means that a counterfactual account of X and a dispositional account of
X will differ in certain cases. Consequently problems with a counterfactual account of,
for example, intentional or free action (Frankfurt 1969) may be overcome by preferring
an account in terms of dispositions or related states, such as capacities (Smith 1997). In
both cases the objections can be seen to be trading on finks or masks/antidotes.
As it stands, (CA) provides an analysis only of overtly dispositional locutions, such
as “is disposed to dissolve when placed in water.” Nonetheless, most of the discussion of
the analysis of dispositional expressions concerns covertly dispositional locutions such
as “soluble,” proceeding on the assumption that the equivalence between the covert
and overt expressions is straightforward and obvious. For example, we found that (CA)
seems to be refuted because something may be fragile, yet it is false that it would break
if stressed. But such a counterexample works only on the assumption just discussed,
that the covertly dispositional “x is fragile” is equivalent to the overtly dispositional “x
is disposed to break when stressed.” However, if that assumption is mistaken then the
existence of finks and antidotes cannot be taken immediately to refute (CA). It might
instead refute the simplistic equation of covertly and overtly dispositional expressions
(cf. Lewis 1997: 153; Choi 2003: 576–7). Indeed Choi (2008) undertakes a defence of
(CA) on precisely this basis.
A problem arises when we combine (CA) with the standard Stalnaker–Lewis seman-
tics for counterfactuals. That semantics include the centering condition: AšB entails
A oB. Thus any two facts are related by the subjunctive conditional ‘ o’. (CA) says
that any two possible or actual states that are subjunctively related are dispositionally
related. Combining these tells us that any two facts are dispositionally related. Since

735
 A LEXA N DER B I R D

that is clearly false, either (CA) needs amendment or the Stalnaker–Lewis semantics
does. Since there are independent reasons for wanting to adjust the latter, it might be
worth considering whether further adjustments to the semantics for ‘ o’ are required.
For example, we may give up not only centering but also weak centering: AB entails
AoB. If we do this, then it is no longer clear that finks and antidotes suffice to refute
(CA). For then a case where the stimulus S occurs but the manifestation M does not
occur is consistent with S oM. That might be justified if what replaces weak centering
is the condition that only in normal worlds where S oM is true, is SoM also true (for
example worlds without finks or masks/antidotes; cf. Gundersen 2002, 2004).
While such moves permit us to strengthen the tie between dispositions and subjunc-
tive/counterfactual conditions, other developments propose movement in the opposite
direction: for example the proposal that dispositions should be aligned with habitual or
generic propositions (Fara 2005).

6 Metaphysical Considerations Again

Carnap’s original analysis was motivated by the (anti-)metaphysical considerations
that underlie his verificationism. The latter is a doctrine concerning the meanings of
expressions, and so the “analysis of dispositions” is conceived of as providing a reductive
account of linguistic expressions or concepts. A central concern was to avoid an analysis
in terms that were equally troubling to a verificationist, such as terms for unobservable
properties or stronger-than-material conditionals.
The view that without suitable analysis dispositions were to be deemed not entirely
respectable remained even after the verificationist impetus to Carnap’s project had
waned, and counterfactual/subjunctive conditionals themselves were accepted. (CA)
tells us that to predicate a disposition of an object is equivalent to asserting a subjunc-
tive conditional of it. But that tells us little about how the world and in particular the
object in question must be for the dispositional or subjunctive predications—D(S,M)a
and Sa Ma—to be true. Ryle’s inference ticket phenomenalism rejects the idea that
there is any particular way the actual world is. But as metaphysics itself returned to
respectability, the question became more pressing: what would make either side of (CA)
true? A common answer is that the subjunctive/counterfactual conditional is made true
by the existence of some categorical property plus the laws of nature. Thus:

‘D(S,M)a’ is made true by, for some categorical property F, Fa and its being a law
of nature that x(FxšSx o Mx)

A categorical property is held to be one that has no troubling dispositional or condi-

tional character, but is necessarily always fully manifest—shape and duration are often
held to be examples, albeit disputed ones. As Mellor (1974: 157) puts it (not his own
view): “Dispositions are as shameful in many eyes as pregnant spinsters used to be—ide-
ally to be explained away, or entitled by a shotgun wedding to take the name of some
decently real categorical property.”
In the light of the above, the following remarks are sometimes made:

(I) The dispositional–categorical distinction is primarily a conceptual or lin-

guistic distinction, not a metaphysical one. The dispositional–categorical
distinction is not a distinction between categories of properties but between

736
 DI SPO SI TI O N A L E X P R E S S I O N S

classes of expression, between those that permit a conditional analysis and

those that do not (Strawson 2008; cf. Mumford 1998: 65).
(II) Dispositional expressions (‘D(S,M)’ in the above) are ones that characterize
properties (such as F in the above) via their typical effects in the actual
world with its actual laws of nature (Armstrong 1969, 1997, Quine 1973;
Mackie 1973). Or they refer to higher-order functional properties (for
example the property of possessing some categorical property such as F
in the above) (Prior et al. 1982; Prior 1985). In the context of the phi-
losophy of mind, where mental states may be seen as dispositions, the
former yields an identity theory, whereas the latter yields the functional-
ist view.
(III) Correspondingly, the failure of (CA) thanks to finks and masks/antidotes
is sometimes held (a) to undermine the distinction between dispositional
and categorical expressions, and (b) to vindicate the metaphysical program
of rehabilitating dispositional properties as real properties as distinct from
being either just as the shadows of conditionals or identical to categorical
properties or higher order functional properties realized by categorical
properties (Wright 1990, Martin 1994: 7; Mumford 1998: 63; Schrenk
2010: 171; cf. Mellor 1974).

Such debates reflect some unclarity about the nature of attempts to analyze dispositions
and the relationship between the semantics and metaphysics of dispositions. As the idea
that philosophical analysis is a matter of investigating our concepts comes under ques-
tion (Williamson 2007), it might appear that the alternative in this case will hold that
the philosophical activity of analyzing dispositions concerns those properties, the dispo-
sitions, rather than our concept of disposition. But to take such a view requires that there
is indeed a distinct class of things, the dispositional properties. That is what (I) above
denies. (I) itself is plausible to the extent that it is plausible that (CA) or something like
it is true. Conversely, if it is true that we cannot find a straightforward analysis of disposi-
tions, it become more plausible that our dispositional expressions do pick out a distinct
class of properties, properties that are dispositional in nature. The position has a rough
analog in the analysis of knowledge. If the simple justified true belief account of knowl-
edge were correct, then that would indicate that “knowledge”is a term that serves simply
to pick our a subclass of beliefs, those that meet certain additional conditions. In which
case, it is plausible that the analysis of knowledge is just a matter of analysis a concept.
On the other hand, as Williamson (1995) argues, the failure of the attempt to analyze
knowledge into belief, plus other conditions, is evidence that states of knowing are not a
subclass of the states of belief but constitute a distinct kind of mental state. In which case
the analysis of knowledge (conceived more broadly than simply supplying necessary and
sufficient conditions) is an investigation into the nature of knowledge itself.
On the other hand there is uncertainty as to whether the dispositional and disposi-
tion-like expressions themselves form a unified class. The distinction between covert
and overt dispositional expressions has already been mentioned; it is not agreed that we
can assimilate these to a single class. Moreover, terms for abilities, capacities, tenden-
cies, and propensities have similarities to disposition terms, but no clear unification of
all these is yet available. Putting that variety on one side, there are questions about what
sort of thing such terms could refer to. The expression “dispositional property”carries
with it a degree of ambiguity. Some terms in basic physics refer to properties that are

737
 A LEXA N DER B I R D

argued to be dispositional in nature or essentially dispositional, for example “inertial

mass,” “charge,” “velocity,” and so forth (Ellis and Lierse 1994; Bird 2007; Lange 2005).
Such properties are fundamental natural properties and so are good candidates for being
sparse universals. On the other hand, the paradigm disposition expressions discussed in
the literature, such as “fragility,” denote properties that are neither fundamental nor
obviously natural (note the diversity of things that can be fragile and the manner of
their being fragile, such as both an old parchment and the economy). If it is appropriate
to think of these as referring to entities at all, the referents will be abundant universals.
If ‘T’ refers to a sparse universal, it is plausible to think that the analyzing T is analyz-
ing the thing, the universal T, and that this is a different exercise from analyzing the
concept T (or the term ‘T’). On the other hand, if ‘T’ denotes an abundant universal,
then it is rather less clear that there is distinction between analyzing T and analyzing
the concept T. For the existence of abundant universals seems to be little more than an
ontological shadow of the possibility of predication by the concept T—if, that is, there
are abundant universals at all. So if properties that are paradigmatic dispositions such
as fragility, solubility, malleability, etc. are not natural properties, then it is difficult to
see how there can be a project of analyzing dispositions that is different from analayzing
the concept disposition.

7 Conclusion
The earliest phase of the history of analyses of dispositions conceives itself as engaged
in the task of analyzing dispositional expressions or concepts, providing substitutions
equivalent in meaning. It was motivated by radically empiricist (anti-)metaphysics and
epistemology which give rise to the verificationist criterion of meaningfulness. Car-
nap’s own failed attempts show that it is impossible to provide a complete analysis of
dispositional expressions that complies with that criterion. The same empiricism limits
Carnap to use of truth-functional connectives, so his “conditional” analysis of disposi-
tions invokes only the material conditional, which is too weak for the purposes required
of it (for example characterizing what is true of a soluble item that is not in water and
distinguishing it from an insoluble item not in water). Some authors noted that a solu-
ble item not in water will often be similar to a soluble item that is placed in water (and
so is dissolving). This approach is of interest because for the first time it introduces the
idea of what we would now call the causal basis of the disposition, and because it raises
questions concerning the naturalness of properties. Metaphysical questions of a kind
inimical to radical empiricism are beginning to come to the fore. The true break with
Carnap’s intended program comes when it is understood that the material conditional
needs to replaced by something stronger and so non-truth-functional. This eventually
settles on the counterfactual/subjunctive conditional, giving the standard (simple)
conditional analysis of dispositions. It settles on this slowly, partly because of resid-
ual empiricist scepticism about such conditionals but also because such conditionals
were poorly understood—a state of affairs remedied by the possible-worlds semantics of
Stalnaker and Lewis in the late 1960s and early 1970s.
The conditional analysis, backed up by a semantics for the conditional, is only the
beginning of the contemporary story of the analysis of dispositions. This debate is pur-
sued not, as it was by Carnap, to avoid substantive metaphysics but in order to aid it, first
in order better to understand the commitments of a dispositional–functional account
of mind and then thanks to dispositional essentialist accounts of natural properties and

738
 DI SPO SI TI O N A L E X P R E S S I O N S

laws. Nonetheless, the business of analyzing dispositions is still seen primarily as one
concerning the nature of disposition expressions and concepts. But as such an approach
is questioned by the recent debate regarding the nature of philosophical analysis itself:
Is analysis concerned with concepts or with the things themselves? The intertwining of
the metaphysics and semantics of dispositions suggests that the plausibility of compet-
ing answers to that question may itself depend on the outcome of the actual process of
analysis itself—a process that is by no means complete.

References
Armstrong, D. M. (1969) Dispositions are causes. Analysis 30: 23–6.
Armstrong, D. M. (1997) A World of States of Affairs. Cambridge: Cambridge University Press.
Ayer, A. J. (1936) Language, Truth, and Logic. London: Victor Gollancz. References to the Pelican edition,
Harmondsworth (1971).
Bird, A. (1998) Dispositions and antidotes. Philosophical Quarterly 48: 227–34.
Bird, A. (2007) Nature’s Metaphysics: Laws and properties. Oxford: Oxford University Press.
Broad, C. D. (1933) Examination of McTaggart’s Philosophy, Volume I. Cambridge: Cambridge University
Press.
Burks, A. (1955) Dispositional statements. Philosophy of Science 22: 175–93.
Carnap, R. (1935) Philosophy and Logical Syntax. London: K. Paul, Trench, Trübner.
Carnap, R. (1936) Testability and meaning (part 1). Philosophy of Science 3: 420–71.
Choi, S. (2003) Improving Bird’s antidotes. Australasian Journal of Philosophy 81: 573–80.
Choi, S. (2008) Dispositional properties and counterfactual conditionals. Mind 117: 795–841.
Ellis, B. and C. Lierse (1994) Dispositional essentialism. Australasian Journal of Philosophy 72: 27–45.
Fara, M. (2005) Dispositions and habituals. Noûs 39: 43–82.
Frankfurt, H. G. (1969) Alternate possibilities and moral responsibility. The Journal of Philosophy 66:
829–39.
Goodman, N. (1954) Fact, Fiction, and Forecast. London: Athlone Press. Page references to 4th edition
(1983).
Gundersen, L. B. (2002) In defence of the conditional account of dispositions. Synthese 130: 389–411.
Gundersen, L. B. (2004) Outline of a new semantics for counterfactuals. Pacific Philosophical Quarterly 85:
1–20.
Johnston, M. (1992) How to speak of the colors. Philosophical Studies 68: 221–63.
Kripke, S. (1982) Wittgenstein on Rules and Private Language. Oxford: Blackwell.
Lange, M. (2005) How can instantaneous velocity fulfill its causal role? Philosophical Review 114: 433–68.
Lewis, D. K. (1973) Counterfactuals. Oxford: Blackwell.
Lewis, D. K. (1997) Finkish dispositions. Philosophical Quarterly 47: 143–58.
Mackie, J. L. (1973) Truth, Probability, and Paradox. Oxford: Oxford University Press.
Malzkorn, W. (2001) Defining disposition concepts: A brief history of the problem. Studies in History and
Philosophy of Science 32: 335–53.
Martin, C. B. (1994) Dispositions and conditionals. Philosophical Quarterly 44: 1–8.
Martin, C. B. and J. Heil (1998) Rules and powers. Philosophical Perspectives 12: 283–311.
Mellor, D. H. (1974) In defense of dispositions. Philosophical Review 83: 157–81.
Mills, S. K. and J. H. Beatty (1979) The propensity interpretation of fitness. Philosophy of Science 46: 263–86.
Mumford, S. (1998) Dispositions. Oxford: Oxford University Press.
Pap, A. (1958) Disposition concepts and extensional logics. In H. Feigl and M. Scriven (eds.), Minne-
sota Studies in the Philosophy of Science vol. 2, pp. 196–224. Minneapolis, MN: University of Minnesota
Press.
Prior, E. (1985) Dispositions. Aberdeen: Aberdeen University Press.
Prior, E., R. Pargetter, and F. Jackson (1982) Three theses about dispositions. American Philosophical
Quarterly 19: 251–257.
Quine, W. (1973) The Roots of Reference. Lasalle, IL: Open Court.
Ryle, G. (1949) The Concept of Mind. London: Hutchinson.
Schlick, M. (1918) Allgemeine Erkenntnislehre. Berlin: Julius Springer. References to the translation by A.
Blumberg, General Theory of Knowledge, published by Open Court, La Salle, IL, (1985).

739
 A LEXA N DER B I R D

Schrenk, M. (2010) Antidotes for dispositional essentialism. In A. Marmodoro (ed.), The Metaphysics of
Powers, Their Grounding and Their Manifestations. Abingdon: Routledge.
Sellars, W. (1958) Counterfactuals, dispositions, and the causal modalities. In H. Feigl and M. Scriven
(eds.), Minnesota Studies in the Philosophy of Science vol. 2, pp. 225–308. Minneapolis, MN: University of
Minnesota Press.
Smith, M. (1997) A theory of freedom and responsibility. In G. Cullity and B. Gaut (eds.), Ethics and Practi-
cal Reason. Oxford: Clarendon Press.
Stalnaker, R. (1968) A theory of conditionals. In Studies in Logical Theory. Oxford: Blackwell.
Storer, T. (1951) On defining ‘soluble’. Analysis 11: 134–37.
Strawson, G. (2008) The identity of the categorical and the dispositional. Analysis 68: 271–82.
Williamson, T. (1995) Is knowing a state of mind? Mind 104: 533–65.
Williamson, T. (2007) The Philosophy of Philosophy. London: Wiley–Blackwell.
Wright, A. (1990) Dispositions, antirealism and empiricism. Proceedings of the Aristotelian Society 91:
39–59.

740
 6.4
P HIL OS OPHY OF
L ANGUAGE F OR
N OR MATI V E ET HIC S
6.4.1
Language, Gender, and Sexuality
Sally McConnell-Ginet

1 Why Might We Care?

Interest in the interaction of language with gender and sexuality has primarily arisen
from moral concerns about unjust social arrangements based on sex, sexual orienta-
tion, or gender. There are certainly people interested in questions in this area who take
women and men to differ profoundly from one another and see gender arrangements
in general, including their linguistic dimensions, as having nothing to do with social
justice and everything to do with biologically fixed natures. Similarly, some see gays
and lesbians as morally deviant and do not see arrangements that tilt towards straights
as involving any kind of morally questionable bias. And even those who do think that
sexism and homophobia are morally repugnant may pursue inquiry into the intersec-
tion of language, gender, and sexuality that is not focused on ethical dimensions. While
not assuming that all men are better off than all women (certainly not the case) or
that sexual minorities everywhere face problematic discrimination (again, matters are
more complicated), this discussion will emphasize the relation of language to power and
privilege, its relation to people’s well-being as members of communities as well as its
cognitive impact.
On the one hand, language can be seen as a mirror of society and culture. It can help
us better understand gender and sexual arrangements and can provide important mate-
rial for demystifying sexism, heterosexism, and related structures deeply embedded in
familiar ways of saying things. On the other hand, language can also be viewed as actu-
ally doing things, as supplying tools and weapons, resources that benefit or harm. The
suspicion is that women and sexual minorities suffer much of the harm and that straight
men are most likely to reap benefits. The second perspective, which takes language as
not simply indicative but partly constitutive of social injustice, is more controversial
 SA LLY M CCO N N E L L - G I N E T

but ultimately more illuminating. Controversy about the possibility of language as a

tool or weapon often results from restricting language to linguistic structures, which in
and of themselves are causally empty. Where language actually does enter into human
affairs, it is because linguistic practices are fundamentally intertwined with other social
practices both at local community and larger institutional levels as well as important
for framing and sustaining ideologies. And, importantly, some of those connections are
mediated by cognitive processes.

1.1 Semantic Derogation and Pejoration Over Time

It is fairly easy to see how linguistic phenomena can be taken as reflecting sexism or
homophobia and then used in diagnosing the unjust social arrangements supporting
these biases. For example, many words for women that were once evaluatively neu-
tral have degenerated into slurs, usually with sexual overtones, whereas words for men,
though they do occasionally undergo pejoration, do not do so with such frequency and
are far less sexualized. Hussy, e.g., was once just a familiar form of huswif ‘housewife’;
it now has an archaic flavor but in my own girlhood it was an insult aimed at girls and
women who were judged ‘forward’ or ‘impudent’ and who either in words or other deeds
indicated a lack of respect for traditional sexual ‘morality’—i.e., for women, chastity,
passivity, and deference to men.
Many readers probably know that it’s been centuries since hussy acquired its censori-
ous meaning. Presumably, that particular word only gives evidence of past attitudes
towards women and their sexuality, attitudes importantly inflected by class and other
social factors. Etymology tells us nothing about current gender arrangements. It does,
however, tell us something about common assumptions at a particular point in the past
(early to mid-17th c) about those labeled huswif that must have made it possible for
people to implicate conversationally by hussy what the word now conveys convention-
ally (McConnell-Ginet 1989 offers a ‘just-so’ account of this shift). But few contem-
porary speakers even use hussy, much less connect it to housewife. Is the example rel-
evant? Although things certainly have changed, studies of terms for designating women
and men conducted very recently continue to document problematic attitudes—to be
labeled stud often counts as an achievement whereas the label slut is seldom embraced.
And it is the wide-scale patterning of which the particular huswif-hussy shift is an exam-
ple that is really significant, not the particular case.

1.2 Language as Tool or Weapon

Less obviously, linguistic practices can be understood as in part constitutive of unfair gen-
der and sexual practices, sometimes directly and sometimes through helping strengthen
the attitudes and ideologies that sustain them. I will briefly consider examples of each.

1.2.1 Epithets and slurs

Epithets seem to work directly, not just to express contempt and hostility but to enact
them. Bitch is sometimes used to refer to arbitrary girls and women but it is also used
to attack particular people. Uppity women who might be asserting themselves forcibly
can expect to hear you bitch, and the association of the slur with female strength and
resistance has led some women to embrace the designation for themselves. But the form

742
 LA N G U A G E, G EN DER, A N D S E X U A L I T Y

is also used to impose servility—you’re my bitch and don’t forget it—few women would
want to reclaim that usage. Such an utterance can also occur as an extremely offensive
move from one male prisoner to another, an assertion of dominance and sexual control
and in that context often a verbal prelude to rape. This kind of usage mixes sexism and
homophobia, a toxic combination that recurs with other slurs.
The once unprintable cunt uses a word for female genitals to attack and here too the
attack can be by a man on another man. (The word can also be used just to designate
women in general, a usage that is probably most common in hypermasculine contexts like
locker-rooms—here the harm comes in indirectly via the way in which such practices
sustain sexist ideologies.) Many find this form so thoroughly offensive that they cannot
say it even to provide an example just as some find the vitriolic force of the “N-word”
leaking through even in indirect speech or didactic contexts (see Anderson & Lepore,
forthcoming). Though the form prick uses a term designating male genitals to insult, its
force is relatively mild, and it is not used to refer to men in general but just those with
particular kinds of characteristics. Furthermore, there is no general practice of attacking
or insulting women by applying male-designating terms to them. Men are, however,
insulted even by terms that in their application to women are not directly insulting:
girl, lady, woman—and it is overwhelmingly men who issue such insults to other men or
boys, usually to impugn their masculinity, which is often taken to require heterosexual-
ity as well as aggression, toughness, and the like. The asymmetry in insult potential of
terms designating females and males is related to the general phenomenon, discussed
in the next section, of equating the ‘normal’ human condition with maleness. Epithets
explicitly targeting sexual minorities are also often used to police gender and sexual
conformity. Dyke and faggot express hostility when used by those who do not identify as
lesbians or gay men, although, like bitch, in-group reclamation is possible as in Alison
Bechdel’s comic strip, Dykes to Look Out For.
Most people find it easy to see slurs hurled at someone directly as harming or at the
very least upsetting that person. But it is not easy to understand just how this works.
As a child, my friends and I would intone “sticks and stones may break my bones, but
words will never hurt me,” as tears ran down our cheeks in the face of verbal onslaughts
from peers (fatty, four-eyes, and nigger-lover were among those I remember trying with
my chant—unsuccessfully, of course—to render powerless). Expanding on (and modify-
ing somewhat) ideas in Kaplan 1999 and the comments on them in Kratzer 1999, Potts
2007 has proposed that epithets and ‘expressive’ language generally involve a dimension
of content orthogonal to the primary or ‘at-issue’ content of the utterance. Although
this approach has virtues, it does not get at the ‘visceral’ character of at least some slurs,
their resistance to rendering in other terms, their capacity to inflict psychic wounds,
none the less real for not involving broken bones or the like. The idea of backgrounded
negative content does, however, help explain why slurs used even in the absence of
their targets might nonetheless be deeply harmful because they help ‘naturalize’ the
highly negative attitudes that underlie them, treating them as defaults that the speaker
presumes are shared by everyone not in the target group. The harm in such ‘naturalizing’
effects is indirect in the sense of not being immediately realized in a particular utterance
situation but such subtle support of problematic ideologies, which is not only hard to
confront but often passes unnoticed, is probably ultimately even more consequential
than direct harm of the kind delivered by targeted slurs .

743
 SA LLY M CCO N N E L L - G I N E T

1.2.2 Masculine generics

A major target of feminist-backed language reform campaigns in the 1970s and 1980s
was he/man language of the sort promoted in English classes for more than a century.
This includes the use of the pronominal forms he/him/his not only in contexts where
potential referents are presumed male but also in contexts where referents might be
female (as in i).

i. Any student having trouble in a course should contact his instructor about whether
to drop.

It also includes the use of man/men to designate both the species homo sapiens (without
an article as in ii) and (with an article as in iii or without as in iv) to designate male
human beings.

ii. Man is thought to be the only species that anticipates death.

iii. A strange man just knocked on the door.
iv. Man without woman will not flourish.

And there are contexts where it is unclear whether man designates the species generally
or just the male members thereof. Sentence v can be interpreted as equivalent to iv but
it need not be, having a possible more inclusive interpretation.

v. Man without a mate will not flourish.

(These examples also seem to presume heterosexuality, although it may now be

possible to construe mate as potentially covering a ‘life partner’ of the same sex.) As
many observed (and a few empirical studies confirmed), he/man talk often rendered
women less visible than if more inclusive language were used—e.g., for pronouns, he
or she or singular they (often frowned on in high school grammar classes but used by
Shakespeare and Jane Austen along with many others) or for the species, homo sapiens
or humans.
Many resisted complaints about generic masculines, explaining that it was ‘just a
grammatical convention’—certainly, many English users have been drilled relentlessly
in the prescription to use he in contexts where women or girls are potential referents
since grammarian Ann Fisher formulated it in 1745 as the way to avoid using they
with a singular antecedent, a widespread usage she deplored. (You, which takes a plu-
ral verb form and was originally opposed to singular thou, did not arouse grammarians’
ire when used with singular reference, perhaps in part because the shift predated the
rise of prescriptive grammars aimed at foreigners and those with relatively little educa-
tion, including many women.) Robin Lakoff (1973, 1975), the first linguist to argue at
length that American women were linguistically disadvantaged, nonetheless thought he
unproblematic as did novelist Ursula LeGuin in The Left Hand of Darkness, her fantasy
novel envisioning a world where people’s sex might change often. Both Lakoff and
LeGuin have since shifted their initial positions as they have become persuaded that
this ‘convention’ is not quite what it seems and is closely connected to a more general
phenomenon of viewing (adult) men as canonical exemplars of humanity, women and
children being marked and somewhat deviant cases.

744
 LA N G U A G E, G EN DER, A N D S E X U A L I T Y

Unfortunately, switching to more inclusive forms of language—using they as a pro-

noun for antecedents that do not specify sex, e.g.—does not mean women are automati-
cally included. Even when apparently generic forms are used, empirical research finds
a tendency to forget about potential female referents except in stereotypically female
contexts (e.g., when the antecedent of they is nurse). Notice what is going on with what
Treichler and Frank 1989 dub ‘false generics’, which were first noted though not so
labeled by Black and Coward 1981.

vi. We awoke in the morning to find that the villagers had all paddled away in their
canoes, leaving us behind with the women and children.
vii. The pioneers and their wives were often ill-equipped to deal with the bone-chilling
winds and the heavy snows they experienced once winter began.

Words like villager and pioneer certainly do not seem to have masculinity as part of their
semantic content and yet apparently writers can presume that the women and children
do not ‘really’ count as potential referents.
Philosopher Elizabeth Beardsley (1976) argued that ‘genderization’ is the crux of the
problem, by which she meant specifying sex when it ought to be irrelevant as in poet-
ess or woman doctor. As Beardsley correctly noted, such specifications tend, by familiar
Gricean principles, to suggest that a poet’s or a doctor’s being female is noteworthy
(whereas being male is expected)—and sometimes as well that the poet or doctor who is
female is somehow not quite ‘up to’ the standard set by men. Unfortunately, matters are
more complicated than we might at first think, given that supposedly inclusive forms
like poet and doctor also tend to suggest men unless something is done to forestall those
suggestions—e.g., talking about Emily Dickinson or Virginia Apgar (the physician who
devised what is now known as the Apgar Test, a quick assessment of infant health used
in American delivery rooms). These complications do not mean that it is unimportant
to avoid he/man language, but they do mean that there is no simple strategy for both
including women and also avoiding irrelevant marking of gender. There have to be
multiple approaches and contexts matter in decisions over which approach to take for
a particular utterance.
Part of the problem in understanding such phenomena is that what is ‘brought to
mind linguistically’ includes more than what is covered by the familiar Gricean trio of
expression meaning, utterance meaning, and speaker meaning. As they process words
uttered, hearers regularly make inferences and build cognitive representations, often
not at readily accessible conscious levels, that contribute to later discourse in the same
way as content added to the common ground on the basis of understanding meaning
that has been linguistically conveyed or communicated by implicature.

2 ‘Hearer Meaning’: Significance Beyond Meaning and Implication

On the model of expression, utterance, and speaker meaning, I will use Hearer Meaning
to include other stuff that linguistic practices may ‘bring to mind’. Conceptual Baggage
covers inferences triggered for hearers by the situated use of particular words. Emer-
gent Significance is more global, covering what hearers glean, often with little or no
direct awareness, on the basis of significant patterns of utterances across many differ-
ent discourses (or perhaps sometimes, significant patterns over the history of a single
exchange). Emergent significance arises from processing what others have said—i.e.,

745
 SA LLY M CCO N N E L L - G I N E T

develops in a hearer’s mind—but this kind of ‘hearer’s meaning’ may manifest itself
when that individual speaks through what s/he presupposes, takes for granted.

2.1 Conceptual Baggage

Back in the early 1970s some feminists told a story along these lines.

A young boy was rushed to the hospital from the scene of an accident in which
his father was killed and prepped for emergency surgery. The surgeon walked in,
took one look, and said “I can’t operate on him—that’s my son.”

Many people, including many self-identified feminists, were puzzled. How could the
boy be the surgeon’s son when his father had been killed? They were assuming that the
surgeon was male and overlooking the possibility that the surgeon was the boy’s mother.
Though recognizing the existence of a stereotype that surgeons are male (perhaps less
robust now than when the story was first told but by no means completely moribund),
semanticists tended to cringe when people took the story to illustrate that surgeon is
semantically ‘male’.
I just recently noticed other assumptions, which are associated with father—people
also don’t tend to entertain the possibility that the boy had two fathers, which can
happen a number of ways, including having two gay fathers. It’s not the definite his
that triggers this ignoring of the possibility of two fathers as can be seen by substituting
brother for father. Just as surgeon triggers assumptions about the maleness of that profes-
sion, so father triggers assumptions about the canonical heterosexual family in which a
child only has one father. (There can, of course, also be multiple fathers if a child has a
stepfather or an adoptive father and also a biological father.)
What McConnell-Ginet 2008 proposed is that words have associated with them cog-
nitively not only information about their linguistic properties, including their conven-
tional meaning, but also further conceptual baggage. This would include stuff like the
stereotypes mentioned in Putnam 1973 as well as frames and scripts (essentially, as Fill-
more 1982 points out, frames that include temporal structure). Cognitive and computa-
tional linguists find such notions critical for modeling language use and understanding:
WordNet, e.g., really builds on frame analyses. And in thinking about many examples
important for understanding issues of the representation of gender and sexuality, we
find conceptual baggage lurking in the background. Although conceptual baggage is
not part of what the words conventionally mean or, usually, of what the speaker means
in uttering the words, it can be essential for understanding how what is said using those
words contributes to the inferences interlocutors draw. At least some times, conceptual
baggage may get loaded onto a word because of the emergent significance of discursive
patterns in which the word figures.

2.2 Emergent Significance

Discussions of language, gender, and sexuality often talk about the emergent signifi-
cance of discursive patterning, although not in those terms. Analysts draw inferences
not from a single utterance or even a single situated discourse but from widespread sta-
tistical patterns across many different discourses. For example, children’s books, syntax
texts, newspapers, and many other media include many more references to men and

746
 LA N G U A G E, G EN DER, A N D S E X U A L I T Y

boys than to women and girls. (See Macaulay & Brice 1994 for the syntax texts and
references to other domains; Gill 2007 is an excellent general discussion of media and
gender, with many useful mentions of representational studies.) Notice that no particu-
lar utterer or utterance need have ‘meant’ that male human beings are more important
than female or even more interesting or less problematic to discuss nor does anyone
have to embrace such beliefs explicitly. Indeed, many people who themselves contrib-
ute to these patterns might be dismayed to realize that they have done so. I speak from
experience: the example sentences in my dissertation featured John and Bill, whom I
thanked for their help in my final paragraph. Why not Joan or Mary? I suspect that, at
some subliminal level, I thought they would be obtrusive, would not serve my need for
‘ordinary’ people, given their ‘special’ or ‘marked’ status as women. I would never have
endorsed the explicit claim that only men were ‘ordinary’ people. Nonetheless, I fell in
unthinkingly with discursive patterns giving undue representation to men, patterns that
are very likely part of what raises the visibility of men as villagers, pioneers, and the like
and decreases women’s visibility overall.
Frequency of appearance is only part of the picture: males are much more often than
females represented as agents, as doers, whereas females, when present, often appear
as patients, recipients of others’ actions. Even when speaking of women prominent
in public arenas, there are still frequent comments on their appearance, their marital
and parental status, and the like, whereas talk of men in similar circumstances far less
often includes such personal details in the absence of their special relevance. (See again
Macaulay & Brice 1994 as well as Gill 2007 for references.) Things have changed since
the dark ages when I was writing my dissertation. Feminism and activism around issues
of sexuality have had an impact: media representations these days often pay at least
lip service to gender equity, though often combining calls for ‘girl power’ with pic-
tures of hypersexualized 8- and 9-year-old girls or talk about women putting their own
needs first with injunctions to present oneself ‘attractively’ (along with ads for costly
‘beauty products’, clothes, and so on). It is also notable that men are often represented
in far from flattering ways—swilling beer, drooling over beautiful women, and generally
behaving like immature jerks. There is also far more use of knowing ‘winks’ and ironic
distancing than was once the case—‘yes, yes, we’ve heard feminist complaints and have
now moved on’. So discursive patterns these days are more mixed and less clear than
they once were, but the gender ideologies on which they draw and to which they con-
tribute are, to a considerable extent, still profoundly problematic.
Heteronormative presumptions can also be seen in contrasts between discourses.
Kitzinger 2002 reports that she and her female partner were regularly asked on entering
the dining room at a hotel where they were staying, “Are you expecting to be joined?” or
“Are you ladies alone?” Noticing this practice and not wanting to jump to conclusions,
Kitzinger stationed herself so as to be able to observe exchanges with other entering
guests: mixed-sex couples, she found, were shown to their tables without inquiries about
possible others. The ideological load of the entry question does not emerge if one simply
considers it in isolation but only by comparing its presence with the absence of such a
query to mixed-sex couples. What is standard for greeting guests at mealtimes is a wel-
come and questions like “would you like a window table?”—that general pattern is what
makes the questions to the pair of women about potential other tablemates significant.
Not only are presumptions of heterosexuality still quite widespread and often not
explicitly noticed. Views about what is standard in heterosexual interactions also often
continue surprisingly skewed. Although women’s sexual agency is more frequently

747
 SA LLY M CCO N N E L L - G I N E T

recognized than it once was, there is still a view of women as essentially passive in typical
heterosexual encounters, men initiating and sustaining contact. At the same time discur-
sive patterns of various kinds support the idea that this supposed passivity masks general
receptivity: women, such patterns suggest, generally want men’s sexual attentions but
they do not (and should not) solicit them actively (though they are often encouraged to
behave so as to interest men sexually but not to appear to be ‘doing’ anything purposive).
It is only against this kind of backdrop that one can make sense of widespread beliefs
that a woman who has not vocally and persistently said “no” has ‘implicitly consented’
to a man’s sexual advances even if one believes her claim that she was afraid of the man
and did not really welcome the sexual activity he initiated. Has a man who hands his
wallet to someone he believes can (and will) hurt him badly if he does not do so ‘con-
sented’ to being robbed? Or has a male prison inmate who lets himself be sodomized
when he thinks the alternative is being stabbed engaged in ‘consensual’ sexual activ-
ity? Susan Ehrlich, a Canadian discourse analyst, has published a number of detailed
and insightful studies examining transcripts from university sexual assault hearings and
criminal court proceedings on sexual assault or rape charges. Although neither Canada
nor the US any longer requires evidence of ‘utmost resistance’ to sustain a charge of
rape, Ehrlich 2002 shows patterns within particular hearings and across a number of
different hearings that seem to assume ‘implied consent’ by a woman who does not keep
repeating her objections to a sexual encounter.
Ehrlich 2007 goes further and examines differing judicial assessments of testimony
from a woman complainant in a Canadian sexual assault case. The 1995 trial judge and
the 1998 Alberta appeals court ruled in favor of the male defendant on the grounds that
the female complainant failed to be explicit and forceful enough in her objections to the
defendant’s actions. They did this while at the same time saying that they believed her
testimony that (1) she did not want the sexual activity to continue and that (2) she was
fearful of what might happen to her if she continued her protests. When the case was
appealed to the Supreme Court, however, the charge was upheld, with the judge writing
the 1999 decision articulating a very different conception of consent than that shown by
lower courts’ decisions. An extract from the decision (Ehrlich 2007, 270):
The question of implied consent should not have arisen. The trial judge’s con-
clusion that the complainant implicitly consented and that the Crown failed
to prove lack of consent was a fundamental error given that he found the com-
plainant credible, and accepted her evidence that she said ‘no’ on three occa-
sions and was afraid. This error does not derive from the findings of fact but from
mythical assumptions. It denies women’s sexual autonomy and implies that women are
in a state of constant consent to sexual activity. The majority of the Court of Appeal
also relied on inappropriate myths and stereotypes. (italics added by Ehrlich)
The myths and stereotypes at issue are, I suspect, still circulated, not so much by explicit
representations of women as ‘always wanting it’ but by the emergent significance of
discursive patterns implicitly presenting women’s sexual passivity as a norm and their
reluctance to admit to their own desires a common phenomenon.
Roger Dworkin (1966) quotes various ‘experts’ in arguing that ‘consent’ is virtually
impossible to establish in cases of allegations of sexual assault or rape (and that some
standard of ‘resistance’ is needed to establish lack of consent). He cites approvingly (p.
682) the following claim: “Although a woman may desire sexual intercourse it is custom-
ary for her to say ‘no, no, no’ (although meaning ‘yes, yes, yes’) and to expect the man

748
 LA N G U A G E, G EN DER, A N D S E X U A L I T Y

to be the aggressor.” (italics added) Of course occasionally a woman might say ‘no’ and
mean ‘yes’, but viewing this as ‘customary’ even in 1966 (when it could well have hap-
pened somewhat more often than it now would) showed more attention to myth than to
realities. Dworkin even speaks of ‘unconscious consent’, a nearly incoherent notion
Now Dworkin was writing over four decades ago. Few law review authors would now
put matters as he did. The Canadian trial and appeals court decisions, however, are
much more recent. And they are not isolated examples of views of women’s sexual pas-
sivity and ‘coyness’ as normative but fit into wider patterns, which we will see in the
next section can constrain women’s effectiveness as speakers.

3 Speech Act Effectiveness

Effectiveness is a matter of how a particular speech act is received, what changes it
brings about in the ongoing course of affairs, including the conversational record of
what interactants are mutually assuming as well as attitudes and actions that might
result. If I issue a directive to you, do you undertake to comply, take on the indicated
obligation? If I make an assertion, do you add its content to your belief set or at least
strengthen your confidence in the proposition I convey? If I make a joke, do you laugh?
If I ask a question, do you take on the job of trying to answer it? If I make a suggestion,
do you consider it seriously? And so on.
In trying to “do things with words,” some are more consistently successful than oth-
ers. Success depends on many factors but the relation of the interlocutors, including
their views of one another, is extremely important. The response or uptake to an utter-
ance is critical; immediate and long-term effects on others must be considered. Grice
1957 initially defined what is was for speaker S to mean that p to hearer H in terms of S’s
intending to produce in H the belief that p—i.e., to produce a certain kind of response.
Similarly, Austin in his 1960 account of illocutionary acts called for a certain kind of
‘uptake’ on the hearer’s part: in order for an assertion to come off successfully, the hearer
need not come to believe what the speaker asserts but must at a minimum recognize that
the speaker is trying to bring it about that the hearer so believe.
Both Grice and Austin made success in meaning or in performing illocutionary acts
depend on others than the speaker. Many critics noted that one could not ‘intend’ to do
what one did not believe one could do and yet I might well mean that p to you even if
I have no expectation that you could be led to believe p. So far as it goes, this criticism
has some merit, but it does not really come to grips with the fact that semantic ‘success’
is not a purely individualistic matter: others play an important role. Although both the
Gricean and the Austinian account do have problems, they are not just confused as
some have assumed. A speaker whose attempts to mean are impeded by others—e.g.,
by others who simply refuse to believe that she means what she says when she rejects
a man’s sexual advances—is semantically handicapped, illocutionarily constrained. I
made this point in McConnell-Ginet 1989, building on a related observation made in
an unpublished talk I’d heard philosopher Sara Ketchum make in the early 1980s.
Philosophers Jennifer Hornsby and Rae Langton make a similar point when they
argue (in, e.g., Hornsby & Langton 1998 and elsewhere) that the notion of ‘speech’
that is involved in advocacy of ‘freedom of speech’ is illocutionary. They suggest that
speakers’ freedom of speech is curtailed if they cannot get their communicative inten-
tions recognized, cannot fully perform certain illocutionary acts because they cannot get
hearers to recognize certain illocutionary intents. This point is developed in the context

749
 SA LLY M CCO N N E L L - G I N E T

of arguing that pornography tends to produce a communicative climate in which wom-

en’s attempts to refuse men’s sexual advances are simply not recognized as such because
it is assumed women’s putative refusals are just a sham and that the women saying ‘no’
are really communicating ‘yes but you should keep on insisting’. I disagree with Hornsby
and Langton that pornography is the main purveyor of this message. As I suggested
in the preceding section, this view of ‘normative’ heterosexual encounters is implicit
across a range of discourses, including legal ones. The explicitness of pornography may,
in my view, render it less powerful in fostering and sustaining such pernicious views
than more respectable kinds of discourse in which these ideas are only implicit (at least
in recent years).
But the fundamental point is independent of whether pornography really does have
the effects Hornsby and Langton posit. Communicative acts are not purely individual-
istic—at a minimum their success requires recognition by the hearer. And what seems
clear is that recognition may be impeded by gender and sexual ideologies that the hearer
draws on in assessing the communicative intentions of particular speakers. Without
appropriate uptake from the hearer—recognition of the speaker’s illocutionary intent—
the speech act is doomed.
Of course, effectiveness requires more. When I make a sincere assertion I want you to
believe what I have said, a directive is issued with an interest in the hearer’s doing what
it enjoins. Formal theories of discourse usually incorporate these desired outcomes in
their models of the impact of utterances on developing conversation: e.g., an assertion
adds the propositional content asserted to the conversational record, a directive places
the requested hearer action on the hearer’s “to-do” list, etc. (See, e.g., Roberts 2004 for
discussion.) In other words, the hearer’s Updating the conversational record or con-
text as intended is fundamental to the speaker’s illocutionary goals. Dynamic semantics
does not assign propositional content to declarative sentences but rather their “con-
text-change potential.” Assuming a possible worlds account, a declarative sentence has
the potential to narrow down the set of worlds still considered ‘live options’ (thinking
propositionally, the assertion can add to the stock of propositions taken for granted) in
the discourse context. Whether this potential is actually realized when a sentence is
uttered depends on social factors operative in the particular discourse, among which the
hearer’s relation to the speaker, including his views of her interests and abilities, may
often play a major role.
It is sometimes said that the desired outcomes involving, e.g., hearers’ beliefs or
intentions to act are not part of what a speaker means or of the illocutionary force of
her utterance but are rather further perlocutionary effects of the same kind as impress-
ing or frightening or surprising. But the kinds of outcomes we are discussing are far
more intimately tied to meaning as is evident from their role in dynamic semantics:
speakers aim (or at least hope) to update the discursive context in certain standard
ways by their utterances. Their capacity to effect such updating, however, is socially
constrained.
So both uptake—recognition of what the speaker means—and appropriate updat-
ing—making the indicated changes in the discourse context—are essential for what I
will call Full Semantic Effectiveness. It is not only gender and sexuality that can limit
semantic effectiveness, of course, but it is in the context of thinking about issues like
women’s difficulties in refusing some men’s sexual advances (and also in some other
communicative attempts) that I have seen the inadequacy of treating meaning without
reference to hearers’ uptake and updating.

750
 LA N G U A G E, G EN DER, A N D S E X U A L I T Y

As noted above utterance effectiveness is not limited to semantic effectiveness: the

hearer might offer the required uptake (comprehension) and make the appropriate con-
textual updates yet further (perlocutionary) effects desirable to the speaker might not
ensue. A speaker offers a suggestion in a business meeting and it is indeed adopted but
the higher-ups adopting it attribute it to a male colleague: the speaker is not credited.
Apportioning credit (or blame) for discursive moves requires keeping accurate tabs on
who has said what and folks are not always good at that, especially when what was said
(including the uptake and update) might not have been expected from that particular
speaker. And of course further aims can be frustrated in many other ways as achieving
them usually involves speaker’s and hearer’s assessments of one another’s immediate
situations and longer-term interests and goals as well as of factors independent of dis-
course participants. But the first essential step is semantic effectiveness, both uptake
and updating. Securing the intended uptake is crucial for illocutionary agency—one
must be understood—but if uptake goes well but the intended updating does not occur,
the illocutionary act is seriously deficient.

4 Conclusion
I have largely ignored discussion of the relation of linguistic variation (different ways
of saying ‘the same thing’) to gender and sexual identities. And I have also not consid-
ered such matters as conversational dynamics, amount of talk in public contexts, and
other staples of the literature on language, gender, and sexuality. My emphasis has been
on the explicit content of what people say and write, hear and read, and the implicit
content speakers communicate as well as that generated by hearers. I have also pointed
to the profoundly social character of “doing things with words” that becomes apparent
when looking at the interaction of language with gender and sexuality.
Note: Section 3 draws heavily from “Linguistics and gender studies,” my contribution
to the forthcoming Handbook of the Philosophy of Linguistics, ed. R. Kempson, N. Asher,
and T. Fernando in the Philosophy of Science series published by Elsevier under the
general editorship of D. Gabbay, P. Thangard, and J. Woods.

References
Anderson, Luvell and Ernest Lepore. Forthcoming. Slurring words. Noûs.
Beardsley, Elizabeth Lane. 1976 Referential Genderisation. In Carol C. Gould and Max W. Wartofsky, eds.,
Women and Philosophy (N.Y.: G.P. Putnam’s Sons), 285–93.
Black, Maria and Rosalind Coward. 1981. Linguistic, social and sexual relations. Screen Education 39:111–
133. Rpt. in abridged form in D. Cameron, ed., The feminist critique of language: A reader (Routledge,
1998).
Ehrlich, Susan. 2001. Representing rape: Language and sexual consent. London and New York: Routledge.
Ehrlich, Susan. 2007. Legal discourse and the cultural intelligibility of gendered meanings. Journal of Socio-
linguistics 11.4: 452–477.
Fillmore, Charles J. 1982. Frame semantics. In In-Seok Yang (ed.), Linguistics in the morning calm (Seoul:
Hanshin for Linguistic Society of Korea), 111–137.
Frank, Francine Wattman and Treichler, Paula A. 1989. Language, gender, and professional writing: Theoreti-
cal approaches and guidelines for nonsexist usage. New York: Modern Language Association.
Kaplan, David. 1999. The meaning of ouch and oops: Explorations in the theory of meaning as use. Unpub-
lished UCLA ms, paper given at Cornell Conference on Theories of Context Dependency.
Kitzinger, Celia. 2002. Doing feminist conversation analysis. In Paul McIlvenny, ed., Talking gender and
sexuality (Amsterdam/Philadelphia: John Benjamins, 2002), 49–78.
Kratzer, Angelika. 1999. Beyond ouch and oops: How descriptive and expressive meaning interact. URL

751
 SA LLY M CCO N N E L L - G I N E T

http://semanticsarchive.net/Archive/WEwNGUyO/, A comment on David Kaplan’s paper, Cornell Con-

ference on Theories of Context Dependency.
Lakoff, Robin. 1975. Language and woman’s place. NY: Harper & Row.
Lakoff, Robin Tolmach, ed. Mary Bucholtz. 2004. Language and woman’s place: Text and commentaries.
Revised and expanded edition of Lakoff 1975, with Lakoff’s original text and commentaries by her and
others. New York: Oxford University Press.
LeGuin, Ursula K. 1989. Dancing at the edge of the world: Thoughts on words, women, places. New York: Grove
Press.
McConnell-Ginet, Sally. 1979. Prototypes, pronouns and persons. In M. Mathiot, ed., Ethnolinguistics: Boas,
Sapir and Whorf revisited (Mouton), 63–83. Rpt. w/ updated notes and references in McConnell-Ginet,
2011.
McConnell-Ginet, Sally. 1989. The sexual reproduction of meaning: A discourse-based theory. In Frank and
Treichler 1989, 35–50. Rpt. in McConnell-Ginet 2011.
McConnell-Ginet, S. 2008 Words in the world: How and why meanings can matter. Language 84.3,
497–527
McConnell-Ginet, S. 2011. Gender, sexuality, and meaning: Linguistic practice and politics. Oxford: Oxford
University Press.
Treichler, Paula and Francine Wattman Frank. 1989. Guidelines for nonsexist usage. In Frank & Treichler,
eds., 137–278.
Potts, Christopher. 2007. The expressive dimension. Theoretical Linguistics 33(2):165–197.

752
 6.4.2
Language and Race
Rae Langton, Sally Haslanger, and
Luvell Anderson

Introduction
What is the point of language? If we begin with that abstract question, we may be tempted
towards a high-minded answer: “People say things to get other people to come to know
things that they didn’t know before” (Stalnaker, 2002, 703). The point is truth, knowl-
edge, communication. If we begin with a concrete question, “What has language to do
with race?” we find a different point: to attack, spread hatred, create racial hierarchy. The
mere practice of racial categorization is controversial: are race terms natural or social kind
terms? What does categorization do to the categorized? (Outlaw, 1990; Omi and Winant,
1994; Appiah, 1996; Andreason, 1998; Kitcher, 1999; Zack, 2002; Haslanger, 2008;
Glasgow, 2009). But there is worse than mere categorization to contend with.
An Alabama court says slaves are not really people: as “rational beings, they are capa-
ble of committing crimes; and in reference to acts which are crimes, are regarded as per-
sons,” but nonetheless “because they are slaves, they are incapable of performing civil
acts, and, in reference to all such, they are things, not persons” (Caterall, 1926, 247). A
restaurant proprietor places a sign, “Whites Only,” using language to discriminate, and
help enact segregation. A Nazi publisher says:

If we do not oppose the Jews with the entire energy of our people, we are lost.
But if we can use the full force of our soul that has been released by the National
Socialist revolution, we need not fear the future. The devilish hatred of the
Jews plunged the world into war, need and misery. Our holy hate will bring us
victory and save all of mankind.
(Streicher, 1943)

White fans of Babe Ruth write hate mail to rising competitor Hank Aaron, using
language to assault, insult, persecute, and threaten (Kennedy, 2003, p. 20):

Dear Nigger Henry, You are [not] going to break this record [for home runs]
established by the great Babe Ruth if you can help it . . . Whites are far more
superior than jungle bunnies . . . My gun is watching your every black move.

Language can be used to express racism, to attack people, to help constitute and cause
racial oppression.
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

There may be implications for law and policy: think of the striking down of Alabama
slave codes; the outlawing of the “Whites Only” sign; the execution of the Nazi propa-
gandist; the Congressional condemnation of the “racial slurs and insults” directed at
Aaron (Rosengren, 2008, 119). There may be implications for epistemology: Think of
the racist perception of accent and idiolect, contributing to an unjust distribution of
credibility (Fricker, 2007); the harnessing of Biblical language to envision escape from
“captivity”, to a “promised land” where “justice rolls down like waters and righteousness
like a mighty stream” (King, 1963). Our topic is large, and fiercely discussed by politi-
cal philosophers, critical race theorists, and philosophers of law. But we shall confine
our focus to three issues in philosophy of language: slurs, or epithets; speech acts of racial
oppression; and racial generics.

1 Slurs
Slurs or epithets, like those in the letters to Aaron, are expressions that function to
demean or derogate people in virtue of their group membership. They derogate classes,
in contrast to more individualized insults like “moron” or “idiot.” Slurs generally have
neutral counterparts; what can be said with a slur can often be said another way with-
out offense. An adequate theory of slurs should account for the variation in offensive
potency among slurs: “nigger” is more offensive than “cracker”; “kite” is more offensive
than “ofay”; “spic” is more offensive than “honkey”. There can be variation among slurs
for the same group: “nigger” is more offensive than “darkie” or “coon.” Besides deroga-
tory uses, slurs are sometimes appropriated by target members in ways that mitigate or
mollify a slur’s offensiveness—for example the use of “nigga” by some African Ameri-
cans (Kennedy, 1999/2000, 2002).
Slurs provoke important questions in the philosophy of language, though only rela-
tively recently have they begun to receive the attention they deserve (Dummett, 1973;
Kaplan, 1999; Hornsby, 2001; Williamson, 2007; Hom, 2008; Richard, 2008; Anderson
and Lepore, forthcoming). Does the derogatory aspect of slurring statements affect their
truth conditions? Where is the offensiveness of slurs located? Do slurs possess expres-
sive content as well as descriptive content? How do slurs and the semantic/pragmatic
distinction bear on each other? Though we cannot do justice to all of these questions,
we shall take a look at some semantic and pragmatic approaches to the truth-value of
slurring sentences, with their corresponding accounts of the location of a slur’s offense.
These approaches locate the offense in the content, whether semantically or prag-
matically conveyed. But we shall finally consider a non-content-based strategy, which
locates the offensiveness of slurs somewhere else entirely—namely, in the fact that they
are socially prohibited.
A first, simple proposal is a semantic one. Slurs have a different literal content from
their neutral counterparts:

(1) Hank Aaron is an African American.

(2) Hank Aaron is a nigger.

On this proposal, the occurrence of the slur in (2) literally expresses more than “African
American” in (1): perhaps “African American and despicable because of it”. On this
understanding, (1) is true, and (2) is simply false. And the additional, derogatory con-
tent in (2) helps explain its offense.

754
 LA N G U A G E A N D R A C E

However there are reasons to think this proposal is not adequate. Suppose someone
wants to deny whatever is ascribed to Aaron in (1). They can simply say

(3) Hank Aaron is not an African American.

(3) denies Aaron is African American: it is false, and inoffensive. Suppose someone
wants to deny whatever is ascribed to Aaron in (2). They say:

(4) Hank Aaron is not a nigger.

On the semantic proposal, (4) denies that Aaron is African-American and despicable
because of it: so it should be both inoffensive and true. But it is not true, or not obviously
true (Anderson and Lepore, forthcoming, p. 6); and (4) is, moreover, offensive. This tells
against the proposal. (Note there are theorists who disagree, saying the negation of a slur-
ring sentence needn’t be derogatory: Dummett, 1973; Hornsby, 2001; Hom, 2008.)
Considerations from indirect speech also undermine the semantic approach. Sup-
pose John says “Aaron is African-American”; and Mark reports him by saying: “John
said that Aaron is an African American”. No problem: this is true, inoffensive, and
follows the general guiding principle of reported speech that one should re-use expres-
sions in reporting their first uses (Anderson and Lepore, forthcoming, p. 8). Suppose,
though, Jim says “Aaron is a nigger”: and Mark reports him by saying, “Jim said that
Aaron is a nigger”. That is just what Mark should say, on the semantic approach. Now,
Jim said something offensive (and false, on the semantic approach): but Mark’s report
of Jim (whatever its truth value) is also offensive. And, interestingly, if we had nothing
but Mark’s report to go by, we might guess that Mark was a racist slurrer, and that Jim
was not. A slurring indirect report may or may lay offense at the reportee’s door: but it
certainly lays offense at the reporter’s door. We have some reasons, then, to doubt the
semantic approach.
As an alternative, one might propose a presuppositional account of slurs. The behavior
of slurs under negation, displayed above, provides a starting point here, since presup-
positions seem to project through negation, as shown in the (unfortunate) classic (5)
and (6):

(5) John stopped beating his wife.

(6) It’s not the case that John stopped beating his wife.

Both statements presuppose “John beat his wife”. A presuppositional approach might
explain the behavior of slurs under denial and negation, since presuppositions project
out of those contexts; and why occurrences of slurs in interrogative statements also tend
to cause offense, as in—

(7) Are most landscapers spics?

—because presuppositions project out of interrogative contexts as well.

On the presuppositional proposal, the slur doesn’t add the offensive content as part
of the explicit truth conditions, but instead presupposes it. To return to our original
examples, whilst (1) is true, (2) suffers from a presuppositional failure (Aaron is not
despicable because of being African American). It therefore lacks a truth value, on a

755
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

Fregean understanding of presupposition; or it suffers whatever other shortcoming one

takes presuppositional failure to give rise to. And on this account, (2)’s offensiveness is
located in its presupposition.
However, indirect reporting presents a problem for this approach as well. Presup-
positions and slurs behave differently in the complement clauses of “say”: the presup-
positionally encoded content of a complement clause following “say” is contained or
isolated within the scope of the verb. But the offense of a slur in the complement clause
projects out of the scope of “say”, so that the offense is carried by the entire sentence.
Notice the difference between (8) and (9):

(8) John said that Bill stopped beating his wife. But Bill never beat his wife.
(9) Maury said that spics are taking over Arizona. But Hispanics are not despi-
cable because of being Hispanic.

Normally, “Bill stopped beating his wife” presupposes “Bill beat his wife”. But in (8), the
speaker does not present the latter as part of the common ground (Stalnaker, 2002): it’s
not assumed that hearers go along with it. “Bill beat his wife” is contained or isolated
within the scope of “say”. In (9), however, the offensiveness of “spic” is not contained
or isolated within the scope of “say”. If the offense were communicated presupposition-
ally, it wouldn’t spread to the whole attribution: the reporter could explicitly distance
himself from the reportee’s presupposition, as in (8). But in the case of a slur, he can’t.
There is something infelicitous about following up the first statement in (9) with the
second. So we have reason to doubt the presuppositional account.
A third strategy denies that the offensive content of slurs affects either their truth
conditions or their presuppositions. A model here is Frege’s understanding of “tone”.
For Frege, “this is my steed” and “this is my nag” have the same truth conditions and
presuppositions, the difference between the two assertions being located elsewhere. One
way of making this idea precise would draw on Grice’s notion of conventional implicature
(Grice, 1961, 1989). Consider:

(10) John is British and brave.

(11) John is British but brave.

For Grice, “but” and “and” are truth-conditionally equivalent, though “but” carries a
notion of contrast that “and” lacks. The contrast is not a part of the asserted content of
“but”; (11) is true as long as John is both British and brave. On this proposal, one might
suggest that the difference between (1) and (2) is that “nigger” conventionally impli-
cates something that “African American” does not. On this account, both are true, and
the offense comes from the implicature. Since a conventional implicature cannot be
cancelled, this explains why a speaker cannot felicitously follow up an utterance of (2)
with, “But I don’t mean to suggest anything offensive about Aaron!”, any more than
a speaker could follow up (11) with an utterance of—“and there’s nothing surprising
about a Briton being brave!”. The proposal helps explain the behavior of slurs under
denial and negation, since these do not affect the conventional implicature.
Despite its virtues, there are reasons to question the conventional implicature pro-
posal. Numerous authors have expressed suspicion about the whole notion of conven-
tional implicature (Bach, 1999; Carston, 2002, 2004). Putting that aside, this proposal
does no better with indirect speech reports than does the presuppositional account.

756
 LA N G U A G E A N D R A C E

Conventional implicatures are contained or isolated within the “says” complement just
as much as presuppositions are. Consider:

(12) Pierre said that John is British but brave. However, there’s nothing sur-
prising about Britons being brave!

The speaker who utters (12) successfully insulates himself from anti-British prejudice;
but the speaker of (9) does not successfully insulate himself from offensiveness towards
Hispanics. So we must have reservations about this approach as well.
We have seen that there are problems with a semantic truth-conditional strategy, a
presuppositional strategy, and a strategy drawing on conventional implicature. There
is much more to be said, and there are arguments to which we cannot do justice here.
(For vigorous defense of a semantic truth-conditional approach see Hom, 2008; for
some important expressivist alternatives, see Kaplan, 1999; Richard, 2008.) But notice
the three content-based approaches considered here also have a problem in common: a
difficulty in accounting for the data with which we began, namely the varying degrees of
offensiveness among slurs. What relevant difference in content—whether explicit, pre-
supposed, or implicated—could account for the different degrees of offensive potency
among slurs for different groups, or among co-referring slurs?
Given these difficulties, it may be worth looking in a completely different place
for an understanding of the offensiveness of slurs: it may be worth looking at a non-con-
tent-based approach. Here is a simple alternative: slurs are prohibited words whose occur-
rences are offensive (Anderson and Lepore, forthcoming). This is a proposal not about
the truth values of slurring sentences (being in principle compatible with a number of
approaches to that question): rather, it is about the source of a slur’s offense. Prohibition-
ism helps explain why offense projects out of the complement of “say” and quotation,
why it is not challenged by denial, and why the speaker of a slurred word bears the
offense, even in indirect reports. It helps explain the phenomenon of appropriation.
The prohibition on a slur is not absolute. Like oil embargos or economic sanctions,
prohibitions on slur use have exceptions, so that members of the targeted group may
use them; and so that discussion in certain contexts, including academic contexts like
this one, manage (we hope) to avoid offense. It helps explain how phonetic and ortho-
graphic accidents provoke offense, so that a word like “niggardly”, though not semanti-
cally or etymologically related to the slur, may nonetheless be found offensive because
the prohibition extends to words that sound like existing slurs (Kennedy, 2002). It
might also explain how a slur may sometimes offend, even when insulated in quotation
marks. To sum up then: whatever one’s story about the content of slurs, whether seman-
tically or pragmatically conveyed, that story has difficulty accounting for the offensive
potency of the slur. The prohibitionist alternative offers an explanation worth taking
seriously.

2 Speech Acts
Language using slurs and epithets, like the letter to Aaron, fits Mari Matsuda’s descrip-
tion of hate speech, whose “message is of racial inferiority”, is “directed against a histori-
cally oppressed group”, and is “persecutory, hateful and degrading” (1993, 36). Matsuda
sees hate speech as a kind of attack, and it includes, of course, speech that does not use
slurs or epithets. Notwithstanding her talk of a “message”, she is less interested in the

757
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

content of the words than in what is being done with these words (cf. Austin, 1962):
hate speech is a speech act. The point of slurs or epithets, from this perspective, is not
their content. But neither is it the simple fact that they are prohibited. It is that slurs
and epithets are apt for performing certain speech acts: as Richard says, “what makes a
word a slur is that it is used to do certain things, that it has . . . a certain illocutionary
potential” (Richard, 2008, 1)—the potential to enable speakers to do things that e.g.
“persecute” and “degrade’.
A speech act perspective is implicit in a United Nations condemnation of “propa-
ganda” based on “ideas or theories of superiority of one race”, that “attempts to justify
or promote racial hatred and discrimination”: speech that is an “incitement to, or act
of, discrimination” or violence (UN, 1965). The UN focuses on speech addressed to
prospective haters, rather than to targets of hate: hate speech incites and promotes racial
discrimination, hatred and violence. Matsuda is interested in assault-like speech, the
UN in propaganda-like speech, but both see hate speech as something that not only
has certain effects (causing distress, spreading hatred), but also is something, in itself.
In Austin’s terms, both are talking about, inter alia, the “illocutionary” aspect of hate
speech.
Austin distinguished the illocutionary aspect of an utterance from two others: the
meaning, content, or “locutionary” act; and the effects on hearers and beyond, or “per-
locutionary” act. Suppose (Austin’s example) a speaker says “Shoot her!” There is con-
tent, a locutionary act: he said “Shoot” meaning “shoot with a gun”, and “her”, refer-
ring to the woman nearby. There are effects, a perlocutionary act: by saying “Shoot
her!” he got his hearer to shoot the woman. But there is more: in saying “Shoot her”,
Smith urged the hearer to shoot. This is the illocutionary act, performed “in saying”
those words.
In Austin’s terms, hate speech can be an illocution that persecutes and degrades; or
incites or promotes hatred, discrimination, violence. It is typically a more ordinary illo-
cution as well: perhaps it asserts that there is a Jewish conspiracy, that hate will bring
victory, that whites are superior; orders blacks to keep away. And it is typically a perlo-
cutionary act that achieves effects, including changes in hearers” beliefs, emotions, and
actions. But its illocutionary aspect is distinct from its locutionary content, and from its
perlocutionary effects. Austin’s distinction between illocutionary and perlocutionary
acts offers a way to distinguish speech that constitutes racial oppression, and speech that
causes racial oppression.
Hate speech is a kind of racially oppressive speech: letters “persecute” and “degrade”
Aaron with assault-like hate speech; Nazi editorials “incite” and “promote” hatred
against Jews with propaganda-like hate speech. But there may be other kinds of racially
oppressive speech: a court says slaves are “incapable of performing civil acts”, are
“things, not persons”; a proprietor says “Whites Only”. Speech like this is not, or not
solely, assault-like or propaganda-like, nor does the label “hate speech” capture the
core of what is done. Its point is to enact, or help enact, a system of racial oppression:
it authoritatively ranks a certain group as inferior, deprives them of powers and rights,
legitimates discrimination against them. Speech that does these things has, perhaps, the
illocutionary force of subordination (cf. Langton, 1993, 2009). Note that its speakers are
authoritative: a mere bystander who said “Whites Only” or “slaves . . . are things, not
persons” would be doing something quite different. This fits an Austinian paradigm for
illocutions that enact norms, or facts about permissibility: speaker authority is typically
a “felicity condition” for their illocutionary success.

758
 LA N G U A G E A N D R A C E

Drawing on Austin, we have sketched a distinction between three potentially over-

lapping classes of racially oppressive illocutions: assault-like hate speech, propaganda-
like hate speech, and authoritative subordinating speech. More recent developments in
pragmatics may shed further light on their workings.
Some have argued that a conversation is a game-like, rule-governed activity: play-
ers make moves, and the “score” or “common ground” of the conversation evolves in
response to those moves (Lewis, 1979; Stalnaker, 2002). This notion of adjustment
to “score” or “common ground” thus has significance not only in its role of tracking
content, described in the previous section, but also, potentially, in a role of tracking
illocution (Lewis, 1979). David Lewis compared conversation to a baseball game, view-
ing “score” as an abstract structure, consisting of propositions and norms, built up over
time, constraining and enabling later moves available to players. But unlike a baseball
game, conversational score tends to evolve in whatever way is required to make a move
count as “correct play”, following rules of accommodation (Lewis, 1979; Stalnaker, 2002;
Atlas, 2004; von Fintel, 2008).
Presupposition introduction and Austinian performatives provide illustrations, for
Lewis. If a speaker says, “Even George could win”, then “straightway” the score or com-
mon ground is adjusted: she adds to the score not just “George could win”, but also
“George is an unpromising candidate”. If an authoritative speaker says, “You are not
permitted to cross the white line”, then “straightway” the score is adjusted, and the
bounds of permissibility are altered. In similar vein, Robert Stalnaker describes how
the “common ground—the shared attitudes of conversational participants—evolves
in response to what is said, following these rules of accommodation. (Lewis’s picture
of “moves” in a “language game”, producing adjustments “straightway” to an abstract
“score”, exemplifies—arguably—Austin’s non-causal picture of an illocutionary act.
Stalnaker’s picture is similar, but subtly different. His picture of adjustments to “com-
mon ground” in the attitudes of hearers exemplifies—arguably—Austin’s causal picture
of a perlocutionary act (Langton, forthcoming).)
The importance of this is that it enables us to see how racially oppressive speech acts
might work in more informal contexts. Some kinds of racially oppressive speech acts
(a “Whites Only” sign, an Alabama court pronouncement) fit a ceremonial Austin-
ian paradigm, but many do not. Speakers often don’t or couldn’t pronounce, authori-
tatively and explicitly, “I hereby subordinate/persecute/discriminate”. Speakers often
don’t intend to perform the relevant speech act, or use a performative tag (“I hereby’),
or a relevant performative verb. They may lack formal authority, and may merely pre-
suppose or implicate certain claims or norms. But this is no bar to their speech acts being
illocutions that are racially oppressive (Langton and West, 1999; Langton, 2010). Thus
propaganda-like hate speech may not explicitly assert that “Good Germans hate Jews,”
may not use a performative tag or illocutionary verb (“I hereby authorize and urge you to
hate Jews”), but may still adjust the score, performing illocutions of authorizing and urg-
ing implicitly: so that it becomes appropriate for hearers to believe that good Germans
hate Jews, and to hate Jews. Likewise assault-like hate speech may merely presuppose or
implicate that its target is subhuman or animal-like, but the speaker may still succeed
in placing that in the score. A restaurant proprietor may enact discriminatory policy,
subtly placing it in the score, not by hanging a “Whites Only” sign but by encouraging
staff to favour customers of a certain race (“Look at Mr. Wilson there—just the sort of
customer we want”).
Mary-Kate McGowan extends Austin’s account on these lines, calling the relevant

759
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

phenomenon a “conversational exercitive” (2003, 2004). Focusing on the normative

aspect of “score” or “common ground”, she observes that conversational moves contrib-
uting to score are illocutions that alter normative facts about what is permissible, and
even possible, in the conversation thereafter. Conversational exercitives work differ-
ently to paradigm Austinian speech acts. They work in covert ways: Speakers needn’t
intend to alter permissibility facts, needn’t have special authority. In a conversation
following “Even George could win,” some moves are later impermissible (“Guess what,
George is an unpromising candidate!”), some are more permissible (for example mock-
ery of George). McGowan ultimately aims to explain how these local norms of a con-
versation hook up to wider social norms, so that enactment of a small racist local con-
versational norm can help constitute enactment of a larger racist social norm, set not
by the law-enacting speech of authorities but by everyday speech of conversational
speakers (2009).
One begins to see, then, how even an informal, conversational utterance could be an
illocutionary act that helps constitute racial oppression, adjusting “score” and setting
norms without being a grand ceremonial Austinian illocution; and how it could be a
perlocutionary act that helps cause racial oppression, by helping cause racist beliefs and
norms in its hearers, via adjustment to common ground.
The speech act perspective explored in this section provokes other questions. Per-
haps it yields a special sense of silencing: If someone is unable to perform certain illocu-
tions because of their race, might that be a kind of illocutionary disablement? (Langton,
1993) Perhaps a pragmatic story helps one see how rebellion is difficult, if blocking a pre-
supposition is costly: must a blocker be uncooperative, derailing the conversation, going
against “what we all take for granted’? (Langton and West, 1999) Perhaps the pragmatic
story itself needs amending: If we are interested in how speech is used to transmit, not
just knowledge, but hate, should “common ground” be extended to include attitudes
“beyond belief”, such as hate? (Langton, 2010) This look at racially oppressive speech
through the lens of speech act theory has been brief but still suggestive, we hope, of
fruitful possibilities.

3 Racial Generics
We have been considering utterances that are clearly racist: racial epithets or slurs,
assault-like hate speech, propaganda-like hate speech, and authoritative subordinating
speech. Now we turn to examples of speech that appear innocent, but may nevertheless
undergird racial subordination.
Consider statements such as:

(15) Latinos are lazy.

(16) Muslims are violent.
(17) Blacks are sexually insatiable.

Those who make these sorts of claims often insist they are simply reporting the facts;
however, counterexamples do not disrupt their confidence. Appiah, in considering this
phenomenon, argues that racial prejudice just is this sort of epistemic failure. Racial
prejudice is the disposition to hold beliefs—

that are to the disadvantage of some race (or races) as opposed to others, and to

760
 LA N G U A G E A N D R A C E

do so even in the face of evidence and argument that should appropriately lead
to giving those propositions up
(Appiah, 1990, 15–16)

However, things are not so simple. Consider the following utterances:

(18) Ducks lay eggs.

(19) Lions have manes.
(20) Mosquitoes carry the West Nile Virus.

All seem to be true, even though one could provide many counterexamples of ducks
that do not lay eggs (all male ducks), lions that don’t have manes (all female lions),
and mosquitoes that don’t carry the West Nile Virus (over 99 percent, see Leslie, 2007,
376).
The utterances in both groups are generalizations, but they do not have the form,
strictly speaking, of quantifications. Generalizations that omit quantifiers such as
“some”, “most”, or “all” fall into the linguistic category of generics, and call for a differ-
ent analysis than ordinary quantified statements. Importantly, for our purposes, generics
need not be undermined by a counterexample, or even many counterexamples.
Although there is no consensus on the best account of generics, Sarah-Jane Leslie
suggests convincingly that generics are the expression of a very primitive “default mode
of generalizing,” that picks up on significant or striking properties and links them to a
psychologically salient kind.
In exploring this, we must be sensitive to different sorts of generalizations. Prasada
and Dillingham (2006, 2009) highlight two sorts. Principled generalizations involve
properties determined by the kind of thing something is (for example having four legs,
for a dog). Statistical generalizations involve properties not determined by the kind of
thing something is but are highly prevalent connections to the kind (for example cars
having radios). Principled generalizations support formal explanations (Fido has four
legs because he is a dog), normative expectations (Fido should have four legs, is defec-
tive if not), and expectations of prevalence (most dogs have four legs) (Leslie et al.,
2009, 479).
Leslie notes that this distinction between principled and statistical generalizations,
although important, is not sufficient, by itself, to accommodate the range of different
generics, for, as noted above, there are true generics that do not ascribe a prevalent
property to the kind, such as:

(18) Ducks lay eggs.

Leslie proposes instead not two, but three, classes (Leslie, 2008, 43). First, there are
characteristic generics, which correspond roughly to Prasada and Dillingham’s “princi-
pled” generics. Cases such as

(21) Tigers have stripes.

and

(22) Dogs have four legs.

761
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

These assert more than a statistical correlation between tigers and stripes, or dogs and
four legs. They purport to tell us what is characteristic of the kind, or what a good exam-
ple of the kind will exhibit. And if we consider what is characteristic of the kind, rather
than of each individual, we can accommodate

(18) Ducks lay eggs.

even if the majority of ducks don’t lay eggs. We have background knowledge that—

provides an outline of information to be gathered about a new kind; charac-

teristic dimensions provide a learner with an informational template. When a
value is found for a characteristic dimension of a kind, it is hereby generalized
to the kind by the basic generalization mechanism, and so the generic that
predicates that property of the kind is accepted. Ducks, being an animal kind,
have reproduction as a characteristic dimension, so the inductive learner looks
for a value to fill the dimension; even limited experience will deliver laying
eggs as the appropriate value, and so the property is generalized to the kind and
“ducks lay eggs” is accepted as true.
(Leslie, 2008, 32–3)

Second, there are striking property generics. This extends Prasada and Dillingham’s “prin-
cipled” generics to cases where the property is not prevalent. To accommodate

(20) Mosquitoes carry the West Nile virus.

even though only a small fraction do, Leslie says: “if the property in question is the sort
of property of which one would be well served to be forewarned, even if there were only
a small chance of encountering it, then generic attributions of the property are intui-
tively true (2008, 15)”.
On her account, being a member of the kind must be a reasonably good predictor of
the striking property; members that don’t have the property must be disposed, under the
right circumstances, to have it (2008, 41).
Third, there are majority generics. This corresponds to Prasada and Dillingham’s “sta-
tistical” generics. Speakers are willing to assent to weaker generics such as

(23) Cars have radios.

(24) Barns are red.

that capture only statistical or enumerative generalizations. However speakers do find

these less natural than characteristic generics, and resist alternative syntactic forms for
the generic. Thus in the case of the indefinite singular form—

(25) A tiger has stripes.

—is considered more natural than

(26) A barn is red.

(See Leslie et al., 2009, 480, 482, 484.)

762
 LA N G U A G E A N D R A C E

Leslie’s account of generics is controversial, and more complex than suggested here
(see Leslie, 2008, 35). The message for our purposes, though, is simple enough. Gener-
ics are a distinctive kind of statement that should not be treated as ordinary quantified
statements: They draw heavily on background knowledge and patterns of inference to
highlight a significant property of a kind—a property which may be characteristic, strik-
ing, or just statistically correlated with the kind.
Leslie is right to complicate Prasada and Dillingham’s view by allowing for striking
property generics in which the property attributed is not prevalent among members
of the kind. However, the distinction between principled and statistical connections
still seems important. In principled connections there is something about being a
member of the kind as such that explains why those members that have the property
do so. Prasada and Dillingham’s mistake was to assume that when a principled connec-
tion holds between members of a kind and some property, that property is prevalent
among members of the kind. But it may not be prevalent because the circumstances
prevent members of the kind from manifesting the property, or because it is typical of
the kind that only certain members manifest it. The crucial fact is this: Those members
which do have the property are disposed to have it by virtue of the fact that they are
members of the kind. If one goes in for essentialist talk, one might say that having the
property is grounded in or flows from the nature or essence of the kind. Laying eggs
is grounded in the nature of ducks (being oviparous is part of what ducks are). Carry-
ing the West Nile Virus is something mosquitoes are disposed to do by virtue of being
mosquitoes. It does not follow, however, that all or most members of the kind have the
property.
If we adjust the idea of principled connections in this way, Leslie’s characteristic and
striking property generics both turn out to be subcategories of principled generics. Such
generics entail, or implicate, claims about natures or essences. Although the utterance
of a majority generic such as:

(23) Cars have radios.

does not imply that it flows from the nature of cars that my VW is disposed to have a
radio, the utterance of what appears to be a characteristic generic such as—

(15) Latinos are lazy.

—does imply that laziness flows from the nature of Latinos, as such.
This helps explain why some speakers refuse to withdraw racially stereotyping claims
when confronted with counterexamples. Even if many Latinos do not demonstrate lazi-
ness, that doesn’t show it isn’t part of their nature. They may just be “defective” Latinos;
they may have been denied the opportunity to express their nature (but it could happen
at any time!). In order to undermine the characteristic generic, one must go beyond
counterexamples and argue against the modal/explanatory claim.
Given the different generics that might be expressed by a bare plural, it is unsurpris-
ing that what they communicate is slippery. Two kinds of slip are highly relevant to the
issue of race: (i) a slippage between characteristic and striking property generics, and (ii)
a slippage between characteristic and majority generics.
To explore the first slippage between characteristic and striking property generics,
consider an ethnic generic such as:

763
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

(16) Muslims are violent.

It is tempting to regard (16) as a striking property generic which, like

(20) Mosquitoes carry the West Nile Virus.

is assertable based on a small number of instances. However, given that the same state-
ment form can also be used to assert a characteristic generic, there is an easy slide from
accepting (16) as a striking property generic to the conclusion that it is characteristic
of Muslims to be violent.

The basic idea is quite simple: just as it takes but a few instances of sharks attack-
ing bathers, or of mosquitoes carrying the West Nile virus, for us to make the
corresponding category-wide generalization, so also a strikingly negative action
on behalf of a few members of a racial minority may lead us to a general belief
concerning their entire racial category. A belief in the essentialized nature of
racial categories makes them appear to be appropriate loci of striking property
generalizations; our disposition to thoughtlessly generalize dangerous or harm-
ful information does the rest . . . Extreme and aberrant actions on behalf of the
few can thus lead to conclusions concerning all.
(Leslie, forthcoming)

Psychological research shows, moreover, that striking property generics are quickly accepted
based on few instances, whereas characteristic generics, once accepted, are resistant to refu-
tation. Given the slippage between these two forms, properties not indicative of a general
pattern get attributed to a group and then stick (Cimpian, Brandone, and Gelman, 2010).
Easy to accept, hard to refute is not an epistemically or politically promising combination.
The second kind of slippage occurs between characteristic and majority generics.
Without taking a stand on the semantics of generics, we offer the following prag-
matic proposal. The utterance of a generic that is not obviously just a majority generic
implicates that the property ascribed is grounded in the nature or essence of the kind
(Haslanger, 2010). The implicature can be canceled explicitly or implicitly by con-
text, but otherwise the implicature enters the common ground of the conversation
(Stalnaker, 2002).
Attending to this slippage between majority and characteristic readings helps us
explain certain shifts in the defense of generics. Consider again:

(15) Latinos are lazy.

Does (15) assert a majority generic or a characteristic generic? Interpret (15) as a major-
ity generic. To combat it, one provides many counterexamples. However, the speaker
can then suggest that, although many Latinos aren’t lazy, they tend to be—thus embrac-
ing the characteristic generic. Instead interpret (15) as a characteristic generic. To com-
bat it one provides evidence that, say, Latinos show no greater tendency towards lazi-
ness than any other group. The speaker can then suggest that, although it is not part of
the nature or essence of Latinos to be lazy, most are. This slide back and forth between
different interpretations of the utterance allows speakers to avoid taking responsibility
for the implications of their claims.

764
 LA N G U A G E A N D R A C E

What are the morals of the story? We suggest three.

First moral: If a characteristic racial generic is asserted, it should be rejected, because is

false, and also politically problematic; it presents social artifact as racial essence, and by
masking oppression is a barrier to social change. Besides masking, it also assists oppres-
sion via social looping effects: If individuals are made to become as they are assumed to
be by nature, then statistical generics are often true, and their truth seems to confirm
the faulty assumption we started with.

Second moral: If a striking property racial generic is asserted, it should also be rejected
because it is expressed in a statement form that is too readily interpretable as a charac-
teristic generic.

Third moral: If a statistical or majority racial generic is asserted it should be rejected

not only if it is false, but also, surprisingly, if it is true—as it might sometimes be, given
that majority generics are no stronger than a quantified “most” claim. A true statistical
or racial generic should, usually, be rejected metalinguistically (Horn, 1985) because it
implicates a characteristic generic. Metalinguistic rejection of the statistical or majority
generic, like the rejection of the characteristic generic, blocks the harmful invocation of
racial natures. A much better approach to true statistical generics is to insist on explic-
itly quantified statements instead (“Most . . .” or “Some . . .”) which avoid slippage to
the characteristic generic.

To sum up: generics may come naturally to us, but their assertion in racial discourse can
do much epistemic and moral harm.

Bibliography
Altman, A. 1993. “Liberalism and Campus Hate Speech: a Philosophical Reflection,” Ethics 103: 302–17.
Anderson, L. and Lepore, E. Forthcoming. “Slurring Words,” Noûs.
Andreasen, R. 1998. “A New Perspective on the Race Debate,” British Journal of the Philosophy of Science
49(2): 199–225.
Appiah, K. A. 1990. “Racismsin David Theo Goldberg,” (ed.), Anatomy of Racism, Minneapolis: University
of Minnesota Press.
—— 1996. “Race, Culture, Identity: Misunderstood Connections,” in A. Appiah and A. Gutmann (eds.),
Color Conscious, Princeton, NJ: Princeton University Press.
Atlas, J. D. 2004. “Presupposition,” in L. R. Horn and G. L. Ward (eds.), The Handbook of Pragmatics,
Oxford: Blackwell. 29–52.
Austin, J. L. 1962. How to Do Things with Words. London: Oxford University Press.
Bach, K. 1994. “Conversational Impliciture,” Mind and Language 9(2): 124–162.
—— 1999. “The Myth of Conventional Implicatures,” Linguistics and Philosophy, 22.4 (Aug1999):
327–366.
Bianchi, C. 2008. “Indexicals, Speech Acts and Pornography,” Analysis 68: 310–316.
Bird, A. 2002. “Illocutionary Silencing,” Pacific Philosophical Quarterly 83: 1–15.
Brontsema, R. 2003. “A Queer Revolution: Reconceptualizing the Debate over Linguistics Reclamation,”
Colorado Research in Linguistics 17, (1): 1–16.
Butler, J. 1997. Excitable Speech: A Politics of the Performative. New York and London: Routledge.
Carston, R. 2002. Thoughts and Utterances. Oxford: Basil Blackwell.
—— 2004. “Relevance theory and the saying/implicating distinction,” in L. Horn and G. Ward (eds.)
Handbook of Pragmatics. Oxford: Blackwell 607–632.
Catterall, H. T. (ed.) 1926. Judicial Cases Concerning Slavery and the Negro. Washington, D.C.: Carnegie
Institute.

765
 RAE LA N G TO N , SA LLY H A SLA N G E R , A N D L U V E L L A N D E R S O N

Cimpian, A., Brandone, A. C. and Gelman, S. A. 2010. “Generic Statements Require Little Evidence for
Acceptance but Have Powerful Implications.” Cognitive Science 34(8): 1452–1482.
Delgado, R. 1993. “Words that Wound: A Tort Action for Racial Insults, Epithets and Name Calling,” in
K.W. Crenshaw et al. (eds.), Words that Wound. Boulder, CO: Westview Press.
Dubois, W. E. B. 1996. “The Name “Negro’” in E. J. Sundquist (ed.) The Oxford W.E.B. Dubois Reader.
Oxford: Oxford University Press 70–72.
Dummett, M. 1973. Frege’s Philosophy of Language. Oxford: Clarendon Press.
Feinberg, J. 1985. Offense to Others New York. London: Oxford University Press.
Fiss, O. 1996. The Irony of Free Speech. Cambridge, MA: Harvard University Press.
Frege, G. 1918/19. “The Thought: A Logical Inquiry,” in R. Harnish (ed.) Basic Topics in the Philosophy of
Language. Englewood Cliffs, NJ: Prentice-Hall 517–535.
Fricker, M. 1998. “Rational Authority and Social Power: Towards a Truly Social Epistemology,” Proceedings
of the Aristotelian Society, 98(2): 159–177.
—— 2007. Epistemic Injustice: Power and the Ethics of Knowing. New York: Oxford University Press.
Glasgow, J. 2009. A Theory of Race. New York: Routledge.
Grice, H.P. 1961. “The Causal Theory of Perception,” Proceedings of the Aristotelian Society, Supp. xxx:
121–153.
Grice, P. “Logic and conversation,” in Studies in the Way of Words. Cambridge, MA: Harvard University
Press.
Habermas, J. 1984. The Theory of Communicative Action, translated by Thomas McCarthy. Boston: Beacon
Press.
Haslanger, S. 2008. “A Social Constructionist Analysis of Race,” in B. A. Koenig, S. Soo-Jin Lee and S.
Richardson (eds.), Revisiting Race in a Genomic Age, New Brunswick: Rutgers University Press 56–69.
—— 2010. “Language, Politics and ‘the Folk’: Looking for ‘The Meaning’ of ‘Race’”. The
Monist93:169–187.
Hom, C. 2008. “The Semantics of Racial Epithets,” Journal of Philosophy 105: 416–440.
Horn, L. 1989. A Natural History of Negation. Chicago: Chicago University Press.
Hornsby, J. 1994. “Illocution and its Significance,” in S.L. Tsohatzidis (ed.) Foundations of Speech Act Theory,
London: Routledge.
—— 1995. “Disempowered Speech,” Philosophical Topics 23: 127–147.
—— 2001. “Meaning and Uselessness: How to Think about Derogatory Words,” in P. French, et al. (eds.)
Midwest Studies in Philosophy XXV: Blackwell 128–141.
James, M. 2008. “Race,” in E. N. Zalta (ed.) The Stanford Encyclopedia of Philosophy (Fall 2008 Edition),
http://plato.stanford.edu/archives/fall2008/entries/race/
Kaplan, D. 1999. “The Meaning of “Ouch” and “Oops”: Explorations in the Theory of Meaning as Use,”
ms. UCLA.
Kennedy, R. 1999–2000. “‘Nigger’? and Other Considerations,” The Journal of Blacks in Higher Education 26:
86–96.
—— 2003. Nigger: The Strange Career of a Troublesome Word, New York: Vintage Press.
King, M. L. 1963. “I have a Dream” speech, MLK Online http://www.mlkonline.net/ Accessed June 8,
2010.
Kitcher, R. 1999. “Race, Ethnicity, Biology, Culture,” in L. Harris (ed.) Racism, New York: HumanityBooks.
Langton, R. 1993. “Speech Acts and Unspeakable Acts,” Philosophy and Public Affairs 22: 305–330.
—— Forthcoming. “Beyond Belief: Pragmatics in Hate Speech and Pornography,” in M.K. McGowan and
I. Maitra (eds.) What Speech Does, Oxford: Oxford University Press.
—— 2009. Sexual Solipsism: Philosophical Essays on Pornography and Objectification. Oxford: Oxford University
Press.
Langton, R. and Hornsby, J. 1998. “Free Speech and Illocution,” Journal of Legal Theory 4: 21–37.
Langton, R. and West, C. 1999. “Scorekeeping in a Pornographic Language Game,” Australasian Journal of
Philosophy 77: 303–19.
Leslie, S. J. 2007. “Generics and the Structure of the Mind,” Philosophical Perspectives 21 (1): 375.
—— 2008. “Generics: Cognition and Acquisition,” The Philosophical Review 117(1): 1.
—— Forthcoming. “The Original Sin of Cognition: Fear, Prejudice and Generalization.” Journal of
Philosophy.
Leslie, S. J., Khemlani, S., Prasada, S. and Glucksberg, S. 2009. “Conceptual and Linguistic Distinctions
between Singular and Plural Generics,” Proceedings of the 31st Annual Cognitive Science Society.
Amsterdam: Cognitive Science Society.

766
 LA N G U A G E A N D R A C E

Lewis, D. 1978. “Truth in Fiction,” American Philosophical Quarterly 15, 37–46, reprinted in Collected Papers,
Vol. I, 261–280.
—— 1979. “Scorekeeping in a Language Game,” Journal of Philosophical Logic 8: 339–59, reprinted in
Philosophical Papers, Vol. I, Oxford: Oxford University Press, 1983, 233–249.
MacKinnon, C. 1987. Feminism Unmodified. Cambridge, MA: Harvard University Press.
——1993. Only Words. Cambridge, MA: Harvard University Press.
Matsuda, M. with Lawrence, C.R. III, Delgado, R. and Williams Crenshaw, K. 1993. Words that Wound:
Critical Race Theory, Assaultive Speech and the First Amendment. Boulder, CO: Westview Press.
McGowan, M. K. 1993. “Conversational Exercitives and the Force of Pornography,” Philosophy and Public
Affairs 31: 155–189.
—— 2004. “Conversational Exercitives: Something Else We Do With Our Word” Linguistics and Philosophy
27: 93–111.
—— 2009. “Oppressive Speech,” Australasian Journal of Philosophy 87: 389–407.
—— and Ishani, M. 2007. “Limits of Free Speech: Pornography and the Question of Coverage,” Legal
Theory, 13: 41–68.
Omi, M. and Winant, H. 1994. Racial Formation in the United States. New York: Routledge.
Outlaw, L. 1990. “Toward a Critical Theory of Race,” in D.T. Goldberg (ed.) Anatomy of Racism, Minneapolis:
University of Minnesota Press.
Potts, C. 2005. The Logic of Conventional Implicatures. Oxford: Oxford University Press.
Prasada, S. and Dillingham, E. M. 2006. “Principled and Statistical Connections in Common Sense
Conception,” Cognition 99(1): 73–112.
—— 2009. “Representation of Principled Connections: A Window onto the Formal Aspect of Common
Sense Conception,” Cognitive Science 33(3): 401–448.
Richard, M. 2008. When Truth Gives Out. Oxford: Oxford University Press.
Rosengren, J. 2008. Hammerin’ Hank, George Almighty and the Say Hey Kid. Naper, IL: Sourcebooks.
Saul, J. M. 2008. “Pornography, Speech Acts and Context,” Proceedings of the Aristotelian Society, 229–248.
Streicher, J. (ed.). 1943. Der Stürmer (Streicher Verlag), from an essay by Ernst Hiemer, “The Holy Hate”;
German Propaganda Archive, translated and edited by Randall Bytwerk, http://www.calvin.edu/academic/
cas/gpa/ww2era.htm Accessed June 3, 2009
Schwartzman, L. 2002. “Hate Speech, Illocution, and Social Context: A Critique of Judith Butler,” Journal
of Social Philosophy 33: 421–41.
Searle, J. R. 1969. Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University
Press.
—— 1979. Expression and Meaning. Cambridge: Cambridge University Press.
Sperber, D., Premack D. and Premack A. J. (eds.). 1994. Causal Cognition: A Multidisciplinary Approach.
Oxford: Clarendon Press.
Stalnaker, R. 2002. “Common Ground,” Linguistics and Philosophy 25(5): 701–721.
Strawson, P.F. 1964. “Intention and Convention in Speech Acts,” Philosophical Review 73: 439–460.
United Nations (UN) 1965. International Convention on the Elimination of All Forms of Racial
Discrimination, Article 4, http://www.hrcr.org/docs/CERD/cerd3.html Accessed June 3, 2009.
von Fintel, K. 2008. “What is Presupposition Accommodation, Again?” Philosophical Perspectives 22:
137–170.
Wieland, N. 2007 “Linguistic Authority and Convention in a Speech Act Analysis of Pornography,”
Australasian Journal of Philosophy 85: 435–56.
Williamson, T. 2007. “Reference, Inference, and the Semantics of Pejoratives,” in J. Almog and P. Leonardi
(eds.)Festschrift for David Kaplan, Oxford University Press: Oxford.
Zack, N. 2002. Philosophy of Science and Race. New York: Routledge.

767
 6.5
AP RI OR IT Y
Sinan Dogramaci

Introduction

What Is Apriority?
Apriority characterizes a way our beliefs acquire justification. Justification acquired
purely in this way is apriori justification, as opposed to empirical (also called “aposteri-
ori”) justification. Apriority then derivatively applies to knowledge or to propositions
when apriori justification is or could be had, respectively. In English, philosophers today
may use either the single or double words, “apriori” or “a priori,” sometimes italicizing
“a priori” to indicate use of the original Latin.
When does a belief have apriori justification? The traditional answer is: when its
epistemic justification is purely acquired independently of experience. However, that
gloss stands in need of several refinements, and there is no consensus as to how to
make them. First, what is epistemic justification? Roughly, it’s justification had when a
person’s belief is likely from her perspective to be true. It’s unclear, though, how to fully
explicate the relevant notion of likelihood in noncircular terms. Second, what counts
as experience here? Perceptual experience, the hallmark of empirical justification, cer-
tainly counts, but whether, for instance, introspection or other modes of self-reflection
should also count is unclear. Finally, and perhaps most famously, no beliefs, much less
justified ones, can be had totally independently of experience. As Kant famously put it,
“how else should the cognitive faculty be awakened into exercise if not through objects
that stimulate our senses and in part themselves produce representations . . . ?” (Kant
(1781/1998): B1) A belief’s apriori justification must somehow hold not in virtue of any
experiences, even if some experiences are causal prerequisites.
Many important philosophers, especially in recent history, have questioned whether
the notion of apriori justification is coherent, has theoretical utility, or has interesting
scope (Quine (1951); Devitt (1998); Harman (2003); Stalnaker (2003); Williamson
(2008).). These challenges shouldn’t be dismissed, but this won’t be where we face up to
them. I’ll assume we have a working grasp of a coherent notion of the apriori and can rely
on a working hypothesis that it is theoretically sound. Any investigation into the apriori
proceeds with a hope to refine and a willingness to revise our initial assumptions.
What is a discussion of a patently epistemological notion doing in a volume on the
philosophy of language? The basic idea of an intimate connection between apriority and
analyticity is centuries old. In the last twenty or so years, though, an interesting connec-
tion has been seen between apriority and conceptual role semantics. After introducing a
little more context, my main aim will be to critically examine that latter connection.
 A PRI O R I T Y

The Traditional Scope of the Apriori

Apriority has been alleged to characterize propositions across a diverse range of top-
ics. A sample of such topics includes: morality, color relations, metaphysical modality,
axioms of recently developed formal theories such as set theory and probability theory,
and philosophy itself. Controversial as the notion of apriority itself already is, its appli-
cation to any particular one of these topics is yet a further controversy. However, if the
notion of apriority is in good standing, then there are two core topics to which it most
uncontroversially applies: mathematics and logic. If these topics aren’t apriori, apriority
is of little interest.
All of mathematics is generally taken to be apriori, but the propositions whose aprior-
ity is most discussed tend to be claims of arithmetic, in particular the most elementary of
such truths. Consider your justification for believing that every number has a successor.
This isn’t something you plausibly know on the basis of any perceptual experience. This
is a paradigmatic example of apriori justification.
Within logic broadly understood, we need to distinguish two areas of apriority. First,
we have apriori justification for believing formal metalogical laws. For example, we’re
justified in believing in the validity of Modus Ponens, i.e. in the claim that every argu-
ment of the form p, if p then q, therefore q is truth-preserving. By contrast, there is our
apriori justification for believing and inferring logical truths. When you reason your
way toward some novel logical truth, reasoning by Modus Ponens and other such rules,
apriori justification is transmitted from premises to conclusion in each inferential step.
In some cases involving zero premises, justification can be generated, not merely trans-
mitted. (An example of a rule that doesn’t simply transmit but generates apriori justi-
fication might be the rule that licenses belief in the logical truth that either alien life
exists or it doesn’t.)
This generation and transmission of apriori justification within logical reasoning is of
special importance. While it’s been argued that our justification in some of these core
mathematical, metalogical, and logical areas reduces to our justification in others, the
one area where our justification seems least susceptible to reduction is this last area of
logical reasoning. (For discussion of such reductions, see Dummett (1973), Boghossian
(2001), Wright (2001) and Dogramaci (2010) on logical reasoning and metalogical
laws, and see Hale and Wright (2001) on mathematics and logic.) For that reason, and
because space is limited, in what follows I will primarily focus on apriority in cases where
we reason toward the logical consequences of what we know.

Two Main Views of What Explains Apriority

In the contemporary debate, there is something of a competition between two main
views about what explains apriori ways of acquiring justification. Though no labels are
completely standard, I’ll call these intuition-based views and meaning-based views. The
latter views are our focus here, but it will be useful to briefly talk about intuition-based
views. The ultimate evaluation of meaning-based views largely comes down to how
their net plausibility compares with that of intuition-based views.
An intuition-based view aims to explain the apriori way a person acquires justification
by appealing to intuitions she has, a kind of mental state modeled on perceptual expe-
rience. Such a quasi-perceptual state of intuition was perhaps most famously invoked
by Descartes, in the notorious form of clear and distinct perception. Intuition-based

769
 SI N A N DO G R A MA C I

views have, in the last twenty years or so, again come into fashion (BonJour (1998);
Huemer (2007); Sosa (2007); Bengson (2010); Chudnoff (forthcoming); Bealer (2000)
is a hybrid intuition- and meaning-based view); at the same time, naturally, criticism of
intuition-based views has also seen a rise (e.g. Weinberg et al. (2001)).
Perhaps the most attractive prospect that intuition-based theories hold is that of a
fully general and unified explanatory theory of epistemic justification: as perceptual
experiences explain empirical justification, intuitions explain apriori justification. To
secure this advantage though, the intuition-theorist cannot construe intuitions as any-
thing too obscure (such as clear and distinct perception). Intuitions should be a kind of
mental state that’s very similar to perceptual experiences. As such, many contemporary
theorists (e.g. Bealer, Huemer, Chudnoff) construe intuitions merely as seemings, the
mental state you’re in whenever something seems to you to be the case. On such views,
just as your belief that blood is red is justified by its visually seeming that blood is red,
your belief that either aliens exist or they don’t is justified by its seeming so, not percep-
tually but merely “intellectually”.
Two major challenges shape the current debate over intuition-based views. One of
these concerns how intuitions help explain inferential justification. Even if we suppose
an intuition justifies belief in a metalogical claim that, say, a certain form of inference is
valid, there’s a question about whether that could explain how justification gets trans-
mitted across beliefs in an inference. The worry is inspired by a famous dialogue by Lewis
Carroll (Carroll (1895)). Carroll’s cryptic dialogue appears to raise a worry that, if infer-
ential justification involves knowledge of validity, then some kind of infinite regress
ensues. In a recent sequence of articles, Paul Boghossian has reinvented the challenge
not as a threat of regress, but as a charge that knowledge of validity is simply explana-
torily redundant when it comes to explaining certain cases of inferential justification. I
already know the conditional if p then, q; it adds nothing explanatorily significant for me
to have an intuition that (necessarily) if: p and if p then q: then: q. (Boghossian (2001);
Boghossian (2003). Wright (2001) critically examines Boghossian’s challenge.)
The other challenge to intuition-based views can be posed in the form of a dilemma.
On the one hand, intuitions had better not be anything supernatural; it’s not plausible
that any faculty of rational insight puts us in, say, causal touch with some heavenly
realm of mathematics or logic. This pressures us toward a nonexotic construal of intui-
tions, something like mere seemings. On the other hand, if intuitions are constituted
just by a seeming, an internal mental state that anyone could have at anytime, then
what reason is there to think intuitions are reliable? We can tell a plausible causal story
about how perceptual experiences are reliable, but a causal story will not do for intui-
tions. The challenge here was first posed by Benacerraf (1973), and later sharpened by
Field (1989) as follows: if you think there is no explanation of how your beliefs about
some topic are reliable, then you are rationally obliged to give those beliefs up.
Meaning-based views of apriority enter the debate advertising themselves as espe-
cially well suited to explain both reliability and inferential justification. What, then, is
a meaning-based view?
The starting point or background premise of any meaning-based view is some form of
conceptual role semantics, henceforth “CRS.” Different meaning-based theorists accept
different forms of CRS. The weakest thesis common to all meaning-based views is that
a person’s disposition to engage in certain patterns of reasoning is a necessary condi-
tion on her understanding certain meanings or grasping certain concepts. Note, this is
stronger than just a claim about what’s required to have representational states or even

770
 A PRI O R I T Y

to have beliefs: the claim is about what’s required to have states with certain contents,
contents involving particular concepts. Note, though, that the claim is weaker than the
common reductive forms of CRS: meaning-based theorists needn’t accept that there are
non-intentionally describable dispositions constituting our intentional mental states.
(Boghossian and Peacocke, prominent defenders of meaning-based views, have both
rejected reductive forms of CRS (Boghossian (1989); Peacocke (1998)). For an exten-
sive critique of CRS in all forms, see Williamson (2008).)
Now, a conceptual role semantics of any kind for any of our words or concepts is
controversial. CRS is perhaps on firmest ground, when it comes to our understand-
ing of logical constants. It certainly seems impossible that a subject fully understands
the standard logical connectives while lacking any disposition to draw the inferences
governed by certain canonical rules. For example, suppose we want to know whether
this linguistic (or mental) symbol in a person’s language (of thought), •, expresses the
concept of conjunction. An extremely intuitive necessary (even sufficient) condition on
that is that she is disposed to infer p • q from the pair of beliefs p and q, and is disposed
to infer p, and to infer q, from the belief p • q. Similarly, anyone who systematically
refused to infer q from the pair of beliefs p and p * q would not plausibly mean the con-
ditional by that connective.
(Could someone grasp a concept, by having a disposition that she never exercises?
We can allow for such a bizarre possibility. The person succeeds in understanding
p * q as a conditional, because if she were also to believe p and to consider whether q,
she would infer q. Since she never gets the opportunity to learn p, she never makes the
inference. This reasoner might even be perfectly rational, though she might not get to
form as many beliefs as she’d like.)
Supposing a CRS for logical concepts is roughly right, then meaning-based views may
begin to look attractive. What we notice is an apparent coincidence, literally speaking,
between the inferences crucial to understanding logical connectives and the inferences
that transmit justification in an apriori way. But we shouldn’t rush to conclusions. CRS
itself has no immediate epistemological implications.
The claim of interest, and the claim that defines something as a meaning-based view,
is that the conditions on understanding a meaning or grasping a concept somehow
explain the apriori way that beliefs acquire justification. What, then, is the explanatory
link from CRS to justification supposed to be? And, how does the link address the chal-
lenges that faced intuition-based views? We’ll now look at the leading contemporary
proposals, and the problems those proposals face.

Meaning-based Explanations of Reliable Inference

Connecting Meaning and Reliable Inference

Each of the leading contemporary advocates of meaning-based views, namely Peacocke,
Boghossian, Bealer, and Hale and Wright, have, at one time or another, endorsed the fol-
lowing powerful claim: When a disposition to engage in certain reasoning is essential to
understanding meaning, this secures that reasoning’s reliability. To see what can be said
in defense of this claim, we have to step back from the issue of apriority for a moment.
Supposing CRS for the logical connectives is roughly right, there is a question that
arises before we even bring up apriority at all. What explains the connection between
certain inferences and certain concepts? In other words, why are these inferences crucial

771
 SI N A N DO G R A MA C I

to grasping conjunction; why are those inferences crucial to grasping the conditional?
There must be some general theory dictating how concepts get paired with their associ-
ated conceptual roles.
A clue to answering that question emerges when we consider certain defective con-
ceptual roles. Prior (1960) concocted the following famous pair of inference rules, alleg-
ing they define a connective he dubbed tonk: from p, infer p-tonk-q; and from p-tonk-q,
infer q. Prior’s intent was to (light-heartedly) mock the claim of a connection between
an inference’s reliability and its being essential to understanding the meaning of a con-
nective; tonk’s inferences, far from being reliable, lead to any absurdity. But a different
and very plausible moral was drawn, first by Belnap (1962) and by many others since,
namely that there simply is no genuine logical operation identified by the connec-
tive tonk; tonk is simply meaningless. That, in turn, suggested there must be substantial
constraints on when an inference is essential to understanding a meaning, at least one
of those constraints being that the inferences are truth-preserving. A truth-preservation
constraint would also begin to explain the pairings between the standard logical con-
nectives and their canonical inferences: Those connectives each make their associated
inferences truth-preserving. For example, if • expressed disjunction or the conditional,
rather than conjunction, then the inference to p (or to q) from p • q would be highly
unreliable.
If the general theory dictating which conceptual roles pair with which concepts does
include such a constraint, then the meaning-based theorist is a step closer to explaining
how apriority ensures reliability. The full explanation isn’t in hand yet, since we still
need the part of the explanation that says why non-truth-preserving inferences cannot
occur in the conceptual roles essential to understanding meanings. If we leave it a pure
coincidence that those inferences are truth-preserving, then the position looks no bet-
ter than that of the intuition-theorist who simply asserts a coincidence between intui-
tions and truths. The Benacerraf–Field challenge to our ability to rationally maintain
our beliefs in this area would remain.
Now, there is a plausible idea the meaning-based theorists can draw on to explain
the connection between meaning and reliability. The idea is that in coming to under-
stand the meaning of a new word or grasp a new concept, we implicitly stipulate that this
word/concept expresses whatever semantic value makes certain inferences truth-preserving.
For example, we may fix the meaning of • as conjunction, in part, by implicitly stipulat-
ing that • expresses whatever truth-function makes certain inferences truth-preserving
(namely, the canonical inferences for conjunction mentioned earlier). The inference
rules for tonk then don’t help to fix any meaning, because the relevant stipulation can-
not be fulfilled.
Much more must be said, of course, to fully develop and defend this sort of strategy
for connecting meaning and reliability. Christopher Peacocke has done extensive work
on this project, engaging with it in many influential works (including Peacocke (1993);
Peacocke (2000); Peacocke (2004)). Boghossian (1996), Bealer (2000), and Hale and
Wright (2001) also defend the strategy, though all in different ways. Against all these
views, Paul Horwich argues that stipulations of truth cannot in any way explain how we
understand meanings, nor can facts about meaning-constitution secure the reliability of
our inferences (Horwich (2005): chapter 6).
The outcome of this debate is important: it decides whether meaning-based views
really are in a superior position to address the Benacerraf–Field challenge. However,
even if that challenge were put to rest, there remain serious doubts about whether a

772
 A PRI O R I T Y

meaning-reliability connection could explain everything we need to explain about

apriority. I turn to those doubts now.

Doubts about Whether a Meaning-Reliability Connection

Fully Explains Apriority
Apriority, the thing we want a theory to explain, is a way of acquiring epistemic justifi-
cation. Now, some philosophers, so-called “reliabilists,” hold the view that the one and
only way for a belief to acquire epistemic justification just is that it be the product of a
reliable belief-forming process. But, that view is highly controversial, and in any case,
all of the leading meaning-based theorists explicitly reject it. Two of them, though,
Peacocke and Bealer, do claim that a guarantee of reliability would explain apriori justi-
fication. Here is Peacocke’s position:

According to the metasemantic theory [Peacocke’s theory], in using an a priori

way of coming to judge that p a thinker is using a method which guarantees,
as a result of the very nature of p and the way in which its truth-condition is
determined, that the thinker judges that p only if it is the case that p. When
the soundness of a method is thus internally related to what it is for the content
to be true, it is hard to see what more could be required for knowledge. Such a
constitutive grounding of the soundness of the method goes far beyond merely
reliabilist conditions for knowledge. (Peacocke (2004): 173)

(The view is also put forward in Peacocke (1993): 190; Peacocke (2000): 265. For a
similar view, see Bealer (2000): 7–10. Bealer doesn’t demand a guarantee, only a “strong
modal tie between [a method’s] deliverances and the truth” (p.9). For brevity, I’ll stick
to using “guarantee”, but the following comments apply equally if we substitute ‘‘strong
modal tie”).
In the passage, Peacocke mentions conditions for knowledge, but I want to set knowl-
edge aside and focus on the weaker condition of being epistemically justified. Should
we agree with the suggestion that if a way of forming a belief is guaranteed to be reli-
able, that explains why it is epistemically justified? I don’t know any simple objection
that would compel a staunch advocate like Peacocke or Bealer to concede this isn’t a
sufficient condition for both knowledge and justification. However, when they offer
the condition as an explanatory one, as they do, their view does face pressure: there are
two heavy burdens the view has to shed before it can be seen as part of the best overall
explanatory theory of justification.
The first burden is as follows. We want the best overall explanation of epistemic
justification. The best explanations are highly general. So, we prefer to give a single
explanation that covers all epistemic justification, apriori and empirical. But, a standard
counterexample to reliabilism seems to show that empirically justified beliefs need not
be the product of any reliable process. It seems possible that a brain-in-a-vat, which is
fed a stream of misleading perceptual experiences indiscernible from some nonenvatted
person’s, can form completely justified beliefs. So, whatever explains why her beliefs are
justified, it seems to not be that they are the product of a reliable process, much less a
process guaranteed to be reliable. So, if we pursue a meaning-based view that explains
apriori justification via a guarantee of reliability, then it seems we will have to settle
for a disjunctive explanatory theory: we’ll have to settle for a theory that says one thing

773
 SI N A N DO G R A MA C I

explains empirical justification, while a completely different thing explains apriori justi-
fication. Thus, the first heavy burden on pursuing such a view is to say why, contrary to
how it just now seemed, we’re not pursuing an explanatory theory that’s less than best.
The second burden has been pressed by Boghossian. He draws on the other standard
counterexample to reliabilism: it seems that an unwittingly reliable clairvoyant would
not be justified in believing the deliverances of her faculty of clairvoyance. Boghossian
(2003) suggests there can be reliable inferences that are essential to understanding some
meanings, but which are as wildly irresponsible as the clairvoyant’s beliefs. To illustrate
this, he claims that there could be a concept, the concept flurg, such that fully under-
standing flurg requires immediately inferring from the arithmetical axioms that some
numbers are flurg, and then from that immediately inferring Fermat’s Last Theorem.
(Here we’re referring to this concept without using it ourselves.) This pair of immediate
inferences leading to Fermat’s theorem obviously doesn’t transmit justification. The
burden on the position in question, then, is to say why the inferences don’t transmit
justification in the same way the inference from p&q to p does. The only apparent
way of doing this is to deny that there is any genuine concept which can only be fully
grasped by making that wild leap to Fermat’s theorem. So, the effective burden is to say
why there is no such concept. That burden is difficult to shed, because the proposed
inferences for flurg are truth-preserving; indeed, as an arithmetical theorem, Fermat’s
theorem is an apriori proposition. So, flurg cannot be dismissed as easily as tonk can be.
We could add new constraints to our general theory linking concepts and conceptual
roles, but it’s unclear how to avoid introducing ad hoc constraints (i.e. the constraints
cannot be motivated by the fact that they handle flurg).
These burdens certainly aren’t the last word on whether a meaning-based view can
explain justification by appealing to a guarantee of reliability (as the presentation of
them as burdens, rather than refutations, was designed to emphasize). But this much
pressure motivates us now to examine a meaning-based route to explaining epistemic
justification that purports to do better.

Meaning-based Explanations of Epistemic Responsibility

Blind-Yet-Blameless Inference
How can the meaning-based theorist explain how beliefs are epistemically justified with-
out drawing on the connection between meaning and reliability? The starting point of
Boghossian’s explanation is as follows. Even if apriori ways of acquiring justification are
reliable, the unwittingly reliable clairvoyant shows that even when beliefs are reliably
formed something can go wrong that leaves them unjustified. Boghossian suggests that
what went wrong in the clairvoyant’s case is that, in forming her beliefs in the way she did,
she was being irresponsible. Boghossian’s guiding idea, then, is that if we can explain why
a thinker is being responsible when forming her beliefs in certain ways rather than oth-
ers, then we will have explained why those are ways of acquiring justification. And that
does seem plausible: The targeted notion of epistemic justification does seem to be tied
to a kind of responsibility, which we can call epistemic responsibility. At a very minimum,
epistemic responsibility is one philosophically worthwhile notion to aim to explain.
A simplistic approach to understanding epistemically responsible belief formation
would be to explicate it as belief formation that the believer knows or somehow recog-
nizes to be reliable. This would be one understandably natural initial reaction to the case

774
 A PRI O R I T Y

of the clairvoyant. Intuitively, the clairvoyant shouldn’t have formed her beliefs blindly
(“blindly” meaning “without any clue as to whether they were formed reliably”). But,
Boghossian firmly rejects that approach, and I earlier mentioned his Carrollian argument
against it. Perhaps if the clairvoyant knew that her faculty of clairvoyance is reliable,
she’d be a responsible believer. But, according to the Carrollian argument, what explains
why reasoning by Modus Ponens is responsible is not that I know my inference is reliable.
(To be sure, if I become philosophically reflective and wonder whether reasoning by
Modus Ponens is reliable, the only way I can responsibly continue reasoning by Modus
Ponens is if I give my question a positive answer. If I give a negative answer, or even if
I adopt an agnostic view, then I am responsible for suspending all reasoning by Modus
Ponens. But, this doesn’t mean that higher-order knowledge of the reliability of rea-
soning by Modus Ponens explains why the reasoning is epistemically responsible. And,
of course, it also doesn’t mean that the responsible person on the street is required to
engage in reflective epistemology in the first place.
There are philosophically interesting problems concerning how the reflective thinker
could positively settle this higher-order question of Modus Ponens’s reliability. Limita-
tions of space prevent me from getting into it here. The classic discussion is Dummett
(1973), further developed in Dummett (1991). Dummett argues, and Boghossian and
many others have agreed, that we can use reasoning by Modus Ponens to infer the con-
clusion that reasoning by Modus Ponens is reliable. This is a so-called rule-circular argu-
ment, familiar also from the soundness proofs used in metalogic. I oppose the legitimacy
of these rule-circular arguments in Dogramaci (2010).)
In looking around for some better way of explaining epistemic responsibility than an
appeal to higher-order knowledge of reliability, Boghossian’s next move is to offer the
following argument. (He will not ultimately accept the argument, but it leads us to a
better one.) Suppose a disposition to draw inferences of a certain form is essential to
understanding the meaning of some basic term (or, if you prefer, grasping some basic
concept). That term (or concept) would need to be used just to state the reliability
claim for the associated form of inference. Therefore, prior to having any disposition to
draw those inferences, it is not possible to evaluate the reliability of those very inferen-
tial transitions. And now an ought-implies-can style of maneuver comes: when it is not
possible to even have the thought that your inference is reliable, it cannot be irresponsi-
ble of you to draw that inference blindly. Thus they are blind-yet-blameless inferences.
This prima facie appealing argument was endorsed in Boghossian (2001), but he
brought it up again in Boghossian (2003) only to decide that it cannot adequately
explain apriority. The problem is that if the argument is any good, it applies to all infer-
ences that are essential to understanding a meaning, but then it applies to the irrespon-
sible inferences for understanding flurg. So, we need a better argument, one that won’t
exonerate flurg’s inferences.
In the last sections of Boghossian (2003), he sketches a path to improving upon the
argument for blind-yet-blameless inferences. Here’s my elaboration of the key ideas
from those last sections.
The clairvoyant was irresponsible, and an appealing explanation is that she’s irre-
sponsible because she formed her beliefs blindly when she did not have to. If we’re to
generalize that explanation, then we need to find some way in which someone reason-
ing by Modus Ponens must proceed blindly, while someone who draws the inferences
essential to understanding flurg does not have to form her beliefs blindly. Boghossian is
going to offer an idea for what that way is.

775
 SI N A N DO G R A MA C I

To get to that idea, he reviews a plausible point that Russell, Ramsey, Carnap, and
David Lewis all made about the conceptual roles essential for understanding theoreti-
cal terms (and concepts) of science. Any inferences essential to grasping these terms
hedge on the existence of a property or entity expressed by the term. Various proposals
about how the hedging happens differ in their details, but the basic idea is simple. One
illustrative way it might happen is that all the inferences essential to understanding,
say, “gravity,” must take place under a supposition: first we suppose there is a property
of gravity, and only then we say gravity is the property with such-and-such a role. That
role might even be the role of being the thing that makes such-and-such a set inferences
truth-preserving (again, only supposing there is such a thing). If the conceptual roles
are hedged in this way, both sides in a debate over the very existence of gravity may fully
understand the terms of the debate. Hedging thus seems to be something responsible
thinkers must do, at least wherever it is possible.
Boghossian doesn’t explicitly put it this way, but we can even say that when inferences
are hedged in the right way, they no longer count as blind; rather than being made “with-
out any clue” as to whether they are reliable, they are made under a supposition that they
are reliable. Seen in that way, hedging provides the key to understanding why someone
who fully understands flurg is being irresponsible in just the same way the clairvoyant was.
We wanted to say the clairvoyant formed beliefs blindly when she didn’t have to. Now,
to fully understand flurg, you do have to draw certain inferences, and you do have to draw
them blindly. But what you don’t have to do is fully understand flurg in the first place. Flurg
has a sister concept, call it flurg*, that can serve any legitimate purpose flurg could serve but
which requires only hedged inferences to be fully understood: first we suppose there exists a
property, flurg*, which some numbers have and which makes the transition from axioms to
Fermat’s theorem valid; then, given the axioms, we immediately infer that some numbers
are flurg*, and from that in turn we immediately infer Fermat’s theorem. Those inferences,
all made under a suitable supposition, aren’t blind and certainly aren’t irresponsible. Thus,
the inferences associated with flurg are indeed blind, but they are not blameless: the infer-
ences associated with flurg*, on the contrary, are not blind, and are blameless.
Finally, Boghossian claims that there is a minimal stock of concepts which a respon-
sible thinker must be permitted to fully understand without engaging in any hedging.
These are the concepts needed before you can hedge anything in the first place. He
doesn’t say exactly what’s in this minimal stock, but he says the concept of the con-
ditional certainly is, probably along with a few more very basic logical concepts. Fully
understanding the meaning we actually express using “if” cannot require making any
suppositions, since that would require already understanding a conditional thought.
Rather, understanding the concept of the conditional requires a disposition to engage
in certain reasoning flat out, including reasoning by Modus Ponens. There is no sister
concept for the conditional; unhedged inferences are the only option when it comes to
these conceptual abilities. Therefore, Boghossian concludes, since the only way to have
a concept like this is to be disposed to draw certain inferences blindly, one is not thereby
being irresponsible. These, and only these, are the blind-yet-blameless inferences.

Other Recent Views

In a, moment I’ll turn to an objection to the above line of argument. First, I want to
briefly mention recent work by other authors. The objection I will be raising applies to
all these views.

776
 A PRI O R I T Y

Schechter and Enoch (2006) is primarily a criticism of meaning-based views. How-

ever, what Schechter and Enoch take issue with is only the thesis that the nature of
epistemic justification can be fundamentally explained by any considerations concerning
meaning. They take meaning-based theorists like Peacocke and Boghossian to task for
false advertising about the role of meaning in the explanation of justification. Meaning-
based theorists say that meaning considerations explain why we have justification, but
then Peacocke appeals to reliability considerations, while Boghossian tacitly appeals to
something like an ought-implies-can principle to supply the fundamental explanation;
meaning only comes into the picture to account for why this or that is (guaranteed to
be) reliable or (im)possible.
Once Schechter and Enoch turn to their own positive view, which they call the
pragmatic account they also allow meaning to play that sort of ancillary role. Their
view is that what fundamentally explains how our beliefs acquire justification is
that we have an epistemic obligation to engage in certain cognitive projects, and this
justifies the belief-forming methods that must be used if those projects are to succeed.
For example, there might be an epistemic obligation to explain the world around you,
and this, on their view, is what justifies a rule for, say, inferring the best explanation.
They roughly characterize our basic epistemic obligations as obligations to explain,
understand, inquire, reason, and deliberate about the world. Meaning comes into the
picture because such obligations can only be successfully discharged if we are able to
have conditional thoughts, and given CRS, this in turn requires us to draw certain
inferences.
(Wright (2004) independently developed an extremely similar pragmatic view at the
same time. I’m focusing on Schechter and Enoch (2006) here because Wright (2004)
doesn’t, at least not explicitly, discuss appealing to CRS.)
Schechter and Enoch even suggest that Boghossian is tacitly presupposing something
like their view. They say:

Indeed, it is plausible that Boghossian’s discussion presupposes a version of the

pragmatic account. . . . Boghossian claims that . . . [b]elief-forming methods
that constitute unconditionalizeable [unhedgeable] concepts are justified . . .
because such concepts are needed to conditionalize [hedge], and hence are
needed to engage in inquiry. This suggests that on Boghossian’s view the justifi-
cation in all of these cases ultimately springs from the needs of rational inquiry,
presumably a rationally required project. (708–9)

Boghossian never explicitly said certain methods are needed to engage in inquiry, only
that they are needed to start hedging in the way a responsible thinker does. But Schech-
ter and Enoch’s suggestion seems a reasonable and fair way of elaborating the complete
argument at work.
A line of argument again in the same family is endorsed in Wedgwood (2011). Wedg-
wood says:

My current proposal, then, is that what makes the primitively rational cognitive
processes a priori is that they are precisely those processes that we must have
some disposition to engage in if we are to possess these basic cognitive capaci-
ties [where those capacities include “one’s possession of each of the various
concepts that one possesses”].

777
 SI N A N DO G R A MA C I

Here, “primitively rational” means justified, and not in virtue of the thinker’s establish-
ing their reliability using other arguments; “a priori” of course indicates that the justifi-
cation is independent of experience. Having said which beliefs and inferences involve
ways of acquiring apriori justification, Wedgwood then goes on to answer the attendant
why-question in a way that’s by now familiar to us:

Can anything more be said to make it plausible that it is precisely these capaci-
ties that explain why the primitively rational processes are a priori? It may be
relevant that these capacities seem to be necessary for even considering the
questions that one actually asks oneself. Clearly, one’s possession of the con-
cepts that appear in these questions is a capacity that one must have in order to
consider these questions. [All quotes from section 4 of manuscript.]

Wedgwood’s overall view, which we’re only scratching the surface of, is of course
unique in a number of subtle ways; the same goes for Schechter and Enoch’s positive
view. But it’s the common core elements that allow us to usefully group them together
with Boghossian’s account of blind-yet-blameless inference. And, as I’ll now show,
there is a single objection that threatens to bring down all these explanations of epis-
temic justification. (The objection threatens Wright (2004) too, but I cannot argue
that here.)
(For a different set of complaints about meaning-based explanations of epistemic
responsibility and justification, see Horwich (2005): chapter 6; and Horwich (2008).
Since space is limited, I will not review Horwich’s points here. I will rather develop an
objection that has not been pressed against meaning-based views, though a paragraph
in Williamson (2003): 252 anticipates the basic problem I develop.)

A Gap in the Meaning-Based Explanation of Responsibility

There’s a familiar distinction, associated with Chomsky (see, e.g., Chomsky (1965))
between competence and performance. Most versions of CRS only say that one must
be competent at drawing certain conclusions from certain premises in order to under-
stand our ordinary logical concepts. What competence involves exactly is debata-
ble. Maybe it just requires a simple disposition, maybe something more intellectu-
ally demanding, like rule-governed thinking, or maybe it’s something else entirely.
Whatever it requires exactly, on any plausible version of CRS, performance errors
are perfectly consistent with understanding. After all, it’s surely possible to affirm the
consequent, but doing that requires understanding a conditional. I can still perfectly
understand a conditional thought, if p then q, even if, say, on some occasion I irre-
sponsibly affirm the consequent, inferring the conclusion p from beliefs in q and in if
p then q. So, whatever level of understanding ordinary people have, it has to be con-
sistent with such errors.
Now, the epistemologist’s task is to explain facts about justification. The things that
are justified or unjustified are beliefs, and the things that succeed or fail to transmit jus-
tification to beliefs are inferences. All this is on the performance side of the Chomskyan
distinction. A cause for worry, now, is that the most that any of the arguments we heard
shows is that if you’re doing anything that’s necessary for understanding basic logical
thoughts, then you’re being responsible. (You might even be meeting an obligation, as
Schechter and Enoch have it.) But then this means what we heard could only explain

778
 A PRI O R I T Y

why it’s responsible (or obligatory) to have competence with certain forms of inference.
How do the arguments we’ve heard explain what we originally wanted to explain, the
status of performances (their status as responsible, obligatory, justification transmitting,
or whatever)?
The worry is not about having competence without ever exercising it. As mentioned
early on, perhaps there could be a perfectly rational thinker who never believes p and if
p then q, and so never has the opportunity to reason by Modus Ponens. Rather, the worry
is about how the fact that an inference manifests a responsible competence should explain
why the inference, the performance, is itself responsible (or obligatory, or justification
transmitting, or whatever).
Let me now make the problem more concrete. Suppose, in a moment of weakness,
I affirm the consequent. This is an irresponsible inference. If I had, in that instance,
reasoned by Modus Ponens, then I would have made a responsible inference. Now, since
affirming the consequent requires that I understand conditional thoughts, I must actu-
ally have whatever competence is required for understanding such thoughts. I’ve com-
mitted a performance error, but, I have also done everything necessary for understand-
ing conditionals. But, if that’s the actual case, then it would not have helped enable me to
understand conditionals had I reasoned by Modus Ponens; I already understand condition-
als. Therefore, my reasoning in the counterfactual scenario falls outside the class of
what the meaning-based theorists argued is responsible. The meaning-based theorists’
arguments fail to explain why my (counterfactual) inference is responsible. Further-
more, in a a case where I actually reason by Modus Ponens but easily could have affirmed
the consequent, the meaning-based theorists cannot explain why my actual inference
is responsible.
The basic source of the problem is that even if having a competence is responsible
(obligatory, whatever), that doesn’t entail that any manifestation of that competence is
responsible (obligatory, whatever). Is this claim too strong? Could the meaning-based
theorist resist it, perhaps by accusing it of “proving too much”? We might imagine the
meaning-based theorist using the following analogy to argue that an explanation of why
having a competence is good surely can always extend to explain why any competent
performances are good. It’s good to have a dust-free home. A dust-free home doesn’t
require any particular token dusting; it only requires a disposition or a habit of dusting
occasionally. But, surely the goodness of a dust-free home explains why it was good that
you dusted yesterday afternoon!
But, the analogy isn’t apt. Every token dusting makes your home less dusty. Not every
token instance of reasoning by Modus Ponens makes you more competent with the
concept of the conditional.
Here’s a better analogy for the relation between inferences and conceptual compe-
tence. Suppose animal overpopulation explains why it’s good to issue a hunting license
to someone. That explanation will entail that token killings are not all unjustified.
However, it will not entail that all killings are justified. Once the population is thinned,
there’s no explanation left for why the next token killing is good; it’s not good. In just
the same way, if my competence at inferring q from p and if p then q is secured, then
there’s no explanation left for why any additional, “extra” inferences are responsible.
There’s a gap between the explanation of why the competence is good and the explana-
tion of why all competent performances are good.
Meaning-based views have sparked a rich and valuable debate that spans nearly
all of contemporary philosophy’s main branches. As I’ve aimed to show here, their

779
 SI N A N DO G R A MA C I

viability depends on addressing some serious doubts about the specific proposals so far
made, doubts concerning meaning’s connections both to reliability and to justification.
There’s no doubt, however, that meaning-based theorists will continue to offer creative
arguments for assigning meaning an important role in the theory of the apriori.

Related Topics
2.4 Concepts
2.5 Analytic Truth
6.1 Philosophy of Language for Epistemology.

References
Bealer, George (2000). “A Theory of the A Priori.” Pacific Philosophical Quarterly 81(4): 1–30.
Belnap, Nuel (1962). “Tonk, Plonk and Plink.” Analysis 22(6): 130–4.
Benacerraf, Paul (1973). “Mathematical Truth.” The Journal of Philosophy 70(19): 661–79.
Bengson, John (2010). The Intellectual Given. Ph.D. thesis, The University of Texas at Austin.
Boghossian, Paul (1989). “The Rule-Following7 Considerations.” Mind 98(392): 507–549.
—— (1996). “Analyticity Reconsidered.’’ Nou s 30(3): 369–392.
—— (2001). “How Are Objective Epiatemic Reasons Possible?” Philosophical Studies 106(1–2): 1–40.
—— (2003). “Blind Reasoning.” Proceedings of the Aristotelian Society, Supplementary Volume 77(1):
225–248.
BonJour, Laurence (1998). In Defense of Pure Reason. Cambridge: Cambridge University Press.
Carroll, Lewis (1895). “What the Tortoise Said to Achilles.” Mind 4(14): 278–280.
Chomsky, Noam (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Chudnoff, Elijah (forthcoming). “The Nature of Intuitive Justification.” Philosophical Studies.
Devitt, Michael (1998). ‘‘Naturalism and the A Priori.” Philosophical Studies 92(1–2): 45–65.
Dogramaci, Sinan (2010). “Knowledge of Validity.” Noûs 44(3): 403–432.
Dummett, Michael (1973). “The Justification of Deduction.” Proceedings of the British Academy LIX:
201–232.
—— (1991). The Logical Basis of Metaphyics. Cambridge, MA: Harvard University Press.
Field, Hartry (1989). Realism, Mathematics and Modality. Oxford: Blackwell.
Hale, Bob and Crispin Wright (2001). The Reason’s Proper Study: Essays towards a Neo-Fregean Philosophy of
Mathematics. Oxford: Oxford University Press.
Harman, Gilbert (2003). “The Future of the A Priori.” In Philosophy in America at the Turn of the Century,
APA Centennial Supplement to the Journal of Philosophical Research, 23–34. Charlottesville, VA: Phi-
losophy Documentation Center.
Horwich, Paul (2005). Reflections on Meaning. Oxford: Oxford University Press.
—— (2008). “Ungrounded Reason.” The Journal of Philosophy 105(9): 453–471.
Huemer, Michael (2007). “Compassionate Phenomenal Conservatism.” Philosophy and Phenomenological
Research 74(1): 30–55.
Kant, Immanuel (1781/1998). Critique, of Pwe Reason. Ed. Paul Guyer and Allen Wood. Cambridge: Cam-
bridge University Press.
Peacocke, Christopher (1993). “How Are A Priori Truths Possible?” European Journal of Philosophy 1(2):
175–199.
—— (1998). “Implicit Conceptions, Understanding and Rationality.” Philosophical Issues 9: 43–88.
—— (2000). “Explaining the A Priori: The Programme of Moderate Rationalism.” In Paul Boghossian and
Christopher Peacocke (eds.), New Essays on the A Priori, 255–285. Oxford: Oxford University Press.
—— (2004). The Realm of Reason. Oxford: Oxford University Press.
Prior, Arthur (1960). “The Runabout Inference-Ticket.” Analysis 21(2): 38–9.
Quine, Willard Van Orman (1951). “Two Dogmas of Empiricism.” The Philosophical Review 60(1): 20–43.
Schechter, Joshua and David Enoch (2006). “Meaning and Justification: The Case of Modus Ponens.” Noûs
40(4): 687–715.
Sosa, Ernest (2007). A Virtue Epistemology: Apt Belief and Reflective Knowledge. Oxford: Oxford University
Press.

780
 A PRI O R I T Y

Stalnaker, Robert (2003). Ways a World Might Be. Oxford: Oxford University Press.
Wedgwood, Ralph (2011). ‘‘Primitively Rational Belief-Forming Practices.” In Reisner and Steglich-
Petersen (eds.), Reasons for Belief, chapter 9. Cambridge: Cambridge University Press.
Weinberg, Jonathan, Shaun Nichols and Stephen Stich (2001). “Normativity and Epistemic Intuitions.”
Philosophical Topics 29(1&2): 429–460.
Williamson, Timothy (2003). “Understanding and Inference.” Aristotelian Society Supplementary Volume
77(1): 249–293.
—— (2008). The Philosophy of Philosophy. Oxford: Blackwell.
Wright, Crispin (2001). “On Basic Logical Knowledge.” Philosophical Studies 106(1–2): 41–85.
—— (2004). “Warrant for Nothing (and Foundations for Free)?” Aristotelian Society Supplementary Volume
78(1): 167–212.

781
 6.6
NE C E S S I T Y AND
ME AN ING
Gillian Russell

We can approach the topic of necessity and meaning by distinguishing three kinds
of view about the relationship between modality—taken to include modal properties
like necessity and possibility, and modal objects such as possible worlds and possi-
bilia—and meaning. The first holds that modality is are rather mysterious and the
best way to explain it (both metaphysically and epistemically) is in terms of more
fundamental and accessible meaning properties. A famous example of this kind of
view is the Linguistic Doctrine of Necessary Truth, according to which all necessary
truth is to be explained in terms of analyticity (Ayer, 1936; Hempel, 1945; Goodman,
1955; Carnap, 1958).
A different kind of view holds instead that it is meanings which are mysterious, and
the best way to understand them is in terms of more fundamental, and better understood,
modal notions, such as those of worlds and possible objects. This approach reverses the
direction of explanation and metaphysical dependence suggested by the positivists and
a version can be found in Lewis (1976).
The third kind of view holds that semantic and modal notions do not mesh well
enough for one to be entirely dependent on, or explained in terms of, the other. This
view is of more recent provenance and reasons for it can be found in Thomason (1974),
Kaplan (1989b), Soames (2004), King (2005), and Russell (2008).
There is room for more skeptical approaches than the three listed above; Quine had
little time for full-blooded modal or semantic properties (Quine, 1951, 1953, 1966). But
if there is no such thing as one, then it is straightforward to see that there will be no
accounting for the other in terms of it, and so the focus of this chapter will be on what
we are to make of the relationship between meaning and modality, on the assumption
that both exist.

1
To say that something is necessary is to say that it could not have been otherwise and
it is customary to distinguish two flavours. De re necessity is attributed to objects, as in:

(1) The number 2 is necessarily half of 4.

(2) Emeralds are necessarily green.
(3) There is something that has the property of being necessarily made of ice.
 N ECESSI TY A N D ME A N I N G

De dicto necessity, by contrast, is attributed to truth-bearers, such as statements, proposi-

tions or beliefs. Examples of ascriptions of de dicto necessity include:

(4) The sentence ‘triangles have three sides’ is necessarily true.

(5) It is necessary that 2+2=4.

Propositions are often said to be the primary bearers of de dicto necessity, where this is to
say that if some other truth-bearer is de dicto necessary, then that is because it expresses
a proposition which is. One reason for this is that propositions are often taken to be the
primary bearers of truth, and from this it seems like a small step to say that they are the
primary bearers of necessary truth. (Cartwright, 1962) Another is that it does not seem
to be true that the sentence 2+2=4 is necessarily true; the sentence could have meant
something else, in which case it might have been false. We might insist on only attribut-
ing necessity to the sentence 2+2=4 under the circumstances in which it expresses the
proposition that 2+2=4, but then we may as well just say that it is propositions that are
the primary bearers, and that at best sentences inherit the property when they express a
necessarily true proposition.
De re necessity has sometimes been regarded as more metaphysically extravagant
than the de dicto kind, but both varieties can be puzzling. To say it is necessary that
snow is white seems to be to say no more about the way the world is than to say snow
is white—it is not as if we are reporting on an empirically detectable halo of necessity
which attaches to the snow’s whiteness if it has it necessarily, and which is absent if it
does not. One view is that when one attributes necessity, one is not saying something
more about the way the world is, but rather saying something about the ways it could
be. But what are these, if they are not facts about the way the world is? Thus there is a
metaphysical puzzle.
There is also an epistemic puzzle. I think I know how one could learn that emeralds
are green; one can look at them. But suppose it is true that emeralds aren’t just green, but
necessarily green. How would one learn that?

2 Linguistic Accounts of Necessity

An old idea is that both puzzles can be solved by appealing to meaning; necessity can
be metaphysically grounded in the meanings of our expressions, and our knowledge of it
is based on our the knowledge that speakers have of the meanings of those expressions.
It is easier to see how this works in the case of de dicto necessity. First, we distinguish
between two different kinds of sentence. Synthetic sentences true in virtue of what they
mean and the way the world is, e.g. the sentence snow is white is true in part because of
what it means and in part because snow has the colour it does. Analytic sentences, by
contrast, are true in virtue of meaning alone. ‘All triangles have three angles’, to take
an oft used example, will be true regardless of the way the world is.
The simplest version of the linguistic account says that a truth-bearer is necessary if
and only if it is analytic. De re necessity, if we are to have any, is to be reduced to the
de dicto kind. This is difficult, but at a first approximation we might suggest that if a
sentence attributing a predicate Fx of an object a is analytic, then a has the property
expressed by F x necessarily.
The linguistic approach is a tempting one. It is easy to understand in its basic outline
and has the virtue of parsimony; it has no need of possible worlds, and we are not required

783
 G I LLI A N RU S S E L L

to possess a mysterious faculty of intuition for apprehending modal truths. I think it is fair
to say, however that it is no longer the dominant view among philosophers of language.
In the next section we will examine some of the reasons the view fell out of favour.

Quine’s Critique
Quine was an influential critic of the linguistic account. Sometimes a complicated argu-
ment is made more succinct, and thus more tractable, if it is presented using symbols. So
let’s use ‘…’ (pronounced like ‘necessarily’) as what Quine called a sentence operator;
that is, like negation, ‘…’ is an iterable operator which may be applied to either open or
closed sentences. The following are then all well-formed:

(6) a) ¬Fa
b) …Fa
(7) a) ¬Fx
b) …Fx
(8) a) x¬Fx
b) x…Fx
(9) a) ¬¬¬xFx
b) ………xFx

In the first pair of sentences, the relevant operator is attached to an atomic sentence.
In the second, the operator is instead attached to an open atomic formula, i.e. one con-
taining at least one free variable. In (8) it appears within the scope of a non-vacuous
quantifier, and in the last pair the operator is iterated, so that we have instances of ™
and … that fall within the scope of another ™ or …, respectively.
About each construction we can ask: what does the construction mean? And what are
its truth-conditions? If our constructions make sense, these questions should be answer-
able. Some argue that the second is just a more formal way of asking the first—i.e. that
to give a sentence’s truth-conditions is already to give its meaning—but intuitively,
sentences with different meanings may have the same truth-conditions (e.g., 2+2=5
and Triangles have four angles) so even after we have answered the truth-condition ques-
tion, we may not yet have an answer to the meaning question. Often an answer to the
meaning question, however, will quickly give us an answer to the truth-condition ques-
tion. For example, if …Fa means that Fa is analytic, then we know that the sentence is
true just in case Fa is analytic. Thus it looks as if the linguistic account of necessity can
answer both questions for the sentences in (6). The problems start when we examine
the more complicated constructions in (7)–(9).

Open Sentences
Applying the linguistic account flat-footedly to …Fx suggests that it means that the
open sentence Fx is analytic. But as normally understood, analyticity entails truth and
open sentences like Fx are not objects of truth or falsity, except relative to particular
assignments of objects to the variable x. Perhaps we could say that …Fx, like ™Fx, is
only true relative to an assignment, i.e. we could say that the open sentence was true of
some objects and false of others. We know what is required in order for ™Fx to be true
of some object; it must be that Fx is not true of it. What is required for …Fx to be true of

784
 N ECESSI TY A N D ME A N I N G

an object? The natural answer is that the object must satisfy Fx necessarily, but if that is
to be the right answer, then we have to make sense of de re necessity before we can make
sense of sentences like (7b).
A linguistic account of de re necessity will say that objects inherit their modal prop-
erties (such as being necessarily F, or merely contingently G) from the analyticity of
sentences. For example, one might say that …Fx is true of an object o just in case …Fa
is analytic and a refers to o. This version of the approach exploits the analyticity of
sentences containing singular terms which refer to the object. An alternative is to use
predicates under which the object falls, i.e. one might say that …Fx is true of o iff o is G
and x(Fx l Gx) is analytic.
The problem with both approaches is that objects may be referred to by many singular
terms, and fall under many predicates, where some of these expressions form analytic
sentences when substituted in the appropriate places and some do not. To use Quine’s
example, if we think that the sentences
b. (10) All mathematicians are rational.
(11) All cyclists are bipedal.
are both analytic, whereas the sentences
c. (12) All cyclists are rational.
d. (13) All mathematicians are bipedal.
are synthetic, we might use our strategy to conclude of any particular mathematician
that she is necessarily rational but merely contingently bipedal, and of a particular
cyclist that she is necessarily bipedal, but merely contingently rational. But the prob-
lem is that some cyclists are also mathematicians, and in such cases our strategy gives
contradictory results: any cyclist mathematician is both necessarily rational and merely
contingently rational.
The singular term strategy runs into similar problems. If …Fx is true of an object o if
and only if t refers to o and …Ft is analytic, then if (within some particular context) we
think that:
(14) The red book is red.
(15) The big book is big.
are both analytic, but
(16) The red book is big.
(17) The big book is red.
are merely synthetic, then our strategy tells us that the referent of ‘the red book’ is neces-
sarily red, but contingently large, whereas the referent of ‘the big book’ is necessarily big,
but contingently red. As before, it may turn out that the red book is the big book, and
then our strategy leads to contradiction.
Several modifications might be attempted, but the most obvious ones turn out not
to work: we might say that o is necessarily F iff at least one predicate G applying to it
was such that (Fx l Gx) is analytic but a moment’s thought shows that this leads to
a trivialisation of de re necessity, making all objects possess all their properties neces-
sarily. We might say that o is necessarily F iff every single predicate ) applying to it was
such that …(Fx l )x) is analytic but this leads to no properties being possessed neces-
sarily. One might say that only certain canonical kinds of predicates or terms can be

785
 G I LLI A N RU S S E L L

used for determining an object’s de re modal properties, where we somehow ensure that
any two canonical expressions delivered consistent results, but how are such canonical
expressions to be identified, unless we say that a canonical predicate (say) is one which
picks out a property which holds of the object necessarily—a strategy which amounts to
abandoning the linguistic approach altogether (Burgess, 1997).

Non-Vacuously Quantified Sentences

Suppose we were to simply accept the problem above. One might think that this is only
a minor concession for the linguistic account. Open sentences do not have truth-values
anyway, and truth on an assignment is a technical notion from logic—it is not such a
huge concession to have to admit that …Fx does not have a truth-value relative to an
assignment. But the costs are greater than this, because we have to be able to interpret
open sentences like …Fx in order to make sense of closed sentences in which they are
embedded, such as x…Fx. Standard truth-condition clauses in first-order logic tell us
that for any formula )x:

x)x¬ is true iff )x¬ is true on some assignment.

x)x¬ is true iff )x¬ is true on all assignments.

So if we are interested in making sense of a quantified sentence like x…Fx or x…Fx

it tells us that:

x…Fx is true iff …Fx is true on some assignment.

x…Fx is true iff …Fx is true on all assignments.

So if we cannot make sense of an open-sentence being analytic relative to an assign-

ment, then we cannot make sense of closed sentences in which the modal operator
comes within the scope of a non–vacuously-binding quantifier either.

Iterated Modal Operators

A third problem arises for sentences which include multiple modal operators, as in (9b)
above. We perhaps don’t use these much in ordinary talk, but the ability to distinguish
between such formulae and their consequences is one of the distinctive advantages
of modal logic. If we had little need to consider iterated modalities, both quantified
and sentential modal logic would seem undermotivated. What does ……p mean on the
linguistic account? …p means that p is analytic. That suggests that ……p expresses a
metalinguistic claim:

(18) ‘p is analytic’ is analytic.

The problem is that on this interpretation it is hard to see why iterated modalities are of
much interest. I am not sure whether

(19) ‘Bachelors are unmarried’ is analytic.

is analytic or not, but if it is, presumably all claims of the same form as (18) are true when
an analytic sentence is substituted for p, and false where a synthetic sentence is. And if it

786
 N ECESSI TY A N D ME A N I N G

is not, then presumably all claims of the form of (18) are false. Either way, the truth-value
of ……p would be entirely determined by the truth-value of …p and it looks as if all the
fancier modal systems that allow for non-trivial iterated modalities are of little interest.

The Necessary A Posteriori

A more recent problem for the linguistic doctrine stems from the phenomenon of the
necessary a posteriori (Kripke, 1980; Sidelle, 1989). At the height of logical positivism
it was common for writers to collapse the ideas of necessity, a priority, analyticity, and
linguistic truth (Burgess, 1997). But Kripke (1980) distinguished necessity (the meta-
physical property of being such that it couldn’t be otherwise) from a priority (the epis-
temic property of being knowable independently of experience) and argued that a truth
could be necessary without being a priori. Kripke’s most famous examples include:

(20) Hesperus is Phosphorus.

(21) Water is H20.

although sometimes the more distracting questions (such as whether H2O really expresses an
a priori truth, and whether the semantic complexity of the expression ‘H2O’ makes a differ-
ence) are avoided by using these related sentences:

(22) Mars is not Phosphorus.

(23) Gorse is furze. (Soames, 2001)

The most straightforward way to see what these examples are supposed to show is to
start with Kripke’s view that names are rigid designators, which is to say that they refer
to the same object with respect to every possible world; if Hesperus refers to the planet
Venus with respect to every possible world, and Phosphorus also refers to Venus with
respect to every possible world, then, with respect to every possible world the proposi-
tion expressed by Hesperus is Phosphorus is simply that Venus is Venus—a necessary
truth. Yet one might not expect to to be able to know this a priori. One would have to
read it in a book, or make some astronomical observations. Hence it is an example of
the necessary a posteriori. (Kripke doesn’t put a lot of emphasis on distinguishing the
objects of these properties in the original text, but one place where it was later done
carefully and explicitly is (Soames, 2001).)
Such a result is also endorsed by standard modal logics, in which (a = b)  (…(a = b))
is a theorem, as is established by the following proof:

(a = b)  (Fa  Fb) (Leibniz’ Law)

(a = b)  (…(a = a)  …(a = b)) (Instance of Leibniz’ Law)
…(a = a) (necessity of identity)
(a = b)  (…(a = b))

If names in English are suitable substituends for a and b, then it would seem that …(Hes-
perus is Phosphorus) is a logical consequence of Hesperus is Phosphorus:

A1 Hesperus = Phosphorus  …(Hesperus = Phosphorus)

A2 Hespersus = Phosphorus
A3 …(Hesperus = Phosphorus)

787
 G I LLI A N RU S S E L L

Since sentences which express a posteriori truths are normally thought to be synthetic,
we seem to have an example of the necessary synthetic, which is prima facie inconsist-
ent with the linguistic theory of necessity.
One attempt to preserve the theory in the face of this problem is (Sidelle, 1989). Sidelle
suggests—plausibly enough—that the linguistic account need not strictly maintain that all
necessary truths are analytic, so long as it maintains that the source of their necessity is ana-
lyticity. With regard to A3, for example, he suggests that though the truth of A3 depends on
A2, the modal force of A3, as being necessary-if-true, comes entirely from the a priori, and
analytic, A1. Though A3 is necessary and a posteriori, its necessity has a linguistic origin,
in the analyticity of A1.
In a well-known review of Sidelle (1989), Yablo responds that Sidelle has underestimated
the problem of explaining how the analyticity of A1 could explain its own necessity, far
less that of A3. For, says Yablo, all the meaning of sentence can do is make it the case that
it expresses a certain proposition. Whether or not that proposition is then true, or neces-
sary, does not depend on the meaning of the sentence expressing it. Hence meaning alone
is never sufficient to establish necessity. (Yablo, 1992) A line of response (though not one
directed at vindicating the linguistic doctrine of necessary truth) can be found in Russell
(2008).

The Contingent Analytic

A further problem for the linguistic account can be found by thinking about context-
sensitive expressions like indexicals. An indexical is an expression that contributes dif-
ferent things to the proposition expressed by a sentence containing it in different con-
texts. Examples include I, now, today, tomorrow and here. If the sentence It is hot today is
uttered on 21st June, it expresses a different proposition (says something different, with
different truth-conditions) than it would have if it were uttered on 22nd June. Once we
notice that sentences containing indexicals can be used to assert different propositions
in different contexts, we have made room for the idea of a sentence with a special prop-
erty, that of expressing different true propositions in different contexts, but nonetheless
a true proposition in every context. Kaplan’s example is:

(24) I am here now.

The idea is that whoever is speaking, wherever, and whenever, the proposition that gets
expressed by (24) is a true one. People have proposed some interesting counterexamples
to this claim but for present purposes the important thing is not that (24) is really the best
example, but that we see that indexicality makes space for there to be a sentence with such
a property. So for argument’s sake, let’s just assume I am here now is such a sentence.
The proposition that (24) expresses will usually be a contingent one. (Kaplan, 1989a) I
am here now, but that didn’t have to be the case—I could have gone in to the university
instead of coming to this cafe, and then I would not have been here now. But the sentence
doesn’t have to express that proposition, and in other contexts it might express a necessary
one. Suppose that God is necessarily in heaven at all times. Then when he utters “I am here
now” he expresses a necessary truth. This theistic example raises some additional questions,
but it has the virtue of making the basic problem clear in a simple fashion: indexical sen-
tences express different propositions in different contexts. The same sentence may express
a contingent proposition in one context, and a necessary one in another. But then what

788
 N ECESSI TY A N D ME A N I N G

explains the necessity of the proposition in the second context? It cannot be the meaning, or
analyticity, of the sentence, since these were not sufficient for necessity in the first context
(Russell, 2010).

3 Possible Worlds Accounts of Meaning

Some people are suspicious of the linguistic doctrine of necessary truth because they are
suspicious of analyticity (Quine, 1935, 1951, 1954; Russell, 2008). If you do not believe
in analyticity, then you will not think that necessity can be accounted for by appeal to
it and that will leave you with two options: find an alternative account, or be skeptical
about necessity.
Skepticism about necessity is difficult to argue against, but it has become an increas-
ingly uncomfortable position to hold. It looks likely that modality has an important role
to play in science and philosophy, the development of modal logics has lent modality a
veneer of respectability, and it is clear that talk of necessity and possibility is not limited
to philosophical and scientific discussion but is prevalent in ordinary conversation as
well. For these reasons skepticism about modality has significant costs.
Sometimes it is not a refutation of a skeptical view that tempts philosophers away
from it, but rather the anticipated fruitfulness of approaches that begin from less skepti-
cal assumptions. The work of Montague, Kamp, Lewis, Stalnaker and others has encour-
aged the idea that rather than giving an account of modality in terms of meaning, we
might instead give an account of meaning in terms of modality (Lewis, 1976; Thoma-
son, 1974; Stalnaker, 1999).
We might think of this as an enhancement of some of Frege’s views (Frege, 1892,
1918). Frege held that expressions—including entire sentences—had referents. The
referent of a singular term was an object, the referent of a whole sentence a truth-value.
The referents of complex expressions are determined by the referents of their parts,
so the referent of a unary predicate needs to be something that could be put together
with an object to yield a truth-value, and a natural choice is a function from objects
to truth-values. Along the same lines, first-order quantifiers are things that can be put
together with a unary-predicate to form a sentence and so their referents will be func-
tions from the referents of predicates to truth-values, that is functions, from (functions
from objects to truth-values) to truth-values.
This is a powerful framework. We started with just two basic types of thing—objects
and truth-values—as the extensions of singular terms and sentences, and then used the
machinery of set theory to generate complex, non-basic types to serve as extensions for
other kinds of expression.
But the idea runs into problems when we consider sentences that contain that-clauses,
most famously, propositional attitude ascriptions such as:

(25) Mary believes that snow is white.

(26) Mary believes that grass is green.

Given what we have said so far, the contribution that the italicised parts make to the
extensions of sentences (25) and (26) should be the same, since the extension of both
snow is white and grass is green is the truth-value true, and given that the containing
sentences are the same in other respects, (25) and (26) must have the same truth-value.
Yet, as Frege realised, that is the wrong result: the truth-values of (25) and (26) could
be different.

789
 G I LLI A N RU S S E L L

His solution was to say that expressions had two kinds of referent, an ordinary direct
referent and an indirect referent which it would take in certain contexts, such as attitude
ascriptions. Frege held that the indirect referent of an expression is its sense. Frege’s
senses were introduced for reasons independent of this problem (Dummett, 1981), and
they play many roles—for example, the sense of a sentence is the thought (or proposi-
tion) it expresses (Salmon, 1982). But what is critical to the solution to the present
problem is that grass is green and snow is white have different senses and hence different
indirect referents, which allows the truth-values of (25) and (26) to come apart.
We could think of Frege’s solution to the problem as involving the introduction of
a new basic type—senses. So far, so good. What is a sense? (Or using less Fregean ter-
minology, a propositional content?) For Frege they were denizens of the third realm,
neither empirical objects, nor subjective ideas, but eternal abstracta (Frege, 1918). Not
everyone has found this satisfactory. But even if one has no problem with it on grounds
of parsimony, it is understandable that we should want to say more about such things if
they are to be put to serious explanatory use (Davidson, 1967; Lewis, 1976).
Although Frege didn’t pursue this line himself (and had little interest in necessity or
possible worlds) a popular approach is to characterise content as a function from pos-
sible worlds to extensions. (It invites problems to use the multiply connoting senses for
such things, so let’s call them intensions.) What we will add to our list of basic types then
is not unanalysed senses, but possible worlds. Possible worlds are ways the world could be,
but what they are beyond that is contentious. For Lewis they were concrete universes,
as real as our own. For others they are properties (Stalnaker, 1976) (Soames, 2001, 23)
or fictions (Rosen, 1990) and there are even those who hold that for present purposes,
figuring out what possible worlds are doesn’t matter; Stalnaker characterises the deci-
sion to theorise using possible worlds as simply a decision to theorise at a certain level of
abstraction (Stalnaker, 1999, 79).
Yet one could be forgiven for thinking that this is not really moving us in the right
direction. It is true that meanings or senses are in need of a philosophical account, but
it is not obviously helping to explain them in terms of possible worlds. What the mean-
ing of the sentence Theatetus flies is might be a little obscure, but it’s familiar enough
an idea—certainly meaning is a part of the ordinary, everyday talk of the folk—and it
is rather surprising to learn of such a familiar thing that it is in fact a function from pos-
sible worlds to extensions.
But to admit this is perhaps only to admit to having little native sympathy with the
thesis that meanings are more mysterious that possibilities. Native sympathies should be
relinquished where there is good reason, and the failure of the linguistic doctrine sug-
gests that it is worth considering a rival approach and, in fact as we will see, the project
of possible worlds semantics has seen considerable success.
Adding possible worlds to our list of basic types adds many non-basic types, such as
functions from possible worlds to truth-values and functions from functions-from-pos-
sible-worlds-to-truth-values to truth-values. Descriptions like this are hard to parse and
some formalism is useful for referring to functions that can take other functions as their
arguments and values. We follow convention in using the lowercase letter e to stand for
the basic type objects (you can think of it as e for entity) and lowercase t for the basic type
truth-values. If V and W are both types, then VW is the type of functions from exten-
sions of the type of V to extensions of the type of W. For example, e, t is the type of
functions from objects to truth-values (intransitive verbs have extensions of this type)
and e, t, , t is the type of functions from [functions from objects to truth-values] to

790
 N ECESSI TY A N D ME A N I N G

truth-values—the type of quantifiers. The intension of a name will be of type w, e , that

of a one-place predicate is w, e, t and that of a sentence w, t .
We can now draw some useful distinctions that we were unable to draw before. Here
are three examples:
Some pairs of non-synonymous predicates are co-extensional—they are satisfied by
all the same objects. A famous example uses the words renate and cordate (Quine, 1986).
Since all animals with kidneys have hearts, and vice versa, these two expressions cor-
respond to the same function from objects to truth-values e, t . But our possible worlds
machinery allows us to tell them apart: the intension of renate is not its extension, but
a function of the type w, e, t . Since there are possible worlds in which there are ani-
mals which are renates but not cordates, there are pairs of objects and possible worlds for
which the intension of renate yields true, but the intension of cordate yields false. Hence
these are not the same function—the two words have different meanings, as desired.
Another advantage of using possible worlds to characterise meanings is that they
allow us to distinguish rigid from non-rigid designators. A rigid designator is an expres-
sion that has the same denotation with respect to every possible world (Kripke, 1980).
This property does important work in explaining the characteristics of certain sentences
and expressions (for example in explaining why identity statements between co-refer-
ring names express necessary truths) but it is an intensional property, and without refer-
ence to modal notions like possible worlds we cannot distinguish rigid from non-rigid
designators: the rigid designators are exactly the ones whose intension is a constant func-
tion from possible worlds to objects.
A third advantage, of course, is that we can use intensions to distinguish the indirect
referents of sentences embedded in propositional attitude ascriptions—even where the
embedded sentences share a truth-value—and hence avoid the unwelcome conclusion
that the truth-values of (25) and (26) must be the same if the truth-values of snow is
white and grass is green are the same.

Problems for Possible Worlds Accounts

The benefits of possible world semantics are significant, but the simple approach out-
lined above faces problems that are strikingly similar to the ones it solves (Kaplan,
1989b), (Soames, 1998), (Soames, 2004) (King, 2005). Consider these correlates of
(25) and (26):

(27) Mary believes that triangles have three sides.

(28) Mary believes that there is no highest prime.

Both embedded sentences express necessary truths and so the indirect referent of each—
on the theory we are considering—is the constant function from each possible world to
the true. Since every other expression in (27) and (28) is shared between the sentences,
our theory holds that the two belief ascriptions share a truth-value. But this is wrong.
Mary might believe that triangles have three sides, but not yet have studied sufficient
arithmetic to have any belief about whether or not there is a highest prime.
There is also a problem with necessarily co-extensional predicates. All and only
quadrangles are quadralaterals, and this is not merely contingently the case, but neces-
sarily. Given this, the words quadrangle and quadralateral are predicted to have the same
content: the function that maps each quadralateral in each possible world to the true

791
 G I LLI A N RU S S E L L

and everything else to the false. Intuitively, however, they express different properties:
a quadrangle is a figure with four corners, whereas a quadralateral is a figure with four
sides; it is a matter of geometrical fact, rather than a mere consequence of their mean-
ings, that the two properties are necessarily co-extensive.
Finally, our theory is not fine-grained enough to distinguish directly referential
expressions from mere rigid designators (Kaplan, 1989a; King, 2005; Russell, 2008).
A directly referential expression is one whose extension is not determined by its con-
tent. Neo-Millian names provide one of the clearest examples. The referent of a neo-
Millian name like Mary might be established via an initial baptism, and then passed
on from speaker to speaker through conversation (the details are non-trivial but not
critical here) but when a speaker utters a sentence like Mary is tall, what the name
contributes to the proposition that gets expressed is not some descriptive information
such as that expressed by ‘the person baptised at such and such a date’ or ‘the firstborn
of Mr and Mrs Green’ but rather Mary herself (making the proposition expressed a
Russellian one).
Such expressions are rigid designators. Since the referent of the name is already con-
tained in the content, there is no opportunity for the content to select different referents
from different possible worlds. By contrast, the description the person baptised at such and
such a time has a descriptive content which will pick out different people given different
possible worlds (Kaplan, 1989a). Descriptions can be rigidified, for example by using the
operators actual or dthat. Yet intuitively the contents of Mary and the actual person baptised
at such and such a time are different, since one is a person, and the other not. If we char-
acterise content using functions from possible worlds to extensions, we leave something
out. There are several ingenious approaches to making possible worlds accounts more
fine-grained in order to respond problems like those above, including the use of:

1. situations (incomplete possible worlds) (Barwise and Perry, 1998)

2. impossible worlds (inconsistent possible worlds) (Takashi, 1988; Lewis,
1998)
3. two-dimensional semantics and epistemically possible worlds (Chalmers,
2004; Soames, 2004)
4. combination possible worlds/structured propositions approaches (Lewis,
1976) (King, 2005)

But three facts should make us cautious if all of these accounts characterise content as a
function from possible worlds-type objects to extensions. The first is that it is hard to see
how they will be able to distinguish rigidity from direct reference, since two terms may
be associated with the same function from (appropriately conceived) possible worlds-type
objects to extensions, whilst one does so via a function that is a part of its content, and
another does so directly. The second is that the problems the simple accounts ran into were
similar to the problems they were intended to solve. Such ‘revenge’ problems suggest that
the attempted solution was never adequate in the first place; an adequate solution ought
to solve these kinds of problems together. Finally, one may also suspect that some of the
techniques that these solutions add to possible world semantics—such as the structuring of
propositions—were sufficient to solve the problems on their own, without the addition of
possible worlds.

792
 N ECESSI TY A N D ME A N I N G

4 The independence of meaning and modality

From here, the Vienna Circle’s time seems like an austere one. Today’s philosopher tends
to be more liberal about his metaphysical commitments, and concurrently more depend-
able when it comes to drawing certain distinctions. The result has been a decline in the felt
strength of the motivation for the linguistic doctrine of necessary truth, and an increase in
the appreciation of the problems that the doctrine faces. But something of austerity of the
positivists’ spirit remains, this time in the reluctance to treat meanings as respectable in
their own right, and the resulting push to account for them in modal terms. But whereas
meanings often determine functions from possible worlds to extensions, the austere thesis
that functions from possible worlds to extensions is all there is to meanings looks increasingly
inadequate.

References
Ayer, A. J. (1990/1936). Language, Truth and Logic. Penguin Books, Harmondsworth.
Barwise, J. and Perry, J. (1998). Situations and Attitudes. CSLI, Stanford, CA.
Burgess, J. P. (1997). Quinus ab omni nævo vindicatus. Canadian Journal of Philosophy, Supplementary Volume
23: Meaning and Reference.
Carnap, R. (1958). Empiricism, Semantics and Ontology, pages 205–221. The University of Chicago Press,
Chicago and London, 2nd edition.
Cartwright, R. (1962). Propositions. In Butler, R., editor, Analytical Philosophy, pages 81–103. Blackwell,
Oxford.
Chalmers, D. J. (2004). The foundations of two-dimensional semantics. In Garcia-Caprintero, M. and Macia,
J., editors, Two-Dimensional Semantics: Foundations and Applications. Oxford University Press, Oxford.
Davidson, D. (1985/1967). Truth and meaning. In Martinich, A. P., editor, The Philosophy of Language, pages
98–121. Oxford University Press, Oxford, 4th edition.
Dummett, M. (1981). Frege: Philosophy of Language. Harvard University Press, Cambridge, Massachusetts,
second edition.
Follesdal, D. (1969). Quine on modality. In Davidson, D. and Hintikka, J., editors, Words and Objections:
Essays on the Work of W.V. Quine, pages 175–185. Reidel, Dordrecht.
Frege, G. (1956/1918). The Thought. Mind, 65(259):289–311.
Frege, G. (1985/1892). On Sense and Nominatum. In Martinich, A. P., editor, The Philosophy of Language.
Oxford University Press, Oxford, 4th edition.
Goodman, N. (1955). Fact, Fiction and Forecast. Harvard University Press, Cambridge, Mass.
Hempel, C. G. (1964/1945). On the nature of mathematical truth. In Benacerraf, P. and Putnam, H., edi-
tors, Philosophy of Mathematics: Selected Readings, pages 85–112. Prentice-Hall Inc, Englewood Cliffs, New
Jersey.
Kaplan, D. (1989a). Afterthoughts. In Almog, J., Perry, J., and Wettstein, H., editors, Themes from Kaplan.
Oxford University Press, New York.
Kaplan, D. (1989b). Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of
demonstratives. In Almog, J., Perry, J., and Wettstein, H., editors, Themes from Kaplan. Oxford University
Press, New York.
King, J. C. (Summer 2005). Structured propositions. In Zalta, E. N., editor, The Stanford Encyclopedia of
Philosophy. The Metaphysics Research Lab, Stanford University.
Kripke, S. A. (1980). Naming and Necessity. Blackwell, Oxford.
Lewis, D. (1976). General semantics. In Partee, B., editor, Montague Grammar, pages 1–50. Academic Press,
New York.
Lewis, D. (1998). Relevant Implication, chapter 8, pages 111–124. Cambridge University Press, Cambridge.
Quine, W. V. O. (1935). Truth by convention. In The Ways of Paradox and other essays, chapter 9, pages
70–99. Random House, New York.
Quine, W. V. O. (1951). Two dogmas of empiricism. Philosophical Review, 60:20–43.
Quine, W. V. O. (1953). Reference and modality, chapter 8. Harvard University Press, Cambridge, MA.
Quine, W. V. O. (1954). Carnap and logical truth. In The Ways of Paradox and other essays, chapter 10, pages
100–125. Random House, New York.

793
 G I LLI A N RU S S E L L

Quine, W. V. O. (1966). Three grades of modal involvement, chapter 15. Harvard University Press, Cambridge,
MA.
Quine, W. V. O. (1986). Philosophy of Logic. Harvard University Press, Cambridge, Mass.
Rosen, G. (1990). Modal fictionalism. Mind, 99(395):327–354.
Russell, G. (2008). Truth in Virtue of Meaning: A Defence of the Analytic/Synthetic Distinction. Oxford Uni-
versity Press, Oxford.
Russell, G. (2010). A new problem for the lingusitic doctrine of necessary truth. In New Waves in Truth,
pages 267–281. Palgrave MacMillan.
Salmon, N. (1982). Reference and Essence. Blackwell, Oxford.
Sidelle, A. (1989). Necessity, Essence and Individuation: A Defence of Conventionalism. Cornell University
Press, Ithica and London.
Soames, S. (1998). Facts, truth conditions and the skeptical solution to the rule-following paradox. Lan-
guage, Mind and Ontology, 12:313–348.
Soames, S. (2001). Beyond Rigidity, chapter 9-11. Oxford University Press, Oxford.
Soames, S. (2004). Reference and Description: the case against two dimensionalism. Princeton University Press,
Oxford.
Stalnaker, R. (1999). Context and Content. Oxford University Press, Oxford.
Stalnaker, R. C. (1976). Possible worlds. Nos, 10(1):65–75.
Stein, H. (1992). Was Carnap entirely wrong, after all? Synthese, 93:275–295.
Takashi, Y. (1988). Beyond possible worlds. Philosophical Studies, 53:175–204.
Thomason, R. (1974). Introduction. In Thomason, R., editor, Formal Philosophy: Selected Papers of Richard
Montague, pages 1–69. Yale University Press, New Haven.
Yablo, S. (1992). Review of Necessity, Essence, and Individuation: A defense of conventionalism. by Alan
Sidelle. The Philosophical Review, 101(4):878–881.

794
 6.7
P ROP OS I T IONAL
A T T ITUD E R EPOR T S
David Shier

Propositional attitudes are cognitive states such as believing, desiring, doubting, and
hoping. Propositional attitude reports (or ascriptions)—i.e., sentences reporting the
propositional attitudes of individuals—are central to our psychological discourse and to
our understanding of the world, since in order to explain and predict behavior, we must
appeal to information about the beliefs, desires, etc. of ourselves and others.
These attitude reports (often generically called “belief reports”) are of special inter-
est to philosophers of language because a number of particularly challenging problems
arise in connection with these linguistic constructions, including (among others) issues
about the substitution of co-referring terms into attitude reports, the rationality of
belief and other attitude states, the status of so-called de re thoughts, and cognitive
anti-individualism.
The sentences we use to report propositional attitudes come in various forms, includ-
ing sentences such as the following (where ‘believes’ can be replaced by other cognitive
attitude verbs).

1. Faith believes that koalas are herbivorous

2. Madeline believes Zorn’s Lemma
3. Isabel believes Phyllis

While these are all attitude reports, the philosophical focus is primarily on locutions
such as (1) that characterize the contents of the reported beliefs, especially when the
report is constructed from a subject, a propositional attitude verb, and an embedded
declarative sentence prefaced with the complementizer ‘that’. Discussion focuses on
such sentences both because they are so central to our explanations and predictions of
behavior and because the aforementioned puzzles about attitude reports arise primarily
with respect to reports of this sort.

The Propositional Theory

Since the pioneering work of Gottlob Frege (1892), the received account of propo-
sitional attitudes has been what might be called The Propositional Theory (its ortho-
doxy reflected in the very label “propositional attitude”). It comprises the following two
claims.
 DAVI D SH I E R

• Relational Account of Attitudes: Propositional attitudes are relations that indi-

viduals can bear to propositions. (Propositions are also typically taken to be the
individuating contents of beliefs, so that to share a belief is to believe the same
proposition.)
• Relational Analysis of Attitude Reports: A (de dicto*) attitude report of the form
‘S believes (desires, hopes, etc.) that P’ ascribes to S the relational property of
bearing the relevant attitude (belief, etc.) to the proposition expressed by the
embedded sentence of its ‘that’-clause. (‘That’-clauses are also typically taken to
be referring expressions designating the propositions expressed by their embedded
sentences, thus making the logical form of the attitude report straightforwardly
relational.)

* See “De Re/De Dicto Distinctions” below.

The Propositional Theory provides an integrated, plausible and rather robust explana-
tion of many of the facts about cognitive states and the sentences we report them with.
Discourse about beliefs, etc. strongly suggests a commitment to abstract entities like
propositions as the contents of these states. For example, Existential Generalization
appears to apply validly to attitude reports, as in the inferences (4)-(5) and (6)-(8).
And as Jerry Fodor (1978) says, “EG may not be criterial for ontological commitment,
but surely it is a straw in the wind.”

4. Phyllis believes that three is prime and four is composite

5. There is something that Phyllis believes
6. Bernard believes that virtue is its own reward
7. Sophie believes that virtue is its own reward
8. Therefore, there is something that Bernard and Sophie both believe

Furthermore, what Phyllis believes must have a number of important semantic proper-
ties such as being true, being about three and four, entailing that three is prime, etc.—the
very sorts of semantic properties that propositions are standardly thought to have. The
Relational Theory’s appeal to propositions also straightforwardly explains the validity
of various other common inferences, such as the following.

9. Daisy said that squirrels are omnivorous

10. Otis believes what Daisy said
11. Therefore, Otis believes that squirrels are omnivorous

Propositionalism is not without its challengers (e.g., the sententialism of Rudolf Car-
nap (1947), W.V.O. Quine (1956), and Donald Davidson (1968), as well as Michael
McKinsey’s (1986, 1999) property theory and Kent Bach’s (1997) descriptivism). Fur-
thermore, the nature of propositions is itself the subject of debate, for example as to
whether they are best conceived as highly structured pieces of information or as sets
of possible worlds, although the latter view faces significant difficulties, summarized
nicely by Soames (1987: 197–200). Nevertheless, Propositionalism and its close vari-
ants dominate the literature.

796
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

Frege
Among the most challenging puzzles about cognitive attitude reports are those involv-
ing the substitution of co-referring terms into ‘that’-clauses. Efforts to resolve these have
played a prominent role in shaping theories of the attitudes as well as semantic theories
of various parts of speech, such as names.
As Frege (1892) realized, if the truth value of a sentence is a function of what its
constituent expressions refer to, then substituting one expression for another referring
to the same item should not produce a sentence with a different truth value. (Jennifer
Saul (2007), though, has argued that substitution puzzles can arise outside of attitude
reports, even in simple sentences.)
For example, consider the Spider-Man comics (and treat them as if they were factual
for present purposes). Since ‘Spider-Man’ and ‘Peter Parker’ refer to the same person,
and (12) is true, then it seems (13) must be true as well.

12. Peter Parker lives in New York

13. Spider-Man lives in New York

But Frege also realized that certain linguistic contexts seem not to permit such substi-
tution salva veritatae of co-refererring expressions. Prominent among these so-called
opaque contexts are the ‘that’-clauses of some propositional attitude reports. For exam-
ple, Peter’s boss is unaware of Spider-Man’s secret identity, so even though (14) is true,
there are strong reasons for thinking (15) must be false. (Among these reasons is the fact
that Jameson himself would not assent to ‘Peter Parker is a vigilante’.)

14. Jameson believes that Spider-Man is a vigilante

15. Jameson believes that Peter Parker is a vigilante

In addressing the puzzle he discovered about informativeness (e.g., how can a true state-
ment of the form ‘a=b’ be informative when ‘a=a’ is not?), Frege famously concluded
that there must be two dimensions to an expression’s meaning –its Bedeutung or refer-
ence (the entity singled out) and its Sinn or sense (a way of singling out the referent).
Frege’s solution to the problem of substitution in attitude reports also makes use of his
theory of Sinn and Bedeutung by postulating that inside the scope of a cognitive attitude
verb, the reference of an expression shifts from its customary reference to its customary
sense.
Thus, on the Fregean view, it is not problematic that (14) and (15) should diverge
in truth value, because the embedded tokens of the names ‘Spider-Man’ and ‘Peter
Parker’ are not co-referential and instead refer to two distinct senses (the web-slinging
costume-wearing way of picking him out and the young photographer way). It is worth
noting that on Frege’s view, then, substitution of co-referentials salva veritatae techni-
cally only appears to fail, since the name tokens inside ‘that’-clauses are not actually
co-referential.
Frege’s solution, while elegant, faces serious difficulties. Among the major problems
specific to the Fregean analysis of attitude reports is one involving anaphoric pronouns,
such as the occurrence of ‘he’ in (16).

16. Jameson believes that Peter Parker is cowardly, but he really isn’t

797
 DAVI D SH I E R

On Frege’s analysis of attitude reports, the token of ‘Peter Parker’ in (16) refers not to
Peter himself, but to a sense—a way of picking out Peter. Since the anaphoric pronoun
‘he’ inherits its reference from the token of the name, the Fregean analysis would say
that the pronoun also refers to this sense, which is plainly wrong, since it surely isn’t the
sense of ‘Peter Parker’ who is said not to be cowardly, but Peter himself.
Furthermore, the Fregean analysis violates the intuitively plausible principle of
Semantic Innocence, according to which the semantic contribution made by a term is
the same whether or not the term is embedded in a ‘that’-clause.

Russell
Russell’s (1905 and 1912) solution to substitution problems originates in his Theory of
Descriptions and differs significantly from Frege’s. While Russell, unlike Frege, held that
logically proper names are genuinely singular terms (i.e., terms that are directly referen-
tial in the sense that they contribute just their referents to the propositions expressed
by sentences containing them), he argued that other types of terms, including definite
descriptions, could not be genuinely singular. One of the lines of thought that led him
to this view had to do with substitution failure.
Consider attitude reports with descriptions in their ‘that’-clauses. Even though the
author of Nobody’s Fool (Richard Russo) is the author of Straight Man, it is entirely pos-
sible for (17) and (18) to have different truth values.

17. Otis believes that the author of Nobody’s Fool won the Pulitzer
18. Otis believes that the author of Straight Man won the Pulitzer

Russell’s explanation of substitution failure, unlike Frege’s, did not entail that terms
in attitude reports function differently than elsewhere. Instead Russell inferred from
substitution failure (and other phenomena) that sentences containing descriptions
have unexpectedly complex logical forms. He famously argued that despite their surface
grammatical appearances, sentences of the form The F is G do not have simple subject-
predicate form and instead have underlying logical forms involving multiple quantifiers,
with the following truth conditions.

19. There is at least one F, there is at most one F, and whatever is F is G

Thus, on Russell’s Theory of Descriptions, the propositions expressed by the ‘that’-

clauses of (17) and (18) are given by the following—where ‘Nx,’ ‘Sx’ and ‘Px’ abbrevi-
ate ‘x authored Nobody’s Fool’ ‘x authored Straight Man’ and ‘x won the Pulitzer.’

20. x (Nx & y(Nyoy=x) & Px )

21. x (Sx & y(Syoy=x) & Px )

Therefore (17) and (18) correspond to (22) and (23), where ‘<S>’ indicates the propo-
sition expressed by ‘S’.

22. Believes (Otis, <x (Nx & y(Nyoy=x) & Px )>)

23. Believes (Otis, <x (Sx & y(Syoy=x) & Px )>)

798
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

But (22) and (23) clearly report belief relations to entirely different propositions, so
there is no reason to expect them to share truth values. Thus Russell neatly resolves the
substitution problem, at least with regard to descriptions.
It should be noted that Russell held a view somewhat more complicated than the
straightforward Relational Analysis. Russell (1912) stated that belief reports do not
ascribe binary relations between believers and propositions, but rather report more com-
plex relations among the believers and the constituents of the proposition believed. In
Russell’s example, Othello’s belief that Desdemona loves Cassio is a relation among
Othello, Desdemona, Cassio, and the relationship of loving. But this complication can
be ignored for our present purposes.
It should also be noted that while (17) and (18) are interpreted above in the most
natural way (with the descriptions understood as characterizing how Otis thinks of the
described individual, and with the associated quantifiers, etc. occurring in small scope,
or secondary occurrence, as Russell called it, relative to the belief operator), there are atti-
tude reports—including less-natural readings of (17) and (18)—in which descriptions
take large scope (or primary occurrence, as Russell called it) relative to belief operators.
These reports would have logical forms such as those in (24) and (25), with respect to
which the substitution of co-referring descriptions is uncontroversially valid. Sentence
(24) states that the author of Nobody’s Fool is such that Otis believes him to be a Pulitzer
winner, and (25) states that the author of Straight Man is such that Otis believes him to
be a Pulitzer winner. (For further discussion, see “De re/de dicto Distinctions” below.)

24. x [Nx & y(Nyoy=x) & Believes (Otis, <Px>)]

25. x [Sx & y(Syoy=x) & Believes (Otis, <Px>)]

Let us now consider how Russell analyzed attitude reports with proper names occurring
in small scope, inside the scope of the cognitive attitude verbs.

14. Jameson believes that Spider-Man is a menace to New York

15. Jameson believes that Peter Parker is a menace to New York

If names like ‘Peter Parker’ and ‘Spider-Man’ function as genuine singular terms, as
Russell thought logically proper names do, he would have to regard (14) and (15) as
expressing the same proposition and hence the same in truth value.
Russell’s (1912) surprising solution was not to challenge the Relational Analysis or
the claim that logically proper names are genuine singular terms. Rather, he postulated
that ordinary proper names are simply not logically proper names and that they function
instead as if they go proxy for definite descriptions. So the embedded sentences of (14)
and (15) would express propositions like those expressed by (26) and (27).

26. The web-slinging, spider-costume-wearer is a menace to New York

27. The young photographer who works for the Bugle is a menace to New York

Therefore, just as with definite descriptions, the belief reports (14) and (15) would
attribute to Jameson belief of very different propositions, thus explaining how their
truth values can diverge.
Russell’s solution (often labeled The Description Theory of Names) has the virtue of being
well integrated into his broader philosophical views. It is a key part of his solutions to

799
 DAVI D SH I E R

several other important problems, such as Frege’s puzzle about informativeness, problems
about negative existence claims, and issues about non-referring names. Additionally, it is
well motivated by his epistemological commitments, such as his Principle of Acquaint-
ance which states that sentences we understand must express propositions composed only
of items we can directly apprehend—i.e., items of our immediate acquaintance. Accord-
ing to Russell (1912), the only sorts of things with which we have this acquaintance are
properties and relations, our own sense data, and (perhaps) ourselves. Therefore, a proper
name of any other sort of item (such as an external physical object or another person) can-
not contribute its referent to the proposition, for we could not understand the propositions
expressed by such sentences, and clearly we do. But he held we can be directly acquainted
with the properties and relations that descriptions contribute to the propositions.
Russell’s solution is superior to Frege’s in several respects. For instance, it does not
violate Semantic Innocence, since on his view names and other terms inside ‘that’-
clauses make the same contributions to propositions that they make elsewhere. Nei-
ther does Russell’s solution face the problems Frege’s does in accounting for anaphoric
pronouns.
Nonetheless, there are persuasive reasons for rejecting Russell’s Description Theory
of Names. Principal among these are arguments advanced by Saul Kripke (1972), Keith
Donnellan (1970), and others against the description theory of names and in support
of the Direct Reference semantic theory that supplanted it. The two main approaches
to propositional attitude reports in the wake of the Direct Reference theory are Naïve
Russellianism and The Hidden-Indexical View.

Naïve Russellianism
Naive-Russellianism—so-called because it resembles Russell’s early semantic views—
holds that ordinary proper names and indexicals normally function as Russell thought
logically proper names did—i.e., as genuine singular terms, contributing only their ref-
erents to the expressed propositions, even when these terms designate external contin-
gent objects.
Naive-Russellianism furthermore holds that attitude reports with singular ‘that’-
clauses report attitude states with those same singular propositions as their contents.
Consider pairs like (14) and (15) once more.

14. Jameson believes that Spider-Man is a vigilante

15. Jameson believes that Peter Parker is a vigilante

According to the Naïve Russellian, (14) and (15) express the same proposition; they
both report that Jameson bears the belief relation to the singular proposition that
attributes the property of vigilantism directly to that individual. Consequently, Naïve
Russellianism is committed to the counter-intuitive view that substitution of co-refer-
ential terms is actually valid. Given that a name is a genuine singular term, its semantic
value is exhausted by its referent; thus co-referential names share their semantic values.
So, assuming the Compositionality principle that the semantic value of an expression
is a function of the semantic values of its component expressions, co-referential names
and indexicals must be mutually substitutable in other expressions like sentences.
Thus the Naïve Russellian must explain away the widespread conviction that substi-
tution into small scope positions in attitude reports is not valid. As Mark Richard (1990,

800
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

p.125) says, it would require “bribery, threats, hypnosis, or the like” to convince most
people” that a pair such as (14) and (15) report the same belief.
The characteristic Naïve Russellian rejoinder—from, e.g., from Tom McKay (1981),
Nathan Salmon (1986), and Scott Soames (1987, 1988 and 2002)—is that while (14)
and (15) semantically express the same proposition, they pragmatically convey different
propositions. Hearers conflate the pragmatically and semantically conveyed informa-
tion and take the two reports to express different propositions. (Not all Naïve Russel-
lians offer such pragmatic explanations. David Braun (e.g. 1998, 2001a and 2001b) is a
notable exception.)
Naïve Russellians have suggested various sorts of information to play this pragmatic
role. For example, utterances of (14) and (15) might pragmatically convey the informa-
tion expressed by (28) and (29). And, since the former is true and the latter is false,
hearers may conclude that (14) is true while (15) is false.

28. Jameson would assent to the sentence ‘Spider-Man’ is a vigilante’

29. Jameson would assent to the sentence ‘Peter Parker is a vigilante’

Salmon, Soames, and most other Naïve Russellians have proposed that propositional
attitudes are mediated by “ways of apprehending” the proposition, or propositional
“guises,” and have furthermore proposed that information about the subject’s way of
taking the proposition can be pragmatically conveyed by an utterance of an attitude
report. That is, they distinguish what someone believes (a proposition) from how he or
she believes it (a way of taking the proposition), and they assert that an attitude report
semantically conveys the “what” and only pragmatically conveys the “how.”
For example, Salmon (1986) proposes that while the ordinary language operator
‘believes’ semantically expresses a dyadic relation between a believer and a proposition,
the psychological reality underwriting the truth of the dyadic statements is a triadic
relation (he calls it ‘BEL’) between a believer, a proposition, and a propositional guise.
So while (14) and (15) both semantically convey that Jameson believes the same singu-
lar proposition, (14) pragmatically conveys that Jameson believes it under ‘Spider-Man’
ways of thinking, while (15) pragmatically conveys that he believes it under ‘Peter-
Parker’ ways of thinking. And this information—that Jameson believes it under the
‘Peter-Parker’ ways, really is false; hence the mistaken, though understandable, belief
that (15) is false.
The pragmatic solutions offered by Naïve Russellians face significant difficulties. It
is implausible that ordinary speakers are normally reasoning by way of meta-linguis-
tic information or information about the believer’s ways of taking propositions. There
are difficulties as well in spelling out the precise pragmatic mechanism by which the
information is allegedly conveyed. For example, Gricean conversational implicature
has been proposed—e.g., by Salmon (1986) and by Kirk Ludwig (1996)—but Stephen
Schiffer (1987) and others, have raised serious objections for this strategy. The pri-
mary problem for Naïve Russellianism remains its counter-intuitive commitment to the
validity of substitution.

Hidden Indexical Theories

Hidden Indexical Theories of attitude reports—e.g., from Schiffer (1977), Mark
Crimmins and John Perry (1989), and Crimmins (1992a and 1992b)—offer a way of

801
 DAVI D SH I E R

accommodating the direct reference view that the ‘that’-clauses of attitude reports refer
to singular propositions while respecting the powerful anti-substitution intuitions. This
approach agrees with typical Naïve Russellians that attitudes toward propositions are
mediated by ways of apprehending them, but it differs by asserting that information
about the subject’s way of apprehending a proposition is semantically, not just pragmati-
cally, conveyed by an attitude report.
These theories derive their name from the fact that they posit context-sensitive
“unarticulated constituents”—i.e., elements of the propositions expressed by attitude
reports for which there are no explicit corresponding syntactical elements in the reports
themselves—that pertain to the subject’s way of taking the proposition. Thus a pair of
reports such as (14) and (15) could semantically express different propositions, despite
having semantically equivalent ‘that’-clauses—e.g., if (14) introduced Jameson’s
‘Spider-Man’ types of ways of apprehending the proposition expressed while (15) intro-
duced his ‘Peter-Parker’ ways of apprehending it.
Related views have been advanced by Mark Richard (1990), who postulates “Russellian
annotated matrices” (the constituents of which are pairs of linguistic entities and their
semantic values, such as names paired with their referents) as the contents of attitude
states, and by Francois Recanati (1993), who postulates “quasi singular” propositions as
contents. The term ‘Contextualism’ is sometimes used for these sorts of theories, but some-
times instead as a more general name to cover these as well as Hidden Indexical Theories.
Despite accommodating both Direct Reference and anti-substitution intuitions, Hid-
den Indexical Theories face substantive problems. Bach (1993) and Schiffer (1992)
have persuasively argued against the psychological reality of speakers making attitude
reports that include ways of apprehending propositions in their truth conditions. Fur-
thermore, the reliance on unarticulated constituents seems ad hoc. (See Jason Stanley
(2002) for example.) While these theories, like Naïve Russellianism, respect Semantic
Innocence, they do so only at the expense of the important principle of Compositional-
ity, since a pair of attitude reports such as (14) and (15) have different semantic value
despite all of their proper parts sharing their semantic values.

De Re/De Dicto Distinctions

The cognitive attitudes literature has supposed, often as a load-bearing part of its theo-
ries, a distinction between de re (“concerning the thing”) attitude reports and de dicto
(“concerning the word/sentence/proposition”) attitude reports.
Following Russell, this has usually been drawn as structural distinction in terms of
relative scope. In this sense, an attitude report is de dicto with respect to the terms (or
positions occupied by terms) occurring inside the scope of its cognitive attitude verb,
and de re with respect to the terms occurring outside that scope. We might call these
“structurally de dicto” and “structurally de re” occurrences.
Let us first consider reports containing descriptions. For example, with respect to
‘the dean’s partner,’ (30) would likely be interpreted as structurally de dicto and (31) as
structurally de re, since the description occurs in small scope in the surface grammar of
(30), and in large scope in (31). Furthermore, (31) is structurally de re with respect to
the anaphoric pronoun ‘he.’

30. Otis believes that the dean’s partner is a sociologist

31. The dean’s partner is such that Otis believes that he is a sociologist

802
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

Given Propositionalism and its Relational Analysis, (30) and (31) would have logical
forms suggested by the following partial analyses.

32. Believes (Otis, <the dean’s partner is a sociologist>)

33. x [x = the dean’s partner & Believes (Otis, <X is a sociologist>)]

Thus, structurally de dicto occurrences appear to function as part of the characteriza-

tion of the contents of the reported beliefs, while de re occurrences do not. Because the
description occurs in small scope in (32), its semantic value is contributed to the propo-
sition expressed by the ‘that’-clause and hence to the proposition that Otis is reported
to believe. The description in (33), having wide scope, does not contribute its semantic
value to the ‘that’-clause, and hence not to the reported belief content.
However, although it seems natural to read a sentence like (30) as structurally de dicto
with respect to the description, matters are more complicated. There is a plausible alter-
native interpretation of (30) according to which the description does not characterize
how Otis is thinking of the individual but only identifies the individual believed to be
a sociologist. For instance, if someone overheard Otis saying, “Today’s guest speaker is a
sociologist,” and she knew that the guest speaker was the dean’s partner, then she might
plausibly report this with an utterance of (30)—even if she thought Otis did not know
that the speaker was the dean’s partner. So it is evident that a “believes-that” sentence
such as (30) can be used to express either a structurally de dicto report or a structurally de
re report, although in most contexts the former would be more natural, and philosophers
of language tend to use “believes-that” sentences as if they were canonically de dicto.
Along with an expression’s role in content characterization, several other important
semantic differences appear to correspond to the structural de re/de dicto distinction. For
example, structurally de re occurrences of terms commit the speaker to the existence of
a referent, while structurally de dicto occurrences do not in general. In the structurally de
re (31), the speaker is using ‘the dean’s partner’ to refer to the object of Otis’s belief, and
is thus committed to there being a partner. But, a de dicto structure like (32) does not
necessarily commit the speaker to the existence of a referent; Otis can bear the belief
relation to the proposition that the dean’s partner is a sociologist even if the dean has no
partner. Similarly, it is not generally valid to apply Existential Generalization (i.e., to
“quantify in”) to structurally de dicto occurrences of descriptions, since from the truth of
(32), the truth of (33) does not follow. (Classic discussions of these issues can be found
in Quine (1956) and Kaplan (1969), among others.)
Furthermore, since a structurally de re occurrence of a term, being outside the ‘that’-
clause, does not contribute its semantic value to the proposition said to be believed,
then any report resulting from replacing the term with a co-referring expression should
share the truth value of the original report. That is, substitution of co-referring terms
should be valid. Such an occurrence is said to be transparent. A structurally de dicto
occurrence, being inside the scope of the ‘that’-clause, contributes to the belief’s propo-
sitional content. Therefore, only substitution of a term that shares its semantic value
(and not just its referent) can insure preservation of the truth value. That is, substitu-
tion can fail. Such an occurrence is said to be opaque.
To summarize: there is an ambiguity involving reports of the form ‘X believes that the
F is G’, which may best be explained as a structural ambiguity between large and small
scope occurrences of ‘the F’. (Some accounts, e.g. Quine’s, instead hold the attitude verbs
like ‘believes’ to be lexically ambiguous, positing two distinct belief operators.) With

803
 DAVI D SH I E R

respect to definite descriptions, at least, the structural distinction is arguably of a piece

with several semantic distinctions, at least on the assumptions of Propositionalism.
But only reports containing definite descriptions have been considered so far. Of
course, if names or other terms are construed along classic Russellian lines—i.e., as
going proxy for descriptions—there is no problem in extending this structural-and-
semantic distinction to reports in which names occur. But if ordinary names, indexi-
cals, etc. are construed as genuine singular terms, then the structural de re/de dicto dis-
tinction and corresponding semantic distinctions come apart from each other. Small
scope occurrences of genuine singular terms carry existential commitment, permit
EG and permit substitution. In short, they behave rather like large scope descriptions.
Consider (34).

34. Daisy believes that Obama is tall

On the Relational Analysis, a structurally de dicto reading of (a) attributes to Daisy the
property of believing the proposition expressed by ‘Obama is tall.’ Given direct refer-
ence semantics for names, that proposition contains Obama himself a constituent. So
the report has existential commitment, since there is no such proposition for Daisy to
believe if ‘Obama’ has no referent. For similar reasons, it appears that Existential Gen-
eralization is valid and one may quantify in to the position occupied by the name. For
instance, (34) entails (35).

35. x [x = Obama & Believes (Daisy, <X is tall>)]

Finally, since direct reference entails that the co-referential names make the same con-
tribution to propositions expressed (their referent), it follows that two ‘that’-clauses
like (14) and (15) differing only in substituted co-referential terms must semantically
express the same proposition, and—given the Relational Analysis—belief reports with
such that clauses must report the same belief. That is, substitution is valid for names.
(Of course, this last point is the subject of much debate. The fact that it would license
such substitution is frequently the basis of objections direct reference views such as
Naïve Russellianism.)

14. Jameson believes that Spider-Man is a vigilante

15. Jameson believes that Peter Parker is a vigilante

Thus there is no univocal answer as to whether a small scope occurrence of a genuine

singular term is de dicto or de re. We might say that such an occurrence is structurally de
dicto, but semantically de re.
Somewhat confusingly, in addition to the de re/de dicto distinction among attitude
reports, the cognitive attitude literature has also made a metaphysical distinction among
attitude states using the labels ‘de re’ and ‘de dicto.’ The distinction is intended to demar-
cate states that are about specific objects (de re) from those that are not (de dicto). For
example, in his well-known disambiguation of ‘X desires a sloop,’ Quine (1956) suggests
that on the one (“notional”) interpretation it ascribes a desire to own some particular
sloop, while on the other (“relational”) interpretation it ascribes a desire to own any
sloop whatever. The former, object-specific, desire would be a de re attitude, while the
latter (the general desire to relieve “slooplessness”) would be de dicto.

804
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

The special nature of such de re states remains controversial. It has frequently been
held to require some particularly close connection between the person and object—
e.g., Kaplan’s (1969) notion of being causally en rapport with an object, or Roderick
Chisholm’s (1976) notion of “epistemic intimacy.” Another common conception of
de re thoughts is that they are attitude states with a special sort of propositional content,
e.g. genuinely singular (“Russellian”) propositions. These two approaches can be com-
bined, if one holds that bearing an attitude to a genuinely singular proposition requires
a special relation to the singular term’s referent.
The two de re/de dicto distinctions are sometimes thought to be connected in some
systematic way—e.g., that de dicto ascriptions can be only be used to ascribe de dicto
thoughts, while de re ascriptions can be employed only to ascribe de re thoughts. But
there are good reasons for thinking that the two distinctions do not line up so neatly.
(See Kent Bach (1987) for a good discussion of such reasons.) Therefore, it is impor-
tant to be clear about whether with the labels ‘de re’ and ‘dicto’ one is characterizing
the reports or the reported attitudes. As Bach (1987: 17) noted, “we must be careful to
distinguish de-re-belief ascriptions from de re belief-ascriptions and de-dicto-belief ascrip-
tions from de dicto belief-ascriptions.”

Kripke’s Puzzle About Belief

Kripke (1979) introduces an important puzzle that resembles substitution puzzles in
some respects, but which does not appear to depend on the assumption of a direct refer-
ence account of nouns or involve substitution of terms into attitude reports.
Pierre, initially a French-only speaker, reads a book showing lovely pictures of
London captioned as ‘Londres’ and thus becomes disposed to assent to the sentence
‘Londres est jolie.’ Kripke concludes, via plausible principles of translation and dis-
quotation (roughly, that assent to sentence S entails belief that S), that Pierre believes
that London is pretty. Pierre then moves to a shabby part of London, where he picks
up English but somehow has no idea that the city he lives in and knows by the name
‘London’ is the city called ‘Londres.’ He is disposed to assent to ‘London is not pretty’
and (again by way of a disquotation principle) Kripke deduces that Pierre believes that
London is not pretty. So both of the following appear true, despite ascribing beliefs with
contradictory contents.

36. Pierre believes that London is pretty

37. Pierre believes that London is not pretty

But, says Kripke, Pierre cannot have genuinely contradictory beliefs, since anyone is in
principle able to detect inconsistent beliefs if he has them, and (as long as he is unaware
that the city called ‘London’ is the one called ‘Londres’) there is no way that Pierre
could possibly recognize that the beliefs in question are inconsistent.
Kripke also constructs a related example that does not involve translation. Peter
believes that there is a pianist named ‘Paderewski’ and that there is a politician named
‘Paderewski,’ but does not believe they are the same individual. The following reports
about Peter are derived from the scenario—another alleged case of contradictory beliefs.

38. Peter believes that Paderewski had musical talent

39. Pierre believes that Paderewski had no musical talent

805
 DAVI D SH I E R

While Kripke’s puzzles must be explained by any account of propositional attitudes,

they pose special challenges for those that combine direct reference with the Propo-
sitional Theory—for instance, most forms of Naïve Russellianism. Bach (1997) and
Shier (1996) have argued that these puzzles are only problematic on the assumption of
the Relational Analysis of Attitude Reports. If one does not assume that the contents
of beliefs can be “read off” from their ‘that’-clauses, then there is no reason to think the
pairs above report belief of contradictory contents. (Brian Loar makes a similar point.)
Bach (1997: 233) points out that any ‘that’-clause can be used, in the right circum-
stances, to generate a Paderewski case, and uses this as the basis of an argument against
the Relational Analysis and its assumption (which he calls “The Specification Assump-
tion) that the ‘that’-clause specifies the complete propositional content of a belief.

Anti-individualism and Attitude Reports

Propositional attitudes and attitude reports also figure prominently in the Individu-
alism/Anti-Individualism (Internalism/Externalism) debates in philosophy of mind
and language. The influential lines of argument against Individualism about meanings
developed by Hilary Putnam (1975), Tyler Burge (1982), and others were extended, by
way of Propositionalism, to argue against cognitive individualism.
Cognitive Individualism is the view that a person’s beliefs and other propositional
attitude states supervene on just her internal features and are thus logically independent
of the features of the environment. This is often put as the claim that such states are
properly individuated by the subject’s narrow psychological properties where, following
Putnam (1975), a narrow property is (roughly) one that “does not presuppose the exist-
ence of any individual other than the subject to whom [it] is attributed.” According
to Cognitive Individualism, any two internally identical persons would share all their
attitude states, despite any differences in their environments.
Individualism is intuitively appealing; how could remote external objects and events
logically determine what one believes? Certainly they play causal roles, but they seem
able to do this only in a mediated way by impinging on the internal determinants of
mental states—i.e., by affecting how the world appears to the individual.
Despite Individualism’s intuitive plausibility, it has come under attack from philoso-
phers who hold the semantic thesis of Direct Reference and take it to entail the exist-
ence of propositional attitude states with wide content—i.e., content that essentially
involves contingent external objects. Consider a propositional attitude report contain-
ing an occurrence in its ‘that’-clause of a directly referential term for a contingent exter-
nal object.

40. Phyllis believes that a volcano is erupting on Jupiter

Given Direct Reference, this seems to ascribe a belief that essentially involves the planet
Jupiter itself—a belief that one simply could not have in a possible world in which Jupiter
never existed. (This is also related to one of the common conceptions of de re thoughts
(above)—that of thoughts with contents that essentially involve particulars.)
The existence of wide beliefs is not entailed by Direct Reference, but by conjoining
it with Propositionalism. (See McKinsey (1991 and 1994) for a thorough discussion of
this issue.) Given the Relational Analysis of Attitude Reports, (40) reports that Phyllis
bears the belief relation the proposition expressed by ‘a volcano is erupting on Jupiter.’

806
 PRO PO SI TI O N A L ATT I T U D E R E P O RT S

And, given the Relational Account of the Attitudes (and the usual further assumption
that the propositional content is individuating content), Phyllis’ belief is type-individu-
ated by that proposition; i.e., one can have the same belief just in case one believes
that proposition. Therefore, the belief-individuating proposition will include the planet
Jupiter itself, contrary to Individualism.
The recognition that the route from Direct Reference to Cognitive Anti-Individu-
alism runs through Propositionalism suggests a strategy for reconciling Individualism
and Direct Reference by challenging Propositionalism, which entails that a report with
a wide ‘that’-clause must report a wide belief. Some variations of this approach are
advanced by Simon Blackburn (1984), McKinsey (e.g. 1991 and 1994), Shier (1996)
and Bach (1997).

Related Topics
1.2 Semantics and Pragmatics
1.7 Meaning and Communication
2.1 Reference
2.3 Propositions
3.1 Names
3.6 Anaphora
3.7 Descriptions
3.11 Indexicals and Demonstratives
7.3 Modern Philosophy of Language
7.4 Frege, Russell, and Wittgenstein.

References
Bach, K. (1987) Thought and Reference, New York: Oxford University Press.
Bach, K. (1993) “Sometimes a great notion: A critical notice of Mark Crimmins’ Talk about Beliefs,” Mind
and Language 8: 431–441.
Bach, K. (1997) “Do belief reports report beliefs?,” Pacific Philosophical Quarterly 78: 215–241.
Blackburn, S. (1984) Spreading the Word, Oxford: Oxford University Press.
Braun, D. (1998) “Understanding Belief Reports,” Philosophical Review 107: 555–595.
Braun, D. (2001a) “Russellianism and Prediction,” Philosophical Studies 105: 59–105.
Braun, D. (2001b) “Russellianism and Explanation” Philosophical Perspectives 15: 253–289.
Burge, T. (1982) “Other Bodies,” in A. Woodfield (ed.) Thought and Object, Oxford: Oxford University
Press.
Carnap, R. (1947) Meaning and Necessity (2e), Chicago: University of Chicago Press.
Chisholm, R. (1976) “Knowledge and Belief: ‘De Dicto’ and ‘De Re’,” Philosophical Studies 29: 1–20.
Crimmins, M. (1992a) “Context in the Attitudes,” Linguistics and Philosophy 15: 185–198.
Crimmins, M. (1992b) Talk about Beliefs, Cambridge, MA: MIT Press.
Crimmins, M. and J. Perry (1989) “The Prince and the Phone Booth: Reporting Puzzling Beliefs,” Journal of
Philosophy 86: 685–711.
Davidson, D. (1968) “On Saying That,” Synthese 19: 130–146.
Donnellan, K. (1972) “Proper Names and Identifying Descriptions,” Synthese 21: 335–358.
Fodor, J. (1981), “Propositional Attitudes” in Representations: Philosophical Essays on the Foundations of Cog-
nitive Science, Cambridge, MA: MIT Press.
Frege, G. (1892) “On Sense and Meaning,” in P. Geach and M. Black (eds.) Translations of the Philosophical
Writings of Gottlob Frege (1952), Oxford: Blackwell.
Kaplan, D. (1969) “Quantifying In,” Synthese 19: 178–214.
Kaplan, D. (1977) “Demonstratives” in J. Almog, J. Perry, and H. Wettstein (eds.). Themes from Kaplan
(1989), Oxford: Oxford University Press.

807
 DAVI D SH I E R

Kripke, S. (1972) Naming and Necessity, in D. Davidson and G. Harman (eds.) Semantics of Natural Language,
Dordrecht: Reidel.
Kripke, S. (1979) “A Puzzle about Belief” in A. Margalit (ed.) Meaning and Use, Dordrecht: Reidel.
Loar, B. (1985) “Social Content and Psychological Content,” in R. Grimm and D. Merrill (eds.), Contents
of Thoughts, Tucson: The University of Arizona Press.
Ludwig, K. (1996) “Singular Thought and the Cartesian Theory of Mind,” Noûs 30: 434–460.
McKay, T. (1981) “On Proper Names in Belief Ascriptions,” Philosophical Studies 39: 287–303.
McKinsey, M. (1986) “Mental Anaphora,” Synthese 66: 159–75.
McKinsey, M. (1991) “Anti-Individualism and Privileged Access,” Analysis 51: 9–16.
McKinsey, M. (1994) “Individuating Beliefs,” Philosophical Perspectives 8: 303–330.
McKinsey, M. (1999) “The Semantics of Belief Ascriptions,” Noûs 33: 519–557.
Putnam, H. (1975) “The Meaning of ‘Meaning’,” in Philosophical Papers II: Mind, Language, and Reality,
Cambridge: Cambridge University Press.
Quine, W. (1956) “Quantifiers and Propositional Attitudes” Journal of Philosophy 53, 177–187.
Recanati, F. (1993) Direct Reference: From Language to Thought, Oxford: Blackwell.
Richard, M. (1990) Propositional Attitudes, Cambridge: Cambridge University Press.
Russell, B. (1905) “On Denoting” Mind 14: 479–493.
Russell, B. (1912) The Problems of Philosophy, Oxford: Oxford University Press.
Salmon, N. (1986) Frege’s Puzzle, Cambridge, MA: MIT Press.
Saul, J. (2007) Simple Sentences, Substitution, and Intuitions, Oxford: Oxford University Press.
Schiffer, S. (1977), “Naming and Knowing,” Midwest Studies in Philosophy 2: 28–41.
Schiffer, S. (1987) “The ‘Fido’-Fido Theory of Belief,” Philosophical Perspectives 1: 455–480.
Schiffer, S. (1992) “Belief ascription,” Journal of Philosophy 89: 499–521.
Shier, D. (1996) “Direct Reference for the Narrow-Minded,” Pacific Philosophical Quarterly 77: 225–248.
Soames, S. (1987) “Direct Reference, Propositional Attitudes, and Semantic Content,” Philosophical Topics
15: 47–87.
Soames, S. (1988) “Substitutivity,” in J. Thomson (ed.) On Being and Saying, Cambridge, MA: MIT Press.
Soames, S. (2002) Beyond Rigidity, Oxford: Oxford University Press.
Stanley, J. (2000) “Context and logical form” Linguistics and Philosophy 23: 391–434.

808
 Section VII

H IST O R I C A L
PER SPEC T I V E S
 7.1
A N C IE NT P H ILOSOPHY
OF L ANGUAGE
Luca Castagnoli and Ermelinda
Valentina Di Lascio

Ancient philosophers did not introduce a discipline or branch of philosophy called,

or straightforwardly identifiable as, “philosophy of language.” Nonetheless, they did
explore a number of topics which we would classify under this heading. This chapter
offers an introduction to the most important of these topics: the nature, origin, and
development of language; the relationship between language, thought, and reality; the-
ories of truth; speech-acts; parts of speech and other grammatical issues; linguistic ambi-
guity. There are a few other relevant topics with which ancient philosophers engaged
but which we have not included, not only for space limitations but also because the
ancients did not inquire into them in primarily language-related contexts: for example,
universals, modalities, logical form, ineffability (for broad introductions cf. Kneale and
Kneale 1962 and Mortley 1986). We have adopted a historical perspective, from the
Sophists to Augustine, focusing only on the major philosophers and schools, and high-
lighting thematic and doctrinal continuity and dialectical development.

The Sophists
In the fifth century BC the Sophists emerged as Greece’s first professional providers of
“higher education”; they were itinerant intellectuals who offered their teaching for large
fees to the upper-class youth of the major cities. Although different Sophists special-
ized in different areas, they all shared a central interest in logos (“language”/“speech”/
“argument”) as a powerful tool within the community. Greeks had always been fasci-
nated by displays of persuasive public speaking and fine oratory to affect political and
judicial deliberations or simply to entertain. The rise of Athenian democracy was a
fertile ground for increased demand of professional training in public speaking, which
the Sophists filled. The most famous Sophist, Protagoras, was notorious for teaching
the art of arguing on both sides of any given matter with equal strength, making the
weaker argument stronger (cf. also the sophistic-style work Twofold Arguments); Gor-
gias was renowned as a master teacher of rhetoric who could argue persuasively on
any matter, regardless of truth and knowledge (cf. Encomium of Helen, Defence of Pal-
amedes). It is exactly because of their concern with rhetoric that the Sophists engaged
in groundbreaking reflection on the nature of language, argument, grammar, etymology
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

and philology (cf. Guthrie 1971; Kerferd 1981). Unfortunately, our primary evidence
is scanty and we rely heavily on limited and fragmentary information from later, and
often hostile, sources.
Sophists are reported to have offered teaching on the subjects of “correct diction”
(orthoepeia) and “correctness of words” (orthotēs onomatōn) (cf. e.g. Plato Cratylus 384b,
391c). It is not clear whether and how the two subjects differed: the former possibly
dealt with the correct use of language in the framework of criticism of poetry. Topics
which were certainly included under either heading are: (1) distinction and revisionary
discussion of three genders of nouns (masculine, feminine, neuter), credited to Protago-
ras, who argued that a word’s gender should mirror the named thing’s natural gender
(cf. Aristotle Sophistical Refutations 14.173b17–22); (2) distinction of different kinds
of speech-acts (command, prayer, narration, address, question, answer), also credited
to Protagoras (cf. Aristotle Poetics 19.1456b9–19); (3) the phenomenon of synonymy:
Prodicus argued for the precise use of words and therefore clarified the different meanings
of quasi-synonyms (cf. e.g. Plato Laches 197d); (4) the equivocal use of words, discussed
by Antiphon (cf. Galen On Hippocrates’ Surgery 18.2), and exploitation of ambiguity for
construction of sophisms; (5) the coinage of neologisms, which should fit exactly the
intended concept, again discussed by Antiphon; (6) whether the correct use of words
depends on use and convention (nomos) or mirrors the nature (physis) of the things they
are meant to signify (cf. Plato Cratylus 384b): (1)–(5) are clearly part of this general
discussion and suggest in various ways rejection of the purely conventional hypothesis.

Plato
A key piece of evidence for the Sophists’ discussion of (6) is Plato’s Cratylus, the dia-
logue most often referred to in analyses of Plato’s (ca. 429–347 BC) views on language.
Socrates’ interlocutors, Cratylus and Hermogenes, offer opposed answers to the question
of what makes “names” correct (onomata include proper names, nouns, adjectives, par-
ticiples, infinitives): according to Cratylus a name is correct when it reflects the nature
of the nominatum (this position has been labeled “linguistic naturalism”), otherwise it
utterly fails to name it; according to Hermogenes, on the contrary, a name is used cor-
rectly provided that the use reflects established convention (sunthēkē) (“linguistic con-
ventionalism”). Hermogenes’ conventionalism is the first to undergo Socrates’ refuta-
tion. It is paired with a form of Protagorean relativism (things do not have fixed natures,
but are relative to each person’s judgment) which Hermogenes himself cannot accept
(incidentally, linguistic conventionalism is attributed by our sources also to Democri-
tus, whom Plato often paired with Protagoras; cf. Proclus Commentary on Plato’s Cratylus
16.5–7). Socrates further supports the rejection of conventionalism through examples
of etymologies, supposedly showing that our ancestors established and used names as
“tools” to mirror a thing’s nature: Names are encoded descriptions of things, either by
semantic composition of simpler names or, at the basic level of the simplest names, by
phonetic imitation of the nature of the nominata (“the primary sounds in a language
are vocal imitations of primary properties like fluidity, stability, hardness, largeness,
etc.” (Sedley 2006: 220)). For example, eithēr (“aether”) is etymologyzed into aei thei
(“that which always runs”), which reveals the essence of aether as something in eternal
motion (Cratylus 410b); in rhein (“to flow”) the letter rho (“r”), produced by vibration
of the tongue, imitates motion. Cratylus welcomes Socrates’ naturalistic etymologies,
but Socrates undermines naturalism as well by arguing that names can never be perfect

812
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

reproductions of the essences of things, but are often largely incorrect imitations (in
these cases the original name-givers showed themselves to be ignorant of the truth).
Moreover, some degree of conventional agreement must have occurred in certain cases
anyway (for example when a word is composed in equal measure by basic sounds imitat-
ing opposite qualities). Therefore, contrary to what Cratylus believes, names are not a
good route to the truth. It is debated whether by the end of the Cratylus Plato himself
remains committed to a definite theory about “correctness of names.” Most scholars
(for example Williams 1981) have interpreted the conclusion as a complete rejection of
naturalism, and thus possibly as an endorsement of a moderate form of conventionalism
(as Crivelli 2008 suggests, in modern terms Plato would reject at least the view that all
names function as definite descriptions used attributively). Others (especially Sedley
2003) have suggested, on the contrary, that Plato did accept the naturalism defended in
the central part of the dialogue, even if mitigated by elements of conventionalism intro-
duced later. However that might be, in the close of the Cratylus Plato himself issues the
warning that the study of language cannot be a privileged means to pursue knowledge of
reality; we should instead inquire directly into the things themselves, i.e. the Forms.
This warning might appear surprising in light of two claims often made by scholars:
(1) names are, primarily, names of the Forms and only indirectly of the things which
participate in the Forms and are therefore named after them (cf. Phaedo 102b1–2; Repub-
lic 10.596a6–7); (2) to understand the meaning of the word “F” is to grasp the Form of
F. However, even if we grant that “there is a linguistic dimension to Plato’s theory of
forms” (Crivelli 2008: 219), Plato’s inquiry into the Forms is not driven by an interest
in the workings of language, nor are its results used to draw definite, explicit conclusions
concerning the semantic relation between language and reality (we might still ask how,
in our terms, Plato would have spelled out that relation). For example, the definitional
inquiry typical of Plato’s dialectical method (“What is F?”) is not after the referent or
sense of a word (“F”), whether “correct” or as determined by use, but after the essence
of a thing (the F), although Plato might of course have recognized that determining the
latter had important dividends for identifying or clarifying the former. Incidentally,
Plato’s definitional inquiries in most dialogues are narrowly focused, especially on moral
qualities (for example courage in the Laches, temperance in the Charmides, justice in the
Republic). This limitation is also reflected in the illustrative lists of Forms which occur
in many passages and in the denial that there are Forms corresponding to all general
terms in other passages (for example Parmenides 130b–e; Statesman 262a–263b). There-
fore, if we took Forms as part of a theory of meaning, we would end up admitting that
for Plato not all general terms had meaning.
Plato’s limited attention to the phenomena of synonymy and homonymy confirms
his secondary interest in language per se. In the dialogues, Socrates often makes fun of
Prodicus’ obsession with semantic distinctions of quasi-synonyms (for example Protago-
ras 340b–341d, 358a–b), and when dealing with sophisms in the Euthydemus he engages
in homonymy disambiguation only once (277e–278a), even if he is clearly aware that
many sophisms exploit linguistic anomalies.
We are not suggesting, of course, that Plato was not interested in language at all.
The philosophical method to achieve knowledge of the Forms, “dialectic,” consists
in reasoned assessment of one’s own and others’ beliefs through question-and-answer
exchanges. That such exchanges can only be carried out through a (shared) language
is no accident; more than once Plato expresses the idea that human thinking and judg-
ing are themselves a sort of inner silent dialogue with oneself (Theaetetus 189e–190a;

813
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

Sophist 263e–264b; Philebus 38c–e; whether for Plato there are also higher forms of non-
discursive cognition is a much debated question beyond the scope of this chapter).
Exactly because of this role of language in philosophical inquiry Plato was aware of the
risk that linguistic errors could engender philosophical errors. The main example is
Plato’s treatment of the puzzle of how it is possible to speak (and judge) falsely. The puz-
zle, which originated with certain readings of Parmenides (fr. 2.7–8: “neither could you
know what is not—for it is impossible—nor could you mention it”) and is attributed by
Plato to Protagoras, is presented in the dialogues in various forms (Cratylus 428e–430d;
Euthydemus 283b–288a; Republic 478b–c; Theaetetus 188c–189b; Sophist 261b–264b; cf.
Denyer 1991). In its simplest formulation the puzzle can be schematized as follows: to
speak falsely is to say what is not; to say what is not is not to say anything; therefore one
cannot speak falsely (one either speaks the truth or does not speak at all; in the Cratylus
the latter is compared to making meaningless noises). Plato’s solution, presented in the
Sophist, consists in clarifying that making a “statement” (logos) is something complex
and thus different from naming something or “saying what absolutely is not.” Even the
simplest statement—the bearer of truth and falsehood—is a combination of a “noun/sub-
ject” (onoma) and a “verb/predicate” (rhēma), in which the latter states something of the
named object. Therefore to make a false statement (“Theaetetus flies”) is not to name a
nonexistent object (flying Theaetetus), but to say of something that exists (Theaetetus)
something which also exists (flying) but does not “combine” with it. Conversely, a true
statement expresses a real combination between the named object and what is said of it
(in the Theaetetus Plato rejected a form of extreme Protagorean relativism according to
which truth depends on individual judgment). Plato’s pioneering distinction between
the two essential components of a statement is a fundamental contribution. It is not
clear, however, whether it amounts to a grammatical distinction (noun/verb), a syntac-
tical one (subject/predicate), or an ontological one (thing/property).

Aristotle
Plato’s distinction in the Sophist was expanded upon by his pupil Aristotle (ca. 384–322
BC). Although his views on the matter are expressed only in embryonic and scattered
form, Aristotle was the first to define important phonetic, grammatical and syntactical
elements. In Poetics 20 he distinguishes:

• three nonsignificant linguistic items: letter, syllable, “connector” (sundesmos

includes particles, conjunctions, prepositions);
• four significant (sēmantikai) linguistic items: “name” (onoma includes proper names,
nouns, adjectives, pronouns, participles), “verb” (rhēma), “modification” of a name
or a verb (ptōsis includes what we call “case” and “inflection”), “phrase”/“sentence”
(logos).

Language is distinctive of human beings (and indeed what makes us human: cf. Politics
1.2). While animal cries are not articulate, human speech is the articulation of the
voice into letters, the basic elements out of which compound sounds (syllables, words,
phrases, and sentences) are composed.
Obviously Aristotle’s views on significant linguistic items are especially relevant here.
These are discussed also in On Interpretation, whose opening offers a general view of
significant language:

814
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

spoken sounds are symbols (sumbola) of affections (pathēmata) in the soul, and
written marks symbols of spoken sounds. And just as written marks are not the
same for all men, neither are spoken sounds. But what these are in the first
place signs of–affections of the soul–are the same for all; and what these affec-
tions are likenesses of–things–are also the same.
(1.16a1–7)
Language is conventional, as suggested by the fact that the same thoughts (“affections in
the soul”) are expressed through different vocalizations and, indirectly, different written
marks in different languages. Oral and written words are adopted primarily as “symbols”
or “tokens” and “signs” of thoughts and, indirectly, of things (cf. Sophistical Refutations
1 for an alternative formulation which makes the symbolic link between words and
things direct). For example, “a name is significant (sēmantikē) ‘by convention (kata
sunthēkēn)’ because no name is a name by nature, but only when it becomes a symbol”
(On Interpretation 2.16a29–30). The term sumbolon is particularly well suited to the idea
of conventionality, since:

the normal use of the Greek word was for a tally or token. A contract or other
agreement might be marked by breaking a knucklebone or other objects in two,
one portion being taken by each of the parties to the agreement. Each person
kept his piece, and could identify the person who presented the other piece by
matching it with his own . . . the meaning of a word is fixed by convention . . .
just as the importance attached to a tally, token, or ticket depends on agree-
ment between the parties concerned.
(Whitaker 1996: 10)
The relation between thought and the world is, on the contrary, natural. Thoughts
are the same for all human beings because so are the things they represent: literally, a
thought is a “likeness” of a thing, because it shares the “form” of the thing thought (for
this aspect of Aristotle’s philosophy of mind cf. On the Soul). Aristotle seems to engage
with the aporetic end of Plato’s Cratylus by endorsing linguistic conventionalism and
transferring the naturalist element from the language-world relation to the thought-
world relation (Aristotle’s terminology alludes to the Cratylus: cf. for example “every
logos is significant not as a tool, but by convention” at On Interpretation 4.17a1–2).
Conventionalism is also the starting point of Aristotle’s specific discussion of the
distinctive features of names, verbs (and their modifications) and phrases/sentences.
All these are first defined as “conventionally significant sounds.” Names and verbs,
albeit composed of letters and syllables, have no part significant by themselves, whereas
phrases/sentences do (the component names, verbs, and modifications). Verbs in turn
differ from names in that they “signify time in addition” and are “signs of things predi-
cated of something else” (On Interpretation 3.16b6–8); when they are taken in isolation,
and thus lack predicative force, verbs are names. Names need not fill the syntactic role
of subjects (“man is an animal”), but they too can be used as predicates when connected
to the subject through a copula expressing the temporal and predicative verbal func-
tions (“Socrates is a man”).
Names, verbs, and those logoi (phrases) which are composed only of names (for exam-
ple a definiens of “man” such as “two-footed terrestrial animal”) are neither true nor
false. Sentences (logoi) are the result of the combination of names and verbs or of names
and names through a copula. Among sentences, the “assertoric” (apophantikoi) ones are

815
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

the bearers of truth and falsehood (and so are the corresponding thoughts), whereas
other kinds of sentences (for example prayers) cannot be true or false. The most basic
assertoric sentences are affirmations and negations. In affirmations, something is said
to be combined with something else; in negations something is said to be divided from
something else. Affirmations are then true when the things which are said to be com-
bined are in fact combined in the world, and false when they are actually divided; nega-
tions are true when the things which are said to be divided are in fact divided, and false
when they are actually combined. In other words, Aristotle adopted a correspondence
theory of truth: Truth consists in the correct match of combination/division in thought
and speech with combination/division in reality (cf. also Metaphysics 4.7.1011b25–7:
“To say of what is that it is not, or of what is not that it is, is false, while to say of what
is that it is and what is not that it is not, is true”). Affirmations having empty names as
subjects are always false, while the corresponding negations are always true (cf. Catego-
ries 10.13b27–35).
One might wonder whether Aristotle’s views on the “signification” relationship
between language, thought, and reality can be construed as a definite theory of mean-
ing. This is a controversial issue, and any reconstruction runs the risk of over-inter-
preting Aristotle’s scattered remarks in light of the conceptual apparatus of modern
philosophy of language. The general picture emerging from some modern readings of
Aristotle’s texts can be sketched as follows. Words signify (sēmainein): (1) our common
thoughts associated with them (“thunder” signifies the thought of a noise accompanied
by lightning) and (2) essences of things in the world (“thunder” signifies the essence
of thunder, i.e. noise in the clouds caused by quenching of fire). Words to which no
essence corresponds, such as empty terms (for example “goat-stag”) and indefinite terms
(for example “non-man”), still have signification in sense (1). All remaining words sig-
nify in both senses, even when the speaker and the hearer lack (full) knowledge of the
essence signified. One main point wherein modern accounts of Aristotle’s theory differ
is in the understanding of which sense of “signification” is primary, and thus of whether
Aristotle’s theory of signification is a theory of meaning at all. For example, Irwin 1981
stresses the primacy of essential signification (what he calls “signification by nature”)
and therefore denies that Aristotle had a theory of meaning, where by “meaning” of
a word Irwin intends “what is entered in a dictionary; . . . what a competent speaker
and hearer grasp but a non-speaker does not grasp; what two synonymous words share”
(242). (This is mere “signification to us,” in which Aristotle was not interested except as
the starting point of the inquiry into essences, i.e. primary signification.) Other scholars
(for example Charles 1994; Modrak 2001), on the contrary, place more weight on the
direct “symbolic” and “semantic” relationship between words and thoughts described in
On Intepretation, i.e. signification in sense (1). This is the primary signification of words
and therefore Aristotle does have a theory of meaning, provided that meaning is not
identified with Fregean sense or reference but with intentional content of thoughts.
What makes attempts to reconstruct Aristotle’s theory of language all the more dif-
ficult is the fact that, just as in Plato’s case, Aristotle’s relevant remarks belong to dif-
ferent contexts in which the primary interest is not language per se. The Poetics account
of significant and nonsignificant linguistic items is linked to the broader discussion of
poetical composition and literary criticism. In On Interpretation Aristotle’s discussion is
guided by his interest in dialectical arguments. For example “his analysis of the structure
of the assertion is meant to apply to the ideal assertions prescribed in dialectic and not
to everyday, idiomatic utterances” (Whitaker 1996: 59). In Metaphysics 4, the concept

816
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

of signification is central to Aristotle’s renowned defense of the Principle of Noncon-

tradiction: if communication is to be possible, words must bear definite signification,
but then the Principle of Noncontradiction must hold good (“X” cannot both signify
and not signify X).
Another instrumental discussion of language occurs in the Sophistical Refutations,
where Aristotle classifies sophisms on the basis of the fallacy involved, with a view to
offering a method for solving them so as to avoid defeat in dialectical exchanges (this
classification is still influential in modern treatments of fallacies). Six of the thirteen
kinds of sophisms identified are based on linguistic fallacies which deceptively exploit
certain anomalies of language: (1) homonymous words signify more than one thing; (2)
amphibolous sentences can be syntactically construed in two semantically different ways;
(3) parts of a sentence can be combined or (4) divided so that the whole has, accordingly,
different significations; (5) words differing only in their “intonation” (i.e. identically
written but pronounced differently) do not signify the same; (6) words having the same
“form of expression,” i.e. belonging to the same grammatical category or playing the
same syntactical role in a sentence, can signify different kinds of things (“Socrates”
and “man” are both names which can function as subjects, but the former signifies an
individual, the latter a universal).
Other discussions of the phenomenon of homonymy occur elsewhere within Aris-
totle’s criticism of rival philosophical views (for example the rejection of the Platonic
univocity of goodness; cf. Nichomachean Ethics 1.6) and defense of his own theories (one
notorious case is that of the homonymy of being; on the various forms of homonymy
cf. Shields 1999).

Hellenistic Philosophy
Inquiries into various aspects of language (for example its origin, grammar, syntax,
semantics) became more focused and systematic during the Hellenistic age (the three
centuries after Aristotle’s death). However, our extant evidence is unfortunately scanty:
most of the vast output of the four major schools (Epicureans, Stoics, Academics, Peri-
patetics) is lost, and often we have to rely on a few short surviving fragments or on much
later reports and paraphrases, second or thirdhand and only partially reliable.

Epicureans
Epicurus’ (341–270 BC) position on the origin of language can be interpreted as a reac-
tion to Aristotle’s main argument for linguistic conventionalism (Sedley 1973: 20).
Aristotle had inferred the conventionality of language from the fact that mental affec-
tions (and the things in the world producing them) are the same for all human beings
but words differ in different languages; Epicurus, on the contrary, takes the existence of
different languages as evidence for the natural, nonconventional origin of language. In
a dense section of his Letter to Herodotus (75–76) Epicurus first outlines a general view
of the origin and development of civilization, articulated into a first phase dominated
by nature and a later one in which human reason perfected nature’s contribution. Then
he applies this view to the paradigmatic case of language, also in two phases. In the first
natural phase men were instinctively led to make utterances by the “impressions” (phan-
tasmata) and “affections” (pathē) generated in them through interaction with the exter-
nal world (just as animals do). Probably the original utterances were already articulate

817
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

to some degree (Epicurus calls them “names”). Since different objects caused different
impressions and affections, which in turn were naturally expressed through different
utterances, different words came to be used for different objects. However, environmen-
tal factors and, possibly, physiological ethnic variations contributed to the formation
of different psychological states in human beings living in different places. This natural
difference explains why different human groups naturally came to use different words
for the same thing, i.e. the variety of natural languages (which thus supports linguistic
naturalism). This sketch of the first phase of the origin of language would be accepted
by most scholars (for example Verlinsky 2005). According to an alternative reconstruc-
tion originally utterances differed for each individual, because so did impressions and
affections, and thus each individual developed a private language, whereas only in the
second phase a common language was adopted by conventional imposition. Brunschwig
1994 has convincingly rejected this private language hypothesis; this, however, does
not mean that for the Epicureans thoughts cannot exist independently of public lan-
guage, and the Epicurean theory seems in fact to be committed to the logical and psy-
chological priority of thought, at least in its basic forms, over language. The naturalistic
view of the origin of language is opposed by Epicurus to a conventionalist one: “names
did not originally come into being by imposition (thesei).” The allusion here (and in the
parallel passage in Lucretius On Nature 5.1028–1090) seems to be to a human inventor
(contra the Cratylus’ name-giver hypothesis), but in a later Epicurean source (Diogenes
of Oenoanda 10.2) the possibility of a divine inventor of language is also discarded and
ridiculed.
In the second phase of language development, reasoning and convention intervened:
Natural languages were refined (“particular words were imposed,” Epicurus says, with
the purpose of making language less equivocal and more concise) and enriched (“the
vocabulary was artificially enlarged, or at any rate existing words were given additional
meanings, by intellectuals turning their attention to abstruse or theoretical entities”
(Long and Sedley 1987: 100)). Thus a novelty of Epicurus’ proposal consists in com-
bining a naturalistic theory of the origin of language with an explanation of the funda-
mental (albeit nonprimary) role played by convention (notice, however, that Epicurus’
naturalism is very different from Cratylean naturalism, according to which words should
mirror the nature of things).
Whereas with their views on the origin and nature of language the Epicureans made
a clear contribution to a well-established tradition, the question of whether they had a
definite theory of meaning is more difficult to answer, just as it was in Aristotle’s case.
Some interpreters (e.g. Atherton 2009) suggest that it would be anachronistic to project
onto the few extant relevant texts any semantic theory, and that the Epicureans were
interested in the phenomenon of linguistic communication rather than our question
of meaning. Others believe that the Epicureans were committed to an “extensionalist”
semantics according to which a direct signifying relation holds between linguistic items
and things in the world. Most scholars suggest that, on the contrary, the Epicureans
were “intensionalists” and some fundamental semantic role was played by intermediate
items between language and things. According to the latter interpreters this intermedi-
ate role was played by mental items, in a way somewhat analogous to that of Aristotle’s
“affections.” In his Letter to Herodotus (37) Epicurus writes that our “primary concepts”
“correspond” to our words and “underlie” them, and that without them our utterances
would be “empty” and no inquiry would be possible; Diogenes Laertius (10.33) reports
that for the Epicureans “we would not have named something if we had not previously

818
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

learned its delineation by means of a preconception (prolēpsis).” “Primary concepts”

or “preconceptions” are in Epicurean jargon the fundamental, naturally formed and
commonly shared universal concepts which arise in the human mind through repeated
experience and memory (for example the concept of horse derived since childhood
from repeated encounters with horses). Intentionalist interpreters believe that these
basic concepts are themselves meanings (Long and Sedley 1987: 101) or, at least, that
words have their meanings in virtue of standing in some relevant relation to these
psychological states of the speakers (cf. Everson 1994). These suggestions are compat-
ible with the texts mentioned above. However, given the epistemological focus of the
context of those texts, an intentionalist interpretation is by no means required, just as
it is not required by Epicurus’ account of glossogenesis (mental states are described as
what prompted natural vocalizations, but not as the meanings of those vocalizations).
The strongest evidence for the rival extensionalist interpretation lies in a couple of
later reports (Plutarch Against Colotes 1119F–1120A; Sextus Empiricus Against the Logi-
cians 2.11–12) according to which, of the Stoic triad words—“sayables”—things (cf. p.
820), the Epicureans rejected the intermediate class of “sayables,” thus leaving words as
truth-bearers and things themselves as their meanings. It has been noted, however, that
the Epicurean rejection of “sayables” does not imply that no alternative intermediary
semantic role could be played by some other item, such as preconceptions. Our sources
possibly drew inferences about the Epicurean position rather than report it. The situa-
tion is further complicated by the possible distinction between word-meaning (which
might have been extensional) and speaker-meaning, in which the speaker’s mental
state could have played a role (cf. Barnes 1999: 195–7); however, the hypothesis that
the Epicureans themselves might have been perceptive to such a distinction is, again,
conjectural. It seems safe to conclude that either the Epicureans had no interest in
semantic theory at all or the evidence is anyway too scanty for us to adjudicate between
intensionalist and extensionalist interpretations.

Stoics
Whereas the Epicureans had little interest in grammar, syntax, and linguistic phenom-
ena such as ambiguity, all these topics were object of intense attention within the Stoic
school, which was founded by Zeno of Citium toward the end of the fourth century
BC and flourished until the end of antiquity. Judging from our limited evidence, it is
not clear whether the Stoics also held a full-blown theory on the origin and nature of
language. According to Origen (Against Celsus 1.24), the Stoics thought that names
are natural and the first human vocalizations imitated things. Augustine (On Dialec-
tic 6) explains that they believed that language originated with onomatopoeic imita-
tion and different principles intervened subsequently in the formation of other words
(presumably the Stoics used these principles in the etymological work in which they
engaged extensively). The idea that first sounds mirrored things, however, differed from
Cratylean naturalism: For the Stoics only the first words were mimetic, and they were
so onomatopoeically (because of the high communicative power of onomatopoeia; cf.
Allen 2005), and not by mirroring the essences of their nominata (even so, etymologies
were considered worth studying as means of information about our ancestors’ under-
standing of reality).
Chrisippus (280–207 BC) was the most original exponent of Stoicism, especially in
the broad area of philosophy labeled “logic”; he wrote at least 130 works on a variety

819
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

of logical subjects, including many which could be catalogued under the heading “phi-
losophy of language” (virtually all these works have been lost). These subjects were
studied by the Stoics as part of “dialectic,” one of the main branches of “logic,” which
encompassed everything that had to do with logos (“rational discourse”). According
to Chrysippus, the subject matter of dialectic was “(1) signifiers and (2) things signi-
fied” (Diogenes Laertius 7.62): (1) utterances, expressions, parts of speech, dialects,
virtues and vices of speech, ambiguity, euphony, poetry; (2) incomplete and complete
“sayables” (lekta), truth, modalities, arguments, validity and invalidity, paradoxes, and
sophisms. This main subdivision already suggests that the concept of signification was
central to the Stoic logical and linguistic theory.
The realm of signifiers includes physical (“corporeal”) entities: at the most generic
level, “utterance” or “sound” (phōnē) is simply air that has been struck and becomes
the object of hearing. Human beings are capable of articulate and alphabetic utter-
ances, i.e. “expressions” (lexeis); but only signifying expressions, issuing from thought,
count as “speech” (logos). The Stoics classified various “parts of speech”: “proper name”
(onoma), “appellative” (prosēgoria includes nouns and adjectives), “verb” (rhēma), “con-
nector” (syndesmos), “article,” and “adverb.” Although the Stoics were interested in
phonetic and purely grammatical aspects of language (for example dialects, defined as
sets of lexeis encompassing the distinctive phonetic practices of a linguistic community,
and solecism, studied among the vices of speech) and were hugely influential upon the
development of Greco-Roman grammar, it is important to note that the parts of speech
were uniquely defined on the basis of what they signified or their syntactic contribution
to signification. For example, a proper name is that part of speech which signifies a cor-
poreal “peculiar quality” (for example “Socrates” signifies the quality of “Socratesness:”
according to the Stoics, each individual has a quality which makes it distinct from eve-
rything else). An appellative is that part of speech which signifies a corporeal “common
quality” (for example “horse” signifies the quality of “horseness”). A verb is that part of
speech which signifies an incorporeal “predicable” (for example “write” signifies writ-
ing) and a connector is that part of speech which binds other parts of speech together.
Central to the Stoic theory of language, and foundational for the part on “things sig-
nified,” is the concept of “sayable” (lekton). When we speak, we utter significant expres-
sions in a certain language, but what thereby gets said and signified are incorporeal
sayables, the meanings of our words that are not understood by those who do not speak
that language (a sayable is, literally, “something which can be said” by uttering a certain
expression). Since in Stoic ontology only what is corporeal exists, incorporeal sayables
(also called pragmata, “things” or “states of affairs”) cannot exist but “subsist” (like the
other incorporeals —space, void, and time). The Stoics thus introduced a three-tiered
account of signification, according to which: (1) linguistic expressions (corporeal signi-
fiers) signify (2) incorporeal lekta, what gets said through those expressions about (3)
corporeal objects in the world and their corporeal qualities. For example, (1) “Dion
walks” signifies (2) that Dion walks about (3) Dion (cf. Sextus Empiricus Against the
Logicians 2.11–12; Seneca Letters 117.13). This account has often been associated with
the Fregean distinctions between word, sense, and reference (cf. for example Mates
1961: 19–26), but the parallelism is much less straightforward than it has been main-
tained (cf. Frede 1994). The role of sayables as significations of words, distinct from
the things in the world of which we speak, can also be compared to the intermediary
role which mental “affections” play in Aristotle’s account of signification (and which
“preconceptions” might have played in the Epicurean theory). Whether or not Stoic

820
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

lekta were introduced to play the same structural role as Aristotelian pathēmata, they are
not mental objects but public entities which secure the possibility of communication:
It is the same sayable that is expressed by a certain speaker and becomes the content of
a listener’s thought.
Although sayables are not thoughts (which for the Stoics are corporeal), they func-
tion as the intentional content of our thoughts (cf. Sextus Empiricus Against the Logi-
cians 2.70: “a sayable is what subsists in accordance with a rational impression”). Human
beings differ from other animals exactly in that their “impressions” are “rational” (logikai
phantasiai), i.e. can be shown forth through significant speech (logos). Whereas some
animals, like parrots, have also “uttered speech” (i.e. produce articulate sounds), only
human beings have “inner speech” (endiathetos logos) which is a function of rationality
(Sextus Empiricus Against the Logicians 2.275–6). Thus a strict relationship between
language and thought holds: There is nothing sayable that cannot be thought, and
nothing which we can think that cannot be said (cf. Barnes 1999: 213). And since logos
was also, for the Stoics, the divine organizing rational principle of the universe, the
relationship extends to reality. Reality is accessible by thought and language, and what
can be said by rational beings is isomorphic with the rational structure of reality.
The origin of the Stoic lekton and the rationale for its incorporeal nature and sub-
sistence are much debated (cf. Long and Sedley 1987: 199–202; Frede 1994; Barnes
1999). Similarly debated is whether sayables are completely mind-independent entities
or whether, while being clearly distinct from signifiers and thoughts, they exist only
as long as the words or thoughts expressing them do (cf. Frede 1994: 118 and Long
1971: 97 for the two positions). The technical details of the theory are also unclear. For
example, although the majority of sources suggest that proper names and appellatives
do not signify lekta but corporeal qualities of objects (cf. Diogenes Laertius 7.58), a case
can be made for them signifying lekta, despite the fact that “predicables” (katēgorēmata),
signified by verbs, are the only examples routinely used for “incomplete sayables” in our
sources. (The Stoics wrote extensively about predicables and classified them: cf. Barnes
1999: 204–6.) The signification of a verb is fixed by what, by uttering it, you can say
about something: For example, “to run” signifies something (running) which can be said
of something (e.g. Socrates). This predicable, running, is an incomplete sayable because
when it gets expressed (for example by the English “runs”) the hearers feel prompted to
ask for a completion (“who runs?”), and no complete state of affairs gets expressed until
the predicable is attached to a “nominative case” (cf. Diogenes Laertius 7.63–4). The
notion of “case” is itself controversial (are cases signified by proper names/appellatives,
and if so what is their relationship with peculiar/common qualities? Are cases incom-
plete lekta?), interestingly straying between grammar, semantics and metaphysics.
“Complete” sayables were classified on the basis of what you do by saying them: for
example “assertibles” (axiōmata) are sayables “saying which we make assertions, and
which are either true or false,” while imperatival sayables are “those saying which we
give a command” (and which are neither true nor false) (Sextus Empiricus Against the
Logicians 2.71–3). The Stoics distinguished other kinds of complete sayables, includ-
ing questions (which require “yes” or “no” as an answer), inquiries (open questions
requiring a discursive answer), puzzlements (questions which do not require answer),
oaths, prayers, invocations, curses, hypotheses and expositions (for example “let this be
a straight line”). The Stoic classification of sayables thus resembles modern classifica-
tions of speech-acts: Each time we express a complete sayable we (1) utter a linguistic
expression, (2) say the associated sayable and, (3) perform a certain speech-act. Unlike

821
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

Aristotle, who believed that nonassertoric speech belongs to the realm of rhetoric and
poetics, for the Stoics the study of all types of sayables belonged to dialectic.
Stoic assertibles have often been associated with propositions: They are the primary
bearers of truth and falsehood (assertoric sentences and impressions are true or false
derivatively, in virtue of having assertibles as their propositional contents). The Stoics
adopted a correspondence theory of truth: “A true assertible is that which is the case
and is contradictory of something, and a falsehood is that which is not the case and is
contradictory of something” (Sextus Empiricus Against the Logicians 2.85). Although
no assertible exists, then, within subsisting assertibles, the Stoics drew a distinction
between true assertibles, which “are the case” or “obtain,” and false assertibles, which
are not the case. They also subscribed to a “temporalized” version of the principle of
bivalence: Assertibles are either true or false at times, and can change their truth-values
over time. For example, the assertible that it is raining is true now, because it is raining,
but will be false in one hour, when it will not be raining. Stoic assertibles also differ from
propositions in that they are tensed and can contain token-reflexive elements, such as
“I” or “this.” The sentence “I am now eating this,” uttered at different times by differ-
ent speakers pointing at different objects, signifies the very same assertible but different
propositions. Incidentally, assertibles are also bearers of modalities (possibility, impos-
sibility, necessity) of which Chrysippus advanced the discussion, initiated by Aristotle
and continued by the Megarians Diodorus Cronus and Philo in the second half of the
fourth and first half of the third century BC.
The Stoics embarked upon a sophisticated classification of assertibles, both “simple”
and “non-simple.” Among simple assertibles they distinguished (cf. Sextus Empiricus
Against the Logicians 2.96–100; Diogenes Laertius 7.69–70):

(1) “definite” assertibles, involving some deictic reference (for example “this one is
walking”; the pronouns “I” and “you” were also considered deictic). Definite asserti-
bles are “destroyed” when the object of the deictic reference no longer exists (for
example “this one is dead,” where “this one” refers to Dion, is false as long as Dion
is alive, and is destroyed, i.e. ceases to subsist, at Dion’s death);
(2) “intermediate” assertibles, composed of a nominative case and a predicable (for
example “Dion is dead,” which is false as long as Dion is alive, and becomes true
from the moment of Dion’s death);
(3) “indefinite” assertibles, governed by an “indefinite particle” (for example “someone
is walking”). An indefinite assertible is true whenever a corresponding definite one
(for example “this one is walking”) is true.

Negations of simple assertibles count as simple assertibles. The Stoics distinguished

three forms of negative assertibles:

(1) “negations,” formed by prefixing the particle “not” (for example “Not: it is day”).
External negation is truth-functional: If an assertible p is true, then its negation
not-p is false; if p is false, then not-p is true. An assertible and its negation are con-
tradictory, and the Principle of Excluded Middle holds good;
(2) “denials,” having a compound negative as subject (for example “no-one is
walking”);
(3) “privations,” having a privative particle prefixed to the predicable (for example
“this one is un-kind”).

822
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

Interestingly, ordinary negative sentences, in which the negation governs the predicate
(for example “Dion is not walking”), were not classified under this scheme, and indeed
counted as affirmations for the Stoics, who insisted that in “negations” the negative
particle should have broad scope, governing the whole assertible. Not only does the
external negation allow recognition of the quality of the sentence at first sight, but it
also guarantees the principle that if an assertible is true (false) then its negation is false
(true) and guarantees the Principle of Excluded Middle. When the subject X is empty,
both “X is F” and “X is not F” are false, because both assertibles have existential import
(for example if Dion does not exist, both “Dion is walking” and “Dion is not walking”
are false), whereas “Not: X is F” (“Not: Dion is walking”) is true. This shows that the
classification of assertibles was not guided merely by observation of existing linguistic
structures but by logical or metaphysical considerations.
This is even clearer in the classification of nonsimple assertibles, which are composed
of two or more simple assertibles (or the same assertible taken two or more times) linked
by some connector. The Stoics made the largest contribution to propositional logic in
antiquity (although the Peripatetic Theophrastus and the Megarians had initiated study
of the subject), and nonsimple assertibles are the building blocks of that logic. The
study of nonsimple assertibles goes beyond the scope of this chapter, and belongs more
properly to the field of philosophy of logic (for an overview cf. Mates 1961; Bobzien
1999). We only mention here that the Stoics distinguished conjunctions, negated con-
junctions, disjunctions, para-disjunctions, conditionals, para-conditionals, causals, and
comparatives, regimenting the natural language usage of these assertibles (for example
connectors are always prefixed) and providing analyses of their truth-conditions (never
truth-functional, except for conjunctions).
As mentioned above, within their dialectic the Stoics investigated linguistic ambigu-
ity (cf. Atherton 1993). Chrysippus held that every expression is ambiguous. Supposedly
he meant that each word signifies both its significate and itself, hinting at something
analogous to our use-mention distinction. He also insisted on the lexical character of
ambiguity, in opposition to the claim of Diodorus Cronus that ambiguity does not exist
because the meaning of a word or a sentence is the meaning intended by the speaker
(this is connected to Diodorus’ own extreme conventionalism). We also inherited a
(possibly Chrysippean) classification of eight types of ambiguities (cf. Galen On Lin-
guistic Sophisms 4).
Although the Stoics do not seem to have discussed the Sorites paradox in terms of
lexical ambiguity, they devoted great attention to the Sorites, which had been devised
by the Megarian Eubulides and used by the Stoics’ skeptical adversaries to undermine
their epistemological optimism. The reconstruction of Chrysippus’ diagnosis and solu-
tion is deeply controversial. Several interpreters have attributed to him an epistemicist
position according to which, although it is impossible for us to determine which of
the conditionals involved is false (for example what precise cut-off value of x falsifies
the conditional “if x grains of sand are not a heap, x+1 grains of sand are not a heap
either”), there is, unbeknown to us, one false conditional in the argument, which
is thereby materially false (cf. Barnes 1984). Other interpreters have rejected this
reconstruction, suggesting that for Chrysippus some of the borderline cases have no
truth-value (possibly because they fail to express assertibles on account of vagueness; cf.
Bobzien 2002).
Although no part of Stoic philosophy was labeled “grammar” or was clearly iden-
tifiable with grammar, there is evidence that the Stoics initiated the study of our

823
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

“traditional grammar” (which includes phonology, morphology, syntax, parts of speech

and their features), labeled “technical grammar” in the Hellenistic period. (It is often
assumed that the Alexandrian school initiated this study, but actually the school’s main
interests were philology and literary criticism.) Not only is the classification of parts
of speech obviously relevant to grammar, but also the discussion of the way in which
simple and compound lekta are structured contributed to the development of syntax in
later antiquity (Dionysius Thrax was the first to write about technical grammar between
the first and second century BC, while Apollonius Dyscolus authored the first extant
treatise on syntax in the second century AD; cf. Frede 1987).

Later Antiquity
The development of grammar is by no means the only example of sustained interest in
language in late antiquity. Peripatetics and Platonists reflected on the views on language
proposed in the Classical and Hellenistic periods, while Pyrrhonian skeptics attacked
the whole previous philosophical tradition, including theories of language (cf. Sextus
Empiricus Outlines of Pyrrhonism 2; Against the Logicians; Against the Grammarians), and
devised at the same time nonassertoric modes of speech to express their skeptical stance.
We can only mention here the closing chapter in this later history, which falls outside
the Peripatetic, Platonist, and Pyrrhonian traditions.
Augustine’s (354–430 AD) insights on language are mostly known to students of
philosophy through the medium of the influential presentation and criticism of the
“Augustinian picture of language” which Wittgenstein sketched in the Brown Book and
the Philosophical Investigations. According to that picture, individual words name objects,
and the meaning of each word is the object for which it stands. Some scholars (e.g. Burn-
yeat 1987) have observed that this picture does not really reflect Augustine’s historical
view. It is true that Augustine wrote that infants begin learning language ostensively by
interpreting adults’ behavior and picking up the names which they observe get associ-
ated with things (Confessions 1.6). There is no indication, however, that all language
acquisition works in this way (Augustine was aware, anyway, that ostension is chroni-
cally and unavoidably ambiguous as a way of teaching a language; cf. On the Teacher
10.29); nor did Augustine, who knew his grammar, fail to realize that language contains
much more than names. What is true is that for Augustine all words signify. Words were
in fact conventionally instituted as “given signs” (as opposed to “natural signs,” which
are not intended to signify) among human beings for the purpose of communication,
teaching, and reminding: “A sign is what shows both itself to the senses and something
beyond itself to the mind” (On Dialectic 5.7). Augustine is not consistent and clear,
however, on what words signify: Sometimes he suggests that they signify incorporeal
states of mind (thoughts, volitions, impulses; for example “if” signifies the speaker’s
doubt); sometimes he claims that words signify things, either perceptible or intelligible
(cf. Kirwan 1994: 193). Although communication is the raison d’être of language, some
form of inner language is the medium of all our thought. Even before being expressed in
a natural language, what is conceived in our souls is already a “silent word,” but “in no
language” (cf. Lectures on the Gospel of John 3.14.7; On the Trinity 15.10.19). Augustine’s
original take on Plato’s influential idea, also criticized by Wittgenstein (Philosophical
Investigations 1.32), discloses the possibility of the existence of mental contents which
are not (fully) communicable in any language (cf. Rist 1997).

824
 A N CI EN T PH I LO SO PH Y O F L A N G U A G E

Related Topics
7.2 Medieval Philosophy of Language.

References

General
Everson, S. (ed.) (1994) Language. Companions to Ancient Thought 3, Cambridge: Cambridge University
Press.
Kneale, W. and Kneale, M. (1962) The Development of Logic, Oxford: Clarendon Press.
Modrak, D. K. W. (2006) “Philosophy of Language,” in M. L. Gill and P. Pellegrin (eds.) A Companion to
Ancient Philosophy, Oxford: Blackwell, 640–63.
Mortley, R. (1986) From Word to Silence, 2 vols., Bonn: Hanstein.
Nuchelmans, G. (1973) Theories of the Proposition: Ancient and Medieval Conceptions of the Bearers of Truth
and Falsehood, Amsterdam: North Holland Publishing.

The Sophists
Guthrie, W. K. C. (1971) The Sophists, Cambridge: Cambridge University Press, 176–225.
Kerferd, G. B. (1981) The Sophistic Movement, Cambridge: Cambridge University Press, 68–77.

Plato
Crivelli, P. (2008) “Plato’s Philosophy of Language,” in G. Fine (ed.) The Oxford Handbook of Plato, Oxford:
Oxford University Press, 217–42.
Denyer, N. (1991) Language, Thought and Falsehood in Ancient Greek Philosophy, London and New York:
Routledge.
Sedley, D. N. (2003) Plato’s Cratylus, Cambridge: Cambridge University Press.
—— (2006) “Plato on Language,” in H. H. Benson (ed.) A Companion to Plato, Oxford: Blackwell,
214–27.
Williams, B. (1981) “Cratylus’ Theory of Names and its Refutation,” in M. Schofield and M. Nussbaum
(eds.) Language and Logos: Studies in Ancient Greek Philosophy, Cambridge: Cambridge University Press,
83–94.

Aristotle
Charles, D. (1994) “Aristotle on Names and their Signification,” in Everson 1994, 37–73.
Irwin, T. H. (1981) “Aristotle’s Conception of Signification,” in M. Schofield and M. Nussbaum (eds.) Lan-
guage and Logos: Studies in Ancient Greek Philosophy, Cambridge: Cambridge University Press, 241–66.
Modrak, D. K. W. (2001) Aristotle’s Theory of Language and Meaning, Cambridge: Cambridge University
Press.
Shields, C. (1999) Order in Multiplicity: Homonymy in the Philosophy of Aristotle, Oxford: Clarendon Press.
Whitaker, C. W. A. (1996) Aristotle’s De Interpretatione: Contradiction and Dialectic, Oxford: Clarendon
Press.

Hellenistic Philosophy
Barnes, J. (1999) “Language: Meaning,” in K. Algra, J. Barnes, J. Mansfeld and M. Schofield (eds.) The Cam-
bridge History of Hellenistic Philosophy, Cambridge: Cambridge University Press, 193–213.
Frede, D. and Inwood, B. (eds.) (2005) Language and Learning: Philosophy of Language in the Hellenistic Age,
Cambridge: Cambridge University Press.
Long, A. A. and Sedley, D. N. (1987) The Hellenistic Philosophers, vol. 1, Cambridge: Cambridge University
Press.

825
 L UCA CA STA G N O LI A N D ERM ELI N D A VA L E N T I N A D I L A S C I O

Epicureans
Atherton, C. (2009) “Epicurean Philosophy of Language,” in J. Warren (ed.) The Cambridge Companion to
Epicureanism, Cambridge: Cambridge University Press, 197–215.
Brunschwig, J. (1994) “Epicurus and the Problem of Private Language,” in Papers in Hellenistic Philosophy,
Cambridge: Cambridge University Press, 21–38.
Everson, S. (1994) “Epicurus on Mind and Language,” in Everson 1994, 74–108.
Sedley, D. N. (1973) “Epicurus, On Nature Book XXVIII,” Cronache Ercolanesi 3: 5–83.
Verlinsky, A. (2005) “Epicurus and his Predecessors on the Origin of Language,” in D. Frede and B. Inwood
2005, 56–100.

Stoics
Allen, J. (2005) “The Stoics on the Origin of Language and the Foundations of Etymology,” in D. Frede and
B. Inwood 2005, 14–35.
Atherton, C. (1993) The Stoics on Ambiguity, Cambridge: Cambridge University Press.
Barnes, J. (1984) “Medicine, Experience and Logic,” in J. Barnes, J. Brunschwig, M. Burnyeat and M.
Schofield (eds.) Science and Speculation: Studies in Hellenistic Theory and Practice, Cambridge: Cambridge
University Press, 24–68.
Bobzien, S. (1999) “Logic: The Stoics,” in K. Algra, J. Barnes, J. Mansfeld and M. Schofield (eds.) The Cam-
bridge History of Hellenistic Philosophy, Cambridge: Cambridge University Press, 92–157.
—— (2002) “Chrysippus and the Epistemic Theory of Vagueness,” Proceedings of the Aristotelian Society 102:
217–38.
Frede, M. (1987) “The Origins of Traditional Grammar,” in Essays in Ancient Philosophy, Minneapolis: Uni-
versity of Minnesota Press, 338–59.
—— (1994) “The Stoic Notion of a Lekton,” in Everson 1994, 109–28.
Long, A. A. (1971) “Language and Thought in Stoicism,” in A. A. Long (ed.) Problems in Stoicism, London:
Athlone Press, 75–113.
Mates, B. (1961) Stoic Logic, Berkeley and Los Angeles: University of California Press.

Augustine
Burnyeat, M. F. (1987) “Wittgenstein and Augustine De Magistro,” Proceedings of the Aristotelian Society s.
v. 61: 1–24.
Kirwan, C. (1994) “Augustine on the Nature of Speech,” in Everson 1994, 188–211.
Rist, J. M. (1997) Augustine: Ancient Thought Baptized, Cambridge: Cambridge University Press, 23–40.

Many thanks to the Leverhulme Trust for supporting Dr Di Lascio’s research through an
Early Career Fellowship in the period October 2009– September 2012.

826
 7.2
M EDIE V AL P HILOSOPHY
OF L ANGUAGE
Gyula Klima

Sources and Issues

Fair warning: the title of this chapter is misleading. There is no such a thing as medi-
eval philosophy of language. At least there was no philosophical discipline recognized
in medieval thought that could be justifiably identified by us as “philosophy of lan-
guage.” This is not to say, however, that medieval thought did not contain systematic,
philosophical reflections on language as the necessary medium of human thought and
communication. On the contrary, as we shall see, the medieval philosophical literature
is shot through and through with such reflections (indeed, ones that have hardly been
matched in sophistication ever since), but they are not to be found in treatises classifi-
able under the heading of a unified philosophical discipline as we conceive of it; rather,
they are to be found in works in their respective medieval disciplines.
Thus, many of these reflections can be found in treatises, commentaries, and quaes-
tiones [Sweeney 2008] in the well-recognized disciplines of grammar and logic, as should
be obvious. However, a great deal of interesting observations relevant to the subject
(rounding out the ex professo considerations of the above-mentioned two) can be found
in works on the soul, as well as metaphysics and theology. Of course, not all of all these
works is relevant to “philosophy of language” as we conceive of it.
The descriptive/normative grammars of Latin used for teaching students the ele-
ments of Latin syntax would certainly not count as anything philosophically interesting
to us. However, the characteristic doctrine of the modists [Marmo 1999], for example,
contained in works on “speculative grammar” (grammatica speculativa) [Ashworth 1977;
Bursill-Hall 1971], attempting to provide some general philosophical underpinning for
the rules of syntax contained in the descriptive/normative works, is certainly relevant
to our contemporary interest in the philosophy of language. Thus, commentaries on
the Roman grammarians, especially Donatus and Priscian, as well as quaestiones and
self-contained treatises inspired by their work, are valuable sources for medieval specu-
lation about the fundamental features any human language must have, based on the
conviction that the modi significandi (the ways linguistic items signify), determining
proper construction or well-formedness (congruitas), are rooted in the modi intelligendi
(the ways we think) common to all humans, which in turn are based on the modi essendi
of things (the ways things are). But this idea of a “universal grammar” and especially
its attempted justification was controversial, and very often challenged, especially by
nominalists, in the Middle Ages [Peter of Ailly 1994].
 G YU LA KL I MA

By contrast, logic, looking at language as the medium of all human thought and rea-
soning, eo ipso deals with what are supposed to be the universal conceptual structures
conveyed variously by different conventional human languages. Accordingly, medieval
speculations concerning the subject matter of logic, and more specific considerations
of exactly how the various syntactical structures of different human languages are sup-
posed to convey the conceptual structures they express, and, in their turn, how those
conceptual structures are related to items of extramental reality, are directly relevant to
our concerns in the philosophy of language. Therefore, much of the immense medieval
logical literature contains valuable insights on the medieval “theory of meaning,” the
theory of signification, most of which was inspired by Aristotle’s Categories (Categoriae or
Praedicamenta) and On Interpretation, especially in their translations and commentar-
ies by Boethius. On the other hand, many important considerations concerning the
medieval “theory of reference,” the theory of supposition, were prompted by Aristotle’s
theory of fallacies in his Sophistical Refutations, giving rise to a burgeoning literature
on this characteristically medieval doctrine, as the most important part of the general
theory of the properties of terms [Read 2008]. One should also mention here the char-
acteristic genre of the treatises on obligations, reflecting on the rules of engaging in the
sophisticated medieval art of disputation [Spade 2008]. Finally, one should not ignore
the immense literature on universals prompted by Porphyry’s Isagoge (‘Introduction [to
Aristotle’s Categories]’), discussing a five-fold classification of the different sorts of uni-
versals (the so-called “predicables”, praedicabilia: genus, species, difference, property,
and accident), organized into the system of Aristotle’s categories (the praedicamenta,
which are taken to be either the most universal genera, namely: substance, quantity,
quality, relation, action, passion, time, place, position, and habit (as some garment is
a “habit”), or the orderings of subordinate universals into the structures traditionally
called “Porphyrian Trees,” dividing genera by their differences into their species, until
the most specific species are reached, whose individuals only differ in accidents, but not
in any essential differences) [Gracia and Newton 2010].
Indeed, “the problem of universals” in and of itself continues to be in one form or
another one of the main concerns of metaphysics as well as the philosophy of language
[Klima 2008]. So it is no wonder that the most refined and subtle discussions relevant
to our topic revolved around this problem (or bundle of problems), despite the fact that
at least by the time of John of Salisbury everybody took a general Aristotelian stance,
accepting the Aristotelian-Boethian criticism of Plato’s “naive theory” of universals.
However, this general Aristotelian stance, rejecting mind-independent universal enti-
ties existing in real being on a par with, or even more perfectly than, other real entities,
admitted ever more refined articulations, leading to issues that we would recognize as
the problems of essentialism, ontological commitment, and presentism vs. eternalism.
Besides logic and metaphysics, the medieval literature on Aristotelian psychology
also contains important, relevant observations that cannot be ignored from our point
of view. In fact, the whole Aristotelian rejection of the “naive” Platonic theory of uni-
versals is predicated on Aristotle’s theory of universal concept acquisition, his theory
of abstraction. Furthermore, Aristotle’s distinction of the three main intellectual opera-
tions (simple apprehension, judgment formation, reasoning) is the basis of the division
of the subject matter of Aristotelian logic, which, in turn forms the basis of medieval
ideas that we would recognize as ones concerning issues such as generativity, composi-
tionality, and mental language, the last of which became particularly important in the
late-medieval nominalist program of “ontological reduction”.

828
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

Finally, theological works contribute at least three important aspects to what may
be regarded as constituting what can come under the label “medieval philosophy of
language”: first, the problems of scriptural exegesis (which is not to say, of course, that
problems of interpretation were only raised in the context of scriptural exegesis) and,
in general, the problems of sensibly talking about God gave rise to much speculation
about nonliteral, metaphorical and analogical uses of language. Second, considerations
of what is happening in the sacraments, such as the Eucharist or Baptism, involved
intriguing reflections on performative speech acts. And third, considerations of the
different languages of the Biblical tradition (Hebrew and Greek), the philosophical
tradition (adding Arabic) and the different religious, cultural, and social backgrounds
they conveyed as well, led to a number of interesting observations in various authors
concerning linguistic diversity and the role of language in cross-cultural rational argu-
ment and understanding (where we must not forget either that scholastic writers had
Latin as their second language, pretty much in the way many of us have English in
contemporary academia).
Given the immense diversity of sources and issues just listed, one might get the
impression that there was no unified theory specifically addressing philosophical issues
concerning language as a system of signs. But we should not forget about important
(parts of) systematic treatises that do address issues concerning language as a part of
some general semiotic reflections on the nature and classifications of signs, such as
those that can be found from the Roman period (see Augustine’s De Doctrina Chris-
tiana), through High Scholasticism (see, for example, Roger Bacon’s Compendium studii
philosophiae, as well as the numerous reflections on Aristotle’s “semantic triangle,” to be
discussed below, in systematic treatises of logic), to very late Scholasticism, inspiring
the semiotic reflections of early and later modern philosophers (see, for example, the
Tractatus de signis of John of St. Thomas) [Meier-Oeser 2008].
As even this sketchy introductory survey should make clear, the diversity and sheer
bulk of the medieval output relevant to our topic make it impossible to provide here
anything like a systematic, historical account of whatever may come under the title
“medieval philosophy of language.” Therefore, in the following sections I will focus
only on some typical medieval treatments of some of the salient topics mentioned in
this introductory section that we would recognize as clearly belonging to our concerns
in philosophy of language.

Meaning: Signification and Imposition

Although Augustine’s general definition of a sign as something that besides present-
ing itself to the senses also presents something to the understanding [Augustine 1975:
86, 1962: 32] was very much in circulation in the Middle Ages (and had a particularly
strong influence on Roger Bacon, for example), medieval considerations of linguistic
meaning usually started with reflecting on Aristotle’s “semantic triangle” of words, con-
cepts, and things described at the beginning of his On Interpretation (16a3–8), and the
corresponding description of ‘signifying’ as establishing some understanding of a thing
(cf. ibid. 16b21–22). These remarks on the “semantic triangle” and signification were
generally understood as asserting that conventional linguistic items, whether written or
spoken, would be meaningless sounds or inscriptions without their being convention-
ally associated with naturally representative acts of our understanding, namely: concepts.
Indeed, medieval logicians, from Boethius onward, used meaningless articulate sounds

829
 G YU LA KL I MA

and the corresponding inscriptions to illustrate the point: if I utter articulate sounds like
blithyri, biltrix, or baf, whoever hears me and does not have a language in which these
sounds happen to be associated with some acts of understanding, they have literally
no idea what I am saying; indeed, since I have no idea that I intend to convey by these
sounds either, I do not really say anything by uttering these sounds, just as I am not writ-
ing anything by idly doodling on a piece of paper. Furthermore, the conventionality of
linguistic signs is also clear from the fact that we could make any of these meaningless
utterances meaningful by simply agreeing what we mean by them, just as we managed to
make the not-too-long-ago meaningless utterance ‘google’ to mean the popular internet
search engine, and later on, by extension, the act or activity of searching on it.
Establishing the conventional relationship between utterances and acts of our under-
standing is what the medievals referred to as imposition, whether this was done (perhaps,
even ceremonially) by some name-giver(s) (as happened with ‘google’, the name) or
simply by introducing and picking up convenient usage (as probably happened with
‘google’, the verb).
Understanding the relations of the Aristotelian semantic triangle in this way may
seem perfectly harmless and uncontroversial. But already some medieval authors were
wary of what we might describe as the ‘representationalism’ apparently involved in this
position. Referring to a ‘magna altercatio’ (great controversy) over the issue, John Duns
Scotus (1265/66–1308) argued that if our words signified our concepts, then we would
all speak our private languages, talking only about our concepts, and never about the
things we live among, which is absurd [Pini 2001, 1999; Perler 1993, 2003].
However, Scotus may have used the phrase magna altercatio only to dramatic effect, to
emphasize the importance of his own position. Actually no medieval thinker interpreted
the Aristotelian semantic triangle as amounting to a commitment to talking only about
our concepts. For whether they described the relation established by imposition between
words and concepts as (immediate) signification or subordination, and whether they took
our words to signify things primarily or secondarily, they all agreed on the basics outlined
above that without associating articulate utterances with acts of understanding, those
utterances are not significative, and that when by an act of imposition they do become
significative, then they will ultimately signify whatever we conceive by the concepts to
which they are subordinated or which they immediately signify. An utterance without being
subordinated to a concept is just an articulate sound that may sound like a word of a
language, but it is not; it will gain its semantic properties through its subordination to a
concept, just as the inscription conventionally recording it will inherit its semantic prop-
erties, and both the utterance and the inscription will signify just those things that we
conceive of by the concept and just in those ways in which the concept represents them.
So, our words signify whatever we conceive of by means of the corresponding con-
cepts. But does this mean, then, that we cannot talk about our concepts, but only about
the things conceived by these concepts? This does not follow, for at least two reasons:
First, we can form concepts of concepts (such as the concepts to which the words ‘con-
cept’, ‘species’, ‘genus’, ‘negation’, etc. are subordinated, as in saying ‘Negation is a syn-
categorematic concept’), and so we can use the words subordinated to these concepts to
talk about our concepts; second, we can even use words subordinated to concepts that
conceive of things that are not concepts to refer to these concepts and not to the things
conceived by these concepts (as when we say, ‘Man is a species’).
The first reason is based on our ability to form concepts of concepts, or as the medie-
vals called them, second intentions. Although in the secondary literature these are some-

830
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

times referred to as “second-order concepts”, the medievals’ second intentions are not
to be confused with what after Frege we call “second-order concepts.” Frege’s second-
order concepts are not concepts whereby we conceive of other concepts; rather, they are
concepts that operate on other concepts, that is to say, they are syncategorematic concepts
[Klima 2006]. By contrast, second intentions are categorematic concepts, by which we
conceive of concepts insofar as they are concepts, that is, in terms of their representa-
tive, and not in terms of their ontological, features. For instance, when I say, ‘Negation
is a syncategorematic concept’ both the subject and the predicate of this sentence are
what medieval authors called “names of second intention” or “names of second imposi-
tion”, that is, names that are subordinated to concepts whereby we conceive of concepts
in terms of their function. However, if I say ‘The concept of negation is a quality of the
human mind’, the subject term of this sentence is a term of second intention, whereas
the predicate is a term of first intention, describing the concept referred to by the sub-
ject in terms of its ontological features (at least on one medieval theory of concepts).
The second reason why we can talk about our concepts, even if our words ultimately
signify the objects of our concepts, and we “normally” use our words to talk about those
objects is that we can use our words to refer to the concepts themselves to which these
words are subordinated. For instance, if I say ‘Man is a species’, the subject term of this
sentence, insofar as it is true, is not used to refer to what it is ordinarily used refer to,
namely, individual human beings (as in ‘Every man is an animal’), but rather to the spe-
cific concept of individual humans (whereby these individuals are conceived precisely
insofar as they are human), for that concept is a species on the medieval interpretation
of ‘species’ (in line with Porphyry’s definition, according to which a species is some-
thing predicable of several numerically distinct individuals). Note that in this case, in
contrast to the former, we do not refer to a concept by means of a term subordinated to
a second intention, a concept of a concept. In this case we use the term subordinated
to a concept to refer to this concept; so, it is not a term of second intention or second
imposition that is used to refer to a concept (possibly other than the concept to which
this term is subordinated) but a term of first imposition that is used with the meaning
(signification) it ordinarily has (i.e., subordinated to the concept that gives its significa-
tion) to refer to the concept to which it is subordinated.
To be sure, since an act of imposition is voluntary and hence subordination is con-
ventional, the same utterance (or the corresponding inscription) may be used not only
with the same signification whenever it is used but in different senses, imposed on dif-
ferent concepts. Thus, the same term may be used according to totally different and
unrelated significations, as imposed on totally different concepts, leading to equivocation
(as in ‘I am going to the bank’ meaning either the river bank or the financial institu-
tion), or on essentially different concepts that evoke similar imagery or some meaning-
ful associations in the hearer’s mind (as in ‘The meadow is smiling’, meaning that it is
flowering), leading to metaphor, or on essentially related concepts, one of which is the
result of some analogical extension of the other (as in ‘I see your point’, meaning that I
understand it), leading to analogy. In fact, relativizing signification to imposition, and
making imposition dependent on the intention of the speaker or interpreter, provided
medieval scholastics with a very flexible framework for handling all sorts of improper
uses and meanings, including irony, sarcasm, humor, or even cases when we use a term
as imposed on a concept of others not shared by us (as when a Christian talks about the
‘gods’ of the pagans).
However, the foregoing example of ‘Man is a species’ illustrated not any of these

831
 G YU LA KL I MA

improper uses, affecting what is meant by the term (its signification), but the case when
the same term, according to the same imposition, i.e., with the same meaning it nor-
mally has, can be used to stand for something other than what it is “normally” used to
stand for. Therefore, since the same term with the same meaning or signification can
be used to refer to different things, meaning must be distinguished from reference, or as
our medieval colleagues would put it, signification must be distinguished from supposition.
Thus, on top of the variability of meaning resulting from the variability of imposition,
the medieval theory of supposition adds another layer of variability resulting from actual
usage in context, insofar as the same term used with the same signification can still be
used to stand for different things, or, as we shall see, even for the same things, but in
different ways.

Reference: Supposition, Appellation, Ampliation and Restriction

Supposition is a property of terms in the context of a proposition, namely: their property
of standing for a thing or several things that the proposition is about. On the basis of
this general characterization, one might think that a medieval theory of supposition is
just the same as a modern theory of reference. However, even on this very general level,
there are some fundamental differences. (1) Medieval authors in general insisted that
supposition is a property of terms in a propositional context, that is to say, supposition is
essentially context-dependent: It simply makes no sense to ask what the supposition of
the term ‘man’ is; although it does make sense to ask what the supposition of the term
‘man’ is in the proposition ‘Every man is an animal’ or in the proposition ‘Every man is
a noun’, and we would get different answers in these two cases. (2) Furthermore, some
medieval authors, most notably Peter of Spain, would assign different types of supposi-
tion to the same term in the same proposition, depending on whether it is the subject
or the predicate of that proposition, while most authors would assign it the same type
of supposition but possibly in a different mode. (3) Again, supposition may only be just
one of the several referential functions a term may have in a proposition whereby it
may contribute to the compositional semantics of the proposition in which it occurs.
On some medieval theories—most importantly, on Buridan’s theory of appellation, the
oblique reference to the connotata or even to the immediate significata of some terms—their
appellation is just as important in determining the truth or falsity of propositions as their
supposition is, since their supposition is in fact determined in part by their appellation.
(4) Finally, supposition may be restricted or ampliated by the propositional context of a
term; indeed, it may be ampliated even to non-existents but still without any “ontologi-
cal commitment” to some weird nonexistent quasi-entities [Klima 2009: 162–174]. To
clarify these summary remarks, we should consider them one by one in their turn.

1. If I say ‘Socrates is a man’ and if I say ‘Socrates is a noun’ the intended referent
once is the man Socrates, once the noun we use to refer to the man in the other
sentence (or indeed that noun in the other sentence and/or any other token noun
of the same type). Of course, in contemporary written English we would indicate
our intended reference to the noun by the use of quotes, but in spoken English
we can at most gesture (and/or say, ‘quote/unquote’) to indicate quotes, and in
the primarily oral culture of Medieval Latin there was no such device available at
all. Instead, medieval logicians distinguished personal and material supposition to
distinguish the cases when a term is taken to stand for what it ultimately signifies

832
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

(according to nominalists—or what actually instantiates what it ultimately signi-

fies, according to realists; more on this later) from the case when it is taken to stand
for itself and/or for any other token of the same type. (Late medieval logicians did
introduce, however, a device similar to modern quotes: They used the prefix ‘li’ or
‘ly’ in front of a term to indicate that it is meant to stand in material supposition;
some even used a slash in the place of an unquote, especially when they used a
complex term or a proposition in material supposition.) Besides these two types of
supposition, most medieval logicians also distinguished simple supposition, when a
common term is taken to stand for the concept it immediately signifies, whether
the concept in question was taken to be some individualized mental act by which its
object is conceived or that which is conceived, precisely as conceived by such a mental
act. For instance, the nominalist William Ockham in his mature theory held that in
the proposition ‘Man is a species’ the subject term stands for individualized mental
acts of individual human minds whereby they conceive of all humans, past, present,
future, or merely possible, indifferently, in a universal manner. In his earlier theory,
however, he held that the concept for which this term in this proposition stands is
a universal object-of-the-mind, a so-called fictum, which (as its name indicates) is
not a real being, for its being consists in nothing but its-being-conceived. So-called
“realists,” however, such as Burley or Scotus, or even the more “moderate” realist
Aquinas would assign some more mind-independence to these objects-as-conceived
as far as their “foundation in reality” is concerned, although in their ontologies they
would still refrain from treating them on a par with really existing singulars. But all
these are just rather subtle metaphysical differences among these authors; semanti-
cally they would all by and large agree that a term in simple supposition stands not
for the particulars that fall under it, but rather for its immediate significate (that
is, the concept to which it is subordinated), whatever the ontological status of that
immediate significate is [Burley 2000; Klima 1993].
2. One of the most influential medieval logicians, Peter of Spain, famously argued that
the predicate terms of categorical propositions always have to stand in simple sup-
position, while the same terms as subjects can have any other kind of supposition.
Accordingly, on his position, in ‘Every man is a man’ the subject term would have
personal supposition, standing for individual humans, the individuals informed by
the ultimate significata of the term, namely, by individualized human natures, but
the predicate would have simple supposition, standing for its immediate significate,
namely, human nature as conceived in abstraction from its individuating condi-
tions. Peter’s position was actually quite unique and going against “the mainstream”
conception which would have accorded personal supposition even to the predicate
term of this sentence, although in a mode different from the subject, characterized
in terms of a different type of “descent to the singulars.” For when a common term
stands for the things that fall under it (whether they are its ultimate significata or
the things informed by its ultimate significata, on a nominalist or a realist concep-
tion, respectively), according to the medieval theory of common personal supposi-
tion, the term can stand for these same things in different ways, depending, again, on
the propositional context. For instance, in ‘Every man is a man’, the subject term
would by all authors be regarded as having confused and distributive personal supposi-
tion, characterizable by means of a “descent to the singulars,” that is, an equivalent
conjunctive proposition, in which the common subject term of the original is replaced
in each member of the conjunction by a singular term referring to one of the sup-

833
 G YU LA KL I MA

posita of the original term, as in ‘This man is a man and that man is a man and
. . .’ and so on, for all humans. By contrast, in ‘Some man is a man’ a similar descent
to the singulars under the subject term is only possible by means of a disjunctive
proposition, as in ‘This man is a man or that man is a man or . . .’, whence this term
was said to have determinate supposition. According to “the common doctrine,”
the predicate term of a universal affirmative proposition could have neither of the
two above-mentioned modes, for no equivalent descent under this term is pos-
sible, whether by means of a conjunctive or by means of a disjunctive proposition.
(Clearly, ‘Every man is this man or every man is that man . . .’ as well as ‘Every
man is this man and every man is that man . . .’ is false, provided there is more than
one man, while the original ‘Every man is a man’ is true.) However, an equivalent
descent by means of a disjunctive term is still possible: ‘Every man is this man or that
man or . . .’ is true just as well as the original is in all possible cases in which the
original is true. (Concerning these equivalences one also has to take into account
medieval considerations concerning the existential import of universal affirmatives,
to be discussed in the next section.) This case, therefore, was regarded as providing
an instance of a different mode of common personal supposition, commonly called
merely confused supposition. Considerations of completeness and symmetry would
also seem to require a fourth mode, namely: one in which descent is made to a con-
junctive term (for instance, descending from ‘Only all men are running’ to ‘Only this
man and that man and that man . . . are running’, and some late medieval authors
did in fact discuss this mode under the name of conjunct supposition (the so-called
suppositio copulatim) [Read 1991a, 1991b]. In fact, the resulting system of descents
and corresponding ascents can be quite naturally expanded into a powerful analytic
tool, matching and even surpassing the analytic abilities of standard quantifica-
tion theory (the different types of descents distinguishing universal and existential
quantifiers with different relative scopes), applicable even to recently often dis-
cussed cases of doubly numerically quantified sentences, such as ‘Two examiners
marked six scripts’ [Klima and Sandu 1990]. Nevertheless, most medieval authors
were quite content with only the three “standard” modes, apparently not motivated
by the concerns we have about such sentences. In any case, returning to Peter’s
position on the predicate term of universal affirmatives, he first remarks that assign-
ing simple as well as merely confused personal supposition to such a term would not
be inconsistent, for these two modes are not incompatible. The term might exercise
the function of referring to human nature in general, conceived in abstraction from
the individuals, as well as the function of referring to the individuals informed by
this nature, not determinately or distributively, but merely confusedly, as by means
of the corresponding nominal disjunction. However, he next argues that a common
predicate term simply does not have this function. His argument can be briefly
restated as follows. In ‘Every man is an animal’ a genus is predicated of one of its
species. But the nature of the genus multiplied in the supposita of the species is not
the genus (which is a universal, that is, something that is one over many, and thus
not multiplied with the multiplicity of the many things participating in it). There-
fore, it is not the nature of the genus multiplied in the supposita of the species that
is predicated here. But then, since the predicate of this sentence stands for what
is predicated, which is not the nature of the genus multiplied in the supposita of
the species, the predicate cannot have confused supposition, which would require
this multiplication. Based on this conclusion, Peter states that although from the

834
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

point of view of logic the nature signified by ‘man’ in its supposita is one, in the
nature of things each man has his own humanity and these humanities are distinct
on account of the matter they inform. Likewise, the nature signified by the term
‘animal’ in individual humans is one from the point of view of logic (secundum viam
logice) but is multiplied in these individuals in the nature of things (secundum viam
nature). So, the multiplication of animalities has nothing to do with the semantic
function of the predicate of ‘Every man is an animal’; indeed, in the nature of
things we find the same multiplication of animalities (namely: the multiplication
of individualized animalities of human beings), even when we consider ‘Every man
is white’ or ‘Every man is black’, where there is no mention of these animalities at
all.
3. In accordance with the foregoing, despite its differences from modern conceptions
of reference, supposition can still be regarded as a term’s context-dependent refer-
ence, insofar as supposition is the context-dependent feature of the term determin-
ing what and how it stands for in a proposition in which it is used. But direct refer-
ence to what a term stands for is not the only possible referential function a term
can have, for it may also obliquely refer to other things in relation to the things it
stands for. The most comprehensive medieval account of this sort of co-reference
is John Buridan’s theory of appellation. On Buridan’s interpretation, connotative (or
appellative) terms signify the things that fall under them not absolutely as absolute
terms do, but in relation to other things, which are their connotata. Therefore, in
a proposition in which they stand in personal supposition, besides suppositing for
the things they signify, such terms also obliquely refer to or appellate the things they
connote. For instance, the term ‘father’ is a connotative term, for it signifies men
not absolutely, but in relation to their children (in fact, it signifies all men, past,
present, future or merely possible, indifferently, just as the term ‘man’ does, but in
relation to their past, present, future, or merely possible children), who are this
term’s connotata. Accordingly, in the context of a present tense proposition, as in
‘Every father loves his child or children’, the subject term will supposit for those of
its significata that are actually related to its connotata in the way the signification of
the term demands, that is to say, it will only supposit for those men who presently
have children. Thus, clearly, appellation, the oblique reference to the connotata of
a connotative term, has an important role in determining the direct reference, the
supposition of the same term, and hence the truth of the proposition in which it
occurs. For, given the fact that the subject term of this proposition stands in com-
mon, personal, confused and distributive supposition, it is true only if the con-
junction ‘This father loves his child or children and that father loves his child or
children and . . .’ is true. However, if in a possible situation no man actually has a
child, than no man can be the suppositum of ‘this father’, whence no man can be a
suppositum of ‘father’. Accordingly, when the subject term of a universal affirma-
tive supposits for nothing (possibly on account of the lack of the relevant appellata,
as in this case), then the proposition is false; that is to say, it must have existential
import. But it is not only appellative terms that can have such oblique reference: For
absolute terms occurring in an oblique case in a complex term also appellate their
ultimate significata, and according to Buridan all terms appellate their immediate
significata (the concepts to which they are subordinated) as the direct objects of
intentional verbs. For example, in ‘Every donkey of some man is running’ (Omnis
asinus alicuius hominis currit), the complex subject term supposits for donkeys, but

835
 G YU LA KL I MA

only for those that belong to some man or another, on account of the appellation of
men by the term in the oblique case (hominis). In this case, the term clearly appel-
lates its ultimate significata. By contrast, in ‘I know the one approaching’ (Cognosco
venientem) or ‘I owe you a horse’ (Debeo tibi equum), Buridan explains the peculiar
behavior of the direct object of the verbs (lack of existential import, breakdown
of the substitutivity of identicals) by claiming that these terms in these contexts
appellate the concepts to which they are subordinated, for the mental acts signified
by these verbs (as opposed to bodily acts) concern their objects only insofar as they
are cognized precisely by means of the concept to which the direct object of the
verb is subordinated.
4. In fact, Buridan argues that as a result of this sort of appellation of concepts (appella-
tio rationis) the supposition of the terms construed with intentional verbs is ampliated,
i.e., extended beyond the range of presently existing things even in a present tense
context, for by means of the concepts appellated by these common terms we con-
ceive of all things that fall under it, past, present, future and merely possible indif-
ferently, in a universal manner. But this sort of ampliation of supposition is achieved
not only by intentional verbs but also by tenses and modalities—in general, what-
ever we would recognize as intensional contexts. These contexts, therefore, would
seem to force what we would recognize as “quantifying over nonexistents.” How-
ever, for medieval logicians, given the variability of supposition, there is not one
fixed domain of quantification in which we are ontologically committed to every
item we quantify over: We are only committed to the existence of those things of
which it is true to affirm in a present tense proposition that they exist. Of course, in
the appropriate intensional contexts we can make reference to things that existed
in the past, or will exist in the future, or can exist in a possible situation, given
our awareness of such items through our relevant mental acts, but reference to
these items in ampliative contexts will never entail a proposition about their actual
existence, so they are not parts of our ontology at all. This is not to say, however,
that medieval realists would have refused to admit in their ontologies items that
medieval nominalists would find to be “weird entities,” which, therefore, they “ana-
lyzed away” by working out the details of how our “mental language” can map our
category-rich spoken and written languages onto a parsimonious ontology, contain-
ing only two or three distinct categories of entities.

Therefore, after these points, which served to illustrate the main differences between
medieval and modern theories of reference and quantification in general, we should
turn in conclusion to some of the fundamental differences between medieval nominal-
ist and realist strategies in constructing their logical-linguistic theories, in particular
with reference to their rather different conceptions of compositionality and mental
language.

Compositionality, Mental Language, and Ontological Commitment

in Medieval Realism and Nominalism
As the foregoing discussion of Buridan’s theory of appellation illustrated, medieval logi-
cians were very much aware of the compositional character of complex expressions.
Indeed, a late-15th-century logician, Hieronymus Pardo, explained linguistic genera-
tivity precisely in terms of compositionality. He argued that in order for a proposition

836
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

to signify what it does, it does not have to be imposed to signify separately as a whole,
for the signification of the whole results from the significations of the parts, and this is
why we are capable of generating a potential infinity of meaningful propositions [Pardus
1505: 1].
However, since in accordance with the generally accepted theory of imposition and
meaning, items in our spoken and written languages inherit their semantic features from
our concepts, the naturally representative units of human thought, compositionality was
also regarded as primarily belonging to our natural mental representational system, our
mental language [Panaccio 1999, 2004; Klima 2009: 37–120]. This medieval “language
of thought hypothesis” was not so much regarded as a “hypothesis”, but as the natural,
most plausible explanation of the facts of how any human language works. Accord-
ingly, the syntactical constructions of natural languages were regarded as exhibiting,
but sometimes as masking, the underlying conceptual constructions, which therefore
could be explicated by using the syntactical constructions exhibiting them to analyze or
expound the constructions masking them.
Accordingly, nominalist logicians, such as Ockham, Buridan, and their ilk, argued
that the syntactic simplicity of some connotative terms may actually mask a complex
conceptual structure on which the simple connotative term is merely imposed for con-
venience, but the actual complex concept to which such a term is subordinated can be
explicated by means of the nominal definition of the term, a complex expression whose
syntactical structure matches the compositional structure of the complex concept in
question. In fact, this type of analysis provided the most powerful conceptual tool for
nominalist philosophers to eliminate what they regarded as the absurd or just unneces-
sary ontological commitments of their realist opponents. Take for instance a privative
term (one that indicates a defect, the lack of a naturally desirable characteristic), such
as ‘blind’. For realist philosophers (even for such a “moderate” realist as Aquinas), this
term, and primarily the concept to which it is subordinated, has to signify the lack of
sight in animal that ought to have it by nature. But what is such a lack of sight, a pri-
vation? It certainly cannot be an entity on a par with a sight, which was taken to be a
positive quality, a natural power enabling an animal to see, for blindness is precisely
the lack, the nonbeing of such an entity. But on the prevailing theory of predication
of the realists, an affirmative predication is true just in case the ultimate significate of
the predicate is actually present in the suppositum of the subject; thus, ‘Homer is blind’
is true, just in case the blindness of Homer is present in Homer’s eyes. But since the
presence or existence of the blindness is nothing but the absence or nonexistence of his
sight, Homer’s blindness cannot be regarded as an entity existing in the same sense as
his sight would, so it was classified as a special kind of entity, a so-called being of reason
(ens rationis), the presence of which is conditioned both on how things are (in this case,
on Homer’s not having sight) and on how they are conceived, namely: in terms of the
concept of blindness whereby we conceive of this situation by forming this privative
concept.
By contrast, nominalists, unwilling to buy into the dubious ontology of such quasi-
entities, argued that the nominal definition of ‘blind’, namely ‘animal not having
sight’ clearly indicates that there is no need to assume the (“halfway”) existence of
such spurious entities, for this nominal definition shows that the concept to which
‘blind’ is subordinated consists of two absolute concepts, whereby we merely conceive
of “ordinary” entities, namely, animals and sights: absolutely, a connotative, relative
concept, whereby we conceive of these same entities in relation to one another, and

837
 G YU LA KL I MA

the syncategorematic concept of negation, which simply has the effect that the result-
ing complex concept, signifying animals and connoting their sights, will supposit in a
mental sentence precisely for those animals that do not have their sight. Thus, on this
analysis there is no need to assume the existence of any spurious entity, only ordinary
substances and their qualities.
There is, however, a curious thing about the compositionality of mental language.
Medieval philosophers were generally committed to the view that our mental acts must
be ontologically simple qualities of our minds. How could they possibly talk, then, about
complex concepts, if they are supposed to be simple entities? And how could the seman-
tic values of these complex concepts be dependent on the semantic values of their
components, if they are simple, meaning that they cannot have any parts? The answer
to these questions is that semantic compositionality in mental language does not have
to go together with syntactic construction in the way it does in spoken and written
languages. For a complex concept does not have to be complex on account of being
some (structured) aggregate of several distinct parts; it is enough if it is merely semanti-
cally complex in the sense that its semantic values are functionally dependent on the
semantic values of other (simpler or simple) concepts, and possibly on other contextual
factors, such as imposition, time, place, speaker, listener, pointing, etc. [Klima 2001:
xxxvii–xxxix]. In fact, quite interestingly, the “realist” Cajetan argues for the same idea
of compositionality in mental language, as not requiring any sort of ontological aggrega-
tion of several simple concepts: The semantically complex concept can be ontologically
as simple as the semantically simple ones, for the semantic complexity of the complex
concept merely consists in the functional dependency of its semantic content on the
semantic content of the semantically simple ones.
What nominalists and realists do disagree on is rather what this semantic content
is, and, accordingly, how one should conceive of the task of the logician in accounting
for it in a compositional semantics. Very briefly, realists are “ontologically nonchalant”
in their semantics, and start worrying about ontological commitments only in their
metaphysics. By contrast, nominalists are “ontologically parsimonious,” eliminating any
unwanted or unneeded ontological commitments in their semantics, although they may
actually allow certain types of entities in their metaphysics or physics for other than
semantic reasons, for as far as their semantics is concerned, they could do just as well
without them. But the reason for this difference in strategies is exactly how they con-
ceive of semantic content. For realists, the semantic content of our mental acts is precisely
the information they carry about their ultimate objects, no matter what sorts of concepts
we are talking about and what sorts of objects these are. So, for realists all our concepts
are representative, whether they are categorematic or syncategorematic, simple or com-
plex. This is how they end up in their ontologies with entities and quasi-entities cor-
responding to all sorts of linguistic categories—such as modes-of-things-as-conceived,
corresponding to syncategoremata, universals, privations, negations, relations of reason,
all beings of reason—corresponding to categorematic terms involving in the formation
of their content some mental operation; and enuntiabilia, dicta, complexe significabilia
or real propositions, corresponding to mental propositions or judgments. By contrast,
for nominalists, the contents of our mental acts are the things conceived themselves.
Accordingly, for a staunch nominalist like Buridan, the semantic contents of our propo-
sitions will not be anything distinct from the contents of their terms: both our terms and
our propositions are about the same things but representing them differently—singularly
or universally, affirmatively or negatively, in simple or a complex fashion—whereas all

838
 M EDI EVA L PH I LO SO PHY O F L A N G U A G E

these differences in the mode of representation are accounted for in terms of conceptual
construction or other genealogical features of our concepts, determined by the process
of their formation to be accounted for in psychology. Thus, for Buridan, logic will have
nothing to do with semantic content in the way realists conceived of it: The truth or
falsity of our propositions will depend only on the identity or nonidentity of the sup-
posita of their terms, whereas the information conveyed by them will be a matter of
psychology. In a word, nominalist logic is extensionalist, as opposed to the essentially
intensional logic of the realists.

Bibliography
Ashworth, E. J. (1977) Tradition of Medieval Logic and Speculative Grammar, Toronto: Pontifical Institute of
Mediaeval Studies.
Augustine, De dialectica, ed. Jan Pinborg, Boston: Dordrecht, 1975
Augustine, De doctrina Christiana, ed. Joseph Martin, Turnhoult: Brepols, 1962.
Burley, W. (2000) On the Purity of the Art of Logic: The Shorter and Longer Treatises, translated and edited by
P.V. Spade, New Haven and London: Yale University Press, 86–90.
Bursill-Hall, G. L. (1971) Speculative Grammars of the Middle Ages: The Doctrine of the partes orationis of the
Modistae, Approaches to Semantics, 11, The Hague: Mouton.
Dominik, P. (2003) “Duns Scotus’s philosophy of language,” in: T. Williams (ed.) The Cambridge Companion
to Duns Scotus, Cambridge: Cambridge University Press,161–192.
Gracia, J. and Newton, L. (2010) “Medieval Theories of the Categories,” The Stanford Encyclopedia of Phi-
losophy (Summer 2010 Edition), Edward N. Zalta (ed.), forthcoming http://plato.stanford.edu/archives/
sum2010/ entries/medieval-categories/
Klima, G. (1993) “The Changing Role of Entia Rationis in Medieval Philosophy: A Comparative Study with
a Reconstruction,” Synthese 96: 25–59.
Klima, G. (2001) Introduction to John Buridan: Summulae de Dialectica, an annotated translation with a
philosophical introduction, New Haven: Yale University Press.
Klima, G. (2006) “Syncategoremata,” in: Keith Brown (ed.) Elsevier’s Encyclopedia of Language and Linguis-
tics, 2nd Edition, Oxford: Elsevier, 353–356.
Klima, G. (2008) “The Medieval Problem of Universals,” in Edward N. Zalta (ed.), The Stanford Encyclopedia of
Philosophy (Winter 2008 Edition), http://plato.stanford.edu/archives/win2008/entries/universals-medieval/
Klima, G.and Sandu, G. (1990) “Numerical Quantifiers in Game-Theoretical Semantics,” Theoria 56:
173–192.
Marmo, C. (1999) “The Semantics of the Modistae,” in Sten Ebbesen and Russell L. Friedman (eds.), Medi-
eval Analyses in Language and Cognition, Acts of the Symposium, ‘The Copenhagen School of Medieval
Philosophy’, January 10–13, 1996, Copenhagen: Royal Danish Academy of Sciences and Letters and
C. A. Reitzels Forlag, 83–104.
Meier-Oeser, S. (2008) “Medieval Semiotics,” in Edward N. Zalta (ed.), The Stanford Encyclopedia of Philoso-
phy (Fall 2008 Edition), http://plato.stanford.edu/archives/fall2008/entries/semiotics-medieval/
Panaccio, C. (1999) Le discours intérieur de Platon à Guillaume d’Ockham. Paris: Éditions du Seuil.
Panaccio, C. (2004) Ockham on Concepts. Aldershot-Burlington: Ashgate.
Pardus, H. (1505) Medulla Dyalectices, Anabat-Gerlier: Paris.
Perler, D. (1993) “Duns Scotus on Signification,” Medieval Philosophy and Theology 3: 97–120.
Perler, D. (2003) “Duns Scotus’s Philosophy of Language”, in T. Williams (ed.), The Cambridge Companion
to Duns Scotus. Cambridge: Cambridge University Press, 161–192.
Peter of Ailly (1994) Destructiones modorum significandi, edited by Ludger Kaczmarek, Amsterdam: John
Benjamins Publishing Company.
Pini, G. (1999) “Species, Concept, and Thing: Theories of Signification in the Second Half of the Thir-
teenth Century,” Medieval Philosophy and Theology 8, 2: 21–52.
Pini, G. (2001) “Signification of Names in Duns Scotus and Some of His Contemporaries,” Vivarium 39:
20–51.
Read, S. (2008) “Medieval Theories: Properties of Terms,” in Edward N. Zalta (ed.), The Stanford Encyclope-
dia of Philosophy (Fall 2008 Edition), http://plato.stanford.edu/archives/fall2008/entries/medieval-terms/

839
 G YU LA KL I MA

Read, S.L. (1991a) “Descensus copulatim: Albert of Saxony vs. Thomas Maulfelt,” in J. Biard (ed.), Itinerai-
res d’Albert de Saxe: Paris-Vienne au XIVe, Paris: J. Vrin, 71–85.
Read, S.L. (1991b) “Thomas of Cleves and Collective Supposition,” Vivarium 29: 50–84.
Spade, P.V. (2008) “Medieval Theories of Obligationes,” in Edward N. Zalta (ed.), The Stanford Encyclopedia
of Philosophy (Fall 2008 Edition), http://plato.stanford.edu/archives/fall2008/entries/obligationes/
Sweeney, E. (2008) “Literary Forms of Medieval Philosophy,” in Edward N. Zalta (ed.), The Stanford Encyclope-
dia of Philosophy (Fall 2008 Edition), http://plato.stanford.edu/archives/fall2008/entries/medieval-literary/

840
 7.3
MODE RN P HILOSOPHY
OF L ANG UAGE
Michael Losonsky

1 Introduction
The history of modern philosophy roughly covers seventeenth- and eighteenth-
century European philosophy, a prolific period spanning from Descartes to Kant. It is in
this period that the philosophy of language first becomes a viable subdiscipline, as the
greatest philosophers of the time turned to the study of language in a manner very dif-
ferent from their forebears of previous centuries. In the modern period, human natural
language became a distinct object of systematic philosophical study expressed in sub-
stantial and self-contained texts. The semantic, grammatical, and pragmatic structures
of natural language come to be seen as domains worthy of the kind of sustained study
that before had been devoted to logic or metaphysics. Moreover, philosophers of this era
turned to language in order to solve problems in other areas of philosophy. The study of
language was not just a side interest or a source of evidence for already established philo-
sophical principles but was seen as necessary for solving philosophical problems. Finally,
central themes and strategies that define the philosophy of language in later centuries,
including contemporary analytic philosophy of language, emerge during this period.
Many factors contributed to the turn to language in the seventeenth century, a major
one of which was a loss of confidence in the received Aristotelian view about the rela-
tions between language, mind, and reality. Aristotle had maintained that the spoken
sounds of natural languages are “signs of affections of the soul” and “these affections
are likenesses of actual things” (De Interpretatione 16a4–7). Moreover, while there is
a diversity of natural languages, affections of the soul as well as actual things are “the
same for all.” All natural languages mirror a shared set of ideas and concepts that in turn
mirror a shared reality.
Toward the end of the sixteenth-century, philosophers worried that ideas are not
the same for all and that they do not reliably mirror reality. The scientific revolution
had cast doubt on the assumption that human ideas represent actual things. Moreover,
philosophers had come to fear that natural languages were inherently vague or ambigu-
ous and that they contain expressions for outdated and inaccurate conceptions of real-
ity that could impede scientific progress. This loss of confidence that there is a deeper
unity below the surface of linguistic diversity was a key factor in the emergence of the
philosophy of language.
There were two basic responses to the ruptures in the Aristotelian unity of mind, lan-
guage and reality. Roughly speaking, linguistic universalists sought to demonstrate that
 M I CH A EL LO S O N S K Y

there indeed was a deeper unity, either a universal grammar or a shared logical structure,
beneath the diversity of natural languages. Linguistic pluralists viewed the diversity of
linguistic structures as an essential feature of language. There is no deeper unity to all
languages because their diversity reflects the diversity of human minds. What follows is
an overview of both types of response.

2 Linguistic Diversity

Francis Bacon (1561–1626)

Bacon does not devote a major text to natural language, but his influential writings
raised issues that turned seventeenth-century philosophy to language. Bacon’s lifelong
project was to single-handedly replace Aristotle’s logic. He called his logic the New
Organon (1620) to indicate that he intended it as a replacement of Aristotle’s treatises
on logic, which had been collected by Aristotle’s students under the heading Organon
(The Instrument). Bacon rejected Aristotle’s focus on the deductive syllogism, believ-
ing that it only serves the justification of what is already known, not the discovery of
new knowledge. He held that logic should give “True Directions for the Interpretation of
Nature,” (the subtitle of the New Organon) and Bacon argued that the proper instru-
ment for discovery is what he called “induction,” that is, a method for systematically
organizing our observations and drawing conclusions from them.
Bacon believed that his method was a remedy for certain systemic errors in human
cognition that he described as the “idols” or illusions of the mind. Bacon identified
human natural languages as a source of illusions, namely the idols of the marketplace or
town square [idola fori]. Language is used to communicate human knowledge, including
scientific knowledge, but the terms of natural languages are defined according to the
understanding “of the vulgar” or the uneducated. Consequently, many words do not
refer to anything and many words that do refer are ambiguous or poorly defined.
Introducing better definitions informed by science cannot stop this illusion. Defini-
tions, according to Bacon, are always verbal definitions that depend on other words,
and inevitably some of these more basic words have meanings and uses determined
by uneducated people because language must “suit the understanding of the common
people” (2000: 42). Ordinary appearances, however, rarely match how things really are.
Only observation directed and organized by a scientific method gives us a better insight
into nature, and only an educated elite is capable of scientific work.
For Bacon, then, human language is not just a mirror of the human mind, but it also
structures the mind. While human beings assume that “their reason governs words,” in
fact “it is also true that words retort and turn their force back upon the understanding”
(2000: 48). Bacon has a striking metaphor for this feedback loop: “Words, as a Tartar’s
bow, do shoot back upon the understanding of the wisest, and mightily entangle and
pervert judgment” (2001: 137–8).
Since the harmful effects of language on human cognition cannot be avoided, Bacon
is reserved about attempts to construct a scientific or “philosophical” language consist-
ing only of words that “are the footsteps and prints of reason.” The congruence between
“words and reason is handled sparsism, brokenly” (2001: 141–2). Nevertheless, Bacon
is not entirely dismissive of the possibility of a philosophical language, and he licensed
such projects. While there are only a scattered few words that are the footsteps of
reason, words are not entirely deficient in this respect and hence Bacon says he

842
 M O DERN PH I LO SO PH Y O F L A N G U A G E

“cannot report it deficient” and recommends that there should be a science devoted
to the development of a philosophical language. However, this language would consist
only of conventional and arbitrary signs imposed voluntarily by human beings. Bacon
is wholly dismissive of the idea that a philosophical language can be constructed out of
“real characters,” such as ideograms or hieroglyphs, where there is a likeness or similar-
ity between the character and what it represents.

Thomas Hobbes (1588–1629)

Hobbes, Bacon’s personal secretary between 1619 and 1623, accepted Bacon’s critique
of natural language. Hobbes famously wrote, “words are wise mens counters, they do
but reckon with them; but they are the mony of fooles” (1996: 28–9). Only a fool uses
words without paying attention to their meanings, namely the ideas they signify. The
“first abuse” of speech is speaking with wrong or no definitions, and correct speech pays
attention to the meanings of words and supplies univocal, clear, and consistent mean-
ings. Accordingly, the first step of any science is to start with “the right Definition of
Names” (1996: 28).
However, unlike Bacon, Hobbes assigns a positive role to language in human cogni-
tion. Natural language is so important for Hobbes that he devotes a full chapter of the
Leviathan to natural language just before the fifth chapter “Of Reason and Science”
(1996: 20). He believes that without language human beings cannot generalize, and
without generalization neither reasoning nor science are possible. For example, without
language human beings cannot think about classes, properties, or kinds. Hobbes main-
tains that there are only individual things and there is “nothing in the world Universall
but Names; for the things named, are every one of them Individual and Singular” (1996:
26). Moreover, language allows human beings to order their thoughts over longer peri-
ods of time and order them with a method involving general principles. For example,
arithmetic calculations as well as logical deductions require the languages of arithmetic
and logic. Accordingly, in the Leviathan Hobbes defines reason as “nothing but the
Reckoning (that is, Adding and Subtracting) of the consequences of generall names
agreed upon, for the marking and signifying of our thoughts” (1996: 32). In other words,
reasoning is a kind of symbolic computation, and for this reason Hobbes has been seen
as the grandfather of Artificial Intelligence (Haugeland 1989). Hobbes even tries to
sketch the rudiments of a language for computing logical inferences on the basis of a
finite set of operations (Hobbes 1981).
Since language is essential to reason and science, creatures without language, includ-
ing young children that have not yet acquired language “are not endued with Reason
at all.” Moreover, reason itself is not innate to human beings but acquired through
“study and industry” because for Hobbes languages are artifacts constructed by human
beings (1996: 21–3). Thus, all that depends on language, including science, has a con-
ventional component. Hobbes believed that science could achieve only “conditionall
Knowledge,” namely knowledge of the consequences of the conventional definitions
that stand at the beginning of any science.
Since all languages are products of voluntary activity, there will be a diversity of
languages answering to the diversity of psychological and physical constitutions. Bacon
already described human psychological diversity as “Idols of the Cave,” namely errors
due to “the peculiar constitution, mental or bodily, of each individual; and also educa-
tion, habit, and accident” (2000: 41). For Hobbes it is a fundamental fact of the human

843
 M I CH A EL LO S O N S K Y

condition that while nature is the same for all people, their individual natures, disposi-
tions, desires, and needs are not the same, and so there is a “diversity of our reception”
of nature (1996: 31, also 25).
Accordingly, Hobbes’s philosophy of language has no room for a philosophical lan-
guage or a universal language other than a language that for political reasons has been
imposed on people by a sovereign ruler. Hence, unlike Bacon, Hobbes does not endorse
universal or philosophical language projects.

John Locke (1632–1704)

In Book III of An Essay Concerning Human Understanding (1689:1975) Locke devel-
ops Hobbes’s line of linguistic thought into a full-fledged philosophy of language that
arguably is not just “the first modern treatise devoted specifically to the philosophy
of language,” but the first treatise devoted specifically to the philosophy of language
(Kretzmann, 1968: 175–6; Losonsky 2001: 5–21). He maintains that questions about
what there is as well as questions about the extent of human knowledge cannot be
answered properly without understanding the nature of natural language, particularly
the meaning of natural language. Locke decided to write a book with the title Of Words
because he found “that it is impossible to speak clearly and distinctly of our Knowledge
. . . without considering, first, the Nature, Use, and Signification of Language” (1975:
401). Language became so important in Locke’s mind, that he recommended that one
of the three branches of science is semantics: “sēmeintikē, or the Doctrine of Signs, the
most usual whereof being Words” (1975: 720).
Locke followed Hobbes in locating the meaning of language in the human under-
standing: “Words in their primary or immediate Signification, stand for nothing, but
the Ideas in the mind of him that uses them” (1975: 405). The meaning of language,
then, rests on ideas that are internal to the individual speaker’s mind. To be sure, words
can also refer to natural objects, for example water or gold, but this reference is only sec-
ondary and mediated by the ideas that words signify. For Locke, human beings succeed
in referring to objects external to their minds only if they have ideas that refer to those
objects, and it is an abuse of language to assume that words directly refer to external
objects without the mediation of human ideas (1975: 497).
Locke assigned the study of language such a central role because he concluded that
language plays a constitutive role in the formation of the ideas of the understanding.
A problem for the meaning of language is that the ideas of the human understanding
are unstable and human beings need greater order and stability than is found in human
experience (1975: 464). Locke believed that human conscious experience begins with a
rich but confusing array of individual simple ideas, for example ideas of particular shapes
and shades of color, but human beings need complex ideas, for example, ideas of objects.
According to Locke, without language the human understanding would not be able to
regularly think about these ideas: “ ’Tis the Name which is, as it were the Knot, that ties
them fast together” (1975: 434).
Complex ideas that require language, according to Locke, are ideas of sorts, species,
or kinds that have necessary and sufficient conditions. Neither nature nor human expe-
rience has such sharp boundaries, and it is only by abstracting certain ideas and annex-
ing a name to them that human beings are able to form sortal ideas with necessary and
sufficient conditions. Locke believes that moral ideas also require language (1975: 432
and 479).

844
 M O DERN PH I LO SO PH Y O F L A N G U A G E

For Locke all ideas are private—they are features of our individual caves, to use
Bacon’s terminology—and so human communication is a source of doubt and uncer-
tainty, especially when it comes to complex ideas (1975: 476–7). Human beings cannot
simply assume that speaker and hearer have the same ideas. Many words for complex
ideas that individuals have cannot be translated (1975: 432–2). At best, speakers and
hearers can aim to signify similar ideas with their words, and this requires effort in the
form of the rectification of ideas. Rectification is the process for determining that two
human beings have ideas that conform to each other by trying to fix ideas to a standard
in nature. While rectification improves communication, it is always fallible and never
yields absolute certainty (Losonsky 2007: 292–6).
Needless to say, Locke did not think that a universal or philosophical language was
possible. While natural languages can be improved by being careful about the significa-
tion of words, the idea of a perfect language that avoids the imperfections of language is
“ridiculous” (1975: 509). Without a system of ideas that is stable across time and shared
by all human beings, a universal language, as long as it is assumed that the meaning of
natural language is determined by the ideas of the human understanding, is impossible.
Not even formal logic can provide an underlying unity. Formal logic is “useless” and just
a “Web of perplexed Words” that undermines natural language, which is “the Instru-
ment and Means of Discourse, Conversation, Instruction, and Society” (1975: 494–5).
Locke’s philosophy of language is a culmination of a development that begins with
a line of thought that emphasizes the diversity of human ideas, the dependence of the
meaning of language on the ideas of the human understanding, the artifactual nature of
natural language, the resulting diversity of human languages, and the role language plays
in structuring the human understanding. For Locke, this means that philosophers must
pay attention to human language when addressing metaphysical and epistemological
issues, and more generally, human beings, including scientists, must curtail their aspira-
tions to knowledge and temper their confidence even where knowledge is possible.

Linguistic Unity

René Descartes (1596–1650)

The responses just sketched make clear that in the modern period it was no longer
possible, without additional justification, to assume a reliable correspondence between
the human mind and reality. Nevertheless, the Aristotelian assumption that linguistic
unity rests on an underlying psychological unity continued to thrive in seventeenth-
century pursuits of a universal language as well as the allied pursuits of a universal gram-
mar or logic.
An influential moment in seventeenth-century linguistic universalism is Descartes’
letter in 1629 (1991, III: 10–13) responding to Marin Mersenne (1588–1648), who had
raised the question of prospects for universal language projects. Descartes replies that
universal language projects are possible only if all human ideas have a combinatorial
structure like the set of natural numbers. Just as an infinite set of numbers can be known
on the basis of a finite set of numbers and operations, a universal language would require
that all human ideas are “compounded” out of a finite set of simple ideas and a finite
set of operations. Given this structure of ideas, the universal language would then be
mapped onto this order of ideas in the way that the language of arithmetic corresponds
to the set of natural numbers. While Descartes dismisses the suggestion that such a

845
 M I CH A EL LO S O N S K Y

universal language will be widely adopted and used, he believes that such a universal
language is possible because human ideas indeed have a combinatorial structure. The
discovery of such a structure would require a “true philosophy,” but Descartes believed
that this is attainable.
Unlike Bacon, Hobbes, and Locke, Descartes does not worry about the instability
of human thinking. What is unstable, Descartes maintains, is human imagination and
sensation, and those who argue for psychological diversity confuse ideas with images
and sensations. Ideas, Descartes maintains, are pure mental conceptions that even a
disembodied being without images and sensations can have, and these mental concep-
tions have a combinatorial structure on their own independently of language. Moreo-
ver, Descartes maintains with Aristotle that different human beings can have the same
mental conceptions: “A Frenchman and a German can reason about the same things”
(1991, II: 126).
Finally, even though natural languages are ambiguous and irregular, they neverthe-
less succeed in representing the mind’s combinatorial structure. In the Discourse on the
Method of Rightly Conducting One’s Reason and Seeking the Truth in the Sciences (1637)
Descartes writes that with language human beings express their thoughts insofar as
they are able to “produce different arrangements of words so as to give an appropriately
meaningful answer to whatever is said in [their] presence” (1991, I: 140). Descartes
believed that only rational beings can have a language with a combinatorial structure,
and thus possession of language distinguishes beings with reason from beings without
reason, including nonhuman animals (1991, I: 140–1).

Antoine Arnauld (1612–1694) and Claude Lancelot (1615–1695)

If reason is common to all human beings and has a determined combinatorial structure
that is captured by the grammatical structure of natural languages, then there is a gram-
matical structure that is common to all human languages—a universal grammar. This is
the hypothesis that was developed by Arnauld, a sympathetic reader and critic of Des-
cartes, together with the grammarian Lancelot in General and Rational Grammar (1660),
commonly referred to as the Port-Royal Grammar. In the Preface Lancelot writes that
his work on Greek, Latin, Italian, and Spanish grammars led him to questions about the
unity and diversity of the grammars of natural languages, and Arnauld provided him
with a strategy for handling these questions.
A key idea of the Grammar is that language’s primary purpose is to express thoughts,
which have a structure that can be examined and understood without words or any
other signs, and in order to serve this purpose, language has a grammatical structure in
common with the universal structure of thought (1975: 66). The main aim of Arnauld’s
and Lancelot’s Grammar is to explain grammatical structure in terms of the structure
of thought. Since they assumed that the structure of thought is rational, the universal
grammar was also a rational grammar.
The pursuit of a universal grammar alone does not distinguish the Grammar from
Scholastic predecessors. The basic strategy of Scholastic universal grammars was to
describe grammatical categories such as noun or gender in terms of properties of real
objects (Law 2003: 173–77). This assumes that human conceptions of these objects
are reliable, but early modern philosophers were under severe pressure to worry about
reliability. Accordingly, what is new about Arnauld’s and Lancelot’s approach is that
they tie linguistic categories explicitly to mental categories rather than ontological cat-

846
 M O DERN PH I LO SO PH Y O F L A N G U A G E

egories. Arnauld and Lancelot develop their mental categories on the basis of a fun-
damental distinction between the “object of our thought” and the “form or manner of
our thought,” both of which “occur in our minds” (1975: 67). The object of thought is
the conceptual content of thought, and content is literally something that is grasped or
intuited by the mind, even in cases where we are thinking about something that does
not exist. Under the heading manner of thought Arnauld and Lancelot include proposi-
tional functions such as conjunction and disjunction, as well as propositional attitudes
such as a belief, desire, question, or command.
The various kinds of objects of thought and manners of thought Arnauld and Lance-
lot identify are mental categories, not linguistic categories. For example, when they dis-
cuss propositions, subjects, and predicates, Arnauld and Lancelot assume that these have
a status that is independent of language and that they can be known without language.
However, natural languages, as instruments for expressing our thoughts, have corre-
sponding grammatical categories. Nouns, articles, pronouns, prepositions, and adverbs,
for example, are grammatical categories for terms that signify the objects of thought
while verbs, conjunctions, and interjections are grammatical categories for words that
signify the manners of thought. Arnauld and Lancelot then proceed in a familiar Aris-
totelian fashion, for instance, explaining grammatical nouns and adjectives in terms of
the distinction between substance and accident. However, the substance/accident dis-
tinction in the Grammar is not intended to be a metaphysical distinction that carves out
what there is, but a distinction between different kinds of ideas or conceptual contents
(1975: 73). Human beings have concepts of substances and attributes, and Arnauld and
Lancelot assume with Descartes that human beings can know what concepts they have
and how they are related to other concepts without assuming that we know whether or
not these concepts correspond to or represent anything real.
Arnauld’s and Lancelot’s Grammar is a self-contained treatise devoted to the nature
of the meaning of natural languages, and insofar as the philosophy of language is con-
cerned with semantic questions, this is an important text in the history of the philosophy
of language. However, there is a sense in which Locke’s Book III of the Essay remains
“the first modern treatise devoted specifically to the philosophy of language.” Arnauld’s
and Lancelot’s discussion of grammar is not intended to make a broader contribution to
philosophy in the way Locke’s discussion of language contributes to his epistemology,
metaphysics, and philosophy of mind. Their discussion of language assumes a broadly
Cartesian conception of the mind, but they do not even make the minimal claim that
the Grammar confirms or corroborates their conception of the nature of the human
mind.

John Wilkins (1614–1672) and Jakob Böhme (1575–1624)

While Arnauld and Lancelot find the unity in the diversity of languages in a universal
and determinate mental structure, other seventeenth-century universalists aimed to tie
language directly to objects. There were two main approaches for preserving the Aris-
totelian view that languages are tied to objects. One approach took its cue from the
natural sciences while the other approach assumed that there was a language of nature
that expressed the essences of objects.
Language of nature doctrines were tied to the religious belief that the language of
nature is the language God used in creation and gave to Adam. They were also com-
bined with a disdain for human volition. But there were other features of the language

847
 M I CH A EL LO S O N S K Y

of nature doctrines that were equally important, including that language is not con-
ventional, that it is acquired by human beings simply in virtue of being human without
learning or instruction, and that language naturally and accurately represents reality.
A major source for seventeenth-century language of nature doctrines was the widely
translated work of Böhme, who believed that certain sounds correspond to the essential
properties of objects, and that these sounds comprise a language that can be under-
stood simply by hearing the sounds. In England this view was defended in 1654 by John
Webster (1610–1682) in a widely read pamphlet calling for the reform of university
education (Debus 1970). Webster laments that the language of nature, which is “not
acquired by us, or taught us by others,” but “infused” and “innate,” is neglected in the
curriculum (26).
A pamphlet defending the university curriculum and attacking the very idea of a lan-
guage of nature was published by Wilkins and the mathematician and astronomer Seth
Ward (1617–1689). They rejected the assumption that what human beings acquire
by their “own industry” is essentially deficient (Debus, 3–4). The natural sciences are
a product of human industry, as are complex ideas and human languages. Words are
always tied to objects by voluntary human decisions.
Wilkins was the author of the best-known attempt to develop a universal language,
his Essay Towards a Real Character and a Philosophical Language (1668). Wilkins, a co-
founder of the Royal Society, dedicated to the “Improvement of Natural Knowledge,”
wanted to develop a language that would serve the advancement of science. Wilkins’
Essay is a prolegomena or preparatory study for a universal and philosophical language.
The key idea is that this language consists of a finite and unambiguous set of simple
terms tied to simple ideas. These terms can be combined using a finite set of rules to
form more complex terms and sentences. The vocabulary and rules of this language
correspond to ideas and structures of the human mind that are common to all human
beings. Wilkins assumes, following Aristotle, that human beings have “the same Princi-
ple of Reason” and “the same Internal Notion or Apprehension of things” (1968: 20).
This universal language is also a philosophical language because the ideas that this
language expresses correspond to reality. Words will “signify things” and the relation-
ships between these words, as expressed in definitions, will be “suitable to the nature
of things” (1968: 21). The reason the terms of this language refer to objects is that the
terms correspond to internal notions that correspond to reality. Although this appears
to rehearse Aristotle’s view that the affections of the soul correspond to real objects,
Wilkins was a product of the scientific revolution and he acknowledged the epistemic
concerns raised by it. Wilkins aimed to tie his language to reality by constructing it on
the basis of a “Theory . . . exactly suted to the nature of things” (1968: 21). For Wilkins,
a theory that best fits the nature of things is a scientific theory, that is, a theory con-
structed according to scientific method, based on empirical evidence, and verified by
the scientific community. For Wilkins, science is the source for the knowledge of things
as well as their structure, including the predicamental structure he uses, that is ordering
everything in terms of genus, difference, and species.
Wilkins was well aware that his classification was defective and only provisional, and
that it would be improved over time and with scientific cooperation (1968: Epistle and
21). This suggests that Wilkins’s method should be seen as anticipating strategies of
contemporary naturalism which holds that the natural sciences are appropriate sources
for information about what there is, and that philosophy, including the philosophy of
language, should rely on this information.

848
 M O DERN PH I LO SO PH Y O F L A N G U A G E

Gottfried Wilhelm von Leibniz (1646–1718)

Leibniz offers an alternative strategy for tying words to things, namely relying on the
correspondence between relationships. A language will have “a kind of complex mutual
relation or order” and this linguistic structure can correspond to the structure of things
(Leibniz 1970: 184; see Losonsky 2001: 64–5). Two distinct structures can have a one-
to-one correspondence in the way a circle and ellipse can correspond to each other:
for every point on a circle there is a point on the ellipse and the points are ordered in
the same way. Thus human beings do not need to rely on a conventional connection
between language and reality. If language represents reality, it will be in virtue of cor-
responding structures, and according to Leibniz whether these correspondences obtain
is a nonarbitrary and nonconventional matter of fact.
By relying on corresponding structures, Leibniz offered a new way of identifying the
universal structure of all languages. There is a universal linguistic structure if the diverse
structures of natural languages correspond to each other. An important consequence
of this approach is that human beings need not be aware of structures and their cor-
respondences. Responding to Locke’s view that human thoughts are private and fleet-
ing, Leibniz distinguished between “actual thoughts” or “noticeable thoughts,” on the
one hand, and the “very form or possibility of those thoughts” (1996, 300–1). Actual
thoughts are the conscious thoughts a person has at a time, and these passing thoughts
are typically momentary and private. But fleeting conscious psychological states can
also have a stable form that need not be noticed by the person in those states. The
search for universality, then, is not a search for some conscious state that all human
beings have but for the form of those states.
One form that concerns Leibniz is semantic content. Actual thoughts have content, or
what Arnauld and Lancelot called “objects of thought.” Leibniz identifies semantic con-
tent in terms of logical relations, specifically logical entailment. One important feature
of a proposition is its logical relations to other propositions: what it entails, what it does
not entail, what it contradicts, and so forth. These logical relations are not constructed—
Leibniz describes them as the “natural order of ideas”—and they are shared if there is a
correspondence between the ideas of different people. Moreover, insofar as sentences of
natural languages have semantic content, they also have a universal structure, namely
their logical form [forme logique] (1996, 480). Thus the search for a universal structure for
all natural languages is the search for a universal logical form of all natural languages.
Leibniz identifies two tools for uncovering the logical form of natural language. The
first is a logical calculus, which he began developing in his Dissertation on the Art of
Combinations (1666). Adopting Hobbes’s suggestion that all reasoning is computation,
Leibniz actually develops a calculus that represents mathematically the formal structure
of valid inferences. With this calculus it is possible to exhibit the entailment relations
between sentences and thus exhibit their logical form. The second tool is the substi-
tution of semantically equivalent terms and phrases. Leibniz believes that an orderly
sequence of substitutions of equivalent terms and phrases can transform an ordinary
sentence of a natural language into a more precise sentence that uses the logical calculus
to explicitly express the logical structure of the original sentence. This transformation
or translation counts as an analysis of the original sentence. Leibniz also offers a tool
for identifying equivalent terms: “Those are the same of which one can be substituted
for the other with truth preserved,” namely the Principle of Substitutivity Salva Veritate
(1966, 34, 53 and 122).

849
 M I CH A EL LO S O N S K Y

However, for Leibniz there is more than just a correspondence between language
and mind. Leibniz departs from the orthodox Aristotelian and Cartesian conception of
human language as a mirror of the mind and agrees with Hobbes and Locke that natural
languages and other symbol systems have a constitutive role to play in human reason-
ing. Leibniz maintains that human beings cannot reason without symbols (1996, 77
and 212; 1966b, 450). Just as a person can calculate quickly and successfully by writing
numerals and function symbols on a sheet of paper without being conscious of the mean-
ing of the numerals, human beings think when speaking even though they do not have
an additional idea or image. This is not only true of public speech, but when thinking
privately without speaking, the train of thoughts can consist of ideas of words (1966b,
451). Although language plays a constitutive role in human thinking, this does not
undermine the thesis that there are universal components to human thought because
language can express a universal order, namely a logical order.

Conclusion
Seventeenth-century philosophers of language developed several basic themes that
endured as the philosophy of language developed in subsequent centuries. Arguably,
there is a tradition beginning with Locke that emphasizes linguistic and psychological
diversity and gives increasing importance to the role human language has in struc-
turing the human mind (Aarsleff 1982). Linguistic diversity and relativity charac-
terize the eighteenth-century philosophies of language of Étienne Bonnot Condillac
(1715–1780) and Johann Gottfried Herder (1744–1803). However, in the eighteenth-
century emphasis shifted from individual psychological differences to broader social
psychological and anthropological differences. For example, Condillac maintains that
nations have different characters that shape language even more than individual char-
acters, and this yields the special character or “genius of a language” (2001, 185–195).
This trend continues in the nineteenth century, especially in the work on non-Indo-
European languages by Wilhelm von Humboldt (1767–1835).
The pursuit of a universal grammar also continues in the eighteenth century. Horne
Tooke (1736–1812) aimed to combine Locke’s philosophy of mind and language with
a universal grammar. James Harris’s (1709–1780) Hermes or a Philosophical Inquiry Con-
cerning Language and Universal Grammar (1751) aims to combine a universal grammar
with anthropological concerns. Not only do individuals have ideas, Harris writes, but
also nations have ideas—“the genius of their nation”—and some nations have better
ideas and better-organized ideas than others (1751, 407–8).
While Leibniz’s logical calculus based on mathematical functions influenced eight-
eenth-century mathematicians Leonhard Euler (1707–1783), Gottfried Ploucquet
(1716–1790), and Johann Heinrich Lambert (1728–1777), the project of exhibiting the
logical form of natural languages was dormant in the eighteenth century. For instance,
Immanuel Kant’s (1724–1804) interest in the logical form of human cognition did not
turn into a concern for the logical form of natural language. This concern reemerges in
the nineteenth century in the writings of Bernard Bolzano (1781–1848), John Stuart
Mill (1806–1873), and Gottlob Frege (1848–1925). All three isolated the proposition
as a unit of logical and linguistic analysis and distilled the features of propositions that
are relevant to logical validity. However, Bolzano and Frege, acknowledging Leibniz’s
influence, sharply separate logic and psychology, while Mill follows Locke in identifying
logical relations with psychological operations.

850
 M O DERN PH I LO SO PH Y O F L A N G U A G E

Despite their differences, Bolzano, Mill, and Frege all agree with Locke that the prob-
lems of philosophy cannot be solved without a careful study of language and its mean-
ing. Insofar as Bolzano, Mill, and Frege are major sources of twentieth-century and
contemporary philosophy of language, particularly the philosophy of language in the
analytic tradition, philosophy of language today is rooted in the turn to language by
seventeenth-century philosophers.

References
Aarsleff, H. (1982) From Locke to Saussure: Essays on the Study of Language and Intellectual History. University
of Minnesota Press.
Arnauld, A. and Lancelot, C. (1975) General and Rational Grammar: The Port-Royal Grammar. Translated
by J. Rieux and B. Rollin. Mouton.
Bacon, F. (2000) The New Organon. Translated by L. Jardine and M. Silverthorne. Cambridge University
Press.
Bacon, F. (2001) The Advancement of Learning. Ed. S. J. Gould. Random House.
Condillac, E. B. de. (2001) Essay on the Origin of Human Knowledge. Cambridge University Press.
Debus, A. G. (1970) Science and Education in the Seventeenth Century: The Webster-Ward Debate. American
Elsevier.
Descartes, R. (1991) The Philosophical Writings of Descartes. 3 Vols. Translated by J. Cottingham, R. Stoot-
hoff, D. Murdoch, and A. Kenny. Cambridge University Press.
Harris, J. (1751) Hermes: or, A Philosophical Inquiry Concerning Language and Universal Grammar. Woodfall.
Haugeland, J. (1989) Artificial Intelligence: The Very Idea. MIT Press.
Hobbes, T. (1981) Computation sive Logica: Logic. Translated by I. C. Hungerland, A. Martinich, and G. R.
Vick. Abaris Books.
Hobbes, T. (1996) Leviathan. Cambridge University Press.
Kretzmann, N. (1968) “The Main Thesis of Locke’s Semantic Theory.” Philosophical Review 77: 175–96.
Law, Vivien. (2003) The History of Linguistics in Europe: From Plato to 1600. Cambridge University Press.
Leibniz, G. W. v. (1966a) Logical Papers. Translated and edited by G. H. R. Parkinson. Clarendon Press.
Leibniz, G. W. v. (1966b) Deutsche Schriften. 2 vols. Ed. G. E. Guhrauer. Olms.
Leibniz, G. W. v. (1970) Philosophical Papers and Letters. Ed. L. Loemker. Reidel.
Leibniz, G. W. v. (1996) New Essays on Human Understanding. 2nd ed. Translated by P. Remnant and J.
Bennett. Cambridge University Press.
Lewis, R. (2007) Language, Mind and Nature: Artificial Languages in England from Bacon to Locke. Cambridge
University Press.
Locke, J. (1975) An Essay Concerning Human Understanding. Oxford University Press.
Losonsky, M. (2001) Linguistic Turns in Modern Philosophy. Cambridge University Press.
Losonsky, M. (2007) “Language, Meaning and Mind in Locke’s Essay.” In The Cambridge Companion to Locke’s
“Essay Concerning Human Understanding.” Ed. L. Newman. Cambridge University Press, 286–312.
Wilkins, J. (1968) An Essay Towards a Real Character and Philosophical Language. The Scolar Press.

851
 7.4
FR E GE , RU SSELL,
A ND WITTG ENST EIN
Michael Potter

Frege, Russell, and Wittgenstein were founders (although not by any means the sole
founders) of the analytic tradition in philosophy; but they did not found the philosophy
of language, which has roots stretching back much further. Their principal contribu-
tion, indeed, could be regarded as being in the opposite direction: it is not so much that
they applied philosophical methods to the study of language as that they applied linguis-
tic methods to the study of certain problems in philosophy. In the course of this work
they did develop ideas which shed light on language and how it functions. However,
even this must be heavily qualified, since their main contributions were much more to
the philosophy of logic (the study of the inferential role of sentences) than to the phi-
losophy of language (the study of how language means what it does). In the summary
that follows we shall focus on the morals that can be drawn from their contribution for
modern work in the philosophy of language.

Frege
It is worth noting straightaway that mathematics played an important role in shaping the
philosophical ambitions of all three of the philosophers we shall be considering. Frege
was a mathematician for all of his professional life; Russell began as one and much of what
he wrote had the philosophy of mathematics as its focus; and Wittgenstein, although he
did not train as a mathematician, announced when he arrived at Cambridge in 1911
that it was the philosophy of mathematics that he wanted to study with Russell. The
desire to secure the philosophical foundations of mathematics which all three of these
philosophers shared plays a large part in explaining the attitude to language which—to
begin with, at least—they chose to take. Frege began with the ambition of formulating
a precise language—a Begriffsschrift, he called it—for the expression of thoughts. Where
the thoughts to be captured are mathematical, Frege’s ambition was an entirely plau-
sible one, which he went some way towards realizing. It was natural, once the method
had been shown to be successful in mathematics, that he and others should consider
applying it more widely. However, the difficulties of formalization are much more severe
where nonmathematical language is concerned, and the relationship between natural
languages and their formalized correlates remains problematic to this day.
Perhaps the most remarkable of Frege’s contributions to philosophy was what is
nowadays called the linguistic turn (a term popularized by Rorty but coined by Gustav
 F REG E, RU SSELL, A N D W I T T G E N S T E I N

Bergmann). This was the movement that placed a concern with language at the center,
not only of the philosophy of mathematics but of all of philosophy. What made it cen-
tral was the realization that we approach the world via thought, but have no access to
thought except via language. Hence, it was claimed, our inquiry into the nature of the
world is best conducted via an inquiry into the nature of language.
One of the most important contributions made by Frege was to place language at
the center of philosophical, and in particular metaphysical, inquiry by recognizing its
importance as a route to the structure of our thinking about the world. This approach
was encapsulated in what has become known as the context principle, namely the
injunction that we should not enquire after the meaning of a word in isolation but only
in the context of a sentence. Frege formulated this principle in order to apply it to the
case of number-words, and hence to rule out the question as to what the numbers are,
posed independently of the question how number-words are used in sentences. Applied
more generally, the context principle led during the twentieth century to an approach
which has become known as internalist. This approach seeks to discuss metaphysical
questions about the structure of the world by means of a discussion of the structure of
the language in which we represent the world.
If it were only a route to internalism, the context principle would not amount to an
important contribution to the philosophy of language. However, it has also been influ-
ential within that area, as it has led to an extended debate about the overall shape that
theories of meaning should have. Applied in its strongest form, the context principle
would require that to understand a single word I need to understand the whole of lan-
guage, and hence that each time I learn a new word my understanding of every other
word must change slightly. Donald Davidson (1967, p. 308), for instance, has advanced
the context principle in this strong form. Many twentieth-century advocates of the con-
text principle, of whom Michael Dummett is the most notable, have preferred to limit
its application so that it does not have this holistic consequence.
Another Fregean contribution which has influenced the philosophy of language is
his notion of thought and, derivatively, of sense. Thoughts are not, according to Frege,
mental entities in the mind of an individual thinker: they are intersubjectively avail-
able and hence cannot be wholly any one person’s property. It is worth stressing here
that the publicity of thoughts is not itself a distinctively Fregean contribution. What
is distinctive is that Frege combined this with a second element, namely that what he
meant by a thought was only that part of the content of the sentence that is relevant to
inference. The residue, which he called “tone,” is no doubt important for the philoso-
phy of language but is not the concern of logic. By means of this restriction Frege hoped
to explain the independence of logic from the mind. His target, that is to say, was in
the first instance the view not that language is psychological but that logic is. What he
railed against was the view, popular in the late nineteenth century, that logic is a codifi-
cation of how we think. Frege held that this is utterly wrong-headed: logic, he claimed,
codifies the laws of truth, not of thought; the laws of how we ought to think if we wish
to aim at the truth, not of how we do actually think.
Much of the later interest in Frege’s notion of sense has focused on its application at
the subsentential level, to proper names in particular. In “On sense and reference” Frege
argued that in order to explain the informational content of identity sentences such as
“Hesperus=Phosphorus” we cannot appeal merely to what the two sides of the equation
refer to, since this is the same in both cases (namely the planet Venus). Nor can we
explain the content solely in terms of the difference between the words “Hesperus” and

853
 M I CH A EL P O T T E R

“Phosphorus,” since that would, according to Frege, make the content linguistic rather
than, as it should be, astronomical. Nor can we base our account on the difference in the
ideas which I associate in my mind with the two words, since these, being mental, are
intrinsically private, whereas the informational content of the equation is something
public—something which you can learn from me when I tell it to you. For these reasons
Frege felt driven to the conclusion that the contributions of the words “Hesperus” and
“Phosphorus” to the meanings of sentences in which they occur—what he called the
senses of the words—are different, but are inter-subjectively available items which are
not to be identified either with the words themselves or with anything mental.
In “On sense and reference” (1980a) Frege motivates the notion of sense for proper
names primarily on the basis of the need to explain intersubjective communication.
The notion that there could be such a thing has seemed puzzling to many philosophers
since then, from Russell to Kripke and beyond. When asked some years later to justify
his view, Frege invoked (1980b, p. 80) the example of two explorers, one of whom
comes across a mountain called Aphla, the other a mountain called Ateb. Only when
they plot their discoveries on a map do they come to realize that Aphla and Ateb are the
same mountain, but seen from different directions. In the form in which Frege told the
story, it seems to make again the point about publicity: the notion of sense is required
if we are to explain how communication between individuals is possible. It is interest-
ing to note, however, that the problem arises even if we revise the story so that only
one explorer is involved. On this retelling, the moral of the story seems to concern the
nature of the world as much as the structure of thought. It is because the world has unex-
pected and unknown aspects that we need a notion of sense capable of making identity
statements nontrivial.
I have said that Frege’s principal concern was with the philosophy of logic, not of
language. In a few places, though, Frege did go beyond these narrow bounds. This is par-
ticularly notable in “On sense and reference,” where the discussion of indirect speech
goes some way beyond anything that would have been necessary if his interest had been
strictly limited to the foundations of mathematics.
One ongoing theme in Frege’s writings is the complicated relationship between lan-
guage and thought. In his late writings he seems largely to have abandoned the idea that
language was a good guide to the structure of thought, but his reasons are not wholly
clear. The most obvious one might be his paradox of the concept horse, but he did not
himself take this paradox as seriously as he might have done: It was left to Wittgenstein
to do that. More important for Frege seems to have been the failure of his logicist project
for grounding arithmetic. He seems to have thought that this was somehow due to a
mismatch between language and thought, but he was unable to articulate the nature of
the mismatch with any clarity.

Russell
What is often taken to be Russell’s most famous contribution to the philosophy of lan-
guage is his theory of descriptions (Russell, 1905), according to which sentences con-
taining definite descriptions should be analyzed so that the description disappears and
its role is taken over by quantifications. However, Russell did not originally intend this
theory as a contribution to the philosophy of language at all, and for two reasons. The
first was that, like Frege, he was at this stage more interested in the logical rather than
the grammatical form of propositions. The difference in form between “John is bald”

854
 F REG E, RU SSELL, A N D W I T T G E N S T E I N

and “The present King of France is bald” is one which emerges when we consider how
these sentences behave when we negate them. The latter sentence has two negations
with different truth conditions (“The present King of France is not bald” and “It is not
the case that the present King of France is bald”), whereas the former only has one
negation. Russell conceived of this difference as a logical rather than a grammatical one.
For him, therefore, the theory of descriptions was a demonstration that the logical and
grammatical forms of a sentence may be different.
The second reason that Russell did not see himself as contributing to the philosophy
of language is one that emerges when we recall the reason why he was interested in
the King of France’s hair in the first place. For him this was a proxy for the much more
important case of the class of all classes which do not belong to themselves. In other
words, if he was concerned with language at all, it was only the language of mathemat-
ics, not that of ordinary life. So his aim was not to give an account of all the various
uses of the word “the” in ordinary language. The fact that his theory does not cope
well with “The whale is a mammal” did not seem to him to be an objection to it. How
far Russell was from conceiving of “On denoting” as a contribution to the philosophy
of language may be gauged from his response to Strawson’s “On Referring.” Strawson
had criticized the theory of descriptions because there are various features of our use
of definite descriptions which it fails to analyze correctly. Russell’s reply was that he
was “totally unable to see any validity whatever in any of Mr Strawson’s arguments”
(Russell, 1957, p, 385).
During the most philosophically influential part of his working life—the decade or so
immediately preceding the First World War—Russell did not begin by being interested
in language in its own right at all. Nonetheless, “On denoting” did have the effect of
forcing him to consider the role of language, even within mathematics, more carefully
than hitherto, because it showed him that the surface grammar of a sentence may mis-
lead us as to the true form of the thought the sentence expresses. He came to realize, as
Frege had by a different route, that if language is a medium for the expression of thought,
it is not a wholly transparent one.
Russell’s work during his most productive phase contributed to a developing research
project which might be thought of as an analysis of language but was heavily influenced
by epistemological concerns deriving from Descartes. The underlying language, that is to
say, was to consist of elements corresponding to whatever is most immediate in experi-
ence. The project then consisted in the construction, from these elements, of linguistic
items corresponding to the objects of the external world as we ordinarily conceive of it.
Even when Russell gave more explicit attention to language in his later work (such as the
Inquiry), the approach he took was shaped by the epistemological concerns of his project,
thus limiting the interest of this work to anyone who does not share these concerns.

The Tractatus
The Tractatus is famous for its advocacy of a so-called picture theory of language, but
this is at best a misnomer, since what constitute pictures of the world are, according to
the theory, not the sentences of ordinary language but the propositions which those
sentences express. For this reason “picture theory of thought” would be a less misleading
title for what Wittgenstein intended.
According to the picture theory what enables a picture to represent a situation is that
the structure of the picture is identical to the structure of the situation. If the aim of

855
 M I CH A EL P O T T E R

semantics is to explain how propositions come to have meanings, then according to the
picture theory there is only one task which semantics needs to perform, namely that of
connecting the components of the proposition with the components of the world. The
further task of explaining how the way in which the components are assembled succeeds
in being expressive of a certain situation drops away, since it is just the same way that
the components of the world are assembled if the proposition is true. For Wittgenstein,
therefore, the reason why thought succeeds in latching onto the world is that its struc-
ture is identical to that of the world.
I have stressed that it is not the sentence but the proposition (or, in Frege’s language,
the thought) which according to the picture theory shares its structure with the world.
In order for sentences also to share this structure it would be necessary for us to construct
a language which mimicked it. In the Tractatus Wittgenstein went some way towards
constructing such a language, although at times he seems to have been more concerned
with explaining how Frege’s and Russell’s formal languages fell short of this ideal than
in the details of how his own construction was supposed to work.
The idea that the world is one that is represented in thought is at the heart of Witt-
genstein’s atomism, since he moves (1992, 2.021–2.022) from an argument that deter-
minate thoughts are atomistic to the conclusion that the world is correspondingly ato-
mistic. Moreover, in a further simplifying move he claimed that these atoms are simple
not only in the sense that they have no parts but also in that they do not have distinct
aspects: An atom which is presented to me in one way cannot later be presented to me
in another. By this means Wittgenstein ruled out the possibility which Frege had used
the example of Aphla and Ateb to highlight. From this Wittgenstein derived his essen-
tially solipsistic conclusion that there cannot be different perspectives on the world.
One theme which lies just below the surface of much of Frege’s philosophical writing
is the relationship between thought and the world. Frege took it as given that the struc-
ture of thought mirrors the structure of the world, so that the distinction between satu-
rated and unsaturated components of the thought corresponds precisely to a distinction
between saturated and unsaturated components of the world. However, Frege’s concep-
tion of objects prevented him from holding that this neat isomorphism holds within
these grammatical categories. Wittgenstein, by rejecting Frege’s notion of objects, was
able to extend Frege’s conception of thought mirroring the world much further.
The picture theory can thus be seen as representing the high point of Frege’s linguis-
tic turn as a method in metaphysics, since it holds out the hope that if we constructed an
ideal language, and thereby uncovered the structure of thoughts, we could then read off
the structure of the world directly from them. However, the logically perfect language
is offered in the Tractatus more as an ideal than a practical reality, and by the time he
finished the book he had begun to doubt whether the ideal was realizable at all.
By sidelining ordinary language in this manner Wittgenstein left rather obscure what
philosophical purpose is served by studying it. He said that “all propositions of our collo-
quial language are actually, just as they are, logically completely in order”(1922, 5.5563),
but it is an interesting question quite what he meant by this, and it is a tension that
remains unresolved in the Philosophical Investigations (see Wittgenstein 1953, §98).
Nonetheless, Wittgenstein still advanced in the Tractatus the ambition of using the
analysis of language to solve philosophical problems. He proposed a model according to
which most of the things we ordinarily say could in principle be expressed in the ideal lan-
guage. Some things we say resist translation into this ideal language and are thus revealed
to be nonsense masquerading as sense. In particular, the Tractatus problematizes much of

856
 F REG E, RU SSELL, A N D W I T T G E N S T E I N

the philosophy of language, since its explanations of how language comes to have mean-
ing inevitably make use of the very features of language which it seeks to explain.
One instance of this is the notion of a grammatical category. In later life Wittgenstein
was fond of quoting, as a guide to good philosophizing, a maxim of Hertz to the effect
that our minds will “cease to ask illegitimate questions” when we remove unnecessary
variables from our specification of a problem. In the Tractatus we can see Wittgenstein
applying this maxim in order to reject the conception of grammatical categories as akin
to pigeonholes into which words can be sorted (a conception which he accused Russell
of holding). According to Wittgenstein the “grammatical category” to which a word
belongs is a label for certain structural features of its meaning, not something which can
be separated from the word and discussed in isolation from it.

Later Wittgenstein
One way of summarizing the difference between early Wittgenstein and late is that in
his later work he abandoned hope of a certain sort of tidying up of language. So what in
the Tractatus is described as “the language which alone I understand” (5.62) is replaced
by a motley of overlapping language games. Whatever the role of ideal language really is
in the Tractatus, it is clear that assertoric sentences are there taken as primary, whereas
in the Philosophical Investigations what is emphasized from the start is how many and vari-
ous are the roles that language has.
The later Wittgenstein also gave up on Frege’s notion of sense, if by that is meant the
idea of something quite determinate which a sentence, on a particular occasion of utter-
ance, can be said to mean. In the Investigations Wittgenstein offers numerous examples
intended to loosen our grip on this kind of notion of meaning. Modern philosophers of
language influenced by Wittgenstein sometimes suggest that meaning is so sensitive to
context as to make the project of supplying a theory to account for it almost hopeless.
Another issue which the later Wittgenstein addressed was the relationship between
the private and the public. I mentioned earlier that Frege characterized the mental as
private and emphasized, by contrast, the intersubjective availability of the content of
our utterances, from which he deduced that these contents cannot be mental. What
Wittgenstein did was to question what role the private realm of the mental can ever play
in explaining linguistic meaning. Understanding the sense in which a private language
is impossible is important, he thought, since it does a great deal to delimit our explana-
tion of how public language is possible.
Indeed, one of Wittgenstein’s main targets in his later work was the whole concep-
tion on which Russell’s Cartesian project was based. He questioned, that is to say, the
assumption that our knowledge of our own ideas is somehow more certain than any-
thing else, and hence that the correct method of epistemological enquiry is to work from
the inside outwards. This critique, if correct, is of methodological importance for the
philosophy of language much more widely, since the Cartesian assumption is often used
as an unargued premise to motivate the direction in which explanations proceed.
In the Investigations Wittgenstein also offered a critique of an “Augustinian” account
of language, by which he meant an account which sees it as possible to build up a theory
of linguistic meaning with names as the starting point. In a way, Wittgenstein’s objec-
tion seems now to be a variant of the one he had against what he saw as Russell’s ten-
dency to think of grammatical categories as being like pigeonholes. Augustine, he says,
“describes the learning of human language as if the child . . . already had a language,

857
 M I CH A EL P O T T E R

only not this one” (1953, §32). The extent to which the child might be said to “have a
language, only not this one” is of course something that remains controversial.
So far, though, all of this is largely negative. Was there anything positive which the
late Wittgenstein had to offer? It is sometimes said that he advanced a “use theory of
meaning,” but on its own this tells us very little, since it is hard to see how a credible
theory of meaning could fail to appeal to how we use words and hence be in some sense
a “use theory.” What Wittgenstein intended, however, seems to have been not so much
a theory as a method. This may well go some way towards explaining the apparent nega-
tivity of much of his later philosophy (philosophy as therapy).
Another suggestion that has been made is that ordinary language philosophy is an
application of Wittgenstein’s methods. It was he, after all, who described his task as
being “to bring words back from their metaphysical to their everyday use” (1953, §116).
However, as a matter of history the influence of the later Wittgenstein on Oxford ordi-
nary language philosophers such as Austin and Ryle seems to have been rather slight,
and the commonality of their concerns somewhat superficial.

Morals
What morals do these authors offer for modern work in the philosophy of language? The
issues we have been discussing are almost all related to the notion of content—whatever
it is that is expressed by a piece of language. As we have noted here, their focus on logic
led these authors to focus on narrow notions of content from which those aspects not
relevant to inference have been excluded. Yet such narrow notions plainly do not tell
the whole story. Moreover, even within the narrow scope of logic, we should take note
of the difficulties our authors found when they tried to determine what this content is.
It is curiously hard to find an appropriate notion of content that is stable enough to have
a determinate structure capable of precise theoretical investigation.
It follows, too, that it is important not simply to assume that the link between lan-
guage and thought, which the founding fathers highlighted, is a wholly transparent one.
The starting point is the idea that a sentence of one language can have the same content
as one in another. The task of the translator, one might say, is to preserve as much of the
content as possible in the process of translation. But as soon as we suppose that there is
a notion of content which can be preserved, we have a puzzle as to its structure, since
languages differ widely in how they say things.
Another moral is the need for consideration of the role of outlying cases. Language is
messy, and almost every generalization one cares to make has exceptions. What is not so
clear is how seriously we should take the exceptions. Our attitude to them will presum-
ably depend on what we take the purpose of our work to be. If we have metaphysical or
epistemological pretensions, then perhaps a few counterexamples are not so troubling.
If, on the other hand, our purpose is to explain language as it exists, any policy of ignor-
ing counterexamples would need to be theoretically motivated. We need, that is to say,
to decide what our purpose is first, and only then proceed to offer a theory.
Above all, though, we need to think through the implications of two points made by
Wittgenstein but deriving ultimately from Frege: first, that it makes no sense to enquire
about a grammatical type independent of its instances; second, that it remains one of
the central challenges for the philosophy of language to explain how its account of
meaning can be related to our own mental lives.

858
 F REG E, RU SSELL, A N D W I T T G E N S T E I N

Bibliography
Davidson, D. (1967), ‘Truth and meaning,’ Synthese, 17: 304–23.
Frege, G. (1980a), “On sense and reference,” in Translations from the Philosophical Writings of Gottlob Frege,
edited. and translated by P. Geach and M. Black, 3rd edn, Oxford: Blackwell, pp. 55–78.
Frege, G. (1980b), Philosophical and Mathematical Correspondence, Oxford: Blackwell.
Russell, B. (1905), “On denoting,” Mind, 14: 479–93.
—— (1957), “Mr Strawson on referring,”, Mind, 66: 385–9.
Wittgenstein, L. (1922), Tractatus Logico-Philosophicus, London: Kegan Paul, Trench, Trubner.
—— (1953), Philosophical Investigations, Oxford: Blackwell.

Further reading
Heck, R., and R. May (2006), “Frege’s contribution to philosophy of language,” in E. Lepore and B. Smith
(eds.), The Oxford Handbook of Philosophy of Language, Oxford University Press.
Potter, M. (2009), chapters 8 and 26 of Wittgenstein’s Notes on Logic, Oxford University Press.

859
 7.5
LO GICAL P OSIT IV ISM
AND Q UINE
Sanford Shieh

In the middle third of the twentieth century logical positivism was a dominant strand
of analytic philosophy. Positivism began with a group of philosophers, mathematicians,
and scientists—associated with the Vienna Circle and the Berlin Society for Scientific
Philosophy—interested in philosophical understanding of modern science. Under the
influence of Ludwig Wittgenstein, the Vienna Circle, especially Moritz Schlick, Hans
Hahn, and Rudolf Carnap, came briefly to espouse a form of Humean empiricism incor-
porating the techniques of the mathematical logic developed by Gottlob Frege and
Bertrand Russell. On this view, call it classical positivism, language is not only a central
subject of philosophy but also the principal instrument of philosophizing. Classical posi-
tivism, as popularized by Ayer (1936), became identified with positivism tout court. But
the positivists themselves were aware of an array of difficulties with the classical doc-
trines, and soon moved away in a number of diverging directions. Carnap, in particular,
reached a thoroughgoing deflationary and pragmatic anti-metaphysical view, advocat-
ing the replacement of traditional philosophy by the construction of formal languages.
W. V. O. Quine, student and friend of Carnap, engaged in a more than decade long
debate with Carnap over Carnap’s mature views. Quine’s arguments came to be widely
accepted as a decisive refutation of positivism, thereby helping to end its influence in
the analytic tradition. Recent work has increasingly shown how little their debate has
to do with classical positivism, and how inconclusive Quine’s arguments against Carnap
are, thereby opening the way to a better appreciation of the depth and philosophical
interest of their disagreements (see, inter alia, Awodey and Carus 2007, Creath 2003,
Ebbs 1997, and Ricketts 2003).

Classical Tractarian Positivism

Positivism is famously associated with a verificationist theory of meaning, on which, in
order for a sentence to be meaningful, it has to be associated with a method of verifica-
tion that is ultimately based on sensory experience. The initial notoriety of the Vienna
Circle stems from their use of this criterion of meaningfulness to argue that metaphys-
ics is not false but cognitively meaningless nonsense. The verification principle comes
from the Circle’s commitment to empiricism, the view that sense experience is the
only source of genuine knowledge about the world. But the distinctive logical form of
empiricism advanced by the positivists arose from an apparent incompatibility between
 LO G I CA L PO SI TI VI S M A N D Q U I N E

empiricism and their attachment to modern science as the paradigm of genuine knowl-
edge. A key problem is that mathematics is an integral part of science, but it is unclear
how mathematical knowledge could originate in sense experience. Experience might
always be different from the way it is, so any truth based on experience is contingent. In
contrast, we have no clear conception of how mathematical truths might be false. The
same point evidently holds of logical truths. Thus, mathematical and logical knowledge
seem to be a priori, and so to require some faculty of rational intuition, paving the way
to metaphysics. Indeed, when combined with verificationism, these features of logic and
mathematics lead to the unwelcome result that they are no more cognitively meaning-
ful than metaphysics. So, “anyone who wants to advocate a consistent empiricism must
face up to” this problem (Hahn 1980: 40).
The Circle saw, in Wittgenstein’s Tractatus (1922), the solution. On their reading of
the Tractatus, sentences in general represent the world by expressing truth functions of
elementary sentences, which, in turn, represent by picturing atomic facts. The truth-
functional mechanism of linguistic representation guarantees the existence of tautolo-
gies, sentences that are true no matter what are the truth-values of elementary sentences
composing them. Logic consists of tautologies, sentences that owe their truth, not to
correct depiction of the world but to the nature of linguistic representation. Since logic
is not about the world, our knowledge of logic does not rest on the sensory sources of
genuine knowledge of worldly facts. Rather, it rests on knowledge of how we represent
the world in language, of the rules or conventions (that we tacitly accept as) governing
the use of language, in short, on knowledge of meaning. The apparent apriority and
necessity of logic can now be taken to have a linguistic, rather than factual, origin.
Clearly, this Tractarian view of logic has to be extended for it to yield the desired
consistent empiricism. It has to cohere with verificationism—this was accomplished
by taking elementary sentences, the fundamental Tractarian unit of representation, to
depict, and so be verifiable by, immediate sensory experience. And it has to be applied
to mathematics—here the hope was to carry out a type of logicist reduction of math-
ematics, to tautologies rather than Frege’s or Russell’s formulations of logic. Modulo
the execution of this project, the positivists had a consistent empiricism and a basis for
rejecting metaphysics as neither verifiable nor tautologous.
Three sets of problems with this view were discussed in the Circle. First, exactly
what notion of verification is relevant to meaning? Is it actual, or merely in principle?
Is it conclusive, or merely probable confirmation? Is it tied to occurrences of sensations
or observations of macroscopic physical objects? Eventually there seemed to be some
convergence on verification as confirmation in principle. But it was never clear how
verification contributes to the empirical meaning of dispositional and theoretical terms
of science (see Hempel 1965).
Second, the main obstacle for Tractarian logicism is the presence of unbounded
infinitary quantification in mathematics; it is, to say the least, unclear how such sen-
tences can be finite truth-functional combinations of elementary sentences. A similar
problem afflicts a Tractarian account of the statements of science: If to be meaningful
they have to be finite truth-functions, then they can be no more than summaries of
extant observations; this rules out laws, unbounded quantifications ranging over poten-
tial observations.
Finally, what is the status of the verification principle itself, and more generally, the
products of positivist philosophizing? Are these sentences empirically contentful? Or
are they tautologies? As Carnap puts it in the form of an objection, “[i]n consequence of

861
 SA N F O RD S H I E H

your view your own writings . . . would be without sense, for they are neither mathemati-
cal nor . . . verifiable by experience” (1935: 36). In the background here is the question
whether classical positivism represents a genuine escape from metaphysics. The appar-
ently meta-philosophical critique of the nonsensicality of first-order metaphysical sen-
tences seems to rest on certain first-order claims about the nature of meaning. Should
these purported claims turn out to be as nonsensical as traditional metaphysics, it’s hard
to avoid the conclusion that the positivist “ready to prove that metaphysics is impos-
sible is a . . . metaphysician with a rival theory” (Bradley 1897: 1).

Carnapian Positivism

A Transitional View
Carnap attempted to work out the details of Tractarian positivism using David Hilbert’s
(1925) conception of infinitary mathematics as contentless sequences of signs whose
only legitimate use is for deriving contentful finitary mathematical sentences. Initially,
this conception was harnessed to solve the twin problems of unbounded quantifica-
tion mentioned above. The role of the contentless was to be played by unbounded
quantifications, and that of the contentful by tautologous truth-functions in the case of
mathematics and nontautologous ones in the case of science. Technical difficulties fac-
ing the mathematical case eventually led Carnap to abandon the Tractarian framework
altogether in favor of the Hilbertian one (see Awodey and Carus 2007: 176–9).
Language became conceived as a calculus, “a system of rules” (Carnap 1937b: 4)
rather than a system for picturing extra-linguistic facts. The study of language thus
conceived was at first called “meta-logic” and eventually “logical syntax.” The model
for this discipline is Hilbert’s meta-mathematics. Hilbert conceives of the standards
governing mathematical proof as relations among sentences considered as meaningless
finite sequences of objects. Analogously, Carnap’s pure syntax studies the rules consti-
tuting a calculus in terms of properties of and relations among “possible arrangements”
of “elements . . . about . . . which nothing more is assumed than that they are distrib-
uted in various classes” (ibid.: 4–5). The rules of a language are divided into two main
categories: Formation rules specify which sequences of elements are expressions of the
language, transformation rules specify the inferential properties and relations among
expressions.
The question now is, how does this conception of language fit with empiricism and
with the supposed nonempirical legitimacy of logic and mathematics? One account pro-
ceeds in terms of an analytic-synthetic distinction (hereafter AS). Analyticity is Carnap’s
successor to the Tractarian positivist notion of tautology. Since language is seen as a set
of rules, the semantic notion of sentences true in virtue of the mechanism of representa-
tion is replaced by a syntactic notion of sentences true in virtue of the rules of the lan-
guage. Empirically meaningful sentences are synthetic. Synthetic sentences are divided
into observational and theoretical: The former acquire their empirical content by direct
sensory observation, the latter via observational consequences. Logic and mathemat-
ics are precisely the rules governing the derivation of observational consequences. So
transformation rules are logico-mathematical axioms and rules of inference.
Meta-mathematics also yields an account of the status of philosophy. Hilbert’s insight
is that properties of and relations among finite sequences are precisely the subject mat-
ter of contentful finitary mathematics, so mathematical justification can be studied

862
 LO G I CA L PO SI TI VI S M A N D Q U I N E

mathematically. Analogously, Carnap’s logical syntax—meta-logic—is the mathemati-

cal study of the rules of language. Once philosophy is “replaced by . . . logical syntax”
(1937b: xiii), philosophy is mathematics, non-empirical but not nonsensical.

Logical syntax
Carnap’s (1937b) (hereafter Syntax) contains a fundamentally different view. To begin
with, transformation rules are not restricted to logic and mathematics: “[w]e may . . .
also construct a language with extra-logical rules of transformation [including] amongst
the primitive sentences the so-called laws of nature. . . . It is possible to go even further
and include . . . observation sentences” (1937b: 180; hereafter all emphases are Carnap’s
unless otherwise specified). Moreover, Carnap does not even require the transformation
rules to be given as divided into logical and physical rules. Purely syntactically specified
transformation rules “determine . . . the distinction between logical and descriptive symbols,
. . . and further, between logical and extra-logical (physical) transformation rules, from which
. . . ‘analytic’ arises” (1937b: 168–9). So in Syntax AS is not merely language-relative but
is grounded on primitive syntactic (inferential) characteristics of each language, rather
than extra-linguistic metaphysical or epistemological notions like fact or verification by
sensory experience.
These points are reinforced by Carnap’s advocacy of the formal over the material
mode of speech. The material mode involves the use of “pseudo-object sentences,”
which are “formulated as though they refer . . . to objects” (1937b: §74, 285). The formal
mode consists of syntactic counterparts of these pseudo-object sentences. For example,
The moon is a thing; five is not a thing, but a number
is in the material mode;
‘Moon’ is a thing-word . . .; ‘five’ is not a thing-word, but a number-word
is its formal mode correlate. Carnap characterizes philosophical claims in the material
mode as “dangerous,” “obscure,” and “misleading,” and asserts that “[t]ranslatability into the
formal mode of speech constitutes the touchstone for all philosophical sentences” (1937b: 313).
Thus a classical positivist thesis such as
Logical and mathematical sentences do not state anything about the facts
should be replaced by a syntactic sentence, perhaps:
Sentences analytic in L are L-determinate.
So, for Carnap, philosophical notions like fact are replaced by syntactic features of lan-
guages like L-determinacy. Indeed, classical positivism’s fundamental contrast between
linguistic convention and extra-linguistic empirical fact is replaced by syntactic distinc-
tions. As a result, there are no longer any language-independent philosophical notions.
Carnap has parted company with classical positivism.
The obvious question here is: why prefer formal mode translations? Carnap states that
the “material mode of speech is not in itself erroneous; it only readily lends itself to wrong
use” (1937b: 312). Indeed, “if suitable definitions and rules for the material mode of
speech are laid down and systematically applied, no obscurities or contradictions arise”
(ibid.; my emphasis). Misuse of the material mode arises because pseudo-object terms
such as ‘number’ “appear to designate kinds of objects, and thus make it natural to ask

863
 SA N F O RD S H I E H

questions concerning the nature of objects of these kinds” (ibid.: 310), unfortunately
these questions are posed in natural languages, which are “too irregular and too compli-
cated to be actually comprehended in a system of rules” (ibid.: 312). Carnap views natu-
ral languages behavioristically, consisting of “dispositions to certain activities, serving
mainly for the purposes of communication and of co-ordination of activities among the
members of a group” (1939: §2, 3), and linguistic behavior does not “determine whether
the use of a certain expression is right or wrong but only how often it occurs and how
often it leads to the effect intended. . . . A question of right or wrong must always refer
to a system of rules” (1939: §4, 7). (See also Carnap 1937b: §2, 5; 1942: §1, 3; §5, 14;
1990: 432, final paragraph).
This diagnosis points to Carnap’s attitude to the contrast between “controversies in
traditional metaphysics” and “discussions in empirical science”: metaphysical disputes
“appeared . . . useless . . .; there seemed hardly any chance of mutual understanding, let
alone of agreement, because there was not even a common criterion for deciding the
controversy” (1963a: 44–45). What underlies this is the view that if indeed there is no
common criterion for deciding a controversy, then it would be pointless, irrational, to
continue the dispute in the form of trying to find out who is right. A criterion of right and
wrong is supplied by a set of rules. Specifically, the abstract rules of pure syntax are
projected onto linguistic behavior by “correlative definitions” (Carnap 1937b: 78), and
this enables one to describe that behavior in the vocabulary of rational justification (see
Ebbs 1997: §§57–9; Ricketts 2003: 262–4). Thus Carnap’s aim, in the construction of
artificial language systems, is to set out criteria that would rationalize pointless philo-
sophical debates. This is the reason for preferring the formal mode.
Rationality would be achieved if the disputants adopted a common set of rules for
adjudicating their disagreement. But this is not the only way. If rational assessment of
claims is possible only on the basis of a system of rules constituting a language, it is not
clear how it would be possible to assess rationally the supposed claim that that system of
rules as a whole is correct or incorrect, true or false of reality. So the adoption of a system
of rules is not constrained by rule-governed assessment. This is Carnap’s Principle of
Tolerance: “In logic, there are no morals. Everyone is at liberty to build up his own logic,
i.e. his own form of language, as he wishes” (1937b: 51–2). But one could still, ration-
ally, recommend one form of language over others, on the basis of pragmatic criteria:
simplicity, convenience, safety from contradiction. So, the parties to a philosophical
conflict could reconceive their opposition, not as a disagreement over the truth of a
doctrine but as different recommendations about what language, what system of rules,
to adopt. So reconceived, the dispute acquires a point.
None of this means that Carnap is no longer committed to empiricism. His com-
mitment takes the form recommending certain types of languages: “it is preferable to
formulate the principle of empiricism not in the form of an assertion . . . but rather in
the form of a proposal or requirement [for] the language of science” (Carnap 1937a: 33).
Language systems for science enable us to conceive of and to pursue science as a rule-
governed rational activity. In particular, Carnap hopes to be able to display the stand-
ards governing the use of theoretical language in science as sensitive to observational
evidence. But this particular verificationist form of rule-governedness is no longer essen-
tial to his opposition to metaphysics. Carnap has parted company with verificationism
as a criterion of meaningfulness. What qualifies a sentence as a metaphysical pseudo-
statement is simply lack of common standards of assessment. Thus Carnap writes, about
the metaphysical dispute between nominalism and realism, “I cannot think of any

864
 LO G I CA L PO SI TI VI S M A N D Q U I N E

possible evidence that would . . . decide [it]. Therefore I feel compelled to regard [it] as
a pseudo-question, until both parties to the controversy offer a common . . . indication
of possible evidence regarded as relevant” (1956b: 219).
A final question that naturally arise is: What is Carnap’s basis for his view of rational
justification? From here, philosophical debates over the true nature of rationality loom.
I suggest that Carnap’s conception of rationality is also not a theoretical claim but a
practical proposal, to conceive of the intractability of philosophical debate, in contrast
to scientific progress, as rooted in the lack of rules for adjudicating disputes, and so to
hope for philosophical progress through the replacement of traditional philosophizing
by logical syntax. Carnapian pragmatism goes all the way down.

Semantics
The fundamental features of the philosophical perspective just presented remains in
place when Carnap moved to semantical language-systems.
For our purposes, the main changes are the following. The primitive rules consti-
tuting linguistic frameworks are rules of designation and satisfaction, on the basis of
which Carnap defines a variety of basically Tarskian truth conditions for sentences.
Logico-mathematical truth is now truth in virtue of semantic rules of language alone,
explicated by the predicate ‘L-true’. The L-truth of sentences of logic is defined in terms
of state-descriptions, maximal consistent sets of sentences (Carnap 1956a: §2). The
analyticity predicate is defined in terms of L-truth and a class of “meaning postulates”
(Carnap 1952). The crucial point is that the primitive semantic notion of reference, i.e.
designation, is language-internal, since the designata of object language expressions are
simply stated by using descriptive expressions of the meta-language translating those
object expressions. So reference no more involves extra-linguistic metaphysical or epis-
temological notions than do the earlier syntactic concepts. This point is clear from
Carnap’s well-known distinction between internal and external questions (1956b). The
former are to be answered by references to the rules of a linguistic framework; the latter
concern the correctness of frameworks as wholes. The same line of thinking underlying
Tolerance tells us that there are no rational bases for answering external questions. In
particular, Carnap fails to see that ontological questions about whether a type of entities
exist have any clearly defined sense when taken in a philosophical way, as independent
of linguistic frameworks.

Quine Against Carnap

Classical positivism’s distinctive doctrine is that logico-mathematical knowledge is non-
empirical but legitimate because based on meaning or linguistic convention. Against
this Quine presents three objections widely found compelling.
First, Quine argues against the thesis that the sentences of logic are true in virtue of
being conventionally adopted:

Each of these conventions is general, announcing the truth of every one of an

infinity of statements conforming to a certain description; derivation of the
truth of any specific statement from the general convention thus requires a
logical inference, and this involves us in an infinite regress.
(1976: 103)

865
 SA N F O RD S H I E H

It follows that one cannot know that the specific sentence is true by convention with-
out knowing that the inference by which one concludes that it is an instance of the
general convention is correct. This in turn requires knowledge of another logical truth,
grounded on another particular instance of a general convention, and this launches the
regress. So the source of knowledge of logic cannot be convention.
Second, in (1953) Quine rejects the notion of analyticity, understood as truth in
virtue of meaning. One argument is that explanations of analyticity—in terms of cogni-
tive synonymy, definition, and interchangeability salva veritate within a necessity opera-
tor—are circular because these other concepts are no clearer than analyticity, or are
explained in terms of analyticity. What appears to be a slight variant rejects Carnap’s
explications of analyticity in terms of semantic rules on the ground that the notion of
semantic rule remains unexplained. Against semantic rules Quine also argues that any
class of truths can be labeled ‘semantic rules’ given an “enterprise of schooling uncon-
versant persons in” truth condition for sentences of some language (1953: 35). But since
any truth whatsoever could be used for this purpose, any truth would qualify as a seman-
tic rule. Quine also writes that a genuine clarification has to spell out the “behavioral
. . . factors relevant to analyticity” (1953: 36).
Finally, also in (1953), Quine argues against reductionism, on the basis of holism in
confirmation. One can take this argument as also directed against analyticity. AS can
be interpreted as an epistemic distinction, marking a difference in the bases for accept-
ance and rejection of sentences. The acceptance or rejection of a synthetic sentence in
a language L is subject to evaluation as correct or incorrect on the basis of the rules of L.
In contrast, the acceptance or rejection of an analytic sentence amounts to acceptance
or rejections of the rules of L, i.e., of L itself, and so cannot be evaluated as correct or
incorrect but is based on pragmatic grounds: expedience, simplicity, fruitfulness. This
distinction is familiar from the argument for Tolerance, and so is plausibly ascribed to
Carnap.
The holism of confirmation goes against this epistemic asymmetry. If the observa-
tional consequences of a theory conflict with actual observations, no single sentence
of the theory need be singled out for rejection or disconfirmation; moreover, nothing
stands in the way of rejecting the (logico-mathematical) rules according to which the
consequences are drawn from the theory, or indeed the observations themselves. The
choice of which sentence to revise, or whether to make any revision at all, is control-
led by the same pragmatic factors that influence choice of language. Quine advances a
“counter-suggestion” to AS, his web of belief model of language, also called “gradual-
ism” (1986: 100). According to this model, each sentence of one’s overall theory, or
even all of science, is more or less firmly accepted, depending on the extent to which
it is inferentially linked to other sentences; in response to observational evidence, the
choice of which sentence to revise is shaped by how much disturbance in attitudes to
other sentences a revision would cause.
Against Carnap none of these arguments is conclusive. From Carnap’s deflation-
ary perspective, the classical positivist purported thesis that conventional adoption of
rules explains what makes logico-mathematical sentences true is a pseudo-statement.
So Quine’s regress doesn’t refute any position Carnap holds. Note Carnap doesn’t deny
that logic is conventional. Indeed, he suggests that “principle of the conventionality
of language forms” is a better name for Tolerance (1963a: 55). But this shows that for
him logic is conventional only in the sense that adopting a system of logic requires no
theoretical justification.

866
 LO G I CA L PO SI TI VI S M A N D Q U I N E

Carnap takes the circularity argument to show that natural language expressions have
no systematic and precise meanings. But he fully accepts this point; one aim of linguistic
frameworks is to remove such unclarities.
Quine’s dismissal of semantic rules is surely one reason why Carnap found (Quine
1953) “rather puzzling” (1963b: 918). Why would one demand, as Quine does, an expla-
nation of what makes a sentence a semantic rule? Is it that true in virtue of semantic rules
is supposed to be a special property of some sentences that explains how we can have a
type of nonempirical knowledge of what they express? So that, if we don’t know what
makes something a semantic rule, then we have no basis for evaluating whether truth
in virtue of such rules explains the putative possession of this special epistemic prop-
erty? But then Quine must seem to Carnap to have missed the fact that the purpose of
frameworks, and so of semantic rules, is explication and replacement of informal philo-
sophical pseudo-conceptions, especially that of logico-mathematical truth. Neither L-
truth nor the semantic rules of a framework is meant to do any philosophical explana-
tory work, to capture extra-framework properties. For Carnap the choice of semantic
rules is constrained by pragmatic considerations alone: for instance, effectiveness for
the “enterprise of schooling unconversant persons in some language.” Thus, just as in
the case of the regress argument, Carnap has already rejected the position that Quine’s
argument seems to oppose.
Carnap tries a rational reconstruction of Quine’s objection. He takes Quine to doubt
that there’s any clear notion to be explicated because “pre-systematic explanations of
analyticity are too vague and ambiguous, . . . basically incomprehensible” (1963b: 919),
and because there is no empirical evidence, in the form of linguistic behavior, for analytic-
ity. In response, Carnap attempts to specify behavioral evidence relevant to adjudicating
competing hypotheses about the meanings speakers associate with expressions (1956c:
237). This opened another line of debate with Quine, leading to Quine’s notorious argu-
ments for the indeterminacy of translation (1960). The upshot of those arguments is that
behavioral evidence fails to fix even the extensions of many predicates, much less their
meanings. Quine’s indeterminacy arguments remain controversial. In particular, since
physical theory is also not fixed by observation, it is unclear why the failure of behavioral
facts to fix linguistic hypotheses should warrant the conclusion that these hypotheses are
are not open to confirmation by behavioral data (see Chomsky 1969). But, even if Car-
nap has to concede that there is no behaviorally distinguished explicatum for analyticity,
it’s not clear why this invalidates his project of providing rationalizing reconceptions of
philosophical disputes and scientific practice. As we have seen, Carnap doubts that the
behavioral dispositions constituting the phenomena of natural languages are actually rule-
governed. Carnapian language systems are not intended to state the rules speakers actually
use but are tools for achieving rationality in the practice of philosophy and science.
As for Quine’s anti-reductionism arguments, the holism of confirmation is no news
to Carnap. Already in Syntax Carnap claims that an empirical “test applies, at bottom,
not to a single hypothesis but to the whole system of physics . . . (Duhem, Poincaré)”
(318). Moreover,

all rules [of the physical language] may be altered as soon as it seems expedient to do
so. This applies not only to the P[hysical]-rules but also to the L[ogical]-rules,
including those of mathematics. In this respect, there are only differences in
degree; certain rules are more difficult to renounce than others.
(Ibid., emphases mine)

867
 SA N F O RD S H I E H

Finally, Carnap accepts “that ‘any statement can be held true come what may’” (Car-
nap 1963b: 921). Carnap goes on to explain why he can take holism in stride. He
distinguishes

between two kinds of readjustment in the case of a conflict with experience, . . .

a change in the language, and a mere change in . . . truth-value [of] a statement
whose truth value i[s] not fixed by the rules of language. . . . A change of the
first kind constitutes a radical alteration, sometimes a revolution, and it occurs
only at certain historically decisive points in the development of science. On
the other hand, changes of the second kind occur every minute.
(1963b: 921)

In order to understand Carnap’s reply, let’s first turn to one of Quine’s criticisms of
what he calls the “linguistic doctrine of logical truth” (hereafter LD): a sentence of
logic is true in virtue of the meanings of the logical constants (1963: 386). Quine con-
siders two arguments in favor of LD, restricted to “elementary logic,” i.e., first-order
logic with identity. First, we are inclined to say, about proponents of systems of logic
apparently consistent but contrary to our logic, that they are “merely using the familiar
particles ‘and’, ‘all’, or whatever, in other than the familiar senses” (ibid.) Second, we
are inclined to wonder, about so-called prelogical people who supposedly “accept cer-
tain simple self-contradictions as true,” whether their affirmations haven’t just been
mistranslated (ibid.) In each case our reaction is supposed to be explained by LD. Quine
argues that the hypothesis that the sentences of logic are “obviously true,” in the sense
of being very firmly accepted, explains our reactions equally well. Let’s prescind from
evaluating Quine’s argument (see Creath 2003 for a negative assessment). I want to
focus on two other points. First, Quine takes the upshot of this argument to be that LD
is “empty” and has no “experimental meaning” (ibid.) Second, Carnap not only agrees
with this upshot, but also says, “I am surprised that Quine deems it necessary to support
this view by detailed arguments,” because LD does not consist of “factual” statements
but rather “proposals for certain explications” (1963b: 917).
This last point shows us how to understand Carnap’s response to holism. AS, like LD,
is a set of proposals for explication, that is, for rationalizing redescriptions of scientific
linguistic practices in terms of certain frameworks. For this to be possible, Carnap needs
no more than for the frameworks to be consistent with the holism of the (actual) prac-
tice of science. That’s the point made in reply to Quine: holism of empirical testing is
consistent with describing some revisions as changes in language and others as changes
in theory. So the data, as it were, do not rule out AS as an explication of our confirma-
tional practices.
All this is from Carnap’s perspective. How does the situation look to Quine? Let’s
begin from the two planks of Quine’s naturalism: “[t]here is no legitimate first philoso-
phy, higher or firmer than physics” (1969a: 303), and, philosophy is “natural science
trained upon itself and permitted free use of scientific findings” (1966: 667). We might
then think that for Quine AS would be a hypothesis about our use of sentences in our
practices of confirmation, on a par with any other scientific hypothesis, and gradualism
a competing hypothesis. Quine should then judge the two according to the criteria—fit
with observation, simplicity, clarity, expedience—that he takes to be operative in sci-
ence. Does Quine then hold that gradualism comes out ahead of AS or LD? If so, he
doesn’t tell us why. Moreover, as we have just seen, according to Quine LD has “no

868
 LO G I CA L PO SI TI VI S M A N D Q U I N E

experimental meaning.” So is Quine claiming that gradualism, in contrast, has experi-

mental meaning? As we saw, the “data” seem consistent with but do not entail AS. Do
the “data” then entail gradualism? If so, Quine gives no arguments in support. But, sup-
posing they do, then, since they are consistent with AS, so is gradualism. So, once again,
why prefer gradualism to AS?
In fact Quine doesn’t take AS or LD to be on a par with gradualism. Consider his
parable of Ixmann the logical positivist, “out to defend scientists against the demands
of a metaphysician”:

The metaphysician argues that science presupposes metaphysical principles,

or raises metaphysical problems, and that the scientists should therefore show
due concern. Ixmann’s answer consists in showing in detail how people . . .
might speak a language quite adequate to all of our science but . . . incapable
of expressing the alleged metaphysical issues. (I applaud this answer, and think
it embodies the most telling component of Carnap’s own anti-metaphysical
representations. . . .) Now how does . . . Ixmann specify that . . . language? By
telling us . . . what these [people] are to be imagined as uttering and what they
are thereby to be understood to mean. Here is Carnap’s familiar duality of for-
mation rules and transformation rules (or meaning postulates). . . . But these
rules are part only of Ixmann’s narrative machinery, not part of what he is por-
traying. He is not representing his hypothetical [people] as somehow explicit
on formation and transformation rules.
(Quine 1963: 401)

Here it’s clear that Quine does see that Carnap’s linguistic frameworks are not intended
as explanations of verbal behavior or scientific practice but rather as serving a practical
anti-metaphysical aim. However, Quine’s view of that aim isn’t quite the same as Car-
nap’s. For Carnap, as we have seen, metaphysical issues arise from a lack of rationality
in colloquial language, and so will hopefully vanish through the imposition of rational-
izing perspectives embodied in frameworks. The standard here is rationality; scientific
practice is fully rational only as seen through a framework. For Quine, in contrast,
showing that metaphysical issues need not arise in a language “adequate for all of sci-
ence” is enough for dispensing with them. The standard here is the needs of science. For
Quine the only use of a linguistic framework would be an “overall pattern of canonical
notation” . . . “shared by all the sciences,” since the “same motives that impel scientists
to seek ever simpler and clearer theories . . . are motives for simplification and clarifica-
tion of the broader framework” (1960: 161). Quine argues that a notation consisting
of the syntax and semantics of elementary logic is sufficient for all the sciences (see
1957; 1960: chapters 3–6 provide much more detail). Thus the problem with Carnapian
frameworks is that they contain more than is needed for science; all the explications of
AS in these frameworks are dispensable, as far as the needs of science are concerned.
Hence AS and LD are on a par with metaphysical issues; Carnap’s ground (as Quine
sees it) for doing without the latter “pseudo-doctrines” (Quine 1963: 390) apply no less
to the former.
This is to judge Carnapian frameworks as canonical notations for science. If they are
not such notations, what are they for? Perhaps here Quine is as puzzled as Carnap was by
Quine’s criticisms. Surely Carnap is not confused, as someone might be about Ixmann’s
fiction, and takes his frameworks to be hypotheses about underlying mechanisms of

869
 SA N F O RD S H I E H

actual scientists’ practices? For Quine philosophizing about science is seeking a scien-
tific account of science. From this point of view science is the production, as a result
of “surface irritation and internal conditions,” of language as physical phenomena, for
example, “concentric air-waves,” taking “the form of a torrent of discourse about” the
world (Quine 1957: 1). Since the aim is a scientific account of these physical processes,
Quine doesn’t see the relevance of regarding science through Carnapian frameworks:
He asks, “why all this creative reconstruction, all this make-believe?. . . Why not settle
for psychology?” (1969b: 75).
We can now understand Quine’s preference for gradualism. Gradualism pictures the
doing of science in (the direction of) naturalized epistemological terms: not justification
but firmness of acceptance, i.e., of assent and dissent behavior. This is of a piece with
the view of science as a physical phenomenon. Neither is a scientific hypothesis, but
Quine is confident of both “because we know in a general way what the world is like”;
they rest on “our very understanding of the physical world, fragmentary though that
understanding be” (1957: 1–2).
This proto-scientific view of science and of language is of course controversial
(Chomsky 1969 outlines a classic critique; Ricketts 1982 gives a Quinean reply.) But
from the perspective of our discussion what is most interesting is that if Carnap’s point
of view rests on a proposal for conceiving of philosophical activity and its relation to
science, surely Carnap can accept Quine’s view as a counter-proposal, to be evaluated
on pragmatic grounds. So let me conclude this discussion with a question: does Car-
nap’s pragmatic perspective have to count as a first philosophy, higher or firmer than
science?

Final Remarks
The Carnap–Quine debate is a clash of two anti-metaphysical strategies, both opposed
to classical positivism but based on opposed conceptions of language, science, and of
philosophy itself. One of the ironies of the history and historiography of analytic phi-
losophy is that Carnap is taken to be a representative of the positivism that he rejected
along with Quine, and Quine’s criticisms of positivism are taken to justify the reinstate-
ment of the metaphysics that he rejected along with Carnap. With a clearer view of
what is at stake we can perhaps reopen an investigation of the relation between Carna-
pian pragmatism and Quinean naturalism.

Related Topics
2.5 Analytic Truth
4.3 The Role of Psychology
7.4 Frege, Russell, and Wittgenstein

References
Awodey, Steve and Carus, André. (2007) “The Turning Point and the Revolution,” in A. Richardson and
T. Uebel eds.) The Cambridge Companion to Logical Empiricism, Cambridge: Cambridge University Press,
pp. 165–92.
Ayer, A. J. (1936) Language, Truth, and Logic, London: V. Gollancz Ltd.
Bradley, F. H. (1897) Appearance and Reality: A Metaphysical Essay, 2nd edn., London: Macmillan.
Carnap, Rudolf. (1935) Philosophy and Logical Syntax, London: K. Paul, Trench, Trubner & Co., Ltd.

870
 LO G I CA L PO SI TI VI S M A N D Q U I N E

——. (1937a) “Testability and Meaning (Continuation),” Philosophy of Science 4: 1–40.

——. (1937b) The Logical Syntax of Language, translated by Amethe Smeaton (Countess von Zeppelin),
London: K. Paul, Trench, Trubner & Co., Ltd.
——. (1939) Foundations of Logic and Mathematics, Chicago: University of Chicago Press.
——. (1942) Introduction to Semantics, Cambridge, MA: Harvard University Press.
——. (1952) “Meaning Postulates,” Philosophical Studies 3: 65–73.
——. (1956a) Meaning and Necessity: A Study in Semantics and Modal Logic, 2nd edn., Chicago: University
of Chicago Press.
——. (1956b) “Empiricism, Semantics, and Ontology,” in Meaning and Necessity, pp. 205–21.
——. (1956c) “Meaning and Synonymy in Natural Languages,” in Meaning and Necessity, pp. 233–47.
——. (1963a) “Intellectual Autobiography,” in Paul Arthur Schilpp (ed.) The Philosophy of Rudolf Carnap,
La Salle, IL,: Open Court, pp. 1–84.
——. (1963b) “W. V. Quine on Logical Truth,” in Paul Arthur Schilpp (ed.) The Philosophy of Rudolf Car-
nap, La Salle, IL: Open Court, pp. 915–22.
——. (1990) “Quine on Analyticity,” in Richard Creath (ed.) Dear Carnap, Dear Van: the Quine-Carnap
Correspondence and Related Works, Berkeley: University of California Press, pp. 427–32.
Chomsky, Noam. (1969) “Quine’s Empirical Assumptions,” in Donald Davidson and Jaakko Hintikka (eds.)
Words and Objections: Essays on the Work of W. V. Quine, Dordrecht: D. Reidel, pp. 53–68.
Creath, Richard. (2003) “The Linguistic Doctrine and Conventionality,” in Gary Hardcastle and Alan W.
Richardson (eds.) Logical Empiricism in North America, Minneapolis: University of Minnesota Press, pp.
234–56.
Ebbs, Gary. (1997) Rule-Following and Realism, Cambridge, MA: Harvard University Press.
Hahn, Hans. (1980) “Empiricism, Mathematics and Logic,” in B. McGuiness (ed.) Empiricism, Logic, and
Mathematics: Philosophical Papers of Hans Hahn, Dordrecht: Reidel, pp. 39–42.
Hempel, Carl Gustav. (1965) Aspects of Scientific Explanation and other Essays in the Philosophy of Science, New
York: Free Press.
Hilbert, David. (1925) “Über das Unendliche,” Mathematische Annalen 95: 161–90.
Quine, W. V. (1953) “Two Dogmas of Empiricism,” in From a Logical Point of View: Nine Logico-Philosophical
Essays, Cambridge: Harvard University Press, pp. 20–46.
——. (1957) “The Scope and Language of Science,” British Journal for the Philosophy of Science 8: 1–17.
——. (1960) Word and Object, Cambridge, MA: MIT Press.
——. (1963) “Carnap and Logical Truth,” in Paul Arthur Schilpp (ed.) The Philosophy of Rudolf Carnap, La
Salle, IL: Open Court, pp. 385–406.
——. (1966) “Russell’s Ontological Development,” Journal of Philosophy 63: 657–67.
——. (1969a) “Reply to Chomsky,” in Donald Davidson and Jaakko Hintikka (eds.) Words and Objections:
Essays on the Work of W. V. Quine, Dordrecht: D. Reidel, pp. 302–11.
——. (1969b) “Epistemology Naturalized,” in Ontological Relativity and other Essays, New York: Columbia
University Press, pp. 69–90.
——. (1976) “Truth by Convention,” in The Ways of Paradox and other Essays, revised and enlarged edn.,
Cambridge, MA: Harvard University Press, pp. 77–106.
——. (1986) Philosophy of Logic, 2nd edn., Cambridge, MA: Harvard University Press.
Ricketts, Thomas. (1982) “Rationality, Translation, and Epistemology Naturalized,” Journal of Philosophy
79: 117–36.
——. (2003) “Languages as Calculi,” in Gary Hardcastle and Alan W. Richardson (eds.) Logical Empiricism
in North America, Minneapolis: University of Minnesota Press, pp. 257–80.
Wittgenstein, Ludwig. (1922) Tractatus Logico-Philosophicus, translated by C. K. Ogden, London: K. Paul,
Trench, Trubner & Co., Ltd.

Further reading
For bibliographies of logical positivism, see O. Neurath, R. Carnap, C. Morris (eds.) Foundations of the Unity
of Science: Toward an International Encyclopedia of Unified Science, 2 vols. (Chicago: University of Chicago
Press, 1970); A. J. Ayer (ed.) Logical Positivism (New York: Free Press, 1959); R. M. Rorty (ed.) The Lin-
guistic Turn, 3rd edn. (Chicago: University of Chicago Press, 1992); and Friedrich Stadler, The Vienna
Circle: Studies in the Origins, Development, and Influence of Logical Empiricism (Vienna and New York:
Springer, 2001). Sahotra Sarkar (ed.), Science and Philosophy in the Twentieth Century: Basic Works of Logi-
cal Empiricism, 5 vols. (New York: Garland, 1996).

871
 SA N F O RD S H I E H

The newer scholarship on positivism in general and Carnap in particular began with Alberto Coffa, The
Semantic Tradition from Kant to Carnap: To the Vienna Station, ed. L. Wessels (Cambridge: Cambridge
University Press, 1991), Michael Friedman, Reconsidering Logical Positivism (Cambridge: Cambridge
University Press, 1999), and papers on Carnap by Warren Goldfarb and Thomas Ricketts: Goldfarb and
Ricketts, “Carnap and the Philosophy of Mathematics,” in D. Bell and W. Vossenkuhl (eds.) Science
and Subjectivity (Berlin: Akademieverlag, 1992) pp. 61–78, Ricketts, “Carnap’s Principle of Tolerance,
Empiricism, and Conventionalism,” in Peter Clark, (ed.), Reading Putnam (Oxford: Blackwell, 1995), and
their papers in the collections below.
Some of the best recent historical work on Carnap are by Steve Awodey and Andre Carus: Awodey and
Carus, “Carnap, Completeness, and Categoricity: The Gabelbarkeitssatz of 1928,” Erkenntnis 54 (2001):
145–72; Awodey and Carus, “Carnap’s Dream: Gödel, Wittgenstein, and Logical Syntax,” Synthese 159
(2007): 23–45; and Carus, Carnap and Twentieth-Century Thought (Cambridge: Cambridge University
Press, 2007).
For more current scholarship see the following collections of papers: Ron Giere and Alan Richardson (eds.)
The Origins of Logical Empiricism (Minneapolis: University of Minnesota Press, 1996); Paolo Parrini, Wes-
ley Salmon, and Merrilee Salmon (eds.) Logical Empiricism: Historical and Contemporary Perspectives (Pitts-
burgh: University of Pittsburgh Press, 2003); Gary Hardcastle and Alan Richardson (eds.) Logical Empiri-
cism in North America (Minneapolis: University of Minnesota Press, 2003); Steve Awodey and Carsten
Klein (eds.) Carnap Brought Home: The View from Jena (Chicago: Open Court, 2004); Richard Creath
and Michael Friedman (eds.) The Cambridge Companion to Carnap (Cambridge: Cambridge University
Press, 2007); Alan Richardson and Thomas E. Uebel (eds.) The Cambridge Companion to Logical Empiri-
cism (Cambridge: Cambridge University Press, 2007); and Pierre Wagner (ed.) Carnap’s Logical Syntax of
Language (New York: Palgrave Macmillan, 2009).
For careful exposition and illuminating discussion of Quine’s anti-analyticity arguments see Gillian Rus-
sell, Truth in Virtue of Meaning (Oxford: Oxford University Press, 2008) and Cory Juhl and Eric Loomis,
Analyticity (London: Routledge, 2009). These authors also provide sophisticated contemporary defenses
of analyticity.
The newer assessment of the Quine–Carnap debate began with Howard Stein, “Was Carnap Entirely Wrong,
After All?,” Synthese 93 (1992): 275–95, and Daniel Isaacson, “Carnap and the Philosophy of Math-
ematics,” in D. Bell and W. Vossenkuhl (eds.) Science and Subjectivity (Berlin: Akademieverlag, 1992),
pp. 100–30. Apart from the works of Creath, Ebbs, and Ricketts cited in the References, one should also
consult Alexander George, “On Washing the Fur without Wetting It: Quine, Carnap, and Analyticity,”
Mind 109 (2000): 1–24; Burton Dreben, “Quine,” in Robert B. Barrett and Roger Gibson Jr. (eds.) Perspec-
tives on Quine (Oxford: Blackwell, 1990); “In Mediis Rebus,” Inquiry 37 (19944): 441–47, and “Quine on
Quine,” in Roger Gibson Jr (ed.) The Cambridge Companion to Quine (Cambridge: Cambridge University
Press, 2004), pp. 287–93; and Peter Hylton, Quine (New York and London: Routledge, 2007.

872
 7.6
O RD INARY LANGUAGE
P HI L OSOPHY
Michael Beaney

our common stock of words embodies all the distinctions men have found worth
drawing, and the connexions they have found worth marking, in the lifetimes
of many generations: these surely are likely to be more numerous, more sound,
since they have stood up to the long test of the survival of the fittest, and more
subtle, at least in all ordinary and reasonably practical matters, than any that
you or I are likely to think up in our arm-chairs of an afternoon—the most
favoured alternative method.
(Austin, “A Plea for Excuses”, p. 182)

‘Ordinary language philosophy’ is a somewhat misleading term for a loosely connected

set of subtraditions of analytic philosophy that had their heyday in Oxford and Cam-
bridge in the 1950s and 1960s, although their roots lie in ideas developed in the first
half of the twentieth century and their influence extended well outside Britain and
is still felt today. In Cambridge, the key figures were G. E. Moore (1873–1958) and
Ludwig Wittgenstein (1889–1951). In Oxford, where the center of gravity of British
analytic philosophy moved after the Second World War, the key figures were Gilbert
Ryle (1900–76), J. L. Austin (1911–60) and P. F. Strawson (1919–2006).

1 Cambridge Ordinary Language Philosophy

Central to the development of analytic philosophy in the twentieth century was what
came to be known as the ‘linguistic turn.’ The origins of this are disputed. Some have
dated it to the use by Gottlob Frege of contextual definition in The Foundations of Arith-
metic of 1884 (see for example Dummett 1991, pp. 111–12), or at least to the more self-
conscious use of contextual definition by Bertrand Russell in his theory of descriptions,
first put forward in 1905. Some have dated it even earlier, to Jeremy Bentham’s use
of paraphrasis in ‘analysing away’ talk of obligations (see for example Hacker 1996, p.
281). But there is little doubt that it received a canonical formulation in Wittgenstein’s
Tractatus Logico-Philosophicus, which was published in 1921. Here Wittgenstein wrote
that “Most of the propositions and questions of philosophers arise from our failure to
understand the logic of our language” (4.003). All philosophy, he went on, is ‘critique of
language’, and he singled out Russell’s theory of descriptions as showing how the gram-
matical form of a sentence can mislead us as to its real logical form (cf. 4.0031).
 M I CH A EL BE A N E Y

The ideas of the Tractatus influenced the development of both logical positivism and
ordinary language philosophy. Following Russell’s lead (in his introduction to the Trac-
tatus), the logical positivists took Wittgenstein to be advocating an ideal logical lan-
guage in which philosophical problems would not arise. In fact, however, Wittgenstein
held that the propositions of our ordinary language are in perfect logical order as they
are (cf. 5.5563). What was necessary was an ideal notation in which the actual logic
of our language could be exhibited clearly (cf. 3.325). This belief in the fundamental
adequacy of ordinary language, even if analysis is needed to reveal this, is the governing
principle of ordinary language philosophy.

1.1 Wittgenstein’s Later Philosophy

After the Tractatus was published, Wittgenstein gave up philosophy for a few years.
On his return, partly prompted by discussions with members of the Vienna Circle, he
started to criticize some of his earlier ideas. In particular, he rejected the assumption
that there is a single logic—the propositional and predicate logic of Frege and Russell—
that underlies our use of language, and the associated view that there is a single account
of the way that language has meaning—that given by his so-called picture theory of
language. However, Wittgenstein never repudiated the idea that our ordinary language
is fundamentally adequate; what he came to stress was the different ways in which words
are used and the multiplicity of our forms of language.
In his Philosophical Investigations, posthumously published in 1953, Wittgenstein com-
pares words with the tools in a toolbox: “there is a hammer, pliers, a saw, a screw-driver,
a rule, a glue-pot, glue, nails and screws.—The functions of words are as diverse as
the functions of these objects”. (§11) He introduces the idea of a ‘language-game’: an
activity or practice in which language and actions are interwoven (cf. §7); and he gives
examples to illustrate the multiplicity of language-games: from giving orders and report-
ing events to telling jokes, cursing and praying (§23).
Each of these language-games may have their own ‘logic’, or as he now prefers to
call it, their own ‘grammar’. There is not a single logic or grammar which fits all uses of
words, no single essence to language (cf. §§ 65, 97ff.). Philosophical confusion typically
arises from failing to appreciate the grammar of the relevant forms of expression. On
Wittgenstein’s later view, the task of philosophy is to describe the different uses of our
words, to provide an ‘Übersicht’ (perspicuous representation) of their grammar, in order
to dissolve philosophical problems (cf. §§ 122ff.). Philosophy, he wrote, “is a battle
against the bewitchment of our intelligence by means of language” (§109).

1.2 Moore
Together with Russell, Moore had rebelled against British idealism around the turn of the
twentieth century, but this rebellion did not inaugurate a linguistic turn. Moore agreed
with Russell in seeing the task of philosophy as the analysis of propositions. But in his
early work, he regarded propositions simply as complexes of concepts, where concepts
were understood as constituents of the world (1899, pp. 5, 8). Later, influenced by both
Russell’s theory of descriptions and Wittgenstein’s work, he came to see that language
can easily mislead us, and his own writing came to be characterized more and more by the
meticulous—even pedantic—clarification of the meanings of ordinary propositions and
the painstaking critique of philosophical claims that conflicted with those meanings.

874
 O RDI N A RY LA N G U A G E P H I L O S O P H Y

Moore is often regarded as an ordinary language philosopher, but it might be more

accurate to call him a ‘common sense philosopher’, as epitomized in his 1925 paper,
“A Defence of Common Sense”. Moore here lists a number of propositions that he
claims are ‘truisms’, such as that he exists and has a body, as well as a further truism to
the effect that other people know corresponding things. He also claims that he knows,
with certainty, that these are all true. He recognizes, however, that many philosophers
have disputed this, and he spends the rest of the paper attempting to show how they
are wrong. A key part of his defense is the distinction he draws (influenced by Wittgen-
stein’s Tractatus) between understanding the meaning of an expression and being able
to give a correct analysis of its meaning (1925, p. 111). It may be that you and I disagree
about what the correct analysis of the meaning of a truism is (or have no idea at all), but
this does not imply that we do not understand what it means, which may be enough for
us to know that it is indeed a truism.
In what sense, then, is Moore an ordinary language philosopher? An answer is pro-
vided in a paper published in 1942 entitled “Moore and Ordinary Language” by Nor-
man Malcolm (1911–90). Malcolm met Moore and Wittgenstein in Cambridge in
1938–9 and became one of the leading practitioners of ordinary language philosophy in
the United States, writing extensively on Moore’s and Wittgenstein’s work. Malcolm
begins his paper by listing twelve philosophical statements that Moore would reject
as false. One example is the statement that “We do not know for certain the truth of
any statement about material things”. Malcolm reports Moore’s typical reply as follows:
“Both of us know for certain that there are several chairs in this room, and how absurd it
would be to suggest that we do not know it, but only believe it, and that perhaps it is not
the case”! (1942, p. 347; cf. p. 351). Malcolm takes such replies to illustrate his central
contention that “The essence of Moore’s technique of refuting philosophical statements
consists in pointing out that these statements “go against ordinary language” (p. 349).
Saying that “I know for certain that there are several chairs in this room” (when there
are) is a paradigm ordinary use of the phrase ‘know for certain’ (cf. pp. 354–5).
Malcolm notes that Moore might not agree with his account of Moore’s method
(1942, fn. 6); but in his short reply to Malcolm’s paper, Moore does not repudiate that
account, merely quibbling over a minor matter (1942, pp. 668–9). So we can take it that
Moore accepted it as basically accurate, and with it, the clarification of his status as an
ordinary language philosopher.

2 Oxford Ordinary Language Philosophy

Both Wittgenstein and Moore had a significant influence on the development of Oxford
ordinary language philosophy. Ryle knew Wittgenstein well: they went on holiday
together, and Wittgenstein is reported to have claimed that Ryle was one of only two
people who understood his philosophy (cf. Monk 1990, pp. 275, 436). Moore’s work
was read and criticized by Oxford philosophers from the early 1900s onward, and Austin
reputedly found Moore’s style of thinking far more congenial than Wittgenstein’s (cf.
Hampshire 1960, p. 246; Ryle 1970, p. 283). But there was also an indigenous Oxford
tradition that was highly influential, especially on Austin, although this influence was
never explicitly acknowledged. This was the tradition of Oxford realism, whose central
figure was John Cook Wilson (1849–1915). At various places in his posthumously pub-
lished Statement and Inference (1926), Cook Wilson urges us to respect the distinctions
embodied in ordinary language. His most explicit statement is worth quoting at length:

875
 M I CH A EL BE A N E Y

The authority of language is too often forgotten in philosophy, with serious

results. Distinctions made or applied in ordinary language are more likely to
be right than wrong. Developed, as they have been, in what may be called the
natural course of thinking, under the influence of experience and in the appre-
hension of particular truths, whether of everyday life or of science, they are
not due to any preconceived theory. . . . the actual fact is that a philosophical
distinction is prima facie more likely to be wrong than what is called a popular
distinction, because it is based on a philosophic theory which may be wrong in
its ultimate principles.
(1926, II, p. 874; cf. I, pp. 34, 46, 151, 299ff., 387–8; II, p. 875)

Austin does not refer to Cook Wilson at any point in his published writings, yet Austin’s
famous remark, as quoted at the beginning of this chapter, is a clear echo of what Cook
Wilson is saying here. That remark might be taken as the motto of ordinary language
philosophy; Cook Wilson anticipated it over forty years earlier.
The view expressed here informed Cook Wilson’s own methodology, in the painstak-
ing analyses he offered of key distinctions such as that between subject and predicate,
and between denotation and connotation (1926, I, Part II, chs. 5, 18). Among the
Oxford realists—as there was, too, among the surviving Oxford idealists—there was also
a general hostility to the logic of Frege and Russell. This was partly born of ignorance
and reluctance to understand it; but it also had a philosophical motivation rooted in
respect for the rich and subtle distinctions of ordinary language.

2.1 Ryle
Ryle’s first significant publication was “Systematically misleading expressions” (1932).
Ryle here endorses the Tractarian/Moorean distinction between understanding the
meaning of an expression and knowing what its correct analysis is. The aim of philoso-
phy, on Ryle’s view, is to find the correct analysis of problematic expressions—the pur-
suit of which, however, does not imply that users of the expressions cannot understand
them or use them intelligibly (cf. p. 43). The expressions he focuses on are those that
he calls ‘systematically misleading’, which he characterizes as “couched in a syntactical
form improper to the facts recorded and proper to facts of quite another logical form
than the facts recorded” (p. 44).
The first type of such expressions that Ryle considers is what he terms ‘quasi-onto-
logical statements’. Consider, for example, the negative existential statement ‘Satan
does not exist’, which suggests—misleadingly—that the property of non-existence is
being attributed to something. If the statement is true, then there is nothing to be
the subject of such an attribution. Instead, writes Ryle, it must mean something like
‘nothing is both devilish and alone in being devilish’ (p. 46).
Ryle considers three other types of expressions, which he calls ‘quasi-Platonic expres-
sions’, which tempt us into reifying universals, ‘descriptive expressions and quasi-
descriptions’, which are often mistakenly treated as names, and ‘quasi-referential “the”-
phrases’, which may be wrongly taken to imply the existence of some object. In each
case, Ryle argues, the expression must be reformulated to prevent the confusion from
arising and to reveal the real logical form of the relevant fact. Ryle suggests, for example,
that the quasi-Platonic statement ‘Unpunctuality is reprehensible’ should be rephrased
as ‘Whoever is unpunctual deserves that other people should reprove him for being

876
 O RDI N A RY LA N G U A G E P H I L O S O P H Y

unpunctual’, which makes overt reference only to the attribute of being unpunctual and
not to any reified universal (p. 50).
The influence of Russell’s theory of descriptions is obvious here, as is the logical
atomism of Wittgenstein’s Tractatus and Russell’s writings around 1920. In his later
work, however, prompted by the criticisms that were increasingly being made through
the 1930s to the various forms of reductive analysis advocated in the early phase of
British analytic philosophy, Ryle came to reject the assumption that there was a single
logical form underlying every meaningful expression. He retained the view, however,
that language contains misleading expressions that need to be clarified to avoid confu-
sion. The paraphrase suggested above of ‘Unpunctuality is reprehensible’, for example,
could still be offered to show that there is no need to reify universals; it is just that the
paraphrase should not be taken to better reflect any ‘underlying’ logical form.
Ryle’s later method can be described as conceptual clarification without reductive
analysis—or, more simply, as connective analysis, to use Strawson’s term (1992, ch. 2).
In his most famous work, The Concept of Mind, published in 1949, Ryle talks of his aim
being “to rectify the logical geography of the knowledge which we already possess” (p.
9). If we see this as connective analysis, then we can regard Ryle as continuing to insist
on the distinction between understanding the meaning of an expression and knowing
what its correct analysis is: one can have the former without the latter, the task of the
philosopher being to provide the latter.
This idea is reflected in the distinction Ryle drew in chapter 2 between knowing
how and knowing that. There are many things we may know how to do, such as ride a
bicycle and speak a language, without being able to explain what we are doing, in other
words, without knowing that such-and-such rules govern these activities. The task of the
philosopher is to make explicit these rules, especially where confusion is likely to arise
from half-hearted or inadequate attempts to articulate them. Here Ryle’s key idea was
that of a category-mistake, which occurs when things of one logical category are mistak-
enly assigned to a different logical category. Ryle argued that the Cartesian dogma of
the ‘Ghost in the Machine’ was a category-mistake, classifying mental phenomena as
physical events. In The Concept of Mind, Ryle offers a rich—if sometimes controver-
sial—account of a whole range of concepts and distinctions embedded in our ordinary
use of mental expressions, attempting to put into practice his methodology of connec-
tive analysis—charting the logical geography of our mental concepts.

2.2 Austin
Although Ryle was the more dominant figure both in Oxford and British philosophy
generally in the two decades following the end of the Second World War, through the
changes he effected to the study of philosophy at Oxford, his influence on appointments
throughout Britain and his editorship of Mind from 1947 to 1972, Austin is often seen as
epitomizing ‘Oxford philosophy’ as it came to be called from outside. Austin published
little during his lifetime, but after his premature death in 1960, his papers and lectures
were edited by two of his close colleagues, J. O. Urmson (1915–) and G. J. Warnock
(1923–95), who made significant contributions in their own right to the development
of Oxford Philosophy. Austin’s Philosophical Papers appeared in 1961, and the two slim
volumes of his lectures, Sense and Sensibilia and How to Do Things with Words, in 1962.
Austin’s paper “A Plea for Excuses” (1956) provides the ideal introduction to his
philosophy, both in explaining his methodology and in illustrating that methodology

877
 M I CH A EL BE A N E Y

in the case of the language we use in offering excuses for our conduct. Austin here
makes the remark that I suggested provides the motto for ordinary language philosophy.
According to Austin, “our common stock of words embodies all the distinctions men
have found worth drawing” (p. 182), and during the course of the paper he shows how
rich and significant are the distinctions between, for example, ‘inadvertently’, ‘mistak-
enly’, ‘accidentally’, ‘absent-mindedly’, ‘unintentionally’, ‘carelessly’, ‘automatically’,
‘aimlessly’ and ‘purposelessly’, even if at times people confuse them.
Sensitivity to the nuances of ordinary language and the multifarious things we do
with language is characteristic of all of Austin’s work. Austin took an interest in this
both for its own sake, with an eye to developing a new type of linguistic theory, and for
the dissolution it could provide of traditional philosophical problems. These two facets
of his work are reflected, respectively, in the two courses of lectures, both of them given
on several occasions, which were posthumously published as How to Do Things with
Words and Sense and Sensibilia. Sense and Sensibilia might be regarded as his masterpiece:
it offers a sustained critique of the doctrine, as Austin states it (in a rather Moore-like
way), that “we never see or otherwise perceive (or ‘sense’), or anyhow we never directly
perceive or sense, material objects (or material things), but only sense-data (or our
own ideas, impressions, sensa, sense-perceptions, percepts, &c.)” (p. 2). He takes Ayer
(1940) as his main target, but the doctrine, in one form or another, was shared by many
others at the time, both in and outside Oxford.
His methodology is laid down in the first lecture:

our ordinary words are much subtler in their uses, and mark many more distinc-
tions, than philosophers have realized . . . It is essential, here as elsewhere, to
abandon old habits of Gleichschaltung, the deeply ingrained worship of tidy-
looking dichotomies.
(p. 3)

Austin attacks, in particular, the supposed dichotomy between ‘sense-data’ and ‘mate-
rial things’, noting that “There is no one kind of thing that we ‘perceive’ but many
different kinds, the number being reducible if at all by scientific investigation and not
by philosophy” (p. 4). He recognizes that the so-called argument from illusion is one of
the major motivations for the sense-datum doctrine but criticizes the argument as itself
an illusion: there is no simple argument but rather “a mass of seductive (mainly verbal)
fallacies” (p. 5). He considers the range of different cases that are often classified as
‘sense-deception’, including perspectival variation, refraction, mirror-images, mirages,
reflections, and hallucinations; and he explains how the argument trades on a confusion
between illusion and delusion (pp. 20–5). A flavour of his approach can be given by
comparing the following three statements (cf. p. 36):

(1) He looks guilty.

(2) He appears guilty.
(3) He seems guilty.

While ‘looks’, ‘appears’ and ‘seems’ can sometimes be used interchangeably, there are
clearly differences here, which the sense-datum theorist ignores in forcing all ‘looks’,
‘appearances,’ and ‘seemings’ into one category of phenomena that supposedly requires
recognition of ‘sense-data’.

878
 O RDI N A RY LA N G U A G E P H I L O S O P H Y

How to Do Things with Words might also be read as offering a sustained critique of a
prevalent philosophical doctrine: the doctrine that the primary function of a statement
is to describe some state of affairs, or represent some fact, which it does either truly or
falsely (cf. p. 2). Austin calls this the ‘descriptive fallacy’, and he introduces the term
‘constative’ to replace ‘descriptive’ when characterizing utterances that consist in mak-
ing a statement (cf. pp. 3, 6). In fact, however, Austin does not directly attack this
doctrine in the way that he attacked the sense-datum doctrine. His main concern is
to explain what he calls ‘performative’—as opposed to ‘constative’—utterances. This
shows, incontrovertibly, that language has many more uses than just describing states of
affairs, but the clarification of performative utterances themselves takes priority. What
Austin offers is a sketch of a theory of speech acts, which was to be developed further by
later philosophers (see below).
A ‘performative’ utterance is an utterance that performs some action. For example,
to say ‘I promise’ is itself (normally) to promise, and to say ‘I bet’ is itself (normally) to
make a bet. Of course, various conditions have to be met for such utterances to genu-
inely count as performing the relevant action, and Austin discusses these—and the
way in which things can go wrong or be ‘infelicitous’—in some detail. He draws further
distinctions to elaborate his account of performative utterances. He distinguishes the
‘phonetic act’ (the act of uttering certain noises), the ‘phatic act’ (the act of uttering
certain words of a language) and the ‘rhetic act’ (the act of saying something meaning-
ful). All these make up what he calls a ‘locutionary act’ (pp. 94–5), which he in turn
distinguishes from an ‘illocutionary act’ and a ‘perlocutionary act’ (pp. 98ff.). Illocu-
tionary acts are what we do in saying something, such as promising, betting, warn-
ing, informing, announcing a verdict, or criticizing. Perlocutionary acts are what we
do by saying something, such as convincing, deterring, or surprising someone. Austin
also distinguishes various types of performative utterance, depending on their illocu-
tionary force, which he calls ‘verdictives’, ‘exercitives’, ‘commissives’, ‘behabitives,’
and ‘expositives’; and the lectures conclude with a list of verbs under each heading
(pp. 151–64).
Austin’s invention of all this new—and as he admits himself (p. 151), somewhat
rebarbative—terminology might seem surprising for an ordinary language philosopher.
He is modest about whether his final classification tracks distinctions operative in ordi-
nary language, but to the extent that it does, inventing names for them is acceptable.
And even if they are contestable, this still raises our awareness of the complexities and
subtleties of ordinary language. According to Austin, this is essential not only in dis-
solving traditional philosophical problems but also in providing the basis for a more
systematic account of language. His early death prevented him from developing this in
the way that had always been his longer-term aim.

2.3 Strawson
In the development of ordinary language philosophy 1950 was a key year. Ryle’s The
Concept of Mind had just been published, Austin and Strawson debated the nature of
truth in the Proceedings of the Aristotelian Society, and Strawson’s most widely read paper
l, “On Referring”, was published in Mind, announcing his own presence on the philo-
sophical scene. I shall concentrate here on the latter.
“On Referring” is a critique of Russell’s theory of descriptions. According to Russell,
a sentence of the form ‘The F is G’ is to be analyzed as ‘There is one and only one F

879
 M I CH A EL BE A N E Y

and whatever is F is G’, which can be formalized in predicate logic as ‘(x) (Fx & (y)
(Fy o y = x) & Gx)’. To say that the present King of France is bald, for example, is to
say that there is one and only one King of France and he is bald. According to Strawson,
however, this theory fails to do justice to our use of definite descriptions in ordinary
language. It may account for some uses but not all. For example, when I say ‘The table
is covered with books’, I am not asserting that there is one and only one table and
whatever is a table is covered with books. I am simply using the definite description ‘the
table’ to refer to a particular table, the context making clear which table I mean; what I
am saying is just that this table is covered with books. (Cf. 1950a, pp. 14–15) Moreover,
where the definite description fails to refer, as in Russell’s example of ‘The present King
of France is bald’ (uttered today), the sentence is not false, as Russell thought, but truth-
valueless, according to Strawson.
Strawson distinguishes between sentences, which can be characterized as meaningful
(significant) or not, and statements (‘assertions’, or ‘uses of sentences’, as Strawson calls
them), which are the appropriate candidates for the assignment of a truth-value; and
he argues that Russell confuses the two. While ‘meaning’ is a function of the sentence
or expression, Strawson writes, “mentioning and referring and truth or falsity, are func-
tions of the use of the sentence or expression” (1950a, p. 9). While ‘The present King
of France is bald’ is meaningful, according to Strawson, it fails to make, if used today,
a true or false statement, just because there is a failure of reference. In his Introduction
to Logical Theory, published two years later, Strawson distinguishes between entailment
and presupposition, his point then being that in stating that the F is a G, one is typically
presupposing that there is an F. Where there is no F, no statement with a truth-value
is made.
The debate between Russell and Strawson has spawned a huge literature, and is famil-
iar to all students of modern philosophy of language. I will just make two points here.
First, at the heart of Strawson’s critique is the distinction between sentence and state-
ment, where making a statement is understood as a speech act. This brings him firmly
on the side of Austin in stressing that a correct account of language must recognize the
important role played by speech acts. Second, a continual refrain in Strawson’s work is
the gap between the complexity of the rules governing the use of ordinary language and
the rules of formal logic, especially the predicate logic of Frege and Russell. This brings
him firmly on the side of Wittgenstein in stressing the dangers of formalization in phi-
losophy. As Strawson concludes “On Referring”: “Neither Aristotelian nor Russellian
rules give the exact logic of any expression of ordinary language; for ordinary language
has no exact logic”. (1950a, p. 27)

3 Assessment and Influence

I will end by offering a brief assessment of ordinary language philosophy and indicating
something of its influence and the developments that occurred after its heyday in the
1950s and 1960s.

3.1 Assessment
In offering an assessment of ordinary language philosophy, in the space available here, I
will focus on Austin, who is often taken as the paradigm ordinary language philosopher.
The objections that have been made to his philosophy, and the responses that both

880
 O RDI N A RY LA N G U A G E P H I L O S O P H Y

he and various commentators have offered, illustrate the issues that ordinary language
philosophy raises. The most fundamental objection is one that Austin confronts head
on in “A Plea for Excuses”:

When we examine what we should say when, what words we should use in what
situations, we are looking again not merely at words (or ‘meanings’, whatever
they may be) but also at the realities we use the words to talk about: we are
using a sharpened awareness of words to sharpen our perception of, though not
as the final arbiter of, the phenomena.
(1956, p. 182)

Austin suggests that his way of doing philosophy might be called ‘linguistic phenom-
enology’ rather than ‘ordinary language’, ‘linguistic’ or ‘analytic’ philosophy, which by
the 1950s were the standard terms. Language is rooted in our activities and practices, so
to consider how words are used in certain situations is precisely to consider what reali-
ties those situations involve. Wittgenstein was equally clear about this (see for example
Philosophical Investigations, §§ 241, 370–3). He, too, might be called a linguistic phe-
nomenologist, a description that helps defuse some of the misunderstandings to which
ordinary language philosophy has been prone.
Austin goes on to consider two further objections. The first he calls ‘the snag of Loose
(or Divergent or Alternative) Usage’. Since we use language in different ways, speak
loosely, and even falsely, how can we appeal to what we say, or would say in imagined
situations, in either dissolving philosophical problems or developing a theory of lan-
guage? Austin’s answer is to deny that we disagree nearly as much as people think—and
certainly not after the considered reflection that he himself encouraged in his discus-
sions with colleagues. Divergences sometimes result, for example, simply from imagin-
ing the situations slightly differently. Far from shying away from disagreement, Austin
suggests, we should pounce upon it, as the explanation of it is bound to be revealing
(1956, pp. 183–4).
The second objection Austin calls ‘the crux of the Last Word’. Why should ordinary
language claim to have the last word on anything? A common distinction may work
well for everyday practical purposes, but different distinctions may be required for sci-
entific purposes or wherever intellectual sophistication is needed. Here Austin’s answer
is concessive: “ordinary language is not the last word: in principle it can everywhere be
supplemented and improved upon and superseded. Only remember, it is the first word”
(1956, p. 185). The terminology introduced in How to Do Things with Words provides a
good example of Austin’s own willingness to go beyond ordinary language in building
on the first words he had so assiduously collected and divided.
Does this mean, as Maxwell and Feigl once put it in a critique of ordinary language
philosophy from the perspective of logical positivism, that “many cases of putative ordi-
nary-usage analysis are, in fact, disguised reformations” (1961, p. 194)? Austin might
indeed be regarded as proposing that we use his refined terminology in recognizing and
explaining performative utterances. But this terminology is not intended to replace ordi-
nary linguistic expressions in the way that the logical positivists conceived the proc-
ess of ‘explication’, as Carnap called it (1950, ch. 1). Austin’s terminology supplements
ordinary language, the detailed clarification of which is needed to motivate and justify
that terminology.

881
 M I CH A EL BE A N E Y

3.2 Influence
As we have seen, Austin’s work had two main purposes: to provide the basis for a sys-
tematic theory of language that recognizes the role of performative utterances and other
speech acts, and to enable the dissolution of traditional philosophical problems. The
theory of speech acts was developed, most notably, by Austin’s student, John Searle
(1932–), in Speech Acts (1969); and speech act theory now occupies a central place in
contemporary philosophy of language. As far as the dissolution of philosophical problems
is concerned, perhaps the most important of those directly influenced by Austin—as well
as by Wittgenstein—is another American philosopher, Stanley Cavell (1926–), who is
probably the leading advocate of ordinary language philosophy in the States. Must we
mean what we say? (1969) and The Claim of Reason (1979) are his two key works.
The detailed investigations of ordinary language, encouraged by Austin and Witt-
genstein, formed the basis of a research program, and there was a whole generation of
philosophers brought up in the 1950s and 1960s who chose their own area of linguistic
discourse to do their fieldwork (a term which Austin himself uses; 1956, p. 183). In the
best work, this was always undertaken to elucidate or dissolve certain traditional philo-
sophical problems, in just the way Austin had done in Sense and Sensibilia and Witt-
genstein had done in Philosophical Investigations. To take just four prominent examples:
Elizabeth Anscombe (1957) analyzed our expressions of intention, Norman Malcolm
(1959) chose dreaming as his fieldwork, Anthony Kenny (1963) investigated our con-
cepts of action, emotion, and will, and Alan White (1975, 1990) explored the jungles
of our modal thinking and of our language of imagination.
H. P. Grice (1913–88) occupies a unique place in the story of ordinary language phi-
losophy. One of his most significant early publications was a paper he wrote with Straw-
son defending the analytic/synthetic distinction against the Quinean critique (Grice
and Strawson 1956). This placed Grice alongside the other Oxford ordinary language
philosophers. He then developed his theory of conversational implicature, which offers
a way of synthesizing semantic theory in the Frege–Russell mold with recognition of the
role played by speech acts (1975, 1981). His main work on this occurred after he moved
to Berkeley in 1967, joining Searle, so we might see this as marking the beginning of the
end of the dominance of Oxford ordinary language philosophy and the development of
new forms of analytic philosophy. Grice’s writings have also been enormously influen-
tial on subsequent philosophy of language.
Ordinary language philosophy has also had an influence outside the analytic tradition.
Perhaps the best example of this is Austin’s influence on Jacques Derrida (1930–2004).
Austin’s attack in Sense and Sensibilia on “the deeply ingrained worship of tidy-looking
dichotomies” (quoted above), and his recognition of the importance and complexity
of what he calls ‘negations and opposites’ in “A Plea for Excuses” finds strong echoes
in Derrida’s project of deconstructing philosophical contrasts and oppositions. Derrida
also had a famous debate with Searle over speech act theory, prompted by Derrida’s
essay “Signature Event Context” (Derrida 1972, 1977; Searle 1977).
Looking back from the twenty-first century, there was certainly a movement in
British philosophy from around 1945 to 1970 for which ‘ordinary language philosophy’
provides a natural ‘first word’. But the simple opposition between ordinary language phi-
losophy and other forms of philosophy—whether ‘ideal language philosophy’, any other
subtradition of analytic philosophy, or indeed so-called continental philosophy—has
long since dissolved into a rich and shifting variety of subtler and smaller-scale divisions

882
 O RDI N A RY LA N G U A G E P H I L O S O P H Y

and oppositions. This suggests that we should not seek to specify any ‘essence’ to ordi-
nary language philosophy. Any opposition that we might be tempted to pose between
ordinary language philosophy and other forms of philosophy needs deconstruction—a
deconstruction that must involve, however, careful attention to the use of ‘ordinary
language’ in both ordinary and philosophical discourse.

Bibliography
Anscombe, G. E. M., 1957, Intention, Oxford: Blackwell
Austin, J. L., 1950, “Truth”, in Austin 1979, pp. 117–33; originally published in Proc. Aris. Soc. Supp., 24
——, 1956, “A Plea for Excuses”, in Austin 1979, pp. 175–204; originally published in Proc. Aris. Soc.,
1956–7
——, 1962a, How to Do Things with Words, eds. J. O. Urmson and M. Sbisà, Oxford: Oxford University
Press
——, 1962b, Sense and Sensibilia, ed. G. J. Warnock, Oxford: Oxford University Press
——, 1979, Philosophical Papers, 3rd ed., eds. J. O. Urmson and G. J. Warnock, Oxford: Oxford University
Press; 1st ed. 1961
Ayer, A. J., 1940, The Foundations of Empirical Knowledge, London: Macmillan
Baldwin, Thomas, 2001, Contemporary Philosophy, Oxford: Oxford University Press
Beaney, Michael, 2006, “Wittgenstein on Language: From Simples to Samples”, in E. Lepore and B. C.
Smith, eds., The Oxford Handbook of Philosophy of Language, Oxford: Oxford University Press, pp. 40–59
——, 2009, “Analysis”, in the Stanford Encyclopedia of Philosophy, online at http://plato.stanford.
edu/entries/analysis
——, forthcoming, The Oxford Handbook of the History of Analytic Philosophy, Oxford: Oxford University Press
Berlin, Isaiah et al., 1973, Essays on J. L. Austin, Oxford: Clarendon Press
Carnap, Rudolf, 1950, Logical Foundations of Probability, Chicago: University of Chicago Press
Caton, Charles E., 1963, ed., Philosophy and Ordinary Language, Urbana: University of Illinois Press
Cavell, Stanley, 1969, Must We Mean What We Say?, Cambridge: Cambridge University Press, 1976
——, 1979, The Claim of Reason, New York: Oxford University Press
Derrida, Jacques, 1972, “Signature Event Context”, Glyph, 1 (1977), pp. 172–97; also in Derrida 1982, pp.
307–30; Derrida 1988, pp. 1–23; originally published in French 1972
——, 1977, “Limited Inc a b c . . .”, Glyph, 2, pp. 162–254; also in Derrida 1988, pp. 29–110
——, 1982, Margins of Philosophy, translated by Alan Bass, Hemel Hempstead: Harvester; orig. publ. in
French 1972
——, 1988, Limited Inc, Evanston: Northwestern University Press
Dummett, Michael, 1991, Frege: Philosophy of Mathematics, London: Duckworth
Frege, Gottlob, 1884, Die Grundlagen der Arithmetik, Breslau: W. Koebner, selections translated in M. Beaney,
ed., The Frege Reader, Oxford: Blackwell, pp. 84–129
Grice, H. P., 1975, “Logic and Conversation”, in Grice 1989, pp. 22–40; originally published 1975
——, 1981, “Presupposition and Conversational Implicature”, in Grice 1989, pp. 269–82; originally pub-
lished 1981
——, 1989, Studies in the Way of Words, Cambridge, MA: Harvard University Press
Grice, H. P. and Strawson, P. F., 1956, “In Defence of a Dogma”, in Grice 1989, pp. 196–212; originally
published in Phil. Rev., 65, pp. 141–58
Hacker, P. M. S., 1996, Wittgenstein’s Place in Twentieth-Century Analytic Philosophy, Oxford: Blackwell
——, forthcoming, “The Linguistic Turn in Analytic Philosophy”, in Beaney forthcoming
Hampshire, Stuart, 1960, “J. L. Austin”, in Rorty 1967, pp. 239–47; originally published in Proc. Aris. Soc.,
60 (1959–60), pp. 2–14
Hanfling, Oswald, 2000, Philosophy and Ordinary Language, London: Routledge
Kenny, Anthony, 1963, Action, Emotion and Will, London: Routledge and Kegan Paul
Malcolm, Norman, 1942, “Moore and Ordinary Language”, in Schilpp 1942, pp. 343–68; repr. in Rorty
1967, pp. 111–24
——, 1959, Dreaming, London: Routledge and Kegan Paul
Marion, Mathieu, 2000, “Oxford Realism: Knowledge and Perception”, Parts I and II, British Journal for the
History of Philosophy, 8, pp. 299–338 (Part I), 485–519 (Part II)

883
 M I CH A EL BE A N E Y

Maxwell, Grover and Feigl, Herbert, 1961: “Why Ordinary Language Needs Reforming”, in Rorty 1967,
pp. 193–200; originally published in J. Phil., 58, pp. 488–98
Monk, Ray, 1990, Ludwig Wittgenstein: The Duty of Genius, London: Jonathan Cape
Moore, G. E., 1899, “The Nature of Judgment”, in Moore 1993, pp. 1–19; originally published in Mind, 8,
pp. 176–93
——, 1925, “A Defence of Common Sense”, in Moore 1993, pp. 106–33; also in Moore 1959, pp. 32–59
——, 1942, “A Reply to my Critics”, in Schilpp 1942, pp. 535–677
——, 1959, Philosophical Papers, London: George Allen and Unwin
——, 1993, Selected Writings, ed. Thomas Baldwin, London: Routledge, 1993
Passmore, John, 1966, A Hundred Years of Philosophy, 2nd ed., London: Penguin; 1st ed. London: Duckworth,
1957
Rorty, Richard, 1967, ed., The Linguistic Turn, Chicago: University of Chicago Press
Russell, Bertrand, 1905, “On Denoting”, Mind, 14, pp. 479–93
——, 1922, “Introduction [to Wittgenstein’s Tractatus]”, in Wittgenstein 1961 [1921], pp. ix–xxii
Ryle, Gilbert, 1932, :Systematically Misleading Expressions”, in Ryle 1971b, pp. 41–65; originally published
in Proceedings of the Aristotelian Society, 32, pp. 139–70; also repr. in Rorty 1967, pp. 85–100
——, 1949, The Concept of Mind, London: Penguin
——, 1953, “Ordinary Language”, in Ryle 1971b, pp. 314–31; orig. in Phil. Rev., 52
——, 1970, “Review of ‘Symposium on J. L. Austin”, in Ryle 1971a, pp. 282–5
——, 1971a, Critical Essays, Collected Papers, Volume 1, Abingdon: Routledge, 2009
——, 1971b, Collected Essays 1929–1968, Collected Papers, Volume 2, Abingdon: Routledge, 2009
Schilpp, P. A., 1942, ed., The Philosophy of G. E. Moore, 3rd ed., La Salle, Illinois: Open Court, 1968; 1st
ed. 1942, 2nd ed. 1952
Searle, John, 1969, Speech Acts: An Essay in the Philosophy of Language, Cambridge: Cambridge University
Press
——, 1977, “Reiterating the Differences: A Reply to Derrida”, Glyph, 1, pp. 198–208
Soames, Scott, 2003, Philosophical Analysis in the Twentieth Century, 2 vols., Princeton: Princeton University
Press
Strawson, P. F., 1950a, “On Referring”, in Strawson 1971, pp. 1–27; originally published in Mind, 59,
pp. 320–44
——, 1950b, “Truth”, in Strawson 1971, pp. 190–213; originally published in Proc. Aris. Soc. Supp., 24
——, 1952, Introduction to Logical Theory, London: Methuen
——, 1971, Logico-Linguistic Papers, London: Methuen
——, 1992, Analysis and Metaphysics: An Introduction to Philosophy, Oxford: Oxford University Press
Stroll, Avrum, 2000, Twentieth-Century Analytic Philosophy, New York: Columbia University Press
Travis, Charles and Kalderon, Mark, forthcoming, “Oxford Realism”, in Beaney forthcoming
Urmson, J. O., 1956, Philosophical Analysis: Its Development between the Two World Wars, Oxford: Oxford
University Press
Warnock, G. J., 1969, English Philosophy since 1900, Oxford: Oxford University Press; 1st ed. 1958
——, 1989, J. L. Austin, London: Routledge
White, Alan R., 1975, Modal Thinking, Oxford: Blackwell
——, 1990, The Language of Imagination, Oxford: Blackwell
Wilson, John Cook, 1926, Statement and Inference, 2 vols., Oxford: Clarendon Press
Wittgenstein, Ludwig, 1921, Tractatus Logico-Philosophicus, translated by D. F. Pears and B. McGuinness,
London: Routledge, 1961, 1974; originally translated by. C. K. Ogden, London: Routledge, 1922; origi-
nally published in German, 1921
——, 1953, Philosophical Investigations, translated by G. E. M. Anscombe, Oxford: Blackwell; 3rd ed. 1978

Further Reading
For further discussion of ordinary language philosophy and its place in the history of philosophy, see Baldwin
2001, ch. 3; Berlin et al. 1973; Caton 1963; Hanfling 2000; Passmore 1966, ch. 18; Rorty 1967, Part III;
Soames 2003, Vol. 2, Parts 2–4; Stroll 2000, chs. 4–6; Urmson 1956, Part III; Warnock 1969, 1989. See
also Hacker forthcoming on the linguistic turn; Beaney 2006 on Wittgenstein’s conception of language;
Marion 2000 and Travis and Kalderon forthcoming on the influence of Oxford realism on Austin and oth-
ers; and Beaney 2009, §6, on analysis in analytic philosophy. Ryle discusses our talk of ‘ordinary language’
in Ryle 1953; and Grice reflects on Oxford philosophy in Grice 1989, chs. 10–11.

884
 7.7
PRAGMAT IC S AND
C ONT E XT : T HE
D E VE L OP M ENT
OF INT E N SIONAL
S E MAN T IC S
Jason Stanley

Introduction
In the twentieth century, logic and philosophy of language are two of the few areas of
philosophy in which philosophers made indisputable progress. For example, even now,
many of the foremost living ethicists present their theories as somewhat more explicit
versions of the ideas of Kant, Mill, or Aristotle. In contrast, it would be patently absurd
for contemporary philosophers of language or logicians to think of themselves as work-
ing in the shadow of any figure who died before the twentieth century began. Advances
in these disciplines make even the most unaccomplished of its practitioners vastly more
sophisticated than Kant. There were previous periods in which the problems of lan-
guage and logic were studied extensively (for example the medieval period), but from
the perspective of the progress made in the last 120 years, earlier work is at most a source
of interesting data or occasional insight. All systematic theorizing about content that
meets contemporary standards of rigor has been done subsequently.
The advances philosophy of language has made in the twentieth century are of course
the result of the remarkable progress made in logic. Few other philosophical disciplines
gained as much from the developments in logic as the philosophy of language. In the
course of presenting the first formal system in the Begriffsscrift, Gottlob Frege devel-
oped a formal language. Subsequently, logicians provided rigorous semantics for formal
languages, in order to define truth in a model, and thereby characterize logical con-
sequence. Such rigor was required in order to enable logicians to carry out semantic
proofs about formal systems in a formal system, thereby providing semantics with the
same benefits as increased formalization had provided for various branches of math-
ematics. It was but a short step to treating natural languages as more complex versions of
formal languages, and then applying to the study of natural language the techniques
 J A SO N STAN L E Y

developed by logicians interested in proving semantic results about formal theories.

Increased formalization has yielded dividends in the philosophy of language similar to
those in mathematics. It has enabled philosophers to provide better and more fruitful
definitions and distinctions.
Progress in philosophy of language and logic has positively affected neighboring dis-
ciplines such as metaphysics and meta-ethics. Because of this, some philosophers have
regarded philosophy of language as some kind of “first philosophy,” as Descartes viewed
what we would now call “epistemology.” That philosophy of language has progressed
significantly does not mean that it provides us with a first philosophy. One can rec-
ognize that a discipline has advanced more than others without thinking that it holds
the key to all advancement. The twentieth century was the century of “linguistic phi-
losophy,” not because all or even most philosophical problems have been resolved or
dissolved by appeal to language, but because areas of philosophy that involved meaning
and content became immeasurably more sophisticated.

Grice and the Semantics-Pragmatics Distinction

In his extraordinarily influential paper “Logic and Conversation,” Grice (1989b) set
out to defend the truth-table analysis of the meaning of the natural language logical
particles from the ordinary language onslaught. When Strawson spoke of the connec-
tion between antecedent and consequent that is part of the “primary use” of an English
conditional statement, he spoke of the acceptability, truth, or reasonability of a use of a
conditional statement. This suggests that Strawson did not distinguish the truth of an
utterance from the acceptability of that utterance. The key to Grice’s defense of the
truth-table analysis of the meanings of “and,” “or,” and “if . . . not” is that these notions
can (and often do) come apart. A given utterance can be true, even though uttering it
is not acceptable because it violates conversational norms. In explaining this distinc-
tion, Grice provided the foundations for a theory of conversational norms. This theory
clearly explains how an utterance may be true, though unacceptable as an assertion, due
to specific facts about the conversation and its participants. Grice then used the distinc-
tion between the truth of a statement and its conversational acceptability in a defense of
the thesis that the connectives of propositional logic were correct explications of their
natural language counterparts. More specifically, Grice assumed that natural language
logical particles have the truth-table meanings of their logical counterparts, and argued
that features of the uses of these expressions that are not explicable by truth-tables are
due to facts about the norms governing conversation, rather than the meanings of the
words.
According to Grice, conversation is a cooperative rational activity; each conversa-
tion has a purpose. This fact about conversations imposes as a norm what Grice (1989b,
p. 26) calls the Cooperative Principle: “Make your conversational contribution such as
is required, at the stage at which it occurs, by the accepted purpose or direction of the
talk exchange in which you are engaged.” The Cooperative Principle is the overarching
principle guiding conversation. Following it imposes a number of more specific norms
on conversational participants. For example, lying involves one kind of uncooperative
conversational behavior, being purposely irrelevant involves another, and not being
sufficiently informative is a yet a third kind of uncooperative behavior. Following the
Cooperative Principle isn’t always a matter of saying something true, relevant, and max-
imally informative; it is also a matter of how one says what one says. According to the

886
 I N TEN SI O N A L S E MA N T I C S

maxim of manner, one should try to list events in the order in which they occur, and to
cite causes before effects.
Using these conversational principles, Grice attempts to explain many of the facts
about standard use cited by Strawson without abandoning the thesis that the same
truth-table analysis for logical connectives also provides the meanings of their natu-
ral language counterparts. Consider Strawson’s point that “or” in ordinary language
is ambiguous between an exclusive use (“but not both”) and an inclusive use. Assuming
that “or” unambiguously means inclusive “or” (the meaning of the logical connective
for disjunction), one can explain the fact that “or” is often used exclusively by general
conversational principles. Suppose Hannah uttered an instance of “P or Q,” but in fact
believed that both P and Q were true. Then Hannah would not be maximally inform-
ative; she would be violating Grice’s conversational maxim of quantity. So someone
believing that P and Q, to follow conversational norms should say the more informative
P and Q rather than the less informative P or Q (which is compatible with the truth
of only one of P and Q). So, when someone utters an instance of ‘P or Q’, they convey
(without asserting, as part of the linguistically determined content) that they do not
know that P and Q. The fact that this is part of what is conveyed by following conver-
sational principles, rather than what is asserted as part of the linguistically determined
content, can be ascertained by appeal to Grice’s central criterion for distinguishing
what is part of what is said (the linguistically determined asserted content) from what
is merely conversationally conveyed, which is the test of cancellability. One can cancel
the implication conveyed by an utterance of “P or Q” (which is that one doesn’t know
both P and Q) by saying “P or Q; in fact, both P and Q are true,” as in “John is with Bill
or he is with Frank; in fact he is with both,” So, consistently with the assumption that
“or” unambiguously means inclusive “or,” one can explain why “or” is often used as if it
meant exclusive “or.”
One can use the very same kind of explanation to dissolve the sense that it is part of
the meaning of a disjunctive statement that the speaker is unaware of the truth of either
disjunct. If Hannah knows that John was at the party, it would be a violation of the
maxim of quantity for her to assert that either John was at the party or he was at home.
She would not be being maximally informative by asserting the disjunctive statement,
and hence would be violating the maxim of quantity. Furthermore, the implication
that the speaker is unaware of the truth of either disjunct can be canceled, as in Grice’s
example (1989c, pp. 44–5): “The prize is either in the garden or in the attic. I know that
because I know where I put it, but I’m not going to tell you.” Thus, one can explain the
fact that a disjunctive statement is usually only proper if the speaker is unaware of the
truth of either disjunct, without making that fact part of the conventional meanings of
any words.
Grice also attempted to provide pragmatic explanations (that is, explanations from
general principles governing conversation) for the divergences between the truth-table
meaning for the conditional and ordinary indicative conditionals (Grice, 1989d). In
attempting to account for the connection thesis, the thesis that a conditional is only
assertible if the antecedent provides a ground or good reason to accept the consequent,
Grice (1989d, pp. 61–2) appealed to the conversational maxims, in particular that of
quantity, which directs interlocutors always to assert the strongest claim consistent with
their evidence, and that of quality, which directs them to have adequate evidence for
their assertions. If the indicative conditional is the material conditional, then it is true
if and only if the antecedent is false or the consequent is true. If the speaker knows that

887
 J A SO N STAN L E Y

the antecedent is false, adherence to the maxim of quantity requires that the speaker
simply assert the negation of the antecedent, rather than the whole conditional; muta-
tis mutandis for the truth of the consequent. So a conditional is only assertible if the
speaker is unaware of the truth values of the antecedent and consequent. But the maxim
of quality requires anyone who asserts a conditional to have evidence for the truth
of the material conditional. Since, for the reasons just given, the evidence cannot be
truth-functional (that is, the speaker’s grounds cannot be knowledge of the truth-values
of the antecedent or consequent), the speaker must have non-truth-functional grounds
for her assertion of the material conditional, if she is adhering to the maxims of quantity
and quality. So, asserting an indicative conditional, on the supposition that it has the
meaning of the material conditional, requires the speaker to have non-truth-functional
grounds for her assertion. More specifically, it requires the speaker to know or believe
that the antecedent would be a good ground for the consequent.
As we shall see, there are a number of problems with Grice’s defense of the material
conditional analysis of indicative conditionals. But Grice’s defense of the thesis that
the meaning of “or” is exhausted by the truth-table for inclusive “or” has been widely
accepted, as have a number of other Gricean explanations of use-facts. The moral of
Grice’s work is that the facts of linguistic use are a product of two factors, meaning and
conversational norms. Failure to absorb this fact undermines many of the main theses
of ordinary language philosophy.
However, there were two aspects of the ordinary language philosopher’s position.
The first involved emphasizing the divergences in use between the logical terms and
their ordinary language counterparts. The second involved the fact that natural lan-
guages involve context-sensitive words (for example “I,” “here,” and “now”), and that
many words only have reference relative to a context of use, and many sentences only
have truth-values relative to a context of use. Since reference and truth-value are only
properties of uses of expressions, they are inappropriate notions to use in the analysis
of the linguistic meanings of expressions. In general, expression types in natural lan-
guage do not have references or truth-values, only uses of them do. So employing the
apparatus of semantic theory, which crucially avails itself of notions such as reference
and truth, is not the right way to give a theory of meaning for natural language; the
meaning of expression types is given by rules of use. Grice’s response to the ordinary
language philosopher only speaks to the first of these aspects of the ordinary language
philosopher’s position. But a response to the second aspect of the ordinary language
philosopher’s position was to emerge from the work of those who developed and refined
intensional semantics.

The Development of Intensional Semantics: From Montague to Kaplan

Carnap’s semantic theory crucially exploits the notion of a possible world in defining
semantic values of expressions. Each expression has, as its primary semantic value, an
intension, which is a function from a possible world to the extension of that expression
at that world. In the case of sentences, the intension of a sentence is a function from
possible worlds to truth-values. Carnap’s semantic theory has, as its “central notion” (in
Michael Dummett’s sense), the notion of truth with respect to a possible world. The logi-
cian Richard Montague, a student of Tarski’s, argued that a theory of meaning should
take the more general form of truth with respect to a context of use, where possible worlds
are but one feature of a context of use (Montague, 1974b, p. 96). Montague treated a

888
 I N TEN SI O N A L S E MA N T I C S

context of use as an index, a collection of semantically relevant aspects of the context

of use. If the language in question contained tenses and modal operators, then the indi-
ces involved in the semantic interpretation of that language would contain times and
worlds. If the language also contained the indexical terms “I” and “here,” the indices
would aklso have persons and places as aspects. Montague then generalized Carnap’s
notion of intension: instead of an intension being a function from possible worlds to
extensions, an intension, for Montague, was a function from indices to extensions. For
example, the intension of a sentence such as “I am tired” would be a function from
indices to truth-values; it would take an index whose aspects were times, worlds, and
persons to be true if and only if the person at the index was tired at the time and world
of the index.
The interpretation of modal operators in Montague’s system was also a generaliza-
tion from their interpretation in modal semantics. In Carnap’s system, the function
of modal operators was to shift the evaluation of a proposition from one possible world
to another; a modal operator took an intension, and evaluated that intension at other
possible worlds. On this account, a sentence such as “possibly S” is true relative to a
world w if and only if the intension of S is true in some (possibly distinct) world w'. So
the function of “possible,” for Carnap, is to shift the evaluation of the intension of S
from w to w'; “possibly S” is true in w if and only if S is true in w' (and the function of
“necessarily” is to shift the evaluation of the content of the embedded sentence to all
possible worlds). In Montague’s system, modal and tense operators evaluate intensions
at indices rather than just possible worlds. On this account, a sentence such as “possibly
S” is true at an index i if and only if the intension of S is true at i', where i' differs from i
at most in its world feature. So rather than truth with respect to a possible world being
the fundamental notion, truth with respect to an index is Montague’s fundamental
notion, with worlds being one element of an index. This apparatus allowed Montague
to generalize the apparatus of intensional semantics to treat context-sensitivity in natural
language without sacrificing the elegant treatment of modal and tense operators. As
we shall see below, this leads to an alternative response to the challenge from ordinary
language philosophy than the one developed by Grice.
Montague’s contributions to the systematic study of language went well beyond gen-
eralizing intensional semantics to capturing tense and context-sensitivity. Montague’s
most influential papers focused on intensional constructions in natural language. In
Montague (1974d), he gave an account of a number of intensional constructions other
than the classic cases of propositional attitude verbs and modal and temporal contexts.
For example, Montague provided a semantic analysis of intensional transitive verbs, such
as “seek” and “worship.” The difference between intensional transitive verbs and exten-
sional transitive verbs (such as for example “kick” and “meet”) is that while one cannot
meet a unicorn or kick a unicorn (since there are no unicorns), one can nevertheless
seek a unicorn. So, whereas satisfying an instance of the predicate “meeting N” requires
that that there is some existent entity that one meets, satisfying an instance of the pred-
icate “seeking N” does not require that there is some existent entity that one seeks.
Beginning with Frege and Russell intensional transitive verbs had generally been
ignored in the literature, largely because of the influence of Russell’s theory of descrip-
tions, the standard method of dissolving apparent reference to nonexistent entities.
Russell’s theory involves providing a contextual definition of definite descriptions;
meanings are assigned only to sentences containing definite descriptions rather than
the definite descriptions themselves. Russell’s theory helps us analyze away apparent

889
 J A SO N STAN L E Y

reference to nonexistent entities in a construction such as “John believes that the foun-
tain of youth is in Peru,” since we can apply the theory to the sentence “the fountain of
youth is in Peru,” and arrive at an object of John’s belief, without there being a fountain
of youth. In contrast, one cannot use Russell’s theory to arrive at an object of seeking for
a construction such as “Pizarro sought the fountain of youth,” since that theory gives us
no way of treating the definite description “the fountain of youth” in isolation. For this
reason, Quine (1960, section 32) regimented intensional transitive verbs away in favor
of propositional attitude verbs (so the intensional transitive construction “x looks for y”
becomes the propositional attitude construction “x endeavors that x finds y”). The fact
that analytic philosophers had not produced a successful analysis of intensional transi-
tive verbs must be viewed as a bit of an embarrassment. The problem of intensional
transitive verbs is one of the original motivations for twentieth-century discussions of
content. For example, it was salient in the minds of Brentano and his students, who
sought to render consistent the thesis that the characteristic feature of mental states was
that they were about things, and that one could have a mental state the object of which
did not exist. Montague’s discussion of intensional transitive verbs was thus a watershed
moment in the theory of meaning. It has subsequently given rise to a lively literature in
semantics and philosophy of language on the topic (for example Partee, 1974; Zimmer-
man, 1993; Forbes, 2000; Richard, 2001).
Montague’s semantic theory was not just distinctive for its focus on intensional con-
structions in natural language. Montague also returned philosophers of language and
semanticists to a tradition that was lost or at the very least obscured in the kind of
semantic theory favored by Tarski (and Davidson). Recall that Frege treated the tradi-
tional relation between the subject of a sentence and its predicate as that of an argument
to a function. That is, Frege regarded the fundamental relationship between the seman-
tic values of expressions in a sentence to be one of functional application. Though quanti-
fiers, for Frege, had the function of binding variables within their scope, they also had
determinate semantic values, namely second-level functions. For example “everything”
denoted a function from first-level functions to truth-values. The denotation of “every-
thing” is a function that takes any first-level function that takes every object to the true,
to the true, and takes every other entity to the false. Similarly, the denotation of “some-
thing” takes to the true any first-level function that has the true for at least one value,
and everything else to the false. So Frege operated with an ontology that was stratified
into types; there were objects, then functions from objects to truth-values (first-level
functions), then functions from first-level functions to truth-values (second-level func-
tions), and on up. In Tarski’s work, by contrast, no use is made of functional applica-
tion as a relation between semantic values. Quantifiers are not assigned functions of
various kinds; an object-language universal quantifier over objects is interpreted via the
use of a meta-language quantifier over sequences. Montague’s semantics returned phi-
losophers of language to the Fregean tradition of treating semantic values as functions
from arguments to values, with functional application as the primary mode of semantic
composition. There are lively foundational debates between advocates of Montague’s
type-theoretic approach to semantics and advocates of the more Tarskian approach, such
as James Higginbotham.
Montague’s marriage of intensional semantics with type-theory was extraordinarily
fruitful, and led (with the help of the work of his distinguished student Barbara Partee)
to the emergence of semantics as a new discipline within linguistic theory. But that is
not to say that the generalization of intensional semantics that was at the heart of his

890
 I N TEN SI O N A L S E MA N T I C S

program has been universally accepted; in fact, the majority of philosophers of language
today regard it as incorrect. The mistake made by Montague was to think that the
study of modality was a branch of pragmatics, the study of context-sensitivity in natural
language. Recall that Montague’s generalization of intensional semantics consisted of
treating possible worlds as features of the more general notion of a context of use. He
then generalized the treatment of operators as shifting the evaluation of the truth of a
content from one world to the next, to shifting the evaluation of the truth of a content
from one context of use (or index) to the next. It is this generalization that is widely
(but not universally) regarded as an error.
The first hint that something was amiss in the assimilation of modal and temporal
operators to the general study of truth relative to a context of use came from Hans
Kamp’s work on the temporal indexical “now.” Kamp, a student of Montague’s, estab-
lished several theses about temporal logic. The first is there are certain natural language
sentences that have readings that are most perspicuously captured via the postulation
of more than just past and future tense operators. For example, consider the sentence in
(1) (from Kamp, 1971, p. 231):

(1) A child was born that will become the ruler of the world.

Sentence (1) means something like “In the past, a child was born who, in the future
of the present moment, becomes the ruler of the world.” In order supply (1) with its
natural reading, one needs to have at one’s disposal an operator with the meaning of the
English word “now,” whose function is to evaluate its embedded content at the present
moment. Kamp then established that a satisfactory semantics for “now” requires having
two times in the Montagovian “index” that is supposed to represent a context of use.
One of the times would be shifted by temporal operators such as “it was the case that”
and “it will be the case that.” The other time would be intuitively the time of the utter-
ance, and would never be shifted by any operators. Its function would be to allow for
the interpretation of any occurrences of “now” in the sentence. The present moment
feature of the Montagovian index could not be shifted by any operators, because oth-
erwise, in interpreting any embedded occurrences of “now” (that is, embedded inside
other temporal operators), one could no longer access the present moment, and thereby
successfully interpret “now”.
Kamp’s insights about “now” carry over directly to the modal indexical “actual.” In
order to provide a successful interpretation of embedded occurrences of “actual” (that
is, occurrences of “actual” inside other modal operators), each Montagovian index must
contain two worlds, one that would be shifted by modal operators, and the other that
would be the world of utterance. Interpreting an embedded occurrence of “actual” (that
is, one that occurs within the scope of other modal operators) requires keeping track
of the world of utterance. For the function of the initial modal operator is to shift the
evaluation of the content of the embedded sentence to another possible world, and one
needs to retain the information about the actual world of utterance, in order to inter-
pret any occurrences of “actual” within that embedded sentence. So Montague’s indices
each would contain two kinds of features. First, they would contain features (worlds and
times) that were shifted by operators. Second, to interpret indexical operators such as
“now” and “actual,” the indices would contain features that were not capable of being
shifted by operators but would always represent features of the actual context of use of
the sentence being uttered.

891
 J A SO N STAN L E Y

Kamp’s work suggests that within a single Montagovian index there are two quite
different sorts of features. First, there are features that are shifted by operators, such
as “necessarily” and “possible” (and the past and future tense, assuming that they are
operators). Second, there are features that intuitively represent features of the actual
context of use. These include the moment at which the utterance was made, and the
world at which the utterance was made, which are required, respectively, to interpret
indexical operators such as “now” and “actual.” Furthermore, these features are not
capable of being shifted by operators, or else one could not interpret embedded occur-
rences of indexical operators. So, for example, in evaluating the truth of the intension
of say “Necessarily S” with respect to an index i, one would evaluate the intension
of S at all indices i’ that differed from i at most in their world index, and shared with
i all the features relevant for interpreting indexical operators—that is, all those fea-
tures that represent aspects of the context of use in which “Necessarily S” was uttered.
This suggests that Montague’s indices are not natural kinds. Each index contained two
kinds of information: information relevant for interpreting modal and temporal opera-
tors, on the one hand, and information that represented features of the context of use
which are relevant for interpreting indexical expressions such as “now,” “actual,” “I,”
and “here.”
There were also other reasons to be suspicious of Montague’s index-theoretic approach.
In 1970, Robert Stalnaker pointed out (Stalnaker, 1999, pp. 36ff.)) that Montague’s
semantics (or as Stalnaker calls it, Montague’s “Semantics-Pragmatics”) did not allow
for the representation of propositions. For Montague, there is only one semantic content
of an utterance (or occurrence) of the sentence “I am tired,” and that is a function from
contexts of use to truth-values. If Hannah utters “I am tired,” and John utters “I am
tired,” the only difference there is between the contents of their utterances is that one
may be true and the other may be false (that is, the value of the semantic content of
“I am tired” may be different, because it is being evaluated relative to distinct indices).
But, as Stalnaker emphasizes, there are additional differences between their utterances.
Intuitively, what Hannah said when she uttered “I am tired” is distinct from what John
said when he uttered “I am tired”; they expressed different propositions. But there is no
semantic value in Montague’s system that represents the different propositions in ques-
tion. There is just a function from indices to truth-values associated with “I am tired,”
and this is not the proposition expressed by either of these utterances of “I am tired”
(since they express different propositions).
Kamp’s work clearly shows the need for “double indexing.” The first kind of index
is required to interpret indexical expressions occurring within a sentence. The second
kind of index is required to give the proper semantics for operators on content, such as
“necessarily” and “possible” (and the tenses, if they are operators). But it took another
student of Montague’s, David Kaplan, to draw out the real moral behind the need for
double indexing. In his seminal work “Demonstratives” (Kaplan, 1989), Kaplan argues
that the two kinds of indices correspond to two kinds of semantic values. The first kind
of index represents the dependence of semantic value upon context. The semantic con-
tent of a context-dependent sentence such as “I am tired” depends upon features of the
context of use. If John is the speaker in the context of use, then “I am tired” expresses
the proposition that John is tired; if Hannah is the speaker in the context of use, then
it expresses the proposition that Hannah is tired. The second kind of index represents
the dependence of truth of a semantic content on a circumstance of evaluation (such as
a possible world, or a time if tenses are operators on contents), and is required to give a

892
 I N TEN SI O N A L S E MA N T I C S

satisfactory semantics for sentence-operators. A proposition may be true at one possible

world but false at another.
Accordingly, expressions are associated with two kinds of semantic values, which
Kaplan called character and content respectively. The character of an expression is a
function from a context of use to the content of that expression relative to that context.
According to Kaplan, the character of an expression is also the linguistic meaning of
that expression. So, the linguistic meaning of the first-person pronoun “I” is a function
from contexts of use to persons (intuitively, the speakers of those contexts). Any use of
“I” has the same meaning as any other use, though a possibly distinct semantic content.
Kaplan took the semantic contents of singular terms, such as proper names and indexi-
cals such as “I” to be their referents, in Frege’s sense, and he took the semantic contents
of sentences relative to contexts to be propositions. Sentences operators such as “neces-
sarily” and “possibly” shifted the world feature of the index that represented the circum-
stance of evaluation. The index that represented the context of use did not contain any
features that were shifted by operators in the language (Kaplan, 1989, pp. 510ff.).
By dividing features of indices into contexts of use and circumstances of evaluation,
Kaplan’s semantic theory represents a clear advance over Montague’s. It explains why
only certain features are shiftable by operators and, more importantly, it gives a seman-
tic representation of propositions (the values of the characters of sentences). As a result,
Kaplan’s distinctions have been widely adopted in philosophy of language over the
last thirty years. In particular, most philosophers have come to accept that context-
dependent expressions show that there are two levels of semantic value; first, linguistic
meaning, and second, the content of an occurrence of an expression on an occasion.
Different occurrences of an expression might have different semantic contents, despite
sharing a linguistic meaning, as is so clearly the case with the first-person pronoun “I”
and other indexicals.
The work done by Montague and then Kaplan allows for another kind of reply to
the ordinary language philosopher’s skepticism about the possibility of giving a rigorous
semantics for a natural language other than the one provided by Grice. Recall that the
ordinary language philosopher’s skepticism arose from the conviction that truth and ref-
erence were properties of uses of expressions, rather than properties of expressions, and
meanings were rules for using those expressions. Kaplan’s semantic theory undermines
these considerations. It does make perfect sense to speak of singular terms having refer-
ence, albeit relative to a context, and it makes perfect sense to speak of sentences having
truth-values, also relative to a context. So it makes perfect sense to attribute reference and
truth to expression types, once contextual relativity is factored into the semantic theory.
Whereas the notion of a rule of use is vague and mystical, Kaplan’s notion of the character
of an expression is not only clear but set theoretically explicable in terms of fundamental
semantic notions; the character of an expression is a function from a context to the refer-
ence of that expression in that context. Far from context-sensitivity being an impediment
to giving a proper account of linguistic meaning in terms of reference and truth, appeal to
these semantic notions allows us to give a considerably more explicit characterization of
linguistic meaning than the ordinary language philosophers were capable of providing.

References
Forbes, G. (2000) “Objectual Attitudes”, Linguistics and Philosophy 23(2): 141–183.
Grice, P. (1989a)Studies in the Way of Words, Cambridge, MA: Harvard University Press.

893
 J A SO N STAN L E Y

——(1989b) “Logic and Conversation”, in Grice (1989): 22–40.

——(1989c) “Further Notes on Logic and Conversation”, in Grice (1989) 41–57.
——(1989d) “Indicative Conditionals”, in Grice (1989): 58–85.
Kamp, H. (1971) “Formal Properties of ‘Now’”, Theoria 37(3): 227–273.
Kaplan, D. (1989) “Demonstratives” and “Afterthoughts”, in J. Almog, J. Perry, and H. Wettstein, (eds.),
Themes from Kaplan, Oxford University Press, New York.
Montague, R. (1974b)Formal Philosophy, New Haven: Yale University Press.
——(1974b) “Pragmatics”, in Montague (1974): 95–118.
——(1974c) “On the Nature of Certain Philosophical Entities”, in Montague (1974): 148–187.
——(1974d) “On the Proper Treatment of Quantification in Ordinary English”, in Montague (1974):
247-270.
Partee, B. (1973) “Some Structural Analogies between Tenses and Pronouns in English”, Journal of Philosophy
70: 601–9.
——(1974) “Opacity and Scope”, in M. Munitz and P. Unger (ed.)Semantics and Philosophy, New York:
NYU Press: 81–101.
Quine, W.V.O. (1960) Word and Object,Cambridge, MA: MIT Press.
Richard, Mark (2001): “Seeking a Centaur, Adoring Adonis: Intensional Transitives and Empty Terms”, in
P. French and H. Wettstein (eds.) Figurative Language: Midwest Studies in Philosophy Volume 25, Oxford:
Basil Blackwell: 103–127.
Stalnaker, Robert (1999) “Pragmatics”, in Stalnaker (1999): 31-46. [First published in Synthese 1970].
Zimmerman, T. E. (1993) “On the Proper Treatment of Opacity in Certain Verbs”, Natural Language
Semantics 1: 149–179.

894
 7.8
A BR IE F HI ST OR Y OF
G ENE R ATI V E GR AM M AR
Robert Freidin

1 The Advent of Transformational Generative Grammar

The modern inquiry into the nature, use, and origins of human language known as
generative grammar emerged in the middle of the 20th century. It began with a con-
ceptual shift concerning the goals of linguistic inquiry, and this first step led to further
fundamental shifts in the study of language involving our understanding of language as a
mental construct, the notion of grammar, and the nature of the human language faculty,
of which a speaker’s grammar constitutes a significant part.
The first step towards the current perspective involved a shift in focus from the exter-
nal forms of a language to the grammatical system that generates it, a system that is
assumed to exist in the mind of the speaker, constituting the core part of a speaker’s
knowledge of his or her language. From this perspective, the study of language is a sub-
part of several areas in philosophy, including first and foremost epistemology and the
related area of philosophy of mind, in addition to philosophy of language.
The brief history of generative grammar presented here follows an essentially single
line of development from the origin of transformational grammar in the 1950s to the
Minimalist Program, which began in the early 1990s and continues in the present.
Because of space limitations, it leaves out many details, including alternative theories
that have been proposed, some still current. Thus it is not intended as a complete his-
tory of the field.
The shift in focus from external language to mental grammar took place in several
steps, the first of which concerns Noam Chomsky’s earliest work in linguistics, when
as an undergraduate at the University of Pennsylvania in 1947 he read the proofs of
Zellig Harris’s Methods in Structural Linguistics (1951), his formal introduction to the
field. Chomsky’s first research project was (at Harris’s suggestion) an attempt to ana-
lyze a particular language following Harris’s methods. The attempt itself was a failure
(see Chomsky 1975:30), but led to the first formulation of modern generative gram-
mar—ultimately Chomsky’s 1951 master’s thesis The Morphophonemics of Modern
Hebrew (henceforth MMH, published as Chomsky 1979). Thus a comparison of the
two approaches is illuminating.
Harris describes Methods as follows:

This volume presents methods of research used in descriptive, or more exactly,

structural, linguistics. It is thus a discussion of the operations which the
 RO BERT F R E I D I N

linguist may carry out in the course of his investigations, rather than a theory
of the structural analyses which result from these investigations. The research
methods are arranged here in the form of the successive procedures of analysis
imposed by the working linguist upon his data.

Note how Harris separates the methodological operations of linguistic analysis from “a
theory of structural analyses” resulting from them. Methods focuses solely on the former,
essentially taxonomic procedures for the segmentation of a linguistic expression into its
constituent parts and the classification of those elements. These operations yield what
can be called a grammar of lists, taking the term “grammar” to be some characterization
of a language. The purpose of this grammar of lists is to state all the regularities in a
language, derived from an exhaustive analysis of a presumably representative corpus of
utterances. Although this assumes that the finite representative corpus can be extended
to the entire unbounded language, how this actually works is not addressed.
In striking contrast, MMH bypasses these taxonomic procedures and focuses instead
on the construction of a grammar as a set of formal rules that enumerates the grammati-
cal sentences in a language, of which the linguist’s analyzed corpus is merely a special
subset. For MMH the linguistic analysis of a language becomes “the process of con-
verting an open set of sentences—the linguist’s incomplete and in general expandable
corpus—into a closed set—the set of grammatical sentences—and of characterizing
this latter set in some interesting way” (MMH: 1). A footnote to the previous quota-
tion addresses the unbounded character of languages: “thus the resulting grammar will
contain a recursive specification of a denumerable set of sentences” (MMH: 67). This
formulation captures the core of modern generative grammar.
In LSLT, written four years after MMH, Chomsky sets out three goals for research in
generative grammar: constructing the grammars of individual languages, formulating a
general theory of linguistic structure, and validating the results of this work. The first
two goals are interdependent, as Chomsky notes (pp. 78–79).

In constructing particular grammars, the linguist leans heavily on a precon-

ception of linguistic structure, and any general characterization of linguistic
structure must show itself adequate to the description of each natural language.
The circularity is not vicious, however. The fact is simply that linguistic theory
has two interdependent aspects. At any given point in its development, we can
present a noncircular account, giving the general theory as an abstract formal
system, and showing how each grammar is a particular example of it. Change
can come in two ways—either by refining the formalism and finding new and
deeper underpinnings for the general theory, or by finding out new facts about
languages and simpler ways of describing them.
(p. 79)

In practice the two goals are mutually reinforcing.

In the early work in generative grammar the concept of a language is taken to be
well understood. In §56.1 of LSLT a language is defined simply as “a set of utterances”
(p. 201). In the introduction to LSLT, written two decades later, the definition is given
as “a set (in general infinite) of finite strings of symbols drawn from a finite “alphabet” . . .
[that is] determined by general linguistic theory, in particular, by universal phonetics,
which specifies the minimal elements available for any human language and provides

896
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

some conditions on their choice and combination” (p. 5). What determines member-
ship in the set is whether a native speaker finds a given sentence acceptable (cf. Chom-
sky 1957: 13). However, defining language as a set of objects attempts to provide an
external characterization—something whose existence is independent from individual
speakers of a language, what Chomsky calls an E(xternalized)-language. (See Chomsky
1986b for a critical discussion of this notion.)
The concept of grammar is more straightforward. A grammar of a language L consists
of “a system of rules that specifies the set of sentences of L and assigns to each a struc-
tural description” (LSLT, p. 5). The structural description of a sentence specifies all of
its elements and their organization, so it follows that set of structural descriptions of the
sentences of a language gives the structure of language. Beyond the linear order of lexi-
cal items, provided overtly in the phonetic representation of linguistic expressions, a
structural description also involves hierarchical structure, in which lexical items group
together as constituent parts. Hierarchical structure plays a major role in interpreta-
tion (for example the ambiguity of a review of a book by two professors), although it is
entirely covert. A grammar is thus an explicit characterization of the structure of a
language.
In its initial conception, a grammar consists of two distinct formal mechanisms for
generating the sentences of a language: phrase structure rules and transformations. The
former constructs initial syntactic structures which the latter could modify in certain
ways, including the combination of pairs of initial structures—yielding complex and
compound sentences. (1) gives the set of phrase structure rules for English discussed in
Chomsky 1957, which for pedagogical purposes constitutes a substantial simplification
of the rules developed in LSLT.

(1) 1. Sentence o NP + VP
2. VP o Verb + NP
3. NP o {NPsing, NPplural}
4. NPsing o T+N+Ø
5. NPplural o T+N+S
6. T o the
7. N o man, ball, etc.
8. Verb o Aux + V
9. V o hit, take, walk, read, etc.
10. Aux o C (M) (have + en) (be + ing)
11. M o will, can, may, shall, must

(Note that C in (10) stands for the element that will come to represent tense and
number agreement.)
Phrase structure rules are based on a top-down analysis where a sentence is divided
into it major constituent parts and then these parts are further divided into constitu-
ents, and so on until we reach the lexical items (possibly including morphemes—cf. rule
10 above) that make up the sentence. They are given as rewrite rules where the symbol
to the left of the arrow is expanded (rewritten) as the string of symbols to the right of
the arrow. There are two kinds of phrase structure rules in (1): those that introduce only
lexical items (6–7, 9 and 11) and those that rewrite symbols as a string of constituents
(1–5, 8 and 10), thereby specifying the hierarchical structure of phrases. The former
introduce lexical items into derivations (lexical insertion).

897
 RO BERT F R E I D I N

Early work on phrase structure led to the conclusion that the kind of phrase structure
rules under consideration are not adequate to handle certain phenomena in natural lan-
guage in a simple way. For example, the discontinuous dependencies between a verbal
auxiliary and the corresponding affix of the following verbal element (see the under-
lined forms in (2)) could not be easily expressed in a phrase structure rule format.

(2) a. Mary is helping us.

b. Bill has briefed us about the problem.

The distribution of the passive auxiliary (be-en) also created problems for a purely phrase
structure rule account. (See Chomsky 1957, chapter 5; Postal 1964 for the original
arguments, and for historical overview and evaluation, Lasnik 2000 and Freidin 2004.)
These limitations were used to motivate the inclusion of transformational rules, which
it was argued were better suited for capturing the abovementioned phenomena in a
natural and revealing way.
In their earliest instantiation, transformational rules operated on the set of strings
generated by phrase structure rules, called a phrase-marker, to produce a derived phrase-
marker. Consider the passive transformation from Chomsky 1957.

(3) Passive—optional:
Structural analysis: NP – Aux – V – NP
Structural change: X1 – X2 – X3 – X4 o X4 – X2 + be + en – X3 – by + X1

As formulated in (3) the passive transformation identifies four consecutive constituents

and rearranges them, including two insertions of lexical material (the passive auxiliary
be+en and the passive by). For example, the passive transformation takes the phrase-
marker underlying (4a), represented as a phrase structure tree in (5) and converts it to
a phrase-marker for (4b).

(4) a. The boy is kicking the ball.

b. The ball is being kicked by the boy.

(5) Sentence

NP VP

T N
Verb NP

Aux V T N

The boy present be + ing kick the ball

898
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

Specifying a phrase-marker for (4b) raises the problem of derived constituent structure.
LSLT §82.1 notes two possible solutions: either specify the effect each transformation
has on constituent structure in the formulation of the transformation itself or assume
that general conditions apply that restrict derived constituent structure for the output
of each transformation. In practice, the latter alternative is generally assumed. Moreo-
ver, it has been noted since the beginning that phrase structure rules place substantial
constraints on derived constituent structure (see Chomsky 1965).
The formulation in (3) is specific to English and tied to a particular grammatical
construction, the passive. In addition to the two lexical insertions, the transformation
involves the transposition of two NPs. Thus this formulation involves three or four ele-
mentary operations, depending on how the transposition of NPs is analyzed (to which we
will return below). The rule applies to a single phrase-marker, thus belongs to the class
of singulary transformations in contrast to generalized transformations, which combine pairs
of phrase-markers to produce complex and compound sentences. Furthermore, it is des-
ignated as optional—i.e., it does not have to apply when its structural description is met,
in contrast to the obligatory rule that shifts verbal affixes to their appropriate positions.
From the beginning, research on transformations revealed empirical problems for
the formulation and application of these rules. For example, LSLT §95.3 notes that
while (6a) constitutes a viable sentence, the related question (6b) that might plausibly
be derived from the underlying phrase-marker of (6a) via an interrogative transforma-
tion was in fact deviant.

(6) a. Your interest in him seemed to me rather strange.

b. *Whom did your interest in seem to me rather strange?

Chomsky’s solution in LSLT is to complicate the structural description of the transfor-

mation so that it does not apply to such structures. Chomsky (1964: 930–32) discusses
another problem involving the interrogative transformation and this time suggests a
general condition on the application of transformations, which “when appropriately
formalized, might then be proposed as a hypothetical linguistic universal” (p. 931).
The condition states that “if a phrase X labeled as category A is embedded within a
larger phrase ZXW which bears the same label A, then no rule applying to the category
A applies to X (but only to ZXW).” This condition is proposed to insure that trans-
formations cannot apply in multiple ways to a single phrase-marker. This constitutes
the first formulation of what comes to be known as the A-over-A Principle. As dis-
cussed below, it opens the way to the richest field of syntactic research: conditions on
transformations.
The model of grammar that emerges has two distinct components, a set of phrase
structure rules and a set of transformational rules. The output of the first yields initial
underlying phrase-markers, which constitute the input to the second and yield final
derived phrase-markers. Chomsky 1965 introduces a significant change to this model
by introducing generalized phrase structure rules that produce embedded clauses, which
allows for the elimination of generalized transformations (see §3 for further discussion).
In this model initial phrase-markers are designated as deep structures, in contrast to
the final derived phrase-marker in a derivation, which is called a surface structure. The
model is given in (7) below.
Of the two levels of representation, surface structure is obviously the one relevant
to phonological interpretation. In the case of semantic interpretation, which levels are

899
 RO BERT F R E I D I N

(7) Phrase structure rules

Deep structure

Transformational rules

Surface structure

relevant has been the subject of considerable controversy from the 1960s on (see below
for further comment). In addition, there are constraints on derived constituent struc-
ture and on the application of transformations (for example the A-over-A Principle),
all of which are part of a theory of syntax.
The early work in generative grammar focuses on the form and function of grammars
as the central topic for developing a general theory of linguistic structure. The other
goal of constructing generative grammars of individual languages has never been fully
realized. Although LSLT itself covers a substantial portion of English syntax, it was not
intended as a complete survey. Matthews 1965 provides one of the earliest efforts to
apply generative grammar to a non-Indo-European language. See Stockwell et al. 1973
for an attempt to survey the major syntactic structures of English. In effect the early
work on transformational grammar shifted the focus of inquiry from the external forms
of languages to formal grammar and the theory of linguistic structure that accounts for
these forms.

2 Generative Grammar as a Theory of Knowledge

The next major development in the study of generative grammar reinforces this shift in
focus. The first chapter of Chomsky 1965, written in 1958–9, interprets grammar as a
model of linguistic knowledge in the mind of a speaker, consisting of a finite number of
grammatical rules, including a finite lexicon. The sentences of a language are derived
from the internal grammar and have no independent physical existence apart from
it. In this way, language becomes a derivative notion, in contrast to grammar. Over
two decades later, Chomsky 1986b: chapter 2 clarifies this in the distinction between
I-language (I for internal (and in Chomsky 1995b also for individual and intentional))
vs. E-language (E for externalized). An I-language is simply the grammar in the mind
of the speaker (i.e. a computational system and a lexicon that constitutes the speak-
er’s linguistic knowledge). In contrast, an E-language is some notion of language that
makes no direct reference to the mind of the speaker (for example. a set of sentences).
An I-language is a finite object capable of infinite output (realizing von Humboldt’s
characterization of “making infinite use of finite means,”; see Chomsky 1964), in con-
trast to the (infinite) set of sentences that is said to comprise an E-language.
Taking a language to be a system of knowledge in the mind of the speaker raises four
basic questions for the study of language (see Chomsky 1988).

900
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

(8) a. What constitutes this system of knowledge?

b. How does this system arise in the mind of a speaker?
c. How is this system put to use (in speech and writing)?
d. What mechanisms in the brain serve as the material basis for this system
of knowledge and for its use?

The first question is fundamental given that answers to (8b–d) depend on the answer to
(8a). In the psychological framework of (8) there is a clear separation between linguis-
tic knowledge (competence) and linguistic behavior (performance)—see chapter one of
Chomsky 1965 for discussion. While linguistic knowledge determines linguistic behav-
ior to a large extent, the latter may be affected by “such grammatically irrelevant con-
ditions as memory limitations, distractions, shifts of attention and interest, and errors
(random or characteristic)” (Chomsky 1965: 3). These effects fall under a theory of
linguistic processing, ultimately leading to the ongoing debate about what effects result
from the grammar vs. the processor (see Phillips 2010).
The answer to (8a) involves a lexicon and the grammatical mechanisms (i.e. phrase
structure rules and transformations) that construct linguistic expressions from the items
it contains. In addition the system of knowledge must include constraints on the appli-
cation and output of these mechanisms—for example, the A-over-A Principle and con-
ditions on derived constituent structure.
How such a system is acquired by a child (first language acquisition) actually involves
two how-questions: “how is it possible?” and “how is it done step by step?” The first,
which constitutes the logical problem of language acquisition, concerns the fact that the
linguistic system acquired cannot be accounted for solely on the basis of the evidence
available from the child’s environment. Thus at the center of this problem lies the pov-
erty of the stimulus, one of the core problems in epistemology beginning with Plato’s
Meno.
Chomsky’s solution is to posit an innate—hence part of the human genome and
thus universal across the species—language faculty (FL) that determines the system of
knowledge acquired on the basis of the impoverished evidence provided by the environ-
ment. The resulting theory of Universal Grammar (UG) resurrects Descartes’ theory of
innate ideas in another form.
First language acquisition starts with the mind of the child in an initial state (i.e.
UG) which on presentation of primary language data (PLD) changes to a steady state,
the system of knowledge for the language the child comes to speak. This process is
schematized in (9), where GL is a grammar of the language in which PLD is the limited
sample the child is exposed to.

(9) UG + PLD oGL

Under this framework, there are two foci for linguistic research, the content of UG and
the grammars that humans acquire. UG constitutes a theory of language (as opposed
to individual languages) and provides an explanatory basis for language acquisition.
In contrast, GL minus the contribution of UG merely describes the content of the lin-
guistic knowledge acquired and therefore may raise problems for learnability if pov-
erty of the stimulus problems remain. Ideally the contribution of UG is maximal and
the remainder can be accounted for on the basis of linguistic data provided by the
environment.

901
 RO BERT F R E I D I N

Perhaps the key factor involving poverty of the stimulus for language acquisition
concerns a speaker’s ability to distinguish deviant utterances from legitimate utterances.
Thus English speakers will judge (6b) to be deviant, in contrast to the legitimate (10).

(10) To whom did your interest in Fred seem rather strange?

Given that speakers are not told to avoid sentences like (6b), their knowledge that (6b)
is deviant does not apparently come from their experience. Another aspect of poverty
of the stimulus arises from the fact that the PLD is finite and quite limited in contrast to
the unbounded nature of the language acquired.
This unbounded character of language is a core property of language use, providing
a basis for its innovative nature; thus much of what we say and hear consists of novel
utterances, new to our experience and possibly to the history of the language. Language
use is also free from stimulus control and moreover it is coherent and appropriate to the
situation. This bundle of properties constitutes what Chomsky calls the creative aspect of
language use, a topic first discussed in Descartes’ Discourse on Method (part V) over three
hundred years ago. (See also Chomsky 1988, 1994 for further discussion, including how
this relates to the mind/body problem.) The properties of coherence and appropriate-
ness remain a serious problem for a theory of language use (8c) in terms of both percep-
tion and production, which ultimately must involve the relation between language and
thought, a relation that can only be described as mysterious given the fact that there is
no formal theory of thought.
The answer to (8d) concerns a future neuroscience of language that is barely on
the horizon. The major overwhelming problem for this enterprise is unification, which
cannot happen when the core concepts and vocabularies of generative grammar and
neuroscience remain disjoint (Poeppel and Embick 2005). Nonetheless, without doubt
the brain provides the material basis for FL (including UG) as well as the I-language
acquired by a speaker. In this framework both the initial state (UG) and the steady state
that develops in the mind of the speaker (an I-language) constitute physical objects
in the world, part of human biology. From this biolinguistic perspective, the goals and
methodology of generative grammar align with those of the natural sciences in the
pursuit of formal theories with a rich deductive structure (see Chomsky 1995c; Jenkins
2000).

3 From Rules to Principles and Parameters

At the outset, work on generative grammar focused primarily on the formulation of
grammatical rules and their interaction in derivations. Much of this work was car-
ried out in studies that attempted to account for the properties of constructions (for
example passives and interrogatives) in various languages. However, it was clear that
descriptive devices, both phrase structure rules and transformations, needed to be con-
strained in some principled way if the explanatory burden of the logical problem of
language acquisition was to be addressed. For this reason, work on generative gram-
mar was also concerned with placing constraints on the form and function of gram-
matical rules. See LSLT for the earliest proposals concerning phrase structure rules and
transformations.
An important step toward constraining the descriptive power of grammatical rules
was the separation of the lexicon from phrase structure rules (Chomsky 1965), in effect

902
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

separating the lexicon of a language from the mechanisms that assigned structural
descriptions to linguistic expressions, what is now called the computational system of a
grammar. This step eliminated substantial redundancy between the intrinsic properties
of lexical items and phrase structure rules. This separation allowed for a reduction in the
descriptive power of the phrase structure rule component by eliminating all context-
sensitive phrase structure rules. The context-sensitive properties of lexical items (for
example intransitive verbs could not occur in structures with NP objects) were handled
by a separate lexical insertion operation, a substitution transformation, that matched
the contextual requirements of a lexical item with the phrase-marker into which it was
being inserted.
Another significant step involved the elimination of the class of generalized trans-
formations in favor of phrase structure rules with clausal recursion. (See Chomsky 1965
for the original proposal and Freidin 1978, 1999; Lasnik 2004 for further discussion.)
The application of transformations to generalized phrase-markers, now the output of
phrase structure rules and lexical insertion, was constrained by the syntactic cycle,
a rule-ordering principle that requires transformations to apply to the most deeply
embedded cyclic domain first and then the smallest cyclic domain containing it, and
so on to the root domain. Note that the syntactic cycle constitutes the first bottom-up
principle of syntactic theory, in contrast to the top-down analysis imposed by phrase
structure rules.
Another major development in the direction of the bottom-up analysis of phrase
structure arrives in the form of X-bar theory (Chomsky 1970), where the relation
between lexical categories and phrasal categories (for example N and NP) is viewed as
projection of a syntactic category label of a lexical item onto the phrasal constituent
formed from that lexical item and some other constituent. Note, however, that for at
least another decade, the X-bar analysis continued to be formulated in terms of top-
down phrase structure rules.
X-bar theory develops out of the discussion of the analysis of nominalizations like
(11); see Chomsky 1968.

(11) a. the enemy’s destruction of the city

b. the city’s destruction by the enemy
c. the destruction of the city by the enemy

Chomsky (1970) argues against an analysis that derives (11a) from a corresponding
sentence (12a) in favor of a general constraint on transformations that prohibits them
from changing the syntactic categories of lexical items.

(12) a. The enemy destroyed the city.

b. The city was destroyed by the enemy.

Therefore, the examples in (11) must all be derived independently of the sentences
in (12). The structural description of the NP (11a) would involve two levels of phrasal
projection, originally differentiated in terms of bars, one for the first level and two for
the second (given in (13) as primes).

(13) [N" the enemy’s] [N' destruction [PP of the city]]]

903
 RO BERT F R E I D I N

The head of this NP, destruction, takes the city as its object in the same way that the verb
destroy takes the same NP as its object.
This parallelism between nominalization and clause extends to the passive sentence
(12b) and its corresponding nominal (11b). Assuming that (11b) must be transforma-
tionally derived (as is (12b)) required a more general formulation of the “passive trans-
formation.” Chomsky 1970 splits the Syntactic Structures formulation (3) into two parts,
a rule that postposes the underlying subject NP and another that preposes the underly-
ing object NP—eliminating the lexical insertion operations of the passive auxiliary,
which of course doesn’t exist in nominals, and the passive by. Each part involves a single
movement between two NP positions. In Chomsky 1976 such movements fall under a
single rule (14).

(14) Move NP

Thus the complicated language-specific and construction-specific passive transforma-

tion of Syntactic Structures, which involves four distinct operations, is reduced to a single
elementary operation that is neither language-specific nor construction specific. (See
Freidin 1995 for a more detailed discussion of the history of the passive transformation.)
Moreover, (14) generalizes to all NP movement constructions—inter-clausal raising as
well as intra-clausal passive, for example subject raising cases as in (15), where Mary is
interpreted as the subject of like but is pronounced as the subject of the main clause.

(15) Mary seems to like John.

The general nature of (14) makes it part of the resources of UG, rather than a language-
particular rule of grammar. Later developments involve further generalizations, where
movement rules are cast as single operation “Move D” (Chomsky 1981a). (See also Las-
nik & Saito 1984, 1992 where movement, deletion, and insertion operations fall under
a single formulation as “Affect D.”)
Highly articulated structural descriptions for transformations (for example in (3))
limit the application of the elementary operations involved. Therefore when structural
descriptions are minimal, as in (14), which mentions only the elementary operation
and the phrase it applies to, other constraints (for example the A-over-A Principle)
are necessary to stop the rule from misgenerating deviant sentences. Work on condi-
tions on transformations is expanded in Ross’s famous 1967 dissertation, which argues
for replacing the A-over-A Principle with a set of constraints with a greater empirical
range, proposing that certain syntactic configurations constitute islands from which
movement is prohibited, primarily wh-movement. This conditions framework is fur-
ther developed in Chomsky 1973 with the additions of constraints that restrict NP
movement—among others including a refinement of the syntactic cycle (the Strict
Cycle Condition (SCC)) and the Subjacency Condition, which subsumes some of the
empirical effects of some of Ross’s island constraints (the Complex NP Constraint and
the Left Branch Condition) and forces movement to occur in small local increments
successively cyclically rather than in a single long-distance step. The crosslinguistic gen-
erality of this conditions framework is demonstrated in Kayne (1975) on the syntax of
French clitic pronouns, Quicoli 1976b on the syntax of French quantifiers, and Quicoli
1976a on the syntax of Portuguese clitic pronouns. This work shows how Chomsky’s
constraints on NP movement in English generalize to constructions in other languages

904
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

that do not exist in English. These analyses illustrate how linguistic variation involving
different constructions actually conforms to the same general principles of grammar, in
this case constraints on rule application.
Another important advance in the development of generative grammar comes shortly
after with the discovery that the formulation of certain conditions may vary within a
restricted range from language to language. Rizzi 1980 observes that in Italian relative
clauses the relative pronoun may be extracted out of an indirect wh-question in certain
constructions, but not in others—in contrast to English where indirect wh-questions
constitute syntactic islands. Rizzi proposes that for Italian the Subjacency Condition is
formulated with a different bounding category S’ (= CP), vs. S (= TP) in English. The
two values are construed as parameters for the formulation of the condition and the
difference between relative clauses in Italian and English is analyzed as an instance of
parametric variation. (See also Lasnik 1991 and Freidin 1992: chapters 7–8 for discus-
sion of parameters for the principles of binding theory.) The notion of parameters is
extended to other instances of crosslinguistic variation, for example the occurrence of
null subject pronouns in indicative clauses (yes in Spanish; no in French and English)
and the placement of interrogative phrases at the beginning of direct and indirect ques-
tions (yes in English, Spanish, French; no in Chinese and Japanese). For discussion of
how parameters account for crosslinguistic variation see Chomsky 1981b; Kayne 2000,
2008; Baker 2001. See also Pollock 1989 for a parametric analysis of the verbal mor-
phology systems of French vs. English, which has an important bearing on the syntactic
analysis of clauses.
Parametric variation appears to be limited to a narrow range of choices, specified
as a part of UG and thus an essential component of a theory of language design that
accounts for crosslinguistic variation and complements conditions on rules of gram-
mar, which account for crosslinguistic uniformity of language design. Taking general
conditions on rules to be principles of grammar, the addition of parameters to UG
creates a Principles and Parameters framework (henceforth P&P), the prevailing
framework in generative grammar since 1979. This framework constitutes a significant
departure from earlier work by replacing construction-specific and language-specific
rules of grammar with general rules (for example Move D) that make no reference to
either syntactic constructions or particular languages. In effect, grammatical opera-
tions of the computational system function as elements of UG. This perspective results
in another shift in focus from grammars of particular languages to the contents and
function of UG.
In another important development in the conditions framework that predates P&P,
Chomsky 1973 and 1976 demonstrate how conditions on NP movement generalize
to rules of semantic interpretation involving anaphor-antecedent relations. The role
of semantics in generative grammar has been an important and controversial topic
from the outset. A key issue is how syntactic structure supports semantic interpreta-
tion. This translates into a question about which syntactic levels of representation are
relevant for interpretation. For example in passives, where the surface subject of the
clause is interpreted as the logical object of the verb, the underlying structure, in which
the surface subject occurs as the verbal object, is obviously relevant to interpretation.
Chomsky 1957 also notes that the interpretation of quantifiers appears to depend on
their linear surface order (see also Chomsky 1965: chapter 3, footnote 9) and that
therefore surface structure may also play a crucial role in interpretation. (For discussion
of this model and the issues concerning semantic interpretation see Chomsky 1972;

905
 RO BERT F R E I D I N

Jackendoff 1972.) An alternative model is proposed in the late 1960s in which signifi-
cantly more abstract initial underlying syntactic structures are claimed to be semantic
representations. (For discussion of this model of generative semantics, see McCawley
1968, Lakoff 1971; for a critique, see Chomsky 1972, especially chapter 3; and for a his-
tory of this period see Newmeyer 1986.) A third model arises in the mid-1970s in which
the output of the transformational component serves as the sole level of representation
for semantic interpretation. This involves a theory of movement operations in which a
moved constituent leaves behind an empty constituent as a trace in the position from
which it is moved (Chomsky 1973, Fiengo 1974). For example, the analysis of an inter-
clausal raising construction like (15) would be (16), where the coindexed t marks the
position of the trace of Mary.

(16) [TP Maryi [VP seems [TP ti to [VP like John]]]]

With traces, the semantic function of moved constituents (for example, the surface sub-
ject of passives) can be recovered from the output of the transformational component.
The relation between a trace and its antecedent was initially interpreted as a form
of binding on a par with the relationship between a lexical anaphor (for example, a
reflexive pronoun) and its antecedent. This created the option to interpret constraints
on NP movement as conditions on representations (i.e., on trace binding) rather than
constraints on the application of rules. Freidin 1978 demonstrates that under the former
interpretation, the empirical effects of the cyclic principle (i.e., the SCC) can be derived
from other independently motivated general principles, thereby eliminating the need
for stipulating the SCC as a principle of grammar. This result shows the cyclic nature
of syntactic derivations to be deeply embedded in the theory of grammar. The syntactic
cycle remains an important topic in generative grammar—see Freidin 1999 and Lasnik
2004 for further discussion of its rich and intricate history.
The model of grammar (17) that develops under trace theory identifies four distinct
levels of representation, the two levels in (7) and two levels that interface with other
cognitive components: Phonetic Form (PF) with sensory-motor functions and Logical
Form (LF) with conceptual-intentional functions, (17), see below.
LF is distinct from S-structure to the extent that there are additional operations that
map the latter onto the former, creating covert syntactic structures necessary for
semantic interpretation—for example, a rule of quantifier raising (May 1977; 1985),
which creates quantifier/variable structures crucial in the interpretation of scope. Thus
in this model the interface level of LF alone provides the syntactic basis for semantic
interpretation.

(17) D-structure

S-structure

PF LF

906
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

In addition to constraints on the locality of movement operations and anaphor

binding, constraints on the output of the transformational component have also been
proposed, beginning with David Perlmutter’s surface structure constraints in his 1968
M.I.T. dissertation (see Perlmutter 1971). This device is developed more extensively
for the analysis of English in Chomsky & Lasnik 1977, which concerns an account of
the distribution overt vs. phonetically null subjects in infinitival clauses among other
topics. In a 1977 letter to Chomsky and Lasnik regarding their manuscript, Jean-Roger
Vergnaud proposes an alternative to their surface structure filter involving notions of
case and syntactic government (see Vergnaud 2008). In Chomsky 1980 (written in
1978) this proposal is formulated as the Case Filter, which prohibits the occurrence of
a NP with phonetic features in a position that is not licensed for structural Case (for
example nominative Case is licensed for the subject of finite clauses). This analysis is
based on the universality of structural Case, regardless of whether it occurs overtly (mor-
phologically) on all NPs (which it does not in English).
The Case Filter provides an explanation for the obligatory nature of NP movements.
For example, the failure of the inter-clausal movement in (15) would yield the devi-
ant (18), where Mary violates the Case Filter because it is a NP with phonetic content
occurring in a syntactic position that is not licensed for Case.

(18) *It seems Mary to like John.

This analysis generalizes to NP movement in passive constructions as well under the

assumption that passive predicates do not license Case on their nominal objects. In this
way the syntactic behavior of Move NP is controlled by an output condition on repre-
sentations rather than a constraint on operation of the rule itself.
The Case Filter has also been used account for certain facts concerning the linear
order of constituents, which is merely stipulated in phrase structure rules. For example,
if the adjacency of a V and its object NP is required for Case licensing, then the fact
that in a VP containing an NP object and a clausal argument (CP), the linear order
must be V-NP-CP and not V-CP-NP follows from the Case Filter (see Stowell 1981:
chapter 3 for detailed discussion). Therefore a phrase structure rule that stipulates this
order involves a significant redundancy. Note that the phrase structure rule would itself
be language-specific, tied to a particular grammar, whereas the Case Filter is part of UG.
Assuming this to be generally the case, Chomsky proposed in a class lecture (September
1979) that there are no phrase structure rules (see again Stowell 1981 for a detailed
analysis of the redundancy between phrase structure rules and principles of UG). It is
a curious fact about the history of generative grammar that although phrase structure
rules were dropped from the P&P framework in 1979, no alternative formal mechanism
for the construction of phrase structure was generally adopted until the proposal of an
operation Merge in Chomsky 1995a (see §4 for details).
The development of Case theory (see especially Lasnik 2008) with its core principle
the Case Filter was instrumental in the development of the P&P framework, leading to
a modular theory of grammar in which rules and principles of grammar are organized
into interacting modules. Chomsky (1981a: 135) distinguishes two classes of subsystems
for the theory of grammar, as shown in (19), one for the lexicon and the rules that
create from it structural descriptions of linguistic expressions at the various levels of
representation (19a), and another for the principles that constrain the operation and
output of the rules (19b).

907
 RO BERT F R E I D I N

(19) a. i. lexicon
ii. syntax
(1) categorial component
(2) transformational component
iii. PF-component
iv. LF-component
b. i. bounding theory
ii. government theory
iii. T-theory
iv. binding theory
v. Case theory
vi. control theory

Bounding theory is concerned with the locality of movement operations for which the
Subjacency Condition is a core principle. T-theory covers predicate/argument structure,
governed by the T-Criterion (Chomsky 1981a), which requires in part that every argu-
ment in a linguistic expression must be assigned an argument function by a predicate.
Government theory employs the concept of syntactic government to further constrain
movement operations by requiring that in addition to satisfying the Subjacency Condi-
tion, empty categories such as traces must be properly governed (the Empty Category
Principle (ECP)). Chomsky 1981a attempts to unify the subcomponents of (17b) by
formulating many of the basic principles in terms of syntactic government, yielding the
so-called Government and Binding (or GB) Theory. See also Chomsky 1986a for an
attempt to reformulate the Subjacency Condition in terms of government.
Although a decade of research in GB theory produced an explosion of new data
from a variety of languages, inaugurating a rebirth of comparative grammar, the concept
of syntactic government—the heart of the theory—proved to be problematic and was
eventually dropped in the early 1990s (see Chomsky 1993). The main weakness resided
in the complex formulations of the relations “govern” and “properly govern.” Various
formulations of the ECP were so complex that they mirrored the complexity of the data
they were supposed to account for. Nonetheless, GB yielded major developments in the
P&P framework, which remains a major framework in generative grammar today.

4 Economy and the Minimalist Program

Notions of simplicity and economy for syntactic theory had been discussed in the first
formulation of generative grammar (see Chomsky 1979 and for discussion, Freidin 1995;
Freidin & Lasnik 2011). In the mid-1980s the notion of economy resurfaces explicitly
in the form of a principle of Full Interpretation (henceforth FI, first proposed in Chom-
sky 1986b), which requires that every element in PF and LF representations receive
an appropriate interpretation. In 1991 Chomsky construes FI as a prohibition against
superfluous symbols in the two interface representations. An argument (NP or clause)
in a syntactic representation that is not assigned a semantic function is uninterpret-
able, hence superfluous and therefore in violation of FI. Thus FI subsumes part of the
T-Criterion.
The inclusion of FI as a part of UG has other far-reaching consequences and therefore
is more than just a restatement of part of the T-Criterion. FI also rules out sentences
containing any superfluous lexical items—verbs, prepositions, and adjectives as well

908
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

as nouns. Under the assumption that a quantifier in natural language to be interpret-

able must bind a variable, FI requires some form of quantifier raising, where the trace
of the raised quantifier functions as a variable (May 1985). This generalizes to wh-
movement in questions where the wh-phrase functions at LF as a quasi-quantifier
(Chomsky 1976) and more generally to other syntactic operators like relative pronouns
(Chomsky 1977).
FI also determines the nature of syntactic derivations. Given that phonetic features
are irrelevant for the interpretation of LF and correspondingly that semantic features
are likewise irrelevant for the interpretation of PF, FI requires that there be a point in
a derivation where the two sets of features are separated, called Spell-Out (Chomsky
1993). Spell-Out, in contrast to S-structure in (17), is not a level of representation
because there are no conditions that define it, nor could there be if derivations involve
multiple points of Spell-Out as proposed in Uriagereka (1999) and Chomsky (2001;
2008).
In addition to the single economy condition on representations, FI, there are also
economy conditions on derivations concerning the nature and function of movement
operations. One involves the “last resort” character of the major movement operations
for wh-phrases and non-wh-NPs, named Last Resort (Chomsky 1986b, 1995b; Las-
nik 1995). In the case of wh-phrases movement is required by FI to create operator-
variable constructions at LF. Thus movement happens because it must. The same holds
for non-wh-NP movement, as in simple passives (for example (4b) and (12b)) and rais-
ing constructions (for example (15)), as required by the Case Filter, possibly FI under
the appropriate analysis. For example, Last Resort would block interclausal NP move-
ment out of finite clause, as in the deviant (20a) in contrast to the legitimate (20b),
where the subordinate clause subject does not move.

(20) a. *[TP Maryi [VP seems [TP ti [VP likes John]]]]

b. It seems Mary likes John.

Since Mary is licensed for Case as the subject of the subordinate clause, the NP has no
reason to move. This analysis generalizes to nonsubject NPs as illustrated in (21), where
John is interpreted as the object of like.

(21) a. *John seems Mary to like.

b. *John seems Mary likes.

Another economy condition on derivations concerns the fact that syntactic movement
seems to involve shortest steps (as in successive cyclic movement), a requirement that
has a “least effort” flavor. One formulation, the Minimal Link Condition (MLC), con-
strains the links of a chain created by traces and their lexical antecedent (Chomsky
1995a, 1995b, 2001, 2008; Boeckx 2003).
The reconsideration of the various modules of GB theory in terms of notions of econ-
omy leads not only to a simplification of some parts of GB but, more importantly, to the
formulation of a minimalist program (henceforth MP) for linguistic theory (Chomsky
1993, 1995b; for a discussion of the roots of the MP, see Freidin & Lasnik 2011, and
for discussion of evolution of the MP within the generative enterprise as a part of the
scientific enterprise, see Freidin & Vergnaud 2001). The MP attempts to address two
fundamental questions (22), first articulated in Chomsky 1998.

909
 RO BERT F R E I D I N

(22) a. To what extent is the computational system for human language

optimal?
b. To what extent is human language a perfect system?

A major task for the MP is to provide explicit interpretations for the terms “optimal”
and “perfect.” Note also that the two questions are interrelated in that an optimal com-
putational system for human language would almost certainly be a prerequisite if human
language were actually a perfect system.
The P&P framework, which serves as a foundation for the MP, is itself based on sev-
eral fundamental assumptions that suggest that the answers to both questions could be
“to a significant extent.” Among the core assumptions is the definition of a language
(i.e. an I-language) as consisting of a lexicon and a computational system, where the lat-
ter is largely determined by UG—i.e. the grammatical operations that construct struc-
tural representations of linguistic expressions and the constraints on their operation and
output. Given that knowledge of a lexicon also involves poverty of stimulus issues, it
appears that some part of the lexicon is, like syntactic structure, determined by proper-
ties of UG (see Chomsky 2000: chapter 2), possibly including parameters (see Borer
1984). From this perspective, given UG a selection S of parameters determines a lan-
guage. Possibly the sole task of language acquisition is acquiring a lexicon, which results
in fixing the value of parameters. If the locus of linguistic variation rests solely in the
lexicon (for example) the phonetic labels of lexical items (what Chomsky calls “Saus-
surean arbitariness”) and parameters, then there may be no variation in overt syntax or
the LF component. This leads to a striking follow-up conjecture that “there is only one
computational system and one lexicon” (Chomsky 1993: 3), apart from idiosyncratic
lexical properties. This conjecture remains controversial.
The initial formulation of the MP in Chomsky 1993 articulates some additional
assumptions that are unique to the MP. Foremost among them, and the one that essen-
tially underpins the rest, is that the theory of grammar must meet a criterion of concep-
tual necessity. In essence the assumption constitutes a version of Ockham’s razor (the
law of parsimony), a methodological consideration that has informed modern science
(including modern generative grammar) for six centuries. However, the MP applies this
more aggressively and rigorously than previous work in the field, yielding a substan-
tive hypothesis that the only conceptually necessary linguistic levels of representation
are the two interface levels PF and LF. Thus the MP discards D-structure and S-
structure, the two levels of representation internal to derivations, and attempts to reana-
lyze the empirical evidence that was employed to motivate them. Without internal lev-
els, the only output conditions possible apply to the two interface levels. Each linguistic
expression is a realization of interface conditions. Furthermore, if the output conditions
imposed on LF and PF representations are determined by the cognitive components
that interface with them, then language would constitute “an optimal way to link sound
and meaning”—designated as the Strong Minimalist Thesis (SMT). “If SMT held fully,
which no one expects, UG would be restricted to properties imposed by interface con-
ditions. A primary task of the MP is to clarify the notions that enter into SMT and to
determine how closely the ideal can be approached” (Chomsky 2008: 135).
The SMT, if correct, is one aspect of language design that points to the optimal nature
of the computational system. Another concerns what grammatical operations the system
contains. Under P&P, the transformational component has been reduced to a small set
of elementary operations, including “Move D” a deletion operation that, in addition to

910
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

eliminating non-trivial chains at PF (see discussion of (28) below), would account for
ellipsis phenomena as illustrated in (23) where the phonetic features of the VP written
about Descartes are deleted in the second conjunct (as indicated by the strikethrough).

(23) Jane has written about Descartes and Bill has too.

This formulation of the transformational component also supports the hypothesis for
the optimal nature of the computational system.
With the elimination of phrase structure rules (see above), the question of what gram-
matical mechanisms replace these rules becomes crucial for the MP. The answer is pro-
vided in Chomsky 1995a, formulated as the operation Merge, which joins two syntactic
objects (lexical items and/or objects formed from lexical items) and projects the syntactic
label (the syntactic category feature) of one of them as the label of the new syntactic
object created. Merge is a recursive operation (i.e. it applies to its own output) that builds
phrase structure bottom-up. Thus consider the derivation under Merge for (24).

(24) Our group will leave early.

The two lexical items leave and early merge to form a VP, which in turn is merged with
the modal will to form a T(ense) Phrase (TP). Independently, our and group are merged
to form a NP. Finally, this NP and T phrase will leave early are merged to form a TP—i.e.
the syntactic category T is projected as the label of clause formed. This last step merges
two independent phrases in essentially the same way that generalized transformations
operated in the earliest transformational grammars. In effect, a general transformational
operation replaces a set of less general and more complicated phrase structure rules,
including those rules that are reduced to their most general schema under X-bar theory.
What this shows is that a pure transformational grammar (i.e. without phrase structure
rules) constitutes a best solution for the construction of phrase structure. It is a sur-
prising development given the abandonment of generalized transformations in favor of
phrase structure rules that occurred in 1965.
If a simple operation like Merge constitutes the only grammatical mechanism for
building basic phrase structure, then it also supports the optimality hypothesis the MP
seeks to establish. Note too that constraining Merge also contributes to the hypothesis.
While Merge applies bottom-up, unconstrained it could presumably insert a syntactic
object inside an existing syntactic object. Such derivations are prohibited by the No
Tampering Condition (NTC) of Chomsky 2007, 2008, formulated in (25).

(25) Merge of X and Y leaves the two S[yntactic] O[bjects] unchanged.

(Chomsky 2008:138)

The NTC constitutes a general constraint on derived constituent structure and at the
same time enforces the cyclic application of Merge by allowing only merger at the root
for all syntactic objects. Moreover, the NTC provides a natural constraint on efficient
computation, another contribution to the optimality hypothesis.
If the NTC imposes cyclic computation in syntax, then it applies not only to the
single generalized transformation Merge but also to the single singulary transforma-
tion Move D. This holds without further stipulations if Move D is actually a form of
Merge, referred to as Internal Merge (IM) to distinguish it from External Merge (EM),

911
 RO BERT F R E I D I N

which operates on two independent syntactic objects. Thus consider the application of
Move D that is involved in generating the passive construction (26).

(26) The review was written by two professors.

Given that the NP the review is first merged as the object of written, Merge constructs
(27) prior to the application of the movement operation.

(27) [T was [VP [V written [NP the review]] [PP by [NP two professors]]]

To generate (26) from (27), all that is required is that a copy of the NP the review be
merged with (27), yielding (28).

(28) [TP [NP the review] [T was [VP [V written [NP the review]] [PP by [NP two
professors]]]]

At PF the copy of the review in VP is deleted, thus linearizing the nontrivial chain in
(28) (see Nunes 2004 for details). Under this analysis, there is a single operation Merge
that accounts for all of phrase structure, including the contribution of singulary trans-
formations. This reduction of phrase structure rules and movement transformations to
a single recursive operation provides a solid foundation for the optimality hypothesis of
the MP.
It is worth noting that the NTC in conjunction with this analysis of Merge makes the
generation of cannonical D-structure impossible. Consider the derivation of (29).

(29) They claimed that the review was written by two professors.

The NTC requires that the copy of the review that occurs as the subject of the subordi-
nate clause be merged before the complementizer that is merged with that clause. Thus
there is no way of segregating IM operations from EM operations and applying the
former as a block after the latter have applied. In this way, an optimal formulation of
the computational system supports the minimalist assumption that the language faculty
involves only interface representations, providing some justification for the rigorous
application of a criterion of conceptual necessity.
Given the plausibility of the optimality hypothesis, there is reason to expect that
the hypothesis that language constitutes a perfect system can be explicated in some
substantive way. One of the earliest proposals is that the computational system utilizes
only the features found in lexical items and therefore “interface levels consist of nothing
more than arrangements of lexical features,” what is called the Inclusiveness Condition
(IC) (Chomsky 1995b: 225 –but see page 228 and footnote 10 for comments on how
inclusiveness might be violated by PF representations). It follows from this condition
that interface representations will not contain bar levels on syntactic categories as in X-
bar theory (cf. (13)), nor will labels distinguish maximal phrasal projections (XP) from
heads (X) or their intermediate phrasal projections. In addition the IC prohibits the
use of indices in representations (cf. (16)), which significantly affects the formulation
of binding theory (Chomsky 1993; Freidin 1997), and also the postulation of traces as
special empty categories (see again (16)). In effect, the IC enforces the copy theory of
IM, where non-trivial chains consist of multiple copies of a phrase. Note that the IC is

912
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

actually the null hypothesis for how the lexicon interacts with the computational sys-
tem to produce structural representations of linguistic expressions. Likewise, the NTC
also functions as a null hypothesis on the output of Merge.
A computational system that contains Merge (i.e. both EM and IM) governed by
the NTC and the Inclusiveness Condition constitutes a highly constrained theory of
derived constituent structure, ultimately interface representations. To the extent that
the functioning of this computational system is largely determined by interface condi-
tions imposed by the cognitive components that interact with the language faculty—i.e.
the components which operate on the interface representations PF and LF and thereby
make language usable, it could be that I-language is a perfect solution to interface con-
ditions. However, the locality conditions on IM, going back to the A-over-A Principle
and Ross’s island constraints, remain to be accounted for. From an economy perspective
it appears that IM involves some requirement for “shortest move” which might translate
into a “minimal link” constraint on chains (see references on p. 909). If this constraint
applies to derivations, specifically the application of IM, then it is part of the compu-
tational system, necessarily a principle of UG. Alternatively, this requirement could
be formulated as a condition on representations (see Freidin 1978 for a formulation
of Subjacency as a condition on trace binding), in which case it would be possible for
the locality of IM, which requires successive cyclic movement, to follow from interface
conditions. Exactly which analysis is correct remains to be determined.
From the discussion above, it should be clear that the development of the MP has
resulted in shrinking the content of UG. To the extent that interface conditions are
legibility requirements imposed on the sole interface representations PF and LF by the
cognitive components that connect with linguistic representations, they are not part
of UG. This leaves the operations Merge and Delete, and conditions on their applica-
tion—including the NTC, the IC, and some version of the MLC, along with a condi-
tion on the recoverability of deletions (Chomsky 1965)—as parts of UG.
This development has a significant impact on the potential answer to the prime
biolinguistic question: Why is it that humans acquire natural languages and other
species do not? Because the shrinking of UG highlights the computational core of
language—i.e. Merge, the recursive procedure that accounts for linguistic structure,
it could be that recursion is the answer to the biolinguistic question (see Hauser et
al. 2002). This answer undoubtedly bears on the question of the evolutionary origins
of language as a system of knowledge, however it remains controversial—see Pinker
& Jackendoff (2005), Fitch et al. 2005, Jackendoff & Pinker 2005, and Larson et al.
2010 for discussion.
The shrinking of UG also raises the question of how much (or how little) needs to
be postulated in UG to account for language design. This question arises in a more
highly articulated framework than was available in earlier work where the dichotomy
of nurture vs. nature in language acquisition was generally conceived of as a two-way
split between the external language environment (PLD) and the innate—hence inter-
nal—UG that largely determined the grammar acquired. More recent discussions (see
Chomsky 2005, 2007) have raised the possibility that a third factor—namely, princi-
ples not specific to the language faculty—could also play a role. Although the possibil-
ity of third-factor explanations for language is briefly mentioned in Chomsky 1965, it is
only with the development of the MP, especially its concern for principles of efficient
computation, that third-factor accounts become plausible. Consider what Chomsky
wrote:

913
 RO BERT F R E I D I N

there is surely-no reason today for taking seriously a position that attributes a
complex human achievement entirely to months (or at most years) of experi-
ence, rather than to millions of years of evolution or to principles of neural
organization that may be even more deeply grounded in physical law—a posi-
tion that would, furthermore, yield the conclusion that man is, apparently,
unique among animals in the way in which he acquires knowledge.
(p. 59)

Principles of neural organization would not be specific to language, hence a third fac-
tor. If such principles follow from physical law applied to neurons, then an account of
language design based on them would almost certainly involve optimal computation in
a perfect system.
This perspective that has developed under the MP is highly speculative. There is still
a considerable way to go in understanding how computation in human language is both
efficient and optimal, how the governing principles for language constitute principles of
efficient computation, how such principles relate to as yet unknown principles of neural
organization, and finally how these latter principles “may be even more deeply grounded
in physical law.” (Note that the idea that some properties of biological systems can be
explained by physical law originates in the work of d’Arcy Thompson (1942)—see Frei-
din & Vergnaud 2001 and Freidin & Lasnik 2011 for further discussion of this idea as it
relates to generative grammar). While it is plausible that answers to the first two parts
can be achieved, answers to the last two parts would require a double unification of lin-
guistics and neuroscience on the one hand (see Poeppel & Embick 2005 for a discussion
of the problems this involves) and neuroscience and physics on the other.
The study of modern generative grammar over the past sixty plus years has yielded an
understanding of the nature of human language, its universal character, and the ways
in which languages can vary, that is far beyond what preceded during several thousand
years. The MP moves the field toward a deeper understanding by raising more challeng-
ing questions while at the same time imposing more rigorous standards on what might
count as an answer.

Note
I am indebted to Howard Lasnik for insightful comments on the previous draft of this chapter.

References
Baker, M. 2001. Atoms of language, New York, NY, Basic Books.
Boeckx, C. 2003. Islands and chains: Resumption as stranding, Amsterdam, John Benjamins.
Borer, H. 1984. Parametric syntax, Dordrecht, Foris.
Chomsky, N. 1957. Syntactic structures, The Hague, Mouton.
Chomsky, N. 1964. The logical basis of linguistic theory. In: Lunt, H. (ed.) Proceedings of the Ninth Interna-
tional Congress of Linguistics. The Hague: Mouton.
Chomsky, N. 1965. Aspects of the theory of syntax, Cambridge, MA: MIT Press.
Chomsky, N. 1970. Remarks on nominalization. Indiana University Linguistic Club Publications, reprinted
in: Jacobs, R. A. & Rosenbaum, P. S. (eds.) Readings in English transformational grammar. Waltham, MA:
Ginn and Co.
Chomsky, N. 1972. Studies on semantics in generative grammar, The Hague: Mouton.
Chomsky, N. 1973. Conditions on transformations. In: Anderson, S. & Kiparsky, P. (eds.) A festschrift for
Morris Halle. New York: Holt, Rinehart and Winston.

914
 A BRI EF H I STO RY O F G ENE R AT I V E G R A MMA R

Chomsky, N. 1975. The logical structure of linguistic theory, New York: Plenum.
Chomsky, N. 1976. Conditions on rules of grammar. Linguistic Analysis, 2, 303–351.
Chomsky, N. 1977. On wh-movement. In: Culicover, P., Wasow, T. & Akmajian, A. (eds.) Formal syntax.
New York, NY: Academic Press.
Chomsky, N. 1979. The morphophonemics of Modern Hebrew, New York: Garland.
Chomsky, N. 1980. On binding. Linguistic Inquiry, 11, 1–46.
Chomsky, N. 1981a. Lectures on government and binding, Dordrecht: Foris.
Chomsky, N. 1981b. Principles and parameters in syntactic theory. In: Lightfoot & N. Hornstein (ed.)
Explanation in linguistics. London: Routledge.
Chomsky, N. 1986a. Barriers, Cambridge, MA: MIT Press.
Chomsky, N. 1986b. Knowledge of language, New York, Praeger.
Chomsky, N. 1988. Language and problems of knowledge, Cambridge, MA: MIT Press.
Chomsky, N. 1991. Some notes on economy of derivation and representation. In: Freidin, R. (ed.) Principles
and parameters in comparative grammar. Cambridge, MA: MIT Press.
Chomsky, N. 1993. A minimalist program for linguistic theory. In: Hale, K. & Keyser, S. J. (eds.) The view
from Building 20: Essays in linguistics in honor of Sylvain Bromberger. Cambridge, MA: MIT Press.
Chomsky, N. 1994. Naturalism and dualism in the study of language and mind. International Journal of Philo-
sophical Studies, 2, 181–200.
Chomsky, N. 1995a. Bare phrase structure. In: Campos, H. & P. Kempchinsky (ed.) Evolution and revolution
in linguistic theory. Washington, D.C.: Georgetown University Press.
Chomsky, N. 1995b. The minimalist program, Cambridge, MA: MIT Press.
Chomsky, N. 1995c. Language and nature. Mind, 104, 1–61.
Chomsky, N. 2000. New horizons in the study of language and mind, NewYork: Cambridge University
Press.
Chomsky, N. 2001. Derivation by phase. In: Kenstowicz, M. (ed.) Ken Hale: A life in language. Cambridge,
MA: MIT Press.
Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry, 36, 1–22.
Chomsky, N. 2007. Approach UG from below. In: Saurerland, U. & H.-M. Gärtner, (ed.) Interfaces + Recur-
sion = Language? Chomsky’s minimalism and the view from syntax-semantics. Berlin: Mouton de Gruyter.
Chomsky, N. 2008. On phases. In: Freidin, R., Otero C. & Zubizarreta M-L. (eds.) Foundational issues in
linguistic theory. Cambridge, MA: MIT Press.
Chomsky, N. & H. Lasnik. 1977. Filters and control. Linguistic Inquiry, 8, 425–504.
Fiengo, R. 1974. Semantic conditions on surface structure. Doctoral dissertation, MIT.
Fitch, T., M. Hauser, & N. Chomsky 2005. The evolution of the language faculty: Clarifications and impli-
cations. Cognition, 97, 179–210.
Freidin, R. 1978. Cyclicity and the theory of grammar. Linguistic Inquiry, 9, 519–549.
Freidin, R. 1992. Foundations of generative syntax, Cambridge, MA, MIT Press.
Freidin, R. 1995. Conceptual shifts in the science of grammar: 1951–1992. In: Otero, C. (ed.) Noam Chom-
sky: critical assessments. vol. 1, London: Routledge.
Freidin, R. 1997. Binding theory on minimalist assumptions. In: Rooryck, P. P. A. J. (ed.) Atomism and bind-
ing. Dordrecht: Foris.
Freidin, R. 1999. Cyclicity and minimalism. In: Epstein, S. D. & Hornstein, N. (eds.) Working minimalism.
Cambridge, MA: MIT Press.
Freidin, R. 2004. Syntactic structures Redux. Syntax, 7, 100–126.
Freidin, R. & H. Lasnik. 2011. Some roots of minimalism in generative grammar. In: Boeckx, C. (ed.) Hand-
book of linguistic minimalism. New York: Oxford.
Freidin, R. and J.-R. Vergnaud. 2001. Exquisite Connections: some remarks on the evolution of linguistic
theory. Lingua, 111, 639–666.
Harris, Z. S. 1951. Methods in structural linguistics, Chicago: University of Chicago Press.
Hauser, M., N. Chomsky, & T. Fitch 2002. The faculty of language: What is it, who has it, and how did it
evolve? Science, 298, 1569–1579.
Jackendoff, R. 1972. Semantic interpretation in generative grammar, Cambridge, MA: MIT Press.
Jackendoff, R. & S. Pinker. 2005. The faculty of language: What’s special about it? Cognition, 95, 201–236.
Jenkins, L. 2000. Biolinguistics: Exploring the biology of language, New York: Cambridge University Press.
Kayne, R. 1975. French syntax: The transformational cycle, Cambridge, MA: MIT Press.
Kayne, R. 2000. Parameters and universals, New York: Oxford University Press.
Kayne, R. 2008. Some preliminary comparative remarks on French and Italian definite articles. In: Freidin

915
 RO BERT F R E I D I N

R., Otero, C., & Zubizarreta, M.-L. (eds.) Foundational issues in linguistic theory. Cambridge, MA: MIT
Press.
Lakoff, G. 1971. On generative semantics. In: Steinberg, D. S. & Jakobovits, L. A. (eds.) Semantics: An
interdisciplinary reader in philosophy, linguistics and psychology. Cambridge: Cambridge University Press.
Larson, R., Viviane Déprez & Hiroko Yamakido (eds.) 2010. The evolution of human language: Biolinguistic
perspectives, New York: Cambridge University Press.
Lasnik, H. 1991. On the necessity of binding conditions. In: Freidin, R. (ed.) Principles and parameters in
comparative grammar. Cambridge, MA: MIT Press.
Lasnik, H. 1995. Last resort. In: Haraguchi, S. & Funaki, M. (eds.) Minimalism and linguistic theory. Tokyo:
Hituzi Syobo.
Lasnik, H. 2000. Syntactic Structures revisited: Contemporary lectures on classic transformational theory, Cam-
bridge, MA: MIT Press.
Lasnik, H. 2004. Conceptions of the cycle. In: Corver, N. & Chen, L.-S. (eds.) WH-Movement: Moving on.
Cambridge, MA: MIT Press.
Lasnik, H. 2008. On the development of Case theory: Triumphs and challenges. In: Freidin, R., Otero, C., &
Zubizarreta, M.-L. (eds.) Foundational issues in linguistic theory. Cambridge, MA: MIT Press.
Lasnik, H.and M. Saito 1984. On the nature of proper government. Linguistic Inquiry, 15, 235–289.
Lasnik, H. and M. Saito 1992. Move a, Cambridge, MA: MIT Press.
Matthews, G. H. 1965. Hidatsa syntax, London: Mouton.
May, R. 1977. The grammar of quantification. Doctoral dissertation, MIT.
May, R. 1985. Logical form: Its structure and derivation, Cambridge, MA: MIT Press.
Mccawley, J. D. 1968. The role of semantics in grammar. In: Bach, E. & Harms, R. T. (eds.) Universals in
linguistic theory. New York: Holt, Rinehart and Winston.
Newmeyer, F. 1986. Linguistic theory in America: The first quarter-century of transformational generative gram-
mar, New York: Academic Press.
Nunes, J. 2004. Linearization of chains and sideward movement, Cambridge, MA: MIT Press.
Perlmutter, D. 1971. Deep and surface constraints in syntax, New York: Holt, Rinehart and Winston.
Phillips, C. 2010. Some arguments and non-arguments for reductionist accounts of syntactic phenomena.
Language and Cognitive Processes (forthcoming).
Pinker, S. and. R. Jackendoff. 2005. The nature of the language faculty and its implications for the evolution
of language. Cognition, 97, 211–225.
Poeppel, D. and D. Embick. 2005. Defining the relation between linguistics and neuroscience. In: Cutler, A.
(ed.) Twenty-first century psycholinguistics: Four cornerstones. Mahwah, NJ: Lawrence Erlbaum Associates.
Pollock, J.-Y. 1989. Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry, 20,
365–424.
Postal, P. 1964. Constituent structure: a study of contemporary models of syntactic description. International
Journal of American Linguistics, 30.
Quicoli, A. C. 1976a. Conditions on quantifier movement in French. Linguistic Inquiry, 7, 583–607.
Quicoli, A. C. 1976b. Conditions on clitic movement in Portuguese. Linguistic Analysis, 2, 199–223.
Rizzi, L. 1980. Violations of the wh-island constraint and the subjacency condition. Journal of Italian Linguis-
tics, 5, 157–195.
Ross, J. R. 1967. Constraints on variables in syntax. Doctoral dissertation, MIT.
Stockwell, R. P., Schachter, P., & Partee, B. H. 1973. The major syntactic structures of English, New York:
Holt, Rinehart and Winston.
Stowell, T. 1981. Origins of phrase structure. Doctoral dissertation, MIT.
Uriagereka, J. 1999. Multiple spell-out. In: Epstein, S. D. & Hornstein, N. (eds.) Working minimalism. Cam-
bridge, MA: MIT Press.
Thompson, D. W. 1942. On growth and form. Cambridge: Cambridge University Press.
Vergnaud, J.-R. 1977/2008. Personal letter to Howard Lasnik and Noam Chomsky. In: Freidin, R., Otero, C.,
& Zubizarreta, M.-L. (ed.s) Foundational issues in linguistic theory. Cambridge, MA: MIT Press.

916
 IND EX

a priori justification 768; see also apriorty apriority: intuition-based views 769–70; of
‘actually’ operator 627–8, 891–2 mathematics and logic 769; meaning-based
adjectives: and adjectives of personal taste 133, 338; views 769–74; pragmatic account 777–8; see also
and attributive modification 330–4, 336; and analytic truths
compositionality 238–9; and contextualism 338; Aristotle 814–17; on assertion 816; on empty
and semantic relativism 338; as predicates 329, names 816; on signification 816–17
333; attributive analysis of 334–5; color artificial languages in semantics 144, 448–53;
adjectives 338; distribution of 328–32; and variety of artificial languages 550; as
intersective adjectives 333–4, 336; nonsubsective models of natural language 549, 551–2; as
adjectives 333; semantic value of 334; subsective notational tools 553; early influences 548–9;
adjectives 333–5; see also gradable adjectives formalization 679, 885–6; vs. natural language 852
adverbs: and functional projections 410; and logical artificial languages in science 546, 862; vs. formal
form 37–8, 411–16, 418–20; and predicate semantics 552–3; vs. philosophy of language 546
abstracts 415; and propositional functions 416; as artificial languages in philosophy 545–8, 552, 679–80;
predicate modifiers 415; conjunctive adverbs 409; as notational tools 545–6; as revealing structure of
diagnostic tests for 409–10; sentence adverbs 410; natural language 546–7; as subject matter 546; vs.
see also adverbs of quantification, and adverbs and paraphrases 545
event semantics assertion: and common ground 47; and
adverbs of quantification 416–22; and conditional presupposition 632; and relevance of
donkey sentences 419–20; and donkey psychology 530–1
sentences 418–19, 421; and prosody 421–2; assessment sensitivity 135–6, 138–9
and the proportion problem 420–1; as binary attitude reports 30; and anti-individualism 806–7;
quantifiers 419; as unselective quantifiers 418–19 and hidden indexical approaches 313; and
adverbs and event semantics 37–8, 413–17, metalinguistic analyses 311–12; and relevance
422; and comparison classes 415; and of psychology 530–1; hidden indexical theories
event individuation 414; and ontological of 801–2; naïve Russelianism 800–1; propositional
commitment 414; as properties of events 413; in theory of 795–6; relational analysis of 796
opaque contexts 415 Augustine 824, 857–8
ambiguity 80, 685–6, 812, 823 Austin, J. L. 877–9
analytic truths 296–301, 865, 867; and a priori
knowledge 231–4; and convention 230, 232, 236; Bacon, Francis 842–3
and holism 233; and linguistic competence 234, bare demonstratives 439, 442–3; and co-
239; and mathematics 236–41, 865;and reference 442–3; and speaker intentions 439, 442,
necessity 231–4; and philosophical inquiry 233– 444, 447; and variable character 442
4; and semantic externalism 233–4; and Benacerraf’s Problem 770
synonymy 233–6; and the Vienna Circle 231, 234 bivalence 43, 145, 148, 822
analytic/synthetic distinction 783, 862, 882 blind-yet-blameless inference 775–6
anaphoric pronouns 27; and the proportion branching quantifiers 589–90; see also quantifiers
problem 371–2, 375; as bound variables 367;
as context-dependent quantifiers 372, 375; cancellability 887; see also conversational implicature
as definite descriptions 369–71; as referring canonical proof 32–3, 35
expressions 367; discourse anaphora 368, 371–8; Carnap, Rudolf 862–70; formal vs. material
donkey anaphora 369–75, 377–8, 582; dynamic mode 863–4; internal vs. external questions 865;
approaches 375–8 logical syntax 863–5; on rationality 864–5
anti-realist semantics 670 Carroll, Lewis 770
 I N DEX

categorical property 736 and slurs 756–7; vs. non-conventional

category mistake 877 implicature 53
causal-historical theory of reference 181, 196, 310, conversation 886; game-like aspects 759–60; maxims
312, 499–502, 506–7 of 54, 58, 87, 886–8;
causation: and discourse interpretation 724–5; conversational implicature 53–60, 95, 165, 882, 887;
context-sensitivity of causal claims 719–20; direct and propositional content 58–9; cardinals and
vs. indirect 723–4; Lewis’s theory of 718–19; the scalar predicates 57; embedded implicatures 60;
language of 716–24; transitivity of 719 empirical work 57, 59–60; explicature 57,
centered worlds 634 59; generalized vs. particularized 60; scalar
coherence theory of truth see truth theories implicature 5–6; vs. contextual implication 54–5
common ground 47–8; and update 48 cooperative principle 53–4, 87
common knowledge 658 correlative definitions 864
communicative intention see speaker intention correspondence theory of truth see truth theories
competence vs. performance 900 Correspondence Theory 606
complex demonstratives 439, 441, 443–4, 447; and counterfactuals see subjunctive conditionals
speaker intentions 439, 444–5; referential vs. creative aspect of language use 901
quantificational treatments 444–5, 447
compositionality 2, 9, 13, 19, 32, 91–101, 190, 195, Davidsonian truth theories see also truth theories
681–2, 837–9; and functional application 682, de dicto/de re distinction 782–3, 802–5; of attitude
890; and linguistic understanding 98; and reports vs. attitude states 804–5
quantified noun phrases 99–101; direct 98; deep structure 23, 39, 899–900
functional formulation of 94; of ideas 845–6; rule- default inferences 59–60
to-rule conception of 93; strong vs. weak 93–4; deflationaism see truth theories
substitutional formulation of 94 definite descriptions 60, 189–95, 196–7, 307–10, 316;
concepts: acquisition of 828; and a priori in attitude ascriptions 308; and Fregean sense 307;
contents 227–8; and Fregean sense 222–3; and and presupposition 383–4; and quantifier
grasp of 221–3; and rule following 226–7; and domain restriction 385–7; as particles 389;
metaphysics-first approachs 229–30; as epistemic as predicates 388–9; attributive vs. referential
intensions 223; uses 384–5; incomplete 385–6; referential
conceptual clarification 877 interpretations 383, 385; Russell’s theory 44, 47,
conceptual role semantics 223–4, 770–1; and 155–6, 381–5, 389, 798–800, 854–5, 877, 889–90;
competence vs. performance 778–9 surface form-meaning mismatch 380; unified
conditional presuppositions 49 theories of descriptions 386–7
conditionals 23, 25, 144, 146, 239, 244, 248–9, demonstratives 18, 21–2, 24; and speaker
371–2, 378, 518–20, 709–10, 713, 779, 823, 887; intentions 26, 442–4; demonstrations 438–44,
see also indicative conditionals, and subjunctive 447; see also bare demonstratives, complex
conditionals demonstratives, and indexicals
consequence 570, 584–5, 592–3; in many-valued deontic modals 247, 459–60, 619, 662, 674, 695, 713
logic 639–40; relation 637 Descartes, René 845–6
constant domain logic 669 descriptive fallacy 879
content/character 16–7, 23–4, 124–5, 622, 630, 892– designated value 639
3; and Lewis’s context/index distinction 631–2; direct reference 95, 155–6, 159, 357, 439, 440–1,
and the a priori/necessary distinction 631; see also 444–5, 447, 515, 519–21, 530, 792, 798, 800, 802,
indexicals 804–7
content of an assertion vs. content of a sentence 702 disagreement 864, 881; moral 708
context 23, 26, 116, 889; of assessment 132; and disambiguator 685
index 134–5; as determining circumstance 136; discourse representation and anaphora 372–5,
as determining content 136, 701; as parameters vs. 378; discourse referents 272–5; discourse
utterance situations 25 representation 372–3; discourse representation
context insensitive language 32 structure 373–4
context principle 832 Discourse Representation Theory 48–9
context set 632; non-defective 632 disjunction: inclusive vs. exclusive 887
context-insensitive expressions 16 dispositions 729–39, 771; and counterfactuals 733;
context-sensitive expressions 16 and verificationism 730; causal basis of 732–3;
context-sensitivity 2, 18, 116–30, 686; and the liar conditional analysis of 734; finks, masks, and
paradox 128–9 antidotes 735; in science 730; test vs. response
contextualism 132–3, 135–6, 164, 172–4 condition 731; vs. dispositional expressions 736–8
contingent a priori 628–9 domain of quantification 836
contingent analytic 788–9 domains of discourse 21–3, 25–6
conventional implicature 60–3; and diagnostics 62–3; dynamic semantics 46–8, 204–5, 653, 750; and

918
 I N DEX

communication 256; and compositionality 254–5; fictionalism 672

and context set 257–8, 261, 266; and context- figurative language 18, 25, 80–1, 165, 831
change potential (update function) 254–6, 258, focus 103–13; and contrastive focus 108; and
269, 272–4, 276; and definite descriptions 268; free focus 105; and givenness 109; and
and discourse 253; and discourse representation presupposition 105–6; and truth conditions 105;
structure (DRS) 253; and discourse vs. types of grammatical focus 111–12
sentence truth-conditions 253; and epistemic formal languages see artificial languages
modals 273–6; and donkey anaphora 259–68; and free pragmatic processes 70–7, 169–73
local context 269; and presupposition 269; and Frege cases 94, 192–3, 787, 632–4, 853–4; and definite
semantic consequence 256–8; and the projection descriptions 193
problem 267–74 Frege-Geach Problem 711; and compositionality 711
Frege, Gottlob 852–4
empirical studies: objections 504–7; semantic Fregeanism see proper names
externalism vs. internalism 503–4; variations in functionally connectives 642–3; precomplete 643
reference intuitions 499–504, 507; variations in future contingents 133, 648
truth-value judgments 501–2 fuzzy semantics 644
empiricist theories of content 296, 297, 860–1
Empty Category Principle (ECP) 907 game-theoretic semantics 662–3; and validity 663;
empty names 153–61, 193; existence vs. signaling games 664
subsistence 820 generalized quantifier theory 343–4, 346, 585–93;
Epicureans 817–19 determiners in 588–9; relational quantifiers
epistemic contextualism 699–702, 718; nonindexical 589–90; see also quantifiers
contextualism 701 generative grammar 896; semantic interpretation
epistemic modals 132, 713 in 904–5
epistemic responsibility 774–9 generic masculines 744–5, 812; false generics 745
epistemicism see vagueness generic operator (Gen) 357–9, 362–4; vs. standard
equivocation 812, 831 quantifiers 359
event semantics 37–8; adverbialization 281, 285–7, generics 761–5; and defeasible validity 360; and
289–93; and Davidsonian logical form 280–1, 284, domain restriction 360–1; and general causal
293; frames of reference 281, 285, 287–9, 292; claims 721–2; and individual level predicates
identity statements 282, 286, 289–92; plural event 356–7; and normalcy approaches 359–60; and
pronouns 281–3; plural event quantification 281, possible worlds semantics 359–60, 363; and
283; supermonadicity 281, 283–4, 286–7, 290, relevant quantification 360; and situation
292–3 semantics 361; and stereotypes and prototypes 361;
existential presupposition 42–3; and implicatures 47, characteristic 761–2; D-generics vs. I-generics
50; and optimality theory 49–50; cognitive 357–9; definite and indefinite singular generics 363;
approaches 49–50; computational approaches 50; probabilistic approaches 361–2; racial 760–5;
in conditionals 45–8; of definite descriptions 44; statistical 761–2; striking property 762; vs.
dynamic semantic theories of 47–50; many-valued quantified statements 355–6
theories of 45, 49; of names 44; pragmatic vs. Government and Binding Theory 908
logical phenomenon 36, 47, 50; in quantified gradable adjectives 18, 23, 25–7, 335–8; and
sentences 45, 49–50 comparison classes 335; and standards 335–6; and
experimental turn: and implications for vagueness 338; semantic value of 336–7
philosophy 498–9, 507–9; in linguistics 495–8 groundedness 207
explicature 89, 167, 169, 172, 174; see also relevance guises 801
theory
expressivism 709–10; and semantics 709; embedding habituals 358–9; see also generics
of mood 710–11; expressing vs. reporting an hearer meaning: conceptual baggage 746; emergent
attitude 709–10 significance 746–9
extension 9–15, 19 Hesperus and Phosphorus see Frege cases
extensional context 13 heteronormativity 747–8, 750
extensional semantic theories 11 hidden constituents 21, 96, 98; as unarticulated
constituents 96, 124–5, 169; hidden
fallacies 817 indexicals 125–6; hidden quantification 21
fiction: and pretence 181–3; and truth 183–4 Hoare calculus 655
fictional names: abstracta as referents 178; Hobbes, Thomas 843–4
existence claims 179–80; non-existent objects holism 299–301, 866–9
as referents 180; possibilia as referents 180–1; hyperintensional context 13
predication in fiction 179; reference of fictional
names 177 ideal language 842–3, 845, 856–7; Begriffsschrift 852

919
 I N DEX

identity theory of truth see truth theories intuitionism 151–2; and disjunction 669–70;
illocutionary act 758, 879; see also speech acts interpretation in modal logics 673–5; choice
illocutionary theories of metaethics 710–11; and sequence 676–7; constructive proof 667–8;
sentential mood 710 epistemological vs. ontological 670–1;
implicature 53–4, 59, 87–9; see also conventional interpretation of the quantifiers 668–9;
implicature, and conversational implicature relation between truth and proof 670; vs.
implied consent 748–9 constructivism 667
imposition 830 intuitions 769–70; about technical terms 509,
incomplete predicates 18, 21–2, 25–6 558–9; and linguistic knowledge 35, 50, 556,
incomplete proposition 125 558; and method of cases in philosophy 499; and
indexicalism 169–71 objective evidence 508, 561; and performance
indexicals 21, 23, 621, 788, 822, 888, 891–2; error 496; as empirical judgments 560–2; expert
and double-indexing 440; and indexes intuitions 496–7, 508–9, 560–1; evidential role in
(Montague) 440; and narrow context vs. wide philosophy 554–6; evidential role in syntax 495–6;
context 446–7; and referential vs. reflexive lay intuitions 497–8, 555–6, 560–1; syntactic vs.
content 446; and the answering machine semantic 558–9
paradox 439, 441–2, 447; and the semantics- island constraints 903
pragmatics debate 445–7; as token-reflexives 446; isomorphism: between language and reality 849,
automatic vs. discretionary 441, 446–7; 855–6
character/content of 441–4, 446–7; circumstance
of evaluation 136; Frege on 440; Kaplan’s kinds 355, 357, 361, 363; see also generics
theory of 439–42, 444–5; pure indexicals vs. knowledge attributions see attitude reports
demonstratives 24, 440–1; referentialist vs. knowledge: context-sensitivity of ascriptions
descriptivist views 445–6 699–702; contrastivism 697–9; of facts or of
indicative conditionals: and counterfactuals 454–5; propositions 695–7; how vs. that 693–5, 877;
and dynamic semantics 463; and epistemic relativism 702
operators 459–60; and implicature 452–4; Kripke model 576
and logical strength 451; and material Kripke’s puzzle about belief 312–16, 805–6;
conditionals 451–4; and non-truth conditional hidden indexical approaches 314–15; pragmatic
accounts 457–9; and probability operators 456–8; approaches 315–16
and restrictor analyses of 461–4; and the Ramsey
test 450, 463–4; as Belnap conditionals 462; lambda abstraction 683
as shifty conditionals 463; as variably lambda calculus 657
strict conditionals 456–7; in embedded language and thought: relationship between 818, 821,
environments 450, 453–4, 459, 462; horseshoe 854
theory of 452–5; vs. biscuit conditionals 449; vs. language acquisition 2, 824, 912
subjunctive conditionals 449 language faculty 895, 901; I-language vs.
indirect discourse report 708, 712 E-language 900
indirect referent 790, 797 language of thought 837; inner speech 821, 824
indirection 80–1, 88 Law of Excluded Middle 148, 668, 671, 823
induction 842 Law of Non-contradiction 817
inflection see prosody Leibniz, Gottfried 849–50
information diagrams 659 Lewisian causal chain 718
ingredient sense 629 lexical causatives 722–4
intended interpretations 147, 150 lexical insertion 897
intension 11–13, 19, 620, 888–9; vs. extension 791 liar paradox 127, 206–7; and correspondence theories
intensional context 13, 797, 803 of truth 207; and deflationism 206; and truth-
Intensional Logic (Montague) 680, 686–7 theories of meaning 206–7
intensional objects 621 Lindström’s Theorem 587
intensional semantic theories 11; see also possible linguistic account of necessity 783–9, 861; Quine’s
worlds semantics critique of 784–6, 868
intensional transitive verbs 889–90 linguistic competence 2–3, 35, 61, 92, 556–60; as
intensionality 686 an ability 559–60; as a priori knowledge 557;
interest-relativism 151 and Cartesianism 558–60, 562; as propositional
interpretations of ‘’ 618–20 knowledge 556, 559
interpretive truth conditions 32 linguistic conventionalism 812–13, 815, 817–18,
interpretive truth theory 32, 34–35, 39, 257 829–30, 843, 848
intersubjectivity of thoughts 853 linguistic naturalism 812–13, 817, 819, 847–8
intonation 103–4, 110–13; neutral intonation 103 linguistic pluralism 842–5
introduction and elimination rules 580 linguistic turn 852–3, 873

920
 I N DEX

linguistic universalism 841–2, 845–9 meaning skepticism 295–303; meaning and

linguistics and philosophy 1–4, 688 rule-following 302–3; in ‘Two Dogmas of
literal meaning 19–20 Empiricism’ 297–301; in Word and Object 300–1
Locke, John 844–5; on reference 844 medieval theories of reference 832–6;
logic: classical propositional 45, 823; dynamic 655– ampliation 836; theory of appellation 835–6;
8, 664–5; epistemic 658–9; fuzzy 644–5; theory of supposition 828
modal 575–8; philosophical 30; probability 647; mental representations: and truth-conditional
provability 673; public announcement logic 660–1; semantics 23; meanings of 27
weak Kleene vs. strong Kleene 45; see also many- metaethics: and theories of reference 706–7;
valued logics, modal predicate logic, and temporal contextualism 707–8; realism 705–7;
logic relativism 708; speaker subjectivism 707, 709;
logic-first theories 150; vs. model theories 151 metalinguistic negation 765
Logical Atomism 30 metamathematics 862–3
logical consequence see consequence Millianism 156–61
logical form 29, 849, 855, 876; of action sentences 37; minimalism 21, 119, 163–4, 171–2; vs.
and canonical proof 33; of definite descriptions 30, contextualism 172
35–7; and descriptive metaphysics 30, 38; and Minimalist Program 895, 909–14; economy
interpretive truth theory 32; and Logical Form 39; conditions on derivations 908–9; internal vs.
and regimented language 31, 37; and validity 38–9; external merge 911–12; Strong Minimalist
assignment of 29; knowledge of 30, 32; of quantified Thesis 910; third-factor explanations 913–14
sentences 36–7; vs. grammatical form 29–31 modal predicate logic 614–18; conceptualism 615;
Logical Form (LF) 39, 170, 906, 908–10, 913 counterpart theory 616–17; objectualism 615;
logical quantifier 585–6, 591–2; Lindström’s variable domains 617; with identity 617–18
criterion 586; Mostowski’s criterion 585 model theory: for 2D logic 625–6; for epistemic
logicality 584, 592 logic 659; goals of model 571; for modal
logicism 861 logic 611–12; for predicate logic 572; for sentential
logos 811, 821 logic 571; for tense logic 573–4, 604–6; see also
artificial languages
many-valued logics 45, 148, 636; applications of 649; model-theoretic semantics see artificial languges
finitely many values 640–4; infinitely many moderate contextualism 119
values 644–7; motivations 648–9; non-truth- monsters (Kaplan) 630–2
functional systems 647; predicate logic 646–7; Montague grammar: vs. formal semantics 687; see also
many-valued semantics 636 artificial languages
mass terms: and abstract mass terms 424, 431; and Montague’s Thesis 654
atomic parts 428–30, 433; and coercion 432–3; and Moore’s paradox 55
cross-linguistic data 433–4; and cumulativity 426;
and diachronic data 434; and divisibility 426, names see proper names
429; and first-order logic 431–2, 435; and narrow vs. wide content 806
homogeneity 426, 433; and lattice theories 428, natural language metaphysics 688
430–1, 435; and measurement 424–6, 432, 434; necessary a posteriori 629, 706, 787–8
and mereology 429–31; and natural language necessity: deep vs. superficial (Evans) 627–9; vs.
metaphysics 424–5, 429, 434–5; and plural apriority 787
morphology 425–6, 432; and vagueness 431, 433; negative existentials 193
as names of natural kinds 430; as predicates 430; neologisms 812
mass and count interpretations 427, 432–3; vs. noncognitivism in metaethics 709–13; approach to
aggregate terms 429, 433 inconsistency 711
mass and count features: as interpretability non-indexical contextualism 136, 140
constraints 226–9; in the verbal domain 421, 435; normative significance of logic 150
as well-formedness constraints 425–8 norms of assertion 139–40
mathematical models: and generative grammars 538–
41; and historical linguistics 537–8; and language object of thought vs. manner of thought 847
as a psychological phenomenon 535–6; and role ontological commitment 837–8
in theory construction 534; vs. applications of opaque context see intensional context
logic 535–6; see also artificial languages, and model optimality theory 49–50
theory ordinary language philosophy 54, 858, 888;
meanings 2, 10–16, 22–3, 35; as abstract vs. adequacy of ordinary language 874; linguistic
psychology entities 527–8; gappy 22–3; knowledge phenomenology 881
of 21, 80; possible worlds approach 244–5, 620, Oxford realism 875–6
789–92, 888–93; vs. reference 832
meaning postulates 684–5 performatives 879

921
 I N DEX

phenomenalism 731 chains of reference 310, 312; and cluster theories

philosophical analysis 737–8 of names 309–10; and Fregean sense 308; and
Phonetic Form (PF) 906 function 154–5; and gappy propositions 156; and
phrase structure rules 897–8 possible-worlds intensions 12; and relevance of
phrase-marker 898–9 psychology 529–31; as descriptions 10, 194–5,
picture theory of language 855–7 309–10; as directly referring expressions 156,
pitch accents 104; nuclear pitch accent (NPA) 104 158–9; as predicates 160; in propositional attitude
Plato 812–14; possibility of false speaking 814 ascriptions 308, 311; as rigid designators 310;
plurals: and collective interpretations 393–4, 397–8, nonexistents as referents 153, 159–61; ordinary vs.
405; and compositionality 293; and contextual logical 799–800; see also Millianism
restrictions on distributivity 395–6; and propositional radical 23, 59, 125
cumulative interpretations 393, 403–5; and propositions 11, 12, 14–16, 20; and relevance of
discourse representation theory (DRS) 292; and psychology 529–31; and truth conditions 209–10,
distributive interpretations 393–6, 404–5; and 215; as definite descriptions 10; as intensions
event semantics 403–5; and first-order logic of sentences 19; as meanings of (declarative)
398–403; and first-order plural logic 398, sentences 31, 695; as objects of thought 31; as
400–2, 405; and implicature 292–3; and possible-worlds intensions 12; as revealing logical
plural variables 400; and quantifier domain form 31; cognitive realist theories of 216–9;
restriction 398–400; and Russell’s paradox complete 20, 23; interpreted relative to context 21;
399–401; and second-order logic 398, 401–3; metaphysical status of 209; minimal 21–3; ordinary
count vs. mass nouns 292; dependent plurals 293; vs. logical 799–800; realist theories of 210–12; in
plural definite descriptions 396; plurals as semantic and cognitive theories 209, 212, 215–18;
collections 398–401 unity of 210–11, 214–15, 219; vs. facts 696
positivism 296, 544, 730–1, 787, 860–2, 865, 870, prosody 103; and information structure 103; and
874, 881 marked prosody 104; stress 104
possibility: metaphysical vs. logical 575 psychologism 853
possible worlds semantics 12–16, 790; and
conditionals 244, 248–9; and impossible quantifiers 18, 23; and cardinality quantifiers 350,
worlds 251; and intensional transitives 247, 352; and categorical propositions 344–5;
251; and model theory 242–5; and Montague and conservativity 348; and dual
grammar 244; and ontology 244–5; and transformation 345; and existential import 345;
philosophical logic 242–4; and proper names 246, and first-order definability 344, 347–8, 352;
250; and propositional attitude ascriptions 248–51; and monotonicity 348; and permutation
and sentential adverbs 247; and the problem of invariance 349–50, 352; and proper names 347;
necessary co-extension 248–51, 791–2; and two and proportionality 348; and quantifier
dimensionalism 250 dependence 349; and the square of opposition 344–
poverty of stimulus 901–2 6; as binary relations 347; as logical terms 343, 344,
pragmatics 18–20, 26, 36, 59–60, 62, 163, 165, 174, 349–50, 352; as second-level concepts 346; first-
250, 255–6, 384, 387, 478, 498, 519–521, 686–7, order vs. second-order 342–3, 348, 350–2; standard
759–60, 777, 887, 881, 891; vs. semantics 23–5, 61, quantifier theory 581–5
68, 71, 75, 77, 163–4, 167–9, 172, 442, 445–447, quantifying in 614–15, 803
570, 661, 688, 720, 754, 801, 886 questions: and alternative questions 484;
presupposition 632; and common ground 47; and and concealed questions 485; and context
discourse situation vs. context of discourse 48; and dependence 482–3, 486; and embedding
presupposed contents 519; and presuppositional verbs 484–5; and fragment answers 480, 483,
theories of names 520; and projection tests 519; 485; and inquiry 488–9; and interrogative speech
familiarity conditions 520; projection problem 46 acts 478, 481, 488; and partition theories of 479–
primary linguistic data 901 80, 481–3, 485, 488; and quantification 487–8;
Principle of Semantic Innocence 798 and syntactic form 478, 481, 485, 487; and yes-no
Principle of Substitutivity Salva Veritate 849 questions 484; as sets of possible answers 480,
Principle of Tolerance 864, 866 482–4, 489; as sets of true answers 481–2, 483–4,
Principles and Parameters 904–6, 910–11 486, 489, 694
private language 818, 824, 857 Quine, Willard van Ormand 865–70; against
PRO 694 analytic/synthetic distinction 866; against
process algebra 657 conventionalism 865–6; confirmation holism 867–
process graphs 655 9; indeterminacy of translation 867
program semantics 654–5; see also dynamic semantics
pronoun of laziness 697 radical contextualism 119–21
proper names 9–10, 12, 14–15, 19, 95, 189–92, 196, reference (Bedeutung) 9, 16, 19, 189–97; and natural
812–13; abstracta as referents 153, 157; and causal kinds 196–7; of general terms 196–7

922
 I N DEX

regimentation 31, 35, 37, 39 structured intensions 14–15

relativism 132–41; content-relativism 136 structured propositions 11, 16, 136, 156, 190, 212,
relevance theory 57–8, 163–74; and definite 215, 527, 796
descriptions 167–9; and modularity 164; and subjunctive conditionals 734; and dynamic strict
proper names 167–9; and the function of linguistic implication analyses of 470; and mood 466–7;
meaning 167 and negative polarity items (NPIs) 517–8; and
reliabilism 773–4 non-monotonicity 469–70; and past-tense
reliable inference 771–3 morphology 467, 474–5; and possible worlds
rigidity: rigid designator 14, 787, 791–2; rigid semantics 467–8; and premise semantics 468–9;
intensions 620; rigidified description 628; vs. direct and restrictor analyses 471–2; and Sobel
reference of terms 792 sequences 470; and strict conditional analyses 517;
Russell, Bertrand 854–5 and strict implication analyses of 469–70; and
Ryle, Gilbert 876–7; systematically misleading the principle of Conditional Excluded Middle
expressions 876–7 (CEM) 472–3; and the Ramsey test 467–8; and
variably strict analyses 468–9, 517–8
safety principle 146 supervaluationism 147–8, 151, 647;
sayables 819, 820–3; incomplete 821; simple vs. subvaluationism 647
non-simple 822–3 surface structure 898–9; see also deep structure
second intensions 830–1 synonymy 812
semantic indecision 147, 151 syntactic arguments: against unarticulated
semantic rules 865, 877 constituents 514–16, 520–1; and complex
semantics 3, 10, 18–19, 20, 23, 25, 27, 34, 35, 39, 43, demonstratives 516; and objects of assertion 516;
50, 75–77, 98, 133, 135, 136, 144, 147, 155, 190, and definite descriptions 516
204–5, 209, 242–51, 253–5, 257–8, 281, 509, 527,
534, 544, 547–53, 556, 563, 569–70, 574, 604, 620, T-sentence 32–4
653, 679–80, 682, 685, 687 Tarski’s theory of truth 32, 203
sense-data theories 878 temporal logic: hybrid tense logic 606–7; instants vs.
sense (Sinn) 15, 154–5, 790, 853 intervals 599; method of temporal arguments
signification 816–17, 818, 820, 824, 829–31, 844, 848 597–9; reified temporal logics 599–602; tense
simple demonstratives see bare demonstratives logic 572–4, 602–4; two approaches 596–7
singular proposition 11, 16 tense logic see temporal logic
Skolem’s Paradox 581 The Compactness Theorem 581–2
slurs 743; and conventional implicature 756– The Completeness Theorem 581
7; language as a weapon 742, 753–4, The Löwenheim-Skolem Theorem 581
757–8; presuppositional account 755–6; thick evaluative terms 712, 743, 754
prohibitionism 757; semantic account 754–5 tonk 722
Sophists 811–12 trace 906
Sorites paradox see vagueness transformation rules 898–900
Soundness and completeness: of K 612–13; of truth at a context of assessment vs. use 134–5, 139
T 613–14 truth conditions 2, 19, 20–1, 22, 25; of non-
speaker intention 26–7, 81–3, 87; and context 26, declarative sentences 35
83; and the reflexive paradox 85–6; as self- truth tables: asymmetric 45–6; Łukasiewicz 639, 641;
referential 80, 82, 85–6; as supplementing literal strong Kleene 638, 641; weak Kleene 638, 640–1
meaning 26; problem of access 27; understanding truth theories: as a theory of meaning 204; coherence
speaker intention 27; vs. perlocutionary theory of truth 199, 200, 201; correspondence 199–
intentions 84 201, 204–6, 816, 822; deflationism 199, 202–6
speaker meaning 22, 53–7, 59, 819, 887; vs. identity theory of truth 199, 201
linguistic meaning 20, 79–81; vs. what is truth vs. acceptability 886
communicated 59–60 truth vs. truth-in-context 22
speech acts 83–5, 88–9, 757–60, 812, 821–2, 879; truth-conditional semantics 23, 652;
effectiveness of 749–51, 760 Davidsonianism 570
static semantics 254; and common ground 255; and truth-functional analysis of connectives 886
communication 255; and epistemic modals 276; truth-value gap 47, 648, 823
and semantic consequence 256 truth-value glut 648
Stoics 819–25; on negation 822–3; on parts of truthmakers 675
speech 820 two-dimensionalism: and sense and intention 633;
Strawson, P. F. 879–80; on Russell’s theory of diagonal 626, 634; homogeneity of paramaters 625;
descriptions 36, 44–5, 47, 855, 879–80 n-dimensional logic 624; scenarios vs. worlds 633;
structural linguistics 895–6 u-worlds and e-worlds 625
structured characters 17 type-lifting 97

923
 I N DEX

type: of a quantifier phrase 586 684–5; logical verbs: and affirmation 325; and argument
type 682–3; of natural language phrases 683, 890 structure 322; and force 325; and lexical
aspect 324–5; as predicates 318–21; as properties of
underspecification 57 events 321–3; as unifying propositions 325–6; neo-
understanding: of context 21; of literal meaning 20; Davidsonian vs. Montagovian theories 322; verbs
of sentences 91 and times 324–5; vs. non-verbal predicates 319–21
universal grammar 827, 846–7, 850, 901–2, 913 verification principle 860–2
universals 828, 843 verificationism 296, 299–300, 860–1
use theory of meaning 858, 888 Vienna Circle 860–1
use/mention distinction 823, 832–3
Wittgenstein, Ludwig 855–8, 874; language-
vagueness 143–52, 686; degree theory of games 857, 874; Philosophical Investigations 824;
vagueness 149, 151; epistemicism 145–7, 151, 823; Tractatus Logico-Philosophicus 855–7, 861
higher-order 148–9; Sorites paradox 128, 143, 823; words: as symbols 815; meaning 21–2, 53
vague sets 145
validity 19–22, 24, 570–571 X-bar theory 903–4

924