Burgess Math Model Modality

This page intentionally left blank
MATHEMATICS, MODELS, AND MODALITY
John Burgess is the author of a rich and creative body of work which
seeks to defend classical logic and mathematics through counter-
criticism of their nominalist, intuitionist, relevantist, and other critics.
This selection of his essays, which spans twenty-five years, addresses
key topics including nominalism, neo-logicism, intuitionism, modal
logic, analyticity, and translation. An introduction sets the essays in
context and offers a retrospective appraisal of their aims. The volume
will be of interest to a wide range of readers across philosophy of
mathematics, logic, and philosophy of language.
JOHN P. BURGESS is Professor in the Department of Philosophy,

Princeton University. He is co-author of A Subject With No Object
with Gideon Rosen (1997) and Computability and Logic, 5th edn with
George S. Boolos and Richard C. Jeffrey (2007), and author of Fixing
Frege (2005).
MATHEMATICS, MODELS,
AND MODALITY
Selected Philosophical Essays
JOHN P. BURGESS
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521880343
© John P. Burgess 2008
This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2008
ISBN-13 978-0-511-38618-3 eBook (EBL)
ISBN-13 978-0-521-88034-3 hardback
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Dedicated to the memory of my sister
Barbara Kathryn Burgess
Contents
Preface page ix
Source notes xi
Introduction 1
PART I MATHEMATICS 21
1 Numbers and ideas 23
2 Why I am not a nominalist 31
3 Mathematics and Bleak House 46
4 Quine, analyticity, and philosophy of mathematics 66
5 Being explained away 85
6 E pluribus unum: plural logic and set theory 104
7 Logicism: a new look 135
PART II MODELS, MODALITY, AND MORE 147

8 Tarski’s tort 149
9 Which modal logic is the right one? 169
10 Can truth out? 185
11 Quinus ab omni naevo vindicatus 203
vii
viii Contents
12 Translating names 236
13 Relevance: a fallacy? 246
14 Dummett’s case for intuitionism 256
Annotated bibliography 277

References 284
Index 297
Preface
The present volume contains a selection of my published philosophical

papers, plus two items that have not previously appeared in print.
Excluded are technical articles, co-authored works, juvenilia, items super-
seded by my published books, purely expository material, and reviews.
(An annotated partial bibliography at the end of the volume briefly
indicates the contents of such of my omitted technical papers as it seemed
to me might interest some readers.) The collection has been divided into
two parts, with papers on philosophy of mathematics in the first, and on
other topics in the second; references in the individual papers have been
combined in a single list at the end of the volume. Bibliographic data for
the original publication of each item reproduced here are given source
notes on pp. xi–xiii, to which the notes of personal acknowledgment,
dedications, and epigraphs that accompanied some items in their original
form have been transferred; abstracts that accompanied some items have
been omitted.
It has become customary in volumes of this kind for the author to
provide an introduction, relating the various items included to each
other, as an editor would in an anthology of contributions by different
writers. I have fallen in with this custom. The remarks on the individual
papers in the introduction are offered primarily in the hope that they may
help direct readers with varying interests to the various papers in the
collection that should interest them most. But such introductions also
serve another purpose: they provide an opportunity for an author to note
any changes of view since the original publication of the various items, thus
reducing any temptation to tamper with the text of the papers themselves
on reprinting. I have made only partial use of the opportunity to note
changes in view, but nonetheless I have felt no temptation to make
substantial changes in the papers, since my own occasional historical
research has convinced me of the badness of the practice of revising papers
on reprinting.
ix
x Preface
I have tried to acknowledge in each individual piece those to whom
I have been most indebted in connection with that item, though I am sure
there are some I have unintentionally neglected, whose pardon I must beg.
Here I would like to acknowledge those who have been helpful specifically
with the preparation of the present collection: Hilary Gaskin, who first
suggested such a volume, and Joanna Breeze, along with Gillian Dadd and
the rest of the staff who saw the work through publication.
Source notes
‘‘Numbers and ideas’’ was first delivered orally as part of a public debate at
the University of Richmond (Virginia), 1999. Ruben Hersh argued for the
thesis ‘‘Resolved: that mathematical entities and objects exist within the
world of shared human thoughts and concepts.’’ I argued against. It was
first published in a journal for undergraduates edited at the University
of Richmond (England), the Richmond Journal of Philosophy, volume 1
(2003), pp. 12–17. (There is no institutional connection between the
universities of the two Richmonds, and my involvement with both is
sheer coincidence.)
‘‘Why I am not a nominalist’’ was first delivered orally under the title
‘‘The nominalist’s dilemma,’’ to the Logic Club, Catholic University of
Nijmegen, 1981. It was first published in the Notre Dame Journal of Formal
Logic, volume 24 (1983), pp. 93–105.
‘‘Mathematics and Bleak House’’ was first delivered orally at a sympo-
sium ‘‘Realism and anti-realism’’ at the Association for Symbolic Logic
meeting, University of California at San Diego, 1999. The other symposiast
was my former student Penelope Maddy, and the Dickensian title of my
paper is intended to recall the Dickensian title of her earlier review,
‘‘Mathematics and Oliver Twist’’ (Maddy 1990). First published in
Philosophia Mathematica, volume 12 (2004), pp. 18–36.
‘‘Quine, analyticity, and philosophy of mathematics’’ was first delivered
orally at the conference ‘‘Does Mathematics Require a Foundation?,’’
Arché Institute, University of St. Andrews, 2002. Identified in its text as
a sequel to the preceding item, this paper circulated in pre-publication
draft under the title ‘‘Mathematics and Bleak House, II.’’ First published in
the Philosophical Quarterly, volume 54 (2004), pp. 38–55.
‘‘Being explained away’’ is a shortened version (omitting digressions on
technical matters) of a paper delivered orally to the Department of
Philosophy, University of Southern California, 2004. (I wish not only to
thank that department for the invitation to speak, but especially to thank
xi
xii Source notes
Stephen Finlay, Jeff King, Zlatan Damnjanovic, and above all Scott
Soames for their comments and questions, as well as for their hospitality
during my visit.) It was first published in the Harvard Review of Philosophy,
volume 13 (2005), pp. 41–56.
‘‘E pluribus unum’’ evolved from a paper ‘‘From Frege to Friedman’’
delivered orally at the Logic Colloquium of the University of
Pennsylvania and the Department of Logic and Philosophy of Science
at the University of California at Irvine. It was first published in
Philosophia Mathematica, volume 12 (2004), pp. 193–221. (I am grateful
to Harvey Friedman for introducing me to his recent work on reflection
principles, to Kai Wehmeier and Sol Feferman for drawing my attention
to the earlier work of Bernays on that topic, and to Penelope Maddy for
pressing the question of the proper model theory for plural logic, which
led me back to the writings of George Boolos on this issue. From
Feferman I also received valuable comments leading to what I hope is
an improved exposition.)
‘‘Logicism: a new look’’ was first delivered orally at the conference
marking the inauguration of the UCLA Logic Center, and later (under a
different title) as part of the annual lecture series of the Center for
Philosophy of Science, University of Pittsburgh, both in 2003. It has not
previously been published.
‘‘Tarski’s tort’’ was first delivered orally at Timothy Bays’ seminar on
truth, Notre Dame University, Saint Patrick’s Day, 2005. It was previously
unpublished. The paper should be understood as dedicated to my teacher
Arnold E. Ross, mentioned in its opening paragraphs.
‘‘Which modal logic is the right one?’’ was first delivered orally at the
George Boolos Memorial Conference, University of Notre Dame, 1998. It
was first published in the Notre Dame Journal of Formal Logic, volume 40
(1999), pp. 81–93, as part of a special issue devoted to the proceedings of
that conference. Like all the conference papers, mine was dedicated to the
memory of George Boolos.
‘‘Can truth out?’’ was first delivered orally under the title ‘‘Fitch’s para-
dox of knowability’’ as a keynote talk at the annual Princeton–Rutgers
Graduate Student Conference in Philosophy, 2003. It was first published
in Joseph Salerno, ed., New Essays on Knowability, Oxford: Oxford
University Press (2007). The paper originally bore the epigraph ‘‘Truth
will come to light; murder cannot be hid long; a man’s son may, but at the
length truth will out’’ (Merchant of Venice II: 2). Thanks are due to Michael
Fara, Helge Rückert, and Timothy Williamson for perceptive comments
on earlier drafts of this note.
Source notes xiii
‘‘Quinus ab omni naevo vindicatus’’ was first delivered orally to the
Department of Philosophy, MIT, 1997. It was first published in Ali
Kazmi, ed., Meaning and Reference, Canadian Journal of Philosophy
Supplement, volume 23 (1998), pp. 25–65. (The present paper is a com-
pletely rewritten version of an unpublished paper, ‘‘The varied sorrows of
modality, part II.’’ I am indebted to several colleagues for information used
in writing that paper, and for advice given on it once written, and I would
like to thank them all – Gil Harman, Dick Jeffrey, David Lewis – even if
the portions of the paper with which some of them were most helpful have
disappeared from the final version. But I would especially like to thank
Scott Soames, who was most helpful with the portions that have not
disappeared.)
‘‘Translating names’’ was first published in Analysis, volume 65 (2005),
pp. 96–204. I am grateful to Pierre Bouchard and Paul Égré for linguistic
information and advice.
‘‘Relevance: a fallacy?’’ was first published in the Notre Dame Journal of
Formal Logic, volume 22 (1981), pp. 76–84. Its sequels were Burgess (1983c)
and Burgess (1984b).
‘‘Dummett’s case for intuitionism’’ was first published in History and
Philosophy of Logic, volume 5 (1984), pp. 177–194. The paper originally bore
the epigraph from Chairman Mao ‘‘Combat Revisionism!’’ I am indebted
to several colleagues and students for comments, and especially to Gil
Harman, who made an earlier draft of this paper the topic for discussion at
one session of his summer seminar. Comments by editors and referees led
to what it is hoped are clearer formulations of many points.
Introduction
ABOUT ‘‘ R E A L I S M ’’
A word on terminology may be useful at the outset, since it is pertinent
to many of the papers in this collection, beginning with the very first.
The label ‘‘realism’’ is used in two very different ways in two very different
debates in contemporary philosophy of mathematics. For nominalists,
‘‘realism’’ means acceptance that there exist entities, for instance natural
or rational or real numbers, that lack spatiotemporal location and do not
causally interact with us. For neo-intuitionists, ‘‘realism’’ means acceptance
that statements such as the twin primes conjecture may be true independ-
ently of any human ability to verify them. For the former the question of
‘‘realism’’ is ontological, for the latter it is semantico-epistemological. Since
the concerns of nominalists and of neo-intuitionists are orthogonal, the
double usage of ‘‘realism’’ affords ample opportunity for confusion.
The arch-nominalists Charles Chihara and Hartry Field, for instance,
are anti-intuitionists and ‘‘realists’’ in the neo-intuitionists’ sense. They do
not believe there are any unverifiable truths about numbers, since they do
not believe there are any numbers for unverifiable truths to be about. But
they do believe that the facts about the possible production of linguistic
expressions, or about proportionalities among physical quantities, which in
their reconstructions replace facts about numbers, can obtain independ-
ently of any ability of ours to verify that they do so. Michael Dummett, the
founder of neo-intuitionism, was an early and forceful anti-nominalist, and
though he calls his position ‘‘anti-realism,’’ he and his followers are ‘‘real-
ists’’ in the nominalists’ sense, accepting some though not all classical
existence theorems, namely those that have constructive proofs, and agree-
ing that it is a category mistake to apply spatiotemporal or causal predicates
to mathematical subjects.
On top of all this, even among those of us who are ‘‘realists’’ in both
senses there are important differences. Metaphysical realists suppose, like
1
2 Mathematics, Models, and Modality
Galileo and Kepler and Descartes and other seventeenth-century worthies,
that it is possible to get behind all human representations to a God’s-eye
view of ultimate reality as it is in itself. When they affirm that mathematical
objects transcending space and time and causality exist, and mathematical
truths transcending human verification obtain, they are affirming that such
objects exist and such truths obtain as part of ultimate metaphysical reality
(whatever that means). Naturalist realists, by contrast, affirm only (what
even some self-described anti-realists concede) that the existence of such
objects and obtaining of such truths is an implication or presupposition of
science and scientifically informed common sense, while denying that
philosophy has any access to exterior, ulterior, and superior sources of
knowledge from which to ‘‘correct’’ science and scientifically informed
common sense. The naturalized philosopher, in contrast to the alienated
philosopher, is one who takes a stand as a citizen of the scientific com-
munity, and not a foreigner to it, and hence is prepared to reaffirm while
doing philosophy whatever was affirmed while doing science, and to
acknowledge its evident implications and presuppositions; but only the
metaphysical philosopher takes the status of what is affirmed while doing
philosophy to be a revelation of an ultimate metaphysical reality, rather
than a human representation that is the way it is in part because a reality
outside us is the way it is, and in part because we are the way we are.
My preferred label for my own position would now be ‘‘naturalism,’’ but
in the papers in this collection, beginning with the first, ‘‘realism’’ often
appears. Were I rewriting, I might erase the R-word wherever it occurs; but
as I said in the preface above, I do not believe in rewriting when reprinting,
so while in date of composition the papers reproduced here span more than
twenty years, still I have left even the oldest, apart from the correction of
typographical errors, just as I wrote them. Quod scripsi, scripsi.
This collection begins with five items each pertinent in one way or
another to nominalism and the problem of the existence of abstract entities.
The term ‘‘realism’’ is used in an ontological sense in the first of these,
‘‘Numbers and ideas’’ (2003). This paper is a curtain-raiser, a lighter piece
responding to certain professional mathematicians turned amateur philo-
sophers who propose a cheap and easy solutions to the problem. According
to their proposed compromise, numbers exist, but only ‘‘in the world of
ideas.’’ Since acceptance of this position would render most of the profes-
sional literature on the topic irrelevant, and since the amateurs often offer
unflattering accounts of what they imagine to be the reasons why profes-
sionals do not accept their simple proposal, I thought it worthwhile to
accept an invitation to try to state, for a general audience, our real reasons,
Introduction 3
which go back to Frege. The distinction insisted upon in this paper,
between the kind of thing it makes sense to say about a number and the
kind of thing it makes sense to say about a mental representation of a
number (and the distinction, which exactly parallels that between the two
senses of ‘‘history,’’ between mathematics, the science, and mathematics, its
subject matter) is presupposed throughout the papers to follow.
Some may wonder where my emphatic rejection of ‘‘idealism or con-
ceptualism’’ in this paper leaves intuitionism. The short answer is that I
leave intuitionism entirely out of account: I am concerned in this paper with
descriptions of the mathematics we have, not prescriptions to replace it with
something else. Intuitionism is orthogonal to nominalism, as I have said,
and issues about it are set aside in the first part of this collection. I will add
that, though I do not address the matter in the works reprinted here, my
opinion is that Frege’s anti-psychologistic and anti-mentalistic points raise
some serious difficulties for Brouwer’s original version of intuitionism, but
no difficulties at all for Dummett’s revised version. Neither opinion should
be controversial. Dummett’s producing a version immune to Fregean
criticism can hardly surprise, given that the founder of neo-intuitionism
is also the dean of contemporary Frege studies. That Brouwer’s version, by
contrast, faces serious problems was conceded even by so loyal a disciple as
Heyting, and all the more so by contemporary neo-intuitionists.
AGAINST HERMENEUTIC AND REVOLUTIONARY

NOMINALISM
‘‘Why I am not a nominalist’’ (1983) represents my first attempt to articu-

late a certain complaint about nominalists, namely, their unclarity about
the distinction between is and ought. It was this paper that first introduced a
distinction between hermeneutic and revolutionary nominalism. The for-
mulations a decade and a half later in A Subject With No Object (Burgess
and Rosen, 1997) are, largely owing to my co-author Gideon Rosen, who
among other things elaborated and refined the hermeneutic/revolutionary
distinction, more careful on many points than those in this early paper.
This piece, however, seemed to me to have the advantage of providing a
more concise, if less precise, expression of key thoughts underlying that
later book than can be found in any one place in the book itself. Inevitably
I have over the years not merely elaborated but also revised (often under
Rosen’s influence) some of the views expressed in this early article.
First, the brief sketches of projects of Charles Chihara and Hartry Field
in the appendix to the paper (which I include on the recommendation of an
anonymous referee, having initially proposed dropping it in the reprinting)
are in my present opinion more accurate as descriptions of aspirations than
of achievements, and even then as descriptions only to a first approxima-
tion; moreover the later approach of Geoffrey Hellman is not discussed at
all. My ultimate view of the technical side of the issue is given in full detail
in the middle portions of A Subject, superseding several earlier technical
papers.
Further, though I still see no serious linguistic evidence in favor of any
hermeneutic nominalist conjectures, I no longer see the absence of such
evidence as the main objection to them. For reasons that in essence go back
to William Alston, such conjectures lack relevance even if correct. Even if we
grant that ‘‘There are prime numbers greater than a million’’ does just
mean, say, ‘‘There could have existed prime numerals greater than a
million,’’ the conclusion that should be drawn is that ‘‘Numbers exist’’
means ‘‘Numerals could have existed,’’ and is therefore true, as anti-
nominalists have always maintained, and not false, as nominalists have
claimed. There is no threat at all to a naturalist version of anti-nominalism
in such translations, though there might be to a metaphysical version.
This line I first developed in a very belatedly published paper (Burgess
2002a) of which a condensed version was incorporated into A Subject.
Finally, I now recognize that there is a good deal more to be said for the
position I labeled ‘‘instrumentalism’’ than I or almost anyone active in the
field was prepared to grant back in the early 1980s when I wrote ‘‘Why I am
not,’’ or even in the middle 1990s, when I wrote my contributions to
A Subject. The position in question is that of those philosophers who speak
with the vulgar in everyday and scientific contexts, only to deny on entering
the philosophy room that they meant what they said seriously. This view is
now commonly labeled ‘‘fictionalism,’’ and it deserves more discussion
than it gets in either ‘‘Why I am not’’ or A Subject. It should be noted that
while I originally opposed fictionalism (or instrumentalism) to both the
revolutionary and hermeneutic positions, Rosen has correctly pointed
out that fictionalism itself comes in a revolutionary version (this is the
attitude philosophers ought to adopt) and a hermeneutic version (this is the
attitude commonsense and scientific thinkers already do adopt). What
I originally called the ‘‘hermeneutic’’ position should be called the ‘‘content-
hermeneutic’’ position, and the hermeneutic version of fictionalism the
‘‘attitude-hermeneutic’’ position, in Rosen’s refined terminology.
On two points my view has not changed at all over the past years. First,
while nominalists would wish to blur what for Rosen and myself is a
key distinction, and avoid taking a stand on whether they are giving a
Introduction 5
description of the mathematics we already have (hermeneutic) or a pre-
scription for a new mathematics to replace it (revolutionary), gesturing
towards a notion of ‘‘rational reconstruction’’ that would somehow manage
to be neither the one nor the other, I did not think this notion had been
adequately articulated when I first took up the issue of nominalism, and
I have not found it adequately articulated in nominalist literature of the
succeeding decades.
Second, as to the popular epistemological arguments to the effect that
even if numbers or other objects ‘‘causally isolated’’ from us do exist, we
cannot know that they do, I have not altered the opinions that I expressed in
my papers Burgess (1989) and the belatedly published Burgess (1998b), and
that Rosen expressed in his dissertation, and that the two of us jointly
expressed in A Subject. The epistemological argument, according to which
belief in abstract objects, even if conceded to be implicit in scientific and
commonsense thought, and even if perhaps true – for the aim of going
epistemological is precisely to avoid direct confrontation over the question
of the truth of anti-nominalist existence claims – cannot constitute knowl-
edge, surely is not intended as a Gettierological observation about the gap
between justified true belief and what may properly be called knowledge. It
follows that it must be an issue about justification; and here to the natu-
ralized anti-nominalist the nominalist appears simply to be substituting
some extra-, supra-, praeter-scientific philosophical standard of justification
for the ordinary standards of justification employed by science and com-
mon sense: the naturalist anti-nominalist’s answer to nominalist skepticism
about mathematics is skepticism about philosophy’s supposed access to
such non-, un-, and anti-scientific standards of justification.
AGAINST FICTIONALIST NOMINALISM
Returning to the issue of fictionalism, in our subsequent work Rosen and

I have generally dealt with it separately and in our own ways. A chapter
bearing the names of Rosen and myself, ‘‘Nominalism reconsidered,’’ does
appear in Stuart Shapiro’s Handbook of Philosophy of Mathematics and Logic
(2005), and it is a sequel to our book adding coverage of fictionalist
nominalism, with special reference to the version vigorously advocated
over the past several years by Steve Yablo; but this chapter is substantially
Rosen’s work, my contributions being mainly editorial.
My own efforts to address a fictionalist position are to be found rather in
‘‘Mathematics and Bleak House,’’ which revisits, in a sympathetic spirit,
Rudolf Carnap’s ideas on the status of ontological questions and nominalist
theses. Neo-Carnapianism is on the rise, and I am happy to be associated
with it, though like any other neo-Carnapian I have my differences with
my fellow neo-Carnapians. ‘‘Quine, analyticity, and philosophy of mathe-
matics’’ can be read as a sequel to the Bleak House paper (it was written much
later, though owing to various accidents both came out in the same year,
2004). It revisits the famous exchange between Carnap and Quine on
ontology, again in a spirit sympathetic to Carnap.
Carnap thought there was a separation to be made between analytic
questions about what is the content of a concept such as that of number,
and pragmatic questions about why we accept such a concept for use in
scientific theorizing and commonsense thought. Quine denied there was in
theory any sharp separation to be made. I argue that there is in practice at
least a fuzzy one. I also argue that Quine had better acknowledge as much if
he is to be able to make any reply to a serious criticism of Charles Parsons.
The criticism is that Quine’s holist conception of the justification of
mathematics – it counts as a branch of science rather than imaginative
literature because of its contribution to other sciences – cannot do justice to
the obviousness of elementary arithmetic.
Though placed in the first half of this volume along with papers about
nominalism, the Quine paper can equally well be read more or less inde-
pendently as a paper in philosophy of language and theory of knowledge
about the notion of analyticity, one that just happens to use mathematics and
logic as sources of examples. The placement of this paper, and more generally
the division of the collection into two parts, should not be taken too seriously.
As any neo-Carnapian will tell you, though Carnap was certainly an
anti-nominalist, his position is perhaps better characterized as generally
anti-ontological rather than specifically anti-nominalist. My own general
anti-ontologism became finally, fully, and emphatically explicit in ‘‘Being
explained away’’ (2005), my farewell to the issue of nominalism. In this
retrospective (written for an audience of undergraduate philosophy con-
centrators) I distinguish what I call scientific ontics, a glorified taxonomy of
the entities recognized by science, from what I call philosophical ontosophy,
an impossible attempt to get behind scientific representations to a God’s-
eye view, and catalogue the metaphysically ultimate furniture of the uni-
verse. The error of the nominalists consists, in my opinion, not in onto-
sophical anti-realism about the abstract, but in ontosophical realism about
the concrete – more briefly, the error is simply going in for ontosophy and
not resting content with ontics.
In taking leave of the issue of nominalism, I should reiterate the point
made briefly at the end of A Subject, that from a naturalist point of view
Introduction 7
there is a great deal to be learned from the projects of Field, Chihara,
Hellman, and others. Naturalists, I have said, hold that there is no
possibility of separating completely the contributions from the world and
the contributions from us in shaping our theories of the world. At most we
can get a hint by considering how the theories of creatures like us in a world
unlike ours, or the theories of creatures unlike us in a world like ours, might
differ from our own theories. The nominalist reconstruals or reconstruc-
tions, though implausible when read as hermeneutic, as accounts of the
meaning of our theories, and unattractive when read as revolutionary, as
rivals competing for our acceptance with those theories, do give a hint of
what the theories of creatures unlike us might be like.
Another hint is provided by those monist philosophers who have recon-
strued what appear to be predicates applying to various objects as predi-
cates applying to a single subject, the Absolute, with the phrases that seem
to refer to the various objects being reconstrued as various adverbial
modifiers. Thus ‘‘Jack sings and Jill dances’’ becomes ‘‘The Absolute
sings jackishly and dances jillishly,’’ while ‘‘Someone sings and someone
else dances’’ becomes ‘‘The Absolute sings somehow and dances other-
how.’’ What is specifically sketched in ‘‘Being explained away’’ is how this
kind of reconstrual can be systematically extended, at least as far as first-
order regimentation of discourse can be extended. Of course it is not to be
expected that we can fully imagine what it would be like to be an intelligent
creature who habitually thought in such alien terms, any more that we can
fully imagine what it would be like to be a bat. Nor insofar as we are
capable of partially imagining what is not wholly imaginable are formal
studies the only aid to imagination. The kind of fiction that stands to
metaphysics as science fiction stands to physics – the example I cite in the
paper is Borges – may give greater assistance.
FOUNDATIONS OF MATHEMATICS: SET THEORY
As long as mathematicians adhere to the ideal of rigorous proof from explicit

axioms, they will face decisions as to which proposed axioms to start from,
and which methods of proof to admit. What is conventionally known as
‘‘foundations of mathematics’’ is simply the technical study, using the tools
of modern logic, of the effects of different choices. Work in foundations
emphatically does not imply commitment to a ‘‘foundationalist’’ philosoph-
ical position, or for that matter to any philosophical position. In Burgess
(1993) I nevertheless argued that work in foundations can be relevant to
philosophy, and tried to explain how. I will not attempt to summarize the
explanation here, except to give this hint: most of the interesting choices of
axioms, especially those that are more restrictive rather than the orthodox
choice of something like the axioms of Zermelo–Frankel set theory, were
originally inspired by positions in the philosophy of mathematics (finitism,
constructivism, predicativism, and others). Foundational work helps us
appreciate what is at stake in the choice among those restrictive philosophies,
and between them and classical orthodoxy.
While the early papers in the first part of this collection are predom-
inantly though not exclusively critical, and the middle papers a mix of
critical and positive – I would say ‘‘constructive,’’ except that this word has
a special meaning in philosophy of mathematics – the last two are, like the
bulk of my more technical work, predominantly though not exclusively
positive. Though they do not endorse as ultimately correct, they present as
deserving of serious and sustained attention three novel approaches to
foundations of mathematics, very different in appearance from each
other, but not necessarily incompatible.
To the extent that there is an agreed foundation or framework for
contemporary pure mathematics, it is provided by something like the
Zermelo–Frankel system of axiomatic set theory, in the version including
the axiom of choice (ZFC). ‘‘E pluribus unum’’ (2004) attempts to combine
two insights, one due to Boolos, the other to Paul Bernays, to achieve an
improved framework.
The idea taken from Boolos is that plural quantification on the order of
‘‘there are some things, the us, such that . . .’’ is a more primitive notion
than singular quantification of the type ‘‘there is a set or class U of things
such that . . .’’ and that Cantor’s transition from the former to the latter was
a genuine conceptual innovation, not a mere uncovering of a commitment
to set- or class-like entities that had been implicit in ordinary plural talk all
along.
Boolos himself had applied this idea to set theory, to suggest, not
improved axioms, but an improved formulation of the existing axioms.
For there is a well-known awkwardness in the formulation of ZFC, in that
two of its most important principles appear not as axioms but as schemes, or
rules to the effect that all sentences of a certain form are to count as axioms.
For instance, separation takes the form
8x9y8zðz 2 y $ z 2 x & fðzÞÞ
wherein f may be any formula. Needless to say, no one becomes con-

vinced of the correctness of ZFC by becoming convinced separately of
Introduction 9
each of infinitely many instances of the separation scheme. But the
language of ZFC provides no means of formulating the underlying single,
unified principle. One proposed solution to this difficulty has been to
recognize collections of a kind called classes that are set-like while somehow
failing to be sets. With capital letters ranging over such entities, and with
‘‘z 2 U ’’ written ‘‘Uz’’ to emphasize that the relation of class membership is
a kind of belonging that is like set-elementhood and yet somehow fails to
be set-elementhood, the separation scheme can be reduce to a single
axiom, thus:
8U 8x9y8zðz 2 y $ z 2 x & UzÞ:
But notion of class brings with it difficulties of its own, leaving many
hesitant to admit these alleged entities.
The suggestion of Boolos (in my own notation) was to replace singular
quantification 8U or ‘‘for any class U of sets . . .’’ over classes by plural
quantification 88uu or ‘‘for any sets, the u’s . . .’’ and Uz or ‘‘z is a member
of U’’ by z / uu or ‘‘z is one of the u’s,’’ thus yielding a formulation in
which the only objects quantified over are sets:
88uu8x9y8zðz 2 y $ z 2 x &z / uuÞ:
One may even take a further step and make the notion x uu or ‘‘x is the
set of the u’s’’ primitive, with the notion y 2 x or ‘‘y is an element of x’’ being
defined in terms of it, as 99uu(x x uu & y / uu) or ‘‘there are some things
that x is the set of, and y is one of them.’’ Such a step was actually taken in
a paper by Stephen Pollard (1996) some years before my own, of which
I only belated became aware, along with Shapiro (1987) and Rayo and
Uzquiano (1999).
The idea taken from Bernays was that an approach incorporating a
so-called reflection principle can provide a simpler axiomatization than
the standard approach to motivating the axioms of ZFC, and permit the
derivation of some further so-called large-cardinal principles that are
widely accepted by set theorists, though they go beyond ZFC. The original
Bernays approach had the disadvantage of involving ‘‘classes’’ over and
above sets, and of requiring a somewhat artificial technical condition in the
formulation of the reflection principle. Boolos’s plural logic was subject to
the objection that, like any version or variant of second-order logic, it lacks
a complete axiomatization. I aim to show how the combination of Boolos
with Bernays neutralizes these objections.
FOUNDATIONS OF MATHEMATICS: LOGICISM
‘‘Logicism: a new look’’ (previously unpublished) provides a concise, semi-

popular introduction to two alternative approaches to foundations each of
which I have examined more fully and technically elsewhere. Each repre-
sents a version of the old idea of logicism, according to which mathematics
is ultimately but a branch of logic. Computational facts such as 2 þ 2 ¼ 4,
on this view, become abbreviations for logical facts; in this case, the fact
that if there exists an F and another and no more, and a G and another and
no more, and nothing is both an F and a G, and something is an H if and
only if it is either an F or a G, then there exists an H and another and yet
another and still yet another, but no more.
One new idea derives from Richard Heck. Frege, the founder of modern
logic and modern logicism proposed to develop arithmetic in a grand
system of logic of his devising. That system is, in modern notation and
to a first approximation, a form of second-order logic, with axioms of
comprehension and extensionality,
9X 8xðXx $ fðxÞÞ
8X 8Y ð8zðXz $ YzÞ ! ðfðX Þ $ fðY ÞÞÞ
supplemented by an axiom to the effect that to each second-order entity X

there is associated a first-order entity X in such a way that we have
8X 8Y ðX ¼ Y $ 8zðXz $ YzÞÞ:
Russell showed that a paradox arises in this system, and also introduced the
idea of imposing a restriction of predicativity on the comprehension axiom,
assuming it only for formulas f(x) without bound class variables. Russell
proposed a great many other changes, and his overall system diverged greatly
from Frege’s. Heck was the first to consider closely what would happen if one
made only the one change just described, and he showed that the resulting
system, though weak (and in particular consistent) is strong enough for the
minimal arithmetic embodied in the system known in the literature as Q to
be developed in it. So a bare minimum of mathematics can be developed on
a predicative logicist basis in the manner of Frege. (More technical details as
to what can be accomplished along these lines are provided in my book
Fixing Frege (Burgess 2005b). Since the book and the paper were written
there have been important advances by Mihai Ganea and Albert Visser.)
Introduction 11
Russell’s version of logicism was opposed by Brouwer’s intuitionism and
also by Hilbert’s formalism. The latter was consciously modeled on instru-
mentalist philosophies of physics, according to which physical theory is a
giant instrument for deriving empirical predictions, though theoretical
terms and laws in physics in general do not admit direct empirical defi-
nitions or meaning. Hilbert’s philosophy of mathematics can be repre-
sented by a simple proportion:
computational : mathematics :: empirical : physics.
For Hilbert, ‘‘real’’ mathematics consists of basic computational facts

like 2 þ 2 ¼ 4, and the rest of mathematics is merely ‘‘ideal,’’ with an
instrumental value for deriving computational results, but no direct com-
putational meaning. My late colleague Dick Jeffrey proposed instead that
one should think of mathematics as being logical in the same sense in
which physics is empirical: the data of mathematics are logical, as those
of physics are empirical, though there can be no question of defining
all mathematical notions in strictly logical terms, or all physical notions
in strictly empirical ones. Mathematics becomes, on this view, a giant
engine for generating logical results, as physics is a giant engine for
generating empirical results. Hilbert’s proportion is modified by Jeffrey
so that it reads thus:
logical : mathematics :: empirical : physics.
The connection between the Jeffrey idea and the Heck idea is that
predicative logicism provides enough mathematics to connect the basic
computational facts that figure in the Hilbert proportion with the logical
facts that figure in the Jeffrey proportion. So, though predicativist logicism
falls far short of the whole of mathematics it would be possible to regard it
as providing the data for mathematics.
The question I raise about all this is whether the engine is doing any real
work: do sophisticated mathematical theories (such as standard Zermelo–
Frankel set theory or the proposed Boolos–Bernays set theory) actually
make available any more logical ‘‘predictions’’ in the way sophisticated
theories in physics make available more empirical predictions? I show how
(and in what sense) certain results of Julia Robinson and Yuri Matiyasevich
in mathematical logic yield an affirmative answer to this question. (More
technical details are provided in a forthcoming paper ‘‘Protocol Sentences
for Lite Logicism.’’)
MODELS AND MEANING
‘‘Tarski’s tort’’ (the other previously unpublished item in the collection) is a

sermon on the evils of confusing, under the label ‘‘semantics,’’ a formal or
mathematical theory of models with a linguistic or philosophical theory of
meaning. Tarski’s infringing on the linguists’ trademark ‘‘semantics,’’ and
transferring it from the theory of meaning to the theory of models, encour-
ages such confusion, which has several potentially bad consequences.
Such a confusion may lead, on the one hand, to erroneous suspicions
that many ordinary locutions involve covert existential assumptions about
dubious entities. (For example, one may fall into a fallacy of equivocation
and argue that since possible worlds are present in the model theory of
modal logic, they are therefore present in the semantics of modality, and
are therefore present in the meaning of modal locutions.) Such a confusion
may lead, on the other hand, to unwarranted complacency about the
meaningfulness of dubious notions. (For example, one may fall into a
different fallacy of equivocation and argue that since quantified modal
logic has a rigorous model theory, it therefore has a rigorous semantics, and
therefore has a rigorous meaning.)
Confusion of models and meaning under the label ‘‘semantics’’ may also
give undeserved initial credibility to truth-conditional theories of meaning,
through their mistaken association with the prestigious name of Tarski.
One point on which I think Dummett is entirely right is rejection of the
truth-conditional theory of meaning, and insistence that meaning must be
explained in terms of rules of use, though my reasons for holding this view
are rather different from Dummett’s. (For me, perhaps the most incredible
feature of the truth-conditional theory is the assumption that truth is an
innate idea, possession of which is a prerequisite for all language-learning.
I find much more plausible the suggestion that the idea of truth is acquired
at the time the word ‘‘true’’ is acquired, and that acquisition of the idea
consists in internalizing certain rules for the use of the word.)
Such a view leads naturally to the ‘‘inconsistency theory’’ of truth advo-
cated in different forms by my teacher Charles Chihara, my student John
Barker, and (not under any influence of mine) my son Alexi Burgess. There
being by Church’s theorem no effective test for the inconsistency of rules, it
would be a miracle if all the rules we ever internalized were consistent. The
simplest and most natural rules for the use of ‘‘true,’’ permitting inference
back and forth between p and it is true that p, are inconsistent. Acceptance
that these inconsistent rules of use are the ones we internalize when we acquire
the word ‘‘true’’ and therewith the idea of truth provides the simplest and
Introduction 13
most natural explanation of the intractability of the liar and related para-
doxes, an explanation favored not only by the persons already mentioned,
but by Tarski himself. All these issues are touched on in ‘‘Tarski’s tort,’’
though none is argued in all the detail it deserves. (Some of the issues were
previously aired in Burgess (2002b) and elsewhere.) The warnings about
fallacies of equivocation are pertinent to other papers in the second part of
the collection, which is one reason for placing this paper first in that part.
MODELS AND MODALITY
Nothing is more important in approaching modal logic than to bear

constantly in mind the distinction between two kinds of necessity: meta-
physical necessity – what could not have been otherwise – and logical
necessity – what it would be self-contradictory to deny. Modal logic has
been characterized from early on by a great proliferation of systems, even at
the level of sentential logic. In a way this is all to the good, since different
conceptions of modality, once one has learned to distinguish them, may
call for different formal systems. But we do need to distinguish the differ-
ent notions before we can meaningfully ask which system is right for which
notion. And though philosophers and logicians nowadays are more aware
of the distinctions among various kinds of modality than they were
formerly, the problem of determining which formal system is appropriate
for which conception of modality is one that still has received surprisingly
little attention.
In ‘‘Which modal logic is the right one?’’ (1999) I take up this question for
the case of the original conception of necessity of C. I. Lewis, the founder of
modern modal logic, for whom necessity was logical. And the first thing that
needs to be said about logical modality is that it comes in two distinguishable
kinds, a ‘‘semantic’’ or model-theoretic notion of validity and a ‘‘syntactic’’ or
proof-theoretic notion of demonstrability. (For first-order logic demon-
strability and validity coincide in extension, by the Gödel completeness
theorem, but they are still conceptually distinct; for other logics they need
not coincide even in extension.) The former makes necessity a matter of truth
by virtue of logical form alone, the latter a matter of verifiability by means of
logical methods alone. The common conjecture is that the system known as
S5 is the correct logic for the former, and that known as S4 for the latter. The
well-known Kripke model theory for modal logic is a useful tool, but hardly
in itself provides a complete proof of either conjecture. (As the originator of
this model theory said, ‘‘There is no mathematical substitute for philoso-
phy.’’) As it happens, the conjecture about S5 admits a fairly easy proof,
which I expound, while for the conjecture about S4 only partial results are
available, which I explore.
Turning from logical to metaphysical necessity, no single tool is more
useful for understanding the logical aspects of the latter than the analogy
between mood and modal logic on the one hand, and tense and temporal
logic on the other. One of the older puzzles about metaphysical modality is
Frederic Fitch’s paradox of knowability, which purports to demonstrate
the incoherence of the view that anything that is true could be known.
‘‘Can truth out?’’ (2006) first considers what light can be shed on this
puzzle by looking at a temporal analogue, and then by applying Arthur
Prior’s branching-futures logic, in which modal and temporal elements are
combined. As with the previous paper, a certain amount of progress is
possible, but a complete solution remains elusive.
MODALITY AND REFERENCE
Logical necessity was originally what modal logicians had generally meant
the box symbol to represent, and I believe that Quine was entirely correct
in asserting that with that understanding of the symbol, quantifying into
modal contexts makes no sense. Quine’s complaint was that to make non-
trivial sense of 9x&Fx one must make sense of so-called de re modality, of
the notion of an open sentence Fx being necessarily true of a thing, and this
is impossible (or at any rate, has not been done by the proponents of
quantified modal logic) if ‘‘necessarily true’’ is to mean ‘‘true by virtue of
meaning.’’ For a thing, as opposed to an expression denoting a thing, has
not got a meaning for anything to be true by virtue of. Truth by virtue of
meaning is an inherently de dicto notion, applicable to closed sentences.
Quine underscored his point by illustrating the difficulty of reducing de re
to de dicto modality. One can’t say &Fx is true of the object b if and only
if &Ft is true, where t is a term denoting b, because &Ft may be true for
some terms denoting the object and false for other terms denoting the
same object.
The early response of modal logicians to Quine’s critique, by which
I mean the responses prior to Kripke’s ‘‘Naming and necessity’’ (1972),
involved an appeal, not to a distinction between metaphysical and logical
modality, but rather to (purely formal ‘‘semantics’’ and/or to) the magical
properties of Russellian logically proper names. In simplest terms, the
‘‘solution’’ would be that &Fx is true of b if and only if &Fn is true,
where n is a ‘‘name’’ of b, it being assumed that if &Fn is true for one
name it will be true for all. I believe this line of response is a total failure,
Introduction 15
and anyone acquainted with the views of Mill ought to have been aware
that it must fail. For it will be recalled that Mill, before Russell, held that
a name has only a denotation, not a connotation. He also held, like the modal
logicians who identified & with truth by virtue of meaning, that all necessity
is verbal necessity, deriving from relations among the connotations of
words. What those responding to Quine ought to have remembered is
that, having committed himself to those two views, he inevitably found
himself committed to a third view, that there are no individual essences: ‘‘all
Gs are Fs’’ may be necessarily true because being an F may be part of the
connotation of G, but ‘‘n is an F ’’ cannot be necessarily true, because n has
no connotation for being an F to be part of. Positing Millian/Russellian
‘‘names’’ may permit the reduction of de re to de dicto modality, where the
relevant dicta involve such names, but only at the cost of depriving de dicto
modality, where again the dicta involve such names, of any sense – so long
as one continues to read & as truth by virtue of meaning.
What is now the fashionable view evaluates the early responses to Quine
much more positively than I do, when it does not outright read Kripke’s
ideas back into earlier texts. That Quine was right as against his early critics
is the view that I, going against fashion, defend in ‘‘Quinus ab omni naevo
vindicatus’’ (1998). The origin of the curious title is explained in the article.
Closely linked with the issue of metaphysical versus logical necessity is
the question of the status of identities linking proper names, as in
‘‘Hesperus is Phosphorus.’’ This question has been the topic of an immense
body of literature in philosophy of language. Like Kripke, I on the one
hand reject descriptivist theories of proper names, but on the other hand
equally reject ‘‘direct reference’’ theories. I am attracted to a third view
based on distinguishing two senses of ‘‘sense,’’ mode of presentation versus
descriptive content, a view rather tentatively (and certainly non-polemically)
put forward in connection with Kripke’s Puzzling Pierre problem in
‘‘Translating names’’ (2005).
But returning for a moment to ‘‘Quinus,’’ since some readers like nothing
better than polemics between academics, and others like nothing less, all
potential readers should be informed in advance that ‘‘Quinus’’ is as polem-
ical as anything I have ever written (though much of the polemic is relegated
to footnotes), and several degrees more so than anything else in this collec-
tion. Even the explanation why the paper is polemical must itself inevitably
be somewhat polemical, and so I will relegate it to a parenthetical paragraph
which those averse to polemics may skip, along with the paper itself.
(My paper is explicitly a response to a paper by Ruth Barcan Marcus
from the early 1960s, but it is also implicitly a response to a widely
circulated letter by the same writer from the middle 1980s, discussed in the
editorial introduction to Humphreys and Fetzer (1998). This letter is the
original source for the claim that Kripke’s ideas were taken without
acknowledgment from the early Marcus paper. The response to such
allegations is that Kripke could not have stolen his ideas from the indicated
source, since neither those important and original contributions nor any
others were present there to be plagiarized. In my paper I do not mince
words in presenting this defense. Marcus’s insinuations are more directly
addressed in my paper (Burgess 1996). The contents of her letter, with
elaborations but without acknowledgment of that letter as a source, reap-
pear in the work of one of its many recipients, Quentin Smith. His version
is addressed in my paper (Burgess 1998a).)
HERMENEUTIC CRITICISM OF CLASSICAL

LOGIC: RELEVANTISM
In Burgess (1992) I offered qualified defense of classical logic, leaving

plenty of room from additions and amendments once one moves beyond
the realm of mathematics. I introduced in the paper a distinction that
seems to me crucial in evaluating certain criticisms of classical logic,
namely, the distinction between prescriptive criticism, according to
which classical logicians have correctly described the incorrect logical
practices of classical mathematicians, and descriptive criticism, which
maintains that classical logicians have incorrectly described the correct
logical practices of classical mathematicians. The distinction parallels that
between revolutionary and hermeneutic nominalism. The remaining two
items in the collection discuss examples of the two types of anti-classical
logics. (Both have modal aspects or admit modal interpretations, and
certainly can be studied by some of the methods, notably Kripke models,
developed for modal logic, and insofar as this is so may be squeezed
under the ‘‘models and modality’’ heading for the second part of the
collection.)
Much descriptive criticism of classical logic, especially that from the old
ordinary language school, was essentially anti-formal. The best example of
descriptive criticism in the service of a rival formal logic was provided by the
‘‘relevance’’ logic of A. R. Anderson and Nuel D. Belnap, Jr. in its original
form, back when it was a more or less unified philosophical school of
thought, denouncing and deriding classical logic, and recommending one
or the other of two specific candidates, the systems E and R, as replacement.
The enterprise has since been renamed ‘‘relevant’’ logic and devolved into the
Introduction 17
study of a loose collection of formal systems linked by family resemblances,
with quite varied intended or suggested applications.
‘‘Relevance: a fallacy?’’ (1992) was devoted to presenting counterexamples
to one key early claim of relevance logic, never up to the time I was writing
explicitly retracted in print. Relevantists rejected, as ‘‘a simple inferential
mistake,’’ the inference from A ~ B and A to B. But inference from ‘‘A or
B’’ and ‘‘not A’’ to B is common in classical mathematics and elsewhere.
Anderson gave little attention to the resulting tension, but Belnap attemp-
ted to resolve it by claiming that ‘‘or’’ in ordinary language generally means
not the extensional ~ but some intensional þ. It was this specific claim
I challenged. I do not have a globally negative view of the relevantist enter-
prise. Not only has there been some impressive technical work by Saul
Kripke, Kit Fine, Alasdair Urquhart, Harvey Friedman, and others, but
the ‘‘first degree’’ and ‘‘pure implicational’’ fragments of several relevantist
systems do have coherent motivations, as do various related logics, such as
Neil Tennant’s idiosyncratic version of relevantism or ‘‘logical perfection-
ism.’’ I do not, however, think either of the main relevantist systems E or R
as a whole has any coherent motivation, and more importantly, I do not
think there was any merit in the original relevantist criticism and caricature
of classical logic.
My paper prompted replies in the same journal by Chris Mortensen and
Stephen Read, to which I in turn responded, again in that journal.
Ironically, though members of a group who pride themselves on their
sense of ‘‘relevance,’’ the two writers who directly replied to my paper
simply could not confine themselves to addressing the specific issue I had
treated, but insisted on offering, as if this somehow refuted my claims,
expositions of motivations for relevantism quite different from those of
Anderson and Belnap (and quite different from each other). This led me to
write, ‘‘The champion of classical logic faces in relevantism not a dragon
but a hydra.’’ It also led me to reiterate my original point and present a
variety of further counterexamples, drawn from several sources. The
polemics do not seem to me worth reprinting, but I do refer any reader
not convinced by the two examples in my original paper to the first list of
examples (borrowed from authors ranging from E. M. Curley to Saul Kripke)
in x2 of the first of my replies to critics (Burgess 1983c).
My second reply (Burgess 1984b) contained one more example: by the
regulations of a certain government agency, a citizen C is entitled to a
pension if and only if C either satisfies certain age requirements or satisfies
certain disability requirements. An employee Z of the agency is presented
with documents establishing that C is disabled. Z transmits to fellow
employee Y the information that C is entitled to a pension (i.e. is either
aged or disabled). Y subsequently receives from another source the infor-
mation that C is not aged, and concludes that C must be disabled.
REVOLUTIONARY CRITICISM OF CLASSICAL

LOGIC: INTUITIONISM
While it may be contentious whether relevantism really does provide an

example of descriptive criticism of classical logic, it is beyond controversy
that intuitionism provides an example of prescriptive criticism. Adherence
to intuitionistic logic would certainly require major reforms in mathe-
matics. Very likely acceptance of the verificationist concerns that motivate
contemporary intuitionism would require still more dramatic reforms of a
nature we cannot yet quite take in, in empirical science. For in the
empirical realm we have to contend with two phenomena that do not
arise in mathematics: first, we generally have to deal not with apodictic
proof but with defeasible presumption; second, it may happen that though
each of two assertions may be potentially empirically testable, performing
the operations needed to test one may preclude performing the operations
needed to test the other. (A DNA sample, for instance, may be entirely used
up by whichever of the two tests we choose to perform first; nothing
analogous ever happens with operations on numbers.) The ultimate veri-
ficationist logic may well have to combine features of intuitionistic, non-
monotonic, and quantum logics.
Michael Dummett is perhaps the single most influential representative
of anti-naturalism in contemporary philosophy, though his contributions
extend far beyond that role. (He is, among many other things, the leader of
Frege studies, and in that capacity motivated, among many other things,
the exploration of the predicativist variant of Frege’s theory discussed in the
last paper of Part I.) His paper (Dummett 1973a) inaugurated a new era in
the classical-intuitionist debate over logic and mathematics, and was the
font from which what is now a vast stream of ‘‘anti-realist’’ literature first
sprang. A noteworthy feature of Dummett’s approach is that, like Brouwer
but unlike almost every writer on intuitionism in-between, he takes the
considerations that motivate intuitionism to apply not just to mathematics
but to all areas of discourse. Mathematics is special only in that we have a
better idea of what a revision of present practice would amount to in that
area than in any other.
For that reason I have placed ‘‘Dummett’s case for intuitionism’’ in this
part of the collection, rather than the part on philosophy of mathematics
Introduction 19
specifically. But to repeat what I said in the introduction to Part I, the
division of the collection into parts is not to be taken too seriously.
‘‘Dummett’s case’’ advances two kinds of countercriticisms of the criticism
of classical logic that appears in Dummett (1973a). In the first part of the
paper, making an explicit mention of Noam Chomsky and an implicit
allusion to John Searle, I object to Dummett’s unargued behaviorist assump-
tions. In this part of the paper I am arguing as devil’s advocate, since I do in
the end agree with Dummett in rejecting truth-conditional theories of mean-
ing. Some followers of Dummett have objected to the label ‘‘behaviorist’’; but
I think this is largely a terminological issue. If I say that Sextus, Cicero,
Montaigne, Bayle, and Hume were all skeptics, I do not imply that their views
were identical; likewise if I say that Watson, Skinner, Quine, Ryle, and
Dummett are all behaviorists. It is clear that Dummett’s position, however
one labels it, remains light-years away from Searle’s, let alone Chomsky’s.
What I object to in the second half of the paper is the lack of explicitness
about how the transition from is to ought, from the premise that a truth-
conditional theory of meaning is incorrect for classical mathematics to the
conclusion that classical mathematics ought to be revised, is supposed to be
made. Hints are thrown out, to be sure, which have been developed in
different ways by Dummett himself in later works (especially The Logical
Basis of Metaphysics) and in more formally terms by Dag Prawitz, Niel
Tennant, and others. The most explicit proposals suggest that the mean-
ings of logical operators should be thought of as constituted by something
like the introduction and/or elimination rules for those operators in a
natural deduction system. Then intuitionistic logic is claimed to be better
than classical logic because there is a better ‘‘balance’’ between its intro-
duction and elimination rules.
In ‘‘Dummett’s case’’ I largely confine myself to saying that this aesthetic
benefit of ‘‘balance’’ could hardly outweigh considerations related to the
needs of applications, objecting to Dummett’s apparent indifference to the
latter in advocating revision of mathematics. How much of the standard
mathematical curriculum for physicists or engineers can be developed
within this or that restrictive finitist, constructivist, predicativist, or what-
ever framework is a topic that has been fairly intensively investigated by
logicians (though to this day I know of no serious study of how a finitist,
constructivist, or predicativist should interpret mixed contexts involving
both mathematical and physical objects and properties, something nomi-
nalists by contrast take to be of central concern). Dummett simply omits to
address this issue at all in his key paper, and that omission would, I think,
raise the eyebrows of any logician.
I have recently written a sequel to ‘‘Dummett’s case’’ (Burgess 2005e),
but on the advice of an anonymous referee have decided not to include it
here, as its tone may in places offend Dummettians (even more so than that
of my older paper, if that is possible). The main point of the paper still
seems to me worth making: neither the view that the meaning of each
logical particle is constituted by its introduction rule (for example, the rule
‘‘from A and B to infer A & B’’), nor the view that it is constituted by the
introduction rule together with the corresponding elimination rule (for
example, ‘‘from A & B to infer A and to infer B’’) is tenable if ‘‘meaning’’ is
supposed to be what guides use. For we constantly ‘‘introduce’’ and ‘‘elim-
inate’’ conjunctions, say, by quite other means than conjunction introduc-
tion and elimination (for instance, we may arrive at a conjunction by
universal instantiation and by modus ponens, which is to say, by the
elimination rules for the universal quantifier and for the conditional).
And these steps cannot be justified from the point of view of one who
really, truly, and sincerely takes the sole direct guide to the use of ‘‘&’’ to be
the introduction and/or elimination rules for that connective. Nor – and
this is the crucial point of the paper – can any metatheorem provide an
indirect justification, since the proof of the metatheorem inevitably
involves introducing and eliminating conjunctions according to rules
that, at least until the proof of the metatheorem is complete, have not yet
been justified from the point of view in question. (Moreover, the applica-
tion of any metatheorem to any particular case would anyhow require
universal instantiation and modus ponens.)
More generally, my paper points out how frequently writers (including
not only neo-intuitionists, but relevantists and nominalists) who profess to
reject certain principles of classical logic (or mathematics), and appeal to
metatheorems supposed to show that nothing much is lost thereby, can be
caught using the supposedly rejected principles in the proofs of those very
metatheorems. To my mind, this is one striking illustration of how
difficult it is to be a genuine dissenter from classical logic and mathematics.
PART I
Mathematics
1
Numbers and ideas
1 REALISM VS NOMINALISM
Philosophy is a subject in which there is very little agreement. This is so

almost by definition, for if it happens that in some area of philosophy
inquirers begin to achieve stable agreement about some substantial range of
issues, straightaway one ceases to think of that area as part of ‘‘philosophy,’’
and begins to call it something else. This happened with physics or ‘‘natural
philosophy’’ in the seventeenth century, and has happened with any
number of other disciplines in the centuries since. Philosophy is left with
whatever remains a matter of doubt and dispute.
Philosophy of mathematics, in particular, is an area where there are very
profound disagreements. In this respect philosophy of mathematics is
radically unlike mathematics itself, where there are today scarcely ever
any controversies over the correctness of important results, once published
in refereed journals. Some professional mathematicians are also amateur
philosophers, and the best way for an observer to guess whether such
persons are talking mathematics or philosophy on a given occasion is to
look whether they are agreeing or disagreeing.
One major issue dividing philosophers of mathematics is that of the
nature and existence of mathematical objects and entities, such as numbers,
by which I will always mean positive integers 1, 2, 3, and so on. The problem
arises because, though it is common to contrast matter and mind as if the
two exhausted the possibilities, numbers do not fit comfortably into either
the material or the mental category.
Clearly numbers are not material bodies. The so-called numbers on the
front of a house, marking its street address, may indeed be made of brass or
wood or plastic. But these ‘‘numbers’’ are not the numbers we speak of when
we say that two is an even number, or that three is an odd number, or that
both are prime numbers. Rather, they are numerals, or names of numbers.
Almost equally clearly, numbers are not mental in the way that, say,
dreams or headaches are. They are not private to an individual. One does
23
not speak of my number two and your number two, his number two and
her number two, but simply of the number two. The individual, say a
school child doing a simple sum, experiences the numbers as something
external, about which he or she is not free to think whatever he or she wants.
But if numbers are not material bodies or private experiences, what (if
anything) are they? Among professional academic philosophers, which is to
say university professors of the subject, the most commonly held views are
two, for want of better terms called realism and nominalism.
Realism maintains that numbers exist, and are of a very different nature
from human ideas: indeed, they differ quite as much from human ideas as they
do from material bodies. They are abstract entities, to which it makes no sense
to ascribe a position in space or date in time, and which are not causally active
or acted upon. There is nowhere to go to look for a number, and you cannot
do anything to a number, any more than a number can do anything to you.
Nominalism maintains that numbers do not exist, and that theorems of
mathematics asserting the existence of numbers are untrue, just like fairy
tales asserting the existence of gnomes. To be sure, much of mathematics is
applicable in science and everyday life in a way that fairy tales generally are
not, but that, according to nominalists, only shows it is a useful fiction, not
that it is non-fiction.
There are problems for both opposing philosophical views, and the
problems of each are cited by the adherents of the other as reasons for
embracing it instead. And formerly there were among philosophers also
many who maintained a third view, conceptualism or idealism, according to
which numbers exist, but only as shared human concepts or ideas.
The view has traditionally been popular among anthropologists and
other social scientists, whose special subject matter is precisely the shared
ideas of a culture. They point out that taking numbers to be such shared or
communal ideas sufficiently explains why the school child doing a simple
sum does not feel free to make up an answer at will. If numbers are ideas
shared by a culture, no one member of that culture has the authority to
change the rules of addition, any more than to change the rules of grammar
of the culture’s language.
The anthropological view has also found adherents among mathematics
educators. Rather more surprisingly, the same view has won adherents
among the minority of professional mathematicians who are also amateur
philosophers.1
1
The classical expression of the anthropological view is that of White (1947). For a recent endorsement by
a mathematician, see Hersh (1997), a book that makes a professional philosopher’s hair stand on end.
Numbers and ideas 25
Conceptualist and idealist views, however, were subjected along with
other nineteenth-century views to a scathing critique by the late nineteenth-
century German mathematician and philosopher Gottlob Frege.2 Largely as
a result of that critique, the anthropological view today has virtually no
adherents among professional academic philosophers. Its rejection is one of
the rare cases of general agreement and consensus on an issue in philosophy.
Precisely because there is such general agreement, philosophers seldom
stop to explain, in language more modern than Frege’s, just what is wrong
with the view that so many anthropologists, sociologists, psychologists,
mathematics educators, and even mathematicians have found attractive. It
is this task of explanation that I will be undertaking in the present essay,
using an example of a kind that definitely would not have been used by Frege.
2 BIGFOOT
Let us begin by considering the proposition that Bigfoot, also known as the
Sasquatch – a cousin of the Abominable Snowman or Yeti – exists in the
realm of shared human ideas and concepts. Now certainly there is some-
thing in the neighborhood that exists in the realm of shared human ideas
and concepts, namely, the shared human idea or concept of Bigfoot. This is
the idea of a large, hairy, humanoid creature inhabiting the wilder parts of
the Pacific Northwest, from northern California to British Columbia.
There are even people who claim to have sighted individual Bigfeet, and
to have formed ideas of these individuals, even to the point of giving them
names like ‘‘Harry’’ or ‘‘Harriet.’’ The idea of an individual Bigfoot
includes the traits that are common to all Bigfeet according to the general
idea of Bigfoot, but also more specific elements: for instance, Harry is male
and Harriet is female. These ideas of individual Bigfeet are less widely
shared than the idea of the species, but we may suppose they are at least
shared among members of the International Society for Cryptozoology,
who take a special interest in such things.
The majority view among zoologists is that there do not, in fact, exist any
large, hairy, humanoid creatures, and that the alleged sightings of Harry,
Harriet, and other individual Bigfeet were either illusions or hoaxes. But I ask
you to join me in assuming, just for the moment, that the majority is wrong,
and that creatures of the kind indicated, including Harry and Harriet, do
exist. On this assumption, I will argue, two things should be clear.
2
See Frege (1884), English translation by Austin (1960). The critical portions (the part of the book
relevant to the present essay) are reprinted in Benacerraf and Putnam (1983).
The first is that Harry, Harriet, and other large, hairy, humanoid
creatures inhabiting the wilder parts of the Pacific Northwest are very
different sorts of things from shared human ideas and concepts, and in
particular are very different sorts of things from the ideas and concepts of
Harry, of Harriet, and of Bigfoot in general. They differ in absolutely
fundamental respects, for instance, in their location in space and time.
Let us consider space, for instance. (Similar considerations would apply
to time.) It is not clear whether or where a shared human idea or concept
should be thought of as located in space, but presumably if it is located
anywhere, it is located where the human beings who share it are located.
Thus if the International Society for Cryptozoology holds its annual
convention on the banks of Loch Ness, the idea of Bigfoot in general,
and the ideas of Harry and Harriet in particular, are located mainly in
Scotland. Harry, Harriet, and the rest of their kind, however, are still
located in Washington or Oregon or thereabouts. The creatures cannot
be the ideas, because the two are located in different places.
The creatures differ from the ideas also in respect of how many of them
there are. People have ideas of Harry, Harriet, and several more Bigfeet that
have allegedly come into contact with human beings; but there are supposed to
be, according to the minority view I have asked you to assume for the moment,
more Bigfeet than just these: more individuals like Harry and Harriet than
there are shared human ideas of individual Bigfeet. So again the creatures
cannot be the ideas, since there are more of the former than of the latter.
A second point I hope will be clear is that it is the flesh-and-blood
creatures, not the ideas, that are the Bigfeet. The term ‘‘Bigfoot’’ refers to
the inhabitants of the wilds of Washington and Oregon, not to the contents
of the minds or brains of the cryptozoologists assembled in Scotland. If we
wish to refer to the latter, we must use some other expression than the word
‘‘Bigfoot,’’ such as the phrase ‘‘the idea of Bigfoot.’’
In short, on the minority view, according to which the flesh-and-blood
creatures do exist, the following is the case: Bigfeet, being flesh-and-blood
creatures, are not ideas, and are more numerous than the ideas of them and
located in a different place from those ideas.
3 NUMBERS
Are things any different on the majority view? It is when one assumes that
there are no such flesh-and-blood creatures that some are tempted to say
that the Bigfoot in general, or Harry and Harriet in particular, are human
ideas. I think this temptation should be rejected.
Let me say straightaway that it would be pointless to object to someone
expressing disbelief in Bigfoot by saying, ‘‘Bigfoot exists only in the imagina-
tion of the credulous,’’ or something of the sort. Someone might well say this –
I might well say it myself, for that matter, when not talking philosophy – and
mean it only as a manner of speaking, as a way of saying ‘‘Bigfoot doesn’t exist
at all, though some credulous persons imagine that it does.’’ The proposition I
want to consider, however, is that Bigfoot literally does exist, but only in the
realm of shared human ideas and concepts, where, according to the anthro-
pological view, numbers also have their being.
To indicate the reasons why I reject this proposition, suppose the
population of some endangered forest or swamp species falls until there
is only one left. So long as this one surviving flesh-and-blood or wood-
and-sap organism lives, considerations of the kind already adduced in the
case of Bigfoot indicate that it is the only member of the species, and it is
not an idea, from which it follows that the members of the species are
not ideas.
Now suppose this last survivor also perishes. Are we now to say that the
species still has members, but that the members of the species are now
ideas? Should we say that the species has not become extinct but rather has
undergone a metamorphosis, transcending its former carnal or xyline
nature, and taken on a conceptual essence: that its members have cast
aside their fleshly or wooden bodies, and are now made of whatever ideas
are made of? Should we say that the species has undertaken a migration,
abandoning the woods or marshes that were once its home, and occupying
now instead a niche in the minds or brains of human subjects?
It seems to me about as plain as anything can be in philosophy – where
admittedly things are never as plain as they are in some other disciplines –
that this is not what we should say, and that the correct way to describe the
situation is by saying that creatures of this animal or plant species simply no
longer exist at all, though of course human ideas about them do exist, and
may perhaps continue to exist as long as the human species does.
Likewise in the case of Bigfoot. If the forest creature exists, then Bigfoot is
that forest creature, and is something very different from an idea. If the forest
creature does not exist, then Bigfoot is, so to speak, even more different from
an idea: for in that case Bigfoot is nothing, while the idea is at least something,
and what could be more different than something and nothing?
The case is the same, I maintain, with our shared human ideas and
concept of number in general, and of individual numbers such as one or
two or three. (Again the individual ideas contain whatever is contained in
the general idea, plus additional distinguishing elements. We no longer
imagine, as did the Pythagoreans, that two is female and three is male, but,
for instance, two is even and three is odd.)
These ideas are clear enough, I maintain, to indicate that one, two, three,
and the other numbers, if they exist at all, do not have the same sort of
spatial or temporal features as human ideas, and above all are more
numerous than human ideas could possibly be.
Taking first issues of time and place, mathematics is used throughout
science, and mathematical objects and entities are referred to in all its
branches, including those like cosmology that deal with times and places
very remote from any inhabited by human beings. Are we to say that a
cosmologist’s estimates of the relative numbers of heavy and light elements
at a certain stage in the early evolution of the universe must be wrong,
because there were no numbers at all back then, no human beings having
yet evolved to create them? Surely not.
And then there is the matter of infinity. It is a crucial feature of the
concept of the number system that it has infinitely many elements, that
there are infinitely many numbers. But surely human beings have formed
ideas or concepts of only finitely many of them. There simply are not
enough human ideas and concepts for each number to be one. Some
numbers at least must therefore either enjoy a mode of existence different
from that of any human idea, as realists maintain, or else must simply fail to
exist, as nominalists hold. And is it not preposterous to maintain that while
one of the pair realism or nominalism gives the correct account of mathe-
matical existence in the case of some numbers, conceptualism is correct for
the rest? Surely the question of the existence and nature of numbers has a
uniform answer, and if conceptualism fails in any case, then it must fail in all.
4 REALISM VS NOMINALISM REVISITED
Such, then, are some of the principal reasons why I and almost all profes-
sional philosophers of mathematics reject conceptualism, and consider the
only real issue to be that between nominalism and realism. This last issue is
far too large to be thrashed out here, but I do wish to say a word about it,
and in particular about the character of the realist position, which very often
tends to be misrepresented. Nominalists do not believe in numbers because
they cannot see them (or see any visible effects caused by them), and tend to
represent their opponents as claiming that they can see them.
According to an old story, Plato was once lecturing in his Academy on
his Forms, and was speaking of the forms of ‘‘tableness’’ and ‘‘cupness.’’
Diogenes the Cynic interrupted and said, ‘‘O Plato, I see the table and the
cup, but the tableness and the cupness I do not see.’’ To this Plato replied,
‘‘Very naturally, Diogenes, since you have eyes, by which material things
are perceived, but lack Intellect, by which the forms are seen.’’3
Nominalists tend to represent their opponents as Platonists, maintain-
ing that if numbers do not emit electromagnetic radiation to which the eye
is sensitive, then they must be emitting something else, perhaps noetic rays,
which can be sensed by some other organ, perhaps the pineal gland. This,
however, is a misrepresentation of realism. Or at least, I have never known
a single realist who was in any meaningful sense a Platonist.
What is actually the case is that anti-nominalists take much more
seriously than nominalists the thought that mathematics is a human crea-
tion, since mathematics is a body of theory expressed in language, and
language is a human creation.
Now creating a language involves creating certain rules for its use.
Among these is, I believe, a rule to the effect that tense and date are not
to be applied to mathematical existence assertions. One can say ‘‘There
exist infinitely many prime numbers,’’ but to ask ‘‘How many of them
already existed in 1000 B C E , or during the Cenozoic Era?’’ is to commit a
kind of grammatical solecism.
Nominalists say they are opposed to the view that numbers are ‘‘eternal,’’
existing ‘‘outside of time.’’ But to say that numbers are ‘‘eternal’’ is a
misleadingly Platonistic way of putting the simple negative grammatical
fact of the inapplicability of tense distinctions in mathematical contexts.
That simple grammatical point is all the realist really believes about the
‘‘timelessness’’ of number.
(By contrast with the case of the numbers themselves, it makes perfect
sense to ask whether the idea or concept of prime number had emerged by
1000 B C E – the issue involved would be that of the interpretation of certain
Babylonian tablets and Egyptian papyri – and it makes perfect sense to
assert that it had not emerged in the Age of the Dinosaurs. This difference
between the ‘‘timeless’’ numbers proper and datable ideas of them was one
of the points I was arguing in rejecting conceptualism.)
Likewise, there are certain rules or standards as to what counts as
adequate or sufficient to establish or prove a mathematical existence
theorem, and by these rules Euclid’s theorem on the existence of infinitely
many prime numbers is as well-established as anything can be.
The nominalists assume that they have an understanding of what it
would be for a mathematical object or entity to exist that is independent
3
See the life of Diogenes the Cynic in Diogenes Laertius (1925).
of ordinary mathematical standards of sufficient proof, by reference to
which understanding they can criticize the ordinary mathematical stand-
ards. So-called realism is really just skepticism about the existence of any
understanding of what ‘‘existence’’ means in mathematics that is independ-
ent of ordinary mathematical standards for evaluating existence proofs.
The nominalist denies the existence of numbers, while the realist denies
that the nominalist understands what is meant by ‘‘existence’’ as applied to
numbers.
Thus the realists think the nominalists are confused. But realists and
nominalists agree that the conceptualists are confused, and while I cannot
hope to have convinced anyone by the foregoing very brief remarks that the
realists are right as against the nominalists, I hope I have convinced some of
you that realists and nominalists are right in their common opposition to
conceptualism.
2
Why I am not a nominalist
INTRODUCTION
The sum of the divisors of 220 is 284, and the sum of the divisors of 284 is
220. The Pythagoreans spoke of numbers so related as being amicable. I do
not know how this ancient teaching should be taken, but surely nobody
nowadays, except perhaps a stray numerologist or two, would imagine that
numbers are literally capable of forming friendships. A number is just not
the sort of thing that can enjoy a social life. And this is but the least of a
number’s lacks.
A number lacks a position in space, such as tables, chairs, and other
material bodies possess. It lacks dates in time, such as dreams, headaches,
and other contents of minds possess. It lacks all visible, tangible, audible
properties. In a word, it is abstract.
Disbelievers in numbers and other abstract entities or ‘‘universals’’ have
come to be called nominalists. Nominalism has always attracted philoso-
phers of the hard-headed, no-nonsense type. But does it not conflict with
modern science, which speaks the language of abstract mathematics?
1 INSTRUMENTALIST NOMINALISM
Some nominalists concede that their philosophy of mathematics conflicts

with science by implying that science, when it speaks the language of
mathematics, is not speaking truly. These nominalists adopt an instru-
mentalist philosophy of science, according to which science is just a useful
mythology, and no sort of approximation to or idealization of the truth.
Truth is to be sought, rather, in a philosophy prior and superior to
science.
The position of the well-known nominalist Nelson Goodman is best
understood as a subtle and sophisticated variation on instrumentalism. For
Goodman, science is less a useful fiction than useful nonsense. But whereas a
31
straightforward, simple-minded instrumentalist would be willing to label
science as untrue and let it go at that, Goodman holds that the philosopher
ought at least to attempt to give some sense to the scientist’s otherwise
senseless productions by reconstructing them nominalistically:
The nominalist does not presume to restrict the scientist. The scientist may use
Platonistic class constructions, complex numbers, divination by inspection of
entrails, or any claptrappery that he thinks may help him get the results he
wants. But what he produces then becomes raw material for the philosopher,
whose task is to make sense of all this: to clarify, simplify, explain, interpret in
understandable terms . . . Nominalism is a restraint the philosopher imposes on
himself, just because he feels he cannot otherwise make real sense of what is put
before him. (Goodman 1956, objection vii)
Goodman’s own steps towards nominalistic reconstruction of science

(taken jointly with Quine in Goodman and Quine (1947)) never led very
far. So presumably for Goodman the bulk of science remains nonsensical.
Most recent philosophers of science, even those nominalistically inclined,
have been hostile toward instrumentalist philosophies like Goodman’s for a
couple of good reasons. For one thing, since science is just an outgrowth of
common sense, there can be no sharp dividing line between them. The most
abstruse theoretical physics is connected in a thousand ways through
experimental and applied science, through engineering and technology, to
everyday belief. And much of everyday belief is couched in the vocabulary of
mathematics, albeit of a sort more elementary than that which figures in
general relativity theory or quantum mechanics. The philosopher who
begins by rejecting theoretical physics as fiction will find no logical place
to stop, and in the end will be unable, without inconsistency and self-
contradiction, to accept commonsense belief as fact.
For another thing, the behavior of instrumentalists when not consciously
philosophizing strongly suggests that their professed disbelief in science is a
sham. Catch them off guard, and you are likely to find them classing the
Steady State theory as false, and the Big Bang theory as true, just like the rest
of us. The instrumentalist seems to be ‘‘engaging in intellectual double-
think: taking back in [his] scientific moments what [he] asserts in doing
science’’ (Field 1980, p. 2). He seems to be ‘‘an irrational person . . . who
is unwilling to accept the consequences of his own theories’’ (Chihara
1973, p. 63).
It is on account of such slippery slope and insincerity objections that
instrumentalism is not a live option for most contemporary nominalists;
and it is certainly not a live option for me.
Why I am not a nominalist 33
2 SCIENTIFIC DISPENSABILITY AND NONEXISTENCE
Some anti-nominalists have argued that the conflict between nominalism

and science is so strong that nothing like modern science as we know it
could survive if the nominalist ban on mathematical abstractions were
accepted. Such a position has been reluctantly maintained by the ex-
nominalist Quine ever since the failure of his joint attempt with
Goodman at nominalistic reconstruction. Such a position was also main-
tained, under Quine’s influence, by Hilary Putnam, during his phase of
enthusiastic realism.
I have explained early and late that I see no way of meeting the needs of scientific
theory . . . without admitting universals irreducibly into our ontology . . .
Nominalism . . . is evidently inadequate to a modern scientific system of the
world. (Quine 1981, pp. 182–3)
It has been repeatedly pointed out that such a [nominalistic] language is
inadequate for the purposes of science . . . The restrictions of nominalism are
devastating . . . It is not just ‘‘mathematics’’ but physics as well that we would
have to give up. (Putnam 1971, p. 35)
In short, Quine and Putnam have maintained that mathematical objects

are scientifically indispensable.
The refutation of this thesis has been the first aim of the most prominent
recent nominalist writers, Charles Chihara and Hartry Field. The pro-
grams of nominalistic reconstruction developed in their books (Chihara
1973; Field 1980) are reviewed in outline in the Appendix to the present
chapter. Suffice it to say here that Chihara and Field draw on results from
advanced research in the foundations of mathematics (predicative analysis,
measurement theory, proof theory), and that Chihara assigns the work
normally done by mathematical abstractions to certain modal notions
(including that of the possibility-in-principle of inscribing tokens of sym-
bols of a certain formal language), while Field assigns it to certain spatio-
temporal objects (admitting as concrete entities regions of space-time that
are irregular, disconnected, and of heterogeneous material content). Their
books cast considerable doubt on the thesis of the scientific indispensability
of mathematical objects.
Does this suffice to establish nominalism? Chihara and Field seem to
think so. While for many readers the most valuable parts of Chihara’s book
will be the chapters on Russell and Poincaré, for the author himself,
to judge by his Introduction, what is most important is the attempt to
refute the anti-nominalist arguments of Quine, and some not dissimilar
arguments of Kurt Gödel. Chihara implicitly presumes that a refutation of
these arguments is tantamount to a proof of nominalism.
As for Field, his book bears the subtitle ‘‘A Defense of Nominalism,’’ but
includes (p. 4) the disclaimer that ‘‘nothing in this monograph purports to
be a positive argument for nominalism.’’ The resolution of the paradox lies
in Field’s presumption that nominalism does not need to be defended by
positive arguments. He explicitly says that if he can accomplish the nega-
tive aim of undercutting the arguments of Quine and Putnam, then he will
have reduced belief in mathematical objects to the status of ‘‘unjustifiable
dogma.’’ Thus Field, like Chihara, presumes the burden of proof to be on
his lotus-eating ‘‘Platonist’’ opponent.
I disagree. Chihara and Field may have gone a long way toward showing
that science could be done without numbers. I maintain, however, that
science at present is done with numbers, and that there is no scientific
reason why in future science should be done without them. And thus it is
not the (continued) acceptance of mathematical objects, but rather the
nominalist’s insistence on their rejection, that constitutes an unjustified
and anti-scientific philosophical dogmatism.
Quine and Putnam have been false friends of numbers in making the
case for their acceptance seem to depend on a claim of indispensability.
Actually, the burden of proof is on such enemies of numbers as Chihara
and Field, to show either: (a) that science, properly interpreted, already does
dispense with mathematical objects, or (b) that there are scientific reasons
why current scientific theories should be replaced by alternatives dispensing
with mathematical objects. I will call the claim (a) about the proper
interpretation of current science hermeneutic nominalism, and the proposal
(b) to replace current science by an alternative revolutionary nominalism.
I have argued that any anti-instrumentalist nominalism must be either
hermeneutic or revolutionary. I will argue that hermeneutic nominalism,
judged by the standards of linguistics, is an implausible hypothesis thus far
unsupported by evidence; and that revolutionary nominalism, judged by
the standards of physics, is a costly proposal thus far without scientific
motivation.
3 HERMENEUTIC NOMINALISM
If we take everyday beliefs at face value, then we must conclude that natural
numbers are posits of common sense dating from prehistoric times. If we
take physics even halfway literally, then we must conclude that science has
been committed to complex numbers for well over a century. According to
hermeneutic nominalism, this is all illusion. General relativity theory may
seem to make statements about vector-valued functions. Quantum
mechanics may seem to make statements about linear operators. But, in
fact, no physical theory asserts or presupposes the existence of such mathe-
matical objects; no branch of science actually posits or commits itself to the
existence of abstract entities.
Hermeneutic nominalism is thus a thesis of a type that has recently been
described by Saul Kripke:
The philosopher advocates a view in patent contradiction to common sense.
Rather than repudiating common sense, he asserts that the conflict comes from
a philosophical misinterpretation of common language – sometimes he adds that
the misinterpretation is encouraged by the ‘‘superficial form’’ or ordinary speech.
He offers his own analysis of the relevant common assertions, one that shows that
they do not really say what they seem to say. (Kripke 1976, p. 269)
Let us imagine a laboratory assistant to Lord Kelvin reporting the data in

some experiments on the conversion of mechanical to thermal energy. It
sounds as if he is speaking of energy-in-joules and temperature-in-degrees-
Kelvin and other such numerical and abstract entities. According to
hermeneutic nominalism, he is actually speaking of something completely
different: perhaps of possible chalk marks on possible blackboards (follow-
ing Chihara). Maybe of so-called basic regions scattered through the vast-
ness of space-time (following Field). Or perhaps of something still less
expected and still more surprising (following some yet unwritten rival to
Chihara (1973) and Field (1980)).
Now this claim is in itself not very plausible, and it becomes even less so
when we reflect that to take anything like what we find in Chihara’s book
or Field’s as an account of what the laboratory technician is saying is to
attribute to that technician a tacit knowledge of such topics in foundations
of mathematics as predicative analysis and measurement theory. These
subjects did not even exist in Lord Kelvin’s day, and even now they are
studied by few pure mathematicians, let alone working physical scientists
and their technical assistants.
Kripke’s words (not specifically directed against nominalism by their
author) seem appropriate here:
Personally I think such philosophical claims are almost always suspect. What the
claimant calls a ‘‘misleading philosophical misconstrual’’ of the ordinary statement
is probably the natural and correct understanding. The real misconstrual comes
when the claimant continues, ‘‘All the ordinary man really means is . . .’’ and gives
a sophisticated analysis compatible with his philosophy.
Certainly the burden of proof is on the proponents of hermeneutic nom-
inalism, who claim to have discovered a radical difference between appear-
ance and reality in scientific discourse.
As a thesis about the language of science, hermeneutic nominalism is,
I presume, subject to evaluation by the science of language, linguistics. For
I am prepared to dismiss those who
write as if, in addition to . . . everyday or ‘‘garden variety’’ rules of English, capable
of being discovered by responsible linguistic investigation carried on by trained
students of language, there were also . . . ‘‘rules’’ capable of being discovered only
by philosophers. (Putnam 1971, p. 5)
In the current technical jargon of linguistics, the hermeneutic nomina-
list’s thesis that scientific statements do not really say what they appear
to say becomes the hypothesis that their deep structure differs from
their surface structure, while the thesis that such statements are not really
about what they appear to be about becomes the hypothesis that certain
noun phrases in the surface structure are without counterpart in the deep
structure.
Now readers of professional linguistics journals will recognize that
hypotheses of this general type (though normally less radical than those
of hermeneutic nominalism) are not seldom entertained by trained stu-
dents of language. Such readers will also be familiar with the kinds of
evidence cited in responsible linguistic investigations to support such
hypotheses. Until some evidence of this kind can be adduced in support
of its implausible hypotheses, I for one will be prepared to dismiss herme-
neuticism as a desperate device of ‘‘ostrich nominalism.’’
4 REVOLUTIONARY NOMINALISM
It is one thing to observe that matters could equally well have been
arranged otherwise than they currently are. It is quite another thing to
urge that a rearrangement would constitute an improvement. To say that
the British convention of driving on the left-hand side of the road is no
worse than our own convention of driving on the right-hand side is not to
advance a criticism of our traffic laws.
‘‘Science,’’ Putnam tells us, lives ‘‘extremely happily on the rich diet of
impredicative sets’’ (1971, p. 56). The work of Chihara and Field suggests
that science could survive on more meager fare, on a diet of inscription-
possibilities or of spatiotemporal regions. But would science be healthier
after such a change of menu?
When scientists abandoned caloric fluid and luminiferous ether, it was
because they had discovered alternative theories that were empirically
superior, of wider scope and greater accuracy in predicting the results of
observations and experiments. Now the alternative theories concocted by
Chihara and Field cannot be claimed to be empirically superior to current
theories, for they have been designed to be empirically equivalent.
Will it be urged that those alternatives are somehow pragmatically
superior? Their awkward and ungainly character makes it difficult to
claim that they are more convenient and efficient as systematizations of
the data of experience. Will it be urged that, despite their unnatural and
artificial character, they somehow contributed to clarity, simplicity, intelli-
gibility, and the like, in ways that matter to working scientists? Something
of the sort must be urged if a nominalistic revolution in science is to be
motivated.
The proviso, ‘‘in ways that matter to working scientists,’’ is crucial, if a
mere instrumentalist opposition to science is to be avoided. It is pointless
and futile to urge a revolution in the practice of physicists motivated only by
considerations appealing only to philosophers of a certain type. Physicists
are too well aware of the dismal historical record of philosophical interfer-
ence in science to accept such dictation from outsiders.
Now the avoidance of ontological commitments to abstract entities does
not seem to have won recognition in the scientific community as being in
itself a goal of the scientific enterprise on a par with scope and accuracy,
and convenience and efficiency, in the prediction and control of experi-
ence. It seems, on the contrary, a matter to which most working scientists
attach no importance whatsoever. It seems distinctively and exclusively a
preoccupation of philosophers of a certain type. Thus Goodman is able to
cite only a few linguists who are nominalistically inclined, and not one
physicist:
Paucity of means often conduces to clarity and progress in science as well as
philosophy. Some scientists indeed – for example, certain works in structural
linguistics – have even imposed the full restriction of nominalism upon themselves
in order to avoid confusion and self-deception.
One would search the physics journals in vain for any expression of nomi-
nalistic qualms and scruples, of reluctance and hesitancy to use mathematical
apparatus, of suspicion that such ‘‘Platonistic claptrappery’’ as complex
numbers may be a source of ‘‘confusion and self-deception.’’
The proposed nominalistic revolution in physics can be scientifically
motivated only by showing that the avoidance of ontological commitments
to abstract entities would somehow serve indirectly to advance us toward
some more recognizably scientific goals. For my own part, I cannot discern
any such scientific benefits to be expected from the proposed revolution,
while I do discern a couple of non-negligible costs.
First, any major revolution involves transition costs: the rewriting of
textbooks, redesign of programs of instruction, and so forth. A reform
along the lines of Chihara (1973) would involve reworking the mathematics
curriculum for science and engineering students, avoiding impredicative
methods in favor of predicative parodies that are harder to learn and not so
easy to apply.
A reform along the lines of Field (1980) would involve reworking the
physics curriculum, so that each basic theory would initially be presented
in qualitative rather than quantitative form. A course on measurement
theory would have to be crammed into the already crowded study plan, to
explain and justify the use of the usual numerical apparatus. This is
educational reform in precisely the wrong direction: away from applica-
tions, toward entanglement in logical subtleties.
Second, the physicist who puts on nominalistic blinders may be unable
to see certain potentially important paths for the development of science. I
have in mind here not an inevitable logical consequence of nominalistic
revolution, but a likely psychological consequence. Chihara (1973, p. 209)
promises that he will recant his nominalism should some future physical
theory turn out to require mathematical objects indispensably. But the
danger I have in mind is that if science goes nominalist today, that future
theory may simply never be discovered. Yuri Manin has noted this point in
connection with intuitionism:
Unfortunately, it seems that it is these ‘‘extremes’’ – bold extrapolations, abstractions
which are infinite and do not lend themselves to a constructivist interpretation –
which make classical mathematics effective. One should try to imagine how much
help mathematics could have provided twentieth century quantum physics if for the
past hundred years it had been developed using only abstractions from ‘‘constructive
objects.’’ Most likely, the standard calculations with infinite dimensional represen-
tations of Lie groups which today play an important role in understanding the
microworld, would simply never have occurred to anyone. (Manin 1977, pp. 172–3)
(Mention of quantum mechanics should remind us that it is unclear
whether the methods of Chihara and Field are adequate even for present-
day science in its entirety. For Chihara the problem is a minor one, and
could probably be solved by adopting a somewhat stronger system of
predicative analysis than the particular weak system Rx he favors. For
Field, the problem is a major one, for he has given us no idea how he
proposes to treat quantum theory, which differs radically (owing to its use
of infinite-dimensional apparatus and to its statistical character) from the
one theory he does treat in detail, Newtonian gravitational theory.)
But I need not enlarge on the costs for present-day and future physics of a
nominalistic revolution. Surely the burden of proof is on the revolutionary,
who proposes a drastic departure from our thus far eminently successful
policy of ontological tolerance in common sense and scientific theory
construction. Until it is shown that nominalism offers physical science
some substantive advantages, I for one am prepared to dismiss its revolu-
tionary proposals as motivated only by medieval superstition (‘‘Ockham’s
razor’’) and fastidious bigotry (cf. Goodman 1964, objection viii).
Chihara and Field have gone a long way toward constructing nominal-
istic alternatives empirically equivalent and pragmatically only slightly
inferior to our current scientific theories. Their work suggests that an
ontology of abstracta may be one feature of those current theories that is
merely conventional, in the best sense of the word (that of David Lewis
1969). This suffices to cast considerable doubt on some more extreme
versions of realism.
It does not suffice to cast doubt on moderate versions of realism, which
merely observe that our current theories seem to invoke abstracta and that
we do not yet have reasons to abandon those theories. For to characterize
some feature of our present ways of doing things (in scientific theorizing or
in driving) as conventional is not in itself to criticize that feature. And
Chihara and Field have not come close to constructing nominalistic alter-
natives that are manifestly superior (empirically or pragmatically) to our
current scientific theories.
5 NOMINALISM, ONTOLOGICAL AND EPISTEMOLOGICAL
I have rejected nominalism in its traditional ontological form, as the doctrine

that there exist no abstract entities. I equally reject it in its currently fashionable
epistemological form, as the thesis that even if there exist any abstract entities,
still we could never come to know about their existence. Epistemological
nominalism is usually supported by an argument of the following form: all
entities of which we can have knowledge are causally connected with our
organism; no abstract entities are causally connected with our organism; ergo,
no abstract entities are entities of which we can have knowledge.
The argument is, of course, valid, a syllogism in Camestres. But the
premises are dubious and debatable. As for the minor premise, of course a
cyclic group does not act on our organs of sight, touch, and hearing in the
same way as an alarm clock. And nobody nowadays, except perhaps a stray
numerologist or two, would imagine that mathematical objects act on us
through some mysterious sixth sense or ESP unknown to orthodox phy-
siology. Nonetheless, as Maddy (1980) skillfully argues, there is a good deal
of research in developmental psychology and neurophysiology that can be
read as showing that we do, in a sense, have causal contact with certain
abstracta.
As for the major premise, it rests on a causal theory of knowledge. That
theory has many opponents, who regard it as a half-truth arrived at by over-
hasty generalization from too narrow a range of cases, to which the cases of
knowledge of mathematical objects, ethical values, other minds, and so
forth are just so many counterexamples. Significantly, that theory has also a
good many half-hearted sympathizers, who do not regard it as wrong-
headed or misguided, but merely as in need of amendment. In many
amended versions, the notion of causality disappears, to be replaced by
that of reliability or explanation or something of the sort, and with it
disappears the major premise of the epistemological nominalist’s syllogism.
Again Maddy (1984) provides a useful survey of the issues.
The more cautious sympathizers with the causal approach to the theory
of knowledge now maintain only that the abstractness and consequent
causal inertness and isolation of mathematical objects create difficulties for
the epistemologist in trying to account for mathematical knowledge. I am
surprised to find Field citing these epistemological difficulties as if they in
themselves constituted some sort of grounds for nominalism:
[Nominalism] saves us from having to believe in a large realm of . . . entities which
are very unlike the other entities we believe in (due for instance to their causal
isolation from us and from everything that we experience) and which give rise to
substantial philosophical perplexities because of those differences. (Field 1980, p. 98)
A footnote to this passage makes it plain that Field’s ‘‘philosophical

perplexities’’ are precisely the epistemological difficulties just alluded to.
(Incidentally, the same footnote provides a good bibliography of works
arguing for epistemological nominalism.)
To bring out just how odd this argument is, I want to consider a parallel:
suppose that Burrhus Skinner were to confess that after all those years of
work with his rats and pigeons he is still ‘‘substantially perplexed’’ by the
ability of freshman students to master calculus and mechanics. Now what
mathematician or physicist would take that as motivation for rewriting the
textbooks in those subjects? What linguist would take it as evidence that the
sentences in those textbooks have some bizarre and outré depth grammar?
No one would take it as an indication of anything but the inadequacies of
behaviorist learning theory.
Likewise, a philosopher’s confession that knowledge in pure and applied
mathematics perplexes him constitutes no sort of argument for nominal-
ism, but merely an indication that the philosopher’s approach to cognition
is, like Skinner’s, inadequate.
CONCLUSION
Unless he is content to lapse into a mere instrumentalist of ‘‘as if ’’ philo-

sophy of science, the philosopher who wishes to argue for nominalism faces
a dilemma: he must search either for evidence for an implausible hypothesis
in linguistics, or else for motivation for a costly revolution in physics. Neither
horn seems very promising, and that is why I am not a nominalist.
APPENDIX
For the reader’s convenience, I here outline the constructions of Chihara

and Field, and the claims which those authors make for their constructions.
I will not advance any technical objections against those constructions
(though in fact I have one small reservation about Chihara’s approach,
and share with Kripke several large reservations about Field’s), since my
aim has been to argue that even if the constructions are technically flawless,
they do not suffice to establish nominalism.
A Chihara’s modal nominalism

I here outline the construction of Chihara (1973), Chapter V and
Appendix. Chihara’s strategy is to reinterpret in a nominalistically accept-
able fashion a portion of pure mathematics: arithmetic first, then so-called
predicative analysis. He then argues that the portion of mathematics so
reinterpreted suffices for scientific applications, and dismisses the rest of
mathematics (the impredicative part) as mythology.
To illustrate Chihara’s approach to arithmetic, consider Euclid’s famous
theorem:
(0) ð8 number mÞð9 number nÞðm5n & n is primeÞ
As a first attempt to avoid mathematical objects, let us rewrite this as:
(1) ð8 numeral aÞð9 numeral bÞ . . .
(I will indicate only the transformation of the prefix of (0); this is not to say
that the transformation of the matrix does not require some caution.) Now if
numerals are taken as types (patterns of inscription), then they are themselves
abstract entities akin to shapes, and (1) is not much of an improvement on
(0). But if numerals are taken as tokens (individual inscriptions), then they
are concrete entities, made of chalk or ink, but there may not be (indeed,
almost certainly are not) enough of them around to make (1) true. To get a
version of (0) that is both true and committed only to concrete entities, we
must introduce the modal notions of necessity (&) and possibility (}).
Then, taking numerals as tokens, our final reinterpretation of (0) is:
(2) &ð8 numeral aÞ}ð9 numeral bÞ . . .
Informally this says: however long a tally you could ever write down, I
could write down a still longer one such that . . . Here we have the idea
behind the approach to arithmetic in Chihara (1973).
Chihara’s approach builds on the work of predicativists (specifically, Hao
Wang), mathematical constructivists somewhat more liberal than intui-
tionists. Predicativists accept uncritically classical arithmetic (theory of
natural, or equivalently rational, numbers) but in analysis (theory of real
numbers, or equivalently of sets of natural numbers) they accept only what
is definable. To begin with, they accept those sets of natural numbers that
are definable by purely arithmetical conditions, conditions quantifying
only over natural numbers. These are the order zero sets. Next they accept
those sets of natural numbers that are definable by conditions quantifying
over natural numbers and order zero sets. These are the order one sets. And
so on, through higher and higher orders. (Just how high to go is a delicate
question.) A surprisingly large portion of classical mathematics can be
‘‘parodied’’ within this framework, as the survey by Feferman (1977)
shows. Intuitively, it is plausible that a theory of definable sets should be
reducible to arithmetic plus truth-predicates, with quantification over
definable sets being replaced by quantification over the code numbers of
their defining conditions, and the membership relation replaced by the
relation ‘‘n is the code number of a formula with one free variable that is
true of m.’’ The details can be worked out, and we get a reduction of
predicative analysis to something that has already been shown to be
nominalistically reinterpretable. Chihara’s account of the application of
mathematics in science is illustrated by Figure 2.1.
While scientific theories are formulated mathematically in terms of
sharply defined functions, at least in the overwhelming majority of
Empirical Basic scientific
conditions theory
define formulated for
FUZZY Idealization SHARP
FUNCTION FUNCTION
Theorems of
analysis
Information about
Consequences of
FUZZY
scientific theory:
FUNCTION
information about
hence information
Approximation SHARP
about empirical
FUNCTION
situation
Figure 2.1
applications empirical conditions define only fuzzy functions. For instance,

the condition:
f(t) ¼ x iff
the projectile is x meters above the floor of the chamber
at t seconds after firing
cannot define a sharp function because of the fuzziness of the projectile and
the chamber (viewed on a scale of micrometers) and of the firing event
(viewed on a scale of nanoseconds). Thus the application of a scientific
theory to empirical conditions typically involves an element of idealization.
Now predicative mathematics provides sufficiently many sharp functions
to serve as idealizations for empirically defined fuzzy functions, because any
classical function can be approximated as closely as desired by a predicative
function. Moreover, the theorems of analysis used in deriving consequen-
ces from basic scientific theories can all be parodied predicatively: this can
be verified by comparing the mathematics curriculum for science and
engineering students as listed in any college catalogue with the survey by
Feferman (1977) of predicative mathematics. Thus predicative mathe-
matics, which we have already seen to be nominalistically reinterpretable,
suffices for scientific applications.
B Field’s spatiotemporal nominalism

I here outline the construction of Field (1980). Field’s strategy is to
reformulate basic scientific theories and their consequences (the special
information they entail about special situations) in a way that avoids all
mathematical vocabulary, and then to argue that the consequences can be
deduced from the basic theories without introducing any mathematics.
To illustrate Field’s approach to the formulation of science, let us con-
sider, as he does, thermodynamics. Here a typical qualitative, math-free,
nominalistically acceptable notion would be the comparative relation R
between point-events of space-time given by ‘‘x is cooler than y.’’ Here a
typical quantitative, mathematical, ‘‘Platonistic’’ notion would be that of
temperature on a given scale, conceived as a real-valued function r on point-
events in space-time. Measurement theory, as surveyed in the compendium
Krantz et al. (1971), is a corpus of theorems to the effect that suitable
assumptions on qualitative relations entail the existence (and uniqueness
up to stated changes of scale) of quantitative functions appropriately repre-
senting them. In thermodynamics, suitable assumptions would include that
R is irreflexive and transitive, appropriate representation would include that
xRy if and only if r(x) < r(y), and stated changes of scale would be like those
used in passing between Fahrenheit and Celsius. Once we have such a basic
representation theorem, it becomes possible to reformulate any scale-invariant
assumption on the quantitative functions as an assumption on the qualitative
relations. In thermodynamics, continuity for the temperature function r can
be reformulated in terms of a notion of temperature-basic region, itself
defined in terms of the cooler relation R.
In this way the nominalist can reformulate the whole of science, both
basic, general theoretical principles, and particular consequences for practical
applications. However, the only route we have seen so far from the quali-
tatively formulated version of a basic theory to the qualitatively formulated
QUALITATIVE Representation QUANTITATIVE

formulation of theorems of formulation of
BASIC BASIC
Measurement
THEORY THEORY
theory
Conservation Theorems
theorems
⇓ of
analysis
QUALITATIVE Representation QUANTITATIVE

formulation of theorems of formulation of
CONSEQUENCES Measurement CONSEQUENCES
of basic theory theory of basic theory
Figure 2.2
versions of its consequences involves a ‘‘Platonistic’’ detour: from qualitative
basic theory by measurement theory to quantitative basic theory, thence by
theorems of analysis to quantitative consequences, and thence by measure-
ment theory again to qualitative consequences, as in Figure 2.2.
It is, however, theoretically possible, though practically inconvenient, to
avoid the introduction of mathematics, to avoid the detour through the
quantitative and abstract:
the conclusions we arrive at by these means are not genuinely new, they are already
derivable in a more long-winded fashion . . . without recourse to the mathematical
entities. (Field 1980, pp. 10–11)
for these purposes [‘‘problem solving’’] the usual numerical apparatus is a practical
necessity. But it is a necessity that the nominalist has no need to forgo: he can treat
the apparatus . . . as a useful instrument for making deductions from the nominal-
istic system that is ultimately of interest; an instrument which yields no conclusions
not obtainable without it, but which yields them more easily. (Field 1980, p. 91)
These claims are supported by appeal to conservation theorems from proof
theory (the most important being perhaps one due to Scott Weinstein).
3
Mathematics and Bleak House
1 ‘‘ N O M I N A L I S M ’’ AND ‘‘ R E A L I S M ’’ 1
Nominalism is a large subject. In our book (Burgess and Rosen 1997) my
colleague Gideon Rosen and I distinguished a negative or destructive side
of nominalism, which tells us not to believe what mathematics appears to
say, from a positive or reconstructive side, which aims to give us something
else to believe instead. We noted that there were a few nominalists who
contented themselves with the negative side, conceding that mathematics is
useful, insisting that what it appears to say is not true, and letting it go at
that, without attempting any reconstrual or reconstruction of mathe-
matics. We expressed some surprise that there were not more such destruc-
tive nominalists, since as compared with reconstructive nominalism,
destructive nominalism has what Russell in another context called ‘‘the
advantages of theft over honest toil’’; and if nothing else was clear from the
work of Hartry Field, Charles Chihara, Geoffrey Hellman, and other
reconstructive nominalists whose work we surveyed, it was clear that the
amount of honest toil that would be required for a nominalistic reconstrual
or reconstruction of mathematics would be quite considerable.
Today, a couple of years after publication, it is beginning to seem that
the main achievement of our book will have been to provide a decent burial
for the hard-working, laborious variety of nominalism. For almost every-
thing that has come forth since from the nominalist camp has represented a
light-fingered, larcenous variety, which helps itself to the utility of math-
ematics, while refusing to pay the price either of acknowledging that what
mathematics appears to say is true, or of providing any reconstrual or
reconstruction that would make it true. The usual label for this variety of
1
I have decided to keep this paper in the form in which it was originally written for oral delivery,
adding footnotes to supply citations of the literature, and for clarification at a few points where
experience has shown misunderstanding of my intended sense may be likely.
46
Mathematics and Bleak House 47
nominalism is ‘‘[mathematical] fictionalism.’’2 Not only has fictionalism
become the most widely pursued form of nominalism, but even some
former anti-nominalists have been wavering or drifting in its direction,
including even the author of one of the most eloquent early criticisms,
‘‘Mathematics and Oliver Twist.’’ I will take the liberty, therefore, of
substituting ‘‘fictionalism’’ for the ‘‘nominalism’’ label in the official title
of our symposium.
While I am at it, I had better say something about the ‘‘realism’’ as well.
It is an even larger subject, since there is hardly any bit of philosophical
terminology more diversely used and overused and misused than the R-
word. There seems to be a systematic difference between the way in which
the word is understood by many of those who describe themselves as
‘‘realists’’ and the way in which it is understood by most of those who
describe themselves as ‘‘anti-realists.’’ For many professed ‘‘realists,’’ realism
amounts to little more than a willingness to repeat in one’s philosophical
moments what one says in one’s scientific moments, not taking it back,
explaining it away, or otherwise apologizing for it: what we say in our
scientific moments is all right, though no claim is made that it is uniquely
right, or that other intelligent beings who conceptualized the world differ-
ently from us would necessarily be getting something wrong. For many
professed ‘‘anti-realists,’’ realism seems rather to amount to a claim that
what one says to oneself in scientific moments when one tries to under-
stand the universe corresponds to Ultimate Metaphysical Reality; that it is,
so to speak, a repetition of just what God was saying to Himself when He
was creating the universe.
The weaker position might be called anti-anti-realism, and the stronger
position capital-R Realism. Quine says somewhere that his ‘‘realism’’ and his
‘‘pragmatism’’ are reconciled by his ‘‘naturalism.’’ This is a hard saying, but I
think it can be explained along the following lines. First, ‘‘realism’’ here
means anti-anti-realism: the refusal to apologize while doing philosophy for
what is said while doing mathematics or science. Second, ‘‘pragmatism’’
2
In earlier publications I called this stance ‘‘instrumentalism,’’ while Rosen called it ‘‘constructive
nominalism,’’ by analogy with van Fraassen’s ‘‘constructive empiricism.’’ (Incidentally, though van
Fraassen is best known for his fictionalism about unobservable physical entities, he is also a fictionalist
about abstract mathematical entities.) However, the analogous positions in other areas of philosophy
(e.g. modality) are today generally called ‘‘fictionalist,’’ and on this ground Rosen adopted the
‘‘fictionalist’’ label, and I now follow him. Unfortunately, there has been another use of ‘‘fictionalist’’
in philosophy of mathematics, by Hartry Field and his students, for whom it includes all rejectionist
views, all views that hold that standard mathematics is false, whether fictionalist in our sense or
reconstructive. This usage seems to be out of alignment with the usage of ‘‘fictionalism’’ in other areas
of philosophy, and for that reason to be avoided.
means rejection of capital-R Realism, rejection as unjustified of any claim
that mathematics and science give us a God’s-eye view of capital-R Reality.
Third, ‘‘naturalism’’ means adherence to a conception of epistemology as an
inquiry conducted by citizens of the scientific community examining science
from the inside, rather than an inquisition conducted by philosophers
foreign to science judging science from the outside.
And the reconciliation? Naturalism teaches us to look at our scientific,
philosophical, and other forms of intellectual endeavor as activities of bio-
logical organisms with cognitive capacities that, though extensive, stop well
short of omniscience. As such, none of these endeavors can succeed in
achieving a God’s-eye view of Reality. And therefore there is no reason to
apologize for one of them, science, failing to achieve such a view, and every
reason not to suppose that another of them, philosophy, could do better.
Should I, therefore, replace ‘‘realism’’ by ‘‘naturalism’’ in the title of our
symposium? No, for unfortunately ‘‘naturalism’’ too has been used and
overused and misused in multiple senses. In particular, while in Quine’s
sense naturalism abstains from imposing philosophical constraints on
science, there is another equally widespread but diametrically opposed
conception in which ‘‘naturalism’’ consists precisely in imposing such a
philosophical constraint, namely, the constraint that no entities are to be
assumed that do not stand in natural cause-and-effect relations with us. So
I will just replace ‘‘realism’’ by ‘‘anti-fictionalism,’’ making the title of our
symposium come down to this: ‘‘fictionalism in philosophy of mathe-
matics: pro and con.’’
2 LITERARY GENRES
To begin with the pro side, it is impossible to quarrel with the proposition
that mathematics is in some respects like fiction. For indeed, anything is like
anything else, in some respect. I even think the comparison may be illumi-
nating, as to the nature of fiction, or rather, as to some of the questions
philosophers raise about fiction, and in particular about the status of
fictional characters. The hard-headed view is that they just do not exist.
Another view is that they are abstract entities of some sort or other. Yet
another view is that they are mental entities. Now the view that mathe-
matical entities are mental entities, though a perennial favorite of amateur
philosophers, has been in disrepute among professional philosophers since
Frege’s trenchant critique of psychologistic philosophies of mathematics
more than a hundred years ago. I think that fiction is enough like mathe-
matics to suggest that the view that fictional entities are mental entities is
equally dubious. For example, one problem for mentalistic theories of
mathematical entities is that there are too many such entities for minds to
have created each one; and a little reflection shows that the situation is
similar with fiction: a novelist may write of the doings of a vast army with
thousands of officers and myriads of soldiers, while in fact describing only a
couple of dozen of them individually. But I am straying from our topic,
which was supposed to be what, if anything, comparison with fiction can
show us about mathematics, and not the reverse.
Reverting to that topic, then, I have said on the one hand that it can
hardly be denied that mathematics is like fiction in some respects, most
obviously in consisting of large bodies of manuscript and printed and now
electronic writing. On the other hand, it must be said that there is a sense in
which mathematics is clearly non-fiction: there is a well-established prac-
tice of classifying writing as ‘‘fiction’’ and ‘‘non-fiction,’’ and setting aside
attempted theoretical definitions and analyses, attending rather to the
criteria by which in actual practice such classifications are made, clearly
mathematics counts as non-fiction. The compilers of the New York Times
best-seller list will never put any mathematical work, however wonderful,
at the top of the fiction column, and not just because nothing even by
Andrew Wiles will ever sell like Stephen King. Nor will any librarian
catalogue, say, the Proceedings of the Cabal Seminar as an ‘‘anthology of
short stories based on the characters created by Georg Cantor.’’ Now of
course misclassifications are sometimes made, and mischievous hoaxes and
pious frauds sometimes succeed, but these represent mistakes in applying
general criteria to specific cases. I do not think it even makes sense to
suggest that the general criteria are themselves mistaken, and it seems
unquestionable that by those criteria mathematical writing is not literally
fictional writing: mathematics is not in all respects the same as fiction.
So the question is: in which respects is mathematics like, and in which
respects is it unlike, fiction? That in part depends on the species of the
genus fiction one considers. The first species one thinks of will probably be
the novel, and comparison with novels is in fact common among professed
fictionalists and their critics, as with Oliver Twist, mentioned earlier. There
has been, however, a minority – including especially the late Leslie Tharp3 –
who have preferred a comparison with mythology. In a slightly different
vein, Steve Yablo, in an interesting paper,4 has suggested a comparison with
metaphor. Metaphor of course is not a genre of fiction but a figure of
3
Rescued from oblivion in Chihara (1989).
4
The reference was to an on-line version, but the paper has since appeared in print as Yablo (2000).
speech, and I think that to speak as Yablo does of a metaphor running on
for volumes and volumes and volumes is to stretch the concept of ‘‘meta-
phor’’ well beyond breaking point. But perhaps much of the content of
Yablo’s suggestion could be preserved if we took the comparison to be
between mathematics and parables or fables.
Be that as it may, I believe the comparison with fables is the most apt of
the candidates I have considered, and comparison with novels the least so.
Novels almost always are attributable to identifiable individual authors:
Proust or Flaubert, Trollope or Dickens. Some fables are attributable to
such authors, Lafontaine for instance; others are traditional. Mathematics
also consists of both traditional elements and elements with identifiable
authors. Novels are almost always unique. Fables tend to be retold over and
over in variant versions by different writers, so that we have Aesop’s
version, Lafontaine’s version, and many latter-day retellings of the fox
and the crow, for instance. Mathematics likewise gets retold by textbook
writer after textbook writer. The characters in one novel seldom reappear in
another, and even those who do reappear, like Swann or Palliser, do so only
in comparatively few stories, all by the same author. This is so with some
characters of fable, but many, like the clever fox, reappear in whole cycles of
tales. The same mathematicalia, p and e, the sine and cosine functions, 0
and 1 and 2 and so on, reappear through whole libraries of mathematical
works. Again, characters encountered in novels are generally of the same
species as those encountered in daily life, while those in fables are, as one
dictionary definition reminds us, beings of a different order: ‘‘animals that
talk and behave like human beings.’’ Mathematics, too, has objects even
more unlike those of any other subject, and it is for precisely that reason
that there is thought to be a philosophical problem about them.
Yet more important is the matter of application, which in literature
typically takes the form of a ‘‘message.’’ The fable typically though not
invariably has a ‘‘moral,’’ while to demand one of the novel is virtually the
definition of Philistinism. I am reminded in this connection of what
Nabokov says in his posthumously published lectures about Bleak House
and its supposed concern with reform of the Court of Chancery.
At first blush it might seem that Bleak House is a satire. Let us see. If a satire is of
little aesthetic value, it does not attain its object, however worthy that object may
be. On the other hand, if a satire is permeated by artistic genius, then its object is of
little importance and vanishes with its times while the dazzling satire remains, for
all time, as a work of art. So why speak of satire at all? . . . Such cases as Jarndyce
did occur now and then in the middle of the last century although, as legal
historians have shown, the bulk of our author’s information on legal matters
goes back to the 1820s and 1830s so that many of his targets had ceased to exist by
the time Bleak House was written. But if the target is gone, let us enjoy the carved
beauty of the weapon. (Nabokov 1980, p. 64)
The question of applications is crucial in the case of mathematics, because
though it would be a kind of Philistinism to demand that every piece of
mathematics have one, many do; and it is precisely because many do that
many philosophers have opposed nominalism, this being the least com-
mon denominator of all ‘‘indispensability arguments.’’
Still more important, however, is a feature common to all genres of
fiction. The most important single respect in which fictionalists hold
mathematics to be like novels or fables or whatever is in being a body of
falsehoods. In particular the existence theorems of mathematics are sup-
posed to be untrue: these say there exist, for instance, prime numbers
10
greater than 1010 , whereas according to mathematical fictionalists, and
indeed all nominalists, there are no such things as numbers at all.5
3 NON-LITERAL LANGUAGE
Nominalism, to repeat, is a large subject. In our book, Rosen and I

distinguished two varieties of reconstructive nominalists. Those of one
variety, the hermeneutic, insist that their reconstruals of mathematics reveal
what, contrary to superficial appearances, deep down mathematical lan-
guage has meant all along: mathematical theorems are true, but while what
mathematical theorems appear to mean implies the existence of mathe-
matical entities, what they really mean does not. This position might be
summed up in the formula, ‘‘There are no numbers, and some of them are
10
primes greater than 1010 .’’ Reconstructive nominalists of the other variety,
the revolutionary, concede that their reconstructions of mathematics are
not analyses of current mathematics, but amendments to it; not exegeses,
but emendations.6
5
I have been concerned here with comparison between mathematics and fiction less for its own sake
than for its bearing on the issue of nominalism. The comparison deserves a much fuller examination.
For an extended discussion of analogies and disanalogies, containing extensive references to the
further literature, see Thomas (2000, 2002).
6
This is not the place to discuss reviews of Burgess and Rosen (1997) at any length, but it may be
mentioned that many adherents of and sympathizers with reconstructive nominalism have wished to
claim that there is some third alternative. For instance, it is sometimes said that a nominalist
interpretation represents ‘‘the best way to make sense of ’’ what mathematicians say. I see in this
formulation not a third alternative, but simply an equivocation, between ‘‘the empirical hypothesis
about what mathematicians mean that best agrees with the evidence’’ (hermeneutic) and ‘‘the
construction that could be put on mathematicians’ words that would best reconcile them with
certain philosophical principles or prejudices’’ (revolutionary).
We did not in the book discuss a division between hermeneuticists and
revolutionaries among the fictionalists, but such a distinction can be
made.7 The hermeneutic fictionalist maintains that the mathematicians’
own understanding of their talk of mathematical entities is that it is a form
of fiction, or akin to fiction: mathematics is like novels, fables, and so on in
being a body of falsehoods not intended to be taken for true. According to the
hermeneutic fictionalist, the anti-nominalist philosopher is being more
royalist than the queen (of the sciences); is being a kind of fundamentalist,
taking literally what was never so meant.
Consider, for instance, the question when the ‘‘reification’’ of numbers
first occurred historically, in the main line of development leading up to
modern mathematics. What I take to be a fairly conventional view dates the
reification of natural numbers to about the time of Archytas of Tarentum
and other Pythagoreans, among whom we begin to see the transition from
the use of numerals as adjectives, as in ‘‘six oxen were sacrificed to Zeus,’’ to
their use as nouns, as in ‘‘six is a perfect number’’; and it dates the reification
of real numbers from about the time of Omar Khayyam and other
medieval Islamic and Hindu mathematicians, who in contrast to Greek
mathematicians treated ratios of geometric magnitudes as numbers, as
things that could be added and multiplied. According to Chihara (1990),
however, these dates are all wrong. The reification of number actually
occurred not circa 500 B C E or 1000 C E , among mathematicians, but some
time in the 1960s and 1970s, and among philosophers: it is a rash and recent
innovation of Quine and other ‘‘literalist’’ philosophers, myself and my
fellow symposiast included.
Chihara’s argument for this claim is essentially that the mathematicians
he has questioned have either expressed puzzlement when asked about their
ontological commitments, or have repudiated suggestions that they are
committed to any ontology at all. Yablo’s argument is on the face of it
different, but I believe at bottom similar. He begins by calling attention to
a very interesting phenomenon, the ‘‘paradox’’ of his title. Let me attempt
my own presentation of it.
According to an old story, when Lindemann settled the ancient problem
of squaring the circle, his colleague Kronecker reduced him to tears by
7
The term ‘‘hermeneutic fictionalism’’ was taken as the title for a large-scale study by Jason Stanley
(2001), which pursues at length a number of the points to be made below, and many more also,
through several different areas of philosophy where fictionalism has become fashionable. In an
alternative terminology, hermeneutic reconstructive nominalism and hermeneutic non-reconstructive
nominalism are called content hermeneuticism and attitude hermeneuticism respectively. This
terminology is used in Rosen and Burgess (2005).
asking, ‘‘What is the value of your investigation of p, since irrational
numbers do not exist?’’ Suppose a nominalist philosopher today were to
say to Wiles, ‘‘What’s all this fuss about your proving there are no natural
numbers x, y, z > 0 and n > 2 such that xn þ yn ¼ zn? Since there are
no numbers at all, a fortiori there are none satisfying that equation.’’8
I imagine most mathematicians would be contemptuous of this speech and
most philosophers – even most nominalist philosophers – embarrassed by it.
According to another story, G. E. Moore famously once argued along
something like the following lines: he held up his hands and said, ‘‘As you
can see, here are two human hands. Since human hands are material bodies
in the external world, there exists an external world of material bodies.’’
Suppose an anti-nominalist philosopher today were to hold up his hands
and say, ‘‘As you can see, the number of my hands is two. Since the number
two is an abstract entity, abstract entities exist.’’ Would not most mathe-
maticians be baffled and bewildered by this argument, and would not most
philosophers – even anti-nominalist philosophers – balk and boggle at it?
On the one hand, objections to what mathematicians and others would
ordinarily say about mathematical entities on the grounds of their alleged
non-existence are regarded with scorn, while on the other hand, purported
proofs of their existence are viewed with suspicion. Yablo argues that this
double attitude is explicable if we assume that mathematical assertions are
meant non-literally, as fiction, or rather, in this preferred terminology, as
metaphor. For that would make both objections on the grounds that they
are not literally true, and attempts to prove that they are literally true,
equally off the mark in opposite directions. As I said earlier, I think Yablo’s
argument ultimately relies on the same kind of consideration as Chihara’s,
namely, mathematicians’ puzzlement at or repudiation of philosophical
theses and arguments about the existence of the entities they study. I also
think the positions of Chihara and Yablo ultimately involve similar mis-
takes about the nature and meaning of ‘‘commitment’’ and ‘‘literalness.’’
Let me begin with commitment. The reason Quine and others have
spoken of the ‘‘commitment’’ of mathematical, scientific, and everyday
thought to mathematical and other abstract entities, is precisely because
they did not want to speak of mathematicians’ or scientists’ or lay-persons’
‘‘assertions’’ of or ‘‘beliefs’’ in the existence of such entities. Mathematicians
qua mathematicians do address questions about whether there are prime
10
numbers greater than 1010 , but they generally do not spend much time
talking, and presumably do not spend much time thinking, about the
8
This example is from a talk by the late George Boolos.
question whether there are any such things as numbers at all: hence the
inappropriateness of speaking of their ‘‘assertions’’ or ‘‘beliefs’’ about such
questions. Quine’s claim was that they are committed to an affirmative
answer to this question, because what they do assert and believe, that there
10
are prime numbers greater than 1010 , implies that there are prime numbers
and therefore that there are numbers, and it has this implication whether or
not they ever acknowledge it, and indeed even if they repudiate it, when
talking philosophy rather than mathematics.
Now for literalness. Quine does allow that someone might say, ‘‘There are
10
prime numbers greater than 1010 ’’ and yet not be committed to its impli-
cations. For one could say it and not really mean it, or not really believe it,
whether or not one were capable of articulating what it is that one does mean
and believe when saying it. One could say it and not intend it to be
understood ‘‘literally.’’ But what does this mean? If the function of the
word ‘‘literal’’ were to indicate the presence of something positive, some extra
enthusiasm perhaps, then it would indeed be very doubtful whether mathe-
10
maticians who assert that there are prime numbers greater than 1010 mean
it ‘‘literally.’’ But I suggest that the function of the word is in actual fact less
to indicate the presence of something positive than to indicate the absence of
something negative: roughly what in the legal phrase is called ‘‘mental
reservation and purpose of evasion.’’ If this is so, then what is in actual fact
very doubtful is whether mathematicians who assert that there are prime
10
numbers greater than 1010 intend their assertion only as something ‘‘non-
literal.’’ To do that, they would have to be more philosophically self-
conscious than they appear to be. To put the matter another way, the
‘‘literal’’ interpretation is not just one interpretation among others. It is the
default interpretation. There is a presumption that people mean and believe
what they say. It is, to be sure, a defeasible presumption, but some evidence is
needed to defeat it. The burden of proof is on those who would suggest that
people intend what they say only as a good yarn, to produce some actual
evidence that this is indeed their intention.9
9
To mean what one says literally is simply to mean what one says, just as to be a genuine antique is
simply to be an antique. The force of ‘‘literally’’ is not to assert that one is doing something more
besides, but to deny that one is doing something else instead: meaning something other than what one
says, as when one speaks metaphorically, hyperbolically, elliptically, or otherwise figuratively. One
does not have to think anything extra in order to speak literally: one has to think something extra in
order to speak non-literally. Such, at any rate, is the not uncommon view of the meaning of ‘‘literal’’ to
which I subscribe. For more on this point, see Searle (1979). Assuming this point, the hypothesis that
someone is writing or speaking literally is the hypothesis that nothing more is going on than meets the
eye or the ear: the null hypothesis about covert intentions. In the text I call it the default hypothesis,
since I take it there is a defeasible presumption in favor of null hypotheses in general. In the specific
And I submit that the fact that mathematicians tend to be perplexed by
and dubious about philosophical argumentation over the existence of
mathematical entities – the ‘‘frog and mouse battle’’ as Einstein once called
a specific instance of such debate – and even the fact that mathematicians
who are badgered with skeptical questions by skeptical philosophers can be
got to make skeptical noises, are not very good evidence. In this connec-
tion, I remember reading an interview in Scientific American some years
back with Murray Gell-Man, who explained that in his original paper on
the quark hypothesis he avoided claiming that quarks were real in order to
avoid trouble with ‘‘philosophers.’’ This self-censorship perhaps does not
say much for Gell-Man’s courage – Galileo, after all, did not recant until he
was ‘‘shown the instruments’’ – but it is rather revealing as to how we
philosophers are viewed by some of our scientific colleagues, and suggests
that there may be serious difficulties with the methodology of pestering
scientists for opinions on philosophical issues to which they may have given
little or no thought, and accepting their answers as indicative of their
intentions in putting forward the affirmations that they do put forward
when philosophers leave the scene and let them get back to work.
4 CONTRASTING CASES
To underscore these points, I would like to contrast the comparative lack of

evidence for attributing a global ‘‘non-literalist’’ intention to mathemati-
cians generally with two cases where we do have good evidence of ‘‘non-
literalist’’ intent. One case is that of certain local usages and idioms, the
other of certain specific mathematicians.
To take the latter case first, among professional mathematicians there
are a tiny minority who have given serious, sustained thought to philo-
sophical questions. Notoriously they tend to disagree with each other quite
as much as professional philosophers do – think of Hilbert and Brouwer –
so that the quickest way to tell whether they are talking mathematics or
talking philosophy is to listen whether they are agreeing or disagreeing.
Probably among this very small and sharply divided group of mathematician–
philosophers can be found adherents of virtually every position found
case of the null hypothesis about meaning, there is the additional consideration, not mentioned in the
text, that word-meaning and speaker-meaning, though distinct, are not independent. It would be
impossible for words to mean what they do if everyone always used them to mean something else, and
difficult for them to mean what they do unless most people most of the time use them to mean that, so
that a randomly chosen person at a randomly chosen time probably means what he or she says.
among philosophers proper, including positions that regard all of mathe-
matics as just a good yarn, or merely a great game.
Certainly the bulk of Hilbert’s own results, beginning with his basis
theorem, were meaningless or ‘‘ideal’’ statements according to his official
philosophy. The topological work, and notably the fixed-point theorem,
which established Brouwer’s reputation among mathematicians, is largely
false or meaningless according to his intuitionist principles. There is of
course always a question, whenever someone says something in one context
and takes it back in another, whether we should regard the original
affirmation as merely pretense and the subsequent denial as revealing the
person’s real opinion, or whether contrariwise we should regard the cere-
mony of recantation as play-acting, and the real opinion the one expressed
originally. But I think that in the case of the field marshal of the batrachians
and the generalissimo of the rodents we may conclude on the basis of their
extensive philosophical writings that they did indeed mean much of what
they said in their mathematical work as ‘‘non-literal’’ or ‘‘fictitious’’ in some
sense. My earlier point was that in the case of the overwhelming majority of
mathematicians there is no such evidence that unspoken philosophical
caveats accompany their mathematical assertions.
The other case I wish to mention is that of the mathematician who, for
instance, asserts on one page that there is only one non-cyclic group of
order four, and on the next page that among the subgroups of some larger
group there are three non-cyclic groups of order four. How can there be
three of them if there is only one of them? That is a mystery whose solution
is that the assertion of uniqueness was not meant literally, but rather
involved a figure of speech, namely ellipsis: ‘‘unique’’ was elliptical for
‘‘unique up to isomorphism,’’ and what is really meant is that all non-
cyclic groups of order four, including the three that turn up as subgroups of
the larger group alluded to, are isomorphic to each other. This isomor-
phism is, of course, what the proof of ‘‘uniqueness’’ proves.
The difference between such cases of mathematicians locally meaning
this or that form of expression as one or another kind of figure of speech,
and the hermeneutic fictionalist’s claim that mathematicians globally mean
none of what they say literally, is that there is evidence within what
mathematicians say while engaged in mathematical research or teaching, to
indicate, for instance, that the claim of uniqueness of the Klein four-group
is intended to be understood as pertaining not literally to uniqueness, but
to uniqueness up to isomorphism. For one thing, if a student who had not
yet learned all the relevant idioms and usages were to raise a question, the
mathematician would explain. Owning up to the non-literal character of
the uniqueness assertion is something the mathematician does on the job,
not just when interrupted and pestered by skeptical philosophers, or after
hours pursuing some philosophical hobby.
To be sure, it is also possible in this particular example that the
mathematician just does not know what ‘‘unique’’ means, like the mer-
chant whose advertisement I once saw, who claimed that ‘‘Every item in the
store is unique, and many are one-of-a-kind.’’ But there are other classes of
examples of local non-literalness. Indeed, my discussion of the case of
uniqueness assertions is largely inspired by Stewart Shapiro’s discussion
of one extensive class of such cases,10 mathematicians’ frequent use of
dynamic language, as if the functions, for instance, were moving mathe-
matical objects around, changing one mathematical object into another,
and so on – a manner of speaking to which, as Shapiro notes, Plato already
objected. I think again, with Shapiro, that there is good evidence within
what mathematicians say while engaged in mathematical research or teaching
for taking this particular kind of language to be meant non-literally. But
the presence of evidence in this case again contrasts with the comparative
absence of evidence for the hermeneutic fictionalist’s global claim that all
kinds of mathematical language are meant ‘‘non-literally.’’
5 UNDECIDABLE QUESTIONS
My conclusion, then, is that hermeneutic fictionalism is implausible, and

that if one is going to be a fictionalist, one had better be a revolutionary
fictionalist, denying while doing philosophy what is asserted while doing
mathematics, but not pretending that it never was asserted, or pretending
that it was only asserted but was not really meant or believed. And there are
indeed similarities between mathematics and fiction that can be cited by
the revolutionaries that would seem to give no support at all to hermeneu-
tics. One such is the apparent incompleteness of mathematics, suggest by the
Gödel theorems. (I trust that in speaking to an audience of logicians I need
hardly add that the question of the philosophical bearing of these theorems
is by no means an easy one, and that such theorems are, in the words of my
fellow symposiast, the ‘‘beginning of the story’’ and not ‘‘the end of the
story.’’) This is often compared to the incompleteness of fictional tales,
which leave many questions, beginning with the proverbial length of the
protagonist’s nose, undecided and undecidable.
10
In public talks since the later 1970s and in Shapiro (1997).
This similarity would seem to provide less than no support to the
hermeneutic view that mathematicians do intend their mathematical asser-
tions as fictions. After all, mathematicians were making mathematical
assertions for centuries before Gödel, in blissful ignorance of any incom-
pleteness or incompletability phenomena, and even while confidently
asserting that ‘‘in mathematics there is no ignorabimus’’ and ‘‘wir müssen
wissen, wir werden wissen.’’ So the fact that fiction rather obviously leaves
many questions undecided would seem to be a powerful argument for the
conclusion that mathematicians were not thinking of their science as a
branch of creative writing. By contrast, the observation that mathematics
resembles fiction in respect of incompleteness might well be thought to
provide support for the revolutionary view that philosophers should regard
mathematical assertions as fictions. At any rate, the observation often is
cited in support of such a position.
To this sort of observation there are two kinds of response: a more specific
and a more general. The more specific notes that, even if we accept that
Gödelian theorems do show that there are fundamental reasons of mathe-
matical principle why some mathematical questions must remain forever
undecided, this by no means establishes a dividing line with mathematics
and novels, fables, and so on, on the one side, and scientific and common-
sense thought on the other side. For are there not, to begin with, also
fundamental reasons of physical principle why some physical questions
must remain forever undecided? The second law of thermodynamics suggests
the growth of entropy will blur the historical record to the point where many
facts about the past will become irrecoverable. General relativity suggests
information may be swallowed up by black holes and hidden from us forever
beyond their horizon. And then there is quantum mechanics. On top of all
this, there are the apparently undecidable questions that arise wherever there
is vagueness, which is just about everywhere except mathematics. But these
considerations are doubtless familiar, and I need not enlarge on them here.
What I want to consider instead is another line of response, suggesting
that even if it were distinctive of mathematics among what pass for sciences
to present us with unanswerable questions, it would be doubtful that this
feature should be taken as a criterion for distinguishing fiction from fact.
More generally, the other line of response I wish to consider – one vaguely
analogous to Hume on miracles – suggests that it is virtually always more
doubtful that philosophy has arrived at the correct understanding of
‘‘truth’’ or ‘‘existence’’ or what have you, than that what well-established
mathematical, scientific, and commonsense principles tell us are facts are
fantasies, or that what they tell us exists is a phantasm.
Revolutionary mathematical fictionalism is an ‘‘error theory,’’ offering a
‘‘correction’’ to mathematics, and especially to mathematical existence
theorems, much as Brouwer once offered an intuitionistic ‘‘correction’’ to
his own Brouwer Fixed Point Theorem.11 At the same time, it is offering a
‘‘correction’’ to science insofar as it is formulated mathematically. The line
of response against such philosophical ‘‘corrections’’ to mathematics and
science that I want to consider is the one expressed in the Credo of my
colleague David Lewis. Rosen and I quoted it in our book, and I will not
quote it again here; but for those who have not read it, its point is that,
given the comparative historical records of success and failure of philoso-
phy on the one hand, and of mathematics on the other, to propose
philosophical ‘‘corrections’’ to mathematics is comically immodest.
And indeed, the historical record of philosophical ‘‘corrections’’ to
mathematics and science from Bellarmine’s ‘‘correction’’ of Galileo – an
early form of fictionalism or ‘‘constructive empiricism’’ – onwards, has
been pretty dismal. One argument against revolutionary fictionalism is
thus just that, given the historical record, on simple inductive grounds it
seems extremely unlikely that philosophy can do better than mathematics
in determining what mathematical entities exist, or what mathematical
theorems are true, and much more likely that for the (n þ 1)st time,
philosophy has got the nature of truth and existence wrong.
6 INTERMINABLE DEBATE
Supplementing this Ludovician line of thought, there is another, deriving

ultimately from Rudolf Carnap’s ‘‘Empiricism, semantics, and ontology’’
(Carnap 1950), that deserves attention. It is a line of thought admittedly in
considerable disrepute today. For instance, the late George Boolos in a
passage in Boolos (1997b) characterizes the Carnapian argument I have in
mind as ‘‘rubbish.’’ And so far as I know, when that paper was delivered
orally this remark raised no substantial murmur from the audience. Any
attempt to rehabilitate an argument that has been thus consigned to the
rubbish bin of philosophical history by so prominent a figure and with so
little protest would be a difficult undertaking, and what I will attempt here
will be something much less ambitious than a full rehabilitation. I will
merely attempt to restate the essence of the argument in a way that strips it
of most of its dated formulations and presuppositions.
11
The theorem can be found in Brouwer (1976) and the ‘‘correction’’ in Brouwer (1975).
To begin with, Carnap notes, as so many others have since, and as I did
above, that the question whether there exist any such things as numbers at
all is one that is never raised by the mathematicians themselves (while
doing mathematics), but only by philosophers (including such few pro-
fessional mathematicians as are also amateur philosophers, when doing
philosophy rather than mathematics). And the most salient fact Carnap
notes about the debate among philosophers is that it goes on and on and on
without ever being settled. The philosophical issue of nominalism and
realism is in this respect like Jarndyce and Jarndyce.
There is an apparent difference in that in Dickens’ novel the parties to
the lawsuit were finally driven mad by the interminable legal wrangling,
whereas few philosophers have been driven over the edge by the issue of
nominalism and realism. But then philosophers are not really in the
position of the suitors. Rather, as Rosen put it when we were once
discussing the matter, ‘‘We’re the lawyers.’’ Of course, in Dickens’ novel,
even the lawyers eventually stopped arguing the case: the case ended
when the entire value of the estate had been eaten up by legal costs,
and there was no money left to pay any more lawyers. So far we philoso-
phers are still being paid, and even to some slight degree paid attention,
and perhaps there is no immediate danger of the value of the estate running
out in our case. Still, Carnap thought that the endless litigation with
no sign of settlement approaching was an indication that something was
badly wrong.
His analysis of the situation was rather complex, and had three strands:
there is a positive phase with two aspects, plus a negative phase. Let me take
the elements one at a time. The two aspects of the positive phase can be
brought out by comparing Carnap’s position on the one hand with that of
one of Yablo’s targets, Crispin Wright, and on the other hand with that of
all nominalists’ target, Quine.
Wright once propounded an argument against nominalism roughly as
follows: I have as many fingers as toes; but as everyone who understands the
concept of ‘‘number’’ knows, to say that I have as many fingers as toes is
equivalent to saying that the number of my fingers equals the number of
my toes; but to say this presupposes that there is such a thing as the number
of my fingers or toes; hence the number ten exists. What the Carnapian
agrees with in this argument is the recognition that concepts come with
rules for their employment, some of which entail affirmative answers to
certain existence questions, so that one has only two choices: either one
rejects the concept, in which case the existence questions cannot even be
asked; or else one accepts the concept, in which case one immediately gets
affirmative answers to those existence questions. One cannot ask the
question and answer it in the negative. One is compelled to accept certain
existence assertions, if one accepts the concept, or the ‘‘framework’’ to
substitute Carnap’s term for Wright’s.
One is not, however, compelled to accept every concept that might be
proposed. In response to criticism by Hartry Field, Wright acknowledged
this point, and in subsequent rounds of debate he and his colleague Bob
Hale have attempted to formulate a general principle that would enforce
acceptance of the concept involved in many though not all cases, including
the particular case relevant to Wright’s original argument, that of the
concept of ‘‘number.’’12 So far they have not come up with a general
principle that has commanded any very widespread support, and the
Carnapian view would be that it is a mistake to attempt to enforce the
acceptance of concepts by a priori arguments.
Quine, and along with him Hilary Putnam at one stage in the evolution
of his views, urged a very different sort of reason for accepting the existence
of numbers (or other abstract mathematical entities to which numbers
could be ‘‘reduced’’). According to Quine, we must (alas, with the greatest
reluctance!) resign ourselves (ah, that it should have come to this!) to
accepting (unbidden and unwelcome!) mathematical entities, because
(most regrettably and unfortunately!) mention of them seems (would
that it were not so!) to be an unavoidable requirement (how cruel a
necessity!) in formulating scientific theories. What Carnap agrees with in
this argument is that the ultimate grounds for accepting the concept
‘‘number’’ is the role it plays in formulating scientific theories, together
with the role scientific theories in turn play in much of our life.
Quine’s argument invites two kinds of objections. On the one hand, there
are the reconstructive nominalists who question that mention of mathemat-
ical entities really is an unavoidable necessity after all. It is the large con-
cessions to nominalism made in the anti-nominalist argument of the ex-
nominalist Quine that invite this sort of objection. On the Carnapian view,
by contrast, a ‘‘framework’’ of mathematical entities is not something to
which we grudgingly resign ourselves because it is a necessary evil, but
something we gratefully adopt because it is an enormous convenience; and
there is no suggestion that if mathematical entities by hook or crook or
somehow could be eliminated, then they should, and hence no invitation to
nominalists to go looking for hooks and crooks.
12
See Hale and Wright (2001), especially ‘‘Responses to Critics.’’
On the other hand, there are objections of the kind especially voiced by
Charles Parsons (1980, p. 150), roughly to the effect that Quine’s argument
does not do justice to the seeming obviousness of elementary arithmetic, and
makes acceptance of ‘‘two plus two is four’’ or ‘‘the number of my hands is
two’’ depend on recondite and abstruse considerations about whether it is
possible to formulate general relativity without referring to tensors, or
quantum mechanics without reference to linear operators. Such objections
are invited by Quine’s refusal to acknowledge any ‘‘conceptual’’ truths, even
conditional ones, and his insistence that all knowledge ‘‘faces the tribunal of
experience’’ as a whole. By contrast, on the Carnapian view, the seemingly
obvious simple statements about the number two just mentioned are taken
to be immediate, given the relevant concept or framework, and consider-
ations of the larger role of that framework in science are only relevant to
explaining why the framework, as well as whatever comes thus immediately
with it, is classified as non-fiction rather than fiction. Moreover, to repeat,
what is important about the larger role in science is merely that it would be
very inconvenient in practice not to use mathematical language, and not that
detailed scrutiny of the most sophisticated theories shows it to be wholly
impossible in principle to do without such language.
It is against the background of this two-sided positive account, accord-
ing to which there are both a local a priori element, the rules constitutive of
the concept, and also a global pragmatic element, motivating acceptance
of the concept, that Carnap attempts to account for the interminability of
philosophical debate between nominalists and anti-nominalists. What is
missing, according to Carnap, when philosophers debate whether numbers
exist – not whether they exist according to standard mathematics, or whether
it is convenient to speak and think as if they existed, but whether they really
exist – is any framework providing rules for assessing assertions of onto-
logical metaphysics about what italics-added really or capital-R Really
exists, in the way that the framework of number-language determines
that ‘‘I have as many hands as feet’’ is a sufficient condition for ‘‘The
number of my hands equals the number of my feet.’’ Or to put the matter
another way, the ontologists are treating the practical question whether to
accept the framework of number-language, to which only considerations of
convenience and the like are genuinely germane, as if it were a theoretical
question of a kind that can only be asked within a preexisting conceptual
framework. Or to put the matter yet another way, and as briefly as possible,
the debate never ends because there is not only a lack of agreement between
the two sides as to what the right answer to the question is, but also a lack of
agreement between the two sides, and for that matter even among those on
the same side, about what counts as a relevant consideration for or against one
or another answer to the question.13
7 MEANINGLESS QUESTIONS
I suspect the reason Carnap’s presentation of the case failed to convince was
largely that he was too much identified with the infamous ‘‘empiricist
criterion of meaningfulness,’’ which certainly has by now been consigned,
if not to the rubbish bin, then at least to archives, where it may be studied
by historians of philosophy, but where it no longer influences current
philosophical debate. According to this criterion – which among other
things sits rather poorly with a recognition of Gödelian phenomena – the
absence of agreed empirical and/or logical criteria for what counts for or
against one or another ontological hypothesis renders argumentation for
these theses meaningless, a kind of nonsense poetry without the poetry.
Instead of a comparison of mathematics with the work of Charles Dickens
or George Elliot, we have here a comparison of philosophy of mathematics
with the work of Lewis Carroll or Edward Lear: it makes the same amount
of sense, though it is less entertaining. Carnap was notorious for the
provocative claim that the issue of nominalism and realism, the ‘‘problem
of universals,’’ is a ‘‘pseudo-problem.’’ This Carnapian thesis is much
stronger than, for instance, the Ludovician claim that induction suggests
it is less likely that philosophy has now solved the problem of universals in a
way that shows mathematics to be in error, than that philosophy has once
again failed to solve the problem. Moreover, the Carnapian thesis implies
or presupposes other very large and controversial claims.
Let me elaborate. One very traditional sort of way to try to make sense of
the question of the ultimate metaphysical existence of numbers would be to
turn the ontological question into a theological question: did it or did it not
happen, on one of the days of creation, that God said, ‘‘Let there be
numbers!’’ and there were numbers, and God saw the numbers, that they
were good? According to Dummett, and according to Nietzsche – or my
13
As mentioned in the review (Burgess 2001), there is a superficial appearance of similarity between
Carnap’s view and that of Mark Balaguer, especially as in Balaguer (1998, pp. 158–79), who maintains
that the question of the existence of numbers is ‘‘factually empty.’’ Really, there is a deep difference
between the two positions, since Balaguer does not distinguish two questions as Carnap does. Like
10
someone who thought that the plain, literal meaning of ‘‘Prime numbers greater than 1010 exist’’ was
1010
‘‘Prime numbers greater than 10 are ingredients of ultimate metaphysical Reality,’’ Balaguer
declines to join Carnap in answering, ‘‘Yes, of course,’’ to the question, ‘‘Do numbers exist?’’ If
doubters as well as deniers of numbers are counted as nominalists, then his refusal to return an
affirmative answer to the question whether numbers exist makes Balaguer a kind of nominalist.
perspective on Nietzsche – this is the only way to make sense of questions
of ontological metaphysics. The Carnapian claim that ontological meta-
physics is meaningless is roughly equivalent to the conjunction of
this Nietzsche–Dummett thesis, ‘‘realism makes sense only on a theistic
basis,’’ with analytic atheism, the thesis that theological language is mean-
ingless. Both these theses are highly controversial: analytic atheism was
explicitly rejected even by outspoken agnostics like Russell, and the
Nietzsche–Dummett thesis is rejected by many philosophers in Australia
who regard themselves as simultaneously ‘‘realists’’ in some strong sense,
and ‘‘physicalists’’ in some sense equally strong.
I myself believe, like Russell, that analytic atheism is false, and suspect,
contrary to the Australians, that the Nietzsche–Dummett thesis is true. If as
I believe the theological question does make sense, and if as I suspect it is
the only sensible question about the italics-added real or capital-R Real
existence of numbers, then I would answer that question in the negative;
but then I would equally answer in the negative the question of the Real
existence of just about anything. For as has been said elsewhere, everything
we have learned about our processes of cognition points in the direction of
the conclusion that even other intelligent creatures, to say nothing of an
Omniscient Creator, would or might well have patterns of language and
thought very different from ours, recognizing categories of objects very
different from those we recognize, or perhaps not even having a category of
‘‘objects’’ at all, if they used a language without a category of nouns, as well
they might.
Since I do not wish to claim that the absence of empirical meaning is
tantamount to the absence of all meaning, where Carnap would put
forward a categorical negative, ‘‘These questions are meaningless,’’ I only
put forward a rhetorical question, ‘‘What are these questions supposed to
mean?’’ But I do agree with Carnap that the question of the Real existence
of mathematical entities does lack empirical meaning, and while I do not
think this settles the question of nominalism, I think it does have an
important bearing on the question of how much mathematics is like
novels, fables, and other forms of fiction.
For consider the question whether, say, the works of Carlos Castañeda
or Rigoberta Menchù are non-fiction or fiction: are they eye-witness
reportage of magic or tragic occurrences, or merely novels masquerading
as anthropology or autobiography? In these cases, we know very well what
it would have looked like for the events in Menchù’s book to have occurred,
and in this age of cinematic special effects, we can even say the same for
Castañeda’s books. By contrast, as regards the question of the ultimate
metaphysical Reality of numbers, we have absolutely no idea of what
difference it would make to how things look; or rather, we have a very
strong suspicion that it would make no difference at all. This is what is
meant by having an empirically meaningful question in one case, and an
empirically meaningless question in the other.14
I think that in view of this radical difference between mathematics and
novels, fables, or other literary genres, the slogan ‘‘mathematics is a fiction’’
is not very appropriate, and the comparison of mathematics to fiction not
very apt. My conclusion is that, whatever may remain to be said for or
against nominalism, about whether we should or should not call ourselves
‘‘nominalists,’’ we should not call ourselves ‘‘fictionalists.’’
14
In other words, I mean no more by saying that the choice between two views is ‘‘empirically
meaningless’’ than that the two views themselves are empirically equivalent. Thus understood, it is
a trivial truism that there is no empirically meaningful difference between any given theory T, and
the fictionalist alternative T * according to which everything observable happens as if theory T were
true, though it is not. What would be highly non-trivial would be the claim that the difference
between T and T * is meaningless tout court et sans phrase. But that would only follow from empirical
meaninglessness assuming a discredited empiricist criterion of meaningfulness.
4
Quine, analyticity, and philosophy

of mathematics
1 TWO SENSES OF ‘‘ F O U N D A T I O N S O F M A T H E M A T I C S ’’
Does mathematics requires a foundation?1 The first thing that must be said
about the question is that the expression ‘‘foundations of mathematics’’ is
ambiguous. Let me explain.
Modern mathematicians inherited from antiquity an ideal of rigor,
according to which each mathematical theorem should be deduced from
previously admitted results, and ultimately from an explicit list of postu-
lates. It also inherited a further ideal according to which the postulates
should be self-evidently true. During the great creative period of early
modern mathematics, there were and probably had to be many departures
from both ideals. But during the century before last, as mathematicians
were driven or drawn to consider less familiar mathematical structures,
from hyperbolic spaces to hypercomplex numbers, the need for rigor was
increasingly felt, and higher standards were eventually instituted. But while
the ideal of rigor may be claimed to have been realized, the ideal of self-
evidence was not.
Considering only the ideal of rigor, the working mathematician’s under-
standing of its requirements, of what is permissible in the way of modes of
definition and modes of deduction of new mathematical notions and
results from old, is largely implicit. Logic, which investigates such matters,
and fixes explicit canons, is a subject in which the algebraist, analyst, or
geometer need never take a formal course. Nor are mathematicians in
practice much concerned with tracing back the chain of definitions and
deductions beyond the recent literature in their fields to the ultimate
1
This question was the title of the Arché Institute conference, August 12–15, 2002, where a preliminary
version of this paper was first delivered. I would like to thank the leadership of the institute and local
organizers of the affair, and especially Crispin Wright and Fraser MacBride, for making my
participation both possible and pleasurable, and my fellow participants for valuable feedback on
the preliminary draft.
66
Quine, analyticity, and philosophy of mathematics 67
primitives and postulates. But there is a place, and one even may go so far as
to say a need, for someone to investigate the choice of postulates, and what
differences a different choice would make. It is these kinds of investiga-
tions, when carried out by mathematical means, that in standard classi-
fications of the branches of mathematics are called ‘‘logic and foundations.’’
Or rather, that label is applied to the kind of investigation just mentioned
plus any other research that can fruitfully apply the same methods.
This, then, is the first and weaker of two senses in which the term
‘‘foundations’’ is used. There will be a need for ‘‘foundations of mathe-
matics’’ in this first sense so long as mathematicians continue to adhere to
an ideal of rigor, and – despite hype from some popularizers about ‘‘the
death of proof ’’ – that would mean for the foreseeable future.
But there is a second and stronger sense, in which one would speak of a
‘‘foundation’’ for mathematics only if, in addition to the ideal of rigor, the
ideal of self-evidence or something like it were realized. Though I have
listed this sense second, it is presumably older, since it is only if something
like the ideal of self-evidence is realized that the metaphor implicit in the
word ‘‘foundations’’ is appropriate. Postulates with something like self-
evidence would provide the firm foundations of the edifice of mathematics,
and this firmness together with the firmness of the rigor by which new
results are built upon old would guarantee the firmness of the higher
stories. This picture is merely the application to mathematics of the picture
offered by epistemological ‘‘foundationalism,’’ according to which the
edifice of knowledge is to be built up from a secure and privileged basis
by secure and privileged means.
Mathematicians have learned to live with the absence of self-evident
postulates, of ‘‘foundations,’’ and in this sense passively acquiesce in the
proposition that foundations (in the foundationalist sense) are not
required. Many philosophers remain more troubled by the situation, and
in consequence either lapse into nominalist, constructivist, or other here-
sies, rejecting orthodox mathematics, or involve themselves in programs to
provide the missing foundations.
For a very familiar specific instance of the distinction between two senses
of ‘‘foundations,’’ we may consider the result we have been taught to call
Frege’s theorem, namely, the deducibility of the basic laws of arithmetic
from the postulate we have been taught to call Hume’s principle, according
to which, if there are as many of these as of those, then the number of
the former is equal to the number of the latter. Uncontroversially Frege’s
theorem is a major contribution to foundations of mathematics in the first
and weaker sense, which is concerned with logical relationships between
postulates and theorems, without too much concern over the status of the
postulates. But there has been a controversy, involving the late George
Boolos and the nominalists on one side, and some of the St Andrews school
on the other, over the status of Hume’s principle, and over whether Frege’s
theorem can provide a foundation for mathematics in the second and
stronger sense.
On one view, Hume’s principle is analytic, Frege’s theorem does provide
such a foundation for arithmetic, and the challenge is to find a way of
providing a similar foundation for more of mathematics. On the other
view, Frege’s theorem does not provide a foundation, and Hume’s princi-
ple is either a synthetic truth, or an untruth. What I want to do here is to
elaborate a third or intermediate position, according to which Hume’s
principle is analytic, but still does not provide a foundation for arithmetic
in the sense of foundationalist epistemology. Naturally this presupposes a
notion of analyticity in which a statement may be analytic but nonetheless
need not be self-evident or a logical consequence of self-evident statements,
or anything of the sort. In sketching the intermediate position I will be
mainly concerned to sketch a conception of analyticity with this feature.
My starting point will be, as my title suggests, the thought of the late
W. V. Quine. His work, by the way, provides another illustration of the
distinction between the two senses of foundations. Quine was, in his
generation, a significant contributor to ‘‘logic and foundations’’ in the
first sense. (I heard it said, by one of my fellow speakers at a memorial
meeting, that asked about his standing on one occasion, he described
himself as ‘‘captain of the B-team’’; and this seems quite a just estimate.)
But Quine was also famously a paradigmatic opponent of epistemological
foundationalism, and the author of the best-known rival to the architec-
tural metaphor. According to Quine, knowledge is not a building but a
web, more or less fixed at the edges by the attachment of observation
sentences to sensory evidence, but underdetermined as to how we spiders
should spin the middle portions, including mathematics, which lies some-
where very near the center.
My reexamination of Quine will be a sequel to an earlier reexamination
of Carnap, entitled ‘‘Mathematics and Bleak House’’ (Burgess, 2004b).
I even thought of entitling the present paper ‘‘Mathematics and Bleak
House II,’’ though in the end I was deterred from doing so when I found
myself having only one occasion to refer to the Dickens novel, mentioning
the police investigator in it, one Bucket, who was for a generation2 the
2
Until the appearance of Sherlock Holmes.
canonical fictional detective in the English-speaking world. The concern of
my earlier paper with the Dickensian title was to re-evaluate Carnap’s
classic ‘‘Empiricism, semantics, and ontology’’ (Carnap 1950). Here I wish
to consider Quine’s reply, ‘‘Carnap’s views on ontology’’ (Quine 1951a) and
more particularly the famous paper of a few months earlier on which that
reply was based, ‘‘Two dogmas of empiricism’’ (Quine, 1951b).
To put the matter very roughly, Quine argued in replying to Carnap
that the position of Carnap presupposed the analytic-synthetic distinction,
the first of the two dogmas Quine took himself to have refuted. Like some
other recent commentators,3 I dissent from the common view that Quine
clearly vanquished Carnap in their exchange. To put matters very roughly
again, my claim will be that Quine almost needs to recognize a notion of
analyticity – and also that he can recognize such a notion, without betray-
ing his core philosophical principles.
2 QUINIANISM VS PLATONISM
Before I examine the differences between Quine and Carnap, I wish to con-
sider what divides them both from the nominalists. And before I consider
what divides them from the nominalists, I wish to consider what Quine,
Carnap, and many nominalists have in common that divides them from
anyone who would really deserve to be called a ‘‘Platonist’’ in anything like a
traditional sense. I will begin with a sympathetic description of – I do not
pretend it is anything like an argument for – a Quinian world-view.
It was the ambition of Galileo, Kepler, and other worthies of their era, by
close reading of the book of nature, to discover the very intentions of its
great Author in writing it; or, to vary the metaphor, to produce a plan of
the universe faithfully reproducing the blueprint used by its great Architect
in constructing it. For Quine, this is a hopeless ambition: no science
produced by human beings can provide a God’s-eye view of the universe,
and this should be evident almost as soon as one begins to view the human
knower as an object of scientific study.
When human knowers are so viewed, human knowledge, including
especially scientific theorizing, is seen as the product of certain organisms
3
Let me in particular cite two useful unpublished works, the Princeton senior thesis of Tom Dixon
‘‘Separating Semantics from Empiricism and Ontology,’’ and the doctoral dissertation of Inga
Nayding, ‘‘Positing Existence.’’ Both see less difference between Quine and Carnap than perhaps
the two saw between themselves, and both attempt, each in his or her own way, to narrow the
difference still further. I derived encouragement from their example, even though my own way of
attempting to narrow the difference is not quite theirs.
in a certain environment, seeking to fulfill certain needs. Beavers build
dams; people first construct hydrological theories, and only then construct
dams (unless, having also constructed ecological theories, they decide not to
construct the dams after all). Scientific theories make intelligible patterns
in the environment as we experience it, and in favorable cases make it
possible to influence (or warn us not to try to influence) the course of that
experience. But what science can make intelligible in our experience, and
how it can make it so, inevitably depends on the nature of our intellects,
and what kinds of experience we are capable of. Thus scientific theory,
product of a certain organism in a certain environment, will inevitably be
the way it is in part because that environment, the universe, is the way it is,
and in part because the organisms, ourselves, are the way we are: there is no
reason to suppose intelligent extraterrestrials, with very different kinds of
sensory experience and very different intellects, would produce the same
scientific theories we have.
For that matter, there is not much reason to believe that even if we
ourselves had it to do all over again we would come up with the very same
theories we have. Thus scientific theories are the way they are partly
because the universe is the way it is, partly because we are the way we
are, and partly because of a third factor: partly because the interaction
between the universe and ourselves has gone the way it has. But if our
theories as they are thus differ from what they might equally well have been
had history gone slightly differently, and differ even more from what the
theories of alien creatures might be expected to be like, then a fortiori they
must differ greatly from the ‘‘theories’’ of an omniscient Creator, and the
ambition of gaining a God’s-eye view of the universe must be unrealizable.
Such is the Quinean picture at the highest level of generality.
To descend to a level of slightly greater specificity, one feature of the way
our intellects work is that language is crucial to scientific thought, and our
language exhibits a comparatively limited range of grammatical forms. In
particular, our scientific theories run very much to sentences of the
noun–verb, subject–predicate, object–property type. As we employ sen-
tences of this type over and over in different contexts and with different
functions within scientific theorizing, our scientific theories will be posit-
ing over and over again different kinds of objects with different kinds of
properties.
To be a little more specific still, all this applies the mathematical
apparatus deployed in our scientific theories. Starting with the use of
numerals as adjectives, we have found that to bring out certain patterns a
shift to using them as nouns is required, and so natural numbers have been
posited. After long speaking only of relations of proportions among geo-
metric magnitudes, we have found it immensely convenient, if not practi-
cally indispensable, to shift to speaking of ratios of magnitudes as objects,
and as objects that can be added and multiplied, and so real numbers have
been posited. Later we have found useful a transition from speaking of real
numbers (plural) with certain properties to speaking of the set (singular) of
such real numbers, thus positing sets as single objects constituted by
pluralities of objects.
We thus end up speaking of different kinds of numbers, sets, functions,
and so forth in sentences of the same grammatical form as ones about
medium-sized dry goods, even though these sentences occupy very differ-
ent positions and roles in the body of our knowledge. ‘‘Septimus is a prig’’
and ‘‘Seventeen is a prime,’’ for instance, have similar grammatical or
logical forms, but very different epistemological positions. The best way
to verify the former would be to locate Septimus in space and time and
interact with him causally: he may sit near us and speak, sending sound-
waves to our ears, by which we detect his priggishness. The number
seventeen does not do anything analogous. It does not sit in Cantor’s
paradise and shine N-rays on our pineal glands, by which we detect its
primality. That arithmetical property is checked by quite different means.
This feature of our actual scientific theories is perhaps one they need not
have had. Whether or not we could have managed to do without any non-
spatial, non-temporal, causally inactive, causally impassive mathematical
objects at all – the partial success of programs of nominalistic reconstruc-
tion of mathematics suggests that in principle this might have been
possible, though examination of the details suggests that in practice it
might not have been feasible – there is no strong reason to suppose that
if we had it to do all over again we would end up with the very same kinds of
mathematical objects. As for the scientific theories of space aliens, not only
is there no strong reason to suppose they would involve distinctive mathe-
matical objects, but what is more, there is no strong reason to suppose they
would involve objects at all, since there is no strong reason to suppose space
aliens would have a language that involved nouns. The Chomskians
maintain that universal grammar is species specific, and in combinatory
and predicate-functor logics we get at least a vague and dim image of what a
language with a grammar radically unlike ours might be like. And as for the
‘‘theories’’ of a Deity, ‘‘we see but through a glass, darkly.’’
Thus Quine – and in this Carnap would surely join him – can concede
to the nominalist that the (particular kinds of) mathematical objects that
figure in our current scientific theories are there largely because of the way
we are (and the way our interaction with the universe has gone) rather than
because of the way the universe is. The positing of numbers may be
extremely convenient and in practice even indispensably necessary for us,
but theories that involve such posits cannot be claimed to give a God’s-eye
view of the universe, to reflect ultimate nature of metaphysical reality, or
anything of the sort.
3 QUINIANISM VS ‘‘ F I C T I O N A L I S M ’’
What Quine – and again Carnap would surely be with him – will not
concede to the nominalist is that any of this gives us any reason at all to
reject current science and mathematics. It gives us no reason why we should
perform, on entering the philosophy seminar room, a ceremony of abjura-
tion, recanting what we habitually say outside that room, when engaged in
scientific work or in daily life. No theory of ours can give a God’s-eye view
of the universe: ‘‘the trail of the human serpent’’ will be over all. But the fact
that any particular theory fails to deliver a reflection of the ultimate nature
of metaphysical reality, uncontaminated by any contribution from us, is no
special grounds for complaint against that particular theory. If there are
grounds for complaint, it is against the human condition.
And thus Quine is willing to speak inside the philosophy room in the
same terms in which he speaks outside it, neither taking back, nor explain-
ing away, nor apologizing for, the use of number-laden or set-laden
language. Rather he is willing to reiterate in his capacity as philosopher
the theorems of mathematics and theories of science. To apply the words
used by the great Scottish philosopher in a somewhat different context,
‘‘Nothing else can be appealed to in the field, or in the senate. Nothing else
ought ever to be heard of in the school, or in the closet.’’4
Here we have direct opposition to the most common kind of nominalist
today. They do take back in their philosophical moments what they assert in
their scientific moments. And for most of them, that is about all they do by
way of expression of their nominalist allegiances: few nominalists any more
are involved in active programs for reconstructing science on a number- and
set-free basis. This kind of inactive nominalism is generally called ‘‘fiction-
alism’’5 and in my earlier paper on Carnap it was the contrast between his
4
David Hume, ‘‘On the practical consequences of natural religion, or, Of a particular providence and
a future State,’’ in Enquiry Concerning the Human Understanding, Part XI.
5
The label is also used by some activists like Hartry Field, whose views I am leaving to one side in the
present discussion.
views and fictionalism that was my topic. Much that I said there about
Carnapian anti-fictionalism applies also to Quinian anti-fictionalism, and I
will recapitulate briefly.
The first thing to be said against ‘‘fictionalism’’ is that the label is
remarkably ill chosen. To say, for instance, that Victorians regarded Bleak
House as fiction, is among other things to say that when Victorians were in
need of the services of a good detective, they did not waste any time looking
for Bucket. They did not use the contents of the novel as a guide to
practice.6 The label ‘‘fictionalist’’ for the dominant contemporary variety
of nominalism is ill chosen because those nominalists who apply the label to
themselves do not in fact regard current mathematically formulated ordi-
nary and scientific lore and theory in the same way they regard the
productions of Dickens or other novelists. They do use such lore and such
theory as a guide to practice.
The analogy ‘‘fictionalists’’ cite ought to be, not with works of imagi-
native literature, but with scientific theories that are regarded as no more
than useful approximations to more complex but truer theories, known or
remaining to be developed. An architect, for instance, designing a modest
private residence, will obviously have to take into account the topography
of the site, the fact that some points on the site are at a higher elevation
above sea level than others, but generally will not take into account the
curvature of the earth, or the fact that points at the same elevation do not lie
on a perfect plane. To this extent, the architect is using as a guide to
practice the primeval theory that the earth is flat, though no architect today
believes any such thing.
Here is a case where a theory is rejected, a theory is not believed, and yet
that theory is not regarded as fiction, as a work of creative writing, but
rather is used as a guide to practice, is employed for practical purposes. The
analogy the mislabeled ‘‘fictionalists’’ ought to be citing is with such cases,
with uses of flat-earth geography when we know the earth is round, or of
Newtonian gravitational theory when we know it is only an approximation
to Einsteinian general relativity, or indeed of this last when we know it
needs a quantum correction, though we do not as yet have a developed
theory of quantum gravity.
But the citation of such examples soon makes evident a serious disanalogy
between the attitude of the nominalist and that of the scientist or engineer
who makes use of a theory known to be only a simplified approximation.
6
At least not in the way they would have used it as a guide had they thought it non-fiction: for all
I know, some readers may have been stirred by it to work for reforms in the Court of Chancery.
This can be brought out by considering an architectural firm suddenly given
a commission for a much larger project than the private homes that are all
they had built before. They would need to take into account the fact they had
been ignoring, that the earth is round, and recalculate whether it is safe to
ignore its curvature on the new and larger scale on which they would be
working. If the project were as vast as the Very Large Array,7 clearly they
could not get away with treating the earth’s surface as flat, they way they can
in building a house on a half-acre lot. For projects of intermediate size, they
would have to think at least for a moment about the question, or ask some
consulting engineer whether flat-earth geography can still be trusted. And
the engineer would base the calculation on round-earth geography: just how
far flat-earth theory can be trusted for architectural purposes is something
that is estimated in terms of round-earth theory.
The situation is similar in all cases of technical and everyday applications
of theories that are not really accepted in the sense of being not really
believed. The scientific and technical application of an approximation is
always framed by some kind of estimate of how good the approximation is,
obtained from some more accurate and credible theory, whether fully
developed or still a work in progress.
The situation is quite dissimilar in the case of the nominalists’ attitude
towards mathematics, pure and applied. Here there is no question of the
untruth of ordinary and scientific theories ever being relevant to practical
questions. Nor is there any nominalistic alternative theory present in the
background, or being sought. Rather, ‘‘fictionalism’’ became the dominant
form of nominalism in the 1990s largely as a result of disappointment with
the search for nominalist alternatives to standard mathematical formula-
tions of scientific theories in the 1980s. The dissimilarity and disanalogy
I have been describing marks nominalism as a non-, un-, or anti-scientific,
and distinctively philosophical concern. For a Quinean, this feature – a
willingness to say that some formulation is acceptable in everyday and all
scientific contexts, however theoretical, but still not acceptable in the
philosophy seminar room – would mark nominalism of the contemporary
kind as involving a species of old-fashioned alienated epistemology as
opposed to the ‘‘naturalized’’ epistemology Quine promoted.
Now, though I cannot discuss Carnap at length here, I believe that the
position he held by the 1950s was not fundamentally different in this respect
7
An arrangement of radio dish-antennas spread out over twenty miles or more, used for radio
astronomy.
from that I have just associated with the name of Quine. Thus, just as Quine
and Carnap and ‘‘fictionalists’’ are all agreed that our current science, partly
owing to its being mathematically formulated, does not present a God’s-eye
view of the universe, so also Quine and Carnap agree that the nominalist
suggestion that current science should therefore be regarded as only a
‘‘useful fiction’’ is to be rejected.
To put the matter another way, the ‘‘fictionalist’’ nominalist considers
that, even when we have answered in the affirmative whether an apparatus
of numbers and/or other mathematical objects does or will figure in the
best physical theory, there remains a further question, whether numbers
really exist, which they answer in the negative. Quine and Carnap agree in
doubting that there is any intelligible question of this form.
4 QUINIANISM VS CARNAPIANISM
Thus the issue between Quine and Carnap on one side, and ‘‘fictionalist’’
nominalism on the other side, is over the intelligibility or appropriateness
of the question, ‘‘Science and common sense aside, are there really num-
bers?’’ The issue between Quine on one side and Carnap on the other can
also be represented as a difference over whether or not a certain question
arises, or more precisely, over whether such a question as, say, ‘‘If I have as
many fingers as toes, is the number of my fingers equal to the number of
my toes?’’ arises in more than one sense. Carnap famously thought there
were two senses to such a question, internal and external to the ‘‘linguistic
framework’’ of number – or, as I will say, to the ‘‘concept’’ of number –
where Quine took there to be only one question.
Taking the concept of number for granted – as the Carnapian is justified
in doing, since the questioner has used the term ‘‘number’’ in framing the
question – the question admits an immediate positive answer: that if
one has as many fingers as toes, then the number of one’s fingers is the same
as the number of one’s toes, is something that ‘‘comes with’’ or ‘‘is part of ’’
the concept, and in this sense is analytic. This is the first, ‘‘internal’’ way of
taking the question.
But there is another, ‘‘external’’ way of taking it. Perhaps in asking the
question what the questioner really means to do is to raise the issue whether
we should accept the concept of number. The ‘‘material mode’’ formula-
tion, which has the appearance of presupposing the concept of number,
may be misleading in this respect. Perhaps what is really being questioned
is, put in the less misleading ‘‘formal mode,’’ simply this: is the concept of
‘‘number’’ to be accepted?
This question is certainly one the Carnapian is willing to discuss, and the
answer to this external question will take longer to give than did the answer
to the internal question. What the Carnapian will insist is that in discussing
why we accept the concept of ‘‘number,’’ questions about the correspond-
ence of that concept to a feature of ultimate metaphysical reality are out of
order. The considerations the Carnapian advances will rather be broadly
speaking ‘‘pragmatic.’’
Thus, for Carnap, the immediate affirmative answer is justified by
appeal to linguistic considerations (by the consideration that Hume’s
principle or something like is analytic); by contrast, the further question
why to accept that concept takes longer to answer, and the ultimate
affirmative answer is justified largely by appeal to pragmatic considerations
(by the consideration of the utility of mathematical formulations in scien-
tific theory). Quine, rejecting the analytic–synthetic distinction, cannot
recognize that there are two questions here.
I have said that I belong to the minority who are not so sure Quine
scored a victory in his debate with Carnap; but what I want now to say is
that if he did score a victory, it was a pyrrhic one. For in rejecting the
distinction between the two ways of taking the question, Quine seems to
deprive himself of any justification for giving it an immediate affirmative
answer. For Quine the answer is ultimately affirmative, to be sure, but his
right to give this answer seems to depend on the whole long story, involving
pragmatic considerations, that has to be told to answer Carnap’s second
question. And this lays Quine open to the objection raised especially by
Charles Parsons, that his account of matters cannot do justice to the felt
obviousness of elementary mathematics. (He acknowledges the existence of
the feeling, but has no apparatus with which to explain it.)
I myself consider this type of objection so serious that a Quinian ought
to want to be able to endorse some notion of analyticity that would allow
an immediate affirmative answer, ‘‘Yes, that’s analytic,’’ or ‘‘Yes, that’s just
part of the concept.’’ It may be an exaggeration to say Quine needs a notion
of analyticity if his position is to be at all plausible, since other means by
which an immediate affirmative answer could be justified have not been
explored; but certainly the most obvious means would be to accept some
notion of analyticity.
(Note that ‘‘fictionalists’’ have no trouble returning, with their fingers
crossed, an immediate affirmative answer to the question, which they will
retract upon entering the philosophy seminar room. Then they will claim
that all they meant was, ‘‘Yes, that’s part of the usual fairy tale,’’ or ‘‘Yes,
that’s how the traditional legend goes.’’)
5 QUINIANISM VS CARNAPIANISM, BIS
Quine, I have said, almost needs to accept a notion of analyticity. The

question is, can he? To answer this question we need to look at Quine’s
arguments against the analytic–synthetic distinction in ‘‘Two dogmas.’’
And what are these two dogmas? The analytic–synthetic distinction is
one. The other is the kind of logical empiricist theory of meaning accord-
ing to which for each meaningful statement there must be a more or less
definite range of verifying and falsifying (or, at least, of confirming and
disconfirming) potential observations. The latter dogma implies the former
or, at least, a theory of meaning of the kind indicated gives one way of
making sense of an analytic–synthetic distinction: the analyticity is the
limiting or degenerate case in which every potential observation counts in
favor of the statement, and none against it. So for Quine, rejection of the
second doctrine is a corollary to rejection of the first.
But Quine’s writings contain also other, more independent, arguments
against the second dogma, based on considerations of underdetermination
of theory by evidence, the lack of crucial experiments, and the like. And
whether on account of these observations or others, by 1950 many veteran
logical empiricists were in the process of giving up their older theories of
meaning. Difficulties with the theory had become notorious by the time of
the Carnap–Quine exchange.8
It is therefore of some interest to consider what kind of notion of
analyticity might survive rejection of the old logical empiricist theory of
meaning. Perhaps the most obvious fall-back theory of meaning would be
of the type philosophers of science have called ‘‘cluster concept’’ theories, of
which Carnap’s later ‘‘Ramsey sentence’’ view can be construed as a variant
version. Let me, avoiding detailed comparisons of different views of the
same general type, merely describe the kind of theory I have in mind at the
highest level of generality.
The background assumption is that there must be something surveyable
and graspable associated with words to guide our usage of them. On almost
any conception, and certainly on Quine’s, we are supposed to be able to
grasp the logical form of a statement, and to grasp the basic laws of logic; for
that is how we grasp the logical implications among statements.9 On almost
8
‘‘Problems and changes in the empiricist criterion of meaning’’ (Hempel 1950) immediately followed
Carnap (1950), ‘‘Empiricism, semantics, and ontology,’’ in the same issue of the same journal.
9
I intend ‘‘laws of logic’’ to be neutral here as between what would correspond in a formal system to
axioms and what would correspond to rules. That there would have to be at least one rule is the lesson
of Quine (1936).
any conception, and again on Quine’s, we are supposed to be able to grasp
the connection between observational terms and predicates and observable
objects and properties. What remains to be considered are theoretical terms
and predicates that are non-logical but also non-observational.
Here the obvious candidate for a surveyable, graspable something asso-
ciated with an item of vocabulary would be some core theory, some basic
laws. For a theoretical term generally is learnt along with a batch of related
theoretical terms, and along with a batch of basic laws involving the given
term and those other terms. The surveyable, graspable, body of basic laws
does in at least one obvious sense guide subsequent usage of the term, and if
one calls this surveyable, graspable, usage-guiding body a meaning, then it
can be said that the basic laws that are members of that body are part of the
meaning of the theoretical term, or part of the concept expressed by that
term, and in that sense analytic.
Now already this fall-back notion of analyticity differs appreciably from
more traditional notions of analyticity, and cannot do all the same work.
Notably, to call something ‘‘analytic’’ in this sense is not at all to call it
unproblematic. What is analytic must be accepted if the concept is accepted,
but then perhaps the concept should not be accepted!10 The basic laws may,
after all, be internally inconsistent, either obviously so, as with Prior’s ‘‘tonk,’’
or less obviously so, as with Frege’s infamous law V.11 Or the basic laws may
have implications clashing with the results of observation. Or the term and
concept may simply be otiose, creating clutter and doing no useful work.
For a Quinian, the fact that the analytic need not be unproblematic
would not be an unwelcome feature. For certainly Quine wants to be able
to say that there is a question whether one should accept a given concept or
not, though indeed such questions are to be resolved by pragmatic consid-
erations, and not on the basis of whether or not the concept corresponds to
an ingredient of ultimate metaphysical reality. To concede that, say,
Hume’s principle, is analytic in the indicated sense would permit Quine
to join Carnap in giving an immediate affirmative answer to the question
whether the number of my fingers is the same as the number of my toes,
10
An alternative closer to the traditional notion would be to count as part of the meaning of the term
only such laws as have the form of an equation or a biconditional and the status of an abbreviatory
definition. But as Quine correctly points out, the same equation or biconditional may have the status
of a definition in one exposition and lack it in another.
11
In such a case, where there is a word, a more or less definite list of basic laws, and an inconsistency in
that list, should we say that there is a concept, but it is an inconsistent and therefore unacceptable
one, or that there simply is no concept? I follow the former usage, but have found some other
philosophers very strongly attached to the latter. I consider this a purely verbal issue in the sense of
‘‘purely verbal issue’’ to be discussed below.
while still doing justice to the thought that somehow its pragmatic role in
scientific theorizing is relevant to the question why we regard Riemann’s
On the Hypotheses Which Lie at the Foundations of Geometry as a contribu-
tion to science, and Dickens’ Bleak House (finished just the year before) as a
contribution to art.
This rather untraditional notion of analyticity would in fact seem to be
just what Quine should want, if he is to be able both to remain faithful to his
core philosophical principle that the ultimate grounds for regarding mathe-
matics or anything else as non-fiction rather than fiction are pragmatic, and
also to explain the felt obviousness of elementary arithmetic. And yet Quine
will accept neither this nor any other analytic–synthetic distinction.
And why not? Why does Quine reject the kind of theory of meaning
I have just been very abstractly and very vaguely describing? Well, he does
not discuss that particular kind of theory specifically, but he thinks he has
reasons for rejecting any theory of meaning at all. He allows that if the
notion of synonymy, or sameness of meaning, could be made sense of, then
meanings could be admitted, being identifiable, if all else fails, with
equivalence classes of expressions under the equivalence relation of syno-
nymy. He allows that synonymy can be made sense of in terms of analy-
ticity, and vice versa. But he famously denies that either of the two can be
made sense of in a scientifically acceptable way.
But as the earliest replies to ‘‘Two dogmas’’ recognized, Quine in making
his case is taking for granted some not-uncontroversial assumptions about
what is scientifically acceptable in a linguistic or other psychological
inquiry. As Grice and Strawson (1956) put it, he is assuming that only
some kind of ‘‘combinatorial’’ or ‘‘behavioral’’ account could make a
linguistic or psychological posit respectable. Quine’s general complaint
about analyticity, as applied to the particular kind of picture of analyticity
vaguely sketched above, amount to just this: that there is no clear combi-
natorial or behavioral criterion for distinguishing which laws count as
‘‘basic’’ and therefore ‘‘constitutive of the meaning’’ of a term, and which
count as ‘‘non-basic’’ and ‘‘additional to the meaning’’ of the term.
The assumption that there would have to be such a criterion before the
notion could be respectable and acceptable is just an instance of the same
behavioristic assumptions that notoriously led Quine to write that ‘‘differ-
ent persons growing up in the same language are like different bushes
trimmed and trained to take the shape of identical elephants. The anatom-
ical details of twigs and branches will fulfill the elephantine form differ-
ently from bush to bush, but the overall outward results are alike’’ (Quine
1960, p. 8). The canonical objection to this assumption is given in
Chomsky’s devastating review of Skinner’s Verbal Behavior (Chomsky,
1959): children’s language grows to resemble that of their parents with
strikingly little input. The resemblance between the two bushes cannot be
simply the result of their being trimmed the same way, because the garden-
ers have done very little trimming.
The conclusion is that if one is to get anywhere thinking about language,
one is going to have to learn to think outside the Skinner box. The brain is
not an organ for cooling the blood, and one is not going to get anywhere
trying to understand the complex relations between sensory input and
behavioral output simply by seeking correlations between the two, treating
everything going on in-between in the brain and elsewhere as a black box.
Nor can one wait until neuroscience finds the relevant physical structures
inside the skull before bringing them into psychological theorizing.
Without some theorizing in advance, one would not even know what to
look for inside the skull, any more than, without Mendel’s positing
‘‘factors’’ and formulating laws of heredity in terms of these, would one
have known what to look for in seeking the physical basis of heredity.
Theoretical posits, including meanings of a kind that bring with them a
distinction between analytic and synthetic, must be allowed even if they are
not directly manifested in observable behavior – provided that they play a
useful role in explaining the phenomena of language and thought.
This, no doubt, most philosophers today will grant, and Quine’s failure
to grant it dates his classic paper more than any other feature. And yet, even
granting that behaviorism of any kind, logical or substantial or methodo-
logical, is to be rejected, and that non-behaviorist programs positing
meanings are to be not just tolerated but even encouraged, there is still
this much to be said for Quine’s skepticism about meanings: no stable
consensus in favor of any one such program has as yet emerged among
either linguists or philosophers. There is nothing that could be called a
body of accepted scientific conclusions about meaning or analyticity that
workers in other areas, such as philosophy of mathematics, can draw upon
and apply to their concerns. And this being so, the question retains some
interest whether a notion of analyticity can be developed without introduc-
ing unobservable theoretical posits, and without stepping outside the
boundaries within which Quine confined himself.
6 QUINIANISM VS INTUITIONISM
Quine’s worry about analyticity, even about the notion of analyticity

sketched earlier that would seem to give him just what he should want
and no more, is that there is no clear, principled way to draw the boundary
between laws that are so ‘‘basic’’ to a term that inquirers who differ over them
would be correctly described as ‘‘attaching a different meaning or concept to
the term’’ or ‘‘speaking of different things,’’ and laws that are not that ‘‘basic,’’
so that inquirers who differ over them would be correctly described as
‘‘attaching the same meaning or concept to the term’’ or ‘‘speaking of the
same things but saying different things about them.’’ Towards indicating a
way to quiet this worry, let me briefly examine some cases of apparent
disagreement. I will consider cases of disagreement in logic, though examples
could also be drawn from theoretical disagreements in empirical science.
At one extreme, let us consider an Australian logician who tells us that,
unlike so many of his or her compatriots who merely claim that there are
some statements of the form ‘‘P and not P’’ that are true, he or she maintains
that all such statements are true. This sounds very radical. But suppose that
in further conversation we find him or her frequently arguing from P or
from Q to ‘‘P and Q,’’ but never to ‘‘P or Q,’’ and from ‘‘P or Q,’’ but never
from ‘‘P and Q’’ to P or to Q. This is a case where one will feel inclined to
say that the radical wannabes are simply attaching a different concept to the
same logical particles, meaning ‘‘and’’ by ‘‘or’’ and vice versa. And it is a case
where a change of terminology would be helpful. Indeed it would actually
put an end to the appearance of disagreement between Bruce or Sheila and
us. That a terminological switch would thus terminate debate is the mark
of a purely verbal dispute.
At the other extreme, consider two logicians Y and X who both accept all
the simple laws of classical logic in textbooks, but who disagree over whether
in a certain complicated case certain premises do or do not imply a certain
conclusion. Y claims he has found a deduction showing they do. X claims
she has found a counter-model showing they do not. This is a case where
one will not feel inclined to say that the two are attaching slightly and subtly
different meanings or concepts to the same words. That kind of suggestion
would, in fact, sound like a bad joke. The obvious way for the two to settle
their differences, given how much they have in common, would be for each
to go carefully over both his or her own and the other’s work, looking for a
flaw in the deduction or in the model, or if necessary to bring in a third
party who shares their classical orientation as referee.
For an intermediate case, consider a group of Dutch logicians who
reject, if not all, then very many, instances of ‘‘P or not P,’’ if not in the
sense of denying them, then in the sense of refusing to affirm them. We
may find that they persistently argue in accordance with the canons of
Brouwer’s or Heyting’s intuitionistic logic rather than of Frege’s or
Russell’s or Hilbert’s classical logic. In this case, sophisticated work by
Gödel and others shows that for substantial parts of discourse, our way of
speaking can be translated into something acceptable to them, and vice
versa.12 But there are definite limits to how far one can go, and there can be
no question of translations putting a complete end to the dispute, which is
therefore not purely verbal.
Nonetheless, one can see that, as a practical, utilitarian matter, it would
be helpful for the sides to distinguish their ‘‘or’’s. And in fact this is
commonly done, the two ‘‘or’’s being called ‘‘classical disjunction’’ and
‘‘intuitionistic disjunction,’’ and similarly for negation and so forth. That
there is something in common is indicated by ‘‘disjunction’’ appearing in
both labels, but that there are significant differences is indicated by the
different qualifying adjectives. The terminological distinction at least dis-
courages partisans of either side from engaging in question-begging
argument – ‘‘we must have P or not P, else we would have not P and not
not P, which is a contradiction’’ – and tends to deflect controversy in the
direction of meta-level discussions, which may not lead to their being very
quickly settled,13 but will at least help clarify what is at stake.
Logicians less wary of ‘‘meanings’’ than Quine seem spontaneously to say
about this case that the meaning of ‘‘or’’ as used by Brouwer is different
from the meaning of ‘‘or’’ as used by Hilbert. And as a matter of fact, Quine
himself says that when the deviant logician tries to deny a classical doctrine,
‘‘he only changes the subject.’’ The appearance of this assertion in Quine’s
Philosophy of Logic,14 however, startled readers of his earlier works, since it
seems to go clean against his whole official doctrine of repudiating mean-
ings. What I would now like to propose is that Quinians can, without
betraying the overarching Quinian principles, incorporate a notion of
meaning that would make what Quine says about the deviationist not a
startling lapse, but exactly right.
My proposal is that the law should be regarded as ‘‘basic,’’ as ‘‘part of the
meaning or concept attached to the term,’’ when in case of disagreement
12
The allusion is to the double negation interpretation, by which classical ‘‘either / or’’ becomes
intuitionistic ‘‘not neither / nor,’’ and to the modal interpretations, by which intuitionistic
‘‘either / or’’ becomes classical ‘‘it is constructively provable that either / or.’’
13
Witness the immense literature spawned by Michael Dummett’s neo-intuitionism, for instance.
14
Quine (1970, p. 81). Two pages later it says that ‘‘he may not be wrong in doing so,’’ meaning that the
reasons or motives for deviations from classical logic must be examined, though Quine in fact
ultimately rejects them on pragmatic grounds. Quine’s position in these passages seems not far from
conceding that the intuitionistically unacceptable classical laws, say, may be called ‘‘analytic’’
provided that term is not taken to imply ‘‘unproblematic’’ or ‘‘incorrigible.’’ This contrasts with
his official view in Chapter 7 of the same work.
over the law, it would be helpful for the minority or perhaps even both sides to
stop using the term or at least to attach some distinguishing modifier to it. Such
basic statements would then count as analytic.15 This proposal makes the
notion of analyticity vague, a matter of degree, and relative to interests and
purposes: just as vague, just as much a matter of degree, and just as relative
to interests and purposes, as ‘‘helpful.’’ But the notion, if vague, and a
matter of degree, and relative, is also pragmatic, and certainly involves no
positing of unobservable psycholinguistic entities, and for these reasons
seems within the bounds of what a Quinian could accept.
There is no denying that utility of the notion is limited by its vagueness,
and yet I think there are some non-trivial cases that are comparatively if not
absolutely clear. The intuitionist case just discussed is one of them, and
I think that the case of greatest interest in the present context, Hume’s
principle, is another. That is to say, I think Hume’s principle can be called
analytic in the sense that it would be helpful if ‘‘fictionalists’’ would stop
saying things like ‘‘I grant that if numbers exist then Hume’s principle is
true of them, but I don’t grant that numbers exist,’’ and instead just
abandoned (inside the philosophy seminar room) the use of the term
‘‘number’’ (with the usual exception of use in negative existentials and in
indirect discourse), and said instead, ‘‘I don’t grant that use of ‘number’ is
accepted, though I grant that if it is accepted, it should be used in accord-
ance with Hume’s principle.’’
What is the difference here? Well, the first formulation tends to turn
discussion in the direction of the question, ‘‘Do numbers exist?’’ And this is
a question that cannot usefully be debated between anti-nominalists and
nominalists, since there is simply no agreement at all between them as to
what would constitute a relevant consideration in favor of or against the
statement that numbers exist. (It was in connection with this point that in
my earlier paper I alluded to the interminable lawsuit Jarndyce and Jarndyce
in Dickens.)
By contrast, much as in the intuitionist case, the second formulation
tends to turn discussion in the direction of issues about what makes it
appropriate or inappropriate to accept a given concept (in a philosophical
as opposed to a scientific context). Though, again as in the intuitionist case,
there is no guarantee that thus turning from the object level to the meta-
level of discourse will result in convergence of opinion, an airing of differ-
ences at the meta-level can at least somewhat clarify why there seems to be
15
As would their logical consequences, at least in contexts where, in contrast to the examples above,
there is no disagreement over logic.
so little chance of achieving agreement at the object level, and debate over
the criteria of acceptability of concepts does not seem as wholly futile as
does debate at the object level, where it seems impossible for either side to
do more than argue in a (vicious or virtuous) circle.
I will not press the point, however, but will leave it to the reader to
ponder. What I want to do instead, before closing, is to consider one all-
too-obvious complaint about the notion of analyticity proposed. The
proposed notion of ‘‘analyticity’’ seems, in its relativity and vagueness, as
well as in its failure to imply unproblematicity, so different from traditional
ones as to make it unhelpful to use the traditional term for it, or at least
unhelpful to use that term without some distinguishing modifier. Hence
by my own principles I ought at least to add a qualifying adjective. Let me
therefore do so, and close by commending to Quinians and non-Quinians
alike a notion that may be called that of the pragmatic analytic.
5
Being explained away
1 A LOGICIAN LOOKS AT NOMINALISM
When I first began to take an interest in the debate over nominalism in

philosophy of mathematics, some twenty-odd years ago, the issue had
already been under discussion for about a half-century. The terms of the
debate had been set: W. V. Quine and others had given ‘‘abstract’’ and
‘‘nominalism’’ and ‘‘ontology’’ and ‘‘Platonism’’ their modern meanings.
Nelson Goodman had launched the project of nominalistic reconstruction
of science, or of the mathematics used in science, in which Quine for a time
had joined him before turning against him. William Alston and Rudolf
Carnap and Michael Dummett had raised doubts about what the point of
Goodman’s exercise could be; and though they had unfortunately been
largely ignored, Quine’s contention that the exercise cannot be successfully
completed had gained wide publicity as the so-called indispensability argu-
ment against nominalism. By contrast, two subtle discussions of Paul
Benacerraf had been appropriated by nominalists and turned into the so-
called multiple reductions and epistemological arguments for nominalism.
While such arguments, if sound, would suffice to establish the nomina-
list position even if Quine were right that mathematical entities cannot be
eliminated from science, nonetheless a number of nominalists were just
then setting out to prove Quine wrong. Reviving Goodman’s project, but
by allowing themselves means Goodman had not been willing to allow
himself, they hoped to succeed where Goodman had failed: they hoped to
find a way of interpreting standardly formulated scientific theories, which
at least appear to imply or presuppose the existence of such things as
numbers and functions and sets, in alternative theories that would not
even appear to do so.
Now a lot of the work of logicians since the time of Kurt Gödel has
consisted in finding interpretations of one theory in another of a super-
ficially quite different appearance. So an experienced logician should be in
85
a good position to give advice as to how what the nominalists were trying
to do should be done. When I entered the field, I attempted just this:
I undertook to tidy up the ongoing work of nominalists of the period, by
indicating the optimal method of interpreting away numbers and sets in
favor of points and regions of space-time, or interpreting away claims
about the actual existence of abstract numbers into claims about the
possible existence of concrete numerals. But while I was thus not impressed
by claims that the nominalist project was infeasible, I was concerned over
the question ‘‘Why bother?’’ What I doubted was not whether what the
nominalists were trying to do could be done, but whether it was worth
doing.
For logicians are used to thinking of the differences between theories
that can be interpreted in each other as less important than the difference
that exists when there is only an interpretation of a first theory in a second,
and not the other way around. In the latter case, the second theory is overall
stronger than the first, as logicians measure strength of theories, while in the
former case the two theories are of equal strength. To me it was a bit
surprising that so many philosophers seemed to attach great importance to
differences between theories whose assumptions were of the same overall
strength, simply because the interpretation of the one in the other involves
switching what Quine called ‘‘ontological’’ commitments (namely, impli-
cations about what sorts of objects exist) for what he called ‘‘ideological’’
commitments (namely, assumptions about what sorts of predicates and
what sorts of logical operators make sense). It seemed to me that this kind
of difference cannot make much of a difference, because it is simply too easy
to interpret and reinterpret and, like a ‘‘creative’’ accountant, move costs
and benefits back and forth between the ontological and the ideological
columns.
I found philosophers mostly dismissive of this attitude. They typically
would suggest that a logician has the impression that it is easy to reinterpret
theories to change their ontologies only because the logician has been
working with theories about abstract entities, and that creative accounting
is much more difficult when the entities with which one is concerned are
concrete. (Indeed, this supposed difference between abstract and concrete is
behind one of the standard nominalist arguments, the ‘‘multiple reductions’’
argument appropriated from the discussion of John von Neumann’s and
Ernst Zermelo’s rival set-theoretic definitions of numbers in Benacerraf ’s
‘‘What numbers could not be.’’) My own impression, by contrast, was that
the reason not much had been accomplished in interpreting away apparent
implications or presuppositions as to the existence of concrete entities of one
Being explained away 87
sort or another was that very little effort had been put into trying to do so. At
the time I did not, however, attempt to show how it could be done –
an omission I will begin to rectify later on in the present note. I did attempt
to express my reservations and explain why I am not a nominalist, but
found myself even more completely ignored than Alston and Carnap and
Dummett had been.
2 REVOLUTIONARY RUMBLINGS
Meanwhile, however, Gideon Rosen was independently arriving at over-

lapping ideas, and a more persuasive way of presenting them; and I had the
good luck that he preferred to join forces with me on a book, rather than
publish his dissertation as a book on its own. Our approach in A Subject
With No Object (Burgess and Rosen, 1997) was two-pronged.1 We began by
distinguishing, in a terminology carried over from my earlier work, two
spirits in which a nominalist reinterpretation of a scientific theory might be
put forward: the hermeneutic, according to which the nominalist version is
a revelation of what the current version really meant all along; and the
revolutionary, according to which the nominalist version is a rival to the
current version intended to replace it henceforth. Rosen has since elabo-
rated this distinction, making some subdivisions, and his elaborated form
appears in a chapter in Stewart Shapiro’s Handbook of Philosophy of
Mathematics and Logic (2005). I do not want to go into great detail here,
but let me review some of the key points.
Revolutionaries claim their nominalist theories are distinct from and better
than current theories. But better by what standards? Revolutionaries may be
subdivided into the naturalized (citizens of the scientific community) and the
alienated (foreigners to the scientific community) according to whether their
appeal is to scientific standards or to some supposed suprascientific philo-
sophical standards. About the latter little need be said here, since few
contemporary nominalists wish to put themselves in a position like that of
1
Lest footnotes and bibliography become longer than the paper proper, I will refer the reader to the
long list of references at the end of Burgess and Rosen (1997) for the full titles and bibliographical data
of the relevant works of all the various authors alluded to in passing here, from Alston to Zermelo.
Exceptions must be made in the case of two more recent authors whom Rosen and I did not discuss
in our book. One is Yablo, whom Rosen does discuss in Rosen and Burgess (2005). (My role in
producing the chapter involved little more than some rewriting of Rosen’s work to make it fit within
the editor’s word limit.) But I advise the reader to visit Yablo’s website (www.mit.edu/yablo/
home.html) for the most up-to-date listing of his works, since he is actively engaged in producing new
material, all of it intriguing whether one agrees with it or not.
The other exception is Azzouni (2004). My review has appeared as (Burgess 2004d).
Cardinal Bellarmine ‘‘correcting’’ scientific theories of planets or of Norman
Malcolm ‘‘correcting’’ scientific theories of dreaming by appeal to the higher
authority of Aristotle or Wittgenstein (as interpreted by themselves).
Generally, revolutionaries profess to be naturalized. But if they are, if
they think their versions of gravitational theory or whatever are superior
scientifically to standard versions, then one might expect them to publish
their work in theoretical physics journals, or at least to attempt to do so. If
‘‘ontological economy’’ of mathematical apparatus really is as important to
scientists as it is to certain philosophers – something I myself very much
doubt, since it is very difficult to find any clear historical instance of such
a preference – such contributions ought to be welcome. Yet the experiment
of submitting a write-up of a nominalist project to a theoretical physics
journal has never been tried, so far as I know, and candid revolutionaries of
a professedly naturalist stamp would probably concede that if undertaken,
the test would very likely be failed: the papers would not make it through
peer review. But if this is admitted, how can a revolutionary profess to be a
naturalist, adhering to scientific standards in judging theories?
One common line is to claim that though nominalist physics is perhaps
not superior to mathematical physics by the standards of physicists, what
really need to be compared are not just the two versions of physics, but
rather two packages of combined physics and epistemology. Somehow
nominalist revision, which may make the job of the physicist more diffi-
cult, is supposed to make the job of the naturalized, scientific epistemol-
ogist easier. In what way? It is at this point that nominalists of the school
I have been alluding to bring forward their appropriation of Benacerraf’s
discussion of knowledge in his famous paper ‘‘Mathematical truth.’’ The
so-called Benacerraf problem is the puzzle, ‘‘How could we come to know
anything asserting or implying or presupposing that there are numbers or
functions or sets, given that it does not make sense to ascribe spatiotemporal
location or causal powers to such mathematical entities?’’ Nominalism
provides a very easy answer to this ‘‘How can we?’’ question – namely, the
answer ‘‘We can’t!’’ – which otherwise would be a difficult one, it is said.
This line of thought involves a serious confusion, which can be brought
out by considering what properties a belief must have in order to rank as
knowledge. These are three: justification, truth, and whatever it takes to
bridge the gap between justified true belief and knowledge that was
discovered by Edmund Gettier. But the epistemological argument for
nominalism is not about Gettierology. Nor is it really about truth. (The
nominalist argues that standard mathematical existence theorems cannot
be known to be true as a way of avoiding directly confronting the question
of whether they are true.) So the issue is one of justification. Once this is
appreciated, it can be seen that the whole idea of trading costs to physics
against benefits to epistemology is a muddle. For providing an explanation
of the historical fact that current mathematical and scientific theories have
come to be believed will be an important task for scientific, naturalized
epistemology regardless of whether or not one takes belief in those theories
to be justified. This task is in no way made easier by the assumption that
the belief in question is unjustified.
The obvious anti-nominalist solution to the Benacerraf puzzle is to
suggest that if you cannot think how we could justifiably come to believe
anything implying, say, ‘‘There are functions,’’ then just look at how mathe-
maticians come to believe, say, Gödel’s result, ‘‘There are solutions to the
field equations of general relativity with closed time-like paths.’’ That’s how
one can justifiably come to believe something implying ‘‘There are func-
tions.’’ The revolutionary nominalist who rejects this answer must think the
actual historical process leading to belief in this theorem of Gödel’s is
somehow not a justifiable process of belief-formation. But it is virtually a
tautology that the belief arrived at is justifiable by mathematico-scientific
standards. And hence the revolutionary nominalist’s position, according to
which it is unjustifiable, must involve covert appeal to suprascientific philo-
sophical standards of justification – must be alienated – after all.
To vary the example, let us consider the claim of nominalists who
maintain that what they are appealing to is just ‘‘what science teaches us
about how we humans obtain knowledge,’’ and see how this applies to, say,
the belief that more than a half-dozen books advocating nominalism have
been published in the past three decades or so. By ‘‘books’’ here I clearly do
not mean concrete book tokens, since there are not just ‘‘more than a half-
dozen’’ but hundreds or thousands of such tokens, scattered through
various institutional and personal libraries. So the belief in question is
one about abstract book types, and hence according to the nominalist must
be something ‘‘science teaches us’’ we cannot know. But is this a teaching of
science, or of some Procrustean epistemological theory? If you asked me for
evidence to justify the belief that more than a half-dozen books advocating
nominalism have been published in the past three decades or so, I could
point to various book tokens on the shelves of my office, with titles like
Ontology and the Vicious Circle Principle and Science without Numbers
and Mathematics without Numbers, and names like ‘‘Hartry H. Field’’ and
‘‘Charles S. Chihara’’ and ‘‘Geoffrey Hellman’’ on the title page, along with
dates like 1973 and 1980 and 1989. Can anyone seriously maintain that science
teaches us that this is insufficient evidence?
3 HERMENEUTICAL HI-JINKS
Turning to hermeneutic nominalism, the most obvious objection to its claim

that what appear to be statements about numbers and sets are really state-
ments about something quite different, is the simple lack of evidence for it.
But there is another problem, which can be illustrated by the case of the
proposal to paraphrase away apparent talk of the existence of abstract
numbers as really being talk of the possible existence of concrete numeral-
tokens. The hermeneutic nominalist who resorts to this kind of paraphrase –
and similar remarks would apply to those who favor other kinds – will want
to say, as the revolutionary nominalist does not, that ‘‘There are prime
numbers greater than 1010’’ is true, justifiably believed, and so on, because
‘‘deep down all it really means’’ is something like ‘‘There could have existed
prime numeral-tokens greater than 1010.’’ The trouble is that parity of
reasoning suggests then that ‘‘There are numbers’’ must equally be true,
justifiably believed, and so on, because ‘‘deep down all it really means’’ is
something like ‘‘There could have existed numeral tokens.’’ But whether
‘‘There are numbers’’ is true, justifiably believed, and so on, was the whole
original issue. Certainly this was the question Goodman and Quine were
asking when they first agitated the issue of nominalism (and not, for
instance, some question about hypothetical ‘‘deep structures,’’ in which
neither Goodman nor Quine ever believed). To concede that ‘‘There are
numbers’’ is true, justifiably believed, and so on, is to concede all the anti-
nominalist maintains. (Perhaps anyone who really deserved to be called a
‘‘Platonist’’ in any historically serious sense would want to claim more; but
I doubt that there are any living Platonists in any such sense of ‘‘Platonist.’’)
Such, in brief, were the kinds of arguments Rosen and I put forward in
A Subject With No Object. Since the appearance of that book it has become
apparent, however, that hermeneuticists, like revolutionaries, are divisible
into two subcategories, which Rosen has called content hermeneuticism
and attitude hermeneuticism. The former is the kind of view I have been
discussing so far, about what mathematically formulated statements ‘‘deep
down really mean.’’ The latter is not a view about meaning in this sense, but
about the attitude of mathematicians and scientists and the lay public
towards scientific and mathematical and commonsense theories apparently
involving abstract entities. Attitude hermeneuticism is the view that, con-
trary to the common assumption of the anti-nominalist, revolutionary
nominalist, and content-hermeneutic nominalist, such theories are not
really believed. As developed by Steve Yablo and others, the attitude-
hermeneutic view has been the dominant version of nominalism over the
better part of the past decade, though the attitude-hermeneuticist line has
zigged and zagged a bit over the course of that period.
First we were told mathematics is like fiction. Well, it is not, and this in
two crucial respects. For one thing, our attitudes towards mathematics and
towards fiction are totally different: we rely on mathematics in important
practical applications, as we do not rely on novels, short stories, and the
like. (If we need the services of a good detective, we do not go to Baker
Street.) For another thing, when there is a dispute about whether some
particular text is fiction or non-fiction – the Don Juan books of the elusive
Carlos Castañeda may serve as an example – we at least have a pretty clear
idea what it would be like for it to be true or not true. For instance, we have
a pretty clear idea of what it would be like for a man to smoke a concoction
of dried mushrooms, turn into a bird, and fly off a cliff. By contrast, once
we join the nominalist in abandoning ordinary mathematical standards for
judging when mathematical existence claims are true, or have been
adequately proved, we are left with no other agreed standards.
Then we began to be told that mathematics is like metaphor or some
related figure of speech. Well, again it is not. For one thing, as Rosen argues
in our chapter in Shapiro (2005), metaphorical usages can almost always be
instantly recognized by the speaker as having been meant non-literally, as
soon as anyone raises the issue, whether or not one is able to say in literal
terms what was meant; and again that is far from being true in the
mathematical case. The ‘‘figuralist’’ or ‘‘figurative’’ interpretation seems
to be attributing to mathematicians and scientists and lay people too
philosophically sophisticated an attitude.
I do not know what we will be told next. Fictionalism and figurativism
do not exhaust the options for the attitude hermeneuticist – in ongoing
work Yablo has some very interesting things to say about ‘‘non-catastrophic
presupposition failure’’ – though I think it is clear by now that attitude
hermeneuticism is not something arrived at by first studying some linguis-
tic phenomenon, then noticing that the conclusions one draws about it
have nominalistic implications. Rather, a commitment to nominalism
seems to be there first, and to be what is driving the search for some
linguistic phenomenon or other whose analysis could somehow or other be
applied to support the nominalist position.
Setting Yablo’s developing views aside, let me turn to the latest book-
length work on the issue of nominalism, Jodi Azzouni’s Deflating Existential
Consequence (2004). This work makes the mind-boggling claim that one
can sincerely assert ‘‘There are such things as numbers’’ and even ‘‘‘There
are such things as numbers’ is literally true’’ and still not be ‘‘ontologically
committed’’ to numbers. Azzouni’s position may perhaps be classed as an
extreme version of attitude hermeneuticism, but owing to its extremism it is
perhaps best put in a class by itself. However it ends up being classified, there
is clearly something radically wrong with it.
In the first place, to repeat an earlier observation, whether it is true that
there are numbers was the whole issue, and in conceding that it is true, the
would-be nominalist of this new style is conceding everything the anti-
nominalist maintains. In the second place, the claim about ‘‘ontological
commitment’’ that the new-style would-be nominalist is making is self-
contradictory. For ‘‘ontological commitment’’ was a phrase without use
and therefore without meaning until Quine gave it a meaning by stipula-
tive definition; and that stipulative definition makes sincere assertion that
there are numbers, or that ‘‘There are numbers’’ is literally true, a more than
sufficient condition for ‘‘ontological commitment’’ to numbers.
In an effort to make some kind of sense of Azzouni’s nonsensical claim
I was led to speculate that what he has done has been to take Quine’s phrase
‘‘ontological commitment’’ and substitute for Quine’s understanding of
‘‘ontological,’’ on which the word is merely a fancy synonym for ‘‘existen-
tial,’’ some other understanding of ‘‘ontological,’’ presumably adopted
from some pre-Quinean tradition. So let me, without making any strong
exegetical claim about Azzouni, examine the contrast between pre-
Quinean and post-Quinean senses of ‘‘ontology’’ and its derivatives. For
this purpose we must turn back to a time before the beginning of the debate
on modern nominalism, and I think we do best to turn quite a way back,
right back to the beginning of the modern era.
4 READING GOD’S MIND OR IMPOSING

A SCHEME ON THE WORLD?
My account of the history will be condensed to the point of being a cartoon,

but nonetheless I hope it may help the woods stand out from the trees.
I begin with a much-quoted passage in William James, describing the
attitude of the heroes of the Scientific Revolution, who hoped for a science
that would be nothing less than a reproduction in our minds of the
blueprint for the universe used by the Great Architect:
When the first mathematical, logical, and natural uniformities, the first laws, were
discovered, men were so carried away by the clearness, beauty and simplification
that resulted, that they believed themselves to have deciphered authentically the
eternal thoughts of the Almighty. His mind also thundered and reverberated in
syllogisms. He also thought in conic sections, squares and roots and ratios, and
geometrized like Euclid. He made Kepler’s laws for the planets to follow: he made
velocity increase proportionally to the time in falling bodies; he made the law of
the sines for light to obey when refracted; he established the classes, orders, families
and the genera of plants and animals, and fixed the distances between them. He
thought the archetypes of all things, and devised their variations; and when we
rediscover any one of these his wondrous institutions, we seize his mind in its very
literal intention. (James 2000, p. 29)
To show James is not just making this up, I could reproduced much-
quoted passages from Galileo’s Assayer and Kepler’s Astronomia Nova, but
let me forbear.
The goal for those who accepted this picture was to produce a descrip-
tion of reality ‘‘just as it is in itself,’’ or equivalently a description of the
universe as God sees it, and not as we see it. (Take the invocation of the
Deity literally or metaphorically as you choose.) Such a description would
necessarily be very different from the description of our environment
which we use in everyday life. (According to the seventeenth-century
worthies I have been alluding to, a chief difference would be that the colors
and sounds and odors we see and hear and smell would be gone, and only
size and shape and position, along with speed and direction of motion,
would be left.)
But as David Hume already saw, if one makes one’s standard for
‘‘knowledge’’ the possession of a representation of reality that describes it
‘‘just as it is in itself,’’ then the consequence will ‘‘an universal skepticism’’
and the conclusion that ‘‘knowledge’’ is impossible. Hence Immanuel
Kant’s Copernican revolution. For Kant, the aim is still to separate out,
in our ordinary and scientific accounts of the world, what is contributed by
the world and what by us; but instead of attempting to do this by
producing an account with nothing contributed by us, Kant proposed to
proceed the other way around, by producing an account with nothing
contributed by the world, an account of the pure forms of sensibility and
categories of the understanding supplied by us, into which the world pours
empirical content.
In the century and a half between Kant and Carnap, which I will leap
over in a single bound, there was really surprisingly little change in the
nature of the project. With Carnap there is more talk of ‘‘linguistic frame-
works’’ and less of ‘‘pure forms of sensibility’’ or ‘‘categories of the under-
standing,’’ and there is a shift from claims about what we inevitably must
impose to claims about what we conventionally do impose on the world. But
even if for Carnap there is no one conceptual scheme we must impose on
the world, yet still we must impose some conceptual scheme or other, and
there can be no question of getting behind any and every conceptual
scheme to the world ‘‘just as it is in itself.’’ Alongside this agreement of
substance between Kant and Carnap, there is a disagreement over termi-
nology, and in particular over the role of the term ‘‘metaphysics.’’
Originally this term applied to the attempt to get behind our conceptual
schemes to a God’s-eye view of reality, something Kant and Carnap agree is
impossible. Kant proposes to use it instead for his own project of articulat-
ing just what the scheme our intuition and understanding impose on the
world amounts to. Carnap, by contrast, proposes simply to retire the term.
Thus for Kant ‘‘the future of metaphysics is critique,’’ while for Carnap
metaphysics has no future.
Against Carnap, Quine claimed that while the fabric of our theory of the
world is ‘‘white with convention’’ and ‘‘black with fact,’’ there are no purely
black threads and no purely white threads in it. The point about black had,
in effect, already been conceded, or rather insisted upon, by Carnap when
he argued, contra Moritz Schlick, that the evidence in science consists of
corrigible reports of observations about the furniture and implements of
the laboratory, and not incorrigible reports about sense-data. The point
about white, about the existence of a purely conventional element, was the
issue between Carnap and Quine.
Quine’s contention was just this. Suppose, as Carnap maintains, we
generally favor one linguistic framework or conceptual scheme over another
on grounds of convenience: in attempting to describe the world, we find it
better suits our purposes to do so in using this framework or scheme rather
than that. Well, what sort of fact is this fact that one scheme is more
convenient than another for us to use in attempting to deal with the world?
It would seem to be a fact not just about us, but also about the world: we are
such and it is such that we can more successfully deal with it in this way rather
than that. So the scheme is not, after all, something contributed purely by us,
since part of the reason we choose it is that the world lends itself to description
in terms of these conceptual resources rather than others.
Rightly viewed, the difference between Quine and Carnap here is one of
detail: much more unites than divides them. In particular, Quine has no more
use than Carnap for the kind of pre-Kantian project of attempting to describe
reality ‘‘just as it is in itself.’’ And yet there is a terminological difference
between the two over the term ‘‘ontology,’’ traditionally a near-synonym for
‘‘metaphysics.’’ Quine agreed with Carnap that ontology in this traditional,
pre-Kantian sense is meaningless. Quine, however, differed from Carnap in
what he called the ‘‘ethics of terminology,’’ insisting that if a word was
meaningless, he had the right to give it a meaning by stipulative definition,
and choosing to exercise this alleged right in the case of the word ‘‘ontology.’’
The new enterprise of ‘‘ontology’’ in the post-Quinean sense is simply a
glorified taxonomy, an attempt to catalogue what sorts of objects there are
in reality, not ‘‘just as it is in itself ’’ but as apprehended by us through our
everyday and technical language, our commonsense and scientific theories.
This untraditional use of ‘‘ontology’’ is of a piece with the historically
dubious use of ‘‘nominalism’’ and the historically absurd use of ‘‘Platonism.’’
(In any traditional sense, it is people James is talking about, people like
Galileo and Kepler, who are the Platonists, while an anti-metaphysical
pragmatist like Quine is no more a Platonist than was James.) Why Quine
chose to apply an old label to a new project is to me something of a mystery.
It is clear that having a synonym, ‘‘ontological’’ or ‘‘ontic,’’ for ‘‘existential’’
must have been useful during the heyday of Jean-Paul Sartre. Readers would
have winced if the section of Word and Object entitled ‘‘ontic decision’’ had
instead been entitled ‘‘existential choice.’’ I fear, however, that Quine may
have chosen to use ‘‘ontology’’ mainly to needle Carnap, who seems to have
more than just disliked the word (perhaps because it was a favorite of
Heidegger, who incidentally also used it in a radically untraditional sense).
The danger posed by Quine’s transferring ‘‘ontology’’ from the old project to
the new – rather than coining contrasting labels – is that some may be led to
confuse the two homonymous enterprises.
And just this is what I suspect may have happened in the case of those
recent nominalists who say in one breath, ‘‘I sincerely believe that it is
literally true that there are such things as numbers,’’ and in the next, ‘‘I am
in no way ontologically committed to numbers.’’ This otherwise non-
sensical double-talk becomes less nonsensical if one takes ‘‘ontology’’ in
the second assertion to be meant in a pre-Kantian rather than a post-
Quinean sense. Indeed, while I myself sincerely believe that it is literally
true that there are such things as numbers, I do not believe that the aim of
traditional, pre-Kantian ontology (namely, the aim of getting behind all
conceptual schemes to reality ‘‘just as it is in itself ’’ and cataloguing what
sorts of objects it contains) is a feasible one. Of course, this being my
attitude, I wish to make ‘‘ontological’’ claims, in a traditional, pre-Kantian
sense, neither for abstract objects nor for concrete ones. It is here that
I differ from what seems to be the attitude of the double-talking nomina-
lists, who go on to say in their third breath, ‘‘But I am ontologically
committed to this table and these chairs, and to the moon and the stars.’’
What I see wrong in this kind of nominalism is not its ‘‘anti-realism’’ about
the abstract, but what appears to be its ‘‘realism’’ (in a traditional, pre-
Kantian sense) about the concrete.
5 ABSTRACT SKEPTICISM VS CONCRETE CREDULITY
What I am inclined to conclude from the tendency observable over these last
decades for nominalism to morph from one form to another is that nomi-
nalism can never be defeated by arguments solely about the abstract, since
what feeds it is an underlying naı̈veté about the concrete and our knowledge
thereof. It is for this reason that I welcome recent epistemological arguments
for what I will call – from the Greek word for ‘‘simple’’ – the haplist position.
As the nominalist holds that everything there is is concrete, and hence that
there are no numbers, no books (in the sense of types rather than tokens),
and so on, so the haplist holds that everything there is is simple, not extended
and composite, and hence that there are no chairs and tables, and no moon
and stars – and no people, and in particular no haplist philosophers! Though
the haplist conclusion is absurd, attention to what haplists have to say may at
least help show that the explanation of our knowledge of the concrete is not
so straightforward as nominalists seem to suppose. This is especially so since
the form of the epistemological argument for haplism is so similar to that of
the epistemological argument for nominalism.
The nominalist’s skeptical argument goes something like this: I look at my
hand and see that (counting the thumb as a finger) there is a finger and another
and another and another and another and no more, and conclude that the
number of fingers on my hand is five. But if we look at what fundamental
physics tells us is really going on here, what we find is just this: light coming
from an external source is reflected off my fingers over there to my eye over
here, beginning a process in my body that ends with my forming the belief that
the number of fingers on my hand is five. But assume what you will about
whether, in addition to the concrete fingers, such an abstract entity as the
number five exists or not, no such alleged thing plays any role in this
explanation. If I end up speaking as if there were such a thing, there actually
being such a thing plays no role in explaining why I do: the explanation why
I do must be sought quite elsewhere, in the convenience of positing such
‘‘useful fictions’’ as numbers and sets for purposes of getting on in the world.
The haplist’s skeptical argument goes rather like this: I look over there
and see something brown and chair-shaped, and conclude that there is a
chair over there. But if we look at what fundamental physics, as in Richard
Feynman’s Q.E.D., tells us is really going on here, what we find is just this:
photons coming from an external source are absorbed by the electrons
among the myriad fundamental particles swarming in chair formation over
there, some of which electrons quickly emit other photons directed over
here, initiating a process – and so on. But assume what you will about
whether, in addition to the simple fundamental particles, such an
extended, composite entity as the chair exists or not, no such alleged
thing plays any role in this explanation, in which the electrons and quarks
do all the work. If I end up speaking as if there were such a thing as the
chair, there actually being such a thing plays no role in explaining why I do:
the explanation why I do must be sought quite elsewhere, in the infeasi-
bility of my keeping track of the complex motions of the myriad tiny
fundamental particles, and consequent convenience of positing such ‘‘use-
ful fictions’’ as chairs and tables for purposes of getting on in the world.
Pointing to the parallelism between the two forms of skepticism,
I submit that if the haplist’s is nothing more than a clever sophism (as
I imagine most nominalists would agree it is), then the nominalist’s is no
better. Still, I would like to make a stronger case against the claim that
while ultimate metaphysical reality ‘‘as it is in itself ’’ does not contain
numbers or books, by contrast it does contain tables and chairs, or the
moon and the stars. This brings me at long last to the topic my title was
intended to herald: the reasons for doubting that ultimate metaphysical
reality ‘‘as it is in itself ’’ contains objects of any sort. These reasons were
adumbrated in a section of A Subject With No Object that has been little
read, but I think it is time to refresh and elaborate upon the suggestion
made there, in the hopes of moving the never-ending but ever-changing
debate over nominalism in a new direction.
6 TALKING OF OBJECTS – OR NOT
We speak of the world as containing objects with properties inasmuch and

insofar as we speak a language with nouns and verbs, and sentences with
subjects and predicates. The position to which I subscribe and that I wish
to defend here is that there is no reason to suppose, just because we speak to
each other in such a language, that God speaks to himself in such a
language, or that the object-property structure is a feature not merely of
reality as apprehended by us, but of reality as apprehended by God, or
equivalently, ‘‘as it is in itself.’’ There is nothing wrong with our speaking as
we do, and certainly I do not myself propose to speak otherwise, but there
is nothing uniquely right about it either, and if other intelligent creatures
do not do so, they are not necessarily making a mistake.
Now as a matter of fact, though I have said that we speak to each other in
a language with certain grammatical features, it is not beyond controversy
that all human languages do in fact share these features. Some linguists
have claimed otherwise, as in the following passage from Whorf:
[I]n Nootka, a language of Vancouver Island, all words seem to us to be verbs, but
really there are no classes [of nouns and verbs]; we have, as it were, a monistic view
of nature that gives us only one class for all kinds of events. ‘‘A house occurs’’ or ‘‘it
houses’’ is the way of saying ‘‘house,’’ exactly like ‘‘a flame occurs’’ or ‘‘it burns.’’
These terms seem to us like verbs because they are inflected for durational and
temporal nuances, so that the suffixes of the word for house event make it mean
long-lasting house, temporary house, future house, house that used to be, what
started out to be a house, and so on. (Whorf 1956, pp. 215–16)
And some literary writers have imagined a whole world of speakers of such
languages, as in the following passage from Borges:
Hume noted once for all time that Berkeley’s arguments did not admit the
slightest refutation nor did they cause the slightest conviction. This dictum is
entirely correct in its application to earth, but entirely false of Tlön. The nations of
this planet are congenitally idealist. Their language and the derivations of their
language – religion, letters, metaphysics – all presuppose idealism. The world for
them is not a concourse of objects in space; it is a heterogeneous series of
independent acts. It is successive, not spatial. There are no nouns in Tlön’s
conjectural Ursprache, from which the ‘‘present’’ languages and dialects are
derived: there are impersonal verbs, modified by monosyllabic suffixes (or pre-
fixes) with an adverbial value. For example: there is no word corresponding to the
word ‘‘moon,’’ but there is a verb which in English would be ‘‘to moon’’ or ‘‘to
moonate.’’ ‘‘The moon rose above the river’’ is hlör u fang axaxaxas mlö, or literally:
‘‘upward behind the on-streaming it moon[at]ed.’’ (Borges 1962, p. 23)
(In what follows let me use ‘‘moonate’’ – or perhaps better, ‘‘lunate’’ – since
the other verb suggested already exists in English in a vulgar sense.)
Whorf is speaking about an actual language, and if he is right, then a
noun-free language is not only possible but actual. Unfortunately, how-
ever, though Whorf is speaking of real people, it has been disputed whether
what he is saying about them is really true. Borges, of course, is only
describing a fictional planet. That does not matter for us philosophers,
since all we are interested in is the possibility of speaking a language without
nouns. But Borges does not really show this, since his description of the
language of Tlön does not go into enough detail to convince one that the
bulk of the things we might like to say could be replaced by saying things
using only verbs and adverbial modifiers.
What I wish to review here is a different approach to showing how a
language like English could be translated into a language with only those
grammatical categories. Of course, if one’s only understanding of this new
language were by way of explanations of how to translate it into English or
English into it, no conclusions about ‘‘ontology’’ would follow; so an effort
of the imagination is still required to convince oneself that children could
grow up being spoken to and speaking such a language and no other. I trust
this will not be too difficult, but ultimately the reader must judge.
To begin with, we need to imagine English translated or, as Quine called
it, ‘‘regimented’’ into what logicians call a first-order language (with
predicates only and no singular terms). The possibility of such regimen-
tation is what lies behind Quine’s slogan ‘‘to be is to be the value of a
variable.’’ Here I must assume familiarity, from Quine’s writings or else-
where, with how such regimentation might be attempted. To give at least
one example, consider the following truth:
(1) Whatever lives, changes.
Now (1) can be regimented as follow:
(2) 8x(x lives ! x changes)
Also (2) admits several equivalents, including one involving only negation,
conjunction, and existential quantification:
(3) 9x(x lives (x changes))
In his paper ‘‘Variables explained away,’’ Quine shows how we can
eliminate variables like the x in (3). We first enrich our language with
new operators, the so-called predicate functors, operators that attach to
predicates to form new predicates, defined thus:
(mF)x1 . . . xm Fx1 . . . xm
(jFG)x1 . . . xmy1 . . . yn Fx1 . . . xm Gy1 . . . yn
(rF)x1 . . . xm–1 9xm(Fx1 . . . xm–1xm)
(qF)x1 . . . xm–1 Fx1 . . . xm–1xm–1
(fF)x1 . . . xm–1xm Fx1 . . . xmxm–1
(yF)x1 . . . xm–1xm Fxm . . . xm–1x1
Going from English to Quinese, each expression in the right-hand column

may be abbreviated by the corresponding expression in the left-hand
column. Writing for short ‘‘F’’ and ‘‘G’’ for ‘‘lives’’ and ‘‘changes,’’ (3)
becomes:
(4) 9x (Fx Gx)
It can then be reduced to an equivalent without variables in the following steps

(5) 9x (Fx (mG)x)
(6) 9x (jF(mG))xx
(7) 9x (q (j F(m G)))x
(8) (r (q (j F(m G))))
(9) m (r (q (j F(m G))))
Going back and restoring ‘‘lives’’ and ‘‘changes’’ for ‘‘F’’ and ‘‘G’’ in (9) we
have
(10) m (r (q (j lives (m changes))))
My personal contributions in this area have been two. First, I thought of
combining Quine’s slogan ‘‘to be . . .’’ with his paper title ‘‘. . . explained
away’’ to produce the title of the present paper. Second, I suggested a way of
pronouncing the predicate functors:
(m talks)x x doesn’t talk
(j walks runs)xy x and y respectively walk and run
(r stares at)x x (just) stares
(q destroys)x x self-destructs
(f eats)xy or (y eats)xy x suffers or undergoes eating by y
Applying this suggestion to (10), we can go back from symbols to words in

the following steps:
(11) m (r (q (j lives (doesn’t change)))
(12) m (r (q (respectively live and don’t change)))
(13) m (r (self-respectively lives and doesn’t change))
(14) m (just self-respectively lives and doesn’t change)
(15) doesn’t just self-respectively live and not change
Here we have a noun-free verb phrase. We may, if we wish, supply a

subject, ‘‘The Absolute,’’ or we may indicate that the verb phrase is a
complete sentence in itself by using the obsolete third-person singular
verbal ending -th, as when one eliminates the pleonastic subject pronoun
in ‘‘it rains’’ by writing, as Quine somewhere suggests, ‘‘raineth.’’ We thus
have the choice between two options:
(16a) Monist: The Absolute doesn’t just self-respectively live and not change.
(16b) Nihilist: Doth not just self-respectively live and not change.
Two subsidiary points should be emphasized. First, one needs some way
not merely of making assertions, but also of carrying out arguments in the
new kind of language – some way other than translating back into first-
order terms and applying text-book rules. In fact, John Bacon and others
have supplied proof procedures, which, however, cannot be gone into here.
Second, the point noted by Johann van Benthem should be emphasized,
that if one starts with a many-sorted first-order language, one can apply the
tricks I have been describing to some of the sorts and not the rest of them,
retaining whatever sorts of objects one likes, and eliminating whatever sorts
of objects one does not.
7 THE DARK SIDE
Thus whether one speaks overtly of abstract objects or concrete objects, of

simple objects or compound objects, or indeed of any objects at all, is
optional. My claim is that if children who grew up speaking and arguing
in Monist or Nihilist or some Benthemite hybrid between one or the
other of these and English, it would be gratuitous to assume that covertly
they are ‘‘committed’’ to a full range of sorts of objects, just as if they
spoke a language like ours, with a full range of sorts of nouns. And any
assumption that the divine Logos has a grammar more like ours and less
like theirs would be equally unfounded, I submit. It is in this sense that
I claim any assumption as to whether ultimate metaphysical reality ‘‘as it
is in itself ’’ contains abstract objects or concrete objects, or simple objects
or compound objects, or again any objects at all, would be gratuitous and
unfounded.
This kind of anti-metaphysical claim, if not quite the kind of reason for
it that I have offered, has been characteristic of pragmatism from James
onward. I have mentioned one recent pragmatist, Quine, whom I believe
to hold essentially this sort of view, despite his very regrettable coquetting
with the modes of expression peculiar to early modern metaphysicians.
I need now to say something about another recent pragmatist, Hilary
Putnam.
Putnam is often cited alongside Quine as the second author of the
indispensability argument against nominalism, but as explained in
A Subject With No Object, there is an important difference between
Putnam and Quine here. This is because when Putnam put forward his
indispensability argument he had already committed himself to a doc-
trine of ‘‘equivalent descriptions,’’ according to which there is nothing
to choose between a conventional formulation of mathematics in set-
theoretic terms and an alternative formulation in modal-logical terms.
Thus what he was really claiming to be indispensable for science was
something of the overall strength of classical mathematics, as opposed to
constructive mathematics of one kind or another. He was not making any
claim about the indispensability of ontological commitments (to sets)
specifically, since he thought these could always be traded for ideological
commitments (to modality). From a logician’s point of view, Putnam’s
claim is a good deal more interesting than Quine’s, but this is not the
place to go into that aspect of Putnam’s views.
An aspect that does require discussion is Putnam’s very regrettable
coquetting, like James before him, with the modes of expression peculiar
to traditional idealist metaphysicians, rather as another recent pragmatist,
Richard Rorty, coquettes with the modes of expression peculiar to
contemporary post-modernes. I am alluding to the tendency to say that
the moon and the stars are ‘‘mind-dependent,’’ or worse, ‘‘socially con-
structed.’’ There is something right in what Putnam maintains, and even
in what Rorty maintains, and my hope is that my sketch of an alternative
kind of language can help us separate this correct element from perni-
cious nonsense about mind dependence and social construction – really
amounting to little more than what Quine called a ‘‘use-mention con-
fusion,’’ with or without the added twist of a confusion of academic
radical skepticism with genuine political radicalism – emanating from the
idealist or po-mo Dark Side.
The view from the Bright Side is that if we do choose the conventional
option, and follow the conventional rules for making and evaluating claims
about objects, we must conclude that the moon and the stars long antedate
human mentality and society, and therefore cannot be dependent on the
former and cannot have been constructed by the latter. On the other hand,
if we choose an alternative option, then we will not be speaking about
objects at all, and among the objects of which we will not be speaking or
saying anything will be the moon and the stars, and among the things we
will not be saying about them is that they are mind-dependent or socially
constructed. One either plays the language game by the rules, or does not
play it all, and in neither case is saying ‘‘The moon is mind-dependent’’ or
‘‘The stars are socially constructed’’ a legitimate move. We may choose
between ‘‘moon’’ and ‘‘stars’’ on the one hand and ‘‘lunate’’ and ‘‘stellate’’ on
the other, but if we take the first option, we must say that the moon and the
stars were there long before there were astronomers or human beings or
primates or mammals or animals or life, while if we take the latter option,
we must say that the Absolute was lunating and stellating long before it
began to astronomize or humanize or primatize or mammalize or animalize
or vitalize.
All this is merely by way of avoiding certain objections to pragmatism
resulting from injudicious diction on the part of some of its most distin-
guished advocates. On issues of substance rather than style, I stand with
the pragmatist tradition: I agree with James that ‘‘the trail of the human
serpent is over all,’’ and think that this – and not some lesson about the
impossibility of mathematical knowledge – really is something ‘‘science
teaches us about how we humans obtain knowledge.’’ From the pragmatist
thesis that it is impossible for human beings to obtain a God’s-eye view of
the world, I infer the anti-nominalist corollary that it is pointless to
complain that, for all we can know, mathematical objects may not be part
of the world as God sees it. To be sure, I am well aware that the consid-
erations I have presented above are very far from constituting a knock-down
argument for that anti-nominalist conclusion. But what the course of the
debate over nominalism seems to me to reveal is that arguments are not
what are needed at this point: nominalists are in the grip of a picture, and
that until that grip is shaken, no argument, however cogent, can hope to
accomplish more than to cause nominalism to morph once again into some
new form: knock down one form of nominalism, and another will pop up
in its place. What is really needed is a Gestalt switch, and this the above
sketch of another way of speaking may perhaps help to induce.
6
E pluribus unum: plural logic and set theory
1 THE ORIGIN OF SET THEORY
If one is interested in how best to formulate and motivate axioms for

set theory, it is worthwhile to take another look at the early history of
the subject, right back to the work of its founder, Cantor. Cantor’s defi-
nition of a set was ‘‘any collection into a whole’’ of ‘‘determinate, well-
distinguished’’ sensible or intelligible objects. According to a well-known
quip of van Heijenoort, this definition has had as much to do with the
subsequent development of set theory as Euclid’s definition of point – ‘‘that
which hath no part’’ – had to do with the subsequent development of
geometry. But in fact the notion of a many made into a one, which is
what Cantor’s definition makes a set to be, will repay some study.
In order to give concrete meaning to Cantor’s abstract definition, our
study should begin with a look at the context in which Cantor first felt it
desirable or necessary to introduce the notion of set. As is well known,
Cantor’s general theory of arbitrary sets of arbitrary elements emerged
from a previous theory of sets of points on the line or real numbers. This
itself emerged from work on Fourier series. The technical details of
Cantor’s theorems on this topic are irrelevant for present purposes, but
the general form of his results should be noted.
Cantor’s first result was a certain (uniqueness) theorem for a given series,
which depended on the assumption that at every point or number x
without exception, a certain (convergence) condition holds. He then was
able to relax this assumption, and show that the theorem still holds if there
is only one exceptional point, or only two, or only three, and so on. A more
substantial generalization was that the theorem still holds even if there are
infinitely many exceptional points, provided they are isolated from each
other, in the sense that for every exceptional point there are two points such
that it is the only exceptional point lying between them.
104
E pluribus unum: plural logic and set theory 105
In connection with this last result and further generalizations, Cantor
introduced a new notion. If X is a set of points, let @X be the set of points in
X that are not isolated from other points in X. Then writing E for the set of
exceptional points, and writing Ø for the empty set, the last result men-
tioned may be restated as saying that the theorem still holds if the following
condition is met:
(1) @E ¼ Ø.
Cantor was then able to prove that the theorem still holds if @E has only
one element, or only two, or three, and so on, or even if there are infinitely
many points in @E, provided they are isolated from each other.
Writing @ 2X for @(@X), this last result says the theorem still holds if the
following condition is met:
(2) @ 2E ¼ Ø.
Similarly, a sufficient condition would be @ 3E ¼ Ø, or @ 4E ¼ Ø, and so on.
Writing @ xE for the intersection of all @ n for n ¼ 1, 2, 3, and so on, the
following is also a sufficient condition:
(3) @ xE ¼ Ø.
Cantor went on to iterate the operation @ beyond @ x, introducing in the
process the transfinite ordinal numbers x þ 1, x þ 2, and so on, but we
need not follow him further.
At what stage in this process of generalization does it become indispen-
sable to think of the infinitely many exceptional points as together forming
a single object, a set E to which an operation @ can be applied and
reapplied? Close examination of the form of the theorems shows that for
any fixed finite n the condition @ nE ¼ Ø can be expressed as a condition
mentioning only the exceptional points, and not the set E or the operation
@, while beyond this stage, restatement in terms of just the points and not
the point-set is impossible.
Consider, for instance, the hypothesis that (1) fails but (2) holds, or in
other words the following:
(4) @E 6¼ Ø but @ 2E ¼ Ø.
Without mentioning E or @, this can be expressed, as follows:
(5) Not every exceptional point is isolated from all other exceptional points, but
every exceptional point that is not isolated from all other exceptional points is
isolated from all other exceptional points that are not isolated from all other
exceptional points.
If we introduce a predicate Ex for ‘‘x is exceptional,’’ and use the usual
symbol < for order, and if we spell out explicitly the definition of isolation,
(5) can even be restated in the formalism of first-order logic, with variables
ranging only over points, as follows:
(6) 9u(Eu ! I(u)) & 8u(Eu & I(u) ! 9v19v28v(v1 < v & v < v2 & Ev &
I(v) « v ¼ u))
wherein I(u) is an abbreviation:
I(u) « 9u19u28u0 (u1 < u0 & u0 < u2 & Eu0 « u0 ¼ u).
This is none too perspicuous, and psychologically it is doubtful that Cantor
could have discovered even those of his results that can be thus reformu-
lated without set language, had he not introduced set language at the stage
he did. But logically the results prior to (3) can be reformulated without set
language, both in words, as in (5), and in first-order logic, as in (6).
From a logical point of view, the existential generalization of (4), which
tells us we really are getting a significantly stronger theorem when the
hypothesis is weakened from (1) to (2), is especially interesting. It reads as
follows:
(7) There is a set X of points such that @X 6¼ Ø but @ 2X ¼ Ø.
This, too, can be restated in terms of points rather than in terms of a point-
set, along the lines of (5), as follows:
(8) There are some points such that not every point among them is isolated from
the other points among them, but every point among them that is not isolated
from all other points among them is isolated from all other points among
them that are not isolated from all other points among them.
In this case, however, formalization in first-order logic, along the lines of
(6), is impossible. The initial quantification, ‘‘there are some points . . .’’ is
irreducibly plural, and cannot be rendered in terms of singular quantifiers
‘‘there is a point . . .’’ or ‘‘for every point . . .’’ The status of (7) is thus
somewhere between that of (1) or (2) or (4) on the one hand, which can be
given first-order formalizations quantifying only over points, and that of
(3) on the other hand which cannot be stated without treating sets as
objects.
The kind of irreducibly plural quantification exemplified by (8) was
made an object of special study by the late George Boolos, who gave
simpler examples of the phenomenon, and developed an extension of
first-order logic to handle them. Analogous to (5) and its existential gen-
eralization (8) are the following:
(9) The critics who write for Exceptional magazine admire only each other.
(10) There are some critics who admire only each other.
Here (9) can be given a first-order formalization, while (10), attributed to
Peter Geach and David Kaplan, cannot.
Irreducibly plural quantification amounts to the very last stop before the
introduction of sets. If we are to understand what is most distinctive about
set theory, we must understand what it adds to and how it goes beyond the
mere plural; and to understand that we must first understand the plural and
its logic.
2 PLURAL LOGIC
Unfortunately notation and terminology connected with plural quantifi-

cation have not been standardized. Boolos (1984), Lewis (1991), and
Burgess and Rosen (1997) all differ. The first item of business must be to
indicate the notation and terminology to be used here. To begin with we
have all the apparatus of singular logic. Thus we have , &, , !, «, 8, 9,
and the logical predicate of identity ¼, along with appropriate logical
axioms and rules of implication. In particular we have the usual laws of
identity, namely, self-identity and indiscernibility of identicals:
(1) u ¼ u
(2) u ¼ v ! ((u) ! (v)).
Note the conventions used in displaying these laws. In both (1) and
(2) initial universal quantifiers have been suppressed. In (2), what we
really have is a scheme or rule according to which, for any formula (t),
writing (u) and (v) for the results of substituting u and of substitut-
ing v for each free occurrence of t therein, (2) is to count as an axiom.
In (2) again, the technical proviso is left tacit that the variables u and v
are free for t in (t), which is to say that no free occurrence of t in (t)
occurs within the scope of a quantifier 8u or 9u or 8v or 9v. In (2) yet
again, it is to be understood that there may be parameters, or free
variables not displayed, so that what is really to count as an axiom is
something like:
(3) 8w1 . . . 8wn8u8v (u ¼ v ! ((u, w1 . . ., wn) ! (v, w1 . . ., wn)))
The derivability of the law of symmetry
(4) u ¼ v ! v ¼ u
depends on allowing parameters, the formula (t) to which (2) is applied
being t ¼ u.
In terms of identity one may define certain other notions, notably
distinctness and unique existence:
(5) u 6¼ v « u ¼ v
(6) 9!x(u) « 9u(u) & 9u19u2((u1) & (u2) & u1 6¼ u2).
Here the distinctness predicate 6¼ and the unique existence quantifier 9! are
to be treated not as primitive but as defined, which is to say that they are not
part of the official notation, but rather are unofficial abbreviations. The
definitions (5) and (6) do not count as substantive assumptions or axioms,
but merely as abbreviations for tautologies, for biconditionals of the form
A « A.
So much for singular or first-order logic. The principles concerning the
formulation of axioms and the status of definitions that have been set out at
some length above in connection with identity are to be tacitly understood
as still applying as we now move on to plural logic, and when we later move
on to set theory. Turning then to plural logic, to begin with we need to add
three items of notation. First, there are plural variables, xx, yy, zz, and so on.
Second, there are plural quantifiers, written 99 and 88. If 9u and 8u are
read ‘‘there is an object, u’’ and ‘‘for any object, u,’’ then 99xx and 88xx may
be read ‘‘there are some objects, the xs’’ and ‘‘for any objects, the xs.’’ Third,
there is a logical predicate of two places, with singular variables going in the
first place and plural variables in the second, u / xx, which may be read ‘‘u
is one of the xs’’ or ‘‘u is among the xs.’’ Much as the symbol used in set
theory for ‘‘element’’ is a stylized epsilon, the symbol used here for ‘‘among’’
is a stylized alpha.
A question immediately arises about the understanding of the plural
quantifier. In a language with a threefold distinction among singular,
dual, and plural, it would be natural to take ‘‘there are some objects . . .’’ to
mean ‘‘there are three or more objects . . .’’ In a language like English,
where we have only the distinction between singular and plural, it is
natural to take it to mean ‘‘there are two or more objects . . .’’ For instance,
in the Geach–Kaplan example (1.10) – that is, displayed item (10) of x1
above – it is natural to understand ‘‘some critics’’ as meaning two or more
critics. But Lewis (1991) takes ‘‘there are some objects . . .’’ to mean ‘‘there
are one or more objects . . .,’’ while Burgess and Rosen (1997) go further
and take it to mean ‘‘there are zero or more objects . . .’’ The last-named
authors, however, show that taking any one reading as official, the other
readings can be defined in terms of it, so in one important sense it does
not matter which reading we take. The ‘‘zero or more’’ reading will be
adopted here.
To frame a basic deductive system for plural quantification, one adds to
complete such a system for singular quantification the following:
(7) Axiom of Comprehension
99xx8u(u / xx « (u))
(8) Axiom of Indiscernibility
8u(u / xx « u / yy) ! ((xx) ! (yy)).
Axiom (7) is a plural analogue of existential generalization, allowing us, for
instance, to infer (1.8) from (1.5). It says that for any condition there are
some objects such that an object is among them if and only if the condition
holds of it. Which objects are these? The objects of which the condition
holds, of course! (If we adopted the ‘‘one or more’’ reading, comprehension
would have to be formulated as the conditional with antecedent 9u(u)
and consequent (7).) Axiom (8) is a plural analogue of the indiscernibility
of identicals (4). It says that if exactly the same objects are among these
objects as are among those objects, so that these and those are the very same
objects, then any condition that holds of these holds of those.
One can introduce an additional notion and notation, not as primitive,
but as defined, namely, ‘‘the xs are the same as the ys,’’ the definition being
as follows:
(9) xx ¼¼ yy « 8u(u / xx « u / yy).
Then (8) can be abbreviated in the following form, to make clear the
analogy with (2):
(10) xx ¼¼ yy ! ((xx) ! (yy)).
No claim is made that the addition of the basic axioms (7) and (8) to a
complete system of first-order logic produces a complete system of plural
logic. For the moment, however, we have all the logical axioms and rules
we will be needing.
3 EXTENSIONALITY
Two additional, non-logical primitives will be introduced in order to allow

us to express the basic notions of set theory: ßu will be used to express ‘‘u is a
set,’’ and u xx to express ‘‘u is the set of (all and only) the xs.’’ (The symbol
ß seems appropriate since as a ligature it is itself a one made out of a many,
or at least out of a two.) Thus the primitive predicates of the language of
plural set theory, to be used in an axiomatization here, are those of plural
logic, ¼¼ and /, together with ß and ; and the formulas of this language
are built up from atomic formulas involving these predicates by means of
, &, , !, « and the singular and plural quantifiers 8, 9, 88, 99. By
contrast, the primitive predicates of the language of singular set theory are
that of singular logic, ¼, together with 2 and ß; and the formulas of this
language are built up from atomic formulas involving these predicates by
means of , &, , !, « and the singular quantifiers 8, 9 only.
The first axiom to be assumed reads as follows:
(1) Axiom of Heredity
ßu « 9xx(u xx).
The right-to-left direction of (1) expresses only the triviality that if an object
is the set of some objects, then it is a set. The left-to-right direction would
seem to express only the equally trivial converse, that if an object is a set, then
it is the set of some objects; and this indeed is all it expresses if it is under-
stood that the range of the quantifiers or universe of discourse includes all
objects. However, what a sentence expresses changes if the universe of
discourse is restricted, and if it is understood that the range of the quantifiers
may be something less than all objects, what the left-to-right direction of (1)
also expresses is that whenever a set is included in the universe of discourse,
its elements are to be included along with it. It follows that if any of these
elements are sets, their elements also are included, and if any of those
elements are sets, their elements are included as well, and so on for gener-
ation after generation. Hence the name proposed for the axiom.
The second axiom involves a biconditional, and expresses in the two
directions of the double arrow the features that distinguish the concept of a
set made of elements from, on the one hand, the concept of a whole made of
parts in mereology, and on the other hand, the concept of a property
instantiated by objects in theories of universals. The axiom reads as follows:
(2) Axiom of Extensionality
u xx & v yy ! (u ¼ v « xx ¼¼ yy).
The two features of the concept of set this axiom expresses are simply that,
given a set, it is uniquely determined what elements it comprises, while given
some elements, it is uniquely determined what set they compose. By contrast,
a whole may be decomposable into parts in several different ways, while two
properties may be instantiated by exactly the same objects and yet be distinct.
While extensionality may not be an utter triviality, it is still in a sense less a
substantive assumption than a partial explication of the concept of set. It would
be inappropriate to use the word ‘‘set,’’ rather than ‘‘whole’’ or ‘‘property’’ or
whatever, for a concept of which extensionality was not a feature.
The relation of element to set, symbolized 2, which in conventional
axiomatizations is taken as primitive, may now be introduced instead as an
abbreviation. Either of the following will do as a definition:
(3) v 2 u « ßu & 9xx(u xx & v / xx)
(4) v 2 u « ßu & 8xx(u xx ! v / xx).
The equivalence of the right sides of (3) and (4) follows from (1) and the
left-to-right direction of (2). It is also useful to introduce the plural version
of 2, symbolized 22, by the following definition:
(5) yy 22 u « 8v(v / yy ! v 2 u).
In terms of the defined notion 2, the right-to-left direction of (2) yields the
following:
(6) ßu & ßv ! (8w(w 2 u « w 2 v) ! u ¼ v).
The last conditional could be strengthened to a biconditional, since the
converse implication is immediate from the indiscernibility of identicals.
In conventional axiomatizations it is (6) or some variant version thereof
that is usually called the axiom of extensionality. The feature that the
elements determine the set is explicit in this formulation, while the inten-
tion to include within the range of the quantifiers or universe of discourse
all the elements of any set that is itself, though it must be assumed since
otherwise distinct sets might have the same elements within the restricted
range or universe, is left implicit. The introduction of plurals, and the
definition of 2 in terms of have, so to speak, permitted what lies beneath
extensionality in its conventional formulation to be exhibited explicitly in
the form of the axioms (1) and (2). Making the assumption of heredity
explicit will prove to be surprisingly important in x7 below.
Before leaving the topic of extensionality, it should be mentioned that the
formulation of extensionality most commonly used in conventional axioma-
tizations is not (6) itself but a variant. Discussion of this issue requires some
machinery we do not yet have, so it will be necessary to return to the issue of
extensionality later on. For the moment we are done with this axiom.
The remaining axioms will be more substantive and will be existence
assumptions. Before indicating what they are, let me recall that one
tempting existence assumption cannot be made. One cannot assume that
for any objects there is a set of just those objects:
(7) 9u(u xx).
For comprehension tells us the following:
(8) 99xx8v(v / xx « v 2 v).
Then (7) and (8) imply the following:
(9) 9u8v(v 2 u « v 2 v).
But (9) is a logical contradiction. All this is just the Russell paradox of the
set of all sets that are not elements of themselves, adapted to the context of
plural quantification.
This may be the best place to interject a word on the role of alleged
entities called ‘‘classes’’ in the axiomatization of set theory. The word
‘‘class’’ rather than ‘‘set’’ was originally used by Frege for the extensions of
concepts, but it has since come to be used for set-like entities that in some
mysterious way fail to be sets. Usually though not invariably classes are
spoken of as having ‘‘members,’’ while sets have ‘‘elements.’’ Thus member-
ship is an elementhood-like relationship that in some mysterious way fails
to be elementhood. There is a mystifying distinction between two funda-
mentally different kinds of collections, sets and classes, and two funda-
mentally different kinds of belonging, elementhood and membership.
Such, at any rate, is the situation described in the language of splitters, for
whom no set is a class, though every set is coextensive with some class in the
sense that the elements of the set are all and only the members of the class.
On a different usage, favored by lumpers, one has not just coextensiveness
but outright identity. Sets simply are some but not all of the classes, the
others being called proper classes; the sets are usually distinguished from the
proper classes in that each set is assumed to be a member of some class,
while no proper class is a member of any proper class. The mysterious
distinction between two fundamentally different kinds of collections, sets
and classes, is replaced by the mysterious distinction between two funda-
mentally different kinds of classes: those that can and those that cannot be
members of other classes.
If one is willing to accept the mystery, then classes have certain uses in set
theory. In conventional axiomatizations, they permit the reduction of
certain schemes (separation and replacement) to single axioms. But as
Boolos urged, in this use plural quantification over sets can be employed
to replace singular quantification over classes, as will be done in the next
section. Classes are also appealed to in heuristic motivations for certain
extensions of the conventional axioms, so-called large cardinal axioms, and
the fact that they are thus appealed to has been argued by Schindler (1994)
and others to weaken the case for large cardinals. Recently Uzquiano (2003)
has pointed out that at least in most cases of such use, plural quantification
over sets can again be employed to replace singular quantification over
classes, as will be done in one important case in the section after next. The
advantage of employing plural quantification in this way is that it leaves us
able to maintain that there is just one kind of collection, and that set theory
is the most general theory of collections.
It might be thought that, conversely, anything that can be accomplished
with plurals can be accomplished with classes, enabling us to avoid appeal
to an extra-classical logic. But plural quantification over sets cannot, in fact,
always be replaced by singular quantification over classes, a point empha-
sized by Boolos and worth reemphasizing here. To begin by restating in
words what has already been stated in symbols in (8) and (9) above:
(10) It is true that there are some objects such that all and only those objects that
are sets and not elements of themselves are among them.
(11) It is false that there is a set of objects such that all and only those objects that
are sets and not elements of themselves are elements of it.
So ‘‘there are some objects . . .’’ cannot in general be replaced by ‘‘there is a

set of objects . . .’’ Class theory indeed tells us that
(12) There is a class of objects such that all and only those objects that are sets and
not elements of themselves are members of it.
But still ‘‘there are some objects’’ cannot in general be replaced by ‘‘there is a
class of objects . . .,’’ and one need only replace each occurrence of ‘‘set’’ or
‘‘element’’ in (10) and (11) by ‘‘class’’ or ‘‘member,’’ respectively, to see why not.
Another limitation of class-and-set theory is that, unlike plural-and-
singular set theory, it is in a sense incapable of making explicit the
assumption of heredity underlying extensionality. Technically, it would
be possible to replace the plural-logical (3.1), asserting that every set is the
set of some elements, by a class-theoretic version, asserting that every set is
coextensive with some class. But to state this last while keeping element-
hood primitive, and coextensiveness as defined, would be merely to restate
one specific instance of the general scheme of comprehension for classes.
For a genuine statement of the heredity assumption, one would need to
take the notion of a set’s being coextensive with a class to be primitive, and
the notion of an object’s being an element of a set to be defined in terms of
it, namely, defined as the object’s being a member of the class with which
the set is coextensive. And that choice of primitives is extremely unnatural
from a class-theoretic point of view, and to my knowledge no class theorist
has ever made that choice. Of course, the importance of this limitation
cannot be clear until it is seen what use is made of the heredity assumption
in the further development of set theory.
4 LIMITATION OF SIZE
Let us return to that development. The name ‘‘extensionality’’ is a reminder

of Frege’s notion of the ‘‘extension of a concept.’’ Frege really did assume
that every condition determines a set, or more precisely, determines a
‘‘concept’’ and thereby an ‘‘extension,’’ and so fell into paradox. Even before
the discovery of any such paradoxes, however, Cantor (1885) had rejected
Frege’s assumptions in a review of Frege’s Grundlagen:
The author’s own attempt to give a strict foundation to the number-concept seems
to me less successful. Specifically, the author has the unfortunate idea . . . to take as
the foundation for the number-concept what in Scholastic logic is called the
‘‘extension of a concept.’’ He completely overlooks the fact that in general the
‘‘extension of a concept’’ is quantitatively wholly indeterminate. Only in certain
cases is the extension quantitatively determinate, in which cases it can then of
course be assigned a definite number, if it is finite, or power, in case it is infinite.
But for this sort of quantitative determination we must already possess the
concepts of ‘‘number’’ and ‘‘power,’’ and it is getting things backwards to try to
found these latter concepts on the concept ‘‘extension of a concept.’’
Exactly what Cantor meant by ‘‘quantitatively indeterminate’’ (quanti-
tativ unbestimmt) is not entirely clear, but he seems to be alluding to the
kind of distinction within the ‘‘actually infinite’’ that he makes elsewhere,
between merely ‘‘transfinite’’ or ‘‘consistent multiplicities,’’ which do form
sets, and the ‘‘absolutely infinite’’ or ‘‘inconsistent multiplicities,’’ which do
not. The Russell paradox does not arise for Cantor because he never
assumes that every condition determines a set, but only those conditions
that do not hold of too many objects, an assumption known as the principle
of limitation of size.
The principle is subject to differing interpretations, based on differing
understandings of what it is to be ‘‘too many’’ to form a set. In advance of
any deep analysis of how to measure ‘‘size,’’ however, it is already clear that
the limitation of size principle motivates one important axiom, namely,
Zermelo’s axiom of separation, according to which if given things are not
too many to form a set, then any things among them are also not too many
to form a set; or to put the matter another way, given any things that are
among the elements of some set, they also may be ‘‘separated out’’ from the
other elements of the set to form a subset.
Formally, one way to state the axiom is as follows:
(1) ßu ! 9v(ßv & 8w(w 2 v « w 2 u & w / zz)).
It is not hard to see that (1) is equivalent to the following:
(2) ßu & 8w(w / yy ! w 2 u) ! 9v(v yy).
And to eliminate the use of the defined symbol 2, it is also not hard to see
that (2) is equivalent to the following:
(3) Axiom of Separation
8w(w / yy ! w / xx) ! 9u(u xx) ! 9v(v yy)).
Separation could equivalently be formulated as a scheme:
(4) ßu ! 9v(ßv & 8w(w 2 v « w 2 u & (w))).
For (1) is implied by (4), being simply the instance with w / zz as (w);
while conversely comprehension, which for any (w) gives us zs such that w
/ zz if and only if (w), together with (1) implies (4). But clearly no one
becomes convinced of a scheme by becoming convinced separately of each
of its instances, one by one; there must be some single underlying principle,
and much as (3.1) made explicit an assumption underlying conventional
formulations of extensionality, so (3) above makes explicit the single
assumption underlying schematic formulations of separation. Hence (3)
will be taken as our official version. Formulating separation as a single
axiom rather than a scheme will prove to be important in x7 below.
However formulated, the assumption of separation is so fundamental to
Cantorian thought that it is arguably inappropriate to apply Cantor’s word
‘‘set’’ (Menge) to theories (such as Quine’s NF and ML) that do not accept
it. In other words, separation may be regarded as a partial explication of the
concept of set, indicating what sets are supposed to be like if they exist. It is
also, however, a kind of existence assumption. Specifically, it is a relative
existence assumption that, given one set, provides us with others. But it is
not a positive existence assumption. The axioms adopted so far, taken
together, do not yet imply the existence of a single set.
Now as has been said, on virtually any understanding of ‘‘many’’ and
related notions, separation follows from the principle that objects form a
set unless they are too many, because if all of these are among those, then
there are no more of these than of those. Likewise, if there are just as many
of these as of those, and if those are not too many, then these are not too
many either. To extract an axiom from this latter thought, however, we
need to adopt some specific understanding of ‘‘just as many.’’
But here Cantor’s analysis in terms of a correspondence between these and
those is the only live option. Understanding ‘‘just as many’’ according to
this analysis, we get Frankel’s axiom of replacement, according to which if
some things form a set, and some other things are in correspondence with
them, then those other things also form a set; or to put the matter another
way, if to each element of a set there corresponds some one other object,
then the elements of the set may be ‘‘replaced by’’ the corresponding
objects, and we will still have a set. The more precise and formal statement
need not detain us here.
To obtain the further axioms of the usual Zermelo–Frankel (ZF) set
theory, one needs further specific judgments or interpretations of how
many objects are ‘‘too many’’ to form a set. Assuming that two is not too
many, we get the axiom of pairing, assuming – contrary to tradition, but
following Cantor – that infinitely many is not too many, we get the axiom of
infinity, and so on. Yet further judgments or interpretations of this kind
suggest axioms beyond those of ZF, the large cardinal or higher infinity
axioms alluded to earlier, in an open-ended series of increasing strength. But
at present such large cardinal assumptions, not included in ZF, generally are
not taken for granted as axioms by mathematicians. If a theorem depends on
such an assumption, it must be stated as part of the hypothesis of the theorem.
By contrast, the famous or notorious axiom of choice (AC) generally is no
longer, as it once was, stated as part of the hypothesis of a theorem that
depends on it. Today it is generally taken for granted as an axiom, and the
usual system of axiomatic set theory is ZF þ AC ¼ ZFC. The main dis-
cussion, and even the statement, of AC will be postponed to a later section,
but it may be mentioned here that von Neumann has argued that even AC
can be motivated as an additional axiom by appeal to a certain interpretation
of the principle of limitation of size. According to this interpretation, called
the maximality principle, given objects form a set unless there are as many of
them as there are objects altogether. How AC follows from this principle is
among the many points explained in the thorough study of the history of the
principle of limitation of size from Cantor through Zermelo to von
Neumann and beyond undertaken in Hallett (1984).
In the light of this study it can be said that the usual axioms of set theory,
and large cardinal axioms also, can be motivated by appeal to a single
principle, that of limitation of size, provided we allow ourselves an open-
ended series of increasingly more specific understandings of that principle.
But what we do not yet have is a single axiom formalizing a single under-
standing of the principle of limitation of size, from which the usual axioms
can be formally deduced. If we are to obtain set theory from a single
positive existence axiom in addition to those already adopted, another
approach must be tried.
5 THE REFLECTION PRINCIPLE
Another approach is available from the work of Paul Bernays (1961),

building on the work of Azriel Levy, and this rival approach seems more
promising than an appeal to the maximality principle. The background is
as follows. Levy derived from the usual axioms of set theory a result known
as the reflection principle, whose precise statement need not detain us.
Bernays showed that most of the usual axioms could be deduced from a
strengthened version of the principle, which thus could be adopted as an
axiom in their place. He also showed that several large cardinal axioms – for
the cognoscenti, those known as the axioms of inaccessible and of Mahlo
cardinals of all orders, an upper bound for the cardinals obtainable in this
way being indescribable cardinals – could also be deduced from his version
of the reflection principle.
One may attempt to motivate the principle by appeal to the idea of
limitation of size. This heuristic motivation is perhaps best presented as a
list of principles, the first of which is Cantor’s vague principle of limitation
of size, each succeeding one of which represents a plausible way of making
the preceding one more precise, and the last of which will provide the basis
for the formal axiom whose consequences are to be explored. Without
further ado, here is the list. All principles have the general form: ‘‘The xs
form a set unless . . .’’
(1) . . . they are indeterminately or indefinitely many.
(2) . . . they are indefinably or indescribably many.
(3) . . . any statement that holds of them fails to describe how many they are.
(4) . . . any statement that holds of them continues to hold if reinterpreted to be
not about all of them but just about some of them, fewer than all of them.
(5) . . . any statement that holds of them continues to hold if reinterpreted to be
not about all of them but just about some of them, few enough to form a set.
(Note that the transition from (4) to (5) would be automatic assuming the
maximality principle.)
Our interest will in fact be limited to the case where the xs are all objects.
If our general understanding is that our quantifiers range over all objects,
then every statement is about all objects, and the principle thus becomes:
(6) Any statement that holds continues to hold if reinterpreted to be just about
the elements of some set t.
With respect to what is expressed by , the macrocosm of all objects is
‘‘reflected’’ in the microcosm of elements of t.
One must be careful about how one applies this reflection principle.
One thing that holds is that no object is a set that has all objects as elements.
Applying reflection might seem to give a set t for which it holds that no
object is a set that has all elements of t as elements; and, of course, there can be
no such set. This example, however, is an improper application. Proper
application only yields a set t for which it holds that no element of t is a set
that has all elements of t as elements; and this is entirely non-paradoxical. But
clearly what is needed at this point is a formal restatement.
As a preliminary we need the notion of relativizing or restricting
quantifiers to a formula (u). By this we mean replacing statements
about some or any object or objects by statements about some or any
object or objects of which the predicate holds. Formally, given any
formula , we form the relativization by replacing each quantifica-
tion of the kind shown on the left side below by the quantification shown
beside it on the right side.
8u 8u((u) ! . . .)
88xx 8xx(8u(u / xx ! (u)) ! . . .)
9u 9u((u) & . . .)
99xx 99xx(8u(u / xx ! (u)) & . . .)
Of special interest will be the case where the formula (u) is u 2 t for some
parameter t. In this case one writes t for and calls it the relativization to t.
In this connection it is often useful to use the abbreviations on the left
side below for the expressions on the right side:
8u 2 t (. . .) 8u(u 2 t ! . . .)
88xx 22 t (. . .) 8xx(xx 22 t ! . . .)
9u 2 t (. . .) 9u(u 2 t & . . .)
99xx 22 t (. . .) 99xx(xx 22 t & . . .)
Then the replacements needed to obtain t amount to the following:
8u 8u 2 t (. . .)
88xx 88xx 22 t (. . .)
9u 9u 2 t (. . .)
99xx 99xx 22 t (. . .)
Our principle then becomes the following reflection axiom:
(7) Axiom of Reflection:
! 9tt.
Note that (7) is meant to apply to formulas written out in primitive
notation, without abbreviations, or at worst, involving only abbreviations
that, like 6¼ and unlike 9!, do not involve any quantifiers in their defini-
tions. An abbreviation involving quantifiers needs to be written out to
make the quantifiers explicit, so that they can be relativized.
6 DEDUCING EXISTENCE AXIOMS OF ZF
The set theory based on plural logic and the non-logical axioms of heredity
(3.1), extensionality (3.2), separation (4.3), and reflection (5.7) will be called
BB for Bernays–Boolos. It turns out that all the usual existence axioms of
ZF, as well as the large cardinals obtainable in the manner of Bernays, can
be obtained from BB by deduction as logical consequences.
Bernays was not especially concerned with the intuitive or heuristic moti-
vation for reflection, and in fact assumed reflection in an ostensibly stronger
version than (5.7), with an additional technical condition on t that it is not
immediately obvious how to motivate by appeal to the idea of limitation of
size. Our first task must be to show that this ostensibly stronger version
actually follows from the version (5.7) that has been taken as axiomatic here.
The first step in the deduction is to replace the version of reflection in
(5.7), which does not even explicitly state that t is a set, by a version that
does state explicitly at least that much. The deduction uses a trick
employed over and over by Bernays. The trick consists in noting that any
statement of course implies its own conjunction with any axiom or
theorem C. And so (5.7) yields the following:
(1) ! 9t(Ct & t ).
Taking as C the trivial truism 9u(u ¼ u) we get the following:
(2) ! 9t(9u(u 2 t & u ¼ u)& t ).
The existence of an element u of t implies that t is a set by (3.3), so we have

the following:
(3) ! 9t(ßt & t ).
The derivation of (3) is a trivial illustration of how (5.7) can yield stronger
versions of reflection.
For a slightly less trivial instance, take as C the axiom (3.1). What we get
is the following:
(4) ! 9t(ßt & 8u 2 t (ßu « 99xx 22 t(u xx)) & t ).
What the left-to-right direction of the second conjunct (3.1)t says is that any
set that is an element of t is the set of some objects that are themselves
elements of t. This is merely a long-winded way of saying that t is a set
such that every element of an element of t is an element of t, a property
called the transitivity of t. Thus we have obtained the following strengthen-
ing of (5.7):
(5) ! 9t(t is transitive & t ).
It was, of course, crucial to the derivation of (5) that has been taken as
primitive. For recall that in relativizing a formula, abbreviations defined
using quantifiers are supposed to be written out in primitive terms. In
principle this includes even the symbol 2, which here is officially an
abbreviation defined by (3.3). What (v 2 u)t amounts to is the following:
(6) 99xx 22 t(u xx & v / xx).
If, however, t is transitive, and u is an element of t, then it will automati-
cally be the case that the objects of which u is the set will all be elements of t,
and therefore (3) is equivalent to the simple v 2 u. In other words, in
practice the abbreviation 2 does not need to be written out in primitive
terms after all, so long as the set t can be taken to be transitive, which by (5)
it always can be.
The principle (5), with allowed to contain 2, is the version of
reflection assumed by Bernays, apart from his using the language of
singular quantification over classes rather than that of plural quantification
over sets, which makes no difference to the subsequent deduction of further
existence axioms. Thus at this point one could simply cite Bernays for the
deduction of further axioms, and close the present section.
It may be instructive, however, to indicate the first few steps in the
further deduction. The first step will be yet another application of the trick
that took us from (5.7) to (3) and from (3) to (5). A transitive set t is said to
be supertransitive if any subset of any element of t, as well as any element of
any element of t, is itself an element of t. By taking as C in (1) the axiom
(4.3), the following further strengthening is also obtainable:
(7) ! 9t(t is supertransitive & t ).
Note that it was crucial for obtaining supertransitivity that separation is
formulated as a single axiom (4.3), not as a scheme (4.4). For the single-
axiom formulation one needs either to use singular quantification over
classes, as Bernays did, or plural quantification over sets, as was done here.
If one attempts to stick with singular quantification over sets, it will be
impossible to obtain supertransitivity, and in consequence impossible to
obtain certain existence axioms.
But having strengthened (5) to (7), Bernays rapidly obtains the axioms of
pairing, union, power, infinity, replacement. For instance, following a
deduction he attributes to Klaus Gloede, given any two objects a and b,
we take as our the following logical truth:
(8) 9u(u ¼ a) & 9u(u ¼ b).
Reflection according to (3) then gives us the following:
(9) 9t(ßt & 9u 2 t(u ¼ a) & 9u 2 t(u ¼ b)).
And this in turn implies the following:
(10) 9t(ßt & a 2 t & b 2 t).
This already is an alternative formulation of the axiom of pairing. A more
usual formulation calls for the existence of a set t whose elements include a
and b and nothing else; but this follows immediately from (10) on applying
separation (4.4) to the condition u ¼ a u ¼ b. The most usual for-
mulation calls for the existence of a unique set, denoted {a, b}, whose
elements include a and b and nothing else; but this follows immediately on
applying extensionality (3.6).
The deductions of union and of power are almost equally easy and
would make good exercises for the reader familiar with the statement of the
axioms. As a hint it may be mentioned that for union, transitivity is
needed; for power, supertransitivity. For replacement, and for infinity
and ‘‘higher infinities’’ or large cardinals, for which last supertransitivity
and hence the single-axiom formulation of separation is again needed, as it
was for power, the reader is referred to Bernays.
7 EXTENSIONALITY REVISITED
To obtain full ZFC we need to obtain three more axioms. The first of these
is a stronger version of extensionality that reads as follows:
(1) 8w(w 2 u « w 2 v) ! u ¼ v.
This will be forthcoming from (3.6) provided we assume the following:
(2) Axiom of Purity
ßu.
From one point of view (2) is an utter absurdity, saying that there are no
objects but sets. Now the mere absurdity of a proposition is no guarantee
that some philosopher will not endorse it. So perhaps there is a philosopher
somewhere who denies that anything but sets exists, and who denies that he
himself exists. Or perhaps there is a philosopher somewhere who denies that
anything but sets exists, and maintains that she herself is a set. But this is
not what is usually intended by those who adopt (2). From the usual point
of view, (2) merely expresses the intention to exclude anything but sets
from the range of the quantifiers. Note, however, that since we are already
assuming that whenever a set is included its elements are included as well,
the assumption (2) actually involves a restriction not merely to sets, but to
sets all of whose elements are sets, and all elements of whose elements are
sets, and so on. Only pure sets are included, hence the name for (2).
Now there are two approaches that might be taken to obtain (2) and
thence (1). One would be to take it as an axiom. Another would be to try to
find an interpretation of set theory with the axiom (2) within set theory
without the axiom (2). That is, one could try to find a formula such that
each axiom of set theory with axiom (2) becomes a theorem of set theory
without axiom (2) when quantifiers are relativized to . Then for every
theorem of set theory with axiom (2), one would have a theorem of
set theory without axiom (2) saying that holds for those sets for which
condition holds.
Ideally, the formula (u) should be one that intuitively expresses that u
is a pure set, or in other words, that u, its elements, the elements of its
elements, and so on, are all sets. Now intuitively the set u, its elements, the
elements of its elements, and so on, are together some objects such that u is
among them and any element of a set among them is among them. Thus
intuitively if u is pure, there will be some objects such that any element of a
set among them is among them, u is among them, and every object among
them is a set. Conversely, if there are some objects such that any element of
a set among them is among them, and if u is among them, and every object
among them is a set, then it follows that u, the elements of u, the elements
of elements of u, and so on, are all sets, and u is pure. Thus a natural
candidate for the formula (u) would be the following:
(3) 99xx(8v8w(v / xx & w 2 v ! w / xx) & u / xx & 8v(v / xx ! ßv)).
Reflection can be used to show (3) is equivalent to the following alter-
native not involving plural quantifiers:
(4) 9t(8v8w(v 2 t & w 2 v ! w 2 t) & u 2 t & 8v 2 t ßv).
The first conjunct of (4) says that t is transitive. Other variations in the
choice of are also possible.
Whatever variant is chosen, what has to be proved is, first, that for each
axiom that is being assumed, one can deduce its relativization from
the axioms that are being assumed, and, second, for the axiom (2) that is not
being assumed, one can also deduce its relativization (2) from the axioms
that are being assumed. It follows that whenever follows from the axioms
that are being assumed plus (2), then follows from the axioms that are
being assumed.
Though the tedious verification of details will not be given here – for
they are quite similar to the details involved in the case of the axiom of
foundation, to be discussed below – this is in fact the case whether by ‘‘the
axioms that are being assumed’’ one means the other axioms of ZFC or
means the axioms of BB. Moreover, it is also true that for any large cardinal
axiom, even if stronger than those provided by BB, that one might decide
to assume in the future, the assumption of such an axiom would render
its relativization deducible as well. Thus one may say that, without
losing any other axiom that one might want, axiom (2) and hence axiom (1)
can be ‘‘obtained,’’ not in the sense of being got by deduction as a logical
consequence, but in the sense of being got by relativization or restriction of
quantifiers.
An arguable advantage of the first approach of assuming purity as an
axiom is that it permits a simpler axiomatization. One can introduce what
amounts to a function symbol ee, which applied to a singular variable u
yields a term eeu of a kind that can be substituted for plural variables, and
whose intended meaning is ‘‘the elements of u.’’ The symbol ß can be
dropped, and the symbol need not be taken as primitive, but rather may
be taken as defined by the following:
(5) u xx « eeu ¼¼ xx.
In general, the use of function symbols builds existence and uniqueness
assumptions into the notation, making it unnecessary to assume them as
axioms. In the present case, with ee there is no need for the axiom of
heredity, and no need for one direction of the axiom of extensionality. All
that needs to be assumed as an axiom of extensionality is the following:
(6) eeu ¼¼ eev ! u ¼ v.
The converse follows from the indiscernibility of identicals.
By contrast, the advantage of the second approach of restricting quanti-
fiers is the philosophical one that it makes explicit what the first approach
leaves implicit, namely, that purity is not a substantive assumption, but a
restriction on the universe of discourse. It is the second approach that will
be adopted officially here. Thus the ZFC version of the extensionality
axiom will not be added to the axioms of BB, and any theorems of ZFC that
depend on it will not be theorems of BB; but for any such theorem , the
relativization will be a theorem of BB, or in other words, it will be a
theorem of BB that holds of pure sets.
The second of the three axioms of ZFC requiring discussion here is the
axiom of foundation, also known as regularity. As axiom (2) implies that all
objects (in the range of the quantifiers) are pure sets, so foundation or
regularity asserts that all objects (in the range of the quantifiers) are well-
founded sets. Since all elements of sets included in the universe of discourse
are themselves included, it follows that if a set u is included, not only is u
well-founded itself, but so are the elements of u, the elements of elements
of u and so on: u is hereditarily well founded.
Several equivalent formal versions of foundation are known, but it will
not be necessary to give any formulation here, since the question of the
status of this axiom is already well explained – with details of the kind
omitted in the discussion of purity above – in introductory textbooks, such
as Hrbacek and Jech (1999). The options are the same as in the case of
purity. Foundation may either be taken as an axiom, while insisting that it
is not a substantive assumption, but merely an expression of the intention
to limit what is included in the range of the quantifiers; or it may be
obtained from the other axioms by relativization. The latter approach has
the advantage of making explicit what the former leaves implicit, and will
be adopted officially here.
8 THE AXIOM OF CHOICE
We now have ‘‘obtained,’’ in one way or another, or one sense or another,

all the axioms of ZF, and it ‘‘only’’ remains to consider AC. Having set
aside von Neumann’s approach based on his maximality interpretation of
the principle of limitation of size, two approaches remain, and one is
exactly similar to the approach to purity and foundation just taken. That
is to say, choice can be obtained from the other axioms, not by deduction as
a logical consequence, but by relativization or restriction of quantifiers.
For Gödel’s famous proof of the relative consistency of choice – his
proof that if ZF is consistent, then ZFC is consistent – proceeds by
defining a technical notion of constructible set, and proving in ZF that all
the ZF axioms, and in addition AC, hold when all quantifications are
replaced by quantifications restricted to constructible sets. If large cardinal
axioms in the range obtained by Bernays are considered, they also continue
to hold when all quantifications are thus restricted, and this is true even for
some larger large cardinals, though for still larger large cardinals one needs
to look for some modification of Gödel’s method, which for all but the
largest large cardinals has indeed been found. It is, moreover, tedious but
routine to adapt this approach to an axiomatization based on plural logic.
At a philosophical level, however, there is a great difference between the
case of choice and that of foundation or purity. For it emphatically cannot
be claimed that when Zermelo originally affirmed the choice axiom, or
when Gödel himself and most later set theorists reaffirmed that axiom, they
merely intended to exclude non-constructible sets from the universe of
discourse. On the contrary, Zermelo was unacquainted with the sophisti-
cated notion of the constructible set, while Gödel and many later set
theorists have explicitly denied that all sets are constructible. Here one
faces a decision of principle. If the goal is to provide an intuitive motivation
for the axioms, the acceptability of relativizing quantifiers will be limited to
cases like purity and foundation, where the relativization arguably repre-
sents the intentions of those who adopt the axiom in question.
By contrast, if the interest lies elsewhere, in finding a reinterpretation of
set theory under which all its axioms will be derivable from a minimal basis,
the method of relativizing quantifiers may be usable without limitation. It
should be noted, however, that if one does allow reinterpretation without
limitation, then it is possible to do very much better in the way of obtaining
set theory from a minimal basis than has been indicated so far. For recent
work of Harvey Friedman shows that without the apparatus of plural logic,
without any axiom of heredity, without any version of an axiom of
extensionality, and without a separate axiom of separation, all of set theory
can be reinterpreted in a system whose sole axiom is a suitably formulated
reflection principle. Moreover, the formulation of the principle can be
modulated so as to obtain, as one wishes, ZFC without the power axiom, or
ZFC exactly, or ZFC with the kind of large cardinals obtained by Bernays,
or ZFC with much larger large cardinals. But since our interest here has
been in the more direct motivation of axioms, Friedman’s impressive
results will be left aside.1 Along with them will be left aside Gödel’s
approach to obtaining AC using his constructible sets.
Having set aside other approaches, one last approach to motivating the
axiom of choice remains to be considered. Or rather, I should say, one last
1
Friedman makes his papers available, between the time of their writing and the time of their
publication, on the preprint server: www.mathpreprints.com/math/Preprint/show. Type ‘‘Harvey
Friedman’’ into the window for access to these preprints.
approach to obtaining the axiom, leaving it to the reader to judge how far
this approach motivates the axiom. But to begin with, the statement of the
axiom should be given. It reads as follows:
(1) 8u(u 2 a ! 9w(w 2 u) & 8u8v(u 2 a & v 2 a ! 8w(w 2 u « w 2 v)
9w(w 2 u & w 2 v)) ! 9v8u(u 2 a ! 9!w(w 2 u & u 2 v)).
If we call the sets in a distinguished, then the first clause of the hypothesis
says that distinguished sets are non-empty, and the second that distin-
guished sets are non-overlapping: a distinguished u and v either have
exactly the same elements (and hence by extensionality are exactly the
same set) or else have no common elements at all. The conclusion asserts
the existence of what is called a choice set, having exactly one element in
common with each distinguished set.
Now even before set-theoretic notions are introduced, there is a version
of the axiom of choice that can be formulated as a purely logical assump-
tion. It is a scheme, reading as follows:
(2) 88xx((xx) ! 9w(w / xx)) & 88xx88yy((xx) & (yy) !
8w(w / xx « w / yy) 9w(w / xx & w / yy)) !
99yy88xx((xx) ! 9!w(w / xx & w / yy)).
If we call objects distinctively related to each other, or distinguished for
short, when condition holds of them, the two conjuncts of the hypothe-
sis of the axiom are non-emptiness and non-overlappingness conditions for
distinguished objects. The conclusion asserts the existence of some objects,
which may be called the chosen objects, such that for any distinguished
objects, there is exactly one chosen object among them.
It is an easy exercise to deduce (1) from (2) in our framework, taking as
(xx) in the condition:
(3) 9u 2 a (u xx).
The hypothesis of (1) easily gives us the hypothesis of (2) for (3) as . What
the conclusion of (2) for (3) as gives us is some chosen objects, the ys,
while what the conclusion of (1) demands is a choice set. What is left to the
reader is to verify that the ys do form a set v as required.
It is not quite so easy an exercise to show, though it is also true, that given
the set theory BB, the set-theoretic version of AC in the form (1) implies
(each instance of) the logical version of AC in form (2). (This derivation of
(2) from (1) is an instance of a more general phenomenon of the derivability
of logical conclusions from set-theoretic assumptions, to be discussed in
the next section.) What one proves is the contrapositive, that if (2) fails (in a
particular instance), then (1) fails. From the assumed failure of (2) for a
particular , one obtains by reflection a supertransitive set t for which
(2)t holds and hence (2)t fails. One can then take as a the set of all subsets
u of t such that
(4) 99xx(u xx & t(xx)).
The non-emptiness and non-overlap clauses in the antecedent of (2)t will
then imply the corresponding clauses in the antecedent of (1) for the set a
defined by (4). But if there were, as per the consequent of (1), a choice set v
for a, then its elements, call them the ys, would be chosen elements, as per
the consequent of (2)t, contrary to the hypothesis that (2)t fails.
It follows that one can obtain the usual set-theoretic version of AC
without adding any new set-theoretic axioms to BB, by adding a version of
AC to the background plural logic. This is the course that will be adopted
here, but the logical version of AC adopted will not be (2) above, but
something else, whose introduction requires some background.
The most important points about AC in the present context would be the
following. On the one hand, AC is useful for proving many mathematical
results in their most general form. This pragmatic consideration has histor-
ically and practically been the most widely persuasive motivation for the
axiom. It is for this reason that, as stated earlier, mathematicians have
generally acquiesced in its assumption: it is the reason why today only
logicians still keep track of which theorems require AC and which do not.
On the other hand, AC is a non-constructive existence assertion, asserting
that something exists for which a given condition holds, without specifying
any particular such thing. That is why when it was first introduced
mathematicians did not at once embrace it. But AC is not the only or
the most basic non-constructive existence assertion in classical mathematics
or logic, since the following basic law of monadic first-order logic is also
such an assertion:
(5) 9u(9v(v) ! (u)).
That is why it is now widely agreed that if one is going to object to non-
constructive existence assertions, one should not wait for AC, but should
begin objecting already at the level of classical logic, if not of classical
sentential logic, thus placing one’s objections entirely outside the scope of
the present paper. And inversely, those who are determined, in the words of
Hilbert, not to be driven out of ‘‘Cantor’s paradise,’’ must begin their
defensive operations no later than the level of monadic first-order logic, or
indeed sentential logic – ‘‘Boole’s paradise,’’ as it might be called.
Hilbert, and with him Bernays, starting from this last observation, pro-
posed a way of building AC even into first-order, singular logic, where it
cannot otherwise be expressed, even as a scheme after the pattern of (2). They
allow the formation of a term eu[(u)] which can be substituted for free
variables. It may be thought of as a description, ‘‘the chosen u such that (u),’’
provided that this is understood in such a way that, when there is in fact no u
such that (u), what the phrase denotes is some arbitrarily chosen object.
Accordingly the following logical axiom for the e-symbol is assumed:
(6) 9u(u) ! (eu[(u)]).
(Actually, a biconditional version of (6) was for Hilbert the very definition
of the existential quantifier, though few have followed him in this, and we
will not.) The connection with (5) is apparent.
The following is also sometimes assumed:
(7) 8u((u) « C(u)) ! eu[(u)] ¼ eu[C(u)].
This says that what the chosen object is depends not on the condition but
on what things it holds of, so that if C holds of exactly the same things, the
same object will be chosen.
If the e-symbol is added to plural logic, then in terms of it we can define
the chosen object among some given objects:
(8) axx ¼ eu[u / xx].
(If we had the ee notation of the preceding section, aeeu would amount to
‘‘the chosen element of u’’ if u is non-empty.) But actually, in this case it is
more natural to take a as the primitive notion, and let e be defined by:
(9) v ¼ eu[(u)] « 99xx(8u(u / xx « (u)) & v ¼ axx).
To derive (6) it is enough to assume the following:
(10) Axiom of Choice
9u(u / xx) ! axx / xx.
It is this version that will be taken officially as an axiom of plural logic here,
alongside comprehension (2.7) and indiscernibility (2.8).
Now (6) follows from (10) and the definition (9). Inversely, (10) follows
from (6) and the definition (8). As for (7), the acceptance of indiscernibility
(2.8) as a scheme means the acceptance of all its instances, whatever
notations are added to the language. We have now added one new nota-
tion, the a-symbol, as we have the following new instance of (2.8) for it:
(11) xx ¼¼ yy ! axx ¼ ayy.
From (11) and the definition (9) (together with comprehension), (7)
follows without the need to assume anything like it as an additional axiom.
The main point is that with an a-symbol and the sole additional axiom
(10) for it, the plural-logical axiom of choice (2) can be deduced. The ys
required by (2) are given as follows:
(12) u / yy « 99xx((xx) & u ¼ axx).
The verification that (2) does follow from (10) given the definition (12) is
left to the reader.
9 SET-THEORETIC MODELS FOR PLURAL LOGIC
So much for the axiomatization of set theory (by heredity, extensionality,

separation, and reflection), starting from a suitable plural logic (with
comprehension, indiscernibility, and choice). We have seen what the
plural-logical perspective of Boolos can contribute to set theory in the
style of Bernays: it enables us to do without classes, and naturally suggests a
choice of primitives that allows the reflection principle to be expressed in a
particularly simple form that makes it arguably an expression of Cantor’s
principle of limitation of size.
Before closing, I should mention what, conversely, Bernays-style set
theory can contribute to Boolos’s plural logic. The point to be made is
one familiar to specialists, though in the idiom of second-order logic and
class-set theories, rather than of plural logic. It has also been treated briefly in
the idiom of plural logic in Boolos (1985), but for the sake of completeness,
and because Boolos’s treatment is rather compressed, and stops short of
explicitly endorsing the reflection principle, let me review the matter here.
But first, since the point to be made pertains to a model theory for plural
logic, a word of caution may be in order. Throughout philosophical logic,
much mischief is caused by a double usage of the word ‘‘semantics.’’ It is
used on the one hand for models, like those provided by Tarski for singular
or first-order logic, or by Kripke for modal logic; and it is used on the other
hand for a theory of meaning. Confusion between these two usages is
manifested in the literature in two different, complementary ways. On the
one hand, if a model theory has not yet been developed for a given logical
notion, it may be alleged that the notion is ‘‘meaningless’’ because it lacks a
‘‘semantics.’’ On the other hand, once a model theory has been developed
for a given logical notion, it may be alleged that problematic ‘‘ontological
commitments’’ are implicit in use of the notion – for instance, that
ontological commitment to ‘‘unactualized possible worlds’’ is implicit in
the use of modal notions – because something like them appears in its
‘‘semantics.’’
Both types of objections could be raised against plural logic. On the one
hand, I have not yet presented a model theory for plural logic, and when
I do it may not immediately be found satisfactory, so it might be claimed
that the meaningfulness of plural quantification is in doubt. On the other
hand, when I do present a model theory and an argument that it is
satisfactory, if that argument is accepted, then since the model theory
will involve an apparatus of sets, it might be claimed that this shows that
an ‘‘ontological commitment’’ to sets is implicit in the use of the plural.
Against the first objection I maintain that even if no one ever did present
a satisfactory model theory for plural logic, the plural was in systematic use
in natural languages long before model theory for anything had been born
or thought of, and such long-standing systematic usages are meaningful if
anything is. Against the second objection I maintain – in addition to
Boolos’s point about the Russell paradox in (3.10) and (3.11) and their
analogues for classes – that the transition from plural language to set-
theoretic language in the work of Cantor and his followers involved an
intellectual struggle more difficult than would have been called for if the
task had been merely one of making explicit something already implicit in
ordinary language.
So much by way of warning against confusing model theory with
‘‘semantics’’ in the sense of a theory of meaning. Now for the model theory
itself. Familiarity will be assumed with the usual notion, due to Tarski, of a
model for first-order logic, though it may be well to begin with a quick
outline even of this. To illustrate the principles involved, it will be enough
to consider formulas with just one non-logical symbol, a two-place pred-
icate R.
Under Tarski’s definition, a model M would consist of a non-empty set
M, the universe of the model, and a set RM of ordered pairs of elements of
M, the relation of the model. In order to define what it is for a formula
without free variables to be true in the model, one must define what it is for
a formula with free variables to be true in the model relative to an assign-
ment of elements of the universe to its free variables. The details of the
inductive definition will not be recalled, except to mention that at the
base step, Ruv is true for the assignment of l and m to u and v if and only
if the ordered pair (l, m) belongs to the set RM; while at the induction step
for the universal quantifier, 8uC(u) is true if and only if C(u) is true
whatever element l of M is assigned to u. In (one version of) the usual
symbolism:
(1) M, l, m |¼ Ruv iff (l, m) 2 RM
(2) M |¼ 8uC(u) iff M, l |¼ C(u) for all l 2 M.
There is a standard extension of this notion of model to monadic second-
order logic, where we have in addition to the variables u, v, . . . a second
style of variable X, Y, . . . and an additional two-place logical predicate
e, and therewith a new kind of atomic formula exemplified by u e X. All
that is changed in the definition is that truth is now relative to an assign-
ment not only of an element l of M to each variable u of the old kind, but
also of a subset of M to each variable X of the new kind. At the base step,
u e X is true for the assignment of l to u and to X if and only if l is an
element of . At the induction step for the universal quantifier, 8XC(X) is
true if and only if C(X) is true whatever subset of M is assigned to X. In
symbols:
(3) M, l, |¼ u 2 X iff l2
(4) M |¼ 8XC(X) iff M, |¼ C(X) for all ˝ M.
Turning now to plural logic, it has what I will call an official model
theory in which truth is defined relative to an assignment of an element l of
M to each singular free variable u, and a subset of M to each plural free
variable xx. At the base step, u / xx is true for the assignment of l and to
u and xx, respectively, if and only if l is an element of . At the induction
step for the universal quantifier, 8xxC(xx) is true if and only if (xx) is true
whatever subset of M is assigned to xx. In symbols:
(5) M, l, |¼ u / xx iff l2
(6) M |¼ 8xxC(xx) iff M, |¼ C(xx) for all ˝ M.
Obviously, the effect is just the same as if each plural variable xx were
replaced by a variable X, and each atomic formula u / xx by an atomic
formula u e X, and then the standard model theory applied to the resulting
second-order sentence. In this official model theory for plural logic validity
is defined in the usual way as truth in all models. The result is that the claim
that some plural formula is valid will be expressible in the language singular
set theory, since plurals are not used in the metalanguage in (5) and (6);
indeed, such a claim will be equivalent to the claim that a corresponding
second-order formula is valid in the standard model theory for second-
order logic, making all the many known results about that standard model
theory applicable to plural logic.
Two questions immediately arise about how satisfactory this official
model theory is, and how validity in the official sense is related to validity
in the intuitive sense of being true in all interpretations. The first question
is whether it would not be more natural to define truth relative to an
assignment of some elements, the s, of M to each free plural variable xx,
with induction clauses as follows:
(7) M, l, |¼ u / xx iff l /
(8) M |¼ 8xxC(xx) iff M, |¼ C(xx) for all 22 M.
Boolos’s answer is that indeed it would more natural to use plurals in the
metalanguage, but that the official singular definition with (5) and (6) is in
fact equivalent, given the comprehension axiom of plural logic and the
separation axiom of set theory, to the more natural plural definition with
(7) and (8). For on the one hand, by separation, for any elements, the s, of
M, there is a subset of M such that a given element l of M is an element
of if and only if it is among the s. And on the other hand, by
comprehension, for any subset of M, there are some elements, the s,
of M such that a given element l of M is among the s if and only if it is an
element of .
The second question is whether it would not be more natural to define
validity in terms, not of models whose universes must be sets, but of
interpretations in a more general sense, in which the objects over which
the variables range might well be all objects, or all pure sets, or all pure and
hereditarily well-founded sets, or the like. (The analogous question arises
even in the case of first-order logic, and was raised long ago by Kreisel
(1967).) To specify an interpretation in this more general sense, one need
only specify, by means of some condition (perhaps involving parameters)
which objects the variables are to be understood as ranging over, and by
another condition (again perhaps involving parameters), which ordered
pairs of such objects are to be understood as R-related. The official
definition would be a special case of this general notion, so that validity
in the natural sense would imply validity in the official sense, but there
would be a question about the converse implication.
An answer extrapolated from Boolos’s remarks would be that indeed
again the proposed definition would be more natural than the official one,
but that again the official definition is equivalent, this time by the reflec-
tion principle, to the more natural definition. To see this, suppose a
sentence is not valid in the general sense, and consider conditions (x)
and C(x, y) specifying an interpretation in which is false. Here (x)
indicates what objects the variables are to be interpreted as ranging over,
and C(x, y) which ordered pairs of such objects are to be interpreted as
being R-related. Now go through and replace each of the items on the left
below by the corresponding item on the right:
8u. . . 8u((u) ! . . .)
9u. . . 9u((u) & . . .)
8xx. . . 8xx(8u(u / xx ! (u)) ! . . .)
9xx. . . 9xx(8u(u / xx ! (u)) & . . .)
Ruv C(u, v)
Calling the resulting formula , the fact that was true under the given
interpretation implies that is true. Now apply reflection to obtain a set
t such that t is true. Let M be the set of all elements of t for which the
condition t holds, and RM the set of all ordered pairs of elements of M for
which the condition Ct holds, thus obtaining a model M. A little thought
shows that the fact that t is true implies that is true in M, showing
that is not valid in the official sense either.
So much by way of argument that the official model theory is satisfac-
tory. I have mentioned that this model theory makes all the many known
results about the standard model theory for second-order logic applicable
to plural logic. One of these results is the incompleteness, or rather, the
incompletablity, of the logic. Validity does not correspond to deducibility
using comprehension, indiscernibility, and choice – the latter being a
sufficient, but not a necessary, condition for the former – nor would the
addition of further axioms enable one to capture validity either, so long as
the set of axioms is recursive. For the sentences deducible from a recursive
set of axioms form a recursively enumerable set, and it is known that the
second-order sentences valid in the standard model theory do not.
That is the bad news. There is also, however, some good news implicit in
the discussion above of why the official model theory is satisfactory. The
good news is that it will never become necessary to add any logical axioms to
the three we already have, even if we become convinced that something
further not deducible from them is intuitively valid, because the effect of
adding a new logical axiom can always be obtained by adding a new set-
theoretic axiom instead, namely, the axiom asserting that the candidate
logical axiom is valid in the official model theory. (In the converse direction,
in many but not all cases adding a new set-theoretic axiom is equivalent to
adding a new logical axiom, namely, in all those cases where the candidate
set-theoretic axiom is equivalent to the assertion of the validity of some
second-order sentence in the standard model theory. This is known to
include very many candidate set-theoretic axioms that have been considered
in the literature, but not all large cardinal axioms.) In this sense, the logical
axioms we have are all the logical axioms we will ever need.
In closing, let me reiterate that what we needed to establish that the
official model theory is satisfactory, and therewith the somewhat reassuring
result just stated, were precisely the set-existence axioms of BB, separation
and reflection. The plural logic of Boolos and the reflection principle of
Bernays, though introduced independently, the one decades after the
other, turn out to be ideally suited for each other. The combination of
the two is a marriage made in heaven – or at least, in Cantor’s paradise.
7
Logicism: a new look
1 NEO-NEO-LOGICISM
After a quick review of the original Fregean logicist program, I would like
to describe two recent revivals of logicist ideas – Richard Heck’s predica-
tivist logicism, and the late Richard Jeffrey’s logicistico-formalism – and
briefly suggest how the two might be combined.
Frege in his Begriffsschrift (1879/1967) presented a deductive system of
second-order logic – with the first-order entities called ‘‘objects’’ and the
second-order ‘‘concepts’’ – including an absolutely unrestricted axiom of
Comprehension, as follows:
(1) 9X 8x(Xx « f(x))
together with its analogues for two-, three-, and many-place relational
concepts or relations. The axiom of Extensionality, in the following form:
(2) X Y ! (f(X) « f(Y ))
is then provable using the following definition of coextensiveness, ‘‘the
analogue of identity’’ for concepts:
(3) X Y « 8z(Xz « Yz).
Since (2) implies the uniqueness of the concept whose existence is asserted
by (1), we may speak of the concept under which fall all and only those
objects x for which f(x) holds, or for short, the concept of being an x such
that f(x), for which I will write Æx: f(x)i.
Frege added, informally in the Grundlagen (1884/1950) and formally in the
Grundgesetze (1893/1903) for the purposes of the derivation of arithmetic
from logic, the infamous Basic Law V, the assumption that to each concept
X is associated an object X, its extension, in such a way that the extensions of
two concepts are identical if and only if the concepts are coextensive:
(4) X ¼ Y « X Y.
135
We may then introduce set-theoretic notation as follows:
(5) Set(y) « 9Y(y ¼ Y)
(6) x 2 y « 9Y(y ¼ Y & Yx)
(7) {x: f(x)} ¼ Æx: f(x)i.
Frege then defines equinumerosity in the familiar Cantorian manner:

(8) X Y « 9R(R is a bijection between X and Y)
where I have not bothered to write out the definition of bijection in purely
logical terms. He then defines number thus:
(9) #X ¼ {y: 9Y(y ¼ Y & X Y )}.
Using this definition he derives what has come to be called (on account of a
passing allusion on Frege’s part to Hume) ‘‘Hume’s Principle’’ or HP:
(10) #X ¼ #Y « X Y.
He then proceeds to define zero, successor, and natural number, and to

derive the Peano postulates.
Russell (1902) famously found a contradiction in Frege’s system. It is
almost too well known to bear repeating, but it needs to be pointed out that
the contradiction about the set of all sets that are not elements of them-
selves uses both comprehension for a formula with a bound second-order
variable and the assumption of the existence of extensions:
= x} ¼ Æx: 9X(x ¼ X & Xx)i.

(11) {x: Set(x) & x 2
To block the paradox, therefore, one might either restrict (1), or restrict
or replace (4).
The first option was not seriously explored until recently. To be sure,
Russell professed a ‘‘vicious circle principle’’ banning definitions of con-
cepts by conditions involving quantification over concepts, or impredica-
tive definitions as they came to be called. But he also introduced an ‘‘axiom
of reducibility’’ whose effect was to circumvent such predicativity restric-
tions, so that as Ramsey (1925) observed he might as well not have imposed
them in the first place.
The second option is what saves Russell from the contradiction. His ‘‘no
classes theory’’ rejected (4) altogether, and with it Frege’s approach to
logicism, on which numbers are objects or first-order entities. (For Russell
they are third-order entities, and he has to make essential use, as Frege did
Logicism: a new look 137
not, of higher-order logic; also, since unlike Frege he cannot prove that there
are infinitely many objects, he has to assume so as the ‘‘axiom of infinity.’’)
Contemporary neo-Fregeanism, whose continuous history begins (though
there were significant precursors) somewhat over two decades ago with
Crispin Wright’s Frege’s Conception of Numbers with Objects (1983), retains
(1) and rejects (4), but in place of the latter assumes (10) as axiomatic, thus
making numbers into objects after all.
The key technical result about this approach was given precise formu-
lation and rigorous proof by the late George Boolos (1987), who showed
that second-order logic with HP, or Frege arithmetic, and second-order
Peano arithmetic, or second-order logic with the Peano postulates, are
interpretable in each other and hence equiconsistent. The direction of the
interpretability of second-order Peano arithmetic in Frege arithmetic he
called Frege’s theorem. There are serious questions how much more of
mathematics beyond second-order arithmetic one can get on a natural
extension of such an approach. (Kit Fine (2002) has a natural-seeming
extension that gives third-order arithmetic.) There is also a question
whether (10) is close enough to being a ‘‘logical’’ principle (as Frege
thought (4) was) to justify neo-Fregeans in calling their position ‘‘neo-
logicism.’’
The latter kind of question seems less of a issue with Richard Heck’s
neo-neo-logicism, which takes the opposite approach of retaining (4) but
restricting (1), assuming only predicative comprehension. Heck (1996)
explored what one would have been left with if one had modified
Frege’s system only by imposition of a ‘‘vicious circle’’ restriction, without
any ‘‘reducibility axiom’’ to cancel it, and without the further ‘‘no classes’’
restriction. He was able to show (building on early work of Terence
Parsons (1987)) that a system of this kind was consistent. He was also
able to show that it is sufficiently strong to interpret Raphael Robinson’s
system Q of minimal arithmetic. Now while that system appears very
weak, an idea of Robert Solovay, further pursued by Edward Nelson and
then a number of others, has shown that ostensibly much stronger
theories can be interpreted in Q. See Hájek and Pudlak (1998) for details.
All these stronger theories can then be interpreted indirectly (via Q) in
Heck’s system.
Further refinements of these results are possible in several directions.
First, interpretability of Q can be proved for weaker predicative systems
than Heck’s. In the very simplest such system, which I call PV (with P for
‘‘predicative’’ and V pronounced ‘‘five’’), one has comprehension in its
simplest predicative form:
(12) 9X8x(Xx « f(x)) provided there are no bound concept variables in f
and in addition Law V in its original form (4). With the definitions of set-
theoretic notions as before we get a little bit of set theory. We of course get
the axiom of extensionality for sets
(13) Set(x) & Set(y) & 8z(z 2 x « z 2 y) ! x ¼ y
from the axiom of extensionality for concepts when sets are introduced as
extensions of concepts. We also easily get empty sets, singletons, pairs and
so on, thus:
(14) ˘ ¼ {x: x 6¼ x}
(15) {u} ¼ {x: x ¼ u}
(16) {u, v } ¼ {x: x ¼ u x ¼ v}.
Here the defining formulas have no concept variables at all. When we recall
that we also may allow free concept variables as parameters, we see that we
also get complements, intersections, and unions:
(17) -X ¼ {x: Xx}
(18) X \ Y ¼ {x: Xx & Yx}
(19) X [ Y ¼ {x: Xx Yx}.
One consequence of (14) and (18) is worth mentioning – the axiom of

adjunction:
(20) 8x(Set(x) ! 8y9z8w(w 2 z « w 2 x w ¼ y)).
This guarantees the existence of x [ {y} for any set x and any object y. That
(13)–(19) and their consequences such as (20) are in effect all we get can be
proved (for the cognoscenti: by using elimination of quantifiers for monadic
first-order logic with identity).
This does not look much like set theory, but in the famous little book
Undecidable Theories by Tarski, Mostowski, and Robinson (1953), which
first introduced Q, a joint result of Tarski and Wanda Smielew is men-
tioned without proof, to the effect that Q is interpretable in the set theory
whose axioms are extensionality, empty set, and adjunction. In order to get
an interpretation of Q we may take 0 ¼ ˘ and for successor use either the
Zermelo definition x 0 ¼ {x} or the von Neumann definition x 0 ¼ x [ {x}.
Interestingly enough, about ten years ago Franco Montagna and Antonella
Mancini (1994) showed that Solovay–Nelson methods for interpreting
other theories in Q can be adapted to prove the Szmielew–Tarski theorem
about the interpretability of Q in another theory.
Second, in the case of stronger systems of arithmetic interpretable in Q
and hence by the Smielew–Tarski theorem indirectly interpretable in PV,
it is sometimes easier to prove interpretability in PV directly. For the most
important specific example, a D0-formula in the language of arithmetic is
one containing only bounded quantifiers 8x < y and 9x < y. A difficult
argument of Alex Wilkie (as reported in Hájek and Pudlak, 1998) shows
that the system called ID0, with the principle of mathematical induction (if
f(x) holds for zero and for the successor of any number for which it holds,
then it holds for all numbers) for D0-formulas, can be interpreted in Q. But
A. P. Hazen showed that we can get interpretability of ID0 in PV without
Wilkie’s difficult argument. See Burgess and Hazen (1998) for details.
Third, one can actually go beyond ID0, which allows only for the
operations of addition and multiplication, to a system which allows also
exponentiation, if one works in a slightly stronger predicative Fregean
theory, which I will call P2 V. In this theory there are objects x, degree-
zero concepts X 0, and degree-one concepts X 1. A formula is of degree zero if
it contains no degree-one variables, and no bound degree-zero variables,
and is of degree one if it contains no bound degree-one variables. We then
have two forms of comprehension, for the two degrees:
(21) 9X 08x(X 0x « f(x)) for f of degree zero
(22) 9X 18x(X 1x « f(x)) for f of degree one.
P2 V, with axioms (20), (21), and (4), can interpret the theory known as
ID0(exp), which in a convenient equivalent of its usual formulation may be
taken to have the following axioms:
(23) 0 6¼ x 0
(24) x 6¼ y ! x0 6¼ y 0
(25) x < 0
(26) x < y 0 « x < y x ¼ y
(27) x þ 0 ¼ x
(28) x þ y 0 ¼ (x þ y)0
(29) x 0 ¼ 0
(30) x y 0 ¼ (x y) þ x
(31) x 0 ¼ 00
(32) x y 0 ¼ (x y) x
(33) (f(0) & 8x(f(x)) ! f(x 0 )) ! 8xf(x)
provided f is a D0-formula.
A formula like the conclusion of (33), consisting of universal quantifiers
preceding a D0-formula is called a 1-formula. Many important theo-
rems of number theory (including the Fermat–Wiles theorem) have this
form. It is known that quite a bit of mathematics can be developed in
such a system as the above (and others known to be interpretable in or
conservative over it). Harvey Friedman has conjectured that every
1-theorem published in the Annals of Mathematics can be proved in
such a system.
The refinements of Heck’s work that I have been discussing so far for
the most part go back to a joint paper by Hazen and myself. That paper
left open how much further one can go. It turns out that, unfortunately,
one is not going to get in this way the whole of first-order arithmetic,
or even all of so-called primitive recursive arithmetic, PRA, which is
generally accepted, following William Tait, as a formalization of finitist
mathematics. This is so even if one goes beyond P2V to P3V and so on,
defined in the obvious way, or even RPV ¼ P1V, the union of all the
Pn V, which amounts to so-called ramified predicative second-order logic
plus Law V. This is on account of considerations related to Gödel’s
second incompleteness theorem, together with the fact that there is a
finitist consistency proof for RPV. (The original consistency proofs of
Parsons and Heck were model-theoretic rather than proof-theoretic in
character, and infinitistic rather than finitistic.) This is the main new
result in this area to be found in my little book Fixing Frege (2005b) where
technical details and bibliographical references for all the material
described so far can be found. (Important improvements have since
been obtained in forthcoming work by Mihai Ganea and by Albert
Visser.) To sum up so far, predicative logicism provides a foundation
for a respectable modicum of arithmetic, but nothing anywhere near the
whole of classical mathematics.
2 LITE LOGICISM
Lite ‘‘beer’’ is a fluid with approximately the taste of a mixture of 50 per cent
real beer and 50 per cent soda water. Lite ‘‘logicism’’ is something brewed
up by my late colleague Richard Jeffrey in his last years (1996, 2002), with
the approximate composition 50 per cent logicism plus 50 per cent formal-
ism. Though this recipe does not, perhaps, make it sound very appetizing,
I myself on tasting it have found it considerably more palatable than I had
expected, and I would like to say enough about it to tempt some of you to
take a sip. The leading idea can be brought out by contrasting the following
Hilbert proportion:
(1) computational : mathematics :: empirical : physics
which is the leading idea behind formalism, with the following Jeffrey
proportion:
(2) logical : mathematics :: empirical : physics.
So let me begin with Hilbert in interpreting whom I follow (Weyl, 1944).
With both Hilbert and Jeffrey we have a view of mathematics self-
consciously modeled on a certain philosophy of physics (more popular
perhaps in Hilbert’s day than in Jeffrey’s). On this view the theoretical
portions of physics are only there to imply empirical laws: universal
generalizations whose instances are empirically decidable. Now it is an
immediate consequence of the definition of empirical decidability that
theoretical physics and empirical laws cannot imply any empirically decid-
able sentences that could not be discovered directly by observation. But
they can yield such sentences more quickly, as predictions of experiences
future rather than records of experiences past. A better description of the
philosophy of physics in question is that it takes physics to be nothing but a
giant engine for generating empirical predictions. While the ‘‘nothing but’’
is controversial, it is comparatively uncontroversial that the data to which
physics is responsible are empirical, and that empirical fruitfulness must be
demanded. It is demanded of higher and higher theories that they should
continue to yield more and more empirical predictions.
On Hilbert’s view the theoretical (or as he called them ‘‘ideal’’) portions
of mathematics are only there to imply universal laws whose instances are
computationally decidable sentences (which he considered the only ‘‘real’’
sentences). In modern terminology these would be 1-sentences. Now
again it is an immediate consequence of the definition of computational
decidability that theoretical mathematics and 1-sentences cannot imply
any computationally decidable sentences that could not be discovered
directly by calculation. But again they can yield such sentences as ‘‘pre-
dictions’’ (for instance, the commutative law of multiplication predicts that
the results of multiplying two hundred-digit numbers in one order and in
the opposite order will be the same far more quickly than we can verify as
much by tedious calculations). Hilbert thought that higher and higher
mathematical theory would be computationally fruitful, yielding new com-
putational predictions, in the sense that it would yield 1-sentences more
quickly, but not in the sense that it would outright yield more 1-sentences.
On the contrary, his program was to try to convince the finitist of the
reliability of classical mathematics, through proving by finitist means that
for any 1-sentences having a classical proof, it is possible in principle,
though perhaps not feasible in practice, to produce a finitist proof.
Gödel’s incompleteness theorems showed Hilbert was wrong: classical
mathematics cannot be finistically proved to be reliable or even consistent;
we must be content with inductive evidence on these points. This is the
negative side of the coin whose positive side is that higher and higher
theories do not just yield quicker proofs of 1-sentences that could be
proved, albeit perhaps very much more slowly, in lower theories, but yield
outright new 1-sentences, so that we have computational fruitfulness in a
stronger sense than Hilbert expected. All this is because a higher theory can
prove the consistency of a lower theory, as the lower theory itself cannot,
and because consistency can be expressed as a 1-sentence.
Though I have so far spoken only of 1-sentences, there are impor-
tant distinctions among them. Hilbert accepted 1-sentences involving
arbitrary primitive recursive functions, including every function in the
sequence addition, iterated addition or multiplication, iterated multipli-
cation or exponentiation, iterated exponentiation, and so on. From certain
philosophical points of view, however, one might wish to stop the series
with exponentiation, or even with multiplication. (Recall, in particular,
that the simplest form of predicative logicism gave only addition and
multiplication, while the ramified form gave exponentiation but not
much more.) In that case, the question of computational fruitfulness, of
whether higher and higher theory does more and more computational
work, would have to be reopened. Do we actually get new 1-sentences
involving only addition and multiplication and exponentiation? Do we get
new 1-sentences involving only addition and multiplication? (We do not
get any involving only addition; for the cognoscenti, this is a consequence of
Pressburger’s theorem.) These questions are non-trivial, but answers are
known. An affirmative answer to the former question is implied by the
work of Julia Robinson (building on the work of Martin Davis and Hilary
Putnam), and an affirmative answer to the latter question is implied by the
work of Yuri Matiyasevich that builds thereupon, for an exposition of
which see Matiyasevich (1993).
No logicist of any stripe can be satisfied with an approach that, like
Hilbert’s as described so far, takes computation to be an end in itself,
without regard to applications. Even Hilbert recognized that some con-
nection must be made between the use in pure mathematics of numerals as
nouns, denoting numerical objects on which we can perform arithmetical
operations, and the doubtless historically far older use in applications of
numerals as adjectives, in numerically definite quantifications. The com-
putational fact that 2 þ 2 ¼ 4 is obviously somehow connected with the
logical fact that if two sheep jumped the fence in the morning and two
sheep jumped the fence in the evening, and no sheep jumped both morn-
ing and evening, then four sheep altogether jumped the fence morning or
evening. But Hilbert gave no formal account of the connection.
Why do I call the fact about the sheep a logical fact? If we think of first-
order logic (with identity) as enriched by (definable) numerical quantifiers
of the type
(3.1) 91xAx « 9xAx
(3.2) 92xAx « 8y9x(x 6¼ y & Ax)
(3.3) 93xAx « 8z8y9x(x 6¼ z & x 6¼ y & Ax)
and so on, along with the quantifiers

(4) 9m!xAx « 9mxAx & 9m þ 1xAx
then the fact about the sheep becomes an instance of the general first-order
logical law
(5) 92!xAx & 92!xBx & 9x(Ax & Bx) ! 94!x(Ax Bx).
Any variety of logicist or set-theoretic approach will supply what Hilbert
does not, a systematic way of connecting arithmetical facts with logical laws,
through the characterization of arithmetical relations and operations that
Fregean and Russellian logicism share with (or rather, borrow from) the
Cantorian theory of cardinal numbers. To recall those characterizations, if
a and b and c are respectively the number of As and Bs and Cs, then we have
the following characterizations of arithmetic notions by existential second-
order formulas, wherein R is two-place and S is three-place:
(6.1) a b «
9R(R gives a bijection between the As and some of the Bs)
(6.2) a þ b ¼ c «
9R(R gives a bijection between the As and some of the Cs and
R gives bijection between the Bs and the rest of the Cs)
(6.3) a b ¼ c «
9S(S gives a bijection between the pairs
consisting of an A and a B and the Cs)
where ‘‘bijection’’ is to be written out in logical terms in the usual way. If

the numbers involved are finite we also have
(7.1) a < b
9R(R gives a bijection between the As and some but not all of the Bs)
(7.2) a b ¼ c «
9S(8x(Cx ! S(–, –, x) gives a function Sx from the Bs to the As)
& 8x8w(Cx & Cw & x 6¼ w ! Sx and Sw are distinct)
& 8x8y8z(Cx & By & Az ! 9w(Cw & Sw(y) ¼ z and
otherwise Sw agrees with Sx)
To illustrate by example how these characterizations provide a bridge
between the computational and the logical, consider the simple law
(8) 8a8b(a < c & c b & b a)
and any particular numerical instance, say that for c ¼ 17:
(9) 8a8b(a < 17 & 17 b & b a).
Using the definitions of the numerical quantifiers and characterizations of

arithmetical relations and operations above, this yields
(10) 8A8B(917xAx & 917xBx &
9R(R gives a bijection between the Bs and some of the As))
which is equivalent to
(11) 8A8B8R(917xAx & 917xBx &
9R(R gives a bijection between the Bs and some of the As))
which amounts to an assertion of the validity of the first-order scheme
obtained by dropping the initial universal quantifiers in (11), or equiva-
lently and more naturally to the validity of the first-order scheme
(12) 917xAx & 917xBx !
(R gives a bijection between the Bs and some of the As).
One instance of (12) would be: if there are fewer than 17 pigeonholes and at
least 17 pigeons, then it cannot be that each pigeon goes into one and only
one hole, with only one pigeon per hole. The general law (8), by taking
more and more specific numerical instances, yields more and more pigeon-
hole principles.
Hilbert’s idea was that just as physics is ‘‘empirical’’ not in the sense that
all its theoretical concepts admit definitions in terms of observation, but in
the sense that its data are empirical, so is mathematics ‘‘computational’’ not
in the sense that all its theoretical concepts admit definitions in terms of
calculation, but in the sense that its data are computational. Jeffrey’s idea is
to substitute ‘‘logical’’ for ‘‘computational’’ here. Note that by the Gödel
completeness theorem, no first-order logical law can be derived from
arithmetic or via arithmetic from higher mathematics, in the manner
sketched above, that cannot in principle be derived by textbook methods,
though perhaps only very much more slowly. It is a question of getting
results more quickly, as logical ‘‘predictions.’’
Now even in the case of pigeonhole principles, which Sam Buss has
shown can be proved relatively quickly by textbook methods, proof by
instantiation of (8) will be quicker for large values of c. But in this particular
example, the 1-sentence (8) involved is one provable already in ID0.
A question of logical fruitfulness now arises, which Jeffrey did not find time
to consider, the question whether higher and higher theory in mathematics
does any logical work, by yielding more and more logical ‘‘predictions,’’ by
yielding more and more 1-sentences that in turn yield logical laws. It turns
out that the answer is affirmative, as a slight extension or corollary of the
work of Robinson and Matiyasevich alluded to earlier. This is the one
formal result in this area to be found in my paper ‘‘Protocol sentences for
lite logicism’’ (Burgess, forthcoming), where technical details and biblio-
graphical references for all the material described so far can be found.
A combination of the two new forms of logicism I have been discussing
can now be contemplated. On such a combined view, the ‘‘real’’ mathe-
matics, including the characterizations needed to derive logical laws from
arithmetic results, and the most basic arithmetical results themselves,
would be provided by predicative logicism, perhaps RPV or alternatively
PV. The attitude towards ‘‘ideal’’ mathematics would be that of lite logi-
cism: it is justified by, on the one hand, inductive evidence that its logical
predictions are reliable, and on the other hand, by a metatheorem to the
effect that higher and higher theory is fruitful in the sense of yielding more
and more logical predictions.
I do not on the present occasion wish to advocate this form of logicism in
any stronger sense than commending it to the reader’s attention as worthy of
consideration. For the combination to be completely satisfactory, we would
need to know that the metatheorem in question can actually be proved in
‘‘real’’ mathematics, RPV or PV. For RPV this is almost certain, though
I cannot claim to have dotted every i and crossed every t. Even leaving this
question unresolved, among various other loose ends, it will I hope be clear
that the potential of logicism is far from having been exhausted even today,
well over a century after it was first introduced by Frege.
PART II
Models, modality, and more

8
Tarski’s tort
1 DEFINABILITY IN DISREPUTE
While what Alfred Tarski labeled his ‘‘semantic conception of truth’’ has
been much discussed, one topic that has not received all the attention it
deserves is his choice of that label. It is this comparatively neglected aspect
of Tarski’s conception that I wish to address here. But first a word about
the situation prior to Tarski.
I begin with a result I learned as a fifteen-year-old student in a summer
mathematics program for high-school students run by the late Arnold
Ross: the theorem that every natural number is interesting. The proof is
by contradiction. Suppose that not every natural number is interesting.
Then the set of uninteresting natural numbers is non-empty. So by the
well-ordering property of the natural numbers, it must have a smallest
element n. But if n is the smallest uninteresting natural number, then n is
interesting for that very reason. Thus we have a contradiction, establishing
that our original hypothesis was false, and that every natural number is
interesting after all. But, of course, some numbers do appear completely
uninteresting to most of us. I suppose a so-called dialethist might claim
that here we have yet another example of a true contradiction, but the more
usual reaction to this bit of adolescent mathematical humor is that ‘‘inter-
esting’’ is too vague or ambiguous, too subjective or relative, a concept to be
admissible in mathematical reasoning. And when Alfred Tarski was begin-
ning his mathematical career, most mathematicians held essentially the
same opinion about the concept of truth. In order to understand what
Tarski was up to with his truth definition, one needs to keep ever in mind
this historical fact.
The first suspect notion that engaged Tarski’s attention was not that of
truth, but rather that of definability. That notion belongs to the same family
as the notion of truth, since an object is definable if there is a condition true
of or satisfied by it and it alone. And eighty years or so ago this notion of
149
definability was in as much disrepute as the notion of truth itself. Thus
Tarski begins his first relevant paper (Tarski 1931), on the notion of defin-
ability for sets of real numbers, by remarking ‘‘Mathematicians, in general,
do not like to operate with the notion of definability; their attitude towards
this notion is one of distrust and reserve.’’ And he goes on to concede
‘‘The reasons for this aversion are quite clear and understandable.’’ Indeed,
the grounds for mathematicians’ suspicions about definability are not far
to seek. The emergence of Russell’s and other set-theoretic paradoxes
had put mathematicians in mind of some of the paradoxes propounded
by ancient philosopher–logicians – Poincaré, as I recall, somewhere
explicitly mentions Zeno the Eleatic and the school of Megara in this
connection – beginning with the paradox of the liar. And soon people
began inventing modern paradoxes of a similar stripe. The closest in spirit
to Russell’s paradox of the set of sets not elements of themselves was
Grelling’s heterological paradox, about adjectives or adjectival phrases not
true of themselves. Paradoxes of this type had an important influence,
in that they helped convince Russell and others that what was responsible
for the set-theoretic paradoxes was not the assumption that actual infin-
ities exist – the heterological paradox has nothing to do with infinity – but
a kind of self-reference or vicious circularity that came to be called
impredicativity.
There were also paradoxes about definability: Berry’s, Richard’s, and
above all König’s. This last was especially important because while no one
had ever taken Berry’s or Richard’s arguments to be anything but ingenious
sophisms, König and presumably at least two others (the referee and the
editor who accepted his note (König 1905/1967) for publication), took his
argument to be a legitimate proof, or more precisely, a legitimate disproof
of the hypothesis that the continuum can be well ordered. The argument, it
will be recalled, is that supposing there exists a well-ordering W of the
continuum, we may consider the set of real numbers that are definable in
terms of W (including those that, like real number zero, for instance, are
definable even without mentioning W ). Since by one theorem of Cantor
there are only countably many finite strings of letters of the alphabet to
serve as definitions, this set must be countable. Since by another theorem of
Cantor there are uncountably many real numbers, the complement of this
set must be non-empty. But then, since W is by hypothesis a well-ordering,
it must have a W-least element: the W-least real number not definable in
terms of W. But this last description provides a definition of the number in
question (in terms of W ), thus yielding a contradiction, and refuting the
supposed theorem of Zermelo.
Tarski’s tort 151
It was, however, not just on account of this paradox of König’s that the
repudiation of definability as an unmathematical notion was especially
firm on the part of the defenders of set theory. For there were other, more
direct arguments than König’s against Zermelo’s axiom of choice, based on
the assumption that the existence of a set or relation or function depends
on its being definable. In replying to these arguments, the defenders of set
theory emphasized the lack of mathematical precision in the notion of
definability. Thus Hadamard, in the exchange with other French analysts
on the axiom of choice (Baire et al. 1905), makes this point and goes so
far as to say the notion of definability belongs to psychology rather than
mathematics.
And yet Tarski saw that the notion of definability had important
mathematical applications, as becomes especially clear in the various
follow-ups to Tarski (1931/1983b) from which emerged the so-called
Tarski–Kuratowski algorithm, which allows one to compute the topolo-
gical complexity of point-sets in the line or plane or space (open, closed, Fr,
Gd, Borel, analytic, co-analytic, projective) by consideration of the logical
complexity (number of alternation of quantifiers when reduced to prenex
form) of the condition that defines the set. The ideas in this short paper,
though of the sort that once absorbed come to seem obvious, were of
the greatest importance, especially after Addison clarified the connec-
tion between the hierarchies of interest to topologists and the hierarchies
introduced, also in the first half of the 1930s, by the recursion theorist
Kleene. These ideas opened the door to the application of ever more
sophisticated logical techniques to the descriptive theory of point-
sets, eventually leading to the explanation by Gödel, Cohen, and their
successors of why so many problems in descriptive set theory had resisted
solution – they are undecidable on the basis of the conventional axioms
of set theory – and in the work of Woodin and his predecessors, showing
how a satisfactory solution was obtainable on the basis of large cardinal
axioms.
It may be noted that Tarski in these early papers was concerned with
something more general than the definability of an element of a domain
(which was what was at issue with the Berry, Richard, and König para-
doxes), namely, the definability of a subset of the domain, which is a matter
of there being a condition satisfied by all and only the elements of that
subset (definability of an element reducing to a special case, the definability
of its unit set). If one-, two-, three-, and many-dimensional sets are in
question, it is necessary to consider satisfaction for conditions with one-,
two-, three-, or many variables. And as Tarski notes, the notion of truth is
simply the degenerate or dimension-zero case of the notion of satisfaction.
Thus there is a direct line between the earlier papers, which defined
satisfaction for a single, specific interpreted language, and the great
Wahrheitsbegriff paper, which defined it for all interpreted languages of a
given kind. What concerned Tarski in that later paper is the same thing
that had concerned him in his earlier papers, namely, the rehabilitation of a
notion in disrepute among contemporary mathematicians.
2 MATHEMATICAL DEFINITION VS LINGUISTIC ANALYSIS
Tarski in his great paper on truth (Tarski 1935/1983c) was not interested in
determining the meaning of the word ‘‘true.’’ He thought he already had a
partial understanding sufficient to determine what the extension of ‘‘true’’
was supposed to be. This understanding is expressed in Convention T, or
rather, in his laying down Convention T as a criterion of ‘‘material
adequacy.’’ And he repeatedly tells us that he has no interest in going
further and determining the intension of ‘‘true.’’ The other requirement he
states for a truth definition, that of ‘‘formal correctness,’’ has nothing to do
with fidelity to the intuitive sense of the term, but merely means mathe-
matical rigor, which is of course essential if the suspicions prevalent among
mathematicians about the notions with which he is concerned are to be
allayed. Material adequacy and formal correctness are his only official
requirements, though naturally he is also interested in the usefulness of
notions he is defining, beyond the applications he has already made of
them in his earlier papers on definability. To repeat, there is no require-
ment of fidelity to the intuitive, pre-theoretic sense of ‘‘true,’’ beyond
conformity to Convention T.
Of course, mathematicians in propounding definitions of mathematical
terms for words already in extra-mathematical use are generally even less
interested than Tarski was in their ordinary meanings. For they care no
more about extensional than about intensional agreement between the
technical sense being introduced and the ordinary sense, and they do not
lay down any criteria of ‘‘material adequacy’’ based on their (total or
partial) understanding of the word in its extra-mathematical sense. Thus
mathematicians feel no compunction whatsoever in speaking about
Hilbert ‘‘space,’’ though it certainly is not the kind of ‘‘space’’ that, say,
astronauts travel around in. But there is also a less superficial sense in which
mathematicians are unconcerned with meanings. For they are not really
concerned even with the meaning of the term in its technical sense, at
least not if ‘‘meaning’’ is understood in the same fine-grained way it is
Tarski’s tort 153
understood by lexicographers propounding definitions or philosophers
propounding analyses. This deeper kind of indifference to meaning, or at
least interest only in a very coarse-grained kind of ‘‘meaning,’’ is best
illustrated by an example.
An analysis text may define the constant e using the well-known repre-
sentation as the limit of a sequence:
(1) e ¼ limn!1 (1 þ n1)n.
Thereafter, all theorems about e in the text will refer back to this definition,
or to previous lemmas based on this definition. This includes, for instance,
the well-known representations as the sum of a series:
(2) e ¼ 1/0! þ 1/1! þ 1/2! þ 1/3! þ 1/4! þ . . .
Surely it is in this practice of fixing a definition and referring all later results
back to it that we should look for the origin of Frege’s notion that in a
properly constructed scientific language, each name should be associated
with a single, fixed definition.
The language of the mathematical community, however, does not con-
form to this requirement. For while one textbook writer may take (1) as the
definition and (2) as a theorem, another writer may do the reverse. Even a
mathematician taught originally from a textbook using one of these
approaches may him- or herself, when he or she comes in due course to
write a textbook, adopt the other. Moreover, even if a sizable majority
favors one approach, which is as may be, they will admit the legitimacy of
the other. There is no question of insisting on one definition as the true,
original one. It would be wrong to say that mathematicians do not care
about distinctions between definitions that are mathematically equivalent,
or even to say that they do not care about distinctions between those that
are provably equivalent, or even to say that they do not care about dis-
tinctions between those that have actually been proved to be equivalent, if
the proof is long and difficult and involves advanced ideas. But they are
indifferent to distinctions between definitions for which there are compa-
ratively well-known, short, simple, elementary proofs of equivalence. (1)
and (2) are certainly not synonymous in any reasonable sense of ‘‘synony-
mous,’’ but they are equally acceptable as ‘‘definitions’’ of e. It is in this
situation that we find the origin of the notion expressed in Quine (1936)
and elsewhere that definitional status is local and transitory – a notion at
the root of his ultimate rejection of the analytic/synthetic distinction.
This same kind of indifference to meaning may be found also in Tarski’s
paper, in addition to the kind of indifference to meaning I mentioned
earlier. For as Wilfrid Hodges reminds us,1 while Tarski’s general defini-
tion of truth is applicable to arbitrary interpreted languages of a certain
kind, for certain special interpreted languages another kind of definition of
truth, very different in appearance from the general definition, is available:
a definition of truth by elimination of quantifiers. The alternative defini-
tion is equally mathematically rigorous or ‘‘formally correct,’’ and equally
‘‘materially adequate,’’ so that it agrees in extension with the general
definition applied to this specific case, though it is by no means synony-
mous with the general definition in any reasonable sense of ‘‘synonymous.’’
Where such an alternative definition is available, Tarski is perfectly
happy with it. Thus Tarski is unconcerned with the meaning of ‘‘true’’ in
a double sense: first, he is unconcerned with pinning down precisely the
meaning of ‘‘true’’ in ordinary language, or in insuring that ‘‘true’’ as a
technical term will agree more than extensionally with ‘‘true’’ as an ordinary
term; second, he is unconcerned with differences between alternative
technical definitions, if these have been proved extensionally equivalent.
And if Tarski is doubly unconcerned with pinning down the meaning of
‘‘true,’’ he is even less concerned with analyzing the meaning of any other
word or symbol. A clause such as either of the following:
(3) (A and B) is true iff A is true and B is true
(4) True(A B) «True(A) True(B)
emphatically cannot be construed as telling us the meaning of the
word ‘‘and’’ or the symbol ‘‘.’’ For in Tarski’s original set-up, the ‘‘object
language’’ for which truth is being defined is contained in the metalanguage
in which the proof is being given. That is why we see ay-en-dees on both
sides of (3) and carets on both sides of (4). In order to understand the
definition, one must understand the metalanguage, and that includes
understanding the object language which is part of it, and therewith each
of the words or symbols of the object language.
3 TARSKI’S TRIUMPH AND TRESPASS
It was not linguistic understanding but mathematical fruitfulness that

Tarski sought with his definition, and in this he was very successful. For
Tarski is the creator of the incredibly rich branch of mathematical logic
1
In his ‘‘Tarski’s truth definitions,’’ in Stanford Encyclopedia of Philosophy, http://plato.stanford.edu/
entries/tarski-truth/ (available only on-line).
Tarski’s tort 155
known as the theory of models. Hodges, however, has noted how very slow
Tarski was to advance beyond the 1933 definition of truth for sentences of
an interpreted language to the definition of the notion that is central to the
theory of models, namely, the notion of truth in a structure for sentences of
an uninterpreted language.
Formally the step is a very short one. An interpreted language is naturally
thought of as an ordered pair consisting of an uninterpreted language and
an interpretation. And an interpretation is simply a set, the domain, and an
assignment of a relation or operation of the right number of places on it to
each non-logical primitive. But that is essentially what a mathematical
structure is: a set, the domain, and certain distinguished relations and/or
operations on it, distinguished from each other by certain symbols asso-
ciated with them. For instance, a ring is, first of all, a set with two binary
operations, one written additively and one written multiplicatively. And
formally, the step from a two-place relation between a sentence and an
ordered pair consisting of an uninterpreted language plus an interpretation
or structure to a three-place relation among a sentence, an uninterpreted
language, and a structure or interpretation is a very short one. But though
Tarski was in effect operating with the latter notion within a few years, he
did not give an explicit definition until over two decades later. Indeed, it is
oversimplifying to say that he gave the definition, since it appears in a joint
paper with a student (Tarski and Vaught 1956).
This notion of a sentence of a given uninterpreted language being true
in a structure, or conversely, of a structure being a model of a sentence of
an uninterpreted language, has proved immensely useful both in appli-
cations to core mathematics (mainly abstract algebra) and in applications
to the metamathematics of set theory. The notion is also needed to
give a fully rigorous statement of the Löwenheim–Skolem and Gödel
completeness theorems, which were proved before they were stated, so
to speak.
Closely related to this last application was another application envi-
sioned by Tarski, that of giving a definition of logical truth. Here, however,
his work (Tarski 1936/1983d) is of more ambiguous status. Tarski himself
pointed to one limitation of his definition of logical truth or truth by virtue
of logical form alone as truth in all models: the definition presupposes a
division of symbols and notions into logical and non-logical, of which
Tarski was not to give an account until a late lecture, published only
posthumously (though since much discussed). Kreisel (1967) pointed
to another limitation: logical truth in an intuitive sense is truth in all
interpretations in an intuitive sense, and is not restricted to truth in
interpretations in the technical sense considered by Tarski, where the
quantifiers must range over a set and not a proper class. For first-order
logic, as Kreisel notes, the completeness theorem shows that it is enough to
consider interpretations where the domain is a set. But for second-order
logic (if one grants that it is logic) the assumption that holding in all
interpretations and holding in all set-interpretations – the intuitive notion
and the Tarski notion – coincide is not provable on the basis of the usual
ZFC axioms for set theory, but is in effect a large cardinal assumption. But
despite the limitations of Tarski’s approach on this one issue, there is no
denying that Tarski’s definition of truth was immensely successful.
Yet Tarski’s achievement was marred by a misdeed on his part that
opened the door to considerable mischief. His violation did not rise to the
level of felony or even misdemeanor – indeed, as my title suggests, it was a
civil rather than a criminal wrong. It was a case of trademark infringement:
his appropriating to his own use the linguists’ term ‘‘semantics.’’ Of course,
it was not really an actionable offense at all, since academic disciplines,
unlike business corporations, are not legal ‘‘persons’’ with standing to sue
anyone, though in a way that is a shame. (It would be a very good thing if,
for instance, the International Seismological Union could collect hefty
punitive damages from any writer who uses ‘‘epicenter’’ as a fancy synonym
for ‘‘center of activity.’’) And if Tarski’s act were actionable, there would be,
as in virtually all cases of this particular offense, something to be said for the
defense. There was a minority usage of ‘‘semantics’’ for something other
than a theory of the meanings of words. Indeed, there were several minority
usages, and in ‘‘The semantic conception of truth’’ (Tarski 1944), he is
quick to disassociate himself from one of them, the ‘‘General Semantics’’ of
another Polish thinker, Count Alfred Korzybski.
But there can hardly be any question that what ‘‘semantics’’ conveyed
and conveys to the mind of the general reader is a theory of meaning,
which Tarski’s theory most emphatically was not. By calling his theory
‘‘semantics,’’ Tarski opened the door to endless misunderstandings on this
point. There has been significant damage to logic arising from such
misunderstandings, from confusion of model theory or ‘‘semantics’’ impro-
perly so-called with meaning theory or ‘‘semantics’’ properly so-called.
Needless to say, if one is careful, one can avoid the confusion even
while keeping the double use of ‘‘semantics’’ by distinguishing formal from
linguistic. But in general usage ‘‘formal semantics’’ is a case of oxymoron
and ‘‘linguistic semantics’’ a case of pleonasm, and it would have been better
not to create a situation where there is a call for distinguishing adjectives
in the first place.
Tarski’s tort 157
4 MODAL MUDDLES: SPURIOUS COMMITMENTS
Tarski’s usage did create such a situation, or at least the decision by most of
his successors to follow him in his usage of ‘‘semantics’’ has done so. For his
usage, though not in fact followed by workers in first-order model theory,
has generally been followed by philosophers, and even more so by com-
puter scientists working in the field of modal, temporal, and related logics.
Hence, for instance, the title ‘‘Semantical considerations on modal logic’’
for the paper Kripke (1963) in which the inventor of Kripke models finally
belatedly published his model theory for modal logic. Not everyone who
follows Tarski’s usage is confused, of course, but everyone who does so
encourages confusion, and there has been confusion enough, and this of
two kinds: spurious attributions of ontological commitment to commonplace
locutions and unwarranted complacency about the intelligibility of dubious
notions. Let me take up the first of these phenomena first, illustrating by the
case of tense logic.
In tense logic we have future-tense and past-tense operators F and P
whose intended meaning is something like ‘‘it (sometime) will be the
case that . . .’’ and ‘‘it (once) was the case that . . .’’ These behave syntacti-
cally like negation: they are one-place connectives. But it is not hard to see
that we cannot treat them model-theoretically like negation. These parallel
clauses:
(5) M |¼ A iff it is not the case that M |¼ A
(6) M |¼ FA iff it will be the case that M |¼ A
will not do because we are looking at mathematical models here,

and mathematical facts about mathematical objects do not change.
If a model does not satisfy a sentence now, there is no use waiting, for it
never will.
The mathematical logician will naturally want to deal with this problem
by adopting the kind of treatment of time that is used in mathematical
physics, where time is in effect represented as an extra spatial dimension.
This amounts to purging the language of tense, replacing sentences of the
type, ‘‘It is raining’’ in the present tense by formulas of the type ‘‘It be
raining at time t,’’ where the ‘‘be’’ is tenseless, or rather by the instantiation
of such a formula where a constant c for the present is put in for the variable
t. Then instead of ‘‘It will rain,’’ or in more stilted and stylized form, ‘‘It will
be the case that it is raining,’’ we get ‘‘There exists a time t such that t is
future relative to c and it be raining at time t.’’ This is, in Quinean terms,
regimentation in first-order logic. Or if the logician does not insist on
translating tense-logical sentences into first-order sentences quantifying
over times, and introducing a relative futurity relation, the logician will
anyhow, in presenting a model theory for tense logic, use exactly the notion
of model one would get if one did translate into first-order terms and use
Tarski-style model theory for the latter.
In other words, in place of a classical model, which at the sentential level
is just an assignment of truth values to sentence letters, one has something
more complicated: a domain X of times and a relative-futurity relation R on
it, and an assignment to each a in X of a model of the classical kind, which is
to say, an assignment of truth values to sentence letters. One then has to
define not truth of a sentence in the model M, but satisfaction of a formula
in the model M by an element a of X. The relevant clauses then become as
follows:
(5 ) M |¼ A[a] iff it is not the case that M |¼ A[a]

(6 ) M |¼ FA[a] iff it there is a b such that aRb and M |¼ A[b].
Now philosophically, thinking tenselessly raises a whole raft of new

questions, and above all this one: according to our ordinary tensed ways
of thinking, I always have been (as long as I have been around) and am
always going to be (for as long as I am still around) three-dimensional. On
this new tensed way of speaking, it seems that I be at different times, or if
one prefers, in different temporary states of the world, or if one really wants
to encourage confusion, in different temporary worlds. But how can I be in
more than one (temporal) ‘‘place’’? Should I conceive of myself as being
four-dimensional, with various three-dimensional stages, ‘‘temporal seg-
ments’’ of me, at various times? Should I conceive of myself as being only at
the present time, and merely having past and future counterparts? Quite a
few puzzles arise. Nonetheless, it may be insisted that the tenseless way of
thinking is more scientific, and perhaps even demanded by certain advan-
ces in physics during the last century.
That may be so, but even if it is, the old-fashioned tensed way of
thinking is going to be around for a long time before the new-fangled
tenseless way is universally adopted, and so we would do well to understand
the relationship between the two. This is what tense logic, as developed by
Prior (1967b), in effect does. Physical assumptions about the structure of
time correspond naturally to assumptions about the structure or ‘‘frame’’ of
times with the earlier–later relation (which is what relative-futurity
amounts to), or in other words, different classes of models. Completeness
theorems relate various axiomatic systems of tense logic to various classes of
Tarski’s tort 159
models, so that, for instance, the assumption that the earlier–later relation
on times is dense corresponds to the assumption that Fp ! FFp is valid.
So far, so good. However, if one thinks of the model theory as a meaning
theory for tense logic, one will be led to the idea that the tenseless way of
thinking is not some new-fangled techno-scientific development of the last
century, but rather has been what, despite superficial appearance to the
contrary, our ordinary tensed ways of speaking ‘‘deep down’’ have meant all
along. The notion of a durationless instant of time is not, on this view, a
sophisticated, advanced, historically late posit, but rather is something that
has been implicitly assumed since time immemorial. In a word – or rather,
in two words – even our ordinary, old-fashioned, commonsense tensed
ways of speaking and thinking are ontologically committed to instants of
time – or worse, to ‘‘temporary worlds’’: ‘‘They aren’t there on the surface,
but they’re there in the semantics.’’ But in fact, their presence in the model
theory – or if you must, in the formal ‘‘semantics’’ – is merely an artifact, an
inevitable consequence of the fact that it is mathematical, and therefore
unchanging models that we are considering. The ‘‘ontological commit-
ment’’ here is entirely spurious, based on a simple fallacy of equivocation
between misnamed formal ‘‘semantics’’ and properly linguistic semantics:
It’s there in the models.
So it’s there in the semantics.
So it’s there in the meaning.
Mathematical facts are necessary as well as permanent, and exactly
parallel considerations apply to the modal case. Mathematical modelers
will reach for a treatment of modality resembling that used in mathemat-
ical probability and statistics, with a space of ‘‘possibilities,’’ or if you prefer
‘‘possible states of the world,’’ or if you really want to encourage confusion
‘‘possible worlds.’’ But this gives absolutely no reason whatsoever to sup-
pose that when I say ‘‘I could have got stranded in an airport owing to a
snow storm’’ I mean there is a possibility or possible state of the world or
possible world where I, or a state or modal segment of me, or a counterpart
of me, is stranded. The fact that what some people insist on calling
‘‘possible worlds’’ are there ‘‘in the semantics’’ tells us nothing about
‘‘ontological commitment’’ to them, if the ‘‘semantics’’ is only formal and
not linguistic. Similar remarks apply to the supposed ontological commit-
ment of plural logic to sets or classes, and there are other examples as well.
The clash between the usage of ‘‘semantics’’ by Tarski and his followers and
the majority use in the educated public is forever misleadingly suggesting
existential implications that are not there.
5 MODAL MUDDLES: UNWARRANTED COMPLACENCY
It is a significant historical fact that the model theory for modal logics was
worked out in the late fifties and early sixties, while the distinctions among
logical demonstrability, logical validity, analyticity, aprioricity, and neces-
sity in the sense of ‘‘would have been no matter what’’ or ‘‘couldn’t have
been otherwise’’ were largely overlooked until the later sixties and early
seventies – not that everyone pays sufficient attention to such distinctions
even today. That is to say, we were in possession of a model theory for
modal logic well before most modal logicians had an unambiguous and
unconfused understanding of what the modalities mean, or even any firm
guarantee that the modalities do mean something. This fact alone should
warn us that it is one thing to have a theory of models, and another to have
a theory of meaning. The following argument is utterly invalid:
These sentences have models.
These sentences have a semantics.
These sentences have a meaning.
Philosophically speaking, the model theory for modal logic has much
less direct value than does the parallel model theory for temporal logic. This
is because different theories about time do naturally present themselves as
theories about the relative-futurity relation on stages of the world, whereas
different conceptions of modality do not naturally present themselves as
theories about the relative-possibility relation on states of the world. It is
very useful to know that a certain class of modal models corresponds to a
certain axiomatic system, if there is reason to be interested in that axiom
system. But philosophically speaking, model theory is of little direct use in
establishing that a given axiom system is appropriate for a given conception
of modality. Indeed, though we have had the model theory for getting on
towards a half-century, the correct axiom system has been convincingly
determined only for a couple of conceptions of necessity: truth by virtue of
logical form, and provability in a given formal system.
I have addressed these topics elsewhere, but let me add just a word here
about provability logic. Here we have a clear understanding or intended
interpretation of what the box and diamond are to mean, and hence a clear
understanding or intended interpretation of what it is for a sentence
involving boxes and diamonds to be a law of logic, true in all instances –
essentially, for all substitutions of sentences of first-order arithmetic for the
sentence letters p, q, r, and so on. We have a proof that a certain axiomatic
system, GL, is sound for this intended interpretation. We have also, thanks
Tarski’s tort 161
to Segerberg, a proof that GL is both sound and complete for a certain class
of modal models. But for a long time this is all that we had, and we did not
have what we want from a theory of models, namely, an assurance that
truth in all models corresponds to truth in all instances, according to our
intended interpretation. This is something like the Kreisel gap noted earlier
in connection with first-order logic, but very much more serious.
But the serious lack was eventually supplied by the genius of Solovay. As
it happened, he made use of Segerberg’s earlier result. That is to say, rather
than ignore the modal models altogether and directly establish that if a
sentence is consistent with GL then it is true in some arithmetical instance,
thus establishing the completeness of GL for the intended interpretation,
he showed how from a model of the appropriate class of a sentence to
produce sentences of first-order arithmetic that, substituted for the sen-
tence letters, would result in an instance that is true according to the
intended interpretation, thus showing that truth in all models does corre-
spond to truth in all instances, and thus indirectly establishing the com-
pleteness of GL for the intended interpretation, given Segerberg’s result
about its completeness for models of the appropriate class. The model
theory is useful, but it is useful only as an auxiliary. Segerberg’s contribu-
tion did not go to waste, but Solovay’s contribution was crucial. For a full
exposition of these matters, and references to the further literature, see
Boolos (1993).
There are other examples of the same phenomenon, where the
model theory plays only an auxiliary role. One pertains to intuitionistic
logic, with Tarski and Kripke in the role of Segerberg and Kreisel in the
role of Solovay. For an exposition see Burgess (1981a). What reflection on
these examples should make clear is that merely possessing a model-
theoretic characterization of a given axiomatic system of modal logic
does not suffice to tell us that the system captures (that is, is sound and
complete for) any interesting interpretation or concept of necessity.
Provision of a purely formal ‘‘semantics’’ for an axiomatic system that
had been without one – for instance, the provision by Fine and by Meyer
of a purely formal ‘‘semantics’’ for the system R of relevance/relevant logic –
does little, and by itself does nothing towards establishing the coher-
ence and intelligibility of any underlying motivating ideas or intended
intuitive interpretation. This last is one case where there certainly was
premature celebration, provoking a paper, ‘‘When is a semantics not a
semantics?’’ by B. J. Copeland (1979), that is still well worth reading to
dispel confusions. So Tarski’s using the word ‘‘semantics’’ in connection
with his truth definition has opened the door to confusion in the area of
modal and more generally intensional or philosophical logic, which people
like Copeland and myself have then had to come along after and try to
straighten out.
6 FROM TARSKI TO DAVIDSONIANISM
Now one who opens a door that should have been left closed cannot be
held responsible for every mischievous thing that then walks through it,
and Tarski was himself the first victim of mischief resulting from his
original offense. I think he was naughty, and should have been made to
stand in a corner by linguists; but he actually suffered something consid-
erably worse: he was stood on his head by Donald Davidson and his
disciples. For Tarski’s usage has not only tended to encourage the modal
muddles I have been bemoaning, but also seems to me to have been in part
responsible for the insufficiently critical reception of the truth-conditional
theory of meaning; and this theory is diametrically opposed to Tarski’s
own views on the meaning of truth and to his views on the paradoxes with
which we began. Now while I hope it will be universally agreed that the
modal muddles Tarski’s usage may have encouraged are bad things, this
certainly will not be agreed about the acceptance of a truth-conditional
theory of meaning or the rejection of Tarski’s view of the paradoxes.
The most I can hope will be agreed is that if Tarski’s usage contributed
to the formation of a predisposition to accept the one and reject the other
that was independent of the real merits of the case, then that development
was a bad thing.
In any case, besides Tarski’s usage of ‘‘semantics’’ there is another feature
of Tarski’s definition of truth, or rather, of the Tarski–Vaught definition of
the model, that may have been operative. To begin with, it may be noted
that once we make the transition to the Tarski–Vaught notion of truth in a
model, we soon find ourselves departing from Tarski’s original notion that
truth for a given language is to be defined in a metalanguage containing
that language.
For the languages for which truth in a model are being defined are
formal languages, while the papers on model theory containing the defi-
nitions are, like other mathematical papers, invariably written in (a mathe-
maticians’ dialect of) a natural language, nowadays usually English. Thus
instead of
(7) M |¼ (A B) « (M |¼ A M |¼ B)
(8) M |¼ (A and B) iff (M |¼ A and M |¼ B)
Tarski’s tort 163
we find
(9) M |¼ (A B) iff (M |¼ A and M |¼ B).
And whereas (7) and (8) do not on the face of it look as if they could be
telling one anything about the meaning of ‘‘’’ or ‘‘and’’ that one did not
already know – after all if one doesn’t know the meaning of ‘‘’’ already one
is not going to understand the right-hand side of (7), and if one does not
know the meaning of ‘‘and’’ already one is not going to understand the
right-hand side of (8) – by contrast (9) does rather look as if it were telling
us something about the meaning of ‘‘,’’ given that we already know the
meaning of ‘‘and.’’
It is tempting to think of the status of (9) as being something like the
status of the following:
(10) ( jai D) is true iff is true and D is true.
And (10) does tell us something – I do not say everything, but I do say
something – about the meaning of the Greek word ‘‘jai.’’ But notice that
(10) tells us something about the meaning of ‘‘jai’’ only because we already
know the meaning of ‘‘and’’ and because we already know the meaning of
‘‘true,’’ at least to the extent of knowing Convention T. So there is a
significant difference between (10) and (9), since the latter involves the
symbol ‘‘|¼’’ rather than the English word ‘‘true.’’
If we consider the clauses in a Tarski–Vaught style definition, thus:
(11a) M |¼ A iff not M |¼ A
(11b) M |¼ (A B) iff (M |¼ A and M |¼ B)
(11c) M |¼ (A B) iff (M |¼ A or M |¼ B)
we cannot take these to be telling us both what the double turnstile means
and what the caret, wedge, tilde, and so on mean. For there are too many
unknowns and not enough equations. The intention is that the double
turnstile is to be read as ‘‘true,’’ the caret as ‘‘and,’’ the wedge as ‘‘or’’; but the
biconditionals would be equally appropriate if instead the double turn-
stile were read as ‘‘false’’ and the caret as ‘‘or’’ and the wedge as ‘‘and.’’ If a
student with no previous knowledge of these matters is told that the
caret, wedge, and so forth are to be read as ‘‘and,’’ ‘‘or,’’ and so forth,
then the student may be able to figure out that the double turnstile might
be read as ‘‘true’’; inversely, if the student is told that the double turnstile
is to be read as ‘‘true,’’ then the student may be able to figure out that the
caret might be read as ‘‘and,’’ the wedge as ‘‘or,’’ and so forth. But the
student has to be given some clue or other.
Now, once the model theorist is past the student stage, and knows that
the double turnstile is customarily to be read as ‘‘true,’’ the model theorist
does learn something about how an addition to the usual list of logical
symbols might be read, if the colleague proposing the addition indicates
how the usual definition of double turnstile is to be extended to the
extended language. For instance, a clause
(12) M |¼ (A # B) iff not (M |¼ A and M |¼ B)
would suggest the reading ‘‘not both’’ for the down arrow. And to give a
perhaps more realistic example, in papers on generalized quantifiers the
same symbol Q gets used over and over, with different clauses in the
definition of double turnstile in different papers, suggesting different read-
ings in different papers, among them ‘‘there exist infinitely many,’’ ‘‘there
exist uncountably many,’’ ‘‘most,’’ and others. The analogue of (12) in a
given paper in such cases tells one something about what the symbol Q is
being used to mean in that given paper. But our ability to guess the readings
is entirely dependent on our familiarity with the meanings of ‘‘infinitely
many’’ and ‘‘uncountably many’’ and ‘‘most,’’ and on our familiarity with
what ‘‘true’’ means, at least to the extent of being familiar with Convention
T, and with the custom of reading the double turnstile as ‘‘true.’’
And in any case, we are not really being told that ‘‘QxA(x)’’ in a given paper
means ‘‘there exist infinitely many x such that A(x),’’ but only that it means
something that is true if and only if there exist infinitely many x such that
A(x), a condition fulfilled not only by ‘‘there exist infinitely many x such that
A(x)’’ but also by ‘‘it is not the case that for all but finitely many x it is not the
case that A(x),’’ and by infinitely many other alternatives. After all, merely to
be told that some sentence of a foreign language we do not understand is true
if and only if, say, snow is white, does not tell us what the sentence means.
Taken together with our knowledge that snow is white, it does tell us that the
sentence means something true. But that is all. Merely being given a list of
items of the following kind, one for each sentence of the foreign language:
(13) ‘‘[FOREIGN SENTENCE]’’ is true iff [ENGLISH SENTENCE]
will not tell us what the sentences mean. If I tell you
(14a) ‘‘To viomi eimai arpqo’’ is true iff snow is white
(14b) ‘‘To jaqbotmo eimai latqo’’ is true iff coal is black
I allow you, given the common knowledge that snow is white and coal is
black, to infer that the two Greek sentences quoted are true; but I do not
divulge the meaning of any Greek sentence.
Tarski’s tort 165
Donald Davidson conjectured, however, in Davidson (1967) and
sequels, that if we require some finite apparatus to generate recursively,
with clauses like (10), a whole list of items of type (12) for every Greek
sentence, and if we impose some suitable further restrictions, then in fact
what appear on the right-hand side of each item on the list will have to be
English sentences with the same meaning as the Greek sentence on the left-
hand side. Or rather – since he was writing during the era of Quinean
suspicion about meaning – the Greek and English sentences will have to be
close enough in meaning, according to our intuitive, pre-theoretic under-
standing of ‘‘meaning,’’ that appearing on opposite sides of a list generated
in the manner indicated can serve as a workable substitute for the intuitive,
pre-theoretic, but to some suspect, understanding of sameness of ‘‘mean-
ing.’’ In short, while in order to learn the meaning of Greek sentences and
the words of which they are composed it is not enough to be told things like
(14a,b), Davidson conjectured that, in order to do so it may be enough to
be told, or to come to know, such things in the right way.
Davidson’s conjecture that theories of truth can in this sense serve as
theories of meaning eventually gave rise to what I will call Davidsonianism,
without intending to imply that Davidson himself fully subscribed to it,
namely, the truth-conditional theory of meaning. Formally, indeed, the
step from Davidson’s theory about how to tell the meanings of foreign
sentences to speakers who already know the meaning of English sentences
to the Davidsonian theory that knowledge of the truth conditions of
English sentences is what knowledge of the meaning of English sentences
consists in, can be a very short one. For the simplest version of
Davidsonianism would simply be the analogue of what Davidson says
about Greek and English applied to English and a hypothetical innate
language of thought. Using the language of Descartes to represent the
language of thought, coming to know the meaning of the English sentences
‘‘Snow is white,’’ ‘‘Grass is green,’’ and so on, amounts to coming in the
right way to know the following:
(15a) ‘‘Snow is white’’ est vraie ssi la neige est blanche
(15b) ‘‘Grass is green’’ est vraie ssi l’herbe est verte.
Davidsonianism as such, it should be emphasized, is not committed to the
language of thought hypothesis; I only say that Davidsonianism is imme-
diate from the conjecture of Davidson if one accepts that hypothesis.
For some of us Davidsonianism seems, not least on account of its
apparent assumption that truth is an innate idea, possession of which is a
prerequisite for all language-learning, to be preposterous. For others, the
Davidsonian assumption is so much taken for granted that it is hardly
recognized as a substantive assumption at all. My concern here will be
not with the enormous question of the merits or demerits of the truth-
conditional theory of meaning, but only with the extent to which Tarski
deserves a share of the blame or credit for its becoming so widely held a
belief among philosophers as it currently is.
The first thing that must be said on this head is, of course, that Tarski’s
responsibility is limited. Davidson himself is not fully responsible for his
disciples’ extrapolations from his conjectures, and Tarski is certainly not
fully responsible for widespread sympathetic reception of those conjec-
tures. It must be acknowledged, for one thing, that their sympathetic
reception was surely in part due to the prestige of the name of Davidson
as a result of quite other achievements. But then again, was it not in part
due to the prestige of the name of Tarski, which Davidson so frequently
invoked?
Well, even if so, it must be acknowledged, for another thing, that the
invocation of Tarski’s name was not entirely appropriate, since as
Davidson, if not every one of his disciples, was aware, those conjectures
amount to an inversion of Tarski. For they make what for Tarski were
clauses in a definition of truth in terms of already understood notions like
negation and conjunction and disjunction, into definitions of a kind of
those operators, in terms of a notion of truth taken as primitive. We
constantly find in the writings of Davidson and disciples mentions of a
‘‘Tarskian’’ theory of truth, where ‘‘counter-Tarskian’’ or ‘‘anti-Tarskian’’
would have been more accurate, if less likely to confer borrowed prestige on
bold (which is to say doubtful) new conjectures. And Tarski, of course, is
not responsible for this usage. But would the idea of invoking Tarski’s
name at all in connection with a theory of meaning have occurred to
anyone, if Tarski had not himself attached to his theory a label ordinarily
used for the theory of meaning, the label ‘‘semantics’’?
7 THE ‘‘ S E M A N T I C ’’ PARADOXES
Well, if so then one important consequence of Tarski’s usage of ‘‘seman-

tics’’ was to help popularize a theory with implications diametrically
opposed to his own view on the ‘‘semantic’’ paradoxes, his so-called
inconsistency theory of truth. This is the last point I wish to bring out,
and it requires a little background. The natural anti-Davidsonian assump-
tion is that understanding the meaning of a word consists, not in knowl-
edge of how it contributes to the truth conditions of sentences in which it
Tarski’s tort 167
appears, but rather in internalization of rules for its use. For a ‘‘use’’
theorist, it would be nothing short of a miracle if we never internalized
any rules having lurking inconsistencies, and the intractability of the liar
and related paradoxes strongly suggests that in the case of ‘‘truth’’ the rules
may be the obvious, inconsistent ones, permitting free passage back and
forth between p and ‘‘it is true that p,’’ in accordance with an absolutely
unrestricted version of Convention T.
Now while Tarski neither explicitly endorses the ‘‘use’’ theory nor
explicitly says that Convention T is all there is to the intuitive meaning
of ‘‘true’’ – though it represents all he is able to understand about the pre-
theoretic meaning of ‘‘true’’ – he does endorse one consequence of those
views, namely, the consequence that the intuitive notion of truth is the
inconsistent one. For Tarski, what is to be expected of a theory of truth is
therefore not a vindication of the intuitive notion, but a restricted replace-
ment for it: a serviceable substitute applicable not to all languages, but only
to languages of a certain comparatively simple structure. This may be the
view of most of us who have worked on the technical comparative study of
the various theories of truth that emerged in the wake of Kripke’s famous
‘‘outline’’ – or if not the majority view, anyhow the plurality view, having
more adherents than any particular one of the more positive theories. But
among philosophers generally it is distinctly a minority view, subscribed to
by my teacher Charles Chihara, my student John Barker, and (not as the
result of any influence of mine) my son Alexi Burgess, and by very few
others. Indeed it seems that many if not most philosophers are so violently
prejudiced against this view that they do not even wish to contemplate
what would be an appropriate response if the theory were accepted.
While there may be various other reasons for this prejudice, surely one of
the most important is that the inconsistency theory of truth is incompatible
with the truth-conditional theory of meaning. Insofar as his own usage of
‘‘semantic’’ tended to encourage the view that ‘‘Tarskian’’ truth conditions
are central to meaning, Tarski has himself ironically helped to create a
prejudice against his own views on the paradoxes and their lessons. (Here as
in so many other cases the malefactor has been himself the chief victim of
his malefaction.) And though the question of merits or demerits of the
inconsistency theory of truth, like the issue of the merits or demerits of the
truth-conditional theory of meaning, of which issue it is but one aspect, is
too large a question to be gone into here, I hope it could at least be agreed
that such questions ought to be examined without prejudice. Since
nothing, perhaps, does more to encourage a bias in favor of the view that
‘‘Tarskian’’ truth conditions are central to meaning than the fact that Tarski
himself calls the approach to truth involving such inductive clauses
‘‘semantic,’’ we have here another reason, additional to the modal muddles
mentioned earlier, for avoiding that usage. Let us therefore, in roaming the
vast field Tarski opened up for us, not follow him in his one terminological
trespass. Let us honor him for every aspect of what he called the ‘‘semantic’’
conception of truth except for his calling it that.
9
Which modal logic is the right one?
1 THE QUESTION
Which if any of the many systems of modal logic in the literature is it whose
theorems are all and only the right general laws of necessity? That depends
on what kind of necessity is in question, so I should begin by making
distinctions.
A first distinction that must be noted is between metaphysical necessity or
inevitability – ‘‘what could not have been otherwise’’ – and logical necessity
or tautology – ‘‘what it is self-contradictory to say is otherwise.’’ The stock
example to distinguish the two is this: ‘‘Water is a compound and not an
element.’’ Water could not have been anything other than what it is, a
compound of hydrogen and oxygen; but there is no self-contradiction in
saying, as was often said, that water is one of four elements along with earth
and air and fire.
The logic of inevitability might be called mood logic, by analogy with
tense logic. For the one aims to do for the distinction between the
indicative ‘‘it is the case that . . .’’ and the subjunctive ‘‘it could have been
the case that . . .,’’ something like what the other does for the distinction
between the present ‘‘it is the case that . . .’’ and the future ‘‘it will be the case
that . . .’’ or the past ‘‘it was the case that . . .’’ The logic of tautology might
be called endometalogic, since it attempts to treat within the object language
notions that classical logic treats only in the metalanguage. However, it
hardly deserves a name, since it immediately splits up into two subjects.
For a second distinction must be between two senses of tautology. On
the one hand, there is model-theoretic logical necessity or validity, the non-
existence of a falsifying interpretation, ‘‘being true by logical form alone.’’
On the other hand, there is proof-theoretic logical necessity or demonstrabil-
ity, the existence of a verifying derivation, ‘‘being recognizable as true by
logical considerations alone.’’ Likewise, there is a distinction between two
notions of contradiction, model-theoretic unsatisfiability and proof-theoretic
169
inconsistency, and between two notions of implication, model-theoretic
consequence and proof-theoretic deducibility. There would be at least a
conceptual distinction even if logic were understood narrowly as first-order
logic, where the model-theoretic and proof-theoretic notions coincide in
extension by the Gödel Completeness Theorem. There may be a difference
in extension between them when logic is understood more broadly: for
instance, if it is taken to include higher-order logic and the mathematics
that goes with it.
Logicians often call model-theoretic and proof-theoretic necessity
semantic and syntactic logical necessity. However, there is a conflict between
this usage and the older usage of linguists on which, roughly speaking,
‘‘semantic’’ means ‘‘pertaining to meaning’’ and ‘‘syntactic’’ means ‘‘per-
taining to grammar.’’ There is a conflict between the two usages of
‘‘semantics,’’ especially, because there is or may be a gap between mathe-
matical modeling and intended meaning. In any case, shorter labels than
‘‘the logic of semantic logical necessity’’ and ‘‘the logic of syntactic logical
necessity’’ would be useful. One might use proplasmatic logic and apodictic
logic, from the Greek for model and proof. But it may be more suggestive
to use validity logic and demonstrability logic, by analogy with provability
logic. The analogy between provability logic and demonstrability logic is
especially close, the one being concerned with what a theory can prove, the
other with what we can demonstrate, the ‘‘can’’ in each pertaining to ability
in principle, regardless of practical limitations.
The question which is the right system of tense logic is not one for the
logician: the logician can indicate how this or that or the other system
corresponds to this or that or the other theory of the nature of time, but
which is the right theory of the nature of time is a question for the physicist.
Similarly, the question which is the right system of mood logic would seem
to be one not for the logician, but for the metaphysician. By contrast, the
question which is the right system of validity or demonstrability logic
cannot be passed off by logic to some other discipline.
The question which is the right validity logic has been answered at the
sentential level, which is the only level that will be considered here: it is the
system known as S5. This result is essentially established already in Carnap
(1946).
The question which is the right demonstrability logic, again at the
sentential level, goes back to the earliest days of modern modal logic. For
though the founder of the subject, C. I. Lewis, did not clearly distinguish
among metaphysical, model-theoretic logical, and proof-theoretic logical
modalities, still he did always write of necessitation as implication, and did
Which modal logic is the right one? 171
often write of implication as deducibility, so that it is reasonable to
conclude that by necessity he primarily meant tautology, by which in
turn he primarily meant demonstrability. No one today, however, takes
seriously his suggestion that the right logic for this notion might be the
feeble S1 or the bizarre S3. To the extent that there is any consensus or
plurality view among logicians today, I take the view to be that the right
demonstrability logic is S4. (Even in ‘‘relevance’’ or ‘‘relevant’’ logic, where
S4 cannot be literally accepted, since the classical sentential logic it is based
on is rejected, still it seems to be a consensus or plurality view that the right
logic should be ‘‘S4-like.’’) The locus classicus for such a view is a paper from
the proceedings of a famous 1962 Helsinki conference on modal logic
(Halldén 1963).
While the argument for the soundness of S4 as a demonstrability logic
given there seems as compelling as an ‘‘informally rigorous’’ argument can
be, there is no real argument for completeness, which remains an open
question. It therefore remains conceivable that the right logic is something
stronger than S4: that it is something intermediate between S4 and S5, such
as S4.2 or S4.3; or that it is something stronger than S4 but incomparable
with S5, such as the logic called Grz after Grzegorczyk (1967) and the logic
that ought to be called McK after McKinsey (1945). (In the literature it has
heretofore been misleadingly called S4.1, though it is not intermediate
between S4 and S5.)
The issues are sufficiently illustrated by the cases of the distinctive axioms
of S4.2 and of McK, which are equivalent respectively to (&&p
&&p), the principle that ‘‘nothing is both demonstrably not demon-
strable true and demonstrably not demonstrably false,’’ and to &&p
&&p, the principle that ‘‘everything is either demonstrably not demon-
strably true or demonstrably not demonstrably false.’’ Halldén rightly says of
the latter – what he could also have said of the former – that it is not an
intuitively plausible principle when the box & is meant as demonstrability.
But to say this is to do something less than to give an ‘‘informally rigorous’’
argument for the claim that either principle outright fails as a general law, let
alone for the claim that any principle not a theorem of S4 does so.
The question which is the right provability logic has been answered, and
though results are often stated for a single-theory, classical first-order
arithmetic, many hold for all true theories satisfying certain minimum
requirements of strength. Actually, one must distinguish the question of
which logic gives all and only those principles about provability all whose
instances are provable by the theory in question from the question which
gives all and only those principles that are valid (or demonstrable by us).
The answer to the former question is given by a system GL, and to the latter
question by a system GLS. Both differ from the Lewis systems S4 and S5 by
lacking the law &(&p!p). The failure of this law is, roughly speaking, the
content of the Gödel Incompleteness Theorems. The standard reference is
of course Boolos (1993).
Below, in x2 I will recall the case for the soundness of S5 as a validity
logic and of S4 as a demonstrability logic. In x3 I will recall the Carnapian
case for the completeness of S5. In x4 I will indicate the minimal require-
ments of strength that are assumed in provability logic, and that I will be
assuming in demonstrability logic also, and attempt to clarify the relation-
ship between the two logics. In x5 I will present a case against McK as a
demonstrability logic; and it will generalize to a case against any system not
contained in S5, such as Grz. Finally, in x6 I will present a case against S4.2;
and this will of course also constitute a case against any stronger system,
such as S4.3. But the case of weaker systems intermediate between S4 and
S5 will be left open, and with it the general question.
2 SOUNDNESS
A key consequence of the step of treating modality in the object language,

treating & as a one-place connective on a par with , is that iterated
modalities, modalities embedded inside modalities, as in &&p, are
allowed. By contrast, when ‘‘valid’’ is expressed only by a word of the
metalanguage, applicable only to formulas of the object language, there can
be no question of iterations like ‘‘it is valid that it is not valid that . . .’’ All
the modal systems most commonly considered in the literature agree with
classical meta-logic, in the sense that where classical meta-logic has a law,
for example ‘‘a valid conclusion is a consequence of any premise,’’ these
systems will have a corresponding law, in the example &p!(q )p).
Agreeing as they all do with classical logic, these systems agree with each
other for formulas without iterated modalities. What distinguishes S5 is
that it has laws that make every iterated formula more or less trivially
equivalent to an uniterated formula.
The first step in establishing S5 as the right logic of validity is to establish
the soundness of S5 as a validity logic: to establish that every theorem of S5 is
correct as a general law about validity, or what comes to the same thing,
that every axiom of S5 is thus correct, and that every rule of S5 preserves
such correctness. This is completely unproblematic for the non-modal
axioms and rules, which are simply those of the sentential component of
classical, non-modal sentential logic. Moreover, though S5 is usually
formulated with a specifically modal rule allowing &A to be taken as a
theorem whenever A is a theorem, this rule can be dispensed with in favor
of adding &A as an axiom whenever A is an axiom of the usual formulation,
which is to say, whenever A is either a classical, non-modal axiom or one of
the following modal axioms:
(1) &p!p
(2) &(p!q)!(&p!&q)
(3) &p!&&p
(4) &p!&&p.
Again for the classical, non-modal axioms this is completely unproble-

matic, while in making a case for – which is to say, in demonstrating – any
one of (1)–(4), one will at the same time be making a case for its demon-
strability, and a fortiori for its validity.
Thus the problem of establishing the soundness of S5 for validity
logic reduces to that of establishing the correctness of (1)–(4), and
similarly the problem of establishing the soundness of S4 for demon-
strability logic reduces to that of establishing the correctness of
(1)–(3). Indeed, for (1)–(3) correctness for the box as validity and for
the box as demonstrability can be established by more or less parallel
arguments. Consider (1), for example. The arguments are simply the
parallel ones that whatever is true by logical form alone must be true,
and that whatever can be recognized to be true by logical consider-
ations alone must be true. But indeed, I need not enlarge on the case
for (1)–(3), which is adequately made by Halldén.
It remains to consider the distinctive axiom (4), which of course is being
proposed only as a general law of validity, not of demonstrability. Here the
main point is just as follows. Consider any particular instance of (4):
(5) If it is not true by logical form alone that p, then it is true by logical form alone
that it is not true by logical form alone that p.
Suppose that the antecedent of (5) is true, which is to say that the following
is false:
(6) It is true by logical form alone that p.
Since (6) is false, there must be some y of the same logical form as p that is
false. Now consider anything else of the same logical form as (6). It will
look like the following, wherein q has the same logical form as p:
(7) It is true by logical form alone that q.
But then y also has the same logical form as q, and since y is false, (7) is
false. In other words, anything of the same logical form as (6) is false, and
hence the following is true:
(8) It is true by logical form alone that it is not true by logical form alone that p.
Thus the consequent of (5) is true, as required.
3 COMPLETENESS FOR VALIDITY LOGIC
In the case of provability logic, as expounded in Boolos (1993), and of

intuitionistic logic, as expounded in Burgess (1981a), once the candidate
logic S has been identified, the argument that it is the right one consists
of three parts: soundness, formal completeness, and material completeness.
That is, it is shown that every theorem of SPis acceptable as a general law
under the intended interpretation; a class of mathematical models is
identified and it is shown that (every theorem
P of S and) no non-theorem of
S comes out true in all models of class ; andPit is shown that any formula
that comes out untrue in some model of class is unacceptable as a general
law under the intended interpretation. This last step, bridging the gap
between ‘‘semantics’’ in the logicians’ sense and ‘‘semantics’’ in something
more like the linguists’ sense, is due in the case of provability logic to
R. M. Solovay, and in the case of intuitionistic logic to Georg Kreisel, who
coined the phrase ‘‘informal rigor’’ in this connection. In both cases, the last
step is the most difficult. The situation is rather similar in the case of
validity logic.
Beginning with formal completeness for the case of validity logic
(soundness having already been discussed), the kind of models now stand-
ard in modal logic first became widely known through another talk at the
1962 Helsinki conference, this one by Saul Kripke; another kind of model
near to those standardly used became widely known through yet another
talk at the same conference, this one by Jaakko Hintikka. A frame model
M, as in Kripke (1963), consists of two parts, a frame and a valuation. The
frame consists of a set W of indices, a two-place relation R on it, and a
designated member w0 of it. A valuation V is a specification for each x in W
and each sentence letter p, q, r, . . . of whether or not the sentence letter
counts as true in that index. The notion of truth in an index is extended to
compound formulas by recursion:
A is true at x if and only if A is not true at x.
A B is true at x if and only if A is true at x and B is true at x.
&A is true at x if and only if A is true at y for every y in W such that Rxy.
It is permitted to have two indices x and y at which exactly the same set of
formulas are true, and such duplication is often important. A formula
counts as holding in M if it is true at w0, and as being valid in a class K
of frames if it holds in every model whose frame is in that class.
The proposal in Hintikka (1963) is less purely ‘‘semantic’’ or model-
theoretic: it is still ‘‘syntactic’’ or proof-theoretic in that, while it has a
relation R, what this relation relates are not abstract indices, but sets
of formulas, and so one does not have duplication. But as it happens,
in connection with the system S5, differences between the approaches
of Kripke and Hintikka, such as permitting or forbidding duplication,
are unimportant; and so for that matter is the main similarity
between the two approaches: the presence of a relation R. For while
the theorems of S5 can be characterized as the formulas valid for the
class of reflexive, transitive, and symmetric frames, this characteriza-
tion reduces, by a series of steps too familiar to bear repetition here,
to a much simpler one.
Consider, for any k, the formulas involving only the sentential variables
or atomic formulas p1, . . ., pk. Then for such formulas, a model may be
taken to consist simply of a non-empty subset W of the set of rows of the
truth table p1, . . ., pk, with one such row w0 designated. The notion of truth
at a row x in the truth table is defined for compound formulas by a
recursion in which the first two clauses are exactly the same as above,
while the third reads as follows:
&A is true at x if and only if A is true at y for every y in W.
A formula counts as holding in such a model M if it is true at w0. The

theorems of S5 may be characterized as the formulas that hold in all such
models.
Turning to material completeness for the case of validity logic, it may be
well to begin by considering an objection to the modeling just described
that has been independently advanced by several writers. One of them put
it as follows:
What is needed for logical necessity of a sentence p in a world w0 is more than its
truth in each one of some arbitrarily selected set of alternatives to w0. What is
needed is its truth in each logically possible world. However, in Kripke semantics it
is not required that all such worlds are among the alternatives to a given one.
It is then suggested that one should adopt not the standard model theory,
or the simplification thereof described above, but rather a deviant model
theory, which after simplification amounts to just this, that the only model
admitted is the one consisting of all rows of the truth table.
There is a fallacy or confusion here. What is wanted is that the technical
notion of coming out true in all models should correspond to the intuitive
notion of coming out true under all interpretations, or all substitutions of
specific p1, . . . , pk for the variables p1, . . . , pk. Since, for instance, among all
the many substitutions available there are ones in which the p1 substituted for
p1 is the same as the p2 substituted for p2, so that it is impossible for p1 and p2
to have different truth values, there must correspondingly be among the
models one available where the only rows of the truth table present are those
for which the value given to p1 is the same as the value given to p2.
The confusion in the objection becomes apparent when one notes that
in the deviant model theory suggested, &(p1 p2) counts as valid,
whereas of course &(p1 p1) does not, so that the standard rule of
substitution fails. But the rule of substitution must hold so long as one
adheres to the standard conception of the role of the variables p1, . . . , pk,
according to which arbitrarily selected p1, . . . , pk may be substituted for
them. Indeed, the deviant model theory corresponds to a deviant concep-
tion on which independent p1, . . . , pk must be substituted for distinct p1, . . . ,
pk. The confusion is worse confounded when it is suggested that the
difference between the standard and deviant model theories somehow
corresponds to a difference between non-logical and logical notions of
necessity. For what is at issue is, to repeat, differences in conceptions of the
role of variables, not in conceptions of the nature of necessity.
Yet, confused as it is, the objection does serve to call attention to an
important question. Each substitution of specific p1, . . . , pk for p1, . . . , pk
determines a non-empty set of rows of the truth table, consisting of all and
only those rows x such that it is not impossible by the logical forms of the pi
alone for them to have the truth values x assigns to the corresponding pi.
The question is, is it the case that for any arbitrarily selected non-empty
subset W of the set of rows of the truth table, there are specific statements
p1, . . . , pk that determine, in the manner just described, exactly that subset?
In other words, if a formula is not a theorem of S5, and therefore fails in
some standard model, is there some specific instance in which it fails? An
affirmative answer to this question is precisely what is needed to establish
the material completeness of S5 as validity logic.
It is a reasonable assumption, and one presumably made by the critics
alluded to, that there exist indefinitely many a, b, g, . . . that are independ-
ent in the sense that any conjunction of some of them with the negations of
the rest of them is possible, in the relevant sense of possibility. For instance,
if a, b, g, . . . are of simple subject–predicate form with distinct subjects
and predicates in each, they will be thus independent. Given this assump-
tion, an affirmative answer to the foregoing question is forthcoming. As
this result has in effect already been expounded several times in the
literature, in Carnap (1946), Makinson (1966), and S. K. Thomason
(1973), there should be no need for me to do more than give an illustrative
example here. Indeed, a simple one, involving just three sentence letters
p, q, r, should suffice.
Consider the set W containing just the three rows in which two of p, q, r
are true and the other false. Call the one where r alone is false x, the one
where q alone is false y, and the one where p alone is false z. What is to be
established is that given independent a, b, . . . , there are truth-functional
compounds p, y, q thereof that might be substituted for p, q, r, for which
the three rows indicated represent all and only the combinations of truth
values that are not false by logical form alone.
To find the required compounds, one first finds three auxiliary compounds
, u, , that are pairwise exclusive and jointly exhaustive, meaning that the
conjunction of any two must be false by logical form alone, while the
disjunction of all three must be true by logical form alone. Setting ¼ a and
u ¼ a b and ¼ a b will do. One next lets the auxiliaries , u,
correspond to the rows x, y, z, and takes as the substitute for a given one of
p, q, r the disjunction of the auxiliaries corresponding to the rows in which it
is true. Thus the substitute p for p should be u or a (a b), which
simplifies to a b. It can be worked out that the substitutes y and q for q and r
simplify to a b and a, respectively. And it can then be worked out that
exactly two of the three, a b and a b and a, must be true, and that
given the independence of a and b it may be any two of the three, as required.
Before leaving the topic of validity logic it may be mentioned that the
fact that S5 is indeed the right logic can be confirmed in a different way.
After soundness is established in order to show that no stronger system than
S5 is acceptable, one would appeal to the result of Scroggs (1951), according
to which the only extensions of S5 are finitely-many-valued logics. One
would then argue that no finitely-many-valued logic can be correct for
semantic logical necessity (given the same reasonable assumption as above,
that there are indefinitely distinct independent statements).
4 DEMONSTRABILITY AND PROVABILITY
A word must be said to clarify the relationship between demonstrability

and provability logics, and to dispel a puzzle about that relationship.
The minimal assumptions of strength needed for provability logic are
three. They may be formulated either as assumptions on the notion of
proof for the theory, or as assumptions on the set of theorems provable. On
the formulation in terms of proofs, the first assumption would be that
whether something is or is not a proof in the theory is decidable, which by
Church’s Thesis implies that the relation of proof-in-the-theory to theo-
rem proved is recursive. The second assumption would be that the rules of
classical first-order logic may be used in proofs in the theory. The third
assumption would be that certain basic, finite, combinatorial modes of
reasoning – whose exact scope need not be gone into here, except to say
that, since we want to get the Second Incompleteness Theorem, the scope
needs to be somewhat wider than it would need to be if we only wanted to
get the First Incompleteness Theorem – may be used in proofs in the
theory.
On the formulation in terms of theorems, it would first be assumed that
the set of theorems provable is recursively enumerable. It would second be
assumed that the set of theorems provable is closed under the rules of
classical first-order logic. And it would third be assumed that the set of
theorems provable includes certain basic, finite, combinatorial results.
Clearly, the list of assumptions on the theory stated earlier yields the list
of assumptions on the set of theorems just stated. And conversely, by
Craig’s Lemma, any set of conclusions satisfying this latter list of assump-
tions coincides with the set of conclusions provable in some theory satisfy-
ing the former list of assumptions.
In demonstrability logic, at least as I will be considering it here, it is to
be assumed that whether something constitutes a demonstration of a
given conclusion is decidable, that the rules of classical first-order logic
may be used in demonstrations, and that certain basic, finite, combina-
torial modes of reasoning may be used in demonstrations. By what has
already been said, it follows that the set of conclusions we can demon-
strate coincides with the set of conclusions that can be proved in some
theory of the kind to which provability logic applies. And yet, provability
logic and demonstrability logic are supposed to be different, in that by
what has been said in earlier sections, (1) below is false, while (2) below is
true:
(1) It can be proved in such-and-such a theory T that if something can be proved

in such-and-such a theory T, then its negation cannot also be.
(2) It can be demonstrated by us that if something can be demonstrated by us,
then its negation cannot also be.
What may be puzzling is how it can be that (1) fails while (2) holds and as
already indicated the above-stated assumptions commit one to the truth of
something of the following form:
(3) What can be demonstrated by us coincides with what can be proved in
such-and-such a theory T.
Indeed, a notorious objection to (3) above, associated with the names
of J. R. Lucas and Roger Penrose, claims that it, together with the true
(2) above, yields the false (1) above.
The solution to the puzzle is to point out that this objection commits the
fallacy of assuming that co-extensive terms, such as ‘‘what we can demon-
strate’’ and ‘‘what such-and-such a theory T can prove’’ can be substituted
without change of truth value everywhere, even in intensional contexts,
such as ‘‘we can demonstrate that . . . ’’ or ‘‘such-and-such a theory T can
prove that . . .’’ To get (1) above from (2) above one would need something
stronger than (3) above, namely the following:
(4) We can demonstrate (3).
The solution of the puzzle is that (3) does not yield (4). (By analogy,
those familiar with the work of Solomon Feferman and William Tait on
the standpoints of the predicativists and the finitists, thinkers who owing to
their philosophical prejudices cannot demonstrate all that we can, will
recall that what is demonstrable from the standpoint of one or the other of
these ’isms can indeed be exactly captured by a theory, though the ’ists
themselves cannot recognize as much.)
The fallacy should become obvious on comparing (1)–(3) above with the
following:
(10 ) The only man with such-and-such a number N of hairs on his head knows
that the only man with such-and-such a number N of hairs on his head is
gray-haired.
(20 ) I know that I am gray-haired.
(30 ) I am the only man with such-and-such a number N of hairs on his head.
Clearly (30 ) above by itself does not, with (20 ) above, yield (10 ) above.
Rather, one would need the following stronger assumption:
(40 ) I know (30 ).
But (30 ) above does not yield (40 ) above.

5 AGAINST McK
Beginning with formal completeness for the case of demonstrability logic,

soundness having already been discussed, the theorems of S4 can be
characterized as the formulas valid for the class of reflexive and transitive
frames; and equally, they can be characterized as the formulas valid for the
class of finite reflexive and transitive frames, a deeper result implying the
decidability of the logic. A historical fact is worth mentioning, that
the result just stated follows immediately from two results already in the
literature two decades before the famous Helsinki conference. One of
these, from McKinsey (1941), characterizes the theorems of S4 in terms
of a class of finite models of a different kind, based not on frame structures
but on algebraic structures of a certain kind. The other of these, from
Birkhoff (1937), connects finite algebraic structures of the kind in question
with finite reflexive and transitive frame structures. (The former paper
makes no mention of frames, and the latter no mention of modal logic.)
This history is worth mentioning among other reasons because the older
algebraic modeling involved, which is sometimes not taught to students of
the subject today, still has its uses even after the development of frame
models, and I will be citing an instance later.
Turning to material completeness, no decisive results have yet been
obtained, and my aim will only be to present, case by case, some partial
results. To begin with, it is not only a reasonable assumption, as already
said in an earlier section, that there exist indefinitely many a, b, g, . . . that
are independent, but also a reasonable assumption that there exist indefin-
itely many a, b, g, . . . that are demonstrably independent. For instance, if a,
b, g, . . . are recognizably of simple subject–predicate form with distinct
subjects and predicates in each, they will be thus demonstrably independ-
ent. It follows that if D is any compound formed by negation and
conjunction or disjunction from p, q, r, . . . such that D is a not a theorem
of classical sentential logic, or in other words, such that D comes out false
in some row of the pertinent truth table, then the result D of substituting
a, b, g, . . . for p, q, r, . . . in D will be demonstrably not demonstrable, or
demonstrably indemonstrable. Similarly, if D is such that D is not a
theorem of classical sentential logic, or in other words, such that D
comes out true in some row of the pertinent truth table, then D will be
demonstrably not demonstrably false, or demonstrably irrefutable. Thus
from any D such that neither it nor its negation is a theorem of classical
sentential logic, we get a counterexample D to the McK axiom that nothing
is both demonstrably indemonstrable and demonstrably irrefutable.
This argument can be generalized to apply to any axiom that is not a
theorem of S5.
Perhaps the easiest route to a generalization is to draw on the work of
Slupecki and Bryll (1973). They pursue the old idea of the Polish school
that a logical system should have in addition to its axioms and rules of the
ordinary kind, its axioms and rules of acceptance, indicating that certain
formulas are acceptable as general laws, some axioms and rules of an
opposite kind, axioms and rules of rejection, indicating that certain for-
mulas are unacceptable as general laws. Just as a formula P is a theorem of
the system, in symbols j- P, if there is a sequence of steps, each an axiom of
acceptance or following from earlier ones by a rule of acceptance ending in
P, so the goal would be to have for each formula Q that is not a theorem of
the system, in symbols -j Q, a sequence of steps involving axioms and rules
of rejection ending in Q.
For classical logic there would be the axiomatic rejection of the constant
false -j and ? rules of rejection that are the reverse of the usual classical rules
of acceptance: if -j P 0 where P 0 is a substitution instance of P, then -j P, and if
-j Q and -j P Q, then -j P. For any modal logic there would be also the rule
of rejection that is the reverse of the usual modal rule of acceptance: if -j &P
then -j P. For each particular modal system additional rules of rejection
would be needed. For S5 Slupecki and Bryll show that just one additional
rule suffices:
if -j P!Q 1 and . . . and -j P!Q N then -j &P!(&Q1 . . . &QN)
where P and the Q i involve no modalities.
In order to show that any non-theorem of S5 should be rejected as a
general law of demonstrability, it will suffice therefore to argue that the
above rule of rejection is acceptable for demonstrability. And indeed, if the
P!Q i are unacceptable as general laws, there must for each be a row xi of
the truth table for the variables p, q, . . . involved on which P comes out true
and Q i comes out false. But then by Carnap’s result there are specific p,
y, . . . that could be substituted for the variables p, q, . . . in P and the Q i to
give sentences and Ci, such that the xi represent all and only the possible
combinations of truth values for the p, y, . . . It follows that is an
instance of a theorem of classical sentential logic, hence demonstrable,
while each Ci is demonstrably indemonstrable by the considerations of the
preceding paragraph. Thus the following:
&!(&C1 . . . &CN)
fails, and the formula of which it is an instance, namely the following:
&P!(&Q 1 . . . &Q N)
is not acceptable as a general law, as required.
6 AGAINST S4.2
The S4.2 principle says that everything is either demonstrably indemon-

strable, or demonstrably irrefutable. An argument against the acceptability
of this principle as a general law can be given.
In addition to assumptions about demonstrability listed in earlier
sections, including the assumption that there are indefinitely many
instances recognizable as being of simple subject–predicate form, I need
the reasonable assumption that there is an instance recognizable as being
of simple subject–verb–object form, with no further pertinent logical
structure. That is, I assume there is a two-place predicate that is
recognizably a two-place predicate with no further pertinent logical
structure, so that it is recognizable that all its pertinent logical structure
is represented when it is represented by a simple two-place predicate
variable F. Given such an example, for any compound formed from
using negation, conjunction or disjunction, and universal or existential
quantification, the logical form of will recognizably be represented by a
formula P of classical first-order logic formed from F using , or ,
and 8 or 9. Let be the set of all such compounds , and L the set of the
corresponding formulas P.
Now suppose a formula P in L fails in no model of the kind used in
classical first-order logic. Then P is a theorem of classical first-order logic
by the Gödel Completeness Theorem; and by the assumption that the
rules of classical first-order logic may be used in demonstrations,
the corresponding in will be demonstrable, and demonstrably
so. Then by general laws represented by theorems of S4, ‘‘it is demon-
strable that ’’ will be demonstrably irrefutable and not demonstrably
indemonstrable.
Now suppose the formula P of L fails in some finite model of the kind
used in classical first-order logic. Then basic, finite, combinatorial reason-
ing shows that it does so, and hence that it does not represent a correct
general law; and by the assumption that basic, finite, combinatorial rea-
soning may be used in demonstrations, the corresponding in will be
indemonstrable, and demonstrably so. Then by general laws represented by
theorems of S4, ‘‘it is demonstrable that ’’ will be demonstrably inde-
monstrable and not demonstrably irrefutable.
For any formula P of L, let C(P) be ‘‘it is demonstrable that ,’’ where
in is the result of substituting for F. Let X be the set of P such that C(P)
is demonstrably irrefutable, and let Y be the set of P such that C(P) is
demonstrably indemonstrable. What has been established so far is that if
P has no counter-model, then P belongs to the difference set X Y; while
if P has a finite counter-model, then P belongs to the difference set Y X.
What the S4.2 principle yields is that the union set X [Y is all of L.
To complete the case against the S4.2 principle, I must invoke the
assumption that the set of demonstrable conclusions is recursively enumera-
ble, from which it follows that the sets X and Y are also recursively enumera-
ble. Then by the Reduction Theorem for recursively enumerable sets it
follows that there are recursively enumerable sets X and Y satisfying the
conditions that X ˝ X and Y ˝ Y and X \ Y ¼Ø and X [ Y ¼ X [ Y.
These conditions imply that X Y ˝ X and Y X ˝ Y . What the S4.2
principle yields is that X and Y are complements of each other in the
recursive set L, which since both are recursively enumerable yields that both
are recursive. What was established earlier yields that if P has no counter-
model, then P belongs to X , while if P has a finite counter-model, then
P belongs to Y and so does not belong to X . And now we have a contra-
diction, since by an elaborated version of Church’s Theorem, there is no
recursive set Z separating the formulas with no counter-models from those
with finite counter-models.
The foregoing argument applies to just one of the infinitely many
formulas that are theorems of S5 but not of S4. Can all such formulas be
rejected? Clearly, if they can, some general argument, not a case-by-case
examination of examples, will be needed to establish that fact. How
might such an argument proceed? Well, rejection principles for S4 have
been formulated by Valentin Goranko (1994), and simpler ones have
been found by Tomasz Skura (1995), who works with finite algebraic
models of the kind alluded to earlier. Skura requires two principles, the
first being a slight variant of the rejection principle for S5 considered
earlier.
Unfortunately, Skura’s second principle, though simpler than Goranko’s
principles, is complex enough that it is not very perspicuous, and it is not easy
to argue why it should be acceptable for syntactic logical necessity. (Goranko
and Skura do not themselves consider such questions.) Fortunately, Skura
does not claim his rules are the simplest feasible, but on the contrary he
explicitly poses it as an open question whether there are any simpler ones. It
may be that this open question will have to be settled before one can settle the
status of the conjecture that S4 provides the answer to the question of which
is the right demonstrability logic.
At present this question, which as I have said goes back to the founders
of modern modal logic, remains after most of a century still without a
definitive answer.
10
Can truth out? 1
1 THE DISCOVERY PRINCIPLE
It is rather discouraging that forty years have passed since Frederic Fitch
first propounded his paradox of knowability without philosophers having
achieved agreement on a solution.2 As a general rule, when modal phe-
nomena prove puzzling, it is a good idea to look at the corresponding
temporal phenomena, and accordingly I propose to examine here not the
knowability principle that whatever is true can be known, but rather the
discovery principle that whatever is true will be known.
As Fitch’s modal paradox attacks the knowability principle, so an
analogous temporal paradox threatens the discovery principle. The for-
mulation of the paradox is as follows. Start with the minimal tense logic
with G and H for ‘‘it is always going to be . . .’’ and ‘‘it always has been . . .’’
as primitive, and F and P for ‘‘it sometime will be . . .’’ and ‘‘it once was . . .’’
defined as G and H.3 Add a one-place epistemic operator K for ‘‘it
is known that,’’ and add as axioms minimal assumptions for this new
operator, expressing that anything known is true, and that if a conjunction
is known, so are both conjuncts:
(1) Kp ! p
(2) K(p & q) ! Kp & Kq.
In an attempt to formalize the discovery principle, add one further

axiom:
(3) p ! FKp.
1
First published in Slaerno (2007).
2
(Fitch 1963). For a summary of recent debates, see B. Brogaard and J. Salerno, ‘‘Fitch’s Paradox
of Knowability,’’ in Stanford Encyclopedia of Philosophy: http://plato.stanford.edu/entries/fitch-
paradox/ (available only on-line).
3
See (Burgess 1984a). The various theorems of tense logic cited below can all be found in this source.
185
The paradox is that one can then derive the following:
(4) p ! Kp.
The derivation of (4) using (3) is, apart from replacing & and à by G and F,
the same as Fitch’s derivation, which is too well known to bear repeating
here.
The operator K is intended to indicate human knowledge, not divine
omniscience. The grounds for belief in the discovery principle have indeed
traditionally involved a belief in divine omniscience, but it is not this belief
alone that supports the principle, but rather this belief plus a further belief
that on some future day God will bring it about that whatever is hidden is
made manifest (quidquid latet apparebit). Obviously that day has not yet
come, and the conclusion (4), that everything true is already humanly
known, is an absurdity, and so we have a reductio of the principle (3).
The ‘‘dialethists’’ and other proponents of radical revisions of classical
logic can be counted on to tout their proposed revisions as solutions to
this paradox, as they have touted them as the solutions to so many others.
But a priori it is overwhelmingly more likely that the problem lies not in
the underlying classical logic, but in the least familiar element, the axiom
(3), the only axiom in which temporal and epistemic operators interact.
And indeed, that is where the problem lies. One has to be careful in going
back and forth between symbolism and English prose, and Fitch, or rather
his hypothetical temporal analogue, was not careful enough.
In tense logic p, q, r, . . . are supposed to stand for tensed sentences,
whose truth value may change with time (or if one wants to speak of
‘‘propositions,’’ then they must be propositions in a traditional rather than
a contemporary sense, propositions that are themselves tensed, and whose
truth value may change with time). FA is supposed to be true at a given
time if A is true at some later time. What (3) actually expresses thus
amounts to this:
(5) If p is true now, then at some later time it will be known that p is true then.
The proposed formalization as (3) has in effect turned the principle that
any truth will become known into the principle that any sentence that
expresses a truth will come to be known to express a truth. But this last
formulation invites the immediate objection that the sentence in question
may cease to express a truth before the knowledge of the truth it once
expressed is acquired.
And so (5) surely does not express what Shakespeare meant in saying
‘‘Truth will out.’’ He meant to imply that if Smith murders Jones secretly,
Can truth out? 187
so that no one knows, then it will become known that Smith murdered
Jones secretly, so that no one knew. He did not mean to imply that if what
the form of words ‘‘Unknown to all, Smith has murdered Jones’’ now
expresses is true, then there will come a time when what that same form of
words then expresses will be known to be true. Thus the temporal analogue
of Fitch’s argument does not discredit the discovery principle, because the
target of that argument is not a correct expression of that principle.
2 INEFFABLE TRUTHS
That one particular objection to a principle fails is no proof of the principle

itself, and indeed no proof that it may not be open to simple, straightfor-
ward objection along other lines. And in fact the discovery principle is open
to two kinds of objection, each of which requires us either to impose a
restriction on the principle, or to assume charitably that a restriction on the
principle was already intended by its advocates.
As background to a first objection consider the timing of the collision of
two ordinary extended material objects. The boundaries of such objects
generally are sufficiently ill-defined on a scale of nanometers as to make
dating their collision on a scale finer than nanoseconds meaningless. If
murders, say, are all the events we want to talk about, we do not need to
conceive of ‘‘times’’ as durationless ‘‘instants,’’ but may conceive of them as
very brief ‘‘moments,’’ of no more than, say, a nanosecond’s duration. In
this case, chronometry – by which I here mean no more than our usual
ways of dating events by year, month, day, hour, minute, second, and on
to milli- or micro- or nanosecond and beyond if one wishes, all tacitly
understood relative to some fixed time zone – supplies a term for every
time.
But it may be otherwise if we wish to speak of point-particles and
their collisions. The worry is that there will be truths that can never
be known because they can never be stated. Suppose, for instance, that
¼ 0.182564793 . . . is an irrational number, and that exactly seconds
before 12:00 p.m., particle i collided with particle j. Can it ever become
known that particles i and j collided at exactly seconds before 12:00 p.m.
on June 1, 2003?
According to the discovery principle, all the following will become
known:
(1) Particles i and j collided at 182 milliseconds before 12:00 p.m. on June 1,
2003.
(2) Particles i and j collided at 182564 microseconds before 12:00 p.m. on
June 1, 2003.
(3) Particles i and j collided at 182564793 nanoseconds before 12:00 p.m. on
June 1, 2003.
Here ‘‘’’ abbreviates ‘‘the nearest unit,’’ to the nearest milli- or micro- or
nanosecond, as the case may be. (For the sake of argument, set aside any
quantum-mechanical doubts about whether the series (1)–(3) really could
be continued indefinitely.)
But for it to be knowable that i and j collided at exactly seconds before
12:00 p.m., would it not have to be sayable that i and j collided at exactly
seconds before 12:00 p.m.? And for this to be sayable, there would have to
be some means in language or thought of referring to the irrational number
– I mean, of course, some means other than referring to it as the number
of seconds before 12:00 p.m. when i and j collided. Mathematics supplies
such means for few irrational numbers, such as p2, p, e, and so forth.
Coincidence may supply a few others: the time when i and j collided may
be describable also as the time when k and l collide, if the two collisions
happen to be simultaneous. But by cardinality considerations we inevitably
lack means of reference to most irrational numbers.
The discovery principle must be understood to exclude ineffable truths.
It must be understood as restricted to truths expressible in our language.
Such a restriction will be built into any tense-logical formalization of the
principle, if the letters p, q, r, . . . are understood as standing for sentences
of our language. Such a restriction seems in one sense not too serious,
because the principle still tells us that for any question we have the language
to ask, the true answer will become known.
3 EPHEMERAL TRUTHS
A second objection to the discovery principle is more subtle. Suppose that

as I write it is 12:00 p.m., June 1, 2003. Then the following is true:
(1) Now, this moment, it is 12:00 p.m., June 1, 2003.
Obviously (1) itself will never be true in the future. And it seems that
no sentence of our language will ever express in the future exactly what
(1) expresses now. Thus the truth that (1) now expresses seems to be one
that will be unknowable in the future because it is unsayable in the future.
Moreover, the demonstrative ‘‘this moment’’ and the indexical ‘‘now’’
are both pleonastic, what they indicate being already sufficiently indicated
Can truth out? 189
by the fact that the verb ‘‘is’’ is in the present tense. Thus what has just been
said about (1) is equally true of the following:
(2) It is 12:00 p.m., June 1, 2003.
And indeed, if now, this minute, Smith is murdering Jones, then the
following is another example subject to the same difficulty as (1) and (2).
(3) Smith is murdering Jones.
The truth that Smith is (now, this moment) murdering Jones seems one
that will be unsayable and therefore unknowable in the future, even though
it is sayable now and even knowable now.
The discovery principle must be understood to exclude not only inef-
fable truths, which are never expressible in our language, but also ephemeral
truths, which are expressible for a moment, and then never again. Such a
restriction seems in one sense not too serious, because it does not leave us
with any question that can always be asked and never be answered. The
ephemeral will be equally inexpressible interrogatively as assertorically.
Such a restriction seems not too serious for another reason, because the
truths it excludes from human knowledge in the future are excluded even
from divine knowledge in eternity, if one follows those theologians who
make the latter knowledge timeless. For (1)–(3) are no more true in a
timeless eternity than they will be true in the seconds and minutes and
hours and days and months and years to come. The old riddle that suggests
an exception to the principle that God can see anything I can see is a joke.4
But the counterexamples (1)–(3) to the principle that God knows anything
I can know are not. This point seemed worth digressing to mention, if only
because a desire to have a formal apparatus in which such issues could be
discussed was an important part of the motivation of the creation of tense
logic by Arthur Prior.
4 A REFORMULATION
We have seen that (1.3) – displayed item (3) of x1 – is not the right
formalization of the discovery principle. What is? It cannot be claimed
4
I mean the riddle:
Q. What is it that God never sees, that the king seldom sees, but that you and I see every day?
A. An equal.
This seems less a problem for theologians than for partisans of ‘‘substitutional quantification.’’
that a complete solution to the paradox has been obtained until this
question is answered.
One answer suggests itself at once. Now that we have restricted the
principle to truths that will remain expressible in our language in the
future, it is tempting to formulate the principle as the principle that any
sentence that will continue to express a truth in the future will come to be
known to express a truth. This goes over into symbols as follows:
(1) Gp ! FKp.
And (1) is, unlike (1.3), immune to Fitch-style paradox, even if one
considerably strengthens the background tense logic. For definiteness, let
us consider the tense logic, call it Llinear, that is appropriate for linearly
ordered time without a last time. Then the immunity of (1) from Fitch-
style paradox is the content of the following proposition.
Proposition. Let T be Llinear plus (1.1), (1.2), and (1). Then (1.4) is not a
theorem of T.
Proof. Consider an auxiliary theory T , obtained from L by adding a
constant p and the following axiom:
(2) Fp.
Then p is not a theorem of T . For if we take any of model of Llinear, and let
p be true at and only at the times later than the present, then (2) will be true
at all times, but p will not, being false at all past times and at the present
time.
Next assign each formula A of the language of T a translation A into the
language of T , by taking Kp to abbreviate p & p. Thus (1.1), (1.2), (1), and
(1.4), respectively, are translated as follows:
(3) p&p!p
(4) (p & q) & p ! (p & p) & (q & p)
(5) Gp ! F(p & p)
(6) p ! p & p.
Note that the translation (6) of (1.4) is not a theorem of T . For if it were,
substituting p for p and applying truth-functional logic, p would be a
theorem, as we have seen it is not.
To show that (1.4) is not a theorem of T, it will suffice to show that the
translation of any theorem of T is a theorem of T . And to show this, it will
suffice to show that the translations (3)–(5) of the three axioms of T are
theorems of T . For the first two axioms this is trivial, since (3) and (4) are
truth-functional tautologies. For the third axiom, the following is a theo-
rem of Llinear:
Can truth out? 191
(7) Gp & Fq ! F(p & q).
And (5) follows by truth-functional logic from (2) and (7), to complete the
proof.
5 CONSEQUENCES OF THE REFORMULATION
The formalization (4.1) has several corollaries worth noting.

Proposition. Let T be as in x4. Then the following are theorems of T:
(1) Pp ! FKPp
(2) p ! FKPp
(3) Fp ! FKPp
(4) Gp ! FKGp.
Proof. First note that each of the following is either an axiom or a

theorem of Llinear:
(5) Pp ! GPp
(6) p ! GPp
(7) Fp ! FGPp
(8) FFp ! Fp
(9) Gp ! GGp.
Also, the following is a derived rule of Llinear:
(10) If A ! B is a theorem, then FA ! FB is a theorem.
(For the cognoscenti, the assumption here is that the rule of temporal
generalization, on which (10) depends, continues to apply after the
formal language has been enriched by the addition of the epistemic
operator K.)
(1), (2), and (4) are immediate from (5), (6), and (9), respectively. As for
(3), it can be derived as follows:
(11) FPp ! FFKPp from (1) by (10)
(12) FPp ! FKPp from (11) and (8).
6 EXAMPLES
To illustrate these corollaries just derived, if Smith has murdered Jones, or

is murdering Jones, or will murder Jones, then according to whichever of
(5.1)–(5.3) is applicable, it will become known that Smith has murdered
Jones. Let us write brackets around present-tense verbs to indicate omni-
temporality, so that, for instance
(1) Smith [murders] Jones.
is to be understood as meaning
(2) Smith has murdered, is murdering, or will murder Jones.
Then we may say that if Smith [murders] Jones, then it will become known
that Smith murdered Jones. And similarly in any other case. Murder
cannot be hid – though (4.1) does not go so far as to join the Bard in
claiming (unfortunately, erroneously) that murder cannot be hid long.
And if the memory of Smith’s victim will never cease to be honored,
then according to (5.4) this fact will become known – though there is
(again, unfortunately) no guarantee it will become known soon enough to
comfort the victim’s grieving friends and relations. And if the universe will
be forever expanding, according to (5.4) this fact, too, will eventually
become known – though there is (yet again, unfortunately) no guarantee
it will become known soon enough to satisfy the curiosity of present-day
cosmologists.
Still, despite its corollaries, (4.1) may look unsatisfactory for the follow-
ing reason. Consider what the corollary (5.2) tells us about a present truth:
(3) If p is true now, then at some later time it will be known that p was true
once.
The ‘‘once’’ here invites the question, ‘‘When?’’ And (5.2) provides no
answer.
Or so it may seem. But in a sense (5.2), taken together with chronome-
try, does provide an answer. If (3.2) and (3.3) are true, then the following
conjunction is true:
(4) It is 12:00 p.m., June 1, 2003, and Smith is murdering Jones.
Applying (5.2) not to (3.3) alone, but to this conjunction, we obtain
It will become known that it was once 12:00 p.m., June 1, 2003, and Smith was
murdering Jones.
or more idiomatically
(5) It will become known that Smith murdered Jones at 12:00 p.m., June 1, 2003.
What more could one want by way of answer to a when-question? Quite
generally, an event occurs at a given time, one can conjoin to a sentence p
Can truth out? 193
asserting the event’s occurrence a sentence q giving the standard chrono-
metric specification of the time, and then apply (5.2), not to p alone, but to
the conjunction.
7 ‘‘ N O W ’’
Nonetheless, it may seem that the most obvious correction of (1.5) would
be the following:
(1) If p is true now, then at some later time it will be known that p was true now.
And (1) seems to tell us more than (4.1) (by way of (6.3)) tells us.
It is known that (1) cannot be expressed using just the temporal oper-
ators G and H and F and P. But tense logicians have considered other
operators. Most to the point in the present context, they have considered a
‘‘now’’ operator J, so interpreted that even within the scope of a past or
future operator Jp still expresses the present, not the past or future, truth of
p. And with this operator (1) can be symbolized, as follows:
(2) p ! FKJp.
One may be tempted to think that (2) would do better as a formalization
of the discovery principle than does (4.1). But this is a misleading way of
putting the issue. For if the operator J is admitted, subject to its usual laws,
then (4.1) implies (2). For one of the usual laws is precisely
(3) p ! GJp
and (2) is immediate from (3) and (4.1). So the temptation here is simply
the temptation to add J to the language.5
I think the temptation should be resisted for a double reason. My first
reason is that introducing the J-operator is unnecessary in order to answer a
when-question. For I have just finished arguing that (4.1) does, after
all, provide answers to such questions. Against this it may be said that
(2) appears to have the advantage of doing so without depending on
chronometry. But my second reason for avoiding the J-operator is that
this apparent advantage comes at the cost of involving us with the problem-
atic notion of a de re attitude towards a time.
This truth is perhaps most easily brought to light by switching tempo-
rarily from regimentations using tense operators to regimentations using
5
I owe this observation to Williamson.
explicit quantification over times. So let t, u, v, . . . range over times. And
let t < u mean that time t is earlier than time u, or equivalently, time u is
later than time t. Let each tensed p be replaced by a one-place p (t) for
‘‘p [is] the case at time t.’’ Every formula A built up from the letters
p, q, r, . . . will similarly be replaced by an open formula A (t). PA and
FA, respectively, will be replaced by
(4a) 9u(u < t & A (u)) (4b) 9u(t < u & A (u)).
In a formula A(t) the parameter t may be thought of as standing for that
time which is now present. Leaving open how to symbolize the epistemic
operator, (5.2) and (2) above go halfway into symbols as follows:
(5) p(t) ! 9u(t < u & it is known at time u that 9v(v < u & p(v)))
(6) p(t) ! 9u(t < u & it is known at time u that p(t)).
There is this difference between the two semi-formalizations, that what

occurs towards the end of (5) can be understood in a de dicto way, thus:
(7) At time u, ‘‘p was true once’’ [is] known to be true.
By contrast, what occurs towards the end of (6) must be understood in a
de re way, thus:
(8) At time u, ‘‘p was true then’’ [is] known to be true of time t.
The symbol-complex KJp in (2) above may be pronounced ‘‘it is known that
p was true now,’’ but what it really amounts to is more like this:
(9) It is known of t that p was true then, where t is that time which is now
present.
8 DE RE ATTITUDES
There are (at least) three major difficulties in making sense of the notion of
a de re knowledge about an object a. Or to put the matter another way,
there is only one obvious strategy for making sense of the notion of a de re
attitude, namely, reduction to a de dicto attitude, and there are (at least)
three major obstacles to this strategy. The strategy is to understand a
subject as knowing of an object a that F(x) holds of it if and only if the
subject knows that F(a) where a is a term denoting a. The three obstacles
or problems relate to the choice of term a.
A first general problem with de re knowledge is that of anonymity. There
may simply be no term a denoting a. This problem has been encountered
Can truth out? 195
in the case of times in x2, and given the restriction on the discovery
principle imposed there, it may be set aside here.
A second general problem with de re knowledge, and one relevant to the
question whether J should be admitted, is the problem of aliases. The
problem is that there may be two terms a and b denoting an object a, and it
may be that the subject knows that F(a) but does not know that F(b), or the
reverse. The star whose common name is ‘‘Aldebaran’’ has also the official
name ‘‘Alpha Tauri.’’ It seems that a subject may have been told by differ-
ent authoritative sources, and hence may know that
(1) Aldebaran is orangish.
(2) Alpha Tauri is the thirteenth brightest star.
and yet, being in ignorance that the two names are names for one and the
same heavenly body, may not know that
(3) Alpha Tauri is orangish.
(4) Aldebaran is the thirteenth brightest star.
And this makes it hard to answer the question whether the subject knows of
the star itself, independently of how it is named, that it is orangish, or the
thirteenth brightest. The existence of aliases is a problem insofar as privi-
leging one of them over the other seems arbitrary.
The same problem can arise for times. Robinson may know that one
rainy day Smith committed murder, and may know that Jones was mur-
dered, and not know that the murder Smith committed was that of Jones.
In this case Robinson will know that
(5) At the time when Smith committed murder, it was rainy.
but not that
(6) At the time when Jones was murdered, it was rainy.
And this makes it hard to answer the question whether Robinson knows of
the time itself, independently of how it is described, that it was rainy then.
Where there exists some standard term for each object of a given kind,
one can always stipulate that a subject is to be credited with de re
knowledge about the object a that F(x) holds of it, if and only if the
subject has de dicto knowledge that F(a) where a is the standard term for
a. Admittedly, such a stipulation may be more a matter of giving a sense
to a kind of locution (ascriptions of de re knowledge) that previously had
none, than of finding out what sense this kind of locution had all along.
Pretty clearly it would be a case of giving rather than finding if
one took as canonical terms for heavenly bodies the official names
adopted by international scientific bodies, preferring ‘‘Alpha Tauri’’
over ‘‘Aldebaran.’’
For times, the obvious candidates for standard terms are those provided
by chronometry. If one is content with (4.1), there is no need to enter into
the problem of de re knowledge about times at all, and so no need to fix on
any standard terms for times. If one adopts (7.2), reliance on chronometry
is the only obvious way to impose a solution on the problem of aliases. But
in that case the one advantage (7.2) appeared to have over (4.1), that of not
depending on chronometry, must be recognized to have been illusory. This
consideration argues, I claim, in favor of the J-free formalization (4.1) and
against the J-laden formalization (7.2).
A third general problem with de re belief is the problem of demonstratives
(and with them indexicals). When the star Alpha Tauri, alias Aldebaran, is
visible in the night sky, one can point to it and say ‘‘that star,’’ and so achieve
reference to it. Now it seems someone looking at the star may well know
(5) That star is orangish.
and yet not knowing the name of the star may well not know either (1)
or (2).
This is, so far, just a special case of the problem of aliases. But demon-
stratives are especially troublesome because, on the one hand, when avail-
able, they seem to provide so direct a way of referring that it is hard to insist
that nonetheless it is some other way of referring that provides the canonical
terms for reduction of de re to de dicto; but on the other hand, demonstra-
tives themselves are not viable candidates for canonical terms, simply
because they are usually not available: if we took demonstratives as canon-
ical terms, most objects would suffer from anonymity most of the time.
Demonstratives act, so to speak, as spoilers, making any other candidates
for the office of canonical term look unworthy, while themselves not being
eligible for that office.
But this problem has been encountered in the case of times in x3, and
given the restriction on the discovery principle imposed there, it may be set
aside here, as the problem of anonymity was set aside. The problem of
aliases, I claim, is enough to make the admission of J undesirable.
9 AN IMPERFECT ANALOGY
I have done with the topic of the discovery principle. But what of the
knowability principle, and the original, modal version of Fitch’s paradox?
Can truth out? 197
Table 10.1
Temporal Modal
Discovery Principle Knowability Principle

present vs future indicative vs subjunctive
tense mood
G, F &, }
Llinear S5
(1.3) p ! FKp p ! àKp (1)
(4.1) Gp ! FKp &p ! àKp (2)
(5.2) p ! FKPp p ! àKàp (3)
now actually
J @
(7.2) p ! FKJp p ! àK@p (4)
times, instants, possibilities, worlds,
moments situations
chronometry ???
I began this essay by recalling that there is a close parallel between temporal
and modal. I should now note that while there are many analogies, in
connection with Fitch’s paradox there is also one glaring disanalogy, that
makes the original, modal problem more refractory than its temporal
analogue. Perhaps the best way of proceeding would be to begin by simply
listing pairs of analogous notions in parallel columns as in Table 10.1.
But returning to what is formally representable, I have recalled in the left
margin in the table the numbers of temporal formulas we have met earlier,
and assigned in the right margin numbers to the analogous modal for-
mulas. Fitch’s (1) is as quickly dismissible as its analogue (1.3).6 The
difficulty comes when one seeks a replacement.
The absence of any obvious analogue for possibilities of standard chro-
nometric specifications for times makes (2) and (3) much less satisfactory
than (4.1) and (5.2) – and (4) correspondingly much more tempting than
(7.2). But the same absence makes the problem of de re knowledge of
possible situations connected with (4) at once more critical and more
difficult to solve or evade than was the problem of de re knowledge of
temporal moments connected with (7.2).
I will not enlarge further here, partly because it would be a good exercise
for readers to work out the analogy for themselves, but mainly because
6
For a full exposition of the essentially grammatical fallacy in the paradox, see Rückert (2004). Rückert
draws on Wehmeier (forthcoming).
I would be largely repeating points that have been made by Dorothy
Edgington in her proposed solution to the paradox, and by Timothy
Williamson in his criticisms thereof.7 A further disanalogy emerges in
discussion of Edgington and Williamson that is not formally representable,
and is therefore not indicated in Table 10.1. It is just this, that generally
speaking the fact that something is only actually true and not necessarily
true tends to matter less to us than the fact that something is only at the
present moment true and not permanently true. Or to put the matter
another way, what will be true when the world is older matters more to us
than what could have been true if the world had been otherwise, since we
hope to live on into ‘‘future worlds’’ but do not expect to transmigrate into
‘‘possible worlds.’’
So far as the present investigation is concerned, it seems that the analogy
between mood and tense takes us only so far, and in the end provides us not
with a solution, but only with a better understanding of just what makes
the problem difficult.
10 COMBINING TEMPORAL AND MODAL FEATURES
Before giving up, however, perhaps we should try the combination of

temporal and modal. That is to say, perhaps instead of considering the
knowability principle as the principle that anything true could have been
known, we should consider it as the principle that anything true could
become known.
The natural setting for such a principle would be a system like Prior’s
logic of ‘‘historical necessity.’’8 In the most elaborate version, which he
calls ‘‘Ockhamist,’’ Prior uses both tense operators G, H, F, P, subject to
the axioms for Llinear, and modal operators &, }, subject to the axioms
of S5. But the modal operators are themselves understood in a tensed
way, as meaning necessity and possibility given the course of history up to
the present.
Prior uses special letters a, b, c, . . . for sentences with the special
property that their truth is independent of the future course of history
in addition to the usual letters p, q, r, . . . for arbitrary sentences. Not all
formulas, but only certain special ones, with the same special property as
7
See Edgington (1985) and Williamson (1987). For further relevant publications see Brogaard and
Salerno (note 2).
8
See Prior (1967b, chapter VII).
Can truth out? 199
the special letters, may be substituted for those special letters. These
include the special letters themselves, any formula beginning with a
modal operator & or }, and any formula obtainable from formulas of
these two kinds using the truth functions and past-tense operators H
and P.
A single axiom links the temporal and modal operators:
(1) a ! &a.
One can obtain by substitution
(2) Pa ! &Pa.
One cannot derive
(3) Fa ! &Fa.
Taking for a in (1)–(3) ‘‘A sea fight is occurring,’’ in Prior’s system one
can conclude that if a sea fight is occurring or has occurred, then the
occurrence of a sea fight is (historically) necessary; but even if a sea-
fight is only going to occur, then its occurrence is (historically) con-
tingent, though once it does occur, it will become (historically)
necessary.
A version of the knowability principle can be expressed in this context by
the formula
(4) &Ga ! }FKa.
And from (4) one can derive, using various tense-logical and modal
theorems, the following rough analogues of the corollaries in the proposi-
tion of x3:
(5) (Pa Ú a Ú Fa) ! }FKPa
(6) Ga ! }FK}Ga.
The details will not be given here, because the system is ultimately
unsatisfactory.
Let me explain how. From (4), by way of its corollaries, one can
conclude the following, wherein I contract ‘‘possibly will’’ to ‘‘may’’:
(7) If Smith is murdering Jones, then it may become known that Smith has
murdered Jones.
(8) If the memory of Smith’s victim will always be honored, then it may become
known that the memory of Smith’s victim may always be honored.
(9) If the universe is always going to be expanding, then it may become known
that the universe may be always going to be expanding.
What one cannot conclude is:
(10) If the memory of Smith’s victim will always be honored, then it may become
known that the memory of Smith’s victim will always be honored.
(11) If the universe is always going to be expanding, then it may become known
that the universe is always going to be expanding.
So (4) seems too weak.
The strengthening of (4) to
(12) Ga ! }FKGa
would provide assurance of (10) and (11), but unfortunately (12) is too
strong. For it would also provide assurance of the absurd
(13) If Smith murdered Jones but will forever escape detection, then it may
become known that Smith murdered Jones but will forever escape detection.
This is Fitch’s paradox, adapted to the present context.
11 VERIFICATIONISM
In sum, Prior’s Ockhamist framework fails to provide a formula that is not

too weak and not too strong, but just right. A glance back at the examples
above will help us localize the difficulty: it is with truths about the actual
future (and more particularly about what will always hold throughout that
actual future). There are philosophers, however, who question the very
meaningfulness of assertions about the actual future.9
And Prior has developed a logic he calls ‘‘Peircean’’ for them. In this logic
one has only the special letters a, b, c, . . ., and only the formulas built up
from them using truth functions, past-tense operators, and four operators
amounting to the combinations &G, }G, &F, }F. Substitution for the
letters a, b, c, . . . is allowed for all formulas so built up. The operators &, },
G, F do not appear separately, apart from the four combinations just
mentioned.
The pertinent feature of this logic in the present context is that it bans as
meaningless the examples that caused trouble in the preceding section, and
(10.4) seems adequate as an expression of the knowability principle for all
such sentences as are still accepted as meaningful.
9
For a recent expression of this view see Belnap and Green (1994).
Can truth out? 201
Banning statements about the actual future is a radical step. Presumably
the friends and relations of Jones know that his memory possibly will always
be honored, and possibly will not always be honored. They know }Ga and
&Ga, where a is
(1) The memory of Smith’s victim is honored.
The Peircean, however, rejects as meaningless
(2) The memory of Smith’s victim will always be honored.
unless ‘‘will’’ is either strengthened to ‘‘necessarily will’’ or weakened to
‘‘possibly will.’’ The Peircean it seems, cannot allow the friends and
relations to hope that (2) is true, or to fear that it is not.10 Likewise,
cosmologists presumably already know that it is possible the universe will
expand forever, and possible that it will not. The Peircean cannot allow
them to wonder if it in actual fact will.
So Peirceanism is a radical doctrine. But then, so is the knowability
principle. The question is, do the two forms of radicalism cohere? If an
adherent of the knowability principle were to embrace Peirceanism,
would the resulting position have any coherent motivation? Or would
embracing Peirceanism be mere ad hoc epicycling, avoiding counter-
examples by declaring them meaningless? This is too large, and too
non-logical, an issue to go into here, but at least a word may be said
about the historical sources of epistemological views like the discovery
and knowability principles on the one hand, and of Peirceanism on the
other.
Belief in the discovery principle, I said at the outset, has traditionally
rested on theological grounds. Belief in the knowability principle has,
by contrast, been mainly an expression of a commitment to a certain
philosophical theory of meaning, verificationism. The radical epistemolo-
gical view that there are no unknowable truths has usually been a conse-
quence of the even more radical semantical view that understanding a
sentence consists in grasping under what conditions it would be known to
be true.
Belief in Peirceanism has had several sources. Prior cites late-medieval
logicians who have held a similar view on theological grounds, but
the more recent proponents of the view seem to base their adherence
on grounds that ultimately are verificationist. Thus combining the
10
This observation is repeated from Burgess (1979d).
knowability principle with Peirceanism could be viewed as combining
two manifestations of an underlying verificationism. Of course, there
are many varieties of verificationism, and it remains to be seen whether a
single variety can cogently motivate both these manifestations simultane-
ously. A key issue will be the verificationist’s attitude towards the reality
of the past.
11
Quinus ab omni naevo vindicatus
1 QUINE’S CRITIQUE
1.1 Quine and his critique

Today there appears to be a widespread impression that W. V. Quine’s
notorious critique of modal logic, based on certain ideas about reference,
has been successfully answered. As one writer put it some years ago: ‘‘His
objections have been dead for a while, even though they have not yet been
completely buried.’’1 What is supposed to have killed off the critique? Some
would cite the development of a new ‘‘possible-worlds’’ model theory for
modal logics in the 1960s; others, the development of new ‘‘direct’’ theories
of reference for names in the 1970s.
These developments do suggest that Quine’s unfriendliness towards
any formal logics but the classical, and indifference towards theories of
reference for any singular terms but variables, were unfortunate. But in
this study I will argue, first, that Quine’s more specific criticisms of
modal logic have not been refuted by either of the developments cited,
and further, that there was much that those who did not share Quine’s
unfortunate attitudes might have learned about modality and about
reference by attention to that critique when it first appeared, so that it
was a misfortune for philosophical logic and philosophy of language that
early reactions to it were as defensive and uncomprehending as they
generally were. Finally, I will suggest that while the lessons of Quine’s
critique have by now in one way or another come to be absorbed by many
specialists, they have by no means been fully absorbed by everyone, and in
this sense there is still something to be learned from Quine’s critique
today.
1
Hintikka (1982, opening paragraph). In context it is clear this is only a description, and not necessarily
an endorsement, of a widespread impression.
203
x3 below will list some lessons from Quine’s critique, after x2 has
examined the early responses to it. Since I will be arguing that most of
these simply missed the point, I should say at the outset that this is easier to
see by hindsight than it was from expositions of the critique available at the
time, and that the early responses were useful insofar as they provoked new
expositions. That there are flaws in Quine’s own presentations is conceded
even by such sympathetic commentators as Dagfinn Føllesdal and Leonard
Linsky, and at least as regards his earliest presentations by Quine himself.2
To remove flaws is the aim of the present x1, and the aim suggested by my
title, which readers familiar with the history of mathematics will recognize
as echoing Saccheri’s Euclides ab omni naevo vindicatus or Euclid Freed from
Every Blemish. Such readers will also recall that though Saccheri’s aim was
to defend Euclid, ironically his work is today remembered as a contribu-
tion to non-Euclidean geometry. While I hope to avoid a similar irony, I do
not hesitate to depart from Quine on occasion, and begin with two
limitations that I think need more explicit emphasis than they get from
Quine.
1.2 Non-trivial de re modality

A first restriction is that Quine’s critique is limited to predicate as opposed
to sentential modal logic, his complaint being that modal predicate logic
resulted from mechanically combining the apparatus of classical predicate
and modal sentential logic, without thinking through philosophical issues
of interpretation.3 Quine does sometimes suggest that engaging in modal
logic would be pointless unless one were eventually going to go beyond the
sentential to the predicate level, so that though his critique deals explicitly
only with predicate modal logic, it is tantamount to a critique of all modal
logic; but the suggestion is not strenuously argued.4
The restriction to predicate logic has two aspects. First, the critique is
limited to de re as opposed to de dicto modality, to modalities within the
scope of quantifiers as opposed to quantifiers within the scope of
2
The most important of Quine’s presentations is ‘‘Reference and modality,’’ in From a Logical Point of
View (Quine 1953/1961/1980). Citations of this twice-revised work here will be by internal section and
paragraph divisions, the same from edition to edition. This work supersedes the earlier Quine
(1947a). For commentary see the editor’s introduction to Linsky (1971a) and Linsky (1971b). See
also Føllesdal (1969) and Føllesdal (1986).
3
A theme in his reviews (Quine 1946, 1947b).
4
See the last paragraph of the third section of ‘‘Reference and modality,’’ ending: ‘‘for if we do not
propose to quantify across the necessity operator, the use of that operator ceases to have any clear
advantage over merely quoting a sentence and saying that it is analytic.’’
Quinus ab omni naevo vindicatus 205
modalities, to modalities applying to open formulas as in 9x&Fx, rather
than modalities applying to closed formulas as in &9xFx. Second, the
critique is limited to non-trivial de re modality. The first point has been
generally understood. Not so the second, which calls for some explanation.
I begin with an analogy. One can contrive systems of sentential modal
logic that admit modalities notationally, but that make every modal for-
mula more or less trivially equivalent to a non-modal formula. It suffices to
add as a further axiom the following, whose converse is already a theorem
in the common systems:
(1) P ! &P.
This corresponds to a definition according to which P holds necessarily just
in case P holds – a definition that could silence any critic who claimed the
notion of necessity to be unclear, but would do so only at the cost of
making the introduction of the modal notation pointless.
Analogously, one can contrive systems of predicate modal logic that
admit de re modalities notationally, but that make every de re formula more
or less trivially equivalent to a de dicto formula. The precise form a trivial-
ization axiom would take depends on whether one is considering monadic
or polyadic predicate logic, and on whether one is admitting or excluding
an existence predicate or an identity predicate or both. In the simplest case
it suffices to add as a further axiom the following, whose converse is already
a theorem in the common systems:
(2) 8x(&Fx ! &8yFy).
This corresponds to the trivializing definition according to which F holds
necessarily of a thing just in case it is necessary that F holds of everything –
a definition that could silence any critic who claimed the notion of de re
modality to be more obscure than that of de dicto modality, but would do
so only at the cost of making the introduction of de re notation pointless.
When Quine complains of the difficulty in defining de re modality, he is
tacitly assuming the trivializing definition above has been rejected; so his
critique is tacitly limited to systems that, like all the common ones, do not
have the trivialization axiom as a theorem. To accept such a system as the
correct system, the one whose theorems give all and only the general laws
necessarily holding in all instances, is to reject the trivialization axiom as
not being such a general law, and hence is to reject the trivializing
definition, which would make it one. Note that Quine’s objection is thus
to the unprovability of something, namely trivialization, not the provability
of anything.
1.3 Strict necessity

A second restriction is that Quine’s critique is limited to what he calls
‘‘strict’’ necessity, identified with analyticity, as opposed to what may be
called ‘‘subjunctive’’ necessity, involved in counterfactuals. For Quine the
former belongs to the same circle of ideas as synonymy and definition, and
the latter to the same circle as similarity and disposition. Quine sometimes
explicitly states this limitation; but he also often suggests that his argu-
ment generalizes to all intensional operators, or at least that there is an
obstacle to making sense of quantification into intensional contexts in
general (which obstacle is insurmountable in the case of quantification
into contexts of strict modality in particular).5 Insofar as I wish to defend
it, I take Quine’s critique to be limited to strict modality, and his
suggestion about generalization to be an attachment to it, not a compo-
nent of it.
In connection with different senses of necessity there is a feature of the
terminology current in the 1940s through 1960s that needs to be explicitly
emphasized, lest one fall into anachronistic misreadings: the tendency to
use interchangeably with each other, as adjectives modifying the noun
‘‘truth,’’ all the expressions in the left-hand column below (and similarly
for the right-hand column).6 Each row merits separate comment.
Necessary Contingent
Linguistic Empirical
A priori A posteriori
Analytic Synthetic
Logical Non-logical
Logical truth and analytic truth. Quine distinguished a narrower notion
of ‘‘logical’’ truth, roughly truth by virtue of syntactic form alone, from a
broader notion of ‘‘analytic’’ truth, roughly truth by virtue of this plus
semantic factors such as definition and synonymy. He notoriously thought
the latter, broader notion unclear, and so had a double objection to the first
of the following formulations:
5
Contrast the opening section of ‘‘Reference and modality,’’ on knowledge and belief contexts, with
the antepenultimate paragraph of the paper, beginning: ‘‘What has been said in these pages relates
only to strict modality . . .’’
6
For a contemporary account deploring such tendencies, Kneale and Kneale (1962, pp. 628ff).
Such tendencies are exemplified by the usage of all the participants in the exchange discussed in x2
below.
(3) It is analytically true that all bachelors are unmarried.
(4) It is logically true that all unmarried men are unmarried.
(30 ) ‘‘All bachelors are unmarried’’ is analytically true.
(40 ) ‘‘All unmarried men are unmarried’’ is logically true.
One objection was to the common feature of (3) and (30 ), involvement with
broadly analytic rather than narrowly logical truth; another, to the com-
mon feature of (3) and (4), treatment of modality as a connective in the
object language applying to sentences, rather than a predicate in the
metalanguage applying to quotations. What is important to understand
is that in his critique of modal logic Quine presses only his objection to the
second feature – a feature presupposed by quantified modal logic, since
quantification into quotation contexts is obvious nonsense – waiving his
objection to the first for the sake of argument. Others of the period shared
neither Quine’s worries about the broad, semantic notion, nor his concern
to distinguish it from the narrow, syntactic notion, and often wrote
‘‘logical’’ when they meant ‘‘analytic.’’
Analytic truth and a priori truth. Quine’s first and foremost target,
Rudolf Carnap, and others of the period, took the distinction between
analytic and synthetic to be central to epistemology because they took it to
coincide with the distinction between a priori and a posteriori. They
recognized not a trichotomy of ‘‘analytic’’ and ‘‘synthetic a priori’’ and ‘‘a
posteriori,’’ but a dichotomy of ‘‘analytic’’ and ‘‘a posteriori.’’
A priori truth and linguistic truth. Quine often complained that others
were sloppy about distinguishing use and mention. If one is sloppy,
quibbles and confusions can result if, as was commonly done, one
uses ‘‘linguistic’’ interchangeably with ‘‘analytic’’ or ‘‘a priori’’ and ‘‘empiri-
cal’’ interchangeably with ‘‘synthetic’’ or ‘‘a posteriori’’ respectively. For
consider:
(5) Planetoids are asteroids.
(6) Ceres is the largest asteroid.
(50 ) In modern English, ‘‘planetoids’’ and ‘‘asteroids’’ refer to the same things.
(60 ) In modern English, ‘‘Ceres’’ and ‘‘the largest asteroid’’ refer to the same thing.
As to (5), discovery that planetoids are asteroids requires (for a fully

competent speaker of modern English) mere reflection, not scientific
investigation. As to (6), discovery that Ceres is the largest asteroid requires
natural-scientific investigation of the kind engaged in by astronomers.
Discovery that (50 ) is the case (understood as about the common language,
not just one’s personal idiolect) requires social-scientific investigation of
the kind engaged in by linguists. Discovery that (60 ) is the case requires
both kinds of scientific investigation. Since linguistics is an empirical
science, using ‘‘linguistic’’ and ‘‘empirical’’ for ‘‘analytic’’ and ‘‘aposteriori’’
can be confusing when dealing with meta-level formulations like (50 ) and
(60 ) rather than object-level formulations like (5) and (6); but such usage
was common.
Linguistic truth and necessary truth. Quine distinguished strict and sub-
junctive modality, but whereas the default assumption today might be that
someone who writes ‘‘necessary’’ sans phrase intends subjunctive necessity,
this was not so for Quine, let alone modal logicians of the period. Originally
the primitive notion of modal logic was ‘‘implication’’ P Q, with ‘‘neces-
sity’’ defined as ØP P; later necessity &P was taken as primitive, with
implication defined as &Ø(P Ù ØQ). But even then, the notion of impli-
cation of primary interest was strict, so that the notion of necessity of
primary interest also had to be, and was often enough explicitly stated to
be, strict. It was commonly assumed, if not that all necessity is linguistic or
semantic or verbal necessity, then at least that the primary notion of
necessity was that of verbal necessity. In reading the older literature, the
default assumption must be that strict necessity is intended when one finds
sans phrase the word ‘‘necessary.’’
1.4 ‘‘Aristotelian essentialism’’

Preliminary restrictions having been enumerated, the critique proper
begins by indicating what would have to be done to make sense of such
notation as 9x&Fx. Given that 9 is to be read in what has always been the
standard way, as an existential quantifier, and that & is to be read in what
was at the time the prevailing way, as a strict modality, the following are
equivalent:
(7a) 9x&Fx holds
(7b) there is some thing such that &Fx holds of it
(7c) there is some thing such that Fx holds necessarily of it
(7d) there is some thing such that Fx holds analytically of it.
The commitment then is to making sense (in a non-trivial way) of the

notion of an open formula or open sentence Fx holding analytically of a
thing.
Now traditional accounts of analytic truth in philosophy texts provide
only an explanation of what it is for a closed sentence to be analytically true,
and do not even purport to provide any explanation of a notion of an open
sentence being analytically true of a thing. (And rigorous analyses of logical
truth in logic texts again supply only a definition of what it is for a closed
formula to be logically true, and do not even purport to supply any
definition of a notion of an open formula being logically true of a thing.)
The notion of analyticity as it stands simply does not apply literally to an
open sentence or formula relative to a thing, and the most one can hope to
do is to extend the traditional notion from de dicto to de re – or to put the
matter the other way round, reduce the notion for de re modality to the
traditional one for de dicto – while remaining faithful to the spirit of strict
modality. This presumably means remaining attached to a conception of
necessity as purely verbal necessity, and confined within the circle of ideas
containing definition and synonymy and the like, not bringing in physical
notions of disposition or similarity, let alone Peripatetic or Scholastic
metaphysical notions of matter and form or potency and act or essence and
accident. Quine expresses pessimism about the prospects for defining de re
modality subject to this restriction by suggesting that quantified modal
logic is committed to ‘‘Aristotelian essentialism.’’
While Quine’s own approach is resolutely informal, there is a technical
result of Terence Parsons that is illuminating here, even though Parsons’
usage of ‘‘commitment to essentialism’’ differs in a potentially confusing
way from Quine’s. Roughly speaking, Parsons shows that though the
common systems are in the sense indicated earlier committed to the failure
of trivialization as a general law, yet no specific instance of such failure is
provable in the common systems even with the addition of any desired
consistent set of de dicto assumptions.7 (On Parsons’ usage the result is
somewhat confusingly stated as saying that though the common systems
are ‘‘committed to essentialism’’ in one, weaker sense, essentially Quine’s,
they are not ‘‘committed to essentialism’’ in another, stronger sense,
Parsons’ own.) This being so, any attempt to make sense of de re strict
modality by reducing it to de dicto faces a dilemma.
On the one hand, if one adopts some general law permitting passage
from de dicto to de re, one will in effect be adding a new general passage law
as an axiom to the common systems. But with any such addition of a new
formal axiom one is already rejecting the common systems as incomplete if
not as incorrect. Worse, there is a threat that the new axiom will yield
trivialization; or worse still, will yield a contradiction. On the other hand, if
one allows passage from de dicto to de re only selectively, one will in effect be
7
For a less rough formulation, see T. Parsons (1969).
adding a new selection principle as an ingredient to the concept of modal-
ity. But with any such addition of a new intuitive ingredient there is a
danger that one will be making one’s conception no longer one of merely
verbal necessity; or worse, that one will be making it arbitrary and
incoherent.
This abstract dilemma is concretely illustrated by Quine’s mathematical
cyclist example, an elaboration of an old example of Mill’s, and his morning
star example, an adaptation of an old example of Frege’s. The only obvious
approach to reducing the application of modal notions to a thing to an
application of modal notions to words, would be to represent or replace a
thing by a word or verbal expression appropriately related to it. In fact,
there are two strategies here, the most obvious one being to take the
expression to be a term referring to the thing, and an only slightly less
obvious one being to take the expression to be a predicate satisfied by the
thing. Hence the need for two examples.
1.5 The mathematical cyclist

One strategy would be to count Fx as holding necessarily of a thing just in
case F is necessarily implied by some predicate(s) P satisfied by the thing.
On the one hand, if we are non-selective about the predicates, this leads
to contradiction with known or plausible non-modal or de re premises,
such as the following:
(8a) It is necessarily the case that all mathematicians are rational.
(8b) It is at best contingently the case that all mathematicians are bipeds.
(8c) It is necessarily the case that all cyclists are bipeds.
(8d) It is at best contingently the case that all cyclists are rational.
(These are plausible at least if we take rationality to mean no more than
capability for verbal thought, and bipedality to mean no more than having
at least two legs, and count mathematicians who have lost limbs as non-
bipeds, and count bicycle-riding circus animals as cyclists.) Non-selective
application of the strategy to (8a–d) yields:
(9a) Any mathematician is necessarily rational.
(9b) Any mathematician is at best contingently a biped.
(9c) Any cyclist is necessarily a biped.
(9d) Any cyclist is at best contingently rational.
Together (9a–d) contradict the known actual existence of persons who are
at once mathematicians and cyclists.
More formally, allowing non-selective application of the strategy amounts
to adopting the following as an axiom, which can be seen to collapse modal
distinctions all by itself:
(10) 8x(Px ! (&Fx $ &8y(Py ! Fy))).
This is the first horn of the dilemma.
On the other hand, the obvious fall-back would be to allow (10) to apply
only selectively, only to certain selected ‘‘canonical’’ predicates. In order
for (10), restricted to canonical predicates, to give an adequate definition of
de re modality, it would suffice for two things to hold. It would suffice to
have first that for each thing there is (or can be introduced) some canonical
predicate it satisfies; and second that for any two canonical predicates A, B
we have:
(11) 9x(Ax Ù Bx) ! &8y(Ay $ By).
This condition would preclude taking both ‘‘x is a mathematician’’ and ‘‘x is
a cyclist,’’ or both Plato’s ‘‘x is a featherless biped’’ and Aristotle’s ‘‘x is a
rational animal,’’ as canonical. But how is one to select what predicates are
admitted as canonical? It seems that making a selection, choosing for
instance between Plato and Aristotle, would require reviving something
like the ancient and medieval notion of ‘‘real definitions’’ as opposed to
‘‘nominal definitions’’; and this is something it seems impossible to square
with regarding the necessity with which we are concerned as simply verbal
necessity.
1.6 The morning star

The second strategy would be to count Fx as holding necessarily of a
thing just in case Ft holds necessarily for some term(s) t referring to the
thing.
On the one hand, if we are non-selective about the terms, applying
the strategy to all terms equally, then whenever two terms s and t refer
to the same thing, Fx holding necessarily of that thing will be equivalent
to Fs holding necessarily and equally to Ft holding necessarily, so that
Fs holding necessarily and Ft holding necessarily will have to be equiva-
lent to each other. But this result leads to inferences from known or
arguably true premises to known or arguably false conclusions, even in
the very simple case where Fx is of the form x ¼ t, since t ¼ t will in all
cases be necessarily true though s ¼ t may in some cases be only contin-
gently true.
For instance, the following are true:
(12) The evening star is the morning star.
(13) Necessarily, the morning star is the morning star.
And the following false:
(14) Necessarily, the evening star is the morning star.
More formally, allowing non-selective application of the strategy
amounts to adopting the following as an axiom:
(15) 8x(x ¼ t ! (&Fx $ &Ft)).
And this can be seen to collapse modal distinctions (at least if enough
apparatus for converting predicates to terms is available). This is the first
horn of the dilemma.
On the other hand, the obvious fall-back would be to allow (15) to
apply only selectively, to certain selected ‘‘canonical’’ terms. In order for
(15), restricted to canonical terms, to give an adequate definition of de re
modality, two things would be required to hold. It would suffice to have
first that for each thing there is (or can be introduced) some canonical
term referring to it; and second that for any two canonical terms a, b
we have:
(16) (a ¼ b) ! & (a ¼ b).
Now the following is a theorem of the common systems:8
(17) (x ¼ y) ! & (x ¼ y).
But (17) involves only variables x, y, . . . , corresponding to pronouns like
‘‘he’’ or ‘‘she’’ in natural language, not constants a, b, . . . or function terms
fc, gc, . . . , corresponding to names like ‘‘Adam’’ and ‘‘Eve’’ or descriptions
like ‘‘the father of Cain’’ and ‘‘the mother of Cain.’’
8
It may be worth digressing to mention that Quine’s one and only contribution to the formal side of
modal logic occurred in connection with this law, though the history does not always emerge clearly
from textbook presentations. The earliest derivations of the law took an old-fashioned approach on
which identity is a defined second-order notion, and on such an approach the derivation was
anything but straightforward. Quine was one of the first to note that on a modern approach with
identity a primitive first-order notion, the derivation becomes trivial, and goes through for all systems
at least as strong as the minimal normal system K. This is alluded to in passing in the penultimate
paragraph of the third section of ‘‘Reference and modality.’’ For the original presentation see (Barcan
1947). For a modern textbook presentation see Hughes and Creswell (1968, p. 190).
So (17) leaves open what terms should be allowed to be substituted for
variables.9
What (16) says is that for the fall-back strategy being contemplated to
work, we must be able to go beyond (17) to the extent of allowing canonical
terms to be substituted for the variables. This condition would preclude
taking both ‘‘the morning star’’ and ‘‘the evening star’’ as canonical. But
owing to the symmetry involved, it would be entirely arbitrary to select
‘‘the morning star’’ as canonical and reject ‘‘the evening star’’ as apocryphal
(or the reverse), and it would seem almost equally arbitrary to reject both
and select some other term such as ‘‘the second planet.’’ This is the second
horn of the dilemma.
And with this observation Quine rests his case, in effect claiming that
since the obvious strategies for doing what needs to be done have been tried
and found to fail, the burden of proof is now on the other side to show, if
they can, just how, in some unobvious way, what needs to be done can be.
And with this observation, I too rest my case for the moment.
1.7 Coda
Quine’s critique was directed toward the strict kind of modality and
toward quantification over ordinary sorts of objects: persons, places,
things. Much of his discussion generalizes to other kinds of modal or
intensional operators and other sorts of objects, to show that for them,
too, the most obvious strategy for making sense of quantifying over such
objects into such modal or intensional contexts faces an obstacle. But
whether this obstacle can be surmounted, by the most obvious fall-back
strategy of identifying an appropriate class of canonical terms or in some
other way, needs to be considered case by case. The most important case of
a non-strict modality for which a reasonable choice of canonical terms
seems to be available (for almost any sort of objects) will be mentioned at
the very end of this study. Here I want to mention a case of a special sort of
object for which a reasonable choice of canonical terms seems to be
available (for almost any kind of intensional operator).
For several writers, beginning with Diana Ackerman, have pointed out
that numerals suggest themselves as non-arbitrary candidates for canonical
9
In the original paper where (17) was derived there were no singular terms but variables, and nothing
was said about application to natural language. For an idea of the range of options formally available,
see the taxonomy in Garson (1984).
terms if one is going to be quantifying only over natural numbers. And the
numerals are in effect taken as canonical terms in two flourishing enter-
prises, intensional mathematics and provability logic, where the modality in
question is a version or variant of strict modality.10
Still, natural numbers are a very special sort of object. Workers in the
cited fields have noted the difficulty of finding canonical terms as soon as
one goes beyond them even just to other sorts of mathematical objects,
such as sets or functions. To avoid difficulties over there simply being too
many objects to find terms for them all, let us restrict attention to recur-
sively enumerable sets of natural numbers and recursive partial functions
on natural numbers, where there is actually a standard way of indexing the
objects in question by natural numbers or the numerals therefor. Even here
there does not seem to be any non-arbitrary way of selecting canonical
terms, since there will be many indices for any one set or function, and two
indices for the same object will not in general be provably indices for the
same object.11
Whatever successes have been or may be obtained for non-strict modal-
ities and ordinary objects, or for strict modalities and non-ordinary objects,
they only make it the more conspicuous how far we are from having any
reasonable candidates for canonical terms in the case to which Quine’s
critique is directed.
2 QUINE’S CRITICS
2.1 Quine and his critics

Today when one thinks of model theory for modal logic, or the application
of theories of reference to it, one thinks first of Saul Kripke, whose relevant
work on the former topic only became widely known after his presentation
at a famous 1962 Helsinki conference,12 and on the latter only after his
10
See Ackermann (1978). Lectures of Kripke have brought this formerly under-appreciated paper
to the attention of a wider audience. See also Shapiro (1985) and especially Boolos (1993, pp. xxxiv
and 226).
11
Workers in the cited fields have in effect suggested that something like indices can serve as canonical
terms for more fine-grained intensional analogues of recursive sets and functions. But these too
would be very special objects. The best discussion of these matters known to me is in some work not
fully published of Leon Horsten.
12
Whose published proceedings make up a memorable issue of Acta Philosophica Fennica, and include
not only Kripke (1963) but Hintikka (1963).
celebrated 1970 Princeton lectures.13 But the impression that somehow an
appropriate theory of models or of reference can refute Quine’s critique can
be traced back a full half-century. For less sophisticated model theories for
quantified modal logic go back to some of the first publications on the
subject, by Rudolf Carnap, in the 1940s;14 and the application of less
sophisticated theories of reference to modal logic goes back to one of the
first reviews of Quine’s critical writings, by Arthur Smullyan, again in the
1940s.15
For purposes of examining the main lines of response to Quine’s
critique prior to the new developments in model theory and the theory
of reference in the 1960s and 1970s, and Quine’s rebuttals to these
responses, it is almost sufficient to consider just three documents,
together constituting the proceedings of a notorious 1962 Boston collo-
quium. The main talk, by Quine’s most vehement and vociferous oppo-
nent, Ruth (Barcan) Marcus, was a compendium of almost all the
responses to Quine that had been advanced over the preceding fifteen
years, plus one new one. The commentary, by Quine himself, marked an
exception to his apparent general policy of not replying directly to critics,
and gives his rebuttal to almost all early objections to his critique. An
edited transcript of a tape recording of a discussion after the two talks
among the two invited speakers and some members of their audience,
notably Kripke, was published along with the two papers, and clarifies
some points.16
2.2 Potpourri
A half-dozen early lines of response to the critique may be distinguished.
Most appear with differing degrees of explicitness and emphasis in the
13
Kripke (1972/1980). 14 Carnap (1946, 1947).
15
Smullyan (1947), with elaboration in Smullyan (1948). Smullyan’s priority for his particular response
to Quine has been recognized by all competent and responsible commentators. See Linsky (1971b,
note 15) and Føllesdal (1969, p. 183).
16
Thus the items are: (i) the compendium (Marcus 1963a); (ii) the comments (Quine 1963) later
retitled ‘‘Reply to Professor Marcus’’; and (iii) the edited discussion (Marcus, Quine, Kripke et al.
1963).They appear together in the official proceedings volume (Wartofsky 1963). The same publisher
had printed them in 1962 in Synthese in a version that is textually virtually identical down to the
placement of page breaks, (i) and (ii) in a belated issue of the volume for 1961, and (iii) in an issue of
the volume for 1962. (There have been several later, separate reprintings of the different items, but
these incorporate revisions, often substantial.) Two of the present editors of Synthese, J. Fetzer and
P. Humphreys, have proposed publishing the unedited, verbatim transcript of the discussion, with a
view to shedding light on some disputed issues of interpretation; but according to their account,
one of the participants, Professor Marcus, has objected to circulation of copies of the transcript or
the tape.
compendium, and most are rebutted in the commentary thereupon. They
all involve essentially the same error, confusing Quine’s philosophical
complaint with some formal claim. Since – despite the best efforts of
Quine himself in his rebuttal and of subsequent commentators – such
confusions are still common, it may be in order to review each response and
rebuttal briefly.
(A) The development of possible-worlds semantics shows that there is no
problem of interpreting quantified modal logic. This response is represented
in the compendium by the suggestion that disputes about quantified
modal logic should be conducted with reference to a ‘‘semantic construc-
tion,’’ in which connection the now superseded approach of Carnap is
expounded (with the now standard, then unpublished, approach of Kripke
being alluded to as an alternative in the discussion). Perhaps Quine
thought the fallacy in this response obvious, since he makes no explicit
response to it in his commentary; but it has proved very influential, albeit
perhaps more as an inchoate feeling than as an articulate thought. The
fallacy is one of equivocation, confusing ‘‘semantics’’ in the sense of a
mathematical theory of models, such as Carnap and Kripke provided,
with ‘‘semantics’’ in the sense of a philosophical account of meaning,
which is what Quine was demanding, and thus neglecting the dictum that
‘‘there is no mathematical substitute for philosophy.’’17 A mathematical
theory of models could refute a technical claim to the effect that the common
systems are formally inconsistent, but without some further gloss it cannot
say anything against a philosophical claim that the common systems are
intuitively unintelligible. In the case of Carnapian model theory this point
perhaps ought to have been obvious from the specifics of the model, which
validates some highly dubious theses.18 In the case of Kripkean model theory
the point perhaps ought to be obvious from the generality of the theory,
from its ability to accommodate the widest and wildest variety of systems,
which surely cannot all make good philosophical sense.
17
These are the closing words of Kripke (1976). The fallacy recurs again and again in other contexts in
the literature. See Copeland (1979).
18
Notably the Barcan or Carnap–Barcan formulas, which give formal expression to F. P. Ramsey’s odd
idea that whatever possibly exists actually exists, and whatever actually exists necessarily exists. (The
‘‘Barcan’’ label is the more customary, the ‘‘Carnap–Barcan’’ label the more historically accurate
according to Cocchiarella (1984), which also explains the connection with Ramsey.) If these formulas
are rejected, one must distinguish a thing’s having a property necessarily (for every possible world it
exists there and has the property there) from its having the property essentially (for every possible
world, if it exists there, then it has the property there). I have slurred over this distinction so far, and
will for the most part continue to do so.
(B) Quantified modal logic makes reasonable sense if 8 and 9 are read as
something other than
P ordinary quantifiers, such as Lesniewski-style substitution
operators and . This is the one substantial novelty in the compendium.
One rebuttal, of secondary importance to Quine, is that if one allows
oneself to call substitution operators ‘‘quantifiers,’’ one can make equally
good or poor sense of ‘‘quantification’’ not only into modal but into
absolutely any contexts whatsoever, including those of quotation. But
quantification into quotation contexts is obvious nonsense – on any
reasonable understanding of ‘‘quantification.’’19 Still, the rebuttal of pri-
mary importance to Quine is a different and more general one, applying
also to the next response.
(C) Quantified modal logic makes reasonable sense if & and } are read as
something other than strict modalities, such as Prior-style temporal operators
G and F. This response is represented in the compendium by the sugges-
tion, made in passing in the introduction, that modal logic is worth
pursuing because of the value of studies of various non-alethic ‘‘modal-
ities.’’ The specific example of temporal ‘‘modalities’’ was suggested by
Quine in his last remarks in the discussion, his purpose being to bring out
his primary point of rebuttal to the previous response, that Lesniewski’s
devices are just as irrelevant as Prior’s devices, given the nature of his
complaint. If his complaint had been that there is a formal inconsistency
in the common systems, then it would have been cogent to respond by
considering those systems as wholly uninterpreted notations, and looking
for some reading of their symbolism under which they would come out
saying something true or plausible. But the nature of the critique is quite
different, the complaint being that the combination 9x& is philosoph-
ically unintelligible when the components 9 and & are interpreted in the
usual way.20
(D) Quantified modal logic is not committed to essentialism because no
formula expressing such a commitment (no instance of the negation of (2))
is deducible in the common systems, even with the addition of any desired set
of consistent de dicto axioms. This response does not explicitly occur as
19
As shown by examples in the opening section of ‘‘Reference and modality.’’ This point seems to be
conceded even by some who otherwise take an uncritically positive view of the compendium, as in
the review Forbes (1995). The last sections of Kripke (1976) in effect point out that the claim that the
ordinary language ‘‘there is’’ in its typical uses is a ‘‘substitutional quantifier’’ devoid of ‘‘ontological
commitment’’ is absurd, since ‘‘ontological commitment’’ is by definition whatever it is that the
ordinary language ‘‘there is’’ in its typical uses conveys.
20
‘‘What I’ve been talking about is quantification, in a quantificational sense of quantification, into
modal contexts, in a modal sense of modality’’ (Wartofsky 1963, p. 116).
such in the compendium, and would have been premature, since the
results of Parsons which it quotes did not come until a few years later. But
it is advanced in a slightly later work of the same author, and has been
influential in the literature.21 It could be construed as merely a general-
ization of the next response on the list, and Quine’s rebuttal to the next
response would apply to this one, too. Basically, the response is the result
of terminological confusion, since its first clause is only relevant if
‘‘commitment to essentialism’’ is understood in Quine’s sense, but its
second clause is only true if ‘‘commitment to essentialism’’ is understood
in a different sense partly foreshadowed in the compendium and expli-
citly introduced as such by Parsons. It has already been noted in the
exposition of the critique both that Quine’s complaint is not about the
provability of anything, and that Parsons’ results substantiate some of
Quine’s suspicions.
(E) The mathematical cyclist example does not show there is any
problem, because no de re conclusions of the kind that figure in the
example (conclusions (9a–d)) provably follow in the common systems
from such de dicto premises as figure in the example (premises (8a–d)).
While the example gives a legitimate counter-instance to the law that
figures in it (law (10)), that law is not a theorem in the common systems.
This response occurs in a section of the compendium where Quine’s
criticisms are said to ‘‘stem from confusion about what is or is not
provable in such systems,’’ and where it is even suggested that Quine
believes & (P!Q)!(P!&Q) to be a theorem of the common sys-
tems!22 This response, which accuses Quine of committing a howler of
a modal fallacy, is itself a howler, getting the point of Quine’s example
exactly backwards. The complaint that we cannot deduce examples of
non-trivial de re modality from plausible examples of non-trivial
de dicto modality by taking something like (10) as an axiom, because
we would get a contradiction, is misunderstood as a formal claim that
something like (10) is an axiom, and we do get a contradiction. Quine’s
rebuttal in his commentary borders on indignation: ‘‘I’ve never said or,
21
Marcus (1967). And about the same time we find even the usually acute Linsky (1971a, p. 9) writing:
‘‘Terence Parsons bases his search for the essentialist commitments of modal logic on Kripke’s
semantics, and he comes up (happily) empty-handed . . . He finds modal logic uncontaminated.’’
The continuation of this passage better agrees with Parsons’ own account of his work and its bearing
on Quine’s critique.
22
See Wartofsky (1963, pp. 90–2). It is just conceivable that this is deliberate exaggeration for effect, a
rhetorical flourish rather than a serious exegetical hypothesis. Marcus (1967) cites some other authors
who have written in a similar vein about the example.
I’m sure, written that essentialism could be proved in any system of
modal logic whatsoever.’’23
(F) The morning star example does not show there is any problem, because
while the law that figures in the example (law (17)) is a theorem of the common
systems, the example does not give a legitimate counter-instance, as can be seen
by applying an appropriate theory of reference. This response is repeated, with
elaboration but without expected acknowledgments – it is described as
‘‘familiar,’’ but no specific citation is given – in the compendium. The
citation ought to have been to Smullyan.24 This response again mistakenly
takes Quine to be claiming to have a counterexample to a formal theorem
of the common systems. (And if Quine had claimed that (12) and (14)
constitute a counterexample to (17), it would have sufficed to point out that
one is not required, just because one recognizes an expression to be a real
singular term, to recognize it as legitimately substitutable for variables
in all contexts. This point has been noted already in the exposition of the
critique, but the response under discussion seems to miss it.) Nonetheless,
response (F) is worthy of more extended attention.
2.3 Smullyanism or neo-Russellianism

While responses (A)–(E) are entirely skew to Quine’s line of argument,
response (F) (when fully articulated) makes tangential contact with it, and
shows that a minor addition or amendment to the critique as expounded so
far is called for. Another reason response (F) calls for more attention than
the others is that for a couple of decades it was the conventional wisdom
among modal logicians. It was endorsed not only by (in chronological
order) Smullyan, Fitch, and Marcus, but also by Arthur Prior and others. It
was the topic of two talks at the famous 1962 Helsinki conference and was
put forward in major and minor encyclopedias.25 Yet another reason
23
And ‘‘I did not say that it could ever be deduced in the S-systems or any systems I’ve ever seen’’
(Wartofsky 1963, p. 113). Despite these forceful remarks, the understanding of Quine’s views has not
much improved in the later Marcus (1967).
24
An earlier paper by the author of the compendium (Marcus 1960) gives a more concise statement of
the response in its last paragraph, where a footnote acknowledges the author’s teacher Frederic Fitch.
The latter, in Fitch (1949) and Fitch (1950), acknowledges Smullyan. (See footnote 4 in the former,
footnote 12 in the latter, and the text to which they are attached.)
25
The major one being Weiss (1967), and the minor one the collection of survey articles (Klibansky
1968). The former contains Prior (1967a) while the latter contains Marcus (1968). The conference
talks (Marcus 1963b, Prior 1963) are to be found in the previously cited Helsinki proceedings.
Another advocate of closely related ideas has been J. Myhill.
response (F) calls for more attention than the others is that it represents an
early attempt to apply a theory of reference distinguishing names from
descriptions to the interpretation of modal logic, and understanding why
this attempt was unsatisfactory should lead to increased appreciation of
more successful later attempts.
The ideas on reference that are involved derive from Russell. The
writings of Ramsey, alluded to in passing in the compendium, and of
Carnap, with whom the author of the compendium at one time studied,
may have served to transmit Russell’s influence, though of course Russell
himself was still writing on reference in the 1950s, and still living in the
1960s, and should not be considered a remote historical figure like Locke or
Mill. But whether his influence on them was direct or indirect, Smullyan’s
disciples are unmistakably Russell’s epigones, even though they seldom
directly quote him or cite chapter and verse from his writings.26
The Smullyanite response, it will be seen, splits into two parts, one
pertaining to descriptions, the other to names. The theory of descriptions
presupposed by the Smullyanites is simply the very well-known theory of
Russell. The theory of names presupposed is the less well-known theory
Russell always took as a foil to his theory of descriptions. This is perhaps
best introduced by contrasting it with the theory of Frege, according to
which the reference of a name to its bearer is descriptively mediated, is
accomplished by the name having the same meaning as some description,
and the description being uniquely true of the bearer. The theory of Russell
is the diametrically opposed one that the reference of a name to its bearer is
absolutely immediate, in a sense implying that the meaning of a name is
simply its bearer, from which it follows that two names having the same
bearer have the same meaning. It is taken to follow (‘‘compositionality’’
being tacitly assumed) that two sentences involving two different names
with the same bearer, but otherwise the same, have the same meaning, and
hence the same truth value (with one sole exception, usually left tacit, the
exception for meta-linguistic contexts, for those sentences, usually involv-
ing quotation, where the names are being mentioned as words rather than
being used to refer).
This theory is Russell’s account of how names in an ideal sense
would function. While Russell illustrated his theory by examples involv-
ing names in the ordinary sense, he actually more or less agreed with
26
Let me not fail to cite chapter and verse myself. For the most relevant pages of the most recently
reprinted work, see Russell (1985), pp. 113–15.
Frege about these (so that the Fregean theory is often known as the
Frege–Russell theory). Moreover, he held that ordinary, complex things
are not even capable of being given names in his ideal sense; that names in
the ideal sense could be given only to special, simple things (such as sense
data). There is an ambiguity running through the writings of all the
Smullyanites as to whether they do or do not wish to claim that names in
the ordinary sense function as names in the ideal sense. But they do
unambiguously wish to claim, contrary to Russell, that whether or not
they are already in existence, names in an ideal sense can at least be
introduced for ordinary things. For this reason, while the Smullyanites
may be called ‘‘Russellians,’’ it is perhaps better to add the distinguishing
prefix ‘‘neo-.’’
So much for the background assumptions of response (F). Its further
articulation has several components:
(F0) Quine’s example is ambiguous, since the key terms ‘‘the morning star’’ and ‘‘the
evening star’’ might be either mere definite descriptions or genuine proper names.
(F1a) If the key phrases are taken to be descriptions, then they are only apparently and
not really singular terms, and (12) is only apparently and not really a singular
identity, so one gets only an apparent and not a real counterexample to (17).
(F1b) Moreover, though the foregoing already suffices, it may be added that (13) and
(14) are ambiguous, and it is not unambiguously the case that they are of opposite
truth value, the former true and the latter false, as the example claims.
(F2) If the key phrases are taken to be names, then (14) means the very same thing as,
and is every bit as true as, (13), contrary to what the example claims.
To dispose of the issue (F0) of ambiguity, the example may be restated
twice:
(12a) Hesperus is Phosphorus.
(13a) Necessarily, Phosphorus is Phosphorus.
(14a) Necessarily, Hesperus is Phosphorus.
(12b) The brightest star of the evening is the brightest star of the morning.
(13b) Necessarily, the brightest star of the morning is the brightest star of the morning.
(14b) Necessarily, the brightest star of the evening is the brightest star of the morning.
2.4 Quine’s rebuttal to neo-Russellianism on descriptions

The main claim (F1a) of the descriptions side of the Smullyanite response is
immediate from Russell’s theory, on which (12b) really abbreviates some-
thing more complex involving quantifiers:
(12c) There exists a unique brightest star of the evening and
there exists a unique brightest star of the morning, and
whatever is the brightest star of the evening and
whatever is the brightest star of the morning,
the former is the same as the latter.
The subsidiary claim (F1b) is also almost immediate, since on Russell’s
theory in all but the simplest cases expressions involving descriptions
involve ambiguities of ‘‘scope,’’ and for instance there is one disambigua-
tion of (14b) that follows by (17) from (12c):
(14c) There exists a unique brightest star of the evening and
there exists a unique brightest star of the morning, and
whatever is the brightest star of the evening and
whatever is the brightest star of the morning,
necessarily the former is the same as the latter.
In rebuttal to all this, the main point is that the example was not
intended as a counter-instance to (17) or any other theorem of the common
systems, but as an illustration of an obstacle to reducing de re to de dicto
modality, so that response (F1a) is wholly irrelevant.
Response (F1b) is partly relevant, however, because it does show that the
example needs to be worded more carefully if the Russellian theory of
descriptions is assumed. The strategy against which the example was
directed was that of defining &Fx to hold of a thing if and only if &Ft
holds where t is a term referring to that thing. But assuming the Russellian
theory of descriptions, there is actually more than one strategy here (when t
is a description) because &Ft is ambiguous between a ‘‘narrow’’ or a ‘‘wide’’
reading.
Also the predicate &Fx used in the example, ‘‘Necessarily x is the
brightest star of the morning’’ is similarly ambiguous. To eliminate this
last ambiguity, take the predicate to be something like ‘‘Necessarily, (if x
exists then) x is the brightest star of the morning.’’ Then on the narrow-
scope reading &Ft and &Fs boil down to:
(13c) Necessarily, if there exists a unique brightest star of the morning then it is the
brightest star of the morning.
(14c) Necessarily, if there exists a unique brightest star of the evening then
it is the brightest star of the morning.
So in this case the reduction strategy fails for the reason originally given,
since (13c) and (14c) are of opposite truth value, the former being true and
the latter false. But on the wide-scope reading &Ft and &Fs boil down
instead to:
(13d) There exists a unique brightest star of the morning and
necessarily, (if it exists then) it is the brightest star of the morning.
(14d) There exists a unique brightest star of the evening and
necessarily, (if it exists then) it is the brightest star of the morning.
In this case the reduction strategy fails for a more basic reason, since (13d)
and (14d) themselves still involve unreduced de re modalities. The claim
that the strategy breaks down thus does not have to be retracted, though the
explanation why it does so needs to be reworded.
Response (F1b) is almost the only significant response to Quine in the
early literature not reproduced in the compendium, and for Quine’s own
statement of a rebuttal to it we need to look beyond his commentary at the
colloquium. We find the following formulation, where ‘‘non-substitutive
position’’ means a position, such as that of x in &Fx, where different terms
referring to the same thing are not freely intersubstitutable:
[W]hat answer is there to Smullyan? Notice to begin with that if we are to bring
out Russell’s distinction of scopes we must make two contrasting applications of
Russell’s contextual definition of description [as in the (c) versions versus the (d)
versions]. But, when the description is in a non-substitutive position, one of the
two contrasting applications of the contextual definition [namely, the (d) versions]
is going to require quantifying into a non-substitutive position. So the appeal to
scopes of descriptions does not justify such quantification, it just begs the
question.27
2.5 Neo-Russellianism on names

Response (F2) is immediate assuming the neo-Russellian theory of names.
Indeed, what neo-Russellianism assumes about names is more than enough
to guarantee that they would have all the properties required of canonical
terms.28 Thus whereas in rebuttal to (F1) Quine did not have to reject
Russell’s theory of descriptions, he does have to reject the neo-Russellian
theory of names.
Response (F2) is so immediate assuming the neo-Russellian theory that
it is stated without elaboration by Smullyan and his early disciple Fitch
27
Reply to Sellars, in Davidson and Hintikka (1969, p. 338). This formulation is the earliest adequate
one known to me, the rebuttal even in the 1961 version of ‘‘Reference and modality’’ being
inadequate.
28
As was pointed out in Kripke’s last few remarks in the discussion at the colloquium. Quine
seems to accept the observation in his last remark. Marcus had apparently ceased to follow by this
point.
as if it were supposed to be self-evident.29 Elaboration is provided by
later disciples in the compendium and elsewhere. The elaboration in
Prior’s talk at the 1962 Helsinki conference is of especial interest because
it anticipates in a partial way a significant later contribution to the theory of
reference.
Since this has not hitherto been widely noted, I digress to quote the
relevant passage:
It is not necessary, I think, for philosophers to argue very desperately about what is
in fact ‘‘ordinary’’ and what is not; but let us say that a name in Russell’s strict sense is
a simple identifier of an object . . .
[T]here is no reason why the same expression, whether it be a single word like
‘‘This’’ or ‘‘Tully,’’ or a phrase like ‘‘The man who lives next door’’ or ‘‘The man at
whom I am pointing,’’ should not be used sometimes as a name in Russell’s strict
sense and sometimes not. If ‘‘The man who lives next door’’ is being so used, and
successfully identifies a subject of discourse, then ‘‘The man who lives next door is
a heavy smoker’’ would be true if and only if the subject thus identified is a heavy
smoker, even if this subject is in a fact a women and doesn’t live next door but only
works there. And if ‘‘Tully,’’ ‘‘Cicero,’’ ‘‘The Morning Star’’ and ‘‘The Evening
Star’’ are all being so used, then ‘‘Tully is Cicero’’ and ‘‘The Morning Star is the
Evening Star’’ both express necessary truths, to the effect that a certain object is
identical with itself.30
The distinctive part of the passage, not in the founder or other members of
the Smullyanite school, is the middle, where it is suggested that even an
expression that is not a name in the ordinary sense may sometimes function
as a name. This is a different point from the trivial observation that names
often have descriptive etymologies, and those familiar with the later liter-
ature will recognize how what is said about ‘‘the man who lives next door’’
partially anticipates what was later to be said about ‘‘referential’’ as opposed
to ‘‘attributive’’ uses of descriptions.
2.6 Quine’s rebuttal

The elaboration in Marcus’s talk at the same conference, a kind of sequel to
the compendium, is of especial interest because it makes more explicit than
any other published Smullyanite work the implication that was to be
29
Fitch (1949) explicitly claims that Quine’s contention is ‘‘clearly’’ false if the key expressions are taken
to be names.
30
Prior (1963, pp. 194–5). Prior was from Balliol, and I have heard it asserted – though I cannot
confirm it from my own knowledge – that there was a tradition of setting examples of this kind in
undergraduate examinations at Oxford in the 1960s.
most emphatically rejected by later work in the theory of reference: the
epistemological implication that discoveries like (14a) are not ‘‘empirical’’ (at
least not in a non-quibbling sense), and are not properly astronomical
discoveries:
[T]o discover that we have alternative proper names for the same object we turn to
a lexicon, or, in the case of a formal language, to the meaning postulates.. . . [O]ne
doesn’t investigate the planets, but the accompanying lexicon.31
The same thought had been expressed in slightly different words – ‘‘dic-
tionary’’ for ‘‘lexicon,’’ for instance – in the discussion at the colloquium.32
The picture underlying such remarks had been sketched in the compen-
dium itself:
For suppose we took an inventory of all the entities countenanced as things by
some particular culture through its own language . . . And suppose we randomized
as many whole numbers as we needed for a one-to-one correspondence, and
thereby tagged each thing. This identifying tag is a proper name of the thing.33
To talk of an ‘‘inventory,’’ and especially to presuppose that we know how
many numbers would be ‘‘needed for a one-to-one correspondence,’’ is to
assume that we are dealing with a known number of unproblematically
identifiable items. If it is a matter of applying tags to such items, then of
course we should be able to keep a record of when we have assigned
multiple tags to a single one of them, though our record would perhaps
more colloquially be called a ‘‘catalogue’’ than an ‘‘accompanying lexicon’’
or set of ‘‘meaning postulates.’’
The rebuttal to the Smullyanites on names consists in observing that
what is said in the last few quotations is false. Take first Prior. If one defines
‘‘names in the strict sense’’ as expressions with the magical property of
presenting their bearers so absolutely immediately as to leave no room for
empirical questions of identity, then there never have been in any histor-
ically actual language and never can be in any humanly possible language
any such things as ‘‘names in the strict sense.’’ As Russell himself noted,
even ‘‘this is the same as this,’’ where one points to the same object twice, is
31
Marcus (1963b, p. 132). Note the characteristically Carnapian expression ‘‘meaning postulates.’’
32
For the published version, too familiar to bear quoting again, see Wartofsky (1963, p. 115). This is one
of the parts of the discussion where comparison with the verbatim transcript could be most
illuminating. It is a shame that the scholarly public should be denied access to so significant a
historical document.
33
Wartofsky (1963, pp. 83–4). This passage has sometimes been misleadingly cited in the later literature
as if it were unambiguously about ordinary names in ordinary language.
not a linguistic and non-empirical truth, if the object in question is
complex, and one points to a different component each time.
Take now the compendium and its sequel. Assigning names to heavenly
bodies may be like tagging, but it is not like tagging individuals from among
a known number of unproblematically identifiable items, since we always
have unresolved questions before us about the identity of asteroids or
comets, as Frege long ago noted. And to resolve such questions one must
investigate not some ‘‘accompanying lexicon’’ or ‘‘meaning postulates,’’ but
the planet(oid)s themselves.
In brief, the following have the same status as (6) and (60 ) respectively,
and not as (5) and (50 ):
(12a) Hesperus is Phosphorus.
(12a0 ) In modern English, ‘‘Hesperus’’ and ‘‘Phosphorus’’ refer to the same thing.
Quine’s own formulation of this rebuttal is almost too well known to
bear quotation. But while what Quine means is what I have just said, what
Quine says may be open to quibbles, since taken with pedantic literalness it
would seem to be about (12a0 ) rather than (12a):
We may tag the planet Venus, some fine evening, with the proper name
‘‘Hesperus.’’ We may tag the same planet again, some day before sunrise, with
the proper name ‘‘Phosphorus.’’ When at last we discover that we have tagged the
same planet twice, our discovery is empirical. And not because the proper names
were descriptions.34
3 QUINE’S LESSONS
3.1 Hints from Quine for the formal logic of modalities

With the wisdom of hindsight it can be seen that there are several impor-
tant lessons about modality and reference directly taught or indirectly
hinted in Quine’s critique. For modal logic, the first lesson from Quine
is that strict or (as many have called it) ‘‘logical’’ modality and subjunctive
or (as we now call it) ‘‘metaphysical’’ modality are distinct. A further lesson
is that quantification into contexts of strict modality is difficult or
34
Wartofsky (1963, p. 101). Quine surely means that (12a0 ) is not just a linguistic empirical discovery but
a properly astronomical empirical discovery. By contrast, Marcus in Wartofsky (1963, p. 115),
distinguishes ‘‘such linguistic’’ inquiry as leads to discoveries like (12a0 ) from ‘‘properly empirical’’
methods such as lead to discoveries about orbits.
impossible to make sense of. A yet further lesson is that quantification into
contexts of subjunctive modality is virtually indispensable.
This last lesson is not as explicitly or emphatically taught as the other two,
and moreover Quine’s remarks are flawed by a tendency to conflate sub-
junctive or ‘‘metaphysical’’ modality with scientific or ‘‘physical’’ modality – as
if we could not speak in the subjunctive of counterfactual hypotheses to the
effect that the laws of science or physics were violated. But due allowance
being made for this flaw, I believe that the work of Quine, supplemented by
that of his student Føllesdal, gives a broad hint pointing in the right direction.
Føllesdal’s treatment of the topic begins by quoting and stressing the
importance of some of Quine’s remarks about the question of the mean-
ingfulness of quantification into contexts of subjunctive modality:
It concerns . . . the practical use of language. It concerns, for example, the use of
the contrary-to-fact conditional within a quantification . . . Upon the contrary-to-
fact conditional depends in turn, for instance, this definition of solubility in water:
To say that an object is soluble in water is to say that it would dissolve if it were in
water. In discussions in physics, naturally, we need quantifications containing the
clause ‘‘x is soluble in water.’’35
Such passages stop just short of saying, what I think is true, that while
quantification into contexts of strict modality may be nonsense, quantifi-
cation into contexts of subjunctive modality is so widespread in scientific
theory and commonsense thought that we could not abandon it as non-
sensical even if we wanted to.
Putting the lessons cited together, it follows that there is a difference between
strict and subjunctive modality as to what expressions should be accepted as
meaningful formulas and so a fortiori as to what formulas should be accepted as
correct laws. The strictly or ‘‘logically’’ possible, what it is not self-contradictory
to say actually is, and the subjunctively or ‘‘metaphysically’’ possible, what could
potentially have been, differ in the formalism appropriate to each.
3.2 A hint from Quine for the theory of reference of names

The article on modal logic in the minor encyclopedia alluded to earlier
devotes a section to objections, of which the very first (3.1) is Quine’s
morning star example. In the next section the following is said:
35
The quotation from Quine is from ‘‘Reference and modality,’’ antepenultimate paragraph. The work
of Føllesdal where it is quoted is Føllesdal (1965). Føllesdal’s final footnote suggests that ‘‘causal
essentialism’’ is better off than ‘‘logical essentialism,’’ and that Quine’s own proposal to treat
dispositions as inhering structural traits of objects is a form of ‘‘causal essentialism.’’
Before proceeding to a summary of recent work in modal logic which is directed
toward clear solutions to [such] problems . . . it is important to realize that the
perplexities about interpretation can only be understood in terms of certain
presuppositions held by Quine and others which I will call ‘‘the received view’’ (rv).
A bit later one finds the assertion that: ‘‘The Russellian theory of descrip-
tions and the distinction between proper names and descriptions is rejected
by rv.’’ This is immediately followed by the assertion that the morning star
example (3.1) is ‘‘resolved on Russellian analysis as was shown by Smullyan . . .
and others,’’36 and somewhat later by the insistence that ‘‘The usefulness
of the theory of descriptions and the distinction between descriptions and
purely referential names was argued long before it proved applicable to
modal logic,’’ so that one cannot simply reject them, as Quine is alleged
to do.
Now some of this account is quite correct, since the theory of descrip-
tions and of the distinction between them and names as one finds it in the
compendium, for instance, did not originate there, or even with Smullyan,
who first applied it to the interpretation and defense of modal logic, but
was indeed argued by Russell long before. But some of this account is quite
incorrect. It is not true that Quine’s rebuttal to Smullyan on descriptions
requires rejection of Russell’s theory of descriptions.37 And it is not
unambiguously true that Quine’s rebuttal to Smullyan on names requires
rejection of ‘‘the distinction between descriptions and proper names.’’ It is
true that it requires rejection of the neo-Russellian conception of that
distinction, but it is not true that Quine insists on rejecting any distinction
between descriptions and proper names. This should be clear from the last
half-sentence of the rebuttal quoted earlier: ‘‘And not because the proper
names were descriptions.’’
Before Quine, difficulties with the theory that the reference of a name to
its bearer is absolutely immediate had been recognized by Føllesdal and
Alonzo Church.38 And before Quine, difficulties with the theory that the
36
Klibansky (1968, pp. 91ff). This echoes Fitch (1950, p. 553) where it is said that: ‘‘Smullyan has shown
that there is no real difficulty if the phrase [sic] ‘the Morning Star’ and ‘the Evening Star’ are
regarded either as proper names or as descriptive phrases in Russell’s sense.’’ The syntactic ambiguity
in this last formulation as to whether ‘‘in Russell’s sense’’ is supposed to modify ‘‘proper names’’ as
well as ‘‘descriptive phrases’’ matches the ambiguity in the formulation quoted earlier as to whether
‘‘Russellian’’ is supposed to modify ‘‘the distinction between proper names and descriptions’’ as well
as ‘‘theory of descriptions.’’ The ambiguity is appropriate, since the theory of names in question is
neo-Russellian.
37
Though this may not yet have been made clear at the time the encyclopedia article was written, since
the formulation of the rebuttal I have quoted dates from two years later.
38
See Føllesdal (1961), x17, pp. 96ff) and Church (1950). Both address Smullyan and Fitch.
reference of a name to its bearer is descriptively mediated had also been
recognized.39 But before Quine, those who recognized the difficulties with
the absolute immediacy theory generally either did not take them to be
decisive or took them to be arguments for the descriptive mediation theory,
and vice versa. But if the first lesson of Quine’s critique for the theory of
reference is that the neo-Russellian theory of names is untenable, the last
half-sentence of his rebuttal suggests a second lesson, that this first lesson is
not in and of itself an argument for the Fregean theory. Putting these
lessons together, it is not to be assumed that there are just two options; there
is space for a third alternative.
3.3 Formal differences between logical and metaphysical modality

A few words may be in order about post-Quinine work on ‘‘logical’’ or
strict versus ‘‘metaphysical’’ or subjunctive modalities. The locus classicus
for the distinction is of course ‘‘Naming and necessity,’’ but my concern
here will be with formal differences, which are not what was of primary
concern there. Three apparent such differences have emerged.
First, there is the difference at the predicate level. The conventional
apparatus allows de re modalities, as in &Rxy, but does not allow
application of different modalities to the different places of a many-
place relation. The conclusion that, if one is concerned with logical
modality, then the conventional apparatus goes too far when it allows
de re modality, has been endorsed on lines not unrelated to Quine’s
by a number of subsequent contributors to modal logic, a notable
recent example being Hartry Field.40 The complementary conclusion
that, if one is concerned with metaphysical modality, then the con-
ventional apparatus of quantified modal logic does not go far enough,
when it disallows the application of different modalities to the dif-
ferent places of a many-place relation, has also been advanced by
a number of modal logicians, a notable recent example being
39
For work on difficulties with the Fregean theory in the 1950s and early 1960s, see the
discussion in Kripke (1972/1980), and Searle (1967). The doctrines in ‘‘Naming and necessity’’
were first presented in seminars in 1963–4, and whereas that work apologizes for being spotty
in its coverage of the literature of the succeeding years, it is pretty thorough in its discussion
of the relevant literature (work of P. Geach, P. Strawson, P. Ziff, and others) from the
immediately preceding years. (Searle discusses work of yet another contributor, Elizabeth
Anscombe.)
40
In Field (1989, chapter 3). Field also cites several expressions of the same or related views from the
earlier literature, and such citations could in a sense be carried all the way back to the ‘‘principle of
predication’’ (von Wright 1951).
Max Creswell.41 (What is at issue in the latter connection is that a
two-place predicate Rxy may correspond to a phrase with two verbs,
such as ‘‘x is richer than y is,’’ each of which separately can be left in
the indicative or put in a non-indicative mood, as in ‘‘x would have
been richer than y is’’ contrasting with ‘‘x would have been richer than
y would have been,’’ so as to allow cross-comparison between how what
is is, and how what could have been would or might have been.)
Second, there may well be a formal difference already at the sentential
level. For logical modality, at least in some of its versions or variants, iterated
modalities make good sense. I allude here again to work on intensional
mathematics and provability logic, where being unprovable is to be distin-
guished from being provably unprovable. For metaphysical modality, it is
much less clear that iteration makes sense. In Prior’s well-known work on
systems combining subjunctive mood operators with past and future tense
operators, for instance, iterated modal operators collapse, unless separated
by temporal operators: there is no distinction recognized between what is as
of today possibly possible and what is as of today possible, though there is a
distinction between what as of yesterday it was possible would be possible as
of today and what after all is possible as of today. In later work also on the
interaction of mood and tense the purely modal part of the logic adopted
amounts to S5, which collapses iterated modalities.42
Third, there is the difference that while logical possibility does not admit
of degrees – a theory cannot be just a little bit inconsistent – metaphysical
possibility seems to, with some possibilities being more remote than others.
At any rate, this is the thought that underlies theories of counterfactuals
since the pioneering work of R. Stalnaker.43 In particular, miraculous
possibilities, involving violations of the laws of physics, are in general
more remote than non-miraculous possibilities, a fact that may make the
error of earlier writers in associating counterfactuals with physical necessity
in some respects a less serious one.
Thus there is a fair amount of work that has been – or can be construed
as – exploration of the formal differences between the two kinds of modality.
41
In Creswell (1990). Cresswell also cites several expressions of the same or related views from the
earlier literature, and such citations could in a sense be carried all the way back (Lewis 1970). This is
the earliest relevant publication known to me, but its author has suggested that there was very early
unpublished work on the topic by A. P. Hazen and by D. Kaplan. The parallel phenomenon for
tense in place of mood was noted even earlier by P. Geach.
42
See Prior (1967b, chapter VII), and among later work R. H. Thomason (1984). The purely modal
part is also S5 for virtually all the workers there cited, as well as later ones like A. Zanardo.
43
Stalnaker (1968). This feature becomes even more prominent in later work on the same topic by
D. K. Lewis and others.
As apparent formal differences accumulate, the situation comes to look like
this: there is one philosophically coherent enterprise of logical modal logic,
attempting to treat in the object language what classical logic treats in the
metalanguage; there is another philosophically coherent enterprise of meta-
physical modal logic, attempting to do for grammatical mood something like
what temporal logic does for grammatical tense; there is a mathematically
coherent field of non-classical logics dealing with technical questions about
both these plus intuitionistic, temporal, and other logics; but there is no
coherent field broad enough to include both kinds of ‘‘modal logic,’’ but still
narrower than non-classical logic as a whole. In this sense, there is no coherent
enterprise of ‘‘modal logic’’ – a conclusion that may be called Quinesque.
3.4 New alternatives in the theory of reference for names

A few words may also be in order about post-Quinine work on theories of
reference for names that reject both the Fregean descriptive mediation and
the neo-Russellian absolute immediacy views. The locus classicus for such an
alternative is of course again ‘‘Naming and necessity.’’ One can perhaps
best begin to bring out how the new theory of that work relates to the old
theory of Quine’s opponents by considering what similarities and differ-
ences are emphasized in the only early extended response to the new theory
by the one former adherent of the old theory who remained living and
active in the field through the 1970s and 1980s and beyond.44
First, the one area of real agreement between the new theory and the old is
emphasized, that both are ‘‘direct’’ (in the minimal sense of ‘‘anti-Fregean’’)
theories; and the new theory is praised for providing additional arguments:
Kripke’s criticism of the ‘‘Frege–Russell’’ view . . . is presented . . . Among the
arguments he musters are that competent speakers communicate about individuals,
using their names, without knowing or being able to produce any uniquely
identifying conditions short of circular ones . . . Unlike descriptions, proper
names are indifferent to scope in modal (‘‘metaphysical’’) contexts . . . Contra
Frege he points up the absurdity of claiming that counterfactuals force a shift in
the reference of a name.
Second, another area of apparent agreement, over the ‘‘necessity of iden-
tity’’ (in some sense), is also emphasized, with the new theory again being
praised for providing additional arguments:
44
Unfortunately this comes in the form of a review of a book by a third party, and is subject to the
limitations of such a form. The third party is Leonard Linsky; the book is Linsky (1977); the review is
Marcus (1978). The three quotations to follow come from pp. 498, 501, and 502–3.
It is one of the achievements of Kripke’s account, with its effective use of the
theory of descriptions, the theory of proper names, the distinction between
metaphysical and epistemological modalities (for example, necessary vs. a priori),
that it provides us with a more coherent and satisfactory analysis of statements
which appear to assert contingent identities.
Third, the contribution most praised is the provision of a novel account of
the mechanism by which a name achieves reference to its bearer:
Kripke provided us with a ‘‘picture’’ which is far more coherent than what had
been available. It preserves the crucial differences between names and descriptions
implicit in the theory of descriptions. By distinguishing between fixing the mean-
ing and fixing the reference, between rigid and nonrigid designators, many
nagging puzzles find a solution. The causal or chain of communications theory
of names (imperfect and rudimentary as it is) provides a plausible genetic account
of how ordinary proper names can acquire unmediated referential use.
All this amounts to something approaching an adequate acknowledg-
ment of substantial additions by the new theory to the old, but what needs
to be understood is that the new theory in fact proposes substantial amend-
ments also. The new theory is not ‘‘direct’’ in anywhere near as extreme a
sense as the old. On the new theory, which is a ‘‘third alternative,’’ the
reference of a name to its bearer is neither descriptively mediated nor
absolutely immediate, but rather is historically mediated, accomplished
through a chain of usage leading back from present speakers to the original
bestower of the name. Also the new theory does not endorse the ‘‘necessity
of identity’’ in anything like so broad a sense as does the old theory, or on
anything like the same grounds. On the new theory, ‘‘Hesperus is
Phosphorus’’ is only subjunctively or metaphysically necessary – not strictly
or logically necessary like ‘‘Phosphorus is Phosphorus.’’ And moreover the
metaphysical necessity of identity is the conclusion of a separate argument
involving considerations peculiar to subjunctive contexts, about cross-
comparison between actual and counterfactual situations – not an
immediate corollary or special case of some general principle of the intersub-
stitutability of coreferential names in all (except meta-linguistic) contexts.45
The gap between the old, neo-Russellian theory and the new, anti-
Russellian theory is large enough to have left space for the development
45
In this connection mention may be made of one serious historical inaccuracy – of a kind extremely
common when authors quote themselves from memory decades after the fact – to be found in the
book review, where it is said that the compendium maintained ‘‘that unlike different but corefer-
ential descriptions, two proper names of the same object were intersubstitutable in modal contexts’’
(p. 502). In actual fact, in the compendium it is repeatedly asserted that two proper names of the same
object are intersubstitutable in all contexts.
of several even newer fourth and fifth alternatives, semi- or demi-semi- or
hemi-demi-semi-Russellian intermediate views, of which the best known
is perhaps Nathan Salmon’s.46 These differ from the Kripkean, anti-
Russellian theory in that they want to say that in some sense ‘‘Hesperus is
Phosphorus’’ and ‘‘Phosphorus is Phosphorus’’ have the same ‘‘semantic
content.’’47 They differ from the Smullyanite, neo-Russellian theory in that
there is full awareness that in some sense assertive utterance of ‘‘Hesperus is
Phosphorus’’ can make a difference to the ‘‘epistemic state’’ of the hearer in
a way that assertive utterance of ‘‘Phosphorus is Phosphorus’’ cannot. How
it could be that utterances expressing the same semantic content have such
different potential effects on epistemic states is in a sense the main problem
addressed by such theories. My concern here is not to offer any evaluation,
or even any exposition, of the solutions proposed, but only to point
out that they all operate in the space between Fregeanism and neo-
Russellianism – and therefore in a space of whose existence Quine was
one of the first to hint.
3.5 Have the lessons been learned?

It would be absurd to claim that Quine anticipated all the many important
developments in modal logic or the theory of reference to which I have
been alluding. But it is not absurd to suggest that some of them might have
been arrived at sooner if the reaction to Quine’s critique had been more
attentive.
Is the matter of more than antiquarian interest today? Well, certainly there
are many workers in philosophical logic and philosophy of language (only a
few of whom I have had occasion to mention) who have long since fully
absorbed every lesson there was to be learned from Quine. And yet, scanning
the literature, it seems to me that specialists in the relevant areas do not
always clearly express these lessons in their writings, and that (surely partly in
consequence) many non-specialists interested in applying theories of modality
or reference to other areas have not yet fully learned these lessons.
Take modal logic first. It is said that when Cauchy lectured on the
distinction between convergent and divergent series at the Académie des
Sciences, Laplace rushed home to check the series in his Mécanique Céleste.
46
Salmon (1986). While the early Marcus followed Smullyan, the later Marcus has developed in
response to Kripke an idiosyncratic theory that may be described as intermediate in degree of
Russellianism between Salmon’s and Smullyan’s. See Marcus (1990).
47
For Kripke’s rejection of this view, see the closing paragraphs of the preface to Kripke (1980).
But when Kripke lectured on the distinction between logical and meta-
physical modality, modal logicians did not rush home to check which
conclusions hold for the one, which conclusions hold for the other, and
which result from a fallacious conflation of the two. It is a striking fact that
the basic article – an article written by two very eminent authorities – on
modal logic in that standard reference work, the multi-volume encyclo-
pedia mistitled a ‘‘handbook’’ of philosophical logic, makes no mention at
all of any such distinction and its conceivable relevance to choosing among
the plethora of competing modal systems surveyed.48
No wonder then that workers from other areas interested in applying
modal logic seem often not fully informed about formal differences
between the two kinds of modality. To cite only the example I know
best, consider philosophy of mathematics, and debates over nominalist
attempts to provide a modal reinterpretation of applied mathematics,
where quantification into modal contexts is unavoidable. Those on the
nominalist side have quite often supposed that they could get away with
quantifying into contexts of logical modality, while those on the anti-
nominalist side have quite often supposed that anyone wishing to make
use of modality must stick to the traditional formal systems, which do not
allow for cross-comparison. Both suppositions are in error.49
Take the theory of reference now. Here a great many people seem to
have difficulty discerning the important differences among distinct anti-
Fregean theories. To mention again the example I know best, many
nominalists seem to think that the work of Kripke, David Kaplan, Hilary
Putnam, and others has established something implying that it is impos-
sible to make reference to mathematical or other abstract, causally inert
objects.50
Such misunderstandings are encouraged by the common sloppy use by
specialists of ambiguous labels like ‘‘causal theory of reference’’; and even
those who carefully avoid ‘‘causal theory’’ in favor of ‘‘direct theory’’ are
often sloppy in their usage of the latter, encouraging other confusions. Of
late, not only has the confused opinion become quite common that
Quine’s critique has somehow been answered by the new theory of
48
Bull and Segerberg (1984). Other articles in the same work, some of which I have already cited, do
recognize the importance of the distinction.
49
It would be out of place to enter into technicalities here. See Burgess and Rosen (1997).
50
In actual fact, on Kripke’s theory, for instance, a name can be given to any object that can be
described, not excluding mathematical objects. But again see Burgess and Rosen (1997). (The theory
of P. Geach probably deserves and the theory of M. Devitt certainly deserves the label ‘‘causal,’’ and
does have nominalistic implications.)
names (the one coming from ‘‘Naming and necessity’’); but so has the even
more confused opinion that Quine’s critique was already answered by an
old theory of names (the one coming from Russell through Smullyan to the
compendium); and so too has the most confused opinion of all, that there
is no important difference between the old and new theories. Confusion of
this kind is found both among those who think of themselves as sympa-
thizers with ‘‘the’’ theory in question,51 and among those who think of
themselves as opponents of ‘‘it.’’52 The latter cite weaknesses of the old
theory as if pointing them out could refute the new theory – a striking
example of how confusion over history of philosophy can lead to confusion
in philosophy proper.
There is hardly a better way to sort out such confusions than by
considering the relations of the old and the new theory to Quine’s critique,
from which therefore some people still have something to learn. Neither
the old theory nor the new provides a refutation of that critique, but the
reasons why are radically different in the two cases. The old theory
attempted to refute that critique, but in doing so it arrived at consequences,
notably the one made explicit in the ‘‘lexicon’’ passage quoted earlier, that
reduced the theory to absurdity. Quine’s rebuttal, pointing out the unten-
ability of these consequences, refuted the old theory. Quine’s critique does
not refute the new theory, but then neither does the new theory refute
Quine’s critique, nor does it even attempt to do so. The new theory would
refute any incautious claim to the effect that ‘‘quantification into any
intensional context is meaningless,’’ since it shows that proper names
have all the properties required of canonical terms for contexts of subjunc-
tive modality. But Quine’s critique was addressed to strict modality, and as
for that, the main creator of the new theory of names has said as I do:53
‘‘Quine is right.’’
51
For comparatively moderate instance see the review Lavine (1995).
52
For an extreme instance see Hintikka and Sandu (1995). This work acknowledges no important
differences among: (i) the neo-Russellian theory of Smullyan as expounded by the early Marcus
(which incidentally is erroneously attributed to Marcus as something original, ignoring the real
authors Smullyan and Russell); (ii) theories adopted in reaction to Kripke by the later Marcus; and
(iii) the theory of Kripke.
53
In context, what is said to be right is specifically the rebuttal to Smullyanism on names quoted
earlier. See Kripke (1972, p. 305).
12
Translating names
1 MILLIANISM AND ANTI-MILLIANISM
Mill taught that the signification of a word has in general two components,
denotation and connotation, but that in the special case of a proper name
there is no connotation, and the signification of the word is just its
denotation. According as ‘‘meaning’’ is aligned with ‘‘connotation’’ or
with ‘‘signification,’’ this doctrine comes out as ‘‘A proper name has no
meaning’’ or as ‘‘The meaning of a proper name is just its denotation.’’
Today ‘‘Millianism’’ is most often used as a label for the latter version:
(1) The meaning of a name is its denotation.
An immediate consequence of (1) is the following:

(2) Two names with the same denotation have the same meaning.
An immediate objection to (2) is that different names for the same item
may be distinguished in level (formal, familiar). Such features are very
important for usage. (Imagine what a diplomatic contretemps would result
if President Chirac were to write President Bush a letter beginning ‘‘Yo,
Dubya!’’) And with words that (unlike proper names) appear in diction-
aries, such features are commonly noted in their definitions, presumably as
part of the meaning of the word. It is therefore, the objector claims,
reasonable to take them to be part of the meaning of a name as well.
To this objection a Millian may reply by insisting that level is a feature of
usage but not of meaning. This reply illustrates in miniature the fact that in
this area there are no conclusive proofs or refutations, but in the end only
judgments as to the overall plausibility of proposals as to where to draw the
line between semantics and pragmatics. Alternatively, a Millian might
concede that (1) and (2) do indeed need to be qualified, while insisting
that the objection does not touch what is really important in Millianism, of
which the unqualified (1) and (2) are inadequate formulations. This
236
Translating names 237
alternative illustrates the fact that it is difficult to find formulations agree-
able to all parties of the main theses in dispute. Here I will for brevity let the
simple (1) and (2) stand as formulations of Millianism. (Features of level
have been mentioned only to illustrate the two facts about the nature of the
debate just indicated, and otherwise will play no role in the discussion
below.)
I will likewise stick with fairly simple formulations of two other theses,
one common to most Millians and anti-Millians alike, the other a premise
of one popular form of anti-Millianism. The common thesis is composi-
tionality, thus:
(3) Short, simple sentences differing only by substituting one word for another
with the same meaning have the same meaning.
Then (2) and (3) together at once give the following:
(4) Short, simple sentences differing only by substituting one name for another
with the same denotation have the same meaning.
The anti-Millianism thesis is so-called transparency, thus:
(5) When short, simple sentences have the same meaning, a subject’s asserting or
assenting to one and denying or dissenting from the other is an indication of a
deficiency in the subject’s linguistic knowledge.
All examples below will involve short, simple sentences (usually three-word
sentences of name–copula–adjective form). In particular, none will involve
such complications as embedding in ‘‘So-and-so believes that . . .’’ contexts.
The anti-Millian argument I wish to consider produces certain exam-
ples, and appeals to pre-theoretic intuitions to support the following claim
about them:
(6) There are cases of two short, simple sentences that differ only by substitution
of one name for another having the same denotation, where a subject may
assert or assent to one sentence and deny or dissent from the other sentence,
not for lack of linguistic knowledge, but rather for lack of some other kind of
knowledge.
The other knowledge in the best-known examples is biographical or
astronomical or geographical or historical (counting etymological knowl-
edge, which is not needed for fluency in the present form of a language, as
historical rather than linguistic).
While the examples most often cited pertain to persons (Marcus Tullius
Cicero) and planets (Venus), there are also examples about cities. Thus
Jones may while doing some tourism have cruised by moonlight past the
famous skyline of the former Ottoman capital, and may while negotiating a
business deal have paid a rapid visit to an undistinguished commercial
district on the outskirts of the same city. If the place was consistently called
‘‘Byzantium’’ by the tour guides and ‘‘Istanbul’’ by the business people,
Jones may end up asserting ‘‘Byzantium is sublime’’ and ‘‘Istanbul is tacky’’
while denying ‘‘Byzantium is tacky’’ and ‘‘Istanbul is sublime.’’ The anti-
Millian’s intuition is that this shows Jones to be a geographical ignoramus,
but does not show Jones to be a linguistic incompetent like the prize-
fighting fan Pete, who applauds whenever someone calls Muhammed Ali
‘‘a great boxer’’ but would take offense if anyone called him ‘‘a great
pugilist.’’
This particular case involves binomialism of the first (Hesperus/
Phosphorus) kind, where an item has two different names tracing back
to two different acts of naming (by Megarian colonists in the seventh
century BC E , and by the Atatürk government in the 1930s). But similar
examples can arise with binomialism of the second (Cicero/Tully) kind,
where two names trace back to the same original act of naming along
divergent paths of transmission. A tourist trip to ‘‘Peking’’ may be confined
mainly to the Forbidden City, while a business meeting in ‘‘Beijing’’ may
be held in a quarter built in the 1950s in high Stalinist style.
2 ANTI-ANTI-MILLIANISM AND TRANSLATION
Now (4) and (5) and (6) cannot all be true, and an accumulation of
examples makes (6) hard to deny. The anti-Millian, assuming (5), con-
cludes that the Millian thesis (4) is false. A militant Millian may simply
insist that since (4) is true, (5) must be false. But many Millians would,
I think, prefer to find some independent argument against (5), not relying
on (4), and one place where they have looked for the materials to construct
such an argument has been in Saul Kripke’s notorious ‘‘A puzzle about
belief’’ (Kripke 1979). That paper contains two examples that, however the
author intended them, anti-anti-Millians might appropriate and exploit
for purposes of arguing against (5), and it is the anti-anti-Millian appro-
priation and exploitation of the first of these, the Pierre example, that I
wish to consider here.
The example involves the translation of proper names, and so I should
acknowledge at the outset the fact that in common parlance one seldom
speaks of ‘‘translating’’ proper names at all. In the broad sense used here,
whatever expression is used in a translation of a sentence at the place
corresponding to the place where a name is used in the sentence being
translated may be called a ‘‘translation’’ of that name. But it must be
acknowledged that most proper names simply do not have non-trivial
translations: typically a name is not replaced by something else when
translating a sentence in which it occurs, but simply taken over for use in
the language into which the sentence is being translated, as a so-called
exonym.
It would be an absurd affectation for a native English speaker, describing
to other native English speakers a recent trip to Italy, to speak of having
been to ‘‘Roma’’ and ‘‘Napoli’’ and ‘‘Firenze.’’ For the famous tourist
destinations of Rome and Naples and Florence are among the minority
of Italian cities whose names do have non-trivial English translations, or
‘‘Anglicizations’’ as they are more ordinarily called. But if I wish to mention
some less famous Italian place, such as Princeton’s sister city Pettoranello,
there is no alternative to using the Italian name.
More precisely, one uses in speech as close an approximation to the
Italian name as one can manage, given the phonetic differences between
Italian and English. In writing, one can just use in English the original
name if translating from a language written in the Roman alphabet,
perhaps modulo some diacritical marks and ligatures. (Thus in the days
of old-fashioned manual typewriters, ‘‘Gödel’’ was often written ‘‘Goedel,’’
and even today ‘‘Gauß’’ is still written ‘‘Gauss.’’) With languages written in
another alphabet one uses what is ordinarily called a ‘‘transliteration,’’ and
with languages not written alphabetically a ‘‘transcription.’’
The Pierre example involves ‘‘London’’ as the name of the great city in
England, which unlike ‘‘Liverpool’’ or ‘‘London’’ as the name of the town in
Ontario does have a non-trivial French translation, ‘‘Londres.’’ The most
direct way to turn the Pierre example into an argument against (5) begins
with the following comparatively uncontroversial assumptions:
(7a) ‘‘Londres’’ is the correct French translation of the English ‘‘London.’’
(7b) ‘‘est joli[e]’’ is the correct French translation of the English ‘‘is pretty.’’
(7) ‘‘Londres est joli[e]’’ is the correct French translation of the English ‘‘London
is pretty.’’1
The next step would be a comparatively uncontroversial inference from (7)
to the following:
1
Should it be ‘‘Londres est joli’’ or ‘‘Londres est jolie’’? Having consulted with quite a number of
informants, French and Québecois, I can report that native Francophones themselves are undecided
as to the genders of most city names, ‘‘Londres’’ included.
(8) The meaning of ‘‘Londres est joli[e]’’ (in French) and the meaning of ‘‘London
is pretty’’ (in English) are the same.
Kripke then describes the case of one Pierre, of whom he claims the
following:
(9) Pierre assents to ‘‘Londres est joli[e]’’ and dissents from ‘‘London is pretty.’’
(10a) There is no deficiency in Pierre’s knowledge of French.
(10b) There is no deficiency in Pierre’s knowledge of English.
(10) There is no relevant deficiency in Pierre’s linguistic knowledge.
Taken together, (8) and (9) and (10) give a counterexample to (5), which
may be claimed to neutralize the anti-Millian force of the Byzantium/
Istanbul and Peking/Beijing examples, and of the better-known Cicero/
Tully and Hesperus/Phosphorus examples.
3 ANTI-ANTI-ANTI-MILLIANISM
I do not wish to question here the truth of (7), or the cogency of the
inference thence to (8). Moreover, in the example as Kripke describes it, (9)
seems indisputable, too. I think, however, that doubts may be raised about
(10), and in order to raise them I wish to question not the truth of but the
grounds for (7a). So let me begin by mentioning a way to argue for the
correctness of translating ‘‘London’’ as ‘‘Londres’’ that would be wholly
inappropriate in the present context.
The argument I have in mind takes as its initial premise the one
undisputed fact in this area:
(11) ‘‘Londres’’ (in French) and ‘‘London’’ (in English) denote the same place.
It then proceeds to this intermediate step:
(12) ‘‘Londres’’ (in French) and ‘‘London’’ (in English) have the same meaning.
And it then proceeds to the final conclusion (7a). Such an argument would
be inappropriate in the present dialectal context, because we are supposed
to be looking for an anti-anti-Millian argument independent of distinc-
tively Millian assumptions, while it is precisely the distinctively Millian
thesis (2) that would be needed to get from (11) to (12).
Questions of dialectic aside, it seems clear that while sameness of
denotation is a necessary condition, it is simply not a sufficient condition
for correctness of translation of place names. If some Francophone coun-
terpart of Jones exclaims, ‘‘Que Byzance est belle!’’ and ‘‘Que Stamboul est
laid!’’ he surely has to be translated as saying ‘‘How beautiful Byzantium is!’’
and ‘‘How ugly Istanbul is!’’ and not ‘‘How beautiful Istanbul is!’’ and
‘‘How ugly Byzantium is!’’ Nor is this just because the person whom we are
translating is confused about the identity of the city. ‘‘Staline restait à
Moscou’’ and ‘‘Djougachvili se cachait’’ surely have to be translated ‘‘Stalin
remained in Moscow’’ and ‘‘Djugashvili was hiding,’’ not ‘‘Djugashvili
remained in Moscow’’ and ‘‘Stalin was hiding,’’ even if the translation is
from the writings of Trotsky and is being prepared for a groupuscule of
Trotskyites all thoroughly aware that Stalin/Staline and Djugashvili/
Djougachvili are one and the same person.
These examples may make it look as if an etymological connection and/
or a resultant phonetic or orthographic relationship were the key. But
such links are neither necessary nor sufficient for correct translation. Some
of the most ancient and famous countries – Egypt, India, Greece – have
long been known to most of the outside world by names having nothing
to do with their native names. Any one of these cases shows etymological
and phonetic and orthographic links are not necessary, as would the case
of ‘‘Deutschland’’ aka ‘‘Germany’’ aka ‘‘Allemagne.’’ The Greek case shows
that they are not sufficient, either. For there actually exists in English a
name for the country in question, the name ‘‘Hellas,’’ that is etymologically
connected and phonetically and orthographically related to the native
name; but nonetheless it is the unconnected, unrelated ‘‘Greece’’ that is
the correct translation of the native name in all common, prosaic contexts.
(Use of ‘‘Hellas’’ and ‘‘Hellenic’’ is appropriate only in the kind of contexts
where ‘‘Hibernia’’ and ‘‘Hibernian’’ might be used instead of ‘‘Ireland’’ and
‘‘Irish.’’)
But if not merely denotation, or that plus etymology, what does make a
given translation of a given name correct or incorrect? A clue is given,
I think, by the case of the translation into western European languages –
more usually called the ‘‘Romanization’’ – of Chinese proper names.
Historically, approximate phonetic transcriptions were used. But since
the differences between the principal Chinese ‘‘dialects’’ are so great, and
the different European languages are so many, that practice led to chaos,
and hence to a demand for a fixed system. Unfortunately not one but two
systems were fixed (both based on Mandarin pronunciation on the Chinese
side, but who knows what on the European side), one by European scholars
and another by the Chinese government. As a result, most important
Chinese geographical places and historical personages can be translated
in either of two ways into English or French. And often, as with ‘‘Hsüan-
tsang’’ aka ‘‘Xuánzàng’’ (the historical personage on whom the character
Tripitaka in A Journey to the West is based), it is difficult for the average
Anglophone or Francophone reader to trace any link between the Wade–Giles
and Pinyin versions. But at least the situation is better than it was when this
same personage was also being called by the names ‘‘Hiouentang’’ and ‘‘Yuan
Chwang’’ and half a dozen others.
While the case of the Romanization of Chinese, where conventions were
consciously and deliberately and explicitly adopted (and where there are
two rival sets of conventions), is unusual, I take it to illustrate a more
general principle. The general principle is that what counts as the (or a)
correct translation of a name from one language to another is determined by the
conventions and customs of the community of bilinguals and specifically of
translators. Needless to say, those conventions and customs are constrained
by the requirement that the name and its translation must have the same
denotation, and may be influenced, though not determined, by etymo-
logical or phonetic or orthographic considerations. In the case of second-,
third-, and fourth-hand borrowings, as when, say, an Arabic or Hebrew
place-name migrates via Greek and Latin and French to English, suffering
distortions at each step, the correct formulation of the general principle
would have to be more complicated. (The case where there is no link of any
kind, even through a chain of intermediaries, need not be considered, since
in this case all names will initially be like ‘‘Pettoranello’’ in lacking non-
trivial translations.) But leaving aside details, if anything like this principle
is granted, there would seem to follow another important principle, to the
effect that bilingual competence involves something more than just competence
in each of the two languages separately, something that can only be acquired by
contact with members of the bilingual community, direct or through their
writings.
Now to return to Pierre, the general principle just formulated raises
doubts whether (10a) and (10b) are enough to imply (10). Kripke describes
Pierre as lacking even the minimal kind of contact with the bilingual
community he could acquire by looking into a French–English dictionary.
Supposing Pierre keeps a diary in his native French of his experiences in the
English city where he finds himself, he will write for the name of that city
‘‘London’’ if he has seen the local name written, or if he has only heard it
spoken, something on the order of ‘‘Lonnedonne.’’ Either way, a single
glance at his diary would suffice to show any bilingual that Pierre lacks the
knowledge, which a French–English dictionary could provide him with,
that ‘‘London’’ is one of the small minority of English place-names that has
a non-trivial French translation, and that this translation is ‘‘Londres.’’
I wish to suggest that what this shows is that Pierre is lacking bilingual
competence, and that this is a lack of a kind of linguistic knowledge,
making (10) false. Thus at bottom, I suggest, Pierre’s state is not after all so
very different from that of Ali’s linguistically challenged fan Pete.
4 ANTI-MILLIANISM AND ANTI-DESCRIPTIVISM
As we have seen, many examples point to the following principle:

(13) Sameness of denotation of a name in one language with a name in another
language is not in general sufficient to make the latter a correct translation of
the former.
Many examples also indicate that sameness of denotation of a name in one
language and a description in another language are not in general sufficient
to make the latter a correct translation of the former. Indeed, something
much stronger holds:
(14) A description is never a correct translation of a proper name.
And it should be emphasized that (14) applies to metalinguistic descrip-
tions every bit as much as to any others. It is always wrong to translate
‘‘Voltaire’’ in a French text as ‘‘the person called ‘Voltaire’’’ or ‘‘the person
who called himself ‘Voltaire’,’’ and this is so even though ‘‘Voltaire’’ was
only a nom de plume.
The observations (13) and (14) suggest that Millianism and descriptivism
are both alike false. Needless to say, these observations do not prove any
such conclusions, for the reason I stated at the outset: in any case of
apparent counterexample, the militant Millian and the die-hard descrip-
tivist can simply suggest that whatever phenomena are cited by their critics
pertain to pragmatics rather than semantics. In the present case, the
Millian or descriptivist alike could simply claim that sameness of meaning
is insufficient to make for correct translation. Yet, to repeat, considera-
tions of translation, while not refuting the widespread assumption that
Millianism and descriptivism are the only alternatives, do suggest doubts
about it. For if one thinks of meaning as that which translation aims to
preserve, then a third alternative suggests itself. This third alternative is
that different names for the same item have different meanings, even
though there is nothing to be said to an English speaker about what
‘‘London’’ (in English) means, except that it means London – nothing to
be said about what ‘‘London’’ (in English) means that might be informa-
tive in the way that it might be informative to tell Pete ‘‘‘pugilist’ means
boxer,’’ or for that matter to tell Pierre ‘‘‘London’ (en anglais) veut dire
Londres.’’
Kripke says that Frege is to be criticized for confusing two senses of
‘‘sense.’’ If we ignore Frege’s concrete examples, where the senses of names
seem to be said to be those of certain associated descriptions, and consider
only Frege’s abstract formulations, according to which the sense of a
name is its mode of presenting its bearer, then the view that different
names with the same denotation have different senses certainly seems
correct. Presenting a fugitive novelist as ‘‘John Doe’’ and presenting him
as ‘‘Salmon Rushdie’’ are certainly, in any ordinary sense of the phrase ‘‘way
of presenting,’’ two very different ways of presenting him – with perhaps all
the difference between life and death between them. The view that differ-
ent names for the same person or place may have different meanings, even
though no name has the same meaning as any description, may thus be
restated as the view that different names have different senses in one sense
of ‘‘senses’’ (modes of presentation), though they have no senses at all in
another sense of ‘‘senses’’ (associated descriptions). Mill’s error lies, on this
way of putting the matter, not in his claiming that names have no
connotation (in the sense of associated descriptions), but rather in his
assuming there is nothing more to signification than denotation plus
connotation; but this is an error of which descriptivists are guilty as well.
In addition to what is signified, there is the way it is signified, which in the
case of names is not by connoting some description of it.
However the view is stated, it is one that gains some support from
consideration of translation, and one that deserves more attention than it
has heretofore been given. It, too, should be taken into account in consid-
ering other puzzling cases beyond the kind I have been discussing here.
Indeed, the real aim of the present note has been not to solve the puzzle of
Pierre, but to call attention to this comparatively neglected view.
5 ENVOI
About the Pierre puzzle there remains one further point to be acknowl-
edged. Suppose Pierre were informed that the French translation of
‘‘London’’ is ‘‘Londres.’’ This might well clear up his confusion, but then
again it might not. For Pierre might conclude that there are two homo-
nymous names ‘‘Londres’’ in French, one denoting a pretty place he saw in
pictures back in France and one the ugly place where he lives now that he
has come to England, just as there are two homonymous names
‘‘Bretagne,’’ one denoting a peninsula that is part of France, and one
denoting a large island across the Channel from France. There is this
difference, that in the case of ‘‘Bretagne’’ there are two different names in
English, ‘‘Brittany’’ and ‘‘Britain,’’ while in the case of ‘‘Londres’’ there is (so
far as Pierre knows) in English just ‘‘London.’’ Pierre may be somewhat
puzzled why the English have not had the wit to add something to one or
both of the names, as the French distinguish their peninsula ‘‘Bretagne’’
from the island ‘‘Grande Bretagne’’ or the German town ‘‘Aix-la-Chapelle’’
from the French town ‘‘Aix-en-Provence.’’ But perhaps after his ugly
experiences living in London he expects no better of the English.
It is clear that if Pierre were to fall into the kind of confusion I have just
been describing, then the Pierre example would reduce to a variant of
Kripke’s other example, the Paderewski example. But this example raises a
quite different set of issues from the issues about translation with which
I have been concerned here, and must be left for another occasion.
13
Relevance: a fallacy?
1 INTRODUCTION
Responding to Harvey’s theories about the circulation of the blood,

Dr. Diafoirus argues (a) that no such theory was taught by Galen, and
(b) that Harvey is not licensed to practice medicine in Paris. Plainly there is
something wrong with a response of this sort, however effective it may
prove to be in swaying an audience. For either or both of (a) and (b) might
well be true without Harvey’s theory being false. So Diafoirus’s argument
can serve only to divert discussion from the real question to irrelevant side-
issues. The traditional term for such diversionary debating tactics is ‘‘fallacy
of relevance.’’
In recent years this tradition has come to be used in a quite untraditional
sense among followers of N. D. Belnap, Jr., and the late A. R. Anderson.
(All citations of these authors are from their masterwork Anderson and
Belnap (1975), and are identified by page number.) According to these self-
styled ‘‘relevant logicians,’’ it is items (IA) and (IIA) in Table 13.1 that
constitute the archetypal ‘‘fallacies of relevance.’’ (In the table , &, and Ú
stand for truth-functional negation, conjunction, and disjunction, respec-
tively.) These forms of argument, say Anderson and Belnap, are ‘‘simple
inferential mistake[s], such as only a dog would make’’ (p. 165). The
authors can hardly find terms harsh enough for those who accept these
schemata: they are called ‘‘perverse’’ (p. 5) and ‘‘psychotic’’ (p. 417).
Needless to say, (IA) and (IIA), which can be traced back at least to
Chrysippus, were not traditionally regarded as fallacious. The Anderson–
Belnap notion of ‘‘relevance,’’ whatever it may amount to, must be some-
thing quite different from the traditional notion, which ‘‘was central to logic
from the time of Aristotle’’ (p. xxi). And yet the authors declare their
so-called ‘‘relevant logic’’ to be a commonsense philosophy, in accord with
the intuitions of ‘‘naive freshmen’’ (p. 13) and others who have not been
‘‘numbed’’ (p. 166) by a course in classical logic. Moreover, whereas other
246
Relevance: a fallacy? 247
Table 13.1
(I) p or q (II) not both p and q

not p p
q not q
(IA) pÚq (IIA) (p & q)
p p
q q
(IB) pþq (IIB) (p q)
p p
q q
dissident logicians (e.g. intuitionists) hold that some forms of argument

always accepted and used without question by mathematicians in their
proofs are in fact untrustworthy, Anderson and Belnap are at some pains
to explain (pp. 17–18 and 261–2) that their brand of non-classical logic does
not conflict with the practice of mathematicians, but only with the classical
logician’s account of that practice.
In view of the fact that everyday arguments and mathematical proofs
abound in instances of (I) and (II), one may wonder how Anderson and
Belnap could hope to reconcile their objections to (IA) and (IIA) with the
claim that their ‘‘relevant logic’’ is compatible with commonsense and
accepted mathematical practice. The answer is that the authors believe
that ordinary-language argument patterns (I) and (II) should be repre-
sented as expressions of the ‘‘intensional’’ schemata (IB) and (IIB), which
are relevantistically acceptable, and not of the ‘‘extensional’’ schemata (IA)
and (IIA), which relevantists reject.
The compound p þ q in (IB) is supposed to be an ‘‘intensional dis-
junction’’ stronger than the truth-functional p Ú q in that mutual rele-
vance of p and q is required for its truth. This p þ q is not entailed by p,
nor even by p & q, since it might be false even though p and q were both
true (or even necessary). This happens in the case of irrelevant pairs such
as p ¼ ‘‘Bach wrote the Coffee Cantata’’ and q ¼ ‘‘The Van Allen belt is
doughnut-shaped’’ (p. 30). Dually, the p q of (IIB) is an ‘‘intensional
conjunction,’’ better called ‘‘cotenability’’ or ‘‘non-preclusion,’’ a com-
pound weaker than the truth-functional p & q in that mutual irrelevance
of p and q is sufficient for its truth. This p q does not entail p, or even
p Ú q, since it might be truth even though p and q were both false (or even
impossible).
The relevantists’ claim that (IB) and (IIB) best represent (I) and (II)
admits of two formulations: a stronger and a weaker. The stronger claim
would be that the ordinary-language ‘‘or’’ and ‘‘and’’ literally mean þ and
rather than Ú and &. The weaker claim would be that anyone basing an
argument on the premise that p or q, or that not both p and q, will at least be
in a position to assert that p þ q and (p q) as the case may be. (The latter
claim is weaker than the former because even if ‘‘or’’ and ‘‘and’’ meant Ú and
&, it might still be that arguments of form (IA) and (IIA) could always be
avoided in practice because in any instance where one might wish to argue
from p Ú q or from ( p & q) the stronger premises p þ q and ( p q) would
be available.)
The stronger of these two relevantistic claims seems quite untenable.
True, Anderson and Belnap do make a feeble attempt (pp. 176–7) to
argue that the ordinary-language ‘‘or’’ usually means þ rather than Ú.
Their argument, however, is scarcely original, amounting to no more
than a repetition of the arguments that used to be used by P. F. Strawson
and other Oxford philosophers in their diatribes against modern
logic. To the serious objections against such Oxonian arguments that
have emerged from H. P. Grice’s work on conversational implicature
the authors of Anderson and Belnap (1975) attempt no reply. In any
case, even if the claim that ‘‘or’’ means þ is credited with a certain
intuitive plausibility, the same cannot be done for the claim that ‘‘and’’
means . To be sure, Strawson and others have claimed that the meaning
of ‘‘and’’ sometimes diverges from that of & (that in some uses ‘‘and’’ does
duty for ‘‘and subsequently’’ or ‘‘and as a result’’); but these divergences
are always in the direction of something stronger than &, not something
weaker. The relevantists themselves shrink from identifying with
‘‘and.’’ While asserting (pp. 344–5) that cotenability is an ‘‘analogue’’
that ‘‘in some ways . . . looks like’’ conjunction, they concede that ‘‘it
isn’t conjunction.’’
The untenability of the strong claim that everyday and mathematical
instances of (I) and (II) are literally meant as instances of (IB) and (IIB)
might already be thought to do considerable damage to the relevantists’
claim to be espousing a commonsense philosophy of logic. My aim in this
paper is to cast further doubt on that claim by presenting counterexamples
to the weaker claim that everyday and mathematical instances of (I) and
(II) can at least be avoided in favor of (IB) and (IIB). I will present some
examples, taken from everyday life and mathematical practice, of argu-
ments of the forms (I) and (II) which can be neither read as nor replaced by
instances of (IB) and (IIB).
2 EXAMPLES
Background to Example 1
The game of Mystery Cards is played thus: the red and black cards from an
ordinary deck are separated. One red and one black ‘‘mystery card’’ are set
aside face down, without having been seen by any player. The remaining
twenty-five red and twenty-five black cards are combined, shuffled, and
dealt out to the players, whose object is to guess the mystery cards. The
players take turns questioning each other. The player whose turn it is
addresses the player of his choice asking a question of the form, ‘‘Is it the
such-and-such red card and the thus-and-so black card?’’ If the player
questioned has either or both of the cards named in his hand, he must
answer ‘‘No’’; otherwise he must answer ‘‘Maybe.’’ Both question and
answer are audible to all players. If a player feels ready to guess the mystery
cards, then on his next turn, instead of asking a question he may make a
statement, saying ‘‘It’s the such-and-such red card and the thus-and-so
black card!’’ He then looks at the mystery cards. If his guess is correct, he
turns them face up and is declared the winner. If wrong, he puts them back
face down, exposes his own hand, and is disqualified from further play.
Admittedly this game is a dull one, but it exhibits in simplified form the
principle at work in several more interesting games (e.g. the one marketed
under the trade-name CLUE, the importance of which was pointed out to
me by D. K. Lewis).
Example 1. Argument
During the course of a game of Mystery Cards, Wyberg hears von Eckes ask
Zeemann, ‘‘Is it the deuce of hearts and the queen of clubs?’’ He hears
Zeemann reply ‘‘No.’’ Later in the game he manages to figure out that it is
the deuce of hearts. He argues: it isn’t both the deuce of hearts and the
queen of clubs; but it is the deuce of hearts; so it isn’t the queen of clubs. He
goes on to use this information to win the game.
Example 1. Analysis
Let p ¼ ‘‘The mystery red card is the deuce of hearts,’’ q ¼ ‘‘The mystery
black card is the queen of clubs.’’ Zeemann’s hint is no more and no less
than that (p & q). Her statement is made on purely truth-functional
grounds: she sees the queen in her hand. Her statement is not made on
the basis of any ‘‘relevance’’ between p and q: the two mystery cards were
chosen entirely independently of each other. Zeemann is justified in
denying a truth-functional conjunction, but would not be justified in
denying cotenability. Since the premise (p q) is not available to
Wyberg, his argument is an instance of (II) that can be neither read as
nor replaced by an instance of (IIB).
Had Wyberg been a relevantist, unwilling to make a deductive step not
licensed by the Anderson–Belnap systems E and R, he would have been
unable to eliminate the queen of clubs from his calculations, and would
have lost the game. A relevantist would fare badly in this game and others,
and in game-like situations in social life, diplomacy, and other areas – unless,
of course, he betrayed in practice the relevantistic principles he espouses
in theory.
Background to Example 2
Dr. Zeemann has just been awarded her degree for a dissertation in number
theory. Her main result is a proof that every natural number n has either a
certain property A(n) or a certain property B(n). As written up in her thesis,
the proof is by induction on n, as follows:
Case n ¼ 0. We show that A(0). [Here follows a proof.]
Case n ¼ 1. We show that B(1). [Here follows a proof.]
Case n 2. We assume as induction hypothesis that either A(n 1) and A(n 2), or
A(n 1) and B(n 2), or B(n 1) and A(n 2), or else B(n 1) and B(n 2).
[Here follows a proof treating each of the four cases separately.]
She remarks that the famous d’Aubel–Hughes Conjecture would imply
that B(0), whereas the equally famous conjecture of MacVee would
imply that A(1), but reports that she has no light to shed on these old
conjectures.
Commentary
Before proceeding, let us note that, following the universal practice of
mathematicians, Zeemann has taken her proof that A(0) to dispose of the
case n ¼ 0 of the general theorem that for all n, either A(n) or B(n). In other
words, she argues from the premise A(0) to the conclusion that A(0) or
B(0). This is worth mentioning because relevantistically inclined writers
have been known to claim that no one ever seriously argues from p to p or q.
Indeed, in everyday conversation we are, in R. C. Jeffrey’s words, ‘‘at a loss
to know what the motive could be’’ for someone to pass from p to the
longer and less informative statement that p or q. ‘‘Knowing the premise,
why not assert it, rather than the conclusion?’’ However, in mathematics we
often have good reason to say less than we know: We will assert less than we
could about the cases n ¼ 0 and n ¼ 1 in order to incorporate these cases in
a generalization holding for all values of n. Now the inference from A(0) to
A(0) or B (0) is only valid if ‘‘or’’ is taken as Ú rather than þ. Hence
Zeemann’s theorem must be formalized as (8n)(A(n) Ú B(n)) and not
(8n)(A(n) þ B(n)). This means that any argument of form (I) in which
the major premise is supplied by Zeemann’s theorem will be an instance of
(I) that can be neither read as nor replaced by an instance of (IB). Let us
proceed to examples.
Example 2a. Argument

Zeemann applies her work to give bounds to the number of solutions to
Tiegh’s equation, thus:
Tiegh himself has shown that the number t of solutions to his equation is 13.
Now a little elementary algebra shows that we cannot have A(t). Hence by our
main result, we must have B(t). But no n with 5 n 16 can satisfy B(n), as is clear
from some more elementary algebra. Hence t 4.
Example 2a. Analysis

This is a typical mathematical argument of the form (I). The premise
A(t) or B(t) must be represented as a truth-functional, not an ‘‘intensional’’
disjunction. The ‘‘unknown’’ t might for all we know be equal to 1, and no
‘‘relevant’’ connection has been established between A(1) and B(1) – indeed,
as Zeemann herself reports, she has been unable to establish anything
about B(1).
Example 2b. Argument

Professor Wyberg has been working for years on the celebrated conjecture
of von Eckes, but has got no further than showing that the conjecture
follows from the assumption that B(1), a result he considers not worth
publishing. Just recently he has given up work on von Eckes’ conjecture in
disgust, and has turned to other matters. In particular, he has just refuted
an old conjecture of MacVee by proving that A(1). Now he reads an
announcement of Zeemann’s result. The details of her proof are not
available – it takes years for theses to come out in print – but he recognizes
the significance of her results. In particular, they enable him to prove von
Eckes’ conjecture at last. He writes a set of notes, ‘‘A proof of von Eckes’
conjecture,’’ with the following structure: First comes his proof that A(1).
Second comes a linking passage:
And so we see that the MacVee conjecture fails. Now Zeemann has recently
announced the result that for all n, either A(n) or B(n). Hence we must have
B(1). We now proceed to put this fact to good use.
Third follows the derivation of von Eckes’ conjecture from B(1).
Example 2b. Analysis

Since what is established by Zeemann is just A(1) Ú B(1), not A(1) þ B(1), we
have here another mathematical instance of (I) that can neither be read as
nor replaced by an instance of (IB). It is a slightly atypical instance. Had he
known the details of Zeemann’s work, had he known that she actually
proves B(1) outright, Wyberg would surely have just cited this fact that B(1)
from her thesis, rather than give the roundabout argument that he did. But
this is not to say that the proof of von Eckes’ conjecture that Wyberg did
give is erroneous. One must distinguish inelegance from incorrectness, as
even most relevantists allow (p. 279). To sharpen the intuition here,
suppose that six months after Wyberg, von Eckes himself notices that his
conjecture can be derived from B(1). Suppose further that von Eckes,
unlike Wyberg, has access to a photocopy of Zeemann’s thesis, and so
knows that she has proved B(1). Von Eckes then writes a paper, ‘‘Proof of a
conjecture in number theory,’’ in which he cites the fact that B(1) from her
thesis and then proceeds to derive his old conjecture from B(1) in a manner
indistinguishable from that of Wyberg. In this situation, nobody in his
right mind would say that von Eckes had produced ‘‘the first correct proof ’’
of the conjecture; the honor of priority goes to Wyberg.
One afflicted with relevantistic scruples could not have argued as
Wyberg did, but would have had to wait for the publication of
Zeemann’s work before claiming to have settled von Eckes’ conjecture.
By that time the less scrupulous Wyberg and the better-placed von Eckes
would already be contending for priority. A follower of Anderson and
Belnap would not prosper in the world of contemporary mathematics –
unless, that is, he sometimes conveniently ‘‘forgot’’ his philosophy of
logic.
3 CONCLUSION
No doubt the reader can construct further examples. One might consider,
for instance, the case of a person who remembers that once upon a time he
was told either that p or that q, but cannot now remember which.
Investigating a bit, he quickly establishes that p, and so concludes that
q. Such examples, I submit, show that as far as negation, conjunction, and
disjunction are concerned, ‘‘classical’’ logic (and with it the whole logical
tradition from Chrysippus onwards) is far closer to commonsense and
accepted mathematical practice than is the ‘‘relevant’’ logic of Anderson
and Belnap.
One ploy the relevantist might use in trying to escape from our counter-
examples may already have occurred to the reader. What if we take the
‘‘relevance’’ required for the truth of p þ q and the falsehood of p q not as
something objective and absolute, but as something subjective and relative?
We might then say this of the Mystery Cards, for example: in objective fact
there is no connection between its being the deuce of hearts and its being
the queen of clubs, the red and black cards having been chosen separately.
In Zeemann’s mind there is no such connection, her statement that it is not
both being based solely on her knowledge that it is not the latter. But
Zeemann’s information establishes such a connection for Wyberg, so that
he is in a position to assert what she is not, namely p þ q. Hence his
argument can be represented as a case of (IIB).
I doubt such a subjectivization and relativization of ‘‘relevance’’ offers a
viable way out to followers of Anderson and Belnap. If (IB) and (IIB) are to
cover all instances of (I) and (II) in mathematics and everyday argumenta-
tion, ‘‘relevance’’ will have to be not just subjectivized but trivialized. Any
grounds for assertion p Ú q short of the simple knowledge that p, or that q,
will have to be taken as sufficient grounds for asserting p þ q: the statement
by a reliable person that she either knows p or knows q though she is not
saying which; the knowledge that p holds for m ¼ 0 and q holds for m ¼ 1,
coupled with ignorance as to whether m ¼ 0 or 1; the simple recollection
that one once knew p or knew q though one has now forgotten which. (And
paradoxically, the acquisition of more information could threaten one’s
right to assert p þ q: if one’s informant decides to provide more specific
information, if the value of m is settled, if one’s memory improves,
one may suddenly lose the right to assert p þ q.) Relevantism would reduce
to the position that (IA) is valid when and only when one’s grounds
for asserting p Ú q are something other than the simple knowledge that q.
Such a position, however, looks suspiciously like a confusion of the criteria
for the validity of a form of argument with the criteria for its utility, a
confusion of logic with epistemology.
Indeed, some writers have been willing to dismiss the whole relevantistic
movement as a simple case of confusion between the logical notion of
implication and the methodological notion of inference. The following
(unpublished) remarks of G. Harman on this point will bear quoting:
By reasoning or inference I mean a process by which one changes one’s views,
adding some things and subtracting others. There is another use of the term
‘‘inference’’ to refer to what I will call ‘‘argument’’, consisting in premises, inter-
mediate steps, and a conclusion. It is sometimes said that each step of an argument
should follow from the premises or prior steps in accordance with a ‘‘rule of
inference’’. I prefer to say ‘‘rule of implication’’, since the relevant rules do not say
how one may modify one’s views in various contexts. Nor is there a very direct
connection between rules of logical implication and principles of inference. We
cannot say, for example, that one may infer anything one sees to be logically
implied by one’s prior beliefs. Clearly one should not clutter up one’s mind with
many of the obvious consequences of things one believes.
Furthermore, it may happen that one discovers that one’s beliefs are logically
inconsistent and therefore logically imply everything. Obviously, one ought not to
respond to such a discovery by believing as much as one can. Some philosophers
and logicians [the reference is to Anderson and Belnap] have imagined that the
remedy here is a new logic in which logical contradictions do not logically imply
everything. But this is to miss the point that logic is not directly a theory of
reasoning at all.
And indeed if ‘‘relevance’’ is taken to be something subjective and relative
(according to the proposal discussed above), I do not see how the relevant-
ists could escape Harman’s charge that they confuse implication and
(useful) inference.
I do not, however, believe that the authors of Anderson and Belnap
(1975) understand by ‘‘relevance’’ something subjective. What little they tell
us about the nature of ‘‘relevance’’ (e.g. pp. 32–3, where they quote with
approval from several sources) strongly suggests that it is a matter of
meaning. Certainly their commonest charge against classical logic (first
raised on p. xxii and repeated ad nauseam) is that it ignores ‘‘intension’’ and
meaning. Meaning, however, is something that, generally speaking, will be
the same for Wyberg as it is for Zeemann. That relevance is meant to be a
semantical, and hence impersonal, notion and not a matter of individual
psychology, is further suggested by the relevantists’ criticism of T. J. Smiley
(p. 217), who is faulted for ‘‘epistemologizing’’ and ‘‘psychologizing’’ the
logical notion of entailment. Thus if the authors of Anderson and Belnap
(1975) intend by ‘‘relevance’’ something less than objective, they are highly
remiss in failing to alert readers to the fact; while if ‘‘relevance’’ is supposed
to be impersonal, then the claim that the relevantistic position is (even in a
weak sense) compatible with commonsense and accepted mathematical
practice succumbs to the counterexamples presented above.
In closing, let me reiterate that I have been concerned here solely with
the original Anderson–Belnap account of ‘‘relevant’’ logic, and with their
claim that their systems E, R, etc., are in better agreement with common
sense than is classical logic. I have not been concerned with other rationales
for developing these systems, nor with the possibility of imposing inter-
pretations on them that were not originally intended by their authors. (It
has been suggested, for instance, that some of the formalism created by
relevantists might be useful in developing a logic of ambiguity, or of truth-
in-fiction.) Workers in category theory, one of the least constructive
branches of modern mathematics, have found certain technical uses for
intuitionistic logic; but no one imagines that this vindicates Brouwer’s
philosophy of mathematics. Similarly, the discovery of serendipitous
applications of some of the formalism created by Anderson and Belnap
would not justify the claim that their logical systems are accurate formal-
izations of current mathematical practice. Still less could it justify the
abusive tone of their remarks about classical logicians.
14
Dummett’s case for intuitionism
1 TEXTS
Some philosophers approach mathematics saying, ‘‘Here is a great and

established branch of knowledge, encompassing even now a wonderfully
large domain, and promising an unlimited extension in the future. How is
mathematics, pure and applied, possible? From its answer to this question
the worth of a philosophy may be judged.’’
Other philosophers approach mathematics in a quite different spirit.1
They say, ‘‘Here is a body, already large and still being extended, of what
purports to be knowledge. Is it knowledge, or is it delusion? Only philos-
ophy and theology, from their standpoint prior and superior to that of
mathematics and science, are worthy to judge.’’ While this inquisitorial
conception of the relation between philosophy and science is less widely
held today that it was in Cardinal Bellarmine’s time, it continues to have
many distinguished advocates.
Prominent among these is Michael Dummett, who has repeatedly
advanced arguments for the claim that much of current mathematical
theory is delusory and much of current mathematical practice is in need
of revision – arguments for the repudiation, within mathematical reason-
ing, of the canons of classical logic in favor of those of intuitionistic logic.
While nearly everything Dummett has written is pertinent in one way or
another to his case for intuitionism, there are two texts especially devoted
to stating that case: his much anthologized article (Dummett 1973a) on
the philosophical basis of intuitionistic logic; and the concluding
philosophical chapter of his guidebooks (Dummett 1977) to the elements
of intuitionism. The present paper offers a critical examination of these
two texts.
1
For more on the contrast between the two approaches to philosophy of mathematics, see the editorial
introduction to Benacerraf and Putnam (1964).
256
Dummett’s case for intuitionism 257
2 DUMMETT’S CASE AGAINST PLATONISM
Dummett has remarked of his case for intuitionism that ‘‘it is virtually
independent of any considerations relating specifically to the mathematical
character of the statements under discussion. The argument involve[s] only
certain considerations within the theory of meaning of a high level of
generality, and could, therefore, just as well have been applied to any
statements whatsoever, in whatever area of language’’ (1973a, p. 226).
Hence it is best to begin an examination of his case by considering some
of his views on meaning.
Especially important for Dummett are what I will call neutrally theories
of language of the first type. On a theory of this type, the meaning of a
sentence is identified with the conditions for correctness of the sentence (or
pedantically: of an assertion made by uttering the sentence) as a represen-
tation of reality. A speaker’s ability to use the sentence is explained by
reference to his grasp of these correctness conditions. Correctness may be
conceived of in more than one way, and hence more than one subtype
within the first type of theory of language is possible.
Especially important for Dummett is the distinction between those
conceptions on which correctness is, and those on which it is not, some-
thing always at least in principle potentially recognizable by human beings.
The best-known theory operating with a conception of correctness as
always recognizable is the intuitionistic proof-conditional theory of meaning
for the mathematical part of language. (The provability of a mathematical
conjecture is recognizable by discovering a proof.) The generalization of
this theory to the whole of language would be a verification-conditional or
verificationist theory.
The best-known correctness conception on which correctness is some-
times recognizable, sometimes unrecognizable, is the usual conception of
truth. (On the usual conception, the truth of a mathematical conjecture need
not imply the existence of a proof of the conjecture or any other means of
recognizing the conjecture as true.) To avoid ambiguities, I will call truth-as-
usually-conceived verity. Many distinguished philosophers have advocated
verity-conditional or verist theories of meaning. Dummett calls the special-
ization of such a theory to the mathematical part of language a Platonist
theory (though I imagine that he would be hard pressed to locate a passage in
the Republic or the Timaeus where such a theory is taught).
How can graspable correctness conditions be assigned to each of the
indefinitely many sentences of a language? The only answer that immedi-
ately suggests itself is: inductively. The theories of language of the first type
considered by Dummett take the sentences of the language to fall into a
hierarchy of degrees of complexity, with the correctness conditions for
those of higher degree being determined inductively from the correctness
conditions for those of lower degree. For example, such theories include an
induction clause indicating how the correctness conditions of a disjunction
are determined from those of its disjuncts. On an intuitionist or verifica-
tionist theory this clause takes the form:
(1) A proof (or verification) of a disjunction consists in the specification of one of
its disjuncts together with a proof (or verification) of that disjunct.
On a Platonist or verist theory this clause takes the form:
(2) A disjunction is true if and only if at least one of its disjuncts is true.
On any theory of language of the first type there is an external standard
against which rules of implication are to be judged. A rule is acceptable if
and only if it preserves correctness, leading in all instances where the
premises are correct to a conclusion that is correct. It is by appeal to such
induction clauses as (1) and (2) above that one can seek to demonstrate that
certain rules are correctness-preserving or sound.
Given the Platonist theory of meaning, the usual soundness proof for
classical logic established that all the rules of that logic are acceptable.
Given the intuitionist theory of meaning, the usual soundness proof for
intuitionistic logic establishes acceptability for all intuitionistic rules, but
not for all classical rules: the acceptability of rules depending on the laws of
double negation or excluded middle is doubtful when such rules are
applied to sentences for which an effective decision procedure is lacking,
such as those involving unbounded quantification over an infinite domain.
Dummett’s strategy is to argue for the repudiation of classical logic in favor
of intuitionistic logic by arguing for the repudiation of Platonist (or more
generally: verist) theories of meaning in favor of intuitionist (or more
generally: verificationist) theories.
On a verist theory, sentences whose correctness need not be recognizable
may be said to represent transcendent features of reality, while sentences
whose correctness must be recognizable may be said to represent immanent
features of reality. Dummett, like Brouwer, denies that any sentence of any
possible language can represent transcendent features of reality. But where
Brouwer sees this denial as expressing a limitation on reality, Dummett sees
it as expressing a limitation on language.
Dummett’s case against verism rests on principles summed up in the
slogan that meaning is use. He offers various formulations of these
principles, some in terms of meaning, others in terms of understanding,
some telling us what these consist in, others through what they are exhaus-
tively manifested. The following are typical (1973a, pp. 216, 217):
The meaning of . . . a statement cannot be, or contain as an ingredient, anything
which is not manifest in the use made of it, lying solely in the mind of the
individual who apprehends that meaning: if two individuals agree completely
about the use to be made of the statement, then they agree about its meaning. The
reason is that the meaning of a statement consists solely in its rôle as an instrument
of communication between individuals . . . An individual cannot communicate
what he cannot be observed to communicate . . .
[T]here must be an observable difference between the behavior or capacities
of someone who is said to have . . . knowledge [of the meaning of an expression]
and someone who is said to lack it. Hence it follows . . . that a grasp of the meaning
of a . . . statement must, in general, consist of a capacity to use that statement in
a certain way, or to respond in a certain way to its use by others.
Shared by all such formulations is an association of meaning with public
and observable use of language as a vehicle of communication, and a
dissociation of meaning from private and hidden use of language as a
vehicle of thought.2 Likewise, any association of meaning or understanding
with something in the conscious or unconscious mind, or in the structure
or functioning of the brain, is rejected. Though he himself avoids the label,
Dummett may be called a behaviorist in his approach to meaning, provided
this label is understood in a broad enough sense to cover not only the
stimulus-response behaviorism of Skinner, but also the logical behaviorism
of Ryle.
A thoroughgoing behaviorist will require that any apparatus posited by a
semantic theory must be identified or directly correlated with some isolable
features of publicly observable verbal behavior. As Dummett formulates it,
the behaviorist demand is that there must be a ‘‘one-one correspondence
between the details’’ of the apparatus posited by a semantic theory and
‘‘observable features of the phenomenon’’ (1977, p. 377). Dummett
describes rejection of this behaviorist demand as one form that the rejection
of the principle that meaning is use might take.
The great majority of contemporary linguists reject this behaviorist
demand as likely to lead only to sterility and stagnation in semantics.
The great majority of contemporary linguists posit in their semantic
theories an apparatus neither identified nor directly correlated with any
2
For more on the contrast between the two sorts of use of language, see the opening paragraphs of
Harman (1982).
set of isolable features of publicly observable verbal behavior. The original
version of Chomsky’s semantic theory, for example, posited an apparatus
of deep structures. Chomsky and the great majority of contemporary
linguists claim that the apparatus posited in their semantic theories is
psychologically real, represented in ways as yet undiscovered in the mind
or brain.3 But they do not claim the apparatus to be directly represented in
behavior.
Thus the principle that meaning is use, on which Dummett bases his
case for a revision of current mathematics, itself already amounts to a
demand for a revision of current linguistics. For this reason Dummett’s
arguments for the principle are of interest apart from their role in his
case for intuitionism. These arguments have often been criticized by
Davidsonians. Here they will be criticized from a viewpoint closer to that
of the Chomskians.
Two arguments for the principle that meaning is use are to be found in the
texts under examination. They are versions of what have come to be called
the acquisition and manifestation arguments. The first (1973a, p. 217) begins:
[O]ur proficiency in making the correct use of the statements and expression of the
language is all that others have from which to judge whether or not we have
acquired a grasp of their meanings. Hence it can only be in the capacity to make a
correct use of the statements of the language that a grasp of their meaning, and of
those of the symbols and expressions which they contain, can consist.
The rather familiar line of this opening4 alerts us that the argument is going
to turn on how an observer Y, say a teacher, can judge that a speaker X, say a
learner, attaches the standard meaning to an expression E. Perhaps before
proceeding further it would be well to review schematically the accounts of
such judgments offered by behaviorists, on the one hand, and by those
anti-behaviorists who associate meaning or understanding with a state of
the mind or brain, on the other.
On the behaviorist account, for X to attach the standard meaning to E is
for X to be able to use E standardly. On this account, Y’s judgment that X
3
A minority of linguists adopt the position advocated in Soames (1985), regarding such claims of
psychological reality as at best premature, but nonetheless insisting on the legitimacy of introducing
an apparatus of deep structures or the like in semantic theory, despite behaviorist objections.
4
Charles Chihara has pointed out in conversation the parallelism between the argument of Dummett
just quoted and the notorious argument of Norman Malcolm against the conception of dreams as
mental or neural activity taking place at specific times during sleep. Malcolm’s argument may be
paraphrased: our telling stories when we wake up is all that others have from which to judge whether
we have dreamt. Hence it can only be in the disposition to tell stories when we wake up that having
dreamt can consist.
attaches the standard meaning to E is a simple inductive inference from the
premise that X has been able to use E standardly in all observed instances to
the conclusion that X will be able to use E standardly in all instances. The
anti-behaviorist account is much more complex.
The first step towards an anti-behaviorist position is acceptance of the
general psychological principle that different people similar in their out-
ward behavior are normally similar also in the inward mental or neural
states that causally underlie behavior. The second step is acceptance, as a
special linguistic instance, of the hypothesis that there exists a mental or
neural state S(E) normally causally underlying the ability to use E stand-
ardly. Thus far a behaviorist may or may not go along. Where the
behaviorist must refuse to follow is at the anti-behaviorist’s third step,
the identification of attaching the standard meaning to E with being in the
state S(E) rather than directly with being able to use E standardly.
To appreciate the rather subtle distinction here, imagine an abnormal
case where a native English speaker X is able to communicate with a native
Chinese speaker W only because X has implanted inside his skull a mini-
supercomputer programmed to translate back and forth between English
and Chinese sentences. There may be no ‘‘difference in the behavior or
capacities’’ of X and W that is observable to those of us lacking telepathic
powers and X-ray vision. X and W may ‘‘agree completely about the use to
be made’’ of various Chinese words and phrases. Yet on account of the
absence of ‘‘an ingredient . . . lying solely in the mind’’ or brain, the anti-
behaviorist will deny that X attaches the standard meanings, or any mean-
ings at all, to those words and phrases.
The status of such science-fiction examples is in itself a matter of slight
importance, but there are more important further differences between
behaviorists and anti-behaviorists. One who identifies attaching the stand-
ard meaning to E with being the hypothetical state S(E) will presumably be
willing to entertain hypotheses about the composition and components of
S(E) and to permit a theory of the standard meaning of E to posit an
apparatus correlated with these hypothetical components of S(E). But while
S(E) itself is normally correlated with the ability to use E standardly, there
is no reason to suppose its components to be directly correlated with
any ‘‘isolable, though interconnected, practical abilities’’ (Dummett 1977,
p. 377). Hence the anti-behaviorist rejection of the requirement that the
apparatus posited in a theory of the standard meaning of E must be directly
correlated with isolable features of publicly observable verbal behavior,
which we have already seen to be the issue dividing Dummettians and
Chomskians.
On the anti-behaviorist account, for X to attach the standard meaning to
E is for X to be in a mental or neural state S(E) posited to underlie, in
normal cases, the ability to use E standardly. On this account, Y’s judgment
that X attaches the standard meaning to E rests on: (a) the evidence for the
presupposition that there exists a mental or neural state underlying, in a
normal case, the ability to use E standardly; (b) the evidence that X’s case is
a normal one; and (c) the evidence that X has been able to use E standardly
in all observed instances. The evidence (c) is the only evidence cited in the
behaviorist account. The evidence (b) may consist in no more than the
absence of evidence that X’s case is an abnormal one. The evidence (a) may
consist in no more than the evidence for the general psychological principle
that different people who are similar in their outward behavior are nor-
mally similar also in their inward mental and neural states.
A behaviorist, of course, may question the strength of the evidence for
this general psychological principle. Quineans, for example, have claimed
that different people identical in their outward behavior may be ‘‘like
different bushes trimmed to resemble identical elephants.’’ Dummett’s
objections to anti-behaviorism, however, do not take this form, being a
priori and philosophical rather than a posteriori and psychological.
Returning now to the argument whose opening was quoted above, it
continues (1973a, pp. 217–18):
To suppose that there is an ingredient of meaning which transcends the use that is
made of that which carries the meaning is to suppose that someone might have
learned . . . [to] behave in every way like someone who understands the language,
and yet might not actually understand, or understand it only incorrectly. But to
suppose this is to make meaning ineffable, that is, in principle incommunicable. If
this is possible, then no one individual ever has a guarantee that he is understood
by any other individual; for all he knows, or can ever know, everyone else may
attach to his words . . . a meaning quite different from that which he attaches to
them. A notion of meaning so private to the individual is one that has become
completely irrelevant to mathematics as it is actually practised, namely as a body of
theory on which many individuals are corporately engaged, an enquiry within
which each can communicate his results to others.
In the earlier parts of this passage, Dummett claims that if there is

anything more to X’s attaching the standard meaning to E than X’s
being able to use E standardly, then Y can never know or have a guarantee
that X attaches the standard meaning to E. Two comments are called for.
First, on both the anti-behaviorist and the behaviorist accounts, Y’s judg-
ment that X attaches the standard meaning to E is an inductive inference.
On the behaviorist account, it is an inference from a limited number of
observed instances of use to an unlimited number of possible future
instances of use. No inductive inference can provide certain knowledge
or an indubitable guarantee.
But if the possibility of skeptical doubt and uncertainty somehow
undermines a theory of meaning, it must undermine not just the anti-
behaviorist theory, on which understanding is something mental or neural
transcending behavior, but also the behaviorist theory, on which under-
standing is an open-ended behavioral ability or capacity, transcending any
finite number of its manifestations. This point has been mentioned in
passing by Susan Haack (1974, pp. 107–8) and developed at length by
Crispin Wright (1980, pp. 123–8).
Second, it is far from obvious that the impossibility of skepticism-proof
guaranteed knowledge in any way undermines a theory of meaning. In the
later parts of the passage under examination, Dummett seems to try to
draw out damaging consequences from the absence of such guaranteed
knowledge. He seems to claim that its absence somehow makes communi-
cation between mathematicians impossible, and hence makes mathematics
as an activity involving communication impossible. Surely such a claim
would be mistaken. For whether mathematicians X and Y succeed in
communicating through their use of expression E surely depends only on
whether X and Y do in actual fact attach the same meaning to E, and not on
whether they possess skepticism-proof guaranteed knowledge that they do
so. One hesitates to accuse a distinguished authority on modal logic of
arguing from àp to àp, but Dummett does almost seem to wish to
move from the (epistemic) possibility that X and Y do not succeed in
communicating to the (metaphysical) impossibility of X and Y succeeding
in communicating.
To avoid fallacies in the ‘‘acquisition’’ argument we must distinguish the
claim that a language learner cannot come to attach the standard meaning
to an expression from the claim that no one can have guaranteed knowl-
edge that the language learner has attached the standard meaning to the
expression. Dummett fails to show that the former follows from the anti-
behaviorist approach to meaning; and while the latter may follow, it is not
obviously unacceptable. This last point has been noted by Dag Prawitz,
who writes (1977, p. 10) of Dummett’s argument that:
One could contest it by arguing that when we learn a language by seeing how its
sentences are used, we only get some hints about their meaning. The samples of
use with which we are presented never completely determine the meaning but only
enable us to form some theories or hypotheses about the meaning. (The fact that
we nevertheless agree rather well about meaning could perhaps be explained by
reference to a genetic disposition to see certain kinds of patterns and hence to form
certain kinds of theories upon seeing a few examples.) Such a view would entail
that we could never be sure that we knew the meaning of a sentence; a new
unexpected use of it could show us that we had misunderstood the meaning and
would force us to revise our theory. And to some extent this may be a correct
picture of our situation.
Dummett’s other argument calls for less comment. It runs as follows (1977,
p. 217):
Now knowledge of meaning . . . is frequently verbalisable knowledge, that is,
knowledge which consists in the ability to state the rules in accordance with which
the expression or symbol is used . . . But to suppose that, in general, a knowledge of
meaning consisted in verbalisable knowledge would involve an infinite regress: if a
grasp of the meaning of an expression consisted, in general, in the ability to state its
meaning, then it would be impossible for anyone to learn a language who was not
already equipped with a fairly extensive language. Hence that knowledge which, in
general, constitutes the understanding of language . . . must be implicit knowl-
edge. Implicit knowledge cannot, however, meaningfully be ascribed to someone
unless it is possible to say in what the manifestation of that knowledge consists:
there must be an observable difference between the behavior or capacities of
someone who is said to have that knowledge and someone who is said to lack it.
Hence it follows, once more, that a grasp of the meaning of a . . . statement must,
in general, consist of a capacity to use that statement in a certain way, or to respond
in a certain way to its use by others.
In the first part of this passage, Dummett invokes infinite-regress consid-
erations to establish that knowledge of meaning is not ‘‘in general’’ verbal-
izable, and even that it is ‘‘in general’’ unverbalizable. If one tries to restate
the argument without the use of the puzzling phrase ‘‘in general,’’ then one
finds that all the infinite-regress considerations seem to establish is that for
the part of language learned first, the most elementary part, knowledge of
meaning is unverbalizable. It would then seem that any behavioristic
conclusions drawn from the argument as a whole ought to be restricted
to this part of language: nothing follows about the part of language learned
later, the more advanced part.5
In the second part of the passage under examination, Dummett invokes
a premise about unverbalizable knowledge to reach a conclusion about
knowledge of meaning. To avoid equivocation, we must distinguish four
claims here, some stronger, some weaker, some more general, some more
specific:
5
Paul Benacerraf suggested to me in general terms that Dummett’s arguments might have force for one
part of language but not another.
(a) Ascriptions of knowledge of meaning must be supported by appeal to observ-
able evidence.
(b) Knowledge of the meaning of an expression consists in no more than the
ability to use it in a certain way.
(c) Ascriptions of implicit knowledge must be supported by appeal to observable
evidence.
(d) Implicit knowledge consists in no more than the ability to behave in a certain way.
The anti-behaviorist rejects (b) but accepts (a). (The anti-behaviorist acc-
ount of the observable evidence supporting an ascription of knowledge of
meaning has been reviewed schematically above.) Since the conclusion
Dummett desires is (b) and not (a), the premise he requires is (d) and not
(c), even if his own formulations are less than unequivocal (1973a, p. 217;
1977, p. 376):
Implicit knowledge cannot, however, meaningfully be ascribed unless it is possible

to say in what the manifestation of that knowledge consists . . .
[A]n ascription of implicit knowledge must always be explainable in terms of
what counts as a manifestation of that knowledge, namely the possession of some
practical ability.
The anti-behaviorist will argue that if – as the implanted-computer exam-

ple suggests – (b) is false, then this implies that (d) is false. Dummett,
however, invokes the controversial premise (d) without supporting consid-
erations, as if it were self-evident. For this reason anti-behaviorists may with
some justice reject the ‘‘manifestation’’ argument as manifestly question-
begging.
Circularities in Dummett’s arguments for behaviorism do not, however,
deprive his case against verism of all its force. Dummett’s complaint against
verism comes down to this, that verists have returned no answer, formulated
in behavioral terms, to the following question: in what can a grasp of the
correctness conditions for a sentence consist if the correctness of that
sentence need not even in principle be potentially recognizable by human
beings? So long as one insists that verists must return an answer, formulated
in psychological terms, to the foregoing question, one will have to sympathize
with Dummett’s complaint against verism, even if one does not sympathize
with his behaviorism, and adopts the approach of introspective or of phys-
iological rather than behavioral psychology. For the best-known advocates of
verism either return no answer at all to Dummett’s question; or worse, they
answer that to grasp the truth conditions of a sentence is to associate
with that sentence the set of possible worlds where it is true, but do not
explain how a mind or brain confined to the actual world can effect such
an association.
One might also sympathize with Dummett’s rejection of verism for any of
a number of reasons quite unlike Dummett’s own, for example, on account
of the apparent conflict between truth-conditional theories of meaning and
theories of truth in the style of Tarski or Kripke. This route to a rejection of
verism is worth mentioning here because Dummett himself sometimes
touches on it tangentially in his writings. In one paper (Dummett 1959) he
notes that there appears to be a conflict between the view that a biconditional
like (2) above constitutes an account of the meaning of ‘‘or’’ and the view that
it constitutes part of an account of the meaning of ‘‘true,’’ a point that Tarski
has also discussed in one of his papers (1944) as an anonymous objection
against his theory. Truth-conditional theories of meaning appear to regard
truth as a primitive concept, possession of which is a prerequisite for any
language-learning, while a theory like Kripke’s appears to regard the concept
of truth as one acquired fairly late in the process of language-learning, when
the learner has acquired a fairly extensive ability to talk of persons, places,
and things, and is beginning to learn to talk of talk.6
For any of a number of reasons, good or bad, like or unlike Dummett’s,
many philosophers of language now reject verism. In inveighing against
verism, Dummett is to a large extent preaching to the converted. Dummett
himself recognizes that Wittgenstein, for one, and Quine, for another, have
rejected verism. Yet somewhat surprisingly he can be found writing: ‘‘[T]he
idea that a grasp of meaning consists in a grasp of truth-conditions was [in
1959] and still is [in 1978], part of the received wisdom among philoso-
phers’’ (1978, p. xxi). A poll of my own department convinces me that
Dummett is wrong here.7 How could a theory rejected by a constellation of
such luminaries as Wittgenstein, Tarski, Quine, Kripke, and Harman, not
6
A not unrelated reason for rejecting truth-conditional theories of meaning is advanced in Harman
(1982):
Davidson, Lewis, and others have argued that an account of the truth conditions of sentences of a
language can serve as an account of the meanings of those sentences. But this seems wrong. Of course,
if you know the meaning in your language of the sentence S, and you know what the word ‘‘true’’
means, you will also know something of the form ‘‘S is true if and only if . . .’’; for example, ‘‘‘Snow is
white’ is true if and only if snow is white’’ or ‘‘‘I am sick’ is true if and only if the speaker is sick at the
time of utterance’’. But this is a trivial point about the meaning of ‘‘true’’, not a deep point about
meaning.
For more on the philosophical significance (or lack of it) of the concept of truth, see Soames (1984).
7
Saul Kripke has suggested that Dummett’s statement may be accurate as an account of local
conditions at Oxford. But surely it would bespeak a certain parochialism to confuse ‘‘fashionable
among Oxford philosophers’’ with ‘‘received among philosophers generally.’’
to mention Dummett himself, be considered ‘‘received wisdom among
philosophers’’?
For many philosophers what is most puzzling about Dummett’s case for
intuitionism will not be the question arising in his case against Platonism:
(a) Why are we supposed to reject verist semantics?
but rather the question arising in his case against formalism:
(b) How is the rejection of verist semantics supposed to lead to the rejection of
classical mathematics?
This latter question will now be taken up.
3 DUMMETT’S CASE AGAINST FORMALISM
Criticism of a theory of language may take any of three forms. The

ordinary descriptive critic advances evidence that we do not actually speak
a language of the sort the theory depicts, whether or not we ought to. The
radical descriptive critic advances evidence that we could not possible speak
a language of that sort, so that the question whether we ought to do does
not arise. The prescriptive critic (or advocate) advances motives why we
ideally ought not (or ought) to speak such a language, whether or not we
currently do.
Dummett is a philosopher not primarily renowned for the clarity of this
prose, and the interpretation of his works will always be a matter of
controversy, not least because he declines to distinguish explicitly factual
or descriptive from normative or prescriptive considerations. As I inter-
preted it in x2, Dummett’s criticism of Platonist or verist theories of
meaning was descriptive and radical, claiming that we could not possibly
possess (because we could not possibly acquire or manifest) a grasp of
correctness conditions that are transcendent rather than immanent. As
I interpret it, Dummett’s advocacy of intuitionist or verificationist theories
of meaning is prescriptive. For surely he cannot claim that such theories
describe and explain the actual, current patterns of usage of any but a tiny
minority of (Dutch) mathematicians. As I interpret it, Dummett’s position
is that no theory of language of the first type, verist or verificationist,
provides a description and explanation of the actual, current patterns of use
of the overwhelming majority of mathematicians.
Looking beyond theories of the first type, especially important for
Dummett are what I will call neutrally theories of language of the second type
or dualist theories. Such theories depict language as containing two different
kinds of sentences: primary sentences, possessing decidable correctness
conditions, and secondary sentences, lacking correctness conditions.
The best-known theories of this type are the theory of mathematical
language associated with the name of Hilbert (1925), and the theory of
scientific language associated with the name of Quine (1951b). On the
former theory, primary sentences, called inhaltlich, consist of simple arith-
metical sentences decidable by computation; secondary sentences, called
ideal, may contain non-computational mathematical vocabulary (e.g. that
of set theory). On the latter theory, primary sentences, said to ‘‘lie on the
periphery,’’ consist of simple empirical sentences decidable by observation;
secondary sentences, said to ‘‘lie in the interior,’’ may contain non-
observational scientific vocabulary (e.g. that of quantum theory). Hilbert’s
overall position in philosophy of mathematics is usually called formalism.
Quine’s overall position in philosophy of science is usually called holism.
Both labels have been used in the literature with so many different con-
notations that they are perhaps best avoided. Dummett repeatedly stresses
the affinities between Hilbert and Quine (1973a, p. 219; 1977, p. 397).8
On either theory, primary sentences are distinguished from secondary
sentences by their restricted vocabulary. Not only is their non-logical,
mathematical or scientific, vocabulary restricted to be computational or
observational, but also their logical vocabulary is restricted. According to
the precise version of the theory being considered, primary sentences are
required to be either atomic, containing no logical particles at all, or else to
be quantifier-free, containing only connectives.
On either theory, secondary sentences serve merely as intra-linguistic
instruments for deducing primary sentences, and not as representations of
any extra-linguistic reality. Computational and observational facts are
represented by primary sentences. It is claimed that the scope, accuracy,
and efficiency of the representation of computational and observational
facts is enhanced by the presence in the language of sentences that do not
themselves represent such facts but that can be used to deduce sentences
that do.
On either theory, the question arises how the ability to use secondary
sentences can be learned. For example, how is the ability to use disjunctions
8
It is something of an oversimplification to describe Quine as a dualist, inasmuch as he often indicates
that he regards the distinction between the observational periphery and the theoretical interior as a
matter of degree rather than kind. But for Dummett the similarities between Quine’s position and
that of the prototypical dualist Hilbert are more important than such differences.
in deductions acquired? On a verist or verificationist theory the answer is:
by grasping the correctness conditions (1) or (2) above. This answer is not
available on a dualist theory, and no explicit answer is offered in Hilbert
(1925) or Quine (1951b). There is, however, an answer that immediately
suggests itself, namely, that the ability is acquired by directly grasping such
rules of implication as the following:
(3) A disjunction is implied by each of its disjuncts.
A disjunction implies whatever is implied by each of its disjuncts. In
other words, the ‘‘meaning of the logical constants’’ – if what determines
their use may be called their ‘‘meaning’’ – consists ‘‘directly in the validity
or invalidity of possible forms of inference’’ (Dummett 1977, p. 363). It
seems to be this answer that Dummett associates with dualism. It is worth
mentioning that quite apart from any general dualist views, the specific
view that an account of the ‘‘meaning’’ of the logical particles is best given
in terms of such implication conditions as (3) rather than such truth
conditions as (2) has had many distinguished advocates, including (accord-
ing to Prior 1960) several of Dummett’s Oxford colleagues.
Dualist theories may be called semi-verificationist. They are not verifi-
cationist in the strict sense, since some sentences are not assigned correct-
ness conditions. They are verificationist in a loose sense, since all sentences
that are assigned correctness conditions are assigned decidable, recogniz-
able, verifiable correctness conditions.
May dualist theories of language be called theories of meaning?
Dummett sometimes takes ‘‘(theory of) meaning’’ in a broad sense, and
insists on an affirmative answer (1973c, p. 378):
A model of language may also be called a model of meaning, and the importance
of the conception of language sketched at the end of ‘‘Two Dogmas’’ was that it
gave in succinct form the outline of a new model of meaning. It is well known that
some disciples of Quine have heralded his work as allowing us to dispense with the
notion of meaning. But even the most radical of such disciples can hardly propose
that we may dispense with the notion of knowing, or having mastery of, a
language; and there is nothing more that we can require of a theory of meaning
than that it give an account of what someone knows when he knows a language . . .
[W]hatever warrant there may be for asserting that Quine has destroyed the
concept of meaning does not appear from the ‘‘Two Dogmas’’ model of language
taken by itself. That has merely the shape of one theory or model of meaning
among other possible ones.
However, Dummett sometimes takes ‘‘(theory of) meaning’’ in a narrow
sense, and insists on a negative answer (1973b, p. 309):
The theory of meaning, which lies at the foundation of the whole of philosophy,
attempts to explain the way in which we contrive to represent reality by means of
language. It does so by giving a model for the content of a sentence, its represen-
tative power. Holism is not, in this sense, a theory of meaning: it is the denial that a
theory of meaning is possible.
When in the narrow, negative mood (as throughout 1977) Dummett is
prepared to join Brouwer and Heyting in declaring that many of the
sentences of classical mathematics are ‘‘incoherent’’ and ‘‘unintelligible.’’
This sounds odd. For Dummett can hardly deny that the sentences of
classical mathematics possess a definite usage within pure mathematics and
a definite utility through applied mathematics. How can he, as a professed
adherent of the slogan that meaning is use, then deny that those sentences
have a meaning? Taken literally, the slogan implies that a sentence having a
use thereby has a meaning. The answer, the explanation of the oddity, is, of
course, that Dummett, as we have already seen, adheres to the slogan that
meaning is use only in a non-literal, almost idiosyncratic, sense.
The narrow, negative terminology need not be misleading provided the
following point is never forgotten: When Dummett says that many sentences
of classical mathematics lack meaning-in-the-narrow-sense, he is only saying
(in highly emotive terms) that the theory of meaning-as-conditions-for-
correctness-as-a-representation-of-reality is inapplicable to those sentences.
This factual claim about how language is cannot by itself imply any norma-
tive claim about how mathematics ought to be. Some extra, tacit premise of a
normative or prescriptive character is needed. As I interpret it, Dummett’s
criticism of dualism is prescriptive, and rests on an extra, tacit premise of
anti-instrumentalism or representationalism, according to which every sen-
tence of a language ideally ought to play a representational rather than a
merely instrumental role. Thus when Dummett writes: ‘‘A sentence is a
representation of some facet of reality’’ (1973b, p. 309), according to my
interpretation he has not quite accurately reflected his own view: ‘‘ought to
be’’ ought to be where ‘‘is’’ is in the quoted formulation.
The requirement of representationality is, of course, accepted by Platonists,
who hold, in opposition to intuitionists and formalists, that this require-
ment is already met by our current language. Representationalism unites
Platonists and intuitionists in opposition to formalists, much as behaviorism
unites Quineans and Dummettians in opposition to Davidsonians and
Chomskians. (There are, however, important differences between the
Harvard behaviorism of Skinner or Quine and the Oxford behaviorism of
Ryle or Dummett. Moreover, it is not obvious that a verificationist or dualist
must be a behaviorist.)
Consider the situation of a philosopher initially sympathetic, for behav-
ioristic or other reasons, to a naive descriptive verificationism like that of
the early positivists, who comes to appreciate that such a theory is inade-
quate as an account of the actual, current patterns of use in our language.
One response would be to revise the theory to fit the facts of language,
perhaps falling back to a semi-verificationist, dualist position. Another
response would be to require a revision of language, to fit the norms of the
theory. Quine and Dummett exemplify these two responses. Dummett’s
against Quine stands or falls with the success or failure of his attempts to
motivate the requirement of representationality.
In both the texts under examination Dummett discusses, by way of
offering such motivation, the following worry about languages of the sort
depicted by dualist theories: in such a language, there is a threat of
deducing incorrect primary sentences by means of secondary sentences.
In one text, the worry seems to be that incorrect primary sentences
might be deduced from (theories composed of) secondary sentences
(1973a, p. 220):
With what right do we feel assurance that the observational statements deduced
with the help of complex theories, mathematical, scientific and otherwise, embed-
ded in the interior of the total linguistic structure, are true, when these observation
statements are interpreted in terms of their stimulus meanings? To this the holist
attempts no answer, save a generalised appeal to induction: these theories have
‘‘worked’’ in the past, in the sense of having for the most part yielded true
observation statements, and so we have confidence that they will continue to
work in the future.
This worry, or rather, the demand for a guarantee against it, is easily
dismissed. Of course there is a threat that a scientific theory about, say,
black holes or quarks may have incorrect observational consequences. We
have seen such threats realized many times in the history of science, and we
have known since the time of Hume that there can be no guarantee against
them. And of course there is a threat that a mathematical theory about, say,
l-adic cohomology or o-complete ultrafilters may have incorrect and even
inconsistent computational consequences. We have seen such threats real-
ized a few times in the history of mathematics (in connection with
infinitesimal calculus and naive set theory), and we have known since the
work of Gödel that there can be no guarantee against them. If and when
such threats are again realized, we will, as we always have in the past on such
occasions, revise our theories. But why even then, let alone now, revise our
logic? It is a delusion to imagine that a preemptive change of logic could
provide a guarantee against such threats, unless, indeed, the new logic were
so restrictive as to make the formulation of any non-trivial theories
impossible.
In the other text, the worry seems to be that incorrect primary sentences
might be deduced from correct primary sentences by way of secondary
sentences. On a sequential formulation of logic, this is the worry that a
sequent
( ) A1, . . . , An ) B
with the Ai primary and correct and B primary and incorrect, might be
deducible by pure classical logic, if secondary sentences are allowed to
appear in the deduction. If only primary sentences are allowed to appear in the
deduction, there is nothing to worry about, since primary sentences are
decidable, and not even intuitionists doubt the trustworthiness of classical
logic as applied to decidable sentences. As Dummett says, it would be a
‘‘severe defect’’ in the classical rules of implication if by means of them ‘‘we
can construct a deductive chain leading from correct premises to an
incorrect conclusion’’ (1977, p. 364). He reminds us that even on a dualist
theory there is an external standard against which the acceptability of rules
of implication is to be judge. A rule is acceptable only if it is sound, only if it
preserves correctness in all instances where the notion of correctness is appli-
cable, that is, in all instances where the premises and conclusion are all
primary sentences. (Thus even if the ‘‘meaning’’ of the logical particles is
given by implication conditions, not just any old particles and conditions
will do. This point has been illustrated by Prior (1960). Dummett does not
claim that classical logic is, in this sense, demonstrably unsound (as is Prior’s
‘‘tonk’’ logic). What worries Dummett is that classical logic seems to be not
demonstrably sound. He desires a guarantee of soundness, or what would, as
we have seen above, be sufficient for this, a guarantee of conservativeness, a
guarantee that the addition of the secondary sentences to the language does
not permit the deduction of any sequences ( ) involving only primary
sentences that were not deducible already (1977, pp. 363–4).
Dummett seems to hold that such a guarantee could only be provided by
a semantic soundness or conservativeness proof, and that such a proof or
‘‘justification’’ will be available only if we revise the language and extend the
assignment-of-correctness-conditions or ‘‘interpretation’’ to ‘‘all statements
or formulas with which we are concerned’’ (1977, p. 220). Dummett seems
to overlook the possibility of a purely syntactic proof of soundness or
conservativeness. As Richard Grandy has pointed out in a perceptive review
(Grandy 1982), just such a guarantee as Dummett seems to desire is
provided by the famous Cut-elimination Theorem of Gentzen, according
to which any sequent ( ) that has a deduction at all has a deduction in
which no symbols occur that do not occur in ( ) already. Moreover, though
Gentzen’s theorem is about classical logic, Gentzen’s proof is given in
intuitionistic metamathematics. Thus the threat that worries Dummett
seems elusive, to say the least.
In any case, the guarantee he desires would be intangible, on his own
admission. For it is precisely the theme of Dummett (1973b) that no
‘‘justification of deduction’’ or soundness of proof can be ‘‘suasive,’’ that
is, can persuade anyone sincerely in doubt as to the soundness of the logic.
For any such proof, being a proof, would itself use logic.
In opposition to Kreisel, Dummett is concerned to argue for intuition-
ism not as one legitimate form of mathematics among others, but as the sole
legitimate form (1977, p. 360). Dummett is concerned to argue for a
revision amounting not to a reform, but to a revolution, in mathematics.
Any revolution involves costs that the benefit of an intangible guarantee
against an elusive threat of unsoundness seems insufficient to outweigh. It
seems that its desirability as a means toward the end of guaranteeing
soundness is not a consideration sufficient to motivate the requirement
of representationality.
Dummett might, of course, rest his case against formalism on the
desirability of representationality as an end in itself. That he indeed
values representationality highly for its own sake is suggested by his apply-
ing the uncalled-for emotive term ‘‘unintelligible’’ to sentences that he
knows perfectly well how to use, but that happen to lack meaning-in-the-
narrow-sense-of-conditions-for-correctness-as-a-representation-of-reality.
Dummett’s value judgment might, however, be questioned by many
mathematicians.
It hardly needs saying that the requirement of representationality will be
rejected by the many pure mathematicians who value mathematics as an
art. From their point of view, there is no sole legitimate form of mathe-
matics. A mathematician may work now in intuitionistic, now in classical
mathematics, just as a painter may work now in a representational, now in
an abstract style. Personal taste will dictate how much time is devoted to
each, though it may be said that the overwhelming majority of mathema-
ticians find more beauty in the classical than in the intuitionistic style.
What does perhaps need saying is that the requirement of representa-
tionality may also be questioned by the many applied mathematicians who
value mathematics for its contribution, through science, to the theoretical
prediction and practical control of experience. From their point of view, it
is essential that language should contain some sentences to serve as records
or predictions of experience, as representations of empirical reality. But
beyond this it seems wisest to accept the advice of Carnap (1950) and be
‘‘tolerant in permitting linguistic forms.’’ It is questionable whether the
scope, accuracy, and efficiency of applications to the empirical world
would be enhanced by imposing the restriction that all sentences must
play a representational rather than a merely instrumental role in language.
Many physicists, mathematicians, logicians, and philosophers have sug-
gested precisely the contrary: that intuitionistic restrictions on mathe-
matics would be detrimental to applications. Such views as the following
are often voiced (Manin 1977, pp. 172–3):
[C]onstructivism is in no sense ‘‘another mathematics’’. It is, rather, a sophisti-
cated subsystem of classical mathematics, which rejects the extremes in classical
mathematics, and carefully nourishes its effective computational apparatus.
Unfortunately, it seems that it is these ‘‘extremes’’ – bold extrapolations, abstrac-
tions which are infinite and do not lend themselves to a constructive interpretation –
which make classical mathematics effective. One should try to imagine how much
help mathematics could have provided twentieth century quantum physics if for
the past hundred years it had developed using only abstractions from ‘‘constructive
objects.’’
I do not pretend to be an expert in such matters, but there are several
studies in the literature that seem to me to indicate that such complaints are
not entirely without foundation. As one example, there is an important
series of papers by Pour-El and Richards (1979–87) establishing that much
of the machinery of functional analysis deployed in quantum physics
cannot be developed in its usual form with recursive analysis. And experi-
ence shows that what can or cannot be done recursively is a usually reliable
(though by no means infallible) guide to what can or cannot be done
intuitionistically.
As another example, there is a paper of Douglas Bridges (1981), examin-
ing quantum physics from an intuitionistic viewpoint. Bridges is, to be
sure, a follower of Bishop’s intuitionism-without-choice-sequences rather
than of Brouwer’s intuitionism-with-choice-sequences, but Bishop’s school
has thus far been able to go further than Brouwer’s school in reconstructing
applicable portions of functional analysis. Bridges is obliged to concede
that ‘‘a constructive examination of the mathematical foundations of
quantum physics does reveal substantial problems.’’ It is also worth men-
tioning that even if the indications cited are misleading and it turns out to
be possible in principle to get by with intuitionistic functional analysis in
applications to quantum physics, getting by in this way is very likely to be
infeasible in practice.
Table 14.1
Philosophies of mathematics
PLATONISM FORMALISM
(Hilbert)
Associated theories of meaning
VERISM HOLISM
(Quine)
Character of objection to theory of meaning
RADICAL DESCRIPTIVE PRESCRIPTIVE
We do not and could not speak a language We ought not to speak a language of the sort
of the sort described by the theory described by the theory
Principle on which the objection is based
BEHAVIORISM REPRESENTATIONALISM
There can be nothing more to knowing the Every sentence ought to serve as a
meaning of a sentence than being able representation of extra-linguistic reality,
to use it not a mere intra-linguistic instrument
for deducing other sentences
Argument for principle
If there were anything more to such Representationality is desirable
knowledge, the extra ingredient could (a) as a means towards guaranteeing
be neither acquired nor manifested. soundness
(b) as an end in itself
Comment
The acquisition and manifestation (a) Soundness can be guaranteed without
arguments involve fallacies and circularities representationality
(b) Representationality may conflict with
the desirable end of applicability
Whether applications to the empirical world are of value is a question on

which philosophers’ judgments vary over a wide spectrum. On the far right
stands Plato, who regarded such applications as evil. On the far left stand
the Gang of Four, who regarded the development of mathematics for any
purpose but such applications as evil. Among intuitionists, Brouwer was, in
this respect, a thoroughgoing Platonist, with an attitude not of passive
indifference, but of active hostility towards applications (see van Stigt
1979). Weyl, however, seems to have been uneasy over his inability to
reconcile his philosophical attraction towards intuitionism with his scien-
tific interest in applications.
One need not be a Maoist to sympathize with this unease, and to be
disturbed by an argument for the claim that intuitionism is the sole
legitimate form of mathematics in which any consideration of widely
held doubts as to the adequacy of intuitionism for applications is omitted.
(The omission is the more surprising in an argument directed against
Quine, since doubts as to adequacy for applications have been central
among Quine’s objections against other revisionist proposals such as
those of the nominalists.) The omission suggests a tacit system of values
so unworldly as to be irresponsible.
Dummett may perhaps be absolved personally from charges of irrespon-
sibility and inquisitorial interference with science. For though he advances
an argument for intuitionistic revisionism, he is cautious enough to dis-
tance himself personally somewhat from that argument. He is not so bold
as to claim that his conclusion ought to be accepted and put into practice.
What he claims is that it is an argument ‘‘of considerable power’’ (1973a,
p. 226). In view of the gaps and weaknesses in the argument that I have
tried to point out, even this more cautious claim might well be challenged.
4 SUMMARY
Table 14.1 summarizes my interpretation of and commentary on

Dummett’s case for intuitionism.
Annotated bibliography
TECHNICAL PAPERS ON SET THEORY
‘‘Forcing’’ (Burgess 1977a)

The continuum hypothesis (CH) states that the continuum, the cardinal of the set
of real numbers, is equal to aleph-one, the least cardinal greater than that of the set
of natural numbers. The two great methods for proving the consistency and
independence relative to the usual Zermelo–Frankel axioms of set theory are the
method of inner models, exemplified by Gödel’s constructible sets, through which
he proved the consistency of CH, and the method of forcing, through which Paul
Cohen proved the independence of CH. Burgess (1977a) is an exposition for
forcing intended to make the method available for use by non-specialists, by
following Robert Solovay and reducing what needs to be understood about
forcing in order to apply the method to three ‘‘axioms of forcing’’ whose proof
can be left to specialists.
‘‘Consistency proofs in model theory’’ (Burgess 1978a)

Just as ordinary arithmetic, the theory of addition and multiplication, has an
analogue in transfinite cardinal and ordinal arithmetics, so ordinary combina-
torics, the theory of permutations and combinations, has a transfinite analogue
in combinatorial set theory. Generalized-quantifier logic is a family of extensions
of first-order logic that retains the same notion of model, but adds additional
clauses to the definition of truth-in-a-model to cover such operators as ‘‘there
exist infinitely many’’ or ‘‘there exist uncountably many.’’ Hypotheses in com-
binatorial set theory turn out to have implications for generalized-quantifier
logic. Ronald Jensen, through a deep analysis of the fine structure of Gödel’s
constructible sets, proved the consistency of many combinatorial principles, and
therewith of certain principles about generalized-quantifier logic implied by
them. This paper shows how one of Jensen’s deepest consistency results about
generalized-quantifier logic can be obtained more easily using the method of
forcing.
277
278 Annotated bibliography
‘‘Descriptive set theory and infinitary languages’’ (Burgess 1977b)

Descriptive set theory is the branch concerned with definable sets of real numbers
or linear points. Infinitary logic is a family of extensions of first-order logic that
retains the same notion of model, but adds additional clauses to the definition of
truth-in-a-model to cover such operations as the conjunction and disjunction of
infinite sets of formulas. Several workers, most notably Robert Vaught, showed
the applicability of some results in descriptive set theory to infinitary logic. This
paper presents my contributions to Vaught’s project, the main one being as
follows. Jon Barwise developed a notion of ‘‘absoluteness’’ for logics, and charac-
terized the logics absolute relative to the weak Kripke–Platek set theory as the
sublogics of a certain well-known infinitary logic. This paper characterizes the
logics absolute relative to standard Zermelo–Frankel set theory, essentially as those
whose sentences admit suitable ‘‘approximations’’ by sentences of that same
infinitary logic.
‘‘Equivalence relations generated by families of Borel sets’’

(Burgess 1978b)
‘‘A reflection phenomenon in descriptive set theory’’ (Burgess 1979a)
‘‘Effective enumeration of classes in a R11 equivalence relation’’
(Burgess 1979b)
The special sets of real numbers (or pairs of real numbers) that are studied in
descriptive set theory are called projective, and they are divided into several classes
of increasing complexity and diminishing tractability: Borel or D11, analytic or R11,
coanalytic or 11, D12, R12, 12, and so on. A basic question of set theory is how many
elements a set of real numbers can have. For projective sets up to analytic it can be
proved that they contain either countably many or perfectly many (a stronger
condition implying but not implied by continuum many); for higher projective
sets the same result requires large cardinal axioms going beyond Zermelo–Frankel
set theory. A related question is how many equivalence classes an equivalence
relation that is projective (considered as a set of pairs of real numbers) can have.
Jack Silver proved that the answer is countably or perfectly many if the equivalence
relation is coanalytic. It had long been known that if the equivalence relation is
analytic there is a third possibility: aleph-one many but not perfectly many. Harvey
Friedman asked whether these are the only possibilities. The first two papers listed
above together supply the affirmative answer that was the main result of my doctoral
dissertation, written under Silver’s direction, while the third offers a refinement. The
first shows that any analytic equivalence relation is the intersection of aleph-one
Borel equivalence relations, and the second uses Silver’s theorem to show that the
intersection of aleph-one Borel equivalence relations has countably many, aleph-one
many, or perfectly many classes, a result later extended by Saharon Shelah to further
projective equivalence relations using large-cardinal assumptions.
‘‘A selection theorem for group actions’’ (Burgess 1979c)

‘‘A measurable selection theorem’’ (Burgess 1980a) ‘‘Sélections mesurables
pour relations d’équivalence à classes G’’ (Burgess 1980b)
‘‘Careful choices: a last word on Borel selectors’’ (Burgess 1981d)
‘‘From preference to utility: a problem of descriptive set theory’’
(Burgess 1985)
Given a function assigning to each real number x a set of real numbers F(x), the
axiom of choice (AC) guarantees that there exists a function defined for those x
for which F(x) is non-empty, and assigning to each such x an element f (x) of
F(x). But AC does not guarantee that this f has any nice properties of the kind
one studies in calculus, such as differentiability or integrability. Measurable
selection theorems are results to the effect that, assuming certain niceness con-
ditions on F, there follows the existence of an f with certain corresponding
niceness conditions. The first four papers in this group provide three examples,
differing in their niceness assumptions about F and niceness conclusions about f.
Measurable selection theorems are used in what I have elsewhere called ‘‘the
(hyper)theoretical fringes of subjects whose core is applied.’’ The last paper in the
group uses one to solve a problem raised by R. D. Mauldin in mathematical
economics.
‘‘What are R-sets?’’ (Burgess 1982a)

‘‘Classical hierarchies from a modern standpoint, parts I & II’’
(Burgess 1983a and b)
One of the main tools in modern descriptive set theory is the ‘‘game quantifier’’
of Yannis Moschovakis (which is also, to allude back to the discussion of
infinitary logic above, the basis for the most important logic that is
Zermelo–Frankel but not Kripke–Platek absolute). It is known that, in a sense
that can be made precise, application of this quantifier to Borel sets yields D12
sets. Borel sets themselves are divided into classes of increasing complexity:
closed or 01 sets, open or R01 sets, D02 sets, 02 sets, R02 sets, D03 sets, and
so on. Moschovakis had already established that applying the game quantifier to
closed and open sets yields analytic and coanalytic sets, respectively. The first
paper listed above is a semi-popular introduction to the pair of papers listed
below it, which show that from D02 and D03 sets one obtains, respectively,
families of sets studied during the 1920s and 1930s under the names of C-sets
and R-sets.
‘‘The truth is never simple’’ (Burgess 1986)

‘‘Addendum to ‘The truth is never simple’’’ (Burgess 1988)
The first paper is concerned with determining the place in the classifications of
descriptive set theorists of the set of all (Gödel numbers of) truths of arithmetic
according to various recent theories of truth: four versions of Kripke’s view (with
the Kleene three-valued or with van Fraassen supervaluation schemes for handling
truth-value gaps, and with the minimal fixed point or with the maximal intrinsic
fixed point) and three versions of the revision view (Gupta’s, Hertzberger’s, and
Belnap’s). A complete set of answers is obtained except for one case that had to
wait for the second paper, and one other case that my work left open. In the course
of working out the answer, examples are provided of sentences that are true on
some of the views but not on the others; indeed, examples of all possible
combinations are given.
‘‘Sets and point-sets’’ (Burgess 1990)

Perhaps the most surprising discoveries to emerge from foundational work in the
last century are two. First, all the natural choices turn out to be, for reasons for
which we at present lack any clear explanation, linearly ordered in strength (for the
cognoscenti, I mean in consistency strength), so that given any two choices, it
always turns out that one is stronger than the other, unless they turn out to be,
despite superficial differences, of equal strength. Second, stronger and stronger
axioms, though about objects of higher and higher level or type, continue to have
more and more implications even about objects of the very lowest level and type.
What this latter cryptic assertion means is spelled out in the case of geometrical
theories in this semi-technical, largely expository paper. The aim is to explode the
myths, which have gained some currency among philosophers not well-trained in
logic, that ‘‘mathematics is conservative over physics’’ and that ‘‘higher mathe-
matics is conservative over lower mathematics.’’
‘‘A remark on Henkin sentences and their contraries’’ (Burgess 2003b)

Jaakko Hintikka has for some time now been advocating what he has called IF (for
‘‘independence-friendly’’ or ‘‘information-friendly’’) logic. This logic is ‘‘non-clas-
sical,’’ not in the way that modal and tense and probability and conditional and
intuitionistic logic are, requiring a different notion of model from that used in
first-order logic, but rather in the way that generalized-quantifier and infinitary
logic are. The logic has been found interesting from a technical point of view even
by many who have not found Hintikka’s large philosophical claims for it con-
vincing. The paper establishes a technical result to the effect that the logic admits
no semantic operation of negation, a result that it is hard to interpret as having any
but negative implications about the philosophical claims for the logic, though this
point is not strenuously argued in the paper.
TECHNICAL PAPERS ON TENSE AND OTHER

NON-CLASSICAL LOGICS
‘‘Basic tense logic’’ (Burgess 1984a)

Basic tense logic adds to classical sentential logic new one-place connectives P for
‘‘it was the case that’’ and F for ‘‘it will be the case that.’’ Soundness and
completeness theorems establish that different sets of axioms for P and F exactly
correspond to different assumptions about the structure of time. (Is it linearly
ordered? Does it have a first or last moment? Are the moments of time densely
ordered or discrete?) When I first read the statements of a group of such theorems,
I worked out for myself proofs based on what has come to be called the ‘‘step-by-
step’’ method, and erroneously concluded that this must be the method everyone
was using. In fact, only a few were, while most were using Segerberg’s ‘‘bull-
dozing’’ and ‘‘unraveling’’ methods instead; but my adoption of the step-by-step
method in this expository chapter helped popularize it, first among philosophical
logicians, then among theoretical computer scientists.
‘‘The unreal future’’ (Burgess 1979d)

‘‘Decidability and branching time’’ (Burgess 1980c)
These two were originally a single item. Division was suggested by Krister
Segerberg, regular editor of Theoria and guest editor of Studia Logica. One of
the first applications of modern tense logic envisioned by its founder, Arthur
Prior, was to the analysis of traditional debates over future contingents, and of
the underlying picture of a time in which the one past behind us branches into
many possible futures before us. As part of this analysis Prior distinguished two
positions called, after two historical figures, ‘‘Peircean’’ and ‘‘Ockhamist.’’ The
former does not, while the latter does, consider it meaningful to speak, even
now, of some unknown one from among the many possible futures as being the
future that will become actual. The first section of ‘‘The unreal future’’ offers a
more formal account of Prior’s conception of the interaction between tempo-
ral and modal operators than he himself gave, then gives an informal summary of
the formal results in ‘‘Decidability and branching time,’’ which shows that the set
of sentences valid in all Peircean models is decidable, while for Ockhamism one
must distinguish standard from non-standard models. The second section of the
same paper was ostensibly a survey of conceptual issues that needed to be
addressed before moving beyond the sentential to a predicate logic of branching
time, but was really my first attempt to come to grips with ‘‘Naming and
necessity.’’ Underlying it was, in embryonic and inchoate form, the thought
that ‘‘metaphysical’’ modality might be demystified by tracing its source to our
sortal classifications of the objects of our thought, and our conventions as to
what does and what does not count as another appearance of the same thing of a
given sort. The idea was not successfully worked out in the paper, and I have not
as yet even today achieved a development of it that is wholly satisfactory even to
myself, though the idea still seems to me to be promising.
‘‘Axioms for tense logic’’ (Burgess 1982b)

The more important first part of this two-part paper concerns the two-place
‘‘since’’ and ‘‘until’’ operators by which Hans Kamp enriched Prior’s tense logic.
The paper provides an axiomatization proved sound and complete by an extension
of the step-by-step method.
‘‘The decision problem for linear temporal logic’’

(Burgess and Gurevich 1985)
Segerberg’s ‘‘filtration’’ method can be used to prove the decidability of the tense
logics appropriate to many different models of time, but not the model on
which the moments of time are ordered like the real numbers. This paper
presents two different proofs of the decidability of that logic, along quite
different lines. The first is mine. Gurevich, the referee for the paper in which
I wrote up this method, suggested the second, using more advanced techniques
and thereby providing some additional information. We agreed it was best for
him to become a co-author, so that both methods could be made available in a
single paper.
‘‘Probability logic’’ (Burgess 1969)

This paper provides a complete axiomatization and a decision theory for a
sentential modal logic enriched with an operator ‘‘it is probable that’’ in addition
to the operator ‘‘it is necessary that.’’ The axioms are those appropriate for a
qualitative notion of probability, though results are also presented on quantitative
notions.
‘‘Quick completeness proofs for some logics of conditionals’’

(Burgess 1981b)
Robert Stalnaker and David Lewis independently developed similar but not
identical views on counterfactual or subjunctive conditionals, and there is a
third variant as well. This paper provides complete axiomatizations and therewith
proofs of decidability for a range of variants. The systems have since been found to
admit another interpretation, in terms of nonmonotonic logics of the kind
pioneered by Kraus, Lehmann, and Magidor.
‘‘The completeness of intuitionistic propositional calculus

for its intended interpretation’’ (Burgess 1981a)
The question addressed here is whether Heyting’s axiomatization of
intuitionistic logic is complete in the sense that every sentential formula all of
whose instances (obtained by substituting formulas of intuitionistic arithmetic or
analysis for its sentential variables) are intuitionistically correct is a thesis of the
system. Such a question is not answered merely by a proof of completeness for
some formal ‘‘semantics’’ on the order of topological models or Kripke models.
Georg Kreisel was able, using topological models as a starting point, to obtain an
affirmative answer, making certain assumptions about intuitionistic analysis (the
theory of ‘‘lawless’’ sequences). The paper shows how, on the same assumptions, to
obtain an affirmative answer taking Kripke models as a starting point.
References
Ackermann, Diana [Ackermann, Felicia Nimue] (1978) ‘‘De re propositional

attitudes toward integers,’’ Southwestern Journal of Philosophy vol. 9,
pp. 145–53.
Anderson, Alan Ross and Belnap, Nuel D., Jr. (1975) Entailment: The Logic of
Relevance and Necessity vol. I (Princeton, NJ: Princeton University Press).
Anderson, C. A. and Zelëny, M. (2002) (eds.) Logic, Meaning, and Computation:
Essays in Memory of Alonzo Church (Dordrecht: Kluwer).
Azzouni, Jodi (2004) Deflating Existential Consequence: A Case for Nominalism
(Oxford: Oxford University Press).
Baire, René, Borel, Émil, Hadamard, Jacques, and Lebesgue, Henri (1905)
‘‘Cinque lettres sur la théorie des ensembles,’’ Bulletin de la Société
Mathématique de France vol. 33, pp. 261–73.
Balaguer, Mark (1998) Platonism and Anti-Platonism in Mathematics (Oxford:
Oxford University Press).
Barcan, Ruth C. [Marcus, Ruth Barcan] (1946) ‘‘A functional calculus of first
order based on strict implication,’’ Journal of Symbolic Logic vol. 11, pp. 1–16.
(1947) ‘‘Identity of individuals in a strict functional calculus of second order,’’
Journal of Symbolic Logic vol. 12, pp. 12–15.
Bar-Hillel, Y., Poznanski, E. I. J., Rabin, M. O., and Robinson, A. (1961) (eds.)
Essays on the Foundations of Mathematics (Jerusalem: Magnes Press).
Barwise, K. Jon (1977) Handbook of Mathematical Logic (Amsterdam: North
Holland).
Belnap, Nuel D., Jr. and Green, Mitchell (1994) ‘‘The thin red line,’’ in Tomberlin
(1994), pp. 365–88.
Benacerraf, Paul and Putnam, Hilary (1964) Philosophy of Mathematics: Selected
Readings (Englewood Cliffs, NJ: Prentice-Hall).
(1983) Philosophy of Mathematics: Selected Readings, 2nd edn (Cambridge:
Cambridge University Press).
Bernays, Paul (1961) ‘‘Zur Frage der Unendlichkeitsschemata in der axiomatischen
Mengenlehre,’’ in Bar-Hillel et al. (1961), pp. 3–49.
(1976) ‘‘On the problem of schemata of infinity in axiomatic set theory,’’
English translation of Bernays (1961) by J. Bell and M. Plänitz, in Müller
(1976), pp. 121–72.
284
References 285
Birkhoff, Garrett (1937) ‘‘Rings of sets,’’ Duke Mathematical Journal vol. 3,
pp. 443–54.
Bochenski, I., Church, A., and Goodman, N. (1956) The Problem of Universals:
A Symposium (Notre Dame, IN: Notre Dame University Press).
Boolos, George (1984) ‘‘To be is to be the value of a variable (or to be some values
of some variables),’’ Journal of Philosophy vol. 81, pp. 430–39, reprinted in
(Boolos 1997), pp. 54–72.
(1985) ‘‘Nominalist Platonism,’’ Philosophical Review vol. 94, pp. 327–44,
reprinted in Boolos (1997a), pp. 73–87.
(1987) ‘‘The consistency of Frege’s Foundations of Arithmetic,’’ in Thomson
(1987), pp. 3–20; reprinted in Demopoulous (1995), pp. 211–33, and in Boolos
(1997a), pp. 182–201.
(1993) The Logic of Provability (Cambridge: Cambridge University Press).
(1997a) Logic, Logic, and Logic (Cambridge, MA: Harvard University Press).
(1997b) ‘‘Must we believe in set theory?’’ in Boolos (1997a), pp. 120–32.
Borges, Jorge Luis (1962) ‘‘Tlön, Uqbar, Orbis Tertius,’’ translated from the
Spanish by Alastair Reid, in A. Kerrison (ed.) Ficciones (New York: Grove
Press).
Bridges, Douglas (1981) ‘‘Towards a constructive foundation for quantum
mechanics,’’ in Richman (1981), pp. 260–73.
Brouwer, L. E. J. (1975) Collected Works, vol. I: Philosophy and Foundations of
Mathematics (Amsterdam: North Holland).
(1976) Collected Works, vol. II: Geometry, Analysis, Topology and Mechanics
(Amsterdam: North Holland).
Bull, R. A. and Segerberg, Krister (1984) ‘‘Basic modal logic,’’ in Gabbay and
Guenthner (1984), pp. 1–88.
Burgess, John P. (1969) ‘‘Probability logic,’’ Journal of Symbolic Logic vol. 34,
pp. 264–74.
(1977a) ‘‘Forcing’’ in Barwise (1977), pp. 403–52.
(1977b) ‘‘Descriptive set theory and infinitary languages,’’ in Proceedings of the
1977 Belgrade Symposium on Set Theory and Foundations of Mathematics,
Mathematical Institute, Belgrade, pp. 9–30.
(1978a) ‘‘Consistency proofs in model theory: a contribution to Jensenlehre,’’
Annals of Mathematical Logic vol. 14, pp. 1–12.
(1978b) ‘‘Equivalence relations generated by families of Borel sets,’’ American
Mathematical Society Proceedings vol. 69, pp. 323–6.
(1979a) ‘‘A reflection phenomenon in descriptive set theory,’’ Fundamenta
Mathematicae vol. 104, pp. 127–39. P1
(1979b) ‘‘Effective enumeration of classes in a 1 equivalence relation,’’
Indiana University Mathematical Journal vol. 28, pp. 353–64.
(1979c) ‘‘A selection theorem for group actions,’’ Pacific Journal of Mathematics
vol. 80, pp. 333–6.
(1979d) ‘‘The unreal future,’’ Theoria vol. 44, pp. 157–79.
(1980a) ‘‘A measurable selection theorem,’’ Fundamenta Mathematicae vol. 100,
pp. 91–100.
286 References
(1980b) ‘‘Sélections mesurables pour relations d’équivalence à classes Gd,’’
Bulletin des Sciences Mathématiques vol. 104, pp. 435–40.
(1980c) ‘‘Decidability and branching time,’’ in K. Segerberg (ed.) Trends in
Modal Logic, Studia Logica vol. 39, pp. 203–18.
(1981a) ‘‘The completeness of intuitionistic propositional calculus for its
intended interpretation,’’ Notre Dame Journal of Formal Logic vol. 22,
pp. 17–28.
(1981b) ‘‘Quick completeness proofs for some logics of conditionals,’’ Notre
Dame Journal of Formal Logic vol. 22, pp. 76–84.
(1981c) ‘‘Relevance: a fallacy?,’’ Notre Dame Journal of Formal Logic vol. 22,
pp. 97–104.
(1981d) ‘‘Careful choices: a last word on Borel selectors,’’ Notre Dame Journal of
Formal Logic vol. 22, pp. 219–26.
(1982a) ‘‘What are R-sets?’’ in Metakides (1982), pp. 307–24.
(1982b) ‘‘Axioms for tense logic, I. Since and until,’’ Notre Dame Journal of
(1983a) ‘‘Classical hierarchies from a modern standpoint, I. C-sets,’’ Fundamenta
Mathematicae vol. 115, pp. 81–96.
(1983b) ‘‘Classical hierarchies from a modern standpoint, II. R-sets,’’ Fundamenta
Mathematicae vol. 115, pp. 97–105.
(1983c) ‘‘Common sense and ‘relevance,’’’ Notre Dame Journal of Formal Logic
vol. 24, pp. 41–53.
(1983d) ‘‘Why I am not a nominalist,’’ Notre Dame Journal of Formal Logic
vol. 24, pp. 93–105.
(1984a) ‘‘Basic tense logic,’’ in Gabbay and Guenthner (1984), pp. 89–134.
(1984b) ‘‘Read on relevance: a rejoinder,’’ Notre Dame Journal of Formal Logic
vol. 25, pp. 217–23.
(1984c) ‘‘Dummett’s case for intuitionism,’’ History and Philosophy of Logic
vol. 5, pp. 177–94.
(1985) ‘‘From preference to utility: a problem of descriptive set theory,’’ Notre
(1986) ‘‘The truth is never simple,’’ Journal of Symbolic Logic vol. 51, pp. 663–81.
(1988) ‘‘Addendum to ‘The truth is never simple,’’’ Journal of Symbolic Logic
vol. 53, pp. 390–2.
(1989) ‘‘Epistemology and nominalism,’’ in Irvine (1989), pp. 1–15.
(1990) ‘‘Sets and point-sets,’’ in Fine and Lepin (1990), pp. 456–63.
(1992) ‘‘Proofs about proofs: a defense of classical logic, I,’’ in Detlefsen (1992),
pp. 79–82.
(1993) ‘‘How foundational work in mathematics can be relevant to philosophy
of science,’’ in Hull et al. (1993), pp. 433–41.
(1995) ‘‘Frege and arbitrary functions.’’ in Demopoulos (1995), pp. 89–107.
(1996) ‘‘Marcus, Kripke, and names,’’ Philosophical Studies vol. 84, pp. 1–47,
reprinted in Humphreys and Fetzer (1998), pp. 89–124.
(1998a) ‘‘How not to write history of philosophy,’’ in Humphreys and Fetzer
(1998), pp. 125–36.
References 287
(1998b) ‘‘Occam’s razor and scientific method,’’ in Schirn (1998), pp. 195–214.
(1998c) ‘‘Quinus ab omni naevo vindicatus,’’ in A. A. Kazmi (ed.) Meaning and
Reference: Canadian Journal of Philosophy Supplement vol. 23, pp. 25–65.
(1999) ‘‘Which modal logic is the right one?’’ Notre Dame Journal of Formal
Logic vol. 40, pp. 81–93.
(2001) Review of Balaguer (1998), Philosophical Review, vol. 101, pp. 79–82.
(2002a) ‘‘Nominalist paraphrase and ontological commitment,’’ in Anderson
and Zelëny (2002), pp. 429–44.
(2002b) ‘‘Is there a problem about deflationary theories of truth?’’ in Horsten
and Halbach (2002), pp. 37–56.
(2003a) ‘‘Numbers and ideas,’’ Richmond Journal of Philosophy vol. 1, pp. 12–17.
(2003b) ‘‘A remark on Henkin sentences and their contraries,’’ Notre Dame
Journal of Formal Logic vol. 44, pp. 185–8.
(2004a) ‘‘Quine, analyticity, and philosophy of mathematics,’’ Philosophical
Quarterly vol. 54, pp. 38–55.
(2004b) ‘‘Mathematics and Bleak House,’’ Philosophia Mathematica vol. 12,
pp. 18–36.
(2004c) ‘‘E pluribus unum: plural logic and set theory,’’ Philosophia Mathematica
vol. 12, pp. 193–221.
(2004d) review of Azzouni (2004) Bulletin of Symbolic Logic vol. 10, pp. 573–7.
(2005a) ‘‘No requirement of relevance,’’ in Shapiro (2005), pp. 727–50.
(2005b) Fixing Frege (Princeton, NJ: Princeton University Press).
(2005c) ‘‘Translating names,’’ Analysis vol. 65, pp. 196–204.
(2005d) ‘‘Being explained away,’’ Harvard Review of Philosophy vol. 13,
pp. 41–56.
(2005e) ‘‘On anti-anti-realism,’’ Facta Philosophica vol. 7, pp. 121–44.
(forthcoming) ‘‘Protocol sentences for lite logicism,’’ in Lindström
(forthcoming).
Burgess, John P. and Gurevich, Yuri (1985) ‘‘The decision problem for linear
temporal logic,’’ Notre Dame Journal of Formal Logic vol. 26, pp. 115–28.
Burgess, John P. and Hazen, A. P. (1998) ‘‘Arithmetic and predicative logic,’’ Notre
Burgess, John P. and Rosen, Gideon (1997) A Subject With No Object: Strategies for
Nominalistic Interpretation of Mathematics (Oxford: Oxford University
Press).
Cantor, Georg (1885) review of Frege (1884), Deutsche Literaturzeitung, vol. 6,
pp. 728–9.
Carnap, Rudolf (1946) ‘‘Modalities and quantification,’’ Journal of Symbolic Logic
vol. 11, pp. 33–64.
(1947) Meaning and Necessity: A Study in Semantics and Modal Logic (Chicago:
University of Chicago Press).
(1950) ‘‘Empiricism, semantics, and ontology,’’ Revue Internationale de
Philosophie vol. 4, pp. 20–40.
Chihara, Charles (1973) Ontology and the Vicious Circle Principle (Ithaca, NY:
Cornell University Press).
288 References
(1989) ‘‘Tharp’s ‘Myth and Mathematics,’’’ Synthese vol. 81, pp. 153–65.
(1990) Constructibility and Mathematical Existence (Oxford: Oxford University
Press).
Chomsky, Noam (1959) review of Skinner (1957) Language vol. 35, pp. 26–58.
Church, Alonzo (1950) review of Fitch (1949) Journal of Symbolic Logic vol. 15,
p. 63.
Cocchiarella, Nino (1984) ‘‘Philosophical perspectives on quantification in tense
and modal logic,’’ in Gabbay and Guenthner (1984), pp. 309–53.
Cohen, R. S. and Wartofsky, M. W. (1965) (eds.) Boston Studies in the Philosophy of
Science, vol. II (New York: Humanities Press).
Copeland, B. J. (1979) ‘‘When is a semantics not a semantics: some reasons for
disliking the Routley–Meyer semantics for relevance logic,’’ Journal of
Philosophical Logic vol. 8, pp. 399–413.
Creswell, Max (1990) Entities and Indices (Dordrecht: Kluwer).
Davidson, Donald (1967) ‘‘Truth and meaning,’’ Synthese vol. 17, pp. 304–23.
Davidson, Donald and Harman, Gilbert (1972) (eds.) Semantics of Natural
Language (Dordrecht: Reidel).
Davidson, Donald and Hintikka, Jaakko (1969) (eds.) Words and Objections: Essays
on the Work of W. V. Quine (Dordrecht: Reidel).
Demopoulos, William (1995) (ed.) Frege’s Philosophy of Mathematics (Cambridge,
MA: Harvard University Press).
Detlefsen, Michael (1992) (ed.) Proof, Logic and Formalization (London:
Routledge).
Diogenes Laertius (1925) Lives and Opinions of Eminent Philosophers, translated
from the Greek by R. D. Hicks, Loeb Classical Library (Cambridge, MA:
Harvard University Press).
Dummett, Michael (1959) ‘‘Truth,’’ in Dummett (1978), pp. 1–24.
(1973a) ‘‘The philosophical basis of intuitionistic logic,’’ in Dummett (1978),
pp. 215–47.
(1973b) ‘‘The justification of deduction,’’ in Dummett (1978), pp. 290–318.
(1973c) ‘‘The significance of Quine’s indeterminacy thesis,’’ in Dummett (1978),
pp. 375–419.
(1977) Elements of Intuitionism (Oxford: Oxford University Press).
(1978) Truth and Other Enigmas (Cambridge, MA: Harvard University Press).
Edgington, Dorothy (1985) ‘‘The paradox of knowability,’’ Mind vol. 94,
pp. 557–68.
Evans, Gareth and McDowell, John (1976) (eds.) Truth and Meaning: Essays in
Semantics (Oxford: Oxford University Press).
Farber, M. (1950) (ed.) Philosophic Thought in France and the United States
(Buffalo, NY: University of Buffalo Press).
Feferman, Solomon (1977) ‘‘Theories of finite type related to mathematical
practice,’’ in Barwise (1977), pp. 913–72.
Field, Hartry H. (1980) Science Without Numbers: A Defense of Nominalism
(Princeton, NJ: Princeton University Press).
(1989) Realism, Mathematics and Modality (Oxford: Basil Blackwell).
References 289
Fine, A. and Lepin, J. (1990) PSA 88 [Proceedings of the 1988 Convention of the
Philosophy of Science Association], vol. II (East Lansing, MI: Philosophy of
Science Association).
Fine, Kit (2002) The Limits of Abstraction (Oxford: Oxford University Press).
Fitch, Frederic (1949) ‘‘The problem of the morning star and the evening star,’’
Philosophy of Science vol. 16, pp. 137–41.
(1950) ‘‘Attribute and class,’’ in Farber (1950), pp. 640–7.
(1963) ‘‘A logical analysis of some value concepts,’’ Journal of Symbolic Logic
vol. 28, pp. 135–42.
Føllesdal, Dagfinn (1961) ‘‘Referential opacity and modal logic,’’ Harvard
University doctoral dissertation, reprinted as Føllesdal (1966).
(1965) ‘‘Quantification into causal contexts,’’ in Cohen and Wartofsky (1965),
pp. 263–74; reprinted in Linsky (1971a), pp. 52–62.
(1966) Referential Opacity and Modal Logic, Filosofiske Problemer, vol. XXXII
(Oslo: Oslo Universitetsforlaget).
(1969) ‘‘Quine on modality,’’ in Davidson and Hintikka (1969), pp. 175–85.
(1986) ‘‘Essentialism and reference,’’ in Hahn and Schlipp (1986), pp. 97–113.
Forbes, Graeme (1995) review of Marcus (1993), Notre Dame Journal of Formal
Frege, Gottlob (1879) Begriffsschrift: eine der arithmetischen nachgebildete
Formelsprache des reinen Denkens (Halle: Louis Nebert).
(1884) Die Grundlagen der Arithmetik: Eine logisch-mathematische Untersuchung
über den Begriff der Zahl (Breslau: Wilhelm Koebner).
(1893/1903) Grundgesetze der Arithmetik, begriffsschriftlich abgeleitet, 2 vols.
(Jena: Pohle).
(1950) The Foundations of Arithmetic, translation of Frege (1884) from the
German by J. L. Austin (London: Blackwell).
(1967) Begriffsschrift, translation from the German of Frege (1879) by S. Bauer-
Mengelberg, in van Heijenoort (1967), pp. 1–82.
French, Peter A. and Wettstein, Howard K. (2001) (eds.) Midwest Studies in
Philosophy XXV: Figurative Language (London: Blackwell).
Gabbay, D. and Guenthner, F. (1984) (eds.) Handbook of Philosophical Logic,
vol. II: Extensions of Classical Logic (Dordrecht: Reidel).
Gabbay, D., Rahman, S., Symons, J., and van Bendegen, J. P. (2004) (eds.) Logic,
Epistemology, and the Unity of Science (Dordrecht: Kluwer).
Garson, James (1984) ‘‘Quantification in modal logic,’’ in Gabbay and Guenthner
(1984), pp. 249–308.
Goodman, Nelson (1956) ‘‘A world of individuals,’’ in Bochenski et al. (1956),
pp. 197–210; reprinted in Benacerraf and Putnam (1964).
Goodman, Nelson and Quine, W. V. O. (1947) ‘‘Steps toward a constructive
nominalism,’’ Journal of Symbolic Logic vol. 12, pp. 97–122.
Goranko, Valentin (1994) ‘‘Refutation systems in modal logic,’’ Studia Logica
vol. 53, pp. 299–324.
Grandy, Richard (1982) review of Dummett (1973a), Prawitz (1977), etc. Journal of
Symbolic Logic vol. 47, pp. 689–94.
290 References
Grice, H. P. and Strawson, P. F. (1956) ‘‘In defense of a dogma,’’ Philosophical
Review vol. 65, pp. 141–58.
Grzegorczyk, Andrzej (1967) ‘‘Some relational systems and the associated topo-
logical spaces,’’ Fundamenta Mathematicae vol. 60, pp. 223–31.
Haack, Susan (1974) Deviant Logic (Cambridge: Cambridge University Press).
Hahn, L. E. and Schlipp, P. A. (1986) The Philosophy of W. V. Quine (LaSalle, IL:
Open Court).
Hájek, Petr and Pudlak, Pavel (1998) Metamathematics of First-Order Arithmetic
(Berlin: Springer).
Haldén, Søren (1963) ‘‘A pragmatic approach to modal theory,’’ Acta Philosophica
Fennica vol. 16, pp. 53–64.
Hale, Bob and Wright, Crispin (2001) The Reason’s Proper Study: Essays towards a
Neo-Fregean Philosophy of Mathematics (Oxford: Oxford University Press).
Hallett, Michael (1984) Cantorian Set Theory and Limitation of Size (Oxford:
Clarendon Press).
Harman, Gilbert (1982) ‘‘Conceptual role semantics,’’ Notre Dame Journal of
Heck, Richard G., Jr. (1996) ‘‘On the consistency of predicative fragments of
Frege’s Grundgesetze der Arithmetik,’’ History and Philosophy of Logic vol. 17,
pp. 209–20.
Heijenoort, Jean van (1967) (ed.) From Frege to Gödel: A Sourcebook in
Mathematical Logic, 1879–1931 (Cambridge, MA: Harvard University Press).
Hempel, Carl G. (1950) ‘‘Problems and changes in the empiricist criterion of
meaning,’’ Revue Internationale de Philosophie vol. 4, pp. 41–63.
Hersh, Reuben (1997) What is Mathematics, Really? (Oxford: Oxford University
Press).
Hilbert, David (1925/1967) ‘‘On the infinite,’’ translated from the German by
Stefan Bauer-Mengelberg, in van Heijenoort (1967), pp. 367–92.
Hintikka, Jaakko (1963) ‘‘Modes of modality,’’ Acta Philosophica Fennica vol. 16,
pp. 65–82.
(1982) ‘‘Is alethic modal logic possible?’’ Acta Philosophica Fennica vol. 35,
pp. 89–105.
Hintikka, Jaakko and Sandu, Gabriel (1995) ‘‘The fallacies of the new theory of
reference,’’ Synthese vol. 104, pp. 245–83.
Hofweber, Thomas and Everett, Anthony (2000) (eds.) Empty Names, Fiction and
the Puzzles of Non-Existence (Chicago: CSLI).
Horsten, Leon and Halbach, Volker (2002) (eds.) Principles of Truth (Frankfurt:
Hänsel-Hohenhausen).
Hrbacek, K. and Jech, T. (1999) Introduction to Set Theory, 3rd edn (New York:
Marcel Dekker).
Hughes, G. E. and Creswell, M. J. (1968) An Introduction to Modal Logic (London:
Methuen).
Hull, D., Forbes, M., and Okruhlik, K. (1993) (eds.) PSA 92 [Proceedings of the
1992 Convention of the Philosophy of Science Association], vol. II (East
Lansing, MI: Philosophy of Science Association).
References 291
Humphreys, P. and Fetzer, J. (1998) (eds.) The New Theory of Reference, Synthese
Library vol. CCLXX (Dordrecht: Kluwer).
Irvine, Andrew (1989) (ed.) Physicalism in Mathematics (Dordrecht: Kluwer).
James, William (2000) ‘‘Pragmatism,’’ in Pragmatism and Other Writings, ed.
G. Gunn (New York: Penguin), pp. 1–132.
Jeffrey, Richard C. (1996) ‘‘Logicism 2000,’’ in Stich and Morton (2002),
pp. 1–132.
(2002) ‘‘Logicism lite,’’ Philosophy of Science vol. 69, pp. 447–51.
Kahle, Reinhard (forthcoming) (ed.) Intensionality: An Interdisciplinary Discussion
(Boston: A. K. Peters, Lecture Notes in Logic).
Katz, Jerrold (1985) (ed.) The Philosophy of Linguistics (Oxford: Oxford University
Press).
Kitcher, Philip (1978) ‘‘The plight of the platonist,’’ Noûs, vol. 12, pp. 119–36.
Klibansky, Raymond (1968) Contemporary Philosophy, 4 vols. (Florence: Editrice
Nuova Italia).
Kneale, William and Kneale, Mary (1962) The Development of Logic (Oxford:
Clarendon Press).
König, Julius (1905) ‘‘Über die Grundlagen der Mengenlehre und das
Kontinuumproblem,’’ Mathematische Annalen vol. 61, pp. 156–60.
(1967) ‘‘On the foundations of set theory and the continuum problem,’’ trans-
lation of König (1905) from the German by Stefan Bauer-Mengelberg, in van
Heijenoort (1967), pp. 145–9.
Kranz, D., Luce, R., Suppes, P., and Tversky, A. (1971) Foundations of
Measurement (New York: Academic Press).
Kreisel, Georg (1967) ‘‘Informal rigour and completeness proofs,’’ in Lakatos
(1967), pp. 138–57.
Kripke, Saul (1963) ‘‘Semantical considerations on modal logic,’’ Acta Philosophica
(1972) ‘‘Naming and necessity: Lectures give to the Princeton University
Philosophy Colloquium, January, 1970,’’ in Davidson and Harman (1972),
pp. 253–355 and 763–9; reprinted with a new preface as Kripke (1980).
(1976) ‘‘Is there a problem about substitutional quantification?’’ in Evans and
McDowell (1976), pp. 325–420.
(1979) ‘‘A puzzle about belief,’’ in Margalit (1977), pp. 239–83.
(1980) Naming and Necessity (Cambridge, MA: Harvard University Press).
(1982) Wittgenstein on Rules and Private Language (Cambridge, MA: Harvard
University Press).
Lavine, Shaughan (1995) review of Marcus (1993), British Journal of the Philosophy
of Science vol. 46, pp. 267–74.
Lakatos, Imre (1967) (ed.) Proceedings of the International Colloquium in the
Philosophy of Science, London, 1965, vol. I (Amsterdam: North Holland).
Lee, O. H. (1936) Philosophical Essays for A. N. Whitehead (New York: Longmans).
Lewis, David K. (1969) Convention (Cambridge, MA: Harvard University Press).
(1970) ‘‘Anselm and actuality,’’ Noûs vol. 4, pp. 175–88.
(1991) Parts of Classes (Oxford: Oxford University Press).
292 References
Linsky, Leonard (1971a) (ed.) Reference and Modality (Oxford: Oxford University
Press).
(1971b) ‘‘Essentialism, reference, and modality,’’ in (Linsky (1971a), pp. 88–100.
(1977) Names and Descriptions (Chicago: University of Chicago Press).
Linström, Sten (forthcoming) (ed.) Logicism, Intuitionism, Formalism. What Has
Become of Them? (Berlin: Springer).
Maddy, Penelope (1980) ‘‘Perception and mathematical intuition,’’ Philosophical
Review vol. 89, pp. 163–96.
(1984) ‘‘Mathematical epistemology: what is the question?’’ The Monist vol. 67,
pp. 46–55.
(1990) ‘‘Mathematics and Oliver Twist,’’ Pacific Philosophical Quarterly vol. 71,
pp. 189–205.
Makinson, David (1966) ‘‘How meaningful are modal operators?’’ Australasian
Journal of Philosophy vol. 44, pp. 331–7.
Manin, Yuri (1977) A Course in Mathematical Logic, translated from the Russian
by N. Koblitz (Berlin: Springer).
Marcus, Ruth Barcan (1960) ‘‘Extensionality,’’ Mind vol. 69, pp. 55–62.
(1963a) ‘‘Modalities and intensional languages,’’ in Wartofsky (1963),
pp. 77–96.
(1963b) ‘‘Attribute and class in extended modal systems,’’ Acta Philosophical
(1967) ‘‘Essentialism in modal logic,’’ Noûs vol. 1, pp. 90–6.
(1968) ‘‘Modal logic,’’ in Klibansky (1968), pp. 87–101.
(1978) Review of Linsky (1977), Philosophical Review vol. 87, pp. 497–504.
(1990) ‘‘Some revisionary proposals about belief and believing,’’ Philosophy and
Phenomenological Research vol. 50 (Supplement), pp. 133–53.
(1993) Modalities: Philosophical Essays (Oxford: Oxford University Press).
Marcus, R. B., Quine, W. V., Kripke, S. A. et al. (1963) Discussion of Marcus
(1963a) in Wartofsky (1963), pp. 105–16.
Margalit, Avishai (1977) Meaning and Use (Dordrecht: Reidel).
Martinich, A. P. (1979) The Philosophy of Language (Oxford: Oxford University
Press).
Matiyasevich, Yuri (1993) Hilbert’s Tenth Problem (Cambridge, MA: MIT Press).
McKinsey, J. C. C. (1941) ‘‘A solution to the decision problem for the Lewis
systems S2 and S4, with an application to topology,’’ Journal of Symbolic
(1945) ‘‘On the syntactical construction of modal logic,’’ Journal of Symbolic
Metakides, George (1982) (ed.) Proceedings of the First Patras Logic Symposion
Montagna, Franco and Mancini, Antonella (1994) ‘‘A minimal predicative set
theory,’’ Notre Dame Journal of Formal Logic vol. 35, pp. 186–203.
Müller, Gert-Heinz (1976) Sets and Classes (Amsterdam: North-Holland).
Nabokov, Vladimir (1980) Lectures on Literature, ed. F. Bowers (New York:
Harcourt Brace Jovanovich).
References 293
Newman, J. R. (1956) (ed.) The World of Mathematics, 4 vols. (New York: Simon
and Schuster).
Parsons, Charles (1980) ‘‘Mathematical intuition,’’ Proceedings of the Aristotelian
Society, vol. 80, pp. 145–68.
Parsons, Terence (1969) ‘‘Essentialism and quantified modal logic,’’ Philosophical
Review vol. 78, pp. 35–52; reprinted in Linsky (1971a), pp. 73–87.
(1987) ‘‘On the consistency of the first-order portion of Frege’s logical system,’’
Notre Dame Journal of Formal Logic vol. 28, pp. 61–8; reprinted in
Demopoulos (1995), pp. 422–31.
Pollard, Stephen (1996) ‘‘Sets, wholes, and limited pluralities,’’ Philosophia
Mathematica, vol. 4, pp. 42–58.
Pour-El, M. and Richards, I. (1979) ‘‘A computable ordinary differential equation
which possesses a computable solution,’’ Annals of Mathematical Logic vol. 17,
pp. 61–90.
(1981) ‘‘A wave equation with computable initial data such that its unique
solution is not computable,’’ Advances in Mathematics vol. 39, pp. 215–39.
(1983) ‘‘Noncomputability in analysis and physics: a complete determination of
the class of noncomputable linear operators,’’ Advances in Mathematics
vol. 48, pp. 44–74.
(1987) ‘‘The eigenvalues of an effectively determined self-adjoint operator are
computable, but the sequence of eigenvalues is not,’’ Advances in Mathematics
vol. 63, pp. 1–41.
Prawitz, Dag (1977) ‘‘Meaning and proofs: on the conflict between classical and
intuitionistic logic,’’ Theoria vol. 43, pp. 2–40.
Prior, Arthur N. (1960) ‘‘The runabout inference-ticket,’’ Analysis vol. 21, pp. 38–9.
(1963) ‘‘Is the concept of referential opacity really necessary?’’ Acta Philosophical
(1967a) ‘‘Logic, modal,’’ in Weiss (1967), vol. V, pp. 5–12.
(1967b) Past, Present, and Future (Oxford: Clarendon Press).
Putnam, Hilary (1971) Philosophy of Logic (New York: Harper).
(1975) ‘‘Truth and necessity in mathematics,’’ in Mathematics, Matter, and
Method (Cambridge: Cambridge University Press), pp. 1–11.
Quine, W. V. O. (1936) ‘‘Truth by convention,’’ in Lee (1936), pp. 90–124.
(1946) Review of Barcan (1946), Journal of Symbolic Logic vol. 11, pp. 96–7.
(1947a) ‘‘The problem of interpreting modal logic,’’ Journal of Symbolic Logic
vol. 12, pp. 43–8.
(1947b) Review of Barcan (1947), Journal of Symbolic Logic vol. 12, pp. 95–6.
(1951a) ‘‘Carnap’s views on ontology,’’ Philosophical Studies vol. 2, pp. 65–72.
(1951b) ‘‘Two dogmas of empiricism,’’ Philosophical Review vol. 60, pp. 20–43.
(1953) From a Logical Point of View (Cambridge, MA: Harvard University
Press).
(1960) Word and Object (New York: John Wiley and Sons).
(1961) From a Logical Point of View, 2nd edn (Cambridge, MA: Harvard
University Press).
(1963) Comments on Marcus (1963a) in Wartofsky (1963), pp. 97–104.
294 References
(1969) ‘‘Reply to Sellars,’’ in Davidson and Hintikka (1969), p. 338.
(1970) Philosophy of Logic (Englewood Cliffs, NJ: Prentice-Hall).
(1980) From a Logical Point of View, 3rd edn (Cambridge, MA: Harvard
University Press).
(1981) ‘‘Response to David Armstrong,’’ in Theories and Things (Cambridge:
Harvard University Press), pp. 182–4.
Ramsey, Frank Plumpton (1925) ‘‘The foundations of mathematics,’’ Proceedings of
the London Mathematical Society vol. 25, pp. 338–84.
Rayo, Augustin and Uzquiano, Gabriel (1999) ‘‘Towards a theory of second-order
consequence,’’ Notre Dame Journal of Formal Logic vol. 40, pp. 315–25.
Rescher, Nicholas (1968) (ed.) Studies in Logical Theory (Oxford: Basil Blackwell).
Richman, F. (1981) (ed.) Constructive Mathematics [Springer Lecture Notes in
Mathematics 873] (Berlin: Springer).
(1975) Logic Colloquium ’73 (Amsterdam: North Holland).
Rosen, Gideon and Burgess, John P. (2005) ‘‘Nominalism reconsidered,’’ in
Shapiro (2005), pp. 460–82.
Rükert, Helge (2004) ‘‘A solution to Fitch’s paradox of knowability,’’ in Gabbay
et al. (2004), pp. 351–80.
Russell, Bertrand (1902/1967) letter to Frege, translated from the German by
Beverly Woodward, in van Heijenoort (1967), pp. 124–5.
(1985) The Philosophy of Logical Atomism, ed. David Pears (LaSalle, IL: Open
Court).
Salerno, J. (2008) New Essays on the Knowability Paradox (Oxford: Oxford
University Press).
Salmon, Nathan (1986) Frege’s Puzzle (Cambridge, MA: MIT Press).
Schindler, Ralf-Dieter (1994) ‘‘A dilemma in the philosophy of set theory’’, Notre
Schirn, Matthias (1998) (ed.) Philosophy of Mathematics Today (Oxford: Oxford
University Press).
Scroggs, Schiller Joe (1951) ‘‘Extensions of the Lewis system S5,’’ Journal of
Symbolic Logic vol. 16, pp. 112–20.
Searle, John R. (1967) ‘‘Proper names and descriptions,’’ in Weiss (1967) vol. VI,
pp. 487–91.
(1979) ‘‘Metaphor,’’ in Martinich (1979), pp. 92–123.
Shahan, R. W. and Swoyer, C. (1979) (eds.) Essays on the Philosophy of W. V. Quine
(Norman: University of Oklahoma Press).
Shapiro, Stewart (1985) (ed.) Intensional Mathematics (Amsterdam: North
Holland).
(1987) ‘‘Principles of reflection and second-order logic,’’ Journal of Philosophical
Logic vol. 16, pp. 309–33.
(1997) Philosophy of Mathematics: Structure, Ontology, Modality (Oxford:
Oxford University Press).
(2005) (ed.) The Oxford Handbook of Philosophy of Mathematics and Logic
(Oxford: Oxford University Press).
Skinner, B. F. (1957) Verbal Behavior (New York: Appleton-Century-Crofts).
References 295
Skura, Tomasz (1995) ‘‘A Lukasiewicz-style refutation system for the modal logic
S4,’’ Journal of Philosophical Logic vol. 24, pp. 573–82.
Slupecki, Jerzy and Bryll, Grzegorz (1973) ‘‘Proof of the L-decidability of Lewis
system S5,’’ Studia Logica vol. 24, pp. 99–105.
Smullyan, Arthur (1947) review of Quine (1947a), Journal of Symbolic Logic vol. 12,
pp. 139–41.
(1948) ‘‘Modality and description,’’ Journal of Symbolic Logic vol. 13, pp. 31–7.
Soames, Scott (1984) ‘‘What is a theory of truth?’’ Journal of Philosophy vol. 84,
pp. 411–29.
(1985) ‘‘Semantics and psychology,’’ in Katz (1985), pp. 204–26.
Stalnaker, Robert (1968) ‘‘A theory of conditionals,’’ in Rescher (1968), pp. 98–112.
Stanley, Jason (2001) ‘‘Hermeneutic fictionalism,’’ in French and Wettstein
(2001), pp. 36–71.
Stich, S. and Morton, A. (2002) (eds.) Benacerraf and his Critics (London:
Blackwell).
Stigt, W. P. van (1979) ‘‘The rejected parts of Brouwer’s dissertation on the
foundations of mathematics,’’ Historia Mathematica vol. 6, pp. 385–404.
Tarski, Alfred (1931) ‘‘Sur les ensembles définissables de nombres réels,’’
Fundamenta Mathematicae vol. 17, pp. 210–39.
(1935) ‘‘Der Wahrheitsbegriff in den formalisierten Sprachen,’’ Studia
Philosophica vol. 1, pp. 261–405.
(1936) ‘‘Über den Begriff der logischen Folgerung,’’ in Actes du Congrès
International de Philosophie Scientifique, vol. VII (Paris: Hermann), pp. 1–11.
(1944) ‘‘The semantic conception of truth,’’ Philosophy and Phenomenological
Research vol. 4, pp. 341–75.
(1983a) Logic, Semantics, Metamathematics, 2nd edn, ed. J. Corcoran
(Indianapolis: Hackett).
(1983b) ‘‘On definable sets of real numbers,’’ translation of Tarski (1931) from
the French by J. H. Woodger, in Tarski (1983a), pp. 110–42.
(1983c) ‘‘The concept of truth in formalized languages,’’ translation of Tarski
(1935) from the German by J. H. Woodger, in Tarski (1983a), pp. 152–278.
(1983d) ‘‘On the concept of logical consequence,’’ translation of Tarski (1936)
from the German by J. H. Woodger, in Tarski (1983a), pp. 409–20.
Tarski, Alfred and Vaught, Robert (1956) ‘‘Arithmetical extensions of relational
systems,’’ Compositio Mathematica vol. 13, pp. 81–102.
Tarski, A., Mostowski, A., and Robinson, R. M. (1953) Undecidable Theories
Thomas, Robert (2000) ‘‘Mathematics and fiction I: identification,’’ Logique et
Analyse vol. 43, pp. 301–40.
(2002) ‘‘Mathematics and fiction II: analogy,’’ Logique et Analyse vol. 45,
pp. 185–228.
Thomason, R. H. (1984) ‘‘Combination of tense and modality,’’ in Gabbay and
Guenthner (1984), pp. 135–65.
Thomason, S. K. (1973) ‘‘A new representation of S5,’’ Notre Dame Journal of
296 References
Thomson, J. J. (1987) On Being and Saying: Essays for Richard Cartwright
(Cambridge, MA: MIT Press).
Tomberlin, James E. (1994) (ed.) Logic and Language [Philosophical Perspectives,
vol. VIII] (Atascadero, CA: Ridgeview Publishing).
Uzquiano, Gabriel (2003) ‘‘Plural quantification and classes,’’ Philosophia
Mathematica, vol. 11, pp. 67–81.
Wartofsky, Max (1963) (ed.) Proceedings of the Boston Colloquium for the Philosophy
of Science 1961/1962 (Dordrecht: Reidel).
Wehmeier, Kai (forthcoming) ‘‘Modality, mood, and descriptions,’’ to appear in
Kahle (forthcoming).
Weiss, Paul (1967) (ed.) Encyclopedia of Philosophy, 6 vols. (New York: Macmillan).
Weyl, Hermann (1944) ‘‘David Hilbert and his Mathematical Work,’’ Bulletin of
the American Mathematical Society vol. 50, pp. 612–54.
White, Leslie A. (1947) ‘‘The locus of mathematical reality: an anthropological
footnote,’’ Philosophy of Science vol. 14, pp. 289–303, reprinted in Newman
(1956), vol. IV.
Whorf, Benjamin Lee (1956) ‘‘Science and linguistics,’’ in Language, Thought, and
Reality: Selected Writings, ed. J. B. Carroll (Cambridge, MA: MIT Press),
pp. 207–19.
Williamson, Timothy (1987) ‘‘On the paradox of knowability,’’ Mind vol. 96,
pp. 256–61.
Woods, John (2002) Paradox and Paraconsistency: Conflict Resolution in the
Abstract Sciences (Cambridge: Cambridge University Press).
Wright, Crispin (1980) Wittgenstein on the Foundations of Mathematics
(Cambridge: Cambridge University Press).
(1983) Frege’s Conception of Numbers as Objects (Aberdeen: Scots Philosophical
Monographs).
Wright, G. H. von (1951) An Essay in Modal Logic (Amsterdam: North Holland).
Yablo, Steven (2000) ‘‘A paradox of existence,’’ in Hofweber and Everett (2000),
pp. 275–311.
Index
Absolute, the, see monism Boolos–Bernays set theory (BB), 119, 123, 124,
abstractness versus concreteness, 24, 31 126–7, 134
Ackermann, Diana (Felicia Nimue), 213 Borges, Jorge Luis, 7, 98
acquisition argument, 260 Bouchard, Pierre, xiii
Addison, John W., 151 Bridges, Douglas, 274
adjunction, axiom of, 138 Brouwer, L. E. J., 3, 11, 18, 55–6, 59, 81, 82, 258,
aliases, problem of, 195 274–5
alpha symbol (a), 128 Bryll, Grzegorz, 181
Alston, William, 4, 85, 87 Burgess, Alexi, 12, 167
analyticity, 6, 77–9, 80, 82, 84, 153, 206 Buss, Sam, 145
Anderson, A. R., 16, 17, 246–8, 250, 252–5 Byzantium and Istanbul, 238, 240
anonymity, problem of, 194
Anscombe, Elizabeth, 229 Cantor, Georg, 49, 104–5, 114, 115, 116, 117, 127,
anti-realism, see Dummett, Michael 129, 130, 143
Archytas of Tarentum, 52 Carnap, Rudolf, 77, 93–5, 220, 274
Aristotle, 88, 209, 211 modal logic and, 170, 172, 177, 181, 215, 216
attitudes, de dicto and de re, 193–6 ontology and, 5–6, 59–64, 68–9, 85, 87
Azzouni, Jodi, 91–2 Quine and, 69, 71–2, 74–6, 78
Castan~eda, Carlos, 64, 91
Bacon, John, 100 Cauchy, Augustin, 233
Balaguer, Mark, 63 Chihara, Charles, 1, 3, 7, 12, 32, 33–4, 35, 36, 38–9,
Barcan, Ruth C., see Marcus, Ruth Barcan 46, 52, 53, 89, 167, 260
Barcan formula, 216 Chinese, 241–2, 261
Barker, John, 12, 167 choice, axiom of (AC), 116, 151
Barwise, Jon, 278 Chomsky, Noam, 19, 71, 79, 260, 270, 271
behaviorism, 19, 79–80, 259–65, 270, 272 chronometry, 187, 192–3, 196–202
Bellarmine, Robert Cardinal, 59, 88, 256 Church, Alonzo, 228
Belnap, Nuel D., Jr., 16, 17, 246–8, 250, 252–5, 280 Church’s theorem, 12
Benacerraf, Paul, 85, 86, 88, 264 Church’s thesis, 178
Benthem, Johann van, 100 Chrysippus, 246, 253
Berkeley, George, 98 Cicero and Tully, 238, 240
Bernays, Paul, xii, 8, 9, 117, 119, 120, 124, 125, 128, classes, 9, 112–13
129, 134 Cocchiarella, Nino, 216
Berry, G. G., 149–51 Cohen, Paul, 151, 277
Bigfoot, 25–7 compositionality, 237
Birkhoff, Garrett, 180 comprehension, axiom of, 109, 135
Bishop, Errett, 274 concept (Begriff ), 114, 135
Bleak House, 50, 60, 68, 69, 73, 79, 83 conceptualism, 3, 24–30
Boole, George, 127 conditional logic, 283
Boolos, George, xii, 8–9, 53, 59, 68, 106, 112, 129, conservativeness, 45, 270, 272, 280
130, 132, 134, 137, 172, 174 constructibility, 124
297
298 Index
continuum hypothesis (CH), 277 Fitch, Frederic, 14, 185–6, 187, 196, 197, 219, 223,
Convention T, 152–3, 163, 164, 167 224, 228
Copeland, B. J., 161 Føllesdal, Dagfinn, 227, 228
Craig’s lemma, 178 formalism, 11, 135, 140, 268, 270, 273
Creswell, Max, 230 Forbes, Graeme, 217
Curley, E. M., 17 foundation, axiom of, 123, 124, 125
cut-elimination, 19, 272 foundations of mathematics, 7, 66–8
Frankel, Abraham, 116
Davidson, Donald and Davidsonianism, 162, Frege, Gottlob, 18, 81
165–6, 260, 266, 270, 271 anti-psychologism and, 3, 25, 48
Davis, Martin, 142 logicism and, 10, 78, 135–6, 137, 143, 145
definability, 149–51 names and, 153, 210, 220–1, 226, 229,
definitions, status of in mathematics, 152–3 231, 244
demonstrability, 13, 169–71, 172, 173, 177, 178–84 Frege’s theorem, 66–8, 114, 137
demonstratives and indexicals, 196 French, 239–41, 242, 244
Descartes, René, 2, 165, 166 Friedman, Harvey, xii, 17, 125, 140, 278
Devitt, Michael, 234
dialethism, 186 Galilei, Galileo, 1, 55, 59, 68, 69, 93, 95
disjunction, Ganea, Mihai, 140
intensional versus extensional, 247 Geach, Peter, 107, 229, 230, 234
meaning of, 258, 269 Gell-Mann, Murray, 55
Dickens, Charles, see Bleak House generalized-quantifier logic, 164, 277, 281
discovery, principle of, 185–96, 201 general relativity, 35, 58, 73
Dixon, Thomas, 69 Gentzen, Gerhard, 271, 272
Diogenes, 28 God, 2, 6, 47–8, 63–4, 69–70, 71–2, 92–3, 94,
dualism, 267–9, 270, 272 186, 189
Dummett, Michael, 3, 12, 18–20, 63, 82, 85, 87, Gödel, Kurt, 85, 89, 124–5, 151, 277
256, 260, 264, 266, 268 completeness theorem, 144, 155, 182
incompleteness theorems, 58, 63, 140, 142,
Edgington, Dorothy, 198 172, 270, 271
Égré, Paul, xiii Goodman, Nelson, 31–2, 33, 37, 85, 90
Einstein, Albert, 55 Goranko, Valentin, 183
epistemology, 5, 39–41, 71, 88–9 Grandy, Richard, 270, 272
naturalized versus alienated, see naturalism Greece and Hellas, 241
epsilon symbol (e), 128 Grelling, Kurt, 150
equivalence, 278–9 Grice, H. P., 78, 79, 248
essentialism, 209, 217 Grzegorczyk, Andrzej, 171
Euclid, 104 Gupta, Anil, 280
extension (Umfang), 114–15, 135, 136 Gurevich, Yuri, 282
extensionality, axiom of, 110–11, 121, 123, 124,
135, 137 Haack, Susan, 263
Hadamard, Jacques, 151
fables, 50 Haldèn, Søren, 171, 173
Fara, Michael, xii Hale, Bob, 61
Feferman, Solomon, xii, 42, 43, 179 haplism, 96, 97
Fermat–(Wiles) theorem, 139 Harman, Gilbert, xiii, 254, 266
Fetzer, James, 215 Hazen, A. P., 139, 140, 230
Feynman, Richard, 96 Heck, Richard, 10, 11, 135, 140
fictionalism, 4, 5–6, 47, 48–51, 52–7, 58, 59, 72–4, Heidegger, Martin, 95
76, 83, 91 Heijenoort, Jean van, 104
Field, Hartry, 1, 3, 7, 32, 33–4, 35, 36, 38–9, 40, 46, Hellman, Geoffrey, 4, 7, 46, 89
47, 61, 72, 89, 229 heredity, 110, 111, 113, 123
figuralism, 91 hermeneuticists, 3–7, 16, 34, 51–7, 58,
Fine, Kit, 17, 137, 161 90–2
finitism, 140, 179 Hersh, Ruben, xi
Index 299
Herzberger, Hans, 280 language of thought, 165
Hesperus and Phosphorus, 15, 221, 226, 232, Laplace, Pierre-Simon Marquis de, 233
238, 240 Lavine, Shaughn, xi
Heyting, Arend, 3, 81, 283 Lesniewski, Stanislaw, 217
Hilbert, David, 11, 55–6, 82, 127, 140–3, 144, Levy, Azriel, 117
268–9 Lewis, C. I., 13, 170, 172, 230
Hintikka, Jaakko, 174–5, 203, 235, 281 Lewis, David, xiii, 59, 63, 230, 249,
Hodges, Wilfrid, 154, 155 266, 283
holism, 11, 268, 270 limitation of size, 114, 116, 117, 129
Horsten, Leon, 214 Lindemann, Ferdinand von, 52
Hume, David, 19, 58, 72, 93, 98, 135, 136 Linsky, Leonard, 218, 231
Hume’s principle (HP), 67–8, 76, 78, 83, 136, 137, literalness, 53, 54, 56–7
270, 271 Locke, John, 220
Humphreys, Paul, 215 logic, descriptive vs prescriptive, 16, 18
logicism, 10–11, 135–40, 142, 143, 145
idealism, 3, 24–30, 98 London, see Puzzling Pierre
ideology, 86, 101, 102 Löwenheim–Skolem theorem, 155
implication versus inference, 254 Lucas, J. R., 179
impredicativity, see predicativity and lumpers and splitters, 112
impredicativity
incompleteness theorems, see Gödel, Kurt Maddy, Penelope, xi, xii, 40, 47, 57
independence-friendly (IF) logic, 281 Makinson, David, 177
indiscernibility of identicals, 107, 109, 123 Malcolm, Norman, 88, 260
indispensability, 33–4, 101 Mancini, Antonella, 138
infinitary logic, 278, 281 manifestation argument, 260
infinity, axiom of, 116, 121 Manin, Yuri, 38, 274
instrumentalism, 4, 11, 41, 47 Maoism, 275
introduction and elimination, 19–20 Marcus, Ruth Barcan, 15, 215, 218, 219, 223, 224–5,
intuitionism and intuitionistic logic, 1, 11, 18–20, 226, 233, 235
81–2, 83, 174, 270, 274, 281, 283 Matiyasevich, Yuri, 11, 142, 145
Dummett and, 1, 3, 257, 258, 267 Mauldin, Daniel, 279
maximality, principle of, 116, 117
James, William, 92–3, 95, 102 McKinsey, J. C. C., 171, 180
Jarndyce and Jarndyce, see Bleak House meaning, 12, 19, 78, 79, 80, 163–4, 257
Jeffrey, Richard, xiii, 11, 135, 140–1, 144, 250 descriptive versus prescriptive theories of,
Jensen, Ronald, 277 267, 270
truth-conditional or ‘‘verist’’ theory of, 12, 162,
Kamp, Hans, 282 257, 258, 266, 267, 269
Kant, Immanuel, 93–4 see also compositionality, disjunction,
Kaplan, David, 107, 230, 234 dualism, names, representationalism,
Kepler, Johannes, 2, 69, 93, 95 semantics, translation, transparency,
Khayyam, Omar, 52 verificationism
Kleene, S. C., 151 Menchù, Rigoberta, 64
knowability, 14, 185, 196–202 Mendel, Gregor, 80
König, Julius, 150–1 Meyer, Robert, 161
Korzybski, Alfred, 156 Mill, John Stuart, 15, 210, 220, 236, 244
Kreisel, Georg, 132, 155, 161, 174, 273, 283 Millianism, 236–44
Kripke, Saul, 14, 16, 17, 35, 167, 214, 215, 223, 233, modality and modal logic, 13, 16, 157, 160, 169–84,
235, 266, 280 185, 203–4, 231
models for modal logic, 13, 129, 161, 175, 216, de dicto and de re, 14, 204–5, 209
218, 283 quantification and, 14–15, 204–27
names and, 15, 174–5, 229, 231–2, 233, 234, 238, models, 12–13, 16, 157–61, 174–6
240, 242, 244 monism, 7, 100–1
Kripke–Platek set theory (KP), 278 Montagna, Franco, 138
Kronecker, Leopold, 52 Moore, G. E., 53
300 Index
Morning Star and Evening Star, 212, 221, 228 predicativity and impredicativity, 10, 11, 41, 42,
see also Hesperus and Phosphorus 136, 137, 145, 179
Mortensen, Chris, 17 Pressburger’s theorem, 142
Myhill, John, 219 primary versus secondary sentences, 268, 270,
mystery cards, game of, 249–50, 253 271, 272
Prior, Arthur, 14, 78, 158, 189, 217, 220–1, 224, 225,
Nabokov, Vladimir, 50 230, 270, 272, 282
Nading, Inga, 69 probability, logic of, 282
names, proper, 14, 15, 221, 224, 231–3, 236–45 provability, logic of, 160–1, 170, 171, 174, 177
naturalism, 2, 7, 48, 74, 87 purity, axiom of, 121–3, 124, 125
necessity, 13 Putnam, Hilary, 33, 34, 36, 61, 101–2, 142, 234
as analyticity, 206 Puzzling Pierre, 15, 238–40, 244–5
logical, 13–14, 169, 229–30, 234
metaphysical, 14, 169, 226–7, 229–30, 234 quantification,
Nelson, Edward, 137, 138 generalized, see generalized-quantifier logic
neo-Fregeanism, 135, 137 modality and, see modality, quantification and
neo-logicism, see logicism plural, see plurals and plural logic
neo-intuitionism, see intuitionism substitutional, 217
Neumann, John von, 86, 116, 138 quantum mechanics, 35, 58, 274
Nietzsche, Friedrich, 63 Quine, W. V., 19, 47–8, 52, 54, 68–9, 70, 74–5, 78,
nihilism, 100–1 80, 82, 85, 92, 99–100, 157–9, 262, 266,
nominalism, 1, 3–7, 20, 23–30, 31–45, 46–7, 51, 52, 268–9, 270–1
71, 72–4, 85–92, 95–7, 103 analyticity and, 76, 77–9, 82–3, 153
hermeneutic, see hermeneuticists Carnap and, 6, 69, 71–2, 75, 78, 94–5
instrumentalist, see instrumentalism modality and, 14–15, 203–29
revolutionary, see revolutionaries nominalism and, 32, 33, 34, 60, 61–2, 71–3, 85,
nonmonotonic logic, 283 90, 101–2, 276
Nootka, 98
numbers, 23–4, 27–8, 51, 52, 70–1, 86, 149 Ramsey, F. P., 136, 216, 220
Rayo, Augustin, 9
Ockhamism, 198–200, 282 Read, Stephen, 17
ontology, 6, 86, 91–2, 94–5, 98, 101, 102 realism, 1–2, 23–30, 47–8, 64, 95
metaphysical, 1, 46, 47, 72
Paderewski, 245 naturalist, see naturalism
pairing, axiom of, 116, 121 reducibility, axiom of, 136, 137
paradise, Cantor’s, 71, 127, 134 reflection, principle of, 117–19, 120, 122, 133, 134
paradoxes, 10, 12, 14, 130, 150–1, 166, 185, 196, 200 regimentation, 157
Parsons, Charles, 6, 62, 76 relativization, 118, 122, 124, 125
Parsons, Terence, 137, 140, 209, 218 relevance, logic of, 16–18, 20, 246–55
Peano postulates, 136, 137 replacement, axiom of, 112, 116, 121
Peirceanism, 200–1, 282 representationalism, 270, 273–4
Penrose, Roger, 179 revolutionaries, 3, 7, 16, 18, 34, 51, 57–8, 59, 87–9
Pi-one (1) sentences, 139, 141–2, 145 Richard, Jules, 150, 151
Plato, 28, 57, 211, 274–5 Richards, Ian, 274
Platonism, 69, 90, 95, 257, 267, 270 Riemann, Bernhard, 79
plurals and plural logic, 9, 106–9, 129–30 Robinson, Julia, 11, 142, 145
Poincaré, Henri, 150 Robinson, Raphael, 137
Pollard, Stephen, 9 Rorty, Richard, 101, 102
positivism, 271, 272 Rosen, Gideon, 3, 5, 46, 47, 51, 59, 60, 87, 90, 91
post-modernism, 102 Ross, Arnold, 149
Pour-El, Marian, 274 Rückert, Helge, xii
power, axiom of, 121 Russell, Bertrand, 10, 46, 64, 81, 136, 143, 150, 224,
pragmatism, 47, 102 225, 228, 235
Prawitz, Dag, 19, 263 Russellianism, 14, 220–1, 223, 228, 231, 232–3
predicate-functor logic, 99–100 Ryle, Gilbert, 19, 270, 272
Index 301
S4 (modal system), 13 tense logic, 157–9, 170, 185–202, 281–2
S5 (modal system), 13 Tharp, Leslie, 49
Salmon, Nathan, 233 Thomason, S. K., 177
Sandu, Gabriel, 235 Tlön, 98
Sartre, Jean-Paul, 95 transitivity, 120
Schindler, Ralf-Dieter, 112 translation, 238–44
Schlick, Moritz, 94–5 transparency, 237
Scroggs, S. J., 177 truth, 12, 149, 151–2, 154, 280
Searle, John, 19, 229
second-order logic, 131, 135, 156 union, axiom of, 121
Segerberg, Krister, 161, 281, 282 Urquhart, Alasdair, 17
selection, measurable, 279 Uzquiano, Gabriel, 9, 112
semantics, 12–13, 129–30, 159, 165, 166, 168,
216, 259 validity, 13, 169–70, 172–4, 176–7
separation, axiom of, 8, 114–15, 134 Van Fraassen, Bas, 47, 280
set theory, 104–29, 277–81 Vaught, Robert, 155, 278
see also Zermelo–Frankel set theory verificationism, 201, 257, 267, 269, 270, 272
Shapiro, Stuart, 5, 9, 57 verism, see meaning, truth-conditional
Shelah, Saharon, 279 theory of
Silver, Jack, 278 vicious circle principle, 136, 137
skepticism, 19, 96, 97, 263 Visser, Albert, 140
Skinner, B. F., 19, 80, 270, 271
Skura, Tomasz, 183 Wang, Hao, 42
Slupecki, Jerzy, 181 Wehmeier, Kai, xii
Smielew, Wanda, 138–9 Weinstein, Scott, 45
Smiley, T. J., 254 Whorf, Benjamin Lee, 97, 98
Smullyan, Arthur, and Smullyanism, 215, 219–20, Wiles, Andrew, 49, 53
221, 223, 228, 233, 235 Wilkie, Alex, 139
Soames, Scott, xiii, 260 Williamson, Timothy, xii, 193, 198
Solovay, Robert, 137, 138, 161, 174, 277 Wittgenstein, Ludwig, 88, 266
Stalin and Djugashvili, 241 Woodin, Hugh, 151
Stalnaker, Robert, 230, 283 Wright, Crispin, 60–1, 137, 263
Stanley, Jason, 52 Wright, G. H. von, 229
Strawson, P. F., 79, 229, 248
supertransitivity, 120 Yablo, Steve, 5, 49, 52, 53, 60, 87, 90, 91
Tait, William, 140, 174 Zanardo, Alberto, 230

Tarski, Alfred, 12–13, 129, 130, 138–9, 149–50, Zeno of Elea, 150
151–2, 153–7, 161, 162, 163, 166–8, 266 Zermelo, Ernst, 86, 114, 116, 125, 138, 151
Tarski–Kuratowski algorithm, 151 Zermelo–Frankel set theory (ZFC), 8–9, 11, 112,
temporal logic, see tense logic 116, 119, 124, 125, 156, 277, 278
Tennant, Neil, 17, 19 Ziff, Paul, 229

Burgess Math Model Modality

Uploaded by

Copyright:

Available Formats

Burgess Math Model Modality

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Burgess Math Model Modality

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

MATHEMATICS, MODELS, AND MODALITY

JOHN P. BURGESS is Professor in the Department of Philosophy,

Cambridge University Press

© John P. Burgess 2008

This publication is in copyright. Subject to statutory exception and to the provision of

ISBN-13 978-0-511-38618-3 eBook (EBL)

ISBN-13 978-0-521-88034-3 hardback

PART II MODELS, MODALITY, AND MORE 147

Annotated bibliography 277

The present volume contains a selection of my published philosophical

AGAINST HERMENEUTIC AND REVOLUTIONARY

‘‘Why I am not a nominalist’’ (1983) represents my first attempt to articu-

AGAINST FICTIONALIST NOMINALISM

Returning to the issue of fictionalism, in our subsequent work Rosen and

FOUNDATIONS OF MATHEMATICS: SET THEORY

As long as mathematicians adhere to the ideal of rigorous proof from explicit

8x9y8zðz 2 y $ z 2 x & fðzÞÞ

wherein f may be any formula. Needless to say, no one becomes con-

8U 8x9y8zðz 2 y $ z 2 x & UzÞ:

88uu8x9y8zðz 2 y $ z 2 x &z / uuÞ:

FOUNDATIONS OF MATHEMATICS: LOGICISM

‘‘Logicism: a new look’’ (previously unpublished) provides a concise, semi-

supplemented by an axiom to the effect that to each second-order entity X

For Hilbert, ‘‘real’’ mathematics consists of basic computational facts

logical : mathematics :: empirical : physics.

MODELS AND MEANING

‘‘Tarski’s tort’’ (the other previously unpublished item in the collection) is a

MODELS AND MODALITY

Nothing is more important in approaching modal logic than to bear

MODALITY AND REFERENCE

HERMENEUTIC CRITICISM OF CLASSICAL

In Burgess (1992) I offered qualified defense of classical logic, leaving

REVOLUTIONARY CRITICISM OF CLASSICAL

While it may be contentious whether relevantism really does provide an

Numbers and ideas

Philosophy is a subject in which there is very little agreement. This is so

4 REALISM VS NOMINALISM REVISITED

Why I am not a nominalist

Some nominalists concede that their philosophy of mathematics conflicts

Goodman’s own steps towards nominalistic reconstruction of science

2 SCIENTIFIC DISPENSABILITY AND NONEXISTENCE

Some anti-nominalists have argued that the conflict between nominalism

In short, Quine and Putnam have maintained that mathematical objects

Let us imagine a laboratory assistant to Lord Kelvin reporting the data in

5 NOMINALISM, ONTOLOGICAL AND EPISTEMOLOGICAL

I have rejected nominalism in its traditional ontological form, as the doctrine

A footnote to this passage makes it plain that Field’s ‘‘philosophical

Unless he is content to lapse into a mere instrumentalist of ‘‘as if ’’ philo-

For the reader’s convenience, I here outline the constructions of Chihara

A Chihara’s modal nominalism

(2) &ð8 numeral aÞ}ð9 numeral bÞ . . .

applications empirical conditions define only fuzzy functions. For instance,

B Field’s spatiotemporal nominalism

QUALITATIVE Representation QUANTITATIVE

QUALITATIVE Representation QUANTITATIVE

Mathematics and Bleak House

Nominalism, to repeat, is a large subject. In our book, Rosen and I

To underscore these points, I would like to contrast the comparative lack of

= x} ¼ Æx: 9X(x ¼ X & Xx)i.