Michael Devitt - Ignorance of Language

I G N O R A N C E O F L A N G UAG E
This page intentionally left blank

Ignorance of Language
M I C H A E L D EV I T T
CLAR ENDON PR ESS O X F O R D

3
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With oYces in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
ß Michael Devitt 2006
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2006
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloguing in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd., King’s Lynn, Norfolk
ISBN 0-19-925096-0 978-0-19-925096-7
1 3 5 7 9 10 8 6 4 2
Preface
This is a book in the philosophy of linguistics. In a way it began a long time ago
in 1968–69 when I was a graduate student and had my Wrst experience of
generative grammar. I audited a syntax class at Harvard given by George LakoV
and then a philosophy and linguistics class at MIT given by Noam Chomsky and
Sylvain Bromberger.
Those were heady revolutionary times in linguistics. Papers were poured out,
passed excitedly from hand to hand, and declared ‘‘way out of date’’ months before
they were published. This samizdat frenzy was easy to understand. Generative
grammar was revealing, at a furious pace, a wealth of fascinating facts about the
syntax of natural languages. I was very impressed (and still am). But I was also
very puzzled. For, along with the claims about deep structure, surface structure,
and transformations, went the claim that this was all, somehow, an account of the
mind. Surely, I thought, all the talk of generations and transformations could not
be descriptions of mental processes. Yet it often seemed as if this was indeed what
linguists intended. And if this wasn’t what they intended, what exactly did
grammatical descriptions have to do with the mind? Surely, I thought, the
grammar is describing the syntactic properties of (idealized) linguistic expres-
sions, certain sounds in the air, inscriptions on paper, and the like. These items
are produced by minds, of course, and presumably get many of their properties
somehow from minds, but they are not themselves mental. It rather looked to me
as if linguists were conXating a theory of language with a theory of linguistic
competence. That was the idea from which this book sprang.
Still, it was a very long time springing. My Wrst book, Designation (1981),
makes passing mention of the idea and criticizes a related philosophical conXa-
tion of the theory of meaning with the theory of understanding. The Wrst real
step in developing the idea was in the mid 1980s when Kim Sterelny and I wrote
Language and Reality: An Introduction to the Philosophy of Language (1987).
Naturally this included a discussion of generative grammar. And, in chapter 8,
we addressed the issue of what linguistics is about and the issue of the place of
language in the mind. We took a very unChomskian position on these issues. We
developed this position into a paper, ‘‘What’s Wrong with ‘the Right View’ ’’
(1989) (‘the Right View’ being Jerry Fodor’s witty name for the received Chom-
skian view of what linguistics is about). The paper did not go down well in
certain circles.
The next step came in June 1996 when I was invited to a conference in
Maribor on naturalized semantics, focused on my book, Coming to Our Senses
(1996). I wrote a long paper for the conference, ‘‘Chomskian Linguistics and
vi Preface
Referential Semantics’’, which had most of the main ideas for this present book.
Within a year or so, I had a draft of most of the book. This was the basis for a
much revised chapter 8 in the second edition of Language and Reality (1999).
Progress since then has been very slow. Partly this was because I got distracted by
other projects, philosophical and administrative. But I think it was largely
because I needed to have some familiarity with a great range of literature,
literature not only in linguistics but in psycholinguistics and in the psychology
of skills. I frequently found this task a bit overwhelming.
The book is concerned with the psychological reality underlying language.
I arrive at some ‘‘major conclusions’’ and some ‘‘tentative proposals’’. The con-
clusions I am fairly conWdent about, the proposals, very much less so. A central
theme of the book is that we have nowhere near enough evidence to be conWdent
about many psychological matters relevant to language; we are simply too
ignorant. Still I do think that the tentative proposals are promising guides to
the future, and more so than their alternatives. These conclusions and proposals
are very much at odds with standard Chomskian views. For example, I urge that
linguistics is not part of psychology; that the thesis that linguistic rules are
represented in the mind is implausible and unsupported; that speakers are largely
ignorant of their language; that speakers’ linguistic intuitions do not reXect
information supplied by the language faculty and are not the main evidence
for grammars; that the primary concern of linguistics should not be with
idiolects; that there is little or nothing to the language faculty; that language
processing is a fairly brute-causal associationist matter; that the rules speciWed by
‘‘Universal Grammar’’ are largely, if not entirely, innate structure rules of
thought; that if the language-of-thought hypothesis is false the rules of a language
are not psychologically real, and the rules speciWed by Universal Grammar are
not innate, in a speaker. Though many of my conclusions and proposals are
radical, the assumptions that lead to them are often familiar, sometimes even rela-
tively uncontroversial; for example, that a competence should be distinguished
from its products; that a rule can govern a process without being represented;
that language expresses thought; that linguistic competence is a skill or
ability. I sometimes felt that I was arguing for something that should be fairly
obvious.
Over the years of writing the book, I have delivered some parts of it in talks.
I gave a talk based on Chapter 2 in very many places, starting at Kings College
London in March 1997. The talk was initially entitled ‘‘Ignorance of Language’’
but in 2000 I changed the title to the more apt ‘‘Linguistics is not Psychology’’.
This turned into the publication, Devitt 2003. In 2001, I delivered a talk in
Dubrovnik, ‘‘Language Processing’’, which was the beginnings of Chapter 11.
I have delivered ‘‘Intuitions in Linguistics’’ (2006), based on Chapter 7, in several
places since 2002. Finally, I wrote ‘‘Deference: A Truth in Linguistic Relativity’’
(2002), the basis for sections 8.6 and 8.7, for the Society for Philosophy and
Psychology conference at Edmonton in 2002.
Preface vii
As one would expect, I received a great variety of responses to this material.

Among this variety, three sorts deserve comment. The Wrst sort is to my many
attributions to linguists, sometimes to Chomsky, sometimes to other linguists.
With surprising frequency I received responses along the lines of: ‘‘Nobody
believes that.’’ An extreme example was the response of an anonymous reviewer,
clearly a linguist, to an attribution in my 2006 of what I call, ‘‘The Representa-
tional Thesis’’: the thesis that the rules/principles of a language are represented in
the minds of its speakers. The response was: ‘‘it is beyond doubt that [the
Representational Thesis] is not any view in linguistics’’. What to say? One does
one’s best to get people right, citing evidence in support. In this case, I cite a great
deal of evidence. If the evidence really is misleading and the attributions false,
one wonders why people write in such misleading ways. In any case, the main
point in making the attributions is not to score points but rather to set the stage
for a positive discussion of the psychological reality underlying language. If we
can clearly identify, and reject, a view of this reality that is at least suggested in the
literature—for example, the Representational Thesis—then we can move on to
consider better ideas. We have made progress. The truth of the matter about
psychological reality is much more important than the truth of the matter about
what people have claimed about that reality.
The second sort of response simply restates a view, vigorously and with
references to Chomsky’s work, that is precisely the one criticized. But, of course,
the issues here, as elsewhere in science, cannot be settled by vigorous restatements
and appeals to authority; they should be settled by answering criticisms and by
oVering evidence and argument.
The third sort of response draws attention to this or that grammatical discov-
ery that I have not mentioned. Usually the discovery is truly fascinating and so it
is not hard to empathize with enthusiasm for it. But what needs to be asked is
whether the discovery is relevant to the issue in question. That issue is the
psychological reality underlying language. So we are concerned with what, in
general, syntactic theories, grammars, might tell us about that psychological
reality. We are concerned, one might say, with ‘‘the interpretation’’ of linguistics.
But we are not much concerned with what is going on ‘‘in the engine room’’ of
linguistics, with the syntactic details. Our concern raises questions like: Suppose
that this or that part of a generative grammar were true, what would it show us
about the place of language in the mind? A fascinating grammatical discovery
might indeed throw some light on such a question, but it might not.
Although I am not ‘‘doing linguistics’’ but talking about it, I am a philosopher
treading on linguistic turf and this can cause resentment. Occasionally, indeed, it
has led to anger and abuse. Part of the problem is, of course, that the views I am
criticizing are held passionately by many. Another part is that my style is brusque.
I have a tendency to, as we say in Australia, ‘‘call a spade a bloody shovel’’. It is too
late for me to try to change that. I don’t mean to oVend and I am sorry if I do. But
I do mean to state views clearly and provocatively. I think that Chomskian views
viii Preface
of the place of language in the mind are seriously mistaken and that we should be
exploring other options.
The inspiration for some of the ideas in this book, particularly the positive
ones in Part IV, came from the Gricean view that thought is explanatorily prior to
language and from the language-of-thought hypothesis. The locus classicus for the
latter is, of course, Jerry Fodor’s, The Language of Thought (1975). I reread this
masterpiece after developing these ideas and was reminded that Fodor also
embraces a Gricean view and urges a view of the relation of thought to language
that is rather similar to mine. Yet we are not led down the same path by these
similar views; indeed, we are in sharp disagreement over the Representational
Thesis, the matter of what a grammar is about, the language faculty, and various
other matters. Despite my tendency in this book to harp on the diVerences rather
than the similarities, I should like now to acknowledge a large debt to Fodor.
I have many others to thank: Kim Sterelny, my co-author in some of the works
that led to this one; already acknowledged helpers with Devitt and Sterelny 1989;
the audiences at the above-mentioned talks; Stephen Crain and Juan Uriagereka,
with whom I taught a course on the philosophy of linguistics in Fall 1997 at the
University of Maryland; students in classes at Maryland and the CUNY Gradu-
ate Center; Sara Bernal, JeV Bub, John Collins, Fiona Cowie, Jerry Fodor,
Cynthia Haggard, Paul Horwich, Dunja Jutronic, Guy Longworth, Bob Mat-
thews, Paul Pietroski, GeoV Pullum, Bill Ramsey, Gurpreet Rattan, Philip
Robbins, Barbara Scholz, Robert Stainton, and Steve Stich, all of whom have
given me advice on parts of the book, and, especially, Alex Barber and Eric
Margolis who commented on drafts of the whole book; my research assistants
Fritz McDonald and James Dow for valuable help. Most of all I am indebted to
Georges Rey for comments on drafts and very many discussions of the topic over
the last few years. Very likely I have forgotten some others who have helped in the
long time it has taken to write this book. If so, I apologize.
It is conventional in a preface to admit to errors in the book to follow (thus
generating the paradox of the preface). I think there is an unusually high chance
of errors in this one because of the risk of errors of commission and omission
concerning linguistic, psycholinguistic, and psychological literature. I just hope
that the errors are not crucial to the arguments.
Contents
I. LINGUISTICS IS NOT PSYCHOLOGY

1. Introduction 3
2. A Grammar as a Theory of Linguistic Reality 17
II. POSITIONS ON PSYCHOLOGICAL REALIT Y

3. Some Possible Positions on Psychological Reality 45
4. Some Actual Positions on Psychological Reality 62
III. ‘‘ PHILOSOPHICAL’’ ARGUMENTS FOR

THE REPRESENTAT IONAL THESIS
5. The Rejection of Behaviorism 87
6. Folk Psychology 89
7. Intuitions 95
IV. THE REL ATION OF L A NGUAGE

TO THOUGHT
8. Thought Before Language 125
9. A Case for the Psychological Reality of Language 142
10. Thought and the Language Faculty 163
V. L AN G UAG E U S E A N D AC QU I S I T I O N
11. Language Use 195
12. Language Acquisition 244
Glossary of Named or Numbered Items 273

References 277
Index 295
Full Contents
I. LINGUISTICS IS NOT PSYCHOLOGY

1. Introduction 3
1.1 Chomsky’s View of the Linguistic Tasks 3
1.2 The Plan 8
1.3 ClariWcations 14
2. A Grammar as a Theory of Linguistic Reality 17

2.1 Competence vs. Outputs 17
2.2 Structure Rules vs. Processing Rules 18
2.3 Respecting Structure Rules 21
2.4 Application to Linguistics 23
2.5 The Contemplated Task and the Linguistic Enterprise 30
2.6 Four Methodological Points 35
2.7 Interesting Psychological Matters 38
2.8 Conclusion 40
II. POSITIONS ON PSYCHOLOGICAL REALIT Y

3. Some Possible Positions on Psychological Reality 45
3.1 Represented vs. ‘‘Simply Embodied’’ Rules 45
3.2 Structure Rules Used as Data in Processing 53
3.3 Rational–Causal vs. Brute–Causal Processes 53
3.4 Some Possible Positions on Psychological Reality 56
3.5 Conclusion 60
4. Some Actual Positions on Psychological Reality 62

4.1 Chomsky and Positions (i) and (iii) 62
4.2 Chomsky and Position (ii) 68
4.3 Fodor, Bever, and Garrett 72
4.4 Bresnan and Kaplan 76
4.5 Berwick and Weinberg 79
4.6 Matthews 81
4.7 Conclusion 83
xii Full Contents
I I I . ‘‘PH I LO S O PH I C A L’’ A RG U M E N TS F O R
T H E R E P R E S E N TATI O N A L TH E S I S
5. The Rejection of Behaviorism 87
6. Folk Psychology 89
6.1 Philosophers vs. Chomsky 89
6.2 The Philosophers 91
6.3 Chomsky 92
6.4 Conclusion 93
7. Intuitions 95
7.1 Introduction 95
7.2 Evidence for Linguistic Theories 98
7.3 Tension in the Linguists’ View of Intuitions 100
7.4 Intuitions in General 103
7.5 Linguistic Intuitions 108
7.6 Comparison of the Modest Explanation with the Standard
Cartesian Explanation 112
7.7 A Nonstandard Cartesian Explanation of the Role of Intuitions? 117
7.8 Must Linguistics Explain Intuitions? 119
7.9 Conclusion 120
IV. TH E R E L ATIO N OF L A NGUAGE TO THO UGHT
8. Thought Before Language 125

8.1 Intentional Realism 125
8.2 Language Expresses Thought 127
8.3 Ontological Priority of Conceptual Competence 128
8.4 Explanatory Priority of Thought 132
8.5 Priority of Theoretical Interest in Thought 134
8.6 Linguistic Relativity 135
8.7 Reference Borrowing 138
8.8 Conclusion 141
9. A Case for the Psychological Reality of Language 142

9.1 Representational Theory of the Mind (RTM) 142
9.2 Language-of-Thought Hypothesis (LOTH) 145
9.3 Linguistic Competence 147
9.4 The Public-LOTH 149
Full Contents xiii
9.5 The Syntax of Mentalese 152
9.6 ‘‘Communicative’’ vs. ‘‘Cognitive’’ Conceptions 160
9.7 Conclusion 161
10. Thought and the Language Faculty 163

10.1 Brain Impairment and the Independence of Language 163
10.2 Brain Impairment and the Language Faculty 165
10.3 Thoughts and the Language Faculty 172
10.4 Chomsky on Thoughts 174
10.5 Chomsky on Conventions and Idiolects 178
10.6 Rey on ‘‘the Intentional Inexistence of Language’’ 184
10.7 Conclusion 189
V. L AN G UAG E U S E A N D AC QU I S I T I O N
11. Language Use 195
11.1 Introduction 195
11.2 ‘‘The Only Theory in Town’’ 198
11.3 Background on Linguistic Competence 201
11.4 Implausibility of Represented Rules in Linguistic Competence 203
11.5 Psychology of Skills 210
11.6 Brute–Causal Processing 220
11.7 Psychology of Language Production 230
11.8 Psychology of Language Comprehension 234
11.9 Conclusion 241
12. Language Acquisition 244

12.1 Innateness Theses 244
12.2 Evidence for Nativism 248
12.3 ‘‘The Only Theory in Town’’ and I-Representationalism
(Thesis 3r) 252
12.4 Embodiment of UG-Rules (Thesis 3) 256
12.5 Language Faculty (Thesis 1) 260
12.6 Implausibility of I-Representationalism (Thesis 3r) 266
12.7 Bearing on the Representational Thesis (RT) 270
12.8 Conclusion 271
Glossary of Named or Numbered Items 273

References 277
Index 295
PA RT I
L I N G U I S T I C S I S N OT
P S YC H O LO G Y
1
Introduction
The major concern of this book is to investigate the respect, if any, in which the
rules or principles of a language are ‘‘psychologically real’’ in its competent
speaker/hearer.1 Naturally enough, the views of Noam Chomsky will loom
large in this enterprise.
1.1 C H O M S K Y ’ S V I EW O F T H E L I N G U I S T I C TA S K S
The wonderfully successful research program in linguistics initiated and sus-

tained by Chomsky starts from the assumption that a person competent in a
language knows that language. The program then deWnes the linguistic tasks in
terms of this knowledge. Thus, at the beginning of a book called, appropriately
enough, ‘‘Knowledge of Language’’ (1986), Chomsky claims that ‘the three basic
questions that arise’ in the study of language are:
(i) What constitutes knowledge of language?
(ii) How is knowledge of language acquired?
(iii) How is knowledge of language put to use? (p. 3)
In general, talk of ‘‘knowledge’’ is very loose. This has led to some initial
diYculty in interpreting (i) to (iii). However, there is a natural interpretation
which takes Chomsky pretty much at his word. On this interpretation, his
answer to question (i) claims that competent speakers of a language have
propositional knowledge of its rules.2 This knowledge underlies the speakers’
1 Earlier versions of generative grammar talked of ‘‘rules’’, later ones emphasize the far more general
‘‘principles’’. Such diVerences as there may be between rules and principles are unimportant to my
discussion. I shall mostly just talk brieXy of rules. I shall abbreviate ‘‘speaker/hearer’’ to ‘‘speaker’’.
2 Thus Chomsky describes the knowledge as a ‘‘system of beliefs’’ (1969a: 60–1; see also 1980a:
225) and says that a child’s acquisition of a language is the discovery of ‘‘a deep and abstract
theory—a generative grammar of his language’’ (1965: 58). See also his more recent 1986, pp. 263–
73, which includes the following: ‘‘Knowledge of language involves (perhaps entails) standard
examples of propositional knowledge’’ (p. 265); ‘‘it is proper to say that a person knows that R,
where R is a rule of his or her grammar’’ (p. 268). Jerry Fodor describes Chomsky’s view as
propositional (1983: 4–10); it is the view that ‘‘your linguistic capacities . . . are . . . explained by
reference to the content of your beliefs’’ (p. 7).
4 Linguistics is not Psychology
intuitive judgments about the syntax of expressions; it underlies their ‘‘linguistic
intuitions’’.
The key point concerning the rules of the language is that a speaker stands in
an unconscious or tacit propositional attitude to these rules which are represented
in her ‘‘language faculty’’, a special ‘‘organ’’ or ‘‘module’’ of the mind. I shall call
this ‘‘the Representational Thesis’’ (‘‘RT’’). Chomsky puts the claim about
representation with characteristic Wrmness: ‘‘there can be little doubt that know-
ing a language involves internal representation of a generative procedure’’
(1991a: 9; see also 1965: 25; 1975a: 304; 1980a: 201; 1980b: 9; 2000a: 50).
The term ‘know’ is mostly used for the propositional attitude in question but,
when the chips are down, Chomsky is prepared to settle for the technical term
‘cognize’ (1975b: 164–5; 1980a: 69–70).
The key point concerning the intuitions about particular syntactic matters is
that speakers derive their intuitive judgments from their representations of rules by a
causal and rational process like a deduction:
it seems reasonably clear, both in principle and in many speciWc cases, how unconscious
knowledge issues in conscious knowledge . . . it follows by computations similar to
straight deduction. (Chomsky 1986: 270)
The intuitions are, we might say, ‘‘the voice of competence’’. So, simply in virtue
of being competent, speakers have propositional knowledge of syntactic facts;
their competence gives them ‘‘privileged access’’ to this reality. Because of this,
these intuitions provide the main evidence about the nature of the rules.3 This is
not to say that the intuitions are infallible: performance error can lead to mistakes
(Chomsky 1986: 36). Still, apart from this ‘‘noise’’, intuitions reXect the under-
lying representations of the rules of the language.
To be competent in a language is to be able to produce and understand the
expressions of that language. According to Chomsky, on our natural interpret-
ation, this competence involves representations of the rules of the language. So
those representations determine what expressions the speaker produces and
understands. According to the point about intuitions, those representations
also determine what the speaker says about those expressions in her intuitive
judgments.
On our interpretation, task (i) for a language comes down to the study of the
system of rules that is the object of the speaker’s knowledge. Chomsky calls this
object, an ‘‘I-language’’. Since the speaker’s knowledge about this I-language
constitutes her competence, task (i) is, in eVect, the study of that competence. In
attempting this task, the linguist produces a ‘‘grammar’’, which is a theory of
the I-language. That theory, hard-won by the linguist, is precisely what the
speaker tacitly knows. Task (ii) is concerned with how the speaker acquires her
competence. How much of her knowledge of the language is innate and how
3 This view of intuitions is widespread in linguistics; see section 7.1 for evidence.
Introduction 5
much learned from experience? Task (iii) is concerned with the role played by this
competence in performance. What role does her knowledge of the language play
in understanding and producing expressions of the language?
It is surely indubitable that speakers of a language do ‘‘know the language’’, in
the ordinary sense that they are competent in the language. Furthermore, the
intuitive judgments that they make about the language are generally reliable.
Still, I think that the just-described view of this knowledge and of its relation to
the judgments is highly dubious. In particular, we should doubt RT, doubt that
speakers must have propositional knowledge of the language or that they must
have representations of linguistic rules in the language faculty or anywhere else in
the mind. Rather, I shall argue, a person could be competent in a language
without representing it or knowing anything about it: she could be totally
ignorant of it.4
Because of this, and the aforementioned looseness of talk of ‘‘knowledge’’,
I think that linguistics would do better to avoid the talk: any purpose served by
talk of knowledge seems better served, by talk of ‘‘competence’’.5
The talk of representing rules raises a question: What sense of ‘represent’ do
I have in mind in RT? The sense is a very familiar one illustrated in the following
claims: a portrait of Winston Churchill represents him; a sound /the President of
the United States/ represents George W. Bush; an inscription ‘rabbit’ represents
rabbits; a certain road sign represents that the speed limit is 30 mph; the map on
my desk represents the New York subway system; the number 11 is represented
by ‘11’ in the Arabic system, by ‘1011’ in the binary system, and by ‘xi’ in the
Roman system; and, most aptly, a (general-purpose) computer that has been
loaded up with a program represents the rules of that program. Something that
represents in this sense has a semantic content, a meaning. When all goes well,
there will exist something that a representation refers to. But a representation
can fail to refer; thus, nothing exists that ‘James Bond’ or ‘phlogiston’ refer to.
Finally, representation in this sense is what various theories of reference—
description, historical-causal, indicator, and teleological—are attempting to
partly explain.6
There are, of course, other senses of ‘represent’ in ordinary English. Thus, if we
were to say that Hillary represents New York, Fido represents dogs, Jerry Fodor
4 I think that we should be similarly dubious of the thesis that speakers must have tacit
propositional knowledge about the meanings and truth conditions of sentences of their language
(1981: 95–110; 1997a: 270–5). This thesis is ubiquitous in philosophy and common in psychology
(see, e.g., Anderson 1983: 268) and elsewhere.
5 Chomsky resists talk of ‘‘competence’’ because it suggests that linguistics is concerned with an
ability. I discuss this resistance later (Ch. 6). Zenon Pylyshyn makes a plea to conWne talk of
knowledge to cases where what is known is represented (1991: 248). This is surely a move in the
right direction but I think that we should go further. I think that we should drop talk of knowledge
from serious science, even from naturalized epistemology. There is always an alternative way of
talking that is less vague and that lacks the unfortunate connotations of talk of knowledge.
6 For a critical discussion of such theories, see Devitt and Sterelny 1999.
represents cognitive science, or Olivier represents Henry V, we would be using
‘represent’ in one of these other ways. But these are not the ways in which I shall
be using ‘represent’. I shall always be using it with the sense illustrated in the last
paragraph. Thus RT is to be understood in light of this.7
Chomsky is naturally interpreted as urging RT, with ‘represent’ understood in
this way, because doing so nicely accommodates his talk of ‘‘beliefs’’, ‘‘theories’’,
‘‘propositional knowledge’’ and of intuitions being ‘‘deduced’’ from representa-
tions. RT is the core of what Jerry Fodor (1981b) calls ‘‘the Right View’’ of what a
grammar is about. RT is certainly widespread in linguistics.8 Still the interpret-
ation of Chomsky may not be right.
Chomsky’s frequent talk of speakers standing in a propositional attitude to the
rules of their language would alone amount to RT if we could be conWdent that
he subscribed to the popular ‘‘Representational Theory of the Mind’’ (‘‘RTM’’).
For, according to RTM, any propositional attitude (or thought) involves stand-
ing in a certain functional relation to a mental representation; thus a certain
belief in my head represents that Nixon was malevolent and a certain hope in
Milosevic’s mind represented that NATO would give up. Fodor conWdently
attributes RTM to Chomsky, taking him to assume ‘‘that the intentional objects
of [propositional attitudes] are ipso facto ‘internally represented’ as a matter of
nomological (or maybe metaphysical) necessity’’ (2001: 109). Still one wonders
whether this conWdence is misplaced.
Even Chomsky’s frequent talk of speakers representing the rules of their
language does not settle the matter because we cannot be conWdent that he is
using ‘represent’ in the familiar and relatively clear sense that I have endorsed
(which is also the sense relevant to RTM). Aside from its other senses in ordinary
English, ‘represent’ (and its cognates) is used fairly indiscriminately in linguistics,
psychology, and AI. Thus the cognitive psychologist John Anderson describes the
behaviorist’s S-R theory of a rat’s maze-running skill as a theory about how that
skill is ‘‘represented’’ in the rat (1980: 235–6). In the sense of ‘represent’ that I am
7 I am attempting to clarify my sense of ‘represent’ (and its cognates) by distinguishing that sense
from other ordinary ones. I am not, of course, attempting a theory of representation, a theory of
what ‘represent’, in that sense, refers to. The concerns of this book do not require such a theory
(which is just as well given how diYcult it is proving to come up with one!).
8 (i) A recent encyclopedia article endorses the view (attributed to Chomsky) that ‘‘the human
language faculty is a system of knowledge’’. ‘‘This knowledge is formal, speciWc to the language
capacity (as distinct from other cognitive capacities), and cerebrally represented. It constitutes a
biological module, putatively distinct from other knowledge bases and mental processes’’ (Grod-
zinsky 2003: 741). (ii) In discussing the distinction between knowing how and knowing that, Fodor
remarks: ‘‘my linguist friends tell me that learning how to talk a Wrst language requires quite a lot of
learning that the language has the grammar that it does’’ (1998a: 125). (iii) All but one of the views
discussed in Chapter 4 seem to take RT for granted. (iv) Two philosophers inXuenced by Chomsky,
Susan Dwyer and Paul Pietroski, base a theory of belief on the view that ‘‘ordinary speakers believe
the propositions expressed by certain sentences of linguistic theory’’ (1996: 338). (iv) Alex Barber
takes Chomsky and others to hold that a linguistic theory is ‘‘the explicit statement of certain aspects
of the content of knowledge states possessed by ordinary speakers’’. This is ‘‘the default conception
many linguists have of the project they are engaged in’’ (2003b: 3).
Introduction 7
deploying, the very essence of the behaviorist’s theory is, of course, that the rat
runs the maze without using representations. And the critics of behaviorism who
argued that the rat uses an internal map of the maze are insisting that the rat does
use a representation in this sense. Chomsky himself often seems to have some
technical sense of ‘represent’ in mind that is diVerent from the familiar one.9
So, natural as it is to attribute RT to Chomsky and other linguists, it may not
be right to do so. And some sympathetic to his research program think that this
natural interpretation is not right (as I have discovered when proposing it).
Indeed, some even go so far as to hold that the view is not to be found in
linguistics, despite the massive evidence to the contrary.10 If the interpretation is
indeed not right, what is? Chomsky must hold that the rules of the language are
embodied somehow in a speaker without being represented (in my sense, of
course) just as, say, arithmetical rules are embodied somehow in a simple
mechanical calculator without being represented (see section 3.1 below). Fur-
thermore, these embodied rules must be responsible somehow for the speaker’s
use of language and intuitive judgments about it. I would, of course, like to
interpret Chomsky correctly but I want to emphasize from the beginning that
interpreting Chomsky is not my major concern. My major concern is to evaluate
a variety of ways in which language might be psychologically real in the speaker,
whether or not they are plausibly attributed to Chomsky (or his followers). So
I shall take no Wrm stand on this matter of interpretation.
The natural interpretation attributes RT to Chomsky. If this thesis is right
then the language is psychologically real in the speaker in a strikingly robust way:
the rules of the language are represented in the speaker’s language faculty;
the language is psychologically real because the theory of the language—its
grammar—is psychologically real. Indeed, a language simply is this system of
rules encoded in the mind. Those represented rules are the reality that a grammar
is theorizing about. If that interpretation is wrong then Chomsky must hold that
a language is a system of rules embodied somehow in the language faculty
without being represented. Then those embodied rules are the reality that a
grammar is theorizing about.
It can be seen that, according to Chomsky, the reality of a language is in the
mind, whether as represented rules or as otherwise embodied rules: the reality is
psychological:
Linguistics is simply that part of psychology that is concerned with one speciWc class
of steady states, the cognitive structures that are employed in speaking and
understanding. (1975b: 160)
9 On the issue of what Chomsky means by ‘represent’ and other apparently intentional
expressions, see the fascinating exchange: Rey 2003a; Chomsky 2003; Rey 2003b. I do not have
the heart to enter into this debate.
10 ‘‘it is beyond doubt that RT is not any view in linguistics’’ (anonymous reviewer of Devitt
2006).
He has persuaded many others of this. As Robert Matthews says:
It is a measure of the depth of the conceptual revolution wrought by Noam Chomsky in
linguistics that few linguists would quarrel with his notion that theoretical linguistics is a
subWeld of psychology. (1991: 182)11
So it is not surprising that Chomsky is irritated by the oft-raised question: ‘‘Are
the rules described by a grammar ‘psychologically real’?’’ (see e.g., 1980a: 189–
201). He points out that a grammar is a scientiWc theory and so should be treated
just like any other scientiWc theory. And a scientiWc theory should be treated
realistically, for the alternative of treating it instrumentally has surely been
discredited. This yields a very fast argument for the psychological reality of the
rules described by the grammar. We have good, though not of course conclusive,
evidence for a grammar’s truth and so we have good evidence for the reality it
concerns. And, in Chomsky’s view, that reality is psychological.
Yet, on the face of it, this view of linguistics seems implausible. In any case,
Kim Sterelny and I (1987, 1989) have argued against it. Jerrold Katz (1981,
1984) and Scott Soames (1984) have argued independently along similar lines.12
Our point seems simple, even rather obvious.13 Chomsky (1986: 34–6; 1991b:
31; 1995b: 33–4; 2000a: 139–40) responded to it brieXy and dismissingly. Susan
Dwyer and Paul Pietroski (1996) made some critical remarks in passing. Louise
Antony (2003) responded critically to Soames in a 1991 talk that has been
published in a recent volume. Stephen Laurence (2003) has mounted a lengthy
attack in the same volume. Some people stopped talking to us. Beyond this, there
is no evidence that our arguments have had any eVect.
1.2 THE PLAN
My earlier argument now seems to me to have many errors. So my Wrst aim, in

the rest of Part I, is to argue the matter somewhat diVerently and, I hope, better.
I claim that there is something other than psychological reality for a grammar to
be true of: it can be true of a linguistic reality.14 One might think that this claim
11 John Collins’ recent claim is typical: ‘‘the job of linguistic theory is to understand the nature of
the language faculty’’ (2004: 516). (With surprising conWdence Collins interprets Chomsky as not
proposing the Fodorian view that the language faculty comprises propositional structures known by
the speaker.) Some do quarrel with the psychologistic interpretation of linguistics, however: ‘‘We
make no claims, naturally enough, that our grammatical theory is eo ipso a psychological theory’’
(Gazdar et al. 1985: 5). The BloomWeldian linguistics that preceded Chomsky’s revolution was very
antimentalist. Chomsky’s psychologistic interpretation was not apparent in his early works. He
claims that this was because the approach seemed to him ‘‘too audacious’’ at the time (1975c: 35).
There is some controversy about this claim; see Harris 1993: 268–9n.
12 See also related views in Dretske 1974; Sampson 1976 (crediting Margaret Gilbert); Cummins
and Harnish 1980; Stabler 1983; and George 1989b.
13 Thus, Fiona Cowie thinks that the point ‘‘should be . . . accepted as utterly uncontroversial’’
(1999: 246).
14 Earlier (1981: 92–5) I argued an analogous point against philosophers who identify semantics
with the explanation of linguistic competence or understanding.
Introduction 9
was uncontroversial and yet Chomsky and others seem to resist it. So I shall start
by arguing for the claim carefully with the help of three quite general distinc-
tions. Next, given the weight of evidence, it is plausible to think that the
grammar is indeed more or less true of that linguistic reality. Furthermore, this
reality is worthy of theoretical study in its own right, whatever the case may be
with psychological reality. So a grammar is about linguistic reality not the
language faculty. Linguistics is not part of psychology. This is my ‘‘Wrst major
conclusion’’. The grammar might also be true of a psychological reality, of course,
but to show that it is requires an explicitly psychological assumption. And, I shall
argue, it is hard to Wnd evidence for an assumption that will do the trick.
If this is right, the very fast argument for the psychological reality of linguistic
rules—we have good evidence for the grammar and so we have good evidence for
the psychological reality it describes—is revealed as not only fast but dirty. It
remains an open question whether the rules hypothesized by the grammar are
psychologically real.
In the rest of the book I shall argue that it is hard to establish psychological
assumptions that show that linguistic rules are indeed psychologically real.
My general concern will be to explore the nature of the psychological
reality underlying language, whether that reality embodies the rules or not.
Putting this another way, my general concern is with the nature of linguistic
competence.
As I have indicated, it is common to believe RT: speakers of a language
embody its rules by representing them. I shall argue that there is no signiWcant
evidence for this thesis and, given what else we know, it is implausible. This is my
‘‘second major conclusion’’. Should a linguistic rule be embodied it is unlikely to be
represented (in the sense illustrated in section 1.1). Those who are already
convinced of this can skip Chapters 5 and 6, and some of 7, 11 and 12.
Part II starts the investigation of the psychological reality issue by describing
some positions on the matter. In Chapter 3, against a background of further
distinctions, particularly the distinction between rules that govern by being
represented and applied and those that govern by being simply embodied,
I describe some alternative possible positions on psychological reality. These
vary according to whether or not the rules of the language are embodied in the
mind; whether or not some processing rules for language are represented in
the mind (cf. RT); whether or not some processing rules operate on metalin-
guistic representations of syntactic and semantic properties of linguistic items.
And there is an uncontroversial minimal position that is committed only to there
being a psychological reality that ‘‘respects’’ the linguistic structure rules. Given
the Part I conclusion that linguistics is not psychology, no position stronger than
the minimal one can be sustained without some powerful psychological assump-
tion that is independent of anything revealed by the grammar of a language. Such
an assumption is needed to conclude that the rules of the language are present
one way or another in the mind.
In Chapter 4, I attempt to place some historically interesting actual positions

among these alternative possible positions on psychological reality. Chomsky’s
own position is hard to determine. There is strong evidence that he is committed
to RT, but even that commitment is open to question. In their classic work, The
Psychology of Language (1974) Jerry Fodor, Tom Bever, and Merrill Garrett
subscribe to RT despite Wnding no work for RT in explaining language use.
Joan Bresnan and Ronald Kaplan (1982) seem to take RT for granted. They seek
a grammar that is psychologically motivated and so abandon ‘transformational’
grammars in favor of ‘lexical-function’ grammars. In contrast, Robert Berwick
and Amy Weinberg (1984), who also seem to take RT for granted, defend the
idea that a transformational grammar is psychologically real. Robert Matthews
(1991) Wnds no evidence for RT and rejects it. Still he claims that the rules of the
language are psychologically real in another way.
I turn next to the assessment of evidence from various quarters for positions on
psychological reality. In Part III I begin my case for my second major conclusion
by rejecting what we might call ‘philosophical arguments’ for RT. (The main
arguments for RT are those from language use and from language acquisition and
will be rejected in Part V.) Chapter 5 considers, very brieXy, the argument from
the rejection of behaviorism. But this rejection is compatible with many alter-
natives to RT. Chapter 6 considers the argument from the folk psychological view
that a person competent in a language ‘knows’ the language. But, despite
Chomsky’s objections, this knowledge could be mere knowledge-how. Chapter
7 considers the most interesting of the three arguments. The intuitive judgments
of speakers are treated as the main evidence for a grammar. Why are they good
evidence? The Chomskian answer involves RT: the intuitions are derived by a
rational process from a representation of linguistic rules in the language faculty.
I argue for a diVerent view based on a view of intuitions in general. Linguistic
intuitions do not reXect information supplied by represented, or even unrepre-
sented, rules in the language faculty. Rather, they are empirical central-processor
responses to linguistic phenomena diVering from other such responses only in
being fairly immediate and unreXective. And I argue that they are not the main
evidence for a grammar. These views about intuitions constitute my ‘‘third major
conclusion’’. The conclusion yields an explanation of the evidential role of
intuitions without any appeal to embodied rules of the language.
Part IV strikes a more positive note by attending to thought. I start in Chapter
8 with the folk idea that ‘language expresses thought’. This idea seems irresistible
once one has accepted that people really have thoughts, once one has accepted
‘‘intentional realism’’. The idea leads to my ‘‘fourth major conclusion’’: the psycho-
logical reality of language should be investigated from a perspective on thought.
The idea also leads to the view that conceptual competence partly constitutes
linguistic competence and so is ontologically prior to it. I next argue, following
Grice, that thought is explanatorily prior to language. These ontological and
explanatory priorities have some interesting temporal consequences. Finally, on
Introduction 11
the basis of these priorities, I argued that our theoretical interest in thought is
prior to that in language. I sum up these priority claims roughly as follows:
thought has a certain priority to language ontologically, explanatorily, temporally,
and in theoretical interest. That is my ‘‘Wfth major conclusion’’. The assumptions
of the arguments in this chapter are, it seems to me, relatively uncontroversial
and yet their repercussions for the issue of the psychological reality of language
are both considerable and strangely unnoticed.
This discussion lays the groundwork for a decidedly speculative proposal in
Chapter 9. The argument for this begins with the popular RTM described a few
pages back. On this view a thought involves a mental representation. The crucial
next step in the argument is the controversial ‘‘Language-of-Thought Hypothesis’’
(‘‘LOTH’’): the representation in thought is language-like. On the basis of this
hypothesis, I argue that the syntactic structure of this representation is likely to be
similar to that of the sentence that expresses it in the thinker’s language. So, a
language is largely psychologically real in a speaker in that its rules are similar to
the structure rules of her thought. Linguistic competence should be seen as an
ability to translate back and forth between the speaker’s ‘Mentalese’ and her
natural language.
The argument for this proposal depends on an assumption about thought that
is far from indubitable. So the proposal is tentative; it is my ‘‘Wrst tentative
proposal’’. If we were to adopt a diVerent assumption about thought—for
example that mental representations are map-like—then we could not draw
this conclusion about the place of language in the mind. Indeed, I doubt that
we could draw any positive conclusion about this. So this argument illustrates the
point that some powerful assumption about the mind is needed to support a
robust view of the psychologically reality of language.
In Chapter 10, I Wrst consider the bearing of these views of thought on a belief
in the language faculty. The view that conceptual competence is part of linguistic
competence already places a considerable part of linguistic competence in the
central processor. I argue that this view is not undermined by the well-known
dissociation of cognitive impairment and linguistic impairment. The rest of
linguistic competence consists in processing competencies for spoken language,
written language, and so on. If there is to be a modality-neutral, relatively central,
language faculty, it must be found in some commonalities between these mo-
dality-speciWc processing competencies. I argue that, contrary to the received
view, the evidence from brain impairment counts much more against there being
the required commonalities than for there being them. Aside from that, if we
adopt my Wrst tentative proposal, we should not expect to Wnd these common-
alities. I am led to the view that there is little or nothing to the language faculty:
this is my ‘‘second tentative proposal ’’. I say more in support of this in Chapter 12.
I turn next to Chomsky’s views of thought and its relation to language. It is
surprisingly diYcult to say what these views are. I oVer some reasons for thinking
that they are very diVerent from the ones I present. I go on to look critically at
some puzzling claims that Chomsky makes against linguistic regularities and
conventions, against the need for shared and discoverable meanings in commu-
nication, and in favor of idiolects rather than common languages as the objects of
research. I emphasize that it is because of shared conventional meanings in a
group that language can play its important role of making the thoughts of each
member of the group accessible to the others. It is the task of linguistics to explain
the nature of these conventional meanings. This yields my ‘‘sixth major conclu-
sion’’: the primary concern in linguistics should not be with idiolects but with
linguistic expressions that share meanings in idiolects. I develop these views by
contrasting them with Georges Rey’s rather startling antirealism about linguistic
entities.
Finally, in Part V, I attend to language use and language acquisition. My main
argument in Chapter 11 is that a consideration of language use provides no
persuasive evidence for RT and that RT is implausible. Not only is RT not now
part of a good explanation of language use, it is unlikely to be so in the future. So,
this argument rejects the apparently popular ‘only-theory-in-town’ abduction for
RT. Indeed, I claim that this abduction fails all the main criteria of a good
abduction. My argument places a lot of weight on what we can learn from the
general psychology of skills and their acquisition. For, linguistic competence is a
skill apparently acquired by ‘‘implicit learning’’. This argument against RT is the
main step in the case for my second major conclusion. Suppose, then, that we
drop RT. It might still be thought that language processing is governed by the
unrepresented structure rules of the language. I argue against this thought: the
structure rules are the wrong sort of rule to govern the translation process of
language use. This is my ‘‘third tentative proposal’’. The proposal has to be
tentative given the consensus that we know so little about this processing. Still,
the proposal seems much more plausible than the alternative view that the
structure rules are processing rules, a view for which the study of language use
seems to provide no evidence. Finally, the chapter considers whether language
processing involves metalinguistic representations of the syntactic and semantic
properties of linguistic expressions. The received view, even among those who do
not clearly hold RT (that the linguistic rules are represented), is that the process-
ing does involve these representations. I present a case that this view is wrong,
that the speedy automatic language processes arising wholly, or at least partly,
from linguistic competence are fairly brute-causal associationist processes. This is
my ‘‘fourth tentative proposal’’. The proposal must be particularly tentative but it
is, in my view, better supported than the contrary view that these processes do
involve metalinguistic representations.
This proposal, together with the second major conclusion rejecting RT,
arises from a principle introduced in Chapter 3: ‘‘Representations are not to
be multiplied beyond necessity.’’ I call this ‘‘Pylyshyn’s Razor’’. I am not against
representations in general, as Part IV shows. However, I think that we
should posit them only if they do explanatory work. I doubt that representations
Introduction 13
of linguistic rules and properties do any work in explaining linguistic compe-
tence.
The received Chomskian view is that the rules of a language are psychologic-
ally real in a speaker, whether they are represented or not. In Chapter 9 I made
my Wrst tentative proposal, based on the controversial LOTH, that a language is
indeed largely psychologically real in a speaker in that its rules are similar to the
structure rules of her thought. The arguments in Chapter 11 are against a
language being otherwise psychologically real. This yields my ‘‘Wfth tentative
proposal’’: if LOTH is false, then the rules of a language are not, in a robust
way, psychologically real in a speaker.
The last chapter, 12, is concerned with language acquisition. I set aside for a
moment the bearing of this issue on the nature of ‘‘the Wnal state’’ of linguistic
competence, the main concern of this book. The issue raises a related concern,
the nature of ‘‘the initial state’’ of competence, that is very interesting in its own
right. The innate initial state enables humans to learn natural languages. What is
it like? This question dominates the chapter. I take the familiar arguments for
nativism to establish the interesting nativist thesis that the initial state is suY-
ciently rich that humans can naturally learn only languages that conform to the
rules speciWed by ‘‘Universal Grammar’’ (‘‘UG’’). Can we establish stronger nativist
theses? For example, can we establish the very exciting thesis that the rules
speciWed by UG are represented in the initial state? According to this thesis the
initial state includes UG itself. So the thesis is the initial state analogue of the
Wnal state Representational Thesis (RT). I call it ‘‘the I-Representational Thesis’’.
Fodor has urged an ‘‘only-theory-in-town’’ abduction in favor of this thesis. This
abduction is poor, lacking detail and evidential support. I propose an alternative
nativist thesis to I-Representationalism based on my Wrst tentative proposal:
humans are predisposed to learn languages that conform to the rules speciWed
by UG because those rules are, largely if not entirely, innate structure rules of
thought. This is my ‘‘sixth tentative proposal’’. The proposal has to be tentative
because it rests on the highly speculative Wrst proposal and because we are
nowhere near a persuasive explanation of language acquisition. Suppose then
that LOTH and hence the Wrst proposal are false and so we have to adopt some
other view of thoughts. This yields my ‘‘seventh tentative proposal’’, the initial-state
analogue of the Wnal-state Wfth proposal: if LOTH is false, then the rules speciWed
by UG are not, in a robust way, innate in a speaker. Next, I argue that, without
RT, Chomsky’s well-known argument for placing the initial state in a distinct
language faculty rather than in the central processor collapses. And if my sixth
proposal is correct that state very likely is largely in the central processor. This
completes the argument for my second proposal (Chapter 10) that there is little
or nothing to the language faculty. The proposal has to be tentative but, I argue,
the contrary view that there is a substantial language faculty seems to lack any
signiWcant evidence. Finally, I return to the I-Representationalist Thesis accord-
ing to which the rules speciWed by UG are not simply embodied in the initial
state, as in my sixth proposal, but represented in it. I have earlier rejected Fodor’s
only-theory-in-town abduction for the thesis. I now argue against the thesis. On
the basis of a comparison of linguistic competence with other skills and attention
to the nature of thoughts and thinking I claim that there is no signiWcant
evidence for the I-Representational Thesis and, given what else we know, it is
implausible. This is my ‘‘seventh major conclusion’’. It arises from another appli-
cation of Pylyshyn’s Razor. If the conclusion is right, the arguments for nativism
do not support the traditional doctrine of innate ideas.
The chapter ends by considering the bearing of this discussion of the nature of
the initial state on the nature of the Wnal state, the main concern of this book. It
has less bearing than one might have expected. Even if I-Representationalism
were true it would give no direct support to RT. And if it is false, as I argue it is,
then RT is left, once again, with no support. That Wnishes my argument for my
second major conclusion.
My fourth major conclusion urges us to investigate the psychological reality
underlying language from a perspective on thought (Chapter 8). We can now
summarize some signiWcant results of that approach. It yields my Wrst tentative
proposal, based on LOTH, that a language is largely psychologically real in a
speaker in that its rules are similar to the structure rules of her thought (Chapter 9);
and it yields my Wfth tentative proposal that if LOTH is false then the rules of a
language are not, in a robust way, psychologically real in a speaker (Chapter 11).
These proposals concern the Wnal state of linguistic competence. The approach
yields two analogous proposals about the initial state: my sixth tentative proposal
that humans are predisposed to learn languages that conform to the rules speciWed
by UG because those rules are, largely if not entirely, innate structure rules of
thought; and my seventh tentative proposal that if LOTH is false then the rules
speciWed by UG are not, in a robust way, innate in a speaker. Finally, the approach
yields the second tentative proposal, which concerns both the initial and Wnal
state: there is little or nothing to the language faculty (Chapters 10 and 12).
In sum, I shall be arguing that a number of Chomskian views about the
psychological reality of language are ill-supported by evidence and argument and
are probably wrong.
For convenience, the seven major conclusions and seven tentative proposals
are listed at the end of the Glossary of Named or Numbered items. The Glossary
also lists, for example, the Representational Thesis (RT) and some possible
positions on psychological reality.
1.3 C L A R I F I C AT I O N S
(i) A person’s I-language, according to Chomsky, supervenes on intrinsic prop-

erties of the person’s brain: it is, as philosophers would say, ‘‘narrow’’ and
individualistic. It does not involve language-world connections and so does not
Introduction 15
involve semantics proper. In eVect, an I-language has only syntactic properties, in
a broad sense of the term. This restriction reXects Chomsky’s doubts about the
scientiWc study of reference.15 I do not share these doubts and so think that the
object of study should not be an I-language but rather something ‘‘wide’’, a ‘‘wide-
I(ed)-language’’: the study should include the referential properties which, to-
gether with syntactic properties, determine the truth-referential meanings of
expressions. But this diVerence of opinion is largely beside the present issue
and so, for the sake of argument, I shall mostly go along with Chomsky’s
restriction to an I-language and to syntax (although, as Chapter 2 makes clear,
my view of the nature of language, whether I- or wide-I(ed)-, is very diVerent
from Chomsky’s).
(ii) A grammar posits various ‘‘levels’’ in an I-language. My concern is largely
with the level of ‘‘Logical Form’’ (‘‘LF’’), a level that describes all the syntactic
properties that contribute to the meanings of sentences, hence a level that makes
the scope of all quantiWers explicit. So the main issue is about the psychological
reality of the rules at that level. I am hardly concerned at all with the (fascinating)
level of ‘‘Phonological Form’’ (‘‘PF’’) partly because, in an important respect, it is
concerned with something that is not essential to language: ‘‘the cognitive
abilities supporting language and its acquisition in humans are not restricted or
specialized to speech but rather permit the development of signed as well as
spoken languages’’ (Newport and Supalla 1999: 759).
(iii) Chomsky’s revolution in linguistics has really been a series of revolutions
moving from the early Standard Theory to Government and Binding and
recently to Minimalism. The concern of this book is not with the linguistic
issue of which rules of a language are right, it is with the place of the right rules,
whichever they may be, in the mind. The concern is not with the linguistic details
but with an issue raised by the whole enterprise. So, for the most part, I shall
abstract from the diVerences between the members of this series.
(iv) Consideration of the psychological reality of language raises issues that are
undoubtedly diYcult. How important are they? What hangs on the outcomes of
this consideration? One accomplishment of generative grammar certainly does
not hang on them: the extraordinary progress in providing explicit statements of
the linguistic rules with the aim of deriving complete structural descriptions of all
the possible sentences of a language. This accomplishment strikes me as a
triumph and none of the outcomes of this book is intended to cast any doubt
on it whatsoever. What about psycholinguistics? The outcomes are clearly at odds
with the early determination of psycholinguistics to show that the linguistic rules—
transformations, for example—themselves govern language processing. But that
determination seems to have largely disappeared. I suspect that the vast majority of
the present day-to-day psycholinguistic work would not be aVected by the
outcomes. One thing is clear, however: our overall picture of the place of language
15 These doubts are nicely expressed and motivated in Pietroski 2003.

in the mind does hang on the outcomes. If I am right, the Chomskian research
program is revealing a lot about language but, contrary to advertisements, rather
little about the place of language in the mind (beyond the idea that, whatever that
place, it may be largely innate). Ordinary speakers are ignorant of language but
theorists are not; they are ignorant of linguistic competence.
Quantum mechanics provides extremely successful physical theories. Yet it is
notoriously diYcult to ‘‘interpret’’ the theories. What picture of reality are they
giving us? Controversy rages over the answer. Similarly, generative grammar is
providing extremely successful linguistic theories. There is not a similar contro-
versy over how to ‘‘interpret’’ these theories but I think that there should be. For,
according to the standard Chomskian interpretation these theories are directly
about the mind and that, I argue, is a mistake. The reality that these grammars
are primarily describing is linguistic not psychological. The grammars leave the
nature of the psychological reality underlying language largely undetermined.
Finally, a brief word about the naming conventions in this book. Philosophers
and linguists follow diVerent conventions for naming a linguistic expression:
philosophers surround the expression with single quotation marks; linguists
italicize the expression. I mostly follow the philosophical convention but occa-
sionally follow the linguistic one where linguists are under discussion. I name a
mental expression by surrounding a linguistic expression with corners. I name
meanings, contents, or propositions by capitalizing and italicizing an expression.
I name properties and relations by simply italicizing an expression.
2
A Grammar as a Theory of Linguistic Reality
My Wrst aim in this chapter is to argue, with the help of three quite general
distinctions, that there is something other than psychological reality for a
grammar to be true of: it can be true of a linguistic reality. Next, given the
evidence, it is plausible to think that the grammar is indeed more or less true of
that linguistic reality. Furthermore, this reality is worthy of theoretical study in its
own right. So, linguistics is not simply a branch of psychology. The grammar
might also be true of a psychological reality, of course, but to show that it is
requires an explicitly psychological assumption.
2.1 C O M PE T E N C E V S . O U T P U TS
A competence is a competence to produce a certain sort of output/product; or it

is a competence to process a certain sort of input; or it is both.
1. Distinguish the theory of a competence from the theory of its outputs/products or
inputs.
For convenience, in the early parts of my discussion, I shall focus on competen-
cies to produce certain sorts of outputs.
I shall draw the distinction Wrst with a simple example, distant from the
concerns of linguistics: the competence of a blacksmith and the horseshoes he
produces. Horseshoes are obvious parts of the physical world. A study of them
will quickly conclude that they are made of iron, have a certain shape, have holes
for nails, and so on. The blacksmith’s competence is some state of his mind or
body that plays the central role in explaining his behavior in producing horse-
shoes. Goodness knows what a study of it would conclude. The key point is that
the ‘‘theory’’ of the horseshoes is one thing, the theory of the competence,
another, because horseshoes are very diVerent from the competence to produce
them. Of course, given the causal relation between the competence and the
horseshoes it produces, we can expect a theory of the one to bear on a theory of
the other. But manifestly this does not make the two theories the same.
With an eye to two important features of grammar construction, we note that
there are two respects in which a theory of the outputs of a competence is not
simply about the actual outputs of that competence. First, there can be perform-
ance errors in the exercise of a competence. Thus sometimes what a blacksmith
produces is not a good horseshoe. The theory is only concerned with the nature
of the outputs of a competence when it performs as it should; the theory idealizes
by abstracting from error. Second, the theory is concerned with any possible
output of the competence (when working well). Thus, the theory of horseshoes is
concerned not only with the actual outputs of competent blacksmiths but with
any of an indeWnitely large number of outputs that they might produce.
The discussion to follow provides several other illustrations of distinction 1.
2.2 S T RU C T U R E RU L E S V S . P RO C E S S I N G RU L E S
The theory of a competence explains what it is about an object that makes it

competent. Part of the explanation must be that the object embodies rules that
govern the process of producing the appropriate output when the competence is
exercised. Call these rules ‘‘processing rules’’. Sometimes the outputs of a com-
petence are also rule-governed, but in a diVerent way: their natures are consti-
tuted by their place in a ‘‘structure’’ deWned by a system of rules. Call these rules
‘‘structure rules’’.
2. Distinguish the structure rules governing the outputs of a competence from the
processing rules governing the exercise of the competence.
In characterizing the output of the blacksmith we will not appeal to rules, but
in characterizing other outputs we will. Thus, consider the output of a chess
player: chess moves. The characterization of chess moves must appeal to a rather
elaborate system of rules: a bishop may only move diagonally; the king may only
move one square; no piece except a knight may move through an occupied
square; and so on. Chess moves are rule-governed in that something counts as a
chess move at all only if it has a place in the structure deWned by the rules of chess.
Something counts as a particular chess move in virtue of the particular rules that
govern it, in virtue of its particular place in the structure. (This was an insight of
the structuralists, of course.) A ‘‘theory’’ of the nature of chess describes these
structure rules.1 In doing so it describes constraints on the appropriate output of
a chess player. A chess player should only make moves that have a place in the
system the structure rules describe. That is, a chess player should make only legal
moves. The structure rules may also be among the rules governing the psycho-
logical process by which she produces chess moves. They may be among the
processing rules activated in the exercise of her chess competence. However, this
is not necessary and may be unlikely. In any case, the key points are that being a
1 An interesting theory of chess will describe good strategies, of course. But that is a diVerent
matter.
Grammar, Theory of Linguistic Reality 19
structure rule, a rule governing outputs, is a very diVerent property from being a
processing rule, a rule governing the psychological production of outputs; and
governing outputs is a very diVerent matter from governing the production of
outputs.
Some examples will help to bring out the distinction. Let us start by consider-
ing chess a bit more. It is a structure rule of chess that bishops move diagonally. If
this rule were also among a player’s processing rules then the player would, when
appropriate in Wguring out a move, go through a process of inferring ‘x moves
diagonally’ from ‘x is a bishop’. A player may not do that. In particular, given
what we know about grand masters, it seems unlikely that they do: they combine
a prodigious memory of chess conWgurations with strategies that may well not
involve any such simple inferences. And we can imagine a machine that has a
massive data base with an appropriate response to every possible chess conWgura-
tion and that plays chess by simply applying this data base. This machine never
goes through any inference that corresponds to a structure rule.2 (If this is a bit
hard to imagine because there are so many possible chess conWgurations, imagine
a machine that plays a simpler game like draughts/checkers in the same way.)
A nice example of our distinction is provided by the distinction between the
formation and transformation rules of a formal logic (the latter are not to be
confused with the very diVerent transformation rules of grammar). The forma-
tion rules are structure rules characterizing the wVs (well-formed formulae) of the
system: nothing counts as a wV unless it accords with those rules. In this way, wVs
are rule-governed. Each wV has its particular syntactic structure in virtue of the
particular formation rules that govern it, in virtue of its particular place in the
structure deWned by the system of rules. The transformation rules are processing
rules governing the move from one wV to another; they govern a process of valid
derivation (if the rules are good). Nothing is both a formation and a transform-
ation rule.
Think of the formal logic as embodied in a ‘‘logic machine’’. The machine
takes wVs as inputs, processes them according to the transformation rules,
yielding wVs as outputs (so it embodies a proof procedure). The outputs of
this machine are all in accord with the formation rules, but those rules are not the
ones that govern the process of producing them. The governing of wVs by
formation rules is a very diVerent matter from the governing of transformations
by processing rules.
Of course, we could build another machine, a ‘‘wV machine’’, that simply
generated wVs: it constructs wVs out of the basic symbols, the lexicon. This
process must be governed by the formation rules although in all other ways,
including selection from the lexicon, it would be random. Thus, in generating a
2 I owe this nice example to suggestions from my students, Michael Maumus and David
Pereplyotchik. It reminds one of ELIZA, the early AI program that (mostly) gives the responses
of a ‘‘competent therapist’’ but does so in a mechanical way without applying any theory of the
mind.
wV, it might start by picking a certain syntactic form. This selection is con-
strained by the formation rules but is otherwise random. Next, for each category
of term in the selected syntactic form, it randomly selects an item of that category
from the lexicon. This process might yield a simple wV like ‘Fa ’ or a more
complex one like ‘Fb & (Gc v Hd )’ or ‘(x)(Gx --> Hx)’. The particular syntactic
structure of each wV would be determined by the particular formation rules
involved in generating it. But notice that the logic machine does not generate wVs
by this process.
Bees provide another good example of the distinction between structure rules
and processing rules. A bee returning from a distant food source produces a
‘‘waggle dance’’ on the vertical face of the honeycomb. The positioning of this
dance and its pattern indicate the direction and distance of the food source.
These dances form a very eVective symbol system governed by a surprising set of
structure rules. It is the task of a theory of the dance symbols to describe these
structure rules. Karl von Frisch completed this task in the 1960s.3 In contrast, the
processing rules by which the bee performs this rather remarkable feat remain a
mystery.4
Here is a description of one of the structure rules of the bee’s dance:
To convey the direction of a food source, the bee varies the angle the waggling run makes
with an imaginary line running straight up and down . . . If you draw a line connecting the
beehive and the food source, and another line connecting the hive and the spot on the
horizon just beneath the sun, the angle formed by the two lines is the same as the angle of
the waggling run to the imaginary vertical line. (Frank 1997: 82)
How might the bee manage this? To start with it must ‘‘remember where the food
source is’’ when it gets back to the hive. How? Two popular ideas are that the bee
uses variations in Earth’s magnetic Weld or in the polarization of the sun’s light.
A wilder idea is that the bee is sensitive to quantum Welds (p. 84). Whatever the
truth of this matter, the real mystery remains: what process does the bee go
through to turn this memory into an appropriate dance, a dance governed by the
structure rule? We should not rush to the judgment that the structure rule itself
must govern this unknown process. It may be the wrong sort of rule to play this
role. Nature faced the design problem of adapting the pre-existing structures of
an insect to produce (and respond to) the message of the bee’s dance. We have no
reason to suppose a priori that nature solved this problem by making the bee go
through the structure rule ‘‘calculation’’. Indeed, it is not at all clear that the bee
could plausibly be seen as performing this calculation: can the bee even manage
3 Any kcepticism there may have been about von Frisch’s discovery should disappear in light of its
conWrmation by a recent study that involved putting radar transponders on bees (Riley et al. 2005).
4 ‘‘Scientists have known of the bee’s dance for more than seventy years, and they have assembled
a remarkably complete dictionary of its terms, but one fundamental question has stubbornly
remained unanswered: ‘‘How do they do it?’’ (Frank 1997: 80).
the necessary representations of the food source, of the spot on the horizon, and
of the angles?5
In sum, to be a processing rule that governs the activities of a chess player, logic
machine, or dancing bee is one thing, to be a structure rule governing the outputs
of such activities is another. And a structure rule of the outputs may have no place
among the processing rules that produce those outputs.
With an eye to important features of grammar construction, we have noted, Wrst,
that our theory of outputs idealizes by abstracting from performance errors. So we
are not concerned with the chess player’s moves when he is drunk, with any ‘‘noise’’
produced by the logic machine, or with the bee’s dance when it is shaken oV course.
We have noted, second, that we are concerned not only with any actual output but
with any possible output. So we are concerned with any of an indeWnitely large
number of wVs that the logic machine might produce and of dances that the bee
might perform.6 We now note, third, that we also abstract from properties of the
outputs that are irrelevant to our concerns. For example, consider a collection of
logic machines each embodying the same formal logic. One machine may produce
a ‘‘written’’ wV in one script, another, in another script; one may produce a fast high-
pitched ‘‘spoken’’ wV, another, a slow low-pitched one. We might be interested in
these diVerences and so distinguish these wVs and the competences that produce
them. But we might well not be. We may be simply interested in the rule-governed
syntactic structures of the wVs, structures shared by the outputs of all these
machines. So in our theorizing we abstract from these diVerences.
Still with an eye to important features of grammar construction, we note,
fourth, that although our theory is of the idealized output we can use it to make
distinctions among the nonideal. Moves that are not chess moves, formulae that
are not well-formed, and maneuvers that are not proper bee dances, can diVer in
their degree of failure. For, they can diVer in the sort and number of structure
rules of chess, wVs, and bee dances that they fail, respectively, to satisfy.
2.3 R E S P E C T I N G S T RU C T U R E RU L E S
Although processing rules need not include any of the structure rules, they must,
I shall say, ‘‘respect’’ them.
5 According to C. R. Gallistel (1990), the bee can: ‘‘the bees must represent the angles and
distances of food sources not only with reference to the sun but also with reference to prominent
features of the terrain surrounding the hive’’ (p. 132). Gallistel’s account of the extraordinary
navigational skills of insects makes this plausible.
6 This talk may appear to commit theories of outputs to the existence of unactualized possibilia, but
the talk can be, and in my view should be, a mere manner of speaking. It is a convenient way of capturing
that these theories, like all interesting ones, are lawlike. Strictly speaking, the theories quantify only over
actual entities but the theories are, in some sense, necessary. So the talk captures the modal fact that if
something were a horseshoe, a chess move, a wV, a bee’s dance, or whatever, then it would have the
properties speciWed by the appropriate theory of outputs. (How are we to explain modal facts? I don’t
know but, pace David Lewis, surely not in terms of unactualized possibilia.)
3. Distinguish the respecting of structure rules by processing rules from the inclusion of
structure rules among processing rules.
I have mentioned that there is a causal relation between a competence and its
output. There is also a ‘‘constitutive’’ relation. This arises from the fact that the
very nature of the competence is to produce its outputs: producing them is what
makes it the competence it is. Thus, the blacksmith’s competence is (partly) the
ability to produce horseshoes; the chess player’s, to produce chess moves, things
governed by the structure rules of chess; the logic machine’s, to produce wVs,
things governed by the formation rules; the bee’s, to produce dances, things
governed by the dance rules. So a theory of the outputs of a competence is
automatically, to that extent, a contribution to the theory of the competence, for
it tells us about the outputs the production of which is deWnitive of the
competence. And we can say that a competence and its processing rules must
‘‘respect’’ the nature of the appropriate output in that, performance errors aside,
the processing rules must produce outputs that have that nature. Where we have
to appeal to structure rules to characterize that nature, as we do with the outputs
of the chess player, the logic machine, and the bee, these structure rules must be
respected by the processing rules. Thus, whether or not the chess player actually
goes through a process of inferring ‘x moves diagonally’ from ‘x is a bishop’,
whatever processes she does go through must respect the structure rule that a
bishop moves diagonally; any moves she makes must be in accord with that rule.
And even if I am right in suggesting that the processing rules governing the bee’s
dancing cannot plausibly be seen as including the previously-described structure
rule for the direction of the food source, the processing rules must respect that
structure rule in that they produce dances that are governed by it.
I emphasize that ‘respecting’, as I am using it, is a technical term applying
primarily to a relation between a competence and its processing rules on the one
hand, and the structure rules governing the outputs of that competence on the
other hand. Occasionally I shall extend the use of the term to talk of the
competent person herself respecting those structure rules, but this should not
be confused with her respecting them in any ordinary sense. She respects them in
that her competence produces outputs that are governed by those structure rules.
Finally, note that this technical sense of ‘respecting’ is very diVerent from either
of the senses of ‘governing’ we have mentioned, the governing of a process by
processing rules and the governing of an object by structure rules.
On the strength of the fact that these structure rules must be thus respected it
may be appropriate to say that the competent object behaves as if those rules were
embodied in the object, but it is surely not appropriate to say solely on those
grounds that the rules are embodied in it. The respecting might, of course, be the
result of the rules being embodied; for example, the rules might also be processing
rules. But the respecting alone does not require that the rules be actually realized
in the speaker; for example, it does not require that they be processing rules. For
there may be many other possible ways that a competence might respect the
rules, as the chess and logic machines illustrated. So the claim that a competence
and its processing rules respect the structure rules is the minimal claim on the
internal reality issue. In a sense, this claim tells us little about the competence
because it tells us nothing about the way in which the competence respects the
structure rules. Still, we should not minimize the minimal claim. We know
something quite substantial about a bee when we know that there is something-
we-know-not-what within the bee that respects the structure rules of its dance.
And were the respected rules richer and more complicated than those of the bee’s
dance we would know something even more substantial.
It follows from the minimal claim that a theory of a competence must posit
processing rules that respect the structure rules of the outputs. Similarly, a theory
of the outputs must posit structure rules that are respected by the competence
and its processing rules. Let us capture this by saying that both theories must
meet the ‘‘Respect Constraint’’.
I have remarked that a theory of the outputs of a competence must be a
contribution to the theory of the competence. I think that we should go further:
the theory of a competence must begin with a theory of its outputs. A competence is
a competence to produce outputs with certain natures; those natures are partly
constitutive of the competence. How then could we make any signiWcant progress
studying a competence until we knew a good deal about the natures of the outputs
that it is supposed to produce? How could we start trying to solve the mystery of the
bee’s competence to dance until we knew the previously-described structure rule for
the direction of the food source? In brief, the theory of outputs has a certain
epistemic and explanatory priority over the theory of competence.
2.4 A P P L I C AT I O N TO L I N G U I S T I C S
I shall now apply this discussion to linguistics, arguing that we should see
grammars as primarily theories of linguistic not psychological reality. In the
discussion I have had an eye to certain important features of grammar construc-
tion. This was in anticipation of a certain objection to the view of linguistics I am
urging. The objection is that this view cannot be right because it cannot account
for those features. We shall see that it can and does.
Observing distinction 1, we distinguish the theory of a speaker’s competence
in a language, a psychological state, from the theory of the outputs of that
competence, sentences in the language. The competence in the language is not
the language any more than the blacksmith’s competence is a horseshoe, the chess
player’s competence is chess, or the bee’s competence is its dance. The linguistic
competence is in the mind/brain, the language is not.7 The construction of the
7 This diVerence from standard ways of talking in linguistics and psycholinguistics is not, of
course, a merely verbal one over the use of the word ‘language’: it reXects a disagreement over the
subject matter of linguistics, whatever that subject matter is called.
theory of the competence is Chomsky’s task (i), described in section 1.1. The
construction of the theory of the language is a diVerent task, one that I wish to
promote. What can we say about it?
Like the theory of the outputs of the blacksmith, chess player, logic machine,
and bee, the theory of the outputs of linguistic competence is not concerned
simply with the actual outputs. It abstracts from performance error to consider
outputs when the competence is working well. Thus we account for the Wrst
important feature of grammar construction. And our theory of outputs is
concerned with any of an indeWnitely large number of these idealized outputs
that the competence might produce, with any possible output.8 Thus we account
for a second important feature. Like the theory of the outputs of the logic
machine, our theory can abstract also from a range of properties of the out-
puts—for example, form of script and pitch of sound—focusing simply on the
syntactic properties that we are interested in. Thus we account for a third
important feature.9 The outputs of a linguistic competence, physical sentence
tokens, are governed by a system of rules, just like the outputs of the chess player,
the logic machine, and the bee. Something counts as a sentence only if it has a
place in the linguistic structure deWned by these structure rules.10 Something
counts as a particular sentence, has its particular syntactic structure, in virtue of
the particular structure rules that govern it, in virtue of its particular place in the
linguistic structure. Like the theory of the idealized outputs of the chess player,
logic machine, and bee, our theory can be used to make distinctions among the
nonideal. Strings that are not sentences can diVer in their degree of failure. For
they can diVer in the sort and number of linguistic structure rules that they fail to
satisfy. Thus we account for a fourth important feature.
Observing distinction 2, we distinguish these structure rules from processing
rules involved in the exercise of linguistic competence. These two sorts of rules
have very diVerent roles. The processing rules produce sentences of the language
in the exercise of linguistic competence. It is because those sentences are governed
by the structure rules that they are indeed sentences of the language. It may be
possible that a structure rule will also be a processing rule, but it is not necessary
that it be.
The linguistic structure rules are like the formation rules for the wVs of a
formal logic. Since we know the formation rules for the wVs, we could build a wV
machine that generated wVs from the lexicon. Similarly, if we knew all the
8 And, as with the earlier theories (note 6), such talk need not be construed as a commitment to
unactualized possibilia but rather as a way of capturing that the linguistic theory is lawlike. So if
something were a sentence, a wh-question, a passive, or whatever, it would have the properties
speciWed for such items by the theory.
9 We might sum up the point of this paragraph so far in the terms of Ferdinand de Saussure
(1916): our interest is in langue not parole.
10 The analogy of language with chess was a favorite of Saussure’s (1916: 20, 107). This reXected
his mistaken view of language as an autonomous system and his related rejection of reference (Devitt
and Sterelny 1999: 266).
linguistic rules, we could build a ‘‘sentence machine’’ that generated sentences

from the lexicon. This process would be governed by the linguistic rules although
in all other ways, including selection from the lexicon, it would be random.
Thus, in generating a sentence, it might start by picking a certain syntactic form.
This selection is constrained by the rules for combining words of the various
syntactic categories but is otherwise random. Next, for each category of word in
the selected syntactic form, it randomly selects a word of that category from the
lexicon. This process might yield a simple sentence like ‘Aristotle admired Plato’,
or a more complex ones like ‘Frege believed that Aristotle taught Alexander the
Great’ or ‘Every dog has Xeas’. The particular syntactic structure of each sentence
would be determined by the particular linguistic rules involved in generating
it. We noted that the processes by which the wV machine generates wVs are very
diVerent from the processes by which the logic machine does. Similarly, I shall
emphasize that the processes by which the sentence machine generates sentences
are very diVerent from the processes by which humans do (4.2).
Finally, observing distinction 3, we note that although the structure rules
governing sentences may not be among the processing rules that govern the
exercise of linguistic competence, they must be respected by the competence and
its processing rules: performance errors aside, the outputs of the process must be
sentences of the language and hence must be governed by the rules of the
language. For, it is the very nature of the competence to produce such sentences.
The claim that the structure rules of the language must be respected by the
competence and its processing rules is the minimal claim on the issue of the
psychological reality of language. In this sense, at least, we might say that
the grammar describes ‘‘what the competent speaker knows’’. And on the
strength of this minimal claim we might say that the speaker behaves as if
those linguistic structure rules were psychologically real in her, as if she embodied
them. But it is surely not appropriate to say solely on the strength of that minimal
claim that those rules are psychologically real in her, are embodied, for the claim
does not require that the rules be actually realized in her. In a sense, the claim tells
us little about linguistic competence because it tells us nothing about the way in
which the competence respects the linguistic rules. Still, we do know something
substantial about a person when we know that there is something-we-know-not-
what within her that respects the rich and complicated structure rules of a certain
natural language.
Both a theory of a person’s linguistic competence, of her knowledge of her
language, and a theory of her linguistic outputs must meet the Respect Con-
straint. A theory of the competence must posit a psychological state that respects
the rules governing the linguistic outputs. And a theory of the linguistic outputs
must posit rules that are respected by the competence and its processing rules.
On my view, a language is composed of the outputs of a linguistic competence,
symbols that are governed by a system of linguistic structure rules. That is the
reality of a language. And the task we have been contemplating, and that I wish to
promote, is the study of the nature of this reality. This is not Chomsky’s task (i),
the study of the nature of the competence itself. Indeed, at Wrst sight, the
contemplated study may seem to be alien to Chomsky’s enterprise. It may even
seem to smack of studying an ‘‘E-language’’, of which Chomsky takes a dim view:
‘‘the concept [of an E-language] appears to play no role in the theory of language’’
(1986: 26); an E-language has ‘‘no corresponding real-world object’’ (p. 27). But
it is not obvious that the outputs of linguistic competence Wt Chomsky’s descrip-
tion of an E-language. According to him an E-language is ‘‘externalized . . . in
the sense that the construct is understood independently of the properties of the
mind/brain’’ (1986: 20). And it sometimes seems as if an E-language for
Chomsky is essentially Platonic. In any case, the outputs I have identiWed,
physical sentence tokens governed by a system of linguistic rules, are certainly
not divorced from the mind/brain since they are the symbolic outputs of the
mind/brain. In studying them our object of study is not the mind/brain, of
course, but their linguistic properties are surely largely determined by the mind/
brain. Finally, the theory of them is as much concerned with real-world objects as
the theories of horseshoes, chess moves, bees’ dances, and wVs. It is often
convenient to talk of the objects posited by these theories as if they were types
not tokens, as if they were abstract Platonic objects, but this need be nothing
more than a manner of speaking: when the chips are down the objects are parts of
the spatio-temporal physical world.
Here I part company with Jerrold Katz (1981, 1984, 1996). He also favors a
linguistic task that is quite diVerent from Chomsky’s task (i), but the one he
favors is the study of a system of Platonic objects.11 For him talk of sentence
types is not a mere manner of speaking but essential to the task. He calls
Chomsky’s view ‘‘conceptualism’’ and my sort of view ‘‘nominalism’’. He takes
nominalism to have been refuted by Chomsky’s criticisms of BloomWeldian
structuralism. Yet, so far as I can see, these criticisms are not of the nominalism
of the structuralists but rather of their taxonomic methodology, a methodology in
the spirit of positivism. According to Chomsky, this methodology imposed
‘‘arbitrary and unwarranted’’ limitations on linguistics: it insisted on deWning
‘‘lower levels’’ before ‘‘higher levels’’; it was inductive instead of explanatory
(abductive); its epistemology was localist instead of Quinean holist. Indeed,
despite the explicit nominalism of the structuralists, Chomsky is prepared to
11 Dwyer and Pietroski take linguistic generalizations to be about non-mental abstract objects
but ‘‘nonetheless . . . hold that linguistics is properly construed as a branch of psychology, in that
linguists ascribe [linguistic] beliefs to agents’’ (1996: 349). This is not a good reason for placing
linguistics in psychology. (They have another reason, discussed in section 2.7 below.) According to
Dwyer and Pietroski, the ascription of beliefs is to explain the use of language and to explain intuitive
judgments about language (p. 340). Yet, the former explanation is the proper concern of psycho-
linguistics and the latter is the proper concern of epistemology (or so I shall argue; 7.7 below). On
their Katzian view of what linguistic generalizations are about, neither explanation should be the
concern of linguistics itself.
take the structuralists as implicitly concerned with the psychological reality of
language and hence not really nominalist at all (Chomsky 1975c: 30–6).12 Yet he
still thinks his methodological criticisms stand. In any case, Chomsky’s meth-
odological criticisms can be and, in my view, should be embraced by the
nominalist. In particular, we should not demand that the linguistic properties
of tokens be reduced to ‘‘brute-physical’’ intrinsic properties of the tokens. The
linguistic properties that concern us are ‘‘high-level’’ relational properties.13
There are likely to be lingering doubts about my contemplated task. One
doubt is about how the domain of study is to be determined: How do we select
the tokens to be studied from all the other behavioral outputs of speakers? And
the answer is: in the way science usually determines domains. That is, guided by
folk linguistics, we start with an intuitive idea of the domain of grammatical
tokens to be studied. We do not include many items that seem ‘‘unacceptable’’ to
speakers. As our linguistics goes scientiWc, we modify our view of the domain,
accepting some strings that we had thought ungrammatical because they were,
say, too hard to parse or ‘‘meaningless’’. We may even reject some strings
previously thought to be grammatical. Linguistics, like other sciences, largely
determines its own domain.
A second doubt may arise from the rather curious view that there aren’t really
any linguistic tokens. This view is a mistaken reaction to two facts. The Wrst is the
just-noted fact that the properties in virtue of which something is a linguistic
token are all relational. The second is the fact that tokens of the one linguistic
expression can appear in a variety of physical forms: a variety of sounds, a variety
of inscriptions, and so on. I shall discuss the view in section 10.6. Meanwhile, I
note brieXy that something can really have a certain linguistic property just as
something can really have the property of being Australian even though neither
have these properties intrinsically and even though things that have them can
diVer greatly in their physical forms.
Another doubt arises out of attitudes to BloomWeldian linguistics. From
the generative perspective, the BloomWeldian approach often appeared to be
somewhat superWcial and instrumentalist, concerned merely with describing
12 In taking this line, Chomsky follows a common and eVective pattern in realist philosophy of
science: arguing that scientists who claim to be instrumentalists follow practices that are implicitly
realist.
13 Katz has another objection to nominalism: grammars are about an inWnite number of
sentences but there cannot be an inWnite number of tokens. If there were a problem for my sort
of nominalism it would lie in its apparent commitment to nonactual possible sentences, a problem
that would arise even if we were dealing with a Wnite language (e.g. English with a limit of one
million words to a sentence). The only signiWcance of any apparent commitment to an inWnite
number of sentences is that it would guarantee that some were nonactual. But talk of there being
nonactual possible outputs of a competence can be a mere manner of speaking (notes 6 and 8). So
too can talk of there being an inWnite number of such outputs. The truth behind the talk of the
nonactual can be simply that the grammar is lawlike. And the truth behind the talk of the inWnite
can be simply that there is no limit to the number of diVerent sentence tokens that might be
governed by the rules the grammar describes.
regularities in the corpus of observed utterances rather than with the language’s
underlying generalizations. The generative focus on the psychological reality of
language is seen as the way to avoid this instrumentalism and be a realist about
linguistic theory.14 So there may be doubts about how my contemplated task can
be realist about language. But, as I have emphasized, the study of linguistic tokens
is not concerned only with actually observed tokens: like any other scientiWc
theory it is modal, concerned with any possible token. And the approach should
indeed be realist, concerned with the underlying generalizations of the language.
Linking language to the mind is important, of course—and I shall do plenty of
it—but it does not require that we collapse the contemplated task into task (i).
And the link to the mind is not needed for realism. We should be realist in
linguistics as everywhere else in science,15 as Chomsky has frequently insisted. But
we can be realist in linguistics without taking the grammar to be true of psycho-
logical reality, but rather taking it to be true of linguistic reality: all being well,
linguistic symbols really do have the properties ascribed to them by the grammar;
some really are c-commanded, some really are co-indexed, and so on.
Here is a more disturbing doubt. I have talked of studying the nature of
a sentence token, a nature that we reach by abstracting from properties that
are irrelevant to our concerns. But what are these concerns? What is our
theoretical interest in the token? It would not be enough to argue for what
Soames (1984) calls the ‘‘conceptual distinctness’’ of this task from the study
of competence. We have to show that the task is worthwhile. I suspect that
the presupposition, often the conviction, that there is no such worthwhile
task is the main reason for thinking that the linguistic task is Chomsky’s (i).
The view is that we need to take the task to be about competence for it to be
worth doing.16
Here are four reasons for thinking that my contemplated task is worthwhile.
First, it must be worthwhile if Chomsky’s task (i) is.17 For, although we have
distinguished the two tasks we have also related them in a way that makes
14 For example, consider the following quotes and the texts that surround them:
On other grounds, it is diYcult to explain why investigators continually found it necessary to revise
and modify their procedures in the light of results that were, in some unexplained sense, ‘‘unaccept-
able’’ though in no way inconsistent with the corpus of data. (Chomsky 1975c: 36)
we are interested in linguistic analyses primarily insofar as they may be claimed to represent the
knowledge speaker–hearers have of the structure of their language. (Fodor, Bever, and Garrett
1974: 40)
The shift of focus from language itself to the native speaker’s knowledge of language is the major
feature of the Chomskian tradition. (Haegeman 1994: 7)
15 See my 1997a. I frequently do missionary work for realism. (I sincerely hope that McGilvray
1999 is wrong in going beyond the usual attribution to Chomsky of the nativist view that our
biology determines our language to the attribution of the appallingly antirealist view that our
biology determines the world itself.)
16 See Laurence 2003, sec. 5, for a vigorous argument to that eVect.
17 I owe this reason to Roblin Meeks.
completing the contemplated task necessary for completing task (i). For, the
nature of the speaker’s competence studied by task (i) involves the nature of the
symbols studied by the contemplated task: those symbols are what the compe-
tence produces. Indeed, our earlier discussion (2.3) shows that the contemplated
task has a certain epistemic and explanatory priority over task (i). How could we
make any signiWcant progress studying the nature of competence in a language
unless we already knew a good deal about that language? Just as explaining the
bee’s dances is a prerequisite for discovering how the bee manages to produce
those dances, so also explaining the syntax of sentences is a prerequisite for
explaining how speakers manage to produce those sentences.
A second reason for thinking that my contemplated task is worthwhile is that
analogous ones are. This may not seem so obvious with the horsehoe, chess, and
the logic machine, but it is surely obvious with the bee’s dance. A serious
researcher spent years ‘‘cracking the code’’ of this dance, working out how it
indicates the direction and distance of the food source. His Wndings were
certainly interesting to scientists.18 The study of human language must surely
be more worthwhile and interesting than the study of the bee’s.
A third reason for thinking the task worthwhile would be that substantial and
interesting theories are fulWlling the task. In the next section I shall argue that
generative grammars are such theories.
The fourth and most important reason starts from the intuition that our
concern with sentence tokens, as with bees’ dances, is with their meanings. This
is a widely held view19 but it is unsatisfactorily vague. I have argued elsewhere
that we should be concerned with the properties of sentence tokens that enable
them to play certain striking roles in our lives, including the role of informing us
about reality; these are the ‘‘meanings’’ of tokens (1996: 2.3–2.8).20 Analogously,
the properties of bees’dances that concern us are the ones that enable them to play
their role of indicating food sources. Sentence tokens have their meanings partly
18 ‘‘Von Frisch’s Dance Language and Orientation of Bees was some four decades in the making.
By the time his papers on the bee dance were collected and published in 1965, there was scarcely an
entomomologist in the world who hadn’t been both intrigued and frustrated by his Wndings.
Intrigued because the phenomenon Von Frisch described was so startlingly complex; frustrated
because no one had a clue as to how bees managed the trick’’ (Frank 1997: 82). Von Fisch’s work was
so interesting that he was awarded the Nobel Prize. We should also note, as Michael Maumus
pointed out to me, the great interest in the work of Jean Francois Champollion and other scholars
who spent years studying the Rosetta stone to crack the code of a ‘‘dead language’’, the language of
Ptolemaic-era hieroglyphics.
19 Randy Harris calls the deWnition of linguistics as ‘‘the study of the links between sound and
meaning’’ ‘‘one that virtually all linguists would agree to’’ (1993: 5).
20 Fodor poses the ‘‘fundamental question’’ of the theory of language thus:
under certain conditions the production by speaker S of an acoustic object U which is a token of a
linguistic type belonging to the language L suYces to communicate a determinate message between
S and any other suitably situated L-speaker. How is that fact to be explained? (1975: 103)
in virtue of their syntactic properties and partly in virtue of the meanings of their
words. So, accepting the restriction to syntax for the sake of argument, the nature
of the sentence token that we need to explain is made up of the syntactic
properties in virtue of which the token can play those striking roles.
Our Wrst reason seemed to make our theoretical interest in the contemplated
task dependent on our theoretical interest in Chomsky’s task (i). On the basis of
our fourth reason, I shall soon argue for the opposite dependency (2.6: ‘‘fourth
methodological point’’).
We need to say much more about the theoretical interest of studying linguistic
symbols and I shall attempt to do so later (8.5). I think that this interest does
indeed arise out of our interest in the mind, in particular from our interest in
thoughts and their role in explaining behavior.21 But, once again, this does not
make our study psychological: in particular, it does not turn it into task (i), the
study of competence.
Doubts about my contemplated task may still linger. One aim of the Part IV
discussion of the relation of language to thought is to set such doubts to rest; see
particularly Chapter 8 and sections 10.4 to 10.6.
Is my contemplated task appropriately characterized as nominalistic? It takes
all the objects that linguistics is about to be concrete tokens, and so to that extent
it is nominalistic. Where it stands ultimately on the nominalism issue depends, of
course, on what we make of its ascription of meaning properties to those objects.
However, it seems unlikely that the nominalist would have any special diYculty
paraphrasing away this property talk. My contemplated task for linguistics is
likely to be as nominalistic as tasks in physics, biology, or economics.
2.5 T H E C O N T E M P L AT E D TA S K A N D T H E L I N G U I S T I C
ENTERPRISE
Whether or not this study of the outputs of competence is the study of an

E-language in Chomsky’s sense, and whatever the case about the psychological
reality of languages, I want to argue that there is nothing alien to the linguist’s
enterprise in the contemplated task.
There are two parts to this question. (a) What property does U have that enables it to convey the
message? (b) How are S and the other L-speaker able to exploit that property to communicate? (a) is
about L. My contemplated task is to answer (a). (b) is about competence in L which is, as I am
arguing, a diVerent but related question.
21 It is this theoretical interest that is likely to make a grammarian of English as concerned with
the outputs of Laurence’s Martians (2003: sec. 5) as with our own outputs. And it will prevent her
concern from spreading to the outputs of parrots, tape recorders, and the like (Devitt and Sterelny
1999: 145), a spread that Laurence argues is a likely consequence of not taking the Chomskian view
(2003: sec. 5).
First, these actual and possible idealized outputs, governed by a system of rules
and Wtting into a structure, are what we would normally call a language. Indeed,
wherever there is a linguistic competence there has to be such a language, for the
language is what the competence produces: the language is what the speaker is
competent in; it is deWnitive of the nature of the competence.
Second, we note that Chomsky himself often describes his task in ways that
suggest it is the one we have been contemplating. For example, consider the
following from the early pages of Syntactic Structures:
The fundamental aim in the linguistic analysis of a language L is to separate the grammat-
ical sequences which are sentences of L from the ungrammatical sequences which are
not sentences of L and to study the structure of the grammatical sequences. (1957: 13;
see also 1980a: 222)
Third, prima facie, a great deal of the work that linguists do, day by day, in
syntax and phonology is studying a language in the nominalistic sense I have
described.22 Work on phrase structure, case theory, anaphora, and so on, talk of
‘‘nouns’’, ‘‘verb phrases’’, ‘‘c-command’’, and so on, all appear to be concerned,
quite straightforwardly, with the properties of symbols of a language, symbols that
are the outputs of a competence. This work and talk seems to be concerned with
the properties of items like the very words on this page. The traditional view of
phonology is that it is concerned with ‘‘external physical phenomena produced
by behavior we call ‘speech’ . . . phenomena external to mind/brain (for ex-
ample, sounds)’’ (Burton-Roberts et al. 2000: 3).23 And, we have already
noted, four important features of grammar construction are also part of the
contemplated study: the idealization of outputs; concern with all possible out-
puts; abstraction from irrelevant properties; the making of distinctions among
the nonideal.
Fourth, the linguistic evidence adduced for a grammar bears directly on a
theory of the language in my sense; evidence about which strings of words are
grammatical; about the ambiguity of certain sentences; about statement forms
and question forms; about the synonymy of sentences that are superWcially
diVerent; about the diVerence between sentences that are superWcially similar;
and so on.
22 Or in Katz’s Platonic sense, which can be taken as simply a convenient manner of speaking of
language in my sense (2.4).
23 Burton-Roberts et al. go on: ‘‘It is diYcult to square the traditional view of phonology and its
rationale’’ with Chomsky’s internalist view of language (p. 4). They describe four responses to this
diYculty. (A) Isolate the externalist phonological part of the language faculty from the internalist
rest; the phonological part is simply ‘‘diVerent’’. (B) Reject the internalist view of language, taking
the theory of language to be ‘‘the theory of communicative behaviour and its perceptual products’’
(pp. 4–5). (C) Reject the externalist view of phonology, thus purging it of ‘‘phonetic substance’’ (p.
5); it is concerned with ‘‘mental objects’’ (p. 9). (D) Accept the internalist view of language but
exclude phonology from the language faculty (pp. 5–6). The view of language I am urging clearly
goes nicely with the traditional view of phonology and response (B). The other responses indicate
the contortions that the Chomskian view of language has brought upon phonology.
Objection: ‘‘But this so-called ‘linguistic’ evidence is largely the intuitions of
the native speaker. These arise from her underlying competence. So the evidence
bears directly on task (i) not your task.’’ Response: It is indeed true that if the
speaker’s knowledge of her language consists in her representation of its rules and
if her intuitions are derived from those representations by a causal and rational
process, then those intuitions are direct evidence for task (i) because they are
direct evidence of what rules are represented. I shall later argue against this view
of these intuitions and against the view that the linguistic evidence largely
consists in these intuitions (Ch. 7). But whatever the truth of these matters, the
point to be made now is that the intuitions are direct evidence about language in
my sense provided that we have good reason to think that they are accurate. It
does not matter to this point whether we think that they are accurate because
they are derived from representations of the rules or for some other reason. If
they are accurate they are evidence about language because language is what they
are about: they are about the grammaticality, ambiguity, etc. of linguistic symbols
or expressions. So if the intuitions are indeed derived from a representation of
linguistic rules, then they will be direct evidence for both task (i) and my
contemplated task. If, on the other hand, they are not so derived but are
nonetheless generally accurate, as I shall argue they are, then they will still be
direct evidence for my task even if only indirect evidence for task (i).
Fifth, the psycholinguistic evidence about language comprehension and acquisi-
tion, oVered to support the view that a grammar is psychologically real, bears
directly on a theory of the language, in my sense.24 Thus, concerning comprehen-
sion, evidence that speakers are sensitive to a proposed syntactic property in parsing
an expression is evidence that the expression really has that property, for it is
evidence that their competence respects the structure rules that determine that
property; see the later discussion of a ‘‘click location’’ experiment for an example
(4.3). The right theory of a language must ascribe rules to the language that
competent speakers of the language respect: the Respect Constraint. In this way,
the psycholinguistic evidence bears directly on our theory of the linguistic reality.25
And, concerning acquisition, evidence about nature and nurture showing that a
24 Adapting Chomsky’s terminology (1965: 24–7), we might say that a grammar that is justiWed
by psycholinguistic evidence as well as linguistic evidence is ‘‘explanatorily adequate’’.
Laurence (2003: sec. 5) names one of my earlier arguments (Devitt and Sterelny 1989: 514) ‘‘The
Martian Argument’’ and takes it ‘‘to question whether in principle [psycholinguistic] data are even
relevant to the evaluation of linguistic theories’’ (p. 95). I doubt that I ever questioned this but I
certainly do not question it now. One of the two advantages that Laurence claims for the Chomskian
view of linguistics over its rivals is that it brings psycholinguistic data to bear on linguistic theory.
The Chomskian view does not have this advantage over the view I am urging.
25 Cf: ‘‘A parser which is well-attuned to the competence grammar can be a source of information
about the properties of the grammar’’ (J. D. Fodor 1989: 174). My point is that the parser has to be
a source of information for the grammar because it has to be suYciently well-attuned to assign the
right syntactic structures, performance errors aside. Of course, on the received assumption that
the grammar is psychologically real and applied in parsing, evidence about parsing will obviously be
seen as bearing on the grammar; for example, see Chomsky 1980a: 200–1; Berwick and Weinberg
1984: 35. My point is that the evidence bears on the grammar even without the assumption.
language with a certain structure could or could not have been learnt by a person
from the ‘‘primary linguistic data’’ is direct evidence for or against any theory that
ascribes such a structure to a language that has been learnt by the person.26
In light of responses to a related point that Soames made about evidence
(1984), I should guard against possible misunderstandings.
(a) I am making the empirical claim that, as a matter of fact, the linguistic and
psycholinguistic evidence bears directly on a theory of language in my nomin-
alistic sense (whatever its bearing on anything else). This sort of claim about the
bearing of evidence on a theory is a familiar part of science and ordinary life. The
claim is not an attempt to impose a priori restrictions on the domain of evidence
relevant to Chomsky’s task (i) or to my contemplated task (cf. J. A. Fodor 1981b:
199–200; Chomsky 1986: 34–6; 1995: 33–4; Antony 2003; Laurence 2003:
101–4). I go along with the Duhem-Quine thesis which allows, roughly, that
anything might be evidence for anything. But it is clearly not a consequence of
that thesis that a piece of evidence bears with equal directness on all theories. It is
not a consequence, for example, that the experience of green grass bears equally
on the theory that grass is green and the theory that echidnas have spikes.
(b) I am not claiming that the linguistic evidence mentioned in my fourth
point is irrelevant to task (i). Indeed, since the processing rules of linguistic
competence must respect the structure rules, any direct evidence about the
structure rules must to that extent bear on task (i). For the same reason, the
psycholinguistic evidence mentioned in my Wfth point must also bear on task (i)
to that extent. Of course, we hope that this evidence will bear on task (i) to a
much greater extent, throwing light on the way in which competence respects the
structure rules. However, I do plan to argue that the psycholinguistic evidence
now available does not in fact throw much light on this matter and gives no
support to the view that competence respects the structure rules by representing
them; it gives no support to the Representational Thesis (RT).27
26 This evidential point seems to me to be the truth underlying the view, attributed to Chomsky,
‘‘that language acquisition is the key to understanding the nature of language’’ (Pinker 1995a: 108).
27 The second of the two advantages that Laurence (2003) claims for the Chomskian view of
linguistics over its rivals is that it confers explanatory power on linguistic theory, in particular the
power to explain language use and acquisition. (1) On the view I am urging, the power of a linguistic
theory is to be found primarily in its explanation of the properties of linguistic tokens. (2) Still, the
theory does contribute to the explanation of language use and acquisition because competence must
respect the linguistic rules ascribed by the theory. So use and acquisition phenomena that would be
predictable if those rules were the ones respected—for example, the phenomena Laurence describes
(sec. 2)—are indeed partly explained by a theory that ascribes those rules. (3) Of course, the theory
would make a greater contribution to the explanation of use and acquisition were it the case that
competence respected the linguistic rules by representing them. I shall be arguing that the psycho-
linguistic evidence does not support this thesis (nor even the more modest thesis that competence
respects the rules by embodying them without representing them). But the point to be made in
response to Laurence is: if the psycholinguistic evidence were ultimately to support the thesis, thus
expanding the explanatory power of linguistic theory, this expansion would not count against the
view of linguistics I am urging. Rather, the expansion would be welcomed as an explanatory bonus:
the theory not only explains language, it plays a larger role in the explanation of language use and
acquisition than we had any reason to expect.
In brief, my evidential point is simply that evidence that has played a big role
in linguistic and psycholinguistic theorizing bears directly on the task that I have
distinguished from Chomsky’s task (i), whether or not that evidence, or any other
evidence, bears on task (i). And my general point is that linguists appear to be
studying, partly at least, a language in my nominalistic sense.
Sixth and Wnally, the appearance that linguists are studying language in this
sense is just what we should expect given Chomsky’s assumption (on the natural
interpretation) that the competence that is the concern of task (i) is knowledge of the
language, involving the representation of its rules; i.e., given RT (1.1). For, the
language that would be thus known and represented would be the very same language
that is the output of the competence. Chomsky assumes that competence consists in
knowledge about the I-language. The point I am emphasizing is that this very I-
language is, indeed must be at the appropriate level of abstraction, the output of
that very competence. So, given Chomsky’s assumption, task (i) requires just the
same study as we have been contemplating. So it is no surprise to Wnd Chomsky
moving straight from an account of the task like the one quoted from Syntactic
Structures to the following version of task (i):
The problem for the linguist . . . is to determine . . . the underlying system of rules that
has been mastered by the speaker-hearer . . . Hence, in a technical sense, linguistic theory
is mentalistic, since it is concerned with discovering a mental reality underlying actual
behavior. (1965: 4)
Given the assumption of RT, task (i) and the contemplated task are much the
same.28 At one and the same time we study the symbolic system that is the output
of the competence and the competence itself which is a representation of that
very system.
If this is so, the contemplated task is not open to objection from Chomsky.
Given his assumption, it is a task that must be performed in performing his task
(i). The contemplated task acknowledges the link between competence and
language but diVers from task (i) in being neutral about the precise psychological
nature of that competence.
Not only must Chomsky accept the contemplated task, we should all accept it.
A competence is a competence to do something. So whenever there is a compe-
tence to investigate there is also a product of that competence to investigate.
When the output is a language, it should go without saying that its investigation
is theoretically interesting. Still, we can say why it is and I have started to do so in
the last section.
Chomskian linguists believe that the grammars they produce are about psy-
chological reality. But believing that it is so does not, of course, make it so. In this
28 A similar assumption yields a similar conXation in some philosophers of language; hence

Michael Dummett’s slogan, ‘‘a theory of meaning is a theory of understanding’’ (1975: 99).
section I have produced evidence that it is not so: the grammars are about
linguistic reality.
2.6 F O U R M E T H O D O LO G I C A L P O I N TS
If this discussion is right, it has a great deal of methodological signiWcance. I have

pointed out (1.1) that Chomsky is irritated by the issue of the psychological
reality of language. For him the only issue here is the truth of the grammar: if the
grammar is true then of course it is true of psychological reality because that is
what the grammar is about.
First methodological point. There is something theoretically interesting for a
grammar to be true about other than the internal reality of speakers just as there is
something theoretically interesting for a theory of chess moves, wVs, or bees’
dances to be true about other than the internal reality of chess players, logic
machines, or bees. The grammar might be true about a symbolic system, a
linguistic reality. Contrary to received opinion, language is not a ‘‘type of mental
entity’’ (cf. Harris 2003: 717). And Chomsky’s claim that ‘‘the language has no
objective existence apart from its mental representation’’ is false (Chomsky 1972:
169). So we can take the grammar realistically without taking it to be true of
psychological reality. Furthermore, given the weight of evidence adduced for a
grammar, it is plausible that it is (more or less) true of linguistic reality.
In light of this methodological point consider the following defense of the
psychological reality of structural descriptions:
Hypotheses about structural descriptions are true when they correspond to the relevant
facts, and the relevant facts concern the internal representations that speaker/hearers
compute when they produce and understand the sentences of their language. But if it
isn’t internal representations that make structural descriptions true, what is it? (Fodor,
Fodor, and Garrett 1975: 244)
On the view I am urging, the answer to the authors’ question is: The facts that
make structural descriptions true are the properties of the expressions described
by the descriptions. Still, my view allows an underlying truth to the authors’
claim preceding their question. For, the competence of the speaker/hearer must
respect the rules that constitute those structural properties of expressions. So
what the speaker produces and the hearer comprehends must have those prop-
erties (performance errors aside). That truth underlies the claim that the facts
that make the structural descriptions true ‘‘concern the internal representations
that speaker/hearers compute’’.
Second methodological point. The view that a grammar has any more to do
with psychological reality than the amount allowed by the minimal claim
requires a powerful psychological assumption about competence, if not Choms-
ky’s assumption then one of similar strength. Without such an assumption, the
grammar simply concerns a language system. This system is the output of
something psychological but it remains to be argued that it is itself psychological.
Of course, this does nothing to show that the grammar is not true of the
psychological reality, that the rules of the language are not actually realized in the
speaker. The point is that whether the grammar is true of psychological reality is a
further question to its being true of the linguistic reality. Settling that further
question depends on settling the truth of a powerful psychological assumption.
The psychological reality of language is not something ‘‘you get for nothing’’ with the
truth of the grammar.29 I fuss about this because I think it is hard to Wnd evidence
for a psychological assumption that will do the trick.30
Third methodological point. According to the Respect Constraint, a theory of
competence in a language must posit processing rules for comprehension and
production that respect the structure rules of the language. And the grammar
must posit structure rules that are respected by the competence and its processing
rules. So, the Respect Constraint makes the justiWcation of the grammar partly
dependent on the justiWcation of the theory of competence, and vice versa.
Beyond that, however, the grammar and the theory of competence are independent
of each other. So there should be no a priori demand that an acceptable grammar
must meet some further constraint concerning psychological reality, for example
what Robert Berwick and Amy Weinberg call ‘‘transparency’’ (1984: 38). And a
grammar should not be dismissed—as, for example, transformational grammars
were by Joan Bresnan and Ronald Kaplan (1982)—for failing to meet such
further constraints. And we should not decide which linguistic theory is right, as
Janet Fodor suggests we might, ‘‘by considering the sorts of parsing procedures
that each linguistic theory implies’’ (1989: 181). A grammar that attributes rules
that are respected by users of the language may be a perfectly adequate theory of
the language however little it can be incorporated into a model of language use.
Nor, in setting out to examine psychological reality, should we look to the
grammar for any insights beyond those arising from the Respect Constraint.
So far as the grammar is concerned, ‘‘the psychological cards can fall where they
may’’, subject only to the Respect Constraint. Similarly, the theory of the bee’s
dance, including the previously-described theory of how the dance indicates the
direction of the food source (2.2), provides no help to the theory of the bee’s
competence to dance beyond that arising from the fact that the competence must
respect the rules of the dance.
29 Cf. ‘‘Obviously, every speaker of a language has mastered and internalized a generative
grammar that expresses his knowledge of his language’’ (Chomsky 1965: 8).
30 In Devitt and Sterelny 1989, our case for the thesis that a grammar is about linguistic reality
rested heavily on the view that it was very likely not true of psychological reality. This is what
Laurence (2003: sec. 4) criticizes as ‘‘The Methodological Argument’’. I am still doubtful that the
grammar is true of psychological reality but my present case for the thesis does not rest on that
doubt.
This bears on a popular criticism of the claim that a grammar’s rules are
psychologically real.31 The criticism is that we lack evidence as to which gram-
mar’s rules are psychologically real: ‘‘If we can come up with one grammar for a
language, we can come up with many which, though they posit diVerent syntactic
rules, are equivalent in their explanation of meaning: they are equally able to
capture all the syntactically determined facts about meaning. We need psycho-
linguistic evidence to show which grammar’s rules are in fact playing the role in
linguistic processing, evidence we do not have.’’ This criticism is not quite right.
We need evidence that the syntactic rules of any grammar are processing rules.
These rules may simply be the wrong sort of rules to be processing rules, just as the
rules of the bee’s dance very likely are, and the rules of the logic machine’s
language certainly are. Suppose, as seems quite likely, that the human language
capacity is an adaptation. Then nature faced the problem of designing this
capacity out of pre-existing structures in our ancestors. We should not think,
in advance of empirical discovery, that nature solved this problem by making
humans go through processes governed by linguistic rules. We should not
suppose a priori that a correct account of the linguistic reality will describe the
psychological reality. A grammar may have nothing more to do with psycho-
logical reality than comes from its meeting the Respect Constraint.
So it was a mistake to assume that psycholinguistic evidence would decide
which of many meaning-equivalent grammars was true of psychological reality:
perhaps none of them are. We might be tempted to think that something
interesting still remains of the criticism in that we need psycholinguistic evidence
to decide which of many meaning-equivalent grammars are true of linguistic
reality.32 For, we need the psycholinguistic evidence to tell us which grammar
meets the Respect Constraint, which one posits rules that are respected by the
competence and its processing rules. The syntactic properties determined by
rules that are respected are the ones that linguistic tokens really have. But this
tempting thought is also mistaken. If two grammars positing diVerent rules really
do ‘‘capture all the syntactically determined facts about meaning’’ then they must
both meet the Respect Constraint. For, there can be nothing more to a compe-
tence respecting rules than its producing and responding to sentences that have
the syntactic properties determined by those rules. The two grammars specify
rules that determine the same syntactic properties because the grammars are
meaning-equivalent. (We shall later Wnd that something interesting does still
remain from the criticism; 11.4.)
Fourth methodological point. We have noted (2.4) that a grammar as a theory
of a language has a certain epistemic and explanatory priority over a theory of the
psychological reality underlying language. We cannot make any signiWcant
31 Devitt and Sterelny 1989 is an example. The criticism is related to what Laurence (2003: sec.
5) calls ‘‘The Martian Argument’’. It has its roots in Quine 1970.
32 I was tempted (2003: 133).
progress studying competence in a language until we know a good deal about
that language. So it is appropriate that, from the start, much of the work in
generative grammar has been directly concerned more with the linguistic than
with the psychological reality.33
I think that we can go further. Our theoretical interest in explaining compe-
tence in a language surely starts from our theoretical interest in that language.
Think of the bee once more. Were it not for our interest in the nature of the bee’s
dance, we would never have become interested in the state that manages to
produce that dance: it is because that state produces something so theoretically
interesting that the state itself is so theoretically interesting. I think that the same
goes for the state that produces language. If so, our theoretical interest in a
language is prior to our interest in its psychological reality.
Earlier (2.4) I suggested that our theoretical interest in language arises from
our interest in thoughts. I will attempt to support this later (8.5) but suppose,
meanwhile, that it is right. Now put it together with what I have just claimed. We
have the following ‘‘direction of theoretical interest’’: from thoughts to language
to linguistic competence. The relation between theories of these three relatively
distinct realities will be discussed in Part IV.
2.7 I N T E R E S T I N G P S Y C H O LO G I C A L M AT T E R S
I trust then that it is obvious that I am not suggesting that the psychological
reality underlying language is unworthy of study. Indeed, the theoretical interest
in a language leads immediately to an interest in two matters psychological.
(i) It is not enough to know that there is something-we-know-not-what within
a speaker that respects the rules of her language, any more than it is enough to
know that there is something-we-know-not-what within a bee that respects the
rules of the bee’s dance. We would like to go beyond these minimal claims to
discover the ways in which the competence of the speaker, and the competence of
the bee, respect these rules.34 But in studying these matters, to emphasize my
third methodological point, it is a mistake to insist on Wnding, or even to expect
to Wnd, embodied in the organism, processing rules that are also structure rules of
its outputs. The processing rules and structure rules have very diVerent jobs to
33 Cf:
many generativists assert that they aim to account for how children master their native languages,
but the vast majority of their analyses do not contribute to that aim. (Hornstein and Lightfoot
1981b: 7)
it is possible, and arguably proper, for a linguist (qua linguist) to ignore matters of psychology. But it
is hardly possible for a psycholinguist to ignore language. (Gazdar et al. 1985: 5)
34 Hence the frustration of entomologists mentioned in note 18.
do. We should keep a totally open mind about how the organism manages to
respect the structure rules.
The rest of this book will in fact be devoted to considering the way in which a
person competent in a language respects the rules of the language.
(ii) The language a person is competent in has one structure and not another.
We should like to know why the person speaks such a language, why the
something-we-know-not-what that she embodies respects the structure rules of
that language and not other structure rules:
we want to know why there are these social regularities and not others, or why we
consider these abstract mathematical structures and not others. Surely the facts might
be otherwise. (Chomsky 1980c: 57)35
The bee’s competence to dance is surely innate. To what extent is this also true of
a person’s linguistic competence and to what extent is that competence the result
of the person’s environment? We shall address this question in Chapter 12.
Our interest in language will surely also lead us in the end to an interest in a
very diVerent psychological matter. (iii) It is impossible to give deep explanations
of linguistic reality without appeal to the psychological: for, surely, psychological
facts together with social and environmental facts determine linguistic facts. So in
the end we will need to study the psychological in order to explain the linguistic.
But in the beginning we do not. Syntactic investigations of being c-commanded,
being doubly embedded and the like, the sort of investigations that linguists do
every day, are not psychological. Even when, in the end, we have to appeal to
psychology to explain in virtue of what tokens have these properties, the object of
our study remains linguistic. Analogously, a study of the property of the bee’s
dance that indicates the direction of the food source is not a study of the bee’s
‘‘psychology’’ even though the explanation of in virtue of what the dance has that
property appeals to inner states of the bee. A linguistic symbol, like a bee’s dance
or a horseshoe, really has its properties whatever the explanation of its having
them. The symbol objectively exists with its linguistic properties ‘‘apart from its
mental representation’’.
There seems to be some confusion on this point. (a) Consider one of the
reasons that Dwyer and Pietroski oVer in support of their view that ‘‘linguistics is
concerned with psychology’’ even though its generalizations are about linguistic
types. The reason is that we must appeal to mental states to answer the question:
‘‘by virtue of what is a (concrete) utterence u an utterance of an (abstract)
linguistic type?’’ (1996: 350). (b) Similarly, Barry Smith, describing ‘‘the cogni-
tivist position advanced by Noam Chomsky’’, takes linguistics to be ‘‘a branch of
cognitive psychology’’ because ‘‘facts about the meaning and form of expressions
are determined by the psychological states of speakers’’ (2001: 284). (c) And
35 So the view of languages that I urge does not lead, as Jerry Fodor anticipates it might, to any
lack of interest in ‘‘counterfactuals about what languages there could be’’ (2001: 118n). (Fodor is
criticizing Cowie for her inclination to endorse the view in Devitt and Sterelny 1989.)
Laurence suggests that ‘‘someone with broadly Chomskian sympathies’’ might
accept that ‘‘linguistics is about symbols’’ and yet still maintain that ‘‘it is, in the
Wrst instance, about competence’’. She can do this because
the important properties of these symbols—the properties in virtue of which symbols
have their linguistic properties—are properties pertaining to our linguistic competence,
and perhaps aspects of how these symbols are processed in language comprehension and
production . . . The important issue is not over whether linguistics is about symbols but
over the nature of the facts which determine the linguistic properties of symbols. (2003:
87–8)
This sort of reasoning is erroneous. Even if symbols had their properties in virtue
of certain mental facts that would not make the theory of those symbols about
those facts and so would not make the theory part of psychology. Indeed,
consider the consequences of supposing it would, and then generalizing: every
theory—economic, psychological, biological, etc.—would be about physical
facts and part of physics because physical facts ultimately determine everything.
A special science does not lose its own domain because that domain supervenes
on another.36
I emphasize that in claiming that linguistics is not part of psychology I am not
divorcing language from the mind. Linguistic competence, a mental state, is
causally related to language: that state produces and responds to linguistic
symbols. Furthermore, the Respect Constraint rests on the fact that the nature
of those symbols is partly constitutive of that competence. Finally, as we have just
noted, the linguistic partly supervenes on the psychological.
2.8 C O N C LU S I O N
Linguistics has something worthwhile to study apart from psychological reality of

speakers: it can study a linguistic reality. This reality is in fact being studied by
linguists in grammar construction. The study of this linguistic reality has a
certain priority over the study of psychological reality. A grammar is about
linguistic reality not the language faculty. Linguistics is not part of psychology.
This is my Wrst major conclusion.
The extraordinary advances in syntax in recent decades leave the question of
the psychological reality of language largely open. To close the question we need
other arguments and evidence, including psycholinguistic evidence about lan-
guage production, comprehension, and acquisition. That will be the concern of
the rest of this book. I shall argue that this evidence leaves the Representational
36 In my view, Laurence is making another error in assuming that the determining facts for
linguistic properties are facts about linguistic competence rather than facts about thoughts and social
and physical environment. I shall propose such a Gricean explanation later (8.4, 9.5).
Thesis (RT) unsupported and implausible. And it makes it hard to choose among
a range of other positions on the question of psychological reality.
In the next part I describe some positions on the psychological reality of
language and attempt to place the positions of Chomsky and others among
them. In doing this, I will often draw attention to ways in which writers diverge
from my methodological points.
The distinctions and methodological points made in this chapter, along with
the Respect Constraint, are listed in the Glossary.
PART II
P O S I T I O N S O N P S YC H O LO G I C A L
REALIT Y
3
Some Possible Positions on Psychological
Reality
3.1 R E P R E S E N T E D V S . ‘‘ S I M P LY E M B O D I E D ’’ RU L E S
Our argument in Part I that linguistics is not part of psychology rested on three
distinctions. To make progress now in describing alternative positions on the
psychological reality of language, we need some more distinctions. The Wrst of
these is not controversial,1 yet it does not have the prominence it should have in
the debate over psychological reality.
4. Distinguish processing rules that govern by being represented and applied from ones
that are simply embodied without being represented.
This is a distinction between two ways in which certain processing rules might
be psychologically real, two ways in which the rules might be embodied in the
speaker. Neither of these ways should be confused with a situation where an
object simply behaves as if it is governed by those processing rules. For that
situation is compatible with those rules not being embodied in the object at all.2
Consider a pocket calculator, for example. Its operations are governed by rules,
the rules of algorithms for addition, subtraction, and so on. Now let R be a rule
for addition. So the calculator should behave as if it is governed by R: its outputs
given inputs should be just what we would expect if it were governed by R. But,
of course, this does not mean that it is governed by R. There are many ways to
add and the calculator may have been so designed that it is governed by a rule
other than R; for example, R might be in the decimal notation and the governing
rule might be in the binary notation. For it to be governed by R it has to not only
1 Pylyshyn says that it ‘‘is not a distinction that Chomsky has endorsed’’ (1991: 233) and yet
occasionally it does seem to play a role in his discussions; see section 4.1 below on his response
(1980c: 56) to Elliott Sober’s peer commentary (1980).
2 On the strength of the fact that the structure rules of an object’s output must be respected by the
object’s competence I allowed earlier (2.3) that it might be appropriate to say that the object behaves
as if those rules were embodied in the object. I am here saying something a bit diVerent that certainly
is appropriate: the object behaves as if certain processing rules are embodied.
46 Positions on Psychological Reality
have the appropriate outputs for inputs but an appropriate internal organization.3
Distinction 4 is between two sorts of appropriate internal organization.
In section 1.1 I pointed out that it is natural, although perhaps not right, to
interpret Chomsky as holding RT. According to RT linguistic rules are psycho-
logically real in that they are represented (in the sense illustrated) in the speaker.
If these rules were processing rules then the representations of them would be
applied in processing; they would be ‘‘read’’. Distinction 4 draws attention to
another possibility: the rules might govern processing without being represented.
The Wrst point to be made is that any system has to have some processing rules
that govern it without being represented and applied.4 For, if there is a rule that
governs by being represented and applied, there has to be another rule that governs
the application.5 That rule might also govern by being represented and applied
but then its application has to be governed by a further rule; and so on. If this
regress is to end and any rule is to govern by being represented, there must be
some rules that govern without being represented, without being encoded; that is
a moral of Lewis Carroll’s famous dialogue between Achilles and the Tortoise
(1895). The upshot of this point is that, given any system that is governed by a
rule, it is an empirical question whether the system represents and applies the rule
or the rule is simply embodied in the system without being represented.
Computers demonstrate the point nicely. Software rules encoded in RAM can
govern the operations of a computer only because there are rules built into the
hardware that enable them to do so.
The folk seem to recognize these two forms of rule governing in two of their
many uses of the loose term ‘know’. Sometimes the term is used for knowing-how,
as in ‘Ralph knows how to swim’. Knowing-how is in the same family as skills,
abilities, and capacities. Sometimes, it is entirely cognitive; for example, knowing
how to play chess. Othertimes, it may be hardly cognitive at all; for example,
knowing how to ride a bicycle. Sometimes, it may involve representation of rules;
for example, some cases of chess know-how. Othertimes—and this is the im-
portant point for us—it is, as the folk would say, ‘‘mere’’ know-how and prima
facie does not involve the representation of rules; for example, knowing how to
ride a bicycle. In contrast, ‘know’ is sometimes used for knowing-that, as in
‘Ralph knows that R is an algorithm for addition’. Knowing-that is essentially
cognitive and propositional. According to the popular, and in my view correct
3 Pylyshyn (1980a, 1991) calls systems or processes yielding the same input/output behavior
‘‘extensionally equivalent’’ or ‘‘weakly equivalent’’. ‘‘Strongly equivalent’’ systems or processes
produce that behavior in the same way, by the same algorithm.
4 Analogously, a logical calculus that has axioms has to have inference rules that govern the
application of the axioms. (It might have no axioms, only inference rules.)
5 ‘‘knowledge doesn’t eventuate in behavior in virtue of its propositional content alone. It seems
obvious that you need mechanisms to put what you know into action.’’ (J. A. Fodor 1983: 9)
Some Possible Positions 47
(1996), Representational Theory of the Mind (RTM), this knowledge involves
representation. For example, it requires that Ralph represent R.6
In the light of this, there can be no swift move from ordinary talk of a speaker’s
‘‘knowing the language’’ to the view that speakers have any knowledge-that about
the language and represent its rules; the knowledge may be mere know-how.
I shall consider the case for a slow move later (Ch. 6).
The distinction I am emphasizing is related to the distinction between ‘‘de-
clarative’’ and ‘‘procedural’’ knowledge involved in performing tasks. This dis-
tinction, originally in AI, is widely acknowledged and frequently applied in
psychology, and is sometimes compared to the folk one between knowing-that
and knowing-how.7 The distinction is described, rather inadequately, as follows:
where declarative knowledge is explicit, accessible to consciousness, and concep-
tual, procedural knowledge is implicit, inaccessible to consciousness, and sub-
conceptual. What about the representation of the processing rules for a task? It is
clear that psychologists think that a subject has declarative knowledge of these
rules only if she represents them. So this knowledge matches one half of my
distinction. Does procedural knowledge match the other half? The answer to this
question is not so simple. First, the psychological consensus is that we still have a
lot to learn about procedural knowledge. Second, although some psychologists
clearly think that procedural knowledge does not involve represented processing
rules (e.g. Sun et al. 2001)—thus matching the other half of my distinction—
some others seem to think that it does, albeit representations of a diVerent sort
from those involved in declarative knowledge (e.g. Anderson 1983, 1993). I shall
discuss this later (11.5).
Finally, the distinction is not one between processing rules that can be
modiWed and those that cannot. Both represented and simply embodied rules
can be modiWed, but they are modiWed in diVerent ways. A represented rule can
by modiWed by changing the representation; in the case of a computer, by
reprogramming the software; in the case of a human, by changing, for example,
a belief. A simply embodied rule can be modiWed by rebuilding; in the case of a
computer, by an engineer changing the hardware; in the case of a human, by
experience changing the wetware.
We shall now consider some systems where the processing rules that govern
their operations are not plausibly seen as represented in the systems, let alone
propositionally known by the systems.
6 This needs qualiWcation because, strictly speaking, this knowledge, like belief, is dispositional:
a person knows a lot of things she has never entertained. So the view should be that ‘‘core
knowledge’’ involves representation; see Dennett 1978 and Field 1978. This qualiWcation is
unimportant to the argument and so will be ignored.
7 ‘‘The distinction between knowing that and knowing how is fundamental to modern cognitive
psychology. In the former, what is known is called declarative knowledge; in the latter, what is known
is called procedural knowledge.’’ (Anderson 1980: 223)
(i) Think of a really simple calculator, a mechanical one. When the calculator
adds it goes through a mechanical process that is governed by the rules of an
algorithm for addition. We have already noted that we cannot assume that the
calculator embodies a particular rule simply because the calculator behaves as if it
does. The present point concerns the rules that do, as a matter of fact, govern the
operations of the calculator. Perhaps these rules operate on representations of
numbers like 28 and functions like addition, but the rules themselves are not
represented in the calculator. The rules are hardwired but not encoded in the
calculator. And, of course, the calculator does not know that the governing rules
are rules for addition, subtraction, etc. We can be quite conWdent about this
because the calculator is not the sort of thing that can know about anything.
Finally, in virtue of the calculator being governed by those rules we can say, if we
like, that the arithmetical ‘‘information’’ that those rules reXect is embodied in
the calculator even though the calculator does not represent the information.8
sinf
m =
sinq
Figure 1. Redrawn from Boden, 1984, p. 153.
8 Dwyer and Pietroski wonder, surprisingly, how having a mechanism in a speaker that is ‘‘correctly
described by a linguistic theory is not tantamount to [the speaker’s] having the relevant beliefs’’ (1996:
341). It is not tantamount because the theory could correctly describe the algorithm that is embodied
in the mechanism in the way that an arithmetical algorithm is embodied in the calculator.
(ii) Consider the cartoon Figure 1. The kingWsher catches Wsh by diving into
water. It does not dive vertically, nor does it pursue Wsh underwater. So, in diving,
it must make suitable allowance for the refraction of light: the light deXects as
it leaves water for the less dense medium of air. The point of the cartoon is that
it would be absurd to suggest that the kingWsher goes through the following
process: it represents (tacitly knows) that m for water-to-air refraction is such and
such; it represents that the angle f is so and so; it represents that m¼ sin f/sin u;
it uses this information to calculate the angle u. Applying this new representation
it dives into the water at the appropriate angle. It is not plausible to suppose that
the kingWsher represents any of these facts about refraction and angles. No more
does it represent whatever rules do govern its Wshing: ‘‘mechanisms that perform
certain tasks need not embody theories of those tasks’’ (Schank 1980: 35). Those
mechanisms, reXecting information about water-to-air refraction, are simply
built into the kingWsher ‘‘by nature’’ just as the rules for the calculator are simply
built into it by an engineer.
(iii) Think of the vast amount of simple behavior, including much insect
behavior, that can be adequately explained in terms of behaviorist dispositions.
This behavior is governed by a simple S-R rule which the organism does not
represent. And think of more complicated insect behavior like the bee’s dance.
Whilst there may be some plausibility to the idea that the bee represents its food
source, there is little to the idea that it represents whatever rules may govern its
dancing (2.2).
(iv) I take the next example from Zenon Pylyshyn’s discussion of ‘‘early vision’’
(drawing on Marr 1982). The visual system accurately recovers a 3-D layout
from a 2-D image. How? It ‘‘behaves as though it were making certain assump-
tions about the nature of the physical world’’; for example,
that most of an image consists of light reXected from surfaces, that the distance of
the surfaces from the perceiver varies gradually in most of the image, that certain kinds of
discontinuous visual features in the image usually arise from physical discontinuities
on the surface of smooth rigid 3-D objects, that contour discontinuities usually arise
from occluding edges, that the light illuminating the object usually comes from above,
and so on.
Yet, ‘‘nobody actually believes that the visual system uses an explicit representa-
tion of these constraints’’ (1991: 237). In Pylyshyn’s terminology, the explanation
of what is going on here appeals to ‘‘intrinsic functional properties’’ not to ‘‘the
content of representations’’ (p. 241), to a ‘‘cognitive capacity’’ not to ‘‘represen-
tation-governed regularities’’ (p. 242).
(v) Consider a vast range of human skills. We know how to swim, ride a
bicycle, touch type, and so on. These activities are governed by rules that have
been built into us largely by learning. Even if representations of some of these
rules can play a role in the acquisition and improvement of the skills, it is not
plausible to suppose that those rules must be represented in a person once she has
achieved competence, nor to suppose that many of them are, as a matter of fact,
usually represented in her. Consider, for example, the skill of catching a Xy ball. An
essential part of this skill is being at the right place when the ball descends to catch
height. An experiment showed that skilled Welders ‘‘ran at a speed that kept the
acceleration of the tangent of the angle of elevation of gaze to the ball at 0’’
(McLeod and Dienes 1996: 531). It is not known how they manage this (p. 542)
but they surely don’t manage it by representing the algorithm for being in the right
place and applying it to a series of representations of the acceleration of the tangent
of the angle. As Pylyshyn says, ‘‘clearly, cognitive processes are relevant to motor
skill. Equally clearly, however, certain aspects of their execution are purely bio-
logical and physical’’ (1980a: 119). This is even true of some cognitive skills.
Ralph may know how to add up by R because he knows that R is an algorithm for
addition and can apply it. If so then this piece of knowing-how is largely knowing-
that. But Ralph is quite likely to be simply governed by R without representing it
to himself: R is psychologically real in him without being represented. Finally,
consider the most cognitive skill of all, the skill of thinking, of inferring one
thought from another. We can, of course, learn ‘‘laws of thought’’—that is what
we do in logic classes—but most of us know hardly anything about these rules.
Most thinking in most people surely does not involve applying representations of
these rules, representations involved in higher-level thoughts about the rule-
governed thoughts; and all thinking could not, as Lewis Carroll showed. I shall
discuss motor and cognitive skills at some length later (11.5).
(vi) Finally, consider the contrast between a universal Turing Machine using a
program for addition and a special-purpose Turing Machine built for addition.
Or, getting more real, consider the contrast between a general-purpose computer
loaded up with a word-processing program and a dedicated word-processor.
The computer is governed by the rules of that program because the rules are
represented in its RAM and it operates by reading and applying them. In contrast,
the rules that govern the operations of the dedicated word-processor—perhaps the
very same rules—are largely hardwired without being encoded. Similarly, the rules
for addition are represented on the universal Turing Machine’s tape but hardwired
into the special-purpose machine. The universal Turing Machine and the com-
puter are analogous to a human who knows that certain rules form an algorithm
for addition and applies them to add. The special-purpose Turing Machine and
the dedicated word processor are analogous to a human who knows how to add by
those rules without having any propositional knowledge of them.
This example points to an important generalization. Any processing rule that
governs the behavior of one object by being represented and applied could
govern that of another by being embodied without being represented as Edward
Stabler points out:
Any program that can be computed by a program-using system can be computed
‘‘directly’’ by a system that we would clearly not want to call a program-using system. . . .
networks of electronic circuits can be hardwired to compute even very complex programs
directly, without having control states to govern their operation according to an encoding
of a program. (1983: 392)
If rules are rules for processing information, then either way of embodying them
in an object enables the object’s behavior to be explained in terms of the Xow of
information.9 So whenever such an explanation applies there is a further empir-
ical question about the way of embodiment.
In contrasting ‘‘connectionist’’ with ‘‘classical’’ mental architecture it is some-
times assumed that the latter is committed to the explicit representation of rules
(see, for example, McClelland 1999). But this is a mistake, as Terence Horgan
and John Tienson point out:
although programs are explicitly represented as stored ‘‘data structures’’ in the ubiquitous
general-purpose computer, stored programs are not an essential feature of the classical
point of view. In some computational devices—including, for example, many hand-held
calculators—the rules are all hardwired into the system and are not explicitly represented.
According to classicism, cognition must conform to representation-processing rules that
constitute a computer program; but a cognitive system could conform to such rules
simply by being hardwired to do so. (1999: 725; see also J. A. Fodor and Pylyshyn
1988: 142–3)
In cognitive science, Pylyshyn has rightly urged, ‘‘one must attribute as much
as possible to the capacity of the system . . . to properties of the functional
architecture . . . one must Wnd the least powerful functional architecture compat-
ible with the range of variation observed’’ (1991: 244). I think that we should
draw a moral from this that I shall call ‘‘Pylyshyn’s Razor’’: ‘‘Representations are
not to be multiplied beyond necessity’’. Sometimes we already know enough about
how a system works to be conWdent that the rules that govern certain behaviors
are represented in the system; thus we know that many rules are represented in
the general-purpose computer. But we are mostly not in this position; we are not,
for example, with the rules that govern the diving kingWsher, the dancing bee,
and various human skills. Pylyshyn’s Razor demands that we posit representa-
tions of these rules in the systems only if the representations do explanatory work. I
have, in eVect, been applying the Razor in suggesting that these rules are not
plausibly seen as represented. In general, the Razor places heavy evidential
demands on any view that a processing rule is represented in a system. It will
play a large role in the rest of this book.
If we could look into the brain and simply ‘‘see’’ if there were representations
of this and that, as we can look in a book and see if there are representations of
9 Fodor seems to overlook this in his defense of information Xow psychology: ‘‘What distin-
guishes what organisms do from what the planets do is that a representation of the rules they follow
constitutes one of the causal determinants of their behavior’’ (1975: 74n). What distinguishes what
organisms do may be that rules operating on representations of their environment constitutes one of
the causal determinants of their behavior. Whether they represent the rules by which they process
that environmental information is surely another matter and not essential to what organisms do.
Peter Ustinov, then that would of course settle the matter. But we cannot: we
don’t even know enough about what to look for. As Jerry Fodor says, ‘‘there isn’t
one, not one, instance where it’s known what pattern of neural connectivity
realizes a certain cognitive content’’ (1998b: 145). So we should only posit
such representations if we can Wnd some serious causal work that they have to
do. Sadly, it is often diYcult to tell when we have found causal work for
representations: the issue is simply not that clear. Consider, for example, these
cautious words from Pylyshyn:
whenever behavior is suYciently plastic and stimulus-independent, we can at least assume
that it is somehow mediated by internal functional states. Such states may be further
viewed as representational, or epistemic, if certain other empirical conditions hold.
For example, we would describe the behavior as being governed by representations and
rules if the relation between environmental events and subsequent behavior, or the
relations among function states themselves could be shown to be, among other things,
a) arbitrary with respect to natural laws, b) informationally plastic, or c) functionally
transparent. (1980a: 120)
The rationale for Pylyshyn’s Razor is not an a priori assumption that the world
is mostly representation-free. The rationale is primarily epistemological. If we fail
to posit representations where there are some, we are likely to come across
evidence that there are some: our explanations are likely to be inadequate. In
contrast, if we posit representations where there are none, it may be diYcult to
come across evidence that there are none, because with enough representations
almost any behavior can be explained.
In sum, the behavior of a machine, an animal or a human can be governed by a
hardwired rule that it does not represent (encode), or know about.10 So, even if
structure rules of the language are processing rules, they may not be represented
(encoded), or known about in the way required by RT. Pylyshyn’s Razor puts an
onus on a theory that posits represented rules: should we have good reasons for
supposing that a system is governed by a rule, we need further reasons for supposing
that the rule governs by being represented and applied.11 The general-purpose
computer is a misleading model for the mind because it leads immediately to
the idea that the mind operates by applying represented rules. Yet the truth may
well be that the mind is more like a set of special-purpose computers.
10 Clearly more needs to be said than I have about what exactly it is for a system to be governed
by a rule that the system does not represent. On this see Davies 1995.
11 Fodor’s delightful 1968 paper, ‘‘The Appeal to Tacit Knowledge in Psychological Explan-
ations’’ (1981a: 63–78), fails to accept this onus. Fodor argues for the intellectualist view that we tie
our shoes by representing the rules for so doing but he gives no reason for supposing that the rules
governing shoe tying are represented rather than simply embodied. He seems to presuppose that
only very simple rules are hardwired (pp. 66–9). But there is no reason a priori to suppose that
complicated rules are not simply embodied in an object. The most plausible thing to say about shoe
tying and other human skills—see (v) above—is that underlying them are complicated but
unrepresented rules of some sort, often acquired by arduous practice.
3.2 S T RU C T U R E RU L E S U S E D A S D ATA I N P RO C E S S I N G
This discussion is all about processing rules. We have left open the general
possibility that a structure rule might also be a processing rule. If one is then,
like any other processing rule, it can either be represented or simply embodied.
Either way the structure rule would be embodied, internally real. Edward
Stabler (1983) has drawn attention to another way in which rules might be
internally real. They might be represented in a system and used by processing
rules as data.
5. Distinguish the representations of structure rules used as data in processing from the
representations of structure rules that are applied in processing.
So RT might hold not because the represented rules are processing rules but
because they are data for processing.
Suppose that we have good reason to think that a system encodes a represen-
tation of structure rules. It may, nonetheless, seem rather implausible that these
rules are processing rules. The problem is that structure rules often seem to be the
wrong sort of rules to be processing rules: they do not seem to be the sort of rules
that would govern a process that the system actually goes through. Consider the
chess player. It is plausible to think that the player represents the rules of chess,
for example, that the bishop can only move diagonally. Yet it seems somewhat
unlikely that the player would actually go through a process of inferring ‘x moves
diagonally’ from ‘x is a bishop’. This does not seem to be the right sort of rule to
be among the strategies and heuristics that the chess player tries out in deciding a
move. However, with the help of distinction 5 we can see another role for the
represented rule. The strategies and heuristics captured by the processing rules
may require ‘‘consulting’’ the chess rules stored in memory. In the process of
assessing possible chess moves, these structure rules are included in the data used
by the processing rules, with the result that only proper moves are contemplated.
In this way the encoded structure rules constrain processes without themselves
being processing rules. This story may well be true of some actual chess player. In
any case it seems more plausible that an encoded structure rule should thus
constrain processes rather than specify them.
3.3 R AT I O N A L – C AU S A L V S . B RU T E – C AU S A L P RO C E S S E S
We have paid a deal of attention to whether processing rules are represented. But
what about whether processing rules operate on representations? The distinction
we need here is hard to characterize precisely but it also is not controversial. And
it also does not have the prominence it should have in the debate over psycho-
logical reality.
6. Distinguish processing rules that govern a ‘‘rational-causal’’ operation on syntactically
structured representations (or symbols) from ones that govern ‘‘brute-causal’’ operations
that may or may not involve representations.12
What do we have in mind as a ‘‘rational-causal’’ operation on representations?
We have in mind a process that manipulates representations in virtue of their
syntactic structures. The rules that govern the process are sensitive to those
structures. The causal link between a stimulus input and a behavioral output is
mediated by representations of each and of stored ‘‘background knowledge’’. The
process starts with some symbols and yields others in ‘‘a Xow of information’’.
The process is, or is analogous to, one of thinking about the input and possible
outputs. The role of background knowledge in this process leads to Xexibility in
the relation between inputs and outputs: the one input can yield diVerent
outputs and diVerent inputs can yield the one output. Pylyshyn illustrates the
point nicely:
seeing that the building you are in is on Wre, smelling smoke coming in through the
ventilation duct, or being told by telephone that the building is on Wre, can all lead to
similar behavior, and this behavior might be radically diVerent if you believed yourself to
be performing in a play at the time. (1980a: 121)
Paradigm examples of rational–causal operations are yielded by the ‘‘classical’’
von Neumann architecture of the standard general-purpose digital computer.
And, according to the ‘‘classical’’ view of the mind, the practical decision making
that takes place in a situation like the one Pylyshyn describes is also a paradigm.
The standard computer operates on symbols because it represents and applies
processing rules that govern such operations. Still, an object’s behavior may be
governed by a rule that operates on symbols even though that rule is not itself
represented. A special-purpose Turing Machine is an example. Our discussion
(3.1) suggests some others: simple electronic calculators that operate on repre-
sentations about numbers; dedicated word-processors that operate on represen-
tations about expressions; and, most interesting of all, human thinkers who
operate on complex mental representations that might be about anything at all.13
A rule for language use might govern a rational–causal process by operating on
metalinguistic representation of the syntactic and semantic properties of linguistic
items; thus it might operate on the likes of ‘This input is a VP’ and ‘That input
means CLINTON ’.
12 I take the terms ‘‘rational–causal’’ and ‘‘brute–causal’’ from Fodor (1981a: 273) but my usage
is somewhat diVerent.
13 Would the operations on cognitive maps that are alleged to explain the navigational behavior
of animals (Gallistel 1990) count as rational–causal? I think not but I confess to being far from
conWdent about this. These operations are sensitive to the ‘‘syntax’’ of the cognitive map but do not
seem to involve any manipulation of symbols. Rather, given an appropriate stimulus—for example,
an ant obtaining food—the map leads ‘‘automatically’’ to a certain behavior—heading straight back
to the nest.
What is the contrast? What operations do not manipulate structured repre-

sentations? Here we have in mind operations that are arational and more ‘‘brute–
causal’’ and ‘‘primitive’’. In such an operation, the processing rule is directly
responsive to the input/output: the input directly triggers a fairly brute–causal
processing operation and the output is the direct result of that brute–causal
operation. There are a variety of examples of such rules.
(i) Associationism provides paradigm examples. The elements of classical
associationism are mental, experiences and ideas. Learning is a matter of pairing
these according to certain laws. Mental processes are simply a matter of moving
from an experience to an associated idea, or from an idea to an associated idea.
Association is a primitive relation in the theory. Behaviorism is nonmentalistic
associationism: the mental elements are replaced by physical ones, environmental
stimuli and behavioral responses. Mental processes are replaced by processes
from stimuli to associated responses, processes that are like reXexes in their brutal
simplicity. Daniel Dennett Wnds what seems to be a nice example in the ‘‘simple,
rigid and mechanical’’ behavior of the wasp Sphex as it drags a previously
paralyzed cricket to its burrow (1978: 65–6). The wasp leaves the cricket on
the threshold of the burrow whilst ‘‘checking’’ inside. If the cricket is moved back
from the threshold whilst the wasp is inside then, on emerging, the wasp will drag
it back to the threshold and repeat the ‘‘checking’’ procedure. If the cricket is
moved again, then the wasp will check again. And so on indeWnitely.
(ii) More complicated examples are posited by the moderate Hull–Tolman
behaviorism: sensory stimuli and behavioral responses can be linked in this direct
way not only to each other but also to internal states, perhaps representational
ones. These are examples because the rules governing operations are not sensitive
to the syntactic properties of representations.
(iii) Moderate behaviorism is the Wrst step toward positing a complex func-
tional organization within the object that intervenes between the stimulus and
the response. This organization, however complicated, might be governed by
rules of the sort in question because its causal operations might not manipulate
structured representations. Perhaps the rules governing addition in an old-fash-
ioned mechanical calculator are of this sort. Perhaps most of the rules governing
motor skills like those of the diving heron, the bee dancing, and human bicycle
riding, swimming, catching, and touch-typing are too (11.5).
(iv) Wherever a system is governed by rules for processing representations
there have to be transducers that take physical inputs and transform them into
representations for processing and take representations that are the result of
processing and turn them into behaviour. The rules governing transducers are
further examples of processing rules that do not manipulate structured represen-
tations.
(v) Finally, the rules governing connectionist networks are notoriously of this
sort for connectionism is a complex form of associationism: ‘‘associative learning
is what neural networks implement’’ (Macintosh 1999: 183); ‘‘simple associa-
tionism is a particularly impoverished and impotent corner of the connectionist
universe’’ (Smolensky: 1991: 165).
We note that representations may be involved in a brute–causal process. Thus
the processes of Hull–Tolman behaviorism and of connectionism may involve
representations14 as surely do those of many skills. But these processes are not
governed by rules that manipulate representations according to their syntactic
structure in the rational–causal way.
If a rule for language use governs a brute–causal process then it may respond to
and produce items that have properties like being a VP and meaning CLINTON
without there being any complex metalinguistic representations of those items as
having those properties. This stands in contrast to our earlier example of a rule
for a rational–causal process that operates on these metalinguistic representa-
tions. Pylyshyn’s Razor (3.1) requires us to Wnd an explanatory need for suppos-
ing that language use is rational–causal before positing the metalinguistic
representations. (Note that, unusually, the inputs/outputs of language use are
representations of a certain sort. Still they might be processed without being
represented as representations of that sort.)
I appreciate that this characterization leaves distinction 6 somewhat vague,15
particularly on the brute–causal side. But this cannot be avoided without an
overcommitment to a particular view of fairly brute–causal processes.
3.4 SOME POSSIBLE POSITIONS ON

P S YC H O LO G I C A L R E A L I T Y
We can apply these six distinctions to yield a range of possible positions on the
psychological reality of the structure rules of language. These are by no means the
only possible positions. Indeed, in the next section, we will discuss actual
positions that ‘‘back away from’’ these positions in interesting ways. Still, the
positions are clear and straightforward and serve as useful benchmarks.
For convenience, our focus so far has been on language production, mostly
ignoring language comprehension. We must now take both types of language use
into account.
14 William Ramsey (1997) argues that connectionist models do not really involve representa-
tions.
15 Note that we cannot characterize it as the distinction between computational and noncom-
putational processes because, although computational processes are frequently taken to be those of a
‘‘classical’’ architecture, there are a variety of views of computation: ‘‘It is surprisingly diYcult to Wnd
an answer within the cognitive science community to the question of whether there is a univocal
notion of computation that underlies the various diVerent computational approaches to cognition
on oVer.’’ (Wilson 1999: xxix; see also Brian Smith 1999)
We have already seen (2.4) that distinction 3 yields a minimal position that
should be uncontroversial:
(m) A competence in a language, and the processing rules that govern its exercise, respect
the structure rules of the language: the processing rules of language comprehension take
sentences of the language as inputs; the processing rules of language production yield
sentences of the language as outputs.
The processing rules are of course psychologically real but position (m) does not
require that those rules involve the structure rules that the processing rules respect
nor that the structure rules are psychologically real in any other way. Perhaps (m)
entitles us to say that the speaker behaves as if her behavior were governed by
those structure rules but not that it is so governed. (m) meets the Respect
Constraint, as any theory of competence must (2.4-2.6), but it does nothing
more.
Now, of course, there must be more to be said about the psychology: there
must be some true account of a speaker’s competence that explains how it respects
the structure rules of the language. And, in light of the claims of Chomskian
linguists, we are particularly interested in positions that say more by giving those
very rules a role in the psychological explanation.
I have argued that, in the Wrst instance at least, the grammar describes
linguistic reality, the structure rules of the language. That was the Wrst methodo-
logical point (2.6). The second methodological point was that any view that the
grammar also describes psychological reality, describes linguistic competence and
its processing rules, needs a powerful psychological assumption. The third point
was that there should be no a priori demand that our theory of psychological
reality be tied any more closely to the grammar than the uncontroversial minimal
position (m). There is no need to constrain that theory in any other way.
We shall now describe Wve possible psychological positions on language use,
some positions with two versions. The Wrst three positions do take the grammar to
throw more light on this psychological reality, in particular to be true of that
reality as well as linguistic reality. And so they each involve a powerful psycho-
logical assumption. Each position is a robust view of the psychological reality of a
speaker’s language, placing its rules one way or another in the mind. The Wnal two
positions explain language use without placing the linguistic rules in the mind.
Distinction 5 yields two possible positions. In both of these, the language is
psychologically real in that the grammar itself is: so they entail RT (the Repre-
sentational Thesis):
(i) The structure rules of the language are also processing rules that are represented in the
speaker and applied in language use.
(ii) The structure rules of the language are represented and used as data by the processing
rules of language use.
Stabler characterizes position (ii) as follows (G being the rules of the language):
the language-processing system has a representation of G in memory to which certain
computational processes are sensitive; parts of G are taken as arguments to functions that
are computed.16 (1983: 400)
(i) identiWes the represented processing rules with the rules of the language but
we shall count a position as (i) if it takes the processing rules to be isomorphic to
or to ‘‘mirror’’ the linguistic rules.
(i) faces an immediate and obvious problem: language use ‘‘runs in two
directions’’, producing sentences in speech and parsing them in comprehension.
How could the one set of structure rules be processing rules for these two
opposite processes? To avoid this problem let us count a structure rule as a
processing rule even if it is ‘‘run backwards’’ in processing. ((ii) does not face the
problem because the representation of the rules can be consulted by processors
running in either direction.)
We could distinguish two versions of each of (i) and (ii). For a reason already
noted (3.1), it cannot be the case that all of a system’s processing rules are
represented. Wherever a processing rule is represented and applied, there must
be underlying processing rules that govern applications without being repre-
sented: there must be some hardwiring. Still, in the case of (i), it might be the
case that some of the processing rules that are not structure rules are represented
or that none of them are. And in the case of (ii), more simply, it might be the case
that some of the processing rules are represented or that none of them are. I shall
not bother to distinguish the two versions.
In considering the chess player earlier (3.2), we noted the unlikelihood of a
represented structure rule being a processing rule and hence found the position
analogous to (ii) more plausible than the one analogous to (i). When we consider
the logic machine and the bee (2.2) the analogues of both positions are surely
false because there is no serious question of the machine or the bee representing
structure rules.
Attention to distinction 4 yields a position which abandons RT:
(iii) The structure rules of the language are also processing rules for language use but the
processing rules are not represented.17
16 One of Stabler’s ‘‘peer commentators’’ points out that we think of a conventional computer as
running a LISP program even though the machine uses the program as data to simulate a LISP
machine (Davis 1983: 404). In such a case, the distinction between doing this and running the
program directly loses some of its interest. Still, the distinction remains. Our discussion of the chess
player and the language user suggests that in other cases where the data used may not be represen-
tations of processing rules—hence not a program—but representations of structure rules, the distinc-
tion is very signiWcant. (Some other commentators make similar points to Davis: Berwick 1983, Gross
1983, Lipton 1983, and Thompson 1983.) Ray JackendoV describes three possible positions on the
relation of grammars to processing. The second is: ‘‘the rules of the mental grammar are explicitly
stored in memory and . . . the language processor ‘consults’ them or ‘invokes’ them’’ (1997: 7).
Consulting the rules seems to amount to position (ii). Perhaps invoking them is position (i).
17 (iii) is JackendoV’s third possible position: ‘‘the rules of mental grammar are ‘embodied’ by
the processor’’ (1997: 8).
So, the language is psychologically real even though the grammar is not. We shall
count a position as (iii) if it takes the embodied processing rules to be iso-
morphic to or to ‘‘mirror’’ the linguistic rules. Such a position faces the same
immediate and obvious problem as (i) and we avoid it in the same way: count a
structure rule as a processing rule even if it is ‘‘run backwards’’ in processing.
In considering positions (i) and (ii), we did not have to attend to distinction 6
because if the rules of the language are represented and applied or used as data, it
goes without saying that processing must operate on metalinguistic representa-
tions of the syntactic and semantic properties of linguistic items. However,
distinction 6 is very relevant to position (iii), yielding two versions. According
to version (a), the unrepresented structure rules process metalinguistic represen-
tations of the syntactic and semantic properties of linguistic items, representations
like ‘This is a VP’ and ‘This means CLINTON ’, and there is a rational
information Xow. According to version (b), the rules are directly responsive to
items that have properties like being a VP and meaning CLINTON. There are no
metalinguistic representations and the process is of a fairly brute–causal associ-
ationist sort.
The analogue of (iii) is certainly false for the logic machine because the
structure rules of its language are the wrong sort to be its processing rules.
I have suggested that the analogue for the bee is very likely false for the same
reason (2.2). The analogue for the chess player is somewhat unlikely but certainly
not inconceivable. Perhaps for some chess players, the structure rule that the
bishop can only move diagonally is a processing rule: the players actually go
through a process of inferring ‘x moves diagonally’ from ‘x is a bishop’.
(i) to (iii) place the rules of the language in the mind of a speaker, represented
in (i) and (ii), unrepresented in (iii). So they are ways to capture what seems to
be the received view of Chomskian linguistics: that these rules are embodied
somehow in the mind. The following positions on language depart from that
view. If they are correct, the grammar throws no more light on language
processing than is captured by (m).
(iv) Some processing rules for language use are represented but they are largely unlike the
structure rules of the language and do not use the structure rules as data.
(For the reason given in discussing (i) and (ii), not all processing rules could be
represented.) Once again we do not need to apply distinction 6: since the
processing rules are represented they must operate on metalinguistic representa-
tions of the linguistic properties.
Analogues of (iv) for the bee and the logic machine are no more plausible than
were the analogues of (i) and (ii). However the analogue of (iv) for the chess
player may have some plausibility.
We move now to a position on language use that not only gives no place to the
structure rules of the language but also no place to represented rules.
(v) The processing rules for language use are unrepresented and largely unlike the
structure rules of the language.
Distinction 6 is relevant here, yielding version (a) that is committed to some of
the rules operating on metalinguistic representations and version (b) that is not.
(For a reason already noted (3.3), it cannot be the case that all of a system’s
processing rules operate on metalinguistic representations. For, if some did,
others would have to supply representations of the input and take the results of
operations on representations to produce a behavioral output.)
An analogue of (v)(a) is very likely true of the logic machine: its transform-
ation rules are unrepresented, are unlike its formation rules, and operate on wVs
to yield an information Xow. An analogue of (v)(b) may be true of the bee’s dance
and other insect behavior.
We have described a range of possible positions on language use, starting with
the highly intellectualist (i) and (ii) and ending with the fairly nonintellectualist
(v)(b).
There is one further position on the psychological reality of the language that I
shall emphasize:
(t) The structure rules of a speaker’s language are similar to the structure rules of her
thought.
This position arises from the controversial ‘‘Language-of-Thought Hypothesis’’
(‘‘LOTH’’) according to which thoughts involve language-like mental represen-
tations. (t) is not a position on language use but on the relation of language to
thought. It diVers strikingly from (i) to (iii) in claiming that the structure rules
get their psychological reality not from being rules used in processing but from
being similar to the structure rules governing mental representations. Analo-
gously, the formation rules for the wVs of a formal language get their ‘‘internal
reality’’ in a logic machine not from being processing rules but from being the
structure rules governing the wVs that are processed by the machine.
Position (t) encourages, although it does not require, the view that the
structure rules are not psychologically real in any other respect than being similar
to the structure rules of thought. And it has lots of other interesting conse-
quences. I shall explore this position in Part IV.
This chapter began with a series of distinctions. First, there was the distinction
between processing rules that govern by being represented and applied and those
that are simply embodied without being represented. Discussion of this led to the
important ‘‘Pylyshyn’s Razor’’: representations are not to be multiplied beyond
necessity (3.1). Next, there was the distinction between representations of
structure rules being used as data in processing and rules being applied in
processing (3.2). Finally, there was the distinction between processing rules
that govern ‘‘rational–causal’’ operations on syntactically structured representa-
tions from ones that govern ‘‘brute–causal’’ operations (3.3). Against the back-
ground of these distinctions, I have described some alternative possible positions
on psychological reality. These vary according to whether or not the rules of the
language are embodied in the mind; whether or not some processing rules for
language are represented in the mind (cf. RT); whether or not some processing
rules operate on metalinguistic representations of syntactic and semantic prop-
erties of linguistic items. And there is an uncontroversial minimal position, (m),
that is committed only to there being a psychological reality that respects the
linguistic structure rules (3.4). Given the Part I conclusion that linguistics is not
psychology, no position stronger than the minimal one can be sustained without
some powerful psychological assumption that is independent of anything
revealed by the grammar of a language. Such an assumption is needed to
conclude that the rules of the language are present one way or another in the
mind. And, we shall see, it is hard to justify such an assumption.
In the next chapter, we shall consider some historically interesting actual
positions on the psychological reality issue and relate them to these possible
positions. This will demonstrate that later criticisms of positions (i) to (iv) are
not criticisms of straw men. In describing an actual position I shall often advert
to the earlier methodological points (2.6). We shall see that most of these writers
seem to subscribe to RT. That is a powerful psychological assumption of the sort
that, according to my second methodological point, is needed in taking the
grammar to be true of psychological reality. This thesis is so taken for granted
that these writers do not seem to contemplate the idea that the grammar might
simply be true of linguistic reality, as urged in my Wrst methodological point.
Finally, we will see that most writers take the constraints that the grammar and
the theory of competence place on each other to be much greater than the
Respect Constraint of my third methodological point.
The distinctions and possible positions mentioned in this chapter, along with
Pylyshyn’s Razor and LOTH, are listed in the Glossary.
4
Some Actual Positions on Psychological Reality
4.1 C H O M S K Y A N D P O S I T I O N S ( I) A N D ( III)
What is Chomsky’s view of the psychological reality of language? This question is

surprisingly hard to answer. Chomsky is insistent that language is psychologically
real but is very noncommittal about how it is. And what he does say suggests
several diVerent positions. In this section and the next I shall present cases for
attributing to him some of the positions described in section 3.4; in this section,
for attributing position (i) or (iii); in the next section, (ii). According to both (i)
and (iii), the structure rules of the language are also rules for language use. So, in
language use a speaker actually goes through a process governed by those rules.
According to (i) this is because the rules are represented in the speaker and
applied in processing; according to (iii) the rules are simply embodied. Accord-
ing to (ii), the structure rules are represented and used as data by processing rules
but are not themselves processing rules. Clearly both (i) and (ii) are committed
to the Representational Thesis (RT).
The evidence that (i) or (iii) is the right interpretation of Chomsky can be put
simply. On the one hand, a language is said to be psychologically real in its
speakers. On the other hand, this language is described as a process. Thus, there is
much talk of ‘‘generation’’, ‘‘computation’’, ‘‘derivation’’, and ‘‘transformation’’,
and there are diagrams with arrows suggesting processes. Consider the following,
for example:
The I-language consists of a computational procedure and a lexicon. The lexicon is a
collection of items, each a complex of properties (called ‘‘features’’). . . . The computa-
tional procedure selects items from the lexicon and forms an expression, a more complex
array of features. (Chomsky 1995b: 15; 2000a: 120)
And these procedures are persistantly presented as psychological (e.g., 1993a:
36). So, the structure rules of the I-language are the rules of these procedures:
I know of no proposed explanation for the fact that our judgements and behavior accord
with certain rule systems other than the assumption that computation involving such
rules and the representations they provide takes place in the mind . . . (Chomsky
1980a: 130)
Some Actual Positions 63
So those rule systems govern processes that really go on in the mind. Indeed, if
language really is a computational process, where else could this process plausibly
be but in the mind?
Consider also the following two pieces of evidence. (i) In an early work (Miller
and Chomsky 1963), Chomsky proposed a very direct relation between the
grammatical rules, in particular the transformational rules, and the rules of the
parser. However the rules could not be the same because grammatical derivations
‘‘run in the wrong direction’’. So the proposal was that the parser runs the inverse
of those rules; it undoes the transformations; it runs the rules backwards. (We
counted this as running the rules in section 3.4.) (ii) In a later work, Chomsky
discusses wh-movement (1980a: 192–202;1980b: 4), a classic transformation
from deep structure (later D-structure) to surface structure (later S-structure)
that forms questions and relatives. He argues for the psychological reality of this
transformation: there is a mental process of moving expressions like ‘who’ and
‘what’ to the left of clauses; ‘‘We attribute psychological reality to the postulated
representations and mental computations’’ (1980a: 197).1
If we take literally such claims about the mental reality of grammatical
computations then we must see Chomsky as demanding a much closer relation
between the grammar and the theory of competence than is appropriate accord-
ing to my third methodological point (2.6). For, according to that point, we
should demand only that the grammar and the theory meet the Respect Con-
straint: the grammar should posit structure rules that are respected by the
competence; and the theory of competence should posit processing rules that
respect the structure rules. And it seems that we must adopt (i) or (iii) as the
interpretation of Chomsky. In the next section, I shall consider the possibility of
not taking the claims literally. Meanwhile, what about the choice between
positions (i) and (iii)? According to (i), the speaker performs an I-language’s
computations as a result of applying representations of the language’s rules,
applying the grammar. I began this work by claiming that the natural interpret-
ation of Chomsky, the one that takes him pretty much at his word, ascribes RT to
him. If he does hold this thesis, then position (i) not (iii) seems to be the right
interpretation. But many think that Chomsky does not hold RT, despite his
words suggesting he does (e.g. Matthews 1991: 192; Collins 2004).2 If not, (iii)
should seem to be the right interpretation. According to (iii), the speaker
performs an I-language’s computations because its rules are embodied without
being represented.
Chomsky seems to acknowledge the distinction between (i) and (iii)3 but he
gives it strangely little attention. An example of his acknowledging the distinction
1 Consider also his comparison of the computations of a language ‘‘to the complex calculations of
the mind/brain that inform me that I am seeing a group of people sitting in a lecture hall’’ (1988: 91).
2 On this matter, I direct the reader once again to the exchange: Rey 2003a; Chomsky 2003; Rey
2003b.
3 Although according to Pylyshyn (1991: 233) he does not.
is to be found in his response to Elliott Sober’s peer commentary (1980) on
Chomsky 1980b. Chomsky supposes that R is a rule and that ‘‘H(R) attributes R
to the mind/brain and asserts that computations eventuating in behavior use R’’.
He then criticizes Sober for equivocating between two distinct psychological
realities: ‘‘psychological reality of a theory and its hypotheses, such as H(R), and
psychological reality of entities such as R attributed by the theory to the mind/
brain’’ (1980c: 56). The distinction between these two psychological realities is in
eVect the distinction between (i) and (iii). Yet his earlier response (p. 44) to
Dennett’s skepticism about the ‘‘claim that universal grammar is innately Wxed in
the form of explicit rules’’ (1980: 19) seems to miss the distinction entirely.
Despite Chomsky’s persistant talk suggesting that linguistic derivations are
psychologically real, there are problems in taking him to believe this and hence to
hold (i) or (iii).
First, scattered through Chomsky’s writings are warnings against taking his
accounts of language as accounts of a psychological process. For example, talking
about the arrows in a diagram depicting the ‘‘levels of representation’’ of a
language, he says that ‘‘their orientation expresses structural relations and entails
nothing about temporal order of speech production or processing’’ (Chomsky
1986: 67). Earlier he remarks that
it seems absurd to suppose that the speaker Wrst forms a generalized Phrase-marker by
base rules and then tests it for well-formedness by applying transformational rules to see if
it gives, Wnally a well-formed sentence. But this absurdity is simply a corollary to the
deeper absurdity of regarding the system of generative rules as a point-by-point model for
the actual construction of a sentence by a speaker. . . . To think of a generative grammar
in these terms is to take it to be a model of performance rather than a model of
competence, thus totally misconceiving its nature. . . . it can be regarded only as a
characterization of the intrinsic tacit knowledge or competence that underlies actual
performance. (1965: 140–1; see also p. 9)
Furthermore, what Chomsky Wnds absurd here surely is so. This may not be so
obvious when we consider what the grammar of that time claims about trans-
formations but it is surely obvious when we consider what it claims about the
formation of the initial D-structure phrase-marker. This formation starts with an
‘S’, rewrites this as ‘NP þ VP’, goes through various other rewrites many of
which are optional, and Wnally terminates with lexical insertion and semantic
interpretation. It is hard to see how this could be the story of a psychological
process because it could not be part of any plausible view of the way a thought is
turned into language, the way a ‘‘message’’ is conveyed, in language production.4
The problem with the story was foreshadowed in our earlier discussion of the wV
machine and the sentence machine (2.2, 2.4). On this story of language produc-
tion, as on that earlier account of wV production and sentence production, the
4 For roughly this reason, according to JackendoV, ‘‘back in the 1960s, we were Wrmly taught not
to think of rules of grammar’’ as embodied processing rules (1997: 8).
selection from the lexicon is random apart from syntactic constraints: these
constraints may demand the selection of a noun not a verb at a certain place in
a string, but hardly the selection of ‘lion’ not ‘tiger’. (Note that meaning—‘‘the
semantic representation’’—does not become part of the story until after the
formation of the D-structure.) Similarly, the selection of one syntactically
appropriate structure over another is random; thus a required ‘NP’ might be a
simple ‘N’ (say, ‘Fred’) or it might be a complex ‘D þ Adj þ N’ (say, ‘the bald
man’). In language production, humans do not generate sentences by making
selections of structures and words that are random apart from the constraints of
syntax. Humans generate a sentence from a thought, intending to convey a
message that is the content of that thought. What selects syntactic structures
and items from the lexicon to form sentences is primarily the thoughts the
sentences express not the language used to express them. (I shall make much of
this point later in Parts IV and V.)
I remarked that treating the grammar as a model of performance in the way
Chomsky criticizes may not seem so absurd when we focus on transformations.
Still, taking transformations to be psychologically real does seem prima facie odd.
Is it psychologically plausible that on the way to producing a passive sentence a
speaker must produce an active-like D-structure which she then transforms into a
passive S-structure by NP-movement? Mightn’t her thought have been passive
from the start? And what about questions, whether wh- or yes-no, which require
transforming an indicative-like D-structure into an interrogative S-structure by
moving items? Surely a person who asks a question would be ‘‘thinking a
question’’ from the start.5 Finally, if we were right a moment ago to agree with
Chomsky that it is absurd to think that a speaker runs through the rules for
forming a D-structure in language use, we should wonder why it is appropriate
even to look for transformations being run in language use. If it is absurd that
some linguistics rules are run by the language processor, why expect others to be?
Why expect that any of these rules will govern processes in the mind?
The absurdity we have been discussing comes from taking grammatical
derivations as psychological processes of language production. But, of course, if
it were appropriate to take them as such processes then it would be appropriate to
take their inverse as psychological processes of language comprehension. So the
idea would be that, in parsing, a person ‘‘runs backwards’’ through the gram-
matical rules. Some think this is a ‘‘natural’’ Wrst guess about parsing (Fodor,
Bever, and Garrett 1974: 313; Berwick and Weinberg 1984: 39). Indeed,
psycholinguists once devoted considerable energy to investigating the psycho-
logical reality of transformations in parsing (as we shall soon see). Certainly, since
parsing always starts with a particular sound, the idea of running the process
backwards does not have the randomness problem that brought absurdity to the
5 ‘‘the input to the process of question-formation is not a declarative sentence, but rather a desire
for information’’ Schank 1980: 36).
idea of running the process forward in production. Still, running transformations

backwards in comprehension seems no more psychologically plausible than
running them forwards in production. And the idea that, in comprehen-
sion, we run backwards through the rules for forming a D-structure until
we reach ‘NP þ VP’ and Wnally ‘S’ would be as absurd as the idea that, in
production, we run forward through those rules. Intuitively, once a hearer has
extracted the D-structure plus semantic interpretation of an utterance she has
gone as far as she needs to go in language comprehension. Finally, if, in language
use, running D-structure rules forward or backward is absurd why should we
even expect transformation rules to be so run? This looks like the wrong model of
language use.
I shall be arguing, in eVect, that the moral we should draw here is broader than
Chomsky’s. Not only does the process talk not provide a plausible model of
performance but, taken literally, it does not provide a plausible model
of competence either. First, there is a prima facie problem in supposing that it
does provide such a model: competence is a state and so it is odd to think of it as
consisting in processes at all. Second, competence is exercised in performance.
And this link to performance is not accidental: the competence is a competence
to use language; that is the very nature of this psychological state. So a theory of
competence must be at the center of a theory of performance. (Chomsky’s task
(iii) in section 1.1 was ‘‘How is knowledge of language put to use?’’) It is hard to
see how the view that a speaker literally selects from structures and the lexicon in
the largely random way described could Wt with any plausible theory of per-
formance. In general, not only are the computational procedures identiWed with
an I-language not plausibly seen as a point-by-point model of sentence construc-
tion, they are not plausibly seen as such a model at any level of abstraction; for
the most part, they are not plausibly seen as psychologically real processes at all,
as processes going on in real time in the mind.
Finally, we should take account of Chomsky’s allusion to David Marr’s famous
‘‘levels’’. Chomsky takes ‘‘the study of grammar to be at the level of the theory of
the computation’’ (1980c: 48). This is the Wrst of Marr’s three levels for under-
standing an information-processing device: it speciWes what is computed and
why. The second is the algorithmic level: it speciWes how the computation is done.
The third is the level of implementation: it speciWes the way the computation is
realized physically (1982: 22–5). Marr also thinks that a grammar is at the
computational level (p. 28). For this to be so, the grammar must be a theory
of a function that the speaker actually computes. That leaves open how the
speaker computes the function—the algorithmic level—but it does entail that
the speaker goes through some psychological process or other of computing the
function speciWed by the grammar.
But if the grammar is a theory of an I-language as described in the quote from
Chomsky (1995b: 15; 2000a: 120) at the beginning of this chapter, it is hard to
see how it could be a theory at the computational level. On the strength of that
description, we might say that the grammar is a theory of a function for
generating sentences from a lexicon but, for the reasons already indicated, that
is not a plausible view of the function that a speaker computes in language use.
What a speaker computes are functions turning sounds into messages in com-
prehension and messages into sounds in production.6
Chomsky has another way of describing a language that may seem a more
promising way to accommodate the allusion to Marr: a language is said to be
‘‘a generative procedure that associates sound and meaning in a speciWc way’’
(Chomsky 1996: 48). Can we then take a grammar to be a theory of a function
for associating sounds and meanings? If so then we could indeed think of a
speaker as computing this function (for associating sounds with messages
amounts to associating them with meanings). We could see a grammar as
describing, at an abstract level, psychological processes actually going on in
speakers during language use.
The problem is that we surely cannot take the grammar to be a theory of a
function for associating sounds and meanings. We might say that the grammar is
a theory of a function for generating sentences from a lexicon, where the lexical
items all have sounds, syntactic characters, and meanings, and where each
sentence generated has its sound, syntactic character, and meaning in virtue of
the way it is generated from the lexicon. That is very diVerent from a function for
associating sounds and meanings.
From my perspective, the underlying truth of the claim that the grammar is a
theory of a function for associating sounds and meanings is that it is a theory of
the rules that are respected by this function. Perhaps we should see the allusion to
Marr as demanding no more.
Let us sum up. First, it is hard to see how the view that a language is literally a
computational process can be combined plausibly with the view that the lan-
guage is psychologically real (whether in virtue of having its rules represented or
in virtue of having its rules simply embodied).
Next, my third methodological point (2.6) was that the grammar and the
theory of competence are largely independent of each other: the only epistemic
constraint that they place on each other arises from the Respect Constraint. It
might still be the case, of course, that some structure rules posited by the grammar
are processing rules posited independently by the theory of competence. How-
ever, I raised the possibility that the structure rules are the wrong sort to be
6 Is a grammar a theory of the speaker’s competence at Christopher Peacocke’s ‘‘level 1.5’’, as he

claims (1986, 1989)? Perhaps so. A level 1.5 explanation states information that the competence
draws on. This information must be a consequence of the grammar. I have emphasized that a
grammar must meet the Respect Constraint: it must posit rules that are respected by the competence
(2.4–2.6). This seems rather like claiming that the grammar must provide level 1.5 explanations of
competence. Peacocke takes the grammar to be psychologically real simply in virtue of this claim
(1989: 114). I do not take it to be psychologically real simply in virtue of its meeting the Respect
Constraint. But this diVerence may be just verbal.
processing rules just as, in our examples, the rules of the bee’s dance very likely are,
and the formation rules of logic certainly are (2.2). The present discussion
suggests that some of the structure rules for language, at least, are indeed the
wrong sort to be processing rules. This gives a motive for attributing position (ii)
rather than (i) or (iii) to Chomsky: the speaker ‘‘consults’’ the represented rule but
need not actually go through the process speciWed by the rule.
4.2 C H O M S K Y A N D P O S I T I O N ( II)
There is an obvious problem in attributing position (ii) to Chomsky: we need to

account for all the talk suggesting that linguistic rules govern real processes in the
mind. But, perhaps we can explain away this talk by treating it as a metaphor
describing the structure of the language.7 To say that a language consists of a
‘‘computational procedure and a lexicon’’ is a metaphorical way of saying that it
consists of a system of structure rules and a lexicon. We can take talk of
transformations in a ‘‘mathematical’’ sense,8 capturing syntactically signiWcant
relations between sentences—for example, between actives and passives, and
between statements and yes–no questions—without supposing that transform-
ations are real processes in the mind.
7. Distinguish actual from merely metaphorical generation, computation, and processing.
The logic machine (2.2) provides a nice analogy. Its derivations are deWned
over the wVs of a formal language. Each wV is governed by the formation rules of
the language. These rules are the structure rules of the language used by the logic
machine. Still, we might describe these rules as a procedure for generating all the
wVs of the language. In talking in this way, we might have in mind the possibility
of a wV machine that is governed by these rules and that actually generates these
wVs by selecting syntactic forms and lexical items, selections that are random
except insofar as they are constrained by the rules. But this process is not, of
course, something that the logic machine does in using the formal language to
generate derivations. The talk of the rules of the language generating wVs is just a
metaphorical way of capturing the way the wVs have structures governed by those
rules. The generation is mathematical not actual. Analogously, perhaps with the
possibility of a sentence machine in mind, we might describe the rules of a
natural language as generating all the sentences of the language. But the claim is
not that humans are actually machines of this sort. The generation of sentences
by the rules of the language is mathematical not actual.
7 I have found that this way of interpreting the process talk appeals to many.
8 Gilbert Harman conWdently interprets Chomsky in this way: ‘‘the relevant sort of generation is
mathematical, not psychological’’ (1983: 408); see also Boden 1988: 4; Bock 1995: 207. Chomsky
himself remarks: ‘‘The term ‘generate’ is familiar in the sense intended here in logic’’ (1965: 9).
It is undoubtedly tempting to take the process talk as metaphorical, hence
making it possible to adopt position (ii) as an interpretation of Chomsky. But
there are problems giving in to this temptation. First, the process talk is presented
so persistantly as a psychological account and never acknowledged to be meta-
phorical. Pylyshyn remarks that, despite the diYculties in understanding
Chomsky’s claims about the place of linguistic rules in the mind,
none of us doubted that what was at stake in all such claims was nothing less than an
empirical hypothesis about how things really were inside the head of a human cognizer. We
knew that we were not speaking metaphorically. (1991: 233)
Second, some of Chomsky’s claims—for example, the allusion to Marr—do not
seem open to the metaphorical interpretation.
There is another way to account for the process talk and open the door to (ii)
as an interpretation. This way is prompted by a certain use/mention sloppiness in
Chomsky’s discussion.
InXuenced by passages like the one quoted earlier (Chomsky 1995b: 15; 200a:
120), we have been working with the idea that the I-language does the computing
and deriving. Yet the claim is often that the grammar, the theory of the I-language,
does so (e.g., Chomsky 1957: 26–7; 1965: 63–6). And this claim may seem more
psychologically appropriate. The idea of derivation within a theory is familiar: it’s
the idea of inferring particular claims from generalizations. But what is derived
from a grammar is not an expression of the language but a description of an
expression, just as what is derived from an astronomical theory is not, say, a star,9
but a description of a star. Clearly, the generation of expressions by the I-language
would be one thing, the generation of descriptions of expressions by the grammar,
another. Yet Chomsky seems uninterested in the diVerence. Sometimes grammars
are said to generate ‘‘structural descriptions’’ (1993a: 36), which is appropriate;
othertimes they are said to generate expressions (‘‘strings’’) (1965: 66), which,
initially at least, is not. And often I-languages are said to generate not expressions
but structural descriptions (1988: 60; 1991a: 14). Finally consider the following
where structural descriptions are identiWed with expressions:
One component of the language faculty is a generative procedure (an I-language, hence-
forth language) that generates structural descriptions (SDs), each a complex of properties,
including those commonly called ‘‘semantic’’ and ‘‘phonetic.’’ These SD’s are expressions
of the language. (1993b: 1)
In brief, Chomsky seems ready to talk of either a grammar or an I-language
generating either a structural description or an expression of the language.
The following diagram illustrates what is appropriate and inappropriate if we
are strict about the use/mention distinction:
9 Despite what neo-Kantian worldmakers tell us; see Devitt 1997a: ch. 13. McGilvray 1999 is a
neo-Kantian interpretation of Chomsky.
I-language ! expressions
Grammar ! structural descriptions

of expressions
The lines represent the process of derivation or generation. The horizontal ones
are in order according to the use/mention distinction; the diagonal ones are not.
Aside: This use/mention issue is relevant to the interpretation of linguistic
texts. For example, the sentence
(1) John hit the boy.
may stimulate something like the following:
(2) [S[NP[NJohn]][VP[V hit][NP[DETthe][Nboy]]]]
or a tree with ‘‘the same information’’ (e.g. in Chomsky 1980a and 1995c; and
see other essays in Webelhuth 1995). Since ‘John’, ‘hit’, ‘the’ and ‘boy’ seem
to be used in (2), it is natural to take (2) as simply a restatement of (1) with its
syntactic structure made fully explicit, just as ‘Cheap (food and wine) can be
interesting’ is a restatement of ‘Cheap food and wine can be interesting’ with its
structure made partially explicit. Interpreted in this way, (2) is the sort of thing
that could be derived from an I-language.
But sometimes (1) will stimulate something like the following, where the
words ‘John’, etc. are italicized:
(3) [S[NP[NJohn]][VP[V hit][NP[DETthe][Nboy]]]]
or a tree with ‘‘the same information’’ (e.g. in Chomksy 1957 and 1965). Since the
words are italicized and this is the linguistic convention for mentioning a word,
and since (3) is called ‘‘a structural description’’ (although also ‘‘a phrase-marker’’),
it is natural to take (3) as a brief version of the following description of (1):
(4) John hit the boy is a sentence made up of a noun phrase consisting of the noun John and
of a verb phrase consisting of the verb hit followed by a noun phrase made up of the
determiner the followed by the noun boy.
(4) is the sort of thing that could be derived from the grammar. (See Baker 1995,
pp. 88–92, for explicit examples of outputs like (4).) Another possibility is to
take (3) as a deWnite description—hence literally a structural description—of John
hit the boy: ‘‘The sentence made up of a noun phrase . . . ’’
Return now to the main discussion. We have seen that strict observation of the
use/mention distinction leaves two candidate stories of generation, the top and
bottom lines in the diagram. It is important to distinguish them:
8. Distinguish the generation of expressions by the I-language from the generation of
structural descriptions of expressions by the grammar.
What is the bearing of this use/mention issue on the interpretation of Chomsky?
If Chomsky required that the generation of expressions by the I-language be
psychologically real, then his position would be (i) or (iii), with the implausi-
bilities already noted. If, on the other hand, he required that the generation of
structural descriptions by the grammar be psychologically real, then his position
might be a version of (ii): the data used in language processing would be not just
the structure rules but descriptions that are derived from those rules. Still, this
interpretation has its implausibilities. It could hardly be the case that each piece
of language processing involves the derivation of all the structural descriptions for
the language. So we wonder how it would be determined which ones are derived
and how such structural descriptions could be used by the processor (on which
more in the next section).
One is left uncertain of Chomsky’s position. My best guess is that, in thinking
about language use, he starts convinced of RT and, on the strength of that alone,
convinced of the psychological reality of linguistic rules. This view of the rules
needs a powerful psychological assumption—my second methodological point
(2.6)—and RT is Chomsky’s assumption. He has no worked out opinion about, or
even much interest in, how that grammar in the head plays a role in language use.
What is puzzling about this is that a strong commitment to RT seems inappro-
priate in the absence of a well-supported theory of language use that gives RT a
central role. Even if there are other reasons for believing in RT—and I shall argue
later that there are no good ones—these should not be suYcient for commitment.
If we can Wnd no place for representations of the rules in language processing,
hence no place for them in an account of linguistic competence, we should surely
doubt the existence of these representations. We should apply Pylyshyn’s Razor
(3.1): representations are not to be multiplied beyond necessity.
Finally, this discussion provides another way to draw the earlier distinction
between a theory of a language, a grammar, and a theory of competence in that
language (2.4). The former is a theory of a metaphorical process of deriving
sentences in the language; the latter is a theory of a capacity for actual mental
processes of moving from thoughts to sentences of the language and moving
from the sentences to thoughts.
The distinctions mentioned in this section are listed in the Glossary, along
with much else.
4.3 F O D O R , B EV E R , A N D G A R R E T T
The Psychology of Language, by Jerry Fodor, Tom Bever, and Merrill Garrett
(1974), is rightly regarded as a classic. The authors describe and evaluate research
to that time into the psychological reality of language, and make some proposals.
First, Fodor, Bever, and Garrett (‘‘FBG’’) clearly take RT for granted: a speaker
represents the rules of her language and has knowledge about them (p. 7). So we
expect their position on the psychological reality of language to be a version of
either (i) or (ii). But this is not what it turns out to be.
In section 4.1 I mentioned psycholinguistic attempts to establish the psycho-
logical reality of transformations, attempts which are contrary to my third
methodological point and which are, I suggested prima facie unpromising.
FBG make an extensive examination of experiments aimed to conWrm this
psychological reality. These experiments generally appeared to assume what
FBG call ‘‘the derivational theory of complexity (DTC)’’: ‘‘the complexity of a
sentence is measured by the number of grammatical rules employed in its
derivation’’ (p. 319). So, the more transformations there are in the derivation
of a sentence the more diYcult it should be to understand. In any case, FBG Wnd
little direct evidence to conWrm the psychological reality of transformations, far
too little to convince them of it (pp. 241, 273–4). So they do not hold to (i)
(linguistic rules are represented and applied in language use).
FBG have a much more positive view about the psychological reality of
structural descriptions:
The parameters of sentences which linguistic descriptions mark enter, one way or another,
into a variety of psychological processes concerning language. There seems no serious
doubt that structural descriptions are, in this sense, psychologically real; they specify at
least some of the descriptions under which linguistic messages are interpreted and
integrated. (p. 273)
FBG Wnd ‘‘a variety of independent experimental evidence’’ supporting this view
(p. 249). This seems to me important and to some degree right. However, from
my perspective, the conclusion we should draw from these experiments needs to
be put diVerently: the experiments do not provide evidence that descriptions of
syntactic properties are psychologically real but rather that the properties them-
selves are, in some sense.
Here is their account of one experiment, which I shall take as my example:
Fodor and Bever (1965) demonstrated that the patterns of errors in ‘‘click location’’ made
by subjects when listening to sentences were related to the constituent structure of the
sentences. For example, Fodor and Bever’s stimulus material contained recordings of the
sentence (5–11) in versions of which a click was located before, after, or in the major
constituent break . . .
5–11 That he was happy was evident from the the way he smiled.
There was a signiWcantly greater error for location of clicks not objectively placed at the
major boundary than for those which objectively occurred at the boundary. There was,
moreover, a signiWcant tendency to place mislocated clicks at the major constituent
break. (p. 252)
Consider Wrst what this shows about the grammar that ascribes that constituent
structure to the language. The experiment provides evidence that, to this extent,
the grammar meets the Respect Constraint. For it provides evidence that the rules
determining the major constituents are respected by the processing rules that
govern the subjects’ use of the language. And if the experiment did not provide
this, it would be bad news for the grammar. For, we should hope to Wnd evidence
that subjects respect all the rules ascribed by the grammar and should expect to
Wnd it for rules that determine something so basic as major constituents. The
experiment shows that the subjects’ behavior is sensitive to major constituents,
thus conWrming that part of the grammar’s account of linguistic reality. We noted
earlier that psycholinguistic evidence can support a grammar (2.5, Wfth point).
This is an example.
What does the experiment show about the psychological reality of language?
At least, it provides evidence that the subjects’ linguistic competence, and its
processing rules, respect the structure rules ascribed by the grammar to determine
major constituents. We knew already, of course, that the competence must
respect the language’s structure rules—that’s the minimal position (m) on the
psychological reality of language (3.4)—but we did not know what those rules
were. The experiment provides evidence of what they are, as we have just noted.
Does it provide any more? Does it support a position on psychological reality
stronger than the minimal (m)? It supports the view that there is something
within the subjects that respects the rules governing major constituent structure,
but does it cast any light on what that something is? I claim not.
I earlier distinguished processing rules that govern rational–causal operations
on syntactically structured representations from ones that govern brute–causal
operations (3.3), leading to versions (a) and (b) of various positions on the
psychological reality issue (3.4). Now to conclude, as FBG do, that the structural
description of (5–11) is psychologically real is to side with version (a): the process
of understanding (5–11) operates on a metalinguistic representation of its
syntactic properties. But this experiment alone provides no evidence for this
over the brute–causal alternative of version (b): the process operates directly on
the syntactic properties. One might argue, of course, that the process demands a
metalinguistic representation, but that argument gets no support from this
experiment. I shall later look skeptically at such arguments (11.6–11.8).
Of course, if we assumed RT, then we would have a ready explanation of why
the competence respects the structure rules: it respects them because it represents
them. And it would then go without saying that the experiment conWrms the
psychological reality of the structural description of (5.11). For, if representations
of the syntactic rules are involved in processing then that processing must operate
on metalinguistic representations of syntactic properties. But why should we
assume RT? The experiment gives it no support—later I shall argue that it has no
signiWcant support—and so gives no support to the psychological reality of
structural descriptions.
I think that what goes for this experiment goes for all the others. They provide
evidence that the major constituent structure ascribed by the grammar, and even
some underlying D-structure, are respected by competent subjects. It is by no
means insigniWcant to have evidence of this, evidence that something-we-know-
not-what within speakers respects some pretty subtle rules of language. But the
experiments do not show more. They do not show how that respect is achieved.
In particular they alone do not support RT and do not show that structural
descriptions are psychologically real.
Although FBG endorse RT, they can often be read as expressing a view along
these lines, along the lines of the minimal position (m). Thus they say that the
experiments I have been discussing
show that linguistic structures engage psychological processes in some way, but they do
not seek to explore the character of the interactions. (p. 264)
And they assess the constraints placed on an ideal sentence recognizer by the
optimal grammar in the following terms:
Since the latter device assigns each utterance one of the structural descriptions that the
former device generates, the output of the optimal grammar constrains the output of the
ideal sentence recognizer. It should be emphasized, however, that the grammar does not,
in that sense, constrain the operations that the recognizer employs in computing its
output. It is an open question . . . what operations an ideal sentence recognizer must
run through in assigning structural descriptions to utterances. (p. 277)
This is almost a statement of what I have called ‘‘the Respect Constraint’’ on a
theory of language comprehension, together with the statement that this is the
only constraint that the linguistic rules place on that theory; cf. my third
methodological point (2.6). I say ‘‘almost’’ because here, as elsewhere, FBG are
committed to the psychological reality of structural descriptions. Even this
commitment could be explained away if we overlooked their Representational
Thesis. For their discussion, like Chomsky’s, has a use/mention sloppiness (see
e.g., pp. 18–21). So we could read their commitment as being to the psycho-
logical reality of structural properties rather than to descriptions of those of
properties. And then we could read that as a commitment to a psychological
reality that respects those properties.
Still, FBG do endorse RT. So we expect RT to feature in their theory of
language use. What is their theory?
In discussing language comprehension FBG Wrst reject one version of position
(i): ‘‘analysis by analysis’’. This is the earlier-mentioned idea that the parser runs
the grammar backwards (pp. 313–16). They then contemplate what is in eVect a
version of (ii) (linguistic rules are represented and used as data in language use)
that has attracted many. This is ‘‘analysis by synthesis’’: ‘‘the grammar is used to
generate a ‘‘search space’’ of candidate structural descriptions which are tested
one by one against the input string’’ (p. 316). But this faces ‘‘a serious diYculty’’:
the large size of the space of candidates that will need to be searched. To solve this
problem they propose heuristics that ‘‘drastically reduce the size of the space to be
searched’’ (p. 317). But, then, the more powerful the heuristic, the less there is for
the internalized grammar to do (p. 318). This raises the possibility that heuristics
do the whole parsing job leaving nothing for the grammar to do. This is, in eVect,
the possibility that FBG explore. They oVer experimental evidence of two
strategies speakers use: one a strategy of clausal analysis (pp. 329–48), and
another, of lexical analysis (pp. 348–53). They point to other clues that speakers
might use (pp. 353–61). These heuristics ‘‘employ the information that is
represented by grammatical rules’’ but not in the form that the grammar presents
that information. Indeed, they conclude: ‘‘There exist no suggestions about how
a generative grammar might be concretely employed as a sentence recognizer in a
psychologically plausible model.’’ (p. 368)
This leaves FBG in a rather strange place, as they acknowledge (p. 369). They
have not abandoned RT and yet, contrary to expectations, have found no place
for its internalized grammar in parsing: the heuristics do all the work. Yet surely
if we really do represent the rules of the language the representation must have a
central role in language use. That is the presupposition of Chomsky’s task (iii) on
our Wrst page. As he says, ‘‘the rules of the grammar enter into the processing
mechanisms’’ (1980a: 200).
Given the strangeness of their view, FBG contemplate some responses. One of
these is, in eVect, position (ii). They comment on this: ‘‘there is, at present, no
positive reason for believing it is true’’ (p. 370). I have already noted that they do
not hold (i). So what is their position in my terms? It seems to be
structure rules of the language and do not use the structure rules as data,
with a large addition. It seems to be (iv) because their explanation of how a
speaker’s competence and its processing rules respect the structure rules—the
explanation in terms of heuristics—does not have a role for the structure rules.
The large addition is, of course, RT. So the structure rules are in the mind but
they don’t parse. In that respect, at least, they are epiphenomenal. Strange indeed!
One wonders why FBG do not apply Pylyshyn’s Razor and simply abandon RT.
Our discussion of FBG has all been about language comprehension but their
line on language production does not alter these conclusions. For the line is
similar, as one would expect, but more cautious. They dismiss the version of (i)
that has the production system simply running the grammar. Their reason
(p. 390) is related to our earlier one for thinking it absurd. They consider a
version of (ii) which is ‘‘analysis by synthesis’’ again, but Wnd the same problem
with it as before: the need for heuristics to cut down the number of candidate
analyses to be tested (p. 393). They discuss the experimental studies. They
conclude that ‘‘practically anything that one can say about speech production
must be considered speculative’’ (p. 434).
The view that the linguistic rules are psychologically real requires a powerful
psychological assumption: that is my second methodological point (2.6). FBG
make such an assumption, RT. But then they can Wnd no work for RT in
explaining language use, thus undermining the assumption. Without the
assumption, their investigation makes it seem likely that, for the most part,
the rules of the language are not psychologically real: that language processing
is handled by a set of heuristic rules that respect the linguistic rules but are mostly
diVerent from them: position (iv) (or possibly (v)).
FBG’s classic made it clear that the early ideas of psycholinguistics—for
example, DTC—had to be abandoned. As Pylyshyn puts it, these ideas were
wrong about how a ‘‘grammar might be incorporated into a theory of compre-
hension/production’’ (1991: 232).
4.4 BRESNAN AND KAPLAN
Joan Bresnan and Ronald Kaplan (‘‘BK’’), like FBG, seem to take RT for granted:
A longstanding hope of research in theoretical linguistics has been that linguistic charac-
terizations of formal grammar would shed light on the speaker’s mental representation of
language. (1982: xvii)
So we also expect their position to be (i) or (ii). And this time we are not
disappointed. Their position is (i).
BK start by embracing ‘‘Chomsky’s competence hypothesis’’ (p. xvii):
a reasonable model of language use will incorporate, as a basic component, the generative
grammar that expresses the speaker-hearer’s knowledge of the language . . . (Chomsky
1965: 9)
So they seek a grammar that is not only linguistically motivated but also
psychologically motivated (pp. xx–xxii, xxxviii): it must describe real processes
in the mind. Indeed, they seem to think that a grammar that is simply linguis-
tically motivated is of no interest (pp. xxiv, xxxviii).10 This is contrary to what
I have argued in Part I and to my Wrst methodological point: a grammar might be
simply true of a linguistic reality. Their insistance that the grammar be psycho-
logically motivated is, of course, at odds with my third methodological point.
I have argued that there is no reason to suppose in advance that a grammar has
any more to do with psychological reality than comes from its meeting the
10 Although elsewhere Bresnan remarks: ‘‘Theoretical linguistics has greatly advanced our
understanding of the abstract structure of human languages.’’ (1978: 3)
Respect Constraint. Linguistic rules that are accurately described by a grammar

may be the wrong sort to govern language use.
BK think that transformational grammars have failed the test of psychological
motivation; see our discussion of FBG in the last section. They reject two
responses to this failure. The Wrst of these is the response of FBG, which is to
reject the competence hypothesis (as we have just seen). The second, is the
response of Chomsky, which is to insist on the psychological reality of the
transformational grammars even in the absence of psychological motivation,
a response that I have also, in eVect, criticized (4.1–4.2). Their own response is
to seek a diVerent sort of grammar:
a more radical decomposition of competence grammars into an expanded lexical and
contracted syntactic component promises to have far greater explanatory power than the
current versions of tranformational grammar, permitting a uniWcation of linguistic and
psycholinguistic research. (p. xx)
In particular they have in mind ‘‘lexical–function grammar’’ (‘‘LFG’’). In this
grammar the ‘‘information’’ about a sentence that was previously captured by
transformations is captured in lexical entries. Thus the relation between an active
and a passive sentence that was captured by transformations involving NP
movement is captured by a lexical rule relating active and passive verbs in the
lexicon: ‘‘apparent NP movement is only an illusion’’ (p. xxx).
What are the psychological implications of this? BK are committed to ‘‘an
information-processing model of language use that includes a processor and a
component of stored linguistic knowledge K’’. Part of K ‘‘prescribes certain
operations that the processor is to perform on linguistic representations’’. They
call this part ‘‘the representational basis’’. So, in my terminology, the representa-
tional basis represents processing rules which are applied in language use. BK
continue: ‘‘a model satisWes the strong competence hypothesis if and only if its
representational basis is isomorphic to the competence grammar’’ (p. xxxi; see
also Bresnan 1978: 3). In my terminology, the processing rules are the structure
rules: position (i).
BK think that their LFG is compatible with the strong competence hypothesis,
for it has eliminated the transformational derivations that have proved so
psychologically elusive. Even lexical rules like the one that derives the lexical
entry for the passive verb from that for the active do not have to be actually run
by the processor. The passive will have been derived from the active when the
active was acquired but after that both active and passive are stored in memory as
part of K (pp. xxxiii-xxxiv).
LFG near enough eliminates computations from phrase structure to phrase
structure. But it still does have ‘‘a single level of phrase structure representing the
surface form of a language’’. This is called ‘‘constituent structure’’ (‘‘c-structure’’)
(p. xxviii). Following Chomsky, we found it absurd that producing a sentence
involved going through the rewrite processes of building a phrase marker.
Further, we found it equally absurd that parsing a sentence involved running
these processes backward (4.1). Do BK think that the processes of building
(unbuilding) c-structures are psychologically real? The implication of Bresnan
1978 (pp. 50–2) is that they do.
BK’s approach, particularly their enthusiasm for the strong competence
hypothesis, stands in nicely clear opposition to my third methodological point.
In my view we have no reason to suppose that the strong hypothesis is true.
Furthermore, we should have no desire for it to be true: when constructing a
grammar we should not be constrained by this hypothesis; when trying to explain
the largely unknown processes of language comprehension it is a mistake to
suppose a priori that the rules governing these processes are isomorphic to the
rules speciWed by the grammar. The disagreement between a lexical-function and
a transformational grammar should be a strictly linguistic one, to be settled
according to which gives a better account of linguistic reality. And it could be the
case, of course, that they are equivalent and hence equally good.11
Bresnan (1978) distinguishes two research objectives. The Wrst is ‘‘the gram-
matical characterization problem’’ which is ‘‘to characterize the grammar that is
to represent the language user’s knowledge of language’’. This description reXects
her commitment to RT. Without that commitment, we can describe the problem
simply as that of characterizing the language. The second is ‘‘the grammatical
realization problem’’ which is ‘‘to specify the relation between the grammar and
the model of language use into which the grammar is to be incorporated as a
basic component’’ (p. 1). She Wnds too many rival solutions to the Wrst problem:
it is ‘‘not very well deWned’’. She proposes that ‘‘the grammatical realization
problem can clarify and delimit the grammatical characterization problem’’
(p. 59). Her idea is that the story of language use will help us choose a grammar.
From the perspective I have argued for, this help can go no further than the
Respect Constraint. The rules ascribed by a grammar must be respected by what
is psychologically real but we should not expect them to be psychologically real.
Indeed, it is possible that none of them are thus real. And if some of them are,
this gives no extra credit to the grammar, for the grammar’s task is to describe the
language not the psychology. Finally, the characterization problem is well enough
deWned in terms of meaning. If two grammars meet the Respect Constraint and
give equally good accounts of meaning, they may well be ‘‘notational variants’’ of
the same theory and so we do not have to choose between them.
Given their support for RT we must take BK to hold to position (i) but it is
worth noting that they do not adduce any considerations in favor of RT and
hence in favour of (i), which require the structure rules to be represented, over
(iii) which takes them to be simply embodied.
11 ‘‘The GB and LFG representations are fairly intertranslatable’’ (Pinker 1989: 34).
‘‘Nontransformational theories . . . characterize the same kinds of language facts as movement
transformations do, but by various other formal means’’ (J. D. Fodor 1989: 179). Insofar as this
is true, we have only pragmatic reasons for choosing between these theories.
4.5 B E RW I C K A N D W E I N B E RG
Robert Berwick and Amy Weinberg (‘‘BW’’) also take RT for granted: ‘‘the rules
and representations of generative grammar serve as a description of knowledge
representation’’. And they take for granted that this knowledge representation in
a speaker provides ‘‘the grammatical basis of linguistic performance’’. These are
the powerful psychological assumptions that they make to place linguistic rules in
the mind; cf. my second methodological point. They start convinced about these
psychological matters despite thinking that ‘‘very little is known about the
machinery actually governing sentence processing’’ (1984: 35). This raises the
old puzzling question: why be so convinced about RT given this ignorance about
its place in a theory of processing? Why not apply Pylyshyn’s Razor?
What do BW have to say about sentence processing? They make the undeni-
able point that the grammar of a speaker’s language must be ‘‘compatable’’ with
the theory of how the speaker parses that language (p. 35). The question then is:
what does this compatability consist in? In my view, it consists in both theories
meeting the Respect Constraint and nothing more, my third methodological
point. Given their convictions, just described, it is not surprising that BW hanker
after something much stronger. They are initially attracted by the Miller and
Chomsky (1963) suggestion that ‘‘grammars be realized more or less directly as
parsing algorithms’’. This is the earlier-mentioned idea (4.1) that the parser ‘‘runs
backwards’’ through the rules, position (i) (or (iii) if RT were dropped). BW call
this very strong principle ‘‘type transparency’’ (p. 39). However, this principle
does not seem to hold for transformational grammars (4.3). BK’s response to
this, as we have just noted, is to maintain transparency by moving from trans-
formational grammars to LFG (4.4). BW argue that this move is unfounded: ‘‘by
allowing a rudimentary kind of parallel computation we can bring a transform-
ationally based parser into line with existing psycholinguistic reaction time
results’’ (p. 56). Using a Marcus parser as their example, they point out that if
the parser computes in parallel, it may be able to parse sentences that are
transformationally complex in the same time as more simple sentences, contrary
to DTC (the derivational theory of complexity) which implicitly assumes serial
computation. So a transformational grammar can accommodate transparency:
the one-to-one correspondence between the rules of the parser and the grammar
can remain.
While this may show that we could parse like a Marcus parser, and would then
have brought the transformation rules into the mind, it does nothing to show
that we do parse that way.12 Perhaps we parse the way BK suggest, bringing LFG
12 ‘‘it seems virtually certain that neither the human adult nor the human child operates like a
Marcus parser’’ (Howe 1986: 87).
rules into the mind. Indeed, according to BW, these two views of parsing ‘‘seem
to be empirically indistinguishable’’ (p. 67).
So far, then, we have seen BW defending transparency and hence position (i).
But they are prepared to weaken their position. First, they claim that transform-
ational grammars can follow BK in allowing a ‘‘precomputed memory retrieval
system’’ (p. 74): the eVects of a transformational rule can be precomputed and
stored in the lexicon and so parsing need not involve any processing according to
that rule. Second, they appeal to ‘‘the notion of grammatical cover’’: ‘‘one
grammar is said to cover another if the Wrst grammar can be used to easily recover
all the parses that the second grammar assigns to input sentences’’. This ‘‘allows
us to hold the structural descriptions of a grammar Wxed and then consider
variations in parsing methods’’ (p. 78). BW claim that the Wrst of these weaken-
ings is compatible with the view that the transformational grammar is ‘‘a central
component of the parsing model’’ (p. 73) and the second with the view that the
parser ‘‘realize(s)’’ the grammar (p. 77).
Two comments. (i) With these claims BW seem to imply that the contem-
plated positions on psychological reality are, in my terms, still versions of (i),
albeit very weak ones. But I wonder if this is really so. I wonder if these positions
really amount to any more than the minimal (m): for example, perhaps all the
covering grammar has to do is respect the grammar covered. (ii) BW claim that
‘‘there is a continuum of more to less direct parsing ‘realizations’ of grammars as
parsers’’. This leads them to the question: ‘‘Why not simply dispense with
grammar and just look at parsing algorithms instead?’’ The answer, they think,
is ‘‘obvious’’: ‘‘we can use competence theory to constrain parsing theory’’ (p. 76).
The context of this claim is, of course, psycholinguistic. Still, it is worth noting
that the only interest in a grammar that they mention is this psychological one.
They do not mention a linguistic interest; cf. my Wrst methodological point. But
the main thing to note is that the constraint on parsing that they actually
contemplate often seems to amount to my Respect Constraint, suggesting that
they may indeed be prepared to settle for (m). Are BW, deep down, and despite
appearances and rhetoric to the contrary, in accord with my third methodological
point?
Their criticism of FBG’s heuristic stategy makes it clear that they are not. The
problem they Wnd with this strategy for passives is that ‘‘it does not crucially
depend on the assumptions of either’’ LFG or a tranformational grammar; ‘‘it
can be made compatible with either’’. In contrast, they claim that the Marcus
parser ‘‘realizes a transformational theory’’ (p. 77). This realization, it has to be
said, is extremely subtle. Although the Marcus parser does not actually
run the rules for passive transformation, the parser is alleged to realize the
transformational theory because the theory explains why the parser drops
‘‘a trace after a verb with passive morphology’’ (p. 76). So BW seem to be aiming
for a very weak version of position (i). In brief, according to BW, the view that
we parse like a Marcus parser has an advantage over the view that we parse
using FBG’s heuristic simply in virtue of being tied in this way to the transform-
ational theory. This advantage is enjoyed quite independent of any empirical
evidence about how we do in fact parse. This is contrary to my third methodo-
logical point.
Finally, I follow Stabler (1983: 401) in noting that nothing in BW’s discussion
of the relation between the grammar and parsing theory warrants commitment to
RT and so nothing motivates a very weak version of (i) rather than (iii).
4.6 M AT T H EW S
Finally, let us consider the position of Robert Matthews, a philosopher who is

close to generative grammar. His position (1991) is similar to BW’s. But it diVers
from theirs, and all the other actual positions we have considered, in a striking
way: Matthews does not embrace RT. Indeed, he can Wnd no evidence for RT
(p. 187; Demopoulos and Matthews 1983: 405). He even doubts that Chomsky
subscribes to it (p. 192).
Matthews’ position starts from the assumption that ‘‘realist scruples . . .
dictate that we take the grammar . . . to have the speaker/hearer as a model’’
(p. 188). This runs counter to my Wrst methodological point according to which
realist scruples require us to take the grammar to be true of linguistic reality.
Despite this, the grammar does dictate something about psychological reality: the
competence of the speaker/hearer must respect the rules ascribed to linguistic
reality by the grammar. Matthews clearly has in mind a closer relation than is
demanded by this Respect Constraint. He seeks a way ‘‘in which a speaker/hearer
might realize a grammar (or, equivalently, the grammar be true of that speaker/
hearer)’’ (p. 188). Since he rejects RT, his position must be a version of (iii). His
demand for the closer relation between the grammar and psychological reality
runs counter to my third methodological point.
Despite this demand, it is hard to see how the theory he oVers diVers from the
minimal position (m). Consider what he has to say about the Marcus parser. He
claims that although it
does not incorporate . . . an EST [Extended Standard Theory] grammar for English,
EST does bear an explanatorily transparent relation to the parsing theory that would have
this parser as one of its models. By this I mean that the syntactic generalizations that are
captured by means of the theoretical constructs of EST . . . are explained in terms of the
organization and operation of the mechanisms postulated by the parsing theory . . . in
the straightforward sense . . . that the generalizations stated in an EST grammar for
English would be true of a speaker who incorporated a Marcus parser. (p. 189)
Yet these generalizations seem to be ‘‘true of a speaker’’ only in the sense that they
are true of the inputs of the speaker: the Marcus parser interprets sentences that
comply with the structure rules speciWed by EST; the parser respects those rules.
But those rules are not the ones that are actually realized in the parser: so EST
does not describe the parser. The relation of the rules of EST to the psychological
reality of the speaker seems to be only that of the minimal position (m): EST
describes the language that the speaker is competent in. And ‘‘the explanatorily
transparent relation’’ between the parsing theory and EST does not seem to add
any more psychological reality to EST because the theory’s explanation of how
the speaker respects the rules of EST does not ascribe any of those rules to the
speaker. If a speaker understands a language, it is trivially the case that the true
theory of her parser will stand in that same explanatorily transparent relation to
the grammar of the language. For the parser must respect the rules of the
language. That is why position (m) should be uncontroversial.
Clearly Matthews wants something stronger than (m). Wherein does the extra
strength lie? In attempting to discover how a position diVers from (m), we must
once again look to criticism of FBG for clues. If Matthews understood ‘‘explana-
tory transparency’’ in the weak sense I have just described, he should hold that
EST might also stand in an explanatorily transparent relation to a theory of
‘‘a parser that incorporates heuristic procedures of the sort hypothesized by’’
FBG. For, the theory of this parser would equally explain how the speaker
respects the rules of the grammar: the speaker does so because those heuristic
procedures do so. Yet, according to Matthews, FBG’s heuristic theory does not
do the explanatory job: ‘‘the relation of parser to grammar would be explanatorily
opaque’’. Why? Because although ‘‘the generalizations stated in the linguistic
theory of the grammar would be satisWed by models of this heuristic theory
. . . they would not be explained by these models’’ (p. 191). The way in which
the Marcus parser unlike the heuristic parser is alleged to realize ESTrests on this
subtle point. So too does the way in which Matthews’ position diVers from (m).
If the subtle point can be be made good, Matthews’ position will be a very weak
version of (iii). I confess that the subtle point is too subtle for me.13
In sum, there are signs in Matthews, as in others we have considered, of
position (m). This is what we should expect and hope for because this minimal
13 A position Matthews contemplates in a later paper seems even less distinguishable from (m).
On the basis solely of a competent speaker’s capacity to map utterances of English sentences onto
beliefs about what was said with them, he is prepared to speak of the mapping process as a
computational implementation of the semantic and syntactic theory of English (2003: 200–2).
But all the mapping requires is that the speaker processes sentences with the semantic and syntactic
properties of English; i.e., it requires only the uncontroversial (m). It does not require that the theory
of the semantic and syntactic properties of English has any role in that mental process other than as
the theorist’s account of the sentences that are processed. The process is ‘‘semantic-involving’’ and
‘‘syntax-involving’’ (Matthews’ terms), in the sense that it involves those semantic and syntactic
properties, not necessarily in the sense that it involves a theory of those properties.
Radford’s position seems also to be indistinguishable from (m). He oVers an analogy to help with
the psychological reality issue: ‘‘Municipal regulations specify certain structural conditions that
houses must meet . . . What they do not do is tell you how to go about building a house . . .
Phrase Structure Rules should be thought of as analogous to municipal building regulations: they lay
down certain structural conditions which sentences must meet’’ (1988: 132).
position should be uncontroversial. But the view that linguistics is part of

psychology demands more than (m) and so Matthews and others are not satisWed
with (m). Yet it is hard to mount a case for the robust positions (i), (ii), or (iii)
on the psychological reality issue. So less robust positions are sought: the
grammar is descriptive of competence in a way stronger than simply positing
rules that are respected by competence but weaker than positing rules that govern
processing or are used as data in processing. According to my third methodo-
logical point, it is a mistake to seek such an intermediate position. Still, perhaps
one can be found and I shall not be arguing that it cannot. I shall be concerned
with the robust positions. If an intermediate position could be found, then the
challenge would be to show that it was theoretically interesting that the grammar
was descriptive of competence in that way.
We have seen that all the writers considered in this chapter apart from Matthews
seem to subscribe to RT. So, that is the powerful psychological assumption
which, according to my second methodological point, they need in taking the
grammar to be true of psychological reality. A commitment to RT strongly
suggests position (i) according to which linguistic rules are represented and
applied in processing, or (ii) according to which the rules are represented and
used as data in processing. And (i) does seem to be the position of BK. But the
others do not give RT a clear enough place in a theory of language use to justify a
conWdent attribution of (i) or (ii) to them. Indeed, FBG seem to hold (iv),
which gives no role to the represented rules in language use. And BW’s version of
(i) is so weak as to be scarcely distinguishable from the minimal position (m)
according to which the linguistic rules are simply respected by the processing
rules. Since Matthews rejects RT, (i) and (ii) are not open to him. He seems to be
urging a version of (iii) according to which the linguistic rules are embodied and
govern processing but his position on psychological reality is so weak as to be also
scarcely distinguishable from (m).
RT is so taken for granted by most of these writers that they do not contem-
plate that the grammar might simply be true of linguistic reality, as urged in my
Wrst methodological point. Finally, most of these writers are strikingly at odds
with my third methodological point. They do not take the grammar and the
theory of competence to be largely independent of each other. Rather, they take
them to place heavy constraints on each other, far heavier than the Respect
Constraint.
FBG made it clear that early theories of how a grammar should be incorpor-
ated into a theory of language use were wrong. As a result grammatical rules were
given less role in that theory in the years that followed: ‘‘During the past Wfteen
years, the role of grammatical theory within models of natural language process-
ing has been continually reduced’’ (Pritchett 1988: 539). Still, the belief that
grammatical rules play a central role in the theory of language use remained, as
Pylyshyn pointed out (1991: 232) and BK, BW, and Matthews demonstrate.
Later work on language processing, to be considered in Chapter 11, indicate that
this belief has weakened in subsequent years. Indeed, this work suggests that
researchers are now pretty much in accord with the methodological points I am
urging.
We have seen no sign of (t), the idea that the grammatical rules have their
place in the mind not as processing rules governing language use but as structure
rules of thought. Indeed, I can Wnd no basis for a conWdent attribution of (t) to
any of these writers. I shall consider Chomsky’s relation to (t) in Part (IV).
Part I of this book argued that linguistics is primarily a study of linguistic
reality. Contrary to received opinion, linguistics is not part of psychology. That
was my Wrst major conclusion. So the great advances in linguistics in recent times
leave the question of the psychological reality of language largely open. The main
concern of this book is to throw light on how this question might be closed. We
have started on this task in Part II by describing some possible positions on the
question, and some historically interesting actual positions on it. We turn next to
the assessment of evidence from various quarters on these positions. In Part III, I
begin with what we might call ‘‘philosophical’’ arguments for RT and hence for
positions (i) or (ii): evidence from the rejection of behaviorism in Chapter 5;
evidence from folk psychology in Chapter 6; and evidence from the role of
intuitions in Chapter 7. I Wnd all this evidence wanting. This is the Wrst step in
establishing my second major conclusion: there is no signiWcant evidence for RT
and, given what else we know, it is implausible. In Part IV we consider what light
the relation of language to thought can throw on the psychological reality
question. Finally, in Part V, against this background about thought, I shall
consider the evidence from language use and language acquisition. This consid-
eration provides the main evidence for my second major conclusion.
All major conclusions are listed at the end of the Glossary.
PA RT I I I
‘‘PH ILOSO PHI CA L’’ A RG U MENTS
FOR THE REPRESENTATIONAL
THESIS
5
The Rejection of Behaviorism
In calling the arguments in this part ‘‘philosophical’’, I do not mean to suggest

that they are not empirical. Indeed, in my Quinean view (1998b), all arguments
are empirical. Still, some arguments have a distinctly philosophical Xavor, par-
ticularly in their distance from the empirical evidence to which all arguments
must ultimately answer.
The Wrst philosophical argument for the Representational Thesis (RT) that
I shall consider is the argument from the rejection of behaviorism. The argument
can be put brieXy: RT, the view that competence in a language involves repre-
senting its rules, follows from the rejection of behaviorism, particularly from
Chomsky’s famous critique (1959) of B. F. Skinner. And my response can be put
just as brieXy: it doesn’t follow; there is a vast amount of logical space between the
rejection of behaviorism and RT.
I do not attribute this argument to any linguist. Indeed, perhaps none holds it,
particularly in the bald form in which I have just stated it. Still, the literature
sometimes seems to suggest the argument; it seems to suggest that once we have
seen the failures of behaviorism, RT is obvious. Fiona Cowie (1999: 159–62)
attributes the argument to Chomsky and, surprisingly, endorses it. All in all, it is
worth a brief consideration.
Behaviorists deny anything to the mind beyond dispositions to behave in
certain ways in response to certain external stimuli. For them the mind, linguistic
competence included, is just a set of input/output functions. This reXects the
crude empiricist dislike of things unseen; an unwillingness to posit theoretical
entities that explain the observed phenomena. Behaviorism seems powerless to
explain our linguistic abilities, as Chomsky points out. Most strikingly, it
founders on the stimulus independence of language: most things we say are not
tied closely to particular stimuli; indeed the link between stimulus and utterance
may be extremely indirect. Furthermore, behaviorism cannot account for the
novelty of language use: right from the beginning of acquiring a language, the
child shows a capacity to produce novel utterances, hence ones that could not
have been learned from association with particular stimuli. To explain our
linguistic competence, we need to ascribe complicated inner states interacting
with each other, as well as with various stimuli, to produce our responses: ‘‘there
88 ‘‘Philosophical’’ Arguments
are, behind the overtly expressed sequences, a multiplicity of integrative pro-
cesses’’ (Chomsky 1959: 55).
I very much agree with this rejection of behaviorism. But our discussion has
already shown that the acceptance of a rich mental life underlying language is a
long way from RT. Perhaps the rules governing our linguistic performance are
embodied without being represented. Perhaps these rules are heuristic rules that
are quite diVerent from the rules ascribed by the grammar. Perhaps competence
consists in a set of dispositions that mostly involve internal stimuli and internal
responses and so mostly involve the fairly brute–causal processing of versions (b)
of positions on psychological reality in section 3.4. There is room for a lot of
mental complication without RT. The rejection of behaviorism shows that the
story of linguistic competence must be a mentalistic one. To establish which
mentalistic story it is requires a lot more. It requires evidence about what
precisely is going on in the mind.
I turn now to the argument from folk psychology.
6
Folk Psychology
6.1 P H I L O S O PH E R S V S . C H O M S K Y
The folk say that a person competent in a language ‘‘knows’’ the language. We
have noted that the knowledge attributed might be of two very diVerent sorts. It
might be knowledge-that, propositional knowledge, or it might be knowledge-
how, in the same family as skills, abilities, and capacities. If it is knowledge-how,
it might still involve some propositional knowledge about the rules but it might
not. And if it is mere knowledge-how then the implication seems to be that it does
not (3.1). Some philosophers, including me,1 have argued against the know-
ledge-that view and, largely for that reason, have claimed that competence is
mere knowledge-how.2 Chomsky is dismissive of the knowledge-how view,
arguing that it is ‘‘entirely untenable’’ (1988: 9). He usually writes as if he
endorses the knowledge-that view.
We should note Wrst that linguistic competence could not be simply
knowledge-that. The competence is a capacity to use language. If that competence
1 1981: 95–103; Devitt and Sterelny 1989.

2 Jason Stanley and Timothy Williamson (2001) deny that ‘‘there is a fundamental distinction
between knowledge-how and knowledge-that. Knowledge-how is simply a species of knowledge-
that’’ (p. 411). On the basis of this they reject my claim in a previous book (1996) that competence
in a language does not require any propositional knowledge about the meanings of expressions. And,
on the same basis, they would surely reject my claim in this book that competence does not require
any propositional knowledge about the syntax of expressions: they would reject my claim that the
competent can be ignorant of language. Their ingenious argument starts from the linguistic claim
that ascriptions of knowing-how in English contain embedded questions. They then argue that,
according to the best semantic theories, such ascriptions attribute propositional knowledge. From
this they arrive at their conclusion that knowing-how is a species of knowing-that. So, on the basis of
assumptions about language, they draw an unlikely metaphysical conclusion. (Just how unlikely is
nicely demonstrated by SchiVer 2002.) I think that we should always be suspicious of this way of
proceeding: our linguistics should be guided by our metaphysics rather than vice versa because we
know far less about our ways of talking about the world than we do about the world talked about
(1996, 1997a, 1998a). And I think that Ian RumWtt (2003) shows that such suspicions about Stanley
and Williamson’s argument are well founded. RumWtt attends to ascriptions of knowing-how in
other languages and shows that the argument does not hold up. Thus, although some ascriptions of
knowing-how in French may Wt Stanley and Williamson’s analysis, those ascribing knowing how to
swim, ride a bicycle, and the like do not. These ascriptions seem to attribute a relation to an activity
not to a proposition, just as those who emphasize the distinction between knowing-how and
knowing-that would hope.
involved some knowedge-that it would also have to involve some knowledge-how:
it would have to involve knowing how to apply the knowledge-that in language
comprehension and production. (This point is related, of course, to the earlier one
that where a rule governs a process by being represented and applied there must be
another rule that governs the application; 3.1.) So we should take the knowledge-
that view to be that linguistic competence is constituted largely, although not
entirely, by propositional knowledge of the rules of the language.
What hangs on this dispute? BrieXy, the initial credibility a view gets from
siding with the folk. The folk think that the competent person knows the
language. One could take the high-handed line that this is just a piece of ‘‘folk
psychology’’, one of a set of unscientiWc folk opinions, and so we need not worry
about it. Certainly folk opinions have often been wrong, but they have very often
been right too. Opinions that have stood up to the experiences of many lifetimes
should not be lightly set aside: the ‘‘wisdom of the ages’’ has often been wise. The
parties to the present dispute think, rightly in my view, that the folk opinion of
linguistic competence is one we should try hard to accommodate. Now suppose,
on the one hand, that this knowledge of language is construed as knowledge-that.
Suppose, further, that we accept the Representational Theory of the Mind
(RTM) as many do (although perhaps not Chomsky). Then the folk view yields
the conclusion that the competent person represents the rules of her language:
RT. So positions like (i) or (ii) on the psychological reality of language are
supported: the represented rules are applied or used as data in language process-
ing (3.4). I don’t claim that this line of argument is explicit in the literature
but it often seems to me to be implicit. Next suppose, on the other hand,
that the knowledge is construed as knowledge-how. Then the folk view gives
no support to RT. And if the knowledge is construed as mere knowledge-how,
hence not involving any knowledge-that, then prima facie RT should be rejected.
So, in arguing against the knowledge-that construal, the philosophers are under-
mining positions (i) and (ii). And in arguing against the knowledge-how
construal, Chomsky is undermining the view that linguistic competence is like
a skill or ability, a view required by position (iii) according to which linguistic
rules are processing rules, and encouraged (as we shall see in Part IV) by position
(t) according to which linguistic rules are similar to the structure rules of
thought.
Of course, even if arguments in this dispute are persuasive, they leave plenty of
room for manoeuver. For example, one could claim, as Chomsky sometimes
does, that a speaker’s knowledge of a language is neither knowledge-that nor
knowledge-how (1969a: 86–7; 1969b: 153).3 Certainly, the ordinary uses of
3 Some phonologists sum up this view mysteriously as follows: ‘‘Rather than being relational [in
the senses of knowledge-how, knowledge-of, or knowledge-that], ‘knowledge of language’ is for
Chomsky constitutive, in the sense that there is no distinction between what is known and the
knowing of it.’’ (Burton-Roberts et al. 2000: 3)
Folk Psychology 91
‘know’ are so many and various that this possibility cannot be dismissed out of
hand. Still, a lot of work then has to be done to establish that there is this other
sort of knowledge and that it supports RT. The immediate support that RT gets
from identifying linguistic competence with knowledge-that is lost. Or one could
accept that the competence is mere knowledge-how but reject the prima facie
consequence that this rules out RT: one could argue that, at some level, this
knowledge-how involves representation of the rules. But, once again, the imme-
diate support for RT is lost.
6.2 T H E PH I LO S O PH E R S
Let us start with the philosophers’ side of the dispute. The philosophers are
mostly inXuenced by the view that a speaker’s knowledge of her language is so
unlike uncontroversial cases of propositional knowledge.4 Stephen Stich (1971,
1978b) made the point nicely. If a person knows that p, we expect him to be
aware of p, or at least to be able to become aware of it when given a suitable
prompt; and we expect him to understand expressions of p. The ordinary speaker
quite clearly lacks this awareness and understanding for most of the grammar. If a
person knows that p, his knowledge should join up with other knowledge and
beliefs to generate more beliefs. If a speaker has knowledge of the grammar it is
clearly not inferentially integrated in this way. Consider an example. Without
tuition, a speaker is unlikely to have the conceptual recourses to understand even
the relatively simple claim that ‘NP > Det þ Adj þ N’ is a rule of English. If
she knows that this is a rule, her knowledge is largely inferentially isolated from
her other beliefs.
Of course, Chomsky’s view is that speakers’ knowledge of the rules of the
language is only tacit. But this is no help to the knowledge-that view because
our knowledge of the language is very diVerent also from ordinary tacit
propositional knowledge. For, such knowledge is knowledge that a person has
not entertained but which he would acknowledge in suitable circumstances;
thus, Ron tacitly knows that rabbits don’t lay eggs, even though the thought has
never crossed his mind, because he would readily acknowledge that they don’t
lay eggs if the question were ever to arise. Clearly, the typical speaker does not
have this relation to claims about the rules of her language. First, she lacks
many of the concepts necessary even to understand such claims. Second, even
if she had the necessary concepts, the truth of the claims would seem far from
obvious to her.
4 Dwyer and Pietroski (1996) argue in the opposite direction. They think that we have such good
reasons (of the sort criticized in Chs. 7 and 10 to 12 below) for the view that speakers believe
linguistic theory that this view should constrain our theory of belief.
In the face of this one might attempt to deWne a technical sense of ‘‘tacit
knowledge’’ that covered a speaker’s relation to the rules of her language.5 But if
that sense does not require that a person who tacitly knows the rules of her language
represents those rules then this knowledge will clearly do nothing to support RT.
And if that sense does require that she represents those rules then we need an
independent argument to establish that she represents those rules. We need inde-
pendent evidence that she tacitly knows the rules in a sense requiring that they be
represented. So, either way, the deWnition can do nothing to support RT.
In sum, a person’s knowledge of her language is very unlike what we ordinarily
think of as propositional knowledge, or even tacit propositional knowledge. So
the indubitable fact that speakers do know their language gives no support to RT.
Philosophers have also been inXuenced by a neat argument proposed by
Gilbert Harman (1967). If a speaker’s competence in a language consists in
having knowledge-that of its rules then, assuming RTM, she must represent
those rules. That representation must itself be in a language. What is it to be
competent in that more basic language? If we suppose the more basic language is
the same as the original language then we are caught in a vicious circle. If we
suppose that it is some other language (‘‘Mentalese’’ perhaps), then its rules also
have to be represented. This requires a still more basic language. And so on. The
only way to avoid a vicious circle or an inWnite regress is to allow that we can be
competent in at least one language directly, without representing its rules. Why
not then allow this of the original language, the one spoken?
6.3 CHOMSKY
Turn now to Chomsky’s side of the dispute. Chomsky takes the knowledge-how
view of linguistic competence to be that competence is a ‘‘practical ability’’ to use
the language in understanding and speech. He objects:
Two people may share exactly the same knowledge of language but diVer markedly in
their ability to put this knowledge to use. Ability to use language may improve or decline
without any change in knowledge. This ability may also be impaired, selectively or in
general, with no loss of knowledge, a fact that would become clear if injury leading to
impairment recedes and lost ability is recovered. (1986: 9)
Let us start with the diVerences in ability to speak. Chomsky gives two
examples of the sort of diVerence that he has in mind. The Wrst is the diVerence
brought about by ‘‘a public speaking course’’ (p. 10). But this is beside the point.
The know-how for public speaking requires ordinary linguistic know-how but is
5 Thus, in an early paper, Fodor counts an organism that knows how to X as tacitly knowing that
S if S speciWes a sequence of operations that the organism runs through in X-ing (1981: 75). As
Graves et al. point out, this ‘‘liberal’’ proposal has the consequence that a bicyclist tacitly knows the
diVerential equations that determine how he maintains balance (1973: 324). Martin Davies (1987;
1989) makes an heroic attempt to deWne a technical sense of ‘‘tacit knowledge’’.
Folk Psychology 93
diVerent from that know-how, as the folk plainly acknowledge. The fact that a
person competent in a language can gain another competence as a result of a
public speaking course or, for that matter, an elocution course or a calligraphy
course, does nothing to show that all of these competences are not mere know-
hows. A point made in the earlier discussion about the concerns of the theory of
linguistic outputs—the theory of language—is relevant. The point was that the
theory, like the theory of the outputs of the logic machine (2.2), abstracts from a
range of properties of the outputs—like form of script and pitch of sound—
focusing simply on syntactic properties. It follows that the theory of the compe-
tence to produce those outputs must make the same abstraction.
Chomsky’s second example is of the diVerence between ‘‘a great poet’’ and ‘‘an
utterly pedestrian language user who speaks in cliches’’ (1988: 10). But, once
again, the diVerence is in another know-how—presumably, largely, a diVerence
in thought—and does not show that knowledge of the language is not know-
how. To suppose that it is know-how is not to suppose that there are no other
skills that depend on it.
Consider next Chomsky’s claim that a person’s ability to use a language can be
impaired by brain damage even though her knowledge of the language remains
relatively stable. There can be no argument with this. But it does not show that
the stable knowledge is not know-how because the same can be said of clear cases
of know-how. A person knows how to ride a bicycle but cannot do so because his
leg is broken; a person knows how to catch but cannot do so because she has
blisters; a person knows how to touch type but cannot do so because he has a
migraine. Indeed, it is presumably the case that exercising any know-how
requires the satisfaction of some internal background conditions.
Chomsky rightly insists that ‘‘to know a language . . . is to be in a certain
mental state, which persists as a relatively steady component of transitory mental
states’’ (1980b: 5). But he writes as if taking this knowledge as mere know-how
must saddle it with a whole lot of irrelevant features of performance (1986: 10)
and must make behavior ‘‘criterial’’ for the possession of the knowledge not
merely evidential (1980b: 5). This is not so. A person’s know-how can be an
underlying steady state abstracted from features of performance. It can be, as
Chomsky insists our knowledge of language is, ‘‘a cognitive system of the mind/
brain’’(1988: 10) and yet still be akin to a skill or ability. Usually, such an ability
gives rise to certain behavior which then counts as evidence for the ability. But the
ability may not give rise to the behavior. The behavior is not ‘‘criterial’’.
It seems to me that Chomsky’s tactic in this dispute with the philosophers—

roughly, ‘‘Don’t give an inch’’—is a mistake. Chomsky thinks that the cognitive
system for language is to be found in a module of the mind, a distinct mental
‘‘organ’’, the ‘‘language faculty’’. This faculty is largely inaccessible to the oper-
ations of the ‘‘central processor’’ which does what we would ordinarily think of as
thinking; hence the ordinary speaker’s inability to state the rules of the language.6
I think that Chomsky would best accommodate folk talk of ‘‘knowing a lan-
guage’’ to this view by accepting the philosophers’ claim that this knowledge is not
knowledge-that but knowledge-how, and then arguing that knowledge-how in
this case, if not others, requires the representation of rules in a module.7 This
tactic has a disadvantage, of course. If RT is understood in this way it gets no
support from the folk talk of knowledge: the argument for RT must start from
scratch. But doubtless this is not a disadvantage that would concern Chomsky.
In sum, I do not think that the folk view that we know the language we speak
gives any evidential support to RT and hence to positions (i) or (ii). The case for
those positions must be developed elsewhere.
6 Some hold to the thesis that the mind is so massively modular that there is little or nothing for
the central processor to do (Barkow, Cosmides, and Tooby 1992; Baron-Cohen 1995; Smith and
Tsimpli 1995; Sperber 1996). I share Kim Sterelny’s skepticism about this thesis (2003: ch. 10); see
also Atkinson and Wheeler 2004. So far as I can see, it would take an extreme, hence highly
implausible, version of the thesis to count against my arguments. In any case, I shall continue to
write as if the thesis were false.
7 This is what Stich recommends in his peer commentary (1980) but Chomsky does not accept
the recommendation (1980c: 57).
7
Intuitions
7.1 I N T RO D U C T I O N
So far then we have found no support for the Representational Thesis (RT) in the
rejection of behaviorism nor in the folk view that speakers ‘‘know’’ the language
they speak. We turn now to a story that linguists tell about linguistic intuitions as
a possible source of another argument for RT.
We should start by clarifying what we mean by ‘‘linguistic intuitions’’. We
mean fairly immediate unreXective judgments about the syntactic and semantic
properties of linguistic expressions, metalinguistic judgments about acceptabil-
ity, grammaticality, ambiguity, coreference/binding, and the like. These judg-
ments are frequently expressed in utterances; for example, ‘‘ ‘Visiting relatives
can be boring’ is ambiguous’’ or ‘‘In ‘Tom thinks Dick loves himself ’ ‘himself ’
must refer to the same person as ‘Dick’.’’ Such metalinguistic utterances are not
to be confused, of course, with the vastly more numerous utterances we make
about the nonlinguistic world. Nor are they to be confused with behavioral
responses that are not metalinguistic utterances; for example, looking baZed by
an ungrammatical utterance, or behaving in a way that clearly takes a pronoun to
corefer with a certain name. The intuitions in question are judgments about
linguistic performances not the performances themselves.
So, what story do linguists tell about these intuitive judgments of competent
speakers? The story starts with the claim that the intuitions of competent speakers
provide the main evidence for linguistic theories. As Chomsky puts it ‘‘linguis-
tics . . . is characterized by attention to certain kinds of evidence . . . largely, the
judgments of native speakers’’ (1986: 36). Indeed, the emphasis on these intu-
itions is sometimes so great as to imply that they are the only evidence the linguist
has. Thus, Liliane Haegeman, in a popular textbook, says that ‘‘all the linguist has
to go by . . . is the native speaker’s intuitions’’ (1994: 8).1 But this is not the
approved position. Evidence is also to be found in linguistic usage, at least.
1 But two pages later Haegeman allows, somewhat grudgingly, an evidential role for usage.
Andrew Radford opens a book (1988) with an extensive discussion of the evidential role of
intuitions. The Wrst mention of the use of the ‘‘corpus of utterances’’ as data does not come until
p. 24. Robert Fiengo starts an interesting paper on linguistic intuitions: ‘‘Intuitions, with the
contents that they have, are the data of Linguistics’’ (2003: 253).
In the next section I shall argue that speakers’ intuitions are not the main
evidence for linguistic theories. Still, I agree that they are evidence. Why are they?
The standard linguistic answer, noted at the beginning of the book (1.1), rests on
RT. These intuitions are thought to be good evidence because the speaker derives
them from a representation of the rules of the language, a representation that
constitutes the speaker’s linguistic competence. The derivation is a causal and
rational process like a deduction:
it seems reasonably clear, both in principle and in many speciWc cases, how unconscious
knowledge issues in conscious knowledge . . . it follows by computations similar to
straight deduction. (Chomsky 1986: 270; see also Pateman 1987: 100; Dwyer and
Pietroski 1996: 342)
we cognize the system of mentally represented rules from which [linguistic] facts fol-
low. (Chomsky 1980b: 9; the facts are expressed in intuitive judgments)
We can use intuitions to conWrm grammars because grammars are internally represented
and actually contribute to the etiology of the speaker/hearer’s intuitive judgments. (J. A.
Fodor 1981: 200–1)
[A speaker’s judgments about the grammatical properties of sentences are the result of ] a
tacit deduction from tacitly known principles. (Graves et al. 1973: 325)
Our ability to make linguistic judgments clearly follows from our knowing the languages
that we know. (Larson and Segal 1995: 10; see also Pylyshyn 1984: 122; Baker 1995: 20)
So, on this explanation, linguistic competence alone provides information about
the linguistic facts; the intuitive judgments are, as I put it, ‘‘the voice of compe-
tence’’. So these judgments are not arrived at by the sort of empirical investigation
that judgments about the world usually require. Rather, a speaker has a privileged
access to facts about the language, facts captured by the intuitions, simply in
virtue of being competent and thus embodying representations of its rules in her
language faculty, a module of the mind. I need a word for such special access to
facts. I shall call it ‘‘Cartesian’’. We would like the details of this Cartesian
explanation spelt out, of course. We would like to know about the causal–rational
route from an unconscious representation of rules in the language faculty to a
conscious judgment about linguistic facts in the central processor. Still, the idea
of one sort of representation leading to another is familiar and so this standard
Cartesian explanation may seem promising.2
We can see in this story an argument for RT and hence for positions (i) or (ii)
on the psychological reality of language (the represented rules are applied or used
as data in language processing; 3.4). For, the standard explanation can be used
as the basis for the following abduction: RT is the core of a good explanation of
why the intuitions are evidence, and there is no other explanation; so RT is
2 Note that the explanation does not suppose that the speaker has Cartesian access to the
linguistic rules, just to the linguistic facts captured by the intuitions.
Intuitions 97
probably true. If the intuitions are really derived from representations of the
grammatical rules then they must be true and hence good evidence for the nature
of those rules. But if they are not so derived, how could they be good evidence?
How could they have this evidential status unless they really were the voice of
competence?3
Various things are required for this abduction to be good. One is that RT has
to be independently plausible given what else we know. If my second major
conclusion can be established, then RT fails this test very badly. But that is not
the concern of this chapter. Here I shall be arguing that there are reasons for
doubting the standard Cartesian explanation quite apart from doubts about RT.
More importantly, I shall be arguing that another explanation of the evidential
role of linguistic intuitions, an explanation that does not rest on RT, is better. So
the abduction fails anyway. My main concern in this chapter is to show this and
hence that RT gets no support from evidential role of intuitions.
I am, of course, understanding the standard explanation and RT in light of the
sense of ‘represent’ illustrated in section 1.1. Although this is the natural way to
interpret the linguists’ talk, I noted that it may not be the right way. ‘Represent’
(and its cognates) is used so widely and loosely in cognitive science that it is hard
to be conWdent about what it means on any one occasion. This is not the place to
try to analyze these uses and come up with other interpretations. However, I take
it that linguists who talk in this way but reject my interpretation, and hence the
standard explanation (as I am understanding it), will nonetheless hold that
the rules are embodied somehow in a speaker without being represented (in the
above sense), and govern processing: position (iii). And these linguists are still
committed to the Cartesian view that intuitions are the voice of competence, the
view that speakers, simply in virtue of being competent, have information about
the linguistic facts. How could this be so if the rules are not represented? The
linguists need what I shall call ‘‘a nonstandard Cartesian explanation’’ of
why linguistic intuitions are good evidence for linguistic theories, an explanation
of how unrepresented rules provide the privileged access to linguistic facts.
3 Against the background of the standard Cartesian explanation, we can see the following as a
statement of the argument:
Linguists normally take the intuitions of speaker/hearers to be the data to which structural descrip-
tions are required to correspond. But this practice would be quite unwarranted unless it were assumed
that speaker/hearers do have access to internal representations of sentences and that these provide a
reliable source of information about the character of the abstract object (the language) which, on any
view, the grammar is ultimately intended to describe. (Fodor, Fodor, and Garrett 1975: 244)
Graves et al. claim that the only plausible explanation of a speaker’s explicit knowledge of
grammatical facts is that she has tacitly deduced that knowledge from tacitly known principles of
the language (1973: 324–9). The argument is implicit in Laurence 2003 (pp. 89–91).
In sections 7.3 to 7.6 I shall argue that the standard Cartesian explanation is
not the best: the evidential role of linguistic intuitions can be better explained
otherwise. This other explanation does not suppose that the intuitions are the
product of embodied linguistic rules, whether represented or not; they are not
the voice of competence. Rather, they are opinions resulting from ordinary
empirical investigation, theory-laden in the way all such opinions are. If this is
right, then the abduction to RT fails. In section 7.7 I shall consider the possibility
of a nonstandard Cartesian explanation. As we have just noted, a linguist who
rejects the view that linguistic rules are represented (in the above sense) in
speakers needs such an explanation. And if he had one, it might, of course, be
the basis for an abduction to position (iii). So far as I know, no such explanation
has ever been proposed. I shall argue that none is likely to be forthcoming.
Finally, in section 8, I shall look critically at the view that it is a task of linguistics
to explain linguistic intuitions.
I turn now to consider brieXy what sorts of evidence we do have for linguistic
theories.
7.2 EV I D E N C E F O R L I N G U I S T I C T H E O R I E S
I began the chapter by noting the received view in linguistics that the intuitions
of competent speakers provide the main evidence for linguistic theories; indeed,
the view is often that these intuitions are near enough the only evidence. These
views greatly exaggerate the evidential role of the intuitions. As recent experi-
menters who did actually test the intuitions of naı̈ve subjects remark, this testing
is ‘‘in contrast to common linguistic practice’’ (Gordon and Hendrick 1997:
326). I suggest that, as a matter of fact, only a small proportion of the evidence
used in grammar construction consists in the canvassed opinions of the ordinary
competent speaker.
So what else constitutes the evidence? (i) One possible source of direct
evidence is ‘‘the corpus’’, the linguistic sounds and inscriptions that the folk
have produced and are producing as they go about their lives without any
interference from linguists. We can observe people and seek answers to questions
like: ‘‘Do people ever say x?’’; ‘‘How do they respond to y?’’; ‘‘In what circum-
stances do they say z?’’ The role such observations have played, particularly in the
beginnings of linguistics, is insuYciently acknowledged. Linguists may well rely
extensively on intuitions now that generative grammars are in an advanced stage
(even though not complete, of course). But think back to the beginning, perhaps
just to the dark days before there were any generative grammars. Surely a lot of
the early knowledge about languages, still captured by generative grammars, was
derived from simply observing linguistic usage, much as the Weld linguist does.
Even now, it is hard to believe that L-speaking linguists surrounded by other L-
speakers are uninXuenced by the data they are immersed in. One would expect
Intuitions 99
them, given their training, to be peculiarly sensitive to these data. Anecdotal
evidence suggests that they are indeed so sensitive; that they are continually
on the watch and noting linguistic oddities. Consider the theory of ‘wanna’
constructions, for example. We can surely be conWdent that the linguists’ obser-
vations of the use of ‘wanna’ have played a considerable role in building this
theory.
(ii) Another possible source of direct evidence in grammar construction, as I
have emphasized (2.5), is psycholinguistic evidence of what syntactic rules
speakers respect. The actual role of this evidence is small, largely conWrming
what is already fairly well established, because our knowledge of linguistic reality
is so much more advanced than our knowledge of psychological reality. Still, if
psycholinguistic evidence showed that a syntactic rule was not respected, that
would be very bad news for the rule.
(iii) Interference by linguists can yield further direct evidence: we can contrive
situations and see what subjects say or understand. Consider, for example, this
description of ‘‘the technique of elicited production’’:
This technique involves children in a game, typically one in which children pose
questions to a puppet. The game orchestrates experimental situations that are designed
to be uniquely felicitous for production of the target structure. In this way, children are
called on to produce structures that might otherwise not appear in their spontaneous
speech. (Thornton 1995: 140)
Although this sort of technique is frequently used on children,4 analogous ones
are doubtless seldom used on adults because contriving these situations is likely
to be laborious. But, clearly, much evidence could be gathered in this way.
We might aptly describe a person’s responses in these contrived situations as
‘‘intuitive,’’ but those responses are very diVerent from the intuitions that are the
main concern of this chapter. They diVer in not being judgments about the
syntactic and semantic properties of expressions and hence not needing to deploy
linguistic concepts.
(iv) Another sort of interference can yield evidence that is less direct: we can
describe situations and ask people what they would say or understand in those
situations. This evidence is less direct because it depends on people’s reXections
on these situations. Once again, we might aptly describe a person’s response as
‘‘intuitive’’ but it is not an intuitive judgment about linguistic properties. Such
responses are another source of evidence that is insuYciently acknowledged.
Linguists ask themselves, and sometimes ordinary speakers, what they would
say or understand in various situations.
(v) Further less direct evidence can be obtained from language acquisition:
evidence about what sorts of languages we could learn is evidence about what
4 See Crain et al. 2005, section 4, for a nice summary.

language we have learnt. This sort of evidence is, of course, frequently acknow-
ledged.
(vi) Finally, I suggest that a good deal of less direct evidence comes from the
intuitive opinions of linguists about the languages they speak. As an entry in
the Encyclopedia of Cognitive Science points out, partly because it is sometimes
‘‘diYcult to replicate relatively agreed-upon judgments of linguists while testing
naive subjects . . . and partly for sheer convenience, linguists rely increasingly
on other linguists for judgment data on some languages’’ (Schutze 2003: 913).
The intuitions of ordinary speakers may seem to be more present than they really
are because linguists take their own intuitions to be representative.
There are surely many other ways to get evidence of linguistic reality apart
from consulting intuitions, just as there are to get evidence of any other
reality.
In sum, the main evidence for grammars is not found in the intuitions of
ordinary speakers but rather in a combination of the corpus, the evidence of what
we would say and understand, and the intuitions of linguists. Still, it cannot be
denied that ordinary competent speakers do have largely reliable intuitions which
do play an evidential role in linguistics, even if that role is greatly exaggerated. So
we still need to explain that role. And the abduction still stands: RT is the best
explanation of this role.
I shall be developing the theory that linguistic intuitions are theory-laden
empirical opinions. But I begin by noting a tension in linguistic discussions of
these intuitions arising, it seems to me, from the attraction of this theory.
7.3 T E N S I O N I N T H E L I N G U I S TS ’ V I EW O F I N T U I T I O N S
A simplistic version of the standard Cartesian explanation would be that since

linguistic intuitions are derived from the rules of the language they must always
be true and so are as good as any evidence could be. Although there are signs of
this version in the literature,5 it is not the approved version. The approved
version allows for errors arising from ‘‘noise’’ (1.1). So, just as there can be
performance errors in using our competence to produce and understand the
sentences of our language, there can be performance errors in producing judg-
ments about such sentences. The explanation is Cartesian in supposing that we
have a nonempirical privileged access to linguistic facts not in supposing that this
access yields infallible judgments.
This retreat from reliance on intuitions seems clearly appropriate on the
standard view. Other retreats are not so clearly so. (i) Although there has been
reliance on intuitions about grammaticality the tendency in recent times has been
5 Fodor, Bever, and Garrett 1974: 82; Baker 1978: 4–5.

Intuitions 101
to emphasize ones about acceptability, goodness, and the like,6 and to oVer
explanations of these intuitions that are often syntactic but sometimes semantic
or pragmatic. Yet grammaticality is the notion from linguistic theory and so if the
intuitions are really derived from a representation of that theory, shouldn’t we be
relying on intuitions about grammaticality? If, in our intuitive judgments,
competence is really speaking, why doesn’t it use its own language? What is the
causal–rational route from an unconscious representation of something’s ‘‘gram-
maticality’’ to a conscious judgment of its ‘‘acceptability’’? (ii) Ordinary speakers
have many intuitions about grammaticality, coreference, and ambiguity, but few
about transitivity, heads, A-positions, c-command, cases, transformations, and so
on. Why is that? Why doesn’t linguistics have a much wider range of intuitions to
rely on? Linguistic theory is very rich. If our competence consists in representing
this theory and our competence speaks to us at all, how come it says so little? Once
again, we wonder how the details of the causal–rational route could be spelt out
to account for this.
The clue to what underlies these retreats from reliance on intuitions is to be
found in many passages like the following:
it is hardly surprising that informants should not be able to tell you whether a sentence is
pragmatically, semantically, or syntactically ill-formed: for these very notions are terms
borrowed from linguistic theory: and like all theoretical terms, they are meaningless to
those not familiar with the theory. (Radford 1988: 13)
Such passages reXect a sensitivity to the highly theoretical nature of linguistic
terms. And they reXect the attractive thought that these terms have their place in
empirical theories that are hard-won by linguists with the result that judgments
involving them are not plausibly attributed to people simply in virtue of being
competent speakers.7 The retreat to acceptability may seem to escape this thought
because ‘acceptable’ is a very ordinary term and not in linguistic theory.
I see linguists as pulled two ways in their treatment of the intuitive judgments
of speakers. On the one hand, the standard view is that speakers represent the
true linguistic theory of their language and derive their intuitive judgments from
those representations. So, those intuitive judgments, deploying terms drawn
from that theory, should be the primary data for the linguist’s theory. On the
other hand, there is the attractive thought that all judgments deploying those
terms are laden with an empirical linguistic theory. Where the judgments are
those of the ordinary speaker, that theory will be folk linguistics. We do not
6 For example, Higginbotham 1987: 123; Radford 1988: 10; Hornstein 1989: 26, 38n; Haege-
man 1994: 7; Baker 1995: 8, 38. Dwyer and Pietroski do not exemplify this tendency (1996: 346).
7 In this respect it is interesting to note Chomsky’s skepticism about ‘‘contemporary philosophy
of language’’ and its practice of ‘‘exploring intuitions about the technical notions ‘denote’, ‘refer’,
‘true of ’, etc.’’ He claims that ‘‘there can be no intuitions about these notions, just as there can be
none about ‘angular velocity’ or ‘protein’. These are technical terms of philosophical discourse with a
stipulated sense that has no counterpart in ordinary language’’. (1995: 24; 2000: 130)
generally take theory-laden folk judgments as primary data for a theory. So we
should not do so in linguistics.
The emphasis on intuitions about acceptability (also goodness and the like)
may seem to remove this tension, but it doesn’t really. First, intuitions about
acceptability are not the only ones playing an evidential role. There are also
intuitions about coreference/binding, ambiguity, and so on, involving terms that
are straightforwardly linguistic. Whatever we say about ‘acceptable’ is no help
with the tension associated with these intuitions. Second, ‘acceptable’ (also
‘good’ and the like) is a highly context-relative term: it might mean acceptable
in polite society, acceptable in a philosophical argument, and so on. What a linguist
is aiming to elicit from an ordinary speaker is, of course, an intuition about what
is acceptable grammatically in her language; he wants the voice of her competence.
He may attempt to make this explicit; for example, ‘‘Is this expression acceptable
in your language?’’ Or it may be implicit; ‘‘Is this expression acceptable?’’ asked in
the right context by someone known to be a linguist. In these situations the
speaker may naturally take ‘acceptable’ to express her notion of grammaticality
(even if she lacks the term ‘grammatical’ for that notion). If she does take it that
way, ‘acceptable’ in these situations acts as a synonym for ‘grammatical’. So we are
still pulled toward seeing her intuitive responses as judgments laden with folk
linguistics. So the tension remains.
In other linguistic contexts ‘acceptable’ (and ‘good’) is likely to be taken in
diVerent ways. Thus the question ‘‘Is this expression acceptable in your commu-
nity?’’ would invite the speaker to consider not only grammatical facts of her
language but also pragmatic ones about etiquette, appropriateness, interest, and
so on. And there is a considerable risk that the simple ‘‘Is this expression
acceptable?’’ and even ‘‘Is this expression acceptable in your language?’’ will
also bring in pragmatic considerations. Yet, clearly, the linguist is concerned to
ask questions that minimize the intrusion of pragmatics into intuitions that he
hopes are the voice of competence.8 In any case, insofar as pragmatic consider-
ations do intrude, the attractive thought encourages the view that the intuitions
are still theory laden: they are laden not only with folk linguistics but also with
pragmatic theories about what is good etiquette, socially appropriate, interesting
enough to be worth saying, and so on. So there is still a tension.
The discussion in this section is the Wrst step in undermining the standard
Cartesian explanation. On the one hand attention to the language in which
competence allegedly speaks and to how little it says raises concern about the
details of the causal–rational route from representations in the language faculty
to a judgment in the central processor. On the other hand, we have begun to see
8 This being so, one wonders why linguists would ever use such a vague, pragmatic, context-
relative term as ‘acceptable’ to seek grammatical intuitions unless pragmatic factors are controlled for
(as in ‘‘minimal-pair’’ experiments; see section 7.5).
Intuitions 103
why the thought that ordinary linguistic intuitions are laden with empirical folk
linguistics is attractive.
In section 7.4 I shall present a view of intuitions in general. In section 7.5 I
shall apply this view to linguistic intuitions, yielding a view of them along the
lines of the attractive thought. This view removes the tension by abandoning
the received idea that the intuitions are derived from a representation of the rules:
competence has no voice. And it yields an explanation of the evidential role
of linguistic intuitions which, I shall argue in section 7.6, is better than the
standard one.
7.4 INTUITIONS IN GENERAL
Questions about the status of intuitions do not arise only in linguistics, of course;
intuitions play a role in ordinary life and science, and seem to dominate
philosophy. What are we to say of them in general? In Coming to Our Senses
(1996: 72–85) I argue for a naturalistic, and nonCartesian, view of intuitions in
general.9 On this view, intuitive judgments are empirical theory-laden central-
processor responses to phenomena, diVering from many other such responses
only in being fairly immediate and unreXective, based on little if any conscious
reasoning. These intuitions are surely partly innate in origin10 but are usually and
largely the result of past reXection on a lifetime of worldly experience’’.11
A clariWcation. It may be that there are many unreXective empirical responses
that we would not ordinarily call intuitions: one thinks immediately of percep-
tual judgments like ‘‘That grass is brown’’ made on observing some scorched
grass, or ‘‘That person is angry’’ made on observing someone exhibiting many
signs of rage. Perhaps we count something as an intuitive judgment only if it is
not really obvious. I shall not be concerned with this. My claim is that intuitions
are empirical unreXective judgments, at least. Should more be required to be an
intuition, so be it.
In considering intuitions and their role in science, it is helpful to distinguish
the most basic intuitions from richer ones. Suppose that we are investigating the
nature of a kind F—for example, the kind gene, pain or echidna. The most basic
intuitions are ones that identify Fs and non-Fs; for example, ‘‘This is an echidna
but that isn’t.’’ It is important to note that to have even these most basic
intuitions a person must have the appropriate concepts: you cannot identify an
9 And in 1994: 561–71. See also Kornblith 1998.
10 In calling the intuitions ‘‘empirical’’ I am claiming simply that they must be justiWed ‘‘by
experience’’. Should any justiWed belief be entirely innate, which I doubt, then beliefs of that sort
must have been justiWed somehow by the experiences (broadly construed) of our distant ancestors,
and we must have inherited that justiWcation via natural selection.
11 ‘‘intuition is the condensation of vast prior analytic experience; it is analysis compressed and
crystallized . . . . It is the product of analytic processes being condensed to such a degree that its
internal structure may elude even the person beneWting from it . . . ’’ (Goldberg 2005: 150)
F if you do not have the concept of an F. The richer intuitions go on to tell us
something about Fs already identiWed; for example, ‘‘Echidnas look like porcu-
pines’’. The richer ones may be much less dependable than the basic ones: a
person may be good at recognizing Fs without having much reliable to say about
them; this is very likely the situation of the folk with pains.
Identifying uncontroversial cases of Fs and non-Fs is only the Wrst stage of an
investigation into the nature of Fs: the second stage is to examine those cases to
see what is common and peculiar to Fs. Sometimes we have a well-established
theory to help with the Wrst stage; thus we had Mendelian genetics to identify the
genes that were examined by molecular genetics in the second stage. But
sometimes we do not have such help: we start pretty much from scratch; we
are at the stage of proto-science. At that stage, the most basic intuitions are
particularly important. In the absence of reliable theory, we must start by
consulting the people who are most expert about Fs to see what they identify as
Fs and non-Fs: we elicit their most basic intuitions about being an F in ‘‘iden-
tiWcation experiments’’. We are then in a position to begin our investigation.
Until recently, at least, this was our position with pains.
When we are starting from scratch, we need the basic intuitions, but we do not
need the richer ones. This is not to say that we should not use them. They may
well be a useful guide to what our investigation will discover about Fs; they are ‘‘a
source of empirical hypotheses’’ (Gopnik and Schwitzgebel 1998: 78).
We should trust a person’s intuitions, whether basic ones or richer ones, to the
degree that we have conWdence in her empirically based expertise about the kinds
under investigation. Sometimes the folk may be as expert as anyone: intuitions
laden with ‘‘folk theory’’ are the best we have to go on. Perhaps this is the case for
a range of psychological kinds. For most kinds, it clearly is not: we should trust
intuitions laden with established scientiWc theories. Consider, for example, a
paleontologist in the Weld searching for fossils. She sees a bit of white stone
sticking through grey rock, and thinks ‘‘a pig’s jawbone’’. This intuitive judgment
is quick and unreXective. She may be quite sure but unable to explain just how
she knows.12 We trust her judgment in a way that we would not trust folk
judgments because we know that it is the result of years of study and experience
of old bones; she has become a reliable indicator of the properties of fossils.
Similarly we trust the intuitions of the physicist over those of the folk about many
aspects of the physical world where the folk have proved notoriously unreliable.
And recent experiments have shown that we should have a similar attitude to
many psychological intuitions. Thus, the cognitive psychologist, Edward Wis-
niewski, points out that ‘‘researchers who study behavior and thought within an
experimental framework develop better intuitions about these phenomena than
12 I owe this nice example to Kim Sterelny. Gladwell 2005 has other nice examples: of art experts
correctly judging an allegedly sixth-century Greek marble statue to be a fake; of the tennis coach, Vic
Braden, correctly judging a serve to be a fault before the ball hits the ground.
Intuitions 105
those of the intuition researchers or lay people who do not study these phenom-
ena within such a framework. The intuitions are better in the sense that they are
more likely to be correct when subjected to experimental testing’’ (1998: 45).
Even where we are right to trust an intuition in the short run, nothing rests on
it in the long run. We can look for more direct evidence in scientiWc tests. In such
a scientiWc test we examine the reality the intuition is about; for example, we
examine the paleontologist’s white stone. These scientiWc examinations of reality,
not intuitions about reality, are the primary source of evidence. The examin-
ations may lead us to revise some of our initial intuitions. They will surely show
us that the intuitions are far from a complete account of the relevant bit of reality.
Intuitions often play a role in ‘‘thought experiments’’. Instead of real experi-
ments that confront the expert with phenomena and ask her whether they are Fs,
we confront her with descriptions of phenomena and ask her whether she would say
that they were Fs.13 These thought experiments provide valuable clues to what the
expert would identify as an F or a non-F. They can do more: the descriptions that
elicit the expert’s response indicate the richer intuitions that, as we have already
noted, can be a useful guide to the nature of Fs. Some experiments may be
diYcult, perhaps impossible, to perform other than in thought. Valuable and
useful as thought experiments may be in practice, they are dispensable in prin-
ciple: we can make do with real experiments. And thought experiments call on the
same empirically-based beliefs about the world as real experiments, and their
results have the same empirical status.
Aside. This account of thought experiments provides a naturalistic explanation
of the characteristic ‘‘armchair’’ method of philosophy. The traditional explan-
ation of this method is that philosophers are conducting thought experiments that
probe their concepts to yield a priori rational intuitions; they are doing ‘‘conceptual
analysis’’.14 The naturalistic explanation accepts that philosophers are conducting
thought experiments but construes these diVerently. The philosophers are not
probing concepts but rather intuitions about kinds. This is just as well because
knowledge of concepts, being a species of semantic knowledge, is very hard to
come by. In contrast, philosophers have acquired considerable knowledge of many
kinds over a lifetime of acquaintance with them. The philosophers’ intuitions that
draw on this knowledge, draw on these theories of the world, are not a priori
but empirical. The philosophers are conducting thought experiments of the sort
described in the last paragraph, counting themselves as experts about the kind in
question. Thus, in a famous example of the method, ‘‘the analysis of knowledge’’,
13 There are other things we might ask—for example, ‘‘What would happen?’’—but these are
beside our concerns. Gendler 2003 is a nice summary of views about thought experiments.
14 The best reason for being dubious of the traditional explanation is that we do not, I have
argued (1996, 1998b), have even the beginnings of an account of what a priori knowledge is. We are
simply told what it isn’t, namely empirical knowledge. Bealer (1998) and BonJour (1998) are
vigorous defenders of rational intuitions; see also Sosa 1998. For an exchange on the subject, see
BonJour 2005a, b, c and Devitt 2005a, b.
the philosopher, as expert as anyone in identifying cases of knowledge, confronts
descriptions of epistemic situations and considers whether the situations are cases
of knowledge. On the basis of these empirical intuitions about cases she constructs
an empirical theory about the nature of knowledge. The naturalist does not deny
armchair intuitions a role in philosophy but does deny that their role has to be
seen as a priori: the intuitions reXect an empirically based expertise in the
identiWcation of kinds.
The view I have presented of the limited and theory-laden role of intuitions
does not need to be modiWed for the situation where what we are investigating
are the products of a human skill or competence (which is the situation in the
philosophy of language and linguistics, of course). This situation would arise if we
were (for whatever reason) investigating the nature of horseshoes, chess moves,
touch typing, or thinking. Someone who has the relevant competence has ready
access to a great deal of data that are to be explained. She does not have to go out
and look for data because her competence produces them. Not only that, she is
surrounded by similarly competent people who also produce them. As a result, she
is in a good position to go in for some central-processor reXection upon the data
produced by herself and her associates. This reXection, often aided by appropriate
education, can yield concepts and a theory about the data. And it can yield the
capacity for sound intuitions, basic and richer, about the data. In brief, she can
become an expert. But this is not to say that she will become an expert. A person
can be competent and yet reXect little on the output of that competence. Or she
can reXect a lot but make little progress. Bicycle riders typically fall into one of
these two categories. It is a truism in sport that great players do not always make
great coaches. The fact that they possess a competence to a superlative degree
does not imply that they can articulate and communicate the elements of that
competence. Knowledge-how may not lead to knowledge-that. In brief, a person
competent in an activity may remain ignorant about it.15 And even if she does
become an expert, we should not assume that her opinions carry special authority
simply because she is competent; her competence does not give her Cartesian
access to the truth. She is privileged in her ready access to data, not in the
conclusions she draws from the data; conclusions of the competent, just like
those of the incompetent, are empirical responses to the phenomena and open
to question; they arise from the empirical observation of data.
Touch-typing provides a nice example of reXecting on the output of one’s own
competence. Ask a touch-typist whether a ‘k’ should be typed with a middle
Wnger and, very likely, he will think to himself, ‘‘How would I type a ‘k’?’’ He will
attend as he goes through the actual or mental motions of doing so and respond
immediately, ‘‘Yes’’. Consider also this report:
15 ‘‘Highly skilled performers are often unable to reXect on or talk about how they achieve their
skilled performance.’’ (Carlson 2003: 38)
Intuitions 107
If a skilled typist is asked to type the alphabet, he can do so in a few seconds and with very
low probability of error. If, however, he is given a diagram of his keyboard and asked to Wll
in the letters in alphabet order, he Wnds the task diYcult. It requires several minutes to
perform and the likelihood of error is high. Moreover, the typist often reports that he can
only obtain the visual location of some letters by trying to type the letter and then
determining where his Wnger would be. (Posner 1973: 25)16
The only privilege enjoyed by the typist’s judgment about which Wnger should be
used to type a ‘k’, or about where a letter is placed on the keyboard diagram, is the
privilege of being based on what is surely a good datum: on how he, a good
touch-typist, types.
Although these typist’s judgments are slow relative to his typing, they would
probably be fast enough for us to count them as intuitive. And they are likely to
be sound, for it is fairly easy to think about typing. Contrast this with thinking
about the outputs of another, much more important, human competence, the
competence to think, to move in a somewhat rational way from one thought to
another. We all have this competence to some degree or other. (Part IV discusses
thoughts.) Most of us reXect a bit on this and have some intuitions about what
follows from what. Still, these intuitions are likely to be sparse and many of them
are surely not sound. Thinking about thinking is so hard.
Now it is, of course, possible that the typist has somewhere in his mind a prior
representation of the keyboard which controls his typing and leads to his sound
judgment about how to type a ‘k’. But why believe this? Set aside (until section
11.5) whether we need to posit this representation to explain his typing. We
surely do not need the posit to explain his judgment. The much more modest
explanation I have just given, making do with cognitive states and processes that
we are already committed to, seems perfectly adequate for the job. Positing the
prior representation is explanatorily unnecessary. Applying Pylyshyn’s Razor
(3.1)—representations are not to be multiplied beyond necessity—we should
not make the posit. Finally, when we turn to the case of thinkers, such positing
would seem worse than unnecessary. The idea would have to be that the thinker’s
mind contains a representation of the ‘‘laws of thought’’ which controls her
thinking and which leads her to, say, the modus ponens intuition that ‘q’ follows
from ‘if p then q’ and ‘p’. But, as Lewis Carroll’s famous dialogue between
Achilles and the Tortoise demonstrates (1895), this view of thinking would lead
16 And consider this report (Sun et al. 2001). Subjects were placed in front of a computer with
the task of navigating a submarine through a mineWeld using sonar. After some episodes, ‘‘subjects
were asked to step through slow replays of selected episodes and to verbalize what they were thinking
during the episode’’ (p. 219). The experimenters sum up the results as follows: ‘‘The subject at Wrst
performed the task on an ‘‘instinctual’’ basis, without conscious awareness of any particular rules or
strategies. Gradually, through ‘doing it’ and then looking at the results, the subject was able to Wgure
out the action rules explicitly. The segment suggested implicit procedural learning at the bottom
level and the gradual explication of implicitly learned knowledge’’ (p. 226). (I discuss implicit
learning in section 11.5.)
to an inWnite regress.17 The modest explanation is the only plausible one: a
person’s thinking is governed by rules that she does not represent and her few
intuitive judgments about thinking are the result of reXecting on the perform-
ances of herself and others.
On the picture of intuitions I am presenting, what should we make of linguistic
intuitions? And whose intuitions should we most trust?
7.5 LINGUISTIC INTUITIONS
The focus in Coming to Our Senses is on the explanation of meanings. I claim that
the folk are as expert as anyone at identifying meanings, expressing the most basic
linguistic intuition of all. They do this in the ubiquitous practice of ascribing
thoughts and utterances to people; the folk say things like ‘‘Ruth believes that
Clinton is sexy’’ and ‘‘Adam said that Bush does not speak a natural language’’.
The ‘that’-clauses of these ascriptions specify meanings (or ‘‘contents’’). Because
these folk ascriptions are generally successful at serving their purposes—particu-
larly, the purposes of explaining behavior and guiding us to reality—we have
reason to think that they are generally true; see the argument for ‘‘inten-
tional realism’’ later (8.1). And note that although part of what is ascribed is a
meaning, no semantic term need be deployed: the ‘that’-clauses above do not
contain such terms and that is normal; a meaning is ascribed by using, in the
‘that’-clause, a sentence with that very meaning, or something close. Not only
are the folk good at identifying meanings but the poor state of semantic theory
gives no reason to think that the theorists will do signiWcantly better.
The situation is diVerent when it comes to deploying the vocabulary of the
philosophy of language and linguistics to make intuitive semantic and syntactic
judgments about an utterance that has the identiWed meaning. Let us now apply
the modest explanation of intuitions in general to the special case of these
linguistic intuitions.
The competent speaker has ready access to a great deal of linguistic data just as
the competent typist has to a great deal of typing data and the competent thinker
has to a great deal of thinking data: the competent speaker and her competent
fellows produce linguistic data day in and day out.18 So she is surrounded by
tokens that may, as a matter of fact, be grammatical, be ambiguous, have to co-
17 Suppose that the inference from the two premises, (a) ‘if p then q’, and (b) ‘p’, to the
conclusion, (z) ‘q’ had to be accompanied by a third premise, (c) ‘if (a) and (b) then (z)’, which
captures the ‘‘law of thought’’ that modus ponens is a good inference. Then in order to infer (z) from
these three premises, we would similarly need a fourth, (d) ‘if (a) and (b) and (c) then (z)’. And so
on ad inWnitum.
18 As Chomsky says, competent speakers ‘‘can easily construct masses of relevant data and in fact
are immersed in such data’’ (1988: 46).
Intuitions 109
refer with a certain noun phrase, and so on.19 So she is in a position to have well-
based opinions about language by reXecting on these tokens. This is not to say
that she will reXect. Indeed, a totally uneducated person may reXect very little
and hence have few if any intuitive judgments about her language. She may be
ignorant of her language.20 Still it is clear that the normal competent speaker
with even a little education does reXect on linguistic reality just as she reXects on
many other striking aspects of the world she lives in. And this education will
usually provide the terms and concepts of folk linguistics, at least. As a result she
is likely to be able to judge in a fairly immediate and unreXective way that a token
is grammatical, is ambiguous, does have to corefer with a certain noun phrase, and
so on. Such intuitive opinions are empirical central-processor responses to
linguistic phenomena.21 They have no special authority: although the speaker’s
competence gives her ready access to data it does not give her Cartesian access to
the truth about the data.22
So, on this modest account, how does a normal competent speaker make a
grammaticality judgment about a novel expression? As a result of education and
reXection, she already has the folk linguistic concept of grammaticality in her
language. And she appreciates the connection between this grammaticality and
competence in the language: roughly, errors aside, competent speakers produce
and understand grammatical sentences. She knows that she is a competent
speaker and so uses herself as a guide to what the competent speaker would do.
So she asks herself whether this expression is something she would say and what
she would make of it if someone else said it. Her answer is the datum. Clearly her
linguistic competence plays a central role in causing this datum about her
behavior. That is its contribution to the judgment that she must then go on to
19 This presupposes a realism about linguistic entities that, as noted in section 2.4, is curiously
denied by some. I discuss this denial in section 10.6.
20 This point is nicely illustrated by the following report: ‘‘As a graduate student I spent a
summer in the Pyrenees (Andorra, Perpignon, etc.) doing Weld research on the phonology of various
dialects of Catalan. Many of our native informants were illiterate peasants. I was forcefully struck
how diYcult it was to elicit linguistic judgments from them regarding their language, which of
course they spoke perfectly well. Just getting the plurals of certain nouns was tough. These folks
seemed to be very hard of hearing when it came to hearing the voice of competence! Their diYculty,
it seemed, was that their native language was largely transparent to them—they had never thought of
it as an object for observation and hence were largely unable to form even the most rudimentary
judgments about its character. Catalan speakers with only a modicum of grade school education, by
contrast, were good informants, presumably because they had learned through their grammar
lessons to think of language as an object with various properties, even if they had no sophisticated
knowledge of what those properties might be, theoretically speaking.’’ (Bob Matthews, in corres-
pondence)
21 Ilkka Niiniluoto urges a similar view: ‘‘Linguistic intuition is . . . largely observational know-
ledge about language’’ (1981: 182).
22 I emphasize that this is a modest explanation of the origins of a speaker’s intuitions about her
language. It is emphatically not an explanation of the origins of her linguistic competence and is
neutral about the extent to which that competence is innate.
make. She does some central-processor reXection upon the datum to decide
whether to apply her concept of grammaticality to the expression, just as she
might reXect upon any other relevant data supplied by the behavior of her fellow
speakers. If the datum shows that she would have no problem producing or
understanding the expression, she is likely to deem it grammatical. If the datum
shows that she has a problem, she will diagnose the problem in light of her
background theories, linguistic and others, perhaps judging the expression un-
grammatical, perhaps judging it grammatical but infelicitous or whatever. Often
these judgments will be immediate and unreXective enough to count as intu-
itions. Even when they do count, they are still laden with such background
theory as she acquired in getting her concept of grammaticality.
What goes for intuitions about grammaticality will obviously go for intuitions
about acceptability insofar as these are nothing but intuitions about grammat-
icality, insofar as ‘acceptable’ in the context is simply expressing the speaker’s
notion of grammaticality (and hence not expressing pragmatic notions like ones
about etiquette).23 And it will go for intuitions about ambiguity and coreference/
binding. Furthermore, we can often be conWdent that such intuitions of normal
educated speakers are right. We often have good reason to suppose that these core
judgments of folk linguistics, partly reXecting ‘‘the linguistic wisdom of the ages’’,
are good, though not of course infallible, evidence for linguistic theories.
Finally, what about intuitions elicited in ingenious ‘‘minimal-pair’’ experi-
ments?24 In these experiments, ordinary speakers are asked to say which of two
word strings is ‘‘worse’’. Since the two strings diVer only in that one fails a certain
hypothesized syntactic constraint, the experiments control for pragmatic factors.
So we can be fairly conWdent that these comparative judgments, in contrast to
simple judgments of acceptability (goodness), are responding only to grammat-
ical facts. And the judgments have another nice feature: they are as close to theory
free as one could get. Still they are lightly laden with a theory, even if only with a
theory constructed during the experiment (see Reply to Objection 2 in section
7.6); they are judgments of which string is worse grammatically. And the judg-
ments are likely to be right.
In sum, it is obvious that a speaker’s own linguistic competence plays some
role in the intuitive judgments she makes about the grammatical properties of
expressions in her language. On the received linguistic view, the competence
supplies information about those properties. On the modest view I am urging, it
supplies behavioral data for a central-processor judgment about those properties. In
particular, the grammatical (sometimes partly grammatical) notions that feature
23 Much the same will go also for acceptability intuitions that are not of this sort and are partly
pragmatic: they are central-processor responses to the data, laden with pragmatic theories as well as a
linguistic one (7.3). And these intuitions may well be reliable, albeit not nearly as useful because of
the pragmatic intrusion.
24 See Crain and Thornton 1998 for a helpful discussion of experiments of this sort.
Intuitions 111
in these judgments are not supplied by the competence but by the central
processor as a result of thought about language. Similarly, the notion of following
from that may feature in intuitive judgments about thinking is not supplied by
the competence to think but by the central processor as a result of thought about
thinking. In neither case does competence have a voice.
Although the intuitions discussed are likely to be right, the intuitions that
linguistics should mostly rely on are those of the linguists themselves because the
linguists are the most expert. This is particularly so when we get beyond the
simple cases to theoretically interesting ones like ‘The horse raced past the barn
fell’ and ‘Who do you wanna kiss you this time?’ The linguists’ skill at identifying
items with and without a syntactic property like, say, the biologists’ skill at
identifying items with and without a biological property, is likely to be better
than the folk’s because their theories are better. Thus linguists have Wrm, and
surely correct, intuitions about the acceptability of many sentences, and about
some matters of co-reference, that the folk do not.25 Linguistic theory is, as
linguists are fond of pointing out, in good shape, far better shape than semantic
theory. As a result of their incessant observation of language, guided by a good
theory, linguists are reliable indicators of syntactic reality; analogously, biologists
are reliable indicators of biological reality. So it is appropriate that linguists do
tend to rely on the intuitions of other linguists, as we have already noted (7.2).26
To say that intuitions, whether those of the linguists or the folk, are good
evidence is not to say that they are the only, or even the primary, evidence.
Indeed, we can look for more direct, less theory-laden, evidence by studying what
the intuitions are about, the linguistic reality itself. In fact, there are many other
sources of evidence, as I have pointed out in section 7.2. If this is right, theory
construction in linguistics could proceed without any appeal to intuitions at all.
This is not to say, of course, that it should so proceed. I have accepted that these
intuitions, particularly those of the linguists, are often good evidence. So, they
should be used. Intuitions are often a very convenient shortcut in theorizing.
It is time to compare my modest explanation of the evidential role of linguistic
intuitions with the standard linguistic one and hence to assess the abduction to
RT.
25 Subjects in an experiment (Spencer 1973) considered 150 sentences that linguists had
categorized as clearly acceptable or unacceptable. The subjects disagreed with the linguists over 73
of these, either Wnding them unclear or giving them an opposite categorization. In another
experiment (Gordon and Hendrick 1997), naive subjects found co-reference between a name and
a pronoun that preceded it unacceptable even where the pronoun did not c-command the name.
This is one of several experiments where folk intuitions were discovered to be at odds with the
linguists’ and with Binding Theory.
26 Still, as my student Francesco Pupa has pointed out to me, we might prefer the intuitions of
the folk to those of the linguists about ‘‘peripheral’’ rather than ‘‘core’’ language (see Chomksy 1986,
p. 147, on this distinction). Pupa used the example, ‘‘How come you left?’’ A linguist may Wnd this
sentence intuitively ungrammatical because it oVends against the rule of Subject–Auxiliary Inversion
for wh-questions. Yet it is clearly acceptable as any member of the English-speaking folk will insist.
7 . 6 C O M PA R I S O N O F T H E M O D E S T E X P L A N AT I O N
W I T H T H E S TA N D A R D C A RT E S I A N E X P L A N AT I O N
I shall start by replying to two objections to the modest explanation. I shall then
give some considerations against the standard one. I shall conclude by claiming
that the modest one is better.
Objection 1.27 Intuitions about touch-typing, thinking, and the like are not
the right analogies for linguistics intuitions. Rather, the right analogy is with
intuitions about perceptual experiences, for example, the intuitions aired in
illusory situations: ‘‘It looks like there is water on the road’’ when experiencing
a mirage; ‘‘The moon looks larger when it is close to the horizon’’ when
experiencing the moon illusion; and so on. These intuitions are immediately
based on the outputs of a module (Fodor 1983) and throw an interesting light on
the nature of that module. They are not covered by the modest theory. No more
are linguistic intuitions.
Reply. (i) Perceptual judgments are not good analogues of linguistic intuitions.
Consider the visual module. Its task is to deliver information to the central
processor of what is seen,28 information that is indeed the immediate and main
basis for judging what is seen; our earlier ‘‘That grass is brown’’, ‘‘That person is
angry’’, ‘‘This is an echidna but that isn’t’’, and ‘‘a pig’s jawbone’’ are examples of
such judgments. Sometimes what is delivered is misinformation; for example,
‘‘There is water on the road’’ when experiencing a mirage. Locutions like ‘looks
like’ enable us to allow for the possibility of misinformation in reporting these
deliverances, as Objection 1 illustrates. The language module has the tasks of
language production and comprehension. The task of production is clearly not
analogous to the task of the visual module but the task of comprehension is: it is
to deliver information to the central processor of what is said, information that is
the immediate and main basis for judging what is said, for judging ‘‘the mes-
sage’’.29 So, intuitions about what the message is are analogous to intuitions
about what is seen. But the former intuitions are not the ones that concern us:
for, they are not intuitions about the syntactic and semantic properties of expressions.
(The contrast is between the intuition that the message is that Tom thinks that
John loves himself and the intuition that in ‘Tom thinks Dick loves himself ’
‘himself ’ must refer to the same person as ‘Dick’.) If the objection is to be
eVective, it would have to be the case that the language module also delivers
27 An objection along these lines has been pressed on me vigorously by Georges Rey.
28 As Fodor says, ‘‘information about the ‘layout’ . . . of distal stimuli’’ (1983: 45). He later
speculates that it ‘‘delivers basic categorizations’’ (p. 97), categories like dog rather than poodle or
animal.
29 Some ‘‘pragmatic’’ abilities supply bases too, of course, determining the reference of indexicals,
removing ambiguities, making Gricean derivations, and so on; see sections 8.3 and 11.8 for more on
this.
Intuitions 113
information of those syntactic and semantic properties, information that would
be the immediate basis for the intuitions that concern us. But the view that the
module does deliver this information is, of course, precisely what is at issue and
so it needs an independent argument. Indeed if the language module did deliver
this information it would be disanalogous to a perceptual module, as we shall
soon see (Further Consideration 1). (ii) The outputs of the language module do
indeed throw an interesting light on the nature of that module; consider, for
example, the signiWcance of garden-path phenomena in comprehension and slips
of the tongue in production. But these phenomena are examples of language
usage; they are not intuitions about the linguistic properties of the expressions
that result from that usage. (iii) Although perceptual judgments are not good
analogues of linguistic intuitions they are covered by the modest theory. I see
them as paradigms of fairly immediate and unreXective empirical responses to
phenomena, as my discussion showed (7.4).
Objection 2.30 The claim that the intuitions of the ordinary competent
speaker are empirical observations deploying theory-laden linguistic vocabulary
is at odds with the following phenomena. (i) Take an English speaker with near
enough no education in linguistics and give her two lists, one of twenty clearly
grammatical sentences of English and the other of twenty clearly ungrammatical
word strings. The Wrst group is labeled ‘A’ and the second ‘B ’. Now give her a set
of strings, some grammatical some not and ask her to classify each one as an A or
a B. She is likely to classify them near enough perfectly. (ii) Next we give her a list
of twenty pairs of sentences that are related as active to passive and a list of twenty
pairs of sentences that are not so related. We tell her that pairs in the Wrst group
are related as X to Y, those in the second are not. We give her many more pairs,
some active–passive related, some not and ask her whether they are related as X to
Y. Once again she is likely to classify them near enough perfectly. These intuitive
judgments are not laden with any theory and deploy no linguistic vocabulary.31
Reply. What happens in these experiments is that the speaker either learns to
use the terms ‘A’ and ‘B ’ for her concepts <grammatical-in-English> and
<ungrammatical-in-English>, and the terms ‘X ’ and ‘Y ’ for her concepts
<active> and <passive>; or, more likely, she acquires these linguistic concepts
30 Based on an objection Stephen Stich made in correspondence on Devitt and Sterelny 1989.
31 I assume that Fodor has something like experiments (i) and (ii) in mind in claiming:
Normal human children are, as far as we know, quite extraordinarily good at answering questions of
the form: ‘‘What grammar underlies the language of which the following corpus is a sample (insert
PLD here)?’’
Yet, he claims, the children ‘‘exhibit no corresponding capacity for answering questions about
bagels’’ (2001: 129). I would have thought they would do just as well in experimental analogues of
(i) and (ii) for bagels. (The questions about both language and bagels would surely have to be much
less sophisticated than Fodor’s.)
while she is learning to use the terms for them.32 These experiments are
analogous to the classic concept learning experiments in psychology and to
solving many puzzles in IQ tests. It is not part of my modest view of intuitive
linguistic judgments that an ignorant person cannot easily learn to make them.
Indeed, most of us do easily learn to make many of them in primary school.
These judgments are typically about basic linguistic facts that are very epistem-
ically accessible. The judgments are theory-laden, but probably not much more
so than many ‘observation’ judgments; for example, ‘Grass is green’; ‘Rocks are
hard’; ‘Elephants are bigger than mice’. Once one has acquired the necessary
concepts, these judgments are easy to make; and the concepts are easy to acquire.
On the Duhem–Quine Thesis, all judgments are theory-laden, but they are not
all laden to the same degree.
So the modest explanation is still in good shape. Turn now to the standard one.
According to the standard Cartesian explanation, the intuitions of someone
competent in a language are good evidence for a theory of that language because
they are derived from her representation of the rules of the language in her
language faculty. The explanation is appealing because the idea of one sort of
representation leading to another is familiar. Still, attention to the language in
which competence allegedly speaks and to how little it says raised concern about
the details of the causal-rational route from an unconscious representation in the
language faculty to a conscious judgment in the central processor (7.3). I shall
now give three further considerations against the standard explanation.
Further Consideration 1.33 According to the standard explanation, the lan-
guage module delivers syntactic and semantic information about expressions to
the central processor. If it did this it would be disanalogous to perceptual
modules (as noted in Reply to Objection 1). For, if it did, the central processor
would have direct access to information that the language module allegedly uses
to fulWll its task of processing language.34 But nobody supposes that the central
processor has direct access to analogous information used by perceptual modules
to fulWll their processing tasks.35 Thus, the visual module simply tells the central
processor what is seen: something along the lines of brown grass, angry person,
an echidna, a pig’s jawbone, water on the road, and so on. It does not deliver
whatever information the module may use to arrive at such conclusions; it does
not deliver ‘‘Marr’s ‘primal’, ‘2.5 D’, and ‘3 D’ sketch’’ (Fodor 1983: 94).
32 Similarly, I would say, subjects in ‘‘artiWcial grammar (AG)’’ tasks in psychology (Reber 1967)
acquire the concept <grammatical-in-AL> where AL is the artiWcial language.
33 I am indebted to my student, David Pereplyotchik, for this point.
34 I later raise doubts that language processing typically involves this information in the sense of
involving representations of syntactic and semantic information (11.6).
35 ‘‘central processors should have free access only to the outputs of perceptual processes,
interlevels of perceptual processing being correspondingly opaque to higher cognitive systems.’’
(Fodor 1983: 60)
Intuitions 115
Further Consideration 2. I have noted in the last section that the intuitions of
linguists often diVer from those of the folk. This should be an embarrassment for
the oYcial line. First, why would the intuitions be diVerent if they were the voices
of competencies in the one language? An easy answer would be that, to the extent
of the diVerences in intuitions, the linguists do not speak the same language as
the folk. But this is very implausible. It is also belied by evidence, both anecdotal
and experimental, that linguistic intuitions change with a linguistic education.36
It is odd that this education should interfere with the causal–rational process by
which intuitions are allegedly derived from the underlying representation of
linguistic rules. Second, and more serious, from the Cartesian perspective it
seems that we should see this interference as the contamination of the pure
voice of competence with theoretical bias. Therefore, rather than relying on
this contaminated evidence linguists should be seeking the intuitions of the
most uneducated folk. So, the actual practice of linguists is mistaken. In contrast,
if my proposal about the place of intuitions is correct, the change of linguistic
intuitions with education is just what we should expect. And the actual practice
of linguists is Wne. The educated intuitions are contaminated only in the way that
all evidence is and must be: it is all theory-laden. Linguistic education should
make a person a better indicator of linguistic reality just as biological education
makes a person a better indicator of biological reality. Of course a person
educated into a false theory may end up with distorted intuitions.37 But that is
an unavoidable risk of epistemic life, in linguistics as everywhere else. We have no
unsullied access to any reality.
Further Consideration 3. If a speaker represents the rules of her language then
that representation must surely control her language use, whether or not her
linguistic intuitions are derived from the representation. And, of course, the
received linguistic view is that the representation does both control use and yield
intuitions. Yet there is persuasive evidence that it does not play both these roles.
If it does not then the intuitions clearly cannot be derived from the representa-
tion (because if they were the representation would have to play both roles). The
evidence is to be found in the study of ‘‘implicit learning’’, learning that takes
place ‘‘largely without awareness of either the process or the products of learning’’
(Reber 2003: 486). I shall later suggest, contrary to a common view in linguistics,
language seems to be a paradigm of such learning (11.5). Suppose that it is. Now,
if implicit learning were largely a matter of acquiring representations of rules that
both govern the performance of a task and yield intuitions about the task, we
would expect improvement in performance to be matched by improvement in
36 In one experiment cited in note 25, subjects with at least one course in generative grammar
agreed more with the linguists than did ‘‘naive’’ subjects (Spencer 1973). In another, subjects who
were encouraged to reXect on a sentence rather than give an immediate reaction agreed more with
the linguists (Gordon and Hendrick 1997). Consider also Matthews’ report in note 20.
37 This is not to say that there is no limit to the distortions that education can bring. Here, as
everywhere else, reality constrains theories and hence distortions.
intuitions. Yet that is not what we Wnd at all: improvement in task performance is
dissociated from improvement in the capacity to verbalize about the task. Con-
sider this summary by Broadbent et al. 1986:
Broadbent (1977) showed that people controlling a model of a city transportation system
gave more correct decisions when they had practiced the task than when they had not.
However, they did not improve in their ability to answer questions about the relationships
within the system. Broadbent & Aston (1978) found that teams of managers taking
decisions on a model of the British economy showed a similar improvement in perform-
ance after practice. Yet the individuals making up the team did not improve on multiple
choice questions about the principles governing the economic model. (p. 34)
The paper reports further experiments that conWrm these results (and go beyond
them). Thus, one experiment found ‘‘no increase in verbalizable knowledge
associated with [a] sudden increase in performance’’ (Stanly et al. 1989: 569).38
So there is good evidence for thinking that what we implicitly learn in acquiring a
language is not a representation of the rules from which we derive our intuitions.
So we have good evidence against the standard explanation. Of course, these
experiments leave open the possibility that, nonetheless, what we learn is a
representation of the rules that controls our language use (without yielding the
intuitions), a possibility I shall consider in Chapter 11.
None of these considerations against the standard Cartesian explanation is
decisive, of course. Still, taken together, they do seem to me to undermine its
plausibility considerably. Furthermore, I think that these considerations together
with the case for the modest explanation are suYcient to establish that the
modest explanation is better than the standard one.
But there is one more important reason for preferring the modest explanation.
It arises from the extreme immodesty of the standard explanation. The standard
explanation rests on RT, a very powerful assumption about the mind. And the
important reason is this: we do not need this powerful assumption to explain the
reliability of linguistic intuitions. We can explain that reliability without positing
representations of linguistic rules in the minds of competent speakers. Consider
the analogous phenomena for typing and thinking. We can explain the reliability
of intuitions about those processes without positing representations of the rules
that govern the processes. Our explanations of these intuitions make do with
cognitive states and processes that we are already committed to. These modest
explanations seemed perfectly adequate for the job and, indeed, much more
plausible than their representational rivals. Similarly, I am urging, the modest
38 The evidence does not, of course, show a total dissociation of verbalizable knowledge and
performance, just a far greater one than would be expected if the knowledge and the performance
stemmed from the one representation of rules. Less dissociation was found in some other experi-
ments (Mathews et al. 1988) in which the verbalizations were largely descriptions of prior sequences
of events in performing the task rather than expressions of knowledge of the rules governing the task.
See also Mathews et al. 1989.
Intuitions 117
explanation in the linguistics case. Language is a very striking and important part
of the human environment. It is not surprising that empirical reXection on
linguistic data, aided by some education, should make people fairly reliable
detectors of the most obvious facts about language. We are surely similarly reliable
about other striking and important parts of the environment, for example,
the physical, biological, and psychological parts. If we can explain the reliabil-
ity without positing representations of the rules, Pylyshyn’s Razor says we
should do so.
In conclusion, the modest explanation is better than the standard linguistic
one. So the standard one cannot be the basis for a successful abduction to RT.
The evidential role of linguistic intuitions does not support RT.
7.7 A N O N S TA N D A R D C A RT E S I A N E X P L A N AT I O N O F
T H E RO L E O F I N T U I T I O N S ?
The standard explanation that I have rejected relies on RT, a thesis that would
support position (i) or (ii) on the psychological reality of language: the repre-
sented rules are applied or used as data in language processing (3.4). Because of
that reliance, I allowed that some linguists may not endorse this explanation
(7.1). Some may believe that the syntactic rules of a language are embodied
somehow in its competent speakers without being represented, and govern
processing: position (iii) on psychological reality. Yet they are still committed
to the Cartesian view that linguistic intuitions are the voice of competence, the
view that competence alone provides information about the linguistic facts. How
could this be so? Can we Wnd what I called a ‘‘nonstandard Cartesian explan-
ation’’ of the evidential role of intuitions (and hence the basis for an abduction to
the thesis that the rules are embodied without being represented)? I know of no
such explanation and I don’t think that any one will be forthcoming.
It helps to note Wrst that even if (iii) were right and the syntactic rules were
embodied without being represented this would make no contribution to the
modest explanation. According to this explanation linguistic intuitions arise from
mostly reliable central-processor reXection on linguistic data. If the rules of the
language were embodied and governed language processing they would have a
role in producing the data that are thus reXected upon (abstracting from
performance error). But this would not be a contribution to the explanation.
There must, after all, be some embodied processing rules that produce the data;
that’s a consequence of the minimal position (m) that a competence in a language
must respect its structure rules (3.4). But, so far as the modest explanation is
concerned, it does not matter what rules produce the data. All that matters to the
explanation is that the data are the product of competent speakers, whatever their
competence consists in and however those data are produced. One might claim,
of course, that only by embodying the structure rules could a speaker produce
data that is governed by those rules. And if that claim were right it would clearly
support position (iii). But it remains to be argued that it is right (Ch. 2). More to
the point, even if this was successfully argued, the support for (iii) would come
from that argument not from the modest explanation of intuitions. That ex-
planation does not depend on any assumptions about what produces the data
beyond that they are produced by a competent speaker. It does not require that
the psychological processing rules involve the syntactic rules of the language.
Any nonstandard Cartesian explanation must of course be diVerent from
the modest one. To be diVerent it must give the embodied but unrepresented
rules a role in linguistic intuitions other than simply producing data for central-
processor reXection. And it must do this in a way that explains the Cartesian view
that speakers have privileged access to linguistic facts. It is hard to see what shape
such an explanation could take.39 The explanation would require a relatively
direct cognitive path from the embodied rules of the language to beliefs about
expressions of that language, a path that does not go via central-processor
reXection on the data. What could that path be? The earlier Further Consider-
ation 1 (7.6) comes into play again. Perceptual modules may well be governed by
embodied but unrepresented rules. And the operation of those rules may yield
information that guides the module in arriving at its message to the central
processor about what is perceived. Yet the central processor has direct access only
to the message, not to any intermediate information involved in arriving at it.
Why suppose that the language module is any diVerent? Consider some other
examples. It is very likely that rules that are embodied but not represented govern
our swimming, bicycle riding, catching, typing, and thinking. Yet there does not
seem to be any direct path from these rules to relevant beliefs. Why suppose that
there is such a path for linguistic beliefs? Why suppose that we can have
privileged access to linguistic facts when we cannot to facts about these other
activities? We do not have the beginnings of a positive answer to these questions
and it seems unlikely that the future will bring answers. Even if we could answer
the questions and come up with the required explanation, we would still need a
persuasive reason to prefer that explanation to my modest one if the abduction to
position (iii) is to be good.
The standard Cartesian explanation of the evidential role of linguistic intu-
itions rests on RT. I have earlier produced reasons for rejecting it in favor of the
modest explanation. So the abduction to RT fails. A nonstandard Cartesian
explanation will rest on the idea that the rules of the language are embodied
without being represented, (iii). I know of no such explanation. I have just given
some reasons for thinking that there is unlikely to be one. So the abduction to
(iii) fails. And we have found no support for the idea that linguistic intuitions
39 Graves et al. dismiss the possibility of such an explanation (1973: 326–7).

Intuitions 119
reXect information supplied by linguistic competence, no support for the idea
that they are the voice of competence.
7.8 MUST LINGUISTICS EXPLAIN INTUITIONS?
The Cartesian view of a speaker’s intuitive judgments leads to the view that
linguistics should explain speakers’ intuitive judgments: ‘‘If a theory of language
failed to account for these judgments, it would plainly be a failure’’ (Chomsky
1986: 37).40 In thinking about this we need to make a distinction.
The distinction is between what the intuitions express and the fact that speakers
have these intuitions (Devitt and Sterelny 1989: 520–1). Now, it is obvious that
linguistic theory must explain linguistic facts. Insofar as intuitions are right, and
linguistically relevant, they express linguistic facts and so the theory must explain
what they express. But insofar as intuitions are wrong or irrelevant, the theory has
no concern with what they express and no need to explain them.41 Linguists
assume that the intuitions are largely right and relevant. I have agreed. It follows
that linguistic theory must indeed explain what these intuitions express.
What about the explanation of how speakers’ come to have the intuitions? It is
no more the concern of linguistics to explain this than it is the concern of biology
to explain how folk come to have their biological intuitions or physics to explain
how folk come to have their physical intuitions. These explanations may well be
worthwhile and interesting but they would be part of descriptive epistemology (and
hence part of psychology). And, if an intuition is right, there is no special
epistemological interest in explaining a person’s having it: we expect the folk to
be fairly reliable detectors of facts about their environment, particularly where
the facts are rather obvious. (The need for an explanation is greater when people
have false intuitions; for example, religious ones.) In this chapter, I have oVered
the beginnings of an epistemological explanation of linguistic intuitions along
these lines.
Although the explanation of speakers having correct linguistic intuitions is
part of epistemology not linguistics, it may be epistemically relevant to linguistics.
For, the abduction that we have considered (but rejected) Wnds evidence for RT
in an explanation of speakers’ having those intuitions. But then it is no surprise to
40 See also, Lees 1957: 36; Chomsky 1969a: 81–2; Baker 1978: 4–5; Dwyer and Pietroski 1996:
340. Consider also these analogous claim about semantics: ‘‘accounting for our ordinary judgments
about truth-conditions of various sentences is the central aim of semantics’’ (Stanley and Szabo
2000: 240); ‘‘Our intuitive judgments about what A meant, said, and implied, and judgments about
whether what A said was true or false in speciWed situations constitute the primary data for a theory
of interpretation, the data it is the theory’s business to explain.’’ (Neale 2004: 79)
41 This is not to say, of course, that the theory has no need to explain performance errors like the
failure to parse centrally embedded relatives. False intuitions about a language are one thing, errors
in using it, another.
Wnd that an epistemological view can be epistemically relevant to linguistics.

Anything can be epistemically relevant to linguistics. That is a consequence of the
Duhem–Quine thesis.
Linguists greatly exaggerate the evidential role of the intuitive judgments of

ordinary speakers. Still these intuitions are good evidence for a grammar. Why
are they? The Chomskian answer is that they are derived by a rational process
from a representation of linguistic rules in the language faculty. I have argued for
a diVerent view that has the great advantage of being theoretically modest.
Linguistic intuitions do not reXect information supplied by represented, or
even unrepresented, rules in the language faculty. Linguistic competence supplies
data for these intuitions but the intuitions are not its voice. Rather, linguistic
intuitions are like intuitions in general. I have arrived at my third major conclu-
sion: Speakers’ linguistic intuitions do not reXect information supplied by the
language faculty. They are immediate and fairly unreXective empirical central-
processor responses to linguistic phenomena. They are not the main evidence for
grammars. This conclusion accommodates the evidential role that intuitions play
in linguistics without any appeal to embodied rules of the language.
Earlier we found no support for RT in the rejection of behaviorism nor in folk
psychology. No support is to be found either in the evidential role of linguistic
intuitions. So, I am on my way to establishing my second major conclusion: there
is no signiWcant evidence for RT and, given what else we know, it is implausible.
Furthermore, the evidential role of linguistic intuitions does not support any
robust thesis about the psychological reality of the linguistic rules: a competent
person has ready access to the output of her competence whatever rules produce
that output. So the role does not support positions (i) or (ii) according to which
the represented rules are applied or used as data in language processing, nor
position (iii) according to which the rules govern by being simply embodied
(3.4). If there is to be any support for these positions on psychological reality, it
must be found elsewhere.
In what way, if at all, is person’s language psychologically real in her mind?
That is the main concern of this book. According to the argument in Part I,
linguistics is not primarily about the mind—Wrst major conclusion—and so the
idea that the language is psychologically real does not ‘‘come for nothing’’: it
needs, as my second methodological point emphasized (2.6), a powerful psycho-
logical assumption to establish it. In the next part, I shall argue for such an
assumption, position (t): a language is largely psychologically real in a speaker in
that its rules are similar to the structure rules of her thought. In general, I shall
urge that the psychological reality issue is best approached from a perspective on
thought. Finally, in Part V, against the background of this discussion of thought,
Intuitions 121
I shall consider the evidence from language use and language acquisition. This
will provide the main case for my second major conclusion.
All named or numbered major conclusions, methodological points, and
positions on psychological reality are listed in the Glossary.
PA RT IV
T H E R E L AT I O N O F L A N G UAGE
TO THOUGHT
8
Thought Before Language
8.1 INTENTIONAL REALISM
My aim in this chapter is to argue for various ways in which thought is prior to
language. The assumptions of this argument are, it seems to me, relatively uncon-
troversial and yet their repercussions for the issue of the psychological reality of
language are both considerable and strangely unnoticed.
This discussion will lay the groundwork for Chapter 9 which will present an
argument for the bold position (t): the structure rules of a speaker’s language are
similar to the structure rules of her thought. The assumptions of that argument,
particularly the Language-of-Thought Hypothesis that thoughts involve lan-
guage-like mental representations (LOTH), are certainly not uncontroversial.
The discussion in these two chapters, particularly the priority claims, seems
in sharp conXict with Chomsky’s commitment to thought and language being
largely independent of one another, with language residing in a distinct language
faculty. I shall set aside addressing this apparent conXict until Chapter 10.
Discussions of the relation between thought and language is often framed
around a distinction between the ‘‘communicative’’ and the ‘‘cognitive’’ concep-
tions of language; see many papers in the helpful collection, Language and
Thought (Carruthers and Boucher 1998a). I do not follow this practice for reasons
that I will make clear in section 9.6.
What are thoughts? In the sense that concerns us, the Wrst step in an answer takes
them to be propositional attitudes, mental states like beliefs, desires, hopes, and
wondering whethers. Further steps in the answer will follow, particularly in Chapter 9.
Clearly, the view that thought is in some way prior to language is committed to
there really being thoughts; it is committed to ‘‘intentional realism’’. This realism
is my Wrst assumption. It is relatively uncontroversial but still some people reject
it: they are eliminativists, regarding thoughts as posits of a discredited folk theory.
Famously, this was the attitude of the behaviorists. More recently, it has been the
attitude of Patricia Churchland (1986) and Paul Churchland (1981, 1989), and
some others drawn to connectionism.1 Many philosophers have responded to
1 I think that Donald Davidson (1980, 1984), under the inXuence of Quinean behaviorism, is at
bottom an antirealist about the mind. However, his views are very hard to discern. I shall not be
discussing Davidson’s views on mind and meaning here. I have argued against them at length
elsewhere (1981: ch. 4; 1996: 66–7, 139–40; 1997a: ch. 10; Devitt and Sterelny 1999: ch. 15).
126 Relation of Language to Thought
this eliminativism with ‘‘transcendental arguments’’ to show that it is ‘‘incoher-
ent’’, ‘‘unstable’’, and so forth. I think that such arguments are very misguided.2
Still, I think that we have a very good reason for supposing that we do indeed
have thoughts.
Begin by considering why we ascribe thoughts. Doubtless we have many
purposes, but I think we should focus on two very important ones:3 Wrst, we
ascribe thoughts to explain and predict the behavior of the subject; and, second, to
use the thoughts and utterances of a subject as guides to a reality largely external
to the subject. I shall consider these in turn.
1. Consider this explanation of behavior:
Why did Granny board the bus? She wants to buy a bottle of scotch. She believes that she
has her pension check in her pocket. She believes that the bus goes to her favorite liquor
store.
Such ‘‘intentional’’ explanations of ‘‘intentional’’ behavior are familiar and cen-
tral parts of ordinary life, of history, of economics, and of the social sciences in
general. They all ascribe thoughts. With the decline of behaviorism, psycho-
logical explanations along these lines are even oVered for much animal behavior;
for example, a rat’s maze running.
2. Ascribing beliefs serves another remarkably valuable purpose. If a person
believes that the world is such and such, and if the person is reliable, then we
have good reason to believe that the world is such and such. Thus, attributing the
property of meaning IT IS RAINING to Mark’s belief not only helps to explain
his rain-avoidance behavior—putting on a coat and picking up an umbrella—
but also gives us evidence about the weather. We can even learn from someone
who is a reliable guide to the way some area of the world is not.
We have a wide range of interests in learning about the world. The direct way
to serve these interests is to examine the world. The indirect way is to use reliable
indicators. Sometimes these indicators are ‘‘natural’’ ones like tree rings. Some-
times they are artifacts like thermometers. Very often they are the beliefs of
others. Some belief ascriptions serve our theoretical interest in explanation. Many,
however, are like ascriptions of desires, hopes, and so on in serving interests that
are not really theoretical at all. We have the most immediate practical interest in
Wnding out quite humdrum facts about the world to satisfy our needs for food,
shelter, a mate, and so on. So it helps to know what is on sale at the supermarket,
2 For examples of transcendental arguments, see Baker 1987: 113–48; Malcolm 1968; Gasper
1986. Barbara Hannan points out how common such arguments are. She discusses one sympathet-
ically but does not endorse it (1993: 171–2). I discuss such arguments brieXy in my 1996, section
4.2, drawing on my 1990c, which is mostly a critique of an extremely complicated transcendentalist
argument in Boghossian 1990a. Boghossian 1990b is a response to the critique. Devitt and Rey
1991 is a response to that response. Taylor 1994 is a nice critique of transcendental arguments. Such
arguments Wnd sympathy in surprising places; see e.g., Fodor and Lepore 1992: 207.
3 I draw on my 1996: 2.5.
Thought Before Language 127
where there is a hotel, who is available, and so on. Ascribing beliefs is a very good
way of Wnding out about anything at all.
We note next that this practice of ascribing thoughts is generally successful at
serving these two purposes. Day in and day out we explain people’s behaviors
with these ascriptions. Almost everything we know about the world—what we
learn at mother’s knee, in classrooms, and from books—we get from ascribing
beliefs to people and assessing them for reliability. If there really were not any
thoughts, this success would be very hard to explain.
That is an argument for thoughts and intentional realism. There are, of course,
arguments against thoughts and for eliminativism. These arguments do not strike
me as persuasive for reasons I have set out elsewhere (1996: 5.3).
So, what is a thought? It is a mental state with a meaning or content (henceforth
I will usually talk simply of meanings). There must be more to a thought than its
meaning, of course: we must distinguish believing that s from hoping that s,
thoughts that have the same meanings but are nonetheless diVerent. But we need
not be concerned with the extra.
8.2 L A N G U AG E E X P R E S S E S T H O U G H T
Intentional realism is my Wrst relatively uncontroversial assumption. The next is

the folk view that ‘‘language expresses thought’’ (‘‘LET’’). As Fodor, Bever, and
Garrett say, ‘‘there is much to be said for the old-fashioned view that speech
expresses thought, and very little to be said against it’’ (1974: 375). Although
relatively uncontroversial, this view has, or should have, signiWcant repercussions
on our view of language. It demands that we give a prominent place to thoughts
in theorizing about the place of language in the mind and it undermines the view
that language and thought are independent.
Once intentional realism is accepted, LET becomes irresistible. LET is
reXected in intentional explanations of linguistic behavior. Consider an example:
Why did Granny say ‘‘I need a drink’’? She believes that she needs a drink. She
wants to express her belief to her audience. In her language ‘I need a drink’ expresses
that belief.
What does LET amount to? Roughly, it is the view that language production is a
matter of uttering a sentence of the language to express a thought with the
meaning that the sentence has in that language: the sentence conveys a ‘‘message’’
that is the meaning of the underlying thought. Related to this is the view that
language comprehension is a matter of assigning to a sentence a thought with the
meaning that the sentence has in the language. The utterance’s message is thus
received. Once one has accepted the reality of thoughts and noted that they are
essentially things with meanings, it is hard to see any alternative view of the
relation between meaningful thoughts and meaningful language.
This is rough in two important ways. (1) The meaning of a sentence that has to
be matched with the thought’s meaning is its meaning in the context of utterance.
Thus suppose that my observation of a particular person leads me to a thought
with the meaning HE IS ON FIRE and I then express this thought by uttering the
sentence ‘He is on Wre.’ Because this sentence contains the indexical, ‘he’, its
meaning in English alone does not match the meaning of my thought: we must
add in that my use of ‘he’ refers to the person I am observing. What matches the
thought for meaning is not the sentence type but the sentence token. That token
gets its meaning partly from the meaning of the sentence type in English and
partly from the token’s context of utterance.4 (2) Sometimes the meaning (in the
context) of the sentence that a person uses to express her thought is not the
meaning of her thought, it is not what she means by uttering the sentence. In the
terms of the distinction that Paul Grice (1989) has made famous, her ‘‘speaker
meaning’’ is not the literal or conventional meaning of the utterance.5 Metaphors
provide clear examples of this distinction. Thus I may mean by my utterance of
‘He is on Wre’ not that the person is literally on Wre but that he is ebullient. Grice’s
distinction is important, and I shall return to it soon (8.4), but we can aVord to
set it aside as we move to our Wrst two priority claims.
8 . 3 O N TO LO G I C A L P R I O R I T Y O F
C O N C E P T U A L C O M PE T E N C E
I have argued that speakers have meaningful thoughts. I have argued further that
their language expresses thoughts. This relatively uncontroversial assumption,
LET, has serious consequences for our view of linguistic competence and pro-
cessing.
The most theory-neutral view of competence in a spoken language comes with
position (m), the minimal position on the psychological reality of language (3.4).
It is the view that this competence is the ability to produce and understand
sentences with the sounds and meanings of that language (analogously, of course,
competence in a language in another medium but, for convenience, I will talk
only of a spoken language).6 This view is so neutral that even an eliminativist
about thoughts could adopt it. Our relatively uncontroversial assumption leads
4 Sperber and Wilson object to the view that ‘‘the communication of a thought is achieved by
uttering a sentence identical to it in content’’—part of what they call ‘‘the code theory of verbal
communication’’—on the ground that ‘‘sentences with pronouns are obvious exceptions’’ (1998:
187–8).
5 I am not inclined to think that a mismatch between speaker and conventional meaning (in the
context) is as common as some think (Sperber and Wilson 1995, 1998; Bach 1987) but, so far as I
can see, nothing I say hinges on this.
6 In certain circumstances we might say of someone who loses this ability in an accident that she
still ‘‘knows’’ the language? The circumstances are ones where we take central parts of the ability to
be still intact (6.3). The competence I am describing encompasses more. (Aphasias are discussed in
sections 10.1 and 10.2.)
to a more theory-laden view: the competence is the ability to use a sound of the
language to express a thought with the meaning that the sound has in the
language in the context of utterance; and the ability to assign to a sound a
thought with the meaning that the sound has in the language in the context of
utterance.7 In brief, the competence is the ability that matches token sounds and
thoughts for meaning.
A qualiWcation is called for: the process from a sound to a thought must
involve abilities other than linguistic competence. Linguistic competence pro-
vides the process only with those aspects of a token’s meaning that come simply
from its being of a certain type that is in the language. The process must also
involve some ‘‘pragmatic’’ abilities; in particular, it must involve the ability to
determine the reference of indexicals and the ability to remove ambiguities. We
shall have to say more about these pragmatic abilities later (11.8). Meanwhile, we
should modify the second half of our account of linguistic competence to read:
an ability, together with certain pragmatic abilities, to assign to a sound a thought
with the meaning that the sound has in the language in the context of utterance.
LET has led to the view that linguistic competence is an ability to match sounds
and thoughts for meaning. If this is right then it is immediately apparent that any
theory of linguistic competence, and of the processes of language comprehension
and production, should be heavily inXuenced by our view of the nature of
thoughts. So, the psychological reality of language should be investigated from
a perspective on thought. That is my fourth major conclusion. We will be exploring
the nature of thought further in Chapter 9. The results of the exploration
will guide my own investigation of the psychological reality of language in the
rest of this book.
According to the view that we are presenting, competence in a language
requires a certain conceptual competence, the competence to have thoughts
with the meanings expressible in the language. For, clearly, a person cannot
match a sound to a thought for meaning unless she can have the thought with
that meaning. The conceptual competence partly constitutes the linguistic com-
petence. What more is there to linguistic competence? What is the other part?
That other part must be a ‘‘processing competence’’, a competence to match for
meaning the thoughts that stem from the conceptual competence with certain
sounds. The linguistic competence is a combination of the conceptual compe-
tence to think certain thoughts together with this processing competence. So our
relatively uncontroversial assumption leads to the interesting conclusion that a
certain conceptual competence is part of the very nature of a linguistic compe-
tence.8 Since the processing competence is surely not part of the very nature of
7 Thus, consider the following: the language understander ‘‘must recognize each of the words in
the sentence and determine the syntactic (and semantic) relationship among them . . . determine
the propositional content or ‘message’ ’’ (Tanenhaus and Trueswell 1995: 218).
8 But couldn’t someone who had only the processing competence be linguistically competent? It
is metaphysically possible for a person to lose her conceptual competence in an accident but retain
the conceptual competence, the conceptual competence has an ‘‘ontological
priority’’9 over the linguistic competence: it is metaphysically possible to have
the conceptual competence without the linguistic competence, but not vice versa.
This ontological priority has temporal consequences. First, a person could not
acquire competence in a language before she had acquired the competence to
think the thoughts expressible in the language. The acquisition of the conceptual
competence must either precede the acquisition of the linguistic one or the
acquisitions must occur together.10
Second, the ontological priority has consequences for our innate capacities.
Wherever one stands on Chomskian nativism, it is obvious that humans, unlike,
say, dogs, have an innate capacity to acquire a competence in a natural language.
And humans also obviously have an innate capacity to acquire the competence to
have the thoughts that such a language can express. There are three points to be
made about the temporal relations between these capacities. (a) The ontological
priority of the conceptual competence over the linguistic has the consequence
that the capacity to acquire a linguistic competence could not have evolved before
the capacity to acquire the conceptual one.11 (b) In light of the constitution of a
linguistic competence, the capacity to acquire one must be constituted by the
conceptual capacity together with the capacity to acquire a processing compe-
tence. The conceptual capacity and the processing capacity might have evolved
separately, of course. But, if the processing capacity was an adaptation, something
that was selected for,12 it must have evolved at the same time as, or after, the
conceptual capacity. For, the processing capacity could only be an aid to Wtness by
being a tool of thought.13 And it follows from this that if the linguistic capacity as
her processing competence. Clearly if she were to regain her conceptual competence she would be
linguistically competent. But isn’t she linguistically competent anyway, simply on the strength of her
processing competence? I think not. First, it is rather obvious that she is not competent to
understand the spoken language if she cannot think the thoughts that the language expresses. It
may not be so obvious that the failure to think these thoughts entails that she is not competent to
speak the language. Still, I think we would not count her as being competent. And I am not counting
her as being so. (I am indebted to Alex Barber for prompting this note.)
9 Adopting Martin Davies’ term (1998: 226–7).
10 So I think that Pinker is being too cautious in the following passage: ‘‘Conceptual devel-
opment . . . too, might aVect language development: if a child has not yet mastered a diYcult
semantic distinction, such as the complex temporal relations involved in John will have gone, he
or she may be unable to master the syntax of the construction dedicated to expressing it’’ (Pinker
1995b: 146).
11 Derek Bickerton (1990, 1995) argues that the capacities evolved together. Indeed, he identiWes
the evolution of the linguistic capacity with the evolution of a cognitive capacity that covers not
simply the conceptual competence to have the thoughts but also the competence to reason with
them as we do. I think that the identiWcation with the evolution of the conceptual capacity is
plausible—see section 12.5 below—but, as Carruthers points out (1998: 108–9), it is hard to see
why the reasoning capacity should have come with the linguistic capacity. It is irresistable to
speculate about the origins of language but such speculations are inevitably very short of evidence;
see Botha 2003 for a helpful discussion.
12 Pinker and Bloom (1990) argue persuasively for the adaptationist view but many, including
Chomsky, reject it; see, for example, Hauser et al. 2002.
13 See the papers in Hurford et al. 1998, part I.
a whole was an adaptation, it must have evolved at the same time as, or after, the
conceptual capacity. (c) On the other hand, if the processing capacity was not an
adaptation, hence the linguistic capacity as a whole was not an adaptation, the
processing capacity could have been strangely present but unused until the
conceptual capacity evolved to exploit it.
An objection to the view I am presenting springs immediately to mind. There
is considerable evidence that cognitive impairment and linguistic impairment do
not go hand in hand. So how could the conceptual competence be constitutive of
linguistic competence? I shall respond to this objection in some detail in Chapter
10, along with discussing the apparently conXicting views of Chomsky. Mean-
while, note that the conceptual competence in question is a competence to think
certain thoughts but it is not a competence to reason well with those thoughts. So
the view allows a linguistically competent person to be stupid.
Finally, here is a consideration in favor of the ontological priority of conceptual
competence over linguistic. A consequence of that priority is that an organism
could, logically, have the conceptual competence to think certain thoughts without
having the linguistic competence to express them. But there are persuasive reasons
for thinking that some organisms do, as a matter of fact, have a primitive sort of
conceptual competence, at least, without any corresponding linguistic compe-
tence. First, consider apes, dogs, even sheep. The best explanation of their
behaviors attributes thoughts to these animals, mostly rather primitive thoughts
perhaps, but thoughts nonetheless (Allen and BekoV 1997). And yet these animals
do not have languages in which they could express these thoughts.
A wide variety of studies indicate that nonhuman mammals and birds have conceptual
representations. Surprisingly, however, there is a mismatch between the conceptual
capacities of animals and the communicative content of their vocal and visual signals
. . . animals acquire and use a wide range of abstract concepts, including tool, color,
geometric relationships, food, and number. (Hauser et al. 2002: 1575)
Second, consider our human ancestors. If we go back far enough we will Wnally
reach some ancestors who spoke no language. Yet they surely had thoughts
(Bickerton 1990, 1995). Third, consider immature humans. Developmental
psychologists have produced persuasive evidence that babies have a rich mental
life. Thus, babies show early signs of having such basic concepts as that of a
physical object (Carey 1985). (A concept is, of course, a part of a thought.) And a
study of deaf children who had not been exposed to any conventional sign
language showed ample signs of thought (Goldin-Meadow and Zheng 1998).
Despite their lack of a language, the children communicated using gestures that
revealed ‘‘thought that has not yet been Wltered through a language model’’ (p.
28), ‘‘untainted manifestations of child ‘thought’ ’’ (p. 50). The authors con-
clude: ‘‘Our data suggest that there are indeed thoughts that children themselves
bring to the language-learning situation . . . conceptual starting points for gram-
matical notions’’ (p. 52).
So we have very good reason to believe that organisms can have thoughts
without having a language to express them. Yet organisms surely cannot have a
language capable of expressing thoughts without having the competence to have
those thoughts.
In conclusion, the relatively uncontroversial assumption LET undermines the
view that language’s place in the mind is largely independent of thought: the
assumption ties language closely to thought because conceptual competence
partly constitutes linguistic competence.
8.4 E X P L A N ATO RY P R I O R I T Y O F T H O U G H T
Our next assumption comes from Grice (1989). The Gricean story builds on
LET. An utterance has a ‘‘speaker meaning’’ reXecting the meaning of the thought
that the speaker is expressing, reXecting its ‘‘message’’. Often that meaning will be
the conventional or literal meaning of the utterance in some language in that
context. But sometimes it will not be: metaphors are an obvious example where
the two meanings come apart (as we noted in section 8.2). Finally, and crucially,
speaker meaning is explanatorily prior to conventional meaning: the regular use
in a community of a certain form with a certain speaker meaning leads, somehow
or other, to that form having that meaning literally or conventionally in the
language of that community.14 So the story is: thought meanings explain speaker
meanings; and speaker meanings explain conventional meanings.15 In this way
thought is explanatorily prior to language.16
(A conventional meaning can also be established by some inXuential people
agreeing to use a certain form with a certain speaker meaning. Icelanders
successfully use this method to keep their language unique and some French
would love to use it to keep out English. Still, the method is rare and I shall
ignore it. In any case, it also demonstrates the explanatory priority of speaker
meaning over conventional meaning.)
This part of the Gricean story is relatively, although certainly not entirely,
uncontroversial17 but other parts are not. (i) Grice and his followers have made
14 This account would need to be modiWed along the following lines to deal with indexicals: the
regular use of a certain form with speaker meanings dependent on the context leads, somehow or
other, to that form having the literal or conventional meaning in the language of yielding such
meanings dependent on the context.
The literal and the conventional meaning of an expression are usually the same, but not always.
A person may have an eccentric idiolect: the literal meaning of her expression may not be a meaning
it has according to any linguistic convention. I shall ignore this until section 10.5.
15 This Gricean explanation of the linguistic in terms of the psychological and social is an
example of the earlier-mentioned ‘‘deep explanation of linguistic reality’’ (2.7 (iii)).
16 Davies calls this ‘‘analytical priority’’ (1998: 227). I resist this name because I don’t think that
the priority involves any a priori analysis of concepts (1996; Devitt and Sterelny 1999).
17 Michael Dummett (1993) gives a priority to language over thought and Donald Davidson
(1984) gives no priority to either. Both these views arise, in my view, from rather behavioristic
assumptions about the mind.
many attempts to explain speaker meaning in terms of a baroque structure of
communicative intentions. (ii) There have also been attempts to say how the
regular use leads to a convention.18 I am not committed to the results of any of
these attempts. Indeed, I totally reject the attempts in (i) which strike me as
misguided in principle (1981: 80–6; Devitt and Sterelny 1999: 146–51). Fur-
thermore, it is important to note that Gricean explanations of speaker meanings
are seriously incomplete because they rest on unexplained thought contents: the
speaker means the thought content that he intends to convey but no account is
given of the nature of that content. Still, what I do take from the Griceans, the
distinction between speaker and conventional meaning and the priority of the
former, is suYcient to establish the explanatory priority of language.
The earlier ontological priority had a temporal consequence. A person
could not acquire competence in a language before she had acquired the com-
petence to think the thoughts expressible in the language (8.3). The present
Gricean explanatory priority also has a temporal consequence. There can be no
conventional meanings in a community until there have been the regularities in
speaker meanings. And utterances with a certain speaker meaning must be
accompanied by thoughts with that meaning. This requires a certain temporal
priority of thoughts over language. We shall give some details of this priority later
(8.6, 9.5).
We noted in the last section that organisms can have thoughts without having a
language to express them and yet cannot have a language capable of expressing
thoughts without having the competence to have those thoughts. The part of
the Gricean story that I have adopted Wts these facts nicely. Furthermore, that
part is prima facie plausible and has been extensively supported in the literature
over the last forty years.19 All in all, the argument for a general explanatory
priority of thought over language seems very persuasive. Indeed, it is hard
to think of a plausible alternative. Once one has accepted that there are thoughts
which language expresses, this priority seems compelling. Why then might it
be resisted?
We shall consider Chomsky’s resistance in Chapter 10. And we shall soon
consider possible resistance arising from two sources, linguistic relativity and
the phenomenon of reference borrowing. But, Wrst, we can use the discussion
so far to argue for our theoretical interest in thought being prior to that in
language.
18 SchiVer 1972 is an example of particularly valiant attempts at both tasks. Lewis 1969 is the
classical attempt to explain conventional meaning, but it is not Gricean. Laurence (1996, 1998)
rejects the view that there are any linguistic conventions of the sort Lewis is trying to explain. I shall
consider his rejection in section 10.5.
19 Aside from Grice 1989 and SchiVer 1972, see for example Armstrong 1971, Bennett 1976,
Avramides 1989, and Neale 1989. (These works include, of course, parts of the Gricean story that I
am not adopting.)
8.5 PRIORITY OF THEORETICAL INTEREST IN

THOUGHT
My earlier argument that the linguistic task is to explain the nature of sentence
tokens raised the important question: What is our theoretical interest in this task?
Why is the task worthwhile? I gave four reasons for thinking it worthwhile (2.4).
First, it must be worthwhile if the study of linguistic competence (Chomsky’s
task (i) ) is worthwhile because that study involves my task. Indeed, my task has a
certain epistemic and explanatory priority over the study of competence. Second,
I noted the interest of an analogous task, explaining the code of the bee’s dance.
Third, I claimed that substantial and interesting theories—generative gram-
mars—are fulWlling the task. Fourth, and most importantly, I claimed that the
properties of tokens that the task studies—meanings, hence the syntactic prop-
erties that partly constitute meanings—play striking roles in our lives. We can
now expand on this claim and place the linguistic task in a broader theoretical
environment that gives priority to interest in thoughts.20
Suppose that Mark produces the sound, /It is raining/. We ascribe to this
sound the conventional meaning IT IS RAINING. If we assume that Mark
literally meant this, then we take that meaning to be its speaker meaning too.
As a result, we have evidence of his thoughts. Most straightforwardly, if we
assume that he is being sincere, we will take him to have a belief with that
meaning. So we ascribe the same meaning to the belief as to the utterance it
causes; we take the utterance to express the belief. We saw earlier how ascribing
that meaning to a belief can directly explain Mark’s behavior and guide us to
reality. So, ascribing that meaning to an utterance can serve those same purposes
indirectly. An essential part of what makes the utterances that humans produce
and react to language is that the utterances are the expressions of meaningful
thoughts. And an essential part of what makes language theoretically interesting
is that these thoughts, hence their expression, can explain behavior and guide us
to reality.21 Language is an extraordinarily eVective way of making the thoughts
of others accessible to us, thoughts that otherwise would be largely inaccessible;
and of making our thoughts accessible to others, often in the hope of changing
their thoughts and hence their behavior. So we have a great theoretical interest in
explaining the properties of linguistic expressions, including their syntactic
properties, that enable the expressions to play this striking role. And just as our
interest in the properties of the bees’ dance leads to an interest in the bees’
competence to produce dances so also does our interest in linguistic expressions
lead an interest in our competence to produce them. We have the following
20 I draw, once again, on my 1996, secs. 2.3–2.8.

21 ‘‘This is the essence of the language instinct: language conveys news.’’ (Pinker 1994: 83)
‘‘direction of theoretical interest’’: from thoughts to language to linguistic com-
petence.
In sum, not only does thought have a certain ontological and explanatory
priority over language, our theoretical interest in thoughts has a certain priority
over our interest in language. I shall return to this theme later in criticizing
Chomsky’s view of the importance of idiolects and the unimportance of conven-
tional meanings (10.5), and Georges Rey’s view of ‘‘the intentional inexistence of
language’’ (10.6).
8.6 L I N G U I S T I C R E L AT I V I T Y 2 2
One source of resistance to our Gricean-based claim of explanatory priority might

come from the ever-popular thesis of linguistic relativity, due originally to Edward
Sapir and Benjamin Lee Whorf. For, this thesis appears to be committed to the
opposite priority, the priority of language over thought. So we need to consider
linguistic relativity.
We might state the thesis of linguistic relativity somewhat vaguely as follows:
Because languages diVer signiWcantly in their representations of reality, speakers of
diVerent languages diVer in some thoughts.
This can yield several more precise versions diVering greatly in power and likely
truth. Two extreme versions are certainly powerful enough to warrant the aura of
excitement and signiWcance that used to surround the thesis of linguistic relativ-
ity. The Wrst, sometimes suggested by the writings of Sapir (1949: 162) and
Whorf (1956: 213), takes a language to determine not only thought but reality
itself. This neo-Kantian idea that we make our world with our concepts or
language is sadly popular but has, in my view (1997), absolutely nothing to be
said for it. The second extreme version, also sometimes suggested (Sapir 1931:
578; Whorf 1956: 213–14, 256), takes a language to force people to think in
certain ways and to constrain them from thinking in other ways; diVerences
between languages lead to thoughts that are insurmountably incommensur-
able.23 This version would certainly be inconsistent with my Gricean priority
claim but it is highly implausible (Devitt and Sterelny 1999: ch. 10) and is not
part of the current debate.
The current debate is over whether a language inXuences its speakers to think in
certain ways, whether it facilitates some thoughts and not others. This version of
linguistic relativity is surely true to some extent. It is much more diYcult to
invent concepts than to use ones already available. A language makes many
concepts available. Thus, it was once very diYcult for anyone to have thoughts
22 This section and the next draw on my 2002.
23 This is the so-called ‘‘strong’’ form of linguistic relativity. It has been argued that Sapir and
Whorf did not really hold this view (Gumperz and Levinson 1996: 33n).
about genes because there was no word for genes. And it is still very diYcult for
the speakers of any language that lacks a word for them. But for the rest of us, it is
very easy. Suppose that the legendary story of the many Eskimo words for snow
were true.24 Then it would doubtless be easier for the Eskimo than for us to think
a whole lot of thoughts about snow. Every specialist group with its own technical
vocabulary illustrates the same point: wine tasters, chess and bridge masters can
all think certain thoughts more easily than the rest of us. The adoption by the
English language of the foreign words ‘kow-tow’ and ‘kosher’ may well have
made it easier for English speakers to think certain thoughts about behavior
toward ‘‘a superior’’ and about what is according to accepted rules or standards.
But such examples, of which there are many, are not interesting examples of
linguistic relativity.
Whorf made very interesting claims about the inXuence of the syntax of
languages on thought. He thought that syntactic diVerences between the Hopi
language and an SAE language like English, including their tense-aspect systems,
led to diVerent conceptions of time. But there are serious problems with this line
of thought. One is that his gloss on the Hopi language is controversial. Another is
that it is far from obvious what our conception of time is, let alone what the
Hopi’s is. And there are more (Devitt and Sterelny 1999: 10.3).
Alfred Bloom (1981) provided what seemed to be powerful evidence of the
inXuence of a language’s syntax. He claimed that the absence of distinct counter-
factual markers in Chinese made it diYcult for Chinese speakers to think
counterfactually. But there were problems with his experiments, and when
these were Wxed the diYculties disappeared (Au 1983).
In recent times, psychologists have investigated some quite exciting examples of
linguistic inXuences. (i) Languages diVer in the way they describe spatial and
temporal relations. Do these diVerences aVect the way people think about space
and time? (ii) Languages diVer in the extent to which they distinguish grammat-
ically between objects (e.g. rabbits) and substances (e.g. gold). Do these diVer-
ences aVect the way people think about what objects are made of and their shapes?
(iii) Languages diVer in the genders they assign to nouns. Do these diVerences
aVect the way people think about the object to which the nouns apply? I gather
that the jury is still out on the signiWcance of such investigations but they are
certainly suggestive of interesting inXuences of language on thought.25
So what bearing do the established truths, even the possible truths, of linguistic
relativity have on the Gricean view that thought is prior to language? Very little.
First, the Gricean view that thought is prior to language in the explanation of
meaning is quite compatible with any amount of causal inXuence of language on
thought. Thus, we are easily able to have thoughts about genes because we have a
24 Pullum (1991) argues that it is a hoax.

25 See Boroditsky 2003 for a nice summary.
word ‘gene’. But it was because of the pioneering thoughts of Mendel that we
have the word ‘gene’ at all. And we get the beneWt of the word ‘kow-tow’ because
long-ago Chinese thoughts led to its introduction. Even if Whorf were right
about the diVerent inXuences of the Hopi language and English on thoughts
about time, conceptual diVerences could have established the linguistic diVer-
ences.
Second, the fact that the presence of a word in a language makes it easy for a
speaker to gain a concept does not show that, contrary to the Gricean view, the
word has a role in determining the nature of the concept thus gained; it does not
show that language is prior to thought in the explanation of meaning. Consider
the concept <kow-tow>, for example. It is very likely that English speakers have
this concept as a result of the introduction into English of the Chinese word,
‘kow-tow’. Yet it is plausible to think that the concept is constituted by its
association with the concept <show servile deference>. The story of how an
English speaker comes to have the concept of <kow-tow> is beside the point of
its nature. What makes it the case that a concept she has is that concept and not
some other one is its association with <show servile deference>. Indeed, we can
say quite generally that the cause of x is one thing but the nature of x may be a
quite diVerent thing; think of a war, for example.
So the likely truths of linguistic relativity are not at odds with the Gricean
story. Nor are two other matters. First is the likely fact that those who have no
acquaintance with a language during the critical period will be cognitively
impaired as well as linguistically so; consider the famous cases of ‘‘wolf children’’
(Malson 1972) and of Genie whose parents deprived her of human contact
(Curtis 1981). Second are the interesting proposals of Andy Clark, revolving
around the following two themes: ‘‘the use of text and/or speech as forms of
external memory and workspace’’; ‘‘the (putative) role of words and sentences
. . . as transformers of the very shape of the cognitive and computational spaces
we inhabit’’ (1998: 174). So far as the Gricean story is concerned our rich
conceptual life could be highly dependent on our having a public language.
This discussion of linguistic relativity enables us to say more about the
temporal priority of thought over language. (a) The ontological priority of
conceptual over linguistic competence (8.3) leads to the conclusion that a person
could not be competent with words—for example, ‘gene’ or ‘kow-tow’—before
she had the appropriate concepts. Still, the presence of a word in a language may
lead a person to gain the word and the concept in the one moment. (b) According
to the Gricean explanatory priority (8.4), this presence requires that people once
had the concept without the word. For, a word is present in a language when its
form is conventionally associated with its meaning by speakers of the language.
On the Gricean story, that association arose out of the form being regularly used
by speakers with that meaning. And those uses require that the speakers already
had the concept that the word expresses. So the concept came before the word’s
presence.
8.7 R E F E R E N C E B O R ROW I N G
A better reason for resisting the Gricean story is that it needs an important
modiWcation. The modiWcation is not about syntax, our main concern, but it is
important enough to warrant discussion. Despite the claims of the last section, I
think that we have very good reason for thinking that the natures of many
concepts are not, as a matter of fact, very diVerent matters from their causes:
the causal story of many a concept is intimately involved in an account of its
nature. Although thought is indeed ultimately prior to the public language in the
order of explanation, the meanings (contents) of many of our concepts are to be
partly explained by the conventions of the language. These are concepts that
partly depend for their natures on reference borrowing (or deference).
Saul Kripke (1980) and Keith Donnellan (1972) suggested a theory of refer-
ence borrowing as part of an historical–causal theory of proper names. The
historical–causal theory starts with reference Wxing. Initial users of a name Wx its
reference, usually in face to face ‘‘groundings’’ in an object. The Wrst of these
groundings is a dubbing but, as I like to emphasize, it is important to the
plausibility of the theory to note that a name is likely to be ‘‘multiply grounded’’
in the object. Now, I doubt that the historical–causal theory alone has the
resources to explain reference Wxing: very likely we will have to appeal also to
ideas from indicator or teleological theories.26 But we can set the problem of
reference Wxing aside because our concern here is with reference borrowing. The
theory is that, after grounding, the name is passed on from person to person in
communication situations: later users who have no acquaintance with the bearer
borrow the reference of the name from earlier users. For example, consider our
current uses of ‘Aristotle’ to designate the famous philosopher. These uses desig-
nate him in virtue of being causally linked to him via centuries of reference
borrowings and the initial groundings.
Kripke extended the historical–causal theory to cover natural kind words like
‘tiger’ and ‘water’. So did Hilary Putnam (1975), who brings out the sig-
niWcance of reference borrowing in his talk of ‘‘the linguistic division of labour’’
(pp. 227–8). Language is a social phenomenon. People are equally able (in
principle) to use each term of language in their interactions with the world
even though they are not equally able to relate that term to the world. How?
Because they each gain the beneWt of their linguistic involvement with others.
Those on whom everyone ultimately depends are, of course, those who have
grounded the term. These grounders may be experts, able to give the identity
26 See Devitt and Sterelny 1999 for a detailed discussion of historical–causal theories (chs. 4 and
5) and for a much briefer assessment of indicator and teleological theories (sec. 7.7). Historical–
causal theories need multiple grounding to deal satisfactorily with cases of reference confusion and
cases of reference change.
conditions of the referent, but it is not essential that they be. What matters is that
they have, as a matter of fact, grounded the term to the world.
Tyler Burge (1979) has gone further, making a plausible case that reference
borrowing extends to words like ‘arthritis’, ‘sofa’, ‘contract’, and ‘brisket’. If so,
reference borrowing is a feature of a signiWcant proportion of the words in the
language. This is not, of course, to say that it is a feature of all words.
The theory of reference borrowing shows how a person can be linguistically
competent with a word despite being largely ignorant, or even wrong, about its
referent. People can be competent with the name ‘Catiline’ despite knowing very
little about Catiline; they can be competent with ‘Einstein’ despite thinking that
Einstein invented the atomic bomb; they can be competent with ‘elm’ and
‘beech’ without having any knowledge that would distinguish elms from
beeches; they can be competent with ‘contract’ despite wrongly thinking that
contracts have to be written.
How does this theory of words bear on concepts? Linguistic competence with a
word is suYcient for conceptual competence with the concept that the word
expresses. So the theory of reference borrowing shows how people can have
concepts like <Catiline>, <Einstein>, <elm>, <beech>, and <contract>
despite being ignorant or wrong about their referents.
In sum, the theory of reference borrowing places very little epistemic burden
on the linguistically and conceptually competent. Contrast this with the rich
burden of associations demanded by some other theories: for example, what
philosophers call ‘‘description’’ theories of words; and what psychologists call
‘‘classical’’ and ‘‘prototype’’ theories of concepts. There is, of course, room for
argument about just how little an epistemic burden should be placed on the
competent, but we need not join this argument. SuYce it to say that the
competent may associate with a concept little that is true of its referent.
In light of this, we must modify the earlier Gricean account of the priority of
thought over language. Consider the concept <Aristotle>. We all have this
concept and as a result are able to think thoughts about the ancient philosopher.
According to the theory of reference borrowing, we have the concept largely in
virtue of a causal chain stretching back to the philosopher. That is, the fact that
one of our mental states is the concept <Aristotle> is largely determined by the
fact that this mental state is plugged into an appropriate causal network
grounded in the philosopher. Think now of this network. It was established
and maintained by a linguistic convention of using the word ‘Aristotle’ in
communication situations to refer to the philosopher. Such uses participate in
the convention. Thus the nature of our concept <Aristotle> is largely deter-
mined by this conventional use of the word ‘Aristotle’. The conventional
meaning of this word, featuring in generations of reference borrowings, is an
essential part of the causal chain that largely constitutes the meaning of our
<Aristotle> concept. What makes our concept an <Aristotle> concept is largely
its relation to a word that conventionally means ARISTOTLE.
There is just the same dependence of a concept on language with any person’s
concept that is based on reference borrowing. So there is likely to be this dependence
with most people’s concept <elm>, most people’s concept <gene>, some people’s
concept <kangaroo>, and perhaps a few people’s concept <arthritis>.
This does modify the Gricean story but it does not come close to abandoning
it, for the following four reasons:
(1) Even where a person’s concept is based on reference-borrowing, and hence
dependent on an established linguistic convention, we have not claimed that it is
entirely so. Perhaps there is some small epistemic burden on the person’s concep-
tual competence so that the concept has some non-linguistic determiners; for
example, perhaps the concept <Aristotle> has to be associated with the concept
<human>.
(2) We have not claimed that all concepts can be based on reference-borrow-
ing. It is plausible to think that concepts expressed by proper names, natural kind
concepts, and some Burgean concepts can be so based. It is not so plausible to
think that others like <pediatrician>, <bachelor>, and <hunter> can. If they
cannot, the natures of these concepts have no direct dependence on linguistic
conventions.
(3) Even where some people have a concept of a certain type based on
reference borrowing and linguistic convention, it is not the case that all people
do. Thus, whereas we now have such an <Aristotle> concept, Aristotle’s con-
temporaries who Wxed the name’s reference by dubbing him ‘Aristotle’, did not.
Furthermore, all those involved in subsequent groundings of a word have a
concept which is, to that extent, dependent for its nature not on linguistic
convention but on direct confrontation with the appropriate object(s). Some of
us have had such confrontations with Kripke, Putnam, elms, and kangaroos, and
so have <Kripke>, <Putnam>, <elm>, and <kangaroo> concepts partly
based on those confrontations, partly on linguistic conventions.
(4) On the Gricean picture, it is the regularities of speaker meanings arising
out of these convention-independent thoughts that establish the convention
in the Wrst place. It is in this way that speaker meaning is prior to conventional
meaning. So even the concepts of those who have made no groundings,
concepts that are therefore dependent on the convention—for example, our
<Aristotle> concepts—have natures that are ultimately explained in terms of
groundings. For, the convention itself is explained in terms of groundings.
The creation of the linguistic convention requires some people to have concepts
that are not dependent on conventions. Once created the convention makes
it possible for other people to gain the concept by reference borrowing,
thus having a concept that is to be explained partly in terms of that conven-
tion. Thought meanings explain the conventions that explain other thought
meanings.
In conclusion, the fact of reference borrowing forces a modiWcation in the

Gricean claim but it leaves intact the view that, ultimately, thought is explana-
torily prior to language.
In this chapter I have argued for the relatively uncontroversial views that there
really are thoughts (intentional realism) and that our language expresses them:
LET (8.1–8.2). This led to my fourth major conclusion: the issue of the
psychological reality of language should be investigated from a perspective on
thought (8.3). It led also to the view that conceptual competence partly consti-
tutes linguistic competence and so is ontologically prior to it. Following Grice,
I next argued that thought is explanatorily prior to language (8.4). These
ontological and explanatory priorities have some interesting temporal conse-
quences. And on the basis of these priorities, I argued that our theoretical interest
in thought is prior to that in language (8.5). The Gricean claim was shown not to
be undermined by such truths as there may be in the thesis of linguistic relativity
(8.6). However, the claim did have to be modiWed in light of the phenomenon of
reference borrowing. Still, thought is ultimately prior to language in the order of
explanation (8.7).
We might sum up these priority claims roughly as follows: thought has a
certain priority to language ontologically, explanatorily, temporally, and in
theoretical interest. That is my Wfth major conclusion.
These priority claims, and the approach that they stem from, are very sign-
iWcant for the investigation of the place of language in the mind, and yet they are
strangely unnoticed. We shall be paying a lot of attention to them in Chapter 9
in arguing for position (t) and in Chapter 10 in considering Chomsky’s appar-
ently conXicting commitment to the independence of thought and language.
9
A Case for the Psychological Reality of
Language
9.1 R E P R E S E N TAT I O N A L T H E O RY O F T H E M I N D ( RT M )
According to my fourth major conclusion, the nature of thoughts is central to the

issue of the psychological reality of language. We will be exploring that nature in
this chapter.
I shall present an argument for (t), the position that the structure rules of a
speaker’s language are similar to the structure rules of her thought. I pointed out
earlier that any view that the language is psychologically real in a robust way
requires a powerful psychological assumption (second methodological point;
2.6). Position (t) is certainly powerful. And the case for it rests on controversial
views about the nature of thoughts, in particular, on the Language-of-Thought
Hypothesis that thoughts involve language-like mental representations (LOTH).
The case for (t) is far from conclusive. En route to (t), we shall present, in this
section, the Representational Theory of the Mind (RTM)
We have already noted that thoughts are mental states with meanings (8.1).
What more can we say about them?
(i) We should start with the popular RTM. A Wrst attempt at stating RTM
runs as follows: to have a thought, say the belief that p, is to stand in a certain
functional relation to a token mental representation that means that p.1 So the
belief has its meaning in virtue of its vehicle, the representation, having that
meaning. In adopting RTM, we have for the Wrst time posited mental represen-
tations. Pylyshyn’s Razor cautions against making such posits beyond necessity
(3.1). Positing thoughts is necessary to explain behavior and to use others as
guides to reality (8.1). And RTM is necessary to explain thoughts.
The positing of mental representations is the important part of RTM for our
purposes and it is clear. Still, there is another part of RTM that is not clear and
that is worth a digression.
What exactly does RTM claim about believing?2 The above statement gives the
impression that RTM identiWes believing with a certain functional relation, one
1 Strictly speaking this applies only to ‘‘core’’ thoughts not to thoughts that person has but has
not entertained.
2 This discussion draws on my 2001.
Case for Psychological Reality 143
that holds between a person and a token mental representation. But do we really
want to say that for a person to have that belief is for them to believe such a token?
That seems a bit odd. The oddness increases if we change the thought. We surely
do not want to say that when a person expects that the Yankees will win she
expects a mental token. The oddness becomes unbearable if we follow Quine
(1960: 151–6) in thinking that ascriptions involving intentional verbs like
‘‘hunt’’ should be treated similarly to thought ascriptions: to hunt Bin Laden is
not to hunt a mental token!3 But if we do not identify believing with the
functional relation that holds to mental tokens what, according to RTM, do
we say about its nature?
At the core of RTM is the view that the psychological reality underlying
(1) Joe believes that the Yankees will win
is that Joe stands in a certain functional relation to a mental token with a certain
meaning. Call that relation ‘‘B’’ and the meaning ‘‘M ’’. So if (1) is true,
(2) Joe Bs an M token.
If believing is not identiWed with B what exactly has (2) got to do with (1) beyond
supplying the meaning of the thought ascribed? More generally, what have
the facts according to RTM got to do with what we ordinarily ascribe to
the mind with the likes of (1)? I think that we have to choose between three
answers.4
1. One answer takes believing to be a relation to an intentional object that
holds in virtue of the relation B to a mental token. So, (1) asserts that Joe believes
a certain intentional object, the proposition THE YANKEES WILL WIN. He
does so in virtue of (2). This is neat but it raises two worries.
First, the answer is committed to intentional objects. What are they? The usual
view is that they are abstract objects outside space and time. But many of us think
that such objects are ‘‘creatures of darkness’’. How could we refer to them? How
could we know about them? Second it is unclear, to say the least, how a person’s
relation to such an abstract object can hold in virtue of a relation to a mental
token.
2. I prefer an answer that denies that ‘believing’ is simply a relative term and
that believing is a relation. Rather, we should think of (2) as a paraphrase that
demonstrates the dual function of ‘believing’ (cf. Quine 1960: 154). One
function is to specify the appropriate relation, B, and is therefore the function
of a relative term. But the other function is that of a quantiWer, ‘‘supplying the
3 The ascriptions, ‘a hunts b’ and ‘a believes that b is F ’ are similar in two respects: (i) they can
be true even if b does not exist; (ii) if they are true they might not remain so if we substituted ‘c’ for
‘b’ even though b ¼ c.
4 James Tomberlin (1998, 2001) thinks that we should favor another answer, one that posits
nonactual entities. I don’t think that this is acceptable (1998a, 2001).
indeWniteness’’ of ‘an M token’.5 The meaning M is, of course, speciWed by ‘that
the Yankees will win’.
From this perspective, although it is appropriate enough to ask about the
nature of B, it is not appropriate to ask about the nature of believing. Asking that
reXects the mistaken view that it is a relation.
3. The previous two answers each allow the facts according to RTM to
support what we ordinarily ascribe to the mind; so (1) can be true. I conclude
by mentioning an answer that removes that support, making (1) false. The
answer follows answer 1 in taking believing to be a relation to an intentional
object. So (1) would be true if Ralph stood in the believing relation to THE
YANKEES WILL WIN. The answer diVers from 1 in not supposing that (1) holds
in virtue of (2). Rather the answer claims that (1) does not hold because there are
no intentional objects. (1) is a convenient manner of speaking but, when the
chips are down, must be abandoned in favor of (2).
This is revisionist but it does not seem objectionably so to me. It is an
acceptable fall-back if linguistics tells us that answer 2 does not work. So far as
I can see, my argument in this book does not depend on which answer is right.
I shall talk vaguely of believing ‘‘involving’’ the functional relation B.6
RTM leads to a further enriching of our view of linguistic competence (8.3).
We can replace our talk of thoughts with talk of mental representations. And
since competence involves moving back and forth between mental and linguistic
representations it is appropriate to describe the moves as translation. So the
competence is the ability to translate back and forth between mental represen-
tations and the sounds of the language. The conceptual competence that, along
5 I aired this idea in my 1996, p. 216. The view that the underlying structure of a sentence might
contain a quantiWer not visible on the surface is not surprising. There is so much underlying syntax.
Thus linguists Wnd an underlying quantiWer in ‘He ate’ (as Georges Rey reminded me) and Lepore
and Ludwig (2000) have argued that complex demonstratives (like ‘that dog’) have underlying
quantiWers.
6 The three answers parallel suggestions made by Eugene Mills in his critical discussion (1997) of
my Coming to Our Senses (1996). The present discussion shows that my response to his criticism
(1997b: 397–8) was far too harsh. He was right to be puzzled by RTM. And I was wrong to say that
‘‘nothing much hinges’’ on whether we call the functional relation that I am here calling ‘‘B’’
‘‘believing’’.
The discussion raises the question: what is the common view of where RTM stands on this issue?
Hartry Field’s account at least suggests the identity-of-relation view: ‘‘belief and desire are not
attitudes toward propositions, but toward meaningful sentences in a system of internal representa-
tion’’ (1978: 98). So too does Fodor’s account: ‘‘propositional attitudes are relations between
organisms and formulae in an internal language’’ (1981a: 187). And so too does Rey’s masterly
presentation of RTM under the name ‘‘CRTT’’ (short for ‘‘Computational/Representational
Theory of Thought’’): he talks of CRTT attempting ‘‘to analyze propositional attitudes’’ (1997:
209), and claims that ‘‘hoping, imagining, wishing, dreading, believing, and preferring that p
consist in diVerent computational relations to a sentence that expresses the proposition [that p]’’
(p. 210). But perhaps these suggestions are misleading and these authors are like my earlier self in
not addressing the issue carefully.
with a processing competence, constitutes competence in the language is the
competence to have mental representations with meanings expressible in the
language. RTM makes an important contribution to our theory of linguistic
competence and hence to our likely position on the psychological reality of
language.
9.2 L A N G U AG E - O F - T H O U G H T H Y P OT H E S I S ( LOT H )
In the last section we have enriched our view of thoughts by adding RTM, in (i),
to the view that thoughts are mental states with meanings. There are more
additions to be made.
(ii) I make the relatively uncontroversial assumption that the meaning of this
representation involved in thought is complex.7 An examination of thought
ascriptions would conWrm this assumption; thus the meanings ORTCUTT and
SPY are parts of the meaning ORCUTT IS A SPY ascribed to Ralph’s thought by
‘‘Ralph believes that Ortcutt is a spy’’.
(iii) Controversy may begin with the assumption that the mental representa-
tion that has this complex meaning is itself complex: it has its meaning in virtue
of its parts having meanings. This is clearly a further step because simple things
can have complex properties. Consider a nonsemantic example: The property
being a bachelor may well be complex, consisting partly of being unmarried, but it
is not the case that part of a bachelor is unmarried. And, in semantics, a simple
thing like a Xag can have a complex meaning; for example, the meaning THIS
SHIP HAS YELLOW FEVER.8
This further step does not quite take us to LOTH, as we shall see, but the
familiar arguments oVered by Jerry Fodor for that hypothesis count in favor of
the step. The Wrst of these arguments is a ‘‘methodological’’ one, inferring the
complexity of the mental representation that causes a behavior from the com-
plexity of the behavior (1987: 141–3). The second argument claims that we need
to see mental representations as complex in order to explain thinking, the process
of inferring one thought from another (pp. 143–7). The third is the classical
argument from productivity, based on the potential inWnity of a person’s thoughts,
and its more persuasive recent relative, the argument from systematicity: A
person’s capacity to think one thought is systematically related to her capacity
7 The assumption is not entirely uncontroversial because it would be rejected, presumably, by

those who think that ‘that’-clauses name structureless propositions which are sets of ‘‘possible
worlds’’. See Richard (1990: 9–37) for a nice critical discussion of this.
8 For those who prefer to think of sentence meanings as propositions, the point can be put like
this: The complexity of a proposition is one thing, whereas the complexity of the tokens that express
it is another. See also Fodor (1987: 136–9) on the distinction between the complexity of an
intentional object and the complexity of the mental state that has it.
to think many others (pp. 147–53).9 Set aside for a moment whether these
arguments establish that mental representations have linguistic complexity and
hence establish LOTH. The arguments do seem to make it overwhelmingly
plausible that the representations have at least the complexity of our further step:
that their complex meanings arise from the meanings of their parts.
This assumption falls short of LOTH because nonlinguistic representations
can have this sort of complexity. Thus, part of a map can mean HUDSON IS
CLOSER TO ALBANY THAN TO NEW YORK and yet the ‘‘syntax’’ of that
part—the way its meaning depends on the meanings of its simpler constitu-
ents—is very diVerent from that of a sentence.10 LOTH requires that the
simplest meaningful parts of the representation involved in a thought be like
words, and that the structure of the representation be like the syntactic structure
of a sentence. So the meaning of the mental representation is determined by its
words and its syntactic structure just as the meaning of a sentence is determined
by its words and its syntactic structure. According to this hypothesis we think in a
mental language.
We may seem to have a simple reason for preferring LOTH to the mental-map
view. Map-like representations can obviously capture meanings that concern
spacial relations but how could they capture the vast variety of other meanings?
At Wrst sight, meanings that concern other relations—for example, NEW YORK IS
MORE LIVELY THAN ALBANY — seem to be no problem. But one wonders how
a map could distinguish among the indeWnitely large number of relations. And
could it capture meanings that concern nonrelational situations like FELIX IS A
TIGER and CLINTON IS SEXY ? And what about TURTLES MAKE POOR
LOVERS and HIGH INTEREST RATES MAY CAUSE A BANKING COLLAPSE?
And how could it capture quantiWcation, counterfactual conditionals, and other
complex thoughts? These meanings seem totally beyond maps. However, David
Braddon-Mitchell and Frank Jackson, who favor a mental-map view, have a neat
response to such worries: ‘‘We can be certain that something map-like can serve to
represent any empirical fact about our world. The world itself is map-like’’ (1996:
172). So I shall suspend judgment on the representational capacities of maps.
(iv) Nevertheless, I think that we should prefer LOTH. Perhaps the view that
a mental representation has some nonlinguistic complexity—for example, the
complexity of a map—could meet the requirement of Fodor’s Wrst methodo-
logical argument and could account for the systematicity that is the concern of
his third argument, but it is diYcult to see how it could account for the mental
process of thinking that is the concern of his second argument. Formal logic gives
9 This systematicity seems sort of obvious and yet Kent Johnson (2004) has done an excellent
job of casting doubt on it.
10 Armstrong 1973 proposes a mental-map view. Lewis 1994 entertains the idea that thoughts
are map-like; see also Cummins 1996.
us a very good idea of how thinking might proceed if thoughts are represented
linguistically. From its very beginning, computer science has used this idea to
build machines that do process linguistic representations. In recent years, com-
puter science has developed ‘‘connectionist’’ machines that use representations of
a very diVerent sort, if they use representations at all.11 Despite the striking
success of these machines with some forms of problem solving, connectionist
processes seem rather far from capturing anything like human inference. We still
have very little idea how thinking could proceed if thoughts were not language-
like but, say, map-like.12
The task of explaining the meanings of mental representations provides further
support for LOTH. According to the thesis that these representations are
complex, the thought that Clinton is sexy, like the utterance ‘‘Clinton is sexy’’,
has a part that means CLINTON. How are we to explain the contribution that
parts like this make to the meanings of wholes? When we are concerned with the
analogous problem for sentences, we seek explanations in terms of syntactic
structures. We are a long way from having all the details, of course, but the
approach seems promising. If LOTH is correct then the same approach is just as
promising for thoughts; for, on that view, the meanings of thoughts are the
meanings of mental sentences. And, once again, we have very little idea of a
satisfactory alternative explanation.13
Finally, we shall see later (9.5) that the relation of language to thought provides
further support for LOTH.
9.3 LINGUISTIC COMPETENCE
Consider the consequences of the discussion so far for our view of competence
in a spoken language. We started with the most theory-neutral view of this
competence: it is the ability to produce and understand sentences with the
11 Ramsey (1997) argues that they should not be thought to use representations.
12 Braddon-Mitchell and Jackson rightly point out that maps can evolve over time. They go on:
‘‘When cartographers . . . put two maps together to make one that incorporates all the information
in a single map, these operations are governed in part by the structures of the maps they are working
on’’ (1996: 173). But it is hard to see how this could be a model for thinking which would seem to
require that the one map evolve as the result of diVerent parts interacting with each other.
13 Braddon-Mitchell and Jackson propose a functional-role semantics for their mental-maps
view: the explanation of meaning/content is ‘‘in terms of how the belief controls behaviour’’ (1996:
181). We might expect this to lead to meaning holism—the meaning of any belief depends on the
meaning of every belief—and indeed it does: ‘‘the approach to content via behaviour is one that Wrst
yields the content of a rich system of belief, and then give the content of individual beliefs only
inasmuch as they are part of this rich system of belief ’’ (pp. 183–4). In my view, the case against
meaning holism is overwhelming (1996: ch. 3). It is worth noting that whether the representations
involved in thoughts are language-like, map-like, or whatever, their structure rules will be very
diVerent from the processing rules governing thinking; cf. the formation and transformation rules of
a logic machine (2.2).
sounds and meanings of that language (similarly for a written language, etc.). We
then moved from this minimal view to increasingly theory-laden views. (a)
Intentional realism (8.1), together with the view that language expresses
thought—LET (8.2)—led us to the following view (8.3): the competence is
the ability to use a sound of the language to express a thought with the meaning
that the sound has in the language in the context of utterance; and the ability
(together with some pragmatic abilities) to assign to a sound a thought with the
meaning that the sound has in the language in the context of utterance (similarly
for inscriptions, etc.). Hence, a certain conceptual competence has an ontological
priority over the linguistic competence. (But, remember, the conceptual compe-
tence is to have certain thoughts, not to think well with them; sec. 8.3.) It became
apparent immediately that a view of the nature of thoughts should be central to
our theory of linguistic competence, and hence to our stance on the psychological
reality issue; the fourth major conclusion. (b) RTM was the Wrst, fairly modest,
step toward a theory of thoughts (9.1). It yielded the view that competence is the
ability to translate back and forth between mental representations and the sounds
of the language. (c) Acceptance of LOTH is a further, fairly huge, step. It yielded
the very theory-laden view that competence is the ability to translate back and
forth between mental sentences and the sounds of the language. So the conceptual
competence that, along with a processing competence, constitutes competence in
the language is the competence to think mental sentences with meanings ex-
pressible in the language.
If LOTH is correct, the process of language production and comprehension is
a translation process between the language of thought and the speaker’s public
natural language. This raises the question: Which language is thought in? It is
tempting to suppose that a person’s language of thought is the same as the public
language in which she expresses her thought. According to this thesis, ‘‘Public-
LOTH’’, her language of thought is her language of talk. Something even
stronger than (t) would be true: the rules of the speaker’s language would be
the same as the structure rules of her thought. And the translation process in
language use should be very easy, like that from spoken to written English. An
alternative hypothesis is that a person’s thought is in a distinct mental language,
‘‘Mentalese’’. If this is right, the question arises: How similar is a person’s
Mentalese to her public language? We can anticipate diYculties answering
these questions because we lack strong evidence about the nature of Mentalese.14
The signiWcance of the questions for our concern with psychological reality
can be made vivid by considering the logic machine again (2.2). This machine
takes wVs as inputs, processes them according to rules for valid deductions and
14 Note that the issue here concerns the language that is involved in beliefs, desires, expectations,
and other such central-processor states. It does not concern any other languages there may be in
modules of the mind.
yields wVs as outputs. So, not only are the machine’s inputs and outputs wVs but
all steps of its derivations are wVs. The language over which its computations are
deWned is the formal language that determines the syntactic properties of any wV.
So, although the rules of that language are not internally real processing rules of
the machine they are still internally real in a very robust way: they are structure
rules of the language the machine ‘‘thinks in’’. Similarly, if a person’s language of
thought were the same as her public language, then the structure rules of her
public language would be psychologically real in a very robust way whether or not
any of them were processing rules for language use: they would be structure rules for
her language of thought. And even if the language of thought were a distinct
Mentalese, then the structure rules of her public language would still be largely
psychologically real provided that her Mentalese was syntactically similar to her
public language. That is position (t) on psychological reality.
It is important to note that the possibility that the rules of a speaker’s public
language might be psychologically real in this way depends on LOTH. If
thoughts are, say, map-like, then the rules governing them will be nothing like
the rules of a natural language. But once we adopt LOTH, we raise the possibility
that the rules of the language we think in will be like those of our natural
language. This is an illustration of what I have been emphasizing: that our
view on the psychological reality of language depends very much on our view
of thoughts.
I once (1981) followed Harman (1973) in arguing for the Public-LOTH. This
hypothesis now seems to me mistaken. But I still think it plausible that there is a
great deal of similarity between the syntactic rules of Mentalese and the speaker’s
public language. In the next section I shall consider what can be said for and
against the Public-LOTH. In the one after, I shall argue for the syntactic
similarity.
9.4 T H E P U B L I C - LOT H
One objection to the Public-LOTH can be dismissed quickly. When the living
brain of an English speaker is examined, we Wnd no cerebral blackboard on which
English sentences are written. Nothing inside the head looks like an English
sentence. This objection is confused. What does an English sentence look like?
Practical problems aside, a sentence might look like anything: it can be physically
realized as a sequence of acoustic vibrations, as gestures in a sign language, as
marks on paper, as a sequence of Xags, as electric pulses of various kinds, and so
on. So there is nothing incoherent in the idea that sentences could be realized in a
neural medium as well. Perhaps tokens of thought, just like tokens of speech or
braille, are sentences of a public language.
It is generally agreed that the Public-LOTH must be qualiWed, at least: all
thinking could not be in a public language. Our earlier discussion of animals,
human ancestors, immature humans, and deaf children not exposed to a conven-
tional sign language indicated that the best explanation of their behaviors often
attributes thoughts to them (8.3). Yet these organisms do not have languages in
which they could express these thoughts. And some mature human thought is
surely not in a public language either: consider our thought about music or chess,
for example. So the thesis must be that most mature human thought consists in
having attitudes of believing, desiring, hoping, etc., to mental sentences in a
public language.
This hypothesis has intuitive appeal because our cognitive capacities seem
closely correlated with our linguistic capacities. The general development of the
two capacities goes hand in hand. Further, it is very plausible to suppose that our
ability to think certain thoughts depends on language. There is evidence that we
could not reason about large numbers without having a language (Dehaene
1997). And could we have had beliefs about aircraft, or desires about nuclear
bombs before we had words for aircrafts and bombs?
The hypothesis also has some introspective support. Speech often seems to be
thinking out loud; thought often seems to be talking to oneself, ‘‘inner speech’’
(Sokolov 1972). Moreover, consider a familiar barrier a person must break
through in learning a foreign language: learning to ‘‘think in the language’’.
Until she crashes through this barrier, she uses the foreign language by translating
back and forth to the familiar home language. Or, so it seems to introspection.
These considerations are far from decisive in support of the Public-LOTH, for
the alternative Mentalese hypothesis may well be able to accommodate them.
A variety of causal stories can explain the simultaneous development in a
person of a language and thought without requiring that the thought be in the
language: the language might play a causal role in developing thought; or
thought, along with environmental features, might play a causal role in devel-
oping a language; or both may have the one independent cause, part nature and
part nurture.
It seems fairly straightforward for the Mentalese hypothesis to account for the
introspective data. First, the phenomenology of inner speech does not dictate
that we are actually thinking in the language rather than simply using the
language as an aid to thinking (Clark 1998). Second, perhaps the person’s
translation of Mentalese into the foreign language takes two steps before breaking
through the barrier: she Wrst translates into her native language and then
translates that language into the foreign one. Whereas the translation into the
foreign language is labored and conscious, the translation into the native lan-
guage is practiced and unconscious. Hence her initial sense of translating the
foreign language but not her native one. She breaks through the barrier when she
collapses the two steps into the one unconscious one. Hence her later sense of
thinking in the foreign language.
Furthermore, there is introspective and experimental evidence that seems to
count against the Public-LOTH. We often seem to have a thought and yet have
trouble expressing it, the ‘‘tip of the tongue’’ phenomenon (‘‘TOT’’): ‘‘Speakers
experiencing TOTs know the meaning they want to express, and often have
particular information about the word’s sound (e.g. it starts with a ‘t’) although
they cannot retrieve the full word form’’ (Vigliocco and Vinson 2003: 184). Yet if
the thought were in the very words that express it why would we have the trouble?
The phenomenon suggests, rather, that we are struggling to translate our thought
from Mentalese into English. But perhaps not. Perhaps we are struggling to form
the thought, a thought which once formed is in English and can be easily
expressed.
Experiments show that when we read a passage, we tend to remember ‘‘the
message’’ rather than the precise wording (Wanner 1977). Thus, take some
English sentence that did not appear in the passage we read. If we are asked
whether that sentence appeared, we are likely to say that it did if it has much the
same meaning as one that did. This suggests that we have stored a representation
in Mentalese and do not remember which of several English translations of it
appeared in the passage. But, again, perhaps not. Perhaps we stored a represen-
tation in English but, given our primary interest in the message not the particular
form of words that presented it, we were not concerned to store it in the form
that presented it: we chose a form that is most ‘‘natural’’ for us, perhaps the one
most suitable for retrieval. When asked the question we do not remember
whether that sentence or an equivalent one appeared in the passage.
Finally, we can understand a string of words that is not in our language and,
indeed, could not be in any human language; for example, ‘‘Who do you think
that loves Mary?’’ So it seems as if we can think thoughts that have a syntax unlike
any sentence in our language.15 But does it really seem like that? Perhaps we
understand the string not by carrying its syntax into our thought but by
translating it into a thought with a syntax that is like a sentence in our language.
After all, we often understand the signs and gestures of someone with whom we
share no public language. If LOTH is correct, we do not carry the ‘‘syntax’’ of
these signs and gestures into our thought but rather translate them into a thought
with a linguistic syntax. This syntax might be that of a sentence in our natural
language.
So far we have found nothing decisive one way or the other on whether the
language of thought is largely a public language or Mentalese. However, I rather
doubt that this is a theoretically meaningful issue. What precisely would be
required for a mental sentence to be a sentence of, say, English? It must at least
have the same syntactic and semantic properties as an English sentence. Our
acceptance of LET, the folk idea that language expresses thought, yields the
sameness of semantic properties. In the next section I shall argue that sense can be
made of the sameness of syntactic properties. The problem with the issue is that
this sameness of syntax and semantics is not enough to make the mental sentence
15 Thanks to Georges Rey for this point.

English. We can see this by reXecting on the inscriptions ‘Nana is a cat’ and
‘Nana est un chat’. These sentences have the same syntax and semantics but the
Wrst is English and the second is French. What makes them so is that the former
inscription is an English ‘‘vehicle’’ of meaning and the latter is a French ‘‘vehicle’’.
So if the mental sentence that expresses ‘Nana is a cat’ is to be English rather than
French or whatever, it has to be an English vehicle. What would make it so?. We
could, of course, answer that the mental sentence is English simply because it is
expressed by an English sentence, but that trivializes the question. There seems to
be no possibility of a theoretically interesting answer. English or French vehicles
can be found in many media, as we noted at the beginning of this section: in
sounds, inscriptions, signs, Xags, and so on. But I doubt that that any sense can be
made of these vehicles being found in thought.16
So I propose to set aside the public vs. Mentalese issue. I shall treat the
language of thought as if it were a distinct Mentalese and focus instead on
what seems a more theoretically productive issue: How similar is a person’s
Mentalese to her public language? In particular, how syntactically similar is it?
If the Public-LOTH had been true then Mentalese would have been syntactically
identical to the public language, a position even stronger than (t). But if
Mentalese is syntactically similar to the public language, then (t) is true. My
Wrst tentative proposal is that it is true and so the public language is largely
psychologically real in its speakers. I shall now argue for this.
9.5 T H E S Y N TA X O F M E N TA L E S E
We should start with a qualiWcation. Our earlier discussion of animals, ancestors,

and immature humans (8.4) shows that (t) must be taken as applying only to
language speakers. An organism that does not speak a language capable of
expressing its thoughts can obviously not have thoughts with structures similar
to the language that it speaks. So, the daring proposal that we are considering is
that the Mentalese syntax of mature humans speaking a natural language is
similar to the syntax of that language.
Let us Wll out the proposal a bit. If it is correct then the Mentalese syntax for an
English speaker is similar to the syntax of English (which is not, as we have just
noted, to say that the Mentalese is English). Any Mentalese sentence, like any
other sentence, has its particular syntactic structure in virtue of the particular
rules that govern it. According to the proposal those rules are similar to those for
English sentences. If this were so, the situation would be analogous to the logic
machine. The rules of English would be largely psychologically real in the most
robust way: they would be similar to the rules of the language that the person
16 Defenders of the public language-of-thought hypothesis, like Devitt 1981 and Carruthers
1996, do not seem to have noticed this problem.
thinks in. The processing rules for translating back and forth between Mentalese
and English would be largely deWned over the syntactic properties determined by
those structure rules even if, as would seem likely, none of the structure rules were
processing rules as required by positions (i) and (iii) (3.4).
What about a person who speaks another language, for example Japanese?
According to this proposal, the syntax of this person’s Mentalese would be similar
to the syntax of Japanese and so the rules of Japanese would be largely psychologic-
ally real in her. What about someone who is bilingual in English and Japanese?
She would be bilingual in Mentalese too, some of her thoughts would have a syntax
similar to English, some to Japanese.
Why should we believe this daring proposal? I shall begin my attempt to
answer this question by confronting an appealing objection, nicely stated by
Stephen Pinker (1994: 78–80).
The crux of this objection is that whereas a public language like English is
ambiguous, Mentalese cannot be; and whereas Mentalese is fully explicit, a
public language is not. So Mentalese could not be syntactically similar to English.
Consider the English sentence:
(1) Tex likes exciting sheep.
This is clearly syntactically ambiguous. It could mean that Tex enjoys causing
sheep to become excited or that he enjoys the company of exciting sheep. In
contrast, the mental sentence involved in any thought that (1) expresses cannot
be ambiguous if the person’s thinking is to proceed in the appropriate rational
manner: the central processor can only operate on a mental sentence that makes
explicit whether it has the syntax appropriate to the one meaning or the other.
Although (1) does not mark, in its brute–physical form, the diVerences between
these two syntactic structures, mental sentences must do so, for only thus can the
central processor operate appropriately on these sentences in the process of
thinking. So the two mental sentences that might underlie (1) might ‘‘look
somewhat like’’:
(2) [S[NP[NTex]][VP[Vlikes][VP[Vexciting][Nsheep]]]]
(3) [S[NP[NTex]][VP[Vlikes][NP[Aexciting][Nsheep]]]].
(2) and (3) diVer syntactically from each other and so it seems from (1), the
English sentence that expresses them both. Consider next the phenomena of
pronouns. A passage of English can contain pronouns that might be deictic or
anaphoric. If they are anaphoric they might depend for their reference on this or
that other expression in the passage. These diVerences are not explicitly marked
in English but they must be in Mentalese.
The diVerences between Mentalese and English in ambiguity and explicitness
seem clear enough. But do these diVerences really show that Mentalese is not
syntactically similar to English? There is persuasive reason for thinking that they
do not. A triumph of generative grammar has been to make us appreciate how
much of the syntax of a sentence is not explicit, ‘‘how unrevealing surface
structure may be as to underlying deep structure’’; ‘‘surface similarities may
hide underlying distinctions of a fundamental nature’’ (Chomsky 1965: 24).
Consider empty categories, for example: ‘‘An empty category . . . is a constituent
that has no phonological substance associated with it; it is inaudible in speech,
and invisible in the standard orthography’’ (J. D. Fodor 1989: 156). Consider
also the scope of quantiWers. This scope is often not explicit on the surface and
yet is revealed at the level of logical form or LF, the level of syntactic structure that
most concerns us here (1.3). As Gennaro Chierchia points out: ‘‘in the inter-
pretation of sentences with quantiWed NPs, we apply scoping to such NPs.
Scoping of quantiWers in English is a covert movement, part of the mental
computation of meaning . . . . The result of scoping . . . is what gets semantic-
ally interpreted and is called Logical Form’’. It ‘‘provides an explicit representa-
tion of scope, anaphoric links, and the relevant lexical information’’ (1999: c).
Considerations of this sort lead Stephen Neale to go further: ‘‘rather than saying
that (30) [‘‘Every poet respects some sculptor’’] is an ambiguous sentence, really
we should say that (30) is the surface representation of two distinct senten-
ces . . . that share an S-Structure representation and in fact look and sound
alike’’ (1993: 119). In sum, once one takes note of the implicit structure of an
English sentence it is far from obvious that this structure diVers from that of the
Mentalese sentence it expresses. Indeed, Cherchia goes on to say: ‘‘It is highly
tempting to speculate that logical form actually is the language of thought’’
(1999: ci).
In light of this, we can see that the objection moves far too swiftly from a
diVerence in explicitness and ambiguity to a diVerence in syntax. There are two
further considerations which enhance this response to the objection. And the
second gives substantial support to (t).
First of all, it is important to distinguish a representation’s brute–physical
formal properties from its syntactic properties.17 Whereas the formal properties
are intrinsic to the representation, the latter are relational.
A formal property of the inscription ‘Tex likes exciting sheep’, in the intended
brute–physical sense of ‘formal’,18 is that of beginning with an inscription
shaped such and such (replace ‘such and such’ with a description of the shape
of ‘Tex’). A formal relation between ‘Tex likes exciting sheep’ and ‘Tex loves
Mary-Lou’ is that of both beginning with an inscription of the same shape.
A formal property of a symbol in a computer is that of being a certain pattern of
on-oV switches. A formal property of a symbol in the brain is that of being a
certain array of neurons. Manifestly, the written token of (1) above shares very
17 There is an unfortunate tendency in the literature to confuse these two sorts of properties. For
more on this, see my 1996: 258–65.
18 Confusingly, there is another sense of ‘formal’ arising out of discussions of ‘‘formal languages.’’
Formal properties in this other sense are very similar to syntactic properties.
few formal properties with any spoken token of (1). Similarly, it shares very few
with a mental token that (1) expresses.
Syntactic properties and relations are ones that bear on the construction of
sentences in a language. A syntactic property of ‘Tex’ is that of being a noun; of
‘likes’, that of being a verb; of ‘Tex likes exciting sheep’, that of being a
sentence. A syntactic relation between one interpretation of that sentence and
the sentence, ‘Exciting sheep are liked by Tex’, is that of the latter sentence being
the passive of the former. Syntactic properties are ones that reXect a token’s
relations to other tokens in the language; they are functional properties and
extrinsic to the representation. Although formally so diVerent, a written and
spoken token of (1) might share all their syntactic properties. Sentences that
‘‘look diVerent’’ can be syntactically alike. Formal diVerences are one thing,
syntactic diVerences, another. In assessing (t), we are considering the extent to
which tokens of (1) might share syntactic properties with the formally very
diVerent mental token that they express.
Consider next the following question: in virtue of what does a sentence have its
syntactic structure, whether explicit or not? What makes it the case that a
particular token, for example a token of (1), has the structure it has, perhaps
the structure revealed by (2) or (3)? This question does not seek a description of
the structure that the sentence has—a description that linguists have been so
successful at providing—but rather an explanation of its having that structure.
Since this structure along with the sentence’s word meanings determines the
sentence’s meaning, this explanation is part of an explanation of the sentence’s
meaning. It is part of an explanation of in virtue of what the sentence has
whatever meaning it has.
One of our earlier priority claims points the way to an answer. This was the
Gricean claim that thought meanings explain speaker meanings; and speaker
meanings explain conventional meanings. In this way, thought is prior to
language (8.4).19 We now wed this Gricean story to LOTH, the hypothesis
that is central to the case for (t). So we should answer questions about public
language meanings by Wrst addressing questions about Mentalese meanings.20
The meaning of a mental sentence is determined by the meanings of its words
and the syntactic structure that contains them. In virtue of what do the words
have their meanings? For basic words, probably including the mental correlates of
proper names and natural kind terms, this must be a matter of some sort of direct,
presumably causal, relation to their referents.21 In explaining these relations we
19 It will be remembered that thought meanings are left unexplained in Gricean stories.
20 Questions about Mentalese meaning are not questions about competence in Mentalese. Just as
a theory of a system of linguistic representations—a language—should not be conXated with the
theory of competence in that system (Ch. 2), a theory of a system of mental representations should
not be conXated with the theory of competence in that system.
21 What about ‘‘empty’’ terms like ‘Pegasus’ that lack a referent? This is a tricky question that
must be set aside. (My 1981, chapter 6, is an attempt at it.)
are likely to have to appeal to ideas from historical–causal, teleological, and
indicator theories of reference. I think, though Fodor (1987) does not, that some
words will surely not be basic but rather covered by a description theory;
<bachelor> is a likely example. Meanings for these words will come from
inferential associations with others; for example, <bachelor>’s association with
<adult>, <unmarried>, and <male>.22 But our concern is more with mental
syntax. In virtue of what does a mental sentence have its syntax? Presumably, in
virtue of the way in which its meaning depends on the meanings of its parts and
in virtue of the structure’s role in determining a sentence’s possible inferential
interactions with other sentences. Thus, the mental sentences <Reagan is
wrinkled> and <Thatcher is tough> share a syntactic structure in virtue of a
similarity in these respects. Similarly, <All politicians are rich> and <All police
are corrupt>, which share a diVerent structure in virtue of their similarities.
In light of this, in virtue of what does a public-language sentence have its
meaning? There are two steps to the answer, reXecting the distinction between
the speaker meaning of a sentence on an occasion of utterance and its literal or
conventional meaning on that occasion. (i) The speaker meaning of a sentence
on an occasion of utterance must be determined by the causal story of its
production. Thus the speaker meaning of a word in that sentence is determined
by the meaning of the mental word it expresses on that occasion. The ‘‘speaker
syntax’’ of a sentence on an occasion of utterance is determined by the structure
of the underlying mental sentence that the utterance expresses on that occasion
and by the way in which the utterance was produced from that mental sen-
tence.23 (ii) The conventional meaning of a sentence on an occasion of utter-
ance—what it conventionally means in a language given the context—is
explained somehow in terms of regularities in speaker meanings: Speakers of
the language regularly use one linguistic word to express one mental word,
another, another; they regularly use linguistic sentences with one speaker syntax
to express one mental syntactic structure, another, another.24
Wedding the Gricean story of the general explanatory priority of thought over
language to LOTH has yielded a sketch of an explanation of the syntactic
properties of public-language sentences. What bearing does this sketch have on
(t) and the objection? (i) Suppose that a public sentence was not ambiguous
either in word or structure and was produced as directly as could be from a
mental sentence. Then surely the speaker syntax of that public sentence would be
the same as that of the mental sentence. But perhaps there is no such public
sentence. So now suppose that the public sentence is ambiguous in word and
structure and that the production process simply reduces the explicitness of the
22 See Devitt and Sterelny 1999, particularly section 7.7, for a discussion of these various theories
of word meaning and reference.
23 In my 1996 (p. 158) I failed to note the importance of the way in which the utterance was
produced to its syntax.
24 I have argued for this view before: 1981a: 80–6; Devitt and Sterelny 1999: 7.5–7.6.
mental sentence it expresses. Consider (1), for example. (a) We have noted that
the mental sentence underlying a particular token of (1) must make explicit
whether the structure is of the sort captured by (2) or (3). So, what happens in
production is that this explicitness is lost. (b) The speaker may know several
people named ‘Tex’ and so that name is ambiguous for him. The mental
correlates of this name will explicitly mark the distinction between these various
uses of the name: for example, <Tex1>, <Tex2>, and so on. This explicitness is
also lost in production. Now if the only eVect of the production of a public-
language sentence on the syntax of a mental sentence is to reduce its explicitness,
it seems as if we must say that it has the same syntax as the mental sentence, albeit
less explicitly.25 How indeed could the speaker syntax of the public sentence
diVer from the syntax of the mental sentence it expresses? Perhaps it could do so if
production does more than reduce explicitness. Thus suppose that a mental
sentence has an SOV syntax and production transforms this into what appears to
be an SVO public sentence; or that the mental sentence is active and production
transforms this into what appears to be a passive sentence. It seems rather
unlikely to me that production would make these sorts of changes but if it did
then perhaps we should say that the syntax of the mental and the public diVer.
But even that isn’t obvious. Perhaps we should say the sentences are implicitly
SOV and active, respectively, and so they really have the same underlying syntax
as the mental sentence they express.
What I think we can conclude from all this is that if LOTH is correct, the
speaker syntax of public-language sentences will be the same as, or similar to, the
syntax of the mental sentences they express. So let us be cautious and say that they
are similar. Turn now to the conventional syntax. Insofar as the regularly used
speaker syntax is similar to the syntax of the mental sentence it expresses, so also
will be the conventional syntax that is determined by that regular use. So the
syntax of a speaker’s language is similar to the syntax of her thought. We have
arrived at my Wrst tentative proposal based on (t): a language is largely psycho-
logically real in a speaker in that its rules are similar to the structure rules of her
thought.26 And the objection to this proposal, pointing to diVerences in ambi-
guity and explicitness, fails.27
How did we reach these conclusions? First, we adopted the controversial
LOTH. Second, we noted, what seems scarcely deniable, that public language
sentences can have syntactic properties that are not explicit. Third, we pointed to
the role of mental sentences in explaining the implicit and explicit syntactic
25 There is a worry about this. Suppose that a sailor starts with the mental sentence <This ship
has yellow fever> and expresses it by Xying a yellow Xag from the mast. Intuitively, the yellow Xag
has no syntax. Yet it seems to have been produced by a process that simply removed all the explicit
syntax of the mental sentence. So why does the Xag not have implicitly the syntax of that mental
sentence? I don’t have a good answer but I’m sure there must be one.
26 We could, of course, be much less tentative about the conditional proposal that if LOTH then (t).
27 My previous response to this objection (Devitt and Sterelny 1999: 143) now strikes me as
rather inadequate.
properties of the public sentences that express them. These three assumptions
seem to make (t) inevitable.
The importance of LOTH to this argument is enormous. Thus, suppose that
we dropped it, committing ourselves only to RTM instead. RTM is neutral on
the nature of the mental representations involved in thought. So the representa-
tions might be map-like. If they were it would still be the case that the syntactic
properties of public sentences would have to be explained by the map-like syntax
of those representations and by the production processes from those representa-
tions to the public sentences that express them. But the explanation could not
yield the conclusion that the sentences are syntactically similar to the represen-
tations because, of course, the sentences have a linguistic syntax whereas the
representations have a map-like one; languages diVer from maps.
This prompts an aside. We might well wonder how the syntax of a public
language could be explained by map-like mental representations and the produc-
tion processes from these to language. It is unclear how a map-like representation
could yield an utterance with a linguistic structure. We noted earlier (9.2) that
the view that mental representations are map-like rather than language-like
seemed to promise no explanation of the meaning of those representations. We
can add now that the view also seems to promise no explanation of in virtue of
what a public-language sentence has its syntactic structure. Once one has
accepted intentional realism and RTM, we seem to need LOTH to explain the
relation of language to thought.
The argument we have presented is the main argument for (t). But there are
other considerations that might be thought to give (t) some support.
1. Let us start with linguistic relativity. In our earlier discussion we pointed to

obvious examples of the inXuence of language on thought: the word ‘gene’
makes it easy for us to have thoughts about genes; the word ‘kow-tow’ makes it
easy for us to have thoughts about showing servile deference. And there is the
possibility that there may be more exciting examples of linguistic inXuence.
However, none of these possibilities seemed likely to provide evidence of the
inXuence of the syntax of language on the syntax of thought. And that, of course,
is our particular interest.
So, the investigation of linguistic relativity does not seem yet to have provided
any clear evidence of the inXuence of language on Mentalese syntax. What is the
bearing of this on (t)? Even if there were clear evidence of the inXuence, that
would not establish (t): for the syntax of public language to inXuence the syntax of
Mentalese is one thing, for them to be similar is another. Still, the inXuence
would be signiWcant evidence for the similarity. (Whorf embraced the Public-
LOTH and so, for him, the inXuence leads to Mentalese syntax being identical to
the public one; 1956: 252.)
So if linguistic relativity could be established it would support (t). What about
the reverse? Would the establishment of (t) support linguistic relativity? No.
A similarity in the syntax of the public language and Mentalese might be the
result of the inXuence of the public language on Mentalese but it also might be
the result of the inXuence of Mentalese on the public language or the inXuence of
something else on both languages.
2. This discussion leads naturally to a question about the origins of a person’s

Mentalese. Mentalese might be innate, as Fodor famously thinks it largely is, or it
might be acquired through experience, or it might be a bit of both. Suppose that
some part of the structure of Mentalese is innate. Then it is plausible to think that
this part will constrain the structure of any learned natural language in ways that
will make its syntax to that extent the same as that of Mentalese. Indeed, I will
later consider the idea that all human languages are governed by the rules of
Universal Grammar (UG) because those rules are largely if not entirely the innate
structure rules of thought (12.4). Suppose, next, that some part of the structure
of Mentalese is acquired through experience. Then it is plausible to think that
this part is indeed inXuenced, as the linguisitic relativists think, by the structure
of the natural language that the person also acquires. So, either way, whether
Mentalese is innate or acquired, it is plausible to think that it is structurely similar
to the natural language.
3. An argument along the following lines might seem to show that the syntax of
the two languages must be fairly similar. (a) The translation process of produc-
tion and comprehension must preserve meanings: when a thought is expressed by
an utterance, the utterance must mean the same as the Mentalese sentence
involved in the thought; when an utterance is understood, it must be assigned
to a Mentalese sentence that means the same. (b) The meaning of an expression is
a function of its syntax. So, the syntax of the speaker’s Mentalese token has to be
close enough to that of its expression in her public language to make them mean
the same.
The problem with this argument is that the requirement that the meanings of
the two tokens be the same is rather vague and thus leaves room for a lot of
diVerence in the syntax of the two tokens. After all, there is a sense in which an
active English sentence means the same as its passive, despite their syntactic
diVerences; and there is a sense in which an English sentence means the same as
its Japanese translation, despite their syntactic diVerences. So this requirement
does not justify the claim that the two languages are closely similar.
4. Fodor has given a reason for thinking that the syntax of a person’s Mentalese is
like that of her language: we have to account for the great speed of language
processing: ‘‘the more structural similarity there is between what gets uttered and
its internal representation, the less computing the sentence understander will
have to do’’ (1975: 152). And the less the computing the less the time needed to
compute. Indeed, if the Mentalese and the language had the same structures, we
can imagine a very direct, very speedy, process from English utterance to
Mentalese interpretation. But, even if the structures are similar, the process
could be fairly direct and speedy.
In sum, LOTH opens up the possibility that the rules of a speaker’s public
language might be psychologically real because they are similar to the structure
rules governing the language of thought. In this section I have presented a case for
their being indeed fairly similar, hence a case for position (t) on psychological
reality. This case is far from overwhelming, particularly as it rests on the
controversial LOTH, but it does seem to me to have a lot of plausibility.
Finally, we should add a word to our earlier discussion of the temporal priority
of thought over language (8.3, 8.4, 8.6). If Mentalese is not entirely innate, then
it is likely that the presence of certain syntactic structures in a natural language
will lead a person to gain the related structures in Mentalese. So, for that person,
there is no temporal precedence of the structure in Mentalese over the structure
in the natural language. But the structure is present in the language at all only
because people once had the related mental structure without beneWt of that
linguistic structure.
As I noted (8.1), the discussion of the relation between thought and language
is often framed around a distinction between the ‘‘communicative’’ and the
‘‘cognitive’’ conceptions of language. I have not followed this practice. Before
concluding this chapter I shall say why.
9.6 ‘‘ C O M M U N I C AT I V E ’’ V S . ‘‘ C O G N I T I V E ’’
CONCEPTIONS
Peter Carruthers and Jill Boucher describe the distinction between these two
conceptions as follows. According to the communicative conception, language is
‘‘a mere adjunct to belief and thought’’; its ‘‘exclusive function and purpose’’ is
‘‘the communication of thought’’; ‘‘thought itself is largely independent’’.
According to the cognitive conception, on the other hand, language is ‘‘crucially
implicated in human thinking’’; ‘‘we think in natural language’’; language is
‘‘constitutively involved in’’, or is ‘‘the medium’’ of thoughts (1998b: 1); ‘‘inner
verbalization [‘‘inner speech’’] is constitutive of our thinking’’ (2002: 657). In my
view, this distinction is too vague to be helpful, conXating many fairly distinct
questions.
First there is the question of how similar the language of thought is to the
public language. This splits into three. How similar are they in their vehicles, in
their syntactic properties, and in their semantic properties? I have argued that
they are very similar in their syntactic and semantic properties but that little sense
can be made of the question of a similarity of vehicle (8.2, 9.4, 9.5). This
conclusion seems to accord more with the cognitive than the communicative
conception.
Next there is a question about whether thought or language is prior. This splits
into as many questions as there are ways in which one might be prior. I have
argued that thought has a certain priority to language ontologically, explanatorily,
temporally, and in theoretical interest. That’s my Wfth major conclusion. The
priorities are quite compatible with a great deal of causal inXuence of thought on
language (ch. 8; 9.5). The ontological priority seems to accord more with the
cognitive conception than the communicative, the explanatory, more with the
communicative than the cognitive.
Finally, there is a question about the function of language. This splits into two.
The Wrst concerns the use we actually make of language. I have gone along with
the folk idea that we use it to express thoughts. But I think Clark’s ‘‘supra-
communicative’’ view (1998) that we also use it in various ways as an aid to
thinking is very plausible. The second question concerns the biological function
of the capacity for language. Now to suppose that this capacity has a biological
function is to suppose that it is an adaptation rather than, say, a spandrel.28 I
think that this is an appealing supposition but I have not made it. If the capacity
is indeed an adaptation and hence has a function, that function must surely be to
express or aid thought (8.3). Perhaps this leans toward the communicative
conception.
I started the last chapter by arguing for intentional realism and for LET, the view
that our language expresses thoughts (8.1–8.2). These views led to my fourth
major conclusion: the issue of the psychological reality of language should be
investigated from a perspective on thought. The views, together with some
Gricean claims, also led to my Wfth major conclusion: thought has a certain
priority to language ontologically, explanatorily, temporally, and in theoretical
interest (8.3–8.7).
In the present chapter, I have adopted the popular RTM, the view that having
a thought involves standing in a certain functional relation to a mental repre-
sentation (9.1). I went on to argue for the controversial LOTH, the hypothesis
that mental representations posited by RTM are language-like. I have empha-
sized the support for LOTH that comes from its role in explaining thinking and
in explaining the meanings of mental representations (9.2). LOTH yields view
(c) of competence in a language: competence is the ability to translate back and
forth between mental sentences and the sounds (etc.) of the language (9.3). If
LOTH is correct, what language do we think in? I do not favor Public-LOTH,
28 This is implied by what Searle takes to be a common-sense assumption about language: ‘‘The
purpose of language is communication in much the same sense that the purpose of the heart is to
pump blood.’’ (1972: 16)
according to which a person largely thinks in her public language (9.4). However,
I have argued that her language of thought, her Mentalese, is syntactically similar
to her public language. I arrived at my Wrst tentative proposal based on (t):
a language is largely psychologically real in a speaker in that its rules are similar to
the structure rules of her thought. Once LOTH is assumed, various consider-
ations count in favor of this proposal but I give most weight to the following one:
the explanation of in virtue of what expressions of a public language have their
implicit and explicit syntactic properties yields the view that the structure of
Mentalese is similar to, if not the same as, the structure of the public language
(9.5). If we are to conclude that language is psychologically real in a robust way,
we need a powerful psychological assumption. Mine is (t).
We turn now to consider the bearing of these views of thought on the issue of
the language faculty, and to a comparison with the apparently very diVerent
Chomskian views of thought and its relation to language.
10
Thought and the Language Faculty
10.1 B R A I N I M PA I R M E N T A N D T H E I N D E PE N D E N C E O F
L A N G UA G E
The positions I have been presenting in this part of the book, climaxing with
(t)—the structure rules of a speaker’s language are similar to the structure rules of
her thought—seem to conXict with received Chomskian views. I shall consider
these apparent conXicts in this chapter. I start, in this section, with my view that
conceptual competence partly constitutes linguistic competence, which seems at
odds with the Chomskian idea of language being largely independent of thought.
There is an objection to my view which I noted earlier (8.3): the view is under-
mined by the well-known dissociation of cognitive impairment and linguistic
impairment. In this section I confront this objection, arguing that my view of
linguistic competence is not undermined. In section 10.2, I shall argue that,
contrary to received opinion, the evidence from brain impairment does not
support the existence of a language faculty. The evidence suggests, rather, that
there may not be such a faculty. Aside from that, I shall argue in section 10.3 that,
if we accept (t), we should not expect there to be a substantial language faculty
because there would not be much for that faculty to do. In sections 10.4 and
10.5, I shall turn to the apparently contrary views of Chomsky. I shall Wnish in
section 10.6 by considering recent claims by Georges Rey that are strikingly at
odds with the views I have presented.
The relatively uncontroversial folk idea, LET, that speakers have meaningful
thoughts which their language expresses, yields view (a) of competence in that
language (8.3): the competence is an ability to use a sound of the language to
express a thought with the meaning that the sound has in the language in the
context of utterance; and an ability (together with some pragmatic abilities) to
assign to a sound a thought with the meaning that the sound has in the
language in the context of utterance. This yielded an ontological priority of
thought over language. Competence in the language requires a certain concep-
tual competence, the competence to have thoughts with the meanings express-
ible in the language. But how could this be right? There is considerable
evidence that cognitive impairment and linguistic impairment do not go
hand in hand. As Pinker puts it: ‘‘There are several kinds of neurological and
genetic impairments that compromise language while sparing cognition and
vice versa’’ (1994: 46). This seems to demand that the linguistic and the
conceptual be independent.
Let us start with cases of linguistic impairment without cognitive impairment.
(1) Broca’s aphasia provides a famous example. People who suVer damage to
Broca’s area in the frontal lobe of the left hemisphere of the brain often cannot
talk properly; in particular, they have problems with grammar. Yet, their intel-
lectual functions that are not closely tied to language are all preserved (Pinker
1994: 48). (2) Next, some children who do not develop language on schedule are
diagnosed with ‘‘SpeciWc Language Impairment (SLI)’’. SLI runs in families and
it is plausible to think that it is hereditary. Indeed, the work of Myrna Gopnik on
SLI was hyped in the press as the discovery of ‘‘the grammar gene’’ (pp. 297–8).
People with SLI ‘‘speak somewhat slowly and deliberately . . . . They report
that ordinary conversation is strenuous mental work . . . . Their speech
contains frequent grammatical errors’’. Yet SLI ‘‘does not seem to impair overall
intelligence’’ (p. 49).
Let us assume that these cases really do demonstrate linguistic impairment
without cognitive impairment.1 It is easy to see that this alone does not under-
mine the story I have been telling. On my view, linguistic competence consists in
conceptual competence together with processing competence: the thoughts that are
the products of the conceptual competence have to be matched for meaning with
sentences in the language (8.3). So the linguistic impairment in question here
might be simply an impairment of processing competence leaving the conceptual
competence unaVected. In this way, a clever person could be linguistically
incompetent. The evidence of Broca’s aphasia and SLI does not show that the
impairment is not simply in processing competence. Indeed, given that these are
cases of people who are not, in general, cognitively impaired, where else could the
impairment be but in processing competence?
We turn next to cases of cognitive impairment without linguistic impairment,
to ‘‘linguistic idiot savants’’ (p. 50). Striking examples are provided by people
suVering from the genetically-based Williams syndrome. These people have an
unusual elWn-faced appearance and are ‘‘signiWcantly retarded, with an IQ of
about 50’’. Yet despite this they are ‘‘Xuent, if somewhat prim conversationalists’’
with a fondness for unusual words (pp. 52–3). A wonderful example of a savant is
provided by the brain-damaged Christopher who, despite severe mental handi-
caps, has superb linguistic skills, mastering foreign languages with ease (Smith
and Tsimpli 1995). Laura is another example: she has ‘‘complex linguistic
abilities’’ and yet ‘‘a testable IQ in the low 40s’’ (Yamada 1990: 3).
1 There is some controversy about this with SLI. Indeed, Cowie (1999) points out that ‘‘almost
nothing about SLI is uncontroversial’’ (p. 290) and goes on to give a nice summary of the
controversies (pp. 290–3).
Thought and the Language Faculty 165
Let us assume that these cases really do demonstrate cognitive impairment
without linguistic impairment.2 Once again, it is easy to see that the story I have
been telling is not undermined. On my view, the conceptual competence that
partly constitutes linguistic competence is a competence to have certain thoughts,
the thoughts expressible in the language. The competence is not one to reason
well with those thoughts, it is not a competence at thinking. So a linguistically
competent person could be stupid. To count against my view, we would need to
establish both (a), that the savants cannot think thoughts with certain meanings,
and (b), that sentences out of their mouths really have those meanings. The cases
provide no evidence of (a). Indeed, it would be hard to provide it. Our usual best
evidence of a person’s thoughts is what she says, and that clearly cannot be used as
evidence here. And it is not going to be easy to Wnd other behavioral evidence: her
failure to act on a putative thought may often arise from a failure of practical
reasoning. The evidence that we do have that the subjects are retarded shows only
that they are very poor at putting thoughts together in a rational manner, not that
they do not have thoughts. Of course, as Eric Margolis has pointed out to me, if
we assume a conceptual-role semantics certain failures of rationality will entail a
lack of certain concepts and hence an inability to have thoughts involving those
concepts. But then we need evidence that this semantics is right3 and also evidence
that the subject’s failures of rationality are ones that, according to this semantics,
entail a lack of concepts. Suppose, nonetheless, that we did have evidence of (a).
Then that would seem to undermine (b). For if a subject lacks the concept <F >
and hence has no thoughts about Fs then ‘F ’ out of her mouth will not be about
Fs and will not have the meaning it has in the language she seems to speak; it will
be ‘‘mere noise’’. To that extent she will not be competent in that language.
In sum, the dissociations demonstrated by cases of impairment are the wrong sort
to count against the view that the competence to have thoughts with the meanings
expressible in a language is a considerable part of competence in that language.
I turn now to the bearing of this view of linguistic competence, and of the
brain impairment evidence, on the established view that there is a language
faculty.
10.2 B R A I N I M PA I R M E N T A N D T H E L A N G UA G E
FAC U LT Y
That language expresses thought (LET) yields the view that conceptual compe-
tence partly constitutes linguistic competence. This conceptual competence is in
the central processor, the ‘‘general learning device’’ that does the thinking. So, an
2 Although, once again, there is a controversy nicely summarized by Cowie (1999: 293–7).
3 My 1996 takes a dim view of conceptual-role semantics, arguing that only such conceptual
roles as may determine reference constitute meanings.
important part of linguistic competence is in the central processor. Of course,
the conceptual competence is not suYcient for the linguistic competence: the
thoughts that are the products of the conceptual competence have to be matched
for meaning with sentences in the language. So linguistic competence requires a
processing competence as well as the conceptual one. That processing compe-
tence is surely not in the central processor and so might well be found largely in a
language faculty, a mental ‘‘organ’’ that is a relatively central module of the mind
distinct from the central processor, and that functions as the cognitive system for
language. But, clearly, by removing a large part of linguistic competence from the
language faculty we are downplaying its role. So even LET, a view committed to
thoughts but noncommittal about their nature, has an important consequence
for our view of the place of language in the mind.
Attention to the fact that a person typically has many competencies in a
language, not just one, leads to a further downplaying. Indeed, it raises a radical
possibility: there may not be a language faculty at all! An English speaker is
typically able not only to speak English but to understand it spoken, to write it,
and to read it. She may even be able to send it in Morse, understand it in Braille,
sign it in naval Xags, and so on. Each of these competencies requires that
conceptual competence be linked by an appropriate processing competence to
a symbolic input or output; each processing competence matches thoughts for
meaning with appropriate inputs or outputs. It is, of course, obvious that each
processing competence must diVer from the others in some respects; thus, the
competence to write must diVer from the competence to speak simply in virtue of
their diVerent outputs. But now suppose that each processing competence
diVered totally from all others. Each one is related to the same conceptual
competence, of course, but other than that it has nothing in common with the
others. Then there would be no language faculty. For, if there is a language faculty
it must be much more than a mere set of unrelated modality-speciWc processing
competencies. It must be a modality-neutral, relatively central, module of the
mind, a cognitive system partly responsible for all language processing.4 It must
be a ‘‘knowledge system’’ speciWc to language, somehow respecting the rules of a
language; indeed, on the favored view, the Representational Thesis (RT), the
language faculty respects the rules by representing them. On our supposition of a
variety of distinct processing competencies (perhaps distinct modules) there
would be no place for such a language faculty. The only central modality-neutral
competence underlying all of the particular linguistic competencies, the compe-
tence that all of these exemplify, would be the conceptual competence that resides
in the central processor. There would be nothing for a distinct, central, modality-
neutral, language-speciWc faculty to do.
This supposition is sure to seem too extreme. Speaking and understand-
ing spoken English obviously have something in common besides conceptual
4 ‘‘the language faculty is not tied to specific sensory modalities’’ (Chomsky 2000a: 121).
competence: they both involve the one linguistic modality, that of phones. But, if
we are to save the language faculty, we need something more signiWcant in
common than that. In particular, we need to Wnd something in common
among linguistic competencies that do not share a modality. For example, does
the competence to understand spoken English have anything in common with
the competence to read English (beyond the competence to think thoughts
expressible in English)? If we could Wnd something in common among all
processing competencies then that would play up the language faculty.
The main argument for the existence of a language faculty comes from
linguistic nativism, to be discussed in Chapter 12. Still, received opinion is
that the evidence from brain impairment supports that existence. And, at Wrst
sight, the evidence does indeed seem to provide this support. For, when damage
to an area of the brain impairs one processing competence it tends to impair
many. This suggests that those competencies are all located in that area. This
would seem to be evidence of the commonality we seek. So, let us consider the
impairment evidence.
Patients with Broca’s aphasia not only have trouble speaking they have trouble
understanding speech. Their comprehension impairment is less sweeping but is
noticeable when understanding requires ‘‘grammatical analysis’’ (Pinker 1994:
48). Furthermore, ‘‘their written communication follows this same production-
comprehension dissociation, with impaired writing but often less severe disturb-
ance to reading’’ (Swinney 1999: 31). Wernicke’s aphasia is also multi-modal. In
this aphasia ‘‘patients utter Xuent streams of more-or-less grammatical phrases,
but their speech makes no sense’’ (Pinker 1994: 310–11). Patients also have
‘‘a profound comprehension deWcit. Furthermore, both writing and (particularly)
reading are standardly highly impaired’’ (Swinney 1999: 31). Finally, Argye
Hillis and Alfonso Caramazza generalize as follows about a wide range of
aphasias, including Broca’s and Wernicke’s: ‘‘reading comprehension is generally
impaired at least to the degree of auditory comprehension, and written output is
typically impaired at least as much as spoken output, often mirroring the content
of speech’’ (2003: 177).
Deaf users of a sign language (like ASL) seem to provide further evidence of
areas of the brain devoted to language in a modality-neutral way. For, the eVect of
aphasias on the deaf is analogous to that on hearers:
right-handed deaf signers like hearing persons, exhibit aphasia when critical left-
hemisphere areas are damaged . . . language impairments following stroke in deaf signers
follow the characteristic pattern of left frontal damage leading to nonXuent output with
spared comprehension, whereas left posterior lesions yield Xuent output with impaired
language comprehension . . . the presence of strong biases that left inferior frontal
and posterior temporal parietal regions of the left hemisphere are well suited to
process a natural language independent of the the form of the language. (Corina
1999: 756–7)
So, at Wrst sight, brain impairments suggest that the various processing
competencies are located in Wernicke’s area and Broca’s area (and the arcuate
fasciculus that joins those areas). This is known as ‘‘the Wernicke–Geschwind
model’’. And this classical model may seem to support the idea that there is a
commonality between linguistic processing competencies and hence support
the existence of a language faculty. However, this Wrst sight is misleading in at
least three ways.
First, it has proved impossible to come up with generalizations that really tie
one brain function, let alone many, to a particular area of the brain. ‘‘The
Wernicke-Geschwind model . . . is now seen as oversimpliWed. Areas all over
the brain are recruited for language processing; some are involved in lexical
retrieval, some in grammatical processing, some in the production of speech,
some in attention and memory’’ (Dronkers 1999: 450). The introduction to a
recent state-of-the-art collection on language-brain research notes that ‘‘no paper
in the present collection focuses on or attributes any special role to Broca’s area or
Wernicke’s area . . . . There is a dramatic increase in attention to cortical areas out-
side the traditional perisylvian language zone . . . . The right hemisphere . . . is
being rehabilitated’’ (Poeppel and Hickok 2004: 9–10). Findings from a range of
very diVerent studies ‘‘do not allow the conclusion that Wernicke’s area (or any
other brain region) is alone responsible for a given cognitive process. For
example, although Wernick’s area appears to be important for the understanding
of words, it is unclear what—if any—semantic information is represented in this
area’’ (Hillis and Caramazza 2003: 181); ‘‘the lexical processing system is dis-
tributed over a large area of the left hemisphere’’ (Caramazza 1999: 470);
‘‘the syntactic capacities are not implemented in a single area . . . they constitute
an integrated system which involve both left and right neocortical areas, as well as
other portions of the brain, such as the basil ganglia and the cerebellum’’ (Moro
et al. 2001: 117). Ultimately, even the location of aphasias has proved diYcult:
‘‘despite early reports documenting a close relationship between aphasia classiW-
cation and site of lesion in chronic stroke, the correlation of aphasia type with
location of lesion has not withstood recent attempts at replication’’ (Hillis and
Caramazza 2003: 180). Pinker expresses his frustration at the situation: ‘‘the role
of Broca’s area in language is maddeningly unclear’’ (1994: 310); ‘‘no one really
knows what either Broca’s or Wernicke’s area is for’’ (p. 311); ‘‘pinning brain areas
to mental functions has been frustrating’’ (p. 314).
Of course, there could still be a language faculty even if the various linguistic
processing competencies are located in one person in one brain area, in another,
another, even if they are located in a person at one time in one area, at another
time, another, or even if each of them is located across many parts of the brain.
Mental processes are one thing, the physical sites at which they are located in the
brain, another. Nonetheless, it would have seemed nice evidence for the language
faculty if the various competencies had been, for the most part, located in all
people at all times in the one brain area; that would have been nice evidence that
there really was a language module. Furthermore, if there is to be a language
faculty it should be the case that in each person at each time the various
competencies are, to a signiWcant extent, located together. But there is no
evidence that they are.
The second way in which the Wrst sight is misleading is that even if the various
processing competencies could be located in one area according to, say, the
Wernicke–Geschwind model, their common impairment when that area is
damaged would still not establish that they have the commonality required for
a language faculty. For, they might all be totally distinct and yet damaged
together. Perhaps their locations are causally related in ways that explain the
common impairment. Hillis and Caramazza are interesting on this score. Con-
sidering Broca’s aphasia in stroke victims, they respond skeptically to the idea that
damage to ‘‘a central syntactic processor’’ might be responsible for the range of
impairment symptoms. ‘‘Rather, it is likely that the frequent co-occurrence of
these various symptoms reXects the fact that large, consistent regions of the brain
are typically supplied by distinct cerebral arteries, the occlusion of which results
in stroke. Suppose the larger area supplied by a vessel such as the superior division
of the left middle cerebral artery (MCA) consists of a number of smaller regions
each responsible for a speciWc language function (e.g. grammatical sentence
formulation, computation of syntactic relations, and articulation); occlusion
of this vessel would typically result in impairment of all three functions.’’
(2003: 179)
The third, and perhaps most important, way in which the Wrst sight is mislead-
ing is that even if a particular processing competence is typically impaired along
with others in an aphasia, the fact that it often is not suggests that there is no
commonality among processing competencies and hence no language faculty. The
tell-tale sign of this fact is that claims in the literature about one linguistic
impairment being accompanied by another nearly always include qualifying
expressions like ‘‘tend to’’, ‘‘standardly’’, ‘‘often’’, ‘‘generally’’, or ‘‘typically’’;
see above for some examples. The literature does not even clearly support the
intuitive view that the production and comprehension of speech have much in
common.
The evidence on overlap between the production and recognition systems is inconclusive.
If there are to be shared resources, the constraints of the two processes entail that the
sharing must be at a central level; however, while the evidence outlined [in this paper] is
compatible with a shared-resource account, it also does not rule out separation of the two
systems. (Cutler 1995: 124)
The cognitive processing systems responsible for comprehension and production
may nonetheless be distinct. Research on language disorders suggests a degree of inde-
pendence between them, because people with disorders of production can display
near-normal comprehension abilities, and vice versa. (Bock 1999: 456; see also Bock
1995: 205–6)
Not surprisingly, then, competencies involving diVerent modalities have been
found to be independent. Thus, the earlier-quoted passage from Hillis and Cara-
mazza, noting that aphasias aVecting the spoken language ‘‘generally’’ aVect
reading and writing, goes on as follows:
However, there are cases of pure (auditory) word deafness, in which comprehension of
spoken language is severely impaired, but comprehension of written language is intact
. . . There have also been reported cases in which written naming accuracy far exceeds
spoken naming accuracy. There are also a variety of patterns of pure reading impairment
(alexia), writing impairment (agraphia) or both (alexia with agraphia), associated with
diVerent lesion sites. (2003: 177)
In another work, Caramazza, notes other selective impairments:
Some brain-damaged patients are selectively impaired in retrieving only the orthographic
from (e.g., the spelling of the word chair) or only the phonological form of words
(e.g. the sound of the word chair). Patients of this type can be entirely normal in their
ability to understand and deWne words, but fail to retrieve the correct word form in one,
but not the other, modality of output . . . . there are patients who are impaired in
producing verbs only in speaking (they can write verbs and can produce nouns both
in speaking and in writing) and patients who are impaired in producing nouns only in
speaking; and there are patients who fail to understand written but not spoken
verbs. (1999: 469–70)
Caramazza rightly sees this account of highly selective impairments as a
challenge to ‘‘the view that there exists a modality-neutral lexical node mediating
between modality-speciWc lexical representations and word meaning’’ (p. 470).
Indeed, the account encourages the idea that there is a direct route between
thoughts and modality-speciWc inputs/outputs, a route that does not go though a
modality-neutral language faculty; i.e., it encourages our earlier supposition that
there is no such faculty.
More encouragement comes from the case of J.B.N.,
who spoke in Xuent jargon and failed to comprehend spoken words or sentences, despite
normal hearing and ‘‘early’’ auditory processing. Nevertheless, J.B.N had intact writing
and comprehension of written language. Thus, this patient did not have impaired
semantics, or word meanings, but was impaired in linking spoken words to their
meanings and vice versa, owing to poor blood Xow in the sylvian branch of the anterior
temporal artery, supplying Wernicke’s area. (Hillis and Caramazza 2003: 182)
Again this suggests fairly direct routes between thoughts and inputs/outputs,
routes that in J.B.N. have been damaged for spoken inputs/outputs but not for
written ones. It suggests that there is no place for a language faculty.
Objection.5 It is not appropriate to consider competencies in a written
language, let alone in Morse and so on, in looking for evidence of the language
5 I have heard this sort of objection a few times but have not come across it in the literature.
faculty. These competencies are ones that the child can only acquire by some
clearly central process; she cannot acquire them ‘‘naturally’’, under the con-
straints of Universal Grammar (UG), as she can competencies in a spoken
English and ASL. So these written competencies are not the business of the
language faculty.
Reply. (i) If there was a language faculty then it would be odd indeed if it was
not involved in our competencies in the written language. For if there was one it
would play a vital central role in processing the spoken communication of
thoughts. Yet the written language can, with one exception, serve to communi-
cate thoughts as eVectively as the spoken language. The exception is, of course,
that it is a rather slow method of communication. But it is hard to see why this
lack of speed would remove the need for the language faculty to play the same
vital central role in processing the written communication of thoughts. (ii) Let us
accept that competencies in a written language cannot be acquired naturally,
although one wonders whether and how this inability has really been established.
It is not clear why this diVerence between the written and spoken languages
should be relevant to the nature of the competencies in them. Certainly, if there
was a language faculty, our inability to naturally acquire the written competencies
would not alone show that the language faculty did not play that vital role in
written communication. For, the inability might be explained not by the lack of
involvement of the language faculty but rather by some incidental practical
problems. Perhaps, the slowness of written communications is what causes the
inability. At least, more work has to be done on the objection if it is to undermine
the view that if there was a language faculty it would be involved in written
communication.
In sum, the evidence from brain impairment seemed, at Wrst sight, to support
the idea of a language faculty. Where brain damage impairs one processing
competence it tends to impair many, suggesting the commonality among com-
petencies that is necessary for there to be a modality-neutral language faculty. But
this Wrst sight is misleading. First, a particular area of the brain cannot be tied
tightly to one brain function, let alone to many. Second, even if the various
processing competencies could be located in one area, their common impairment
when that area is damaged is quite compatible with their being distinct. Their
functioning might depend, for example, on a particular cerebral artery that has
been damaged. Third, even if a particular processing competence is typically
impaired along with others, the fact that it often is not suggests that there is no
commonality among processing competences. So it turns out that the evidence
from brain impairment provides no persuasive evidence in favor of the existence
of a language faculty. Furthermore, the highly selective nature of some impair-
ments suggests that there is a direct route between thoughts and modality-speciWc
inputs/outputs that does not go though a modality-neutral language faculty. All
in all, the evidence from brain impairment counts much more against the
existence of a language faculty than for it.
10.3 T H O U G H TS A N D T H E L A N G U AG E FAC U LT Y
The evidence against the language faculty that we have been considering is only
suggestive, of course: it is very far from conclusive.6 So in this section, I want to
consider the likeliness of there being a substantial language faculty, given certain
assumptions about the nature of thoughts. Given these assumptions, should we
expect to Wnd substantial commonalities among processing competencies?
Suppose that we move beyond LET, which has governed the discussion so far,
and adopt the minimal RTM view of thought: having a thought involves
standing in a relation to a mental representation. With RTM goes view (b) of
competence: competence is the ability to translate back and forth between mental
representations and the sounds etc. of the language (9.1). RTM takes thoughts to
involve representations but is noncommittal about the nature of those represen-
tations. So the representations might be map-like. If they were, the various
processing competencies would have the job of translating between map-like
structures and very diVerent linguistic structures. This job seems rather formid-
able. Perhaps then we should expect that there would be a modality-neutral
linguistic ‘‘knowledge system’’ that each of the processing competencies could call
on to perform the job. Perhaps we should expect that there would be substantial
commonalities constituting a substantial language faculty. For, there would seem
to be a lot for a language faculty to do.
However, the likeliness of a substantial language faculty seems to disappear if we
move beyond LET in two further ways, as I have argued we should. LETyielded an
ontological priority of thought over language. The Wrst move is to follow Grice in
accepting the relatively, although certainly not entirely, uncontroversial explana-
tory priority of thought over language (8.4, 8.6–8.7). The second move weds this
acceptance to the decidedly controversial Language-of-Thought Hypothesis
(LOTH) and view (c) of linguistic competence: competence is the ability to
translate back and forth between mental sentences and the sounds etc. of the
language (9.3). This wedding yields the following version of the Gricean explana-
tory priority: the meanings of words in a language are ultimately explained by the
meanings of the mental words that they express; and the syntactic structures of
sentences in the language are explained by the structures of mental sentences that
the sentences express. If this is so, I argued (9.5), we have good reason to think that a
speaker thinks in a Mentalese that has a syntax very like that of her natural
language; we have good reason to accept (t) and hence my Wrst tentative proposal.
That greatly aVects the job that processing competencies have to do. Instead of the
formidable job of translating between, say, map-like structures and linguistic
6 One piece of evidence that may count in favor of a language faculty is evidence that visual word
recognition is not ‘‘direct’’ but ‘‘phonologically mediated’’. But the evidence is not conclusive, the
whole issue being ‘‘a source of seemingly endless controversy’’ (Seidenberg 1995: 151).
structures, the competencies have the job of translating between similar language-
like structures. Intricate as this job doubtless is for each competence, it does not
seem that it would be helped by a modality-neutral linguistic ‘‘knowledge system’’;
it is hard to see what a substantial language faculty would have to do.
In the last section, I described evidence suggesting that there are not the
commonalities between the various processing competencies required for there to
be a language faculty. In this section, I have argued that, if we accept (t), we should
not expect to Wnd commonalities that would constitute a substantial language
faculty. For, given (t), it is likely that there is not much to an English speaker’s
syntactic competence in her language beyond her competence in her Mentalese, her
conceptual competence. This conceptual competence can carry all this linguistic
weight because it is an English-oriented competence. We have arrived at my second
tentative proposal: there is little or nothing to the language faculty.
Let me summarize the key steps to this proposal. First, LET alone downplays
the language faculty because according to LET a certain conceptual competence
is an important part of a person’s competence in a language. That part is in the
central processor not a language faculty. If there is a language faculty it must be
found in the other part of a linguistic competence, a processing competence.
Second, Wnding this depends on the various processing competencies having a
substantial part in common. Yet the evidence from brain impairments suggests
that they do not have this commonality and hence that there is no language
faculty. Third, if we wed the Gricean explanatory priority of thought over
language to LOTH yielding (t), we should not expect there to be a substantial
language faculty because there is nothing much for the language faculty to do.
This case for the second tentative proposal rests a bit on LOTH but does not
totally depend on it as does the case for the Wrst (9.5).
The proposal that there is not a substantial language faculty has to be tentative
because the case for it is far from overwhelming. Still, if my line of reasoning is
correct, the contrary view that there is a substantial language faculty should be
even more tentative because the case for it is even weaker. That contrary view
requires that there be a substantial commonality among processing competen-
cies. Yet, the evidence from brain impairment does not support this common-
ality. And if (t) is true the commonality seems unlikely. Even the fairly innocuous
LET leads to a downplaying of the language faculty.
Chomsky’s nativist discussion of language acquisition may seem to provide a
powerful objection to this downplaying, for Chomsky argues against placing our
linguistic competence anywhere but in a distinct language faculty. He argues that
the rules (principles) of language are so peculiar that we would not expect to Wnd
them anywhere but in a distinct language faculty, in particular not in the central
processor. Furthermore, no general learning device of the sort found in the
central processor could explain the acquisition of a language, a point that
seems to be supported by Ray JackendoV ’s neat ‘‘Paradox of Language Acquisi-
tion’’. I shall later consider and reject these arguments (12.5). Indeed, I shall
argue that the discussion of the last two chapters has important consequences for
nativism and language acquisition that count further against there being a
substantial language faculty (12.4).7
I turn now to consider Chomsky’s apparently contrary views on some other
matters discussed in this part of the book. In the next section I consider his views
on thoughts. In the following section I consider his views on conventions and
idiolects.
10.4 C H O M S K Y O N T H O U G H TS
I have no Wrm view about the extent of Chomsky’s disagreement with the claims
made in this part of the book. He does seem to accept LET, the folk idea from
which I started: ‘‘a language is a particular way of expressing thought and
understanding the thought expressed’’ (1991a: 8). He is highly critical of various
aspects of the Gricean story but these are not aspects that I have adopted (8.4).
Thus, against attempts to explain meanings in terms of a speaker’s communica-
tive intentions he rightly points out:
Under innumerable quite normal circumstances—research, casual conversation, and so
on—language is used properly, sentences have their strict meaning, people mean what
they say or write, but there is no intent to bring the audience (not assumed to exist, or
assumed not to exist, in some cases) to have certain beliefs or to undertake certain
actions. (1975b: 62)
And he emphasizes, as I do, the Gricean reliance on unexplained thought
contents (1975b: 65–7). I haven’t found in Chomsky any discussion of the
part of the Gricean story that I have adopted, the part that yields the explanatory
priority of thought over language. Indeed, Chomsky has surprisingly little to say
about thoughts at all. What he does say certainly does not suggest that he would
accept that thought has the various priorities over language summed up in my
Wfth major conclusion. Indeed, he is naturally construed as thinking that lan-
guage is largely independent of thought. And any downplaying of the language
faculty is decidedly unChomskian. Finally, the signs are that he does not hold
LOTH.8 If he does not, he could not relate the syntax of language closely to that
of thought, as (t) does.
7 ‘‘Cognitive linguistics’’, which is opposed to generative linguistics, also rejects the language
faculty: ‘‘language is not an autonomous cognitive faculty . . . conceptual structures and processes
proposed for language should be essentially the same as those found in nonlinguistic human
cognition’’ (Croft and Cruse 2004: 328). Cognitive linguistics arose out of an opposition to
truth-conditional semantics that seems to me very mistaken (1996). However, its idea of a
‘‘production grammar’’, stimulated by the need to accommodate idioms, is certainly interesting.
8 Although Carruthers and Boucher claim that ‘‘Chomsky (1957) argued that thought had to be
translated out of language into a ‘language of thought’ in the process of linguistic understanding’’
(1998b: 9). Chomsky is certainly dubious of a Fodorian ‘‘universal Mentalese’’ (2000a: 176–8). This
is not the sort of Mentalese that I am urging because I accept that there may be a great deal of causal
Any theory of language that accepts the reality of meaningful thoughts must
relate those thoughts somehow to meaningful language. Making this connection
seems particularly important for a mentalistic theory of language like Chomsky’s.
For, thoughts are clearly and indubitably in the mind.9 We would expect the
relation between thought and language to be at center stage in a discussion of the
psychological reality of language. So, it is strange to Wnd that Chomsky does not
say much about the connection.10
I am talking here of thoughts as mental states. Frege talked of thoughts as
abstract objects that are grasped by people in these mental states. These putative
objects are more usually called ‘‘propositions’’. Chomsky clearly doubts that such
objects have any place in the study of language.11 I agree. What is surprising is
that he says so little about thoughts as mental states.
Passages like the following indicate a rather diVerent view of the relation
between language and thought from the ones we have outlined:
[The language faculty] is internally highly modularized, with separate subsystems

of principles governing sound, meaning, structure and interpretation of linguistic
expressions. These can be used, to a suYcient degree, in thought and its expres-
sion . . . (1991b: 51)
Elsewhere he includes ‘‘conceptual–intentional systems’’, along with ‘‘articulatory

and perceptual systems’’, among ‘‘performance systems’’ that use ‘‘expressions
generated by the I-language’’. The conceptual–intentional systems use the
‘‘meaning’’ provided by the LF properties of an expression (1995b: 19–20;
2000a: 124–5). A message I take from such passages is that the rules or principles
of the I-language are distinct from any structure rules governing thought but
they, and the expressions they generate, are nonetheless used in thought. But how
could they be thus used? We can see, of course, how language is used by thought
in the articulatory and perceptual systems, for that use is involved in expressing
thoughts and interpreting utterances. So we understand the last ‘‘in expression’’
part of Chomsky’s ‘‘used . . . in thought and its expression’’. But what other uses
influence of language upon thought (9.5). The reasons for Chomsky’s doubts do not count, so far as
I can see, against my picture of Mentalese and its relation to language.
9 In claiming this I do not mean to deny, what indeed I believe, that the identity of a thought
depends partly on its relations to the world outside the mind.
10 Nor do others. Fodor remarked some time back that ‘‘relatively little attention has been paid
the question of how models of language articulate with theories of cognition’’ (1975: 100). And this
has continued: ‘‘the relationship between language and thought has been relatively little discussed in
recent decades’’ (Carruthers and Boucher 1998b: 2).
11 ‘‘Such understanding as we have of these matters does not seem to be enhanced by invoking
thoughts that we grasp . . . ’’ (1994: 25–6). ‘‘The notion ‘common store of thoughts’ has no
empirical status, and is unlikely to gain one even if the science of the future discovers a reason,
unknown today, to postulate entities that resemble ‘what we think (believe, fear, hope, expect, want,
etc.)’ [The view that there is a common store of thoughts] seems groundless at best, senseless at
worst’’ (1996: 47).
are there that he could have in mind for the distinct conceptual–intentional
system? What is the Wrst ‘‘in thought’’ part of this claim? We need to use our
grasp of the I-language to express thoughts and understand their expression but
what else could we use it for? If the popular RTM is correct, thought must be in a
representational system, either a language or a system of some other sort. So
thought itself is already governed by a system of structure rules. What use could
thought then have for a distinct system of linguistic rules? According to (t), of
course, the two systems are similar and so one might perhaps say, on the strength
of this, that the linguistic rules are thus used in thought. But that is presumably
not what Chomsky has in mind. Even if we do not adopt RTM, we must
see thoughts as already having meanings. So what use could thoughts make of
meanings generated by the I-language other than the use of expressing
them? Finally, we wonder how the picture of one module governing the meaning
and another, the structure of a linguistic expression, can yield a plausible view of
the way a thought, which already has a meaning and structure, leads to that
expression?12
A diVerent view of the relation between language and thought may also
underlie Chomsky’s view of the prospects for his task (iii), the psycholinguistic
task of explaining language perception and production. ‘‘The perception prob-
lem has to do with how we interpret what we hear. . . . The production
problem, which is considerably more obscure, has to do with what we say and
why we say it. We might call this latter problem Descartes’s Problem’’ (1988:
4–5). Chomsky’s thinks that this problem, unlike the part of the perception
problem that concerns the parser, may well be a mystery, ‘‘beyond the range of
our understanding’’ (1991b: 41). The diYculty with the problem, pointed out by
Descartes, is ‘‘that normal human speech is unbounded, free of stimulus control,
coherent and appropriate . . . what we might call ‘the creative aspect of language
use’ ’’ (p. 40).13 But it is surely a mistake to identify the production problem of
task (iii) with Descartes’s problem.
We should start by distinguishing the creativity that Chomsky has in mind
from another that he clearly does not, what we might call ‘‘the creativity of
language’’. The latter creativity is to be found in the productivity of language:
someone competent in the language has the capacity to produce any one of an
indeWnitely large number of novel sentences. Descartes’ problem is concerned
with a creativity in language use: roughly, we might say just about anything at just
about any time. But this problem seems to reduce to two other problems, neither
of which are the production problem of task (iii). The Wrst of these is the problem of
explaining the creativity of thought. For our thoughts are indeed, in some sense,
‘‘unbounded, free of stimulus control’’ and, as a result, so too are the sentences
12 Chomsky’s modularized picture is, as Larson and Segal say, ‘‘broadly assumed in modern
linguistic theory’’ (1995: 72).
13 On some of this creativity problem, see Kasher 1991b.
that express our thoughts.14 If we could explain this creativity of thought we
would have partly explained ‘‘the creative aspect of language use’’. The creativity
of thought may well be mysterious but it surely has nothing to do with psycho-
linguistics and task (iii). The second problem underlying Descartes’ problem is
that, whatever we think, we seem free to express it or not to express it. If we could
explain this apparent freedom of speech as well as explain the creativity of
thought, we would have completed the explanation of the creativity of language
use. Yet explaining this apparent freedom of speech is again nothing to do with
psycholinguistics. The apparent freedom is just an example of the apparent
freedom of action in general; as Chomsky puts it, ‘‘our actions are free and
undetermined’’ (1988: 147). So this second part of Descartes’ problem is nothing
but the famous problem of freewill. This problem may also be beyond us but it is
a philosophical problem not a psycholinguistic one. The strictly psycholinguistic
problem of task (iii) starts only after the thought to be expressed and the
intention to express it have been formed: ‘‘the process begins with a communi-
cative intention, a message, that stands at the interface between thought and
language’’ (Bock 1999: 453). Descartes’ problem, the problem of explaining the
formation of the thought and the formation of the communicative intention, are
simply not the concern of psycholinguistics. Just as the perception part of task
(iii) seeks to explain how we move from hearing an utterance to the thought that
is its interpretation, the production part of the task seeks to explain how we move
from the intended expression of a thought to the utterance that is its expression.
This task may well be very hard, but it does not seem particularly mysterious.
And it does not seem to pose any interesting problem of creativity.
If one had anything like the view of the relation of language to thought that
I have been urging, even just LET, one would surely give pride of place to the
creativity of thought in discussing the creative aspect of language use. Yet
Chomsky does not even mention it.15
In sum, it is clear that Chomsky would reject position (t) with its dramatic
lessening of the role of the language faculty. Even the much less radical claims of
ontological and explanatory priority for thought seem at odds with his views. But
perhaps this is an illusion, given that his view of thoughts and their relation to
language remains obscure.16
14 The unboundedness of thought is strangely overlooked by Hauser et al. (2002) in arguing that
the human recursive capacity to generate an infinite range of expressions, unlike the human
conceptual–intentional system, is unique to the species. Yet humans are unique in their recursive
capacity to generate an infinite range of thoughts, however much else humans may share conceptu-
ally with other animals. And if they did not have that conceptual capacity for ‘‘discrete infinity’’ it
would be very odd that they had the linguistic capacity for it (8.3).
15 However, he has this to say later in speculating about the origins of language ‘‘this small
mutation . . . giving us the capacity to think creatively and to speak creatively’’ (1988: 183–4;
emphasis added).
16 In a peer commentary, Roger Schank criticizes Chomsky for divorcing ‘‘the conceptual from
the language’’ (1980: 37). Chomsky does not reply to this criticism in his response (1980c: 53–4).
One wonders why Chomsky says so little about thoughts. Perhaps he thinks,
reasonably enough, that it is diYcult to Wnd anything that can be conWdently
said. But, as I am arguing, it is also diYcult to Wnd anything that should be
conWdently said about the place of language in the mind. Indeed, if my fourth
major conclusion is right, the latter is more diYcult, because, according to that
conclusion, a view of the place of language in the mind should be heavily
dependent on a view of thoughts.
10.5 C H O M S K Y O N C O N V E N T I O N S A N D I D I O L E C TS
In this section I shall explore Chomsky’s views on four other matters that seem
somewhat at odds with the Gricean view of the relation of thought to language
that I have urged (8.4, 9.5).
(i) Chomsky seems to doubt that there are regularities of the sort needed for
‘‘the norms and conventions of language’’. ‘‘If by ‘conventions’ we mean some-
thing like ‘regularities in usage’, then we can put the matter aside; these are few
and scattered’’ (1996: 47; see also 1980a: 81–3).
(ii) Furthermore, such conventions as there are, in an ordinary sense, do
not have ‘‘any interesting bearing on the theory of meaning or knowledge of
language’’ (1996: 48).17
(iii) Chomsky says of communication that ‘‘it does not require shared ‘public
meanings’. . . . Nor need we assume that the ‘meanings’ . . . of one participant
be discoverable by the other. Communication is a more-or-less matter, seeking a
fair estimate of what the other person said and has in mind’’ (1994: 21).18
(iv) Chomsky emphasizes that linguistics should not be concerned with a
language like Chinese or English. A language in that sense has a ‘‘sociopolitical
dimension’’ which makes it an unsuitable notion for the science of language
(1986: 15–16); ‘‘the notion ‘common language’ has no place in eVorts to
understand the phenomena of language and to explain them’’ (1996: 47). Rather
our primary concern should be, in eVect, with idiolects (1986: 16–17; 1996: 48).19
17 Fodor does not agree with (i) and (ii); see his1975, ch. 3 and the following: ‘‘think of a natural
language as a system of conventional vehicles for the expression of thoughts (a view to which I know
of no serious objections)’’ (1981a: 196).
18 Cf: ‘‘we should give up the attempt to illuminate how we communicate by appeal to
conventions’’ (Davidson 1986: 446).
19 The view that linguistics should be about idiolects is standard in Chomskian circles: ‘‘in
linguistic theory . . . the object of study is the idiolect. . . . References to community languages, or
to dialects of languages, if they are needed at all, are in any case references to derivative things,
characterized loosely in terms of the overlapping idiolects of members of groups whose individuals
are in frequent serious communication with each other’’ (Higginbotham 1989: 155); ‘‘A linguistic
theory is correct exactly to the extent that it is the explicit statement of a body of linguistic
knowledge possessed by a designated individual language-user’’ (Barber 2001: 263); ‘‘the proper
object of linguistic inquiry is a speaker’s idiolect’’ (Barry Smith 2001: 285).
From my perspective these claims are puzzling. To bring out why, I need some
more background. On the Gricean story I urged, conventions seemed to come
into the picture when we take account of the distinction between the speaker
meaning of an utterance—what the speaker means by it—and its conventional
meaning on the occasion—what, in the context, it means according to linguistic
conventions that the speaker is participating in. Clearly, the speaker meaning is
the one that matches the thought meaning. I have said, rather vaguely, that the
conventional meaning of an utterance in a language is explained somehow in
terms of regularities in speaker meanings (8.4). And I have not bothered to
distinguish it from the literal meaning of the utterance. It is time to say a little
more about these matters.
Mostly we can identify the conventional meaning of an utterance with its
literal meaning, but we cannot always do so. Idiolects are occasionally a bit
eccentric: the literal meaning of an expression in a person’s idiolect may not be a
meaning it has according to any linguistic convention. Donald Davidson brings
this out nicely in his discussion of Mrs. Malaprop’s ‘‘a nice derangement of
epitaphs’’ (1986): what she literally means is ‘‘a nice arrangement of epithets’’ and
yet her words do not mean this according to any convention.
What is it for a certain expression to have a certain literal meaning in a person’s
idiolect? On my view, the expression has that meaning in virtue of that person
being disposed to associate the expression with that meaning in the production
and comprehension of language: she is disposed to use that expression to express
a concept (a part of a thought) with that meaning; and she is disposed to interpret
that expression by assigning a concept with that meaning to it.20 Occasionally a
person will not do what she is normally disposed to do; she will deliberately
assign another meaning to an expression, as in a metaphor; or she will make
a performance error; or she will make an adjustment in understanding
Mrs. Malaprop. In these cases, an expression will have a speaker or audience
meaning that is diVerent from its literal meaning in the person’s idiolect.
How does the conventional meaning of an expression in a community relate to
the literal meanings it has for speakers in that community? Suppose that speakers
in the community share a disposition to associate the expression with a certain
concept meaning, thus generating a regularity of so associating it. Then the
speakers share a literal meaning. If this sharing is partly explained by the
appropriate causal relations between the speakers’ dispositions, then that literal
meaning will be the conventional one in the community. It is, of course, hard to
say precisely what causal relations are appropriate for the shared literal meaning
to be conventional, but the center of what has to be said is that any speaker has
20 This account would need to be modified to deal with an indexical which has a type meaning
that yields token meanings that vary with contexts. We might say that the indexical has that type
meaning in virtue of the person being disposed to use it to express concepts with token meanings
that are thus dependent on the context; and being disposed to interpret indexicals by assigning a
concept to it that is thus dependent on context.
her disposition because other speakers have theirs and hence regularly use the
expression with that meaning; there is some sort of mutual understanding (Lewis
1969, SchiVer 1972). The norm is for speakers in a community to share a literal
meaning because they stand in the required causal relations. As a result, the literal
meaning of most expressions for most speakers will be the conventional meaning
of those expressions in the speakers’ community. Mrs. Malaprop is an exception.
Before considering Chomsky’s four claims, we should attend to Stephen
Laurence’s surprising view of these matters (1996, 1998). He rejects the idea
that there are any linguistic conventions of the sort just described and hence
rejects ‘‘convention-based semantics’’. He thinks that our semantic theory should
be concerned simply with literal meanings in an idiolect. His rejection is largely
based on criticisms of Lewis’ account of conventions in general and of linguistic
conventions in particular. But this is not a good reason for the rejection. Lewis
starts by pointing out that it is a ‘‘platitude that language is ruled by convention’’
(1969: 1). Laurence takes the platitude to be that linguistic properties are not
intrinsic properties: ‘‘I completely agree that language is conventional in the
trivial sense: it isn’t an intrinsic property of the noise ‘chocolate’ that it means
chocolate’’ (1996: 272). But this is clearly not the right way to understand the
platitude because very many properties that are not intrinsic are not conven-
tional: being a parent, being a moon, and so on. And it also clearly not what Lewis
has in mind. Lewis does not mention intrinsic properties in describing the
platitude, saying rather that ‘‘we who use [words] have made them mean what
they do because somehow, gradually and informally, we have come to an
understanding that this is what we shall use them to mean’’ (1969: 1). In brief,
the platitude is that linguistic conventions are created and sustained by regular
practice together with some sort of mutual understanding. (They might also be
created by explicit agreement, of course, but they rarely are.) I sympathize
with Laurence’s criticisms of Lewis’ highly intellectualized account of these
conventions. And I have just acknowledged the diYculty of coming up with a
satisfactory account. But this should not shake our conviction that there are such
conventions, not only in language but in many other facets of life. That there are
is indeed a platitude, something that should be abandoned only in the face of
powerful evidence.
Against this background, I turn to Chomsky’s four claims. Consider claim
(i): there are few regularities in usage and hence few linguistic conventions.
Chomsky thinks that this should be a truism (1996: 48). Yet it is surely
astounding. Although it is possible in principle to have an idiolect that is not
based on regularities in usage and hence is entirely unconventional, nobody in
fact comes close to having such an idiolect. Of course, if Chomsky is right, a great
deal of syntax is innate and hence not conventional. (We should note that the
innate syntax will certainly lead to regularities albeit not conventional ones.) Still,
the syntactic diVerences between public languages show that much syntax is not
innate. These diVerences are captured, on the received Chomskian view, by
diVerent settings of ‘‘parametric values’’. Very occasionally an idiolect’s parameter
settings may be eccentric but almost always they will be conventional. Thus most
people in the USA participate in parameter-setting conventions that lead them
to speak an SVO language; most people in Japan participate in parameter-setting
conventions that lead them to speak an SOV language. The ubiquity of regular-
ities in usage and linguistic conventions is even more apparent when we turn
to words. Despite Mrs. Malaprop, almost any word in anyone’s idiolect is
conventional:21 if we take any such word, we are almost certain to Wnd that its
literal meaning is the meaning it has conventionally among some group of people
with whom the person is in touch. This does not, of course, imply that the group
will be the same for each expression. Indeed, we can expect to Wnd considerable
variation; thus, I share my meaning of ‘wowser’ with nearly all Australians and
hardly any Americans; I share my meaning of ‘chutzpah’ with many Americans
and few Australians; I share a meaning of ‘wet’ with most English, many
Australians, and few Americans; I share my meaning of ‘disinterested’ with a
rapidly diminishing subgroup of Australians, Americans, and English. In brief,
I would say, in contrast to (i), that almost all noninnate syntax and almost all the
word meanings of anyone’s idiolect are conventional.
There is indeed something a little paradoxical about denying the frequency of
linguistic conventions. The linguistic method of consulting the intuitions of
linguists and other speakers to discover facts about a language, discussed in
Chapter 7, presupposes masses of conventional regularities among them, even
while allowing for some diVerences in idiolects. Books are written and papers are
given about expressions in this or that language, all of which presuppose a great
deal of regularity in usage among speakers.
Turn next to Chomsky’s claim (ii): conventions have no interesting bearing on
the theory of meaning or knowledge of language. Chomsky thinks that this also
should be a truism (1996: 48). Yet, if it were true it would seem to be at odds
with what most linguists are actually doing. For, what they are mostly doing is
theorizing about the largely conventional syntactic and semantic properties of
expressions (Ch. 2). And they are right to be doing so, in my view. Conventional
meaning is important to theory in at least four ways.
First, conventional meaning plays a role in explaining the acquisition of a
person’s idiolect: an expression typically has a certain literal meaning in a person’s
idiolect as a causal result of its having that meaning conventionally in a local
group. Acquiring a language is almost entirely a matter of moving, under the
causal inXuence of primary linguistic data that are (performance errors aside)
instances of local linguistic conventions, from an innate ‘‘initial state’’ of readi-
ness for language to a ‘‘Wnal state’’ of participation in those very linguistic
conventions.
21 In claiming this I do not mean to deny that there may be some innate constraints on word
meanings. However, for ease of exposition, it will do no harm to ignore this possibility.
Second, as I have argued earlier (8.7), if a person’s concept ‘‘borrows’’ its
reference—a common occurrence nicely captured by Putnam’s talk of ‘‘the social
division of labor’’ (1975: 227–8)—then its meaning is partly explained by the
conventional meaning of the public word that lends the reference.
Third, and more important, consider why we are theoretically interested in
language in the Wrst place. I have argued that we are interested in it because
linguistic expressions play a certain extraordinarily important role in our lives
(8.5). That role is as a guide to thoughts, thoughts which explain behavior, serve
as our main source of information about the world, and perhaps have other
functions as well (8.1). The property of a linguistic expression that enables it to
perform its important role is its largely conventional meaning. It is because it has that
meaning that it is so worthwhile to produce and so readily understood. That
meaning is constituted by conventional word meanings and a syntax that is
conventional to a considerable extent. On hearing an utterance, a person who
participates in the conventions it involves can, with the help of pragmatic abilities
that determine indexical references and remove ambiguities, immediately grasp
the thought or message that the utterance expresses. It is in virtue of those largely
conventional properties that the utterance is such a quick and eVective guide to
the speaker’s thought. Our theoretical interest in language is in explaining the
nature of these largely conventional properties that enable language to play this
guiding role.22
Finally, the fourth way in which conventional meaning is important emerges
in the discussion of claim (iii) to which I now turn. According to this claim,
communication does not require that a person shares a meaning with another nor
that the other’s meaning is discoverable. The position I argue for in Coming to
Our Senses (1996) is at odds with this. According to that position it is, as a matter
of fact, common for people to share thought meanings and utterance meanings.
These shared meanings are ascribed by attitude ascriptions for our semantic
purposes of explaining behavior and being guided to reality. Because there is an
‘‘intimate link’’, sometimes an identity, between the meaning ascribed and a
meaning of the content sentence of the ascription, ascribing meaning in this
usual way requires the ascriber and the ascribee to, near enough, share a meaning.
The eYcacy of using thought ascriptions to serve those semantic purposes
requires that thought meanings be both discoverable and widely shared. The
communication of thoughts by language is certainly possible without shared
utterance meanings—think of many cases where speaker meaning diverges
from conventional meaning—but the eYcacy of using language as evidence of
thought meanings requires that utterance meanings be widely shared, as we have
in eVect noted in the last paragraph. Finally, the importance of shared meanings
is nicely demonstrated by a type of misunderstanding: people take themselves to
share a meaning but they are wrong, often with disastrous consequences.
22 So ‘‘convention-based semantics’’ does not lack motivation; cf. Laurence 1996: 270.
If all this is right, claim (iv) is largely mistaken. First, linguistics should not be
primarily concerned with idiolects. The concern should be, and I think is, with
the meanings of linguistic expressions, hence their syntactic properties, that are
shared in the idiolects of a group of people, shared largely because the meanings
are conventional in that group but also, in the case of syntactic properties,
perhaps partly because they are innate. Although the idiolects of two members
of a group, X and Y, could in principle be the same, they are obviously in fact
always a bit diVerent. But any meaning that X ’s idiolect does not share with
Y ’s will typically be shared with the idiolect of someone else, Z, and many others.
The concern of linguistics should be with meanings shared between the idiolects
of X and Y, other meanings shared between the idiolects of X and Z, and so on.
For, it is because of these shared meanings that the expressions play their
important role of making people’s thoughts accessible to others in communi-
cation. And it is because the meanings play this role that explaining them is so
theoretically interesting. Insofar as an idiolect does not share meanings with any
other idiolect it is of little linguistic interest (although it may be of medical or
psychological interest).This yields my sixth major conclusion: the primary concern
in linguistics should not be with idiolects but with linguistic expressions that
share meanings in idiolects.
Second, the dismissal of concern with ‘‘a common language’’ like English
seems overblown, at least. The members of any group that share a meaning of
one linguistic expression tend to share meanings of a vast number of others and it
is convenient, on the basis of this, to follow the custom of classifying sets of these
expressions with shared meanings as English, Spanish, and so on. The classiWca-
tion is bound to be a bit vague but no more so than many scientiWcally
appropriate ones. And such classiWcations seem to be useful in linguistics, for
linguistics books and articles are replete with them.23 Are these classiWcations
mere manners of speaking that can be paraphrased away when the serious
linguistic work is to be done? I think not: they are necessary for the linguist to
identify what she is talking about, to identify the subject matter. For, the subject
matter is the shared meanings and syntactic properties of linguistic expressions in
a certain group of people. To identify this subject matter she has to identify the
23 Thus, picking four books almost at random from my shelves I found the following. (i) The
first few pages of Haegeman’s GB textbook (1994) have many uses of ‘‘English’’ to classify shared
meanings. Then ‘‘English’’ is compared with ‘‘Italian’’; for example, ‘‘In Italian a subject of a
subordinate clause can be moved to the main clause domain across the overt conjunction che,
corresponding to that; in English this is not possible’’ (p. 20). And then with Spanish and French (p.
23). And so on throughout the book. (ii) A book following the generative approach is called ‘‘English
Syntax’’ (Baker 1995). (iii) A long article, ‘‘X-bar Theory and Case Theory’’ (Webelhuth 1995b) is
full of references to various languages (and a language group); for example, ‘‘English’’, ‘‘German’’ (p.
40), ‘‘Icelandic’’ (p. 50), ‘‘French’’ (p. 51), ‘‘the Australian aboriginal language Warlpiri’’ (p. 65),
‘‘Japanese’’ (p. 66), ‘‘Arabic’’ (p. 76), ‘‘Welsh’’ (p. 78), ‘‘Breton’’ (p. 78), ‘‘Hindi’’ (p. 80), and
‘‘Romance’’ (p. 81). (iv) Early in Chomsky’s Managua Lectures we find: ‘‘These sentences illustrate a
certain feature of Spanish not shared with such similar languages as Italian’’ (Chomsky 1988: 12);
‘‘Here we observe a difference between English and Spanish’’ (p. 13).
group. Mostly the group can be identiWed well enough with a term like ‘English’
but sometimes we need a less precise term like ‘Romance’ and sometimes a more
precise one like ‘Australian–English’ or ‘the Somerset dialect of English’.
A ‘‘sociopolitical dimension’’ does occasionally intrude into such classiWcations
but the intrusion can be resisted by linguists; for example, a linguist may think
that, for almost all expressions, there is no theoretical point to the politically
inspired division of Serbo-Croatian into Serbian and Croatian and can simply
refuse to go along (unless she is a Serb or a Croat). In any case, the point is not that
linguistics should be focusing on expressions in, say, Italian rather than Romance, or
in, say, English rather than x-English for various values of ‘x’. (And the point is
certainly not about ‘‘who gets to own’’ a term like ‘English’.) The point is that the
primary focus should be on linguistic expressions that share meanings in the
idiolects of a group of people and so the groups will have to be identiWed one way
or another. All linguistic generalizations, save those of UG that cover all expressions,
must make use of these identiWcations. So these identiWcations are important.
This concludes my discussion of Chomsky’s puzzling claims (i) to (iv). I turn
Wnally to some rather startling claims that have been made recently by Georges
Rey. These claims are strikingly at odds with the views I have presented in this
section. I will develop my views further by contrasting them with Rey’s claims.
10.6 R EY O N ‘‘ T H E I N T E N T I O N A L I N E X I S T E N C E O F
L A N G UA G E ’’
Rey begins by noting:

It would seem to be a commonplace that people, when they talk, produce tokens of such
things as words, sentences, morphemes, phonemes and phones—I’ll call tokens of all
such types, ‘‘Standard Linguistic Entities’’ (‘‘SLE’’s). Part and parcel of this commonplace
would be the presumption . . . that these entities can be identiWed with some sorts of
acoustic phenomena, e.g., wave patterns in space and time. (Rey 2006: 237)24
He correctly observes that this view has been repeatedly challenged by linguists
over the years, and undertakes to defend what he takes to be the hypothesis
implicit in the challenge, what he calls a ‘‘folie a´ deux’’ view. This view arises out
of the following negative picture of SLEs:
the human ability to speak natural languages is based on the existence of a special faculty
that includes a system for the intended production and recovery of SLEs. To a Wrst
approximation, instructions issue from speakers’ phonological systems to produce certain
24 He wrongly cites Devitt and Sterelny (1987: 59) as an example of this view. In the passage
cited we are talking about the sounds and inscriptions that realize SLE’s. We emphatically do not
think that SLEs—for example, phones—can be identified with a sound type: we do not think that
a sound token is an SLE simply in virtue of its overt physical properties. On this see our later
discussion (ch. 13) of Saussure.
SLEs, and these instructions cause various motions in their articulatory systems, which in
turn produce various wave-forms in the air. These wave-forms turn out, however, not to
reliably correspond to the SLEs speciWed in the instructions. . . . Indeed, were SLE
tokens actually to exist, it would be something of an accident. Their existence is
completely inessential to the success of normal communication and to the needs of
linguistic theory. . . . the apparent objects . . . are best regarded as (in Franz Brentano’s
1874/1973 phrase) ‘‘intentional inexistents’’: ‘‘things’’ that we think about, but that (we
often know very well) don’t actually exist, such as Santa Claus and Zeus. (pp. 239–40)
So Rey has two conclusions. First, he has the antirealist conclusion that, apart
perhaps for a few accidents, the world does not really contain linguistic entities,
entities with the usual range of linguistic properties including, of course, con-
ventional meanings in a language.25 We have encountered this antirealism
before: it was the second doubt about my view that the linguistic task concerns
linguistic reality not psychological reality (2.4). Second, Rey concludes that
language does not need such entities to play its role and hence these entities
are irrelevant to the theory of language. For, that theory isn’t really about any
acoustic phenomena in space-time. This conclusion stands opposed, of course, to
my claims about the theoretical interest of explaining the conventional meanings
of such acoustic phenomena (8.5, 10.5).
This is not the place to attempt a full response to Rey’s arguments for these
conclusions. My main concern is to show why I reject the conclusions. Still, I shall
brieXy indicate the lines of my response to his arguments.
(i) My response to his argument that the various syntactic properties of SLEs
cannot be regarded as properties of the acoustic stream would build on the
following ideas. The argument shows that the linguistic structure of an utterance
is not obvious and superWcial. But this structural property is relational not
intrinsic and relational properties are typically not obvious or superWcial. Yet
objects really have relational properties; for example, some objects really are
paperweights, moons, echidnas, Australians, and so on. Sometimes, it is easy to
tell that an object has a certain relational property because that property is well
correlated with superWcial properties. This makes it quite easy to tell an echidna,
but not an Australian (if she keeps her mouth shut). And it makes it quite easy to
tell many English adverbs, the ones that end in ‘ly’. It can also be easy to tell that
an object has a certain relational property if learning to identify the object
involves learning to identify it as having that property. This makes it quite easy
to identify the other English adverbs; identiWcation comes with word recogni-
tion. One way or another, it is quite easy to tell the explicit structural properties
25 Several conversations and the following comment from an anonymous reviewer of Devitt
2006 make me wonder if this sort of antirealism is common among linguists: ‘‘The sound waves
produced by a speaker, or the gestures of an ASL user, or the ink marks of a text, understood as
environmental features, have no linguistic properties whatsoever, not lexical, not phonological, not
syntactic, not semantic.’’
of utterances although sometimes hard to tell the implicit ones. But utterances
still really have both (9.5).
(ii) My response to his argument that the various phonological properties of
SLEs cannot be regarded as properties of the acoustic stream would build on the
following ideas. The argument shows that no naı̈ve brute–physical account of the
relation between sounds and phonemes is possible. Phonology shows that there
are many complicated ways in which sounds can instantiate a phoneme, includ-
ing relations to other sounds; and that a sound may be able to instantiate more
than one phoneme.26 Similarly, there are many complicated ways in which
inscriptions can instantiate a letter; and so on for other linguistic media. But
this does not show that the sounds, inscriptions, etc. do not instantiate SLEs.
Quite the contrary. The property of being Australian is instantiated by a vast
variety of physical forms; for example, the forms of the capitalist Rupert Mur-
doch, the runner Cathy Freeman, the horse Phar Lap, the city of Sydney, a bottle
of Penfolds Grange, and the many forms of the saying ‘‘No worries, mate’’. The
property of being the word ‘cat’ is instantiated by a much smaller variety of
physical forms, a variety of sounds, inscriptions in many diVerent fonts and
handwritings, and so on. Just as all the former instantiations really are Australian,
all the latter really are the word ‘cat’. And note that just as some things do not
count as Australian, some things do not count as the word ‘cat’.27
Turn now to Rey’s conclusions. He thinks that the intentional inexistents that
he identiWes with SLEs have the usual linguistic properties but that the actual
entities—particular sounds, inscriptions, etc.—that we would ordinarily think of
as SLEs lack these properties, including conventional meanings. From my per-
spective, if these entities really lacked conventional meanings they could still have
speaker and audience meanings (and perhaps literal meanings). We would then
be in a prelinguistic state, a state analogous to that of a foreigner in a country
where everyone speaks a language that is totally alien to her. Communication
would be possible, but hard and very elementary, pretty much limited to
conveying the crudest messages about food, drink, sex, and shelter. It would
depend on the ‘‘natural meaning’’ of signs—of ‘grrhh’, play actings, and the
like—and a great deal of insight into other minds. (One thinks immediately of
the discussion in Bennett 1976.) If that were indeed the situation, there would be
no theoretical point to attributing a speaker meaning to the signs produced: we
26 ‘‘In natural speech, properties of the acoustic signal associated with one phoneme often
overlap or co-occur with the properties of adjacent segments . . . Researchers have been unable to
find invariant sets of acoustic features or properties that correspond uniquely to individual linguistic
units.’’ (Nygaard and Pisoni 1995: 65–6)
27 I can’t resist noting that worries about the physical realization of phones led to the metaphys-
ical disaster of Saussurian structuralism, hence to the metaphysical catastrophe of post-structuralism
(Devitt and Sterelny 1999: ch. 13). Antirealism seems to be an occupational hazard of phonology.
Rey is at pains to resist this hazard, distinguishing what he takes to be the unreality of SLEs from the
reality of, e.g., automobiles.
might as well simply talk of the contents of the thoughts and the intentions to
express them using those signs. Note that this is not to say, as presumably Rey
would say, that the signs do not have speaker meanings—for we can say that they
have speaker meanings in virtue of being the expression of those thoughts—but
simply that there is no theoretical interest in their having them.
From my perspective, then, if Rey were right we would be in this ‘‘primitive’’
situation for communication and theory. Contrast this situation with the actual
situation, considering communication Wrst. The actual situation is obviously very
diVerent from the primitive one as, of course, Rey would agree. Rather than
being hard and elementary, communication is actually easy and extraordinarily
sophisticated, conveying complicated messages about a limitless variety of topics;
as Quine says, about ‘‘tables, people, molecules, light rays, retinas, air waves,
prime numbers, inWnite classes, joy and sorrow, good and evil’’ (1966: 215). So,
how does Rey think such communication is possible given that there are no
SLEs? On his view, what is going on that saves us from being in the primitive
situation? This is where he appeals to his folie a´ deux hypothesis:
when [the wave-forms] impinge on the auditory system of an appropriate hearer, this
hearer’s phonological system will be able to make an extremely good guess about the
intentional content of the speaker’s instructions, not about any actual SLEs, which, ex
hypothesi, never actually got uttered. Indeed, this sort of guessing in general is so good,
and the resulting perceptual illusion so vivid, that it goes largely unnoticed, and speakers
and hearers alike take themselves to be producing and hearing the SLEs themselves. It is in
this way that it’s a kind of folie a´ deux (or a´ n, for the n speakers of a common language):
the speaker has the illusion of producing an SLE that the hearer has the illusion of
hearing, with however the happy result that the hearer is usually able to determine
precisely what the speaker intended to utter. (2006: 239–40)
Despite all this illusion the ‘‘happy result’’ is achieved because the hearer makes
‘‘an extremely good guess’’. Something has gone very wrong here.
On my view, there is only a small need for guesswork in communication
because, normally, the conventional meaning of an expression in the linguistic
community of the speaker and hearer provides a rich clue to the speaker’s
thought. That clue is often not suYcient, of course: the hearer has to determine
the reference of indexicals and remove ambiguities. These pragmatic skills (about
which we shall say more in section 11.8) require insight into other minds but
seldom anything that is appropriately called ‘‘guesswork’’ because the conven-
tional meaning gives the hearer such a big start in understanding. That big start is
precisely what Rey removes. If the sounds and inscriptions people produce do
not have conventional meanings grasped by all parties, how could communica-
tion be so eVective? We can see, of course, how it can sometimes take place
without any conventions—consider the primitive situation, for example—but
how would the rich and complex communications that dominate our daily lives
be possible? On Rey’s view, communication seems to rest on miraculous guesses.
Rey would surely object that I have overlooked something crucial to commu-
nication in the actual situation that distinguishes that situation from the
primitive one. On his folie view, the ‘‘extremely good guess’’ is made by a hearer
who shares ‘‘a common languge’’ with the speaker. This raises a number of
questions.
First, what makes the languages of the speaker and hearer a common, or the
same, language? For Rey, the answer is that the linguistic objects of thought for
speaker and hearer are the same intentional objects. We wonder, then, what
makes them the same in the absence of any actual objects that are the same. This
is, of course, another example of the old problem of ‘‘intentional identity’’, the
problem of explaining a common focus of thought where there is nothing at the
focus; see the delightful Geach 1967 for a discussion. But it would seem to be a
particularly diYcult example.
Second, consider the signiWcance of the intentional identity of linguistic
objects to communication on Rey’s view. On anyone’s view, successful commu-
nication requires that the message intended by the speaker be (near enough) the
same as the message understood by the hearer. On my view, this matching of
messages is usually and largely achieved because the superWcial properties of the
physical entities (sounds, inscription, etc.) that make up the message-conveying
SLE provide excellent clues to the hearer about the SLE’s conventional meaning,
a meaning that largely constitutes the message intended by the speaker. The
superWcial properties provide those clues, of course, because there is a conven-
tional regularity of using such an entity with those properties to express a
meaning that is part of the message. So, contrary to Rey’s claim (2006: 251),
linguistic properties do play a causal role.28 In contrast, the matching of messages
is achieved on Rey’s view by the intentional object that is the speaker’s SLE being
(near enough) the same as the intentional object that is the hearer’s. That is the
signiWcance of the intentional identity. But then how is this happy identity
achieved? Clearly the superWcial properties of the physical entity—the entity
that is the SLE on my view—must provide the hearer with clues to the speaker’s
intentional object. How could the superWcial properties do that? Not by being
excellent clues to the conventional meaning of the physical entity, as I think they
are, because that entity has no conventional meaning on Rey’s view. We are still
left with a miracle: the success of hearers at guessing speakers’ intentional objects
without having the beneWt of conventions that relate physical entities to mean-
ings. And we are left with the closely related miracle of all parties to a successful
communication being under the same illusion about the linguistic properties of a
physical entity that has none.
28 This raises two well-known philosophical problems: the problem of properties that are
relational not intrinsic having causal powers; and the problem of properties that supervene on
others having causal powers. Linguistic properties are both relational and supervening. But then so
are almost all properties outside physics and even, I’m told, some in physics. Linguistic properties
are in very good company.
Finally, how did speaker and hearer come to speak the same language? Some of
a person’s language may well be innate but she learns a good deal of it. On my
view, this learning is a matter of acquiring conventions of a language as a causal
result of experiencing the regular exercise of the conventions in a community.
Once again linguistic properties have a causal role. A consequence of this learning
process is that a person, more or less, shares a language with others in the
community. But this story depends on what Rey denies, the existence of linguistic
entities with conventional meanings. Without those entities, language learning
becomes a mystery.29
Turn Wnally to the contrast between our theoretical needs in the primitive
situation and in the actual one. In the primitive situation there is no interest in,
or point to, attributing meanings to the signs that attempt to convey messages.
Rey thinks that this is true in the actual situation too. Indeed he thinks that the
actual existence of meaningful linguistic entities is irrelevant to linguistic theory.
This is not so. The study of the (largely) conventional meanings of actual
linguistic entities, meanings constituted by a (partly) conventional syntax and
conventional word meanings, is the concern of linguistic theory. Our theoretical
interest in language is in explaining the nature of these conventional meanings
that enable language to play such an important role in our lives.
In sum, in response to Rey’s argument, I would argue that linguistic entities
really exist even though they are relational and even though the one expression
can appear in a variety of physical forms. In response to his conclusions, I have
argued that it is in virtue of the conventional meanings of linguistic entities that
the entities play their extraordinarily important role in communicating messages.
It is in virtue of those meanings that a language is acquired. Linguistic properties
do play causal roles. If there really weren’t any linguistic entities, communication
would be miraculous and language learning a mystery. Our theoretical interest in
language is in conventional meanings.
10.7 CONCLUSION
The piece of folk wisdom that language expresses thought (LET) leads to the
view that conceptual competence is a considerable part of linguistic competence.
I argued that this view is not undermined by the well-known dissociation
of cognitive impairment and linguistic impairment (10.1). Since conceptual
29 These two paragraphs bear on Laurence’s claim (aimed at Devitt and Sterelny 1987) that the
shared meaning of an expression could arise from a ‘‘coincidence in literal idiolect meaning’’ (1998:
208). Indeed it could and would. But it would be a miracle that such shared meanings played the
role that they do in communication if the sharing were not a convention that speakers and hearers
participate in. And it would be a mystery how people came to share meanings of so many
expressions if not by experiencing the conventions of using those expressions with those meanings.
These ‘‘coincidences’’ would cry out for explanation.
competence is in the central processor, it follows from LET that a good deal of
linguistic competence is too. The rest of linguistic competence consists in
processing competencies for spoken language, written language, and so on. If
there is to be a modality-neutral, relatively central, language faculty, it must be
found in some commonalities between these modality-speciWc processing com-
petencies. At Wrst sight, the evidence from brain impairment may seem to
support the idea that there are these commonalities and hence a language faculty.
I argued that this is misleading and that this evidence counts much more against
the existence of a language faculty than for it (10.2). Aside from that, if we move
beyond LET to (t)—the view that the structure rules of a person’s language are
similar to the structure rules of her thought—we should not expect to Wnd the
commonalities that would constitute a substantial language faculty because there
would not be much for the language faculty to do. I arrived at my second
tentative proposal: there is little or nothing to the language faculty (10.3).
Where does Chomsky stand on thoughts and their relation to language? It is
surprisingly diYcult to say. I oVered some reasons for thinking that his stance is
very diVerent from the one presented in this part of the book (10.4). I went on to
look critically at some claims that Chomsky makes against linguistic regularities
and conventions, against the need for shared and discoverable meanings in
communication, and in favor of idiolects rather than common languages.
I emphasize that it is because of shared conventional meanings in a group that
language can play its important role of making the thoughts of each member of
the group accessible to the others. This yields my sixth major conclusion: the
primary concern in linguistics should not be with idiolects but with linguistic
expressions that share meanings in idiolects (10.5). Finally, I consider Rey’s view
that linguistic objects are merely intentional. I argue, in contrast, that these
objects are real parts of the world. And the task of linguistics is to explain the
nature of their conventional meanings (10.6).
I argued for (t) and my Wrst tentative proposal in Chapter 9: a language is
largely psychologically real in a speaker in that its rules are similar to the structure
rules of her thought. If that is correct then the rules of the language will be
psychologically real in a robust way. Furthermore, position (t) encourages
doubts that the language is psychologically real in any other robust way described
in section 3.4. It particularly encourages doubts that speakers represent the rules
of the language, as required by the Representational Thesis (RT) and positions (i)
and (ii). According to position (i), the speaker applies those represented rules in
language processing; according to (ii), she uses those representations as data in
processing. Yet resorting to represented rules in these ways seems an indirect and
ineYcient way to solve the problem of translating between an English and a
Mentalese sentence with very similar structures. (t) also encourages doubts that
the rules of the language are simply embodied processing rules, as required by
position (iii), because the rules of the language seem to be the wrong sort of rules
to govern the translation process. These doubts remain, although they are less
pressing, if we have to settle for the less-committal Representational Theory of
Mind (RTM). And I don’t think they should disappear even if we have to settle
for the almost noncommittal view that language expresses thought (LET).
Aside from these doubts about views of the psychological reality of language,
I think (t), perhaps even RTM, should make us a little dubious that the
processing rules, whatever they may be, operate on metalinguistic representations
of the syntactic and semantic properties of linguistic symbols, as versions (a) of
positions on psychological reality require, rather than being directly responsive to
those syntactic and semantic properties, as versions (b) require (3.4). Perhaps the
move to metalinguistic representations is an ineYcient way of handling the
translation. Perhaps this is a place to apply Pylyshyn’s Razor: representations
are not to be multiplied beyond necessity (3.1).
Part I argued for my Wrst major conclusion, that linguistics is not part of
psychology. It follows that the truth of a grammar alone provides insuYcient
reason to suppose that the rules it ascribes to a language are psychologically real
in speakers of the language. A fortiori, it provides insuYcient reason to suppose
that those rules are represented (encoded) in the speakers, and hence insuYcient
reason to adopt RT as part of a theory of linguistic competence. The supposed
truth of the grammar leaves the issue of the psychological reality of language
fairly open. In Part III we found no evidence for RT, and hence for positions (i)
and (ii), in various ‘‘philosophical’’ arguments: the arguments from the rejection
of behaviorism, from the folk truism that the competent speaker knows her
language, and from the role of linguistic intuitions. We had thus begun the task
of establishing my second major conclusion: there is no signiWcant evidence for
RT and, given what else we know, it is implausible. And I claimed to have
established my third major conclusion: Speakers’ linguistic intuitions do not
reXect information supplied by the language faculty. They are immediate and
fairly unreXective empirical central-processor responses to linguistic phenomena.
They are not the main evidence for grammars. The focus of Part IV was on
thoughts and led to two more major conclusions: the fourth one that the
psychological reality of language should be investigated from a perspective on
thought; and the Wfth one that thought has a certain priority to language
ontologically, explanatorily, temporally, and in theoretical interest. Guided by
those conclusions, a case was presented for (t), the view that the structure rules of
a speaker’s language are similar to the structure rules of her thought. In this way
her language is largely psychologically real in her. That was my Wrst tentative
proposal. My second was that there is little or nothing to the language faculty.
It is time now to see if there is any evidence for RT and positions (i) and (ii) in
the psycholinguistic studies of language use and language acquisition. And it is
time to see what light those studies cast on other positions on psychological
reality. It is clear that the fate of all positions rests heavily on the psycholinguistic
evidence. It is interesting to note, in contrast, that the fate of position (t) rests
very little on this evidence. The psychological assumptions that do nearly all the
work in the case for (t) are the view that language expresses thought (LET), the
Language-of-Thought Hypothesis (LOTH), and the Gricean explanatory prior-
ity of thought over language. These assumptions are agreeably independent of
the details of language use and acquisition. Of course, LET is a theory of
what language use consists in—the matching of sentences and thoughts for
meaning—but our reason for believing it does not come from psycholinguistic
discoveries. Our reason comes from our acceptance of thoughts. And we accept
thoughts because they successfully explain behavior and guide us to reality.
We are drawn to the Gricean priority because it is the best explanation of the
relation of thought to language, and to LOTH because it is part of the most
promising explanation of thoughts. Neither of these reasons are matters of
language use at all.
Named propositions like RTM and LOTH, and views of competence, are all
listed in the Glossary, along with the major conclusions and tentative proposals.
Let us now turn to the psycholinguistic evidence.
PART V
L AN G UAG E U S E A N D
AQU I S I T I O N
11
Language Use
11.1 I N T RO D U C T I O N
The aim in this chapter is to assess a range of positions on language use. In the
next chapter we will consider language acquisition.
Let us start by recapitulating the possible positions on language use set out in
section 3.4, starting with the uncontroversial minimal position:
(m) A competence in a language, and the processing rules that govern its exercise, respect
the structure rules of the language: the processing rules of language comprehension take
sentences of the language as inputs; the processing rules of language production yield
sentences of the language as outputs.
This position is silent on how the competence respects the structure rules of the
language. In particular, it does not say whether the competence respects those
rules because the rules are psychologically real in some way. The remaining
positions are not silent on this matter:
(i) The structure rules of the language are also processing rules that are represented in the
speaker and applied in language use.
(ii) The structure rules of the language are represented and used as data by the processing
rules of language use.
(iii) The structure rules of the language are also processing rules for language use but the
processing rules are not represented.
structure rules of the language and do not use the structure rules as data.
In Chapter 4, we considered some historically interesting actual positions on
language use and related them to these possible positions. We started with some
tantalizing, and hard to reconcile, remarks by Chomsky, drawn from several
writings. We then moved to more detailed positions proposed by Fodor, Bever,
and Garrett (1974); Bresnan and Kaplan (1982); Berwick and Weinberg (1984);
and Matthews (1991). All of these writers take the grammar of a language to
be somehow true of the psychological reality of its speakers. They do not
196 Language Use and Aquisition
contemplate that the grammar might simply be true of linguistic reality, as urged
by the Wrst of my methodological points in section 2.6. All of the writers except
Matthews seem to take the Representational Thesis (RT) for granted. So,
that is the powerful psychological assumption which, according to my second
methodological point, they need in taking the grammar to be true of
psychological reality. Finally, most of these writers are strikingly at odds with
my third methodological point. They do not take the grammar and the theory of
competence to be largely independent of each other. Rather, they take each to
place heavy constraints on the other. According to my third point, the only limit
on the independence of these two theories arises from ‘‘the Respect Constraint’’: a
theory of a linguistic competence must posit processing rules that respect the
structure rules of the language, and a grammar must posit structure rules that are
respected by the competence and its processing rules. Because of this, the
justiWcation of the grammar and the justiWcation of the theory of competence
are partly dependent on each other. But in all other ways the two theories are
independent.
Fodor, Bever, and Garrett made it clear that early theories of how a grammar
should be incorporated into a theory of language use were wrong. As a result,
grammatical rules had a reduced role in the theory of language use in the decade
or so that followed. Still, as Pylyshyn pointed out (1991: 232), and the above-
mentioned writers demonstrate, the orthodox Chomskian belief that grammat-
ical rules play a central role in the theory of language use remained.1 However,
the discussion of language processing in sections 11.7 to 11.8 indicate that this
belief has weakened in subsequent years. The psycholinguistic interest in trying
to Wnd the rules of the language embodied in a speaker’s mind seems to have
steadily diminished without, so far as I can see, much explicit acknowledgement
that it has;2 the interest is simply withering away.3 Psycholinguists mostly now
approach the study of language processing as if it were, except for the Respect
Constraint, independent of grammars. And so they should.
The question of whether or not the rules of the language are embodied in the
mind is one of the two that dominate positions (i) to (v). (i) to (iii) say that they
are; (iv) and (v) have no such commitment. Rules might be embodied by being
1 Consider also the following: ‘‘Linguistic theory is a theory of native speakers’ underlying
knowledge of their language . . . . The goal of psycholinguistics is to explain how linguistic
knowledge is acquired and how it is mentally represented and used to actually produce and perceive
language’’ (Frazier 1995: 2–3).
2 Kathryn Bock is an exception. She concludes an article with some remarks very much in the
spirit of the third methodological point: ‘‘It should be clear by now that production theory is a
theory of using language, not an account of how language is represented, and that linguistic
frameworks whose province is the static organization of language knowledge are unlikely to provide
the theoretical machinery for explaining how that knowledge is deployed in time.’’ (1995: 207)
3 Avery Andrews has drawn my attention to the near disappearance of concern for the psycho-
logical reality of rules in discussions of LFG (lexical–function grammar) even though that concern
was the initial motivation for this approach to grammars, as noted earlier (4.4).
Language Use 197
represented or they might be simply embodied (3.1). This leads to the other
dominating question: Are some processing rules represented in the speaker? (i),
(ii), and (iv) say that they are; (iii) and (v) say that they are not. A distinction
between rules that govern ‘‘rational–causal’’ operations and rules that govern
‘‘brute–causal’’ operations (3.3) led to another question and to two versions of
(iii) and (v): Do some processing rules operate on metalinguistic representations
of the syntactic and semantic properties of linguistic items? Version (a) says that
some do. Version (b) says that none do: the rules are directly responsive to those
properties of linguistic items; the process is fairly brute–causal and associationist.
These three questions will be the concern of this chapter.
A consequence of positions (i) and (ii) would be that RTwas correct. We have
so far found no evidence for this thesis. My Wrst, and main, goal in this chapter is
to argue that language use provides no persuasive evidence for RT and that RT is
implausible. This is the most important step in arguing for the second major
conclusion of this book: there is no signiWcant evidence for RT and, given what
else we know, it is implausible. Hence, (i) and (ii) should be rejected. The Wnal
step in arguing for this conclusion is in the next chapter on language acquisition.
Much of the argument in the present chapter against RT counts also against
position (iv), the view that processing rules other than the structure rules of the
language are represented.
The popularity of RT can often seem overwhelming. Still, the following quotes
show that I am not alone in my doubts about it. Stabler has this to say about the two
positions on language processing that involve this thesis: of (i) he says that it is ‘‘not
supported by any available evidence’’; of (ii), that it is ‘‘not supported by available
data either’’ (1983: 395). Roger Schank and Lawrence Birnbaum make a more
sweeping claim covering (iii) as well: that among psychologists ‘‘even the strongest
partisans of generative linguistics’’ are convinced that ‘‘there is no evidence that
people make use, in comprehension or generation, of the kinds of rule devised by
generative linguistics to describe linguistic phenomena’’ (1984: 221). Pylyshyn
claims that, to his knowledge, ‘‘there have been no arguments that the rules of
language . . . are explicitly encoded’’ (1991: 247).4 Matthews says that ‘‘nothing,
so far as I can see, suggests that . . . grammars are internally represented by
speaker/hearers, if by this one means explicitly represented or tokened’’ (1991:
187). My main goal is to add to the case against RT.
Positions (i) and (ii) take the structure rules of the language to be involved in
processing by being represented. According to position (iii), the structure rules
are otherwise psychologically real: they govern processing without being en-
coded. I have already raised doubts about this (2.6, 3.2). My second goal is to
cast further doubts on this and hence give support to (v): the processing rules are
largely unlike the structure rules.
4 Pylyshyn seems to have overlooked his own earlier argument to this eVect (1980a: 121; 1980b:
163). This argument is address in section 11.4, point 5, below.
To reject (i) to (iii) is not, of course, to deny that the structure rules are
embodied in the speaker’s mind at all. Indeed, if (t) is true these rules are similar
to the structure rules of thought and so, to an extent, obviously embodied. And
they will even be relevant to language processing because the processing rules will
operate on the syntactic properties of thoughts determined by those structure
rules. Still the structure rules will not be embodied processing rules. The rules will
characterize the structure of the thoughts that are involved in the processing, they
will not govern the processing from thoughts to language and from language to
thoughts. If (i) to (iii) are rejected, and we are to save the orthodox Chomskian
view that grammatical rules play a more central role in language use than is
provided by the minimal (m), then I think that (t) is the only hope.
My third goal concerns the metalinguistic representations of the syntactic and
semantic properties of linguistic entities. According to version (a), the processing
rules operate on these representations; according to version (b), they do not. If
RT is true then version (a) must be but even if RT is false, (a) might be true:
processing might be a rational information Xow even if the rules governing it are
not represented. (a) seems to be the received view even among many who reject,
or do not clearly hold, RT.5 My goal is to cast doubt on this view, trying to make
(b) seem plausible. I conclude that we have good, although very far from
conclusive, reasons to think that the speedy automatic part of language process-
ing is indeed fairly brute–causal associationist.
My discussion will be guided by the fourth major conclusion: the psycho-
logical reality of language should be investigated from a perspective on thought.
And it will be guided as always by Pylyshyn’s Razor: representations are not to be
multiplied beyond necessity (3.1).
11.2 ‘‘ T H E O N LY T H E O RY I N TOW N ’’
It seems that many linguists are inXuenced toward RT by an inference to the best
explanation, or abduction, about language use: RT is thought to provide ‘‘the
only theory in town’’. The abduction runs along the following lines. We can
explain language comprehension if we see it as a rational process of testing
hypotheses about a person’s speech input. And there is no other way to explain
it. On this view, if the psycholinguistic evidence to be considered later showed
that a certain linguistic rule was playing a role in this process (and so was
psychologically real), it would have to be represented. For hypothesis testing is
a process of testing one representation against others.6 (Note that this argument
concerns language comprehension. It is striking that the analogous argument for
production does not have any appeal.)
5 See Frazier 1995 for example.
6 I have often heard such arguments but have been unable to Wnd a clear case in the literature. For
a case that comes close, see J. A. Fodor 2001: 113–15.
Language Use 199
In recent years, abduction has come in for considerable criticism from philo-
sophers of science, particularly those who are dubious of scientiWc realism. This
criticism seems to me unwarranted (1997a: 111–13; 2005c). So I think that
many abductive inferences are good. But, of course, not all are good. And
the inference from the evidence of language use to the conclusion that
this use involves representations of the structure rules of the language—RT—
seems to me an example of one that is not good. Indeed, I think that it is really
rather bad.
My judgment here cannot be supported by appealing to an algorithm for
deciding which abductions are good for no such algorithm is available. Still, we
do know some factors that should inXuence our decisions.
One thing that worries people about abductive arguments is that they can
seem too easy. Whenever we have some previously unexplained phenomena, we
can easily come up with some putative explanation if we do not operate under any
constraints; thus, if we are prepared to tolerate absurdity, we can appeal to
gremlins, acts of God, or Martian invaders. So, criterion (A) of a good abduction
is that it involves a good explanation. As is well known, it is hard to say much
about what makes an explanation good. But goodness requires at least two things
(A1). The explanation must not only not be absurd but it must be plausible given
what else we know, it must be plausible relative to background knowledge.
A corollary of this requirement is that if none of our candidate explanations
meet this requirement, then we should not accept an abduction involving the best
of this implausible bunch. Rather than such a rush to judgment we should follow
a course of action the virtues of which are sadly underappreciated: we should
suspend judgment and keep looking. (A2) Next, a good explanation must have an
appropriate level of detail: if the explanation posits x as the cause of y, it must say
enough about the mechanism by which x causes y to not leave this mysterious; a
wave of the hand is not suYcient.7 If we do not have the details we need, we
should, once again, suspend judgment.
Criterion (B) of a good abduction is that the explanation it features must be
better than any actual alternative or even any alternative that is likely, given what
we already know. Attention to likely alternatives is important. Even if the
featured explanation does not face any worked-out alternatives, we may have
some ideas for alternatives that, given our background knowledge, seem prom-
ising. Until those ideas have been explored suYciently to be set aside, we should
suspend judgment on the abduction. (This is not to be confused with the extreme
skeptical view that we should not embrace an abduction until all possible
alternative explanations have been set aside. We need wait only to set aside
alternatives that are likely, given what we already know.)
7 Berkeley rightly criticizes an abduction for Locke’s ‘‘representative realism’’ on this score. The
abduction is that the existence of a material world causing and resembling our ideas provides the
best explanation of those ideas. Berkeley points out that we do not have any idea of the mechanism
by which a material body can cause a resembling spiritual idea (1710: sec. 19).
I shall argue that explanations of language use that appeal to RT do not meet
any of these criteria. Concerning (A1), these explanations are surely not absurd
but I shall argue that they are implausible given what else we know. Concerning
(A2), the explanations of language use involving RT that have been provided so
far are singularly lacking in details. Fodor, Bever, and Garrett (1974) showed how
early attempts to provide such details failed badly (4.3) and the situation has not
improved signiWcantly since. Indeed, given the consensus that, at this stage, we
know very little about language processing,8 we are surely not in a position to
give a general explanation of this processing, whether involving RT or not, that
is suYciently precise, complete, and successful to be an appropriate candidate
for a good abduction. Chomsky has dismissed the view that linguistic compe-
tence is ‘‘a set of dispositions’’ as ‘‘merely a promissory note’’ (1980b: 5). But
all views, including the view that this competence involves representation of
the linguistic rules, are really just promissory notes at this time. Concerning
(B), there is a range of likely alternative explanations of language use that need
to be investigated before we should embrace an explanation appealing to RT.
That explanation takes the processes of language use to be highly ‘‘rational’’.
The likely alternatives take it to be less so, perhaps not even rational at all.
There is no basis for expecting that the best explanation of language use is
likely to involve RT. Indeed, the evidence suggests strongly that the best
explanation is unlikely to.
So I will be arguing that positions (i) and (ii) are implausible. What else can
we hope to discover at this point about language use? Not that one of the other
positions is the conclusion of a persuasive abduction. The best we can hope for,
given our present level of ignorance, is an assessment of the likelihood that a
position will, or will not, be the conclusion of a persuasive future abduction.
Criterion (A1) emphasizes that abductions are judged against a background of
what we already know. In the next section I shall summarize what we already
know about linguistic competence, which should guide us in seeking a good
abduction about language use. In the following three sections, 11.4–11.6, further
background will be introduced and preliminary assessments will be made of the
likelihood of a future abduction supporting each of the various positions. This
assessment will be made prior to any consideration of the psycholinguistic
evidence from language use itself. The idea is to see what direction seems
promising for the explanation of language use given what else we know and
don’t know about other matters. I shall be mainly concerned to demonstrate
8 Some expressions of this consensus: ‘‘Very little is known about how [a device for sentence
comprehension] might operate, though I guess that, if we started now and worked very hard, we
might be able to build one in Wve hundred years or so’’ (J. A. Fodor 1975: 167); ‘‘we know so little
about the actual machinery engaged in human sentence parsing’’ (Berwick and Weinberg 1984:
35); the relation between the grammar and the parser ‘‘remains to be discovered’’ (Pritchett 1988:
539); ‘‘we know very little about the computational machinery involved in language processing’’
(Matthews 1991: 190–1).
Language Use 201
the prima facie unlikelihood of support for positions (i) and (ii) and hence for
the idea that RT has a place in the explanation of language use: not only is it not
now part of the conclusion of a good abduction it is unlikely to be so in the
future. I am also concerned to throw doubt on position (iv) and the idea that
language use involves any represented rules at all, and on position (iii)
and the idea that linguistic rules are also embodied processing rules. These
considerations against (i) to (iv) are considerations for position (v). In section
11.6 I will present some considerations in favor of version (b) of (v), in favor of a
radical brute–causal view of processing. Finally, in sections 11.7 and 11.8, I turn
to the psycholinguistic evidence on language use. My discussion of this is
inevitably very brief but it is still, I think, enough to show that the evidence at
this point does not give any reason to modify these preliminary assessments of
how the future is likely to go.
So my focus will be on whether, given our background knowledge, positions
are likely to be part of the future. But if they aren’t—and I shall be arguing that
(i) to (iv) aren’t—then they shouldn’t be part of the present.
11.3 B A C KG RO U N D O N L I N G U I S T I C C O M PE T E N C E
Various views of linguistic competence have emerged to form the framework for
theorizing about language use. These views vary in the amount of their theoret-
ical commitment. I shall present them in order of increasing commitment.
The most theory-neutral view of competence in a spoken language is that it is
the ability to produce and understand sentences with the sounds and meanings
of that language. We then moved from this minimal view to increasingly theory-
laden views. (a) Acceptance of the folk view that language expresses thought
(LET; 8.2), led to the view that the competence is the ability to use a sound of
the language to express a thought with the meaning that the sound has in the
language in the context of utterance; and the ability (together with some
pragmatic abilities) to assign to a sound a thought with the meaning that
the sound has in the language in the context of utterance. So competence in the
language requires a certain conceptual competence, the competence to have
thoughts with the meanings expressible in the language (8.3). (b) Acceptance
of the Representational Theory of Mind (RTM) led to an even more theory-
laden view: competence is the ability to translate back and forth between mental
representations and the sounds of the language. So competence in the language
requires the competence to think mental representations with meanings expressible
in the language (9.1). (c) Acceptance of the Language-of-Thought Hypothesis
(LOTH; 9.2) adds still more theory: competence is the ability to translate back and
forth between mental sentences and the sounds of the language. So competence in
the language requires the competence to think mental sentences with meanings
expressible in the language (9.3).
On views (a) to (c), conceptual competence is the essential core of linguistic

competence and enjoys a certain ontological priority over it. On (b) the required
conceptual competence is competence in some system of mental representations,
perhaps a map-like one, perhaps a language-like one, or perhaps something else
altogether; on (c), it is competence in a language-like Mentalese. So our view of
the nature of thoughts becomes central to our theory of linguistic competence,
and hence to our stance on the psychological reality issue. In particular, LOTH
makes it plausible that the structure rules of a person’s Mentalese are similar to
those of her public language, position (t), and hence that the rules of her public
language will be largely present in her mind even if they are not processing rules
for language use. The processing rules will operate on the syntactic properties of
mental and public representations even if the rules that determine the nature of
those syntactic properties are not among the processing rules.
We should note that just as the theory of competence in a system of
linguistic representations—a language—should not be conXated with the the-
ory of that system, the theory of competence in a system of mental represen-
tations should not be conXated with the theory of that system (Ch. 2). Thus
the theory of mental representations is the theory of their meanings, of the
properties in virtue of which those representations, as parts of thoughts, play a
role in explaining behavior and informing us about reality. The theory of the
competence is the theory of the ability to produce such representations, the
ability to think.
If we assume LOTH and view (c) then we can say something about the nature
of Mentalese competence, reXecting what we said earlier about the nature of
Mentalese meaning on the same assumption (9.5). First, Mentalese competence
is complex. One part is syntactic competence. That consists in an ability to
combine Mentalese words of the various syntactic categories into the sentences of
Mentalese; for example, the ability to combine <Reagan> and <wrinkled> to
form the sentence <Reagan is wrinkled>; and the ability to form a sentence like
<All police are corrupt> using the quantiWer <all>. The other part of Menta-
lese competence is lexical competence, a competence with mental words. We
need to distinguish basic words, probably including the mental correlates of
proper names and natural kind terms, from the nonbasic ones. Competence with
a basic word must consist in having thoughts involving that word that are directly
linked, presumably causally linked, in the appropriate way to the word’s referent.
I think (1996), unlike Jerry Fodor (1987), that some words will surely not
be basic but rather covered by a description theory; <bachelor> is a likely
example. Competence with such a word will consist in associating it inferentially
to others; for example, associating <bachelor> with <adult>, <unmarried>,
and <male>.
Next, consider competence in the spoken language. Syntactic competence is
the ability to translate back and forth between the syntactic structures of the
sounds of the language and structures of mental representations. And lexical
Language Use 203
competence is the ability to translate back and forth between the words of the
language and mental words.
If we do not assume LOTH but rather assume that mental representations are
map-like then we would, of course, have to give a diVerent account of conceptual
competence and, as a result, a diVerent account of linguistic competence. It is
hard to see what account we could come up with just as it is hard to see how we
could explain the meanings that these map-like representations would be
required to have (9.2). And if we remain more neutral altogether, resting with
RTM, or even just LET, we can make little further progress at this point in
explaining linguistic competence.
It is worth noting that acceptance of the relatively uncontroversial LET alone
has a signiWcant consequence for the place of RT. For, according to LET, our
linguistic competence is partly, perhaps even largely, constituted by a certain
conceptual competence: a person cannot match sounds and thoughts for mean-
ing unless she has the capacity to have thoughts with those meanings (8.3). This
downplays the language faculty, as we have already noted (10.2), and thereby
downplays the signiWcance of RT. Even if competence in a natural language does
partly consist in representing the rules of the language and making use of those
representations somehow in language processing, an important part of it consists
in something altogether diVerent: the capacity to think certain thoughts.
Finally, consider the task in language comprehension. Given LET, the task is to
pick up clues of the syntactic and semantic properties of the input so that it can
be matched with an appropriate thought. Given RTM, the task is to Wnd clues
to a correct translation of the linguistic input. Similarly, given LET, the task in
language production is to select an appropriate linguistic output for the thought.
Given RTM, the task is to produce a correct translation of the thought. So, in
theorizing about language use, we should not look for more than seems to be
required for these tasks.
In what follows I shall always start from the assumption of LOTH and view (c)
of competence; that will be my default background. I shall then consider the
consequences of replacing this assumption with other views about the nature of
thoughts and linguistic competence.
11.4 I M P L A U S I B I L I T Y O F R E P R E S E N T E D RU L E S I N
L I N G U I S T I C C O M PE T E N C E
Against this background, and more to be introduced, let us now begin a

preliminary assessment of the likelihood of persuasive future abductions support-
ing the various positions, particularly positions (i), (ii), and (iv) that involve
representations of rules. We shall see that the prospects of such abductions are
not promising, given that background, and given the likelihood of abductions to
other less representational conclusions.
The considerations I shall present move from relatively ‘‘a priori’’ ones toward
more empirical ones. I put most weight on the most empirical of all, the
psychology of skills, discussed in the next section. None of these considerations
alone comes close to being decisive but the cumulative eVect of them seems to me
to be fairly telling. And the only-theory-in-town abduction is left looking totally
unpersuasive.
1. It is clear that computers run software programs by representing rules and it is
clear that computers can use represented rules as data. There are also uncontro-
versial examples of humans representing rules. We represent the rules of games
like chess and baseball. We represent the rules of etiquette and of processions.
We represent some of the rules of a foreign language in trying to learn it. And so
on. But these uncontroversial examples do not encompass much of human life.
And, we should note, a large part of what makes these examples uncontroversial
is that we can, and do, state the rules. In this respect, a person’s relation to the
rules of her native language is very diVerent (6.2). Now, of course, the inability of
an English speaker to state the rules of English may simply reXect the fact that her
representation of rules is in a language module that is largely inaccessible to
consciousness. But is there any reason to believe that this is a fact? Indeed, is
there any uncontroversial example of rules being represented in an inaccessible
module?9 According to Stabler, ‘‘no [neurophysiological] evidence has been
found . . . either for linguistic processing or any other cognitive processing’’
governed by representations of the rules (Stabler 1983: 399). We have noted
before that should a rule govern a cognitive process, it could be the case that it
governs by being embodied without being represented (3.1). So, where we have
evidence that a certain rule does govern, Pylyshyn’s Razor demands further evi-
dence before we conclude that it does so by being represented; we need further
evidence that the rule plays its role like a softwired rule in a general-purpose
computer rather than like a hardwired rule in a special-purpose computer.
I suggest that there is a striking lack of this further evidence with human
cognitive processing. Indeed, as I shall try to demonstrate, the evidence is
much more against represented rules than for them.
9 In recent years ‘‘the theory–theory’’ has become a popular explanation of folk psychological
judgments; see, for example, many discussions in Davies and Stone 1995. Our minds are thought to
contain a representation of a largely true theory of human psychology which we then apply to yield
our psychological judgments about one another. The explicit inspiration for this theory of mind is
the standard Cartesian explanation of intuitive linguistic judgments: just as those judgments are
derived from a representation of a theory of the language so also are the psychological judgments
derived from a true psychological theory. I rejected that explanation of linguistic intuitions in
Chapter 7. If I am right in this rejection then the theory–theory loses its inspiration. Without that
inspiration, it is unlikely that the theory–theory’s explanation of psychological judgments will seem
appealing. In any case, I think it should be rejected for a similar reason: a more modest explanation is
more plausible. A few platitudes like ‘‘If x humiliates y then y is likely to hate x’’ may well be
represented and play a role in psychological judgments. But, according to the more modest
explanation, these judgments are largely the result of the folk being ‘‘hardwired’’ by nature and
nurture to make them intuitively, without reXection or inference, and fairly accurately.
Language Use 205
2. Call the processing rules for Oscar’s production of (his idiolect of ) English,
‘‘P’’. So, according to position (i), P includes the structure rules of English which
are represented and applied in language production. According to position (ii), P
consults a representation of those rules in production. Either position will, of
course, guarantee that P respects those rules as it must; see (m) (3.4).
Assume LOTH and consider what must happen in a simple case of language
production. Oscar intends to express the thought that water is not good to walk
on. As a result he moves a mental sentence <Water is not good to walk on> that
means WATER IS NOT GOOD TO WALK ON into ‘‘the Message-Out box’’.
The task of expressing this message is to ‘‘translate’’ that mental sentence into an
English sentence with that meaning; for example, into ‘Water is not good to walk
on’: P must function as a translation manual. According to (i) and (ii) this
translation involves the metalinguistic theory of English. How might that theory
be brought to bear? The input for the translation task has to be the mental
sentence <Water is not good to walk on>. So we have an immediate problem:
this is not something that the metalinguistic theory can do anything with at all.
For, that theory is a theory of English not Mentalese. So, it can yield syntactic and
semantic information about the likes of ‘walk’ and ‘water’ but has nothing to say
about the likes of <walk> and <water>.
(i) One proposal might be that Oscar has another (tacit) metalinguistic theory,
a theory of Mentalese. The idea would be that having this theory largely
constitutes his competence in Mentalese just as having the earlier theory largely
constituted his competence in English. Indeed, if language use requires that
Oscar’s English competence involves representing its rules, we might expect
that language use would also require that his Mentalese competence involves
representing its rules. Now this further metalinguistic theory does yield infor-
mation about <walk> and <water>. The theory of Mentalese could, of course,
handle the input, <Water is not good to walk on>, and we can perhaps then
imagine how it might be possible for this theory to work with the theory of
English to yield the desired translation. For example, the theory of Mentalese
supplies Oscar’s processor with the information that <water> means WATER
and the theory of English, that ‘water’ does too. We can imagine how the
processor could combine these two bits of information to conclude that ‘water’
is the way to translate <water>.
But, of course, this proposal is outlandish and has surely never been made.
(Note that Fodor insists that the rules of Mentalese are not represented; 1975:
65–6.) A major problem for the proposal becomes obvious when we remember
Harman’s argument discussed in section 6.2 and ask the question: What language
would these metalinguistic theories be in? Suppose that the language were
Mentalese (so, on position (i), P would be analogous to an English-French
dictionary written in English). But then competence in Mentalese could not
consist in representing its rules because that representation presupposes the
competence; we would be caught in a vicious circle. Suppose next that the

language was not Mentalese but, say, Metamentalese (so, on position (i),
P would be analogous to an English-French dictionary written in German).
But then what about the competence in Metamentalese? To avoid a regress we
must suppose that this competence does not require representing the rules. But
then why not suppose this for competence in Mentalese, thus abandoning the
proposal?
(ii) Abandoning this hopeless proposal, a much better one suggests itself:
P takes the mental sentence as input and yields its meaning in some fairly
brute–causal way. For example, it takes <water> and yields WATER. The
metalinguistic theory of English, represented presumably in Mentalese, can then
be brought to bear on this meaning, concluding that ‘water’ is the way to express it.
According to (i) there are two theoretical ‘‘middlemen’’ between Mentalese
and English, between <water> and ‘water’. (ii) drops the Wrst, the tacit theory of
Mentalese. A further proposal leaps to mind: (iii) we should also drop the second,
the tacit theory of English. So in language production we go in a fairly brute-
causal way from <water> to ‘water’ without applying any theory of either; the
process is not rational but more like an inner transduction or reXex. Of course,
(iii) gives RT no role in explaining language production. Yet (iii) seems prima
facie more plausible than (ii): representing the rules seems an unlikely way to
accomplish what the system needs to do. According to (ii) Oscar does not need a
representation of Mentalese rules to go from <water> to WATER. Why then
suppose that we need a representation of English rules to go from WATER to
‘water’? Pylyshyn’s Razor counts against this supposition. Indeed, the Razor even
counts against the supposition that P must operate on metalinguistic represen-
tations of the properties of ‘water’, it counts against version (a). What use would
it be to P to represent the fact that ‘water’ means WATER in going from WATER
to ‘water’? (We shall say more along these lines in section 11.6.) Evidence from
language production might, of course, show that we do need these representa-
tions but, I shall argue, it does not.
We have been considering language production but the point could also be
made for language comprehension. Call the processing rules for Oscar’s English
comprehension, ‘‘U’’. It seems prima facie unlikely that U will need the help of a
representation of the rules of English.
For the sake of argument I have assumed LOTH. But, so far as I can see, the
argument would not be aVected if we assumed that the representations involved
in thoughts are not language-like. Thus, suppose they are map-like, as RTM
allows. The metalinguistic theory of English can do no more with a mental map
that represents that water is not good to walk on than it can with a mental
sentence that does. An analogue of (i) according to which the translation involves
a theory of the mental map is just as outlandish as (i) according to which the
translation involves a theory of Mentalese. And the analogue is just as open to the
Harman argument. Once that ‘‘middleman’’ theory is dropped and we move to
Language Use 207
an analogue of (ii), the further move to drop the other ‘‘middleman’’ theory, the
theory of English, and adopt the analogue of (iii) will seem enticing.
What about if we abandon RTM and hence the idea that thoughts involve
mental representations? If we are that noncommittal about the nature of thoughts,
we will have diYculty saying anything much about their role in language
processing. It is hard to see how this diYculty will make RT more plausible.
3. I suggested earlier that structure rules governing a representational input or

output often seem to be the wrong sort of rules to be processing rules: they do not
seem to be the sort of rules that would govern a process that the system actually
goes through (2.4, 2.6, 3.2, 4.1). This is certainly the case with the logic machine
and is very likely the case with the bee’s waggle dance (2.2). Similarly, I suggested
that some linguistic structure rules—for example, the rules for forming D-
structure phrase markers and transformation rules—seem to be the wrong sort
of rules to be the processing rules governing language use: the computational
procedures that Chomsky identiWes with an I-language are not plausibly seen as
psychologically real processes (4.1). These suggestions become particularly telling
given RTM and the view that language use is a translation process: the structure
rules seem particularly unsuitable for the process of translation. Thus, if (t) is
correct, each translation task is between a linguistic representation that has a
certain structure and a mental representation that has a similar one. It is hard to
see how the rules determining that these representations have those structures
could govern the process of matching the structures. So my third tentative proposal
is that the structure rules of a language are the wrong sort of rule to govern the
process of language use. This proposal counts not only against position (i)
according to which the structure rules are represented and applied in processing
but also against (iii) according to which those rules govern the processing
without being represented. It thus gives support to (v) according to which the
processing rules are unlike the linguistic rules. The case for the third tentative
proposal continues in 4 below and in sections 11.6 to 11.8.
4. A further problem for positions (i), (ii), and (iii) arises from an old criticism
of the claim that a grammar’s rules are psychologically real. I expressed the
criticism earlier as follows: ‘‘If we can come up with one grammar for a language,
we can come up with many which, though they posit diVerent syntactic rules, are
equivalent in their explanation of meaning: they are equally able to capture all the
syntactically determined facts about meaning. We need psycholinguistic evi-
dence to show which grammar’s rules are in fact playing the role in linguistic
processing, evidence we do not have’’ (2.6). According to (i), (ii) and (iii), ‘‘the
structure rules of the language’’ are psychologically real. These rules must, of
course, be the rules described by a true grammar of the particular linguistic
reality. Call our preferred grammar, ‘‘G1’’. But, as the old criticism points out,
there will be other grammars just as true of that linguistic reality. Call one of these
‘‘G2’’. (Perhaps G1 is a transformational grammar and G2 is a lexical-function
grammar; 4.4–4.5.) This poses a problem additional to those raised in 1 to 3 above.
Thus, suppose that we set aside our doubts that a representation of rules plays a role
in language processing and our doubts that those rules are the right sort to govern
the processing. Why should we suppose, as (i) must, that representations of G1’s
rules govern processing rather than representations of G2’s? Why should we
suppose, as (ii) must, that representations of G1’s rules are used as data in processing
rather than representations of G2’s? Why should we suppose, as (iii) must, that
the unrepresented but embodied rules of G1 govern processing rather than those of
G2? Psycholinguistic evidence might, of course, support one of these suppositions
but we have no reason in advance to choose which one it will support.10
5. The next consideration against RT arises in dismissing something that may

seem to support RT and hence (i) or (ii). It might be thought that RT gets
support from the apparent fact that language comprehension, language produc-
tion, and intuitive linguistic judgments draw on the same linguistic knowledge.
Pylyshyn, for one, once thought so:
we appear to have multiple access to [grammatical] rules. We appeal to them to account
for production, for comprehension, and for linguistic judgments. Since these three
functions must be rather thoroughly coordinated . . . , it seems a reasonable view that
they are explicitly represented and available as a symbolic code. (1980a: 121; he seems
to have changed his mind later: 1991: 247).
I have argued against this idea of multiple access to the grammatical rules. First,
we do not need to see a speaker’s intuitive judgments as derived from a repre-
sentation of those rules. Rather, we should see these judgments as ordinary
empirical central-processor responses to linguistic phenomena. That was part
of my third major conclusion (Ch. 7). And I have argued that the competencies
may share little more than a conceptual competence, the competence to think the
thoughts expressible in the language. The evidence from brain impairment
suggests that they do indeed share little more (10.2). And if we accept the
appealing LOTH and hence position (t)—the structure rules of a speaker’s
language are similar to the structure rules of her thought—that is just what we
should expect. This led to the second tentative hypothesis: there is little or
nothing to the language faculty (10.3). It is hard to see a place for the represented
rules in the little of the language faculty that may be left. Indeed, if the fairly
brute–causal view of language use to be explored in the sections 11.6–11.8 is
correct, there is no place. This brute–causal view yields a ready explanation of the
10 Note that equivalent grammars do not pose a similar problem for my Wrst tentative proposal
that the rules of a speaker’s language are similar to the structure rules of her thought (9.5). For
the purposes of this proposal, the rules of G1 and G2 count as the same because they determine the
same syntactic properties. What matters to the proposal is that the syntactic properties of the
speaker’s language are similar to those of her thought.
Language Use 209
little shared by the competencies: the association between conceptual content
and language ‘‘runs both ways’’: ‘‘it may be the case that the lexical network
allows activation to spread in both directions because it is used for both produc-
tion and comprehension’’ (Dell 1995: 195–6).
6. The discussion in Fodor, Bever, and Garrett 1974, summarized in section 4.3,
brings out the implausibility of position (ii). How can the represented rules be
used as data in language use? Consider language comprehension. Suppose that,
somehow or other, the processing rules come up with a preliminary hypothesis
about the structure of the input string. In principle, the represented rules might
then play a role by determining whether this hypothesis could be correct (assum-
ing that the input is indeed a sentence of the language). The problem in practice is
that to play this role the input would have to be tested against the structural
descriptions generated by the rules and there are just too many descriptions. The
‘‘search space’’ is just too vast for it to be plausible that this testing is really going
on in language use. This led Fodor, Bever, and Garrett to explore the idea that
heuristic rules not representations of linguistic rules govern language use.
7. What advantage does an object get from having its processing rules repre-
sented rather than just embodied? A consideration of the standard general-
purpose computer provides one answer. Because the computer represents the
rules of software programs it is relatively easy to change what the computer does:
we simply change the program; i.e., load the computer up with a program which
represents diVerent rules. In contrast, it is relatively hard to change what a
machine like a special-purpose Turing Machine, a dedicated wordprocessor, or
a calculator does by changing an unrepresented rule, a rule built into its
hardware: we have to call on an engineer to do some rebuilding. The Xexibility
and ‘‘plasticity’’ that comes from representing rules is also apparent in human
aVairs. Because we represent the rules of games and of etiquette, it is fairly easy to
change those rules. Think particularly of the way children’s games often proceed:
the rules for a game are invented and agreed upon; the game is played for a bit
and found wanting in certain ways; a modiWcation of the rules is proposed; there
is more play followed by further modiWcations. These changes are easy because
the children represent the rules, indeed, they state them. Now the problem for
positions (i), (ii), and (iv) is that linguistic competence is singularly lacking in
this sort of plasticity. As Stabler points out:
we do not Wnd any evidence of plasticity with regard to linguistic competence that would
indicate that an encoding of the grammar inXuences the operation of any sort of human
computing system. (1983: 399; see also Pylyshyn 1991: 243–8)11
11 A sign of this lack of plasticity, pointed out to me by my student Francesco Pupa, is that people
learning a second language mistakenly import features of their Wrst language into the second. Thus
speakers of a Slavic language like Croatian tend to omit the deWnite and indeWnite articles of
English; and Italians tend to add vowels to the ends of English words that should end with
8. Language processing is extraordinarily fast. Indeed, as Fodor points out, ‘‘it
may be that . . . the eYciency of language processing comes very close to
achieving the theoretical limit’’ (1983: 61). Representing rules which have to
be applied or consulted is an unlikely way to achieve this eYciency. This also
counts against (i), (ii), and (iv)
Matthews brings these two considerations together nicely: ‘‘the limited plasti-
city (and lability) of acquired grammatical competence would seem to render
explicit representation unnecessary, while the relatively greater eYciency of
processors whose programs are ‘hardwired’ rather than explicitly represented
would seem to render it undesirable’’ (1991: 187).
The Wnal and most important consideration, 9, against the representation of
rules in linguistic competence deserves a section of its own. This consideration
arises from the psychology of skills in general.
11.5 P S Y C H O LO G Y O F S K I L L S
9. Animals, birds, and insects, exhibit a great range of skills. So too do humans;
for example, ball catching, bicycle riding, piano playing, touch typing and
adding. Many human skills are motor but some are cognitive. Indeed, it is
important to note the large role of cognitive skills: ‘‘much of our mental activity
can be understood in terms of skill acquisition—we acquire skills for reading,
solving problems within a particular domain, recognizing particular patterns,
and so on’’ (Reisberg 1999: 460). Our linguistic competence has all the marks of
a cognitive skill. Thus, we have just noted its limited plasticity and its speed.
These are typical characteristics of skills (Anderson 1980: 226–30; Logan 1988:
492). Another such characteristic is that the process of exercising a skill is
unavailable to consciousness (Logan 1988: 493). This also is a feature of linguistic
competence. Finally, once a skill has become established, it is ‘‘automatic’’ with
the result that it can be performed whilst attention is elsewhere (Anderson 1980:
230–5; Reisberg 1999: 460).12 Once again, this is a characteristic of linguistic
competence. As Fodor points out (1983: 52–5), we normally cannot help but
hear a piece of language as a piece of language with a certain meaning: we cannot
simply hear it as a sound and we cannot, for the most part, choose our interpret-
ation of it. All in all, it is very plausible to think that linguistic competence is a
cognitive skill. So, we can expect to learn something about its nature from
considering the nature of cognitive skills in general. For example, if the rules
consonants. This inXexibility would be puzzling if mastery of the second language were a matter of
learning to represent its rules. The representation of the Wrst language should be simply set aside
whilst representing the second just as one piece of software is set aside whilst using another.
12 A consideration of ‘‘the Stoop eVect’’ demonstrates ‘‘that automaticity is not an all-or-none
phenomenon’’ (Palmeri 2003: 300).
Language Use 211
that govern the exercise of skills in general are typically embodied without being
represented that will count heavily against (i), (ii), and (iv).
Earlier (3.1), in discussing the dancing of the bee, the diving of the king-
Wsher, the ball catching and thinking of humans, I pointed out how prima facie
implausible it is to think that such skills involve representations of the rules
that govern them. Pylyshyn’s Razor lay behind this assessment: if we have good
reasons for supposing that a system is governed by a rule, the Razor demands
that we produce further reasons before supposing that the rule governs by
being represented and applied. So the assessment presumes that this demand
has not been met. Does the psychological study of skills support this presump-
tion? I shall argue that it does: there seems to be no persuasive evidence for the
representation of the rules. Do we Wnd evidence for a stronger conclusion,
evidence that the rules are not represented? I think that we do but the evidence
is far from decisive. The psychological study is still at an early stage and
we simply do not know enough about skills to draw the stronger conclusion.
Aside from that, although such progress as the study has made does mostly
support the stronger conclusion, one part of the study does not and has to be
explained away.
Psychologists agree about several matters. (i) As pointed out earlier (3.1), they
distinguish, rather inadequately, between two sorts of ‘‘knowledge’’, ‘‘declara-
tive’’, and ‘‘procedural’’. Where declarative knowledge is explicit, accessible to
consciousness, and conceptual, procedural knowledge is implicit, inaccessible to
consciousness, and subconceptual. And where declarative knowledge involves
explicit or declarative memory, procedural knowledge involves implicit or pro-
cedural memory. Declarative memory holds factual knowledge such as that
Washington is the capital of America, while implicit memory holds rules that
govern processes, ‘‘routinized skills, . . . priming, and classical and operant
conditioning’’ (Bjorklund et al. 2003: 1059). The evidence for the dichotomy
between implicit and explicit memory, the dichotomy between procedural and
declarative knowledge, and other related dichotomies ‘‘lies in experimental data
that elucidate various dissociations and diVerences in performance under diVer-
ent conditions’’ (Sun 2003: 698).13 (ii) Although declarative knowledge may play
a role in learning a skill, there is consensus that the skill itself is a piece of
procedural knowledge. (iii) There is consensus also that cognitive skills are like
motor skills: ‘‘recent work across a wide range of disciplines now provides
evidence for the view that ‘skills of mind’ and ‘skills of eye, ear, and muscle’ are
fundamentally similar’’ (Rosenbaum et al. 2001: 454; see also Masson 1990;
Carlson 2003); motor skills are ‘‘set apart by their emphasis on the movement of
the limbs and torso as well as on the outcome of the movement in terms of the
goal of the act’’ (Newell 1996: 441).
13 See also Schacter 1999 (p. 394), the many results cited by Sun et al. 2001 (p. 207); Cleeremans
2003 (p. 492); Mulligan 2003 (pp. 1115–17); Reber 2003 (p. 491).
In light of this, our view of the nature of skills depends on our view of the
nature of procedural knowledge and implicit memory. Here there are lots of
interesting ideas but no consensus. Indeed, our knowledge of this matter is at an
early stage: ‘‘a great deal remains to be learned about the cognitive and neural
mechanisms of implicit memory’’ (Schacter 1999: 395); ‘‘the overall picture that
emerges from just over a century of scientiWc research is that human long-term
memory . . . is varied, dynamic, and constructive, and quite unlike current human-
made memory devices in virtually every important respect’’ (Richardson-Klavehn
and Bjork 2003: 1104); ‘‘there is no consensus regarding the details of the
dichotomies’’ mentioned in the last paragraph (Sun 2003: 698).14 The key
issue for us is whether or not such knowledge as we do have gives any support
to the view that the rules that govern a skill are represented. This is not the place,
of course, for an exhaustive discussion of the literature on skills. Still, we can learn
something helpful from a brief discussion.
Consider Wrst what we can learn from the literature attending particularly
to motor skills. A recent encyclopedia article describes four challenges facing
research on motor control and describes many models proposed to meet these
challenges. It is striking that there is no suggestion that the rules involved in these
models are represented in the organism. The Wrst challenge is the redundancy
problem; there are, for example, a near inWnite number of ways to pick up a
particular apple. So, ‘‘we need to explain how one movement is chosen from the
plethora that are possible’’ (Brown and Rosenbaum 2003: 127). Four types of
model have been proposed: ‘‘models that emphasize properties of the peripheral
neuromotor system; models that emphasize eVector interactions; models that
emphasize geometric restrictions; and models that emphasize cost reduction’’
(p. 128). The only mentalistic talk in the discussion of these models is of ‘‘desired
positions’’ being ‘‘represented as modiWable equilibrium positions’’. The second
challenge is to model relations between movements and their eVects. Here there
is talk of ‘‘predicting sensory outcomes’’ and ‘‘comparing’’ these to ‘‘obtained
feedback’’, ‘‘predicting what change in motor commands will correct a mismatch
between intended and obtained results’’. The third challenge is to capture the way
an organism, even an insect, anticipates ‘‘the perceptual consequences’’ of actions
‘‘(feedforward)’’ and to capture the ‘‘rapid correction for perceived errors (feed-
back)’’ (p. 129). The proposal for feedforward—the ‘‘Smith predictor’’—is that
‘‘a controller not only sends signals to the muscles to bring about immediate
perceptual changes; it also preserves a copy of the expected perceptual changes
and compares them with actual feedback signals when they arrive’’. We seek
models for feedback that take account of ‘‘Fitts’ law’’15 in trying to ‘‘optimally
compensate’’ for ‘‘the inherent variability of motor control’’ (p. 131). The fourth
14 Axel Cleeremans (2003) refers to ‘‘as many as eleven diVerent deWnitions’’ of the ‘‘implicit
learning’’ that is thought to be involved in skill acquisition (p. 491).
15 ‘‘MT ¼ a þ b log2(2A/S)’’ where MT is the movement time, S is the size of the target, A is the
distance to be moved, and a and b are constant parameters (Mon-Williams et al. 2003: 122).
Language Use 213
challenge is ‘‘movement sequencing’’ ‘‘studied mainly in connection with such
tasks as walking, speaking, typewriting, and drawing’’ (p. 127). There is behav-
ioral evidence of ‘‘advance planning’’; for example, ‘‘performance errors (e.g. slips
of the tongue) reveal implicit knowledge of what will be said or done’’.
Heirarchical models have been proposed according to which ‘‘serial ordering of
behavior is achieved by the unfolding of high-level goals into lower-level con-
stituents’’ (p. 131). In sum, this article contains some rather mentalistic language
(even in talking of insects)—‘‘desired’’, ‘‘represented’’, ‘‘predicting’’, ‘‘comparing’’,
‘‘intended’’, ‘‘anticipates’’, ‘‘planning’’, etc.—but no talk of represented rules.
Similarly, another article (Mon-Williams et al. 2003) talks of ‘‘the nervous
system’’ having ‘‘preplans’’ (p. 124) and ‘‘storing information’’ (p. 126), and
another (Wolpert and Ghahramani 2003) of its mapping, knowing (p. 138),
estimating, and predicting (p. 139), but neither talk of its representing the rules
by which the system performs these actions.
Turn next to ‘‘dynamical systems theories’’, which ‘‘de-emphasize mental
representations’’ altogether whilst attempting to ‘‘provide substantial detail con-
cerning how skills are actually performed’’. These theories have been ‘‘prominent
among researchers concerned with motor skills’’ (Carlson 2003: 41); they are
‘‘particularly appropriate to account for motor control’’ (Port 2003: 1028). But
theories of this sort aim to cover cognitive skills as well: ‘‘a provisional consensus
seems to be emerging that some signiWcant range of cognitive phenomena will
turn out to be dynamical’’ (van Gelder 1999: 345); the theories have ‘‘a record of
success across a wide range of cognitive abilities, including perception, sensori-
motor activity, language, attention, decision making, and development’’ (Garson
2003: 1036). They are also concerned with skill learning treating it ‘‘as the
process of discovering ways of coordinating activity’’ (Carlson 2003: 41). (So
they bear also, as do the theories to follow, on the concerns of the next chapter.)
Coordination is, indeed, central to the dynamic approach. Thus, J. A. Scott
Kelso, in his inXuential book, Dynamic Patterns (1995), claims that ‘‘coordina-
tion . . . is a fundamental feature of life’’ (p. xi) and aims to discover ‘‘the laws
and principles of coordination’’ that govern it (pp. 287–8). So, what is a
dynamical system? It ‘‘is a set of quantitative variables changing continually,
concurrently, and interdependently over quantitative time in accordance with
dynamical laws, described by some set of equations’’. Modeling it makes
‘‘heavy use of calculus and diVerential or diVerence equations’’ (van Gelder
1999: 245). A key feature for our purposes is that these equations have no
place for representations of rules and very little for representations at all:
These equations are not ordinarily deWned over representations, but instead over the
variables for the system’s properties. From the [Dynamical Systems] point of view,
representations are not essential to an explanation of the mechanisms of cognition,
since what matters is the way in which system variables evolve according to the system’s
equations of motion. Although dynamical explanation may mention representations,
these are conceived of as emergent aspects of system activity. (Garson 2003: 1035)
Two other approaches have a lot in common with the dynamical systems
approach.16 First, the Gibson-inspired ecological approach (Fowler and Turvey
1978; Kugler and Turvey 1987) ‘‘seeks the solution to motor skills through
the mapping of perception to action with minimal appeal to representational
processes . . . . Rather the movement form may reXect emergent properties of
the self-organizing biological system, in a fashion that is consistent with pattern
formation principles of complex physical systems that drive, for example cloud
formations, sand-dune formations, and vortices in streams’’ (Newell 1996: 442).
Second, connectionist networks (Rumelhart and McClelland 1986) have had
some of their most conspicuous successes in accounting for skills. As is well
known, these networks store information not in represented rules but in patterns
of connection weights among units. ‘‘Neural network models are capable of
learning complex sets of input-output patterns, including those in which some
logical rule governs the transformation of input to output patterns . . . they are
able to do so without developing an explicit representation of the rule’’ (Masson
1990: 233–4). Many think that a general account of skills might be based on a
connectionist architecture (see, for example, Sun et al. 2001).
Next, consider the inXuential ‘‘instance theory’’ proposed by Gordon Logan.
This theory is friendly to representations but makes no mention of the repre-
sentation of rules. On this theory, the Wnal ‘‘automatization’’ stage of learning a
skill is ‘‘the acquisition of a domain-speciWc knowledge base, formed of separate
representations called ‘instances’ of each exposure to the task’’. Processing at that
stage ‘‘relies on retrieval of stored instances which will occur only after practice in
a consistent environment’’ (1988: 492). ‘‘Attending to a stimulus is suYcient to
retrieve from memory whatever has been associated with it in the past’’ (p. 493).
Crudely, you perform the action appropriate to the stimulus because you
remember doing so before.
So far, then, we have found no sign in discussions of skills of rules being
represented. That changes when we consider theories proposing ‘‘production
systems’’. I shall take John Anderson’s ‘‘adaptive control of thought’’ (‘‘ACT’’)
theories as my example (1983; 1993; Anderson and Lebiere 1998).17 The basic
idea for skill acquisition, which has gone through several developments, is that
declarative knowledge in working-memory, and practice, leads over time to the
accumulation in long-term memory of representations of production rules,
rules that if a certain condition obtains then a certain action is to be performed
(an IF–THEN rule), and to speeding up the application of these rules. Consider
16 Tim van Gelder sees ecological psychology as an example of ‘‘dynamical work’’ (1999: 244)
and Kelso expresses sympathy for it (1995: 34–7). Van Gelder thinks that the dynamical and
connectionist approaches ‘‘overlap’’: ‘‘Connectionism is best seen as straddling a more fundamental
opposition between dynamical and classical cognitive science’’ (1999: 245); ‘‘neural networks are
examples of dynamical systems’’ (Garson 2003: 1038; see also Port 2003: 1029).
17 SOAR theories (Laird et al. 1987) provides another example. Jones 2003 is a helpful account
of production systems.
Language Use 215
learning to change gears in a stick-shift car, for example. You start with declara-
tive knowledge acquired from a set of instructions: ‘‘First, take your foot oV the
accelerator, then disengage the clutch’’, and so on. With practice, ‘‘procedulariza-
tion’’ occurs and all these processes become automatic. What then is the skill
itself, on this view? It is the accumulated representations of production rules, and
the capacity to apply the represented rules speedily in performance without
calling on working-memory. Indeed, procedularization may well be accompan-
ied by the loss of the declarative knowledge in working-memory. This theory ‘‘de-
emphasized the perceptual-motor details of how skills are performed’’ (Carlson
2003: 41). In this respect, and in its emphasis on representations, it is the
opposite of the dynamical systems theory.
ACT theory does a nice job of accounting for many of the phenomena of skill
learning including the transfer of training from one skill to another (Singley and
Anderson 1989; Masson 1990: 222–8; Sun et al. 2001: 205–6). ACT-R theory,
the most recent ACT theory, ‘‘accounts for a wide range of cognitive phenomena,
including perceptual-motor tasks, memory, and problem-solving’’ (Johnson et al.
2003: 32).18 The key question for us is whether this achievement requires that
the production rules that are learnt be represented. Descriptions of ACTare often
vague on this matter but the received view is that the rules are represented: ‘‘in
ACT both declarative and procedural knowledge are represented in an explicit,
symbolic form (i.e. semantic networks plus productions)’’ (Sun et al. 2001: 235;
see also Masson 1990: 223). Yet it is notable that Christian Lebiere, one of the
developers of ACT, does not list this view of production rules among ACT’s
‘‘three theoretical assumptions’’ (2003: 8) nor among its ‘‘four claims related to
production rules’’ (p. 9). And since the production rules constitute procedural
not declarative knowledge there seems to be no immediate and pressing need to
take them as represented. Because ACT theories are based on general-purpose
computer models (Anderson 1983: 2) it is perhaps not surprising that the
cognitive architecture they propose involves the representation of production
rules. Still, we wonder whether we should take this aspect of the model seriously
if we are looking for a simulation of skills that exist in real organisms. Is there any
reason to think that the IF–THEN rules that become embodied in an organism
as a result of practice are represented rather than merely embodied? Perhaps we
can suppose that the organism has simply learnt to respond to the working-
memory representation of a certain condition with the appropriate action. Is
there any explanatory gain in supposing further that it does this by representing
the rule governing this response and applying it? Anderson himself remarks that
‘‘the production is very much like the stimulus-response bond’’ (Anderson 1983: 6).
And Pinker calls it ‘‘a knee-jerk reXex’’ which is ‘‘triggered’’ (1997: 69). Should
18 It also seems to explain skill at chess. The idea is that this skill consists in acquiring production
rules each of which speciWes an action given a certain chess position. Chess masters may have as
many as 50,000 such rules (Newell and Simon 1972). However this picture of chess playing is
controversial (Holding 1992).
we then apply Pylyshyn’s Razor and take the rule to be simply ‘‘Wred’’ when the
representation of the condition matches the rule’s IF-component?19
The general-purpose computer may mislead here by encouraging the idea that
learning a skill is a matter of reprogramming, a matter of changing the rules
represented in the software. Yet it seems more likely that this learning is a matter
of rebuilding, a matter of building new rules into the hardware; a new special-
purpose computer is acquired. Experience may do for us what the engineer does
for the computer.
So, I am doubting that we should be led by the production-systems approach
to conclude that a skill must involve representations of production rules. This
doubt is supported by the following considerations. (i) I have often harped on the
fact that the human skill of thinking, the most cognitive skill of all, could not
consist generally in applying representations of the ‘‘laws of thought’’ (3.1,
7.3, 7.5). So, this skill, at least, need not involve represented rules. (ii) The
focus of the production-systems approach is on skill learning. But many skills
seem to be largely, if not entirely, innate. One thinks immediately of the skills of
insects—the bee’s dance is an elaborate example—and of animals. The primitive
motor skills of humans—grasping objects, walking, etc.—also seem to be of this
sort. And, once again, there is our skill at thinking, which is surely largely innate.
The motivation for represented production rules that comes from the produc-
tion-systems approach to learning is absent with these largely innate skills. (One
might insist that skills are learned by deWnition,20 but this would be a pointless
maneuver. Something with the nature of a skill that is, as a matter of fact, learnt,
could, in another possible world, be innate. Its nature does not depend on its
source. And we expect what I am calling ‘‘largely innate skills’’ to have natures
importantly similar to acquired skills. A general theory of skills should be
interested in that similarity.) (iii) The production-systems approach yields a
totally ‘‘top-down’’ picture of skill learning: it all starts from declarative know-
ledge. Yet, intuitively, a lot of skill learning seems to be ‘‘bottom-up’’: rather than
starting with declarative knowledge, we observe, practice, and ‘‘just pick the skill
up’’. Even when we start with declarative knowledge, it often seems, intuitively, as
if that knowledge plays a relatively small part in the acquisition of the skill; think,
for example, of learning to throw a frisbee or ride a bicycle. These intuitions are
conWrmed by the literature on implicit learning, to which we now turn.
A. S. Reber deWnes implicit learning as follows: ‘‘the capacity to pick up
information about complex stimulus displays largely without awareness of either
the process or the products of learning’’ (2003: 486). There is much evidence that
a lot of skill learning is of this bottom-up sort. Classical conditioning is an
19 Note that this claim has no bearing on whether the mind is unitary as Anderson (1983) holds
or modular as Fodor (1983) holds.
20 ‘‘Skill refers to an acquired ability that has improved as a consequence of practice.’’ (Carlson
2003: 36)
Language Use 217
example. And implicit learning has been famously exempliWed in sequential (or
serial) reaction time (‘‘SRT’’) and artiWcial grammar (AG) tasks. In SRT tasks,
subjects presented with a sequence of target stimuli learn to exploit the structure
in the sequence even though generally unaware of that structure (p. 487). In AG
tasks, subjects memorize ‘‘exemplary strings’’ and are then asked to classify novel
strings for well-formedness. Their performance at this task shows that they have
‘‘a considerable amount of knowledge about the underlying structure of the
symbol strings. The learning appears to be taking place largely independent of
awareness, and subjects Wnd it particularly diYcult to communicate to others the
knowledge they are applying’’ (p. 488; see also Reber 1989). Sun et al. (2001) cite
a great deal of evidence that ‘‘individuals may learn complex skills without Wrst
obtaining a large amount of explicit declarative knowledge . . . and without
being able to verbalize the rules they use’’ (p. 207).
What are the implications of the study of implicit learning? In particular, what
does the study add to our understanding of the nature of skills? According to Axel
Cleeremans, computer models show that ‘‘elementary, associative learning
processes (as opposed to rule-based learning) are in fact often suYcient to account
for the data’’ of implicit learning (2003: 496). ‘‘it is clear that the knowledge
acquired in typical implicit learning situations need not be based on the
unconscious acquisition of symbolic rules’’ (p. 497). Stanley et al. suggest that
the knowledge that is exploited in performing a task is a ‘‘memory for past
sequence of events related to the task’’ (1989: 571). Mathews et al. suggest that
it is ‘‘memory-based processing, which automatically abstracts patterns of family
resemblance through individual experiences with the task’’ (1989: 1098). Reber
notes that the bottom-up systems of implicit learning ‘‘are rather easily simulated
by connectionist architectures’’ (2003: 487). Mathews et al. (1988) assume
something like a connectionist model. Reber points out that even a sea slug can
exhibit implicit learning in Pavlovian conditioning (2003: 489). These discus-
sions clearly count heavily against the idea that skills are governed by represented
rules. Finally, if implicit learning were largely a matter of acquiring representa-
tions of rules that govern the performance of a task and that yield verbal reports
about the task—which, it will be remembered, is the received view of language
learning—we would expect improvement in performance to be matched by
improvement in verbal reports. Yet that is not what we Wnd at all, as we pointed
out earlier in discussing the source of linguistic intuitions (7.5).
Consider, Wnally, the following thoughts on how implicit learning is imple-
mented in the nervous system. Reber thinks that this learning is ‘‘probably
supported by neurological systems that are old evolutionarily and antedate
those ‘top-down’ systems that are dependent on complex encoding and conscious
control, the hallmarks of explicit systems’’ (2003: 486). He airs the hypothesis
that ‘‘implicit learning and memory occur within the cortical areas involved in
processing the stimuli for which the learning is occurring, and hence will be
found virtually throughout the brain’’ (p. 491). Yamadori et al. report as follows:
‘‘In recent years it has been made clear that neurological substrata of skill
acquisition is diVerent from those engaged in data-related information acquisi-
tion . . . . The former is named procedural memory and the latter declarative
memory’’ (1996: 49). ‘‘Unlike declarative memory which related mainly with the
thalamus, hippocampus, and neocortical structures, this perceptual-verbal skill
[mirror reading] acquisition is very likely related with the subcortical structures’’
(p. 51). Finally, consider the following summary of the Wndings of Posner et al.
(1997) based on brain imaging:
there are four mechanisms that work together to generate skill acquisition. Firstly,
automaticity is a result of an increase in the links between previously isolated subsystems.
Secondly, when skills are acquired, the size or the number of brain areas involved
increases. As a result, the computation within modules is improved. Thirdly, skill
acquisition could be due to the replacement of the initial components by other more
eYcient components. Finally, certain brain areas can start performing tasks formerly
implemented by diVerent areas, indicating circuit change. ( Johnson et al. 2003: 34)
These Wndings about the implementation of skills do not, of course, rule out that
the rules governing skills are represented in the mind, but they certainly do not
encourage that view. And they are congenial to the earlier idea that skill learning
involves ‘‘rebuilding the hardware’’ rather than ‘‘reprogramming the software’’.
So what conclusions should we draw about the nature of skills from these
psychological studies? The studies yield a range of interesting ideas but no very
conWdent conclusion about this nature. First, the ideas are seriously incomplete
in their details. Second, it is hard to see a rational basis at this time for a sweeping
acceptance or rejection of the ideas of one or other theoretical camp. Rather, we
should entertain many ideas whilst often suspending judgment on their truth.
We are clearly at an early stage in our understanding of skills. Nonetheless, we
can say conWdently that the studies have not provided persuasive evidence that
skills typically involve represented rules. This is not to say, of course, that the
evidence shows deWnitively that the skills do not involve represented rules. Still,
the literature on motor skills, on dynamical systems theories, on connectionist
theories, on instance theories, on implicit learning, and on the implementation
of skills counts heavily against the idea that skills do.21 The only support for the
idea that they do may come from production-systems theories. But, as I have
noted, there are reasons for doubting the appropriateness of these theories’
apparent commitment to represented rules in a general theory of skills.
What is the bearing of all this on the matter of linguistic competence? Well, we
are surely right in assuming that linguistic competence is a skill. So we should
expect to form conclusions from the study of that competence that are similar to
21 So does the literature on the related issue of animal navigation and cognitive maps; see, for
example, Gallistel 1990. In an article on this topic, Bruno Poucet claims that ‘‘our understanding of
memory has considerably improved in recent years . . . largely due to the use of animal models’’
(2003: 150). So one wonders why the literature discussed in this section contains no references to
these models.
Language Use 219
those from the study of skills in general. So, we should expect to discover that it is
early days in our understanding of the nature of linguistic competence, as indeed
we have discovered (11.2), with the result that we should mostly suspend
judgment about the matter. And rather than Wnding persuasive evidence for
RT we should expect to Wnd a weight of evidence against it. Perhaps our study of
language in particular confounds these expectations. But I shall argue later that it
does not (11.7–11.8).
Finally, the psychological study suggests that if learning a Wrst language is
implicit learning then we should be particularly dubious of RT. And language
learning seems to be a paradigm of implicit learning: ‘‘Natural languages are
acquired with substantial contributions from implicit acquisitional mechanisms’’
(Reber 2003: 486; see also Cleeremans 2003: 492).22 In support of this, we note
that linguists are fond of emphasizing how little explicit instruction the child gets
whilst acquiring a language at a remarkable rate. Of course, acquisition could still
begin with declarative knowledge if the child, all on its own, arrived at suYciently
many explicit hypotheses about the language. Anderson assumes that the child
does (1983: ch. 7) but this seems rather dubious. Finally, although competent
speakers typically show some capacity to describe this knowledge in their
intuitive linguistic judgments, this capacity is very limited (ch. 7).
All in all, the psychological study of skills—of their nature and of how they are
acquired—is bad news for the view that language processing rules are
represented. This adds to the bad news for various positions presented in the
last section. That evidence was brieXy as follows: should we Wnd evidence of a
processing rule, there is always the possibility that it is hardwired rather than
represented; it is implausible that the translation task of language processing
will involve a theoretical middleman; the linguistic rules are the wrong sort of
rules to be processing rules; we have no reason to suppose that the rules of our
preferred grammar are represented rather than those of another meaning-equiva-
lent grammar; it is hard to see a place for the represented rules in the little that
competencies may share; it is unclear how represented rules could be used as data;
language processing is not plastic enough and is too fast for it to involve
represented rules. In light of these considerations, it seems unlikely that positions
(i), (ii), or (iv) will be the conclusion of a persuasive future abduction; such
an abduction is unlikely to seem plausible given what else we know. And the
only-theory-in-town abduction (11.2) seems totally unpersuasive.
In my initial discussion of this abduction I noted the consensus that we know
very little about language processing at this stage. In particular, we do not have a
worked out and persuasive theory of this processing that includes RT. So the issue
with that thesis is whether it is likely to be part of the best worked out and
22 Many linguists would doubtless disagree: ‘‘There has been something like a consensus in
linguistics for about twenty years that [language acquisition is not a case of implicit learning]’’
(Anonomous referee of Devitt 2006).
persuasive future theory (11.2). In pondering this, mindful of criterion (B)’s

requirement that a good abduction should feature an explanation that is better
than any that is likely given what we already know, we should note that if we
knew more about how the processes underlying other skills worked without
representing rules we could likely explain how language processing worked
similarly. It would be rational to accept RT if we had a good worked-out theory
of language use involving RT and little prospect of a good theory that does not
involve it. This is very far from our present situation.
11.6 B RU T E – C A U S A L P RO C E S S I N G
The main focus of our preliminary assessment in the last two sections has been on
positions (i), (ii), and (iv) and the issue of the representation of rules. We have
also attended brieXy to position (iii) and the issue of linguistic rules governing
processing without being represented. We have barely touched on the idea,
arising from the distinction drawn in section 3.3, that language use may be a
fairly brute–causal associationist process—version (b)—rather than a process
involving metalinguistic representations of the syntactic and semantic properties
of linguistic expressions—version (a). We turn now to that idea. My fourth
tentative proposal is that version (b) is correct.
First, a clariWcation. Some parts of language use involve abilities other than
those strictly of linguistic competence, abilities that we have been calling
‘‘pragmatic’’. Thus, as we have already noted (8.2), language comprehension
involves assigning referents to indexicals. It also involves using contextual clues to
remove ambiguities. And language use in general involves Gricean processes: in
production, the process of tailoring the expression of thought to the hearers and
circumstances; in comprehension, the process of Wguring out what thought the
speaker is likely to be expressing. Some of these processes may well involve
central-processor reXections on linguistic items. Version (b) is not concerned
with any such process. The processes that are brute–causal according to version
(b) are the speedy automatic ones which may arise solely from a person’s linguistic
competence, from her ability to speak and understand her language, or may arise
partly from that ability and partly from pragmatic abilities like that of assigning
referents. My tentative proposal is that these processes are fairly brute–causal.
Any position that is committed to the representation of rules playing a role in
language processing must be committed also to the version (a) view that meta-
linguistic representations of the syntactic and semantic properties of linguistic
inputs and outputs play a role. So, positions (i), (ii), and (iv) have this com-
mitment automatically. Positions (iii) and (v) do not hold that rules are repre-
sented and so leave the brute–causal alternative of version (b) as a possibility.
Nevertheless (b) does not seem to have much popularity even among those who
may hold (iii) or (v). Indeed, the received view in linguistics and psycholinguis-
Language Use 221
tics seems to be the version (a) view that processing operates on metalinguistic
representations. Thus Fodor, Bever, and Garret describe language ‘‘perception as
the matching of stimuli to descriptions’’ (1974: xvi). Fodor claims that ‘‘a repre-
sentation of a message must have, among, its inputs, a representation of the
grammatical relations exhibited by the sentence to which the message is assigned’’
(1975: 167). And, according to Matthews, ‘‘language understanding, it is widely
assumed, involves the recovery and representation for later use of the syntactic
structural descriptions associated with the uttered sentence (or phrase)’’ (2003:
198). In this section, arguing for my fourth tentative proposal, I shall present
some reasons for doubting this received view and for favoring a brute–causal
alternative, particularly version (b) of position (v). The point is not, of course, to
oVer the alternative as a complete explanation of language use. Like the received
view, it is far far too lacking in details for that. The point is rather to suggest that
the best explanation is more likely to comply with the brute–causal alternative
than the received view. And the considerations favoring the alternative are, it goes
without saying, far from decisive. The fourth proposal really is tentative.
I mentioned earlier (3.1) that we cannot look into the mind and simply ‘‘see’’ if
there are representations of this or that. We don’t even know enough about what
to look for. So we should only posit such representations if we can Wnd some
serious causal work that they have to do: Pylyshyn’s Razor. Sadly, it is often
diYcult to tell when we have found causal work for representations: the issue is
simply not that clear. Still, I am inclined to doubt that there is causal work for
representations of syntactic and semantic properties in the speedy automatic part
of language processing.
We have just noted that the psychological literature on skills, particularly that
on motor skills, dynamic systems theories, connectionist theories, instance the-
ories, implicit learning, and the implementation of skills, counts against the view
that the rules that govern language use are represented. I think this literature
should also make us doubt that language use involves representing syntactic and
semantic properties; it makes such a view of language use seem too intellectualist.
Those doubts are particularly encouraged by the sheer speed of language process-
ing. This is not to say that the literature on skills is hostile to talk of represen-
tation in general. There is indeed plenty of such talk. It is not obvious, of course,
whether we should take this talk to be about representations in the sense that
concerns us here, as explained in section 1.1. As I noted there, ‘represent’ and its
cognates are often used fairly indiscriminately in psychology.23 Aside from the
23 One wonders, for example, about the uses in Sun et al. 2001 (p. 208); Reber 2003 (p. 489);
Cleeremans 2003 (p. 493); and Taatgen 2003 (p. 822). Gallistel remarks: ‘‘Few other concepts generate
such heated discussion among psychologists as the concept of representation’’ (1990: 15). He is careful
in explaining his own concept: ‘‘I use the term representation in its mathematical sense. The brain is said
to represent an aspect of the environment when there is a functioning isomorphism between an aspect
of the environment and a brain process that adapts the animal’s behavior to it’’ (p. 3). But isomorphism
alone will not do the job for him: it misses that each point in the cognitive map represents a particular
aspect of the environment; and that the map represents the environment but not vice versa.
explicit talk of representations, we noted that even the literature on motor skills
includes a deal of mentalistic talk that seems to demand representations. Perhaps
we should not take this talk too seriously. Much of the literature suggests that
skills have fairly brute–causal associationist natures and so does not encourage the
idea that language use involves representations of the syntactic and semantic
properties of language.
It is worth noting that the positions of Bresnan and Kaplan and Berwich and
Weinberg discussed earlier (4.4–4.5) have a place for a fairly brute–causal process
in language parsing. Thus, Bresnan and Kaplan, in arguing for their lexical-
function grammar, allow that the derivation of the passive lexical entry from the
active can be stored in memory so that it does not have to be actually run in
language processing (1982: xxxiii–xxxiv). And Berwick and Weinberg, in
defending transformational grammars, allow for a ‘‘precomputed memory re-
trieval system’’ (1984: 74) so that transformations do not have to be run in
processing.
I now want to present a relatively ‘‘a priori’’ reason for favoring a brute–causal
view of language processing. But I start with a caveat. My concern is with whether
language use involves representation of the syntactic and semantic properties of
the sounds, inscriptions, etc., of the language (which I will often brieXy call ‘‘the
linguistic properties’’). For convenience, let us set aside inscriptions, etc., taking
sounds to be representative of linguistic items. My concern is not with whether
language use involves representation of the phonetic and phonological properties
of the sounds. Clearly, these properties have to be recovered somehow from the
acoustic signal at the beginning of speech perception and bestowed somehow on
the signal by the end of speech production.24 I assume that this must involve, at
least, representations of the phonetic properties, the ‘‘physical’’ properties of the
sound, as a result of a transduction. Beyond that I shall (wisely) have nothing to
say on the vexed question of what else it involves,25 including on the inXuence of
lexical knowledge on the recovery process.26 I suppose, although I shall not argue,
24 ‘‘Speech perception processes . . . require a transformation of the auditory input from the
peripheral auditory system to a spectral representation based on more generalized auditory patterns
or properties, followed by the conversion of this spectral representation to a more abstract feature
(phonological) representation, and ultimately the mapping of this sound structure onto its lexical
representation’’ (Blumstein 1999: 644).
25 Burton-Roberts et al. 2000 paint a nice picture of just how vexed phonological theory is:
‘‘There seems to be almost no conceptual assumption in the foundations of phonological theory
that is not controversial.’’ There is conXict ‘‘even about what kind of entity phonological entities are’’
(p. 1), about whether they are ‘‘grounded in phonetic substance’’ or ‘‘mental objects’’ (pp. 8–9).
Even on the assumption that the entities are so grounded, ‘‘there is scope for disagreement as to what
phonetic content consists in . . . , the exact nature of the relation between the phonological and the
phonetic, and how we are to draw the boundary between them’’ (p. 9). Some of these problems arise,
in my view, from trying to accommodate phonology to the Chomskian internalist view of language;
see note 23 in section 2.5.
26 ‘‘phonetic perception is itself inXuenced by input from higher-order linguistic levels, most
notably information from the lexicon’’ (Miller 1999: 788).
Language Use 223
that my reasons for doubting that syntactic and semantic properties are repre-
sented would carry over to phonological properties.
In considering the brute–causal alternative, I shall assume the Language-of-
Thought Hypothesis (LOTH) and hence (t): the structure rules of a speaker’s
language are similar to the structure rules of her thought. Some of the case for the
alternative may survive if we assume only RTM, the minimal theory of thoughts
that allows that thoughts might be map-like. But the alternative certainly seems
most plausible assuming LOTH because that assumption lessens what has to be
done in language processing (10.3). So the case for the fourth tentative proposal,
like that for the second (10.3), rests a bit on LOTH.
What, given LOTH, does the brute–causal alternative add to the background
assumption about linguistic competence (11.3)? The alternative yields a view of
competence that is like the earlier theory of intuitions (Ch. 7) in its modesty. It
does not posit any representations of linguistic properties. So, competence in a
language does not involve the speaker in propositional knowledge about the
syntax and semantics of the language. Linguistic competence consists in a
conceptual competence, grounded in the external world, together with a set of
translation skills that are mere dispositions (cf. Chomsky 1980b: 5), although not
of course behaviorist dispositions.
This modest view stands opposed to the received view that language processing
involves representing the syntactic and semantic properties of linguistic inputs
and outputs. Thus, on the received view, in understanding
(1) John hit the boy
the processor deploys a whole lot of syntactic vocabulary—like ‘‘NP’’ and
‘‘DET’’—to arrive at a mental metalinguistic statement about the input along
the lines of the structural description:
(2) [S[NP[N John]][VP[V hit][NP[DETthe][Nboy]]]].
Translated into ordinary English, this statement amounts to something like:
(3) John hit the boy is a sentence made up of a noun phrase consisting of the noun John and
of a verb phrase consisting of the verb hit followed by a noun phrase made up of the
determiner the followed by the noun boy.
And, in understanding (1), the processor deploys some semantic vocabulary to

arrive at statements like
(4) hit means HIT
where ‘hit’ refers to a word and ‘HIT ’ refers to a meaning.
Now this received view cannot be the whole story of understanding. For what
we have to end up with in ‘‘the message-in box’’ is not a mental metalinguistic
statement about the linguistic input, however detailed, but a mental statement
that means the same as the input. That is, on hearing (1), what we end up with is a
translation of (1), something that means JOHN HIT THE BOY. That is what
understanding the sentence amounts to. Understanding does not amount to
metalinguistic analyses like (3) and (4), which tell us about the syntactic and
semantic properties of (1). The task is not to generate a theory of the linguistic
input but to generate a synonymous mental representation. Of course, generating
(3) and (4) might be a way to the synonymous representation. But in that case we
need a further story about the Wnal transition from the metalinguistic (3) and (4)
to the synonymous representation. How do we get from a theory of the input
sentence to a mental representation that is synonymous with the sentence?
Presumably, the Wnal transition would have to be some brute–causal process:
once a person has this full analysis of the input, she just goes straight to the
synonymous representation. And, presumably, the initial steps in forming the
analysis have to be brute–causal too: thus, hearing the sound of hit and forming a
representation of its phonetic properties simply prompts the analysis (4) (or
perhaps a representation of phonological properties that leads to (4) ). But this
raises the idea: perhaps we never ascend into a mental metalanguage to make this
analysis but rather go brute–causal from the beginning so there does not have to be
any such transitions. The language processor assigns syntactic and semantic
properties not in the sense that it describes the linguistic input using terms for
those properties but in that it transforms the input into mental expressions having
those properties. Analogously, a certain key stroke is processed by a calculator as a
‘4’ without being described as a ‘4’: that stroke simply and brute–causally yields a
representation of 4. So the idea is that the processor receives a phonetic represen-
tation of (1) as input, a representation of a sound that represents John hitting the
boy, and transforms it through however many stages, governed by some hard-
wired rules, into a mental representation of John hitting the boy. It doesn’t ‘‘think
about’’ the input provided by hit, describing it as a V and as something that means
HIT, it simply responds to it with something that is a V and means HIT. There are
no representations of the linguistic properties of the sentence. There are simply
representations of what it represents, John hitting the boy.
Suppose that the received view is true of some processor. So, the processor
ascends into a metalanguage, goes through several stages of theorizing about the
input before Wnally descending into a mental representation that is synonymous
with the input. Then presumably another processor could go through ‘‘the same
stages’’ without the ascent and descent. We should wonder then if there is any
psycholinguistic evidence favoring the received view over this brute–causal idea.
If not, Pylyshyn’s Razor favors the brute–causal idea.
Consider analysis (4). The brute–causal idea makes us wonder whether any
explanatory work is done by supposing that this analysis takes place. What work
does it do to suppose that the processor represents this semantic property of hit?
It seems more plausible that in processing that word we simply come up with a
representation that has the property of meaning HIT. And if that is what we do,
presumably that representation will also have the property of being a verb. So we
Language Use 225
have no need to come up with a representation of its being a verb, as in (3). The
simpler story is surely that hearing the word hit does not prompt a person to
come up with a mental representation about the word but rather with a repre-
sentation, <hit>, that is a verb and means HIT. The natural and eYcient design
for translating hit into <hit> surely accords with this simple story.
This point about language comprehension becomes more vivid if we consider
the analogous point about language production. If we are to suppose that com-
prehension starts with analysis of the properties of linguistic inputs, shouldn’t we
also suppose that production starts with the analysis of the properties of mental
inputs? Consider someone saying (1). The person had a thought, <John hit the
boy>, and expressed it by producing (1). The analogous Wrst step to (4) would be
<hit> means HIT.
But the supposition that we go through this Wrst step is almost as outlandish as
the earlier one that language production starts with an application of a metalin-
guistic theory of Mentalese (11.4, point 2). For one thing, what language would
this representation be in? Presumably it could not be in the language of <hit>
itself, Mentalese. It would have to be in some meta-Mentalese. If we can get from
the mental <hit> to the linguistic hit in production without analyzing <hit>,
we wonder why we cannot get from the linguistic hit to the mental <hit> in
comprehension without analyzing hit?
Objection: The way a word should be processed depends on its syntactic
category, on whether it is a verb, noun, adjective, or whatever. What the processor
must have then are rules for each syntactic category. It surely does not have a rule
for each word of that category. So, it has rules for processing all adjectives, not a rule
for each adjective. And such a general rule must operate on a representation of a
word as a member of that category, on a representation like <That is an
adjective>. Response: Why must the general rule operate on a representation?
To represent a word as an adjective, the processor must Wrst recognize that it is
an adjective. Whatever clue enables this recognition could simply trigger rules
appropriate for adjectives: the processor responds to it as an adjective without
representing it as one. Being an adjective is not a local physical property of a word,
as Rey likes to emphasize (1997: 128; 2003b: 178–9), and so recognizing a word as
having that property may not always be an easy matter (although it mostly is; see
10.6). But it is not made any easier by supposing that it involves representing the
word as an adjective rather than simply responding to it as an adjective.
Let me Wll out the brute–causal view a bit more, starting with words. The
view is that there is a simple association stored in memory between a linguistic
word—more accurately, a representation of its phonetic properties—and the
corresponding mental word with the result that the one leads straight to the other
without any analysis. This association is established by the regular use in the
community of the linguistic word to express the mental word, a regularity that
also establishes the conventional meaning of that linguistic word (9.5). The same
story also applies to familiar sentences—for example, ‘What’s the time?’ and
‘How are you?’—and familiar expressions, including idioms—for example,
‘kick the bucket’, ‘butterXies in the stomach’.27 The view is not, of course, that
these simple associations are all there is to the use of these expressions. Thus,
pragmatic abilities will be called on in language comprehension to determine the
reference of ‘you’, to remove ambiguities, and so on (see the clariWcation that
began this section). But the view is that simple associations are at the core of the
processes. This story for words and familiar expressions seems very plausible.
And there seems to be no reason why it cannot be extended to syntax. The basic
idea again is of simple associations stored in memory, in this case of syntactic
structures in thought with similar (implicit and explicit) syntactic structures in
language so the presence of the one prompts the other. Each of the structural
features of any linguistic sentence—the features captured by structural descrip-
tions like (2) or by phrase-structure trees—is associated with a similar structural
feature of mental sentences. This association is established by the regular use in
the community of that linguistic structure to express the mental structure, a
regularity that also establishes the conventional syntax of that linguistic structure
(9.5). In comprehension, a person must identify the structure from clues pro-
vided by the syntactic category of words (which comes with word recognition),
word order, and the like.28
I have frequently remarked that the structure rules of a language seem the
wrong sort to govern processing (2.6, 3.2, 4.1, 11.4). This is certainly so on the
brute–causal picture I am presenting. The translation task requires that a mental
syntactic structure be associated with a linguistic one. The rules governing such
associations—along the lines of structure x causing structure y—are nothing like
linguistic structure rules. This supports my third tentative proposal and counts
against positions (i) and (iii). There is more support in the next two sections.
This brute–causal associationist picture of language processing risks the fury of
Fodor. For, Fodor has a very dim view of associationism: ‘‘associationism . . . is,
and always has been, an intellectual disaster’’ (2001: 104; see also 1975: 173).
Its return makes him feel old and depressed (2001: 100). But it is important to
note that, although I am suggesting that there may be a place for associationist
processes in language use, I am obviously not embracing associationism in
general.29 For, inXuenced by Fodor himself, I favor LOTH and with it a highly
27 ‘‘Idioms must be stored in the mental lexicon’’ for language production. They ‘‘do not require
simultaneous selection of several lexemes.’’ (Wheeldon et al. 2003: 761)
28 Kintsch 1984 gives some examples of potential clues (113–15).
29 Gallistel (1990) makes telling criticisms of associationist models of classical conditioning. He
oVers a rival representational model which ‘‘assumes that the animal has a record of what occurred
when, that it can segment this record into the temporal intervals over which a given CS was present,
that it can sum these intervals, that it can count the number of occurrences of the US in each such
interval, and that it can sum the number of occurrences of the US during diVerent intervals of CS
presence to obtain the total number of USs observed in the presence of a given CSs’’ (p. 424). He
produces some impressive evidence for these assumptions.
Language Use 227
rational–causal view of thinking. So the suggestion is that we need a hybrid

architecture for the mind, a rational–causal part for the higher functions like
thinking and a brute–causal associationist one for lower functions like language
processing. There is, of course, nothing novel about such a proposal:
I will argue for the plausibility of a particular kind of hybrid architecture, whose abstract
levels are classical . . . I assume that connectionist networks, or close variants thereof,
provide adequate models of moderately low level cognitive processes, which may include
pattern recognition and some motor control. (Hadley 1999: 200; see also Johnson et al.
2003; Sun 2003; and Cleeremans 2003)
Despite his contempt for associationism, even Fodor allows a place for
associationist connections in his account of word recognition: ‘‘these connections
have a real, if modest, role to play in the facilitation of the perceptual analysis of
speech’’ (1983: 82). But one wonders why the role has to be so modest. Why
could it not extend to syntax recognition? In an earlier work, he follows the claim
that ‘‘English speakers can infer messages from representations of grammatical
relations’’ with the following: ‘‘But though they presumably can, they presum-
ably don’t. What apparently happens is that grammatical relations are computed
only when all else fails. There exist heuristic procedures for sentence recognition
which, in eVect, ignore grammatical relations and infer messages directly
from lexical content, accepting, thereby, the penalties of fallibility’’ (1975:
167–8). One wonders why these heuristic procedures could not be associative
and why they could not be all there is to the speedy automatic process of language
processing.
I emphasize that in entertaining the brute–causal associationist view of lan-
guage use I am not entertaining the return of empiricism as a general theory of
mental processes. I would not contemplate for a moment that thinking, for
example, was an associationist process (Part IV). Still some processes surely are
associationist, as even Fodor allows. The issue is whether the processes of lan-
guage use are among them. Can associationism be extended that far?
We are not concerned here with what seem intuitively to be performance
failures. Thus, there will be sentences of the language that a person cannot
translate satisfactorily into thought in the brute–causal way. And there will be
sentences she cannot produce because she cannot think the corresponding
thought. (But will there be any thoughts that she can think but cannot translate
into sentences of the language?) Presumably the central processor helps deal with
the sentences that she cannot immediately translate. If the ‘‘message-in box’’
receives an unacceptable message, then the central processor orders a rerun: using
memory of the starting point, another attempt at translation is made. If even this
fails to yield a satisfactory message in comprehension, then the central processor
may have to take a serious hand. Perhaps this will involve deploying such de-
clarative linguistic knowledge as the person has acquired at school and elsewhere
to try to analyze the input. But it may mostly consist in trying several times to
get ‘‘a feel’’ for the troublesome sentence by comparing it to others until the
syntactic structure is recognized by the associative memory. And this may fail:
there may be sentences of the language that a speaker simply cannot translate into
Mentalese or, at least, cannot without a great deal of help from a teacher. Thus, I
haven’t yet managed to parse Fodor’s example, ‘Bulldogs bulldogs bulldogs Wght
Wght Wght’ as a sentence despite the clue ‘‘Take the Wrst two verbs as transitive’’
(1975: 168).
Despite these limitations in what a person can produce and understand, there
is a sense, of course, in which a person is competent with any of the inWnite
number of sentences in her language; her competence is productive. In what
sense? To answer, we must draw on our earlier discussion of competence (11.3).
We note, Wrst, that her capacity to think thoughts, her conceptual competence, is
productive: if it weren’t for resource limitations in her memory, time, computa-
tional power, and so on, she could think any of an inWnite number of thoughts.
LOTH has a ready explanation of this: the structure rules governing her Menta-
lese allow, by multiple embeddings, connectives, and so on, the generation of an
inWnite number of thoughts; there are a Wnite number of rules for combining a
Wnite number of mental words—concepts—into mental sentences but no limit
on the number of rule applications that might govern a sentence. And any
successful theory of thought must pass the test of explaining the productivity
of thought.30 Turn next to her language. It is productive, of course, in just the
same way as her thought is according to LOTH: its structure rules can yield an
inWnite number of sentences. The productivity of her linguistic competence is
then easily explained: each of the structure-determining rules for thought is asso-
ciated with a structure-determining rule for language. If it weren’t for resource
limitations she could think any of the inWnite number of thoughts and exploit
those associations to express them, and she could exploit those resources to
understand any of the inWnite number of sentences of the language. Given the
productivity of thought and language the brute–causal story of language processing
is quite compatible with the productivity of linguistic competence.
Finally, Pylyshyn may have a problem with the brute–causal story. At the
beginning of this chapter I quoted his doubts about RT. He is, nonetheless,
somewhat tentatively committed to the view that the syntactic properties of
expressions are represented in parsing:
Although there may be some doubt as to whether grammatical rules are explicitly
encoded, there appears to be good evidence that both the output of the analysis (i.e.
LF) and certain intermediate steps in the parsing are so encoded. These have to do with
the fact that certain universal properties of the language faculty appear to be stateable only
in terms of certain properties of the parse tree. For example, the various constraints on
movement are stated in relation to certain properties of the analysis of the sentence, and
thus imply that such an analysis is actually available to the system in the course of parsing
30 This is a test that connectionism seems to fail as Fodor and Pylyshyn (1988) emphasize.
Language Use 229
and/or generation. Attempts to design parsing systems have also suggested that not only
the logical form itself, but also various intermediate stages of the grammatical analysis
may be explicitly encoded. In other words it is likely that parts of the analyzed structure of
the sentence appears as a symbolic code, although the rules themselves may not appear in
such a form. In computer jargon, although the rules may simply be compiled into the
functional architecture of the system and not accessed in interpreted mode, the data
structures to which these rules apply are explicitly represented and their form is empir-
ically signiWcant. (Pylyshyn 1991: 247–8)
But one wonders why the fact that movement constraints and the like have to be
stated in relation to syntactic properties requires an encoded analysis of the
sentences. And one wonders whether the parsing systems designed so far can
plausibly be seen as throwing much light on how humans actually parse.
This completes our preliminary assessment of positions on language use, an
assessment made prior to considering the psycholinguistic evidence from lan-
guage use itself. In this section, I have argued for my fourth tentative proposal:
the speedy automatic language processes arising wholly or, at least, partly from
linguistic competence are fairly brute–causal associationist processes that do not
operate on metalinguistic representations of the syntactic and semantic proper-
ties of linguistic expressions. So the persuasive future abduction on language
processing is likely to support version (b) not (a). I think that this view is
plausible although the case for it is very far from decisive, of course. If the view
is indeed plausible, this adds to the already powerful case, presented in the
previous two sections, against positions (i), (ii), and (iv). For, if processing
involved representations of the rules, as those positions claim, then it would have
to operate on representations of the syntactic and semantic properties adverted to
in those rules rather than brute–causally. I conclude that the likelihood of a future
abduction supporting any of these three positions is remote. I have urged also
that the structure rules of a language are the wrong sort of rule to govern what is,
given RTM, the translation process of language use; my third tentative proposal.
This counts against (iii). So I think that we can predict with a fair amount of
conWdence that the explanation of language use will support (v): the processing
rules are unrepresented and largely unlike the structure rules of the language. And
I think that we can predict with some small conWdence that language processing
will not operate on metalinguistic properties of the linguistic expressions but will
be more brute–causal.
What about the-only-theory-in-town abduction (11.2)? It should be dis-
missed. It fails every criterion of a good abduction. The discussion of the last
three sections shows that its appeal to RT to explain language use is very im-
plausible given what we already know. So it fails (A1). We noted at the beginning
that the explanation fails (A2) because of a serious lack of details. Finally, it
fails (B). The discussion of skills in the last section makes it seem likely that the
future will yield a good alternative explanation that does not involve the repre-
sentation of rules. The discussion in this section suggests that this alternative
may be brute–causal. Such alternatives are much more promising than explan-
ations appealing to RT.
We turn now to an inevitably brief consideration of the psycholinguistic
evidence on language use. I think that this evidence should not persuade us to
modify these preliminary assessments.
11.7 P S YC H O LO G Y O F L A N G U AG E P RO D U C T I O N
We shall see that a good deal of day-by-day work on language processing does
not address the architecture questions that most concern us: whether this
processing involves the rules of the language, either represented or otherwise
embodied. There seems to be little interest in this issue that was once so pressing.
Psycholinguists mostly now approach the study of language processing as if it
were, except for the Respect Constraint (2.4–2.6), independent of grammars.
There is not even much focus on whether processing involves representations of
syntactic and semantic properties. However, the favored models of this process-
ing mostly seem to support a fairly brute–causal view of the processing; they seem
to support the fourth tentative proposal.
I shall discuss language production before language comprehension for two
reasons. First, our focus is on processes arising from linguistic competence. All
language use involves processes, including central processes, arising wholly or
partly from pragmatic abilities, but such processes are more prominent with
language comprehension. For, language comprehension requires the assignment
of referents to indexicals and the removal of ambiguities. Language production
does not face these problem and so the role of factors other than competence is
not so important: ‘‘The direct apprehension of the message sets speakers apart
from their listeners, for whom ambiguity is rife’’ (Bock 1999: 456). Second, I
think that work on language production speaks more clearly to the architectural
issues that concern us than does the work on language comprehension: it not
only allows our preliminary assessments to stand, it strengthens them.
I shall begin by describing, in broad terms, some fairly uncontroversial facts
about language production. I shall then go into some details. Finally, I will
describe the main areas of controversy.
Language production consists in processes ‘‘from mind to mouth’’. These
processes start with intentions to communicate messages, intentions to express
thoughts. The processes go from these intentions to the articulation of linguistic
expressions: ‘‘the cognitive processes that convert nonverbal communicative
intentions into verbal actions. These processes must translate perceptions or
thoughts into sounds, using the patterns and elements of a code that constitutes
the grammar of a language.’’ There are two steps in language production, a
grammatical one (producing a ‘‘lemma’’) and a phonological one (producing
a ‘‘lexeme’’). ‘‘Grammatical encoding refers to the cognitive mechanisms for
Language Use 231
retrieving, ordering, and adjusting words for their grammatical environments,

and phonological encoding refers to the mechanisms for retrieving, ordering,
and adjusting sounds for their phonological environments’’ (Bock 1999: 453).
At the grammatical stage words with the appropriate meanings have to be
retrieved and put into a syntactically appropriate frame. It is ‘‘generally assumed
that syntactic properties of words are retrieved as a consequence’’ of word
retrieval. The retrieval of a word is ‘‘competitive’’ and occasionally leads to errors
because ‘‘other words similar in meaning are also activated’’ (Vigliocco and
Vinson 2003: 184). Similarly, the phonological stage may produce a word simi-
lar in sound to the desired one. ‘‘When words exchange, they exchange almost
exclusively with other words from the same syntactic class. . . . When sounds
exchange, they exchange almost exclusively with other sounds from the same
phonological class’’ (Bock 1999: 455). Two sorts of evidence have been adduced
in theorizing about language production: spontaneously occurring errors, slips of
the tongue and dysXuencies (‘um’s, ‘ah’s, repetitions, etc.); and the results of
laboratory experimentation.
Let us now consider some more details of the Wrst step, grammatical encoding:
Assuming that the speaker wants to express the event ‘The ferocious dog bites the boy’,
the stored meaning-based representations for ‘ferocious’, ‘dog’, ‘to bite’, and ‘boy’ are
retrieved. These representations would further specify that ‘dog’ and ‘boy’ are nouns
(hence usable as subject and object), ‘to bite’ is a verb (hence specifying a relation
between the nouns), and ‘ferocious’ is an adjective (hence it can modify a noun). These
representations also specify that ‘dog’ and ‘boy’ are common, countable nouns, and may
be pluralized, and that ‘to bite’ may be used to refer to the past, present, or future,
depending on the speaker’s intentions. The message-level information controls the
assignment of the diVerent lexical elements to their intended grammatical functions
(e.g. ‘dog’ is assigned to be the subject, and ‘boy’ as the object and not vice versa,
which would result in the unintended sentence: ‘The boy bites the ferocious dog’).
Message-level information also ensures that the appropriate speciWcation of number
for the noun ‘dog’ (singular) and tense for the verb ‘to bite’ (present) are passed
along. (Vigliocco and Vinson 2003: 185)
This account exempliWes just what we would expect: the message, what the
speaker intends to convey, controls the process (4.1). Thus, the speaker’s thought
involving <dog> meaning DOG leads to the retrieval of the word ‘dog’ with
that same meaning. The message also determines that the word appears as subject
and not object and comes in the singular. What goes on in this retrieval process?
There is no suggestion that the process is governed by represented rules. Does it
even involve representations of syntactic and semantic properties? Researchers
talk of the process retrieving ‘‘stored information’’ (p. 184; see also Levelt et al.
1999: 2), but this storage may be in the form of implicit memory, about which
much remains to be learnt (11.5). There seems to be nothing here that requires
the representation of syntactic or semantic properties of ‘dog’. Indeed, what
would be the use of retrieving the information that ‘dog’ means DOG ? The
account is compatible with the following brute–causal picture: ‘dog’ is selected

because it is associated with <dog>; its subject position is associated with the
place of <dog> in the message; its singularity is associated with the singularity of
<dog>. What about the syntax of the sentence? The retrieval of ‘dog’ is the
retrieval of something that is a noun and hence something that must be Wtted into
the sentence in the way appropriate for a noun. But this does not require a
representation that it is a noun. Of course, more has to be done to Wx the syntax of
the sentence; for example, the singularity of the subject demands the singularity
of the verb. But researchers see this sort of grammatical constraint as ‘‘procedural
knowledge concerning how words can be used in sentences’’ (Vigliocco and
Vinson 2003: 185) and so there is no clear need for these constraints to be
represented.
The dominant role of connectionist models in psycholinguistic research into
language production supports a brute–causal picture. Thus, in a survey article,
Joseph Stemberger (2003) discusses three sorts of model. Two are overtly con-
nectionist, ‘‘Network Models: Local Models’’ and ‘‘Network Models: Distributed
Models’’, and he has this to say about third, ‘‘Symbolic Models’’: they have not
been accompanied by ‘‘detailed proposals about processing’’ but ‘‘the spirit of
their proposals has been instantiated in network models’’ (p. 158). Consider also
the talk of ‘‘activation’’ in the work of some leading researchers. Gary Dell
explains production errors as follows: ‘‘semantically and phonologically similar
words become activated in the process of retrieving a target word; thus, these have
some chance of erroneously replacing the target’’; ‘‘the phonological eVects on
word slips reXect the spreading of activation from a target word to the target
word’s phonemes, and from there to other nontarget words that share those
sounds.’’ ‘‘But why should we want to propose that activation moves from sounds
to words during language production?. . . . It may be the case that the lexical
network allows activation to spread in both directions because it is used for both
production and comprehension’’ (1995: 195–6). William Levelt, Ardi Roelofs,
and Antje Meyer explain lexical selection in the following way. ‘‘An active lexical
concept spreads some of its activation to ‘its’ lemma node, and lemma selection is
a statistical mechanism, which favors the selection of the highest activated
lemma’’ (1999: 4). Their ‘‘theory is modeled in terms of an essentially feed-
forward activation-spreading network’’ (p. 6). Dell, Franklin Chang, and Zenzi
GriYn conclude that ‘‘the PDP approach oVers the best chance to explain
production as a skill. . . . Perhaps most importantly, a PDP approach to
language production expresses its commonalities with other linguistic, and
even nonlinguistic, skills’’ (1999: 539–40).
Finally, consider where there is controversy in the theory of language produc-
tion. Bock mentions three controversial areas. (a) There is a dispute over whether
‘‘regularly inXected words are stored and retrieved from memory in the same way
as uninXected words . . . or require a separable set of speciWcally inXectional
operations’’ (1999: 455; see also Vigliocco and Vinson 2003: 184). (b) Theories
Language Use 233
diverge over ‘‘the relationship between phonological encoding and the higher
level processes of lexical selection and retrieval. The discrete-stage view [Levelt et al.
1999] argues that each step is completed before the next is begun’’. So there is no
feedback, just feedforward. ‘‘In contrast, interactive views (Dell et al. 1997)
embrace the possibilities of partial information from one stage aVecting the
next (cascaded processing) and of information from lower levels aVecting higher
levels of processing’’. There is feedback as well feedforward (Bock 1999: 455;
see also Vigliocco and Vinson 2003: 185). (c) Finally, there is controversy over
whether language production and comprehension involve distinct cognitive
systems. The fact that they seem to draw on the same linguistic knowledge
suggests that they are not distinct although, as we noted (10.2–10.3), there
may be little overlap beyond the sharing of a conceptual competence. We have
just seen that some further overlap is supported by the interactive view of Dell
according to which activation moves from sounds to words even in production
just as it must do in comprehension.31
It is noticeable that these areas of controversy do not concern RT at all, nor do
they even talk of representations. They do not seem to bear on any of the
positions on language use that we have described.
In sum, the psycholinguistic study of language production that I have sum-
marized does not undermine our preliminary assessment that there is no place for
RT in the explanation of language use. Indeed, the study gives further support to
that assessment. The study, particularly the prominence of connectionist models,
also supports our preliminary assessment in favor of the brute–causal view,
version (b), over the view that production involves representations of syntactic
and semantic properties, version (a); it supports my fourth tentative proposal.
Still, it is clearly far too early to be conWdent about either view. Finally, the rules
governing the retrieval processes of language production look nothing like
the language’s structure rules. This supports my third tentative proposal that
the structure rules are the wrong sort to govern processing, a proposal that
stands opposed to position (iii). The case for the third proposal concludes in
the next section.
Our focus here has been on processes arising from linguistic competence.
Still we should acknowledge the role of other abilities in language production.
Deciding how to convey a message is a surprisingly complicated pragmatic pro-
cess: ‘‘the tailoring requirements are legion’’ (Bock 1999: 456). In tailoring an
utterance, ‘‘production accommodates its processing to the needs of comprehen-
sion’’ (Ferreira 2003: 747). Thus, we have to choose what language to speak,
whether to whisper or shout, whether to speak fast or slow, which referential
expression to use for an object in mind, and so on. We take into account our
relationship with the listeners, what we think they are capable of understanding,
31 Levelt et al. (1999: 7) remain largely neutral on the issue but their peer commentators, Anne
Cutler and Dennis Norris (1999) urge a common architecture for production and comprehension.
what we think they will Wnd interesting, and what we suppose to be mutually
known. We follow the Gricean cooperative principle. Some of these tailoring
processes seem to be brute–causal—some ‘‘speciWc modiWcations occur uninten-
tionally and automatically’’ (p. 749)—but some are surely central and rational.
This does not demand any revision in the view that the speedy automatic pro-
cesses of language production, whether arising from linguistic competence alone
or partly from pragmatic abilities, are brute–causal, involving no metalinguistic
representations.
This concludes our discussion of language production. We turn now to the
other half of language use, language comprehension.
11.8 P S YC H O L O G Y O F L A N G U AG E C O M P R E H E N S I O N
Work on language comprehension, like that on language production is not

mostly addressing the architecture problems that concern us. There seems little
interest in Wnding the rules of the language somehow embodied in the mind.
Although the work on comprehension does not undermine our preliminary
assessments, we shall see that it does not give them quite the support provided
by the work on production.
We have just acknowledged the role in language production of abilities other
than linguistic competence. There are very many of these ‘‘pragmatic’’ abilities
involved in language comprehension and it will be helpful to start our discussion
by mentioning some of them. The ultimate task in language comprehension is to
grasp the thought that the speaker means to convey by a sentence. Part of this task
is (typically) understanding the literal meaning of the sentence uttered. Parts
of this are obviously not fulWlled simply by exercising lexical and syntactic
competence. Thus, a person’s lexical competence explains why she processes a
sound /he/ as a pronoun but it does not explain why she assigns it one referent
rather than another. Her lexical competence will explain why she might assign
one of the two meanings of an ambiguous word like ‘bank’ but it will not explain
which one she assigns; similarly, her syntactic competence, and an ambiguous
structure like ‘Visiting relatives can be boring’. Processing indexicals and am-
biguous expressions requires abilities other than those that constitute mastery of
the language. And much else that goes on in comprehension does too: ‘‘sentence
processing necessarily involves the eYcient use of many diVerent types of infor-
mation’’ (Gorrell 1999: 749). It is common to call the other abilities, ‘‘world
knowledge’’, in distinction from the ‘‘linguistic knowledge’’ of lexical and syn-
tactic competence (Gernsbacher and Kaschak 2003: 723). Some of this world
knowledge is surely to be found in the central processor.
The world knowledge that plays a role in comprehension needs to be
construed broadly to include not only general knowledge about the world but
also knowledge of the discourse. As an example of the inXuence of discourse,
Language Use 235
consider how we often Wx the referent of a pronoun: we take it as anaphoric on
some noun phrase that was prominent in prior discourse. As an example of the
inXuence of general knowledge, consider how we would understand, ‘‘Fiona
poured wine into Alex’s favorite glasses’’. Our knowledge that drinking glasses
rather than eye glasses are appropriate receptacles for wine will determine our
choice of interpretation. Part of our worldly knowledge is our knowledge of other
minds. This plays an obvious role in our interpretation of deictic pronouns.
We use our understanding of a speaker who says, ‘‘He is crazy’’, to determine the
referent of ‘he’: given what she is looking at, gesturing at, or what is otherwise
salient, given what we know about her, we Wgure out who she is likely to be
referring to. We go through a similar process to interpret ambiguous names.32 In
general, in interpreting a speaker’s utterance we take account of what we know of
the speaker’s interests and knowledge, what we think she would think interesting
and appropriate to say, and so on. The Gricean cooperative principle plays a role
in comprehension as in production.33 Finally, we should note that we tend to
favor the most frequent meaning in interpreting an ambiguous word or structure.
Although there is general agreement that comprehension involves all this
world knowledge—what I have been picking out with my talk of ‘‘pragmatic
abilities’’—as well as strictly linguistic competence, there is not of course
agreement about how it does. It is clear that linguistic competence alone can
only match linguistic words and structures with concepts and thought structures
that are parts of possible interpretations, given the language. We must look to
pragmatic abilities to complete the interpretation where there are indexicals and
to choose between possible interpretations where there is ambiguity. A key
disagreement is over whether pragmatic abilities play a role in determining choice
of structure from the beginning or whether there are some initial processes
governed solely by syntactic competence. We shall consider this disagreement
in a moment. But Wrst we should continue with some further uncontroversial
views about the role of linguistic competence in comprehension.
Our linguistic competence is deployed in word recognition and parsing: ‘‘the
comprehender must recognize each of the words in the sentence and determine
the syntactic (and semantic) relationship among them, as deWned by the gram-
mar of the language’’ (Tanenhaus 2003: 1142). Parsing assigns a syntactic analy-
sis to the sentence. That together with the assignment of meanings in the process
32 See Pietroski 2003 for some nice examples of the pragmatics of reference determination and
related matters. It is worth noting that worldly knowledge plays a similar role with motor skills.
Thus, given the state of a baseball game and beliefs about the pitcher, a batter expects a fastball and
so ‘‘interprets’’ a pitch as a fastball unless presented with evidence, probably too late, that it is not.
33 This is not to say that we should embrace the standard Gricean account of how people
understand Wgurative language. Edward Wisniewski cites much evidence against this account.
Thus, ‘‘given an appropriate discourse context, people do not take longer to understand Wgurative
utterances’’. And it is not the case that ‘‘people seek nonliteral meanings if and only if the literal
meaning makes no sense in context’’. ‘‘Just exactly how people understand Wgurative language is not
resolved’’ (1998: 52).
of recognizing words yields a semantic interpretation. The sentence is not

presented all at once: in speech, there is a stream of sounds; in reading (English),
the eye focuses on the beginning of the sentence at its left and moves in jumps to
the end of the sentence at its right. How soon does processing begin? ‘‘One thing
that we can now be certain of is that sentence comprehension begins almost
immediately’’ (Pickering 2003: 463). Thus, consider our processing of the
spoken sentence. The student spotted by the proctor was expelled. ‘‘By the middle
of the Wrst vowel in student, potential lexical candidates, such as student, stool,
and stoop are becoming active in memory, along with their associated syntactic
and semantic representations.’’ The choice between these candidates is made
almost immediately: ‘‘convergence on the most likely lexical candidate occurs
shortly after enough phonetic input is received to distinguish the input from
other likely alternatives—often well before the word ends’’ (Tanenhaus 2003:
1143). Consider also the processing of the highly ambiguous word bug. ‘‘Re-
search has shown that multiple meanings of bug are often initially activated by
the language comprehension system. Then, within milliseconds, the appropriate
meaning is selected, and the other meanings are discarded.’’ Often this process
involves the enhancement of some meaning by, for example, prior discourse: ‘‘by
selectively suppressing and enhancing information, comprehenders can select the
appropriate meanings of the words they encounter’’ (Gernsbacher and Kaschak
2003: 724). These examples illustrate the problem of word recognition. An
analogous problem faces the parser. The presentation of a sentence often leads
to temporary structural ambiguities. Thus, the initial fragment, The student
spotted, in the sentence above is ambiguous until the sentence continues. The
continuation makes it clear that the fragment is a relative clause but up to that
point it could have been a main clause, as in The student spotted the proctor and
hid his crib sheet ; spotted could have been ‘‘attached’’ into the structure of the
fragment in more than one way. Several models have been proposed of how
speakers resolve such temporary ambiguities. ‘‘The assumption common to these
models is that syntactic analysis is accomplished by a special computational
device that makes direct use of grammatical information’’ (Tanenhaus and
Trueswell 1995: 228). Finally, ‘‘working memory is important in language
comprehension because words and syntactic structures need to be held in
memory until a sentence has been processed fully’’ (Gernsbacher and Kaschak
2003: 725).
Turn now to disagreements. The one already mentioned has been the main
focus of research: ‘‘researchers in sentence processing have been primarily con-
cerned with how readers and listeners resolve the structural ambiguity that arises
in assigning grammatical relationships to words and phrases as a sentence unfolds’’
(Tanenhaus 2003: 1143). ‘‘But can information such as semantic plausibility
be used to ‘guide’ parsing in cases of ambiguity? This has been the dominant
question in parsing research’’ (Pickering 2003: 463). To think that parsing is
not guided in this way is to favor the modular view of parsing: the parser
Language Use 237
is autonomous. This view has the consequence that some comprehension pro-
cesses arise wholly from syntactic competence. The view is the basis for the
‘‘garden-path model’’, one of ‘‘three broad theoretical frameworks’’ for research
in this area. The other frameworks are ‘‘referential theory’’ and ‘‘constraint-based ap-
proaches’’ (Tanenhaus and Trueswell 1995: 229). I shall say a brief word about
each.
The garden-path model is a serial one with two stages. The initial stage is
guided by two principles. Lyn Frazier and Janet Fodor describe the Wrst of these,
‘‘Minimal Attachment,’’ as follows: ‘‘each lexical item (or other node) is to be
attached into the phrase marker with the fewest possible number of nonterminal
nodes linking it with the nodes which are already present’’ (1978: 320; see also
Frazier and Rayner 1982). In following this rule ‘‘the parser chooses to do
whatever costs it the least eVort’’ (p. 295); it builds the simplest grammatical
structure.34 The other principle is ‘‘Late Closure’’: ‘‘make an attachment into the
phrase that is currently being processed as long as the alternative attachments are
equally simple’’ (Tanenhaus and Trueswell 1995: 230). The second stage of
parsing is one of combination and revision. Only there can information about
the discourse context and general knowledge of the world play a role; the Wrst
stage is governed by linguistic knowledge alone.
The referential theory has a diVerent view. Stephen Crain and Mark Steedman
argue that ‘‘the primary responsibility for the resolution of local syntactic am-
biguities in natural language processing rests not with structural mechanisms’’, as
in the garden-path model, ‘‘but rather with immediate, almost word-by-word
interaction with semantics and reference to the context’’ (1985: 321; see also Ni
et al. 1996). Interpretations are constructed in parallel and the parsing route
is determined by whichever interpretation is the most plausible in the context
(p. 328). So world knowledge and pragmatic factors are involved from the start.
There are three principles for this. ‘‘The Principle of A Priori Plausibility’’ states:
‘‘If a reading is more plausible in terms either of general knowledge about the
world, or of speciWc knowledge about the universe of discourse, then, other
things being equal, it will be favored over one that is not’’ (p. 330).‘‘The Principle
of Referential Success’’ states: ‘‘If there is a reading that succeeds in referring to
an entity already established in the hearer’s mental model of the domain of
discourse, then it is favored over one that does not’’ (p. 331). Finally there is
‘‘A Principle of Parsimony’’ to deal with a situation where ‘‘there are two or more
possible senses neither of which succeeds in referring in the hearer’s model’’
(p. 333).
Constraint-based models are similar to referential ones in some respects. They
allow parallelism and multiple interactive sources of constraint on parsing, includ-
ing general knowledge of the world. They do not require an initial stage guided
simply be syntax. ‘‘Ambiguity resolution is viewed as a constraint-satisfaction
34 The more recent ‘‘dependency locality theory’’ uses a similar principle (Gibson 2003: 1138–40).
process, involving competition among incompatible alternatives (Tanenhaus
and Trueswell 1995: 232). ‘‘Models incorporating constraint-based ideas are in-
creasingly realized within connectionist architectures’’ (Tanenhaus 2003: 1145).
Not surprisingly, connectionists make some bold claims about what they can
accomplish with language. Thus, James McClelland claims that ‘‘a class of con-
nectionist networks known as the simple recurrent net [SRN] . . . could learn to
become sensitive to long-distance dependencies characteristic of sentences with
embedded clauses, suggesting that there may not be a need to posit explicit,
inaccessible rules to account for human knowledge of syntax’’ (1999: 139).
However, Steedman thinks that ‘‘claims to model human language acquisition
using SRNs must be treated with some caution’’ (2003: 767). Anne Cutler has
this to say about the state of play with word recognition: ‘‘The currently most
explicit models are TRACE . . . and SHORTLIST . . . , both implemented as
connectionist networks. . . . They both propose that the incoming signal acti-
vates potential candidate words that actively compete with one another by a
process of interactive activation’’ (1999: 797).
There is no consensus on which of the models for comprehension should
guide us to the future. Martin Pickering sums up the evidence as follows:
‘‘Overall, there is good evidence that people use many relevant sources of
information rapidly, but it is unclear precisely which model of comprehension
is correct’’ (2003: 464). According to Edward Gibson, ‘‘the details of the relative
timing and strengths of the resource and informational constraints are currently
not known’’ (2003: 1140). Steedman has this to say in a judicious comparison of
the strengths and weaknesses of connectionist and symbolic approaches to
language: ‘‘Neurocomputational mechanisms have proved their worth in the
Weld of pattern recognition and classiWcation, where they can extract structure
latent in inputs such as images of faces, handwritten letters, and speech, and
embody that structure in recognizers that would be impossible to specify by hand,
or that are orders of magnitude more eYcient than rule-based mechanisms, even
when these are statistically optimized.’’ But they have ‘‘been much less successful
at demonstrating the kind of recursive productivity that rule-based systems are
good at . . . . However, it has proven very diYcult to build rule-based linguistic
or computational-linguistic systems with coverage on the scale characteristic
of human linguistic and reasoning abilities’’ (2003: 765). (The recursive
productivity of linguistic competence may not be as big a problem for connec-
tionism if I am right in section 11.6 in attributing this productivity to the
productivity of thought.)
What are the implications of each model for our architectural concerns? The
main debate in the literature we have been summarizing is about the interpret-
ative clues that the hearer gets and the order in which they are used. There is
much talk about knowledge and information but little attention to the way in
which this knowledge and information is embodied in a hearer; see, for example,
Tanenhaus and Trueswell 1995 and Pickering 2003. And, we should note, that
Language Use 239
there is much about the choice between interpretations that are possible given
the language, but little about how linguistic competence makes those interpret-
ations available for choice in the Wrst place. Before there can be a resolution of a
lexical or syntactic ambiguity, competence must generate some candidate inter-
pretations. The literature seems to take it for granted that this step, at least, is
brute–causal. Indeed, what else could it be? We can go further down this brute–
causal path if we set aside garden-path models for a moment. Constraint-based
theories, with their links to connectionism, certainly give support to a brute–
causal associationist picture of the contribution of linguistic competence to
speedy automatic language comprehension. And that picture also seems suitable
for the rather similar referential theory, even though the theory is not presented
in those terms. On all these theories, we can see an ambiguous word of an
utterance as activating the several concepts with which it is linguistically associ-
ated, and an ambiguous structure as activating the several conceptual structures
with which it is linguistically associated. At the same time multiple factors arising
from general knowledge and context suppress some associations and enhance
others, all in a fairly brute–causal way.
These considerations push us toward a brute–causal picture and my fourth
tentative proposal, but a feature of the garden-path model may push us away
from this. Frazier and Fodor take their theory to support the view that a hearer
mentally represents the well-formedness conditions of her language:
when making its subsequent decisions, the executive unit of the parser refers to the
geometric arrangement of nodes in the partial phrase marker that it has already con-
structed. It then seems unavoidable that the well-formedness conditions on phrase
markers are stored independently of the executive unit, and are accessed by it as needed.
That is, the range of syntactically legitimate attachments at each point in a sentence must
be determined by a survey of the syntactic rules for the language, rather than being
incorporated into a Wxed ranking of the moves the parser should make at that particular
point . . . (1978: 322n)
So they are proposing a version of position (ii): rules of the language are rep-
resented and used as data.35 But it is unclear why the well-formedness conditions
have to be represented and surveyed. Why could not the rules governing the
‘‘subsequent decisions’’ be embodied but unrepresented rules that respect (in my
technical sense) the well-formedness conditions? It is hard to see what ‘‘pay oV ’’
there is in having the conditions represented. Pylyshyn’s Razor counts against our
supposing that they are (11.4).
Even if these conditions need not be represented, we may wonder whether, on
this theory, the syntactic properties of the input have to be. The ‘‘executive unit’’
35 Note also Fodor’s later commitment to the processor being ‘‘transparent’’ in the following
sense: it ‘‘makes use of information about linguistic structure in the form in which the mental
(‘‘competence’’) grammar provides it, so that statements from the grammar do not have to be
modiWed, translated, or ‘pre-compiled’ before they can be applied to sentences’’ (1989: 177).
has to take account of earlier parsing decisions which have been held in working
memory (Gernsbacher and Kaschak 2003: 723). Could this be a brute–causal
process? The answer to this must depend in part on the nature of working
memory. This is another place where there are many models but it is far too
early to draw conclusions. Thus Barbara Dosher summarizes three of these
models, ‘‘the feature model’’, ‘‘the primacy model’’, and ‘‘the distributed associa-
tive memory model’’, noting that each has successes and failures. Yet each
‘‘assumes a quite diVerent form of representation and makes quite diVerent
assumptions about the nature of memory’’ (Dosher 2003: 575). This seems to
hold also for other models (Lewandowsky and Farrell 2003). Only one these many
models, the ACT-R theory of Anderson discussed earlier (11.5), would be clearly
committed to the representation of the syntactic properties of linguistic inputs.
Still, it may well be that the garden-path model does require such representations.
Where does this brief discussion leave our preliminary assessments? Most
importantly, it does not undermine them. And I think that it gives further
support to our negative view of RT. Clearly constraint-based connectionist ap-
proaches support our assessment that the future will be fairly brute–causal, hence
support our fourth tentative proposal. These approaches look promising but it is
too early in the study of language comprehension to be conWdent that they are
right. That assessment does not get the boost here that it got with language
production. So the fourth proposal must be particularly tentative. Still it remains,
in my view, better supported than the contrary view that the speedy automatic
part of language processing operates on metalinguistic representations of syntac-
tic and semantic properties.
Finally, in support of my third tentative proposal but contrary to position
(iii), we have seen no sign of the structure rules of the language governing the
process of comprehension. Given the translational nature of the task, if RTM is
correct, it would be odd if we had. Whether these rules for comprehension arise
simply out of linguistic competence, like Minimal Attachment, or arise partly
out of pragmatic world knowledge, like The Principle of A Priori Plausibility and
the (implicit) rules governing a connectionist network, they seem nothing like
the structure rules of the language. These rules for comprehension help us select
among possible interpretations. The rules have to cope with the fact that the
sentence is not delivered all at once. There have to be rules about how much to
take account of before reacting and how to adjust in light of the next bit, etc.
A rule for comprehension will not be a grammatical rule although it will, of
course, respect grammatical rules. Consider, for example, the following proposal
for handling empty categories in parsing: ‘‘Do not posit a trace at sentence
position P if the sentence does not contain a suitable antecedent for a trace at P’’
(J. D. Fodor 1989: 160).36 It is not surprising that ‘‘few attempts [at establishing
36 See also the grammar-based rules proposed by Bradley Pritchett (1988) to handle garden-path
phenomena.
Language Use 241
the ‘psychological reality’ of speciWc syntactic frameworks] have been convin-
cing’’ (Tanenhaus and Trueswell 1995: 223). The rules of such frameworks
characterize the structures of linguistic items and so are the wrong sort to govern
the processes of translating expressions with those structures into thoughts.37
This concludes the case for my third tentative hypothesis, a case that has been
developing slowly throughout the book (2.4, 2.6, 3.2, 4.1, 11.4, 11.6, 11.7). The
proposal has to be tentative, of course, because we know so little about language
processing. Still, the proposal seems much more plausible than the alternative
that the structure rules are processing rules. Our brief consideration of language
use has provided much more evidence against this alternative than for it. And the
alternative seems a priori unlikely given the translational task these rules have to
perform.
11.9 CONCLUSION
In earlier chapters I have found no support for the Representational Thesis, the
thesis that the structure rules of the language are represented in the mind (RT).
My Wrst, and main, goal in this chapter has been to argue that language use
provides no persuasive evidence for RT and that RT is implausible. The only-
theory-in-town abduction to establish RT is far too lacking in detail to be
successful: it fails criterion (A2) of a good abduction (11.2). But its failures are
much worse than that. It fails criterion (A1) because it is not plausible given what
we already know and it fails criterion (B) in not being better than actual or likely
alternatives. Indeed, I have argued that not only is it not now part of a good
abduction it is unlikely to be so in the future. The preliminary stage of this
argument was prior to addressing evidence from psycholinguistics. First, I pre-
sented the following considerations against RT: should we Wnd evidence of a
processing rule there is always the possibility that it is hardwired rather than
represented; it is implausible that the translation task of language processing
will involve a theoretical middleman; the linguistic rules are the wrong sort of
rules to be processing rules; we have no reason to suppose that the rules of
our preferred grammar are represented rather than those of another meaning-
equivalent grammar; it is hard to see a place for the represented rules in the
little that competencies may share; it is unclear how represented rules could be
used as data; language processing is not plastic enough and is too fast for it to
37 As Fodor notes: ‘‘The study of sentence parsing and production is concerned to a large extent
with processes and procedures about which ‘pure’ linguistics makes no claims.’’ However, she
continues: ‘‘Its closest contact with linguistic theory concerns the properties of mental representa-
tion. How are sentence structures mentally represented by speaker/hearers? How is the grammar
mentally represented?’’ (1992: 3). My second major conclusion, aimed at RT, implies that the
grammar is not mentally represented by speaker/hearers. My fourth tentative proposal in favor of
brute–causal processing implies that sentence structures are not either.
involve represented rules (11.4). I then considered what we can learn from the
psychology of skills and their acquisition. For, linguistic competence is a skill
and seems to be acquired by implicit learning. That psychology counts heavily
against RT (11.5). The Wnal stage of the argument brieXy addressed the psycho-
linguistic evidence on language use. I claim that this evidence does not under-
mine our preliminary assessment that there is no place for RT in the explanation
of language use. Indeed, the study gives further support to that assessment
(11.7–11.8).
This argument provides the most important step in the case for my second
major conclusion: there is no signiWcant evidence for RT and, given what else we
know, it is implausible. The Wnal step in the case is in the next chapter.
According to position (i), the structure rules of the language are also process-
ing rules that are represented and applied in language use. According to position
(ii), the structure rules are represented and used as data in processing. So both (i)
and (ii) are committed to RT. So the argument against RT obviously counts
against these positions. But much of the argument mostly counts also against
position (iv) which is committed to the representation of the processing rules for
language use that are unlike the structure rules for the language.
My second goal in this chapter has been to cast doubt on position (iii) and the
view that the structure rules for the language are unrepresented processing rules.
The structure rules are the wrong sort of rule to govern what is, given RTM, the
translation process of language use. The rules governing the retrieval processes of
language production and the interpretative process of language comprehension
look nothing like the language’s structure rules: my third tentative hypothesis
(11.4, 11.6–11.8).
These arguments against (i) to (iv) count in favor of (v). So I think that we can
predict with a fair amount of conWdence that the explanation of language use
will support (v): the processing rules are unrepresented and largely unlike the
My Wnal goal concerned the metalinguistic representations of the syntactic and
semantic properties of linguistic items. According to version (a) of (v), the
processing rules operate on these representations; according to version (b), they
do not. If the Representation Thesis were true then version (a) would be but even
if RT is false, (a) might be true: processing is a rational information Xow even if
the rules governing it are not represented. My goal has been to cast doubt on this
view, trying to make (b) seem plausible. I presented a case for the view that the
speedy automatic language processes arising wholly or, at least, partly from
linguistic competence are fairly brute–causal associationist processes. The case
is relatively a priori and far from decisive but I think it is appealing. It gets
support from the prominence of connectionist models in the theory of language
production and, at least, is not undermined by the state of play in theorizing
about language comprehension. I think that we can predict with some small
conWdence that we will discover that language processing does not operate on
Language Use 243
metalinguistic properties of the linguistic expressions but is more brute–causal:
my fourth tentative proposal (11.6–11.8).
My argument for this proposal, like the argument against RT that dominates
this chapter, reXect my commitment to Pylyshyn’s Razor. I think that RT is
implausible because the speaker’s representation of linguistic rules does no ex-
planatory work. And the diYculties in Wnding explanatory work for representa-
tions of linguistic properties leads me to the fourth tentative proposal.
The received Chomskian view is that the rules of a language are psychologic-
ally real in a speaker—whether they are represented or not—and play a role in
language processing. My fourth major conclusion, arising from the view that
language expresses thought (LET), was that this issue of psychological reality
should be investigated from a perspective on thought (8.3). This has already
yielded some signiWcant results, both positive and negative. On the positive side,
I have argued Wrst for the controversial Language-of-Thought Hypothesis
(LOTH) and then on that basis for (t) and my Wrst tentative proposal: a language
is largely psychologically real in a speaker in that its rules are similar to the
structure rules of her thought (9.5). And to the extent of that similarity, the
linguistic rules play a role in language processing because the rules governing that
processing operate on syntactic properties determined by the structure rules of
thoughts. On the negative side, if LOTH and hence (t) are false, the arguments
in this chapter against (i) to (iii) imply that the rules of the language are not
psychologically real in any other robust way. We have arrived at my Wfth tentative
proposal: if LOTH is false, then the rules of a language are not, in a robust way,
psychologically real in a speaker.38
The fourth major conclusion will yield further results in the next chapter when
we consider language acquisition. This consideration does not change the assess-
ment of language use in this chapter.
38 I noted the possibility earlier (4.6) of a less robust ‘‘intermediate’’ view on the psychologically
real issue: the grammar is descriptive of competence in a way stronger than simply positing rules that
are respected by competence—position (m)—but weaker than positing rules that govern process-
ing—positions (i) and (iii)—or are used as data in processing—position (ii). I have not argued that
there is no such position. But if there were, then the challenge would be to show that it was
theoretically interesting that the grammar was descriptive of competence in that way.
12
Language Acquisition
12.1 I N N AT E N E S S T H E S E S
Chomsky is famous for his strongly nativist answer to his question (ii), ‘‘How is
knowledge of language acquired?’’ (1.1). There is a rich innate ‘‘initial state’’ for
language acquisition that heavily constrains the languages that humans can ac-
quire in the way we naturally acquire our native language. On the basis of this
initial state and ‘‘primary linguistic data’’ (‘‘PLD’’), a person’s competence in some
language ‘‘grows’’ until it reaches a ‘‘Wnal state’’. Because of the innate constraint,
there is a certain system of rules (or principles) shared by all natural human
languages.
‘UG’ (‘Universal Grammar’) is used to refer, at least, to a theory of this
shared system of rules. But, because the received view is that the initial state
embodies these shared rules, ‘UG’ is often used as if it were deWnitive that it refers
also to a theory of the linguistic rules innately embodied in the mind (or, perhaps
to the rules themselves innately embodied). For reasons that will quickly become
apparent, it suits my purposes not to follow that usage. I shall take it as deWnitive
of ‘UG’ that it refers to a theory of the shared rules, leaving it as an open question
whether or not those rules constitute the initial state.1
Our main concern in this book has been with the Wnal state of a person’s
competence in a language and with the place of the rules, described by the
grammar of that language, in that state; our main concern has been with
positions on this described in section 3.4. We have so far found no persuasive
evidence to support the view that this Wnal state consists in a representation of the
rules of the language, no persuasive evidence for the Representional Thesis (RT).
So we have found no persuasive evidence for position (i) according to which
those rules are represented and applied in processing, nor for position (ii)
according to which those are represented and used as data in processing. Indeed,
1 Although there is a lot of agreement about UG, agreement about what rules are shared and
innately embodied, there is still some disagreement. The disagreement is beside the point of this
chapter and so I shall ignore it, writing as if there were a settled UG.
Language Acquisition 245
the evidence suggests that the rules are not represented. This evidence includes
the evidence on the nature of skills, discussed in section 11.5. For, competence in
a language is a skill and the weight of evidence is that skills do not typically
involve representations of the rules that govern them. The evidence discussed in
that section also included evidence from skill learning. Language learning seems
to be a paradigm of implicit learning and the evidence strongly suggests that
implicitly learning a skill does not involve representing the rules that govern it.
However, that earlier brief discussion of language learning did not address the
very important innateness issue. Perhaps attention to this issue will show that RT
is true after all. We shall address that question at the end of the chapter. We shall
see that the bearing of the innateness issue on RT, and on other views of the Wnal
state, is not as direct as one might have expected.
Aside from considering this bearing, an interest in the psychological reality of
language leads to a concern with the innateness issue in its own right. What is the
initial state from which the Wnal state of competence arises? In particular, are the
rules described by UG embodied, perhaps even represented, in the initial state?
Most of this chapter will be concerned with these questions and Chomsky’s
nativist answer to them.
Chomsky’s claims about innateness are thought by him and others to force a
reassessment of the traditional debate between empiricists and rationalists over
‘‘innate ideas’’. Rationalists thought that many concepts were innate. Empiricists
rejected this. It is thought that Chomsky’s nativism strongly supports the ration-
alist side.
In considering this nativism, it is important to distinguish a range of diVerent
theses that are not adequately distinguished in the literature. First, there is a
boring thesis. This is the thesis that human beings are innately predisposed to
learn languages; it is because of some innate initial state that, given linguistic
data, almost every human learns a language. The thesis is boring because every
informed person, even the crudest empiricist, should believe it and, so far as I
know, does believe it. How else, for example, could one explain the fact that
humans can, but dogs cannot, learn English? One needs to go further to make an
innateness claim interesting, saying something about the innate initial state.
Chomsky, of course, typically goes a lot further.2
Second, there is a set of three interesting theses:
1. The initial state is a language-speciWc learning device in a distinct ‘‘language
faculty’’ not in the central processor. The innate state that makes language learning
possible is not simply a general learning device that makes all learning possible, it
is a special, relatively central, module of the mind.
2 However in one place, amazingly, Chomsky does seem to take the innateness thesis to be the
boring one: ‘‘To say that ‘language is not innate’ is to say that there is no diVerence between my
granddaughter, a rock, and a rabbit’’ (2000b: 50).
This is the minimal interesting nativist thesis. It tells us where the initial state is
in the mind. But it does not tell us anything about the constraints that this state
puts on language learning. The next thesis does.
2. The initial state constrains humans to learn only languages that conform to
UG. So any language that they can learn naturally will have rules speciWed by
UG. Let us describe this as the initial state ‘‘respecting’’ the rules described by
UG—the ‘‘UG-rules’’—on analogy with our earlier talk of the Wnal state of
competence respecting the rules described by the grammar of the language learnt.
We pointed out that this thesis about the Wnal state should be uncontroversial
(2.4). Not so the thesis about the initial state. For, if it is right, the universality of
UG-rules is not a mere accident of human history but is determined by our
biological heritage: it is ‘‘in our genes’’.
This seems to me a more interesting thesis than 1. Where 1 simply tells us
where the initial state is, 2 tells us that this state constrains our languages in a very
signiWcant way. But 2 does not tell us what innate language-constraining rules in
the initial state make this state respect the UG-rules. The next thesis does.
3. The initial state respects the UG-rules because it embodies the UG-rules. Not
merely do we inherit some language-constraining rules that makes us respect the
UG-rules, which is all that thesis 2 requires, we inherit the UG-rules themselves.
This addresses the ‘‘what’’ issue and so is clearly an even more interesting thesis
than 2. I suspect that linguists Wnd the move from 2 to 3 easy because of their
conXation of a theory of syntax—a grammar—with a theory of competence.
I emphasized that the conXation encourages the view that the psychological
reality of grammatical rules ‘‘comes for nothing’’ with the grammar (Ch. 2).
Similarly it encourages the view that the psychological reality of UG-rules ‘‘comes
for nothing’’ with universal grammar, UG. Suppose that this reality did come for
nothing. Then the idea that the initial state respects the UG-rules because the
UG-rules are innate would be very inviting, even if not compelling. But the
psychological reality does not come for nothing: the conXation is a mistake.
The psychological reality of UG-rules is not a ‘‘free lunch’’ but something that
requires psychological evidence.3
Combining 1 and 3 we get the view that the language-constraining rules are UG-
rules and they are in a language faculty. Chomsky certainly embraces theses 1 and 3,
at least (1991a: 22). Now it is important to note that neither of these theses entail
that speakers have innate representations of linguistic rules or innate propositional
knowledge—knowledge-that—about them: the innate rules might be simply
embodied and any innate knowledge might be simply knowledge-how (3.1). As a
3 Two nativists inspired by Chomsky recently stated their nativism in a form that entails 2 but
not 3: ‘‘there are substantive ‘universal’ principles of human grammar and, as a result of human
biology, children can only acquire languages that conform to these principles’’ (Crain and Pietroski
2002: 163).
result, these theses alone do not entail the existence of any innate concepts and so do
not seem to bear on the traditional debate over innate ideas. In any case, interesting
as these theses undoubtedly are, they are not at the center of the debate. For,
Chomsky and his followers seem to go signiWcantly beyond them in claiming that
linguistic rules are indeed innately represented and propositionally known.
Third, this move to knowledge and representation turns the interesting thesis
3 into a very exciting thesis that clearly involves a commitment to innate ideas.
This is a thesis that no empiricist, traditional or contemporary, could allow:
3r. The UG-rules that are innately embodied according to 3 are so because they
are innately represented in Mentalese: UG itself, the theory of those rules, is
innately known. This ‘‘initial-state representationalism’’ is, of course, the ana-
logue of the ‘‘Wnal-state representationalism’’ expressed by the Representational
Thesis (RT). So, let us call 3r the ‘‘I-Representational Thesis’’.
Combining 3r with 1 yields the view that the language-constraining rules
represented in the language faculty in Mentalese are UG-rules: UG itself is
innately known and in the language faculty.
There can be no doubt of Chomsky’s commitment to a distinct language
faculty and so to thesis 1. And it is natural to take him to be committed to 3r and
I-Representationalism just as it was natural to take him to be committed to RT
(1.1); see Chomsky 1965, pp. 25–33, for example. Certainly, Jerry Fodor takes
him to be:
what Chomsky thinks is innate is primarily a certain body of information: the child is, so to
speak ‘‘born knowing’’ certain facts about universal constraints on possible human
languages. (1983: 4)
Indeed, as Fodor points out, only by construing Chomsky’s nativism in this way
can we see it as in the rationalist tradition of innate ideas:
It’s because Chomsky holds that the innate information available in the initial state of
language acquisition is ipso facto among the intentional object of the learner’s proposa-
tional attitudes that Chomsky’s theory of mind is indeed continuous with the tradition-
ally rationalist postulation of innate ideas. (2001: 109–10)
Graves et al. make the same point in attributing innate tacit knowledge to the
child:
In order for rationalists to argue from the existence of innate structures to the existence of
innate knowledge, some notion of tacit knowledge is required so that they can claim that
the grammatical structures represented in transformational grammars are objects of
knowledge. (1973: 318)
So it seems that Chomsky believes the combination of 1 and 3r, the most
exciting thesis of all. A great deal of the attention given to Chomsky’s claims
about innateness comes from construing them in this very exciting way. So I shall
call this combination ‘‘Chomskian Nativism’’.
12.2 EV I D E N C E F O R N AT I V I S M
In this section, I shall brieXy consider some arguments, mostly very familiar ones,
noting which of the various nativist theses they seem to support.
Sui generis: It is pointed out that language learning is sui generis: it is quite
unlike the acquisition of other cognitive skills. (i) It is done very young and has to
be done before about age twelve, within ‘‘the critical period’’. (ii) The level of
achievement is quite uniform by comparison to other intellectual skills: all
normal children acquire a language. (iii) All children, whatever their language,
acquire elements of linguistic capacity in the same order. Acquisition is develop-
mentally uniform across individuals and cultures.4
This evidence might well be thought to support thesis 1, the minimal inter-
esting nativist thesis that language learning is controlled by a language-speciWc
faculty, distinct from the central processor. I shall later argue that this evidence is
in fact quite compatible with language learning being controlled by the central
processor (12.5). What needs to be emphasized now, however, is that this
evidence alone gives no support to thesis 2, hence none to 3 or 3r, because it
throws no light on the constraints that the initial state places on the languages
that we can learn. In particular, it does not show that they must comply with the
UG-rules.
I turn to some arguments that, taken as a package, give persuasive support to
thesis 2: humans are innately constrained to learn only languages that conform to
UG; the initial state respects the UG-rules. This is certainly an interesting nativist
thesis, but it is not as strong as 3, let alone 3r. What needs to be emphasized is
that these familiar arguments alone do not justify adopting a stronger thesis than
2, because they do not show us what it is about the initial state that makes it
respect the UG-rules. Nor do these arguments support thesis 1, for they throw no
light on where the initial state is to be located in the mind. In sum, these
arguments leave us way short of Chomskian Nativism.
Universality: The Wrst argument is simple. Suppose, as I am prepared to, that
the linguists are right in thinking that all human languages are governed by UG-
rules. Then this is a very striking universality. UG-rules are quite unobvious and
it is not hard to invent a language that is not governed by them; for example, one
that does not observe the structure-dependency principle. How come no human
language is like this? Thesis 2 provides a plausible explanation: humans are
innately constrained to speak only UG languages. Indeed, how else is the
universality to be explained? Putnam (1967) has an answer: human languages
are all governed by UG-rules because they are all, as Stephen Stich puts it,
‘‘descended from a single common ur-language whose details have been lost
in pre-history’’ (1978a: 283). Putnam points out that if nativism were correct
4 See Laurence and Margolis 2001, sec. 4 for a nice summary.

then our language capacity would have come in just one evolutionary leap.
Now if there was that one leap, it very likely yielded a single language which
was the common ancestor of all human languages. But then having that common
ancestor alone explains why all languages share certain rules. So, we have no need
to see humans as innately constrained to follow those rules: they follow
them because, ‘‘by chance’’, the Wrst language followed them and has passed
them on. However, this ingenious explanation has a crucial Xaw. Not all human
languages are descended from a common ancestor: some are created almost from
scratch, as we shall see in a moment. So the nativist explanation still seems the
best.
Poverty of Stimulus: The nativist argument that receives the most emphasis
concerns the poverty of stimulus received by the language learner. The gram-
matical rules picked up by the child are abstract, subtle, and unobvious. Yet,
Chomsky claims, the child learns these rules from data of degenerate quality and
limited range (1965: 58). The problem with the quality is that the data include
many ungrammatical strings: false starts, slips of tongue, ‘um’s and ‘ah’s, and so
on. The problem with the range is that the data seem to provide no evidence
bearing on many of the rules that the child masters; in particular, they contain
little explicit instruction and almost no negative evidence, evidence that some-
thing is not grammatical. It is hard to see how the child could derive the linguistic
rules from these impoverished data available to her. She must have a head start,
being tightly constrained to favour UG-rules.
This argument for thesis 2 has a lot of appeal at Wrst sight but that appeal is
greatly diminished by a closer look. First, Fiona Cowie (1998: chs 8 and 9) has
argued persuasively that the linguists’ claims about the data available to the child,
plausible as they may seem, are actually rather dubious.5 For, they are largely
based on the intuitions of linguists about the available data rather than on
empirical evidence. And where attempts have been made to gather the evidence
they suggest that the data available to the child is much richer than Chomskians
suppose; for example, that it includes evidence for the structure-dependency
principle; and that it includes lots of negative evidence. In sum, we lack evidence
about just how impoverished the child’s data really is.
Next, I think that the possibility of an alternative nonnativist explanation of
learning from impoverished data has been dismissed too quickly. Chomsky tends
to restrict his opponents to crude empiricist methods of learning that are
hopelessly inadequate to explain learning in general. So one possible response
to Chomsky is to posit a richer and more sophisticated innate general learning
5 Laurence and Margolis (2001) are not convinced. Nor are Crain and Pietroski (2001). See
Lewis and Ellman 2001; Pullum and Scholz 2002; and Scholz and Pullum 2002 and 2006, for
further criticisms of poverty of stimulus arguments. Lewis and Ellman draw attention to the
importance of statistical information available to the language learner: ‘‘the statistical structure of
language provides for far more sophisticated inferences than those which can be made within a
theory that consides only whether or not a particular form appears’’ (p. 369).
device. In eVect, Putnam made a response of this sort (1967: 297–8). Cowie has
several interesting further suggestions along these lines. But our earlier discussion
of the psychology of skills (11.5) suggests a rather diVerent response. Where
Putnam and Cowie follow Chomsky in treating language learning as if it were a
largely top-down rational process, we should rather treat it as the largely bottom-
up implicit learning of a skill, more brute–causal than rational.6 Although it is
plausible to think that many skills are acquired in the implicit way, we know very
little about how they are. In particular, we know next to nothing about how a
skill might be thus acquired on the basis of what seems to be impoverished data.
(How do we learn to throw a frisbee or ride a bicycle?) If we did know more about
this in general, we might be able to see how a child could learn a language from
impoverished data.
So I think that we should be a bit dubious of the claim that the data available
to the child is so impoverished and a bit dubious of the claim that the child could
not learn a language from such data as is available. Despite these doubts, the
poverty-of-stimulus argument does still seem to me to have some force. For, it
does still seem to me likely that the data available to the child is too impoverished
for her to learn the language unless she is tightly constrained. I just think the
poverty-of-stimulus argument we have been considering is some way from
having established this. However, the arguments to follow, which one might
also label ‘‘poverty-of-stimulus arguments’’, do seem to me to advance the case
for nativism considerably.
Language Creation: (i) Some natural languages are sign languages. These are
also governed by UG-rules but are nonetheless often very diVerent from the local
spoken language: ‘‘natural sign languages show all the structural properties of
other human languages yet have evolved independently of the spoken languages
that surround them’’ (Newport and Supalla 1999: 758). Thus, ASL, widely used
among the deaf in America, is very diVerent from English but similar to some
other spoken languages.This strongly suggests that the initial state for human
language is innately constrained to respect the UG-rules, as thesis 2 claims. This
is further conWrmed by the fact that children do not readily acquire derivative
sign languages, ones intentionally invented, ‘‘presumably because [these lan-
guages violate] natural structural principles for human languages’’ (p. 758). (ii)
The deaf provide another example of language creation: ‘‘isolated deaf children
not exposed to signed languages spontaneously generate gestural systems that
share many formal and substantive features with received languages’’ (Gleitman
and Bloom 1999: 435). The New York Times recently described an example. The
al Sayyid Bedouin Sign Language, used by about 150 deaf people in a village in
the Negev desert, ‘‘developed spontaneously and without outside inXuence. It is
not related to Israeli or Jordanian sign languages, and its word order diVers from
6 And this treatment gets some support from studies on early mappings of words to referents;
Samuelson and Smith 1998, 2000.
that of the spoken languages of the region’’. The language ‘‘which started only 70
years ago’’ is still ‘‘under development’’ (Wade 2005: f3). (iii) There is striking
evidence, discussed by Pinker (1994: 32–9), that children brought up speak-
ing Pidgin—a simple makeshift language with little grammar—‘‘creolize’’ that
language in one generation into a language with the complex grammar of any
normal human language. Thus, young Nicaraguan deaf children brought into a
school where the older children used a pidgin sign language, LSN, developed a
creole sign language, ISN. All of this suggests that ‘‘children do not merely
learn language; they create it’’ (Gleitman and Bloom 1999: 436). And the fact
that the languages they create all conform to UG-rules is powerful evidence for
thesis 2.
The Continuity Hypothesis: Stephen Crain and his colleagues have urged
‘‘the continuity hypothesis’’, maintaining ‘‘that child language can diVer from the
language of the linguistic community only in ways that adult languages can diVer
from each other’’ (Crain and Pietroski 2001: 146). Children do not immediately
arrive at the language that surrounds them, they make ‘‘errors’’; for example,
many insert ‘‘ ‘extra’ Wh-words in their long-distance questions’’ as in ‘What do
you think what pigs eat?’ This ‘‘error’’ is ‘‘presumably not a response to the
children’s [English-speaking] environment’’ (p. 179) but it exempliWes a con-
struction found in German and some other languages. This seems to be evidence
that ‘‘children are trying out grammars with features found in adult languages
elsewhere on the globe’’ (pp. 178–9). If it can be established that children never
try out languages that violate principles of Universal Grammar, then this would
indeed be further strong evidence for the nativist thesis 2.7
In sum, I think that Universality, Poverty of Stimulus, Language Creation, and
the Continuity Hypothesis, jointly present a persuasive case for thesis 2.8 How-
ever, it is important to note, they alone are not suYcient to establish anything
stronger. They support the view that the initial state respects the UG-rules but
tell us nothing about what makes that state respect these rules. Thus, they alone
do not establish that the initial state respects the rules by embodying them, thesis
3, and even less that it does so by representing them, thesis 3r. Nor do the
arguments tell us that the initial state is in a language faculty, thesis 1. Only Sui
Generis seems to give direct support to thesis 1.
7 See Crain and Pietroski 2002, sec. 3 for a summary of further evidence for the continuity
hypothesis.
8 I don’t think that the case is conclusive, however: more work needs to be done. And I join
with Scholz and Pullum (2006) in being skeptical of recent triumphalist claims for linguistic
nativism. A particular cause of concern about 2 arises from the distinction between ‘‘core
language’’ and ‘‘periphery’’ where the latter is not ‘‘determined by Wxing values for the parameters
of UG’’ (Chomsky 1986: 147). On the basis of a study of the hard cases for linguistic theory, ‘‘the
syntactic nuts’’, Peter Culicover (1999) has argued that the periphery is so extensive as to cast
doubt on the whole idea that there is a core determined by UG; see J. D. Fodor 2001 for a
judicious assessment of the argument. (Thanks to my student Francesco Pupa for drawing my
attention to this issue.)
In what follows I shall take thesis 2 for granted. So, an interesting nativism will
not be in question. I shall be concerned with whether we should go further down
the nativist path. I shall start by dismissing the argument that we should go much
further to 3r and I-Representationalism because that is ‘‘the only theory in town’’
(12.3). But, on a positive note, I shall present a case for adopting thesis 3, or
something close (12.4). I shall then turn negative again, looking skeptically at
thesis 1 (12.5). And I shall argue Wrmly against adopting the I-Representational
Thesis (12.6). I shall conclude by considering the bearing of this discussion on
the main concern of this book, the nature of the Wnal state (12.7). This will
conclude my argument against RT. So I will be concluding that a person in the
initial state is totally ignorant of the rules of UG and a person in the Wnal state
can be totally ignorant of the rules of her language.
1 2 . 3 ‘‘ T H E O N LY T H E O RY I N TOW N ’’ A N D
I - R E P R E S E N TAT I O N A L I S M ( T H E S I S 3 R )
We are assuming that the initial state respects the UG-rules—thesis 2—but why
should we take the large step to the ‘‘very exciting’’ I-Representationalist thesis
3r? Even if thesis 3 is correct and humans respect the UG-rules because the initial
state embodies them, why suppose that it does so by representing them? Why
suppose that UG itself is psychologically real?
Once again we face an only-theory-in-town argument (cf. 11.2). Many lin-
guists think that we can explain language learning if we see it as a rational process
of hypothesis testing or, recently, of parameter setting, for the language of the
child’s community. And there is no other way to explain it. So, any linguistic rule
that plays a role in language learning has to be represented, for that is what the
rational process demands.
Fodor has oVered the most explicit version of this argument. He characterizes
Chomsky’s account of language learning as follows:
[It] is the story of how innate endowment and perceptual experience interact in virtue of
their respective contents: the child is viewed as using his primary linguistic data either to
decide among the candidate grammars that an innately represented ‘‘General Linguistic
Theory’’ enumerates (Chomsky 1965) or to ‘‘calibrate’’ endogenous rule schemas by
Wxing parameter values that the innate endowment leaves unspeciWed (Chomsky 1982).
This sort of story makes perfectly good sense so long as what is innate is viewed as having
propositional content: as expressing linguistic universals, or rule schemas, or whatever.
But it makes no sense at all on the opposite assumption. (1983: 5)
And Fodor emphasizes, with characteristic exhuberance, ‘‘that no account of
language learning which does not thus involve propositional attitudes and
mental processes has ever been proposed by anyone, barring only behaviorists’’
(1981a: 194).
I have earlier (11.2) suggested two criteria for a good abduction. (A) The
abduction must involve a good explanation. This has at least two parts. (A1) The
explanation must be plausible given what we already know. (A2) The explanation
must have an appropriate level of detail. (B) The explanation must be better than
any alternative that is actual or likely given what we already know. With these
criteria in mind, it is interesting to consider some of Fodor’s responses (2001) to
Cowie’s critique of Chomskian Nativism, a nativism that she rightly takes to
include I-Representationalism.9
As Fodor notes, Cowie accepts that Chomskian Nativism ‘‘is the best available
theory of language acquisition’’ because rival theories ‘‘do not yet exist’’. Yet she
still rejects an abduction along the lines illustrated above (Cowie 1999: 249).
Fodor Wnds this ‘‘puzzling’’: ‘‘One might have thought that I just couldn’t have a
better reason for preferring my theory to yours than that yours doesn’t exist’’
(2001: 115n). There is no puzzle. Cowie is prepared to accept that the Chom-
skian explanation is the best currently available. But she also clearly thinks, for
reasons described brieXy below, that it is not good enough; it is not ‘‘suYciently
well articulated at this time’’ (p. 210); ‘‘it provides no real explanation of
language acquisition at all’’ (p. 249). In brief, she thinks that it badly fails my
criterion (A). Indeed, she may also think that it fails criterion (B), for she thinks
that the development of alternative explanations has been stiXed by the domin-
ance of the Chomskian paradigm. So she may not think that the Chomskian
explanation is better than any alternative that is likely given what we already
know. In any case, it is clear that what she is urging is that we suspend judgment on
Chomskian Nativism.
We might anticipate that Fodor would be unimpressed with the virtues of this
course of action. Years ago he claimed that ‘‘remotely plausible theories are better
than no theories at all’’ (1975: 27). And he seems never to have met an only-
theory-in-town argument he didn’t like. So it is no surprise that he is quite
unmoved by Cowie’s criticism that Chomskian theories do not show how a
language learner exploits the information allegedly provided by UG. ‘‘Cowie
needs a principled reason for doubting that the problem about how UGs function
in language acquisition can be solved’’ (2001: 117).10 No she doesn’t! Fodor
thinks that ‘‘all she’s got is that, to date, nobody has solved it’’ (ibid). And that,
near enough, is all she needs to suspend judgment. She would need a principled
9 In his enthusiasm for Cowie’s critique, which knows every bound, Fodor strangely misrep-
resents her as rejecting (what I have called) RT (2001: 111–13). I only wish that she did but
she didn’t. She takes Chomsky’s famous argument against Skinner to have ‘‘established’’ that
‘‘language mastery involves a wealth of syntactic, semantic, and pragmatic knowledge . . . . It
requires the endorsement of (R)’’ (1999: 162). (R) is a thesis, ‘‘Representationalism’’, that is
similar to my RT.
10 Consider also the following response: ‘‘the entire force of [Cowie’s] argument comes down to
the observation that speciWc UG-based accounts of particular aspects of language acquisition are
controversial’’ (Laurence and Margolis 2001: 260).
reason for doubting that an innate UG could explain language acquisition if
she were arguing that it couldn’t. But she is not. Rather, she is arguing that we
have insuYcient reason now to say that an innate UG does explain language
acquisition.11
It may be better to entertain remotely plausible theories rather than entertain
none but that does not show that it is better to believe those theories rather than
believe none. And it is not better unless there is good evidence for them. Fodor’s
abductive standards are too low.
Further conWrmation of this emerges later. According to the famous poverty-
of-stimulus argument for Chomskian Nativism, the PLD is too impoverished for
a person to learn a language from it without the help of rich innate linguistic
constraints. Cowie’s extensive study of the empirical literature leaves her skeptical
about this. Again, Fodor is unimpressed. ‘‘What she needs, but clearly doesn’t
have, is an argument that the available data suggests, even remotely, a PLD so
rich that the child can, as it were, squeeze through with lots of room to spare’’
(2001: 121). And, once again, she doesn’t need this. To suspend judgment she
does not have to argue that the PLD is that rich. She just has to argue, as she does,
that there is a lack of evidence that the PLD is as poor as the poverty-of-stimulus
argument demands. She is arguing that there has been an unwarranted rush to
judgment, not that the judgment is wrong.
Cowie’s reasons for having a dim view of the currently available Chomskian
nativist explanations of language acquisition are brieXy as follows. These explan-
ations come in two sorts, the older hypothesis-testing models and the currently
more popular parameter-setting models. She discusses the latter Wrst. She rightly
points out that ‘‘there is no parameter-setting theory of language acquisition’’ but
really only ‘‘theory fragments’’ (1999: 257). She argues, convincingly in my view,
that these fragments have many problems (pp. 257–63). She then considers
hypothesis-testing models. Formal learning theory has developed sophisticated
versions of these which arguably show how languages might be learnable but they
are not, and do not claim to be, psychologically plausible. They do not show how
a child could learn a language ‘‘within some developmentally credible amount of
time’’ (p. 266) and they ascribe assumptions to the learner that cannot plausibily
be ascribed to a child (pp. 266–9). Years ago, Pinker gave six conditions that a
theory of language acquisition must satisfy and argued that ‘‘no current theory of
language learning satisWes, or even addresses itself to, all six conditions’’ (1979:
218). Cowie concludes, I think rightly, that ‘‘the situation is little better today’’
(1999: 270).
11 Collins also misses this point: ‘‘Cowie (1999, p. 272) scolds the nativist for ‘I’m the only
president you’ve got’-style arguments; this is quite jejune. No-one seriously involved in linguistics
and related disciplines is trying to gain any knock-outs’’ of empiricism (2003: 187). Cowie’s main
criticism of this style of nativist argument is not that it fails to ‘‘knock out’’ empiricism but that it
does not justify Chomskian Nativism.
Fodor’s response to Cowie’s criticism of Chomskian Nativism is an extreme
example of a common response to such criticisms. Time and again the lack of a
detailed alternative empiricist explanation of language acquisition is trotted out
as if it were a near decisive point in favor of Chomskian Nativism. Consider
another three responses to Cowie, for example.12 (i) Crain and Pietroski have the
following plonking response to Cowie’s suggestions about the data that may be
available to the child and about what the child might be able to extract from the
data: ‘‘These would be interesting suggestions, were they accompanied by a
proposal about how children extract the various kinds of constraints . . . from
the data available to them. But Cowie does not, in this sense oVer an alternative
to positing a Universal Grammar’’ (2001: 159–60). (ii) Stephen Laurence and
Eric Margolis clearly give a lot of weight to the following: ‘‘No one has a fully
articulated account of language acquisition to oVer and between the two camps
nativists clearly have the more detailed theories’’ (2001: 244). (iii) John Collins
defends Chomsky’s demand that the empiricist produce an alternative as follows:
‘‘It is because empiricism and its behaviourist progeny so grievously underesti-
mated the complexity of what a speaker knows that it is apposite to demand
concrete proposals which are sensitive to the many data. Otherwise, there is
‘nothing to discuss’ ’’ (2003: 187). But, to repeat the methodological point, the
defense of Chomskian Nativism requires much more than pointing to the poor
state of alternatives. It requires that the nativist explanation be good. And that
requires that it be worked out in suYcient detail to be convincing. Until that has
been done, we should suspend judgment on it even if we had no ideas for an
alternative.
Finally, it is worth noting that there are some well-known problems for
parameter-setting models. On these models, the language acquisition device
has to set twenty to thirty binary switches on the basis of PLD. That’s all. This
might seem fairly simple at Wrst sight. Yet, in fact, it is very diYcult. Here is the
problem:
No input sentence ever exhibits the action of one and only one parameter. Even the
simplest sentences involve combinations of parameter settings. A simple transitive sen-
tence in English, for example, involves the action of parameters regulating the direction
and form of Case assignment, the attachment of verbal morphology, agreement, and
theta-role assignment. Thus, sentences look the way they do because principles and
parameters form coalitions to derive the language. In order to represent the input stream
and properly set parameters to their target values, the learner must untangle these
coalitions of parameters. (Clark 1994: 483)
Some ingenious attempts have been made to solve this problem but none has yet
succeeded. Janet Fodor sums up the situation as follows:
12 Consider also Lasnik and Uriagereka 2002 (p. 150) and Crain and Pietroski 2002 (pp. 173–6),
responding to Pullum and Scholz 2002. Scholz and Pullum aptly reply: ‘‘It is possible to evaluate an
argument for a view without raising questions of theory comparison’’ (2002: 187).
Despite years of eVort, it has proven extremely diYcult to show how the syntax of a
natural language could be acquired even if it were completely innate except for 20 to 30
binary choices to be determined by the input sample. (2001: 368)13
In sum, Cowie Wnds the abductive argument for 3r and I-Representationalism
unconvincing. I think that she is right to do so. Still, as noted, she is inclined to
think that I-Representationalism yields the best explanation of language acqui-
sition currently available, even though not one that is good enough. I think that
this is far too generous. In the next three sections I shall say why: given what we
already know, explanations along other lines seem much more promising.
12.4 E M B O D I M E N T O F U G - RU L E S ( T H E S I S 3 )
In these three sections, I shall be arguing about which way the future may lie in
the explanation of language acquisition. I shall certainly not be aiming to
establish an abduction that is a rival to the only-theory-in-town one. That
would be a foolish aim at this time because, so far as I can see, we are nowhere
near an explanation of language acquisition with suYcient detail and evidential
support to be the basis of such an abduction. So, the best we can hope for at this
time is an explanation sketch that seems likely to be the basis for a persuasive
future abduction.This is not surprising. Linguistic competence is a skill and, as
we noted earlier (11.5), it is very early days in our understanding of skills and
their acquisition. Given how little we know about the acquisition of skills in
general it would indeed be surprising if we could explain language acquisition.
Not so long ago, Reber bemoaned the fact ‘‘that the topic of learning should be so
poorly represented in the contemporary literature in cognitive psychology’’.
Indeed, he thought that this may explain the move to nativism. ‘‘Failure to
explicate how complex knowledge is acquired invites the supposition that ‘it was
there all the time’ ’’ (1989: 219).
Far as we are from an explanation of language acquisition, psycholinguists are
of course making steady progress toward one. I want to suggest a place where
some other progress can be made. According to thesis 2, which we are assuming,
humans are innately predisposed to learn languages that conform to UG; the
initial state respects the UG-rules. Clearly this calls for an explanation: What is it
about initial state that makes it respect those rules? How do we explain the
respect? The answer must form the core of any explanation of language acqui-
sition. So, any light that we can throw on the answer should guide us in seeking
an explanation of language acquisition, it should guide us to the future.
13 I am indebted to Janet Fodor for instruction in these matters. She and her colleagues at CUNY
think that they are on the way to a solution. The key is the ‘‘parametric decoding’’ of the input
sentences: ‘‘Sentences must be parsed to discover the properties that select between parameter
values.’’ (Fodor 1998: 339)
Thesis 3 provides an answer: the initial state respects UG-rules because it
embodies them; those very rules are innate in us. In this section I shall explore the
prospects of this explanation from the perspective of views on the nature of
thoughts presented in Part IV. I shall also very brieXy consider the prospects of
some alternative explanations. In the next, section 12.5, I shall consider the
consequences of this discussion for the language faculty and thesis 1. None of this
discussion suggests that the initial state represents UG-rules as thesis 3r, and
hence Chomskian Nativism, requires. I shall continue the case against this
I-Representationalist thesis in section 12.6. 3r is not only not now part of a
good explanation of language acquisition (12.3) but, given what we already
know, it is not likely to be part of a good explanation in the future.
A view of how a competence is acquired needs to be guided by a view of the
nature of the competence that is acquired.14 So, our view of language acquisition
should be guided by our view of the Wnal state of linguistic competence. The
relatively uncontroversial assumption that language expresses thought (LET)
yields the view that linguistic competence is partly constituted by conceptual
competence. So our view of the Wnal state depends on our view of the nature of
thoughts. This led to my fourth major conclusion: that the psychological reality
of language should be investigated from a perspective on thought (8.3). Position
(t), based on the Language-of-Thought Hypothesis (LOTH), is the most com-
mitted position on this nature presented in Part IV: a person thinks in a
Mentalese governed by structure rules that are similar to those of her language.
What explanation does this suggest about why the initial state respects the UG-
rules? It suggests that the state respects those rules because they are largely, if not
entirely, innate structure rules of Mentalese. Thus, it suggests something close to a
version of thesis 3. This suggestion is my sixth tentative proposal.
Why the qualiWcations in the proposal? Well, part of the explanation of the
respect may be found not in the innate structure of thought but in the innate
processing constraints on the way thoughts can be expressed in language and on
how language can be understood in thought. Still, position (t) should make us
wonder whether the innate constraints on Mentalese alone explain why the initial
state for language acquisition respects the UG-rules. Furthermore, if there were
these initial-state processing constraints, they should be reXected in substantial
commonalities among Wnal-state processing competencies. Yet, our earlier dis-
cussion of brain impairment suggests that there are no such commonalities
(10.2).15
What we are contemplating here is the idea that we are innately disposed to
think thoughts that are governed by the structure rules of UG. It is because of the
14 As Mark Singley and John Anderson point out (1989: 1), citing Newell and Simon 1972.
15 And the evidence from brain injury does not seem to help with the innateness issue.
‘‘Absolutely no current theory of brain localization or recovery can accommodate [data on hydro-
cephalus]. In general, the brain injury data show surprising robustness of language, without giving
straightforward answers to questions of language innateness.’’ (Maratsos 2003: 696)
constraints on thought thus imposed that all human languages share certain
features. Thus, it is because all structure rules for thought accord with the
principle that grammatical rules are structure-dependent (‘‘the structure depend-
ency principle’’) that all linguistic rules do. It is because of such constraints on
thought that language learning has the special features much emphasized by the
linguists. On the basis of these innate rules and, according to the usual story, only
debased and impoverished experience of her linguistic environment, a person
comes to have thoughts largely governed by the rules of a complicated language,
rules that underlie her own linguistic performance. Language learning appears to
be sui generis because the development of thought under the stimulus of language
(8.6–8.7) is sui generis.
We have good reason to suppose that all the higher animals have innate
capacities to think but that these capacities diVer; in particular that the human
capacity is far greater than that of, say, the dog. What do these innate diVerences
consist in? They consist in diVerences in the structure and processing rules
governing the systems of representation in which animals think. These innate
rules constrain an animal’s thinking. If the animal has the capacity to learn a
language, as the human does, then the innate structure rules of thought must
constrain the languages it can learn: it cannot learn a language that expresses
thoughts it is innately constrained not to think. So, some innate constraints on
language, at least, must come from the innate constraints on thought.16 Now add
to this picture the following view of the Wnal state of language learning for a
human: the human thinks in a Mentalese with structure rules that are similar to
the structure rules of the language learnt, position (t). On this view the innate
structure rules that constrain thoughts and hence language are the innate syn-
tactic rules of Mentalese. And a nice explanation of why we are innately
constrained to acquire languages that respect the UG-rules is that those rules
are largely the innate structure rules of thought: thought constrains language by
embodying the UG-rules.
I earlier considered an objection to the thesis that a person’s language of
thought is her public language (Public-LOTH). The objection pointed to the
fact that we can understand a string of words that is not in our language and,
indeed, could not be in any human language (9.4). It might be objected now that
this fact counts against the view that the innate constraints on language stem
from the innate structure rules of Mentalese. For, the fact seems to show that
there are greater constraints on human language than on thought. But does it
really show this? As I responded before, perhaps we understand the string by
‘‘translating’’ it into a ‘‘well-formed’’ thought with the syntax like that of our
natural language. But let us suppose that we sometimes do not, and accept that
we can think thoughts that are not ‘‘well-formed’’. This need not undermine the
16 Note that Pinker moves without comment from talk of the ‘‘language instinct’’ to talk of ‘‘the
instinctive nature of thought’’ (1994: 20).
proposed view of innateness. The view need not be that a person cannot think at
odds with the rules of her Mentalese—rules that are largely those of her lan-
guage—or at odds even with the rules of the initial state. The claim could be that
the speaker is not disposed to think such thoughts. So it is not ‘‘natural’’ to think
them. But perhaps one can deliberately think them; and perhaps performance
error leads one accidentally to think them from time to time.
In sum, assuming thesis 2—the initial state respects the UG-rules—position
(t), based on LOTH, suggests that something close to thesis 3 is right: the state
respects the UG-rules because it largely embodies them as structure rules of
thought. That’s my sixth tentative proposal. Of course, we do not now have
anything close to a good explanation of language acquisition along these lines.
That’s one reason why the proposal is tentative.17 Still, if thesis 2 and position (t)
are right, we can expect something close to thesis 3 to be the core part of a
persuasive future explanation of language acquisition.
This sort of proposal is far from novel. As Chomsky himself notes (1966:
52–3), a proposal like this is to be found in the eighteenth-century Port-Royal
grammarians. And Michael Maratsos (1989: 122) has aired a similar one more
recently.18 These proposals should surely be attractive to the nativist. So one
wonders why they have not been more popular. Perhaps Chomsky’s argument to
be considered in the next section is the explanation.
(t) arises from a bold conjecture about the nature of thought, LOTH. But,
suppose now that we drop LOTH, and hence (t), and have some other view of
thoughts. Then this route to thesis 3 is lost. Whatever the nature of thoughts, the
innate constraints on thoughts must provide some of the constraints on language, as
we emphasized. But if thoughts are not language-like but, say, map-like then the
innate structure rules governing them must be very diVerent from UG-rules because
UG-rules govern languages not maps. Insofar as the innate structure of thought
constrains language that constraint would not arise from embodied UG-rules.
So, without LOTH and (t), we must look elsewhere for a place for thesis 3 in
the explanation of why the initial state respects the UG-rules. Even if (t) is
correct, we allowed that some constraints on language may come from innate
processing rules. And, doubtless, the likelihood increases that some will if (t) is
not correct because thoughts are not language-like and LOTH is false. However,
as we noted above, these sorts of initial-state processing constraints should yield
substantial commonalities among Wnal-state processing competencies which our
earlier discussion of brain impairment did not reveal (10.2). In any case, it is hard
to see how the UG-rules could be these innate processing rules for learning
language from PLD: they seem to be the wrong sort of rule; the structure-
dependency principle, for example, is simply not that sort of processing rule.
17 We could, of course, be much less tentative about the conditional proposal that if LOTH then
the UG-rules are largely, if not entirely, innate structure rules of thought.
18 I owe these references to Cowie 1999: 276–81.
This point is, of course, analogous to that of my third tentative proposal against
explaining language use in terms of embodied structure rules (11.4). In sum, if
(t) is correct then 3 provides an explanation of why we are innately predisposed
to learn languages that respect the UG-rules. But if LOTH and hence (t) are not
correct the explanation fails. And it is hard to see how 3 could be otherwise
justiWed. So we would have no good reason for thinking that the UG-rules are
innately embodied in a robust way. We have arrived at my seventh tentative
proposal: if LOTH is false, then the UG-rules are not, in a robust way, innate in a
speaker.19 This is the initial-state analogue of the Wnal-state Wfth tentative
proposal: if LOTH is false, then the rules of a language are not, in a robust
way, psychologically real in a speaker (11.9).
If (t) is false, what then is the likely shape of a future explanation of why
the initial state respects the UG-rules? The explanation must be in terms of the
innate structure of thought and innate processing rules for converting thoughts
into language and vice versa. That much is clear. But, absent some theory of the
nature of thoughts—for example, that they are like maps—we can barely say
anything more. Only against a background theory of thought can we assess the
plausibility of explanations of why the UG-rules are respected; cf. my fourth
major conclusion.
In sum, given thesis 2 and position (t), we have good reason to suppose that
something close to thesis 3 will explain why the UG-rules are respected and
hence that something close to thesis 3 will be a core part of a persuasive future
explanation of language acquisition: sixth tentative proposal. But (t), based on
LOTH, is highly speculative. Without it, thesis 3 is unlikely to be part of that
future: seventh tentative proposal. And without it, we will need an alternative
theory of thought to predict that future.
12.5 T H E L A N G UA G E FA C U LT Y ( T H E S I S 1 )
In Part IV I noted that the view that language expresses thought (LET) alone
downplays the language faculty because according to LET a certain conceptual
competence is an important part of a person’s competence in a language. That
part is in the central processor not a language faculty. So, if there is a language
faculty it must be found in the other part of a linguistic competence, processing
competencies. Finding this then depends on the various processing competencies
having a substantial part in common. For, a set of modality-speciWc processing
19 The qualiWcation, ‘‘in a robust way’’, is to allow for the possibility, analogous to the one noted
earlier for the grammar (4.6, 11.9), of a less robust intermediate view on this innateness issue: UG
might be descriptive of the initial state in a way stronger than simply positing rules that are respected
by that state—thesis 2—but weaker than positing rules that are literally embodied in that state –
thesis 3.
competencies does not a language faculty make: the faculty must be a relatively
central, modality-neutral, cognitive system partly responsible for all language
processing. Yet the evidence from brain impairments suggests that processing
competencies do not have the required commonality and hence that there is no
language faculty (10.2). Finally, I argued that if we wed the Gricean explanatory
priority of thought over language to LOTH yielding (t), we should not expect
the commonality that would constitute a substantial language faculty because
there would be nothing much for the language faculty to do. This led to my
second tentative proposal: there is little or nothing to the language faculty (10.3).
I shall start this section by showing how the present discussion gives further
support to the second tentative proposal by casting doubt on the thesis 1 view
that the initial state is a language-speciWc learning device in a distinct language
faculty. I shall then respond to two objections to the second proposal. The Wrst
objection arises from the apparently sui generis nature of language acquisition.
The second is a well-known argument of Chomsky’s for placing the initial and
Wnal state of linguistic competence in a distinct language faculty.
I have emphasized that, whatever the nature of thoughts, there must be
innate constraints on them, and those constraints must provide some of the
constraints on language acquisition. Insofar as they do, the initial state for
language is not in a language faculty, a language-speciWc learning device, but in
the central processor, the general learning device. For, the constraints are on
thoughts and thoughts are in the central processor. Of course, the language-
constraining rules may not be entirely there: some of them may be innate
processing rules for language. If these were appropriately central and modality-
neutral they would constitute a language faculty. If thoughts were not lan-
guage-like but rather, say, map-like, then we might expect that there would be
such innate constraining rules. For, the task of moving from innate map-like
thoughts to language seems formidable enough to need some innate help. So
we might expect that innate processing rules, along with the innate constraints
on thought, would explain why the initial state respects the UG-rules. So we
might expect a good part of the initial state to be in a language faculty. But
these expectations are undermined by something we have already noted twice:
if there were such innate processing rules, they should give rise to Wnal-state
commonalities among processing competencies that the literature on brain
impairment does not conWrm. Furthermore, the expectations disappear if we
adopt LOTH and hence (t). For then, I have just argued, the best explanation
of why the initial state respects the UG-rules is that those rules are largely
innate structure rules of Mentalese. (t) thus encourages the idea that the initial
state is entirely in the central processor.
All in all, I think we have good reason to believe that the main constraints
on language will come from the innate structure of thought. Steedman goes a
bit further: ‘‘the only plausible source for the innate component lies in the
conceptual structure with which the child comes to language learning, and
which either evolved or was learned for more general cognitive purposes’’
(2003: 770).20
This idea is supported by some recent psycholinguistic research that gives a
central role to the central processor in language acquisition. Dedre Gentner and
Laura Namy (2005) sum up the situation as follows:
Current theories of word learning, and of language acquisition more generally, have
turned increasingly toward domain-general cognitive and social explanations of children’s
acquisition of language . . . There is increasing support for the idea that general learning
mechanisms, guided by social-interactional knowledge, operate in encoding and processing
both the incoming stream of language and the informational structure of the environment.
These mechanisms appear to facilitate all aspects of language learning—speech segmenta-
tion, word learning, and perhaps even the acquisition of grammar. (p. 533)
The emphasis of their research is very much on the acquisition of words. Based
on research on analogy and similarity they propose that the process of compari-
son—structural alignment and mapping—facilitates lexical acquisition. They
Wnd persuasive evidence that it does. They wonder whether this process might
also ‘‘contribute to children’s ability to derive grammatical regularities’’. They
claim that ‘‘some intriguing lines of evidence suggest that this may be a direction
worth pursuing’’ (p. 557).
It is time now to consider objections to the proposal that there is little or
nothing to the language faculty. First, I noted that evidence of the sui generis
nature of language learning might seem to support thesis 1 and commitment to a
language faculty (12.2). Language acquisition is alleged to be quite unlike the
acquisition of other cognitive skills: it is done very young in the critical period; all
normal children do it; and they acquire elements of linguistic capacity in the
same order. But once one has accepted, as I already have (8.6), that the Gricean
picture allows any amount of causal inXuence of language on the capacity to have
certain thoughts, this evidence no longer counts in favor of the language faculty.
For, the evidence is then quite compatible with the view that not only is
competence in a language normally acquired in the critical period in a con-
strained uniform way under the inXuence of that language but so also is the
competence to think the thoughts expressible in the language. On this view, the
acquisition of one cognitive skill is not unlike the acquisition of language: the skill
20 Steedman does not note, however, that this view lessens the case for a language faculty. Indeed
he writes, surprisingly, as if the view is implicit in Chomsky. Steedman is also strikingly noncom-
mittal on the extent to which the structure of thought is innate:
The question of how much of [the grounded conceptual structure upon which both reasoning about
the world and development of language depend] is actively learned by the individual prelinguistic
child, and how much of it has been compiled into heritable ‘hard-wired’ components during the
process of evolution of humans and their animal ancestors, and the question of what further
apparatus is needed for the development of language and whether its origins can also be traced to
more generally useful cognitive abilities, remain open. (2003: 770–1)
in question is the skill to think the thoughts the language expresses. Indeed, the
acquisition of that cognitive skill normally brings the linguistic skill with it.
(Note that the cognitive skill in question is one of having thoughts not of thinking
well with them.)
This view predicts that people who have no acquaintance with a language
during the critical period will be cognitively impaired as well as linguistically so.
And that is indeed what we Wnd as we have already noted in our discussion of
linguistic relativity (8.6), citing the famous cases of ‘‘wolf children’’ (Malson
1972) and of Genie (Curtis 1981).
Now evidence against this view could come from dissociations, during and
after the critical period, between the development of a capacity to use sentences
of a certain structure and the development of a capacity to think the thoughts
that those sentences express. Thus, on the one hand, it would be evidence against
the view if those who develop into linguistic idiot savants could not think
thoughts that their words seem to express. But there seems to be no evidence
of this, although there is of course evidence that they are not good at the process
of thinking (10.1). On the other hand, it would be evidence against the view if
there were people who, in the critical period, learnt neither a language nor the
capacity to think the thoughts that language expresses, but afterwards learnt to
think the thoughts (without, of course, learning the language). But, we should
note, it would not be clear evidence against the view if there were people who,
in the critical period, did not learn a language but did learn to think thoughts
that they could not express. For the failure to learn the language could be
attributable to some failure in processing between thought and language. In
any case, I know of no evidence of either sort that might be thought to count
against the view.21 And it is hard to know what is lost in the critical period by the
late learner. ‘‘Such ‘critical period’ or ‘sensitive period’ eVects . . . cannot by
themselves reveal just what is being lost or diminished in the late learner: This
could be some aspects of learning speciWc to language itself, general capacities for
structured cognitive learning, or some combination of the two’’ (Gleitman and
Bloom 1999: 436).
Turn next to Chomsky’s well-known argument for placing the initial and
Wnal state of linguistic competence in a distinct language faculty. It is to be
found in passages like the following, accompanying Chomsky’s presentation of
nativism:
21 The case of Genie does not supply evidence of the second sort. Genie had no signiWcant
experience of language until she was thirteen. She then acquired a ‘‘relatively well-developed
semantic ability’’ but ‘‘very little syntax or morphology.’’ Her speech consisted in ‘‘the stringing
together of content words’’ (Curtis 1981: 21). Despite this she performed quite well on a range of
cognitive tasks (pp. 21–2). There is no evidence here that she developed the capacity to think
thoughts she could not express nor even that the primitive syntax of her utterances is not replicated
in the thoughts those utterances express.
It is hardly conceivable that principles of the sort we have been considering have any
general applicability beyond the language faculty, and no general mechanisms or prin-
ciples have been proposed that shed any light on the questions we have been addressing,
to my knowledge. (Chomsky 1986: 150)
The idea is that the principles of the initial and Wnal states of competence in a
language are so peculiar that we would not expect to Wnd them anywhere but in a
distinct language faculty, in particular not in the central processor. Furthermore,
no general learning device of the sort found in the central processor could explain
the acquisition of a language. Ray JackendoV’s neat ‘‘Paradox of Language
Acquisition’’ seems to provide support for this latter claim:
If general-purpose intelligence were suYcient to extract the principles of mental gram-
mar, linguists (or psychologists or computer scientists), at least some of whom have more
than adequate general intelligence, would have discovered the principles long ago. The
fact that we are all still searching and arguing, while every normal child manages to extract
the principles unaided, suggest that the normal child is using something other than
general-purpose intelligence. (1997: 5)
If RT were correct, with the result that the Wnal state involved a set of
propositions about the language, then these arguments might seem to be based
on proper contrasts. For, the propositions about language are indeed very
diVerent from others. And the arduous and incomplete central-processor learn-
ing of these propositions by linguists would contrast strikingly with the easy and
complete learning of the propositions by children. But we have so far found no
persuasive reason to believe RT and many reasons to disbelieve it. If we abandon
it, then the contrasts are not proper and the arguments collapse.22
What are the proper contrasts then? We have noted (11.5) that learning a
language is acquiring a skill, the skill, according to the noncommittal position (a)
on competence, at matching sounds and thoughts for meaning. So we should
think of acquiring it as like acquiring other skills, particularly cognitive skills.
These skills are, primarily at least, pieces of procedural knowledge or knowledge-
how, not pieces of declarative knowledge, knowledge-that, or theory. So one
proper contrast is between the acquisition of cognitive skills and the acquisition
of theories. These acquisitions are surely by diVerent mechanisms. Indeed, many
skills, probably including linguistic competence, diVer from theories in being
acquired by implicit learning. But this alone does not show that the mechanism for
acquiring a skill is not in the general learning device. We would surely expect the
mechanism for a cognitive skill—for example, playing chess or adding—to be in
that device. And, if position (t) is right, that is where the mechanism for language
learning may largely be because there may not be much to a speaker’s ability with a
22 Similarly Chomsky’s contrast between the uniform achievement of knowledge of language and
the nonuniform achievement of knowledge of physics (1975: 144) is not proper if RT is incorrect.
language beyond her ability to think thoughts—thoughts about anything—in a
Mentalese which is governed by the rules similar to those of the language.
The acquisition of theories draws attention to another proper contrast: that
between the structure rules for thoughts and the processing rules that govern
inference, the rules that take us from thought to thought in the acquisition of
knowledge-that. I have made much of this contrast before, using the analogy of
the contrast between the formation and transformation rules of logic (2.2). But,
we note, both these sorts of rule are embodied in the general learning device.
The upshot of this is that if we do not adopt RT, a thesis for which we have so
far found no persuasive support, Chomsky’s well-known argument against
placing linguistic competence in the central processor collapses. And the central
processor is where the competence very likely is if LOTH and hence (t) are
correct: there would be little or no truth in thesis 1. Even if some other view of
thought is correct there may not be much of our innate capacity for language to
be placed in a language faculty; there may not be much truth in 1. It goes without
saying that none of this establishes that there is no language faculty, but it does
give some reasons for doubting its role and signiWcance. It supports my second
tentative proposal.23 The case for this proposal rests a bit on LOTH but does not
totally depend on it.
In Wrst making this proposal I pointed out that the contrary view that there is a
substantial language faculty should be even more tentative because the case for it is
weaker (10.3). That contrary view required that there be a substantial common-
ality among processing competencies, something for which we found no evidence.
And if (t) is true the commonality seemed unlikely. Now we see that if (t) is true,
the best explanation of why the initial state respects the UG-rules is that those rules
are largely the innate structure rules of thought and hence in the central processor
not a language faculty. Of course (t) might not be true. (t) depends very heavily on
LOTH, which is certainly speculative, but I know of no persuasive argument
against LOTH. And even if (t) were false, innate constraints on thought in the
central processor must be part of the explanation of why the initial state respects the
UG-rules (provided that there are are thoughts). All in all, we seem to lack any
signiWcant reason for believing in the language faculty. Things would look diVer-
ent, of course, were RT true. But we have found no reason to believe that it is.
My fourth major conclusion (8.3) urges us to investigate the psychological
reality underlying language from a perspective on thought. It is appropriate now
to summarize some signiWcant results of that approach. The approach yielded my
Wrst tentative proposal, based on LOTH, that a language is largely psychologic-
ally real in a speaker in that its rules are similar to the structure rules of her
thought (9.5); and it yielded my Wfth tentative proposal that if LOTH is false
23 We should note that innate phonology must be found a place in the mind and that place is
surely not in the central processor. But it alone can hardly constitute a language faculty. Indeed, in an
important respect, it seems accidental that languages are typically spoken.
then the rules of a language are not, in a robust way, psychologically real in a
speaker (11.9). These proposals concerned the Wnal state of linguistic compe-
tence. In this chapter, the approach has yielded two analogous proposals about
the initial state: my sixth tentative proposal that humans are predisposed to learn
languages that conform to the UG-rules because those rules are, largely if not
entirely, innate structure rules of thought (12.4); and my seventh tentative
proposal that if LOTH is false then the UG-rules are not, in a robust way, innate
in a speaker (12.4). Finally, in this section, the approach has yielded the con-
clusion of the argument for the second tentative proposal, which concerns both
the initial and Wnal state: there is little or nothing to the language faculty.
In this section I have cast doubt on thesis 1 which is one half of Chomskian
Nativism. In the next section I shall cast doubt on thesis 3r which is the other half.
12.6 I M P L A U S I B I L I T Y O F I - R E P R E S E N TAT I O N A L I S M
(THESIS 3R)
If we assume thesis 2 and a robust position on thoughts, position (t), I have

argued (12.4) that a persuasive future abduction concerning language acquisi-
tion is likely to support something close to thesis 3: the initial state embodies
the UG-rules because they are largely the innate structure rules of thoughts.
But without (t), the prospects for 3 in a future theory of language acquisition
seem poor. Insofar as the prospects for 3 are poor, so too are the prospects for
the I-Representationalist version of 3, 3r. So, the prospects for 3r also seem to
depend on (t). I have earlier gone along with Cowie in rejecting the only-theory-
in-town abduction for 3r: the explanation it involves is far too lacking in detail and
evidential support. But is 3r likely to be part of a persuasive future explanation? My
discussion of 3 so far has made no mention of represented rules. Is that a mistake? In
this section I shall argue that it is not. I-Representationalism is not only not now part
of a good explanation of language acquisition, it is not part of the best. And it is
unlikely to be part of a good future explanation.
A. The Wnal state of linguistic competence is a skill. So we should look to psycho-
logical studies of skills, summarized in section 11.5, for guidance about this
competence. We have already found no signiWcant support there for RT as an
account of the Wnal state. What light do those studies throw on the initial state of
competence and on the I-Representational Thesis?
Let us start with I-Representationalism. I have been unable to Wnd
any suggestion in the psychological literature on skill acquisition that innate
representations of the rules underlie the acquisition of a nonlinguistic skill.24
On the top-down production-systems view, skill acquisition starts from declara-
tive knowledge but it is not claimed that this knowledge is innate. Quite the
24 See Johnson et al. 2003, for example.

contrary.25 In any case, language acquisition seems to be a paradigm of bottom-
up implicit learning. We noted that if it really is a case of implicit learning, then
we should be particularly dubious of RT as a view of the Wnal state of lin-
guistic competence. Indeed, the literature on skills suggest that skills have a fairly
brute–causal associative nature and so does not encourage the idea that language
processing involves representations even of the syntactic and semantic pro-
perties of language let alone of the linguistic rules (11.6). If we are right to be
thus dubious of RT as a view of the Wnal state, we should be even more dubious of
the I-Representational Thesis as a view of the initial state. In sum, innate
representations of rules do not seem to underlie nonlinguistic skills. This is
bad news for the idea that they underlie linguistic skills and hence bad news
for I-Representationalism.
This is not to say, of course, that innate structures play no role in skill
acquisition. Indeed, an analogue of the boring innateness thesis for language is
obviously true for skills in general: we can only acquire them because of some
innate capacities. Perhaps some analogues of thesis 2 are true: perhaps there are
some interesting innate constraints on the sorts of skills we can learn. Doubtless,
these will typically not be as interesting as thesis 2. Finally, even if the innate basis
for one of these skills is quite rich, it will be like that for a language skill in not
being so rich that the skill is totally innate: the skill still has to be learned.26
These remarks about the innate structures for skill acquisition do not take us
very far, of course. But, at this stage, we simply do not know enough about skill
acquisition to go much further. The psychological study of skill acquisition
throws little light on the innate structures underlying skills, hence little light on
the innateness of language. If the study were advanced enough to provide good
detailed theories of the acquisition of skills like typing and adding, we might hope
that it would provide a lot of insight into language acquisition. But the study is not
that advanced. Still, the insight that the study does provide gives no support to the
idea that the acquisition of linguistic competence involves the innate representa-
tion of rules, hence no support to the I-Representational Thesis.
I noted earlier (4.6) that Bob Matthews rejects RT as an account of the Wnal
state. We might expect that he would also reject its analogue, the I-Representa-
tional Thesis, as an account of the initial state. And so he does. Indeed, he urges
that we abandon ‘‘a rationalist account of language learning’’ (1980: 25); the
constraints ‘‘on the form of acceptable grammars . . . are so severe as to render
grossly inappropriate any characterization of the acquisition process in terms of
hypothesis testing. Data serve not so much as evidence for a hypothesized gram-
mar as they serve as a series of triggering events that determine the course of
25 Indeed, it would be preposterous to claim that the declarative knowledge involved in favorite
production-systems examples like learning to use a stick shift was innate.
26 Mons-Williams, Tresilian, and Wann 2003 (p. 125) takes it as deWnitive of a skill that it is
learned—a pointless stipulation in my view (11.5)—and Carlson 2003 (p. 36), that it is improved
by practice.
grammatical development’’ (p. 26). I think that the literature on skill acquisition
encourages this idea of language learning as a fairly brute–causal matter.27
B. Normal mature humans have two very impressive cognitive skills. A consider-
ation of the acquisition of these skills counts further against I-Representationalism
and throws more light on the initial state for language acquisition. The skills in
question are: (i) the skill of having complicated thoughts covering a limitless range of
subjects; (ii) the skill of thinking, of moving in a somewhat rational manner from
one thought to another.
(i) Assume LOTH for a moment. Then our skill at having thoughts has the
following two parts: the syntactic skill of combining mental words of the various
syntactic categories into mental sentences; and the skill of having the mental
words, the concepts, that Wt into those sentences (11.3). We have very good
reason to suppose that some of the syntax of thought is innate (12.4). Still,
nobody would suggest that what is innate is a representation of syntactic rules. A
version of Harman’s objection (6.2) would loom: the suggestion would lead
either to a vicious circle or to an inWnite regress. And even those who believe, as
Jerry Fodor does (1981), that our concepts are largely innate will not hold that we
innately represent the rules of the concept; for example, that <echidna> means
ECHIDNA. So, our skill at having thoughts is an uncontroversial example of one
that does not involve any innate representation of its rules.
Now, these remarks assume LOTH. But the conclusion would be just the same
if we assumed only the Representational Theory of Mind (RTM) and left the
nature of mental representations open. Thus, if those representations are map-
like, their structure rules will still not be innately represented.
(ii) Turn next to our skill at thinking, at moving from one thought to another.
I think it plausible that some of the rules governing this process are learnt: there is
a bit of practice making better. But one thing we can surely be conWdent about is
that many of these rules are innate. How else could we even get started in forming
our world view? And another thing we can surely be conWdent about is that these
innate ‘‘laws of thought’’ are not represented. Our initial thinking is not governed
by some higher-level thoughts about thinking (3.1, 7.3). Our skill at thinking is
another uncontroversial example of one that does not involve innate representa-
tions of its rules.
(i) and (ii) then serve as nice examples of cognitive skills without innate
representations of rules. But they do more. I have emphasized that, whatever
one’s view of thoughts, the innate constraints on their structure are one part, at
27 On this score, it is worth noting that whereas Chomsky’s early hypothesis-testing theory of
language acquisition (1965) may seem to demand the representation of innate rules, the later
parameter-setting theory (1982) does not. As Stabler points out, this theory ‘‘appears to be
particularly amenable to the view that the grammar is not represented’’ (1983: 398). The theory
leads Collins to conclude: ‘‘There is nothing rational about acquiring language’’ (2004: 512).
Indeed, the idea of parameter setting as ‘‘triggering’’ suggests a fairly brute–causal process.
least, of the innate constraints on language. This leaves open the possibility that
innate constraints on language processing may be another part. The literature on
brain impairment and position (t) led me to the conclusion that this possibility
was not likely (12.5). Still, it remains a possibility and so let us suppose, for the
sake of argument, that it is actual. Our discussion of (i) bears on the part of
the innate linguistic constraints concerning structure, our discussion of (ii), on the
part concerning processing. In discussing (i), we saw that the innate constraints
on thought do not involve innate representations of rules. So, to that extent,
I-Representationalism is false: to that extent the initial state does not respect the
UG-rules because it represents them. It still might be the case, of course, that any
further innate constraints that there may be on language processing are in the
form of represented rules. Thus, adopting LOTH, it might be the case that there
were innate Mentalese representations of these further constraints. But our
discussion of (ii) provides a more inviting model: just as our thinking is partly
governed by innate processing rules that are embodied without being represented
so too is our language use. There seems to be no basis for insisting that the
rules governing language use are represented. This is a time to apply Pylyshyn’s
Razor.
In conclusion, a comparison of linguistic competence with other skills and
attention to the nature of thoughts and thinking suggest that we should abandon
the I-Representationalist thesis 3r. Assuming position (t) and the interesting
nativist thesis 2—the thesis that the initial state respects the UG-rules—we have
seen that the prospects for thesis 3—the thesis that the initial state embodies
those rules—look good: sixth tentative proposal (12.4). But the evidence counts
against going further to 3r. And without (t), the prospects for even 3 look bleak:
seventh tentative proposal (12.4). So, whether or not (t) turns out to be true, it is
unlikely that I-Representationalism will be part of a persuasive explanation of
language acquisition. There is no signiWcant evidence for the I-Representational
Thesis and, given what else we know, it is implausible. That is my seventh major
conclusion. A human does not innately represent the UG-rules nor does she
innately know anything about those rules; she is totally ignorant of them. If
this is right then the view, much-hyped on the intellectual cocktail circuit,
that language acquisition supports the traditional doctrine of innate ideas is
mistaken.
Of course, this discussion of the I-Representational Thesis about the initial
state has proceeded without assuming RT about the Wnal state. This is appropri-
ate because we have so far found no persuasive reason for believing that thesis and
many reasons for disbelieving it. But suppose, despite this, that RT were true: the
Wnal state consists in representations of the rules of the language. Then presum-
ably such representations must be involved in all the learning steps between the
innate initial state and the Wnal state. This does not, of course, show that the
initial state involves such representations: the rules of protocol and of chess are
often represented in the mind but surely none of these representations are innate.
Perhaps the view that the initial state involves represented rules gains plausibility
if the Wnal state does. But unless we can Wnd another example of innate rep-
resentations of rules underlying a skill, that gain is not going to be great.
In the last section I cast doubt on thesis 1 and the commitment to a language
faculty. In this section I have cast doubt on thesis 3r. So I think that we should
accept neither part of Chomskian Nativism.
12.7 B E A R I N G O N T H E R E P R E S E N TAT I O N A L T H E S I S
( RT )
So far in this chapter we have been concerned with the nature of the initial state
of linguistic competence. It is time now to return to the main concern of this
book, the Wnal state of that competence. What bearing does our discussion of the
initial state have on that concern? What else can we conclude about the nature of
the Wnal state from a consideration of language acquisition?
The argument in earlier chapters strongly suggests that RT is false. If it is, I
have just argued, the I-Representational Thesis is surely false also. And it would
quite likely be false even if RT were true. So, I am conWdent about the falsity of
I-Representationalism. Let us now reverse the order of argument. What does the
falsity of I-Representationalism show about RT? It surely does not entail the
falsity of that thesis: my competence in chess may consist partly in my declarative
knowledge of its rules and yet that knowledge was surely not innate. Nonetheless,
the falsity of I-Representationalism certainly does nothing to support RT. And it
adds to the attractiveness of the view that RT is also false.
But suppose the argument of this chapter were mistaken and I-Representa-
tionalism were true. Would that show that RT was true? It seems not. In dis-
cussing skills in general (11.5) we noted that the automatic procedural stage of
a skill involving no declarative knowledge could arise out of a cognitive stage
involving such knowledge. So it seems that a skill involving no representation
of its rules can arise out of a representation of its rules. Still, it might be
objected that the cognitive stage of that sort of skill was learned not innate.
How could a skill involving no representation of its rules in the Wnal state
arise out of an innate representation of its rules in the initial state? Could the
innate representation just disappear so that it is not part of the Wnal state?
Perhaps so. The natural learning of a language has to be achieved before the
end of the critical period. So something that enables language acquisition is
clearly lost by that point. If I-Representationalism were true then presumably
what would be lost would be the innate representation of UG-rules. The mind
is ‘‘programmed’’ to acquire a language according to these representations
during the critical period and then to delete the representations. So even
I-Representationalism does not directly support RT. Still, it would surely
make RT more attractive.
In sum, although the I-Representationalist Thesis about the initial state of

linguistic competence and RT about the Wnal state are natural bed-fellows,
neither thesis provides signiWcant evidential support to the other. More import-
antly, our discussion of the innateness issue counts against the I-Representationalist
Thesis and provides the Wnal step in the long-developing argument against RT.
My seventh major conclusion is that we should abandon the former, my second,
the latter. In abandoning the former we conclude that a person in the initial state
is totally ignorant of the rules of UG. In abandoning the latter we conclude that a
person in the Wnal state can be totally ignorant of the rules of her language.
Finally, in abandoning RT we are, of course, rejecting positions (i) and (ii) on
language use. But our discussion of nativism does not seem to have any other
signiWcant bearing on positions on language use.
12.8 CONCLUSION
There are some fairly familiar arguments for linguistic nativism: arguments from
the sui generis nature of language learning, from the universality of UG-rules,
from the poverty of stimulus, from language creation, and from the continuity
hypothesis. The sui generis argument seems to support thesis 1—the innate initial
state for language acquisition is in a distinct language faculty—although I argue
that the evidence here is actually compatible with the initial state being in the
central processor (12.5). I think that the other arguments, taken jointly, present a
persuasive case for thesis 2: humans are innately predisposed to learn languages
that conform to the UG-rules; the initial state respects the UG-rules (12.2).
Can we then establish any other nativist thesis? At this stage, we are far short of
a good explanation of language acquisition that could form the basis for a sound
abduction in favor of any other thesis. So, Fodor’s only-theory-in-town abduc-
tion for the very exciting I-Representationalist thesis 3r should be dismissed, as
Cowie argues (12.3). Still, we can make some progress by seeking an explanation
of the initial state’s respect for the UG-rules. If we assume position (t)—a person
thinks in a Mentalese governed by structure rules that are similar to those of her
language—we have good reason to suppose that something close to thesis 3 will
explain that respect and hence be part of a persuasive future explanation of
language acquisition; we have good reason to suppose that the UG-rules are, for
the most part, innate structure rules of thought. This is my sixth tentative
proposal. But if LOTH and hence (t) are false, then it seems unlikely that thesis
3 will be part of the future. My seventh tentative proposal is that if LOTH is
false, then the UG-rules are not, in a robust way, innate in a speaker (12.4).
Is linguistic competence in a language faculty? Without RT, for which we have
so far found no signiWcant support, Chomsky’s well-known argument for placing
linguistic competence in such a distinct language faculty—hence, for thesis 1—
collapses. If (t) is correct that competence very likely is in the central processor,
and even if (t) is not correct that competence may be largely there. This adds to
the earlier case for my second tentative proposal that there is little or nothing to
the language faculty (10.3). The contrary view that there is a substantial language
faculty seems to lack any signiWcant evidence (12.5).
Assuming (t), we have found a case for thesis 3 as part of the future. Should we
go further to 3r? A comparison of linguistic competence with other skills and
attention to the nature of thoughts and thinking, suggests that we should not.
I-Representationalism is not only not the conclusion of a present good abduc-
tion, it is unlikely to be of a future one. My seventh and Wnal major conclusion
is that there is no signiWcant evidence for the I-Representational Thesis and,
given what else we know, it is implausible. So both parts of Chomskian Nativism,
thesis 1 and thesis 3r, are likely false.
Finally, the issue of nativism has less bearing on our view of the Wnal state of
linguistic competence than one might have expected. In Part III we found no
evidence for RT in the arguments from the rejection of behaviorism, from the
folk truism that the competent speaker knows her language, and from the role of
linguistic intuitions. Our Chapter 11 discussion of language use provided more
evidence against RT than for it. The consideration of language acquisition in this
chapter has given no support to RT. My second major conclusion is established:
there is no signiWcant evidence for RT and, given what else we know, it is
implausible.
In conclusion, in this book I have argued that a number of Chomskian views
about the psychological reality of language are ill-supported by evidence and
argument and are probably wrong. My major conclusions and tentative proposals
are to be found at the end of the following Glossary.
Glossary of Named or Numbered Items
(Parenthetical references are to the section in which an item is best explained)
NAMED THESES
The Representational Thesis (RT) A speaker of a language stands in an uncon-

scious or tacit propositional attitude to the rules or principles of the language
which are represented in her language faculty (1.1).
The Respect Constraint A theory of a competence must posit processing rules
that respect the structure rules of the outputs. Similarly, a theory of the outputs
must posit structure rules that are respected by the competence and its
processing rules (2.3).
Pylyshyn’s Razor Representations are not to be multiplied beyond necessity
(3.1).
Intentional Realism People really have thoughts (8.1).
LET Language expresses thought (8.2).
The Representational Theory of the Mind (RTM) Having a thought involves
standing in a certain functional relation to a mental representation (9.1).
The Language-of-Thought Hypothesis (LOTH) Thoughts involve language-
like mental representations (9.2).
Public-LOTH A person’s language of thought is the same as her public language
(9.3).
The I-Representational Thesis The rules specified by UG are innately repre-
sented in the mind (12.1).
DISTINCTIONS
(1) Distinguish the theory of a competence from the theory of its outputs/
products or inputs (2.1).
(2) Distinguish the structure rules governing the outputs of a competence from
the processing rules governing the exercise of the competence (2.2).
(3) Distinguish the respecting of structure rules by processing rules from the
inclusion of structure rules among processing rules (2.3).
(4) Distinguish processing rules that govern by being represented and applied
from ones that are simply embodied without being represented (3.1).
(5) Distinguish the representations of structure rules used as data in processing
from the representations of structure rules that are applied in processing (3.2).
274 Glossary
(6) Distinguish processing rules that govern a ‘‘rational-causal’’ operation on
syntactically structured representations (or symbols) from ones that govern
‘‘brute-causal’’ operations that may or may not involve representations (3.3).
(7) Distinguish actual from merely metaphorical generation, computation, and
processing (4.2).
(8) Distinguish the generation of expressions by the I-language from the gener-
ation of structural descriptions of expressions by the grammar (4.2).
M E T H O D O L O G I C A L P O I N TS
First There is something theoretically interesting for a grammar to be true about

other than the psychological reality of speakers: it can be true about a linguistic
reality (2.6).
Second The view that a grammar has any more to do with psychological reality
than that it must posit structure rules that are respected by the competence
requires a powerful psychological assumption about competence (2.6).
Third The Respect Constraint makes the justification of the grammar partly
dependent on the justification of the theory of competence, and vice versa.
Beyond that, however, the grammar and the theory of competence are inde-
pendent of each other (2.6).
Fourth A grammar as a theory of a language has a certain epistemic and
explanatory priority over a theory of the psychological reality underlying
language (2.6).
SOME POSSIBLE POSITIONS ON

P S YC H O LO G I C A L R E A L I T Y
(m) A competence in a language, and the processing rules that govern its exercise,
respect the structure rules of the language: the processing rules of language
comprehension take sentences of the language as inputs; the processing rules of
language production yield sentences of the language as outputs (3.4).
(i) The structure rules of the language are also processing rules that are repre-
sented in the speaker and applied in language use (3.4).
(ii) The structure rules of the language are represented and used as data by the
processing rules of language use (3.4).
(iii) The structure rules of the language are also processing rules for language use
but the processing rules are not represented (3.4).
(iv) Some processing rules for language use are represented but they are largely
unlike the structure rules of the language and do not use the structure rules as
data (3.4).
Glossary 275
structure rules of the language (3.4).
Versions (a) Some processing rules operate on metalinguistic representations of
the syntactic and semantic properties of linguistic items and there is a rational
information flow (3.4).
Versions (b) Processing rules do not operate on metalinguistic representations of
the syntactic and semantic properties of linguistic items but are directly
responsive, in a fairly brute–causal associationist way, to these properties (3.4).
(t) The structure rules of a speaker’s language are similar to the structure rules of
her thought (9.5).
V I E W S O F L I N G U I S T I C C O M PE T E N C E
(a) Assuming LET: the ability to use a sound of the language to express a thought
with the meaning that the sound has in the language in the context of
utterance; and the ability (together with some pragmatic abilities) to assign
to a sound a thought with the meaning that the sound has in the language in
the context of utterance (similarly for inscriptions, etc.) (8.3).
(b) Assuming RTM: the ability to translate back and forth between mental
representations and the sounds of the language (9.1).
(c) Assuming LOTH: the ability to translate back and forth between mental
sentences and the sounds of the language (9.3).
N U M B E R E D I N N AT E N E S S T H E S E S
1 The initial state is a language-specific learning device in a distinct ‘‘language

faculty’’ not in the central processor (12.1).
2 The initial state constrains humans to learn only languages that conform to
UG; it respects the UG-rules (12.1).
3 The initial state respects the UG-rules because it embodies the UG-rules
(12.1).
3R (The I-Representational Thesis above) The UG-rules are innately represented
in the mind (12.1).
Chomskian Nativism Theses 1 and 3R (12.1).
MAJOR CONCLUSIONS
First Linguistics is not part of psychology (2.8).

Second There is no significant evidence for the Representational Thesis (RT)
and, given what else we know, it is implausible (4.7).
276 Glossary
Third Speakers’ linguistic intuitions do not reflect information supplied by the
language faculty. They are immediate and fairly unreflective empirical central-
processor responses to linguistic phenomena. They are not the main evidence
for grammars (7.8).
Fourth The psychological reality of language should be investigated from a
perspective on thought (8.3).
Fifth Thought has a certain priority to language ontologically, explanatorily,
temporally, and in theoretical interest (8.8).
Sixth The primary concern in linguistics should not be with idiolects but with
linguistic expressions that share meanings in idiolects (10.5).
Seventh There is no significant evidence for the I-Representational Thesis and,
given what else we know, it is implausible (12.6).
T E N TAT I V E P RO P O S A L S
First (based on position (t) above) A language is largely psychologically real in a

speaker in that its rules are similar to the structure rules of her thought (9.5).
Second There is little or nothing to the language faculty (10.3).
Third The structure rules of a language are the wrong sort of rule to govern the
process of language use (11.4).
Fourth The speedy automatic language processes arising wholly, or at least partly,
from linguistic competence are fairly brute–causal associationist processes that
do not operate on metalinguistic representations of the syntactic and semantic
properties of linguistic expressions (11.6).
Fifth If LOTH is false, then the rules of a language are not, in a robust way,
psychologically real in a speaker (11.9).
Sixth Humans are predisposed to learn languages that conform to the rules
specified by UG because those rules are, largely if not entirely, innate structure
rules of thought (12.4).
Seventh If LOTH is false, then the rules specified by UG are not, in a robust way,
innate in a speaker (12.4).
References
Allen, C., and M. Bekoff. 1997. Species of Mind: The Philosophy and Biology of Cognitive
Ethology. Cambridge, MA: MIT Press.
Anderson, John R. 1980. Cognitive Psychology and its Implications. San Francisco: W. H.
Freeman and Company.
—— 1983. The Architecture of Cognition. Cambridge, MA: Harvard University Press.
—— 1993. Rules of the Mind. Hillsdale, NJ: Lawrence Erlbaum Associates.
—— and Christian Lebiere. 1998. The Atomic Components of Thought. Mahwah, NJ:
Lawrence Erlbaum Associates.
Antony, Louise. 2003. ‘‘Rabbit Pots and Supernovas: On the Relevance of Psychological
Data to Linguistic Theory’’. In Barber 2003a: 47–68.
—— and Norbert Hornstein, eds. 2003. Chomsky and His Critics. Oxford: Blackwell
Publishers.
Armstrong, D.M. 1971. ‘‘Meaning and Communication’’. Philosophical Review 80:
427–47.
—— 1973. Belief, Truth and Knowledge. Cambridge: Cambridge University Press.
Atkinson, Antony P., and Michael Wheeler, 2004. ‘‘The Grain of Domains: The
Evolutionary–Psychological Case against Domain-General Cognition’’. Mind and
Language 19: 147–76.
Au, T. K. 1983. ‘‘Chinese and English Counterfactuals: The Sapir–Whorf Hypothesis
Revisited’’. Cognition 15: 155–87.
Avramides, A. 1989. Meaning and Mind, An Examination of a Gricean Account of
Language. Cambridge, MA: MIT Press.
Bach, Kent. 1987. Thought and Reference. Oxford: Clarendon Press.
Baker, C. L. 1978. Introduction to Generative–Transformational Syntax. Englewood Cliffs,
NJ: Prentice-Hall, Inc.
—— 1995. English Syntax, 2nd edn. Cambridge, MA: MIT Press. 1st edn, 1989.
—— and J. J. McCarthy, eds. 1981. The Logical Problem of Language Acquisition.
Cambridge, MA.: MIT Press.
Baker, Lynne Rudder. 1987. Saving Belief: A Critique of Physicalism. Princeton: Princeton
University Press.
Barber, Alex. 2001. ‘‘Idiolectal Error’’. Mind and Language 16: 263–83.
—— (ed.). 2003a. Epistemology of Language. Oxford: Oxford University Press.
—— 2003b. ‘‘Introduction’’. In Barber 2003a: 1–43.
Barkow, J., L. Cosmides, and J. Tooby, eds. 1992. The Adapted Mind. Oxford: Oxford
University Press.
Baron-Cohen, S. 1995. Mindblindness: An Essay on Autism and Theory of Mind. Cam-
bridge, MA: MIT Press.
Bealer, George. 1998. ‘‘Intuition and the Autonomy of Philosophy’’. In DePaul and
Ramsey 1998: 201–39.
Bennett, J. 1976. Linguistic Behaviour. Cambridge: Cambridge University Press.
Berkeley, George. 1710. Principles of Human Knowledge
278 References
Berwick, R. C. 1983. ‘‘Using What You Know: A Computer-Science Perspective’’.
Behavioral and Brain Sciences 3: 402–3.
—— and A. S. Weinberg. 1984. The Grammatical Basis of Linguistic Performance:
Language use and Acquisition. Cambridge, MA.: MIT Press.
Bever, T. G., J. M. Carroll, and L. A. Miller, eds. 1984. Talking Minds: The Study of
Language in Cognitive Science. Cambridge, MA.: MIT Press.
Bickerton, Derek. 1990. Language and Species. Chicago: University of Chicago Press.
—— 1995. Language and Human Behaviour. Seattle: University of Washington Press.
Bjorklund, David F., Wolfgang Sneider, and Carlos Hernandez Blasi. 2003. ‘‘Memory’’.
In Nadel 2003, vol. 2: 1059–65.
Block, Ned, ed. 1981. Readings in Philosophy of Psychology, Volume 2. Cambridge, MA.:
Harvard University Press.
Bloom, A. H. 1981. The Linguistic Shaping of Thought: A Study of the Impact of Language
on Thinking in China and the West. Hillsdale, NJ: Erlbaum.
Blumstein, Sheila E. 1999. ‘‘Phonology, Neural Basis of ’’. In Wilson and Keil 1999:
643–5.
Bock, Kathryn. 1995. ‘‘Sentence Production: From Mind to Mouth’’. In Miller and
Eimas 1995: 181–216.
—— 1999. ‘‘Language Production’’. In Wilson and Keil 1999: 453–6.
Boden, M. A. 1984. ‘‘Animal Perception from an Artificial Intelligence Viewpoint’’. In
Minds, Machines and Evolution, ed. C. Hookway. Cambridge: Cambridge University
Press.
—— 1988. Computational Models of the Mind. New York: Cambridge University
Press.
Boghossian, Paul A. 1990a. ‘‘The Status of Content’’. Philosophical Review 99: 157–84.
—— 1990b. ‘‘The Status of Content Revisited’’. Pacific Philosophical Quarterly 71:
264–78.
BonJour, Laurence. 1998. In Defense of Pure Reason: A Rationalist Account of A Priori
Justification. Cambridge: Cambridge University Press.
—— 2005a. ‘‘In Defense of A Priori Reasons’’. In Steup and Sosa 2005: 98–105.
—— 2005b. ‘‘Reply to Devitt’’. In Steup and Sosa 2005: 115–18.
—— 2005c. ‘‘Last Rejoinder’’. In Steup and Sosa 2005: 120–2.
Boroditsky, Lera. 2003. ‘‘Linguistic Relativity’’. In Nadel 2003, vol. 2: 917–21.
Botha, Rudolf P. 2003. Unravelling the Evolution of Language. Amsterdam: Elsevier.
Braddon-Mitchell, David, and Frank Jackson. 1996. Philosophy of Mind and Cognition.
Oxford: Blackwell Publishers.
Bresnan, Joan. 1978. ‘‘A Realistic Transformational Grammar’’. In Halle, Bresnan, and
Miller 1978: 1–59.
—— ed. 1982. The Mental Representation of Grammatical Relations. Cambridge, MA:
MIT Press.
—— and Ronald Kaplan. 1982. ‘‘Introduction: Grammars as Mental Representations of
Language’’. In Bresnan 1982: xvii–lii.
Broadbent, Donald E. 1977. ‘‘Levels, Hierarchies, and the Locus of Control’’. Quarterly
Journal of Experimental Psychology 29: 181–201.
—— and B. Aston. 1978. ‘‘Human Control of a Simulated Economic System’’. Ergo-
nomics 21: 1035–43.
References 279
—— Peter FitzGerald, and Margaret H. P. Broadbent. 1986. ‘‘Implicit and Explicit
Knowledge in the Control of Complex Systems’’. British Journal of Psychology 77:
33–50.
Brown, Liana E., and David A. Rosenbaum. 2003. ‘‘Motor Control: Models’’. In Nadel
2003, vol. 3: 127–33.
Burge, Tyler. 1979. ‘‘Individualism and the Mental’’. Midwest Studies in Philosophy,
Volume IV: Studies in Metaphysics, eds. Peter A. French, Theodore E. Uehling Jr.,
and Howard K. Wettstein: 73–121.
Burton-Roberts, Noel, Philip Carr, and Gerard Docherty. 2000. ‘‘Introduction’’. In
Phonological Knowledge: Conceptual and Empirical Issues, eds. Noel Burton-Roberts,
Philip Carr, and Gerard Docherty. Oxford: Oxford University Press: 1–18.
Caramazza, Alfonso. 1997. ‘‘How Many Levels of Processing are there in Lexical Access?’’
Cognitive Neuropsychology 14: 177–208.
—— 1999. ‘‘Lexicon, Neural Basis of ’’. In Wilson and Keil 1999: 469–71.
Carey, Susan. 1985. Conceptual Change in Childhood. Cambridge, MA: MIT Press.
Carlson, Richard A. 2003. ‘‘Skill Learning’’. In Nadel 2003, vol. 4: 36–42.
Carroll, Lewis. 1895. ‘‘What the Tortoise Said to Achilles’’. Mind 4: 278–80.
Carruthers, Peter. 1996. Language, Thought and Consciousness: An Essay in Philosophical
Psychology. Cambridge: Cambridge University Press.
—— 1998. ‘‘Thinking in Language? Evolution and a Modularist Possibility’’. In Car-
ruthers and Boucher 1998: 94–119.
—— 2002. ‘‘The Cognitive Functions of Language’’. Behavioral and Brain Sciences 25:
657–74.
—— and Jill Boucher, eds. 1998a. Language and Thought: Interdisciplinary Themes.
Cambridge: Cambridge University Press.
—— and Jill Boucher. 1998b. ‘‘Introduction: Opening Up Options’’. In Carruthers and
Boucher 1998a: 1–18.
Chierchia, Gennaro. 1999. ‘‘Linguistics and Language’’. In Wilson and Keil 1999:
xci–cix.
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton and Co.
—— 1959. Review of Skinner 1957. Language 35: 26–58.
—— 1965. Aspects of the Theory of Syntax. Cambridge, MA.: MIT Press.
—— 1966. Cartesian Linguistics: A Chapter in the History of Rationalist Thought. New
York: Harper & Rowe.
—— 1969a. ‘‘Linguistics and Philosophy’’. In Hook 1969: 51–94.
—— 1969b. ‘‘Comments on Harman’s Reply’’. In Hook 1969: 152–9.
—— 1972. Language and Mind. Enlarged edn. New York: Harcourt Brace Jovanovich.
—— 1975a. ‘‘Knowledge of Language’’. In Gunderson 1975: 299–320.
—— 1975b. Reflections on Language. New York: Pantheon Books.
—— 1975c. The Logical Structure of Linguistic Theory. New York: Plenum Press.
—— 1980a. Rules and Representations. New York: Columbia University Press.
—— 1980b. ‘‘Rules and Representations’’. Behavioral and Brain Sciences 3: 1–14.
—— 1980c. ‘‘Author’s Response’’ to peer commentary on 1980b. Behavioral and Brain
Sciences 3: 42–58.
—— 1982. Some Concepts and Consequences of the Theory of Government and Binding.
Cambridge, MA.: MIT Press.
280 References
Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. New York:
Praeger Publishers.
—— 1988. Language and Problems of Knowledge: The Managua Lectures. Cambridge,
MA: MIT Press.
—— 1991a. ‘‘Linguistics and Adjacent Fields: A Personal View’’. In Kasher 1991a: 3–25.
—— 1991b. ‘‘Linguistics and Cognitive Science: Problems and Mysteries’’. In Kasher
1991a: 26–53.
—— 1993a. ‘‘Mental Constructions and Social Reality’’. In Knowledge and Language
Volume I: From Orwell’s Problem to Plato’s Problem, eds. Eric Reuland and Werner
Abraham. Dordrecht: Kluwer Academic Publishers: 29–58.
—— 1993b. ‘‘A Minimalist Program for Linguistic Theory’’. In The View from Building
20: Essays in Honor of Sylvain Bromberger, eds. Kenneth Hale and Samuel Jay Keyser.
Cambridge, MA: MIT Press.
—— 1994a. Language and Thought. London: Moyer Bell.
—— 1995a. The Minimalist Program. Cambridge, MA: MIT Press.
—— 1995b. ‘‘Language and Nature’’. Mind 104: 1–61.
—— 1995c. ‘‘Base Phrase Structure’’. In Webelhuth 1995a: 383–439.
—— 1996. Powers and Prospects: Reflections on Human Nature and the Social Order.
Boston: South End Press.
—— 2000a. New Horizons in the Study of Language and Mind. Cambridge: Cambridge
University Press.
—— 2000b. The Architecture of Language. New Delhi: Oxford University Press.
—— 2003. ‘‘Reply to Rey’’. In Antony and Hornstein 2003: 274–87.
Churchland, Patricia S. 1986. Neurophilosophy: Toward a Unified Science of the Mind–
Brain. Cambridge, MA: MIT Press.
Churchland, Paul M. 1981. ‘‘Eliminative Materialism and the Propositional Attitudes’’.
Journal of Philosophy 78: 67–90. Reprinted in Churchland 1989: 1–22.
—— 1989. A Neurocomputational Perspective: The Nature of Mind and the Structure of
Science. Cambridge, MA: MIT Press.
Clark, Andy. 1998. ‘‘Magic Words: How Language Augments Human Computation’’. In
Carruthers and Boucher 1998a: 162–83.
Clark, R. 1994. ‘‘Finitude, Boundedness, and Complexity’’. In Syntactic Theory and First
Language Acquisition: Cross-Linguistic Perspectives. Vol. 2: Binding, Dependencies, and
Learnability, eds B. Lust, G. Hermon, and J. Kornfilt. Hillsdale, NJ: Lawrence
Erlbaum: 473–89.
Cleeremans, Axel. 2003. ‘‘Implicit Learning Models’’. In Nadel 2003, vol. 2: 491–9.
Collins, John. 2003. ‘‘Cowie on the Poverty of the Stimulus’’. Synthese 136: 159–90.
—— 2004. ‘‘Faculty Disputes’’. Mind and Language 19: 503–33.
Corina, David P. ‘‘Sign Language and the Brain’’. In Wilson and Keil 1999: 756–8.
Cowie, Fiona. 1998. What’s Within: Nativism Reconsidered. New York: Oxford University
Press.
Crain, Stephen, Andrea Gualmini, and Paul Pietroski. 2005. ‘‘Brass Tacks in Linguistic
Theory: Innate Grammatical Principles’’. To appear in a volume produced by the
Sheffield Innateness Project.
—— and Paul Pietroski. 2001. ‘‘Nature, Nurture and Universal Grammar’’. Linguistics
and Philosophy 24: 139–86.
References 281
—— and Paul Pietroski. 2002. ‘‘Why Language Acquisition is a Snap’’. The Linguistic
Review 19: 163–83.
—— and Mark Steedman. 1985. ‘‘On Not Being Led up the Garden Path: The Use of
Context by the Psychological Syntax Processor’’. In Natural Language Parsing: Psycho-
logical, Computational, and Theoretical Perspectives, eds. David R. Dowty, Lauri Kart-
tunen, and Arnold M. Zwicky. Cambridge: Cambridge University Press.
—— and Rosalind Thornton. 1998. Investigations in Universal Grammar: A Guide to
Experiments on the Acquisition of Syntax and Semantics. Cambridge, MA: MIT Press.
Croft, William, and D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge
University Press.
Culicover, Peter W. 1999. Syntactic Nuts: Hard Cases, Syntactic Theory, and Language
Acquisition. Oxford: Oxford University Press.
Cummins, Robert. 1996. Representations, Targets, and Attitudes. Cambridge, MA: MIT
Press.
—— and Robert M. Harnish. 1980. ‘‘The Language Faculty and the Interpretation of
Linguistics’’. Behavioral and Brain Sciences 3: 18–19.
Curtis, Susan. 1981. ‘‘Dissociations Between Language and Cognition: Cases and Impli-
cations’’. Journal of Autism and Developmental Disorders, 11: 15–30.
Cutler, Anne. 1995. ‘‘Spoken Word Recognition and Production’’. In Miller and Eimas
1995: 97–136.
—— 1999. ‘‘Spoken-Word Recognition’’. In Wilson and Keil 1999: 796–8.
—— and Dennis Norris. 1999. ‘‘Sharpening Ockham’s Razor’’. Behavioral and Brain
Davidson, Donald. 1980. Essays on Actions and Events. Oxford: Clarendon Press.
—— 1984. Inquiries into Truth and Interpretation: Oxford: Clarendon Press.
—— 1986. ‘‘A Nice Derangement of Epitaphs’’. In Lepore 1986: 433–46.
Davies, Martin. 1987. ‘‘Tacit Knowledge and Semantic Theory: Can a Five per cent
Difference Matter?’’ Mind 96: 441–62.
—— 1989. ‘‘Tacit Knowledge and Subdoxastic States’’. In George 1989a: 131–52.
—— 1995. ‘‘Two Notions of Implicit Rules’’. In Philosophical Perspectives, 9, AI, Con-
nectionism, and Philosophical Psychology, 1995, ed. James E. Tomberlin. Cambridge
MA: Blackwell Publishers:
—— 1998. ‘‘Language, Thought, and the Language of Thought (Aunty’s Own Argu-
ment Revisited)’’. In Carruthers and Boucher 1998: 226–47.
—— and Tony Stone. 1995. Folk Psychology: The Theory of Mind Debate. Oxford:
Blackwell Publishers.
Davis, Martin. 1983. ‘‘Church’s Thesis and Representation of Grammars’’. Behavioral
and Brain Sciences 3: 404.
Dehaene, S. 1997. The Number Sense: How the Mind Creates Mathematics. New York:
Oxford Universtiy Press.
Dell, Gary S. 1995. ‘‘Speaking and Misspeaking’’. In Gleitman and Liberman 1995: 83–208.
—— M. F. Schwartz, N. Martin, E. M. Saffran, and D. A. Gagnon. 1997. ‘‘Lexical
Access in Aphasic and Nonaphasic Speakers’’. Psychological Review 104: 801–38.
—— , Franklin Chang, and Zenzi M. Griffin. 1999. ‘‘Connectionist Models of Language
Production: Lexical Access and Grammatical Encoding’’. Cognitive Science 23: 517–42.
282 References
Demopoulos, William, and Robert J. Matthews. 1983. ‘‘On the Hypothesis that Gram-
mars Are Mentally Represented’’. Behavioral and Brain Sciences 3: 405–6.
Dennett, Daniel. 1978. Brainstorms. Cambridge, MA: Bradford Books.
—— 1980. ‘‘Passing the Buck to Biology’’. Behavioral and Brain Sciences 3: 19–20.
DePaul, Michael R., and William Ramsey, ed. 1998. Rethinking Intuition. Lanham:
Rowman & Littlefield Publishers, Inc.
Devitt, Michael. 1981. Designation. New York: Columbia University Press.
—— 1990. ‘‘Transcendentalism about Content’’. Pacific Philosophical Quarterly 71:
247–63.
—— 1994. ‘‘The Methodology of Naturalistic Semantics’’. Journal of Philosophy, 91
(1994): 545–72.
—— 1996. Coming to Our Senses. Cambridge: Cambridge University Press.
—— 1997a. Realism and Truth. 2nd edn with new afterword. Princeton: Princeton
University Press.
—— 1997b. ‘‘Responses to the Maribor Papers’’. In Jutronic 1998: 353–411.
—— 1998a. ‘‘Putting Metaphysics First: A Response to James Tomberlin’’. In
Philosophical Perspectives, 12, Language, Mind, and Ontology, 1998. James E. Tomber-
lin, ed. Cambridge MA: Blackwell Publishers (1998), pp. 499–502.
—— 1998b. ‘‘Naturalism and the A Priori’’. Philosophical Studies, 92, 45–65.
—— 2001. ‘‘Sustaining Actualism’’. In Philosophical Perspectives, 15, Metaphysics, 2001,
ed. James E. Tomberlin. Cambridge MA: Blackwell Publishers: 415–19.
—— 2002. ‘‘Deference: A Truth in Linguistic Relativity’’. Paper delivered at the annual
conference of the Society for Philosophy and Psychology, Edmonton, June 2002.
—— 2003. ‘‘Linguistics is not Psychology’’. In Barber 2003a: 107–39.
—— 2005a. ‘‘There is no A Priori’’. In Steup and Sosa 2005: 105–15.
—— 2005b. ‘‘Reply to BonJour’’. In Steup and Sosa 2005: 118–20.
—— 2005c. ‘‘Scientific Realism’’. In The Oxford Handbook of Contemporary Philosophy,
eds Frank Jackson and Michael Smith. Oxford: Oxford University Press: 767–91.
—— 2006. ‘‘Intuitions in Linguistics’’. British Journal for the Philosophy of Science
(forthcoming).
—— and Georges Rey. 1991. ‘‘Transcending Transcendentalism: A Response to Boghos-
sian’’. Pacific Philosophical Quarterly 72: 87–100.
—— and Kim Sterelny. 1987. Language and Reality: An Introduction to the Philosophy of
Language. Cambridge, MA: MIT Press.
—— and Kim Sterelny. 1989. ‘‘What’s Wrong with ‘the Right View’ ’’. In Philosophical
Perspectives, 3: Philosophy of Mind and Action Theory, 1989, ed. James E. Tomberlin.
Atascadero: Ridgeview Publishing Company, 497–531.
—— and Kim Sterelny. 1999. Language and Reality: An Introduction to the Philosophy of
Language. 2nd edn. Cambridge, MA: MIT Press.
Donnellan, Keith S. 1972. ‘‘Proper Names and Identifying Descriptions’’. In Semantics of
Natural Language, eds. Donald Davidson and Gilbert Harman. Dordrecht: Reidel:
356–79.
Dosher, Barbara A. 2003. ‘‘Working Memory’’. In Nadel 2003, vol. 4: 569–77.
Dretske, Fred. 1974. ‘‘Explanation in Linguistics’’. In Explaining Linguistic Phenomena,
ed. D. Cohen. Washing DC: Hemisphere: 21–41.
Dronkers, Nina F. 1999. ‘‘Language, Neural Basis of ’’. In Wilson and Keil 1999: 448–51.
References 283
Dummett, Michael. 1975. ‘‘What is a Theory of Meaning?’’ In Guttenplan 1975:
97–138.
—— 1993. The Seas of Language. Oxford: Oxford University Press.
Dwyer, Susan, and Paul Pietroski. 1996. ‘‘Believing in Language’’. Philosophy of Science
63: 338–73.
Ferreira, Victor S. 2003. ‘‘Production–Comprehension Interface’’. In Nadel 2003, vol. 3:
747–53.
Field, Hartry. 1978. ‘‘Mental Representation’’. Erkenntnis 13: 9–61. Reprinted with
Postscript in Block 1981: 78–114. [Citations are to Block]
Fiengo, Robert. 2003. ‘‘Linguistic Intuitions’’. Philosophical Forum, 34: 253–65.
Fodor, Janet Dean. 1985. ‘‘Deterministic Parsing and Subjacency’’. Language and Cogni-
tive Processes 1: 3–42.
—— 1989. ‘‘Empty Categories in Sentence Processing’’. Language and Cognitive Processes
4: 155–209.
—— 1992. ‘‘Learnability of Phrase Structure Grammars’’. In Formal Grammar: Theory
and Implementation, Robert Levine, ed. New York: Oxford University Press: 3–68.
—— 1998. ‘‘Parsing to Learn’’. Journal of Psycholinguistic Research, 27: 339–74.
—— 2001. ‘‘Parameters and the Periphery: Reflections on Syntactic Nuts’’. Journal of
Linguistics 37: 367–92.
—— Jerry A. Fodor, and Merrill F. Garrett. 1975. ‘‘The Psychological Unreality of
Semantic Representations’’. Linguistic Inquiry 6: 515–31. Reprinted in Block 1981:
238–52. [Citations are to Block]
Fodor, Jerry A. 1975. The Language of Thought. New York: Thomas Y. Crowell.
—— 1981a. Representations: Philosophical Essays on the Foundations of Cognitive Science.
Cambridge MA.: Bradford Books/MIT Press.
—— 1981b. ‘‘Introduction: Some Notes on What Linguistics is Talking About’’. In
Block 1981: 197–207. Reprinted in Katz 1985.
—— 1983. The Modularity of Mind: An Essay on Faculty Psychology. Cambridge, MA.:
MIT Press.
—— 1987. Psychosemantics: The Problem of Meaning in the Philosophy of Mind. Cam-
bridge, MA: MIT Press.
—— 1990. A Theory of Content and other Essays. Cambridge, MA: MIT Press
—— 1998a. Concepts: Where Cognitive Science Went Wrong. Oxford: Clarendon Press.
—— 1998b. In Critical Condition: Polemical Essays on Cognitive Science and the Philoso-
phy of Mind. Cambridge, MA: MIT Press.
—— 2001. ‘‘Doing Without What’s Within: Fiona Cowie’s Critique of Nativism’’. Mind
110: 99–148.
—— and T. G. Bever. 1965. ‘‘The Psychological Reality of Linguistic Segments’’. Journal
of Verbal Learning and Verbal Behavior 4: 414–20.
—— T. G. Bever and M. F. Garrett. 1974. The Psychology of Language: An Introduction
to Psycholinguistics and Generative Grammar. New York: McGraw-Hill Book
Company.
—— and Ernest Lepore. 1992. Holism: A Shopper’s Guide. Oxford: Basil Blackwell.
—— and Zenon W. Pylyshyn. 1988. ‘‘Connectionism and Cognitive Architecture: A
Critical Analysis’’. Cognition 28: 3–71. Reprinted in McDonald and McDonald 1995:
90–163. [Citations are to McDonald and McDonald]
284 References
Fowler, C. A., and M. T. Turvey. 1978. ‘‘Skill Acquisition: An Event Approach with Special
Reference to Searching for the Optimum of a Function of Several Variables’’. In Infor-
mation Processing and Motor Control, G. E. Stelmach, ed. New York: Academic Press.
Frank, Adam. 1997. ‘‘Quantum Honey Bees’’. Discover : 80–7.
Frazier, Lyn. 1995. ‘‘Issues of Representation in Psycholinguistics’’. In Miller and Eimas
1995: 1–27.
—— and Janet Dean Fodor. 1978. ‘‘The Sausage Machine: A New Two-Stage Parsing
Model’’. Cognition 6: 291–325.
—— and K. Rayner. 1982. ‘‘Making and Correcting Errors during Sentence Compre-
hension: Eye Movements in the Analysis of Structurally Ambiguous Sentences’’.
Cognitive Psychology 14: 178–210.
Gallistel, C. R. 1990. The Organization of Learning. Cambridge, MA: MIT Press.
Garfield, Jay L. ed. 1987. Modularity in Knowledge Representation and Natural-Language
Understanding. Cambridge, MA: MIT Press.
Garson, James. 2003. ‘‘Dynamical Systems, Philosophical Issues about’’. In Nadel 2003,
vol. 1: 1033–9.
Gasper, Philip. 1986. Review of 1st edn of Devitt 1997. Philosophical Review. 95:
446–51.
Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized Phrase
Structure Grammar. Oxford: Basil Blackwell.
Geach, Peter. 1967. ‘‘Intentional Identity’’. Journal of Philosophy 64: 627–32.
Gendler, Tamar Szabo. 2003. ‘‘Thought Experiments’’. In Nadel 2003, vol. 4: 388–94.
Gentner, Dedre, and Laura L. Namy. 2004. ‘‘The Role of Comparison in Children’s Early
Word Learning’’. In Weaving a Lexicon, D. G. Hall and S. R. Waxman, eds. Cam-
bridge, MA: MIT Press: 533–68.
George, Alexander, ed. 1989a. Reflections on Chomsky. Oxford: Basil Blackwell.
—— 1989b. ‘‘How Not to Become Confused About Linguistics’’. In George 1989a:
90–110.
Gernsbacher, Morton Ann, and Michael P Kaschak. 2003. ‘‘Language Comprehension’’.
In Nadel 2003, vol. 2: 723–5.
Gibson, Edward. 2003. ‘‘Sentence Comprehension, Linguistic Complexity in’’. In Nadel
2003, vol. 3: 1137–41.
Gladwell, Malcolm. 2005. Blink: The Power of Thinking Without Thinking. New York:
Little, Brown.
Gleitman, Lila R., and Paul Bloom. 1999. ‘‘Language Acquisition’’. In Wilson and Keil
1999: 434–8.
—— and Mark Liberman. 1995. Language: An Invitation to Cognitive Science, Second
Edition, Volume 1. Cambridge, MA: MIT Press.
Goldberg, Elkhonon. 2005. The Wisdom Paradox: How Your Mind Can Grow Stronger as
Your Brain Grows Older. New York: Gotham Books.
Goldin-Meadow, Susan, and Ming-Yu Zheng. 1998. ‘‘Thought Before Language: the
Expression of Motion Events Prior to the Impact of a Conventional Language Model’’.
In Carruthers and Boucher 1998a: 26–54.
Gopnik, Alison, and Eric Schwitzgebel. 1998. ‘‘Whose Concepts Are They, Anyway? The
Role of Philosophical Intuitions in Empirical Psychology’’. In DePaul and Ramsey
1998: 75–91.
References 285
Gordon, P. C., and R. Hendrick. ‘‘Intuitive Knowledge of Linguistic Co-reference’’.
Cognition 62: 325–70.
Gorrell, Paul. 1999. ‘‘Sentence Processing’’. In Wilson and Keil 1999: 748–51.
Graves, Christina, J. J. Katz, Y. Nishiyama, Scott Soames, R. Stecker, and P. Tovey. 1973.
‘‘Tacit Knowledge’’. Journal of Philosophy 70: 318–30.
Grice, H.P. 1957. ‘‘Meaning’’. Philosophical Review 66: 377–88. Reprinted in Grice
1989.
—— 1989. Studies in the Way of Words. Cambridge MA.: Harvard University Press.
Grodzinsky, Yosef. 2003. ‘‘Language Disorders’’. In Nadel 2003, vol. 2: 740–6.
Gross, Maurice. 1983. ‘‘A Few Analogies with Computing’’. Behavioral and Brain Sciences
3: 407–8.
Gumperz, John J., and Stephen C. Levinson, eds. 1996. ‘‘Introduction to Part I’’. In
Rethinking Linguistic Relativity, eds. Gumperz and Levinson. Cambridge: Cambridge
University Press: 21–36.
Gunderson, Keith, ed. 1975. Minnesota Studies in the Philosophy of Science, Volume VII:
Language, Mind, and Knowledge. Minneapolis: University of Minnesota Press.
Guttenplan, Samuel, ed. 1975. Mind and Language. Oxford: Clarendon Press.
—— ed. 1994. A Companion to the Philosophy of Mind. Cambridge, MA: Blackwell
Publishers.
Hadley, Robert F. 1999. ‘‘Connectionism and Novel Combinations of Skills: Implica-
tions for Cognitive Architecture’’. Minds and Machines 9: 197–221.
Haegeman, Liliane. 1994. Introduction to Government and Binding Theory, 2nd edn.
Oxford: Blackwell Publishers. 1st edn 1991.
Halle, Morris, Joan Bresnan, and George A. Miller, eds. 1978. Linguistic Theory and
Psychological Reality. Cambridge, MA.: MIT Press.
Hannan, Barbara. 1993. ‘‘Don’t Stop Believing: The Case Against Eliminative Material-
ism’’. Mind and Language 8: 165–79.
Harman, Gilbert. 1967. ‘‘Psychological Aspects of the Theory of Syntax’’. Journal of
Philosophy 64: 75–87.
—— 1973. Thought. Princeton: Princeton University Press.
—— ed. 1974. On Noam Chomsky: Critical Essays. Garden City, N.Y.: Anchor Press/
Doubleday.
—— 1983. ‘‘Internally Represented Grammars’’. Behavioral and Brain Sciences 3: 408.
Harris, Catherine L. 2003. ‘‘Language and Cognition’’. Nadel 2003, vol. 2: 717–22.
Harris, Randy Allen. 1993. The Linguistics Wars. New York: Oxford University Press.
Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch. 2002. ‘‘The Faculty
of Language: What Is It, Who Has It, and How Did It Evolve?’’ Science 298:
1569–79.
Higginbotham, James. 1987. ‘‘The Autonomy of Syntax and Semantics’’. In Garfield
1987: 119–31.
—— 1989. ‘‘Knowledge of Reference’’. In George 1989a: 153–74.
Hillis, Argye E., and Alfonso Caramazza. 2003. ‘‘Aphasia’’. In Nadel 2003, vol. 1:
175–84.
Holding, Dennis H. 1992. ‘‘Theories of Chess Skill’’. Psychological Research 54: 10–16.
Hook, Sidney, ed. 1969. Language and Philosophy: A Symposium. New York: New York
University Press.
286 References
Horgan, Terence, and John Tienson. 1999. ‘‘Rules and Representations’’. In Wilson and
Keil 1999: 724–26.
Hornstein, Norbert. 1989. ‘‘Meaning and the Mental: The Problem of Semantics after
Chomsky’’. In George 1989a: 23–40.
—— and David Lightfoot, eds. 1981a. Explanation in Linguistics: The Logical Problem of
Language Acquisition. London: Longman.
—— 1981b. ‘‘Preface’’. To Hornstein and Lightfoot 1981a: 7–8.
Howe, Christine J. 1986. Review of Berwick and Weinberg 1984. Mind and Language 1:
83–8.
Hurford, James R., Michael Studdert-Kennedy, and Chris Knight, eds. 1998. Approaches
to the Evolution of Language: Social and Cognitive Basis. Cambridge: Cambridge
University Press.
Jackendoff, Ray. 1997. The Architecture of the Language Faculty. Cambridge, MA: MIT
Press.
Johnson, Kent. 2004. ‘‘Systematicity’’. Journal of Philosophy 101: 111–39.
Johnson, Todd R., Hongbin Wang, and Jiajie Zhang. 2003. ‘‘Skill Acquisition: Models’’.
In Nadel 2003, vol. 4: 30–6.
Jones, Gary. 2003. ‘‘Production Systems and Rule-Based Inference’’. In Nadel 2003,
vol. 3: 741–7.
Jutronic, Dunja ed. 1997. The Maribor Papers in Naturalized Semantics, Maribor:
Pedagoska fakulteta Maribor.
Kasher, Asa, ed. 1991a. The Chomskyan Turn. Oxford: Basil Blackwell.
—— 1991b. ‘‘Pragmatics and Chomsky’’. In Kasher 1991a: 122–49.
Katz, Jerrold J. 1981. Language and Other Abstract Objects. Totowa, N. J.: Rowman &
Littlefield.
—— 1984. ‘‘An Outline of a Platonist Grammar’’. In Bever et al. 1984: 17–48. Rep-
rinted in Katz 1985.
—— ed. 1985. The Philosophy of Linguistics. Oxford: Oxford University Press.
—— 1996. ‘‘The Unfinished Chomskyan Revolution’’. Mind and Language 11: 270–94.
Kelso, J. A. Scott. 1995. Dynamic Patterns: The Self-Organization of Brain and Behavior.
Cambridge MA: MIT Press.
Kintsch, Walter. 1984. ‘‘Approaches to the Study of the Psychology of Language’’. In
Bever et al. 1984: 111–45.
Kornblith, Hilary. 1998. ‘‘The Role of Intuition in Philosophical Inquiry: An Account
with no Unnatural Ingredients’’. In DePaul and Ramsey 1998: 129–41.
Kripke, Saul A. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press.
Kugler, P. N., and M. T. Turvey. 1987. Information, Natural Laws, and the Self-Assembly of
rhythmic Movement. Hillsdale, NJ: Laurence Erlbaum.
Laird, J. E., A. Newell, and P. S. Rosenbloom. 1987. ‘‘Soar: An Architect for General
Intelligence’’. Artificial Intelligence 33: 1–64.
Larson, Richard, and Gabriel Segal. 1995. Knowledge of Meaning: An Introduction to
Semantic Theory. Cambridge, MA: MIT Press.
Lasnik, Howard, and Juan Uriagereka. 2002. ‘‘On the Poverty of the Challenge’’.
Linguistic Review 19:147–50.
Laurence, Stephen. 1996. ‘‘A Chomskian Alternative to Convention-Based Semantics’’.
Mind 105: 269–301.
References 287
—— 1998. ‘‘Convention-Based Semantics and the Development of Language’’. In
Carruthers and Boucher 1998: 201–17.
—— 2003. ‘‘Is Linguistics a Branch of Psychology?’’ In Barber 2003a: 69–106.
—— and Eric Margolis. 2001. ‘‘The Poverty of Stimulus Argument’’. British Journal for
the Philosophy of Science 52: 217–76.
Lees, Robert B. 1957. ‘‘Review of Syntactic Structures’’. Language 33: 375–407. Reprinted
in On Noam Chomsky: Critical Essays ed. Gilbert Harman. Garden City, N. Y.: Anchor
Press/Doubleday: 34–79. [Citation is to Harman]
Lepore, Ernest, ed. 1986. Truth and Interpretation: Perspectives on the Philosophy of Donald
Davidson. Oxford: Basil Blackwell.
—— and Kirk Ludwig. 2000. ‘‘The Semantics and Pragmatics of Complex Demonstra-
tives’’. Mind, 109: 199–240.
Levelt, William J. M., Ardi Roelofs, and Antje S. Meyer. 1999. ‘‘A Theory of Lexical
Access in Speech Production’’. Behavioral and Brain Sciences 22: 1–38.
Lewandowsky, Stephan, and Simon Farrell. 2003. ‘‘Working Memory, Computational
Models of ’’. In Nadel 2003, vol. 4: 578–83.
Lewis, David K. 1969. Convention: A Philosophical Study. Cambridge, MA.: Harvard
University Press.
—— 1994. ‘‘Reduction of Mind’’. In Guttenplan 1994: 412–31.
Lewis, John D., and Jeffrey L. Ellman. 2001. ‘‘Learnability and the Statistical Structure of
Language: Poverty of Stimulus Arguments Revisited’’. Proceeding of the 26 th Annual
Boston University Conference on Language Development. Somerville, MA: Cascadilla
Press: 359–70.
Lipton, Michael R. 1983. ‘‘Levels of Grammatic Representation: A Tempest in a Teapot’’.
Behavioral and Brain Sciences 3: 409–10.
Logan, Gordon D. 1988. ‘‘Toward an Instance Theory of Automatization’’. Psychological
Review 95: 492–527.
McClelland, James L. 1999. ‘‘Cognitive Modeling, Connectionist’’. In Wilson and Keil
1999: 137–41.
McDonald, Cynthia, and Graham McDonald, eds. 1995. Connectionism: Debates on
Psychological Explanation, Volume Two. Cambridge, MA: Blackwell Publishers.
McGilvray, James. 1999. Chomsky: Language, Mind, and Politics. Cambridge: Polity Press.
Macintosh, Nicholas J. 1999. ‘‘Conditioning’’. In Wilson and Keil 1999: 182–4.
McLeod, Peter, and Zoltan Dienes. 1996. ‘‘Do Fielders Know Where to Go to Catch the
Ball or Only How to Get There?’’ Journal of Experimental Psychology: Human Perception
and Performance 22: 531–43.
Malcolm, Norman. 1968. ‘‘The Conceivability of Mechanism’’. Philosophyical Review 77:
45–72.
Malson, L. 1972. Wolf Children and the Problem of Human Nature. London: Monthly
Review Press.
Maratsos, Michael P. 1989. ‘‘Innateness and Plasticity in Language Acquisition’’. In The
Teachability of Language, M. L. Rice and R. L. Schiefelbusch, eds. Baltimore: Paul H.
Brookes: 105–25.
—— 2003. ‘‘Language Acquisition’’. In Nadel 2003, vol. 2: 691–6.
Marr, David. 1982. Vision: A Computational Investigation into the Human Representation
and Processing of Visual Information. San Francisco: W. H. Freeman and Company.
288 References
Masson, Michael E. J. 1990. ‘‘Cognitive Theories of Skill Acquisition’’. Human Move-
ment Science 9: 221–39.
Mathews, R. C., R. R. Buss, W. B. Stanley, and R. Chinn. 1988. ‘‘Analysis of Individual
Learning Curves in a Concept Discovery Task: Relations among Task Performance,
Verbalizable Knowledge, and Hypothesis Revision Strategies’’. The Quarterly Journal of
Experimental Psychology 40a: 135–65.
—— R. R. Buss, W. B. Stanley, F. Blanchard-Fields, J. R. Cho, and B. Druhan. 1989.
‘‘Role of Implicit and Explicit Processes in Learning From Examples: A Synergistic
Effect’’. Journal of Experimental Psychology: Learning, Memory, and Cognition 15:
1083–100.
Matthews, Robert J. 1980. ‘‘Language Learning versus Grammar Growth’’. Behavioral
and Brain Sciences 3: 25–26.
—— 1991. ‘‘The Psychological Reality of Grammars’’. In Kasher 1991a: 182–99.
—— 2003. ‘‘Does Linguistic Competence Require Knowledge of Language?’’ In Barber
2003a: 187–213.
Miller, G. A., and N. Chomsky. 1963. ‘‘Finitary Models of Language Users’’. In Hand-
book of Mathematical Psychology, Vol. ii, eds. R. D. Luce, R. R. Bush, and E. Galanter,
eds. New York: Wiley.
Miller, Joanne L. 1999. ‘‘Speech Perception’’. In Wilson and Keil 1999: 787–90.
—— and Peter D. Eimas, eds. 1995. Speech, Language, and Communication. San Diego:
Academic Press.
Mills, Eugene. ‘‘Devitt on the Nature of Belief ’’. In Jutronic 1997: 310–17.
Mon-Williams, Mark, James R. Tresilian, and John P. Wann. 2003. ‘‘Motor Control and
Learning’’. In Nadel 2003: 121–6.
Moro, A., M. Tettamanti, D. Perani, C. Donati, S. F. Cappa, and F. Fazio. 2001. ‘‘Syntax and
the Brain: Disentangling Grammar by Selective Anomalies’’. NeuroImage 13: 110–18.
Mulligan, Neil W. 2003. ‘‘Memory: Implicit versus Explicit’’. In Nadel 2003, vol. 2:
1114–20.
Nadel, Lynn, ed. 2003. Encyclopedia of Cognitive Science. London: Nature Publishing
Group.
Neale, Stephen. 1989. ‘‘Paul Grice and the Philosophy of Language’’. Linguistics and
Philosophy 15: 509–59.
—— 2004. ‘‘This, That, and the Other’’. In Descriptions and Beyond, Marga Reimer and
Anne Bezuidenhout, eds. Oxford: Clarendon Press: 68–182.
Newell, A., and H. Simon. 1972. Human Problem Solving. Englewood Cliffs, NJ:
Prentice-Hall.
Newell, Karl M. 1996. ‘‘Motor Skills’’. In Encyclopedia of Leaning and Memory, ed. Larry
R. Squire. Macmillan Library Reference: 441–3.
Newport, Elissa L., and Ted Supalla. 1999. ‘‘Sign Languages’’. In Wilson and Keil 1999:
758–60.
Ni, Weijia, Stephen Crain, and Donald Shankweiler. 1996. ‘‘Sidestepping Garden Paths:
Assessing the Contributions of Syntax, Semantics and Plausibility in Resolving Ambi-
guities’’. Language and Cognitive Processes 11: 283–334.
Niiniluoto, Ilkka. 1981. ‘‘Language, Norms, and Truth’’. In Essays in Philosophical
Analysis Dedicated to Erik Stenius on the Occasion of his 70 th Birthday, Ingmar Porn,
ed. Helsinki: The Philosophical Society of Finland: 168–89.
References 289
Nygaard, Lynne C., and David B. Pisoni. 1995. ‘‘Speech Perception: New Directions in
Research and Theory’’. In Miller and Eimas 1995: 63–96.
Palmeri, Thomas J. 2003. ‘‘Automaticity’’. In Nadel 2003, vol. 1: 290–301.
Pateman, Trevor. 1987. Language in Mind and Language in Society: Studies in Linguistic
Reproduction. Oxford: Clarendon Press.
Peacocke, Christopher. 1986. ‘‘Explanation in Computational Psychology: Language,
Perception and Level 1.5’’. Mind and Language 1: 101–23.
—— 1989. ‘‘When is a Grammar Psychologically Real?’’ In George 1989a: 111–30.
Pickering, Martin J. 2003. ‘‘Parsing’’. In Nadel 2003, vol. 3: 462–5.
Pietroski, Paul M. 2003. ‘‘The Character of Natural Language Semantics’’. In Barber
2003a: 217–56.
Pinker, Stephen. 1979. ‘‘Formal Models of Language Learning’’. Cognition 7: 217–82.
—— 1989. Learnability and Cognition. Cambridge MA: MIT Press.
—— 1994. The Language Instinct: How the Mind Creates Language. New York: William
Morrow and Co.
—— 1995a. ‘‘Why the Child Holded the Baby Rabbits: A Case Study in Language
Acquisition’’. In Gleitman and Liberman 1995: 107–33
—— 1995b. ‘‘Language Acquisition’’. In Gleitman and Liberman 1995: 135–82.
—— 1997. How the Mind Works. New York: W. W. Norton and Company.
—— and Paul Bloom. 1990. ‘‘Natural Language and Natural Selection’’. Behavioral and
Brain Sciences 13: 707–27.
Poeppel, David, and Gregory Hickok. 2004. ‘‘Introduction: Towards a New Functional
Anatomy of Language’’. Cognition 92: 1–12.
Port, Robert F. 2003. ‘‘Dynamical Systems Hypothesis in Cognitive Science’’. In Nadel
2003, vol. 1: 1027–32.
Posner, M. I. 1973. Cognition: An Introduction. Glenville IL: Scott, Foresman.
—— G. J. DiGirolamo, and D. Fernandez-Duque. 1997. ‘‘Brain Mechanisms of Cog-
nitive Skills’’. Consciousness and Cognition 6: 267–90.
Poucet, Bruno. 2003. ‘‘Animal Navigation and Cognitive Maps’’. In Nadel 2003, vol. 1:
150–6.
Pritchett, Bradley L. 1988. ‘‘Garden Path Phenomena and the Grammatical Basis of
Language Processing’’. Language 64: 539–76.
Pullum, Geoffrey. 1991. The Great Eskimo Vocabulary Hoax and Other Irreverent Essays on
the Study of Language. Chicago: University of Chicago Press.
—— and Barbara Scholz. 2002. ‘‘Empirical Assessment of Stimulus Poverty Arguments’’.
Linguistic Review 19: 9–50.
Putnam, Hilary. 1967. ‘‘The ‘Innateness Hypothesis’ and Explanatory Models in Linguis-
tics’’. Synthese 17: 12–22. Reprinted in Block 1981: 292–9. [Citations are to Block]
—— 1975. Mind, Language and Reality: Philosophical Papers Vol. 2. Cambridge: Cam-
bridge University Press.
Pylyshyn, Zenon W. 1980a. ‘‘Computation and Cognition: Issues in the Foundations of
Cognitive Science’’. Behavioral and Brain Sciences 3: 111–32.
—— 1980b. ‘‘Author’s Response’’ to peer commentary on 1980a. Behavioral and Brain
—— 1984. Computation and Cognition: Toward a Foundation for Cognitive Science.
290 References
—— 1991. ‘‘Rules and Representations: Chomsky and Representational Realism’’. In
Kasher 1991a: 231–51.
Quine, W. V. 1960. Word and Object. Cambridge MA: MIT Press
—— 1966. The Ways of Paradox and Other Essays. New York: Random House.
—— 1970. ‘‘Methodological Reflections on Current Linguistics’’. Synthese 21: 386–98.
Radford, Andrew. 1988. Transformational Grammar: A First Course. Cambridge: Cam-
bridge University Press.
Ramsey, William. 1997. ‘‘Do Connectionist Representations Earn Their Explanatory
Keep?’’ Mind and Language 12: 34–66.
Reber, A. S. 1967. ‘‘Implicit Learning of Artificial Grammars’’. Journal of Verbal Learning
and Verbal Behavior 6: 855–63.
—— 1989. ‘‘Implicit Learning and Tacit Knowledge’’. Journal of Experimental Psych-
ology: General 118: 219–35.
—— 2003. ‘‘Implicit Learning’’. In Nadel 2003, vol. 2: 486–91.
Reisberg, Daniel. 1999. ‘‘Learning’’. In Wilson and Keil 1999: 460–1.
Rey, Georges. 1997. Contemporary Philosophy of Mind: A Contentiously Classical Approach.
Cambridge, MA: Blackwell Publishers.
—— 2003a. ‘‘Chomsky, Intentionality and a CRTT’’. In Antony and Hornstein 2003:
105–39.
—— 2003b. ‘‘Intentional Content and Chomskyan Linguistics’’. In Barber 2003a:
140–86.
—— 2006. ‘‘The Intentional Inexistence of Language—But Not Cars’’. In Contemporary
Debates in Cognitive Science, ed. R. Stainton. Oxford: Blackwell Publishers: 237–55
Richard, Mark. 1990. Propositional Attitudes: An Essay on Thoughts and How We Ascribe
Them. Cambridge: Cambridge University Press.
Richardson-Klavehn, Alan, and Robert A. Bjork. 2003. ‘‘Memory, Long-Term’’. In Nadel
2003, vol. 2: 1096–1105.
Riley, J. R., U. Greggers, A. D. Smith, D. R. Reynolds, and R. Menzel. 2005. ‘‘The Flight
of Honey Bees Recruited by the Waggle Dance’’. Nature 435: 205–7.
Rosenbaum, David A., Richard A. Carlson, and Rick O. Gilmore. 2001. ‘‘Acquisition of
Intellectual and Perceptual-Motor Skills’’. Annual Review of Psychology 52: 453–70.
Rumelhart, D. E., and J. L. McClelland. 1986. Parallel Distributed Processing: Explorations
in the Microstructure of Cognition, vol. 1, Foundations. Cambridge, MA: MIT Press.
Rumfitt, Ian. 2003. ‘‘Savoir Faire’’. Journal of Philosophy 100: 158–66.
Sampson, Geoffrey. 1976. ‘‘The Simplicity of Linguistic Theories’’. Linguistics 167: 51–66
Samuelson, L. K., and L. B. Smith. 1998. ‘‘Memory and Attention Make Smart Word
Learning: An Alternative Account of Akhtar, Carpenter, and Tomasello’’. Child Devel-
opment 69: 94–104.
—— 2000. ‘‘Grounding Development in Cognitive Processes’’. Child Development 71:
98–106.
Sapir, Edward. 1931. ‘‘Conceptual Categories in Primitive Languages’’. Science 74: 578.
—— 1949. Selected Writings in Language, Culture and Personality, ed. David G. Man-
delbaum. Berkeley: University of California Press.
Saussure, Ferdinand de. 1916. Course in General Linguistics, eds. Charles Bally and Albert
Sechehaye, trans., Wade Baskin, New York: McGraw-Hill Book Co, 1966. 1st French
edn., 1916.
References 291
Schacter, Daniel L. 1999. ‘‘Implicit vs Explicit Memory’’. In Wilson and Keil 1999:
394–5.
Schank, Roger. 1980. ‘‘An Artificial Intelligence Perspective on Chomsky’s View of
Language’’. Behavioral and Brain Sciences 3: 35–7.
—— and Lawrence Birnbaum. 1984. ‘‘Memory, Meaning, and Syntax’’. In Bever et al.
1984: 209–51.
Schiffer, Stephen. 1972. Meaning. Oxford: Clarendon Press.
—— 1987. Remnants of Meaning. Cambridge, MA: MIT Press.
—— 2002. ‘‘Amazing Knowledge’’. Journal of Philosophy 99: 200–02.
Scholz, Barbara, and Geoffrey Pullum. 2002. ‘‘Searching for Arguments to Support
Linguistic Nativism’’. Linguistic Review 19: 185–223.
—— 2006. ‘‘Irrational Nativist Exuberance’’. In Contemporary Debates in Cognitive
Science, ed. R. Stainton. Oxford: Blackwell Publishers.
Schutze, Carson T. 2003. ‘‘Linguistic Evidence, Status of ’’. In Nadel 2003, vol. 2:
910–17.
Searle, John. 1972. ‘‘Chomsky’s Revolution in Linguistics’’. The New York Review of
Books. Reprinted in Harman 1974: 2–33. [Citation is to Harman]
Seidenberg, Mark S. 1995. ‘‘Visual Word Recognition: An Overview’’. In Millar and
Eimas 1995: 137–79.
Singley, Mark K., and John R. Anderson. 1989. The Transfer of Cognitive Skill. Cam-
bridge MA: Harvard University Press.
Smith, Barry C. 2001. ‘‘Idiolects and Understanding: Comments on Barber’’. Mind and
Language 16: 284–9.
Smith, Brian Cantwell. 1999. ‘‘Computation’’. In Wilson and Keil 1999: 153–5.
Smith, N. and I-M Tsimpli. 1995. The Mind of a Savant: Language-Learning and
Modularity. Oxford: Blackwell Publishers.
Smolensky, Paul. 1991. ‘‘Connectionism, Constituency and the Language of Thought’’.
In Meaning in Mind: Jerry Fodor and His Critics, Barry Loewer and Georges Rey, eds.
Oxford: Basil Blackwell: 201–27. Reprinted in Macdonald and Macdonald 1995:
164–98. [Citation is to Macdonald and Macdonald]
Soames, Scott. 1984. ‘‘Linguistics and Psychology’’. Linguistics and Philosophy 7:
155–79.
Sober, Elliott. 1980. ‘‘Representation and Psychological Reality’’. Behavioral and Brain
Sciences 3: 38–9.
Sokolov, A. 1972. Inner Speech and Thought. New York: Plenum Press.
Sosa, Ernest, 1998. ‘‘Minimal Intuition’’. In DePaul and Ramsey 1998: 257–69.
—— and Mattias Steup, eds. 2005. Contemporary Debates in Epistemology. Cambridge,
MA: Blackwell Publishers.
Spencer, N. J. 1973. ‘‘Differences Between Linguists and Nonlinguists in Intuitions of
Grammaticality–Acceptability’’. Journal of Psycholinguistic Research 2: 83–98.
Sperber, Dan. 1996. Explaining Culture: A Naturalistic Approach. Oxford: Blackwell
Publishers.
—— and Deirdre Wilson. 1995. Relevance, Communication and Cognition. 1st edn.
1986. Oxford: Blackwell Publishers.
—— 1998. ‘‘The Mapping Between the Mental and the Public Lexicon’’. In Carruthers
and Boucher 1998a: 184–200.
292 References
Stabler, Edward P., Jr. 1983. ‘‘How are Grammars Represented?’’ Behavioral and Brain
Stanley, Jason, and Zoltan G. Szabo. 2000. ‘‘On Quantifier Domain Restriction’’. Mind
and Language 15: 219–61.
—— and Timothy Williamson. 2001. ‘‘Knowing How’’. Journal of Philosophy 98:
411–44.
Stanley, William B., Robert C. Mathews, Ray R. Buss, and Susan Kotler-Cope. 1989.
‘‘Insight Without Awareness: On the Interaction of Verbalization, Instruction and
Practice in a Simulated Process Control Task’’. The Quarterly Journal of Experimental
Psychology 41a: 553–77.
Steedman, Mark. 2003. ‘‘Language, Connectionist and Symbolic Representations of ’’. In
Nadel 2003, vol. 2: 765–71.
Stemberger, Joseph P. 2003. ‘‘Speech Error Models of Language Production’’. In Nadel
2003, vol. 4: 156–62.
Sterelny, Kim. 2003. Thought in a Hostile World.: The Evolution of Human Cognition,
Oxford: Blackwell Publishing.
Steup, Mattias, and Ernest Sosa, eds. 2005. Contemporary Debates in Epistemology.
Cambridge, MA: Blackwell Publishers.
Stich, Stephen P. 1971. ‘‘What Every Speaker Knows’’. Philosophical Review 80: 476–96.
—— 1978a. ‘‘Empiricism, Innateness, and Linguistic Universals’’. Philosophical Studies
33: 273–86.
—— 1978b. ‘‘Beliefs and Subdoxastic States’’. Philosophy of Science 45: 499–518.
—— 1980. ‘‘What Every Speaker Cognizes’’. Behavioral and Brain Sciences 3: 39–40.
—— 1983. From Folk Psychology to Cognitive Science: The Case Against Belief. Cambridge,
MA.: MIT Press.
Sun, Ron. 2003. ‘‘Connectionist Implementationalism and Hybrid Systems’’. In Nadel
2003, vol. 1: 697–703.
—— Edward Merrill, and Todd Peterson. 2001. ‘‘From Implicit Skills to Explicit
Knowledge: A Bottom-Up Model of Skill Learning’’. Cognitive Science 25: 203–44.
Swinney, David A. 1999. ‘‘Aphasia’’. In Wilson and Keil 1999: 31.
Taatgen, Niels A. 2003. ‘‘Learning Rules and Productions’’. In Nadel 2003, vol. 2:
822–30.
Tanenhaus, Michael K. 2003. ‘‘Sentence Processing’’. In Nadel 2003, vol. 3: 1142–8.
—— and John C. Trueswell. 1995. ‘‘Sentence Comprehension’’. In Miller and Eimas
1995: 217–62.
Taylor, Kenneth A. 1994. ‘‘How Not to Refute Eliminative Materialism’’. Philosophical
Psychology 7: 101–25.
Thompson, Henry. 1983. ‘‘Computation Misrepresented: The Procedural/Declarative
Controversy Exhumed’’. Behavioral and Brain Sciences 3: 415.
Thornton, Rosalind. 1995. ‘‘Referentiality and Wh-Movement in Child English: Juvenile
D-Linkuency’’. Language Acquisition 4: 139–75.
Tomberlin, James E. 1998. ‘‘Naturalism, Actualism, and Ontology’’. In Philosophical
Perspectives, 12, Language, Mind, and Ontology, 1998, ed. James E. Tomberlin.
Cambridge MA: Blackwell Publishers: 489–98.
—— 2001. ‘‘How Not to be an Actualist’’. In Philosophical Perspectives, 15, Metaphysics,
2001, ed. James E. Tomberlin. Cambridge MA: Blackwell Publishers: 421–5.
References 293
Van Gelder, Tim. 1999. ‘‘Dynamic Approaches to Cognition’’. In Wilson and Keil 1999:
244–6.
Vigliocco, Gabriella, and Daniel P. Vinson. 2003. ‘‘Speech Production’’. In Nadel 2003,
vol. 4: 182–9.
Wade, Nicholas. 2005. ‘‘A New Sign Language Arises, and Scientists Watch it Evolve’’.
New York Times, February 1: F3.
Wanner, Eric. 1977. Review of Fodor, Bever, and Garrentt 1974. Psycholinguistic Research
6: 261–70.
Webelhuth, Gert (ed.). 1995a. Government and Binding Theory and the Minimalist
Program. Oxford: Blackwell Publishers.
—— 1995b. ‘‘X-bar Theory and Case Theory’’. In Webelhuth 1995a: 15–95.
Wheeldon, Linda R., Antje S. Meyer, and Mark Smith. 2003. ‘‘Language Production,
Incremental’’. In Nadel 2003, vol 3: 760–4.
Whorf, Benjamin Lee. 1956. Language, Thought, and Reality, ed. and intro. John B.
Carroll. Cambridge, MA. MIT Press.
Wilson, Robert A. 1999. ‘‘Philosophy’’. In Wilson and Keil 1999: xiii–xxxvii.
—— and Frank C. Keil, eds. 1999. The MIT Encyclopedia of the Cognitive Sciences.
Wisniewski, Edward J. 1998. ‘‘The Psychology of Intuitions’’. In DePaul and Ramsey
1998: 45–58.
Wolpert, Daniel M., and Zoubin Ghahramani. 2003. ‘‘Motor Learning Models’’. In
Nadel 2003, vol. 3: 138–42.
Yamada, Jeni E. 1990. Laura: A Case Study for the Modularity of the Mind. Cambridge
MA: MIT Press.
Yamadori, Atsushi, Takashi Yoshida, Etsuro Mori, and Hikari Yamashita. 1996. ‘‘Neuro-
logical Basis of Skill Learning’’. Cognitive Brain Research 5: 39–54.
Index
Abduction: criteria for 199, 253; from Bever, Tom 10, 28 n. 14, 65, 72–7, 80–3,
evidential role of intuitions 96–8, 100 n., 127, 195–6, 200, 209, 221
100, 112–19; ‘‘only-theory-in-town’’ Bickerton, Derek 130 n. 11, 131
12, 13–14, 198–201, 204, 219, 229, Birnbaum, Lawrence 197
241, 252–6, 266, 271 Bjork, Robert A. 212
adaptive control of thought (ACT), see Bjorklund, David F. 211
production systems blacksmiths and horseshoes 17–18, 21 n.
Allen, C. 131 6, 22, 23, 26, 39, 106
Anderson, John R. 5 n. 4, 12, 47, 210, Blanchard-Fields, F. 116 n., 217
214–16, 219, 240, 257 n. 14 Blasi, Carlos Hernandez 211
Andrews, Avery 196 n. 3 Bloom, A. H. 136
Antony, Louise 8, 33 Bloom, Paul 130 n. 12, 250–1, 263
Armstrong, D. M. 133 n. 19, 146 n. 10 Bloomfieldian linguistics 8 n. 11, 26, 27–8
artificial grammar (AG) 114 n. 32, 217 Blumstein, Sheila E. 222 n. 24
associationism 55–6, 226–7 Bock, Kathryn 68 n. 8, 169, 177, 196 n.
Aston, B. 116 2, 230–3
Atkinson, Antony P. 94 n. 6 Boden, M. A. 68 n. 8
Au, T. K. 136 Boghossian, Paul A. 126 n. 2
Avramides, A. 133 n. 19 BonJour, Laurence 105 n. 14
Boroditsky, Lera 136 n. 25
Bach, Kent 128 n. 5 Botha, Rudolf P. 130 n. 11
Baker, C. L. 70, 96, 100 n., 101 n. 6, Boucher, Jill 125, 160, 174 n. 8, 175 n. 10
119 n. 40, 183 n. Braddon-Mitchell, David 146, 147 nn.
Baker, Lynne Rudder 126 n. 2 12–13
Barber, Alex viii, 6 n. 8, 129–30 n. 8, brain impairment: cognitive 11, 131, 137,
178 n. 19 163–5, 189–90, 263; linguistic 11,
Barkow, J. 94 n. 6 92–3, 131, 137, 163–71, 173,
Baron-Cohen, S. 94 n. 6 189–90, 208, 257, 259, 261, 263, 269
Bealer, George 105 n. 14 Bresnan, Joan 10, 36, 76–80, 83–4,
bees and their dances 20–1, 22, 23, 26, 195–6, 222
29, 35, 36, 37, 38, 39, 49, 60, 67–8, Broadbent, Donald E. 116
134, 207, 216 Broadbent, Margaret H. P. 116
behaviorism 6–7, 10, 55–6, 87–8, 120, Bromberger, Sylvain v
125–6, 132 n. 17, 191, 252, 272 Brown, Liana E. 212
Bekoff, M. 131 ‘‘brute–causal’’ processing vs ‘‘rational–
Bennett, J. 133 n. 19, 186 causal’’ processing 12, 53–6, 59–60,
Berkeley, George 199 n. 61, 73, 88, 197, 198, 201, 206,
Bernal, Sara viii 208–9, 220–34, 239–41, 242–3,
Berwick, Robert 10, 32 n. 25, 36, 58 n. 250, 267–8, 274, 275, 276; see also
16, 65, 79–81, 83–4, 195–6, 200 n., psychological reality, some possible
222 positions on, versions (a) and (b)
296 Index
Bub, Jeffrey viii 81, 87–96, 247, 253 n. 9; on
Burge, Tyler 139–40 RTM 6; on thoughts 11, 125, 163,
Burton-Roberts, Noel 31, 90 n., 222 n. 174–8, 190
25 Chomsky’s tasks:
Buss, R. R. 116 n., 217 (i) 3, 4, 24, 26, 28–9, 30, 32–4, 134
(ii) 3, 4–5, 244
Cappa, S. F. 168 (iii) 3, 5, 66, 75, 176–7
Caramazza, Alfonso 167–70 Churchland, Patricia S. 125
Carey, Susan 131 Churchland, Paul M. 125
Carlson, Richard A. 106 n., 211, 213, Clark, Andy 137, 150, 161
215, 216 n. 20, 267 n. 26 Clark, R. 255
Carr, Philip 31, 90 n., 222 n. 25 Cleeremans, Axel 211 n., 212 n. 14, 217,
Carroll, Lewis 46, 50, 107–8 219, 221 n., 227
Carruthers, Peter 125, 130 n. 11, 152 n., cognitive linguistics 174 n. 7
160, 174 n. 8, 175 n. 10 ‘‘cognizing’’ 4, 69, 96
Cartesian access 96–8, 100–3, 106, 109, Collins, John viii, 8 n. 11, 63, 254 n.,
112–19, 204 n. 255, 268 n.
central processor, see linguistic ‘‘communicative’’ vs. ‘‘cognitive’’
competence, relation to conceptual conceptions of language 125, 160–1
competence and the central processor competence distinguished from outputs
Champollian, Jean Francois 29 n. 18 and inputs 8–9, 17–21, 23–41, 71,
Chang, Franklin 232 273; see also major conclusions, first
chess players and chess moves 18–19, 21, competence hypothesis 76–7;
22, 23, 24, 26, 29, 35, 36, 53, 58, 59, strong 77–8
106 computer analogies 5, 46–7, 50–2, 54,
Chierchia, Gennaro 154 56 n. 15, 58, 62–9, 144 n. 6, 204,
Chinn, R. 116 n., 217 207, 209, 215–16
Cho, J. R. 116 n., 217 conceptual analysis 105–6, 132 n. 16
Chomskian Nativism 247–8, 253–7, 266, conceptual competence: nature of, see
270, 272, 275 thoughts, nature of; relation to
Chomsky, Noam v, vii, 14–16, 39, 45 n. linguistic competence, see linguistic
1, 74, 108 n. 18, 131, 154, 166 n.; competence, relation to conceptual
on conventions and idiolects, 11–12, competence and the central processor
178–84, 190; on E-languages 26, 30; conditioning 211, 216–17, 226 n. 29
on Gricean semantics 174; on connectionism 51, 55–6, 125, 147, 214,
intuitions 3–5, 10, 95–6, 101 n. 7, 217, 218, 221, 227, 228 n., 232–3,
119; on knowledge of language 3–5, 238–40, 242
10, 89–94; on the language continuity hypothesis 251, 271
faculty 13, 173–4, 262 n., 263–5; conventional meaning and syntax 12,
on LET 174; on LOTH 174–5; on 128, 132–41, 155–8, 178–90,
language use 10, 62–71, 76–7, 79, 225–6; see also literal meaning and
195, 200, 207, 223; on linguistics: syntax; speaker meaning and syntax
what it is about 3–9, 26–36; on Corina, David P. 167
nativism 13–14, 130, 173–4, 177 n. Cosmides, L. 94 n.6
14, 244–7, 249–50, 251 n. 8, 252–6, Cowie, Fiona viii, 8 n. 13, 39 n, 35, 87,
259, 268 n., 27; on realism 8, 28, 164 n., 165 n. 2, 249–50, 253–6,
35; on RT 3–7, 10, 34, 46, 62–4, 71, 259 n. 18, 266, 271
Index 297
Crain, Stephen viii, 99 n., 110 n. 24, E-languages 26, 30
237, 246 n., 249 n., 251, 255 Ellman, Jeffrey L. 249 n.
creolization 251 evidence: for grammars 4, 8–9, 10, 17,
Croft, William 174 n. 7 31–4, 35, 37, 73–4, 95–7, 98–103,
Cruse, D. Alan 174 n. 7 108–15, 120, 191; for nativism 13,
Culicover, Peter W. 251 n. 8 246, 248–52, 254, 256, 262–3, 266,
Cummins, Robert 8 n. 12, 146 n. 10 269, 271–2; see also intuitions
Curtis, Susan 137, 263 (intuitive judgments); Respect
Cutler, Anne 169, 233 n., 238 Constraint
Davidson, Donald 125 n., 132 n. 17, Farrell, Simon 240

178 n. 18, 179 Fazio, F. 168
Davies, Martin 52 n. 10, 92 n., 130 n. 9, Fernandez-Duque, D.
132 n. 16, 204 n. 218
Davis, Martin 58 n. 16 Ferreira, Victor S. 233
Dehaene, S 150 Field, Hartry 47 n. 6, 144 n. 6
Dell, Gary S. 209, 232–3 Fiengo, Robert 95 n.
Demopoulos, William 81 Fitch, W. Tecumseh 130 n. 12, 131,
Dennett, Daniel 47 n. 6, 55, 64 177 n. 14
Descartes’ problem 176–7 Fitzgerald, Peter 116
Devitt, Michael v–vi, viii, 5 nn. 4 and 6, Fodor, Janet Dean 32 n. 25, 35, 36, 78 n.
7 n. 10, 8, 24 n. 10, 29–30, 32 n. 11, 97 n., 154, 237, 239–40, 241 n.
24, 36 n. 30, 37 n. 31, 39 n. 35, 37, 251 n. 8, 255–6
46–7, 69 n., 89 nn. 1–2, 103–8, Fodor, Jerry 28 n. 14, 29–30 n. 20, 33,
113 n. 30, 119, 125–7, 132 n. 16, 35, 35, 46 n. 5, 51, 52, 54 n. 12,
133, 134–6, 138 n., 142–4, 152 n., 92 n. 5, 126 n. 2, 127, 144 n. 6,
154 n. 17, 156 nn. 22–4, 157 n. 27, 156, 175 n. 10, 178 n. 17, 202,
165 n. 3, 174 n. 7, 182, 184 n., 216 n. 19, 228; vs. associationism
185 n., 186 n. 27, 189, n., 199, 202, 226–7; on intuitions 96, 97 n.,
203, 219 n. 100 n., 112, 113 n. 31, 114; on
Dienes, Zoltan 50 LOTH viii, 145–6, 159–60, 174 n.
DiGirolamo, G. J. 218 8, 205; on language use 65, 72–7,
Docherty, Gerard 31, 90 n., 80–3, 195–6, 198 n. 6, 200, 209,
222 n. 25 210, 221; on linguistics: what it is
Donati, C. 168 about v, viii, 6, 8 n. 11, 39 n. on
Donnellan, Keith S. 138 nativism 13–14, 247, 252–5, 268,
Dosher, Barbara 240 271; on RT viii, 3 n. 2, 6, 8 n. 11,
Dow, James viii 10, 72–6, 83, 195–6, 198 n. 6
Dretske, Fred 8 n. 12 formal vs syntactic properties 154–5
Dronkers, Nina F. 168 Fowler, C. A. 214
Druhan, B. 116 n., 217 Frank, Adam 20, 29 n. 18
Duhem–Quine thesis 33, 114, 120 Frazier, Lyn 196 n. 1, 198 n. 5, 237,
Dummett, Michael 34 n., 132 n. 17 239
Dwyer, Susan 6 n. 8, 48 n., 91 n., 96,
101 n. 6, 119 n. 40; on linguistics: Gagnon, D. A. 233
what it is about 8, 26 n., 39–40 Gallistel, C. R. 21 n. 5, 54 n. 13, 218 n.,
dynamical systems 213–14, 218, 221 221 n., 226 n. 29
298 Index
Garrett, Merrill 10, 28 n. 14, 35, 65, Hadley, Robert F. 227
72–7, 80–3, 97 n., 100 n., 127, Haegeman, Liliane 28 n. 14, 95, 101 n.
195–6, 200, 209, 221 6, 183 n.
Garson, James 213 Haggard, Cynthia viii
Gasper, Philip 126 n. 2 Hannan, Barbara 126 n. 2
Gazdar, Gerald 8 n. 11, 38 n. 33 Harman, Gilbert 68 n. 8, 92, 149, 205–7,
Geach, Peter 188 268
Gendler, Tamar Szabo 105 n. 13 Harnish, Robert M. 8 n. 12
generation: actual vs. merely Harris, Catherine l. 35
metaphorical 68–71, 274; by the Harris, Randy Allen 8 n. 11, 29 n. 19
I-language vs. by the Hauser, Marc D. 130 n. 12, 131, 177 n. 14
grammar 71, 274 Hendrick, R. 98, 111 n. 25, 115 n. 36
Gentner, Dedre 262 Hickok, Gregory 168
George, Alexander 8 n. 12 Higginbotham, James 101 n. 6, 178 n. 19
Gernsbacher, Morton Ann 234, Hillis, Argye E. 167–70
236, 240 Holding, Dennis H. 215 n.
Ghahramani, Zoubin 213 Horgan, Terence 51
Gibson, Edward 237 n., 238 Hornstein, Norbert 38 n. 33, 101 n. 6
Gilbert, Margaret 8 n. 12 Horwich, Paul viii
Gilmore, Rick O. 211 Howe, Christine J. 79 n.
Gladwell, Malcolm 104 n. Hurford, James R. 130 n. 13
Gleitman, Lila 250–1, 263
Goldberg, Elkhonon 103 n. 11 idiolects 12, 132, 178–84, 189 n., 190,
Goldin-Meadow, Susan 131 205; see also major conclusions, sixth
Gopnik, Alison 104 ignorance of language 5, 16, 89 n. 2, 109,
Gopnik, Myrna 164 139, 252, 269, 271; see also major
Gordon, P. C. 98, 111 n. 25, conclusions, second, third and seventh
115 n. 36 inference to the best explanation, see
Gorrell, Paul 234 abduction
Graves, Christina 92 n., 96, p7 n., innate ideas 14, 245–7, 269
118 n., 247 instance theory 214, 218, 221
Greggers, U. 20 n. 3 intentional realism 10, 108, 125–7, 141,
Grice, Paul 10, 128, 132–3, 172; see also 148, 158, 161
Gricean semantics intuitions (intuitive judgments) 3–7, 10,
Gricean semantics viii, 10–11, 40 n., 26 n., 32, 95–121, 181, 191, 204 n.,
112 n. 29, 132–3, 141, 161, 178–9, 208, 217, 219, 223, 272; see also
192, 220, 234–5; and linguistic major conclusions, third
relativity 135–7; and reference I-Representational Thesis 13–14, 247,
borrowing 138–41; wedded to 252–7, 266–72, 273, 275, 276; see
LOTH 155–6, 162, 172–3, 261, also major conclusions, seventh
262; see also conventional meaning;
literal meaning; speaker meaning Jackendoff, Ray 173, 264
Griffin, Zenzi M. 232 Jackson, Frank 146, 147 nn. 12–13
Grodzinsky, Yosef 6 n. 8 Johnson, Kent 146 n. 9
Gross, Maurice 58 n. 16 Johnson, Todd R. 215, 218, 227, 266 n.
Gualmini, Andrea 99 n. Jones, Gary 214 n. 17
Gumperz, John J. 135 n. 23 Jutronic, Dunja viii
Index 299
Kant, Immanuel 69 n., 135 language-of-thought hypothesis
Kaplan, Ronald 10, 36, 76–80, 83–4, (LOTH) viii, 11, 13–14, 60, 92,
195–6, 222 125, 142, 145–60, 161–2, 172–3,
Kaschak, Michael P. 234, 236, 240 174–5, 190–2, 201–3, 205–6, 208,
Kasher, Asa 176 n. 13 223, 225–7, 228, 243, 247, 257–60,
Katz, Jerrold 8, 26–7, 31 n. 22, 92 n., 96, 261, 264–6, 268–9, 271, 273, 275,
97 n., 118 n., 247 276 ; Public- 149–52, 158, 161–2,
Kelso, J. A. Scott 213, 214 n. 16 258, 273; see also psychological
Kintsch, Walter 226 n. 28 reality, some possible positions on,
Klein, Ewan, 8 n. 11, 38 n. 33 (t); tentative proposals; thoughts,
Knight, Chris 130 n. 13 nature of
Knowledge: declarative (explicit) vs. language use (processing), as brute–causal,
procedural (implicit) 47, 211, see tentative proposals, fourth; not
214–19, 227–8, 264, 266–7, 270; governed by the structure rules of
-how vs. -that (propositional) 10, language, see tentative proposals,
46–7, 50, 89–94, 106, 246–7; of third; see also linguistic competence
language 3–8, 10, 34, 47, 89–94, Larson, Richard 96, 176 n. 12
109, 246–7, 264; tacit 4–5, 49, Lasnik, Howard 255 n.
52 n. 11, 64, 91–2, 96, 97 n., 205–6, Laurence, Stephen 97 n.; on
247; see also learning, explicit vs. conventions 133 n. 18, 180, 182 n.,
implicit; memory, explicit 189n.; on linguistics: what it is
(declarative) vs. implicit (procedural) about 8, 28 n. 16, 30 n. 21, 32 n.
Kornblith, Hilary 103 n. 9 24, 33, 36 n. 30, 37 n. 31, 39–40; on
Kotler-Cope, Susan 116 n., 217 nativism 248 n., 249 n., 253 n.10,
Kripke, Saul A. 138 255
Kugler, P. N. 214 learning, explicit vs implicit 12, 107 n.,
115–16, 211–12, 216–19, 221, 242,
Laird, J. E. 214 n. 17 245, 250, 264, 266–7; see also
Lakoff, George v knowledge, declarative (explicit) vs.
language acquisition 3, 4–5, 10, 12–14, procedural (implicit); memory,
15, 32–3, 40–1, 77, 84, 87, 99–100, explicit (declarative) vs. implicit
115–16, 120–1, 130, 133, 171, (procedural)
173–4, 181, 189, 191–2, 197, 219, Lebiere, Christian 214–15
221, 238, 242, 243, 244–72; see also Lees, Robert B. 119 n. 40
major conclusions, seventh; tentative Lepore, Ernest 126 n. 2, 144 n. 5
proposals, sixth and seventh Levelt, William J. M. 231–3
language expresses thought (LET) 10, Levinson, Stephen C. 135 n. 23
127–32, 141, 148, 151, 161, 163, Lewandowsky, Stephan 240
165–6, 172–4, 177, 189–92, 201–3, Lewis, David 21 n. 6, 133 n. 18, 146 n.
257, 260–1, 273, 275 10, 180,
language faculty 4–14, 31 n. 23, 40, 69, Lewis, John D. 249 n.
93–4, 96, 102, 112–14, 120, 125, lexical function grammar (LFG) 10,
163–92, 203, 208–9, 228–9, 245–8, 77–80, 196 n. 3, 208, 222
251, 260–6, 271–2; see also linguistic Lightfoot, David 38 n. 33
competence; major conclusions, linguistic competence: acquired by
second and seventh; tentative implicit learning 12, 115–16, 219,
proposals, second 221, 242, 245, 250, 264, 266–7;
300 Index
linguistic competence: (cont.): second 9, 12, 14, 84, 97, 120–1, 191,
minimal view of 4, 128, 147–8, 201; 197, 241 n., 242, 272, 275;
relation to conceptual competence argument for 87–121, 195–272
and the central processor 10–11, 13, third 10, 120, 191, 208, 276; argument
94, 96, 106, 109–14, 117–18, for 95–121
129–32, 135–41, 144–5, 148, 153, fourth 10, 14, 129, 141, 142, 148, 161,
163–7, 173–8, 189–91, 201–3, 178, 191, 198, 243, 257, 260,
208–9, 220, 223, 227–8, 233–4, 265–6, 276; argument for 125–9
245, 248, 257, 260–6, 271–2, 275; fifth 10–11, 141, 161, 174, 191, 276;
as a skill or ability 12, 14, 89–94, argument for 37–8, 125–41
106, 108–11, 218–20, 221–30, 232– sixth 12, 183, 190, 276; argument
43, 245, 248, 250, 256, 262–4, 266– for 178–92
70, 272; view (a) of, assuming seventh 14, 269, 271, 276; argument
LET 128–9, 148, 163–6, 189–90, for 244–72
201–3, 275; view (b) of, assuming Malcolm, Norman 126 n. 2
RTM 144–5, 148, 172, 201–3, 275; Malson, L. 137, 263
view (c) of, assuming LOTH 11, map-like representations 11, 54 n. 13,
148, 161, 172, 201–3, 275; see also 146–7, 149, 158, 172, 202–3, 206–7,
language faculty; major conclusions, 218 n., 221 n., 223, 259–61,
second and seventh; psychological 268
reality, some possible positions on; Maratsos, Michael P. 257 n. 15, 259
tentative proposals Marcus parser 79–82
linguistic relativity 135–7, 141, 158–9, 263 Margolis, Eric viii, 165, 248 n., 249 n.,
linguistics, not part psychology, see major 253 n. 10, 255
conclusions, first; methodological points Marr, David 49, 66–7, 69,
Lipton, Michael R. 58 n. 16 114
literal meaning and syntax 128, 132–4, Martin, N. 233
156, 179–81, 186, 189 n., 234, Masson, Michael E. J. 211, 214, 215
235 n. 33; see also conventional Mathews, R. H. 116 n., 217
meaning and syntax; speaker Matthews, Robert viii, 8, 10, 63, 81–4,
meaning and syntax 109 n. 20, 115 n. 36, 195–6, 197,
Locke, John 199 n. 200 n., 210, 221, 267
Logan, Gordon D. 210, 214 Maumus, Michael 19 n., 29 n. 18
logic machines and wffs 19–21, 22, 23, Meeks, Roblin 28 n. 17
24–5, 29, 35, 37, 58, 59, 60, 68, 93, memory, explicit (declarative) vs implicit
147 n. 13, 148–9, 152, 207 (procedural) 211–12, 217–18, 231;
Longworth, Guy viii see also knowledge, declarative
Ludwig, Kurt 144 n. 5 (explicit) vs procedural (implicit);
learning, explicit vs. implicit
McClelland, James L. 51, 214, 238 Mentalese, see language-of-thought
McDonald, Fritz viii hypothesis (LOTH)
McGilvray, James 28 n. 15, 69 n. Menzel, R. 20 n. 3
Macintosh, Nicholas J. 56 Merrill, Edward 47, 107 n.16, 211 n.,
McLeod, Peter 50 214–15, 217, 221 n. 23
major conclusions: methodological points:
first 8–9, 40, 84, 120, 191, 275; first 35, 57, 61, 76, 80, 81, 83, 195–6,
argument for 17–41 274
Index 301
second 35–6, 57, 61, 71, 76, 79, 83, Pisoni, David B. 186 n. 26
120, 142, 196, 274 Platonism 26, 31 n. 22; see also
third 36–7, 38, 57, 61, 63, 67, 72, nominalism; realism
74, 76, 78, 79, 80–1, 83, 196, about language
274 Poeppel, David 168
fourth 37–8, 274 Port, Robert F. 213, 214 n.
Meyer, Antje S. 226 n. 27, 231–3 16
Miller, G. A. 63, 79 Posner, M. I. 107, 218
Miller, Joanne 222 n. 26 Poucet, Bruno 218 n.
Mills, Eugene 144 n. 6 poverty of stimulus 249–51, 254, 271
Mon-Williams, Mark 212 n. 15, 213 pragmatic factors in language use 101–2,
Mori, Etsuro 217–18 110, 112, 129, 148, 163, 182, 187,
Moro, A. 168 201, 220, 226, 230, 233–5, 237,
Mulligan, Neil W. 211 n. 240, 253 n. 9
priority of conceptual competence over
Namy, Laura L. 262 linguistic competence,
Neale, Stephen 119 n. 40, 133 n. 19, 154 ontological 10–11, 128–32, 137,
Newell, A. 214 n. 17, 215 n., 257 n. 14 141, 148, 161, 163–5, 172–3, 202;
Newell, Karl M. 211, 214 Chomsky’s views on 174–7; see also
Newport, Elissa L. 15, 250 major conclusions, fourth and fifth
Ni, Weijia 237 priority of theoretical interest in thought
Niiniluoto, Ilkka 109 n. 21 over language 11, 30, 38, 134–5,
Nishiyama, Y. 92 n., 96, 97 n., 118 n., 247 141; Chomsky’s views on 174–7; see
nominalism 26–7, 30, 31, 33–4; see also also major conclusions, fifth
Platonism; realism about language priority of theory of language over theory
Norris, Dennis 233 n. of competence 23, 29, 37–8, 40,
Nygaard Lynn C. 186 n. 26 134–5; see also methodological
points, fourth
Palmiri, Thomas J. 210 n. 12 priority of thought over language,
Paradox of Language Acquisition, 173, 264 explanatory 10–11, 132–3, 135–41,
Pateman, Trevor 96 155–8, 161, 172–3, 192, 261;
Peacocke, Christopher 67 n. Chomsky’s views on 174–7; see also
Perani, D. 168 major conclusions, fifth
Pereplyotchik, David 19 n., 114 n. 33 Pritchett, Bradley L. 84, 200 n.,
Peterson, Todd 47, 107 n. 16, 211 n., 240 n.
214–15, 217, 221 n. 23 production systems 214–18, 240, 266,
phonology 15, 31, 90 n., 109 n. 20, 154, 267 n. 25
170, 172 n., 184–7, 222–3, 224, psychological reality, some possible
230–3, 265 n. positions on
Pickering, Martin J. 236, 238 (m) 9, 57, 59, 61, 73–4, 80, 82–3, 117,
Pietroski, Paul viii, 6 n. 8, 15 n., 48 n., 120, 128, 195, 198, 205, 243 n.,
91 n., 96, 99 n., 101 n. 6, 119 n. 40, 274
235 n. 32; on linguistics: what it is (i) 57–61, 62–84, 90, 94, 96–7, 117,
about 8, 26 n., 39–40; on 120, 153, 190–1, 195–243,
nativism 246 n., 249 n., 251, 255 244–5, 270–1, 274
Pinker, Stephen 33 n. 26, 78 n., 130 nn. (ii) 57–61, 62, 68–72, 75–6, 83–4, 90,
10 and 12, 134 n. 21, 153, 164, 94, 96–7, 117, 120, 190–1,
167–8, 215, 251, 254, 258 n. 195–243, 244–5, 270–1, 274
302 Index
psychological reality (cont.): Reisberg, Daniel 210
(iii) 58–61, 62–8, 71, 79, 81–3, 90, ‘represent’, senses of, 5–7, 97, 221
97–8, 117–18, 153, 190–1, Representational Theory of Mind
195–8, 207–8, 220–42, 260, 274 (RTM) 6, 11, 46–7, 90, 92, 142–5,
(iv) 59, 61, 75–6, 195–243, 274 148, 158, 161, 172, 176, 191, 201,
(v) 60, 76, 195–243, 275 203, 206–7, 223, 229, 240, 242,
versions (a) 59–61, 73, 191, 197–8, 268, 273, 275; see also thoughts,
206, 220–1, 229, 233, 242, 275 nature of.
versions (b) 59–61, 73, 88, 191, Representational Thesis (RT) vii, 4–7,
197–8, 201, 220–43, 275 9–10, 12–14, 33, 34, 40–1, 46, 52,
(t) 60, 84, 90, 120, 125, 141, 142–62, 53, 57–8, 61, 62–4, 71–84, 87–121,
163, 177, 190–2, 198, 202, 208, 166, 190, 191, 196, 197–220, 228,
257–60, 264–6, 269, 271, 275 229–43, 244–5, 247, 252, 253 n. 9,
see also linguistic competence 264–5, 266–7, 269, 270–2, 273,
Pullum, Geoffrey viii, 8 n. 11, 38 n. 33, 293, 275; see also major conclusions,
136 n. 24, 249 n., 251 n. 8, 255 n. second; psychological reality, some
Pupa, Francesco 111 n. 26, 209–10 n., possible positions on, (i) and (ii)
251 n. 8 Respect Constraint 23, 25, 32, 36–7, 40,
Putnam, Hilary 138, 182, 248–50 57, 61, 63, 67, 73–83, 196, 230, 273,
Pylyshyn, Zenon 5 n. 5, 45 n. 1, 46 n. 3, 274; see also psychological reality,
49, 50, 51–2, 54, 63 n. 3, 69, 76, 84, some possible positions on, (m); rules
96, 196, 228–9; on RT 197, 208, (or principles), ‘‘respected’’ by
209, 228 competence (final state)
Pylyshyn’s Razor 12–13, 14, 51–2, 56, 60, Rey, Georges viii, 7 n. 9, 12, 63 n. 2,
71, 75, 79, 107, 117, 142, 191, 198, 112 n. 27, 126 n. 2, 135, 144 nn.
204, 206, 211, 216, 221, 224, 239, 5–6, 151 n., 163, 184–90, 225
243, 269 Reynolds, D. R. 20 n. 3
Richard, Mark 145 n. 7
Quine, W. V. 26, 37 n. 31, 87, 125 n., Richardson-Klavehn, Alan 212
143, 187 Riley, J. R. 20 n. 3
Robbins, Philip viii
Radford, Andrew 82 n., 95 n., 101 Roelofs, Ardi 231–3
Ramsey, William viii, 56 n. 14, 147 n. 11 Rosenbaum, David A. 211, 212
‘‘rational–causal’’ processing, see Rosenbloom, P. S. 214 n. 17
‘‘brute–causal’’ processing vs rules (or principles): represented vs.
‘‘rational–causal’’ processing ‘‘simply embodied’’ 7, 9, 45–52,
Rattan, Gurpreet viii 58–61, 63–4, 196–7, 273;
Rayner, K. 237 represented and used as data 53,
realism: about language 8, 12, 26–8, 35, 57–8, 60–1, 273; see also
81, 109 n. 19, 184–9; about psychological reality, some possible
thoughts, see intentional realism; see positions on, (ii); ‘‘respected’’ by
also nominalism competence (final state) 9, 21–3, 25,
Reber, A. S. 114 n. 32, 115, 211 n., 32–3, 35–9, 45 n. 2, 57, 61, 63, 67,
216–17, 219, 221 n., 256 73–83, 99, 117, 166, 195–6, 205,
reference: borrowing 138–141, 182; doubts 239, 240, 243 n., 246, 273, 274; see
about 15, 24 n. 10; fixing 138; also psychological reality, some
theories of 5, 138–9, 155–6 possible positions on, (m);
Index 303
‘‘respected’’ by initial state 246–52, Sneider, Wolfgang 211
256–61, 265, 269, 271 275; Soames, Scott 8, 28, 33, 92 n., 96, 97 n.,
structure vs processing 18–21, 24–5, 118 n., 247
273; see also psychological reality, Sober, Elliott 45 n. 1, 64
some possible positions on, (iii) Sokolov, A. 150
Rumelhart, D. E. 214 Sosa, Ernest 105 n. 14
Rumfitt, Ian 89 n. 2 speaker meaning 128, 132–4, 140,
155–8, 179–90, 225–6 see also
Saffran, E. M. 233 conventional meaning and syntax;
Sag, Ivan 8 n. 11, 38 n. 33 literal meaning and syntax
Sampson, Geoffrey, 8 n. 12 Spencer, N. J. 111 n. 25, 115 n. 36
Samuelson, L. K. 250 n. Sperber, Dan 94 n. 6, 128 n. 4
Sapir, Edward 135 Stabler, Edward P. 8 n. 12, 50–1, 53, 58,
Saussure, Ferdinand de 24 nn. 9–10, 81, 197, 204, 209, 268 n.
184 n. Stainton, Robert viii
Schacter, Daniel L. 211 n., 212 Stanley, Jason 89 n. 2, 119 n. 40
Schank, Roger 49, 65 n., 177 n. 16, 197 Stanley, W. B. 116 n., 217
Schiffer, Stephen 89 n. 2, 133 nn. 18–19, Stecker, R. 92 n., 96, 97 n., 118 n., 247
180 Steedman, Mark 237, 238, 261–2
Scholz, Barbara viii, 249 n., 251 n. 8, Stemberger, Joseph P. 232
255 n. Sterelny, Kim v, viii, 5 n. 6, 8, 24 n. 10,
Schutze, Carson T. 100 30 n. 21. 32 n. 24, 36 n. 30, 37 n.
Schwartz, M. F., 233 31, 39 n. 35, 89 n. 1, 94 n. 6,
Schwitzgebel, Eric 104 104 n., 113 n. 30, 119, 125 n.,
Searle, John 161 n. 132 n. 16, 133, 135–6, 138 n.,
Segal, Gabriel 96, 176 n. 12 156 nn. 22 and 24, 157 n. 27,
Seidenberg, Mark S. 172 n. 184 n., 186 n. 27, 189 n.
sentence machine 25, 64–5, 68 Stich, Stephen viii, 91, 94 n. 7, 113 n. 30,
sequential (or serial) reaction time 248
(SRT) 217 Stone, Tony 204 n.
Shankweiler, Donald 237 structuralism 18, 26–7, 186 n. 27; post-
sign languages 131, 149–50, 167, 186 n. 27
250–1 Studdert-Kennedy, Michael 130 n. 13
Simon, H. 215 n., 257 n. 14 sui generis, language as 248, 251, 258,
Singley, Mark K. 215, 257 n. 14 261, 262, 271
skills and abilities, cognitive and Sun, Ron 47, 107 n. 16, 211–12,
motor 6–7, 12, 14, 21 n. 5, 45–52, 214–15, 217, 221 n. 23, 227
89–94, 106–11, 204, 210–20, Supalla, Ted 15, 250
221–30, 232–43, 245, 248, 250, Swinney, David A. 167
256, 262–4, 266–70, 272 Szabo, Zoltan G. 119 n. 40
Smith, A. D. 20 n. 3
Smith, Barry C. 39, 178 n. 19 Taatgen, Niels A. 221 n.
Smith, Brian Cantwell 56 n. 15 Tanenhaus, Michael K. 129 n. 7, 235–8,
Smith, L. B. 250 n. 241
Smith, Mark 226 n. 27 Taylor, Kenneth A. 126 n. 2
Smith N. 94 n. 6, 164 temporal consequences of priority claims
Smolensky, Paul 56 about thought 10–11, 141, 161,
304 Index
temporal consequences (cont.): Tomberlin, James E. 143 n. 4
174, 191; for acquiring competence Tooby, J. 94 n. 6
in a language 130, 133, 137; for Tovey, P. 92 n., 96, 97 n., 118 n., 247
conventional meanings and transcendental arguments 125–6
syntax 133, 137, 160; for innate Tresilian, James R. 212 n. 15, 213
linguistic capacities 130–1; see also Trueswell, John C. 129 n. 7, 236–8, 241
major conclusions, fifth Tsimpli, I-M. 94 n. 6, 164
tentative proposals: Turvey, M. T. 214
first 11, 13, 14, 152, 157, 162, 172,
190, 191, 208 n., 243, 265, 276; Universal Grammar (UG) 171, 184,
argument for 142–62 244–8, 252–6, 260 n.; rules (or
second 11, 14, 173, 190, 223, 261, principles) specified by, (UG-
265, 266, 272, 276; argument rules) 13–14, 159, 244–52, 256–61,
for 163–74, 256–66 265–72, 273, 275, 276
third 12, 207, 226, 229, 233, 240, 260, Uriagereka, Juan viii, 255 n.,
276; argument for 18–21, 24–5, Use/mention sloppiness, 69–71, 74
37, 53, 59, 62–8, 76–7, 190–1,
207–8, 220–43 Van Gelder, Tim 213, 214 n. 16
fourth 12, 220–1, 223, 229, 230, 233, Vigliocco, Gabriella 151, 231–3
239, 240, 241 n., 243, 276; Vinson, Daniel P. 151, 231–3
argument for 220–43 Von Frisch, Karl 20, 29 n. 18
fifth 13, 14, 243, 260, 265, 276;
argument for 195–243 Wade, Nicholas 250–1
sixth 13, 14, 257, 259, 260, 266, 269, Wang, Hongbin 215, 218, 227, 266 n.
271, 276; argument for 256–60 Wann, John P. 212 n. 15, 213
seventh 13, 14, 260, 266, 269, 271, Wanner, Eric 151
276; argument for 256–60 Webelhuth, Gert 70, 183 n. 23
Tettamanti, M. 168 Weinberg, Amy 10, 32 n. 25, 36, 65,
thinking 50, 106–8, 111–12, 116, 118, 79–81, 83–4, 195–6, 200 n., 222
145–7, 150, 153, 161, 165, 211, wff machines and wffs 19–20, 24–5,
216, 226–7, 258, 263, 268–9, 272; 64–5, 68
see also thoughts Wheeldon, Linda R., 226 n. 27
Thompson, Henry 58 n. 16 Wheeler, Michael 94 n. 6
Thornton, Rosalind 99, 110 n. 24 Whorf, Benjamin Lee 135–7, 158
thoughts: realism about, see intentional Williamson, Timothy 89 n. 2
realism; nature of, viii, 6, 11, Wilson, Deidre 128 nn. 4–5
13–14, 46–7, 60, 125, 127, 142–62, Wilson, Robert A. 56 n. 15
174–8, 190–2, 201–3, 205–6, 226–7, Wisniewski, Edward J. 104–5, 235 n. 33
228, 247, 257–60, 268–9, 271, 273, Wolpert, Daniel M. 213
276; see also language expresses
thought (LET); priority of theoretical Yamada, Jeni E. 164
interest in thought over language; Yamadori, Atsushi 217–18
priority of thought over language, Yamashita, Hikari 217–18
explanatory; temporal consequences Yoshida, Takashi 217–18
of priority claims about thought;
tentative proposals, first and sixth Zhang, Jiajie 215, 218, 227, 266 n.
Tienson, John 51 Zheng, Ming-Yu 131

Michael Devitt - Ignorance of Language

Uploaded by

Copyright:

Available Formats

Michael Devitt - Ignorance of Language

Uploaded by

Document Information

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

Michael Devitt - Ignorance of Language

Uploaded by

Copyright:

Available Formats

I G N O R A N C E O F L A N G UAG E

This page intentionally left blank

CLAR ENDON PR ESS O X F O R D

As one would expect, I received a great variety of responses to this material.

I. LINGUISTICS IS NOT PSYCHOLOGY

II. POSITIONS ON PSYCHOLOGICAL REALIT Y

III. ‘‘ PHILOSOPHICAL’’ ARGUMENTS FOR

IV. THE REL ATION OF L A NGUAGE

Glossary of Named or Numbered Items 273

I. LINGUISTICS IS NOT PSYCHOLOGY

2. A Grammar as a Theory of Linguistic Reality 17

II. POSITIONS ON PSYCHOLOGICAL REALIT Y

4. Some Actual Positions on Psychological Reality 62

IV. TH E R E L ATIO N OF L A NGUAGE TO THO UGHT

8. Thought Before Language 125

9. A Case for the Psychological Reality of Language 142

10. Thought and the Language Faculty 163

12. Language Acquisition 244

Glossary of Named or Numbered Items 273

The wonderfully successful research program in linguistics initiated and sus-

1.2 THE PLAN

My earlier argument now seems to me to have many errors. So my Wrst aim, in

In Chapter 4, I attempt to place some historically interesting actual positions

(i) A person’s I-language, according to Chomsky, supervenes on intrinsic prop-

15 These doubts are nicely expressed and motivated in Pietroski 2003.

A competence is a competence to produce a certain sort of output/product; or it

The theory of a competence explains what it is about an object that makes it

linguistic rules, we could build a ‘‘sentence machine’’ that generated sentences

Whether or not this study of the outputs of competence is the study of an

28 A similar assumption yields a similar conXation in some philosophers of language; hence

If this discussion is right, it has a great deal of methodological signiWcance. I have

Linguistics has something worthwhile to study apart from psychological reality of

Figure 1. Redrawn from Boden, 1984, p. 153.

What is the contrast? What operations do not manipulate structured repre-

3.4 SOME POSSIBLE POSITIONS ON

What is Chomsky’s view of the psychological reality of language? This question is

idea of running the process forward in production. Still, running transformations

6 Is a grammar a theory of the speaker’s competence at Christopher Peacocke’s ‘‘level 1.5’’, as he

There is an obvious problem in attributing position (ii) to Chomsky: we need to

Grammar ! structural descriptions

4.4 BRESNAN AND KAPLAN

Respect Constraint. Linguistic rules that are accurately described by a grammar

Finally, let us consider the position of Robert Matthews, a philosopher who is

position should be uncontroversial. But the view that linguistics is part of

In calling the arguments in this part ‘‘philosophical’’, I do not mean to suggest

1 1981: 95–103; Devitt and Sterelny 1989.

It seems to me that Chomsky’s tactic in this dispute with the philosophers—

4 See Crain et al. 2005, section 4, for a nice summary.

A simplistic version of the standard Cartesian explanation would be that since

5 Fodor, Bever, and Garrett 1974: 82; Baker 1978: 4–5.

7.4 INTUITIONS IN GENERAL

7.5 LINGUISTIC INTUITIONS

39 Graves et al. dismiss the possibility of such an explanation (1973: 326–7).

7.8 MUST LINGUISTICS EXPLAIN INTUITIONS?

Wnd that an epistemological view can be epistemically relevant to linguistics.

Linguists greatly exaggerate the evidential role of the intuitive judgments of

8.1 INTENTIONAL REALISM

Intentional realism is my Wrst relatively uncontroversial assumption. The next is

8.5 PRIORITY OF THEORETICAL INTEREST IN

20 I draw, once again, on my 1996, secs. 2.3–2.8.