Truth and Words

This page intentionally left blank

Truth and Words

Gary Ebbs

1
1
Great Clarendon Street, Oxford  
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 offices 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
 Gary Ebbs 2009
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2009
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 Cataloging in Publication Data
Ebbs, Gary.
Truth and words / Gary Ebbs.
p. cm.
Includes bibliographical references and index.
ISBN 978–0–19–955793–6
1. Truth. 2. Predicate (Logic) 3. Language and logic. 4. Grammar, Comparative and
general—Sentences. I. Title.
BC171.E23 2009
306.4401—dc22
2008048453
Typeset by Laserwords Private Limited, Chennai, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd, King’s Lynn, Norfolk

ISBN 978–0–19–955793–6

10 9 8 7 6 5 4 3 2 1
To Hilary Putnam
This page intentionally left blank
Preface

This book is an attempt to fuse together two apparently independent

ideas—the idea that truth and satisfaction are disquotational and the idea
that unless we see good reason in a given context for not doing so, we
are entitled to trust the non-deliberative identifications of sentences and
words on which we rely when we take ourselves to agree or disagree with
others, to learn from what they say, or to express a new discovery. It is
not my first attempt to link these ideas. In Rule-Following and Realism (Ebbs
1997), I recommended that we describe what it is to share a language
from our perspective as participants in actual linguistic interactions with
other speakers, and proposed that we take such descriptions to license us
to apply our disquotational definitions of truth and satisfaction directly to
other speakers’ words and to our own words as we used them in the past.
I still think this proposal points in the right direction. But I now think it
does not go far enough. If we wish to describe our linguistic practices in
a way that fits with a disquotational account of truth and satisfaction, we
need a more radical approach.
My approach in this book is shaped at every step by the assumption that
we desire to clarify and facilitate our inquiries by regimenting our sentences
and formulating logical generalizations. Given this assumption, I ask, ‘‘Do
we need a truth predicate?’’ and ‘‘If so, what sort of truth predicate do we
need?’’ I answer these questions in a way that clarifies the non-deliberative
identifications of words on which we rely when we take ourselves to
agree or disagree with others, to learn from what they say, or to express
a new discovery. I explain how to resist our tendency to think of words
as individuated by their spellings or pronunciations together with facts
about how speakers use them, and show how to construct an alternative
conception of words that fits with the non-deliberative identifications of
words on which we rely in our inquiries. I propose that we combine
this alternative conception of words with our disquotational definitions of
truth for sentences as we now use them. The result radically transforms
our understanding of truth and related topics, including anti-individualism,
self-knowledge, and the intersubjectivity of logic.
viii 

During the planning and writing of this book I was fortunate to receive
excellent advice and criticism from many friends and colleagues. When I
started planning the book in 1998, Tom Ricketts warned me, in a tone of
voice that he reserves for his most urgent and provocative philosophical
remarks, ‘‘There will have to be regimentation.’’ I knew he was right, but
it took me years to figure out exactly how to fit regimentation into my
account (see Chapters 1 and 3). Also in 1998, Hilary Putnam convinced
me that Tarski’s definitions of ‘truth-in-L’ for sentences individuated ortho-
graphically (as strings of letters and spaces) are at best incomplete because
they leave us in the dark about the relationship between those definitions
and our ordinary non-deliberative applications of truth to other speak-
ers’ utterances. Yet another central part of the book started coming into
focus in 1997 and 1998 when I first formulated various versions of my
gold–platinum thought experiment (see Chapter 6). Conversations with
Adrian Cussins, Brian Loar, Tom Ricketts, Mark Wilson, and George
Wilson were especially helpful to me at that stage, but I also learned a great
deal from discussions at Arizona State University, Florida State Univer-
sity, Grinnell College, UC-San Diego, UC-Irvine, University of Illinois
at Urbana-Champaign, and Smith College, where I presented papers that
feature the gold–platinum thought experiment. In my 2000 Central APA
symposium paper ‘‘Denotation and Discovery’’, I further developed my
view of sameness of satisfaction across time. Mark Wilson was the com-
mentator for that paper; his comments prompted me to clarify how my view
applies to natural kind terms found in texts written centuries ago. Starting
around 2003, I had several helpful conversations with Henry Jackman,
whose temporal externalism is in some ways similar to my view of sameness
of satisfaction across time, but in other ways fundamentally different from
it (see Chapter 8). In my 2004 Eastern APA symposium paper ‘‘Truth
and Words’’, I first presented an alternative to the standard conception
of words. This time Steven Gross was the commentator; he urged me to
clarify my pragmatic grounds for thinking that a disquotational definition
of ‘true-in-L’ is satisfactory only if it implies that we are directly licensed
to apply ‘true-in-L’ to other speakers’ sentences and our own sentences
as we used them in the past. Steven later sent me references to a number
of articles that I found very helpful when I wrote Chapter 4. From 2003
to 2006, I participated in a reading group on linguistics and philosophy of
language with Peter Lasersohn and Lenny Clapp. In bi-weekly discussions
  ix

with Peter and Lenny I learned a great deal about recent work in formal
semantics, and gradually developed my current view (see Chapter 1) that
much of this work can be detached from the controversial explanatory
ambitions of its various authors and pressed into service as part of a prag-
matic account of regimentation. In the spring semesters of 2005 and 2007,
I presented drafts of the book in seminars at the University of Illinois
at Urbana-Champaign and Indiana University, respectively; the graduate
students in these seminars raised many challenging criticisms that helped me
to clarify my arguments. In the final stages of the writing I received excel-
lent criticisms and advice from Katy Abramson, Imogen Dickie, Michael
Glanzberg, Steven Gross, David Hills, Henry Jackman, John MacFarlane,
Joan Wiener, and two anonymous readers for Oxford University Press.
Many others also raised objections or gave me advice that helped me to
write this book, including Ben Bayer, Stephen Biggs, Susan Blake, Bill
Brewer, Kyle Broom, Jessica Brown, Nancy Cartwright, Hugh Chandler,
Yajun Chen, Lenny Clapp, Jim Conant, Mike Dunn, Daniel Estrada,
Anthony Everett, Kit Fine, David Finkelstein, Brie Gertler, Sandy Gold-
berg, Warren Goldfarb, Robert Gooding-Williams, Tara Gilligan, Richard
Heck, Jon Jarrett, Darryl Jung, Mark Kaplan, John Koethe, Scott Kim-
brough, Phil Kitcher, Phil Kremer, Michael Kremer, Wolfgang Künne,
Mark Lance, Peter Lasersohn, Michael Liston, Adam Leite, Brian Loar,
Pen Maddy, Ruth Barcan Marcus, Patricia Marino, Mohan Matten, Tim
McCarthy, David McCarty, Art Melnick, Tom Meyer, Nathalie Morasch,
Michael Morgan, Erica Neely, Charles Parsons, Terry Parsons, Oliver
Pooley, Michael Resnick, Sam Rickless, Bill Robinson, Joseph Rouse,
Dick Schacht, Fred Schmitt, Peter Schwartz, David Shwayder, Sanford
Shieh, Barry Smith, Tom Stoneham, Tadeusz Szubka, Alessandra Tanesini,
William Taschek, Kevin Toh, Charles Travis, Steve Wagner, and Chuang
Ye. Warm thanks to all.
Most of the material in Chapter 5 was previously published under the
title ‘‘Learning from Others’’ (Ebbs 2002a), and Chapter 6 draws heavily
from my papers ‘‘The Very Idea of Sameness of Extension across Time’’
(Ebbs 2000) and ‘‘Denotation and Discovery’’ (Ebbs 2003). I thank the
editors and publishers of these articles for their permission to include large
parts of them in this book.
I received a fellowship at the Center for Advanced Studies at the
University of Illinois at Urbana-Champaign (UIUC) to write a first draft
x 

of this book (2001–2), several travel grants from UIUC that made it
possible for me to present some of the central arguments in this book
at other institutions, two grants from the Freeman Fellowship Exchange
Program to present some of these arguments at universities in China (2002,
2004), and funding from the British Academy that covered the costs of my
attendance at Tom Stoneham’s workshops on sameness of extension across
time (2004, 2005). I am grateful to these institutions and programmes for
their support.
I thank Peter Momtchiloff for his early interest in this project and his
patience while I worked on it.
Above all, I thank my wife, Martha, for her unfailing support and
understanding throughout this long project.
Gary Ebbs
Bloomington, Indiana
23 September 2008
Contents

Introduction 1

. Regimentation 14
1.1. Regimentation as Linguistic Policy 14
1.2. Ambiguity 17
1.3. Is Regimentation Possible? 20
1.4. Vagueness 22
1.5. Quantifier Domains, Tense, and Time 25
1.6. Descriptions and Proper Names 27
1.7. Pronouns and Demonstratives 30
1.8. Why Ordinary Language is Indispensable 32
1.9. Limitations of First-Order Logic 33

. The Tarski–Quine Thesis 40

2.1. The Indispensability Argument 40
2.2. Why Generalize on Valid Sentences? 48
2.3. Three Attempts to Generalize on Sentences without
Using a Truth Predicate 52
2.4. Horwich’s Minimal Theory 58
2.5. A Naive Theory of Why it is Epistemically Reasonable for
us to Accept T-Sentences 63
2.6. Surrogate T-Sentences and Explication 65
2.7. Tarski’s Convention T 67
2.8. ‘True-in-L’ Defined in Terms of Satisfaction 70
2.9. How (Tr) Satisfies Convention T and Enables us to
Derive ST-Sentences 72
2.10. Schematic Definitions of ‘True-in-L’ Rejected 74
2.11. Adopting the Tarski–Quine Thesis 78
2.12. Two Objections 79
xii 

. The Intersubjectivity Constraint 82

3.1. A Preliminary Formulation of the Intersubjectivity
Constraint 82
3.2. Practical Identifications of Words (PIWs) 84
3.3. Practical Judgements of Sameness of Satisfaction (PJSSs) 87
3.4. Agreement and Disagreement 90
3.5. Learning from Others 93
3.6. Discoveries 94
3.7. A Reformulation of the Intersubjectivity Constraint 95
3.8. Trust without Trustworthiness? 97
3.9. A Quinean Objection: PJSSs are not Factual 98
3.10. Realism as Integral to the Semantics of the Predicate
‘True’ 102

. How to Think about Words 105

4.1. Is the Tarski–Quine Thesis Incompatible with the
Intersubjectivity Constraint? 105
4.2. Use versus Mention (Transparent Use) 106
4.3. The Orthographic Conception of Words 107
4.4. Explanatory Use (Ex-Use) 108
4.5. The Token-and-Ex-Use Model of Words 112
4.6. Types and Tokens 114
4.7. Kaplan’s Common Currency Conception of Words 120
4.8. The Context Principle and the PJSS-Based Conception of
Words 127
4.9. How to Satisfy the Intersubjectivity Constraint without
Rejecting the Tarski–Quine Thesis 133
4.10. Preliminary Objections and Replies 140

. Learning from Others, Interpretation, and Charity 144

5.1. Is the Intersubjectivity Constraint Compatible with the
Negation of the Tarski–Quine Thesis? 144
5.2. Language Ex-Use and Interpretation 145
5.3. A Case in which One Person Learns from Another 148
5.4. Two Conditionals 151
  xiii

5.5. Strategy 153

5.6. What is Davidson’s Principle of Charity? 154
5.7. Davidson’s Framework for Evaluating (3) and (4) 158
5.8. Why the Conjunction of (3) and (4) Violates Davidson’s
Principle of Charity 160
5.9. My Conclusion Drawn, Generalized, and Explained 165
5.10. Is the Principle of Charity Optional? 169
5.11. An Alternative to Davidson’s Principle of Charity 171
5.12. Frontiers of Translation 174
5.13. The Method behind these Conclusions 177

. A Puzzle about Sameness of Satisfaction across Time 179

6.1. An Intuition about Sameness of Satisfaction across Time 179
6.2. Methodological Analyticity 182
6.3. Causal-Historical Theories 186
6.4. A Thought Experiment 191
6.5. The Standard Conception of the Options for the Thought
Experiment 194
6.6. A Preview of why Options (1) and (2) are Unacceptable 196
6.7. A Dilemma for the Causal-Historical Theory 197
6.8. Dispositions 201
6.9. Epistemic Possibilities and Primary Intensions 204
6.10. Problems with Primary Intensions 208
6.11. Implicit Conceptions 213

. Sense and Partial Extension 217

7.1. Option (3) 217
7.2. Dummett on Sameness of Satisfaction across Time 218
7.3. Dummett on Sense and Implicit Knowledge 220
7.4. Why Dummett’s Constraints Rule out Options (1) and (2) 225
7.5. Dummett’s Version of Option (3) 228
7.6. Two Problems for Dummett’s Version of Option (3) 230
7.7. Field on Partial Extension 236
7.8. A Field-Style Argument against Options (1) and (2) 238
7.9. A Field-Style Defence of Option (3) 240
7.10. A Problem for the Field-Style Defence of Option (3) 241
xiv 

. The Puzzle Diagnosed and Dissolved 247

8.1. The Puzzle Reviewed 247
8.2. The First Gold–Platinum Thought Experiment and the
Context Principle 249
8.3. The Second Gold–Platinum Thought Experiment 252
8.4. The Thesis that the Extension of a PJSS-Based Word
Type is Determined by Facts about the Ex-Uses of Some
of its Tokens 257
8.5. George Wilson’s Riverdale Olympics Case 260
8.6. Henry Jackman’s Temporal Externalism 264
8.7. Counterfactuals about the Past and the Third
Gold–Platinum Thought Experiment 269
8.8. Temporal Externalism and Relative Truth 276
8.9. Immanent Realism 286

. Applications and Consequences 288

9.1. Introduction 288
9.2. A Deflationary Alternative to the Causal Theory of
Reference: Predicates 289
9.3. A Deflationary Alternative to the Causal Theory of
Reference: Proper Names 291
9.4. What is Minimal Self-Knowledge? 294
9.5. Minimal Self-Knowledge as Second Order 295
9.6. Basic Self-Knowledge and Containment 297
9.7. Minimal Self-Knowledge as First Order 303
9.8. Minimal Self-Knowledge as Practical Knowledge 306
9.9. The Division of Epistemic Labour 308
9.10. Judging Minimal Linguistic Competence across Time 311
9.11. Anti-Individualism, Externalism, and Linguistic
Communities 313
9.12. Truth and Logic 316

References 320
Index 331
 Introduction

My goal in this book is to develop an account of truth that we can use to

clarify and facilitate our inquiries. Unlike most recent accounts of truth,
the goal of my account is not to analyse a concept we express when we use
the ordinary English word ‘true’, to say what ‘true’ means, or to explain all
our uses of it. Instead, my account of truth is addressed to those of us who
desire to clarify and facilitate our inquiries by regimenting our sentences
so that we can express logical generalizations, such as ‘‘Every sentence of
the form ‘S ∨ ∼ S’ is true’’. I ask, ‘‘Do we need a truth predicate to
express logical generalizations?’’ and ‘‘If, so, what sort of truth predicate
do we need?’’ I answer these questions by showing how to reconcile two
appealing yet apparently incompatible answers to them.

The Tarski–Quine Thesis

The first answer, due in essentials to Alfred Tarski and W. V. Quine,
is that to express logical generalizations we need a disquotational truth
predicate—one that entitles us to affirm such sentences as
‘Snow is white’ is true if and only if snow is white,
which are instances of the disquotational pattern
(T) ‘ ’ is true if and only if .
This answer takes for granted that the regimented part of our language
comprises infinitely many sentences, so we cannot express our acceptance
of all regimented sentences of a given form, such as ‘S ∨ ∼ S’, by writing
out and affirming an infinitely long conjunction of all of the instances of that
2 

form. To express our acceptance of all those sentences and thereby affirm
a logical generalization we therefore need some practical way to affirm, in
effect, an infinite conjunction of sentences. And for this purpose, according
to the first answer, we need a truth predicate that entails all and only
(non-paradoxical) instances of (T).¹ Following Tarski, we can each satisfy
this need by replacing ‘true’ with ‘true-in-L’, where L is the regimented
part of our language, and defining ‘true-in-L’ in terms of satisfaction. To
define ‘true-in-L’ for quantified sentences, we must define satisfaction in
terms of sequences of objects. I shall explain how to define satisfaction in
terms of sequences for a regimented fragment of English in Chapter 2. For
present purposes, however, it is enough to see that we can each define
something like satisfaction for our own one-place predicates by accepting
such sentences as
x satisfies ‘white’ if and only if x is white,
which are instances of the disquotational pattern
(S) x satisfies ‘ ’ if and only if x is .
When our definitions of satisfaction for predicates are disquotational in this
way, I call the resulting Tarski-style definition of ‘true-in-L’ disquotational.²
Without supplementation, a Tarski-style disquotational definition of
‘true-in-L’ applies only to the sentences for which it is disquotationally
defined. But this very limitation of a Tarski-style disquotational definition
of ‘true-in-L’ is what guarantees that the definition is clear. As Quine
emphasizes, a disquotational definition of ‘true-in-L’ has ‘‘every bit as
much clarity, in any particular application, as is enjoyed by the particular
expressions . . . to which we apply it. Attribution of truth in particular to
‘Snow is white’, is every bit as clear to us as attribution of whiteness to
snow’’ (Quine 1961a: 138). If our goal is conceptual analysis, it may seem
that this clarity comes at too high a price—the price of changing the
subject from a concept (if there is one) expressed by our ordinary English
word ‘true’ to an ersatz concept expressed by ‘true-in-L’ that may be

¹ One might object that we can express logical generalizations without using a Tarski-style truth
predicate. I raise and answer several versions of this objection in Chapter 2.
² What I call a Tarski-style disquotational definition of satisfaction also typically includes disquota-
tional satisfaction clauses for two- (and, in general, n-) place predicates, as well as non-disquotational
clauses for logical constants, such as ‘not’, ‘or’, and ‘all’. I give the details in Chapter 2.
  3

useful for some purposes, but does not deserve to be called truth. If our
leading questions are ‘‘Do we need a truth predicate?’’ and ‘‘If so, what
sort of truth predicate do we need?’’, however, then the utility and clarity
of Tarski-style disquotational definitions of ‘true-in-L’ might motivate
us, following Quine, to use them in place of our ordinary English word
‘true’, and thereby to endorse what I call the Tarski–Quine thesis: there
is more to truth than what is captured by a Tarski-style disquotational truth
predicate.

The Intersubjectivity Constraint

According to the second answer, we need a truth predicate that we are
licensed to apply not only to our own sentences, as we now use them, but
also to other speakers’ sentences and to our own sentences as we used them in the
past. This answer apparently conflicts with the Tarski–Quine thesis. To see
why, note that when we are defining truth for our own sentences as we
now use them, we may stipulate that in, for instance,
‘Snow is white’ is true if and only if snow is white,
the sentence that appears within quotation marks on the left of ‘if and
only if ’ is the same (in relevant respects) as the one that appears without
quotation marks on the right. In contrast, if another speaker utters ‘Snow
is white’, we cannot stipulate that the sentence she utters is the same (in
relevant respects) as our own sentence ‘Snow is white’, as we now use it,
and so we have no formal guarantee that her sentence is true if and only
if snow is white (Putnam 1983, Davidson 1994a). For the same reason,
we have no formal guarantee that our own sentence ‘Snow is white’,
as we used it in the past, is true if and only if snow is white. Hence
it seems that a Tarski-style disquotational truth predicate is not defined
disquotationally for others’ sentences or our own sentences as we used them
in the past. As I mentioned above, however, without supplementation,
a Tarski-style disquotational definition of ‘true-in-L’ applies only to the
sentences for which it is defined disquotationally. Hence it seems that
without supplementation a Tarski-style disquotational definition of ‘true-
in-L’ does not apply to others’ sentences or our own sentences as we used
them in the past.
4 

If the problem were one of how to analyse the concept of truth, a

defender of the Tarski–Quine thesis could simply deny that a full account
of the concept of truth must include an account of the extension of
‘true-in-L’ for other speakers’ words or his own words as he used them
in the past. To avoid fruitless disputes about how to capture a supposedly
unique concept we express when we use the ordinary English word ‘true’,
however, we should focus instead on the question ‘‘Do we need a truth
predicate that applies to others’ sentences and to our own sentences as
we used them in the past?’’ To this question, unlike the one about how
to analyse the concept of truth, it is relevant that we typically want to
formulate and apply logical generalizations partly because we want to
collaborate fruitfully with others. As Gottlob Frege emphasized (Frege
1964: 17), to collaborate fruitfully with others we need to formulate logical
laws that can serve as arbiters of conflicting beliefs. For this purpose, we
need a truth predicate that applies not only to our own sentences as we now
use them, as the Tarski–Quine thesis suggests, but also to other speakers’
sentences and our own sentences as we used them in the past.
It might seem that a person who is only interested in her own solitary
inquiries and does not desire to collaborate with others need not worry
about this pragmatic objection to the Tarski–Quine thesis. But even a
person who does not desire to collaborate with others presupposes that she
may someday discover that a sentence she previously accepted is false. And
to make sense of this presupposition, she needs a truth predicate that she
can apply to her own sentences as she used them in the past.
Judged by these pragmatic criteria, it seems that a Tarski-style disquo-
tational truth predicate is not by itself satisfactory. The pragmatic criteria
therefore motivate us to adopt what I call the intersubjectivity constraint: a
Tarski-style disquotational truth predicate is satisfactory only if it is supplemented by
an account of why it is epistemically reasonable for one to apply it to other speakers’
sentences and to one’s own sentences as one used them in the past.

Practical Judgements of Sameness of Satisfaction

(PJSSs)
To clarify the intersubjectivity constraint, we need to examine how we
identify other speakers’ words and ascribe satisfaction conditions to them.
  5

Suppose my friend Alice is telling me about a house she would like to buy.
I have not seen the house, and ask her ‘‘What colour is it?’’ She replies,
‘‘It’s white.’’ In this context, without hesitation or deliberation, I will take
her to have said that the house is white, and in so doing, I will take her
word ‘white’ to be the same as my word ‘white’—the word I would use
to express what I take her to have said. In short, I will accept
(1) Alice’s word ‘white’ is the same as my word ‘white’.
If I also accept
(2) x satisfies my word ‘white’ if and only if x is white,
then I commit myself to
(3) x satisfies Alice’s word ‘white’ if and only if x is white.
And by accepting (2) and (3), I commit myself to accepting their conjunc-
tion, namely,
(4) (x satisfies my word ‘white’ if and only if x is white) and (x satisfies
Alice’s word ‘white’ if and only if x is white)
A commitment of this kind is what I call a practical judgement of sameness
of satisfaction (or a PJSS, for short). I call such judgements practical because
they are exercises of a learned yet typically non-deliberative ability that is
acquired over a period of years. I call an exercise of this ability a judgement
because we will reject it if we come to think it is wrong. I will reject (4),
for instance, if I come to make other PJSSs for Alice’s word ‘white’ that
conflict with and discredit (4).
As this example suggests, we regard our PJSSs as true or false, and
we assume it is epistemically reasonable for us to trust sentences that
express our PJSSs unless we have particular, local reasons in a given
context for not doing so. In short, we regard our PJSSs as both factual
and trustworthy. In so doing, we take them to license our applications
of our own Tarski-style disquotational truth predicates to other speakers’
sentences and to our own sentences as we used them in the past. We may
therefore reformulate the intersubjectivity constraint as follows: a Tarski-
style disquotational truth predicate is satisfactory only if it is supplemented by an
account of why it is epistemically reasonable for one to regard one’s PJSSs as both
factual and trustworthy.
6 

The Token-and-Ex-Use Conception of Words

The intersubjectivity constraint appears incompatible with the Tarski–
Quine thesis, for reasons I sketched above: it seems that without sup-
plementation, a Tarski-style disquotational definition of ‘is true’ applies
only to our own sentences, as we now use them. Hence it seems that to
supplement a Tarski-style disquotational truth predicate with an account
of why it is epistemically reasonable for one to regard one’s PJSSs as both
factual and trustworthy would be to commit ourselves to the thesis that
there is more to truth than what is captured by a Tarski-style disquotational
truth predicate, and thereby to abandon the Tarski–Quine thesis. In my
view, however, to understand why we feel so sure that the constraint is
incompatible with the Tarski–Quine thesis, we must reflect on how we
conceive of words.
When we first ask ourselves what words are, we are strongly inclined to
start with an orthographic conception of word types, according to which
the ink marks ‘white, white’, for example, comprise two tokens of the
same word type w-h-i-t-e, and the word tokens themselves are strings of
tokens of letter or sound types. But from this perspective, word tokens
appear lifeless and insignificant. How do they come to have meanings and
satisfaction conditions? The only answer, it seems, is that there are facts
about the word tokens—facts that are additional to and independent of the
tokens’ spellings or pronunciations—that determine whatever meanings
and satisfaction conditions they have.
Facts about a word token that are supposed to determine whatever
meanings and satisfaction conditions it has are facts about what I shall call
the token’s explanatory-use (ex-use, for short). Every substantive theory of
meaning and reference includes an account of ex-use. According to some
of these theories, for instance, the meaning and satisfaction conditions of
a given word token are determined by facts about why the word token
was uttered, together with facts about the speaker’s mental states and
causal-historical relations between the word token, objects in the speaker’s
environment, and other speakers’ word tokens. Such theories appeal to us
because we presuppose

The token-and-ex-use conception of words. Two word tokens are of the

same word type if and only if
  7

(a) they are spelled or pronounced in the same way, and

(b) facts about the ex-uses of the word tokens determine that they have
meanings and satisfaction conditions and that their meanings and
satisfaction conditions are the same.
A crucial consequence of the token-and-ex-use conception of words is
(P) Sentences that express our PJSSs are factual (true or false) only if
they have truth values that are determined by facts about the ex-uses
of the word tokens they contain.
For instance, recall (4), the sentence that expresses my PJSS for Alice’s
word ‘white’:
(4) (x satisfies my word ‘white’ if and only if x is white) and (x satisfies
Alice’s word ‘white’ if and only if x is white)
The token-and-ex-use conception of words implies that (4) is factual only
if it has a truth value that is determined by facts about the ex-uses of Alice’s
and my tokens of the orthographic word type w-h-i-t-e.
This consequence of the token-and-ex-use conception of words explains
why we feel so sure that the Tarski–Quine thesis and the intersubjectiv-
ity constraint are incompatible. We see that a Tarski-style definition of
truth for one’s own sentences tells us nothing about how another speak-
er ex-uses a given string of letters or sounds. From this we infer there
is no way to explain why it is epistemically reasonable to regard one’s
PJSSs as both factual and trustworthy without concluding that there is
more to truth than what is captured by a Tarski-style disquotational
truth predicate defined for one’s own regimented sentences. In short,
we feel sure that the Tarski–Quine thesis and the intersubjectivity con-
straint are incompatible because we presuppose the token-and-ex-use
conception of words, which implies that the thesis and the constraint are
incompatible.

The PJSS-Based Conception of Words

A central goal of this book is to develop an alternative conception of words
that does not imply that the Tarski–Quine thesis and the intersubjectivity
8 

constraint are incompatible. I start this task by showing that the type–token
distinction presupposed by the orthographic conception of words is not
as straightforward as it initially appears. To see why, consider letter types
and their tokens. We may suppose that all letter tokens are particulars,
or what I shall call marks—inscriptions in ink, pencil, chalk, grooves in
wood, stone, or plastic, electronic events on a computer screen, uttered
sounds, and so on.³ But what are letter types? It is tempting to think of
them as context-independent patterns, or sign designs.⁴ But a careful look
at how we actually identify marks as letters suggests that there is no single,
context-independent pattern that settles when a given mark counts as a
token of some given letter type. We ordinarily have no difficulty identifying
handwritten tokens of the letter type w, for instance, yet each person has
his own unique way of writing such tokens.
I propose that we try to accommodate these context-sensitive aspects
of our practice of identifying letters by using ‘replica’ in place of ‘token’,
taking ‘is a replica of ’ to express an equivalence relation on replicas,⁵ and
identifying the letter type w with {x: x is a replica of the mark w}, where
the mark in boldface is used as a sample replica of the letter type w.⁶ The
plausibility of the proposal depends on how we explain what is meant by
the phrase ‘is a replica of ’. I propose that we explain this by looking at our
actual practices of classifying marks for various purposes. For instance, we
will not count w a replica of w if we are sorting the two marks by font,
but we will do so if we are sorting them by letter type. Now suppose, as
suggested above, that our established practical criteria for sorting replicas
by letter type, including our criteria for identifying handwritten tokens of
letter types, cannot be replaced by context-independent definitions. Then
when we use such phrases as ‘{x: x is a replica of the mark w}’ to refer to
the sets that explicate our ordinary notion of letter types, we can provide

³ The term ‘mark’ is from Goodman 1976: 131.

⁴ The term ‘sign design’ is from Carnap 1942: §3.
⁵ See Goodman 1966: 361–2, and 1976: 138–43. The term ‘‘replica’’, but not Goodman’s nom-
inalistic use of it, is due to Charles Sanders Peirce, who writes, ‘‘Every legisign signifies through an
instance of its application, which may be termed a Replica of it’’ (Peirce 1933: 143).
⁶ Since by supposition we use ‘x is a replica of y’ to express a reflexive relation, the boldface mark
in ‘{x: x is a replica of the mark w}’ is a replica of itself, and is therefore a member of the set that is the
letter type w. See Goodman 1966: 361–2.
  9

at most an elucidation of the relation we express by using the phrase ‘is a

replica of ’ on that occasion.
Similar observations hold for our explications of word types. Here we
may draw on the fact that a single orthographic type may have different
satisfaction conditions. For instance, the British word ‘robin’ and the
American word ‘robin’ are spelled in the same way, but a bird that satisfies
the British word ‘robin’ does not satisfy the American word ‘robin’, and vice
versa—British robins and American robins are birds of different species.
Now suppose that in the following sentence
(∗ ) A robin is not a robin.
the first boldface mark robin is a token of the British word ‘robin’ and
the second boldface mark robin is a token of the American word ‘robin’.
Then (∗ ) is true, but we will not see this if we take both boldface marks to
be tokens of either the British word ‘robin’ or the American word ‘robin’.
For some purposes, such as copy editing, we group the first boldface mark
in (∗ ) with the second, but for others, such as applying a truth predicate to
another’s words, we do not. Corresponding with these different purposes
are different ‘‘is a replica of ’’ relations. There are many different ‘‘is a
replica of ’’ relations, each of which can be used to specify a different
class of letter or word replicas. Some of these relations can be defined in
more primitive terms. But it seems that many of the relations that are most
valuable to us cannot be defined in more primitive terms. As I noted earlier,
for instance, it may be that even the simple letter types that we use in
everyday linguistic interactions must be elucidated by careful investigations
of our actual identifications of letter replicas.
To elucidate the ‘‘is a replica of ’’ relation that is relevant to our
applications of Tarski-style definitions of satisfaction and truth to others’
sentences and to our own sentences as we used them in the past, we must
examine our ordinary criteria for identifying words. One might think,
following David Kaplan (in Kaplan 1990), that the relevant criteria are the
ones we rely on when we judge whether one speaker has repeated a word
uttered by another speaker. But a conception of words rooted in such
criteria would not by itself license us to make PJSSs. If I utter an ambiguous
word, such as ‘bank’, just to provide you with a word to repeat, you may
10 

repeat the word, but you will not make a PJSS for my token of it, since
you will see that I did not use that token to say anything, and hence that I
did not disambiguate it.
To elucidate an ‘‘is a replica of ’’ relation in a way that fits with the
intersubjectivity constraint, I propose that we adopt
The context principle: never ask for the word type of a word token in isolation,
but only in the context of one’s PJSSs for that word token.
The context principle amounts to a constraint on one’s explication of what
I call PJSS-based word types, which license our intersubjective applications
of Tarski-style disquotational definitions of satisfaction and truth. The idea
behind the principle is that my PJSS for Alice’s word ‘white’, for instance, is
simultaneously (i) licensed by my unhesitating judgement that Alice’s word
token ‘white’ is of the same type as my word token ‘white’ and (ii) a
constraint on my understanding of the word types of which I take Alice’s
word token ‘white’ and my word token ‘white’ to be members. Hence to
obey the context principle is to conceive of words in such a way that we
each accept
(1) Alice’s word ‘white’ is the same as my word ‘white’,
if and only if we take it to license a corresponding PJSS, namely,
(4) (x satisfies my word ‘white’ if and only if x is white) and (x satisfies
Alice’s word ‘white’ if and only if x is white).
In the midst of our inquiries we are always committed to some PJSSs or
other. To obey the context principle is to seek a reflective equilibrium
between our specification of word types and our PJSSs—an account of
word types that fits with all and only the PJSSs we see no reason to
reject.⁷ The context principle thereby disciplines our understanding of
the ‘‘is a replica of ’’ relation that we can use to group word tokens
into PJSS-based word types. To group word types in a way that obeys

⁷ The reflective equilibrium I have in mind is modelled on Nelson Goodman’s classic account of
the justification of deductive inferences: ‘‘How do we justify a deduction? . . . A rule is amended if it yields
an inference we are unwilling to accept; an inference is rejected if it violates a rule we are unwilling to amend. The
process of justification is the delicate one of making mutual adjustments between rules and accepted
inferences; and in the agreement achieved lies the only justification needed for either’’ (Goodman
1983: 63–4).
  11

the context principle is to adopt what I call the PJSS-based conception of

words.

Why the Tarski–Quine Thesis is Compatible

with the Intersubjectivity Constraint
If we adopt the PJSS-based conception of words, we can see that the
Tarski–Quine thesis and intersubjectivity constraint are compatible. There
are two crucial considerations.
First, if we adopt the PJSS-based conception, we will regard any criterion
for sameness of word types that violates the context principle as irrelevant
to truth and satisfaction. Any criterion for sameness of word types that
implies (P) violates the context principle, because it presupposes that there
are criteria for sameness and difference of word types that can be stated
without using any sentences that express our PJSSs. Hence if we adopt the
PJSS-based conception of words, we will reject (P), and thereby discredit
the reasoning that initially led us to think the Tarski–Quine thesis and
intersubjectivity constraint are incompatible.
Second, if we adopt the PJSS-based conception, we will recognize no
legitimate perspective independent of all of our PJSSs from which to
question them, and so we will be entitled to regard them as factual and
trustworthy. We will be entitled to regard them as factual because we will
take ourselves to have no grip on sameness of satisfaction apart from them,
and we will therefore be entitled to trust them unless we see special reason
in a given context for not doing so.
I propose that we adopt the PJSS-based conception of words. By doing
so, we transform our understanding of the word types to which our Tarski-
style disquotational definitions of satisfaction and truth apply: we come to
see our PJSSs as integral to our grasp of our Tarski-style disquotational
truth predicates. Hence if we adopt the PJSS-based conception of words,
we supplement our understanding of our own Tarski-style disquotational
truth predicates with an account of why it is epistemically reasonable for
us to regard our PJSSs as both factual and trustworthy without undermining
the Tarski–Quine thesis that there is more to truth than what is captured
by a Tarski-style disquotational truth predicate defined for one’s own
12 

sentences. We thereby come to see that the Tarski–Quine thesis and the
intersubjectivity constraint are compatible.

Costs and Consequences

The compatibility comes at a cost. If we adopt the account of truth that
I propose, we will commit ourselves to the context principle. We will
therefore be unable to explain what makes it the case that our PJSSs are
true—that particular word tokens have the satisfaction conditions we take
them to have. Why should we be willing to pay this price? Why not
stick with the token-and-ex-use conception of words, reject both the
Tarski–Quine thesis and the PJSS-based conception of words, and try to
show that facts about ex-use make it the case, or determine, that particular
word tokens have the satisfaction conditions we take them to have?
The answer, I shall argue, is that for any substantive theory of what
determines the satisfaction conditions of word tokens—any theory that
presupposes the token-and-ex-use conception of words—there are cases
in which speakers rely on PJSSs that conflict with the theory but that
we cannot reject without giving up the intersubjectivity constraint. I shall
examine several of the most influential and initially promising substantive
theories of what determines the satisfaction conditions of word tokens, and
present examples and thought experiments in which speakers rely on PJSSs
that conflict with these theories. I shall argue that we can make sense of
these PJSSs, and thereby satisfy the intersubjectivity constraint, only if we
adopt the PJSS-based conception of words. I conclude that if we want to
clarify and facilitate our inquiries by formulating logical generalizations, we
need a truth predicate, and the truth predicate we need is one that results
when we combine Tarski-style disquotational definitions of truth with the
PJSS-based conception of words.
The resulting account of truth transforms our understanding of a number
of central topics in the philosophy of language. It fits with and makes sense
of the observations that motivate many philosophers to embrace the causal
theory of reference for predicates. It enables us to eliminate names in favour
of descriptions whose denotations remain the same despite the sorts of rad-
ical changes in belief that led Kripke and others to abandon Russellian and
Fregean accounts of names. It enables us to understand anti-individualism
  13

without positing or presupposing social norms for language use or explicit

identity conditions for linguistic communities. Finally, and perhaps most
important of all, given the pragmatic considerations that motivate the
account, it enables us to formulate and affirm logical laws that we can apply
intersubjectively, without embracing a substantive theory of meaning or
reference.
 1
Regimentation

1.1. Regimentation as Linguistic Policy

As I emphasized in the introduction, my account of truth begins with
the assumption that we want to clarify and facilitate our inquiries by
regimenting our sentences so that we can formulate and apply logical
generalizations. My goal in this opening chapter is to sketch a pragmatic
account of regimentation that I will presuppose in the rest of the book.
The basic idea is that to regiment a given natural language sentence S,
as one used it on a particular occasion O, is to adopt a certain kind of
linguistic policy—to decide to use a sentence S with a perspicuous logical
grammar in place of S, as one used it on occasion O—for the purpose
of clarifying and facilitating one’s inquiries. One’s desire to show, for
instance, that
(1) If there is a sea in which every ship sails, then every ship sails in some
sea or other.
is logically valid might motivate one to replace (1) with a sentence
in artificial notation, such as,
(2) ∃x((x is a sea) ∧ ∀y((y is a ship) → (y sails in x))) → ∀y((y is a ship)
→ ∃x((x is a sea) ∧ (y sails in x))),
and hence to regard (2) as a paraphrase of (1).
To use (2) in place of (1), however, one must be able to explain
the artificial notations that appear in (2). As W. V. Quine observes, our
explanations of these notations ‘‘amount to the implicit specification of
simple mechanical operations whereby any sentence in the logical notation
can be directly expanded, if not into quite ordinary language, at least into
semi-ordinary language’’ (Quine 1960: 159). For instance, when we learn
first-order logic we are taught to expand (2) into
  15

(3) If (for something x such that x is a sea and for all things y (if y is a
ship, then y sails in x)) then for all things y (if y is a ship then there
is something x such that x is a sea and y sails in x).
This sentence contains ordinary words whose use is disciplined by the
insertion of parentheses, variables, quantifiers, and the decision to take
‘and’ and ‘if . . . then ’ to conform to the standard truth tables for ‘∧’
and ‘→’, respectively. By opting to use ordinary words in such artificially
circumscribed ways, we in effect extend ordinary language so that it includes
semi-ordinary words and the sentences we can construct with them. Once
we extend our language in this way, our decision to use (2) in place of
(1) looks similar to ‘‘what we all do every day in paraphrasing sentences to
avoid ambiguity’’ (Quine 1960: 159).
When a speaker decides to use a sentence S of an artificial first-order
language in place of a sentence S of his ordinary language, the relation of
S to S ‘‘is just that the particular business that the speaker was on that
occasion trying to get on with, with the help of S among other things,
can be managed well enough to suit him by using S instead of S’’ (Quine
1960: 160). We generally defer to a speaker’s judgements about what his
goals are and whether a given paraphrase will help him to achieve his goals.
If he judges that by using a regimented sentence S in place of an ordinary
language sentence S, he can more easily achieve his goals, then we grant
that he is entitled to use S in place of S. I shall call this the pragmatic
account of regimentation.
One might think we should add the further constraint that a speaker’s
decision to use S in place of S, as he used S on some previous occasion
O, is satisfactory only if as the speaker now uses S , it expresses the same
statement that S did when he used it on O, where the ‘same-statement’
relation is assumed to be objective and context independent. For suppose
a speaker’s goal is to determine whether or not S, a sentence he used
yesterday, is logically valid, and to do so he uses sentence S in place of S
as he used it yesterday. If S as he now uses it does not express the same
statement as S did yesterday, where ‘‘same-statement’’ is assumed to be
objective and context independent, then his present evaluation of S is not
guaranteed to achieve his goal.¹

¹ Some philosophers, such as Emma Borg, in Borg 2004: 72, are apparently inclined to assume
that the only legitimate purpose of regimentation is to figure in an explanatory semantic theory for
16 

This reasoning overlooks the fact that our ordinary standards for ‘‘same
statement’’ are shaped by our goals, and our goals may shift from context
to context. If the speaker aims to evaluate S, as used on occasion O,
and does so by evaluating S , then given his present goals, he takes his
use of S to express the same statement that S expressed when it was
used on O. One might think that there is some standard for sameness of
statement independent of such context-sensitive goals. But even if there
is such a standard, and we know that there is, it is too strong to say
that a speaker is entitled to use S in place of S, as used on O, only if,
according to the supposed standard, his present use of S expresses the same
statement that S expressed when it was used on O. For if the speaker’s
goal is not to find an S that expresses the same statement that S expressed
when it was used on O according to the supposed standard, but simply to
paraphrase S, as used on O, relative to his present purposes, then if using
S in place of S serves those purposes, he is entitled to use S in place of
S, even if, according to the supposed standard, S does not express the
same statement that S expressed when it was used on O. And even if the
speaker’s goal is to find an S that, according to the supposed standard,
expresses the same statement that S expressed when it was used on O, the
most we can say is that he is entitled to use S in place of S only if he
believes that S satisfies that goal. But this truism is a consequence of more
general deliberative norms, and should not be built into one’s account of
regimentation.
You may therefore accept the pragmatic account of regimentation,
provisionally at least, even if you believe that there is a standard for
sameness of statement that is independent of individual speakers’ decisions
about how to paraphrase their sentences. In later chapters I will argue that
the relevant standards for sameness of statement are not as independent

a natural language, and hence that the logical form of a natural language sentence is an objective,
context-independent property of it. But sometimes we just want to evaluate an argument we’ve made
by using a given natural language sentence, without theorizing about which descriptions of logical form
best explain the structure of natural language. In such contexts, we use the methods of modern logic to
regiment arguments expressed in natural language without committing ourselves to claims that could
conflict with current linguistic theory. In any case, there is at present no agreement among linguistic
theorists about the logical forms of natural language sentences. A person who wishes to regiment
her sentences today could not defer to linguistic theory generally for an account of logical form, but
would have to choose from among a variety of such accounts, all more or less incomplete. This is at
best impractical. To avoid these problems, I recommend that we embrace the pragmatic account of
regimentation.
  17

of our context-sensitive judgements about how to paraphrase sentences

as some philosophers believe, and that, properly viewed, the pragmatic
account of regimentation is of a piece with a satisfactory account of truth.
Between now and then I ask only that you be willing to investigate some
of its consequences.
To begin with, given the pragmatic account of regimentation, we may
define a regimented language as an artificial language whose sentences a
person has decided to use for various purposes in place of his or her natural
language sentences. Suppose one’s regimented language L comprises all
and only sentences constructed in the usual way from a finite list of basic
predicates, such as ‘robin’ and ‘is smaller than’, the truth-functional symbols
‘∼’ and ‘∨’, the quantifier symbol ‘∀’, and an unlimited stock of variables
‘x1 ’, ‘x2 ’, . . . .² Then L contains no proper names or indexicals, and its
predicates, we suppose, are neither ambiguous nor vague. L is too austere
and inflexible to serve many of our everyday and scientific purposes, but
it is ideal for formulating logical laws. I assume that to formulate logical
laws, we will each be willing to restrict ourselves to the vocabulary of
an artificial language like L: we will each be willing to use unambiguous
predicates with the grammar of first-order logic in place of ambiguous or
vague predicates, and to replace proper names and indexicals, as applied
on particular occasions, with definite descriptions that contain regimented
predicates specially introduced for the purpose.
My goal in this chapter is to explain what this assumption comes to.
For this purpose I review and amend a few well-known methods for regi-
menting ambiguous or vague words, quantifications, definite descriptions,
proper names, pronouns, and demonstratives. I emphasize that if we wish to
use these methods to clarify and facilitate our inquiries, we cannot dispense
entirely with ordinary (or, at least, semi-ordinary) language.

1.2. Ambiguity
Suppose that on one occasion I utter the sentence ‘Riverside is a bank’,
taking this utterance to be interchangeable on that occasion with an
utterance of ‘Riverside is a river bank’, but that on another occasion I utter

² L does not contain the symbols for ‘∃’, ‘→’, or ‘∧’, but ‘∃’ may be expressed in terms of ‘∼’ and
‘∀’; ‘ →’ and ‘∧’ may expressed in terms of ‘∼’ and ‘∨’.
18 

the sentence ‘Riverside is not a bank’, taking this second utterance to be

interchangeable on that (second) occasion with an utterance of ‘Riverside
is not a money bank.’ If I overlook the ambiguity in my word ‘bank’, I
may be in danger of reasoning as follows:
(P1) Riverside is a bank.
(P2) Riverside is not a bank
(C) Riverside is a bank and Riverside is not a bank.
The conclusion looks contradictory, hence unacceptable, and so it may
appear that I must reject either (P1) or (P2). I am rarely tempted to make
such simple errors in reasoning, but sometimes I want to take extra steps
to guard against making them. At such times, in place of (P1), (P2), and
(C), I may write
(P1 ) Riverside is a river bank.
(P2 ) Riverside is not a money bank.
(C ) Riverside is a river bank and Riverside is not a money bank.
There is no temptation to regard (C ) as contradictory, hence no reason to
take this inference to raise doubts about (P1 ) or (P2 ).
One may introduce paraphrases like this to avoid drawing bad inferences,
whenever one notices the ambiguities that may tempt one to draw them.
As an extra precaution, one may decide to use (what one takes to
be) unambiguous regimented predicates in place of ordinary ambiguous
predicates, such as ‘bank’, as used on particular occasions. To avoid drawing
bad inferences that turn on the ambiguity of one’s ordinary word ‘bank’,
for instance, one may decide to use ‘bank1 ’ in place of uses of ‘bank’ that
one would paraphrase by ‘river bank’, and ‘bank2 ’ in place of uses of ‘bank’
that one would paraphrase by ‘money bank’.³
When I regiment my uses of ‘bank’, I select from what I take to
be a finite number of standard ways of paraphrasing ‘bank’, including
‘money bank’ and ‘river bank’. This suggests that the English word
‘bank’ is lexically ambiguous, in the sense that any given utterance of a

³ This way of handling ambiguity is sketched by Scott Soames: ‘‘Ambiguity can . . . be treated as a
case of homonomy. For example, instead of thinking that English contains a single (ambiguous) word
type ‘bank’, one can take English to contain two different words, ‘bank1 ’ and ‘bank2 ’, whose tokens
are phonologically identical . . . contextual factors . . . can then be thought of as determining whether
particular utterances are tokens of the type ‘bank1 ’ or the type ‘bank2 ’ ’’ (Soames 1984: 427 n. 26).
  19

sentence containing it may be paraphrased in one of several standard

ways. In contrast, some words seem to be ambiguous in ways that cannot
be resolved by selecting from an established list of standard paraphrases.
Consider, for instance, the following two contexts in which I use the word
‘green’:⁴
(1) The leaves on my Japanese maple are reddish-brown. I want the
leaves to be green, so I paint them. When I am finished, I say to
you, ‘‘The leaves are green.’’
(2) A few minutes later, a botanist friend phones me to ask if she can
use the leaves on my Japanese maple for her study of green-leaf
chemistry. I reply, ‘‘No, the leaves are not green.’’ Neither you nor
I take my new utterance to conflict with my previous utterance of
‘The leaves are green’.
In addition to these two uses of ‘The leaves are green’, and ‘The leaves are
not green’, there are many other possible uses of these sentences. Here, for
instance, are two more:
(3) You and I notice that the leaves on my Japanese maple, which were
reddish-brown a week ago, are now covered with a green mould. I
say to you, ‘‘The leaves are green.’’ A few minutes later, a botanist
friend phones me to ask if she can use the leaves on my Japanese
maple for her study of green-leaf chemistry. I reply, ‘‘No, the leaves
are not green.’’
(4) I want the reddish-brown leaves on my Japanese maple to be green.
Our friend Ali paints them in pointillist style. When Ali is finished,
I say to you, ‘‘The leaves are green.’’ You reply, ‘‘The leaves may
look green, but only from a distance; up close one can see they are
still not green.’’
These several uses of ‘green’ suggest that the number of different ways in
which one might paraphrases sentences containing ‘green’ is not fixed in
advance of particular contexts in which one uses that word.⁵ Words that

⁴ The following four cases are due in essentials to Travis 1997. Travis’s cases, in turn, are inspired by
similar ones in Austin 1962.
⁵ Charles Travis writes that ‘‘There is no reason to think that there is any limit to possible
understandings of [what it would be for leaves to be green], each of which might be invoked by some
words which spoke on that topic’’ ( Travis 1997: 90).
20 

have this feature are not lexically ambiguous, but (as I shall say) contextually
ambiguous.⁶
Like particular uses of lexically ambiguous words, however, particular
uses of contextually ambiguous words may sometimes make it difficult to
avoid drawing mistaken inferences from premises that we accept. If I do
not pay attention to the different contexts of utterance in (1) and (2), for
instance, I might be tempted to think my utterance of ‘The leaves are
green’ in (1) conflicts with my utterance of ‘The leaves are not green’ in
(2). While this appearance is not due to my use of a lexically ambiguous
word, as in the case involving the word ‘bank’ described above, the
practical remedy for it is the same: to replace the two uses of ‘green’ with
the ordinary English phrases ‘looks green from close up under normal
lighting conditions’ and ‘is naturally green’, respectively, and introduce
regimented words ‘green1 ’ and ‘green2 ’ that I may use in place of these
phrases.
Such efforts to avoid ambiguity by regimentation are not guaranteed to
succeed: it is always possible that we will need to disambiguate one of our
own regimented predicates, such as ‘green1 ’, to accommodate new uses of
the word that we had not anticipated. As Geach once wrote, ‘‘the price
of freedom from fallacy is eternal vigilance’’ (Geach 1962: 71). We are
nevertheless guided by a regulative ideal: adopt a regimentation only if it doesn’t
leave unresolved ambiguities that prevent us from keeping track of the inferences and
logical laws on which we rely.

1.3. Is Regimentation Possible?

Examples (1)–(4) of the previous section are due in essentials to Charles
Travis. For reasons explained above, the examples convince me that ‘is
green’ is contextually ambiguous. Similar examples could be found to
show that many other natural language predicates are also contextually
ambiguous. But Travis goes further: he takes his examples to show that no
predicate, on any particular use of it, has an extension.⁷ This radical claim

⁶ Many of the English predicates that Charles Travis uses in Travis 2000 to illustrate what he calls
occasion sensitivity are contextually ambiguous.
⁷ Travis writes that ‘‘No understanding we could have of being green, so none that might attach to
a particular use of ‘is green’, would be one on which ‘is green’ spoke of a property, if a property must
  21

may look plausible if we focus on occasions in which a speaker applies a

given natural language predicate to an object that she sees, touches, or feels
on that occasion. For in such cases one might think that to take what is
expressed to be true it is enough to settle which objects in the immediate
context are supposed to satisfy the predicate in the contextually relevant
sense. This apparently leaves it open whether the predicate, as it is used
(understood) on that occasion, applies to any object that is not present on
the occasion. In such cases, then, it is tempting to equate the particular
use (understanding) of the predicate with our conscious (or, at least,
semi-conscious) beliefs about what satisfies the predicate on that occasion.
This way of thinking about predicates is much less tempting, however,
when we are evaluating arguments involving premises that contain universal
quantifiers. For instance, consider the following argument
(P1) All dolphins are mammals.
(P2) Some dolphins live in the sea.
(C) Therefore, some mammals live in the sea.
A standard paraphrase of the first premise, (P1), is
∀y((y is a dolphin) → (y is a mammal))
where ‘∀y’ is understood to range over everything (in the relevant domain).
If we understand the symbol ‘→’ in the usual truth-functional way, then
this paraphrase of (P1) commits us to supposing that the predicate ‘is a
dolphin’ is true or false of every individual (in the relevant domain). More
generally, regimented predicates are true or not true (on the understanding
assumed by that regimentation) of every individual (in the relevant domain),
and therefore also have extensions—associated sets of objects (from the
relevant domain) of which the predicates are true.
If, as Travis claims, no predicate, on any particular use of it, has an
extension, then regimentation in the sense just described is not possible.
Travis’s radical claim will look appealing if we conflate our beliefs about
how to apply our predicates on particular occasions, on the one hand, with

have an extension . . . . Where ‘is green’ has made different contributions to different wholes, we may
identify different things for it to have spoken of each time—being green on this understanding, and
being green on that one. So we may see the predicate as varying its reference across speakings of it.
But we must not mistake these different things for properties [or predicates] with extensions’’ ( Travis
1997: 99–100).
22 

the satisfaction conditions for our predicates—the conditions that must be

satisfied for our predicates to be true of or not true of particular objects—on
the other. The conflation is even more tempting if, following Travis, we
describe a particular use of a natural language predicate as a particular
‘‘understanding’’ of that predicate on that occasion. For it is natural to
understand ‘‘understanding’’ in a way that is closely connected with our
current beliefs. If we are to make sense of regimentation, however, we must
not conflate the satisfaction conditions for a natural language predicate, as
used on a particular occasion O, with our beliefs about the conditions under
which objects in the intended domain of discourse satisfy that predicate, as
used on O. We must distinguish between our beliefs about those conditions,
and the conditions themselves. Starting in Chapter 3, I shall highlight and
elucidate aspects of our inquiries that commit us to drawing this distinction.
In the meantime I shall take the distinction for granted, and assume that
regimented predicates have extensions.

1.4. Vagueness
Suppose that Herbert has a full head of hair on Monday, and shaves it all off
on Tuesday. Seeing the result on Wednesday, but knowing that Herbert
had a full head of hair two days before, you say, ‘‘Herbert is bald, like a
skinhead.’’ A few moments later, while discussing the kind of baldness that
results from a natural and irreversible ageing process, you say, ‘‘Herbert is
not bald, he just shaved his head.’’ You do not take your utterances of
‘Herbert is bald’ and ‘Herbert is not bald’ to conflict.
Now suppose you want to avoid any appearance of conflict between
your utterances of ‘Herbert is bald’ and ‘Herbert is not bald.’ For this
purpose, you introduce the predicate ‘bald1 ’ as a paraphrase of ‘bald’ as
you used it when you uttered ‘Herbert is bald’, and ‘bald2 ’ as a paraphrase
of ‘bald’ as you used it when you uttered ‘Herbert is not bald.’ After the
regimentation, you use ‘Herbert is bald1 ’ in place of your utterance of
‘Herbert is bald’, and ‘Herbert is not bald2 ’ in place of your utterance
of ‘Herbert is not bald’, and thereby remove the appearance of conflict
between the two utterances.
In this case, let’s suppose, the goal of your regimentation is to resolve a
contextual ambiguity, not to remove the vagueness of your particular uses
  23

of your unregimented word ‘bald’. You take your regimented words ‘bald1 ’
and ‘bald2 ’ to inherit the vagueness of the two applications of ‘bald’ that
‘bald1 ’ and ‘bald2 ’, respectively, replace. The sense in which the predicates
‘bald1 ’ and ‘bald2 ’ are vague can be explained by examples. If it takes fifteen
minutes for Herbert to shave off his hair, for instance, there will be times
during the fifteen-minute interval when Herbert is neither clearly bald1
nor clearly not bald1 . And if it takes Herbert twenty years to lose his hair
by a natural ageing process, there will be times, such as 5 p.m. GMT on
28 March 1985, when Herbert is neither clearly bald2 nor clearly not bald2 .
A predicate is vague if and only if its application can be unclear in the ways
just exemplified.⁸
Many philosophers accept theories of vagueness that imply that a vague
sentence such as ‘At 5 p.m. GMT on 28 March 1985, Herbert is bald2 ’
is neither true nor false.⁹ In contrast, I propose that we take that sen-
tence to have a truth value (true or false) even though no one knows its
truth value.¹⁰ To accept this proposal is to accept a theory of vagueness
that goes beyond the pragmatic decision to use regimented predicates,
such as ‘bald1 ’ and ‘bald2 ’, in place of ordinary English predicates, as
used on particular occasions. But there is no reason we cannot com-
bine such pragmatic decisions with other pragmatic decisions, such as the
decision to adopt the theory that our vague sentences are true or false.
The advantages are clear: if one adopts the theory, one can introduce
vague predicates into one’s regimented languages without giving up clas-
sical logic, hence without falling prey to sorities paradoxes, such as the
following:¹¹
(P1) Every person with 0 hairs growing naturally on his head is bald2 .
(P2) For any n, if every person with n hairs growing naturally on his
head is bald2 , then every person with n + 1 hairs growing naturally
on his head is bald2 .

⁸ Here I follow Timothy Williamson; see Williamson 1994: 2. However, I do not endorse
Williamson’s view that we can in principle never know the precise boundaries of a predicate we
now regard as vague. In my view, to call a predicate vague is to say something about our current
epistemological position: there are cases we regard as clear, and others we regard as borderline, and we
see no likely prospect of settling the borderline cases. It is not to say that nothing that would settle
those cases could be discovered.
⁹ According to Fine 1975, for example, vagueness is a deficiency of meaning.
¹⁰ Horwich 1998a, Sorenson 2001, and Williamson 1994 each present versions of this view.
¹¹ This particular sorities paradox is modelled on Paul Horwich’s formulation of a sorities paradox
about heaps. See Horwich 1998a: 81.
24 

(C) Therefore, for any n, every person with n hairs growing naturally
on his head is bald2 .

This argument seems to establish the obviously unacceptable conclusion

that everyone is bald2 . We find (P1) undeniable. We also find it difficult
to reject (P2), because we are strongly inclined to suppose that due to
the vagueness of ‘bald2 ’, there is no number n of hairs such that every
person with n hairs growing naturally on his head is bald2 , but some
person with n + 1 hairs growing naturally on his head is not bald2 . If
we accept bivalence for sentences containing ‘bald2 ’, however, we can
reject premise (P2), and affirm that for some n, every person with n hairs
growing naturally on his head is bald2 , but some person with n + 1 hairs
growing naturally on his head is not bald2 , without being able to identify
such an n.¹²
Many will be inclined to insist without argument that sentences such
as ‘At 5 p.m. GMT on 28 March 1985, Herbert is bald2 ’ are neither
true nor false. This reaction is partly sustained, I believe, by our tendency
to conflate our beliefs about how to apply our predicates, on the one
hand, with the satisfaction conditions for our predicates, on the other.
The reaction also overlooks our desire to avoid paradoxical inferences. By
accepting the principle of bivalence for sentences that contain ‘bald2 ’, for
instance, we can reject premise (P2) of the above sorities argument and
bypass the sorities paradox without giving up the deeply entrenched laws
of classical logic. This is a good pragmatic reason for accepting the principle
of bivalence for regimented sentences that contain vague predicates. If we
view regimentation as the adoption of a linguistic policy that furthers our
goals, including our goal of avoiding the sorities paradox, we should not be
swayed by the feeling that sentences such as ‘At 5 p.m. GMT on 28 March
1985, Herbert is bald2 ’ are neither true nor false.¹³

¹² For reasons I explain in Ebbs 2001, this does not imply that baldness supervenes on the number
of hairs on a person’s head. Other factors, such as the thickness and distribution of the hair may be
relevant as well.
¹³ Quine 1981 offers a pragmatic motivation for accepting bivalence across the board, although,
unlike me, he proposes that we actually specify particular sharp boundaries for vague predicates. In
contrast, the view I advocate is more like what Roy Sorenson calls epistemicism. Sorenson writes that
‘‘A few trivial inference rules can derive the heart of epistemicism from the empty set of premises.
Consequently, the right attitude to epistemicism is to presuppose it. The philosophical problem is to
explain why the previous assertion seems so dogmatic and question-begging’’ (Sorenson 2001: 184). I
agree with Sorenson that we need an explanation of the widespread rejection of bivalence for sentences
  25

1.5. Quantifier Domains, Tense, and Time

Suppose that on one occasion I utter the sentence ‘All are enrolled in
P505,’ taking this utterance to be interchangeable on that occasion with an
utterance of ‘All first-year Indiana University philosophy graduate students
in Fall 2006 are enrolled in P505,’ but that on another occasion, even
though no student has dropped P505, I utter the sentence ‘Not all are
enrolled in P505’, taking this second utterance to be interchangeable on
that (second) occasion with an utterance of ‘Not all philosophy graduate
students at Indiana University in Fall 2006 are enrolled in P505.’ If I
overlook the implicit domains associated with these two uses of the phrase
‘All’, I may be in danger of reasoning as follows:
(P1) All are enrolled in P505.
(P2) Not all are enrolled in P505.
(C) All are enrolled in P505 and not all are enrolled in P505.
Since (C) looks contradictory, I may be tempted to reject (P1) or (P2). This
would be a mistake, because they are not incompatible. Again, I am rarely
tempted to make such simple errors in reasoning, but sometimes I want to
regiment my sentences in a way that guards against them. At such times, I
may use the symbol ‘∀’, understood as ranging over a fixed domain, and
replace (P1) and (P2) by
(P1 ) ∀x((x is a first-year Indiana University philosophy graduate student
in Fall 2006) → (x is enrolled in P505 in Fall 2006)).

(P2 ) ∼∀x((x is an Indiana University philosophy graduate student in
Fall 2006) → (x is enrolled in P505 in Fall 2006)).
(C ) ∀x((x is a first-year Indiana University philosophy graduate student
in Fall 2006) → (x is enrolled in P505 in Fall 2006)) ∧ ∼∀x((x
is an Indiana University philosophy graduate student in Fall 2006)
→ (x is enrolled in P505 in Fall 2006)).
Unlike (C), (C ) does not look like a contradiction. Hence if we regiment
(C) as (C ), we can guard against the mistake of taking (C)—the conjunction

containing vague terms. My explanation is that we are tempted to reject bivalence for vague predicates
because we are tempted by misguided intuitions about meaning to conflate our beliefs about when a
given predicate applies to an object, on the one hand, with the satisfaction conditions for that predicate,
on the other.
26 

of (P1) and (P2), as uttered on the two occasions described above—to be

a contradiction. More generally, to avoid mistakes in reasoning due to the
context sensitivity of quantifier domains, we may use the symbol ‘∀’ in
place of ordinary occurrences of ‘All’ or ‘Every’, where ‘∀’ is understood
as ranging over a fixed domain, and the different domains over which the
ordinary occurrences of ‘All’ or ‘Every’ may seem to range are specified
by the antecedents of conditionals bound by ‘∀’. Similarly, the different
domains over which the ordinary occurrences of ‘Some’ or ‘there are’ may
seem to range can be specified by adding conjuncts to the matrixes bound
by ‘∼∀x ∼’.¹⁴
Here it is important to note that ‘∀’ should be read as timeless—as
ranging over all objects, past, present, and future. This reading goes hand
in hand with the fact that sentences of regimented English are tenseless;
or, to be more precise, they are all in the present tense, understood as
timeless. Some, but not all, English sentences are in the present tense, and
we sometimes, but not always, take such sentences as timeless: when we
say, ‘‘All dolphins are mammals,’’ for instance, we typically take ourselves
to be generalizing over everything, past, present, and future.¹⁵ In contrast,
every regimented English sentence has the surface grammatical form of
the present tense, and we must always take such sentences as timeless. In
regimented English, to make distinctions indicated in English by tense, we
treat time as just another dimension, like space. We regard times as objects
and add regimented predicates, such as ‘x is before y’, that are true or
false of ordered pairs of objects, including times. This treatment of time is
presupposed by all standard paraphrases of ordinary language sentences into
sentences with the grammar of first-order logic.¹⁶ Surprisingly, however,
it often goes without saying, even in elementary logic texts. Perhaps the

¹⁴ The phenomenon of the context-sensitivity of quantifier domains is not restricted to utterances

of quantifier words in ordinary language. For instance, the sentence ‘∀x∃y(y2 = x)’ might appear on
two occasions in a mathematics lecture, on the first of which it is paraphrased as ‘∀x((x is a natural
number) → ∃y((y is a natural number) ∧ (y2 = x))’, and on the second of which it is paraphrased as
‘∀x((x is a real number) → ∃y((y is a real number) ∧ (y2 = x))’. For a similar example, see McGee
2004.
¹⁵ Does this imply that we can generalize over ‘‘absolutely everything’’? For two classic discussions
of this issue, Dummett 1981 and Cartwright 1994. For more recent discussions of it, see Williamson
2003 and Rayo and Uzquiano 2006. I have sympathies with both sides of the debate about whether
one can generalize over absolutely everything, and will not try to adjudicate it in this book.
¹⁶ For summaries of this standard treatment of time, see Quine 1960: 170–6, Quine 1982: 195, and
Barwise and Etchemendy 2002: 398.
  27

main reason is that most competent English-speakers pick it up readily,

with little or no explicit instruction.¹⁷

1.6. Descriptions and Proper Names

Indefinite descriptions such as ‘Some sea’ and ‘Every ship’ do not correspond
to logically significant parts of the regimented sentences that paraphrase
ordinary English sentences in which they occur. For instance,
(1) Every ship sails in some sea or other.
is usually regimented by a sentence such as
(2) ∀y((y is a ship) → ∃x((x is a sea) ∧ (y sails in x))).
Such phrases as ‘Some sea’ and ‘Every ship’ therefore disappear under
regimentation into quantifications.
Definite descriptions of ordinary language can also be eliminated by
paraphrase. Consider the sentence
(3) The author of The Principles of Mathematics won a Nobel Prize for
literature.
Let’s say that a singular term is a term that can coherently be replaced by
a variable of the sort that could be bound by an objectual quantifier.¹⁸ It
seems coherent to put such a variable in the place of ‘The author of The
Principles of Mathematics’ in (3) to yield the open sentence
(4) x won a Nobel Prize for literature.
Thus it seems as though ‘The author of The Principles of Mathematics’ is a
singular term. To regiment sentences that contain such singular terms, we
can use Russell’s method of paraphrasing definite descriptions. We can first
introduce a description operator symbolized by ‘the x:’ (read as ‘the x such
that’), which we can prefix to a predicate in regimented notation to create

¹⁷ As any experienced logic instructor knows, it is much more difficult to explain to the begin-
ner such comparatively trivial points as that the English phrase ‘only if ’ should be paraphrased
by ‘→’.
¹⁸ This criterion of singular termhood is due to Quine. For his own summary of it, see Quine
1982: 260.
28 

a definite description. Prefixing ‘the x:’ to the predicate ‘x is an author

of The Principles of Mathematics’, for instance, yields ‘the x: x is an author
of The Principles of Mathematics’. Putting this description in place of ‘x’ in
(4) yields
(5) (The x: x is an author of The Principles of Mathematics) won a Nobel
Prize for literature.¹⁹
To regiment this into the austere language sketched above, we paraphrase
(5) as
(6) ∃x ((x is an author of The Principles of Mathematics) ∧ ∀y ((y is an
author of The Principles of Mathematics) → x = y) ∧ (x won a Nobel
Prize for literature)),
in which the definite description no longer appears as a singular term.
We may also choose to regiment ordinary proper names as definite
descriptions. Consider any ordinary sentence containing a proper name,
such as
(7) Bertrand Russell won a Nobel Prize for literature.
Quine points out that a sentence of the form (7) is equivalent in truth
value to
(8) ∃w((w = Bertrand Russell) ∧ (w won a Nobel Prize for literature)).
Quine suggests that we replace the predicate ‘= Bertrand Russell’ with
a simple, unstructured, one-place predicate specially introduced to be
satisfied only by Bertrand Russell.²⁰ For this purpose, following Quine
1982: 278–83, we may use the predicate ‘is-Russell’. In place of (8), we
can then write
(9) ∃w((w is-Russell) ∧ (w won a Nobel Prize for literature)).

¹⁹ According to Russell’s theory of descriptions, a description of the form ‘the x : Fx’ is not a singular
term, but a sentence in disguise. Nevertheless, Russell proved that in extensional contexts we are
licensed to treat a description of the form ‘the x: Fx’ as if it were a singular term that can be substituted
for positions in which it makes sense to put a variable of the sort that could be bound by an objectual
quantifier. See Neale 1990: 131.
²⁰ It is not widely known that Tarski anticipated this key move of Quine’s treatment of names as
descriptions in 1936: ‘‘To say that the name x denotes a given object a is the same as to stipulate that
the object a (or every sequence of which a is the corresponding term) satisfies a sentential function
of a particular type. In colloquial language it would be a function which consists of three parts in the
following order: a variable, the word ‘is’ and the given name x.’’ ( Tarski 1936: 194 n. 1)
  29

If we regiment (7) in this way, then the name ‘Bertrand Russell’ no longer
appears in the position of a singular term. On the other hand, (9) is logically
compatible with the claim that there is more than one Bertrand Russell.
To express our belief that there is only one thing that is-Russell, we can
add a clause to our regimentation, as follows
(10) ∃w((w is-Russell) ∧ ∀z((z is-Russell) → z = w) ∧ (w won a Nobel
Prize for literature)).
Thus regimented, the proper name ‘Bertrand Russell’ becomes, in effect, a
sentence unto itself, namely,
(11) ∃w((w is-Russell) ∧ ∀z((z is-Russell) → z = w)).
If we like, we can abbreviate the description so that it looks like a singular
term, rewriting (11) as ‘The w : w is-Russell’ and (10) as:
(12) (The w : w is-Russell) won a Nobel Prize for literature.
But this would be a mere abbreviation for (10), in which the definite
description is not a singular term.
To replace a predicate of the form ‘= n’ with an unstructured predicate
‘is-n’, we don’t need to make any assumptions about how we came to
be able to use the predicate ‘= n’ that we rely on to introduce ‘is-n’.
Quine’s method of regimenting sentences that contain proper names is
therefore compatible with just about any account of how we came to
be able to use ‘= n’. In particular, to apply Quine’s method, we need
not accept Russell’s theory of how we understand sentences that contain
proper names (Quine 1982: 276–7). In addition, as I’ll argue in Chapter 9,
Quine’s method is not vulnerable to the semantical arguments that Kripke
and others have used against the classical description theory of proper
names.²¹

²¹ First-order logic does not have modal operators, but we can still ask how Quine’s method of
paraphrasing names would fare if we were to add modal operators. If we prefix the right sort of
rigidifying operator R to a definite description d that results when we apply Quine’s method to
regiment a name n, the resulting rigidified description, Rd, will behave like n in modal contexts.
The operator I favour for this purpose is Tyler Burge’s @-operator, which rigidifies a predicate ‘F’
in such a way that in all worlds, actual or counterfactual, a use of ‘the @ F’ denotes the actual F
(Burge 2005: 229–30). Burge points out (at Burge 2005: 240) that his @-operator escapes a problem
Soames 1998 raises for the superficially similar proposal that we use ‘‘actually’’ to rigidify definite
descriptions.
30 

1.7. Pronouns and Demonstratives

The pronoun ‘he’ in
(1) Maurizio broke the bank in Monte Carlo and he admires Tarski
is what Geach calls a pronoun of laziness, because we can eliminate it by
paraphrasing (1) as
(2) Maurizio broke the bank in Monte Carlo and Maurizio admires
Tarski.
No pronouns occur in (2), and so we can use the methods already
introduced to regiment (2), putting regimented definite descriptions in
place of the proper names of (2) and regimented predicates in place of the
ordinary predicates of (2). In contrast, however, most of us would not be
willing to paraphrase
(3) Just one man broke the bank in Monte Carlo and he admires Tarski
as
(4) Just one man broke the bank in Monte Carlo and just one man
admires Tarski.
Instead, we see the pronoun in (3) as bound by an implicit quantifier in
‘Just one man broke the bank in Monte Carlo.’ We see (3) as equivalent to
(5) The man who broke the bank in Monte Carlo admires Tarski.
And if we regiment (5) by using descriptions in place of the proper names
and descriptions in (5), we will view the pronoun in (3) as in effect bound
by an initial existential quantifier.²²
In addition to pronouns of laziness and bound pronouns, there are also
indexical pronouns, which do not simply stand in for singular terms that
have appeared earlier in the same sentence or some relevant previous
discourse, and are not usually seen as playing a role similar to bound
variables in first-order logic. For instance, suppose that you and I are in a

²² Sentences (1) and (3) are modifications of examples (20) and (21) on page 125 of Geach 1962. The
terminology ‘‘pronouns of laziness’’ is introduced on page 124 of Geach 1962.
  31

room filled with people unknown to you, and, while pointing at one of
these people, I turn to you and utter the sentence
(6) He admires Tarski.
In these circumstances, you and I might both be willing to paraphrase
(6) by
(7) That man over there admires Tarski
thereby replacing the indexical use of the pronoun ‘he’ by the description
‘that man over there’, which contains the demonstrative ‘that man’ and the
demonstrative ‘there’, both of which we understand as supplemented, in
the circumstances described, by my pointing gesture.
We may paraphrase sentences such as (6) and (7) into an artificial first-
order language that contains no demonstratives, but only if we introduce
new regimented predicates that take the place of particular applications of
demonstratives. To paraphrase (6), for instance, we may first paraphrase
it as (7), and replace the occurrence of ‘That man over there’ in (7) by
‘The x: x is a man and x is over there.’ We can then introduce a new
regimented predicate to replace our particular application of ‘is over there’,
which we take to express a condition that is uniquely satisfied by one
person to whom I pointed when I uttered (6). Finally, we may replace ‘The
x: x is a man and x is over there’ by a definite description that contains
only our regimented predicate for ‘man’ and a newly introduced predicate
that expresses a contextually salient condition that we take to be uniquely
satisfied by one man to whom I pointed.
The contextually salient conditions on which we rely when we refer to
a particular person or object by using indexical pronouns or demonstratives
in the way just exemplified always involve our conscious perceptual attention
to the relevant person or object, and often involve our conscious visual
attention to it. But this is not to say that conscious perceptual attention,
understood as independent of demonstrative or indexical reference, explains
or grounds such reference relations, as Evans 1982 and Campbell 2002 argue.
One need not endorse this philosophical thesis to paraphrase sentences such
as (6) and (7) in the way just sketched.²³

²³ There is much more to be said about how to paraphrase sentences containing demonstratives
and pronouns, in particular, and about regimentation more generally. The vast literature about
32 

1.8. Why Ordinary Language is Indispensable

The methods of regimentation I have canvassed tell us nothing about the
context-invariant meanings of contextually ambiguous predicates, proper
names, and indexical pronouns and demonstratives. They provide no
account, for instance, of how the English word ‘green’ enables us to
introduce and explain an indefinite number of regimented predicates, or
of ‘‘the rules by which [indexical pronouns and demonstratives] adjust
their reference to circumstances’’ (Davidson 1984: 35). As I have stressed,
however, I assume that our primary motivation for regimenting our
sentences is to clarify and facilitate our inquiries. For this purpose, we do not
need a theory of the context-invariant meanings of contextually ambiguous
predicates, proper names, and indexical pronouns and demonstratives. We
need, instead, a method for introducing and explaining regimented sentences
that we can use in place of particular utterances of ordinary English sentences
that contain such expressions.
The methods I have sketched satisfy this need. If we choose to use them,
we should find no insurmountable obstacle to paraphrasing many of our
ordinary language sentences into an artificial language L that comprises all
and only sentences constructed in the usual way from a finite list of basic
predicates, truth-functional symbols ‘∼’, ‘∨’, quantifier symbol ‘∀’, and an
unlimited stock of variables ‘x1 ’, ‘x2 ’, . . . . No matter how many utterances
of ordinary language sentences we have regimented, however, we may
need to introduce and explain new regimented predicates that we can use
in place of new applications of contextually ambiguous words. We may
also need to introduce and explain new regimented predicates in order
to construct definite descriptions that we can use in place of new uses
of indexical pronouns or demonstratives. And to introduce and explain
new regimented predicates we need to use ordinary language. In short, if
we wish to use the methods of regimentation that I have sketched, we

demonstratives and pronouns is filled with ingenious proposals, any of which one might adopt in
particular contexts for particular purposes. For instance, some would follow Gareth Evans, who
argued trenchantly in Evans 1985 that in addition to pronouns of laziness and bound pronouns, we
must acknowledge what he called E-type pronouns. For useful bibliographies and detailed discussions
of many other important views of demonstratives, pronouns, and descriptions, see Neale 1990 and
Bezuidenhout and Reimer 2004. For treatments of pronouns and descriptions within a first-order
logical framework associated with Donald Davidson’s theory of interpretation, see Larson and Segal
1995 and Lepore and Ludwig 2007.
  33

will always need to use ordinary language to introduce and explain new
regimented predicates.

1.9. Limitations of First-Order Logic

At the start of this chapter I assumed that to formulate logical laws, we
will each be willing to restrict ourselves to the grammar of an artificial
first-order language. This may seem too restrictive, since there are many
English sentences that it is either difficult or impossible to paraphrase by
structured sentences of a first-order language. The qualification ‘structured’
is crucial here: trivially, any sentence that one can understand can be
paraphrased into an unstructured sentence of a first-order language, if, as
is usual, we suppose that first-order logic includes simple sentential logic
as a part. Hence the worry is not that we cannot paraphrase any given
English sentence S, as used on a given occasion O, into a sentence of a
first-order language. The worry is that we cannot always do so in a way
that by itself makes plain the logical relations that S, as used on O, bears
to other sentences. We can of course always mimic those logical relations
case by case, adding premises that, together with S, imply the sentences
that we take S, as used on O, to imply. But we may also seek a deeper,
more systematic understanding of the logical grammar of such sentences.
And this desire may lead us beyond first-order logic. Let us look at some
examples.

The Geach–Kaplan sentence

David Kaplan proved that the sentence:
(a) Some people admire only one another
cannot be paraphrased by any structured sentence of a first-order language
(Quine 1982: 293, Boolos 1984: 432–3). One might try to paraphrase (a) as
(b) There are at least two people who admire only one another.
But unlike (b), (a) ‘‘might hold because of some irreducible group of
say eleven mutual admirers’’ (Quine 1982: 293). It is agreed that (a) can
be paraphrased by a second-order sentence that is not equivalent to any
structured first-order sentence (Boolos 1984: 432). This does not mean
34 

that to paraphrase (a) with a structured sentence we must use second-order

quantification, however, because we can also paraphrase (a) by means of
first-order logic and set theory, as follows (Quine 1982: 293):
(c) ∃z[∃x(x ∈ z) ∧ ∀x{(x ∈ z) → (∃y(x admires y) ∧ ∀y((x admires
y) → (x = y ∧ y ∈ z)))}].
More generally, by allowing first-order variables to range over classes,
and using ‘ε’ for ‘is a member of ’, we can increase the expressive power
of a first-order grammar without using second-order quantification. To
paraphrase sentences such as (a), as well as sentences that explicate the
ancestral relation (Quine 1982: 292–3) and other complex mathematical
notions, a first-order logical language conjoined with set theory is enough.
But why not use second-order logic, as some authors (Boolos 1984,
Shapiro 1991) urge? My answer mixes pragmatic with theoretical con-
siderations. Classical first-order logic with identity became dominant in
the mid-twentieth century, after years of dispute among leading logicians,
including David Hilbert, Ernst Zermelo, Leopold Löwenheim, Thoralf
Skolem, Kurt Gödel, Alfred Tarski, Rudolf Carnap, and W. V. Quine.²⁴
Unlike its main rivals, first-order logic is compact and complete, and has a
finite proof procedure.²⁵ It is the logic most widely used in set theory. It
is also deeply entrenched in mathematics, linguistics, and philosophy cur-
ricula.²⁶ If we desire to clarify and facilitate our inquiries, as I assume, then
the historical pedigree, entrenchment, theoretical appeal, and convenience
of first-order logic are good pragmatic reasons for restricting ourselves, as
far as possible, to generalizations we can express in a first-order language.²⁷

Binary quantification: ‘most’

As we’ve seen, to paraphrase a sentence such as ‘All dolphins live in the
sea’ it is enough to prefix the unary quantifier ‘∀x’ to the matrix ‘(x is a

²⁴ For detailed accounts of this history, see Moore 1980 and 1988; for a summary of the history, see
Shapiro 1991: chapter 7. For a dissenting view, see Eklund 1996.
²⁵ For proofs of these basic results for first-order logic, and of their failures for one of its main rivals,
namely second-order logic, see Boolos, Burgess, and Jeffery 2002.
²⁶ This is not to say that only first-order logic deserves to be called logic, or that it is ‘the’ right
logic. For a stimulating discussion of what these claims might mean and why it would be so difficult to
establish them, see Tharp 1975.
²⁷ My account of truth could perhaps be extended to define disquotational Tarski-style truth
predicates that we could use to express higher-order logical generalizations, but I do not try to extend
it in that way here.
  35

dolphin) → (x lives in the sea)’. It is therefore natural to expect that in

sentences such as
(d) Most dolphins live in the sea
‘most’ can also be paraphrased by a unary quantifier ‘Most (x)’, which,
when prefixed to a matrix A in which there are free occurrences of ‘x’, is
understood to mean that most of the members of the relevant domain of
discourse satisfy A. To treat (d) on analogy with ‘All dolphins live in the
sea,’ however, one would have to find a truth-functional connective ∗ that
could yield a satisfactory paraphrase of (d) of the form:
(e) Most (x) ((x is a dolphin) ∗ (x lives in the sea)).
The problem is that there is no such connective (Rescher 1962, Wiggins
1980: 326, Davies 1981: 123–4, Barwise and Cooper 1981, Neale 1990:
40–1). Those who wish to provide a systematic, compositional theory of
meaning for English must therefore find some other way to understand the
contribution of ‘most’ to natural language sentences in which it occurs.²⁸
This is not the most pressing aspect of the problem for us, however, since
by assumption our goal is not to construct a theory of meaning for natural
language, but to express logical generalizations that we can use to clarify
and facilitate our inquiries. Given this goal, for instance, we could try
directly paraphrasing (d), as used on a particular occasion, with
(f ) More than half of the members of the set of all and only dolphins
live in the sea.
which does not contain the word ‘most’, thereby (for the moment, at least)
avoiding the problem of providing a systematic account of the contribution
of ‘most’ to natural language sentences in which it occurs. To paraphrase
(f ) into a sentence with a first-order grammar, however, we would again
need to add some set theory to our first-order language. Moreover, we may
want some assurance that we will always be able to come up with adequate
paraphrases of sentences that contain ‘most’. If so, we could adopt one of
several strategies for systematically specifying truth conditions for natural

²⁸ Such a desire leads Martin Davies, for example, to seek a systematic compositional account of the
contribution of ‘most’ and other quantifier words to the truth conditions of natural language sentences
in which they occur—an account that fits as closely as possible with the surface grammatical structure
of natural language sentences (Davies 1981: 123–4).
36 

language sentences that contain ‘most’ (see, for instance, Neale 1990: 41–2,
Glanzberg 2006, and Lepore and Ludwig 2007: 62–3), provided we keep
in mind that, insofar as our project is to clarify and facilitate our inquiries,
our motivation for adopting one of the standard strategies is not that we wish
to construct a systematic theory of meaning for natural language.

Relativity of adjectives: ‘short’

Consider the sentence
(g) John is a short basketball player.
One might at first be tempted to paraphrase this in first-order grammar as
follows:
(h) ( John is a basketball player) ∧ ( John is short).
But this paraphrase misses the relativity of the attribution of shortness in
(g) to basketball players. A better paraphrase of (g) would express the idea
that there is at least one person who is short for a basketball player. One way
to express this idea would be to paraphrase (g) as
(i) ( John is a basketball player) ∧ ( John is shorter than most basketball
players).
where the ‘most’ that occurs in ‘John is shorter than most basketball play-
ers’ is paraphrased in one of the several ways mentioned above.²⁹ There
is apparently no general obstacle to producing paraphrases, case by case,
that respect this kind of relativity of adjectives to particular nouns they
modify.

Belief sentences
The difficulty of paraphrasing sentences such as
( j) Frege believes that Mont Blanc is in France.
into first-order logical notation is well known. The context following ‘that’
in ( j) is opaque, in the sense that not all substitutions of coextensive terms

²⁹ For details, see, for instance, Heim and Kratzer 1998: 68–73 and Lepore and Ludwig 2007:
162–73. A number of other approaches could also work. There is no need to settle between them
here.
  37

for terms that follow ‘that’ in ( j) and not all substitutions of coextensive
sentences for the sentence that follows ‘that’ in ( j) preserve the truth value
of ( j). But first-order logic is extensional: in a language with a first-order
logical grammar, substitutions of coextensive terms and sentences always
preserve truth value. How then can one paraphrase belief sentences by
sentences with a first-order logical grammar? Given the pragmatic purposes
that I have described, this question, though serious, is less problematic than
it may at first appear, for two main reasons.
First, there is no obstacle to accepting Quine’s proposal that we regiment
‘x believes that Mont Blanc is in France’ with a one-place fused predic-
ate ‘x believes-that-Mont-Blanc-is-in-France’ (Quine 1960: 216). Donald
Davidson objects that Quine’s proposal would prevent us from providing
a theory of meaning for English, because it would entail that English has
infinitely many semantical primitives—one for each predicate constructed
by concatenating ‘believes that’ with an English sentence (Davidson 1984:
13–14). If our goal is not to construct a theory of meaning for natural
language, however, but, as I have assumed, to regiment our sentences, then
in principle we are free to adopt Quine’s proposal, or any other proposal
that yields a distinct regimented predicate for each new occasion on which
we ascribe a belief by using a natural language sentence.
Second, if we want a general method of paraphrasing belief sentences
into a language with a first-order logical grammar, one that also illuminates
the logical structure of belief sentences, we may adopt some version
of Davidson’s paratactic account of the logical form of belief sentences
(Davidson 1984: 93–108). There are several apparently powerful objections
to Davidson’s original proposal. Most of the objections presuppose that
an account of how to paraphrase belief sentences into first-order logic
is satisfactory only if it is faithful to all aspects of our ordinary concept
of belief, as revealed by our intuitions about which belief reports to
accept in various circumstances and what those belief reports imply.
Since our present project of regimenting our sentences to clarify and
facilitate our inquiries does not require us to be faithful to all aspects of
our ordinary concept of belief, or to preserve all of the intuitions that
are usually associated with it, however, many of the standard objections
to Davidson’s proposal will not count against our using it, at least on
some occasions, to further our goal of clarifying and facilitating our
inquiries. I conjecture that the remaining objections could be addressed by
38 

modifications of Davidson’s original proposal, but I will not try to establish

this here.³⁰

Necessity and possibility

It is now widely accepted that the expressions ‘necessarily’, ‘possibly’, ‘it
is necessary that’, and ‘it is possible that’ are typically used in English to
express notions that cannot be paraphrased by sentences of an extensional
first-order language. Thus, consider the sentence
(k) Necessarily, 9 > 7.
Perhaps in some special contexts one might use this sentence in a way that
could be replaced by a metalinguistic sentence, such as
(l) ‘9 > 7’ is analytic,
where ‘is analytic’ is understood to mean ‘x follows from the semantical
rules for the language of which x is a part’. To make sense of such
a paraphrase, however, we would need to make use of the notion of
semantical rule—a notion that cannot be explained in purely extensional
terms. In addition, however, there are other uses of (k) that we would
not be willing to paraphrase by (l). The most important challenges to
paraphrases such as (l) are those in which (k) is used on a particular occasion
in such a way that it is taken to imply
(m) There is something x such that necessarily, x > 7.
A sentence such as (m), while not strictly speaking part of ordinary language,
makes explicit use of a notion of de re necessity that many philosophers take
to be expressed by some uses of modal adverbs and corresponding phrases
in ordinary language. How are we to paraphrase such sentences?
There are two main strategies. The first is to add modal operators
to a regimented first-order language, taking the operators as primitives
synonymous with the ordinary language modal adverbs ‘necessarily’ and

³⁰ For a list of the main objections to Davidson’s proposal, plausible replies to these objections,
and a new proposal inspired by Davidson, see Lepore and Ludwig 2007: chapter 11. Since Lepore and
Ludwig share Davidson’s assumption that an account of belief sentences is satisfactory only if it can be
part of a finitely stated compositional theory of meaning—an assumption to which the project of this
book does not commit us—some of their replies concede to the objections more than is necessary for
my purposes.
  39

‘possibly’, or ‘it is necessary that’ and ‘it is possible that’ (Peacocke 1978,
Davies 1981: 187–93). The second is to paraphrase sentences that contain
ordinary language modal operators by regimented sentences that quantify
over possible worlds (Lewis 1968, Davies 1981: 193–201, Larson and
Segal 1995: 426–9). I find the second strategy more appealing than the
first, mainly because it is analogous to paraphrasing sentences that contain
second-order quantifiers by a first-order language with set theory added.³¹
The idea is to start with a regimented first-order language and add a theory
of possible worlds over which the quantifiers of that language range, and
then to explain necessity and possibility in terms of quantification over
possible worlds. It is a huge task to work out the details of either of these
strategies. In this book I shall remain officially agnostic about how to handle
ordinary language sentences that contain modal operators, and focus on
those aspects of ordinary language that can be paraphrased into a first-order
language without modal operators and without quantifying over possible
worlds. This attitude is justified, if at all, by the fact that it is difficult
enough to reconcile the Tarski–Quine thesis with the intersubjectivity
constraint even if we restrict ourselves to the extensional fragments of
ordinary language. I conjecture that any adequate treatment of the modal
sentences will be an extension of the treatment I offer for the non-modal
ones. Unfortunately, however, in this book I shall have to leave this as
a conjecture, since I have not yet worked out an adequate treatment of
modal sentences. In any case, as I shall try to show, the account of truth that
I shall propose solves a pressing pragmatic problem that arises if we restrict
our attention to sentences of ordinary language that can be paraphrased by
structured sentences with a first-order logical grammar.

³¹ Another reason is that unlike the first strategy (see Peacocke 1978: §III), the second strategy does
not require that we employ infinite conjunctions.
 2
The Tarski–Quine Thesis

2.1. The Indispensability Argument

Regimentation is not an end in itself. We regiment our sentences in
order to clarify and facilitate our inquiries by formulating and applying
logical generalizations such as ‘‘Every regimented sentence of the form
‘S ∨ ∼S’ is true.’’¹ As I explained briefly in the Introduction, however,
we cannot actually write out all of the sentences of any given logic-
al form, and so we cannot express logical generalizations by writing
out and affirming conjunctions of all of their instances. We therefore
need some way to express such conjunctions without actually writing
them out. Let us now consider in more detail how this need may
motivate us to use a truth predicate, and, ultimately, to embrace the
Tarski–Quine thesis that there is no more to truth than what is captured
by a Tarski-style disquotational truth predicate defined for one’s own
sentences.
To begin with, consider Quine’s classic account of why we need a truth
predicate to express logical generalizations:
We can generalize on ‘Tom is mortal’, ‘Dick is mortal’, and so on, without talking
of truth or of sentences; we can say ‘All men are mortal’. . . . When we want to
generalize on ‘Tom is mortal or Tom is not mortal’, ‘snow is white or snow is
not white’, and so on, we ascend to talk of truth and of sentences, saying ‘every
sentence of the form ‘p or not p’ is true’, or ‘every alternation of a sentence with
its negation is true’. What prompts this semantic ascent is not that ‘Tom is mortal
or Tom is not mortal’ is somehow about sentences while ‘Tom is mortal’ and
‘Tom is Tom’ are about Tom. All three are about Tom. We ascend only because

¹ Such generalizations are among the truths that we take to be worth formulating if we participate
in rational inquiry. By formulating logical generalizations, we can also clarify corresponding logical
consequence relations on which we may then choose to rely.
   ‒  41

of the oblique way in which the instances over which we are generalizing are
related to one another. (Quine 1986: 11)

In this passage Quine does not explicitly limit his reasoning to logical
generalizations on sentences in a regimented language. But if we are to
be clear about the logical generalizations that we aim to express, we
must first regiment our sentences. I shall therefore assume that Quine’s
reasoning applies in the first instance to logical generalizations on sentences
of one’s own regimented first-order language, and that by combining a
recursive grammar of our regimented language L with standard syntactic
criteria for substitution of regimented predicates or sentences for schematic
term and sentence letters, we can specify the set of all and only the
regimented sentences of L with a given form, such as ‘p or not p’.² Given
these assumptions, it would be needlessly indirect to define truth first for
propositions, and then to extend that definition of truth to sentences, as
Horwich 1998a and Soames 1999 propose. To use a truth predicate defined
for sentences is also indirect, but not needlessly so. When we wish to
generalize on sentences such as ‘Tom is mortal or Tom is not mortal’,
‘snow is white or snow is not white’, and so on, our primary goal is not
to say anything about such sentences, but to affirm that Tom is mortal
or Tom is not mortal, snow is white or snow is not white, and so on.
Nevertheless, according to Quine, due to ‘‘the oblique way in which the
instances over which we are generalizing are related to one another’’, to
affirm that Tom is mortal or Tom is not mortal, snow is white or snow
is not white, and so on, we must use a truth predicate that applies to
sentences.
Brian Loar rejects Quine’s argument that a truth predicate is indispensable
for generalizing on sentences. Loar writes:
As an explanation of the point of ‘true’, [the Quine passage cited above] leaves me
puzzled. Evidently ‘when we want to generalize . . .’ does not bear its meaning on
its face. One way of generalizing with regard to sentences of the form ‘p or not
p’ is to say we would accept all instances; but that does not require ‘true’. No, ‘we
want to generalize’ must mean ‘with regard to their truth’. But how does that tell
us what is important about truth? Why do we want to generalize in terms of T and
not some other T ? Why any truth predicate? (Loar 1981: 171–2)

² For syntactical criteria for substituting regimented predicates or sentences for schematic term and
sentence letters, see Quine 1982: chapters 26 and 28.
42   ‒ 

But these questions reveal a misunderstanding. To say that we would

accept all sentences of the form ‘S ∨ ∼S’ is one way of saying something
general about sentences of the form ‘S ∨ ∼S’, but it is not to generalize
on sentences of that form, in the sense of ‘generalize on’ that is relevant
to Quine’s reasoning. Quine notes that ‘‘we can generalize on ‘Tom
is mortal’, ‘Dick is mortal’, and so on, without talking of truth or of
sentences; we can say ‘All men are mortal’.’’³ The crucial consideration
is that the generalization ‘All men are mortal’ implies each of ‘Tom is
mortal’, ‘Dick is mortal’, and so on, given the corresponding tacit premises
‘Tom is a man’, ‘Dick is a man’, and so on. These elementary implications
serve as Quine’s model for what it takes to generalize on sentences, and
are therefore worth examining in detail. The generalization ‘All men are
mortal’ may be paraphrased as ‘For all x, if x is a man, then x is mortal’,
which, like any universal generalization, implies each of its instances. In
particular, ‘For all x, if x is a man, then x is mortal’ implies ‘if Tom is a
man, then Tom is mortal’. Finally, ‘if Tom is a man, then Tom is mortal’
and the tacit premise ‘Tom is a man’ together truth-functionally imply
‘Tom is mortal.’ In the same way, we can derive any sentence of the
form ‘N is mortal’, where the position marked by ‘N’ is filled by a proper
name, if we accept a corresponding premise of the form ‘N is a man.’ This
example suggests that to generalize on ‘Tom is mortal’, ‘Dick is mortal’,
and so on, is to write or utter a (non-contradictory) generalization that,
together with some independently plausible auxiliary premises that are not
identical to the sentences that we wish to generalize on, implies all those
sentences. Similarly, I suggest, to generalize on regimented L-sentences
of a given form is to write or utter a (non-contradictory) generalization
that, together with some independently plausible auxiliary premises that
are not identical to any of the L-sentences of that form, implies them
all.⁴ For instance, to generalize on L-sentences of the form ‘S ∨ ∼S’,
such as
(Tom is mortal) ∨ ∼(Tom is mortal)
(Snow is white) ∨ ∼(Snow is white)

³ Quine 1986: 11. Loar does not quote this sentence; perhaps he just overlooked it.
⁴ It is likely that Quine’s use of the phrase ‘generalizes on’ is not taken from ordinary language, but
from the natural deduction rule known as universal generalization (UG). In standard systems of natural
deduction, one ‘generalizes on’ sentences in Quine’s sense when one applies the rule UG, which allows
one to infer ‘∀x(x = x)’, for instance, from ‘z = z’.
   ‒  43

and so on, is to write a (non-contradictory) generalization that, together

with auxiliary premises that are not identical to any of those sentences,
implies them all.
Do Loar’s proposed alternatives generalize on sentences in this sense?
Consider Loar’s generalization, ‘‘We would accept all sentences of the
form ‘S ∨ ∼S’,’’ which we may rephrase as ‘‘We would accept any given
sentence of the form ‘S ∨ ∼S’.’’ This generalizes on such sentences as
‘‘we accept the sentence ‘(Tom is mortal) ∨ ∼(Tom is mortal)’ ’’ and ‘‘we
accept the sentence ‘(Snow is white) ∨ ∼(Snow is white)’ ’’. Hence Loar’s
generalization does not imply ‘(Tom is mortal) ∨ ∼(Tom is mortal)’, for
instance, unless we also accept the conditional
(a) If we accept the sentence ‘(Tom is mortal) ∨ ∼(Tom is mortal)’,
then (Tom is mortal) ∨ ∼(Tom is mortal).
Granted, (a) is true, because its consequent is (logically) true. But to
generalize on sentences by using Loar’s method, we would need to suppose
that it is epistemically reasonable to accept any given instance of the pattern
(A) If we accept the sentence ‘ ’, then .
To see what is attractive, yet misleading about (A), consider the corres-
ponding first-person singular version:
(A ) If I accept the sentence ‘ ’, then .
If it were epistemically reasonable to accept every instance of (A ), it would
be epistemically reasonable to accept
(b) If I accept the sentence ‘There are black holes’, then there are black
holes.
It would then be epistemically reasonable to infer that there are black holes
from (b) and ‘I accept the sentence ‘There are black holes’.’ Why would
anyone be tempted to trust such an inference? One possible explanation is
that after reflection, no one can accept the sentence
(c) I accept the sentence ‘There are black holes’, but there are no black
holes.
The problem is that to accept (c) is to suppose that one accepts the sentence
‘There are black holes’, and thereby to suppose that one believes that there
44   ‒ 

are black holes; yet to accept (c) is also to assert that there are no black
holes, and thereby to express one’s belief that there are no black holes.
To accept (c), then, is to express, and thereby commit oneself to, beliefs
that one knows to be incompatible. Hence anyone who utters (c) expresses
incompatible beliefs.⁵ An analogous problem affects some statements made
by using the first person plural pronoun ‘we’, such as
(d) We accept the sentence ‘There are black holes’, but there are no
black holes.
But if anyone who utters (d) expresses incompatible beliefs, and any
reflective person can come to see this by going through the reasoning just
sketched, then one might think that those who go through this reasoning
are entitled to accept
(e) If we accept the sentence ‘There are black holes’, then there are
black holes.
Since this reasoning applies to any sentence, it may seem that anyone who
goes through the above reasoning is entitled to accept all instances of the
form (A), just as Loar’s objection suggests.
But this appearance is illusory. To see why, consider a sentence, such as
‘The Earth is flat’, that we do not actually accept. By asserting either
(f ) John accepts the sentence ‘The Earth is flat’, but the Earth is not
flat.
or
(g) They accept the sentence ‘The Earth is flat’, but the Earth is not
flat.
we do not commit ourselves to incompatible beliefs. Moreover, we take
for granted that no one’s acceptance of ‘The Earth is flat’ implies that the
Earth is flat. Granted, we can’t simultaneously express a belief and assert that
what we believe is not so. But the intelligibility of assertions of (f ) and
(g), together with our understanding of acceptance of sentences and the
beliefs such acceptances express, show that sentences of the form (A) are
not generally acceptable, despite the problems with asserting (d) and (e).

⁵ This is a metalinguistic analogue of G. E. Moore’s paradox about belief.

   ‒  45

Hence the auxiliary premises on which we would have to rely to derive

the sentences that interest us from Loar’s generalization are not generally
acceptable. I conclude that while Loar’s generalization says something general
about all sentences of the form ‘S ∨ ∼S’, it does not generalize on them in
the sense relevant to Quine’s reasoning.
In contrast, if we assume standard syntactic criteria for substitution of
regimented predicates or sentences for schematic term and sentence letters,
then from the sentence ‘Every regimented sentence of the form ‘S ∨ ∼S’
is true’ we may infer
‘(Tom is mortal) ∨ ∼(Tom is mortal)’ is true
‘(Snow is white) ∨ ∼(Snow is white)’ is true
and so on. To get from these sentences to
(Tom is mortal) ∨ ∼(Tom is mortal)
(Snow is white) ∨ ∼(Snow is white)
and so on, we need to accept conditionals of the form
‘ ’ is true → .
Together with these conditionals, we can derive every sentence of the form
‘S ∨ ∼S’ from ‘Every regimented sentence of the form ‘S ∨ ∼S’ is true’.
And to get from sentences in the second list to sentences in the first list we
need to accept conditionals of the form
→‘ ’ is true.⁶
Hence it seems that to generalize on sentences by using a predicate ‘is true’
we need to accept T-sentences such as
‘(Snow is white) ∨ ∼(Snow is white)’ is true ↔ ((Snow is white)
∨ ∼(Snow is white)),

⁶ If we assume that our metalanguage ML is consistent and contains an ω-rule that would license
us to derive ‘Every regimented sentence of the form ‘S ∨ ∼S’ is true’ from the set of all and only
regimented sentences of the form ‘S ∨ ∼S’, we can conclude that the set of all and only regimented
sentences of the form ‘S ∨ ∼S’ and ‘Every regimented sentence of the form ‘S ∨ ∼S’ is true’ are
interderivable in ML. As Volker Halbach points out, however, if we add such an ω-rule, we would
have in effect a definition of ‘is true’ for sentences of L (see Halbach 1999: 11–12). Those who want to
add a truth predicate without defining it will therefore not want to accept the ω-rule. They will prefer
simply to add all the T-sentences to L. For a clarification of the sense in which, given the T-sentences,
the sentence ‘Every regimented L-sentence of the form ‘S ∨ ∼S’ is true,’ for instance, is ‘equivalent’
to the set of all and only regimented L-sentences of the form ‘S ∨ ∼S’, see Halbach 1999: 12–15.
46   ‒ 

which are of the form

(T) ‘ ’ is true ↔ .
with the blanks uniformly filled by sentences of our regimented lan-
guage.⁷
We may summarize this reasoning as follows. Suppose we wish to
generalize on sentences of the form ‘S ∨ ∼S’. The recursive grammar
of our regimented language generates infinitely many such sentences. We
cannot actually write out all of those sentences, and so we cannot directly
affirm the conjunction of all of those sentences. We therefore need some
way to affirm all of those sentences without actually writing them out. We
can generalize on the regimented sentences of the form ‘S ∨ ∼S’, if we
assert ‘Every regimented sentence of the form ‘S ∨ ∼S’ is true,’ provided
that we accept any given sentence that results from the pattern (T) when its
blanks are uniformly filled by a given sentence of our regimented language.
Moreover, we know of no way of generalizing on all sentences of a given
form without using a truth predicate. (I will consider two serious challenges
to this claim in the next section.) Hence to generalize on all sentences
of a given form, a truth predicate is indispensable. I shall call this the
indispensability argument.
Let me digress briefly to highlight an assumption of the indispensability
argument that usually goes without saying, but that I will scrutinize,
reject, and modify in Chapters 3 and 4. The assumption is that when we
generalize on sentences of a given logical form, such as ‘S ∨ ∼S’, we identify those
sentences solely by their syntactical shapes. We accept this assumption, in effect,
if we restrict ourselves to the standard syntactic criteria for substitution
of regimented predicates or sentences for schematic term and sentence
letters that occur in our generalizations on sentences. As I noted above,
it is by relying on such accounts of substitution that we can infer, for
example, that

⁷ To draw the distinction between saying something general about sentences, on the one hand, and
generalizing on them, on the other, we need at least a working grasp of logical implication for sentences
of our language. I assume we grasp the notion of logical implication simply in virtue of following
deductive inference rules for the sentences of our language, and, in particular, for the language in
which we express the T-sentences. We could define a truth predicate for those sentences, and then
define implication in terms of the preservation of truth (or the logical truth of certain conditionals).
But I assume that we need not define implication in terms of truth for sentences of our metalanguage
in order to distinguish between saying something general about sentences of a given object language L,
on the one hand, and generalizing on sentences of L, on the other.
   ‒  47

‘(Tom is mortal) ∨ ∼(Tom is mortal)’ is true

and
‘(Snow is white) ∨ ∼(Snow is white)’ is true
are both instances of the generalization ‘‘Every sentence of the form ‘S ∨ ∼S’
is true.’’⁸ For reasons I previewed in the Introduction, I shall propose in
Chapter 4 that we conceive of words in a way that licenses us to apply
our own disquotational truth predicate to sentences that we do not identify
solely by their syntactical shapes. If we adopt my proposed alternative
conception of words, then we can each still identify sentences of our own
regimented language solely by their syntactical shapes, but we can no longer
claim that the italicized assumption holds for all sentences to which our
own disquotational truth predicate applies. In particular, if we adopt my
proposed alternative conception of words, there are some instances of the
generalization ‘‘Every sentence of the form ‘S ∨ ∼S’ is true’’ that do not
count as instances of that generalization given the italicized assumption.
In this Chapter I shall nevertheless provisionally accept that the italicized
assumption holds for all sentences to which our own disquotational truth
predicate applies.
I return now to my clarification and evaluation of the indispensability
argument. If the argument is correct, then to generalize on sentences of a
given form, we need to accept every T-sentence. Since there are infinitely
many T-sentences, however, we cannot express our acceptance of every
one of them by writing them out. How then do we express our acceptance
of the T-sentences themselves? We cannot do so by affirming ‘Every
T-sentence is true,’ since this generalization presupposes that we have
already accepted T-sentences for the T-sentences themselves, and would
therefore either be viciously circular, or start us on an infinite regress of
generalizations and corresponding T-sentences.⁹ Hence the indispensability

⁸ To handle other generalizations, such as ‘‘Every regimented sentence of the form ‘∀x(Fx ∨ ∼Fx)’
is true’’, we will of course also need a purely syntactical (orthographic) account of substitution of
regimented predicates for schematic predicate letters, such as ‘F’, ‘G’, ‘H’, and so on. Because of
the relationship between free and bound variables, the rules for substituting predicates for schematic
predicate letters are trickier to state than the rules for substituting closed sentences for schematic
sentence letters, but such rules are nevertheless widely known and can be found in any good text on
first-order logic, such as Quine 1982.
⁹ In Putnam 1983, Hilary Putnam raises a related concern about circularity. Unlike my concern in
the text, however, his is not directed at the indispensability argument, but instead presupposes that the
goal of an account of truth is to capture the meaning of the ordinary word ‘true’.
48   ‒ 

argument presupposes that if we can use a truth predicate to generalize

on infinitely many sentences, there must be a way for us to affirm every
T-sentence without using a truth predicate. But now it may seem that the
argument faces another version of Loar’s objection. For if we can affirm
every T-sentence without using a truth predicate, then we should also be
able to affirm every sentence of the form ‘S ∨ ∼S’ without using a truth
predicate, just as Loar suggests.
This objection rests on a subtle but crucial conflation of (a) generalizing on
infinitely many sentences of a given form with (b) committing oneself to affirming any
sentence of a given form. To see the difference between (a) and (b), recall that
in first-order predicate logic, the logic of identity is completely captured
by accepting ‘∀x(x = x)’ and any sentence of the form ‘∀x∀y((Fx ∧ x =
y) → Fy)’ as axioms, together with any legitimate reletterings of the bound
variables in either of these axioms. We may regard the relettering rules
as additional to ‘∀x(x = x)’, in which case it is not schematic, but directly
generalizes on the infinitely many regimented sentences ‘x = x’, ‘y = y’,
‘z = z’, and so on, that it implies. When the axiom ‘∀x(x = x)’ is viewed
in this way, to accept it is to state, in effect, that everything is self-identical.
In contrast, the second of these axioms is schematic; by accepting it we
do not state anything at all, but instead commit ourselves to affirming any
regimented sentence of the form ‘∀x∀y((Fx ∧ x = y) → Fy)’. To adopt
the axiom schema ‘∀x∀y((Fx ∧ x = y) → Fy)’ is therefore not to generalize
on all sentences of that form, but only to commit ourselves to affirming any
sentence of that form. Similarly, contrary to the objection sketched in the
previous paragraph, to regard the schema (T) as an axiom schema is not
to generalize on all T-sentences, but only to commit ourselves to affirming any
T-sentence. I conclude that to characterize a truth predicate that we can
use to generalize on infinitely many sentences of a given form, it is enough
to commit ourselves to affirming any T-sentence.

2.2. Why Generalize on Valid Sentences?

One might be convinced that to generalize on infinitely many sentences of
a given form, we need to use a truth predicate, but wonder why we need to
generalize on sentences of a given form, instead of just committing ourselves
to affirming each of the sentences. To generalize on sentences of the form
   ‒  49

‘S ∨ ∼S’, for instance, we would need to affirm a non-contradictory

generalization that implies all of them, and for this, according to the
indispensability argument, we would need a truth predicate. But if we
treat ‘S ∨ ∼S’ as an axiom schema, we commit ourselves to affirming
each regimented sentence of the form ‘S ∨ ∼S’ without generalizing on all
those sentences. Hence if we treat ‘S ∨ ∼S’ and other valid logical schemata
as axiom schemata, we can accept the indispensability argument without
concluding that we need to use a truth predicate. What then motivates us
to generalize on the instances of such schemata?
I assume that our desire to clarify and facilitate inquiry brings with it
a desire to classify particular regimented sentences as valid (logically true).
Given this desire, we can begin to answer the question just posed by asking
how generalizing on sentences is related to classifying particular regimented
sentences as valid. An answer is suggested by the following observation:
If one’s first-order regimented language L includes the identity sign and predicates
that express addition and multiplication, then to classify one’s L-sentences as
valid is equivalent to generalizing on them.
This observation is a consequence of standard definitions of validity and
a theorem in recursion theory. The definitions are as follows (note the
subscripts):
(D1) A first-order logical schema is valid1 if and only if it is true for all
interpretations.
(D2) A regimented sentence is valid2 if and only if (a) there is a first-
order logical schema S from which the regimented sentence results
by substitution of regimented sentences or predicates for sentence
or term letters, and (b) schema S is valid1 .
From (D1) and (D2) we may infer that a particular sentence is valid2
if and only if it results by substitution from a schema that is true for
all interpretations. The sentence ‘∀x((x is white) ∨ ∼(x is white))’, for
instance, is valid2 , since it results by substitution from the schema ‘∀x(Fx ∨
∼Fx)’, which is true for all interpretations.
An interpretation of a closed first-order schema S is usually taken
to be an ordered pair of a non-empty domain of discourse D and an
assignment A of objects in D to the extensions of the predicates in
S, where it is assumed that A may be specified without substituting
50   ‒ 

regimented predicates for term letters in S. When A is specified without

substituting regimented predicates for term letters in S, truth-relative-to-
an-interpretation is not the same as disquotational truth, as characterized
by T-sentences. Nevertheless, a theorem in recursion theory relates the
two. The Löwenheim–Hilbert–Bernays theorem, due to David Hilbert
and Paul Bernays, extends results of Leopold Löwenheim. It establishes
that if one’s first-order regimented language L includes the identity sign
and predicates that express addition and multiplication, then a schema S
of L is true for all interpretations in all non-empty domains of discourse
if and only if every L-sentence that results by substitution from schema S
is true.¹⁰ The L-sentence ‘∀x((x is white) ∨ ∼(x is white))’, for instance,
is valid2 if and only the schema ‘∀x(Fx ∨ ∼Fx)’ is valid1 ; and the schema
‘∀x(Fx ∨ ∼Fx)’ is valid1 if and only if every L-sentence of the form
‘∀x(Fx ∨ ∼Fx)’ is true. Hence if one’s first-order regimented language
L includes the identity sign and predicates that express addition and
multiplication, then one way to classify particular L-sentences of a given
schematic form S as valid2 is to classify S as valid1 by using a disquotational
truth predicate.¹¹

¹⁰ According to Quine, the Löwenheim–Hilbert–Bernays theorem was first proved in Hilbert and
Bernays 1939: 234–53, and later strengthened, as Theorem 35, in Kleene 1953. Quine explains the
central idea behind the theorem in Quine 1995a. In Quine 1982, ch. 33, Quine claims that Hilbert and
Bernays give ‘‘a general rule for writing out, for any schema, an arithmetical interpretation which can be
depended upon to come out true in case the schema does happen to be consistent’’ (Quine 1982: 211).
This is misleading, however, as Carl Jockusch (a recursion theorist at the University of Illinois-Urbana)
explained to me: the Hilbert–Bernays routine for generating an arithmetical interpretation of a schema
depends on treating a non-constructive assignment of truth values to simple sentences as an ‘‘oracle’’. If
we don’t know a verifying assignment of truth values to simple sentences, we cannot actually write
out the arithmetical interpretation, even if we know it exists. The dependence in the general case of
the arithmetical interpretation on a non-constructive assignment of truth values to simple sentences is
reflected in George Boolos’s careful formulation of what Hilbert and Bernays actually proved: ‘‘any
satisfiable schema is satisfied by a model whose domain is the set of natural numbers and whose
predicates are assigned relations on natural numbers that can be defined in arithmetic’’ (Boolos 1975:
525, emphasis in original). Quine exploits this model-theoretic result to establish what he calls the
Löwenheim–Hilbert–Bernays theorem.
¹¹ Strictly speaking, by this criterion of validity, sentences and schemata that can only be proved from
the identity axioms noted in §2.1 do not count as valid. As Quine points out, however, there are at
least two reasons for counting such sentences and schemata as valid: first, complete proof procedures are
available for quantification theory supplemented with the identity axioms, which together constitute
the standard logical identity theory; second, this identity theory, like quantification theory, ‘‘treats all
objects impartially’’ (Quine 1986: 62). To count schemata that can be proved only from the identity
axioms as valid by generalizing on sentences, however, the best we can do is to show that particular
explications of them are valid. To construct the relevant explications for a given regimented language L,
it is enough to define ‘xi = xj ’ as a mere definitional abbreviation for an open sentence of L that contains
no occurrences of ‘=’ but is nevertheless satisfied by a sequence s if and only if si = sj . A recipe for
   ‒  51

But there are also ways of classifying a particular L-sentence s of a

given schematic form S as valid1 without using a truth predicate at all. For
instance, a given schema S is valid1 if and only if its negation generates
a contradiction by a standard method of proof for first-order logic. By
definition (D2), then, s is valid2 if and only if the negation of S generates
a contradiction by a standard method of proof for first-order logic. Why
then should we use a truth predicate to classify schematic form S as
valid1 ?
The answer is that we cannot simply take for granted that standard
methods of proof for first-order logic are sound and complete. To prove
that they are sound and complete, however, we need an independent
characterization of what it is to be valid1 . And for this we need a truth
predicate.
Given (D1), however, why not just classify a schema as valid1 if and
only if it is true under all interpretations, and leave it at that? One reason
is that this preliminary characterization of when a schema is valid1 does
not tell us how the notion of truth that we need to classify a schema as
valid1 is related to the T-sentences that we need to generalize on sentences.
If an interpretation I of S is given by substituting regimented predicates
P1 , . . . , Pn for term letters in S, then S is true relative to I if and only if
the regimented English sentence that results from substituting P1 , . . . , Pn
for term letters in S is disquotationally true. In such cases, we can explain
the two-place relation ‘S is true relative to I’ in terms of a one-place
predicate ‘true’ characterized by T-sentences, which are as clear to us as
the regimented sentences they contain. If we had no general guarantee that
an interpretation I of S can be given by substituting regimented predicates
for term letters in S, however, then we would have no guarantee that we
could explain ‘S is true relative to I’ in terms of a one-place predicate
‘true’ characterized by T-sentences. Lacking such a guarantee, we might
conclude that the notion of truth that we need if we are to classify schemata
as valid1 cannot be explained disquotationally, hence cannot be shown to

constructing such an open sentence is given on page 63 of Quine 1986. By using Quine’s recipe, we
can define ‘xi = xi ’, in particular, as a definitional abbreviation for an open sentence q of L that contains
no occurrences of ‘=’ but is nevertheless satisfied by a sequence s if and only if si = si . Given this
definition, ‘∀xi (xi = xi )’ is a definitional abbreviation for ‘∀xi q’, which ‘‘is a truth purely of the logic
of quantification and truth functions’’ (Quine 1986: 63), which is to say that (a) there is a first-order
logical schema S from which ‘∀xi q’ results by substitution of regimented sentences or predicates for
sentence or term letters and (b) every L-sentence of the form S is true.
52   ‒ 

be as clear to us as our own regimented sentences. Fortunately, however,

to classify schemata as valid1 we do not need to make use of a notion of
truth that cannot be explained disquotationally. For we can easily include
in our regimented languages the identity sign and predicates that express
addition and multiplication. The Löwenheim–Hilbert–Bernays theorem
then ensures that we can identify interpretations with particular sentences
of our regimented language, and classify schemata as valid1 by generalizing
on regimented L-sentences using a disquotational truth predicate.¹²

2.3. Three Attempts to Generalize on Sentences

without Using a Truth Predicate
Let us now return to the indispensability argument. Is our confidence in
it due to our failure to think of another way of generalizing on sentences?
To address this question, I shall now examine three ways of generalizing
on sentences without using a truth predicate.
To begin with, one might try to generalize on sentences by using first-
order quantification. Instead of affirming ‘Every regimented sentence of
the form ‘S ∨ ∼S’ is true,’ and ‘Every regimented sentence of the form
‘∀x(Fx ∨ ∼Fx)’ is true,’ for instance, one might try
(D) ∀x(x ∨ ∼x)
and
(E) ∀x∀y(xy ∨ ∼xy)
A superficial problem with this attempt is that the standard rules for
constructing sentences of L do not yield either (D) or (E). A deep problem
with the attempt is that even if we adopt new grammatical rules that imply
that (D) and (E) are sentences, if we take ‘∀x’ to be a first-order quantifier
that ranges over objects, we will not be able to make sense of (D) or (E).¹³

¹² Quine emphasizes that, unlike (D1), as it is usually understood, a definition of validity in terms of
disquotational truth ‘‘renders the notions of validity and logical truth independent of all but a modest bit
of set theory; independent of the higher flights’’ (Quine 1986: 56). This is an important consideration,
but not, I think, as important as the fact that a definition of validity in terms of disquotational truth is
as clear to us as the sentences for which our disquotational truth predicate is defined.
¹³ For reasons that Joan Weiner explains in Weiner 2005, we can make sense of something like
(D) if, following Frege, (a) we accept that among the objects in the domain of discourse for L are the
   ‒  53

A more promising strategy is to use second-order quantification into

predicate and sentence positions. Instead of affirming ‘Every regimented
sentence of the form ‘S ∨ ∼S’ is true,’ and ‘Every regimented sentence of
the form ‘∀x(Fx ∨ ∼Fx)’ is true,’ one might affirm such sentences as
(F) ∀p(p ∨ ∼p)
and
(G) ∀F∀y(Fy ∨ ∼Fy)
where the quantifiers ‘∀F’ and ‘∀p’ are understood as second order, and so
the variables ‘F’ and ‘p’ occur in positions in a sentence where it would
not make sense to place a first-order variable or a singular term.
Most discussions of second-order quantification focus on quantification
into n-place predicate positions, where n ≥ 1. According to the standard
semantics for second-order languages, a variable assignment s is a function
that assigns a member of a given domain of discourse D to each first-order
variable and a subset of Dn to each n-place relation variable. If we accept
this semantics, then we will take (G) to be true relative to a given domain
D if and only if for every variable assignment s, either s(y) is an element
of s(F) or s(y) is not an element of s(F) (Shapiro 1991: 72). The standard
semantics treats 1-place predicates as of the form ‘is an element of S’ for
some set S of elements of D; predicates are therefore viewed extensionally:
their meanings—the properties or attributes, if any, that they express—are
irrelevant to the truth or falsity of second-order quantifications.
We can see (F) as a special case of quantification into predicate positions if
we see sentences as 0-place predicates. To extend the standard semantics to
apply to 0-place predicates, we must assign extensions, namely, true or false,
to them. But if this is the proper semantics for a sentence such as (F), then

true and the false, (b) we treat sentences as proper names of the true or the false, (c) we add Frege’s
horizontal (symbolized by ‘—’), defined as ‘= the true’, to our language, and (d) we rewrite (D) as

(DF ) ∀x(—x ∨ ∼—x)

( To make sense of (DF ), we must also regard the logical connectives ‘∨’ and ‘∼’ not as symbols
for operations, in Wittgenstein’s sense, but as names of Fregean functions from truth values to truth
values.) Even if we are willing to recast (D) as (DF ), however, this will not help us to make sense of
(E). To generalize on sentences of the form ‘∀x(Fx ∨ ∼Fx)’ in the way that Frege did, one must add
second-order quantifiers to one’s language. I discuss this strategy in the next two paragraphs. For a
model-theoretical interpretation of the first-order part of Frege’s logic, including definitions of the true
and the false, see Parsons 1987.
54   ‒ 

second-order quantification over sentences can be simulated in our first-

order language L. Since L is extensional, we can simply choose an arbitrary
closed sentence of L, say ‘∀x1 (x1 is mortal)’, and define ‘∀p( . . . p . . . )’
as ‘( . . . ∀x1 (x1 is mortal) . . . ) ∧ ( . . . ∼ ∀x1 (x1 is mortal) . . . )’.¹⁴ There
is no extensional difference between any two true sentences or any two
false ones, so this definition is guaranteed to yield, for each second-order
universal quantification over sentences, an L-sentence that has the same
truth value as that quantification.
Such simulations of second-order quantification do not generalize on
sentences, however. If we wish to use (F) to generalize on sentences, we
should not presuppose the standard semantics for second-order quanti-
fication. Instead, we should try to regard second-order quantification as
primitive. We will still need to explain the notations we use to express it,
but for this purpose it may be enough to elucidate the inference rules and
axioms for a particular artificial language that allows second-order quanti-
fication. We may also be able to appeal to second-order quantifications in
ordinary language.¹⁵
It is an open question whether we can give elucidations or explanations
of second-order quantification that satisfy us. But suppose we can. The
strategy for generalizing on sentences by using second-order quantification
still faces two closely related problems. First, unlike first-order logic,
second-order logic is incomplete—there are complete proof procedures
for validity in first-order languages, such as L, but no complete proof
procedures for validity in second-order languages. If we wish to state
logical generalizations for a first-order language, such as L, for which there
are complete proof procedures, we should not do so by adding second-
order quantifiers and variables. Second, ‘∀p( p ∨ ∼p)’ does not generalize
exclusively on sentences of L.¹⁶ To see why, consider again the first-order
quantification, ‘∀x(x = x)’. This implies ‘w = w’ for any free variable ‘w’
in L. We might not have a first-order name or description to put in
place of the variable ‘w’, yet we can still regard the open sentence as true

¹⁴ See Quine 1986: 74–5. For a similar simulation of second-order quantification, see Hilbert and
Ackerman 1950: 127–8. The significance of Quine’s simulation for our understanding of whether we
can use second-order quantification to generalize on sentences without using a truth predicate is noted
in Forbes 1986.
¹⁵ For one interesting attempt to show that ordinary language makes use of second-order quantific-
ation, see Boolos 1984.
¹⁶ As Mike Dunn pointed out to me.
   ‒  55

relative to an assignment of some object from our domain of discourse to

‘w’. Analogously, if we accept second-order quantification as primitive, we
commit ourselves to inferences from ‘∀p( p ∨ ∼p)’ to ‘p ∨ ∼p’, even when
we do not put a sentence in place of the free variable ‘p’. Hence if we wish
to generalize exclusively on sentences of L, perhaps for the reasons explained
in the previous section, we should not use second-order quantification into
sentence positions.¹⁷
The most promising way to generalize exclusively on sentences of L
without using a truth predicate is to use substitutional quantification. By
hypothesis, L has the grammar of first-order predicate logic, so we can
exploit the grammatical structure of L to specify all (and only) the sentences
of L recursively. We can think of the set thus specified as the domain of
a universal substitutional quantifier ‘’ which binds an expression-variable
‘S’ that marks places in a sentence of L for which we may substitute any
sentence of L. Now consider
(H) S(S ∨ ∼S)
Can we use (H) in place of the generalization ‘Every regimented sentence
of the form ‘S ∨ ∼S’ is true’? We cannot do so unless we can understand
the universal substitutional quantifier that occurs in (H). One might first
try to explain the contribution of the universal substitutional quantifier as
follows:
(E1) ‘S( . . . S . . . )’ is true if and only if, for every sentence s of L, the
sentence that results from replacing the free occurrences of ‘S’ in
‘( . . . S . . . )’ by s is true.¹⁸
One might object that this explanation makes use of a truth predicate, and
therefore does not help us to understand how we can use substitutional
quantification to express logical generalizations without using a truth pre-
dicate. But (E1) should not be understood as an account of the meaning
of ‘S’: a sentence of the form ‘S( . . . S . . . )’ does not mention (refer
to) any of the sentences over which the quantifier ranges (Soames 1999:
91–2). A proponent of (E1) might therefore claim that once we understand

¹⁷ Some philosophers think we want and need to be able to make sense of an open second-order
sentence, such as ‘p ∨ ∼p’, even when we do not put sentences of L in place of the free variable ‘p’. I
shall examine one version of this view in the next section.
¹⁸ This formulation is due in essentials to David 1994: 86.
56   ‒ 

substitutional qualification by means of (E1), we can simply use a sentence

such as (H) to express generalizations without using a truth predicate. But
this reply just presupposes that we can use and understand (H) without using
a truth predicate. We are so far at a loss to understand (H) without relying
on (E1), which contains a truth predicate.¹⁹
To avoid this problem, we need to provide a different explanation of
the universal substitutional quantifier. One might try:
(E2) The notation ‘S( . . . S . . .)’ encodes the conjunction of the open
or closed sentences that result from replacing the free occurrences
‘S’ in ‘( . . . S . . .)’ by sentences of L.²⁰
If we explain the notation ‘S( . . . S . . .)’ in this way, then we must
concede that no finite, first-order sentence of English or Regimented
English paraphrases (H), which we must therefore regard as an abbreviation
for the infinitely long conjunction of all and only the sentences of the form
‘S ∨ ∼S’. Similarly, suppose we try to use a substitutional quantifier on
predicate-variables to generalize on sentences of the form ‘∀x(Fx ∨ ∼Fx)’
without using a truth predicate. For instance, we may write
(I) F∀x(Fx ∨ ∼Fx)
where
(E3) The notation ‘F( . . . F . . .)’ encodes the conjunction of the open
or closed sentences that result from replacing the free occurrences
‘F’ in ‘( . . . F . . .)’ by predicates of L. (David 1994: 99)
If we explain the notation ‘F( . . . F . . .)’ in this way, then we must
concede that no finite, first-order sentence of English or Regimented
English paraphrases (I), which we must therefore regard as an abbreviation
for the infinitely long conjunction of all and only L sentences of the form
‘∀x(Fx ∨ ∼Fx)’.²¹ In general, for each logical law that we can state by using

¹⁹ This criticism of (E1) is adapted from David 1994: 86, where it is used to reject substitutional
definitions of ‘true-in-L’. For a lucid investigation of some of the issues at stake here, see Kripke
1976.
²⁰ See David 1994: 99. Van Inwagen 2002 traces the idea back to a book review by Hartry Field in
the 1980s. Before that, however, the idea was already suggested in footnote 9 of Leeds 1978.
²¹ This substitutional approach only works if the resulting conjunctions contain no free variables.
This is not a serious limitation, however, because an open sentence is logically true if and only if its
corresponding universal closure is logically true. Thus we can say that an open sentence is logically true
if and only if the appropriate substitutional generalization on its universal closure is true.
   ‒  57

a truth predicate, there is a corresponding substitutional generalization like

(H) or (I) that contains no truth predicate, but abbreviates the corresponding
infinite conjunction of sentences we wish to affirm.
Does this show that we can generalize on sentences without using
a truth predicate? The answer depends on whether we can understand
explanations such as (E2) and (E3) without relying on a truth predicate.
We can understand such explanations only if (a) the language that we use
to paraphrase the substitutional quantifications contains the infinitely long
sentences for which the substitutional generalizations are to be regarded
as abbreviations, and (b) we understand those infinitely long sentences. But
it is unclear whether English or Regimented English contains infinitely
long sentences.²² And even if the grammar of English or Regimented
English recursively generates such sentences, it is unclear whether we can
understand those sentences directly, without relying on a truth predic-
ate.²³ One might try to lay down inference rules, on analogy with the
standard instantiation and generalization rules for objectual quantifiers, in
the hope of explaining our understanding of substitutional quantifications
without presupposing a direct grasp of infinite conjunctions or disjunctions.²⁴
But to stipulate that such rules together implicitly define the quantific-
ations is, in effect, to regard the quantifications as infinite conjunctions
or disjunctions. The two strategies are at root one, and hence leave it
equally unclear whether we can make sense of the substitutional quanti-
fications without using a truth predicate. A corollary of this conclusion
is that if we are unclear about whether we understand infinitely long
sentences, it will not help us to use a truth predicate defined by the
axiom
(S) (‘S’ is true if and only if S)
as Hartry Field proposes (Field 1994: 259). I shall henceforth assume that to
generalize on sentences, we need to use a truth predicate that is not itself
defined in terms of substitutional quantification.

²² Speaking about an infinitely long sentence ‘‘generated’’ from a multiply substitutionally quantified
generalization, Peter Van Inwagen writes that ‘‘It is really very hard to believe that there are such
sentences as these’’ (Van Inwagen 2002: 215).
²³ For instance, Van Inwagen says he cannot even understand very long sentences, such as those that
contain 3,999 words (Van Inwagen 2002: 216).
²⁴ As suggested in Hill 2002: 17–22.
58   ‒ 

2.4. Horwich’s Minimal Theory

Even if one grants that to generalize on sentences of our own regimented
language L we need to use a truth predicate, one might still ask why we
should define our truth predicates so that they apply in the first instance to
sentences of L, not to the propositions expressed by those sentences. Paul
Horwich observes that we typically apply our ordinary English word ‘true’
to propositions expressed (what is said ) by particular utterances of English
sentences (Horwich 1998a: 16, 35). When ‘true’ is applied to propositions
expressed by particular utterances of English sentences, however, it might
just as well be applied to the sentences directly, for reasons I will explain
below. There is significant difference between a disquotational definition
of ‘true’ for sentences we can use, on the one hand, and a definition of
‘true’ for the propositions expressed by sentences we can use, on the other,
only if the latter definition is also intended to yield a definition of ‘true’ for
propositions that we cannot express. A central task of Horwich’s Minimal
Theory is to yield such a definition. I shall argue that the Minimal Theory
fails in this task, so we should not prefer it to a definition of truth for
sentences we can use.
Suppose that if we put a declarative sentence between the brackets
‘<’ and ‘>’, we form a singular term that denotes the proposition, if any,
expressed by (a particular utterance of ) that sentence. Then we can express
Horwich’s Minimal Theory as follows: the meaning of ‘true’ in English is
constituted by our disposition to accept any given instance of the schema
(MT) <p> is true if and only p. (Horwich 1998a: 128)
If we are disposed to affirm any given instance of (MT), as Horwich claims,
then it seems we can rely on instances of (MT) to express generalizations,
such as the logical generalization
(i) Every proposition of the form <p or not p> is true.
The basic idea—an extension, to generalizations on propositions, of an idea
originally due to Tarski—is that for a given proposition, such as <Snow is
white>,
(ii) <Snow is white or not Snow is white> is a proposition of the form
<p or not p>
   ‒  59

and hence
(iii) <Snow is white or not Snow is white> is true
follows by modus ponens from (ii) and a universal instantiation of
(i), namely,
(iv) If <Snow is white or not Snow is white> is a proposition of the
form <p or not p>, then <Snow is white or not Snow is white>
is true.
Since by assumption we are disposed to accept
(v) <Snow is white or not Snow is white> is true if and only Snow is
white or not Snow is white,
we can infer (from (iii) and (v)) that
(vi) Snow is white or not Snow is white.
According to Horwich, the Minimal Theory guarantees that a derivation
of this kind is in principle available even for propositions we cannot
express. As I shall now try to show, however, the Minimal Theory does
not explain how we make sense of there being such derivations for
propositions we cannot express, because the Minimal Theory does not
explain how we can make sense of there being instances of (MT) that
explain the meaning of ‘true’ when it is applied to propositions that we
cannot express.
To see why, one must first clear away a grammatical confusion that
undermines Horwich’s official statement of the Minimal Theory. According
to the Minimal Theory, as we’ve seen, the meaning of ‘true’ in English is
constituted by our disposition to accept any given instances of the schema
(MT). On the official statement of the theory, one of its axioms is
(1) <<Snow is white> is true iff snow is white>
Horwich says this is a proposition whose structure is
(E∗ ) << p > is true iff p>,
which he takes to denote a function (Horwich 1998a: 17). How should we
think of this function? What are its arguments and values? Horwich takes
(E∗ ) to be a function from propositions to propositions: he writes that ‘‘the
60   ‒ 

axiom (1) is the result of applying the propositional structure (E∗ ) to the
proposition
(3∗ ) <Snow is white>.’’ (Horwich 1998a: 18)
But this cannot be correct. To see why, suppose that (E∗ ) denotes a function
whose argument places are marked by the propositional variable ‘p’. Then
if we apply the function to (3∗ ), namely, <Snow is white>, the result is
not (1), but
(1∗ ) <<<Snow is white>> is true iff <snow is white>>.
This is nonsense, as we can see when we try to express it in English:
‘‘the proposition that the proposition that the proposition that snow is
white is true iff the proposition that snow is white.’’ The problem is that
the singular term ‘<snow is white>’ occurs in positions (one within the
singular-term forming operator ‘<>’, and another directly to the right of
the biconditional) that can only be meaningfully filled by a sentence. More
generally, to make sense of (E∗ ), we must view the variable ‘p’ that occurs
in it as a second-order variable. We therefore cannot take (E∗ ) to denote
a first-order function that takes objects (propositions) as arguments and
yields objects (propositions) as values. Horwich nevertheless asserts that the
axioms of his Minimal Theory ‘‘are given by the principle
(5) For any object x: x is an axiom of the minimal theory if and only if,
for some [object] y, when the function E∗ is applied to y, its value is
x’’ (Horwich 1998a: 19)
where the quantifiers in (5) range over objects. If we accept (5), then (1∗ )
refers to an axiom of the minimal theory. But, as we have seen, (1∗ ) is
nonsense. Hence we cannot accept (5).
How then should we try to make sense of the minimal theory? The best
way, I think, is to reject Horwich’s official explanation of the axioms of
the minimal theory, and adopt an alternative explanation that he offers in
a footnote. We should ‘‘characterize the axioms of the minimal theory as
anything that is expressed by instances of the sentence schema
(E) ‘<p> is true iff p’.’’ (Horwich 1998a: 19 n. 3)
Can we make sense of this formulation of the minimal theory? The answer
depends on whether we can understand Horwich’s further stipulation that
   ‒  61

‘‘the theory cannot be restricted to instantiations of (E) by English sentences;

for presumably there are propositions that are not expressible in current
English, and the question of their truth must also be covered. So further
‘equivalence axioms’ are needed, one for each unformulatable proposition’’
(Horwich 1998a: 18–19 n. 3). To understand this aspect of the minimal
theory, we must somehow grasp the idea of an unformulatable value for the
second-order free variable ‘p’. The idea is at best elusive. Since the variable
is second order, and we wish to make sense of the resulting instance of
schema (E), in which the variable ‘p’ occurs in positions where it only
makes sense to put a sentence, we cannot think of the values that can be
assigned to the variable ‘p’ as objects. This makes it difficult to understand
how there could be unformulatable values for the second-order free variable
‘p’. To see why, consider instances of (E) that we can formulate, such as
<Snow is white> is true iff snow is white.
Here the ‘value’ assigned to ‘p’ is not an object, but a sentence-in-use,
which does not refer to an object or a proposition, even if it may express
one. For such cases, it would be clearer to write
The proposition expressed by ‘Snow is white’ (as used in English) is
true iff snow is white.
This formulation enables us to exploit the logician’s use-mention distinc-
tion, putting the sentence in quotation marks on the left-hand side of the
biconditional and simply using it as a sentence on the right-hand side. As
long as we have sentences to fill the positions marked by the variable, we
need not ask what ‘values’ we thereby assign to it. Hence we can avoid
the awkward question of whether we are assigning objects to the variable
‘p’. If we regard formulatable instances of (E) in this way, however, the
reference to propositions on the left-hand side can be eliminated. For
example, given that
The proposition expressed by ‘Snow is white’ (as used in English) is
true iff ‘Snow is white’ is true in English,
we can derive
‘Snow is white’ is true in English iff snow is white.²⁵

²⁵ This derivation is suggested in Davidson 1996: 32.

62   ‒ 

The problem for Horwich is that the use-mention distinction is only defined
for sentences that one can actually formulate and use. We understand the
positions marked by the variable ‘p’ as places where we can write such
sentences. There is no use-mention distinction for propositions, so we
cannot rely on the use-mention distinction to make sense for the supposed
instances of (E) that we cannot formulate. It is unclear whether we can
make sense of there being such instances at all.
Let us waive this worry, however, and suppose we can somehow make
sense of there being unformulatable values for the second-order free variable
‘p’. Can we then make sense of there being unformulatable instances of
schema (E)? In other words, can we understand what contribution the
word ‘true’ makes to such instances? It is difficult to see how. According to
Horwich, ‘‘the minimalist thesis is that the meaning of ‘true’ is constituted
by our disposition to accept those instances of the truth schemata that we
can formulate. In that way, the word is provided with the constant meaning
wherever it appears—even when ascribed to untranslatable statements’’
(Horwich 1998a: 128). But if, as Horwich claims (see Horwich 1998a:
1–2), the concept expressed by uses of ‘true’ in English has no hidden
structure—no nature beyond what is settled by our disposition to accept
any given instance of the schema (E)—then it seems that our understanding
of what it means to apply ‘true’ to a given proposition must be exhausted
by our understanding of the corresponding instance of (E). If we cannot
express the proposition p that yields a given instance of (E), we cannot state
conditions under which ‘true’ applies to p. Hence even if we suppose we
can somehow make sense of the idea that there are unformulatable values
for the second-order free variable ‘p’, the Minimal Theory does not explain
what it means to apply ‘true’ to a proposition that we cannot formulate.²⁶
On the other hand, as we’ve seen, if we can use some sentence S to express
the supposed proposition p that yields a given instance of (E), we don’t need
to ascribe truth to propositions at all, but can define truth directly for S.

²⁶ Field 1992 raises a similar criticism of Horwich’s Minimal Theory, but without remarking on
the grammatical problems with and limitations of the Minimal Theory that I explain above. It was
only after I saw these problems and limitations that I was able to formulate Field’s criticism in a way
that I find persuasive. My version of the criticism applies also to the following attempt to define truth
for propositions: ‘‘For all propositions x [x is true iff (∃P)(x = the proposition that P & P)]’’ (Soames
1999: 48). Here there is explicit second-order quantification over propositions. ‘True’ is given no
independent explanation, so we are left to rely on an instance-by-instance understanding of the matrix
‘x = the proposition that P & P’. The trouble, again, is that we have no understanding of a given value
for the free variable ‘P’ if we cannot formulate and use a sentence that expresses that value.
   ‒  63

I began this section by asking why we should define our truth predicates
so that they apply in the first instance to sentences we can use, and not
to the propositions expressed by those sentences, as Horwich’s Minimal
Theory implies. I have argued that the Minimal Theory fails at precisely the
point where it departs from a disquotational definition of truth for sentences
we can use. I conclude that for our purposes in this book, we have no
reason to adopt Horwich’s Minimal Theory.

2.5. A Naive Theory of Why it is Epistemically

Reasonable for us to Accept T-Sentences
Recall that according to the indispensability argument (§2.1), to generalize
on all sentences of a given form, we must commit ourselves to affirming
any given T-sentence without generalizing on all T-sentences. Let us now
consider whether, and, if so, why, it is epistemically reasonable for a person
to affirm any given T-sentence. One preliminary answer is that we have
decided to adopt a schema that licenses us to affirm any given T-sentence.
But why is it reasonable to adopt the T-schema as an axiom schema?
Before we try to answer this question, let us first consider the analogous
question for the axioms of identity. Why does it seem reasonable for us
to adopt these axioms? An initially appealing answer is that the axioms of
identity capture the core meaning of ‘=’, and hence anyone who both grasps the
meaning of ‘=’ and understands the axioms is epistemically justified in accepting
them. Similarly, an initially appealing answer to the question, ‘‘Why it is
reasonable to adopt the T-schema as an axiom schema?’’ is that T-sentences
together capture the core meaning of the English word ‘true’, and hence anyone
who grasps the meaning of the English word ‘true’ and understands the T-schema is
epistemically justified in affirming every T-sentence. This answer to the question,
‘‘Why it is reasonable to adopt the T-schema as an axiom schema?’’ relies
heavily on naive, pre-theoretical intuitions about the meaning of ‘true’. I
shall therefore call it the naive theory of why it is epistemically reasonable for us
to accept T-sentences, or the naive theory, for short.
Given our pragmatic goals in this book, the naive theory is unacceptable.
The main problem with the naive theory is that it rests on the notion
of meaning. To clarify and defend the naive theory, one would have to
provide an account of the meaning of the English word ‘true’ and explain
64   ‒ 

why that account shows that anyone who grasps the meaning of the
English word ‘true’ is thereby epistemically justified in accepting any given
T-sentence. But we are much more likely to agree that it is epistemically
reasonable for us to accept T-sentences than we are to agree on how to
explain the meaning of the English word ‘true’. For any given explanation
of this kind, there is likely to be only a minority of inquirers who find it
convincing; the rest will not regard the proposed explanation as adequate
justification for their assumption that it is epistemically reasonable for them
to accept T-sentences. Partisans of particular theories of the meaning of
the English word ‘true’ may of course accept those theories despite the fact
that they are controversial. But if we seek a truth predicate that we can use
for the purposes of clarifying and facilitating our inquiries, we must avoid
partisan debates about the meaning of the English word ‘true’. We must
seek a truth predicate that we can agree to use for generalizing on our
sentences, not a correct theory of the meaning of the English word ‘true’,
even if we are confident that there is such a theory.
A related problem with the naive theory is that it apparently implies that
we are epistemically justified in accepting paradoxical sentences. Suppose,
for instance, that one of the sentences of our regimented language is
(1) The sentence that occurs on lines 45–6 of §2.5 of Truth and Words
is not true.
According to the naive theory, we are each entitled to accept
(2) ‘The sentence that occurs on lines 45–6 of §2.5 of Truth and Words
is not true’ is true if and only if the sentence that occurs on lines
45–6 of §2.5 of Truth and Words is not true.
As a matter of fact,
(3) ‘The sentence that occurs on lines 45–6 of §2.5 of Truth and Words
is not true’ = The sentence that occurs on lines 45–6 of §2.5 of
Truth and Words.
But (2) and (3) commit us to
(4) The sentence that occurs on lines 45–6 of §2.5 of Truth and Words
is true if and only if the sentence that occurs on lines 45–6 of §2.5
of Truth and Words is not true.
   ‒  65

This is a contradiction.²⁷ The problem apparently lies with (2). Yet it

seems that according to the naive theory we are epistemically justified in
accepting (2).
One might try to defend the naive theory by amending it so that it
prevents us from inferring (2). But there is apparently nothing about the
ordinary use of the English word ‘true’ that rules out our acceptance of
(2). We find the reasoning of the sort just sketched paradoxical precisely
because we do not see what is wrong with any of the steps. To defend
the naive theory, one would have to develop an account of the meaning
of the English word ‘true’ that rules out our acceptance of sentences
such as (2). Theories of this kind are vulnerable to the charge that
they do not capture the meaning of the English word ‘true’.²⁸ Partisans
will naturally try to defend their favourite theories of the meaning of
‘true’. Given our pragmatic goals in this book, however, we must do
our best to avoid such partisan debates.²⁹ We therefore need a different
approach.

2.6. Surrogate T-Sentences and Explication

Let us say that a surrogate T-sentence, or an ST-sentence, for short, is one
that results when we write sentences of our own regimented language L
uniformly in the blanks of the pattern
(ST) ‘ ’ is true-in-L ↔ .

²⁷ This version of the liar paradox, due in essentials to J. Lukasiewicz, is based on Alfred Tarski’s
presentation in Tarski 1936: §7.
²⁸ For a similar observation about attempts to dissolve the paradox of scepticism about empirical
knowledge by appealing to a theory of the meaning of ‘know’, see Schiffer 1996. For Tarski’s argument
that the meaningfulness of unrestricted applications of the word ‘true’ in English leads to the liar
paradox, see Tarski 1936: §1. See also Heck 2004. Some authors who are more optimistic about the
prospects for solving the liar paradox by developing a theory of the meaning of the English word
‘true’ nevertheless also reject the naive theory. For a sophisticated recent example of such a view, see
Glanzberg 2005.
²⁹ Although my argument here is officially agnostic about the existence of a correct and consistent
theory of the meaning of ‘true’, I agree with Chihara 1979 that the prospects for preventing the
derivation of the liar paradox by ‘analysing’ the meaning of the English word ‘true’ are dim. Even
Saul Kripke, whose theory of truth is still one of the main contenders today, writes that ‘‘I do not
regard any proposal . . . as definitive in the sense that it gives the interpretation of the ordinary use of
‘true’, or the solution to the semantic paradoxes’’ (Kripke 1975: 699). For a sampling of other views
of the roots of the liar paradox, including less pessimistic ones, see the classic essays reprinted in
Martin 1984.
66   ‒ 

(The semi-ordinary word ‘true-in-L’ that occurs in ST-sentences should

not be identified with the ordinary English word ‘true’ that occurs in
T-sentences. ST-sentences are so named because they are at best surrogates
for T-sentences.) I propose that we replace the question, ‘‘Is it epistemically
reasonable for us to accept all the non-paradoxical T-sentences, and no
paradoxical ones?’’ with the more tractable question, ‘‘Is there a predicate
‘true-in-L’ from whose definition we can derive all the non-paradoxical
ST-sentences we need, and no paradoxical ones, and, if so, can we use this
predicate in place of the English word ‘true’?’’
My answer to this question is an application of the pragmatic method-
ology of explication. To explicate a linguistic expression e that one finds
useful in some ways yet problematic in others is to decide to use, in place
of e, a different linguistic expression e that preserves and clarifies what one
takes to be useful about e, yet avoids what one takes to be the problems
with e.³⁰ When one regiments an ordinary English statement S by deciding
to use a partly artificial sentence S in its place, one thereby also typically
explicates some of the linguistic expressions that occur in S. When one
regiments
(1) Every ship sails in some sea or other.
as
(2) ∀y((y is a ship) → ∃x((x is a sea) ∧ (y sails in x))),
for instance, one thereby also chooses to use ‘∃’ and ‘∀’ in place of the
occurrences of ‘some’ and ‘every,’ respectively, in (1). Just as one’s goal in
regimenting a given natural language sentence is not to capture its meaning,
but to clarify and facilitate one’s inquiries, so one’s goal in explicating a
given linguistic expression is not to capture its meaning, but to fill those of
its functions that one finds useful. Hence just as we are entitled to regiment
(1), as used on a given occasion, by (2), if we judge (as Quine would say)
that the particular business that we were on that occasion trying to get
on with, with the help of (1) among other things, can be managed well
enough to suit us by using (2) instead of (1), so we are entitled to explicate

³⁰ This account of explication is essentially the same as Quine’s (see Quine 1960: §53), which is a
modification of Carnap’s (see Carnap 1956: 8). Unlike Carnap’s account of explication, Quine’s makes
no mention of concepts.
   ‒  67

‘some’ as ‘∃’, if we judge that ‘∃’ fills those functions of ‘some’ that we find
useful.
I assume that we desire to express logical generalizations on the sentences
of a given regimented language L that does not contain its own truth
predicate or any predicate coextensive with its own truth predicate. For
this purpose, the useful ST-sentences are those sentences which result from
substituting L-sentences for the blanks in (ST), where the L-sentences
themselves contain no occurrences of the predicate ‘true-in-L’ or any
predicate coextensive with it. In the same pragmatic sense in which we
may choose to explicate ‘some’ and ‘there is’ as ‘∃’, we may choose to
explicate ‘true’ by a predicate ‘true-in-L’ that is defined expressly so that
from its definition and the logical principles to which we are independently
committed we may derive any given ST-sentence of the kind just described
and no paradoxical ones.
One way to achieve this goal is to adopt (ST) as an axiom schema,
subject to the restriction that the L-sentences that we substitute for the
blanks in (ST) contain no occurrences of the predicate ‘true-in-L’ or any
predicate coextensive with it. But there is a better way to achieve this goal.
In the next three sections (§§2.7–2.9) I explain how we can use Tarski’s
methods to define ‘true-in-L’ for sentences of our own regimented language
without presupposing a substantive relation of sameness of meaning or
translation. I then argue (in §2.10) that there are reasons to prefer a Tarski-
style disquotational definition of ‘true-in-L’ to a schematic definition of
‘true-in-L’.

2.7. Tarski’s Convention T

Alfred Tarski developed a method we can use to explicate the English
word ‘true’ by defining a truth predicate ‘true-in-L’ that entails only
non-paradoxical ST sentences. His method presupposes that the object
language —the language L for which we define ‘true-in-L’—is distinct
from the metalanguage —the language ML in which we define ‘true-in-L’.
Truth predicates constructed by using Tarski’s method are part of ML, but
not of L, so the liar paradox cannot be formulated in L.³¹

³¹ Kripke 1975 outlines a method for defining a truth predicate that applies to some sentences that
contain it. Call this predicate ‘K-true-in-L’. The basic rule for it is that ‘‘we are entitled to assert (or
68   ‒ 

As we’ve seen, we need a non-paradoxical truth predicate that we can

use to generalize on sentences. It turns out that to satisfy this need it is
enough to define ‘true-in-L’ in a way that satisfies Tarski’s Convention T:
Convention T. A formally correct definition of the symbol ‘true-in-L’,
formulated in the metalanguage, will be called an adequate definition
of truth if and only if it implies all sentences which are obtained from
the expression ‘X is true-in-L if and only if S’ by substituting for
the symbol ‘X’ a structural-descriptive name of any sentence of the
object language and for the symbol ‘S’ the expression which forms the
translation of this sentence into the metalanguage.³²
Tarski does not say exactly what he means here by ‘formally correct’.³³ He
does emphasize, however, that
The question how a certain concept is to be defined is correctly formulated only
if a list is given of the terms by means of which the required definition is to be
constructed. If the definition is to fulfill its proper task, the sense of the terms in
this list must admit of no doubt. ( Tarski 1936: 152–3)

deny) of any sentence that it is K-true-in-L precisely under the circumstances when we can assert
(or deny) the sentence itself ’’ (Kripke 1975: 701, with ‘K-true-in-L’ in place of Kripke’s ‘true’). By
using Kripke’s method, we can define ‘K-true-in-L’ for a language L in L itself without paradox. To
apply the method, however, we must leave some predicates of L undefined for some of the objects
over which L’s variables range, and allow that not all sentences of L ‘‘make a statement’’ or ‘‘express
a proposition’’ (Kripke 1975: 699). An apparent advantage of Kripke’s method of defining truth is
that it mirrors some uses of the English word ‘true’ that we cannot mirror if we use Tarski’s method
of defining truth (Kripke 1975: 694–8). If our goal is not to be faithful to our intuitions about the
English word ‘true’, but to define a clear and consistent predicate we can use to generalize on sentences
of our own regimented language, however, then I believe that Tarski’s method of defining truth is
preferable to Kripke’s. Tarski’s method requires that all predicates of L be defined for all the objects
over which L’s variables range; if we use Tarski’s method, we have no need for a distinction between
meaningful sentences of L and sentences of L that ‘‘make a statement’’ or ‘‘express a proposition’’.
Hence Tarski-style definitions of ‘true-in-L’ conform to the lucid disquotational paradigm for defining
truth better than Kripke’s definitions of ‘K-true-in-L’ do. I believe this advantage of Tarski’s approach
outweighs the limitations of it that Kripke identifies, given the goals I have assumed in this book.
Unfortunately, however, I do not have the space to justify this claim here.
³² This version of Convention T is adapted from Tarski 1936: 187–8. I leave out part (b) of Tarski’s
version of Convention T, and use ‘true-in-L’ in place of his term ‘Tr’, which denotes the set of
sentences that are true-in-L. More important, I replace Tarski’s ‘if ’ with ‘if and only if ’. Against this,
Patterson 2002 argues that Tarski’s Convention T should not be taken to state a necessary condition on
defining ‘true-in-L’. I see this as a dispute about how best to explicate ‘true’, and hence as a question
of policy, not fact. I return to this point below.
³³ We should not assume that Tarski means by ‘formally correct’ what others mean by it. Two
standard requirements for the formal correctness of a definition are eliminability and non-creativity (see
Suppes 1957: 154). But Tarski-style implicit definitions of ‘true-in-L’ are not eliminable from all
contexts, as the indispensability argument shows. And, as Michael Glanzberg pointed out to me, they
are not non-creative (in the sense defined by Suppes 1957: 154), either.
   ‒  69

This condition is not formal, but (as one might say) pragmatic: if our goal
is to define a truth predicate ‘true-in-L’ that we can use to formulate
logical generalizations, then we will take a definition of ‘true-in-L’ to
satisfy Convention T only if we understand and have no doubts about
the clarity of the terms in which it is expressed. That Tarski’s method
of defining ‘true-in-L’ satisfies this vague yet intuitive requirement will
become clear when we examine a particular application of the method
below. Before we get into these details, however, let us now briefly address
two other questions that are relevant to applying Convention T: ‘‘What
is a structural-descriptive name of a sentence of the object language?’’ and
‘‘When does a sentence of the metalanguage translate a sentence of the
object language?’’ I shall address these in order.
A structural-descriptive name of a sentence of the object language is one
that names the sentence by describing its syntactical structure. To form
a structural-descriptive name of a sentence, we must use names in the
metalanguage of the letters of the object language, and describe sentences
of the object language as strings of letters and spaces. By this method,
the object language sentence ‘Snow is white’ might be named as follows:
‘‘The L-sentence consisting of ess, en, oh, doubleyu, space, eye, ess, space,
doubleyu, aiche, eye, tee, ee, in that order.’’ Such structural-descriptive
names may be conveniently abbreviated by treating letters and spaces as
names of themselves and a hyphen as a sign of concatenation, as follows:
S-n-o-w i-s w-h-i-t-e. One can also form structural-descriptive names of
object-language sentences by means of Gödel numbering. It is much easier
to form a quotation-mark name of a given sentence, however—all one has
to do is write single quotation marks on the left and right of it. This
method is illustrated by the left-hand sides of ST-sentences. To save space,
I shall often use quotation-mark names instead of structural-descriptive
names, and when I use the latter, especially in later chapters, I shall usually
abbreviate them as strings of letters, spaces, and hyphens.
It is not as easy to interpret the term ‘translation’ as it occurs in
Convention T. We may assume that the meta-language contains the object
language, and it is therefore tempting to explicate ‘translation’ as syntactical
identity, using ST-sentences as a model. The idea would be that each
sentence translates itself, and that no other notion of translation is needed
for Convention T. But Convention T does not require that the sentence
that replaces ‘S’ be identical to the sentence named by X. Hence we cannot
70   ‒ 

avoid saying something about how to explicate the term ‘translation,’ as it

occurs in Convention T.
The term is often taken to stand for a fine-grained relation of equivalence
in meaning between different natural language expressions, a kind of equi-
valence that cannot be completely characterized in syntactical terms, and
to which our own explanations of the vocabulary of our own regimented
languages—what Quine describes as the ‘‘simple mechanical operations
whereby any sentence in the logical notation can be directly expanded,
if not into quite ordinary language, at least into semi-ordinary language’’
(Quine 1960: 159)—are irrelevant. But this understanding of the term
‘translation’, as it occurs in Convention T, overlooks two crucial points.
First, to apply Convention T, we do not need a fully general account
of translation; an account of the ‘‘translation’’ relation between sentences
of L and sentences of the metalanguage in which we define ‘truth-in-L’
is enough. Second, when we define ‘truth-in-L’ in terms of satisfac-
tion, we thereby codify our semi-ordinary-language-explanations of the
vocabulary of L —explanations whose indispensability I stressed in §1.7.
These two points suggest that we may take our Tarski-style definitions
of ‘true-in-L’ in terms of satisfaction to define syntactically the relation
of ‘translation’ relevant to Convention T.³⁴ To explain this suggestion, I
will explain how we may each construct, for our own regimented lan-
guage L, a Tarski-style disquotational definition of ‘true-in-L’ in terms of
satisfaction.

2.8. ‘True-in-L’ Defined in Terms of Satisfaction

By design our regimented languages have a very simple structure: they
comprise all and only sentences constructed in the usual way from a finite
list of basic predicates, such as ‘mortal’ and ‘loves’, variables ‘x1 ’, ‘x2 ’, . . .,

³⁴ Saul Kripke suggests that if we seek an implicit Tarski-style definition of ‘true-in-L’ that satisfies
Convention T of the sort Davidson recommends, we can ‘‘let the truth theory itself determine
the translation of the object language into the metalanguage’’ (Kripke 1976: 338). Kripke notes,
however, that ‘‘the truth theory doesn’t really determine a translation of the object language into the
metalanguage, since in general for a given φ, there is more than one formula φ  , satisfying all the given
criteria, such that T(φ) ≡ φ  is provable’’ (Kripke 1976: 338 n. 14). The approach I shall recommend
is based in the observation that we may use a truth theory to codify our decisions about how to explain
the artificial notations of the regimented object language sentences in semi-ordinary language.
   ‒  71

truth-functional symbols ‘∼’ and ‘∨’, and a quantifier symbol ‘∀’. To

define a predicate ‘true-in-L’ for one’s own regimented language L by
using Tarski’s method, one needs to be able to name each element of L’s
vocabulary. As I noted earlier, to form names of predicates and constants,
one may simply put single quotation marks around them. To form names
of variables, let us stipulate that the ith variable in the sequence ‘x1 ’,
‘x2 ’, . . ., is var(i). One also needs sequences of objects in the intended
domain of discourse for L. (These need not be infinite sequences, but I will
assume they are.) To form names of the objects in the intended domain of
discourse, one can stipulate that the ith object in an infinite sequence s of
those objects is si .
By using these resources, I can specify, for instance, that
(r) For every sequence s, s satisfies my predicate one place ‘robin’
followed by var(i) if and only if si is a robin.
This fits the disquotational pattern:
(Sat1) For every sequence s, s satisfies my one-place predicate ‘ ’
followed by var(i) if and only if si is .
Similarly, for the two place predicate ‘is smaller than’, for instance, I can
specify that
(s) For every sequence s, s satisfies my two-place predicate ‘is smaller
than’ followed by var(i) and var( j ) if and only if si is smaller than sj .
If one abbreviates ‘si is smaller than sj ’ as ‘is smaller than si sj ’, (s) fits the
disquotational pattern
(Sat2) For every sequence s, s satisfies my two-place predicate ‘ ’
followed by var(i) and var( j ) if and only if si sj .
In general, for any n-place predicate one can specify satisfaction conditions
that fit corresponding disquotational patterns for n-place predicates.
Suppose I have specified the satisfaction conditions for each of the n
basic predicates of L in the way just described. Then I can complete my
definition of satisfaction for L by adding the following clauses:
(Neg) For all sequences s and sentences S: s satisfies
∼S (the negation
of S ) if and only if s does not satisfy S.
72   ‒ 

(Disj) For all sequences s and sentences S and S : s satisfies
S ∨ S 

(the disjunction of S with S ) if and only if either s satisfies S or s
satisfies S .
(All) For all sequences s, sentences S, and numbers i: s satisfies
∀xi S
(the universal quantification of S with respect to var(i)) if and
only if every sequence s that differs from s in at most the ith place
satisfies S.
Clauses (Neg) and (Disj) are straightforward. The key to understanding
clause (All) is to see that if s is any sequence of objects from the domain of
discourse D for the universal quantifier of the object language, then the set
of all and only the ith things of any sequence s that differs from s in at most the ith
place is identical with D.³⁵
Together with the disquotational satisfaction clauses for the simple pre-
dicates of my language L, clauses (Neg), (Disj), and (All) define satisfaction
for all sentences of L.³⁶ Using this definition of satisfaction, I can then
define ‘true-in-L’ as follows:
(Tr) A sentence S of L is true-in-L if and only if S is satisfied by all
sequences.³⁷

2.9. How (Tr) Satisfies Convention T and Enables us

to Derive ST-Sentences
I propose that we take (Tr) to settle the translation-relation that is relevant
to Convention T. When translation is understood in this way, (Tr) is
guaranteed to satisfy Convention T. Consider, for instance, the following
derivation of a statement of conditions under which the sentence ‘∀x1 ((x1
is mortal) ∨ ∼(x1 is mortal))’ of L is true-in-L:

³⁵ One need not think of D as a set. The question of whether D is a set is important if one wishes to
make sense of the idea that sentences of the object language can generalize over absolutely everything.
See Dummett 1981, Cartwright 1994, Williamson 2003, and Rayo and Uzquiano 2006. As I noted in
Chapter 1, n. 15, I have sympathies with both sides of the debate about whether one can generalize
over absolutely everything, and will not try to adjudicate it in this book.
³⁶ This sketch of Tarski’s method of defining satisfaction is modelled on Quine 1986, chapter 3.
³⁷ The definition of satisfaction is inductive, not explicit. It can be converted into an explicit
definition if there is enough set theory in the metalanguage. For details, see Quine 1986: chapter 3. For
present purposes, the inductive definition is sufficient.
   ‒  73

(1) ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is true-in-L if and only if

‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is satisfied by every sequence.
[By (Tr)]
(2) ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is satisfied by every sequence
if and only if for every sequence s, every sequence s that differs from
s in at most the 1st place satisfies ‘(x1 is mortal) ∨ ∼(x1 is mortal)’.
[By (All)]
(3) For every sequence s, s satisfies ‘(x1 is mortal) ∨ ∼(x1 is mortal)’ if
and only if s satisfies ‘(x1 is mortal)’ or s satisfies ‘∼(x1 is mortal)’.
[By (Disj)]
(4) For every sequence s, s satisfies ‘∼(x1 is mortal)’ if and only if s does
not satisfy ‘(x1 is mortal)’. [By (Neg)]
(5) For every sequence s, s satisfies ‘x1 is mortal’ if and only if s1 is
mortal. [By the satisfaction clause for ‘mortal’]
(6) For every sequence s, s satisfies ‘(x1 is mortal) ∨ ∼(x1 is mortal)’ if
and only if s1 is mortal or s1 is not mortal. [By (3)–(5)]
(7) For every sequence s, every sequence s that differs from s in at most
the 1st place satisfies ‘(x1 is mortal) ∨ ∼(x1 is mortal)’ if and only if
everything (in D) is mortal or not mortal. [From (6) and the fact that
the set of all and only the 1st objects of any sequence s that differs
from any sequence s in at most the 1st place is identical with D.]
(8) ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is true-in-L if and only if
everything (in D) is mortal or not mortal. [From (2) and (7)]

Note that the sentence on the right-hand side of (8), namely, ‘everything
(in D) is mortal or not mortal,’ is not identical to the sentence on the left-
hand side of (2), namely, ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’.³⁸ Hence
if ‘translation’ as it occurs in Convention T is restricted to the relation of
syntactical identity, (Tr) does not satisfy Convention T. But if I take (Tr) to
settle a translation of sentences of L into the metalanguage in which I state
(Tr), then (Tr) settles, for instance, that ‘everything (in D) is mortal or not
mortal’ translates ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’. Thus viewed, (8) is
of the form X is true-in-L if and only if S, where X stands in for ‘∀x1 ((x1 is
mortal) ∨ ∼(x1 is mortal))’ and S stands in for ‘everything (in D) is mortal

³⁸ As Quine notes, ‘‘What we usually come out with under Tarski’s truth definition is not literally
the sentence to whose quotation the truth predicate had been attached, but another sentence that is
equivalent to it under the logical laws of quantification and identity’’ (Quine 1976a: 313).
74   ‒ 

or not mortal’, which (by stipulation) translates X. More generally, if we

may each take our own Tarski-style definitions of true-in-L to settle a
translation of sentences of L by sentences of our metalanguage, then for
each sentence s of L, we may derive a biconditional of the form X is
true-in-L if and only if S, where the symbol ‘X’ is replaced by a name of s,
and the symbol ‘S’ is replaced by a sentence s that (by stipulation) translates
s. If we understand ‘translation’ in the way that I propose, therefore, we can
see that our own applications of (Tr) to sentences of our own regimented
languages satisfy Convention T.
We need now to convince ourselves that from a definition of ‘true-in-L’
that satisfies Convention T, we may infer ST-sentences, such as
(9) ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is true-in-L if and only if
∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal)).
To make this final step, we need only make use of our assumption that
our metalanguage contains the sentences of the object language. For the
translation relation settled by (Tr) is symmetrical. Hence if our metalan-
guage contains the sentences of the object language, we can use sentences of
the object language interchangeably with those sentences of the metalan-
guage that translate them in the sense of ‘translate’ stipulated above. The
disquotational definitions of satisfaction for our basic predicates, together
with the satisfaction clauses for the logical words, amount to our decision
to expand ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ into the semi-ordinary
language sentence ‘Everything is either mortal or not mortal.’ Given this
decision and the symmetry of the translation relation, from (8) we may
infer (9). More generally, for any given sentence S of L, we can derive all
the ST-sentences—those sentences of the form ‘X is true-in-L if and only
if S’, where ‘X’ is replaced by a quotation-mark name of S. In this way,
(Tr) enables us to derive ST-sentences, and thereby to state logical laws for
sentences of L.

2.10. Schematic Definitions of ‘True-in-L’ Rejected

We are now in a position to evaluate the suggestion, mentioned briefly
above, that we take (ST) as an axiom schema, understood so that the L-
sentences that one is permitted to substitute for the blanks in (ST) contain
   ‒  75

no occurrences of the predicate ‘true-in-L’ or any predicate coextensive

with it. When (ST) is taken as an axiom schema, understood in this way,
I shall call it the ST-schema. From the ST-schema we can derive only
non-paradoxical ST-sentences and thereby satisfy Convention T, where
translation is understood as syntactical type-identity. Hence to take (ST) as
an axiom schema is to define ‘true-in-L’ schematically. The ST-schema is
more economical than a definition of ‘true-in-L’ in terms of satisfaction.
Should we then prefer the ST-schema, viewed as a definition of ‘true-in-L’,
to (Tr)?
On the contrary, for three main reasons, I think we should prefer
(Tr) to the ST-schema, viewed as a definition of ‘true-in-L’. First, the
ST-schema does not enable us to explain the contributions of the logical
constants to L-sentences in which they occur inside the scope of an
objectual quantifier. To see why, note first that from the ST-schema we
can deduce the contributions of the logical connectives ‘∼’ and ‘∨’ to
any L-sentence in which they occur outside the scope of an objectual
quantifier.
For instance, suppose that ‘snow is white’ is a sentence in L. Then we
can argue as follows:
(1) ‘∼(snow is white)’ is true-in-L if and only if ∼(snow is white).
(2) ‘snow is white’ is true-in-L if and only if snow is white.
(3) ‘∼(snow is white)’ is true-in-L if and only if ∼(‘snow is white’ is
true-in-L).
Premises (1) and (2) can be obtained by substituting L-sentences for the
blanks in the ST-axiom-schema; the conclusion, (3), results from (1), (2),
and the standard laws of truth-functional interchange.
In the same way, we can deduce the contribution of ‘∨’ to any sentence
in L in which ‘∨’ occurs outside the scope of a quantifier. For instance, if
that ‘grass is green’ is also a sentence in L, we can reason as follows:
(4) ‘snow is white ∨ grass is green’ is true-in-L if and only if snow is
white ∨ grass is green.
(5) ‘snow is white’ is true-in-L if and only if snow is white.
(6) ‘grass is green’ is true-in-L if and only if grass is green.
(7) ‘snow is white ∨ grass is green’ is true-in-L if and only if ‘snow is
white’ is true-in-L ∨ ‘grass is green’ is true-in-L.
76   ‒ 

The premises (4), (5), and (6) can be obtained by substituting L-sentences
for the blanks in the ST-axiom-schema; the conclusion, (7), results from
(4), (5), (6), and the standard laws of truth-functional interchange.³⁹
We can use the ST-schema in this way to derive the contributions of
‘∼’ and ‘∨’ to the truth conditions of sentences in which ‘∼’ and ‘∨’ occur
outside the scope of a quantifier. But the ST-axiom-schema sheds no light
on the contributions of ‘∼’ and ‘∨’ to the truth or falsity of sentences
in which ‘∼’ and ‘∨’ occur within the scope of an objectual quantifier.
Consider again, for instance, the logical truth
(L) ∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))
We can use the ST-axiom-schema to derive
(M) ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ is true-in-L if and only if
∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal)).
By the usual laws of interchange, we can replace any of the components
in the matrix of the sentence ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ with
their truth-functional equivalents. But such replacements occur within the
scope of the universal objectual quantifier ‘∀x1 ’, and do not by themselves
elucidate the contribution of the connectives ‘∼’ and ‘∨’ to the truth of
the whole sentence. Moreover, we cannot use the ST-schema to explain
why we accept the laws of interchange. If the universal quantifier ‘∀x1 ’
were substitutional, then the sentence would be equivalent to ‘n ((n is
mortal) ∨ ∼(n is mortal))’, which is an abbreviation of the conjunction
of all the results of replacing the occurrences of ‘n’ in the matrix with
names in the language; given the ST-schema, this conjunction is truth-
functionally equivalent to the result of replacing all occurrences in it of
sentences of the form ‘n is mortal’, where ‘n’ is replaced by a name,
by ‘ ‘n is mortal’ is true’. Hence we could use the ST-schema in the
way illustrated above to derive an explanation of the contribution of
‘∼’ and ‘∨’ to the truth value of the whole sentence ‘n ((n is mortal)
∨ ∼(n is mortal))’. If the universal quantifier ‘∀x1 ’ is objectual, however,
as we suppose, we cannot take for granted that there are enough names
to yield a conjunction that has the same truth value as the universal
quantification ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’. As a result, the

³⁹ For a similar derivation, see Field 1994: 258–9.

   ‒  77

contribution of the connectives ‘∼’ and ‘∨’ to the truth value of the
whole sentence ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ cannot be derived
solely from the ST-axiom-schema. Similar reasoning shows that neither
the contribution of the open sentence ‘x is mortal’ nor the contribution
of the corresponding predicate ‘mortal’ to the truth value of the whole
sentence ‘∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))’ can be derived solely from
the ST-schema.
A second reason for preferring (Tr) to the ST-schema, viewed as a
definition of ‘true-in-L’, is that to regiment our sentences is to decide
how to replace ordinary sentences with regimented sentences. If there are
infinitely many sentences in the regimented language, our decisions cannot
consist in an infinite number of separate decisions, one for each sentence.
The only clear and uncontroversial way to codify these decisions is to
adopt a Tarski-style disquotational definition of ‘truth-in-L’ in terms of
satisfaction.⁴⁰
Third, as I emphasized in §1.7, to clarify and facilitate our inquiries we
need to be able to introduce and explain new regimented predicates for our
regimented language while leaving many of our previous explanations of
basic vocabulary of L unchanged. As we have seen, from the ST-schema
we cannot explain the contributions of our basic predicates to the truth
values of all sentences in which they occur, and hence we cannot use the
ST-schema to introduce and explain new regimented predicates for our
regimented language.⁴¹
These reasons for preferring (Tr) to the ST-schema are not decisive.
One’s choice between (Tr) or the ST-schema must be made on pragmatic
grounds, and can therefore always be resisted by a person who does not
share one’s goals. I will nevertheless assume from here on that despite its
greater complexity, (Tr) is preferable to the ST-schema for the reasons just
sketched.

⁴⁰ Donald Davidson’s use of Tarski is incompatible with the one I recommend, since Davidson aims
to clarify meaning by assuming a primitive notion of truth. But Davidson also emphasizes that the
recursive structure of a Tarski-style theory of truth enables us to encode infinitary consequences.
⁴¹ This is a pragmatic argument based in our need to be able to add new regimented predicates to
an already existing regimented language. It makes no assumptions about how we actually learn new
vocabulary in a natural language, or about whether it is possible for a person to understand a sentence
containing a given predicate without understanding other sentences in which that predicate occurs.
The pragmatic argument does not support or presuppose, for instance, that an adequate theory of
meaning for a natural language ‘‘represents the structure of the ability to speak [the] language’’ (Lepore
and Ludwig 2005: 124).
78   ‒ 

2.11. Adopting the Tarski–Quine Thesis

Let me summarize the steps I have taken so far. I began by asking ‘‘Why
do we need a truth predicate?’’ and ‘‘What sort of truth predicate do we
need?’’ and took for granted that our answer to these questions should be
shaped by our desire to clarify and facilitate our rational inquiries. This
desire motivates us to regiment our words so that we can formulate logical
laws in the notation of first-order logic.
I reviewed and modified (§§1.2–1.7) some well-known methods by
which we may each regiment our own ambiguous and vague words,
quantifications, definite descriptions and proper names, and pronouns and
demonstratives, and emphasized (§1.8) that to regiment one’s sentences
by using the methods I describe, one must use ordinary (or, at least,
semi-ordinary) language. I then argued (§§2.1–2.9) that if we use the
methods of regimentation that I sketched, and we wish to formulate logical
generalizations on sentences of the resulting regimented language L, we
will need to define a truth predicate for sentences of L that incorporates
and codifies our ordinary and semi-ordinary language explanations of the
vocabulary of L. My first step was to emphasize (§§2.1–2.4) that our desire
to clarify and facilitate our rational inquiries motivates us to express logical
generalizations on sentences of our regimented language L, but that we can
express such generalizations only if it is epistemically reasonable for us to
affirm any given T-sentence. I argued (§2.5) we cannot explain why it is
epistemically reasonable for us to affirm any given T-sentence by appealing
to the naive theory that our acceptance of such sentences is justified by
our grasp of the meaning of the English word ‘true’. I then sketched
(§2.6) a better strategy: the strategy of defining ‘true-in-L’ in a way that
enables us to derive non-paradoxical surrogate T-sentences (ST-sentences)
that mirror the useful functions of the English word ‘true.’ I proposed
(§2.7) that we adopt Tarski’s Convention T, where we stipulate that the
translation relation is settled by a recursive definition of ‘true-in-L’ in terms
of satisfaction. I explained (§2.8) how to construct such a definition, and
showed (§2.9) that from it we may derive, for every sentence s of L, a
biconditional of the form X is true-in-L if and only if S, where the symbol
‘X’ is replaced by a name of s, and the symbol ‘S’ is replaced by a sentence
s that (by stipulation) translates s. I also showed (§2.9) that we are then
   ‒  79

licensed to derive ST-sentences by replacing s by s. Finally, I argued (§2.10)

that even though we could derive the purely disquotational ST-sentences
directly by adopting (ST) as an axiom schema, we should nevertheless
prefer (Tr), from which we can derive systematic, semi-ordinary language
explanations of the basic vocabulary of L.
This reasoning shows how our desire to express logical generalizations on
sentences of our own regimented language L may motivate us to explicate
the ordinary English word ‘true’ by using a Tarski-style disquotational
definition of ‘true-in-L’ in its place, and thereby to adopt the Tarski–Quine
thesis that there is no more to truth than what is captured by a Tarski-style
disquotational truth predicate defined for one’s own sentences.⁴²

2.12. Two Objections

This completes my account of what I regard as our strongest pragmatic
motivation for embracing the Tarski–Quine thesis. I will end this chapter
by considering two objections to the thesis. The first objection begins
with the observation that we typically do not apply our ordinary English
word ‘true’ to sentences. If Alice says, ‘‘He admires Tarski,’’ pointing at
a contextually salient man, and Zed says, ‘‘That’s true,’’ Zed does not
apply his predicate ‘true’ to the sentence ‘He admires Tarski’, a different
utterance of which, on a different occasion, might have prompted Zed to
say, ‘‘That’s not true.’’ Instead, it is natural to say, Zed applies his predicate
‘true’ to Alice’s utterance of ‘He admires Tarski’, as accompanied by her
pointing gesture, and perhaps also to what Alice said —the proposition she
expressed—when she uttered ‘He admires Tarski.’⁴³ We therefore cannot
in general assume that a given declarative sentence of English is true or false
independent of any particular occasion on which it is used.⁴⁴ In contrast,
our own Tarski-style disquotational definitions of ‘true-in-L’ are supposed
to apply directly to the declarative sentences of our regimented language L.
Such predicates are not directly defined for utterances or propositions, and

⁴² To adopt the Tarski–Quine thesis for the reasons just given is to accept that in this case, anyway,
explication is elimination. See Quine 1960: 260.
⁴³ Donald Davidson prefers the first option, whereas Paul Horwich and Scott Soames prefer the
second. See Davidson 1984: essay 2, Horwich 1998a and Soames 1999.
⁴⁴ For an argument to this effect, see Soames 1999: chapter 1.
80   ‒ 

without further stipulations cannot be meaningfully applied to them. It

may therefore seem that ‘true in L’ is a poor substitute for the English word
‘true,’ and that, contrary to the thesis, we should not use a disquotational
definition of ‘true in L’ in place of our English word ‘true’.
The convictions behind this objection run deep, and cannot be
adequately articulated, much less evaluated, without an investigation of
the sort I will begin in the next chapter. But the objection itself is based on
a misunderstanding of the thesis, which does not aim to mirror all our uses
of the ordinary English word ‘true’ or to analyse a concept we express when
we use that word.⁴⁵ According to the pragmatic reasoning I have sketched,
we need a truth predicate to express logical generalizations formulated in a
regimented language devoid of sentences like ‘He admires Tarski’ to which
we cannot apply a truth predicate independent of particular occasions of
utterance. To express logical generalizations formulated in such a regimen-
ted language L, we need to commit ourselves to affirming non-paradoxical
ST-sentences. A Tarski-style disquotational predicate ‘true-in-L’enables us
to commit ourselves to affirming all the non-paradoxical ST-sentences that
we need to commit ourselves to affirming if we are to generalize on our
own regimented sentences. This leaves it open, for now, how we are to
understand the relationship between ‘true-in-L’ and utterances of ordinary
language sentences like ‘He admires Tarski’, or the propositions, if any,
that such utterances express. One possibility is that our utterance of an
ordinary language sentence s on a particular occasion is true-in-L if and
only if (a) there is some sentence s of L that we decide to use in place of
our utterance of s on that occasion, and (b) s is true-in-L. To challenge,
not just dismiss, the Tarski–Quine thesis, one would have to challenge this
and other possible ways of relating the Tarski–Quine thesis to ordinary
language. I shall examine and ultimately reject one such challenge—the
most powerful one I can formulate—in Chapters 3 and 4.
A second, related objection is that even if all of our declarative sentences
are properly regimented, ‘true-in-L’ must be understood to apply primarily
to utterances or propositions, not sentence types, because, as Tarski himself
pointed out, ‘‘the same expression which is a true sentence of one language
can be false or meaningless in another’’ (Tarski 1944: 342). That there are

⁴⁵ In this respect, the thesis is different from what David (1994) calls disquotationalism, which is
supposed to capture the meaning and use of the ordinary English word ‘true’.
   ‒  81

possible languages in which a given meaningful sentence of our regimented

language is false or meaningless may seem to be enough to show that
‘true-in-L’ does not apply directly to L-sentences themselves, considered
as strings of letters and spaces, but only to our utterances of L-sentences,
which express just those meanings or propositions that we decided to use
the L-sentences to express.⁴⁶
This objection is based in misunderstanding the role of sentence types
in a Tarski-style disquotational definition of ‘true-in-L’ for one’s own
regimented language L. As I explained briefly in the Introduction, the
question whether another speaker is actually using one or more of the
sentence types of our own regimented language cannot be answered by
formal considerations alone.⁴⁷ It does not follow, however, as the objection
assumes, that one cannot rely on sameness and difference of spelling to
identify and distinguish between the sentence types and ST-sentences of
one’s own regimented language. To draw this mistaken inference is to
overlook the fact that one can disquotationally define a predicate ‘true-
in-L’ for sentences of one’s own regimented language L —regimented
sentences that one can actually use. One’s disquotational definitions of
‘true-in-L’ apply directly to one’s own sentence types under the conditions
one specifies by using those very sentences. The formal (recursive) structure
of one’s disquotational definition of ‘true-in-L’ licenses one to derive all of
the ST-sentences one needs to express logical generalizations on one’s own
regimented sentences, considered as types. Contrary to the objection, any
assumptions about ‘meaning’ that one needs to license the formal derivations
of one’s ST-sentences are built into the recursive, disquotational definitions
of satisfaction. Hence at any time t, one’s disquotational definition of
‘true-in-L’ presupposes that one can identify and distinguish intrasubjectively
between the sentence types of one’s own regimented language L, as one
uses it at t, by their spelling alone.⁴⁸

⁴⁶ For related objections, see Putnam 1983, and David 1994: chapter 5, §6, §9.
⁴⁷ The same reasoning is responsible for what Marian David calls the problem of foreign
intruders—the problem that our definition of ‘true-in-L’ applies to sentence tokens solely on
the basis of their orthographic type, whether or not they mean the same. See David 1994: 159–60. I
shall examine the assumption and sketch an alternative to it in Chapter 4.
⁴⁸ Quine 1961a: 135 makes a similar point, though only in passing and in much less detail.
 3
The Intersubjectivity Constraint

3.1. A Preliminary Formulation

of the Intersubjectivity Constraint
We saw in Chapters 1 and 2 that our desire to clarify and facilitate our
inquiries motivates us to regiment our sentences and define a Tarski-style
disquotational predicate ‘true-in-L’ that we can use to formulate logical
generalizations on sentences of our regimented language L. If we assume
on pragmatic grounds that there is no more to truth than what we need to
clarify and facilitate our inquiries, we may each be tempted to endorse the
Tarski–Quine thesis that there is no more to truth than what is captured
by such a Tarski-style disquotational predicate defined for sentences of our
own regimented language.
As I explained in the introduction, however, if we are motivated
to adopt a Tarski-style truth predicate by our desire to express logical
generalizations, as proponents of the Tarski–Quine thesis claim, then it
should occur to us to ask why we want to express logical generalizations
in the first place. A complete answer to this question should mention our
desire to collaborate fruitfully with other inquirers—a desire that motivates
us to regiment our words and sentences so that we can state logical
generalizations that clarify our agreements and disagreements. For this
purpose, we need to be able to generalize not only over our own sentences
as we now use them, as the Tarski–Quine thesis suggests, but also over
other speakers’ sentences. As Gottlob Frege emphasized, we regard logical
generalizations as arbiters of collaborative inquiry.¹ But we cannot regard

¹ Against the idea that words stand for private psychological items, Frege writes that ‘‘if we could
not grasp anything but what was within our own selves, then a conflict of opinions based on a mutual
understanding would be impossible, because a common ground would be lacking. . . . There would be
no logic to be appointed arbiter in the conflict of opinions.’’ (Frege 1964: xix).
    83

logical generalizations as arbiters of collaborative inquiry unless we believe

it is epistemically reasonable for us to apply logical generalizations to other
speakers’ sentences. Our relationship to other speakers’ sentences is not in
this respect different from our relationship to sentences we used in the
past. For the same reasons, then, we cannot regard logical generalizations
as arbiters of inquiry unless we believe it is epistemically reasonable for us
to apply logical generalizations to our own sentences as we used them in
the past. But it is at best unclear how one can use a truth predicate that is
only defined for one’s own sentences as one now uses them to express logical
generalizations that one also takes to apply to other speakers’ sentences and
to one’s own sentences as one used them in the past.²
I emphasize our desire to collaborate with other inquirers not because it is
essential to this criticism of the Tarski–Quine thesis, but because it makes
the criticism seem especially pressing. At its most general, the criticism
applies even if we do not desire to collaborate with other inquirers. To see
why, suppose that out of choice or necessity we desire to engage in solitary
inquiries only, and hence do not actually need to generalize over other
speakers’ sentences. We will nevertheless presuppose that our inquiries may
lead us to reject (disagree with) some sentences that we previously accepted.
If we wish to clarify and facilitate our inquiries, then, we will need a truth
predicate that we can use to express logical generalizations that apply not
only to our sentences as we now use them, but also to our sentences as we
used them in the past. But, again, it is unclear how one can use a truth
predicate that is only defined for one’s own sentences as one now uses them
to express logical generalizations that one also takes to apply to one’s own
sentences as one used them in the past.
In short, to regard logical generalizations as arbiters of inquiry, whether
collaborative or solitary, we need a truth predicate that we are licensed to
apply to other speakers’ sentences and to our own sentences as we used them
in the past. As we’ve seen, however, without supplementation, a Tarski-
style disquotational truth predicate is not defined for other speakers’ sen-
tences or for our own sentences as we used them in the past. The problem is

² This unclarity is just a symptom of a much deeper problem, if, as Hilary Putnam argues, the
Tarski–Quine thesis is compatible with psychologistic accounts of assertion that, if adopted, would
prevent us from making sense of agreement or disagreement between speakers, and hence prevent us
from regarding logical generalizations as arbiters of collaborative inquiry. See Putnam 1983. I evaluate
Putnam’s argument, as directed against Quine, in Ebbs 2002b.
84   

not conceptual, but pragmatic—it is not that a Tarski-style disquotational

truth predicate fails to capture the concept (if there is one) expressed by
the English word ‘true’, but that it fails to provide an adequate explication
of ‘true’ (in the sense of ‘explication’ defined in §2.6), because it does not
capture and clarify some central uses of ‘true’ on which we rely in our
inquiries. This pragmatic criticism of the Tarski–Quine thesis motivates
what I call the intersubjectivity constraint: a Tarski-style disquotational truth
predicate defined for one’s own sentences is satisfactory only if it is supplemented by
an account of why it is epistemically reasonable for one to use it to generalize over
other speakers’ sentences and one’s own sentences as one used them in the past.
My goal in this chapter is to present in more detail the pragmatic motiv-
ation for adopting the intersubjectivity constraint. I shall (first) describe
how we actually apply logical generalizations to other speakers’ sentences
when we take ourselves to share words with them, (second) reformulate
the intersubjectivity constraint in light of these descriptions of how we
apply logical generalizations, and (third) consider two objections to the
reformulated intersubjectivity constraint.

3.2. Practical Identifications of Words (PIWs)

To begin with, let us look at a few simple examples of how we unreflectively
identify words when we take ourselves to understand what a person says.
Suppose I am on a walk with Alice, a bird lands on a branch in front of us,
and Alice says, ‘‘That is a robin,’’ pointing to the bird on the branch. Then
without hesitation or deliberation, I will take her to say that the bird is a
robin, and, in so doing, I will take Alice’s word ‘robin’ to be the same as
my word ‘robin’—the word that I would use to express what I take her to
have said.³ The sense of ‘‘same word’’ in question here is bound up with
my commitment to use my word ‘robin’ to express what I take Alice to
have said: if I later discover that Alice did not say that the bird is a robin,
I will regard as false and try to revise my initial, immediate identification

³ This unhesitating response to Alice’s utterance illustrates a deep and widespread phenomenon
that others have noted before, though for different reasons. For instance, Jerry Fodor observes that
‘‘You can’t help hearing an utterance of a sentence (in a language you understand) as an utterance of
a sentence . . . You can’t hear speech as noise even if you would prefer to’’ and ‘‘It’s what’s said that one
can’t help hearing, not just what is uttered’’ (Fodor 1983: 55, cited in Borg 2004: 91).
    85

of her word ‘robin’. In short, when I take Alice to have said that the bird
is a robin, there is a sense of ‘‘same word’’—a sense that I shall investigate
and explicate more fully in Chapter 4—in which I accept
(1) The word ‘robin’ that Alice used when she said, ‘‘That is a robin,’’
is the same as the word ‘robin’ that I now use to express what I take
Alice to have said—namely, that the bird is a robin.
To accept (1) in the context described above is to make what I call a
practical identification of a word (or a PIW, for short). I call such identifications
practical because they are exercises of a learned yet non-deliberative ability
that is acquired over a period of years. An ability of this sort is more or less
refined according as the person whose ability it is acquires it in more or
less varied circumstances and is responsive to more or less subtle features of
those circumstances. I call an exercise of this ability a practical identification
of a word because we will revise it if we come to think it is wrong. We will
revise (1), for instance, if we come to think that it prevents us from being
able to state in our own words what Alice’s utterance of ‘That is a robin’
expressed on the occasion described above.
We typically rely on such practical identifications of words when we
state what we take ourselves to have said in the recent past. For instance,
suppose that a few moments after Alice spoke, I said, ‘‘Yes, that is a robin,’’
pointing to the bird on the branch. Then without hesitation or reflection,
I will take myself to have said that the bird is a robin, and, in so doing, I
will accept
(2) The word ‘robin’ that I used when I said, ‘‘Yes, that is a robin,’’ is
the same as the word ‘robin’ that I now use to express what I take
myself to have said—namely, that the bird is a robin.
This is what I call a practical identification of a word across time (or a PIW
across time, for short). It is not essential to a PIW across time that it be
made for one of my own previous utterances. If I reaffirm (1), for instance,
several minutes after Alice said, ‘‘That is a robin,’’ I thereby make a PIW
across time for the word ‘robin’ that Alice used when she said, ‘‘That is
a robin.’’ By contrast, when I first accept (1)—roughly simultaneously
with Alice’s utterance of ‘‘That is a robin’’—I thereby make what I call a
practical identification of a word at a given time (or a PIW at a given time, for
short).
86   

In some contexts we will take two or more PIWs to form a chain of

PIWs that commits us to making additional PIWs. For instance, suppose
that a few moments after Alice says, ‘‘That is a robin,’’ indicating the
contextually salient bird in the story above, Hermione says, ‘‘That is not a
robin,’’ indicating the same bird. Suppose also that I don’t hear Hermione
say this, but Alice does. Then without hesitation or reflection, Alice will
take Hermione to have said that the bird is not a robin, and, in so doing,
Alice will take Hermione’s word ‘robin’ to be the same as her word
‘robin’—the word that she would use to express what she takes Hermione
to have said. Finally, suppose I later learn from Alice that Hermione said,
‘‘That is not a robin,’’ while pointing to the same bird that Alice pointed
at in the first story above. In this context, if I continue to accept (1) and I
suppose that Alice accepts (2) for her own previous use of ‘robin’, then if I
have no reason to question Alice’s way of taking Hermione’s word ‘robin’,
I will accept
(3) The word ‘robin’ that Hermione used when she said, ‘‘That is
not a robin,’’ is the same as the word ‘robin’ that I now use to
express what I take Hermione to have said—namely, that the bird
is not a robin.
In the circumstances just described, however, I accept (3) because I accept
that there is an independently existing chain of PIWs that implies (3). Since
the PIWs in the chain are revisable, so is (3). For instance, if I come to
reject Alice’s PIW for the word ‘robin’ that Hermione used when she said,
‘‘That is not a robin,’’ and infer that there is no chain of PIWs from my
word ‘robin’ to Hermione’s, I will come to reject (3).
One might think that if a speaker has not regimented her words, one
should not identify any of her words (as used on particular occasions) with
words of one’s own regimented language. As I stressed in Chapters 1 and 2,
however, we introduce and explain the words of our regimented language
by using words of ordinary language. Just as we use ordinary language
to express what another speaker says on a particular occasion, so we can
use a regimented language, whose words are explained by particular uses
of ordinary language, to express what another speaker says on a particular
occasion. Hence we can and often do make a PIW that identifies a particular
use of an unregimented word w with a word w of our own regimented
language.
    87

If one makes a PIW of the sort just described for words w1 . . . wn

that occur in an utterance of a (regimented or unregimented) sentence
s on a given occasion O, thereby identifying those occurrences of words
w1 . . . wn with occurrences of words w1  . . . wn  , respectively, of one’s own
regimented language L, then one commits oneself to identifying s, as
used on O, with the sentence s of L (if there is one) that results from
combining w1  . . . wn  , in the order in which one takes them to occur in s,
according to the standard rules for constructing sentences of L from words
(predicates, connectives, and quantifiers) of L. This is what I shall call a
PIW-based identification of s with s .

3.3. Practical Judgements of Sameness of Satisfaction

(PJSSs)
If we accept Tarski-style disquotational definitions of satisfaction for our
own regimented words as we now use them, then our PIWs commit
us to applying these disquotational definitions of satisfaction to other
speakers’ words, and to our own words, as we used them in the past. Our
PIW-based identifications of sentences, in turn, commit us to applying our
disquotational definitions of truth-in-L to other speakers’ sentences. Let us
now examine these commitments in detail.
To begin with, consider how a PIW commits us to applying disquo-
tational definitions of satisfaction to other speakers’ words. Suppose that
I accept Tarski-style disquotational definitions of satisfaction for my own
regimented words. Strictly speaking, only sequences of objects, not objects
themselves, satisfy words of my regimented language L. Nevertheless, to
simplify my exposition I will use in place of (Sat) the pattern
(Sats) x satisfies my word ‘ ’ if and only if x is .⁴
Thus suppose that I accept the following application of (Sats) to my
predicate ‘robin’:
(4) x satisfies my word ‘robin’ if and only if x is a robin.

⁴ Read ‘x’ as bound by a universal objectual quantifier that ranges over the objects in the intended
domain of discourse for L.
88   

Now recall the situation described above, in which I commit myself

without deliberation or hesitation to (1) and thereby make a PIW for the
word ‘robin’ that Alice used when she said, ‘‘That is a robin’’:
(1) The word ‘robin’ that Alice used when she said, ‘‘That is a robin,’’
is the same as the word ‘robin’ that I now use to express what I take
Alice to have said—namely, that the bird is a robin.
Putting ‘Alice’s word ‘robin’ ’ in place of ‘the word ‘robin’ that Alice used
when she said, ‘‘That is a robin’’ ’, and ‘my word ‘robin’ ’ in place of ‘the
word ‘robin’ that I now use to express what I take Alice to have said,
namely, that the bird is a robin’, we can rewrite (1) as
(1 ) Alice’s word ‘robin’ is the same as my word ‘robin’.
Together with (4), my unhesitating acceptance of (1 ) will commit me to
(5) x satisfies Alice’s word ‘robin’ if and only if x is a robin.
By accepting (4) and (5), I commit myself to accepting their conjunc-
tion, namely,
(6) (x satisfies my word ‘robin’ if and only if x is a robin) and (x satisfies
Alice’s word ‘robin’ if and only if x is a robin)
In this way, my unhesitating commitment to (1 ) and my explicit commit-
ment to (4), together with the inference from (1 ) and (4) to (5), commit
me to (6). A commitment of this kind, incurred and sustained in the way
just described, is what I call a practical judgement of sameness of satisfaction (or
a PJSS, for short).
Note that if I’m committed to (6), I am also committed to
(7) x satisfies my word ‘robin’ if and only if x satisfies Alice’s word
‘robin’.
This commitment might also reasonably be described as a ‘‘practical judge-
ment of sameness of satisfaction’’. As I use this phrase, however, my
commitment to (7) is an immediate consequence of my practical judgement
of sameness of satisfaction for Alice’s word ‘robin’; the practical judge-
ment itself is my commitment to (6) given my unhesitating commitment
to (1 ), my explicit commitment to (4), and the inference from (1 ) and
(4) to (5).
    89

Recall that one can make a PIW that identifies a particular use of an
unregimented word w with a word w of one’s own regimented language.
When one combines such a PIW with an application of (Sats) to w , the
result is a PJSS for w. Hence one can make a PJSS for a particular use of an
unregimented word.
Let’s say that a licensing PIW for a given PJSS is a PIW the expression
of which figures in an explicit account of what incurs and sustains that
PJSS. For instance, my commitment to (1 ) in the above story expresses a
licensing PIW for my commitment to (6)—one of my PJSSs for Alice’s
word ‘robin’. A PJSS whose licensing PIW is a PIW at a given time may
be called a PJSS at a given time, and a PJSS at least one of whose licensing
PIWs is a PIW across time may be called a PJSS across time. Recall, for
instance, that when I first accept (1), which we now write as (1 ), I make
a PIW at a given time for Alice’s word ‘robin’. Together with (4), this
PIW commits me to (6), which, accordingly, amounts to a PJSS at a given
time for Alice’s word ‘robin’. If I later reaffirm (1 ) I thereby make a PIW
across time for the word ‘robin’ that Alice used when she said, ‘‘That is a
robin.’’ Together with (4), this PIW across time commits me to (6), which,
accordingly, amounts to a PJSS across time for Alice’s word ‘robin’.
If we make a PJSS for words w1 . . . wn that occur in an utterance
of a (regimented or unregimented) sentence s on a given occasion, by
combining our PIWs for w1 . . . wn with applications of (Sats) to w1  . . . wn  ,
respectively, of our own regimented language L, then, as we saw above,
our PIWs commit us to making a PIW-based identification of s with s , the
sentence of L (if there is one) that results from combining w1  . . . wn  , in the
order in which we take them to occur in s, according to the standard rules
for constructing sentences of L from words (predicates, connectives, and
quantifiers) of L. This PIW-based identification of s with s then commits
us to judging that s is true-in-L if and only if s is true-in-L. And since we
have a Tarski-style disquotational definition of true-in-L in terms of words
of L, we can derive a biconditional that specifies the conditions under
which both s and s are true-in-L. This is what I shall call a PJSS-based
judgement of sameness of truth value. Like the PIWs and PJSSs on which they
are based, a PJSS-based judgement of sameness of truth value may be classified
as a judgement of sameness of truth value at a given time, or across time,
according as the utterances or inscriptions of the sentences in question are
contemporaneous or separated by some period of time.
90   

3.4. Agreement and Disagreement

To see how our PIWs and PJSSs, hence also PJSS-based judgements of
sameness of truth value, are relevant to our understanding of agreement
and disagreement, let us revise the robin story slightly. Suppose as before
that Alice says, ‘‘That is a robin,’’ and I accept (1 ) and (4), thereby making
a PJSS for her word ‘robin’. In this revised version of the story, however, I
believe that all robins have yellow beaks, but the bird on the branch does
not have a yellow beak, and I see this, so I say, ‘‘That is a not a robin,’’
thereby expressing my belief that the bird to which Alice pointed is not
a robin. Then Alice will make a PIW for my word ‘robin’: she will take
my word ‘robin’ to be the same as her word ‘robin’, in the sense that she
will take me to have said what she expresses by affirming her sentence
‘The bird is not a robin’. If she also accepts applications of (Sats) to her
own words, then she will make a PJSS for my word ‘robin’. Suppose, in
addition, that Alice and I each introduce a predicate—I will abbreviate it
as ‘salient bird’—that expresses the contextually salient features and causal
connections that uniquely identify the bird in question, and we construct a
definite description—I will abbreviate it as ‘the salient bird’—that uniquely
denotes the bird at which I pointed. (I explained in Chapter 1 why we
cannot do without such context-sensitive additions to the vocabulary of
our regimented language.) Then if she and I take our respective utterances
to be sincere, we will each make PJSS-based judgements of sameness of
truth value for our respective utterances. She will take her sentence ‘The
salient bird is not a robin’ to have the same truth value as my sentence ‘The
salient bird is not a robin’, and so will I. We will thereby take ourselves to
disagree about whether the salient bird is a robin.⁵ In this context, we will
not take ourselves to disagree about whether the salient bird is a robin if we

⁵ This account of why Alice and I will take ourselves to disagree relies on an inference from sincere
assertion to belief that accords with Saul Kripke’s simple disquotational principle, which Kripke states as
follows:
If a normal English-speaker, on reflection, sincerely assents to ‘p’, then he believes that p. The sentence replacing
‘p’ is to lack indexical or pronominal devices or ambiguities, that would ruin the intuitive sense of the
principle . . . (Kripke 1979: 248–9)
Kripke adds that
When we suppose we are dealing with a normal speaker of English, we mean that he uses all words in
the sentence in a standard way, combines them according to the appropriate syntax, etc.: in short he
uses the sentence to mean what a normal speaker should mean by it. (Kripke 1979: 249)
    91

do not trust our PJSSs for one another’s words, including our respective
tokens of ‘robin’.
To understand this observation it helps to consider a context in which
Alice may take herself to disagree with me even though she makes no
PJSSs for any of my words. For instance, suppose we alter the context
described in the last two paragraphs so that Alice does not regiment her
words or apply (Sats) to them, but everything else that is unaffected by this
change remains the same. In the new context, when I say, ‘‘That is not
a robin,’’ it is still natural to suppose that Alice will nevertheless make a
PIW for my word ‘robin’ and take me to have said something that conflicts
with what she said. Moreover, in the new context I will still take Alice’s
word ‘robin’ to be the same as my word ‘robin’. And since I regiment
my words and apply (Sats) to them, I will still make a PJSS for Alice’s
word ‘robin’. (Recall that one can make a PJSS for a particular use of an
unregimented word.) I will therefore take myself to disagree with Alice
about whether the bird is a robin only if I trust my PJSSs for her tokens of
‘robin’, even though in this new context she need not make any PJSSs for
my word ‘robin’ in order to take herself to disagree with me. The crucial
point, however, is that if both Alice and I regiment our words and apply
(Sats) to them, then we will take ourselves to disagree about whether the
bird is a robin only if we both trust our PJSSs for one another’s tokens of
‘robin’.
We must also trust these judgements if we want to formulate logical laws
that show that we cannot both be right. Suppose that Alice and I each use
‘∼’ in place of ‘not’, ‘∧’ in place of ‘and’, and, as before, we construct
the description ‘the salient bird’ to pick out the bird at which we both
pointed. Then I will take Alice to have claimed that the salient bird is
a robin, and she will take me to have claimed that ∼(the salient bird is
a robin). We will see immediately that the conjunction ‘(the salient bird
is a robin) ∧ ∼(the salient bird is a robin)’ is inconsistent. For reasons
I sketched in §2.2, the clearest way to articulate the sense in which the
conjunction is inconsistent is to formulate the law that every sentence
of the form ‘∼(S ∧ ∼S)’ is true, or, equivalently, that every sentence of
the form ‘S ∧ ∼S’ is not true. And to state this law and apply it to the

I can agree with this only if it is understood in a pre-theoretical way, so that it does not imply that
there is an account of what normal speakers mean by their words that is independent of their PJSSs.
The reasons for this qualification will become clear in Chapter 4.
92   

sentences ‘the salient bird is a robin’ and ‘∼(the salient bird is a robin)’,
we may each use Tarski-style disquotational truth predicates defined for
our own sentences in terms of satisfaction clauses for our symbols ‘∼’ and
‘∧’. Alice and I will not be able to use our respective formulations of that
law to clarify the sense in which our respective claims about the bird are
inconsistent, however, if we do not also trust our PJSSs for one another’s
tokens ‘robin’. Our judgement that we disagree about whether the bird is
a robin rests on the PJSS-based judgements of sameness of truth value for
our respective uses of the sentences ‘the salient bird is a robin’ and ‘∼(the
salient bird is a robin)’.
Although we typically trust these judgements, we also sometimes revise
them. Suppose I know Alice is British, but I nevertheless unreflectively
accept (1 ), thereby making a PIW for her word ‘robin’. If I later learn
that her word ‘robin’—the British word ‘robin’—is not true of robins,
but of birds of a different species, I will no longer take her word ‘robin’
to be the same as mine, and so I will no longer take her to have
disagreed with me about whether the bird is a robin. I will revise the
PIW that I took to license my PJSS for her word ‘robin’ and regard
both the PIW and the PJSS as false. For different reasons I may also
revise a PJSS for one of my own previous uses of a given word, and
thereby revise my own current evaluations of beliefs I expressed by using
that word.
These preliminary observations remind us that when we identify and
clarify our agreements and disagreements with others by applying logical
laws that we express for sentences of our own language by using a Tarski-
style disquotational truth predicate, we make PJSS-based judgements of
sameness of truth value. We regard the PJSSs on which these judgements
of sameness of truth value are based as themselves true or false, and we
assume it is epistemically reasonable for us to trust sentences that express
our PJSSs unless we have particular, local reasons in a given context for not
doing so. In short, we regard our PJSSs as both factual and trustworthy. For
reasons I shall sketch in the next two sections, we also regard our PJSSs as
both factual and trustworthy when we take ourselves to learn from what
others tell us and when we take ourselves to be using an old familiar term
to express a new discovery.
    93

3.5. Learning from Others

Suppose, as before, that there is a bird on the branch in front of Alice and
me, and Alice says ‘‘That is a robin.’’ Trusting my practical identification of
her word ‘robin’ with mine, I take her to have said that the bird is a robin;
and given my disquotational definitions of satisfaction for my own words,
I take my practical identifications of her words to license corresponding
PJSSs for her words. In this revised version of the story, however, I believe
that all robins have yellow beaks. The bird on the branch does not have
a yellow beak, and I see this, so I say, ‘‘That is a not a robin,’’ thereby
expressing my belief that the bird is not a robin. Finally, suppose again that
in this situation Alice and I trust our respective PJSSs for one another’s
words, and take our respective utterances to be sincere, so that we make
PJSS-based judgements of sameness of truth value for the sentences we
uttered, and thereby take ourselves to disagree about whether the bird is
a robin.
Now suppose that when I say to Alice, ‘‘All robins have yellow beaks,
but the bird on the branch does not have a yellow beak, so it is not a
robin,’’ she relies, ‘‘No, as Hermione told me yesterday, not all robins
have yellow beaks.’’ Suppose, finally, that I take myself to have learned
from Alice (and indirectly, from Hermione) that not all robins have yellow
beaks, and I agree with Alice that the bird looks in all other ways like a
robin. Then I will naturally defer to Alice (and, indirectly, to Hermione),
and say, ‘‘Okay, perhaps I was wrong—the bird we just saw may be a
robin, after all.’’
To arrive at these conclusions I rely on my PJSS-based judgements
of sameness of truth value for Alice’s sentences, and thereby also on
my PJSSs at a given time for Alice’s words. When I conclude that my
previous utterance of ‘‘That is not a robin’’ is false, however, I also rely
on my PJSSs for the words in that previous utterance, including my PJSS
that
(6) (x satisfies ‘robin’, as I used it yesterday, if and only if x is a robin)
∧ (x satisfies my word ‘robin’, as I use it now, if and only if x is
a robin)
94   

This is a PJSS across time, because it relates my current disquotational account

of satisfaction for my words as I use them now, to my words as I used them
in the past.⁶ If I trust my PJSSs at a given time for Alice’s word ‘robin’ and
my PJSSs across time for my word ‘robin’ PJSSs, then, given my PJSSs for
‘the salient bird’, I am committed to a corresponding PJSS-based judgement
of sameness of truth value for Alice’s claim. Hence I am committed either
to rejecting Alice’s claim, or revising my own previous belief, and taking
myself to have learned from Alice that the salient bird is a robin. In short,
my trust in my PJSSs yields a contextually fixed framework for evaluating
my own belief that the bird is not a robin: if I continue to trust these PJSSs,
and also accept Alice’s claim, I will revise my belief and take myself to have
learned from Alice that the bird is a robin.
The situation in which I learn from Alice that the bird may be a robin
is typical of situations in which we learn from others by accepting what
they tell us and revising our beliefs accordingly. When we learn from
others in this way, we regard our PJSSs as true or false, and we assume
it is epistemically reasonable for us to trust sentences that express our
PJSSs unless we have particular, local reasons in a given context for not
doing so.

3.6. Discoveries
Chains of PIWs extend across time, from moment to moment, and, in some
cases, for centuries. For instance, if I learn that in 1820, John Audubon
pointed to a bird and said, ‘‘That is a robin,’’ I will take for granted that
there is a chain of practical identifications of words (PIWs) that can be
traced from my word ‘robin’ all the way back to Audubon’s unregimented
word ‘robin’. I will therefore make a practical identification of Audubon’s
word ‘robin’ with my word ‘robin’, and take him to have asserted that the
bird was a robin. This practical identification of Audubon’s word ‘robin’
with mine will commit me to a corresponding PJSS for his word ‘robin’.
Together with PJSSs for some of his other words, my PJSS for his word

⁶ It may seem odd to qualify ‘‘my word’’ by the time at which I use it. It will become clear in
Chapter 4 that this oddity is a consequence of the standard way of individuating words, not of the idea
of a PJSS.
    95

‘robin’ will commit me to PJSS-based judgements of sameness of truth

value for those sentences of his in which tokens of ‘robin’ occur.
We also rely on our PIWs and PJSSs when we take ourselves to have
made a discovery, and not simply stipulated a new use for an old term.
For instance, John Locke believed that pure gold is yellow. It was later
discovered that pure gold is white. The scientists who made this discovery
took themselves not to have changed the topic, but to have discovered that
gold is white. To accept this description of the case, we must trust the
later scientists’ practical identifications of the word ‘gold’ more than we
trust any previous judgement that gold is yellow. The later scientists might
be wrong. But we accept their practical identifications of ‘gold’. These
practical identifications are embodied in their use of that word to express
what they take to be discoveries about gold. Their use of that word links it
to earlier uses of the word, and those earlier uses of the word are linked
to even earlier uses of it. Taken together, these uses of ‘gold’ constitute
a trans-temporal chain of practical identifications of the word ‘gold’ that
support our current practical identifications of Locke’s use of ‘gold’ with
our word ‘gold’. And our current practical identifications of Locke’s use
of ‘gold’ with our word ‘gold’, in turn, license corresponding PJSSs for
Locke’s word ‘gold’. Together with PJSSs for some of Locke’s other
words, our PJSSs for his word ‘gold’ commit us to PJSS-based judgements
of sameness of truth value across time for those sentences of his in which
tokens of ‘gold’ occur.
In a similar way, every inquiry brings with it some PIWs, PJSSs, and
PJSS-based judgements of sameness of truth value across time. When we
take ourselves to have made a discovery, and not just to have stipulated a
new use for an old term, we regard our PJSSs as true or false and we assume
it is epistemically reasonable for us to trust sentences that express our PJSSs
unless we have particular, local reasons in a given context for not doing so.

3.7. A Reformulation of the Intersubjectivity

Constraint
As I explained earlier (§3.1), our desire to collaborate fruitfully with
others motivates us to adopt the intersubjectivity constraint: a Tarski-style
disquotational truth predicate defined for one’s own sentences is satisfactory only if
96   

it is supplemented by an account of why it is epistemically reasonable for one to

use it to generalize over other speakers’ sentences and one’s own sentences as one
used them in the past. We have now seen that when one generalizes over
other speakers’ sentences and one’s own sentences as one used them in the
past, one makes PJSS-based judgements of sameness of truth value, and
thereby regards one’s PJSSs as true or false. One assumes it is epistemically
reasonable for one to trust sentences that express one’s PJSSs unless one
has particular, local reasons in a given context for not doing so. In short,
one regards one’s PJSSs as both factual and trustworthy. This suggests
the following reformulation of the intersubjectivity constraint: a Tarski-
style disquotational truth predicate defined for one’s own regimented sentences is
satisfactory only if it is supplemented by an account of why it is epistemically
reasonable for one to regard one’s PJSSs as both factual and trustworthy.⁷ I shall
henceforth use this formulation in place of the previous one. In the rest of
this chapter I shall consider whether we have good pragmatic grounds for
adopting it.
Recall that I do not seek a conceptual analysis of the English word ‘true’,
but pragmatic answers to the questions ‘‘Do we need a truth predicate?’’
and ‘‘If so, what sort of truth predicate do we need?’’ In this context
it is not helpful to ask whether the intersubjectivity constraint captures
something essential to the concept of truth. Instead, the intersubjectivity
constraint should be viewed as a proposed constraint on a satisfactory
explication of ‘true’, where the purpose of the explication is to fill the
functions of ‘true’ that we find useful for clarifying and facilitating our
inquiries. We saw above (§§2.6–2.9) that a Tarski-style disquotational
truth predicate fills some of the functions of ‘true’ that we find useful for
clarifying and facilitating our inquiries. From this pragmatic perspective,
the question whether we have good grounds to embrace the reformulated
intersubjectivity constraint amounts to the question whether a Tarski-
style disquotational truth predicate that does not satisfy the reformulated

⁷ I assume that to use one’s own Tarski-style truth predicates to generalize over other speakers’
sentences and over one’s own sentences as one used them in the past one must make practical
judgements of sameness of satisfaction for other speakers’ words and for one’s own words as one used
them in the past. Some may have intuitions about truth that conflict with this assumption, but it is not
part of my pragmatic project to accommodate such intuitions. As I see it, the intersubjectivity constraint
is motivated by our desire to use logical generalizations to clarify agreements and disagreements with
other speakers in contexts in which our identifications of such agreements and disagreements rest on
PJSSs for each other’s words, and for our own words as we used them in the past.
    97

intersubjectivity constraint fills all the functions of ‘true’ that we need to

fill in order to clarify and facilitate our inquiries.
One’s Tarski-style disquotational truth predicate might fail to satisfy the
intersubjectivity constraint in either (or both) of two ways: (first) it might
not be supplemented by an account of why it is epistemically reasonable
for one to regard one’s PJSSs as trustworthy and (second) it might not
be supplemented by an account of why it is epistemically reasonable for
one to regard one’s PJSSs as factual. To clarify the pragmatic grounds for
adopting intersubjectivity constraint, I examine these two ways in which
one’s Tarski-style disquotational truth predicate might fail to satisfy the
intersubjectivity constraint, presenting versions of each as objections to that
constraint.

3.8. Trust without Trustworthiness?

One might object to the intersubjectivity constraint because one thinks
that to satisfy our need to generalize on other speakers’ sentences by
using one’s own Tarski-style disquotational truth predicate, it is enough to
suppose that it is epistemically reasonable to trust one’s PJSSs unless one has
particular, local reasons in a given context for not doing so; one need not
also provide an account of why it is epistemically reasonable to suppose this,
as the intersubjectivity constraint requires. We typically regard our PJSSs
as trustworthy—we assume it is epistemically reasonable to trust our PJSSs
unless we have particular, local reasons in a given context for not doing
so—without even considering the question whether this attitude towards
our PJSSs is itself epistemically reasonable. As theorists, however, we cannot
be completely comfortable with this unreflective attitude towards our PJSSs.
We know that to satisfy our desire to clarify and facilitate our inquiries, we
need to explicate ‘true’ in a way that fills and clarifies the functions of ‘true’
that we find useful. One of the most useful functions of the word ‘true’
is that it enables us to express logical generalizations on other speakers’
sentences and on our own sentences, as we used them in the past. To make
the case that a proposed explication of ‘true’ is satisfactory, then, we need to
make the case that it fills the useful function just described. But if we do not
understand how it could fill that function, then we do not understand how
the proposed explication could be satisfactory. Hence we don’t understand
98   

how a Tarski-style disquotational truth predicate defined for our own

sentences could be satisfactory unless we supplement it with an account of
why it is epistemically reasonable for one to trust one’s PJSSs unless one has
particular, local reasons in a given context for not doing so. Contrary to
the objection, to establish that a Tarski-style disquotational truth predicate
defined for one’s own sentences is satisfactory we must supplement it with
an account of why it is epistemically reasonable for one to trust one’s PJSSs
unless one has particular, local reasons in a given context for not doing so.
The account need not be one that every inquirer is required to be able to
offer in support of her own trust in her PJSSs; instead, it must be one that
we theorists could use to explain why a given speaker’s trust in her PJSSs is
epistemically reasonable.

3.9. A Quinean Objection: PJSSs are not Factual

One might grant that a Tarski-style disquotational truth predicate is
satisfactory only if it is supplemented by an account of why it is epistemically
reasonable for one to regard one’s PJSSs as trustworthy, yet doubt that such
a truth predicate is satisfactory only if it is also supplemented by an account
of why it is epistemically reasonable for one to regard one’s PJSSs as
factual. This is not just a logically possible doubt: for reasons I will rehearse
below, Quine’s empirical theory of translation implies that our PJSSs are
trustworthy but not factual. If one were to combine one’s own Tarski-style
disquotational truth predicate with Quine’s empirical theory of translation,
it would be epistemically reasonable for one to regard one’s PJSSs as
trustworthy, but it would not be epistemically reasonable for one to regard
them as factual. The resulting combination is therefore compatible with the
conclusion of the previous section, yet challenges the second part of the
intersubjectivity constraint, according to which a Tarski-style disquotational
truth predicate defined for one’s own sentences is satisfactory only if it is
supplemented by an account of why it is epistemically reasonable for one
to regard one’s PJSSs as factual.⁸

⁸ The idea that a disquotational account of truth is satisfactory only if it is supplemented with an
account of why it is reasonable for one to regard one’s PJSSs as trustworthy is compatible with Quine’s
commitments, but as far as I know Quine himself never explicitly endorsed it.
    99

Before I can review the reasons why Quine’s empirical theory of

translation implies that our PJSSs are trustworthy but not factual, I must
explain in more detail what I mean when I say that PJSSs are factual.
I emphasized above that one’s PJSSs are factual if and only if they are
true or false. The explication of ‘factual’ that I recommend rejects the
traditional idea that a sentence is true or false according as it ‘‘corresponds’’
or fails to ‘‘correspond’’ with facts construed as special kinds of things in
the world. Following Quine, I assume that there are no fact-things to
which our declarative sentences must ‘‘correspond’’ if they are to be true,
and no indispensable uses of the word ‘‘fact’’. (‘‘It is a fact that snow is
white,’’ for instance, is equivalent to, and can be replaced, by ‘‘Snow is
white.’’) In this Quinean spirit, I propose that we explicate ‘factual’ as
follows:
(F) If S is a sentence we can use, then S (as used on occasion O) is factual
if and only if either
(i) we can construct a Tarski-style disquotational definition of ‘true-
in-L’ for some language L that contains S (as used on O), or
(ii) there is a sentence S such that (a) we can construct a Tarski-style
disquotational definition of ‘true-in-L’ for some language L that
contains S , and (b) we take S to paraphrase S (as used on O).
We saw in Chapter 2 that
(G) We can each construct a Tarski-style disquotational definition of
‘true-in-L’ for some language L that contains a given sentence S if
and only if S is a regimented declarative sentence S that we can use.
(F) and (G) imply
(H) If S is a sentence we can use, then S (as used on occasion O) is
factual if and only if either
(i) S is a regimented declarative sentence, or
(ii) there is a regimented declarative sentence S that we can use
and that we take to paraphrase S (as used on O).
It is crucial to see that we cannot determine whether or not one of our
sentences is factual according to (H) on formal grounds alone. Sentences of
ours that look like regimented declarative sentences may seem unclear to
100   

us in ways that prevent us from regarding them as factual.⁹ For instance,

sentences that express our practical judgements of sameness of satisfaction
look like regimented declarative sentences; hence one might think that if
we accept (H) we are thereby committed to saying that such sentences are
factual. But we may accept (H) without thereby deciding how we will
regiment sentences that express our practical judgements of sameness of
satisfaction, hence without thereby deciding whether to regard them as
factual or as non-factual.
Quine himself is committed to regarding them as non-factual. His theory
of empirical translation implies his indeterminacy thesis, according to which
a speaker’s language can be mapped onto itself (and any other language
that it translates can be mapped onto it) in a variety of inequivalent ways,
each of which preserves the net behaviouristic association of sentences
with sensory stimulation. He explicates ‘empirical content’ in such a way
that the empirical content of any given sentence is exhausted by the
behavioural dispositions that link it to sensory stimulation, and hence
that the behaviouristic association of sentences with sensory stimulation
is all that matters to empirical translation. A translation, or mapping of
words to words, preserves the behaviouristic association of sentences with
sensory stimulation, according to Quine, just in case the mapping allows
for ‘‘fluency of dialogue’’, described behaviouristically. Our linguistic
dispositions do not uniquely determine translation, either of full sentences
or of the predicates they contain, Quine argues, because countless radically
different translations would each pass his behaviouristic test for ‘‘fluency
of dialogue’’. He concludes that translation is not settled by all the facts
relevant to it, and is therefore objectively indeterminate.¹⁰
A corollary of his conclusion is that it is not epistemically reasonable
for us to regard sentences that presuppose particular mappings between
one’s own words and another speaker’s words as part of an object language
for which we can define a Tarski-style disquotational truth predicate. But
sentences that we use to express our PJSSs presuppose such mappings. In
the robin case described above, for instance, when I affirm
⁹ Boghossian 1990 argues that if we accept a disquotational deflationism about truth then we are
committed to saying that all apparently meaningful sentences are factual. I believe this conclusion
is undermined by Quine’s sophisticated disquotational deflationism about truth and factuality. For a
criticism of Boghossian from a broadly Quinean point of view, see Kraut 1993.
¹⁰ I present a more complete account of Quine’s argument for his indeterminacy thesis in Chapter 2
of Ebbs 1997.
    101

(6) (x satisfies my word ‘robin’ if and only if x is a robin) and (x satisfies

Alice’s word ‘robin’ if and only if x is a robin),
I presuppose, in effect, what Quine calls a homophonic translation manual
between Alice’s word ‘robin’ and my word ‘robin’. According to Quine,
translation is objectively indeterminate, and so we should not regard such
sentences, or the PJSSs we use them to express, as factual in the sense
defined by (H).
Nevertheless, if we accept Quine’s account of translation it is epistem-
ically reasonable for us to regard the sentences that we use to express our
PJSSs, and, indirectly, our PJSSs themselves, as trustworthy, in the pragmatic
sense that we can count on them to facilitate communication with other
inquirers. Quine emphasizes that
Different persons growing up in the same language are like different bushes
trimmed and trained to take the shape of identical elephants. The anatomical
details of twigs and branches will fulfill the elephantine form differently from bush
to bush, but the overall outward results are alike. (Quine 1960: 8)

Similarly, when a person learns a foreign language, her dispositions to relate

sentences of that language with sensory stimulation come to match the
dispositions of native speakers of the language well enough to facilitate
communication. In both contexts, according to Quine, our behavioural
conditioning ensures that the sentences that we use to express our PJSSs
preserve all the relevant speech dispositions and are therefore trustworthy, in
one good pragmatic sense of that word.¹¹
If we combine these consequences of Quine’s view of translation with
Tarski-style disquotational definitions of truth, we may be tempted to
conclude that even though it is epistemically reasonable for us to regard
our PJSSs as trustworthy, because our homophonic and homographic
translations usually pass the fluency test, it is not epistemically reasonable for
us to regard our PJSSs as factual, because there are countless other equally
acceptable translations of others’ words into our own, and of our own
words into themselves. The resulting Quinean explication of truth implies
that a Tarski-style disquotational definition of truth is satisfactory if it is

¹¹ For Quine there is no sharp distinction between epistemological and pragmatic considerations:
‘‘Each man is given a scientific heritage plus a continuing barrage of sensory stimulation; and the
considerations which guide him in warping his scientific heritage to fit his continuing sensory
prompting are, where rational, pragmatic’’ (Quine 1961b: 46).
102   

supplemented by an account of why it is not epistemically reasonable for one

to regard one’s PJSSs as factual, and hence (assuming it is consistent) even if
it is not supplemented by an account of why it is epistemically reasonable for
one to regard one’s PJSSs as factual. In this way, the Quinean explication
of truth yields a powerful objection to the intersubjectivity constraint.

3.10. Realism as Integral to the Semantics

of the Predicate ‘True’
To evaluate the objection, we must ask whether the proposed Quinean
explication of truth is satisfactory. Does it fill the functions of ‘true’ that
we find useful for clarifying and facilitating our inquiries? We stand the
best chance of reaching agreement about this if we focus on central aspects
of our inquiries. I propose that we focus on Frege’s observation that
we rely on logic to arbitrate between conflicting opinions. For reasons I
explained above (§§3.4–3.6), to take ourselves to be able to apply logical
generalizations to arbitrate conflicting opinions, to learn from others by
trusting what they say, or to use an old familiar term to report a discovery
that conflicts with claims we or others had previously made by using that
term, we need to regard our PJSSs as factual. Quine himself emphasizes
that
We should and do currently accept the firmest scientific conclusions as true, but
when one of these is dislodged by further research we do not say that it had been
true but became false. We say that to our surprise it was not true after all. Science
is seen as pursuing and discovering truth rather than as decreeing it. Such is the
idiom of realism, and it is integral to the semantics of the predicate ‘true’. (Quine
1995b: 67)

In this passage Quine is in effect drawing our attention to the fact that our
scientific re-evaluations of past conclusions typically presuppose PIWs and
PJSSs across time. A related, more general, point is that to regard truth
as independent of our own current beliefs we need to rely on our PIWs
and corresponding PJSSs, both at a given time and across time. This is
especially salient when we apply logic to arbitrate disagreements, for to do
so we must concede that an interlocutor with whom we take ourselves to
disagree may be right, despite our confidence that she is not. The question
    103

is whether the proposed Quinean explication of truth does justice to these

important aspects of our inquiries.
The answer is that it does not. As I explained above, Quine’s indeterm-
inacy thesis implies that PIWs on which our PJSSs rely are not factual:
they merely express our immediate, unhesitating subjective preference for
one kind of translation manual over another. This in turn implies that
whether or not we take ourselves to have been wrong or to agree or
disagree with other speakers ultimately rests on our subjective preference
for some translations (of our own words as used as some previous time,
or of the other speakers’ words) over other quite different translations (of
our own words as used as some previous time, or of the other speakers’
words) that are equally correct according to Quine’s empirical theory of
translation. Suppose we take ourselves to disagree with a particular speaker,
making PIWs for her word, and thereby, in effect, ‘‘translating’’ her words.
Quine’s indeterminacy thesis implies that we could have chosen to translate
her words in many other equally acceptable but radically different ways.
Relative to some of the acceptable alternative translations of her words,
we will not take ourselves to disagree with her. Our identifications of
agreements and disagreements are therefore not objective; they reflect
our PIWs, which amount to subjective preferences about how to translate
another’s words. Similarly, when we take ourselves to have been wrong,
we presuppose PIWs and PJSSs across time for our words. But Quine’s
account of translation implies there are other acceptable translations of our
past utterances relative to which we would not take ourselves to have been
wrong. And if the relationship between earlier and later uses of the term
is always mediated by subjective preference for one manual of translation
over other equally acceptable ones, then so is our understanding of the
claim that what we accept now may not be true.
With this in mind, consider again Quine’s observation that the idiom
of realism—the independence of truth and belief—‘‘is integral to the
semantics of the predicate ‘true’ ’’. We cannot do justice to this observation
unless we regard sentences such as (6) that express our PJSSs as declarative
regimented sentences for which we can define truth disquotationally,
and hence, as factual, in the sense defined by (H). But the Quinean
proposal sketched in §3.9 combines a Tarski-style disquotational definition
of ‘true-in-L’ with Quine’s empirical theory of translation, which implies
that sentences that express our PJSSs are not factual. Hence the Quinean
104   

explication of truth described in the previous section fails to preserve the

minimal sort of realism that, as Quine himself observes, is ‘‘integral to the
semantics of the predicate ‘true’ ’’.
A principled Quinean should insist that despite Quine’s own claim to the
contrary, the idiom of realism is not integral to the semantics of the predicate
‘true’—our strong feeling that our PJSSs are integral to the semantics of
the predicate ‘true’ is just a psychological accompaniment of our habitual
applications of ‘true’. If we accept this Quinean account, we can at best
simulate the role of our PJSSs in collaborative inquiries by highlighting the
psychological fact that we typically prefer homophonic ‘‘translations’’ of
our fellow inquirers’ words and of our own words as we used them in
the past. There may be long-term stability in our subjective preferences
about how to ‘‘translate’’ our fellow inquirers’ words and our own words
as we used them in the past, but the principled Quinean will insist that
this is a merely psychological stability that will end as soon as a number
of inquirers choose to adopt radically different (yet, by Quine’s standards,
equally correct) ‘‘translations’’ of their fellow inquirers’ words or of their
own words as they used them in the past.
The problem for the principled Quinean is that we take our PJSSs to be
trustworthy and factual. To accept the Quinean explication of truth would
be to discredit our ordinary practice of taking our PJSSs to be factual, and
thereby to discredit an attitude towards our PJSSs that partly constitutes our
commitment to the stability of our inquiries. My criticism of the Quinean
explication is not that our ordinary practice of taking our PJSSs to be factual
is part of the concept (if there is one) expressed by the English word ‘true’,
but that it is a useful feature of our inquiries that we should seek to preserve
in our explication of ‘true’. I conclude that a Tarski-style disquotational
truth predicate defined for one’s own sentences is satisfactory only if it is
supplemented by an account of why it is reasonable for one to regard one’s
PJSSs as factual.
 4
How to Think about Words

4.1. Is the Tarski–Quine Thesis Incompatible

with the Intersubjectivity Constraint?
We have now seen how our initial pragmatic reasons for adopting
The Tarski–Quine thesis: there is no more to truth than what is captured by a
Tarski-style disquotational truth predicate defined for one’s own sentences
may, after further reflection, lead us to conclude that we also have pragmatic
reasons for adopting
The Intersubjectivity Constraint: a Tarski-style disquotational truth predicate
defined for one’s own sentences is satisfactory only if it is supplemented by an
account of why it is reasonable for one to regard one’s PJSSs as both factual and
trustworthy.
The intersubjectivity constraint is not equivalent to the negation of the
Tarski–Quine thesis. Yet we feel confident that the constraint and the
thesis are incompatible. They are incompatible if and only if
(C) There is no way of explaining why it is epistemically reasonable for
one to regard one’s practical judgements of sameness of satisfaction
as both factual and trustworthy without claiming that there is more
to truth than what is captured by a Tarski-style disquotational truth
predicate defined for one’s own regimented sentences.
We are strongly inclined to accept (C), and hence to conclude that the
intersubjectivity constraint is incompatible with the Tarski–Quine thesis.
But why are we so strongly inclined to accept (C)?
The answer, as I shall now try to show, is that we tend to conceive of
words in a way that implies (C). It turns out, however, that this tendency,
106     

while strong, is resistible. I shall sketch an alternative conception of words

that implies that we have no grip on sameness of satisfaction apart from our
PJSSs. If we adopt this alternative, as I recommend, we can simultaneously
embrace the Tarski–Quine thesis and show that it is epistemically reasonable
for us to regard our PJSSs as both factual and trustworthy.

4.2. Use versus Mention (Transparent Use)

When we are doing logic and semantics, our theorizing about words
starts with the unimpeachable logical distinction between use and mention.
Standard expositions of the use-mention distinction (starting with Quine
1940: §4) typically focus on the difference between using and mentioning
proper names. For instance, if I affirm the sentence
Chicago is a populous city.
I use the name ‘Chicago’ to mention (refer to) Chicago, but do not thereby
mention (refer to) ‘Chicago’. In the sense of ‘language use’ that logicians
contrast with ‘language mention’, to use ‘Chicago’ is to refer to Chicago,
not to describe the reference relation between ‘Chicago’ and Chicago,
or to explain how it is possible for one to use ‘Chicago’ to refer to
Chicago.¹
The use-mention distinction also holds for predicates and logical con-
stants for which we have defined a Tarski-style truth predicate in the way
I explained in Chapter 2. For instance, when I entertain, assume, or affirm
a sentence such as
∀x1 ((x1 is mortal) ∨ ∼(x1 is mortal))
in which the predicate ‘mortal’ occurs, I thereby use that predicate in such
a way that it is satisfied by all and only mortal things, but do not thereby

¹ It is not difficult to construct a sentence any utterance of which both uses and mentions a word
that occurs in the sentence only once. For instance, there is only one occurrence of ‘Chicago’ in

(c) The last word in this sentence is causally related to Chicago.

But when I affirm (c) I both use the word ‘Chicago’ to refer to Chicago and mention the word ‘Chicago’,
saying that it is causally related to Chicago. Nevertheless, when I affirm (c), I use the word ‘Chicago’
to refer to Chicago, not to the word ‘Chicago’; I use a different singular term, namely, ‘The last word
of this sentence’, to refer to ‘Chicago’.
      107

mention the word ‘mortal’. And when I entertain, assume, or affirm the
sentence just displayed, I thereby use it in such a way that it is true-in-L if
and only if everything is mortal or not mortal, but I do not thereby mention
the sentence itself.
Insofar as one is using one’s regimented predicates, logical constants, and
closed sentences, one does not refer to them or consciously attend to them
as linguistic items—they are, one might say, transparent.² This metaphor
allows for a convenient shorthand. In place of the phrase, ‘‘language use of
the sort that contrasts with mention’’, we can write, ‘‘transparent use’’. And
instead of saying that a person ‘‘is using her words in the sense that contrast
with mentioning them’’ we can say she is using her words ‘‘transparently’’.
One should resist any temptation to take the shorter, metaphorical phrases
to have special explanatory content of their own. I shall use them only as a
convenient shorthand for the longer phrases they replace.
Despite the practical indispensability of our transparent uses of words, the
logician’s use-mention distinction leads in three apparently irresistible steps
to the conclusion that our transparent uses of words are not fundamental
to our conception of the word types relevant to truth and logic. In
the first step we group word tokens into types by their spelling or
pronunciation alone; in the second step we conjecture that the meanings
and satisfaction conditions of word tokens are determined by facts about
the tokens; and in the third step we conjecture that two word tokens
are of the same type if and only if they are spelled or pronounced in
the same way and facts about them determine that they have the same
meanings and satisfaction conditions. Let us examine these three steps in
detail.

4.3. The Orthographic Conception of Words

The first step is an immediate consequence of the standard way of drawing
the use-mention distinction. When logicians draw that distinction for a
regimented language, they identify words by their spelling alone, and
regard them as strings of tokens of letter types. According to this orthographic
conception, the ink marks ‘Chicago, Chicago’, for example, instantiate two

² It is for this reason, I think, that Quine calls disquotational truth transparent (Quine 1992: 82).
108     

tokens of the same word type C-h-i-c-a-g-o.³ The orthographic conception

of word types provides explicit criteria of sameness and difference of word
types that we exploit intrasubjectively when we apply Tarski’s method
of defining truth and satisfaction for our own regimented sentences and
words. Recall that Tarski’s method requires (as a regulative ideal) that one
be able to identify and distinguish between the words and sentences of
one’s own regimented language by spelling alone. For instance, we are
each in a position to affirm
(4) x satisfies my word ‘robin’ if and only if x is a robin
taking for granted (as a regulative ideal) that we can identify our own
regimented word ‘robin’ from all other words of our current regimented
language by spelling alone. When we affirm (4), we mention (refer) to our
word ‘robin’ on the left side of the biconditional in (4), and use it on the
right side. If we start with an orthographic conception of words, it is also
natural to add a corresponding phonetic conception of words, according
to which we map uttered sounds to spellings and spellings to words when
we identify words in speech.⁴ Given such a mapping, when we each
affirm (4) we each thereby specify conditions under which a given object
satisfies our own current written or spoken tokens of the orthographic
type: r-o-b-i-n.

4.4. Explanatory Use (Ex-Use)

In the second step of the reasoning previewed above, we ask how word
tokens individuated orthographically or phonetically come to have mean-
ings and satisfaction conditions. Once we adopt both the orthographic and
phonetic conceptions of words, we are inclined to infer that (i) any inscrip-
tion or utterance can be viewed as an inscription or utterance of tokens of

³ Peirce 1933: iv. 423 is the classic source of the distinction between a type and its tokens. I get the
term ‘‘orthographic’’ from Kaplan 1990.
⁴ For languages that have no orthography, the closest analogue to sameness of spelling is sameness of
the sounds or gestures that speakers of the language use to communicate. Those who are attracted to the
orthographic conception may want to extend their account of words to encompass languages without
orthography by specifying analogous criteria for sameness and difference of words in terms of sounds or
gestures. I will not try to do this here, since I am primarily interested in how we identify the words of
regimented languages.
      109

word types, considered as strings of letters and spaces. We then notice that
(ii) tokens of a given string of letters and spaces, such as r-o-b-i-n, may have
different satisfaction conditions depending on where, when, and by whom
they are inscribed or uttered. And (i) and (ii) naturally lead us to ask, ‘‘How
do such tokens, which appear lifeless and insignificant, come to have the
meanings and satisfaction conditions we ordinarily take them to have?’’
The only plausible answer, it seems, is that there are facts about our word
tokens that determine their meanings and satisfaction conditions, to the
extent that their meanings and satisfaction conditions are determined at all.
But what sorts of facts? And do such facts uniquely determine the meanings
and satisfaction conditions of our word tokens? Every substantive theory
of linguistic meaning offers its own distinctive answers to these questions.
W. V. Quine’s theory of linguistic meaning focuses on facts about how
impacts at a speaker’s nerve endings prompt her to assent to sentence tokens
that contain her word tokens, and implies that the totality of such facts
about word tokens does not uniquely determine meanings and satisfaction
conditions for the word tokens. Similarly, Donald Davidson’s theory of
linguistic meaning focuses on facts about how observable circumstances
in a speaker’s surrounding environment prompt her to hold-true sentence
tokens that contain her word tokens, and implies that the totality of such
facts does not uniquely determine meanings and satisfaction conditions of
the word tokens. H. P. Grice’s theory of linguistic meaning focuses on facts
about speakers’ intentions to convey their beliefs to others by instituting
and exploiting linguistic conventions. Unlike Quine and Davidson, Grice
assumes that such facts uniquely determine the meanings and satisfaction
conditions of a speaker’s word tokens. The causal-historical theories of
linguistic meaning inspired by the work of Saul Kripke and Hilary Putnam
focus on facts about causal-historical relations between a speaker’s word
tokens, other objects, and word tokens uttered by other speakers. There are
many other substantive theories of meaning, each of which implies its own
distinctive account of which facts about word tokens, if any, determine
their meanings and satisfaction conditions.
The facts about word tokens that figure in such theories are supposed to
describe or explain linguistic events that occur simultaneously with speakers’
own transparent uses of their words. When I say, ‘‘That is a robin,’’ while
pointing at a particular robin, thereby using tokens of my words ‘That’,
‘is’, ‘a’, and ‘robin’ to say that the bird is a robin, there are simultaneous
110     

linguistic events of my uttering or inscribing tokens of these words. An

event of uttering or inscribing a word token t that is simultaneous with a
particular transparent use of t is what I call an explanatory use (or ex-use) of
t. An account of ex-use is a theory of linguistic meaning that specifies which
linguistic events count as ex-uses of word tokens, and which facts about the
ex-use of a given word token determine whatever meanings or satisfaction
conditions the word tokens have.
There are as many different accounts of ex-use as there are substantive
theories of linguistic meaning. According to Quine’s theory of linguistic
meaning, for instance, a speaker’s ex-use of a word token t is an uttering
of sounds or a writing of marks, and the facts that determine the meaning
and satisfaction conditions of t, to the extent that these are determined
at all, are facts about which patterns of impacts at the speaker’s nerve
endings trigger her dispositions to assent to her sentence tokens, including
the particular sentence token in which t occurs (Quine 1960). Similarly,
according to Davidson’s theory, an ex-use of a word token t is an uttering
of sounds or a writing of marks, and the facts that determine the meaning
and satisfaction conditions of t, to the extent that these are determined
at all, are facts about the observable circumstances in which the speaker
holds-true her sentence tokens, including the particular sentence token in
which t occurs (Davidson 1984). In contrast, according to Grice’s account,
a particular ex-use of a word token t is not (or not only) the result of a
triggering of linguistic dispositions, or the holding-true of particular sentences
under particular circumstances, but (also) the result of a speaker’s intention
to utter a sentence token that contains t in order to convey one of her
beliefs to someone else, and the meaning and satisfaction conditions of
t are determined by the facts about the particular intention with which
the speaker utters t (Grice 1957). Kripke’s and Putnam’s causal-historical
accounts of linguistic meaning also characterize ex-use partly in terms of
speakers’ intentions, but stress that what is expressed by an ex-use of a
word token t is determined in part by facts about how t is causally related
to word tokens uttered by other speakers on previous occasions, as well as
to objects in the speaker’s environment, and objects in the environment of
speakers who lived long ago (Kripke 1980, Putnam 1975).⁵

⁵ In later chapters I shall discuss several of these examples in much more detail. There are many
other substantive theories of linguistic meaning currently available in the literature, but I assume that
      111

Accounts of ex-use disagree about how to describe the relationship

between a transparent use of a word token t and an ex-use of t with which
it is simultaneous. According to Davidson’s account of ex-use, for example,
each transparent use of a word token t is identical to an ex-use of t. To
describe a transparent use of t as an ex-use of t, or to describe an ex-use of
t as a transparent use of t, is, according to Davidson’s account, to redescribe
one and the same action event.⁶ According to some other accounts of
ex-use, in contrast, each transparent use of a word token t is expressed by,
but not identical to, an ex-use of t.⁷
Despite such differences of detail, however, every account of ex-use implies
that when I use a word token transparently, I do not thereby characterize its ex-use.
For instance, when I affirm
(4) x satisfies my word ‘robin’ if and only if x is a robin.
I do not thereby say anything about the ex-use of my word ‘robin’.
Similarly, when I affirm
(6) (x satisfies my word ‘robin’ if and only if x is a robin) and (x satisfies
Alice’s word ‘robin’ if and only if x is a robin).
I express a practical judgement of sameness of satisfaction (PJSS) for Alice’s
word ‘robin’, but I do not thereby state facts about the ex-uses of tokens
of Alice’s word ‘robin’ that together determine that tokens of Alice’s word
‘robin’ are satisfied by an object x if and only if x is a robin. In general, to
make a PJSS for a word token w is not to say anything about the ex-use of
w, but to specify satisfaction conditions for w by transparently using a word
token w that one takes to be of the same word type as w. To state facts
about a word token’s ex-use one must affirm some sentences that do not
express PJSSs for it.

one does not need a complete list of such theories to understand what I mean by ‘‘an account of
ex-use’’.
⁶ Hence Davidson writes that: ‘‘We interpret a bit of linguistic behavior when we say what a
speaker’s words mean on an occasion of use. The task may be seen as one of redescription. We know
that the words ‘Es schneit’ have been uttered on a particular occasion and we want to redescribe
this uttering as an act of saying that it is snowing’’ (Davidson 1984: 141). The same contrast can
be drawn for one of our own current utterances of ‘It is snowing’: viewed transparently, it is an
act of saying that it is snowing; viewed as an ex-use of ‘It is snowing’, it is ‘‘a bit of linguistic
behavior’’.
⁷ These other accounts presuppose accounts of act individuation similar to the one presented in
Goldman 1970.
112     

4.5. The Token-and-Ex-Use Model of Words

If we accept the reasoning of the last two sections, we will be strongly
inclined to take the third and final step in the reasoning previewed
above—to conjecture that two word tokens are of the same semantic type
if and only if they are spelled or pronounced in the same way and facts
about them determine that they have the same meanings and satisfaction
conditions. In other words, we will be strongly inclined to think of the
word types that are relevant to satisfaction and truth in the following way:
(U) Two word tokens t and t are of the same word type if and only if
(i) t and t are each tokens of the same orthographic or phonetic
type, and
(ii) facts about the ex-uses of t and t determine that they each have
semantic values and that their semantic values are the same.
This is what I shall call a token-and-ex-use conception of words. There is a
family of such conceptions, each member of which defines the word types
relevant to satisfaction and truth in terms of word tokens and facts about
their ex-uses, where the facts about the ex-uses of two word tokens are
supposed to determine whether the tokens each have semantic values, and,
if so, whether they have the same semantic values. To conceive of the word
types relevant to satisfaction and truth in this way is to accept what I shall
call the token-and-ex-use model of words.
Although some token-and-ex-use conceptions of words, such as Quine’s,
presuppose that the ex-uses of word tokens can in principle be completely
specified in non-intentional, scientific terms, other token-and-ex-use con-
ceptions of words require only that the ex-uses of word tokens can in
principle be specified without using any sentences that express our prac-
tical identifications of words (PIWs) or practical judgements of sameness
of satisfaction (PJSSs) for the word tokens.⁸ To fix ideas, I shall take
(U) as a representative of the token-and-ex-use conceptions of words
that make only this minimal demand on our specifications of ex-uses of
word tokens. We can therefore adopt (U) without committing ourselves

⁸ And without using any sentences whose content depends anaphorically on a sentence that expresses
a practical identification of a word or a corresponding PJSS for that word token. Thanks to Steven
Gross for pointing out that I need to close this legalistic loophole.
      113

to reductionism about meaning, hence without ruling out that among the
facts about the ex-uses of word tokens are irreducible facts about what
the word tokens mean. For instance, we can adopt (U) without rejecting
the anti-reductionist views of meaning that Boghossian 1989, Horwich
1995, McGinn 1984, and Wright 1984 defend in response to Kripke’s
Wittgenstein-inspired scepticism about meaning in Kripke 1982. We can
also adopt (U) without rejecting that the ex-use of a word token t is fixed
by the dispositions and functional states, if any, that are causally responsible
for a speaker’s applications of t and facts about causal relations that exist
between t, other speakers, objects, and events in the social and physical
environment in which t is uttered. Hence we can adopt (U) without
thereby rejecting any of the standard projects of explaining meaning in
naturalistic terms, including projects of this kind that are inspired by Grice
1957, Putnam 1975, and Kripke 1980. One can even adopt (U) without
ruling out the possibility that no two word tokens are tokens of the same
word type, in the sense of ‘‘word type’’ specified by (U), as Quine is
compelled to conclude by his argument for indeterminacy of translation.
Regardless of our particular views about meaning, we are all strongly
inclined to accept some token-and-ex-use conception of words or other,
for reasons I sketched above. The diagnostic significance of this fact is that
all token-and-ex-use conceptions of words, including (U), imply
(P) Sentences that express our PJSSs are factual only if they have truth
values that are determined by facts about the ex-uses of the word
tokens they contain.
Since we each tend to theorize about words in a way that supports some
token-and-ex-use conception of words or other, such as (U), we are each
tacitly committed to (P), and this explains why we feel so sure that the
Tarski–Quine thesis and the intersubjectivity constraint are incompatible.
We see that a Tarski-style definition of truth for one’s own sentences
does not settle the question whether truth values of sentences that express
our PJSSs are determined by facts about the ex-uses of the word tokens
they contain. From this observation and our tacit commitment to (P), we
infer that a Tarski-style definition of truth for one’s own sentences leaves
open the possibility that our PJSSs are not both factual and trustworthy,
and hence that (C) is true—there is no way of explaining why it is
epistemically reasonable for one to regard one’s PJSSs as both factual and
114     

trustworthy without claiming that there is more to truth than what is

captured by a Tarski-style disquotational truth predicate defined for one’s
own regimented sentences.⁹

4.6. Types and Tokens

My first step towards developing an alternative to the token-and-ex-use
conception of words is to show that the orthographic conception of word
types is not as straightforward as it initially appears. Here, again, my
approach is pragmatic. I ask, ‘‘Do we need a type–token distinction,’’ and
‘‘If so, what sort of type–token distinction do we need?’’
To begin with, recall that according to the orthographic conception of
word types, the ink marks ‘Chicago, Chicago’, for example, instantiate two
tokens of the same word type C-h-i-c-a-g-o, and each token of the type
C-h-i-c-a-g-o is a string of tokens of letter types. Let us focus first on the
relationship between letter types and their tokens. We may suppose that all
letter tokens are concrete particulars, or what I shall call marks—inscriptions
in ink, pencil, chalk, grooves in wood, stone, or plastic, electronic events
on a computer screen, utterances, such as ‘‘Tee,’’ ‘‘Double-yu,’’ and so
on.¹⁰ But what are letter types? It is tempting to think of them as abstract
patterns or forms. If we think of letter types in this way, we will think that
what counts as a token of a letter type is somehow settled by the type’s
abstract pattern or form. I shall call this the naive theory of letter types.
There are two related problems with the naive theory of letter types.
First, it is unclear how any object—abstract or not—could settle whether
or not a given mark is a token of some specified letter type. It is not enough
to say that the pattern or form that (for now) we assume to be identical

⁹ In the philosophical literature today there are two entrenched but opposing attitudes towards
sentences like (6), both of which presuppose (P). The first attitude is exemplified by Quine’s
indeterminacy thesis, according to which sentences like (4), and any sentences that contain them,
such as (6), have no determinate truth values, and hence cannot be regarded as factual. If we accept
(H) of §3.9, and Quine’s indeterminacy thesis, then we will not regard such sentences as among our
regimented declarative sentences. The opposing view, held by a majority of philosophers today, is that
sentences that express our PJSSs have truth values that are settled by the truth values of other sentences,
including sentences that describe how the relevant speakers use their words. If we accept (H) and this
majority view, we will regard sentences like (6) as declarative sentences of our regimented language,
and hence as factual.
¹⁰ I take this use of the term ‘‘mark’’ from Goodman 1976: 131.
      115

with the letter type is similar to the mark. Any two marks are similar in
some respect or other. One might hope to avoid this problem by appealing
to similarity in the relevant respects. But how could a pattern or form settle
which respects are relevant? The idea of an abstract form or pattern that
somehow by itself singles out its tokens is puzzling, at best.¹¹
Second, even if we could solve the first problem, the naive theory of
letter types would not fit with our actual practice of classifying marks. A
careful look at how we actually identify marks as letters suggests that there
is no single, context-independent pattern or form that settles when a given
mark counts as a token of some given letter type, such as A. Is ‘a’ a token of
the letter type A? That partly depends on whether we want to take the letter
type A to encompass both upper and lower case tokens. Even when we
distinguish between upper and lower case letter types, however, questions
remain. Are the marks a, a, and a, for instance, all tokens of the lower case
letter type a? One might hope to make a list of all context-independent
patterns instances of which count as tokens of the upper or lower case letter
types A and a. But we have no reason to think that such a list could ever
be completed. We ordinarily have no difficulty identifying handwritten
tokens of the letter types A or a, for instance, yet each person writes letter
tokens in a slightly different way. As Nelson Goodman points out, ‘‘it may
happen that one of two marks that looks in isolation more like an ‘a’ may
count as a ‘d’ while the one that looks more like a ‘d’ counts as an ‘a’ ’’
(Goodman 1976: 138). Goodman illustrates this possibility with tokens of
‘‘bad’’ and ‘‘man’’ written roughly as follows:

(Goodman 1976: 138).¹² No context-independent criterion for being a

letter token could match our actual classifications of letter tokens into
types.

¹¹ These doubts about the naive theory of types are similar to doubts that Ludwig Wittgenstein
raises about the idea that a picture that comes to mind when we use a word somehow guides our
application of the word. See Wittgenstein 2001: §§138–42. For a more dogmatic dismissal of the role
that similarity is supposed to play in the naive theory of types, see Goodman 1972.
¹² The same sort of problem arises for a naive theory of phones (i.e. those aspects of an acoustic
stream produced by a given speaker that count in context as the speaker’s utterance of particular letters).
116     

Despite these problems with the naive theory of letter types, it is difficult
to imagine how we could communicate in writing or speech if we could
not in practice distinguish between types and tokens. Our practical ability
to distinguish between letter types and letter tokens is indispensable to
our system of writing, which depends on our recognition of a fixed
alphabet, tokens of which can be combined into words that themselves
can be classified into types. The type–token distinction is therefore worth
preserving, despite the problems just described.
A first step towards preserving and clarifying the distinction is to explicate
types as classes of all and only their tokens.¹³ Consider the letter type A,
where this is understood to encompass all As, whether upper or lower case,
italicized, large, small, handwritten, electronic, and so on. We may identify
the letter type A with the set {x: x is of the same letter type as the mark
a}, where the mark in boldface is used as a sample token of the letter type
A. When we explicate a letter type in this way, we suppose that the phrase
‘is of the same letter type as’ expresses an equivalence (reflexive, symmetric,
and transitive) relation. By identifying the letter type A with the set {x: x
is of the same letter type as the mark a}, we do not state explicit criteria
for membership in the set, so we are still left with the problem of clarifying
what it is for a mark x to be of the same letter type as the mark a. Clearly,
we cannot take ‘{x: x is of the same letter type as the mark a}’ to signify the
set {x: x is of the same letter type as the mark a}, reading ‘x is of the same
letter type as the mark a’ as ‘for some letter type y, x is a member of y and
a is a member of y’, on pain of circularity. Still, we are now in a position
to see how to say what a letter type is without committing ourselves to the
existence of a pattern or form that somehow by itself settles what counts
as one of its tokens: we must view ‘x is of the same letter type as the mark

Linguists noticed long ago that among American English-speakers, in any two utterances of ‘writer’ and
‘rider’, for instance, the part of an acoustic stream that counts in context as an utterance of the t in ‘writer’
may be indistinguishable when heard out of context from the part of an acoustic stream that counts in
context as an utterance of the d in ‘rider’ (Rey 2006: 247). If we suppose that phone types are context-
independent sound patterns, and we see no alternative to that supposition, then, like Rey, we will take
the data to imply that no phone tokens are produced in ordinary speech. For reasons I’ll explain soon,
however, there is an alternative to the supposition that phone types are context-independent sound
patterns. I propose that we continue to believe that phone tokens are produced in ordinary speech, and
take the data to undermine the supposition that phone types are context-independent sound patterns.
¹³ In Carnap 1942: §3, Rudolf Carnap distinguishes between sign-designs and sign-events, and
proposes that we view sign-designs as classes of sign-events. He proposes that we group the sign-events
together into classes by their design. As I explained above, however, for many of our purposes, it is not
helpful to classify sign-events by their design.
      117

a’ as of the form ‘Rxa’, where R is logically simple. It would help to find

some way of replacing the phrase ‘is of the same letter type as’ by another
expression that does not contain the words ‘letter type’ at all. Following
Nelson Goodman, I propose that we replace the phrase ‘is of the same
letter type as’ by ‘is a replica of ’, and stipulate that a replica is what we
misleadingly call a token, and ‘is a replica of ’ expresses an equivalence
relation between replicas.¹⁴ With this new terminology, we can rewrite ‘{x:
x is of the same letter type as the mark a}’ as ‘{x: x is a replica of the mark
a}’ and identify the letter type A with {x: x is a replica of the mark a}.¹⁵
We introduce sets such as {x: x is a replica of the mark a} to solve or
dissolve the problems with the naive theory of types, not to provide rules
for deciding when two marks are replicas of each other. Nevertheless, we
must now face the question ‘‘What counts as a replica of a given mark?’’
We cannot answer this question by examining how speakers use the phrase
‘is a replica of ’, since few speakers, if any, use that phrase at all. Moreover,
if we wish to use the phrase ‘is a replica of ’ to specify sets that explicate
(replace) the naive theory of types, then, for reasons noted above, we
should not equate ‘is a replica of ’ with ‘is physically similar in shape to’
(as Hugly and Sayward (1981) suggest). If we wish to use ‘is a replica
of ’ to solve or dissolve the problems with the naive theory of types, we
need a different approach. I propose that we look at our actual practices of
distinguishing between and classifying marks for various purposes.
Still sticking to letters, then, let us consider how we might elucidate
such sentences as ‘a is a replica of a’. We will not count the first of these
boldface marks as a replica of the second if we are sorting them by font,
but we will count the first as a replica of the second if we are sorting
them by letter type. Hence we can rely to some degree on our established
practice of distinguishing between letter types to elucidate the relation that
is expressed by a particular use of ‘is a replica of ’ in such phrases as ‘{x: x
is a replica of the mark a}’. The alternative of classifying the two marks by
font type shows that the phrase ‘is a replica of ’ is contextually ambiguous
in the sense explained in §1.2; we need to introduce regimentations of

¹⁴ See Goodman 1966: 361–2 and 1976: 138–43. The term ‘‘replica’’, but not Goodman’s
nominalistic use of it, is due to Charles Sanders Peirce, who writes, ‘‘Every legisign signifies through
an instance of its application, which may be termed a Replica of it’’ (Peirce 1933: ii. 143).
¹⁵ Since ‘‘x is a replica of y’’ is reflexive, the boldface mark in ‘‘{x: x is a replica of the mark a}’’ is
a replica of itself, and is therefore a member of the set that is the letter type A. See Goodman 1966:
361–2.
118     

the phrase before we can use such phrases as ‘{x: x is a replica of the
mark a}’. In some cases we will be able to define the regimented relation
explicitly, and in other cases, we will not. As Nelson Goodman observes, for
instance, ‘‘The letter-classes of our alphabet . . . are established by tradition
and habit; . . . defining them would be as hard as defining such ordinary
terms as ‘desk’ and ‘table’ ’’ (Goodman 1976: 138). Hence when we use
such phrases as ‘{x: x is a replica of the mark a}’ to refer to the sets that
explicate our ordinary notion of letter types, we can provide at most an
elucidation of the relation we express by using the phrase ‘is a replica of ’ on
that occasion; we cannot explicitly define that phrase.¹⁶
New replicas of a are produced each day. Hence if we suppose that a use
of ‘{x: x is a replica of the mark a}’ at a given time t specifies the set of all and
only the replicas of a that have been produced at t, then the set specified
by a use of ‘{x: x is a replica of the mark a}’ on one day may differ from the
set specified by a use of that expression on the next day. This consequence
does not fit with our pretheoretical view that letter types are timeless. We
should therefore reject the supposition that a use of ‘{x: x is a replica of the
mark a}’ at a given time t specifies the set of all and only the replicas of
a that have been produced at t. But we in effect already rejected it when
we decided (see §1.5) to take first-order variables to range over all objects,
past, present, and future. Hence just as the quantified variable in ‘∀x1 ((x1 is
mortal) ∨ ∼(x1 is mortal))’ ranges over all objects, past, present, and future,

¹⁶ Goodman notes a complication with this strategy: there may be ‘‘a mark that, equivocally, reads
as different letters when placed in different contexts at different times’’ (Goodman 1976: 139). For
instance, in a crossword-puzzle configuration of marks, a single, enduring, mark may count as a ‘‘d’’
when read horizontally from left to right, as part of a token of ‘‘bad’’, but as an ‘‘a’’ when read
vertically from top to bottom, as part of a token of ‘‘man’’. One could construct such an example by
rearranging the handwritten letters displayed on page 115 above, using just one token of the enduring
mark that ambiguously reads as an ‘‘a’’ or a ‘‘d’’ according as it is read in the left-to-right context or the
top-to-bottom context. To handle such cases, Goodman proposes that ‘‘not such enduring marks but,
rather, unequivocal time-slices of them must be taken as . . . inscriptions of letters’’ (Goodman 1976:
139). This proposal does not handle all unequivocal cases, however. Dilworth 2003 points out that
even a time-slice of a mark may count as a ‘‘d’’ for a person reading it from one point of view, and a ‘‘b’’
for a person simultaneously reading it from a different point of view. He imagines a transparent plastic
sign with the letters ‘‘din’’ painted on it; a person reading this sign from one point of view will read the
marks as spelling the word ‘‘din’’, a person simultaneously reading it from the other side will read the
marks as spelling the word ‘‘nib’’. Dilworth takes this to undermine Goodman’s strategy of explicating
letter types as sets of equivalence classes of particulars. But we could amend Goodman’s account to
accommodate Dilworth’s transparent sign by taking inscriptions of letters to be ordered pairs of time-slices
of marks and space-time points, where the points represent spatio-temporal locations of persons reading
or writing those time-slices of marks. This is highly artificial, but that by itself is no objection, given
that our aim is explication (see §2.6), not conceptual analysis.
      119

and hence each use of ‘{x: x is mortal}’ refers to the same set—the set of all
and only the past, present, or future mortal things, so the variable in ‘{x: x
is a replica of the mark a}’ ranges over all objects, past, present, and future,
and hence each use of ‘{x: x is a replica of the mark a}’ refers to the same
set—the set of all and only the past, present, or future replicas of the mark a.
Similar observations hold for our explications of word types. Here we
may draw on the fact that a single orthographic type may have different
satisfaction conditions. For instance, the British word ‘robin’ and the
American word ‘robin’ are spelled in the same way, but a bird that satisfies
the British word ‘robin’ does not satisfy the American word ‘robin’, and vice
versa—British robins and American robins are birds of different species.
Now suppose that in the following sentence

(∗ ) A robin is not a robin

the first boldface mark robin is a token of the British word ‘robin’ and
the second boldfaced mark robin is a token of the American word ‘robin’.
Then (∗ ) is true, but we will not see this if we take both boldface marks to
be tokens of either the British word ‘robin’ or the American word ‘robin’.
For some purposes, such as copy editing, we group the first boldface mark
in (∗ ) with the second, but for others, such as applying a truth predicate to
another’s words, we do not. Corresponding with these different purposes
are different ‘‘is a replica of ’’ relations. For purposes of copy editing, we
may be able to define the relevant replica relation orthographically (or
phonetically), in terms of strings of classes of letter replicas. But some of
the word types that matter to us cannot be defined in this way. When we
explicate these word types, we can reasonably require at most an elucidation
of the phrase ‘is a replica of ’ that we use in such terms as ‘{x: x is a replica
of the mark robin}’ to refer to the relevant equivalence classes of marks.
The method just sketched requires that letter and word classes be classes
of marks, defined in relation to particular letter or word tokens (marks)
that occur in the definitions themselves. Hence it cannot be applied to
define classes of strings of letters or words, including sentences, that no one
has yet identified, displayed, or used.¹⁷ Yet in logic we are committed to
the existence of strings of letters or words that no one has yet identified,

¹⁷ In Wetzel 2006: §4.1, Linda Wetzel calls this one of the two main problems with explicating types
as sets. She does not mention Quine’s proposed solution to it.
120     

displayed, or used. As Quine points out, for instance, ‘‘fruitful work in the
mathematical theory of proof . . . hinges on the existence and distinctiveness
of strings of signs of all finite lengths’’ (Quine 1987: 217; see also Quine
1960: 194–5). We therefore need a way of extending our explication of
letter and word classes to encompass strings of letter and word classes of
all finite lengths.¹⁸ For this reason, following Quine, I propose that we
explicate strings of letter and word classes as finite sequences of letter and
word classes (Quine 1987: 218).
One may choose the method just sketched to explicate the type–token
distinction while simultaneously affirming a token-and-explanatory-use
conception of word types, such as (U) of §4.5, and thereby committing
oneself to accepting (P) of §4.5, where ‘token’ is understood as ‘mark’ and
‘type’ as an equivalence class of marks defined by some regimented relation
expressed by a particular use of the phrase ‘is a replica of ’. Nevertheless, for
reasons I shall explain in the next three sections, the method just sketched
also enables one to explicate the word types relevant to satisfaction and
truth in a way that does not commit one to accepting (P).

4.7. Kaplan’s Common Currency Conception

of Words
In a seminal paper titled ‘‘Words’’ (Kaplan 1990), David Kaplan articulates
a conception of words that can help us to take another step towards
developing an alternative to the token-and-explanatory-use model of
words. To begin with, he highlights aspects of our actual practice of
individuating words that should lead us to reject (U), the token-and-
explanatory-use conception of words that I specified above:
(U) Two word tokens t and t are of the same word type if and only if
(i) t and t are each tokens of the same orthographic or phonetic
type, and

¹⁸ Georges Rey claims that standard linguistic entities (SLEs), such as word tokens, sentence tokens,
and phone tokens, are unlike abstract objects, such as numbers, because ‘‘our best theories of the
nonpsychological world seem to be committed to numbers,’’ whereas in linguistic theorizing, SLEs ‘‘have
no role to play independently of our representations of them’’ (Rey 2006: 250). What Rey apparently
overlooks is that our best theories of some non-psychological topics (e.g. mathematic proofs) are
committed to SLEs.
      121

(ii) the explanatory uses of t and t determine that they each have
semantic values and that their semantic values are the same.
Although Kaplan does not explicitly state (U), he in effect challenges
condition (i) of (U) by elucidating what he calls the common currency
conception of words, which is rooted in our ordinary criteria for judging
when a speaker has repeated a word. To highlight these ordinary criteria
he describes the following thought experiment: ‘‘I say the name of an
individual, possibly a name known to the person to whom I’m speaking.
The subject is to wait for a count of five, and then repeat the name . . .’’
(Kaplan 1990: 102). Kaplan remarks that in these circumstances, ‘‘we
are very strongly inclined to say that when this person speaks, he is
repeating the very name that he heard. . . . it’s clear that we would agree
to describe his output as a repetition of that name’’ (Kaplan 1990: 102–3).
He emphasizes that if repetition of the sort he describes is central to
our understanding of the same-word relation, then a speaker’s output
could be a repetition of a word even if it sounds very different from
the input she heard: ‘‘In view of the fact that individual differences
in psychological processing . . . may affect the resemblance of output to
input, . . . we [can] get differences in sound just about as great as we would
like between what comes in and what goes out’’ (Kaplan 1990: 105).
Similar thought experiments support the conclusion that one could get
differences in spelling between what comes in and what goes out. Such
thought experiments therefore conflict with clause (i) of (U).¹⁹ Kaplan
concludes that
The identification of a word uttered or inscribed with one heard or read is
not a matter of resemblance between the two physical embodiments (the two
utterances, the two inscriptions, or the one utterance and one inscription). Rather
it is a matter of [intrapersonal and] interpersonal continuity, a matter of intention:
Was it repetition? We depend heavily on resemblance between utterances and
inscriptions . . . in order to devine these critical intentions. If it sounds like ‘‘duck’’,
it probably is ‘‘duck’’. But we also take account of accent and idiolect and all

¹⁹ Linda Wetzel emphasizes that ‘‘there is an important and very common use of the word ‘word’
that lexicographers and the rest of us use frequently. It is, roughly, the sort of thing that merits a
dictionary entry’’ (Wetzel 2006: §4.2.1). She concludes that ‘‘the final nail in the coffin for the suggestion
[that] all tokens of the same word have the same sound is that words can be mispronounced’’
(Wetzel 2006: §4.2.4); in support of this conclusion she cites David Kaplan’s word-repetition thought
experiment.
122     

the usual clues to intention. It is the latter that decides the matter. (Kaplan
1990: 104)

In this last sentence, I take ‘the latter’ to be short not for ‘intention’, but for
‘accent and idiolect and all the usual clues to intention’, and infer that word
repetition on Kaplan’s understanding is not settled by a speaker’s intention
unless the intention is of the appropriate sort—the sort that others can
identify by taking account of ‘the usual clues to intention’. I shall return to
this point below.
To accommodate his observations about word repetition, Kaplan pro-
poses that we think of words not as abstract types that classify their
tokens by their resemblance to each other in certain respects, but as
‘‘trees . . . stemming out from their creations, with physical and mental
segments . . .’’ (Kaplan 1990: 117). These are what I shall call repetition trees.
Kaplan’s fundamental idea is that common currency words are repetition trees
generated by actual, not merely possible, acts of repeating words, as these acts are
understood and elucidated by our ordinary criteria for word repetition. Our ordin-
ary criteria for word repetition allow for the possibility of mistakes: we
distinguish between our belief that x is a repetition of y, on the one hand,
and x’s being a repetition of y, on the other. Hence Kaplan’s fundamental
idea is that common currency words are repetition trees generated from an
original mark or marks by chains of actual word repetitions, understood as
logically independent of our beliefs.
Kaplan suggests that if words are not abstract types that classify their
tokens by their resemblance to each other in certain respects, then they are
not types at all, but ‘‘natural objects,’’ or ‘‘continuants’’—particular repetition
trees whose nodes are utterances or inscriptions (marks) (Kaplan 1990: 98,
116–17). This is, at best, misleading. For if we are willing to accept certain
constraints on what counts as a repetition tree, it is not difficult to explicate
word types in a way that fits with Kaplan’s observations about our ordinary
criteria for word repetition. To do so, we can apply the strategy for
explicating types that I introduced above (§4.6). Starting with the relation
expressed by ‘is a repetition of ’, we can explicate a common currency
word as a certain sort of equivalence class of marks (word replicas). We are
assuming that the relation expressed by ‘is a repetition of ’ is asymmetric,
irreflexive, and intransitive, so we cannot use this phrase directly to define
an equivalence class of word replicas. But if we stipulate that as we shall
      123

use the term ‘repetition tree’, every word replica is a node of one and only
one repetition tree, it follows that ‘for some repetition tree z, x is a node
of z and y is a node of z’ is an equivalence relation.²⁰ We can then identify
the common currency word ‘robin’, for instance, with {x: x and robin are
nodes of the same repetition tree}. If we introduce a new phrase ‘x is a
repetition replica of y’, defined as follows:
x is a repetition replica of y if and only if for some repetition tree z, x
is a node of z and y is a node of z
we can rewrite ‘{x: for some repetition tree z, x is a node of z and robin is
a node of z}’ as ‘{x: x is a repetition replica of robin}’ and identify {x: for
some repetition tree z, x is a node of z and robin is a node of z} with {x: x
is a repetition replica of robin}.
To elucidate ‘{x: for some repetition tree z, x is a node of z and robin is
a node of z}’ as ‘{x: x is a repetition replica of robin}’ we must elucidate
the phrase ‘is a repetition replica of ’ by investigating our actual context-
sensitive practices of identifying repetitions of words.²¹ We can survey
some of the issues that we would have to settle in order to elucidate ‘is a
repetition replica of ’ by considering two objections to Kaplan’s conception
of words. Both objections focus on a consequence of my explication of
Kaplan’s proposal—the consequence that an object is a word replica if and
only if it is produced with an appropriate kind of intention.

²⁰ This stipulation may seem too artificial, since our ordinary criteria for word repetition might
sometimes seem to commit us to count a single mark as a member of two different words. But we
should judge the stipulation by its fruits: the question is whether it yields a satisfactory explication of
words as equivalence classes of marks, not whether it captures all our pre-theoretical judgements about
when a person repeats a given word.
²¹ If we explicate Kaplan’s common currency conception of words in the way I propose, it is a
type–token conception of words, where ‘token’ is understood as ‘mark’ and ‘type’ as an equivalence
class of ‘marks’ defined by some regimented relation of the form ‘x is a replica of y’. Does the
explication then undermine Kaplan’s claim to be offering an alternative to what he calls the type–token
model of words? That depends on how one understands the main target of his criticisms. One might
take Kaplan to be trying to discredit every conception of words that could plausibly be called a
type–token conception. Gregory McCulloch apparently reads Kaplan in this way, and emphasizes that
‘‘Kaplan’s point is that token word-occurrences are bound together into word types by criteria which
are neither geometrical nor phonological [an observation that] . . . leaves the type/token vocabulary,
apparently without strain, in place . . .’’ (McCulloch 1991: 74–5). McCulloch does not actually provide
an explication of Kaplan’s conception that shows it to be a type–token view, but I assume that he
would accept my explication, and see it as a vindication of his criticism of Kaplan. Perhaps Kaplan
was confused about this point, but in any case, as I shall try to show at the end of this section, his
common currency conception of words is incompatible with the token-and-explanatory-use model of
words.
124     

Consider first the ‘‘only if ’’ part of the claim. One might think that
there can be word replicas that are not produced with any intention
at all. Suppose, for instance, that you and I see ripples in a sand dune
caused by the wind, and that the ripples seem to spell out the English
sentence, ‘‘We are lost.’’ I point to the ripples and say to you, ‘‘I believe
that,’’ and you take me to have thereby expressed my belief that we
are lost. One might think that in this context the ripples in the sand
count as words of English even before I or anyone else even notices
their similarity with words of English. One might therefore take the
example to undermine the consequence that it is necessary for an object
to be a word that the object be produced with an appropriate kind of
intention.²²
There are two main problems with this first objection. First, even if it
were generally accepted and uncontroversial that the ripples in the dunes
were words of English before I or anyone else noticed, the objection
wrongly presupposes that the goal of Kaplan’s (or any other) explication of
words is to capture all generally accepted and uncontroversial beliefs about
words. The objection overlooks the possibility that to explicate words in
a way that we find fruitful, we may have to accept some counter-intuitive
consequences. Second, one might agree with Peter Alward that ‘‘in our
speech acts, we do not use preexisting tokens; we produce new ones’’
(Alward 2005: 183). On this view, when I say ‘‘I believe that,’’ gesturing
at the ripples in the dunes, I thereby intentionally produce contemporary
tokens of the English words ‘We are lost’ by my gesture towards the
ripples that only seem to spell out those words; it is the new tokens
I thereby produce, not the ripples themselves, that count as nodes of
repetition trees.
A second objection to Kaplan’s conception of words begins with the
observation that according to it, an object is a word replica—a node of
a repetition tree—if it is produced with an appropriate kind of intention.
One might think that if we were to accept this consequence, we would be
committed to the conclusion that a speaker could count as having uttered
a word even if no other speaker could identify what word she uttered.
This conclusion does not fit with our public, intersubjective practice of
identifying words. One might therefore object that contrary to Kaplan’s

²² For a similar objection that I used as a model for this one, see Cappelen 1999: 95.
      125

proposal, it is not sufficient for a mark to be a word replica that it be

produced with an appropriate kind of intention.²³
The problem with this objection is that it rests on a dubious and
uncharitable understanding of the role of intentions in Kaplan’s account. As
I noted above, word repetition on Kaplan’s understanding is not settled by
a speaker’s intention unless the intention is of the appropriate sort—the sort
that others can identify by taking account of ‘‘the usual clues to intention’’.
More important, we have no grip on the idea that a speaker’s output is the
repetition of a word apart from our ordinary public criteria for taking it to
be the repetition of a word, criteria that are bound up with our ordinary
public criteria for identifying a speaker’s intentions to use the same words
that other speakers use. Hence we must rely solely on these ordinary public
criteria if we are to make sense of Kaplan’s proposal.
As I see it, the underlying target of Kaplan’s criticisms is the token-and-
explanatory-use model of words. His emphasis on word repetition discredits
(i) of (U). For instance, suppose that in one of Kaplan’s experimental
situations, I say, ‘‘dolphin,’’ and a test subject, Alice, is supposed to
count to five and then repeat the word I uttered. If Alice follows the
instructions, and says something that sounds like DULL-FIN, she will count
as having repeated the word I uttered, even if I pronounced it DAWL-FIN
not DULL-FIN. Thus her performance counts as a repetition despite a
significant difference in pronunciation. This discredits the phonetic part of
(i). Similarly, suppose I write a token of the English word ‘dolphin’ and ask
Alice to write another token of the same word. If Alice, intending to follow
my instructions, writes a token of d-o-l-f-i-n, she will nevertheless by our
ordinary standards count as having rewritten (but misspelled) the English
word ‘dolphin’. This application of the common currency conception of
words discredits the orthographic part of (i).
A proponent of (U) could try to accommodate this criticism by changing
clause (i) to allow for alternative spellings and pronunciations. To remain in
the spirit of (U), a clause that replaces (i) must include a fixed and finite list of
admissible spellings or pronunciations for each word type. But for any such
list, one could likely generate a new alternative spelling or pronunciation
that is acceptable by ordinary criteria, but is not on the list. Anticipating
this, a proponent of (U) might be inclined to dismiss Kaplan’s criteria

²³ For a similar objection, see Cappelen 1999: 94–5.

126     

for word repetition. But Kaplan’s criteria for word repetition are just our
own ordinary criteria for word repetition. We cannot simply dismiss these
criteria. Hence his common currency conception convincingly discredits
both (U) and any modifications of (U) whose first clauses include a fixed
and finite list of admissible spellings or pronunciations for each word type.
One might still think one can still salvage the token-and-ex-use model of
words if one builds Kaplan’s common currency conception of words into
a token-and-explanatory-use conception of words, as follows:
(U ) Two word tokens (marks) t and t are of the same word type (in
the sense relevant to truth and logic) if and only if
(i) t and t are members of the same common currency word (t
and t are nodes of the same repetition tree), and
(ii) facts about the ex-uses of t and t determine that they each have
semantic values and that their semantic values are the same.
To make sense of (U ), however, we must suppose that the question
whether two word tokens (marks) t and t are members of the same
common currency word is independent of the questions whether they each
have semantic values and, if so, whether their semantic values are the same.
But Kaplan seems to understand word repetition in such a way that we
cannot settle whether two word tokens (marks) t and t are members of the
same common currency word without thereby also settling that they each
have semantic values and, if so, whether their semantic values are the same.
Kaplan emphasizes, for instance, that two marks count as the same common
currency name only if they name the same individual.²⁴ This treatment of
names suggests that he would also endorse the more general conclusion that
two marks count as the same common currency word only if they each have
semantic values and their semantic values are the same.²⁵ If we understand
Kaplan’s criteria for word repetition in this way, we cannot accept any
token-and-explanatory-use conception of words, for all of them require
that we be able in principle to identify word tokens prior to and independently

²⁴ See Kaplan 1990: 110–11, where Kaplan emphasizes that his common currency name ‘‘David’’ is
different from David Israel’s common currency name ‘‘David’’.
²⁵ This interpretation of Kaplan’s common currency conception of words fits with what Kaplan
elsewhere calls consumerist semantics, according to which ‘‘we are, for the most part, language consumers.
And words come to us prepackaged with a semantic value. If we are to use those words, the words we
have received, the words of our linguistic community, then we must defer to their meaning’’ (Kaplan
1989b: 602).
      127

of settling that they do or do not each have semantic values and, if they do,
whether their semantic values are the same. Hence Kaplan’s remarks about
how we keep track of words suggest that his common currency conception
of words is incompatible with any token-and-ex-use conception of words.

4.8. The Context Principle and the PJSS-Based

Conception of Words
From the pragmatic perspective that motivates the intersubjecivity con-
straint, the central problem with Kaplan’s common currency conception
of words is that it does not actually keep track of the word types relevant
to generalizing on other speakers’ sentences. If Kaplan’s test subject does
not know English and is asked to repeat an English word, she will repeat
that word without having any idea what one might use it to say, or even
whether it is a predicate or a logical constant, and hence without being
in a position to make a PJSS for it.²⁶ But, as I argued above, if we are to
generalize on other speakers’ sentences, we need to rely on our PJSSs for
their words. The problem is that Kaplan’s common currency conception of
words does not license us to apply our own disquotational truth predicates
to other speakers’ sentences.
One might try to avoid this problem by restricting Kaplan’s repetition
test to subjects who know, or are competent in the use of, the natural
language of which the test word is a part.²⁷ But if one accepts the common
currency conception of words, then to regard a speaker as competent in
the use of a natural language is just to count some of her utterances as
repetitions of words of the language. Hence the proposed restriction is
circular. Even if we were to adopt the restriction, however, we would still
have a problem with ambiguity. If you ask me to repeat the English word

²⁶ Adèle Mercier also objects to Kaplan’s conception of words. She claims that ‘‘you have not
repeated my word, indeed that you cannot be said to have repeated any word, unless you can make
a mental commitment as to the sort of word you are repeating (proper name, common noun, verb,
and the like). For without this committing yourself syntactically, you will not be able to use the word
productively yourself, to generate sentences using it’’ (Mercier 1999: 93; see also Cappelen 1999: 97).
Mercier’s objection is similar in spirit to the one I raise in the text, but there is a fundamental difference
between them: my objection is not that Kaplan’s conception of words fails to fit with our intuitions
about what it is to repeat a word, but that it yields a conception of words that does not license our
PJSSs, and hence does not show us how to satisfy the intersubjectivity constraint.
²⁷ This is how Peter Alward proposes that we deal with the problem. See Alward 2005: 183.
128     

‘‘bank’’, for instance, I will do so without being able to make a PJSS for
it. Hence the circular restriction would still leave us with the problem that
Kaplan’s common currency conception of words does not license us to
apply our own disquotational truth predicates to other speakers’ sentences.
To avoid the problem yet still preserve Kaplan’s insight that our criteria
for sameness and difference of words are at root public and practical, we
need to elucidate practical criteria for sameness of word types that keep track
of the word types relevant to generalizing on other speakers’ sentences,
and hence relevant to our pragmatic explication of truth and satisfaction.
We actually keep track of the relevant word types by relying on our PJSSs.
When we make such practical judgements, we take ourselves to be using,
hence in some sense repeating, words that we read or hear, and so we are
relying on our ordinary criteria for word repetition. But we are doing so
in a context that disciplines our understanding of the word types that we
repeat. To conceive of word types in a way that is relevant to satisfaction
and truth, we need to pay careful attention to such contexts. I therefore
propose that we conceive of words in a way that obeys what I call
The context principle ( first version): never ask for the word type of a word token
in isolation, but only in the context of one’s PIWs and PJSSs for that word
token.
This principle is meant to refine Kaplan’s insight that we have no grip
on sameness and difference of words apart from our practical criteria
for identifying and distinguishing between words. The idea behind the
refinement is that my PJSS for Alice’s word ‘robin’, for instance, is
simultaneously (i) licensed by my PIW that Alice’s word token ‘robin’ is
of the same type as my word token ‘robin’ and (ii) a constraint on my
understanding of the word type of which I take Alice’s word token ‘robin’
and my word token ‘robin’ to be tokens. Hence to obey the context
principle is to conceive of words in such a way that we accept a sentence
that expresses a PIW, such as
(1 ) Alice’s word ‘robin’ is the same as my word ‘robin’,
if and only if we take it to license a corresponding PJSS.
Against this, one might be inclined to reason as follows: ‘‘If I have no
understanding of PIWs expressed by sentences like (1 ) independent of
the PJSSs that I take them to license, then if I want to know whether to
      129

accept (1 ), I am caught in a vicious circle: to obey the context principle, I

need to know whether to make a PJSS for Alice’s word ‘robin’, but I will
make a PJSS for it only if I already accept (1 ).’’ This reasoning overlooks
the fact that both PIWs and PJSSs are non-deliberative—we make them
independent of any conscious decisions about whether to make them. We
can subsequently revise them, should we see reason to do so. But to get
started, one should ask not whether to make a PIW and corresponding PJSS
for Alice’s word ‘robin’, but whether one actually does make them. Given
an actual record of PIWs and PJSSs, the context principle constrains our
understanding of word types. To obey the context principle is to seek a
reflective equilibrium between our understanding of word types and our
PIWs and PJSSs—an account of word types that fits with all and only the
PIWs and PJSSs we see no reason to reject.²⁸
To keep track of the word types relevant to truth and satisfaction,
however, it is not enough to obey the first version of the context principle.
For our practical criteria for sameness and difference of word type may also
include other speakers’ PIWs and PJSSs. Suppose, for instance, that Alice
mumbles as she utters the sentence ‘There is a goldfish in the bird bath’
and, as a result, her utterance sounds to me like an utterance of ‘There is
a goldfinch in the bird bath.’ Then I will I take her to have said that there
is a goldfinch in the bird bath. I will take the fourth word of the sentence
she mumbled to be of the same type as my word ‘goldfinch’, and thereby
conclude that like my word ‘goldfinch’, the fourth word of the sentence
she mumbled satisfies x if and only if x is a goldfinch. Now suppose that
after looking for a goldfinch in the bird bath, I say to Alice ‘‘Where is
the goldfinch? I don’t see it,’’ and she replies, with clearer diction, ‘‘I said
there is a goldfish in the bird bath!’’ I will then take Alice to have corrected
my PIW and PJSS for the mumbled word by repeating what she said more
clearly. In this case I rely not only on my own PIW and PJSS for the word
‘goldfish’ that Alice uses to correct me, but also, and crucially, on Alice’s
PIWs (and corresponding PJSSs, if she makes any) for the word I use in
my reply to her, and for the words in the sentence she mumbled.

²⁸ The reflective equilibrium I have in mind is modelled on Nelson Goodman’s classic account of
the justification of deductive inferences: ‘‘How do we justify a deduction? . . . A rule is amended if it yields
an inference we are unwilling to accept; an inference is rejected if it violates a rule we are unwilling to amend. The
process of justification is the delicate one of making mutual adjustments between rules and accepted
inferences; and in the agreement achieved lies the only justification needed for either’’ (Goodman
1983: 63–4).
130     

In some contexts we make a PIW solely because we accept that there is a

chain of PIWs made by other speakers that implies it. Recall, for instance,
that if I learn that, in 1820, John Audubon pointed to a bird and said, ‘‘That
is a robin,’’ I will take for granted that there is a chain of PIWs that can be
traced from my word ‘robin’ all the way back to Audubon’s unregimented
word ‘robin’. In these circumstances, I will make a PIW for his word
‘robin’. I cannot ask Audubon whether I’ve understood him correctly. But
I assume that English-speakers who interacted with Audubon in 1820 and
made PIWs for his word ‘robin’ were in a position to correct those PIWs if
they had any reason to doubt them, and that their tokens of that word are
related to mine by a chain of PIWs across time. Similarly, when we take
ourselves to have made a discovery about gold, for instance, we assume
there are chains of PIWs that links our current uses of ‘gold’ to uses of ‘gold’
long ago. In such cases, our non-deliberative (yet revisable) assumption that
such chains exist commits us to making PIWs (and corresponding PJSSs)
for words used by speakers long ago.
These cases illustrate that in practice we typically keep track of words
in part by relying on other speakers’ PIWs and PJSSs. And when we
rely on other speakers’ PIWs and PJSSs, we commit ourselves to making
PJSS-based judgements of sameness of truth value for other speakers’
sentences. To explicate truth in a way that captures this aspect of our
inquiries, we need to replace the first version of the context principle
with
The context principle (second version): never ask for the word type of a word
token in isolation, but only in the context of one’s own PIWs and PJSSs for
that word token and other speakers’ PIWs and PJSSs for that word token, if
such judgements exist.
I shall henceforth take this to be the context principle. Since there are other
English-speakers, and we rely on their PIWs (and corresponding PJSSs, if
they exist) for all of the word types we use, the context principle implies
that we have no conception of the word types we use apart from our
reliance on other English-speakers’ PIWs (and corresponding PJSSs, if they
exist) for those word types. The context principle does not by itself imply
that I do not use any word that no one else uses, but only that if I do use
such a word, then my own PIWs and PJSSs for it exhaust my criteria for
keeping track of it.
      131

As before, I propose that we regard word tokens as marks (word replicas)

and word types as equivalence classes of marks (word replicas). The context
principle then becomes: never ask for the equivalence-class of a mark (word replica)
in isolation, but only in the context of one’s own PIWs and PJSSs for that mark
(word replica) and other speakers’ PIWs and PJSSs for that mark (word replica), if
such judgements exist. When it is restated in this way, the context principle
amounts to a constraint on one’s explication of what I call PJSS-based word
types, which license our intersubjective applications of truth and logic. To
explicate Kaplan’s common currency word types, we used the phrase ‘is a
repetition replica of ’. To explicate PJSS-based word types, I propose that
we use the phrase ‘is a PJSS-based replica of ’.
From the pragmatic perspective that I have stressed, the question of
how to explicate our PJSS-based word types arises only after we have
each regimented our own words, so that we can distinguish between the
word replicas we ourselves utter or write by their spelling alone. After
regimentation, for instance, I can use the string of letters r-o-b-i-n to
produce and identify my own current replicas of my word ‘robin’. To
explicate PJSS-based word types as equivalence classes of replicas, however,
I must respect the context principle. I must therefore identify my PJSS-
based word type ‘robin’ with {x: x is a PJSS-based replica of robin},
where I identify the word replica in boldface by its spelling alone and I
understand the phrase ‘is a PJSS-based replica of ’ in accordance with the
second context principle.
But how does the context principle constrain my understanding of the
phrase ‘is a PJSS-based replica of ’? To address this question, let us first
stipulate that a statement of the form ‘x satisfies my PJSS-based word
type ’ is an abbreviation of ‘x satisfies every word replica in the set {x:
x is a PJSS-based replica of }’. For instance, ‘x satisfies my PJSS-based
word type robin’ is an abbreviation for ‘x satisfies every word replica in the
set {x: x is a PJSS-based-replica of robin}’. Now if I make a PJSS of the
form ‘x satisfies my PJSS-based word type robin if and only if x is a robin
and x satisfies A’s word type r if and only if x is a robin’, and I accept the
context principle, I thereby take my word type robin to be identical to A’s
word type r, and conclude that r is a PJSS-based replica of robin. I might
later revise the PJSS and regard it as false, but if I accept the context principle
I have no grip on whether or not a given mark is a PJSS-based replica of
another given mark apart from my PIWs and PJSSs for those marks.
132     

The central virtue of the context principle is that it respects the PIWs and
PJSSs on which we rely when we identify our agreements and disagreements
with others, take ourselves to learn from others, or make discoveries—PIWs
and PJSSs of the sort that I highlighted above (§§3.2–3.6) in support of
the second formulation of the intersubjectivity constraint. Consider again,
for instance, the first situation I described in Chapter 3: I am on a walk
with Alice, a bird lands on a branch in front of us, and Alice says, ‘‘That
is a robin,’’ pointing to the bird on the branch. Without hesitation or
reflection, I take her to have said that the bird is a robin, and, in so doing,
I take her word ‘robin’ to be the same as my word ‘robin’—the word
that I use to express what I take her to have said. In this sense of ‘‘same
word’’—the sense the I propose to explicate as an equivalence class of
word tokens—I accept
(1 ) Alice’s word ‘robin’ is the same as my word ‘robin’
Now suppose, again, that I also (already) accept
(4) x satisfies my word ‘robin’ if and only if x is a robin.
Then my unreflective acceptance of (1 ) will lead me to accept
(5) x satisfies Alice’s word ‘robin’ if and only if x is a robin.
I thereby commit myself to accepting the conjunction of (4) and (5), namely,
(6) (x satisfies my word ‘robin’ if and only if x is a robin) and (x satisfies
Alice’s word ‘robin’ if and only if x is a robin).
In this way, my unreflective commitment to (1 ) and my explicit commit-
ment to (4), together with the inference from (1 ) and (4) to (5), commit
me to (6), a PJSS. Let ‘r’ name the word token that Alice uttered and that
I take to be a token of my PJSS-based word type robin. I propose that we
explicate (6) as
(6 ) x satisfies my PJSS-based word type robin if and only if x is a robin
and x satisfies Alice’s PJSS-based word type r if and only if x is
a robin.
and (1 ) as
(1 ) Alice’s PJSS-based word type r is the same as my PJSS-based word
type robin.
      133

But (1 ) abbreviates

(1 ) {x: x is a PJSS-based-replica of Alice’s mark r} = {x: x is a
PJSS-based replica of my mark robin},
and (6 ) abbreviates
(6 ) For every word replica w in the set {x: x is a PJSS-based replica
of my mark robin}, x satisfies w if and only if x is a robin, and
for every word replica w in the set {x: x is a PJSS-based replica of
Alice’s mark r}, x satisfies w if and only if x is a robin.
If I accept the context principle, then when I accept (6 ) for the reasons
given above, I will regard Alice’s mark r to be a PJSS-based replica of my
mark robin. I will therefore accept (1 ). At the same time, I will take
my unreflective acceptance of (1 ) to license my acceptance of (6 ). To
accept the context principle is to have no understanding of sentences like
(1 ) independent of the sentences like (6 ) that one takes them to license.
In general, to obey the context principle is to eschew any criteria for
sameness and difference of words that can be specified without using any
sentences that express our PJSSs, and hence to view one’s understanding
of sameness and difference of words as inextricably linked to one’s PJSS-
based use of words. It is to adopt what I call the PJSS-based conception of
words.

4.9. How to Satisfy the Intersubjectivity Constraint

without Rejecting the Tarski–Quine Thesis
Let us now consider whether the Tarski–Quine thesis and intersubjectivity
constraint are compatible if we adopt the PJSS-based conception of words.
There are three crucial considerations. First, if we adopt the PJSS-based
conception, we will regard any criterion for sameness of word types that
violates the context principle as irrelevant to truth and satisfaction. Any
criterion for sameness of word types that implies
(P) Sentences that express our PJSSs are factual only if they have truth
values that are determined by facts about the ex-uses of the word
tokens they contain
134     

violates the context principle, because it presupposes that there are criteria
for sameness and difference of word type that do not presuppose the truth
of any sentences that express our PJSSs. Hence if we adopt the PJSS-based
conception of words, we will regard any criterion for sameness of word
types that implies (P) as irrelevant to truth and satisfaction. But, as we saw
in §4.5, all token-and-explanatory-use conceptions of words imply (P).
Hence if we adopt the PJSS-based conception of words, we will regard
all token-and-explanatory-use conceptions of words as irrelevant to truth
and satisfaction, and thereby discredit the reasoning (explained in §4.5)
that initially led us to think the Tarski–Quine thesis and intersubjectivity
constraint are incompatible.
Second, if we adopt the PJSS-based conception, then we can explain
why it is epistemically reasonable for us to regard our PJSSs as factual and
trustworthy. For this purpose, I propose that we place three constraints on
our understanding of ‘epistemically reasonable’ and related epistemic terms.
The first two of these constraints are
(T1) A person has a reason for believing that S only if she can say why
she believes that S without presupposing that S.²⁹
(T2) A person has an entitlement (or is entitled ) to believe that S if and
only if she has no reason (in the sense of ‘reason’ constrained by
(T1)) for believing that S —she cannot say why she believes that S
without presupposing S —but it is epistemically reasonable for her to
believe that S.
One might think that
(∗ ) For every statement S, it is epistemically reasonable for a person to
accept S only if she has a reason for believing that S,
which, together with (T1) and (T2), implies that there is no statement S
that anyone has an entitlement (or is entitled ) to believe. But there are some
statements, such as the statement that not every statement is both true and false,
that no one has any (independent) reason to accept, yet no one can make
sense of doubting, either. I assume it is epistemically reasonable to accept
such statements. But we have no reason (in the sense of ‘reason’ constrained

²⁹ Although we sometimes say that a person has a reason for believing that S even if all her best
attempts to explain why she believes S presuppose that S, I will not use ‘reason’ in this way.
      135

by (T1)) for believing them—we cannot say why we believe them without
presupposing them. By (T2), then, we are epistemically entitled to accept
these statements, including the statement that not every statement is both
true and false.
This argument for the existence of epistemic entitlements depends
heavily on the assumption that there are some statements we cannot make
sense of doubting. Let us try to make this assumption more precise. To
make sense of doubting a statement S, one must be able to specify a way
in which S may be false. For this it is not enough just to say ‘it is not the
case that’ before an uttered token of S, or to prefix a negation sign to a
written token of S. We are unable to specify how ‘Not every statement is
both true and false’ could be false, even though we are able to write and
utter the sentence ‘It is not the case that not every statement is both true
and false’. In short, to specify how a statement may actually be false, one
must do more than just negate it.³⁰
One might be tempted to conclude that for every statement S, if
a person cannot specify a way in which S may be false, then she is
epistemically entitled to accept S. But this goes too far. For if a person has
no understanding of a statement, then it trivially follows that she cannot
specify a way in which it may be false. And of course it is not epistemically
reasonable for a person to believe any statement of which she has no
understanding just on the grounds she is unable to specify a way in which
it may be false. The saving consideration is that we do not take ourselves or
others to believe or disbelieve a statement if we or they have no understanding
of it. (How much understanding is required for belief/disbelief depends
on the statement and the circumstances. I shall discuss these nuances in
Chapter 9.) If we have no understanding of a statement S, then the question
of whether our supposed belief in S is epistemically reasonable does not
arise, since we do not satisfy minimum conditions for believing or disbelieving
a statement if we have no understanding of it.
To accommodate these observations I propose the following additional
constraint:
(T3) If a person understands a statement S well enough to raise the
question of whether or not she believes S and she cannot specify

³⁰ If it is unclear whether a given speaker is able to specify a way in which her statement that S may
be false, then it will also be unclear whether ( T3) applies to her statement that S.
136     

a way in which a statement S may actually be false, then it is

epistemically reasonable for her to accept S.³¹
If we adopt (T1)–(T3), we may conclude that we are each epistemically
entitled (in a sense constrained by (T2)) to accept that not every statement
is both true and false, since we cannot provide a reason (in the sense of
‘reason’ constrained by (T1)) for accepting that not every statement is both
true and false, but by (T3) it is nevertheless epistemically reasonable for us
to accept that not every statement is both true and false, because we cannot
specify a way in which the statement that not every statement is both true and
false may actually be false.
One might worry that there will be circumstances in which (a) a person
P satisfies minimal conditions for believing or disbelieving a statement S,
(b) P has so little understanding of S that P cannot specify a way in which
S may be false, but (c) it is not epistemically reasonable for P to accept S.
For instance, one might be inclined to reason as follows: ‘‘Suppose that
without any proof, or other grounds, you simply claim (not having heard
of Goldbach’s conjecture) that every even number greater than 2 is the sum
of two primes. You know that a prime number is a whole number larger
than 1 that is divisible only by itself and 1. But you know too little to be
very specific about how it may actually be false. By (T3), apparently, it is
epistemically reasonable for you to accept the claim. But it is obviously not
epistemically reasonable for you to accept the claim, which was for you no
better than a guess. Hence (T3) is false.’’
This reasoning involves a misapplication of (T3). For even if you don’t
know much mathematics, if you understand what a prime number is, you
can specify a way in which that statement may be false. You understand
the statement well enough to see that it may be false if there is a number
k greater than 2 such that k is not the sum of two prime numbers. This is
not just a matter of negating the statement. For you can be more specific.
A quick Google search can lead you to a list of the known prime numbers
less than 10,000. You can then specify that the statement is false if 6,121,
for instance, is not the sum of two prime numbers. In this case, then, you
can specify a way in which the statement may actually be false. You might

³¹ This constraint is inspired by (but is not meant as an interpretation of ) Ludwig Wittgenstein’s

remark that ‘‘From its seeming to me—or to everyone—to be so, it doesn’t follow that it is so. What
we can ask is whether it can make sense to doubt it’’ (Wittgenstein 1969: §2).
      137

later discover that 6,121 is the sum of two primes. But unless you have
a proof of Goldbach’s conjecture, at any given time you will be able to
specify some prime number (or other) that at that time you don’t know to
be the sum of two primes. The moral is that in applying (T3), we must pay
careful attention to whether we can specify a way in which the statement
in question may be false. Although I do not have the space to list and
reply to them here, all the alleged counter-examples to (T3) that I have
encountered overlook some of the ways in which the subject may specify
how the claim in question may actually be false.
Even if (T3) is not undermined by alleged counter-examples like the
one just described, however, many will be inclined to reject it, reasoning
as follows. ‘‘It is epistemically reasonable for a person to accept a particular
statement only if she has epistemic grounds for accepting it. But a person’s
inability to specify a way in which a statement may actually be false gives
her no reason to think it is likely to be true, hence no epistemic grounds
for accepting it. Therefore, if the epistemic role of the statement for her
is exhausted by her inability to specify a way in which the statement
may actually be false, it is not epistemically reasonable for her to accept
it.’’ Those who find this reasoning compelling typically assume that if we
want to show that it is epistemically reasonable to accept some statement
S without evidence, we must try to explain how it is possible for a
person to have grounds for thinking S is likely to be true that are not
based on any evidence she could offer in support of S. Motivated by this
conclusion, some philosophers would argue that we can have such grounds
for accepting a given statement if the psychological processes that led us
to accept it reliably yield true beliefs (Rey 1998). Others hypothesize that
we have a capacity for ‘‘rational insight’’ that enables us to know directly,
without reasons, that a given statement is likely to be true (Bonjour 1998,
Katz 1998). Yet others argue that we are entitled to accept some statements
without providing any reasons for accepting them, because our acceptance
of them is constitutive of the meanings of the words we use to express
them (Boghossian 2000 and 2001, Peacocke 2000).
These positions are all problematic in well-known ways that we need
not rehearse here. For present purposes the crucial point is that these
problematic positions are motivated by the orthodox assumption that if we
want to show that it is epistemically reasonable to accept some statement S
without evidence, we must to try to explain how it is possible for a person
138     

to have grounds for thinking S is likely to be true that are not based on
any evidence she could offer in support of S. But I see no good reason to
accept this assumption. In my view, if one cannot specify a way in which
a statement S may be false, it is epistemically reasonable to accept S, even
if the fact that one cannot specify a way in which S may be false does not
imply that S is likely to be true. Hence, contrary to the orthodox positions
just sketched, I propose that we accept (T3) as a further constraint on our
understanding of ‘epistemically reasonable’.
These reflections bear directly on the question whether it is epistemically
reasonable for us to regard our PJSSs as both factual and trustworthy. For
if we adopt the PJSS-based conception of words, we will take ourselves
to have no grip on sameness of satisfaction apart from our PJSSs, and
hence we will be epistemically entitled to regard them as both factual
and trustworthy. To see why, let us first consider the question whether
we are epistemically entitled to regard our PJSSs as factual. If we adopt
the PJSS-based conception of words, we will have no conception of how
sentences that express our PJSSs could fail to be declarative sentences of a
regimented language for which we can define a Tarski-style disquotational
truth predicate. Hence we will be unable to specify a way in which the
statement
(f ) Sentences that express our PJSSs are declarative sentences of a
regimented language for which we can define a Tarski-style disquo-
tational truth predicate
could be false. By (T3), then, it is epistemically reasonable for us to
accept (f ). But to accept (f ) is to accept that sentences that express our
practical judgements of sameness of satisfaction are factual, in the sense of
‘factual’ defined by (H) above (§3.9). Hence it is epistemically reasonable
for us to regard sentences that express our PJSSs as factual. Since the
sentences that express our practical judgements of sameness of satisfaction
are factual if and only if our PJSSs are factual, it is epistemically reasonable
for us to regard our PJSSs as factual. Nevertheless, we have no reason
to regard our PJSSs as factual, in the sense of ‘reason’ constrained by
(T1). Hence, by (T2), we are epistemically entitled to regard our PJSSs as
factual.
Let us now consider the question whether we are epistemically entitled
to regard our PJSSs as trustworthy. I assume that
      139

(def ) Our PJSSs are trustworthy if and only if it is epistemically reasonable

for us to trust sentences that express our PJSSs unless we have
particular, local reasons in a given context for not doing so.
Now we may reason as follows. If we adopt the PJSS-based conception of
words, we will take ourselves to have no grip on sameness of satisfaction
apart from our PJSSs. More specifically, we will take ourselves to be unable
to raise a doubt about any particular PJSS unless we simultaneously trust a
number of other PJSSs. We will suppose that in principle we may come
to doubt any given PJSSs, but we will not be able to make sense of
simultaneously doubting all of them. We will regard our trust in the PJSSs
that we presuppose when we raise doubts about other PJSSs as epistemically
reasonable, since we cannot see how to assess particular PJSSs without it.
In short, we will be unable to specify a way in which the statement
(t) It is epistemically reasonable for us to trust sentences that express our
PJSSs unless we have particular, local reasons in a given context for
not doing so
could be false. By (T3), then, it is epistemically reasonable for us to accept
(t).³² By (def ), it is epistemically reasonable for us to accept (t) if and only if
it is epistemically reasonable for us to accept that our PJSSs as trustworthy.
Hence, we may conclude that it is epistemically reasonable for us to accept
that our PJSSs are trustworthy. Nevertheless, we have no reason to accept
that our PJSSs are trustworthy, in the sense of ‘reason’ constrained by
(T1). Hence, by (T2), we are epistemically entitled to regard our PJSSs as
trustworthy.
Third, if we adopt the PJSS-based conception, we will regard our PJSS-
based use of words, including our PJSSs, as integral to our grasp of the
word types to which our own Tarski-style disquotational definitions of
satisfaction and truth apply. Recall that if I make a PJSS of the form ‘‘x
satisfies my word type robin if and only if x is a robin and x satisfies A’s word
type r if and only if x is a robin,’’ I will thereby take my word type robin to
be identical to A’s word type r, and conclude that A’s mark r is a PJSS-based

³² This reasoning may seem circular, because (t) itself contains an occurrence of the phrase
‘epistemically reasonable’. But I do not aim to define this phrase; ( T1)–( T3) are constraints on our use
of it, not definitions. Hence they leave open the possibility that other considerations, such as whether
we can make sense of doubting (t), can help us to discover what it is or is not epistemically reasonable
for us to believe.
140     

replica of robin. In general, if we adopt the PJSS-based conception we

will identify the word types to which our own Tarski-style disquotational
definitions of satisfaction apply with the word types that we take others to be
using when we make PJSSs for their words.³³ We will therefore distinguish
between the orthographic criteria that we each set up when we regiment
our words, and that we use to keep track of our own regimented word types
intrasubjectively when we construct disquotational definitions of satisfaction
for them, on the one hand, and the PJSS-based word types to which our
disquotational definitions of satisfaction apply, on the other. Finally, if we
adopt the PJSS-based conception and we explicate our regimented sentence
types as finite sequences of our regimented word types, as I proposed above
(§4.3), then we will identify the sentence types to which our own Tarski-
style disquotational definitions of true-in-L apply with the sentence types
that we take others to be using when we make PJSSs for their words.
I propose that we adopt the PJSS-based conception of words. By doing
so, we transform our understanding of the word types to which our own
Tarski-style disquotational definitions of satisfaction and truth apply: we
come to see our PJSSs as factual and trustworthy, yet also integral to our
grasp of our Tarski-style disquotational truth predicates. Viewed in this
way, the Tarski–Quine thesis does not leave open the possibility that our
PJSSs are not both factual and trustworthy. By adopting the PJSS-based
conception of words, therefore, we discredit (C) (of §4.1): we supplement
our understanding of our own Tarski-style disquotational truth predicates
with an account of why we are entitled to regard our PJSSs as both factual
and trustworthy without undermining the Tarski–Quine thesis that there
is no more to truth than what is captured by a Tarski-style disquotational
truth predicate defined for one’s own sentences.

4.10. Preliminary Objections and Replies

One might doubt that the PJSS-based conception of words, which is at root
practical, is clear enough to be considered integral to our understanding of

³³ How then should we think of quotation, as it is used in our definitions of satisfaction and truth? I
favour a modified Davidsonian account of quotation, according to which a quotation in which a given
mark, such as robin, is displayed refers to the corresponding PJSS letter or word type, such as {x: x is a
PJSS-based replica of my mark robin}. (See Davidson 1984: 79–92 and Bennett 1988 for considerations
that favour a ‘‘display’’ conception of quotation.)
      141

truth. I will address this concern indirectly in later chapters by elucidating

in more detail the central idea behind the PJSS-based conception of
words—the idea that we have no understanding of the word types relevant
to satisfaction and truth independent of the PJSSs on which we rely when
we apply our disquotational definitions of satisfaction and truth to other
speakers’ sentences and to our own sentences as we use them in the past.
Given my proposed explication of the word types relevant to satisfaction
and truth, the task is to elucidate what it is for a given mark to be a PJSS-
based replica of another mark by looking at how we actually keep track of
word types in the midst of our inquiries. This task can seem worth pursuing
only if one already has a preliminary understanding of how the PJSS-based
conception of words differs from token-and-explanatory-use conceptions
of words, and that is what I have tried to convey in this chapter.
I will end this chapter by considering four objections that are easier to
address. First, one might object that if we take the PJSS-based conception
of words to be integral to our understanding of truth, as I propose, then
we cannot make sense of generalizing over sentences of foreign languages,
such as French or German. This objection overlooks the fact that we
can learn how to use words of other languages, and thereby come to
make PJSSs for those words.³⁴ An English-speaker may also be a French
speaker, so an English-speaker may be in a position to construct Tarski-style
disquotational definitions of satisfaction for words of French and to apply
the definitions unreflectively to other French speakers’ words. The same
reasoning applies to any language that we can learn.³⁵
Second, one might object that my proposal conflicts with our intuition
that even before we learn a given foreign language, we can make sense of
applying a truth predicate to its sentences. I concede that my proposal does
not respect this intuition. But I don’t see this as a problem for the proposal,
since we do not need to respect the intuition in order to use logic to clarify
our agreements and disagreements with other inquirers. For this purpose,
it is sufficient that we be able to learn the language of our fellow inquirers,
if we do not already know it.

³⁴ Quine proposed this way of applying Tarski-style definitions of satisfaction and truth to expressions
of a foreign language in Quine 1961a: 130–8. Hartry Field endorses Quine’s proposal in Field 1994:
273–4.
³⁵ Michael Resnik suggests that ‘‘we think of our truth-predicate as applying immanently to the
human polyglot’’ (Resnik 1997: 38). I have argued, in effect, that to make sense of this suggestion
without giving up the Tarski–Quine thesis, we need to embrace the PJSS-based conception of words.
142     

Third, one might think we need a general explanation of how we make

the transition from not knowing anything about a language to being able
to make PJSSs for its words, and that any such explanation will presuppose
or support the token-and-explanatory-use conception of words. I grant
that for every word of every language we learn, there is some (and usually
more than one) way in which we learn to use that word. But this truism
does not imply or even suggest that there is a general explanation of how
we learn to use words, or that to describe how we learn to use a given
word we must presuppose the token-and-explanatory-use conception of
words.
Finally, if our understanding of truth is inextricable from our PJSS-based
conception of words, as I believe, then every sentence to which we can
apply a truth predicate contains word types for which we can make PJSSs.
This will seem counter-intuitive to philosophers who think there may
be true sentences (used by a super-intelligent alien species, perhaps) that
contain word types for which we could never make PJSSs.³⁶ As I argued
in §2.5, however, given my purposes in this book I do not find it fruitful
to evaluate theses about truth by asking whether they successfully capture
a concept that our ordinary uses of ‘true’ express. I recommend that we
ask, instead, ‘‘Do we need a truth predicate?’’ and ‘‘If so, what sort of
truth predicate do we need?’’ My answer to the first of these questions
is that we need a truth predicate to formulate logical laws that we can
use to clarify our agreements and disagreements with other inquirers. For
this purpose, we do not need a truth predicate that applies to sentences
that contain word types for which we could never make PJSSs. Instead,
we need to be entitled to affirm every given sentence of the form ‘X is
true-in-L if and only if S’, where ‘X’ is replaced by a name of a declarative
sentence of our regimented language L, ‘S’ is replaced by the sentence
that ‘X’ names, or a paraphrase of it that we explicitly adopt when we
decide to use sentences of L in place of sentences of ordinary English,
and the sentence types and word types of L are conceived of as PJSS based, in
accordance with the context principle. We can satisfy this need by merging our

³⁶ To make this intuition vivid, Wolfgang Künne imagines that ‘‘there could be intelligent beings,
Alpha-Centaurians, say, endowed with modes of sensory awareness and conceptual abilities that we
and our descendants constitutionally lack. . . . Why should the fact that some of their utterances are
forever incomprehensible to members of our species prevent them from being true?’’ (Künne 2003:
246)
      143

own Tarski-style disquotational definitions of satisfaction and truth with

our PJSS-based conception of words. If we understand truth and words in
this way, we cannot help but regard our PJSSs as factual and trustworthy.
We thereby come to see that the Tarski–Quine thesis and intersubjectivity
constraint are compatible.
 5
Learning from Others,
Interpretation, and Charity

5.1. Is the Intersubjectivity Constraint Compatible

with the Negation of the Tarski–Quine Thesis?
As I stressed in Chapter 2, our desire to clarify and facilitate our rational
inquiries motivates us to regiment our sentences so that we can formulate
logical generalizations, such as ‘‘Every regimented sentence of the form
‘S ∨ ∼ S’ is true.’’ A Tarski-style truth predicate defined disquotationally
for our own regimented sentences enables us to formulate such gener-
alizations for the regimented sentences we can directly use. One might
therefore be inclined to embrace the Tarski–Quine thesis that there is
no more to truth than what is captured by a Tarski-style truth predicate
defined disquotationally for our own regimented sentences. But to clarify
and facilitate our rational inquiries we also need to generalize on the other
speakers’ sentences, so that we can apply logical generalizations to their
utterances. As I emphasized in Chapter 3, however, when we generalize on
other speakers’ sentences, we regard our practical judgements of sameness
of satisfaction (PJSSs) for their words as both factual and trustworthy. This
may lead us to adopt the intersubjectivity constraint: a Tarski-style disquotational
definition of truth for one’s own sentences is satisfactory only if it is supplemented
by an account of why it is epistemically reasonable for us to regard our PJSSs as
both factual and trustworthy. But a Tarski-style disquotational definition of
truth for one’s own sentences tells us nothing about our PJSSs. Hence the
intersubjectivity constraint appears incompatible with the Tarski–Quine
thesis.
I argued in Chapter 4 that this appearance is illusory. If one adopts the
PJSS-based conception of words, one can endorse the Tarski–Quine thesis
    145

yet still explain why it is epistemically reasonable to regard one’s PJSSs

as both factual and trustworthy. In short, the intersubjectivity constraint
is compatible with the Tarski–Quine thesis. But my arguments for the
compatibility of the Tarski–Quine thesis and intersubjectivity constraint
apparently leave it open that one can explain why it is epistemically
reasonable to regard one’s PJSSs as both factual and trustworthy if one
endorses a token-and-ex-use conception of words, and one thereby rejects
both the Tarski–Quine thesis and the PJSS-based conception of words. In
short, my arguments so far apparently leave it open that the intersubjectivity
constraint is compatible with the negation of the Tarski–Quine thesis.
It appears that we now face a choice: either (i) we adopt the PJSS-
based conception of words and embrace both the Tarski–Quine thesis
and intersubjectivity constraint, as I recommend, or (ii) we embrace a
token-and-ex-use conception of words, reject both the Tarski–Quine
thesis and the PJSS-based conception of words, and try to explain why
it is epistemically reasonable to regard one’s PJSSs as both factual and
trustworthy by hypothesizing that facts about language ex-use somehow
constrain or determine the truth values of sentences that express our PJSSs.
For all I have argued so far, many (perhaps even most) philosophers will
still be inclined to choose option (ii) over option (i).
My goals in Chapters 5–8 are (first) to present examples and thought
experiments that discredit option (ii), and (second) to elucidate the PJSS-
based conception of words by showing how it makes sense of examples
and thought experiments that seem puzzling or paradoxical if we assume a
token-and-ex-use conception of words. By the end of Chapter 8, I hope
to have convinced you that contrary to what you might have thought
at first, option (ii) is unacceptable—the intersubjectivity constraint is not
compatible with the negation of the Tarski–Quine thesis—and (i) is the
only genuine option we have if we seek to fit our disquotational truth pre-
dicates with our PJSSs. I begin in this chapter by examining the relationship
between our PJSSs and Donald Davidson’s influential principle of charity.

5.2. Language Ex-Use and Interpretation

If we adopt the PJSS-based conception of words, as I propose, we will
regard all token-and-ex-use conceptions of words as irrelevant to truth and
146   

satisfaction. It would be absurd to claim that language ex-use is irrelevant

to our practical judgements of sameness of satisfaction, however. These
judgements reflect our robust but subtle ability to distinguish between
situations in which we would take a word uttered by another English
speaker to be of the same type as one of our own words, and similar
situations in which we would not do so. From these truisms it is tempting
to infer both that there is a systematic relationship between language
ex-use and our PJSSs and that we need a philosophical theory of that
relationship.
A number of such theories have been proposed, but I shall focus in this
chapter on what is perhaps the most influential of all of these theories—the
theory of interpretation developed by Donald Davidson. Davidson assumes
that an interpreter who does not already know a given language L can
nevertheless describe circumstances under which a given speaker of L holds
her sentences true. In effect, Davidson presupposes a token-and-ex-use
conception of words according to which an interpreter’s specifications of
satisfaction for another speaker’s words are constrained by facts about the
circumstances under which the speaker of L holds her sentences true. The
main goal of Davidson’s theory of interpretation is to explain how an
interpreter can construct a satisfactory interpretation of a language based
only on his knowledge of facts about when its speakers hold its sentences
true. For this purpose, Davidson posits a principle of charity that takes
up the slack between the evidence about language ex-use available to an
interpreter and interpretations of sentences of L. In its simplest form, the
principle of charity requires that we interpret a speaker’s words in such a
way that if the speaker holds a sentence true, then that sentence is true
under the assigned interpretation. A crucial consequence of the principle
of charity is that error only makes sense against a background of massive
agreement, understood as an optimal fit between sentences held true and
truth (Davidson 1984: Essays 2, 9, 10).
To evaluate Davidson’s theory of interpretation, I shall ask whether it
is compatible with our PJSSs. It is clearly incompatible with many of our
PJSSs across time. Suppose we learn, for instance, that in 1750 David Hume
uttered the sentence ‘‘This is gold’’ while pointing at a bit of platinum.
Relying on our PIWs and PJSSs across time for our word ‘gold’, we
unreflectively judge that Hume’s tokens of g-o-l-d are of the same word
type as our tokens of g-o-l-d, and hence that x satisfies both his and our
    147

tokens of g-o-l-d if and only if x is a bit of gold. We know that no bit

of platinum is a bit of gold. Hence we judge that Hume’s utterance was
false. Since by hypothesis, Hume held his utterance of ‘‘This is gold’’ true
while pointing at a bit of platinum, our PJSS for his word ‘gold’ will
lead us to judge that his utterance was false, and hence (for reasons I shall
clarify below) to violate Davidson’s principle of charity. In a similar way,
many of our PIWs and PJSSs across time violate Davidson’s principle of
charity.
Such conflicts between our PIWs and PJSSs across time and Davidson’s
principle of charity are unlikely to persuade anyone who accepts Davidson’s
principle of charity to abandon it, however, since the most compelling
applications of Davidson’s theory of interpretation do not involve cases
in which we try to interpret speakers who lived long ago, but cases
in which we try to interpret our contemporaries. Hence, to discredit
Davidson’s principle of charity, one must show that it prevents us from
accommodating the PIWs and PJSSs on which we rely in our linguistic
interactions with our contemporaries. I shall focus on cases in which we
learn from our contemporaries by relying on our PIWs and PJSSs for their
words and trusting what they write or say.
Many philosophers assume that Donald Davidson’s principle of charity
is at least compatible with trusting what others write or say.¹ Against this, I
will argue that if Davidson’s principle of charity is a constraint on correctly
interpreting what others write or say, then our ordinary impression that we
can learn from others by relying on our PIWs and PJSSs for their words
and trusting what they write or say is an illusion created by habitual but
unjustified misinterpretations of their utterances.² The reason is that in a
large number of cases in which we take ourselves to be learning from others

¹ And some philosophers are committed to the stronger claim that we can justify our trust in what
others write or say by appealing to Davidson’s principle of charity. See, for instance, Coady 1992:
chapter 9. For a similar argument, see Burge 1993. Burge argues that our trust in testimony is justified
by an a priori principle that is ‘‘clearly similar to what is widely called a ‘Principle of Charity’ for
translating or interpreting others’’ (Burge 1993: 487). Burge rejects Davidson’s assumption that speakers
of the same natural language should use the methods of radical interpretation to interpret each other.
But Burge suggests that something like a Principle of Charity provides an a priori entitlement to accept
what others say as true.
² In Fricker 1995, Elizabeth Fricker also raises doubts about Coady’s attempt to use Davidson’s
principle of charity to justify our reliance on testimony, but not by raising doubts about Coady’s
assumption that Davidson’s principle of charity is a constraint on correctly interpreting what others
write or say. Fricker’s point is that testimony may be unreliable even if many of our beliefs are true
(Fricker 1995: 409–10).
148   

by accepting what they write or say, what we take them to write or say,
given (what Davidson sees as) our tacit interpretations of what we read or
hear, is not true by our own lights. In such cases, when we take ourselves
to be learning from others, we violate Donald Davidson’s principle of
charity.
We must therefore either reject Davidson’s principle of charity, or
conclude that what others write or say cannot conflict with what we
believe. I propose that we reject Davidson’s principle of charity. For
reasons I’ll explain, this requires that we also reject the token-and-ex-
use conception of words that underlies his conception of the problem
of interpretation. I propose that we adopt the PJSS-based conception
of words and trust our PIWs and PJSSs, both at a given time and
across time, as in the Hume case described above, unless we have good
reason in a given context not to do so. Unlike Davidson’s principle
of charity, my proposal that we combine the PJSS-based conception of
words with our own Tarski-style disquotational of truth and satisfaction
both allows for and illuminates the familiar phenomenon of learning from
others.

5.3. A Case in which One Person Learns

from Another
My discussion of learning from others will focus on the following case.³
A competent English speaker named Al accepts a number of English
sentences that contain the word ‘arthritis’, including such sentences as
‘Arthritis can cause pain in one’s joints’, ‘Several members of my family
have arthritis’, and ‘I have arthritis in my left knee.’ Al does in fact have
arthritis in his left knee. He also has a pain in his thigh that resembles the
arthritic pain in his left knee. He tells his friend Joe, ‘‘I have arthritis in
my thigh.’’ Joe replies, ‘‘You don’t have arthritis in your thigh—arthritis
afflicts the joints only.’’ ‘‘I bet you’re wrong, Joe,’’ Al says, ‘‘but I’ll ask my

³ This is an elaboration on the arthritis case that Tyler Burge first presented in Burge 1979. I chose
to elaborate on Burge’s arthritis case, and not to construct a completely new case of my own, because
many philosophers now accept our initial, commonsense description of Burge’s case, even though this
description conflicts with some theories of meaning that were once widely accepted. This consensus
about how to describe Burge’s case aids my argument.
    149

doctor—I have an appointment with her tomorrow.’’ The next day, Al

asks his doctor, ‘‘Do I have arthritis in my thigh?’’ She replies ‘‘You don’t
have arthritis in your thigh—arthritis afflicts the joints only.’’ Later that
day, Al tells Joe, ‘‘I was wrong—the pain in my thigh is not arthritis.’’
To begin with, let me draw your attention to several obvious but
important aspects of this case. First, since Al accepts the sentences ‘‘Arthritis
can cause pain in one’s joints,’’ ‘‘Several members of my family have
arthritis,’’ and ‘‘I have arthritis in my left knee,’’ it is natural to say that
Al believes that arthritis can cause pain in one’s joints, that several members
of his family have arthritis, and that he has arthritis in his left knee.
Second, it is natural to suppose that since Al, Joe, and the doctor are all
at least minimally competent English-speakers, they are at least minimally
competent in the use of the English words ‘thigh’ and ‘joint’, so that
if they were asked, ‘‘Is a thigh a joint?’’ they would each say, ‘‘No, a
thigh is not a joint,’’ thereby expressing their belief that a thigh is not
a joint.
Third, Al makes practical identifications of words (PIWs) for Joe’s and the
doctor’s words ‘arthritis’, ‘in’, ‘thigh’, ‘only’, and ‘joint’. In Chapter 3 I
emphasized that speakers of the same language typically take one another
to utter words tokens of which they can each use to express one another’s
assertions, thereby making PIWs for one another’s words. In doing so, they
attribute beliefs to each other and identify points on which they agree and
disagree. In this case, Al takes Joe and the doctor to disagree with his belief
that he has arthritis in his thigh.
Against this background, two aspects of the case stand out. First, Al
does not think he can learn anything from Joe about arthritis, and so
when Joe says, ‘‘You don’t have arthritis in your thigh—arthritis afflicts
the joints only,’’ Al continues to believe that he has arthritis in his
thigh. In contrast, Al assumes that his doctor knows more than he does
about arthritis, and so when the doctor says, ‘‘You don’t have arthritis
in your thigh—arthritis afflicts the joints only,’’ Al takes himself to have
learned from her that arthritis afflicts the joints only. The key observation
is that
(1) Al makes PIWs for his doctor’s words and accepts what she says.
Al thereby takes himself to have learned from her that arthritis afflicts
the joints only. Since Al believes that a thigh is not a joint, he takes
150   

himself to have learned from his doctor that he does not have arthritis
in his thigh.
Second, after his conversation with the doctor, Al tells Joe, ‘‘I was
wrong—the pain in my thigh is not arthritis.’’ Al and Joe remember that
the previous day, Al said to Joe, ‘‘I have arthritis in my thigh.’’ When
we hear this story, we naturally assume that after Al takes himself to have
learned from his doctor that arthritis afflicts the joints only, he also makes
PIWs for words that occur in his previous utterance of ‘I have arthritis
in my thigh’, but rejects the belief that he thereby takes himself to have
expressed. Hence Al takes for granted that for several days, up until he
spoke with his doctor, he believed that the pain in his thigh was arthritis,
and that he learned from his doctor that that belief was false. The key
observation here is that

(2) After Al talks with his doctor, he makes PIWs for words that occur
in his previous utterance of ‘I have arthritis in my thigh’, thereby
taking himself to have expressed the belief that he had arthritis in his
thigh, but he rejects that belief.

Al thereby takes himself to have learned from his doctor that his belief that
he had arthritis in his thigh was false.
It seems a small step from these observations to the conclusion that by
making PIWs for his doctor’s words and accepting what she says, Al learns
from her. But there is more to learning from another than making PIWs
for her words and accepting what she says. What she says must also be
true. From
A learns that p, one may infer
p.⁴ I shall largely ignore
this further condition and focus on cases in which we take for granted
that what a given subject is told is true. I shall highlight and explore the
consequences of aspects (1) and (2) of Al’s situation, which is typical of
situations in which we take ourselves to have learned from others that
some of our previous beliefs were false. In such cases we make PIWs for
others’ words and accept what we thereby take them to have said, while
at the same time relying on PIWs for words in our previous utterances,

⁴ There is also an ordinary sense in which one can be said to ‘‘learn’’ that p, even if p is false.
This sense of ‘‘learn’’ goes with the ordinary sense in which a person may be said to ‘‘teach’’ another
that p, even if p is false. For example, a person who accepts Darwin’s theory of evolution may say
‘‘Zack is teaching Mary (and Mary is learning) that Darwin’s theory of evolution is false.’’ I will not be
concerned with this sense of ‘‘learn’’ in this chapter.
    151

thereby taking ourselves to have expressed beliefs that we now reject

because they conflict with what we take ourselves to have learned.
These aspects of ordinary cases in which we take ourselves to be learning
from others, encapsulated for Al’s case by (1) and (2), seem almost too
obvious to state. We ordinarily take such observations for granted without
any sense that they are in tension with each other, or that they violate any
commonsense constraints on understanding others. It is therefore natural
to assume that the conjunction of (1) and (2) is compatible with Davidson’s
principle of charity.
In fact, however, the conjunction of (1) and (2) violates Davidson’s
principle of charity. To see why, one must focus on the intimate relationship
between (1) and (2), on the one hand, and Al’s PJSSs for his doctor’s words,
as well as his PJSSs for his own words as he used them in the past, on the
other. I shall describe this relationship in the next section, and then sketch a
strategy for showing that the conjunction of (1) and (2) violates Davidson’s
principle of charity.

5.4. Two Conditionals

Recall the following pattern for defining satisfaction for one’s own words
disquotationally:
(Sats) x satisfies my word ‘ ’ if and only if x is .⁵
Anyone who understands (Sats) can apply it to his own regimented words.
For instance, I accept the results of writing my word ‘arthritis’ in the
blanks of (Sats), so I accept that x satisfies my word ‘arthritis’ if and only
if x is (an instance of ) arthritis. If, in addition, I apply (Sats) to another
speaker’s word ‘arthritis’, I make a practical judgement of sameness of
satisfaction (PJSS). More generally, any speaker who accepts applications
of (Sats) to his own words, and who makes a PIW for a speaker A’s word
w while he is talking with A, thereby makes a PJSS for A’s word w. And
any speaker who accepts applications of (Sats) to his own words, and who

⁵ Applications of (Sats) are short for corresponding disquotational definitions of satisfaction of

sentences by sequences of objects. See §3.3.
152   

makes a PIW for a given word w that he used at some previous time t,
thereby makes a PJSS across time for the word w that he used at t.
Let’s assume that Al accepts applications of (Sats) to his own words, so
that, in particular, Al accepts the following sentence:
(a) x satisfies my word ‘arthritis’ if and only if x is (an instance of )
arthritis.
Then we can establish two conditionals, the first with (1) as antecedent,
and the second with (2) as antecedent.
To establish the first conditional, suppose that
(1) Al makes PIWs for his doctor’s words and accepts what she says.
This trivially implies that Al makes PIWs for his doctor’s words. Given that
Al accepts (a), when he makes PIWs for his doctor’s words he accepts the
following sentence
(b) x satisfies the doctor’s word ‘arthritis’ if and only if x is (an instance
of ) arthritis
and thereby commits himself to the conjunction of (a) and (b), which is
a PJSS at a given time for the doctor’s word ‘arthritis’. Now, given that Al
accepts (a) and (b), we can infer that
(3) Al accepts that x satisfies his word ‘arthritis’ if and only if x satisfies
the doctor’s word ‘arthritis’.
These reflections show that whether or not we accept (1), if Al accepts
applications of (Sats) to his own words, then
(C1) If (1) then (3).
This is the first conditional.
To establish the second conditional, suppose that
(2) After Al talks with his doctor, he makes PIWs for words that occur
in his previous utterance of ‘I have arthritis in my thigh,’ thereby
taking himself to have expressed the belief that he had arthritis in his
thigh, but he rejects that belief.
This implies, in particular, that Al makes a PIW for the word ‘arthritis’ that
occurred in his previous utterance of ‘I have arthritis in my thigh.’ Given
    153

that Al accepts (a), when he makes a PIW for his past use of the word
‘arthritis’, he accepts the following sentence
(c) x satisfies the word ‘arthritis’ that I used before talking to the doctor
if and only if x is (an instance of ) arthritis.
and thereby commits himself to the conjunction of (a) and (c), which
is a PJSS across time for the word ‘arthritis’ that he used before talking
to the doctor. Now, given that Al accepts (a) and (c), we can infer
that
(4) Al accepts that an object x satisfies the word ‘arthritis’ that he used
before talking to the doctor if and only if x satisfies the word ‘arthritis’
that he uses after talking with her.
These reflections show that whether or not we accept (2), if Al accepts
applications of (Sats) to his own words, then
(C2) If (2) then (4).
This is the second conditional.

5.5. Strategy
The conjunction of (1) and (2), together with (C1) and (C2), truth
functionally implies the conjunction of (3) and (4). I will use this implication
to argue that if we accept Davidson’s principle of charity, then we must
reject the conjunction of (1) and (2). The heart of my argument, which
I will present in the next several sections, is that if we accept Davidson’s
principle of charity, then we must reject the conjunction of (3) and (4).
Since the conjunction of (1) and (2), together with (C1) and (C2), truth
functionally implies the conjunction of (3) and (4), we may infer that if we
accept Davidson’s principle of charity, then we must reject the conjunction
of (1), (2), (C1), and (C2). That is, we must reject either (1) or (2) or
(C1) or (C2). But, as we shall see, (C1) and (C2) are unproblematic
consequences of Davidson’s methodology of interpretation. Hence if we
accept Davidson’s principle of charity, we must reject the conjunction of
(1) and (2).
154   

5.6. What is Davidson’s Principle of Charity?

To understand Davidson’s principle of charity, one must see how it
fits with his project of constructing empirically testable theories of
interpretation for natural languages. Davidson’s project is to try to bridge
an assumed gap between data about linguistic behaviour, on the one
hand, and semantical interpretations of that data, on the other. In practice
speakers of the same natural language typically ignore this gap—they
typically make PIWs for one another’s words without reflecting about
whether they are justified in doing so. But Davidson thinks that if we
want to interpret another speaker’s words fairly and accurately, it is almost
always wrong to rely on our PIWs for her words. Instead, he thinks,
we should suspend our unreflective trust in our PIWs, acknowledge
the epistemological gap between data about linguistic behaviour, on the
one hand, and semantical interpretations of that data, on the other, and
interpret speakers in accordance with a principle of charity.⁶
To explain how Davidson arrives at this conclusion, I will answer three
questions about his approach: (i) What are the philosophical roots of
Davidson’s project of constructing a theory of interpretation? (ii) What is
Davidson’s conception of the task and test of a theory of interpretation?
(iii) How is Davidson’s conception of the task and test of a theory of
interpretation related to his principle of charity? The answer to question
(i) puts constraints on the answer to question (ii), which in turn puts
constraints on the answer to question (iii). I’ll address them in order,
starting with question (i).
Like many others, Davidson is convinced by W. V. Quine that traditional
philosophical attempts to make sense of the idea of meaning are hopelessly
obscure. Quine argues that all such attempts make use of notions such as
synonymy (sameness of meaning), analyticity (truth ‘‘in virtue of ’’ meaning),
and semantical rule that are themselves no clearer than the idea of meaning.
This would not be a problem if the idea of meaning were clear. But Quine
argues that unlike the sentences used in the mature sciences, including
logic, psychology, and physics, sentences that contain the term ‘meaning’

⁶ In this respect Davidson’s theory of interpretation differs from Quine’s account of translation,
which implies that our PIWs and PJSSs are trustworthy, for reasons I explained in §3.9. The difference
is due to Davidson’s principle of charity, which is related to, but, much stronger than Quine’s.
    155

have no explanatory import. For Quine, explanatory import is the criterion

of clarity, and so he concludes that the terms ‘meaning’, ‘synonymy’,
‘analyticity’, and ‘semantical rule’ are hopelessly obscure.
Quine recommends that we replace the obscure idea of meaning with the
more tractable idea of translation, understood as a mapping from sentences
of one language to sentences of another. To explicate the translation
relation, Quine constructs an empirical theory of translation that makes no
essential appeal to the notion of meaning. In Quine’s view, as I explained in
§3.6, a translation of one language into another is acceptable if it preserves
the relevant speakers’ dispositions to assent to and dissent from sentences
under various stimulus conditions.⁷
Like Quine, Davidson proposes that we replace the obscure idea of
meaning with an idea that has clearer empirical consequences.⁸ But David-
son seeks a theory of interpretation that states the meanings of expressions
of a natural language. A translation of one language A into another B com-
prises a syntactical correlation of the expressions of A with the expressions
of B, and so it does not actually state the meanings of the expressions of A
(Davidson 1984: 129). Davidson agrees with Quine that the notions of truth
and satisfaction are clearer than the traditional idea of meaning,⁹ he thinks
that ‘‘to give truth conditions is a way of giving the meaning of a sentence’’
(Davidson 1984: 24), and he endorses Quine’s suggestion that ‘‘in point
of meaning . . . a word may be said to be determined to whatever extent

⁷ For a more thorough and accurate account of Quine’s views on meaning and translation, see
chapter 2 of Ebbs 1997.
⁸ For a different interpretation of Davidson, see Lepore and Ludwig 2005, especially chapters 3–6.
Lepore and Ludwig claim that Davidson seeks not an explication that replaces the ordinary notions
of translation and meaning, but an account of truth for natural languages that is faithful to our prior
and independent intuitions about translation and meaning. Their interpretation implies that Davidson’s
account of how to test theories of meaning is wrong by his own standards. In my view, however,
Davidson’s account of how to test theories of truth for natural languages is part of an explication
that is meant to replace the ordinary notion of meaning. On the pragmatic view of explication that
I endorse, this distinction is itself blurred, since there is no more to meaning than what can be
salvaged by an explication we find useful for a particular purpose, such as clarifying and facilitating
rational inquiry. Whether or not the distinction is blurred, however, Davidson presents his account
of how to test theories of truth for natural languages as a new standard for judging all questions in
the theory of meaning, and embraces its counter-intuitive consequences, including the one I shall
soon highlight in this chapter. I think this shows that Davidson’s goal is not to construct an account
of truth for natural languages that is faithful to our prior and independent intuitions about translation
and meaning, but to explicate what he finds useful and clear in the ordinary notion of meaning and
translation.
⁹ But note that unlike Davidson, Quine thinks it makes no sense to apply a truth predicate to
sentences that we have not translated and do not understand.
156   

the truth or falsehood of its contexts is determined’’ (Quine 1976b: 82).

Combining these considerations, Davidson claims that the ‘‘meaning’’ of a
sentence S of language L may be specified by an empirically tested theory
of truth for L.
This brings us to question (ii)—‘‘What is Davidson’s conception of the
task and test of a theory of interpretation?’’ Davidson is impressed by Noam
Chomsky’s approach to constructing an empirically testable theory of the
syntax of a speaker’s language:
While there is agreement that it is the central task of semantics to give the semantic
interpretation (the meaning) of every sentence in the language, nowhere in the
linguistic literature will one find, so far as I know, a straightforward account of
how a theory performs this task, or how to tell when it has been accomplished.
The contrast with syntax is striking. The main job of a modest syntax is to characterize
meaningfulness (or sentencehood). We may have as much confidence in the
correctness of such a characterization as we have in the representativeness of
our sample and our ability to say when particular expressions are meaningful
(sentences). What clear and analogous task and test exist for semantics? (Davidson
1984: 21–2)

Davidson’s answer is that the task of a theory of meaning for an infinitary

natural language L is to specify the meanings of all the sentences of L on the
basis of the meanings of a finite number of simple expressions of L without
using the obscure notion of meaning. His conception of the test of such a
theory is modelled on Chomsky’s approach to testing theories of syntax.
Chomsky’s approach is to test a proposed theory of the syntax of a given
speaker’s language by checking evidence about whether or not the speaker
himself regards particular sentences that the theory entails as grammatical.
We can clarify the obscure notion of meaning, Davidson argues, if we can
propose a conception of meaning and evidence that is structurally similar
to the linguists’ conception of theories of syntax and the evidence to which
such theories must be faithful.
Davidson begins with the idea that ‘‘to give truth conditions is a way of
giving the meaning of a sentence’’. He proposes that the task of a theory
of meaning for a natural language L is to construct a Tarski-style truth
theory for L that has a finite number of clauses which together specify the
denotations of all the simple expressions of L. He stipulates that ‘‘a theory of
meaning for a language L shows ‘how the meanings of sentences depend on
    157

the meanings of words’ if it contains a (recursive) definition of truth-in-L’’

(Davidson 1984: 23).
This proposed clarification of the task of a theory of meaning suggests
a simple test of such a theory. According to Davidson, a Tarski-style
truth theory for a natural language L, relativized to persons, times, and
circumstances, entails biconditionals of the following general form:
(T) Sentence s is true-in-L (speaker p’s language) at t if and only if
conditions c obtain at t.
In order to test a Tarski-style truth theory for a natural language, ‘‘all that
is needed is the ability to recognize when the required biconditionals are
true’’. This conception of the test for empirical theories of meaning puts
them ‘‘on as firm a footing empirically as syntax’’ (Davidson 1984: 62).¹⁰
To answer question (iii)—‘‘How is Davidson’s conception of the task and
test of a theory of interpretation related to his principle of charity?’’—we
must look more closely at Davidson’s account of how an empirical theory
of truth is tested. A theory of truth for a German speaker, stated in English,
might entail the following biconditional:
(t) ‘Es schneit’ is true-in-L (speaker p’s language) at t if and only if it is
snowing in the vicinity of p at t.
But how can we test (t) if we do not (yet) know what the sentence ‘Es
schneit’ means? According to Davidson,
A good place to begin is with the attitude of holding a sentence true, of accepting
it as true. This is, of course, a belief, but it is a single attitude applicable to
all sentences, and so does not ask us to be able to make finely discriminated
distinctions among beliefs. It is an attitude an interpreter may plausibly be taken
to be able to identify before he can interpret, since he may know that a person
intends to express a truth in uttering a sentence without having any idea what
truth. (Davidson 1984: 135)

Suppose that statements of our evidence about which sentences a given

speaker holds true have the following general form:
(E) Sentence s is held-true-in-L by speaker p at t if and only if conditions
c obtain at t.

¹⁰ For a later affirmation of the same basic view of the task and test of a theory of interpretation, see
Davidson 1994b: 126–7.
158   

Then one bit of evidence might be:

(e) ‘Es schneit’ is held-true-in-L by speaker p at t if and only if it is
snowing in the vicinity of p at t.
To test a proposed theory of truth for a language L, we must determine
whether the biconditionals it entails are true. This is where Davidson’s
principle of charity comes in: to determine whether (t) is true, we must in effect
treat the phrase ‘true-in-L’ as roughly equivalent to ‘held-true-in-L’. As Davidson
puts it, ‘‘I propose that we take the fact that speakers of a language hold a
sentence to be true (under observed circumstances) as prima facie evidence
that the sentence is true under those circumstances’’ (Davidson 1984: 152).
According to this proposal, for instance, (e) is evidence for (t).
The key point is that to test a theory of truth for a natural language, we
must link our understanding of when the phrase ‘true-in-L’ applies to a
given sentence to our understanding of when the phrase ‘held-true-in-L’
applies to that sentence. The link we need is the principle of charity.
In Davidson’s view, this principle is not optional: the gap that Davidson
describes between data about linguistic behaviour, on the one hand, and
semantical interpretations of that data, on the other, can only be bridged
‘‘by assigning truth conditions to alien sentences that make native speakers
right when plausibly possible, according, of course, to our own view of
what is right’’ (Davidson 1984: 137). Davidson sometimes recommends that
we ‘‘maximize’’ the number of alien sentences that come out true by our
own lights, and sometimes that we ‘‘optimize’’ that number. No matter
how the principle of charity is formulated, however, its role is to give
content to Davidson’s leading idea that the consequences of a particular
theory of truth for a natural language L can be tested against evidence
available to an interpreter who does not already know what the sentences
of L mean.

5.7. Davidson’s Framework for Evaluating (3) and (4)

If we view Al as a Davidsonian interpreter of the doctor’s language (idiolect)
and of the language (idiolect) he used on the previous day, then (3) and
(4) (introduced in §5.4) imply that Al accepts particular interpretations of
the doctor’s words at the time he is learning from her, and of his own words
    159

on the previous day, respectively, so those interpretations are acceptable

or not according as they are permitted by Davidson’s principle of charity.
This can be seen in three steps.
First, suppose that Al has the resources to construct a Davidsonian
theory for another speaker’s language. Then he has the resources to define
satisfaction for his own words disquotationally. Hence we may assume (as
before) that Al accepts applications of (Sats) to his own words, and, in
particular, that Al accepts the sentence
(a) x satisfies my word ‘arthritis’ if and only if x is (an instance of )
arthritis.
We would ordinarily make a PIW for Al’s word ‘arthritis’ and assume that
Al’s assertion of (a) expresses Al’s belief that x satisfies his word ‘arthritis’
if and only if x is (an instance of ) arthritis. But Davidson thinks that if we
are trying to understand Al, we should at first suspend our inclination to
make a PIW for Al’s word ‘arthritis’, and interpret Al in that way only if it
is justified by the principle of charity.¹¹ Even if we suspend our inclination
to make a PIW for Al’s word ‘arthritis’, however, we know that when Al
makes a PIW for his doctor’s word ‘arthritis’, he accepts
(b) x satisfies the doctor’s word ‘arthritis’ if and only if x is (an instance
of ) arthritis
and thereby makes a PJSS at a given time for the doctor’s word ‘arthritis’.
Similarly, we know that when Al makes PIWs for the word ‘arthritis’ that
he used before talking to the doctor, he accepts
(c) x satisfies the word ‘arthritis’ that I used before talking to the doctor
if and only if x is (an instance of ) arthritis
and thereby makes a PJSS across time for the word ‘arthritis’ that he used
before talking to the doctor. From here the reasoning presented earlier
(§5.4) will take us the rest of the way to (C1) and (C2). In this sense,
(C1) and (C2) are unproblematic consequences of Davidson’s approach to
interpretation.

¹¹ This is the main reason that I have emphasized (3) and (4), which we (theorists describing Al’s
situation) can state without making a PIW for Al’s word ‘arthritis’, not the practical judgements that Al
would express by affirming sentences (a), (b), and (c), the meanings of which, according to Davidson,
we cannot take ourselves to know in advance of interpretation.
160   

Second, suppose Al is constructing a Davidsonian theory of truth for

the doctor’s language (idiolect) at the time he is learning from her, or for
his own language (idiolect) at some previous time. Then his PJSSs for the
doctor’s word ‘arthritis’ at the time he is learning from her, and for his own
word ‘arthritis’ at some previous time, imply what Davidson calls ‘‘base
clauses’’ that define satisfaction for, and thereby interpret, the doctor’s word
‘arthritis’ at the time Al is learning from her, and Al’s own word ‘arthritis’
at that previous time. These are the same PJSSs that imply the conditionals
(C1) and (C2), which, together with (1) and (2), imply (3) and (4).
Finally, Davidson thinks his principle of charity governs all interpretation.
From Davidson’s perspective, then, (3) and (4) are acceptable or not
according as the PJSSs that they presuppose are permitted by his principle
of charity.

5.8. Why the Conjunction of (3) and (4) Violates

Davidson’s Principle of Charity
We are now in a position to see that the conjunction of (3) and (4) is
not permitted by Davidson’s principle of charity. I’ll show (first) that if
(4) is permitted, then (3) is not, and (second) that if (3) is permitted, then
(4) is not. These two conditionals are truth-functionally equivalent, but it
is nevertheless instructive to see the symmetrical arguments that support
them.
Suppose that Al is trying to construct a Davidsonian theory of truth for
his doctor’s language at the time when she says, ‘‘You don’t have arthritis in
your thigh—arthritis afflicts the joints only,’’ and that Davidson’s principle
of charity does not rule out (4), according to which Al accepts that x
satisfies the word ‘arthritis’ that he used before talking to the doctor if and
only if x satisfies the word ‘arthritis’ that he uses after talking with her. It’s
part of the story that before Al talks with his doctor, Al takes the ailment
in his thigh to satisfy his word ‘arthritis’. And, as I stressed above, Davidson
thinks we can solve the problem of interpretation only ‘‘by assigning truth
conditions to alien sentences that make native speakers right when plausibly
possible, according, of course, to our own view of what is right’’ (Davidson
1984: 137). Hence, since Davidson thinks that radical interpretation begins
at home, his principle of charity implies that Al must judge satisfaction and
    161

truth by his own lights. Given (4), then, according to Davidson’s principle
of charity, Al must take the ailment in his thigh to satisfy his own word
‘arthritis’ even after he talks with his doctor.
Now suppose in addition that

(3) Al accepts that x satisfies his word ‘arthritis’ if and only if x satisfies
the doctor’s word ‘arthritis’.

Then Al accepts that the ailment in his thigh satisfies the doctor’s word
‘arthritis’. Hence, in effect, Al interprets the doctor’s word ‘arthritis’ in
such a way that the ailment in Al’s thigh satisfies it. But Al can see
from the doctor’s linguistic behaviour that she believes that the ailment
in Al’s thigh does not satisfy her word ‘arthritis’. Thus Al interprets
the doctor’s words in such a way that under the interpretation, the
doctor’s utterance expresses the belief that Al doesn’t have (what Al
calls) arthritis in his thigh. This interpretation of the doctor’s utterance is
unacceptable, according to Davidson, because it attributes what Al takes
to be a false belief to the doctor, and thereby violates the principle of
charity.
Davidson sometimes says that the principle of charity permits us to
attribute false beliefs to other speakers, as long as the error is, by Davidson’s
standards, explicable. (Davidson 1984: 136, 153, 168–9). In this case,
however, the error that Al would be attributing to the doctor if he were
to trust his PIWs for the doctor’s words is not, by Davidson’s standards,
explicable. To see why, consider the following passage, in which Davidson
describes a problem of interpretation that is similar to Al’s:
If you see a ketch sailing by and your companion says, ‘Look at that handsome
yawl’, you may be faced with a problem of interpretation. One natural possibility
is that your friend has mistaken a ketch for a yawl, and has formed a false belief.
But if his vision is good and his line of sight favorable it is even more plausible that he does
not use the word ‘yawl’ quite as you do, and has made no mistake at all about the position
of the jigger on the passing yacht. (Davidson 1984: 196, my emphasis)

In other words, Davidson thinks that under the circumstances described, if

your friend’s vision is good and his line of sight is favourable, you should
not take his word ‘yawl’ to be satisfied by an object x if and only if x is a
yawl, since that would be to attribute an inexplicable error to him—the
error of believing that the ketch sailing by is a yawl.
162   

You might object to Davidson’s treatment of this example on the grounds

that it is natural to assume that your friend intends to speak ‘‘correctly’’—to
apply the word ‘yawl’ in the same way that other competent English-
speakers apply it. In Davidson’s view, however, your friend’s intention
to speak ‘‘correctly’’ in this sense does not settle how his word ‘yawl’
should be interpreted. Davidson would agree that we should interpret
speakers in the way that they intend to be interpreted.¹² But according
to Davidson, our grasp on how the speaker intends to be interpreted is
ultimately rooted in our evidence of which sentences he holds true. By this
criterion, together with the principle of charity, we must conclude that
your friend intends to be interpreted in such a way that the ketch sailing
by is what he calls a ‘yawl’. Your friend may also intend to apply the word
‘yawl’ in the same way that other competent English-speakers apply it,
but according to Davidson this metalinguistic intention is less fundamental
to our understanding of what he means. As Davidson puts it, ‘‘A failed
intention to speak ‘correctly’, unless it foils the intention to be interpreted
in a certain way, does not matter to what the speaker means.’’¹³
Davidson’s treatment of his ketch–yawl example reflects his methodolo-
gical assumption that we can’t explain a speaker’s error simply by attributing
other false beliefs to her, since our attribution of any false beliefs to the
speaker is precisely what needs explaining. Instead we must appeal to factors
that can be checked independently of the interpretation, such as whether
or not a speaker’s vision is good and his line of sight is favourable. In
the ketch–yawl case, by hypothesis, the speaker’s vision is good and his
line of sight is favourable, so, Davidson suggests, if you were to attribute
to your friend the mistaken belief that the passing yacht is a yawl, you
would be attributing an inexplicable error to him. The clear implication

¹² Davidson writes that ‘‘if the speaker is understood he has been interpreted as he intended to
be interpreted’’ (Davidson 1986: 436). In Davidson 1986, he endorses some of H. P. Grice’s views
about the relationship between a speaker’s intentions and the literal meanings of her words. But
Davidson thinks that a speaker’s intentions to be interpreted in a certain way cannot have the status
that Grice attributes to them, since ‘‘making detailed sense of a person’s intentions and beliefs cannot
be independent of making sense of his utterances. . . . an inventory of a speaker’s sophisticated beliefs
and intentions cannot be the evidence for the truth of a theory for interpreting his speech behaviour’’
(Davidson 1984: 144).
¹³ Davidson 1992: 261. Davidson elaborates on this attitude towards ‘‘incorrect’’ usage in ‘‘A Nice
Derangement of Epitaphs’’, where he writes that ‘‘error or mistake of this kind, with its associated
notion of correct usage, is not philosophically interesting. We want a deeper notion of what words,
when spoken in context, mean . . .’’ (Davidson 1986: 434).
    163

of the passage is that you should interpret your friend’s word ‘yawl’ in
such a way that, unlike your word ‘yawl’, it is satisfied by the ketch
sailing by.
By hypothesis, both Al’s and the doctor’s vision is good and their lines
of sight are favourable—they can both see Al’s thigh clearly. Hence,
by Davidson’s standards, Al could not cite these facts to explain what
(according to Davidson) Al should regard as the doctor’s error, if he accepts
that x satisfies his word ‘arthritis’ if and only if x satisfies the doctor’s
word ‘arthritis’. Moreover, I don’t see how there could be any facts about
Al’s and the doctor’s situation that by Davidson’s standards Al could cite
to explain this. I conclude that just as according to Davidson you should
interpret your friend’s word ‘yawl’ in such a way that, unlike your word
‘yawl’, it is satisfied by the ketch sailing by, so according to Davidson Al
should interpret the doctor’s word ‘arthritis’ in such a way that, unlike Al’s
word ‘arthritis’, it is not satisfied by the ailment in Al’s thigh.¹⁴
These considerations show that if (4) is permitted by Davidson’s principle
of charity, then (3) is not. One might think, however, that before Al speaks
with his doctor, Al’s word ‘arthritis’ is satisfied by the ailment in his
thigh, but after he speaks with her and accepts the sentence ‘One can’t
have arthritis in one’s thigh’, Al’s word ‘arthritis’ is no longer satisfied
by the ailment in his thigh. For instance, one might think that when

¹⁴ One might think that Davidson can avoid this conclusion by claiming that ‘arthritis’ is a theoretical
term. Davidson himself has claimed that ‘‘Disagreement about theoretical matters may (in some cases)
be more tolerable than disagreement about what is more evident . . .’’ (Davidson 1984: 169). One might
try to use this commonsense observation to argue that Davidson could allow both Al’s word ‘arthritis’
does not denote the ailment in Al’s thigh, and that Al believes that he has arthritis in his thigh. There
are two main problems with this objection. First, given Davidson’s conception of the task and test
of a theory of interpretation, he has no grounds for thinking that Al’s word ‘arthritis’ is a theoretical
term. Second, even if we did have some reason to regard Al’s word ‘arthritis’ as theoretical, that would
not show that the error that we would be attributing to him if we were to take his word ‘arthritis’
to be satisfied by an object x if and only if x is arthritis is, by Davidson’s standards, explicable. Recall
that to explain a given false belief of a speaker, according to Davidson, it is not enough simply to
attribute other false beliefs to the speaker in light of which her mistake makes sense. Despite Davidson’s
occasional suggestions to the contrary, his principle of charity implies that it is no easier to accept error
among theoretical beliefs than among observational ones.
Simon Evnine thinks that distinction between theoretical beliefs and observational beliefs can be
invoked to defend Davidson against the charge that on his view error is impossible. See Evnine 1991:
chapter 6, especially section 6.5. At the crucial point, however, Evnine simply quotes Davidson’s
commonsense remark about the likelihood of error among our theoretical beliefs, and concludes that
Davidson can accommodate error. For the reasons I just sketched, I don’t see how Davidson’s remark
can help him to avoid the consequence that according to his theory of interpretation, Al can’t be
mistaken about whether he has arthritis in his thigh.
164   

Al accepts the doctor’s assertion, he unwittingly exchanges his old word

‘arthritis’ for a new word that is spelled the same way but is satisfied
by a different range of objects. If we accept this description of what
happens when Al accepts the doctor’s sentence ‘One can’t have arthritis
in one’s thigh’, then (3) may be permitted by Davidson’s principle of
charity.
Let’s turn now to the second way of formulating the conflict between
the conjunction of (3) and (4) and Davidson’s principle of charity: if (3) is
permitted by Davidson’s principle of charity, then (4) is not.
Suppose that Al is trying to construct a Davidsonian theory of truth for
the language he used just a few minutes before he spoke with his doctor,
and that Davidson’s principle of charity does not rule out (3), according
to which Al accepts that x satisfies his word ‘arthritis’ if and only if x
satisfies the doctor’s word ‘arthritis’. Al can see from the doctor’s linguistic
behaviour that she believes that her word ‘arthritis’ is not satisfied by the
ailment in Al’s thigh. Hence Al believes that his own word ‘arthritis’ is not
satisfied by the ailment in his thigh.
Now suppose, in addition, that
(4) Al accepts that an object x satisfies the word ‘arthritis’ that he used
before talking to the doctor if and only if x satisfies the word ‘arthritis’
that he uses after talking with her.
Then Al accepts that before he talked to his doctor, his word ‘arthritis’ was
not satisfied by the ailment in his thigh. But Al remembers that he accepted
the sentence ‘I have arthritis in my thigh’, so he can see that he believed
that his word ‘arthritis’ was satisfied by the ailment in his thigh. Thus Al
interprets his own words in such a way that under the interpretation, his
previous utterance of ‘I have arthritis in my thigh’ expressed the belief
that he has (what Al now calls) arthritis in his thigh. But when Al said, ‘‘I
have arthritis in my thigh,’’ he did not make any mistakes that we could
discern simply by observing the pattern of sentences he held true and using
this data to construct a Davidsonian truth theory for his language (idiolect)
at that time. By Davidson’s standards, then, the mistake that Al attributes
to himself—the mistake of believing that he had arthritis in his thigh—is
inexplicable. In short, Al’s interpretation of his own previous utterance
violates the principle of charity. Hence if (3) is permitted by Davidson’s
principle of charity, then (4) is not.
    165

I conclude that if (4) is permitted by Davidson’s principle of charity,

then (3) is not, and if (3) is permitted by Davidson’s principle of charity,
then (4) is not. In other words, the conjunction of (3) and (4) violates
Davidson’s principle of charity.¹⁵

5.9. My Conclusion Drawn, Generalized,

and Explained
We can now see that if we accept Davidson’s principle of charity, then
we must reject the conjunction of (1) and (2). Recall that the conjunction
of (1) and (2), together with (C1) and (C2), truth functionally implies
the conjunction of (3) and (4). We have now seen that the conjunction
of (3) and (4) violates Davidson’s principle of charity. We may infer that
the conjunction of (1), (2), (C1), and (C2) violates Davidson’s principle
of charity. But, as we saw earlier (§5.7), (C1) and (C2) are unproblematic
consequences of Davidson’s methodology of interpretation. Hence we
must conclude that the conjunction of (1) and (2) violates Davidson’s
principle of charity.
There is no way of avoiding this conclusion while still embracing
Davidson’s conception of the task and test of a theory of interpretation.
Recall that according to that conception, the task of such a theory is to
interpret another speaker’s words by constructing a truth theory for her
language without relying on any evidence about what her words mean,
and the test is to determine whether the consequences of the theory—the
biconditionals it entails—are true. The very idea of such a test builds in
Davidson’s principle of charity, which links evidence plausibly available
to an interpreter—evidence about when speakers hold their sentences
true—with particular biconditionals entailed by the theory. When we

¹⁵ One anonymous reader for OUP suggested to me that since Davidson’s principle of charity
must be understood as a global constraint on the interpretation of the speaker’s utterances, it does not
have the consequences I describe in this section. As I observe in the text, however, given Davidson’s
conception of the task and test of a theory of interpretation, Davidson does not have the resources to
explain away the error that Al must attribute either to himself or the doctor if he accepts both (3) and
(4). This observation about Davidson’s approach to interpretation does not imply that the principle
of charity has only local application, since what Davidson call’s ‘‘formal’’ constraints, especially the
compositionality of a Davidsonian truth theory, remain in place, and require that the application of
charity be systematic and global.
166   

apply this principle we cannot take for granted that we understand what
others are telling us. If we are to obey the principle, we must suppose that at
any given time, our entire grip on truth is exhausted by what we currently
believe at that time. This does not imply we cannot acquire new beliefs
by using words in the way that others do. As I noted above, Davidson
allows for the possibility that when Al accepts the doctor’s assertion, he
unwittingly exchanges his old word ‘arthritis’ for a new word that is spelled
the same way but is satisfied by a different range of objects. But, as I also
noted above, Davidson’s principle of charity does imply that if Al accepts
the doctor’s assertion, he should not also take himself to correct the belief that
he expressed yesterday by saying, ‘‘I have arthritis in my thigh.’’ Instead, he
should conclude that his word ‘arthritis’, as he used it yesterday, is satisfied
by a range of objects that is different from the range of objects that satisfy
his word ‘arthritis’, as he uses it now, after talking with the doctor. Hence
Davidson’s principle of charity implies that we cannot learn from others
by trusting our PIWs for their words and accepting what we thereby take
them to say, where this is understood to involve correcting one’s previous
beliefs.
Why have so many philosophers missed or ignored this consequence
of Davidson’s approach to interpretation? The main reason, I think, is
that our practice of trusting our PIWs for one another’s words is so
deeply entrenched that it is largely unresponsive to Davidson’s a priori
criticisms. Even those who have studied and absorbed Davidson’s approach
to interpretation apparently overlook the extent to which their PIWs
for other speakers’ words shape their ‘‘interpretations’’ of other speakers’
utterances. Most interpreters who think they are following Davidson’s
recommendations in fact unwittingly rely on their PIWs for other speakers’
words even when those PIWs commit them to PJSSs that conflict with
Davidson’s principle of charity. They take comfort in Davidson’s claim that
his principle of charity makes room for error, but they do not realize that
the errors that they are inclined to attribute to themselves or to others are
not in fact compatible with his principle of charity.
There is also a tendency to conflate charity with trust —to assume that it
is charitable to rely on one’s PIWs for another speaker’s words and accept
what one thereby takes her to have said solely because we trust her. Unlike
trust, however, charity is something we think of ourselves as exercising
only if we take ourselves to be in a position superior in some respects to the
    167

position of the person to whom we aim to be charitable. Hence Davidson’s

principle of charity is well named: if we obey it, we will have no choice but
to regard ourselves as ultimate authorities on truth, so we will always interpret
others in such a way that what they say or write is compatible with what
we already believe.¹⁶ As we have seen, however, when I make PIWs for
another speaker’s words and accept what I thereby take her to have said
solely because I trust her, I do not always interpret her words in accordance
with Davidson’s principle of charity. When I trust another speaker in such
contexts, I take her position to be superior to mine for judging truth about
the salient topic of discussion, and in that respect, do not treat her with
charity.
Many philosophers of language nevertheless take for granted that to
understand or make sense of other speakers just is to interpret them with
charity. For Davidson, these come to the same thing, because in his view
we should explicate (and, in effect, replace) the vague words ‘meaning’,
‘synonymy’, and ‘belief ’ by describing how to construct an empirically
testable theory of interpretation governed by his principle of charity. One
who dislikes the consequences of Davidson’s explication described above
could still suppose that to understand or make sense of other speakers just is
to interpret them with ‘‘charity’’. To suppose this in a way that challenges
my conclusions, however, one would need to articulate a principle of
charity that is different from Davidson’s, yet can nevertheless be used at
least to clarify the vague words ‘meaning’, ‘synonymy’, and ‘belief ’ that
Davidson sought to clarify. I know of only three main types of alternatives
to Davidson’s principle of charity, however, and none of them satisfies
the condition just stated. The first main alternative is to explicate the

¹⁶ This consequence of Davidson’s view—that we have no choice but to regard ourselves as ultimate
authorities on truth—does not imply that for Davidson our confident applications of our words make
true the assertions that we express by using them. My point here is epistemological, not metaphysical:
it follows from Davidson’s theory of interpretation that we are never justified in interpreting another’s
assertions in such a way that those assertions conflict with our own beliefs. It may be tempting to try
to explain this by saying that our confident applications of our words make true the assertions that we
express by using them. But even if Davidson’s principle of charity implies that I must take myself to be
an authority on whether or not my word ‘arthritis’ is satisfied by a given object x, it does not follow that
Davidson is committed to the metaphysical claim that my confidence that a given object x is an instance
of arthritis makes true my judgement that x is an instance of arthritis. At most the point about authority
raises the question of whether Davidson is entitled to the distinction between belief and truth. This
is an important and interesting question, but I need not settle it to show that Davidson’s principle of
charity precludes learning from others. I grant for the sake of argument that Davidson is entitled to the
logical distinction between belief and truth.
168   

phrase ‘principle of charity’ in terms of our ordinary intuitions about what

expressions mean, when two expressions are synonymous, and related
intuitions about what speakers believe. (See Lewis 1974 for what is, in
effect, if not in rhetoric, one version of this alternative. Grandy 1973 offers
another version, but uses a different phrase—‘the principle of humanity’.)
If we accept this alternative, then the phrase ‘principle of charity’ cannot
be explained and applied independently of our ordinary intuitions about
meaning, synonymy, and belief.
A second alternative is to suppose that the expressions ‘principle of
charity’, ‘meaning’, ‘synonymy’, and ‘belief ’ can somehow be used to
clarify each other, even if our pre-theoretical uses of any one of them
are not completely independent of our pre-theoretical uses of the others.
According to a version of this alternative proposed by Gareth Evans and
John McDowell, for instance,
the idea would be that a detailed account of the complex network of beliefs and
intentions involved in conversation, in the characteristic Gricean style (but purged
of the reductionist refusal even to entertain the possibility that semantic matter
might figure in the specification of the content of the beliefs and intentions),
might yield a richer set of simultaneous equations for us to consider ourselves
as solving when we construct a theory of meaning: and hence a richer set of
constraints, imposed, not necessarily in a reductive spirit, by bringing general
psychological principles to bear upon determinations of meaning in order to make
the constructed theory fit the data on the basis of which is constructed. Apart
from this greater richness, the constraints would operate in much the same way as
that already considered in connection with the Davidsonian picture. (Evans and
McDowell 1976: xix.)

In this passage, Evans and McDowell reject Davidson’s conception of

the task and test of a theory of interpretation, calling it ‘‘reductive’’.
While they do not explicitly mention a principle of charity, they claim to
model their account of constructing an interpretation on ‘‘the Davidsonian
procedure’’, and may thereby be taken to be proposing that we understand
(some revised version of ) Davidson’s principle of charity as among ‘‘the
constraints that would operate’’ when we construct a ‘‘non-reductive’’
theory of interpretation for a given speaker. The problem with this sort of
proposal is that there is no established pre-theoretical practice of applying
‘‘charity’’ in interpretation, and hence no such use to appeal to when trying
to clarify the phrases ‘meaning’, ‘synonymy’, and ‘belief ’.
    169

A third alternative is to take the fact that a speaker holds-true sentences

that express her PIWs and PJSSs as a prima-facie reason for an interpreter
to suppose that such sentences are true.¹⁷ If we adopt this alternative, we
will take the fact that Al holds-true sentences (3) and (4), for instance, as
a prima-facie reason to suppose that (3) and (4), as Al uses them, are true.
The main problem with this alternative to Davidson’s principle of charity
is that to accept it is to reject the assumption that semantic claims such as
the ones expressed by (3) and (4) need empirical support by evidence that
is available independent of them, and thereby leave no substantive role for
charity to play in semantic interpretation. (I examine and criticize a version
of this alternative in more detail in Chapter 8.)
In short, I see no way to articulate a plausible substantive principle of
charity that does not have the consequences of Davidson’s principle of
charity that I described above. I conclude that the shortcomings with
Davidson’s principle of charity illustrate shortcomings with any plausible
substantive principle of charity. Davidson points out that in his view
‘‘The methodology of interpretation is . . . nothing but epistemology seen
in the mirror of meaning’’ (Davidson 1984: 169). We can now see that his
methodology of interpretation, and, more generally, any methodology of
interpretation that relies on a substantive principle of charity to clarify the
ordinary vague phrases ‘meaning’, ‘synonymy’, and ‘belief ’, is individualist-
ic: it precludes learning from others by making PIWs for another speaker’s
words and accepting what we thereby take her to say solely because we
trust them.¹⁸

5.10. Is the Principle of Charity Optional?

Is this a criticism of the idea that interpretation is governed by a principle of
charity? The answer to this question depends on whether or not interpreting
in light of a principle of charity is optional. Davidson insists that we have no
choice but to accept a principle of charity in interpretation, since otherwise

¹⁷ Bill Brewer, Henry Jackman, and an anonymous reader for OUP each independently urged me
to consider this objection.
¹⁸ This shows what is wrong with C. A. J. Coady’s attempt (in Coady 1992: chapter 9) to use
Davidson’s principle of charity to justify our practice of taking ourselves to learn from others. Coady
does not realize that Davidson’s principle of charity presupposes a radical epistemological individualism.
170   

we would not be able to solve the problem of interpretation. He would

insist that even though we are inclined to trust our PIWs and PJSSs, careful
reflection about linguistic interpretation shows we should not trust them.
If that undermines our assumption that we learn from others by making
PIWs for other speakers’ words and accepting what we thereby take them
to say solely because we trust them, he reasons, then so much the worse
for this assumption, which must be rejected along with other tempting but
unfounded assumptions, such as the assumption that synonymy (sameness
of meaning) is an objective relation, or that some sentences are analytic
(true ‘‘in virtue of ’’ their meaning).
I agree that we cannot accept Davidson’s conception of the task and
the test of a theory of interpretation without also accepting his principle
of charity. If Davidson’s token-and-ex-use conception of words, and the
conception of the task and the test of a theory of interpretation that goes
with it, were not optional, then his principle of charity would not be
optional either.¹⁹ The question, therefore, is whether Davidson’s token-
and-ex-use conception of words and the conception of the task and his
proposed test of a theory of interpretation are optional.
Like most philosophers, Davidson simply presupposes a token-and-ex-
use conception of words. Given this presupposition, he was driven to his
conception of the task and the test of a theory of interpretation by his
disenchantment with traditional approaches to meaning, which rely on
such obscure notions as meaning, synonymy, and analyticity. I agree with
Davidson that an adequate philosophical description of our practice of
interpreting each other should not make use of such notions. The question,
therefore, is whether we can find a way of understanding linguistic
interpretation that accommodates our PIWs and PJSSs, and thereby makes
sense of the familiar phenomenon of learning from others, without relying

¹⁹ Thus I disagree with some critics of Davidson’s principle of charity, such as Richard Grandy
(in Grandy 1973) and David Lewis (in Lewis 1974), who think they can avoid counter-intuitive
consequences of that principle by reformulating it slightly, without questioning Davidson’s conception
of the problem of interpretation. In fact, both Grandy and Lewis reject Davidson’s conception of the
problem of interpretation. Grandy takes many of our PJSSs for granted in his characterizations of our
evidence for interpretations, and Lewis posits the existence of whatever constraints on interpretation
are needed in order to rule out interpretations he doesn’t find intuitive, or to break a tie between
two that he does find intuitive. These alternative ‘‘principles of charity’’ simply presuppose ordinary
notions of sameness of meaning and reference—the very notions that Davidson’s principle of charity
is supposed to help us to clarify in other, independent terms—and are therefore not my target in this
chapter.
    171

on such obscure notions as meaning, synonymy, and analyticity. The

answer to this question, I believe, is ‘‘yes’’.

5.11. An Alternative to Davidson’s Principle

of Charity
I proposed in Chapter 4 that we combine the PJSS-based conception of
words with our own Tarski-style disquotational definitions of satisfaction
and truth. To adopt this proposal is in effect to regard our PIWs and PJSSs
as part of the data that our account of interpretation must accommodate.
As I explained in Chapter 4, if we adopt the PJSS-based conception of
words it is epistemically reasonable for us to trust our PJSSs unless we have
some reason in a given context for not doing so. Hence if we adopt the
PJSS-based conception of words, we will conclude that neither our PIWs
nor the PJSSs they commit us to need special independent justification.
Instead, we will conclude that it is local divergences from our PIWs and PJSSs
that need justification. But the sort of justification that such divergences
need cannot be derived from an abstract principle of interpretation; it must
be based in a description of how we actually proceed when we are trying
to understand utterances that puzzle us.
The first thing we typically do is ask our interlocutor what he or she is saying.
This is only possible if we take for granted some shared vocabulary with
which we may discuss the question. Davidson describes the problem of
interpretation in such a way that no common vocabulary can be assumed,
and so every identification of one person’s word with another’s must be
justified by some abstract principle that links linguistic behaviour to theory.
The key to my alternative is to build some of what Davidson regards
as ‘‘interpretations’’ into the data with which we begin when we try to
understand others.²⁰
To see in more detail how my approach differs from Davidson’s, recall
first that if we combine applications of (Sats) to our own regimented words

²⁰ My approach to interpretation is in some ways similar to Rudolf Carnap’s approach to it in his

paper ‘‘Meaning and Synonymy and Natural Languages’’, reprinted in Carnap 1956: 233–47. Carnap
assumes that to determine what a speaker’s words mean, we must ask him whether he would apply it in
various different possible circumstances—circumstances that we describe by using words of his whose
translations we take as settled. At the same time, however, Carnap also assumes (wrongly, in my view)
that the extensions and intensions of a speaker’s words are fixed by her linguistic dispositions.
172   

with our PIWs for a speaker’s words, we thereby make corresponding PJSSs
for the speaker’s words. This simple observation can help us to describe the
arthritis case. If Al accepts the results of writing his word ‘arthritis’ in the
blanks of (Sats), he can see that when he takes his word ‘arthritis’ to be
the same as the doctor’s word ‘arthritis’, he in effect takes for granted that
an object x satisfies the doctor’s word ‘arthritis’ if and only if x is arthritis,
and thereby makes a PJSS for the doctor’s word ‘arthritis’. Similarly, if the
doctor accepts the results of writing her word ‘arthritis’ in the blanks of
(Sats), she can see that when she takes her word ‘arthritis’ to be the same as
Al’s word ‘arthritis’, she in effect takes for granted that an object x satisfies
Al’s word ‘arthritis’ if and only if x is arthritis, and thereby makes a PJSS
for Al’s word ‘arthritis’.
If we adopt the PJSS-based conception of words, as I recommend, then
we will accept such PJSSs unless we have some concrete reason in a given
context for not doing so. We do not and should not begin with the
assumption that we cannot trust our PIWs for another speaker’s words
unless we can provide some independent justification for doing so. Since
we need not begin with this assumption, our practice of interpreting each
other does not commit us to Davidson’s principle of charity. If there are
reasons in a given context for suspending or rejecting a PIW or PJSS, they
are not abstract reasons derived from a principle of charity, but concrete
reasons rooted in our ordinary practice of interpreting each other.
But what counts as a ‘‘concrete reason’’ for suspending or rejecting
a PJSS? The context principle implies that there can be no correct,
informative, and general account of the reasons that should lead us to
suspend or reject our PJSSs, so I can only offer examples. I have given some
examples already (see §3.2), but here is another. Suppose John invites his
friend Sally over for dinner, and asks her if she has any dietary restrictions.
Sally says, ‘‘I don’t eat meat; anything else would be fine.’’ John serves
chicken. Sally, who is fastidious about her diet, objects, saying, ‘‘John, I
thought I told you that I don’t eat meat.’’ John replies, ‘‘Chicken is not
meat, Sally—it’s poultry.’’ Assume that both are sincere and at first make
PIWs for one another’s words. They consult the Shorter Oxford English
Dictionary (SOED), and discover that according to one of the SOED
entries for ‘meat’—‘The flesh of animals used as food, now esp. excluding
fish and poultry . . .’—John is right. Sally regards this dictionary entry for
‘meat’ as the definition of a word that is different from, even if related
    173

to, her word ‘meat’. Both Sally’s and John’s uses of the word ‘meat’ are
established among competent English users; there is good reason to think
that the word is ambiguous between a looser use that encompasses chicken,
and a more restricted use that does not. In this case, when Sally says, ‘‘John,
I thought I told you that I don’t eat meat,’’ and regards the dictionary
definition of ‘meat’ as the definition of a word that is different from her
word ‘meat’, it becomes clear to John that he was wrong to identify his
word ‘meat’ with Sally’s word ‘meat’.²¹
In this and other cases in which we suspend or revise a previous PJSS,
we presuppose other PIWs and PJSSs. For instance, Sally and John rely
on their PIWs for one another’s words ‘chicken’ and ‘poultry’ in their
discussion of whether or not poultry is meat. But we typically rely on our
PIWs and PJSSs, doubting or revising them only if there are good reasons
for doing so. We trust our PIWs and PJSSs unless we have some concrete
reason in a given context for not doing so.²²
If we adopt the PJSS-based conception of words, as I recommend, then
we can make sense of our trust in our PIWs and PJSSs, including Al and
his doctor’s PIWs and PJSSs. And if we also think Al is entitled to accept
what his doctor tells him, we can describe Al as having learned that arthritis
applies to the joints only, and as being relieved that what he previously
believed about his thigh is false. We can also describe the doctor as having
corrected Al’s false belief about arthritis. We therefore can describe this as
a clear case of what I call learning from others.
Davidson’s argument against describing the case in this way presupposes
his conception of the problem of interpretation, as we have seen. We
can resist Davidson’s argument by adopting a different description of our
practice of interpreting each other. As I see it, our understanding of
satisfaction and truth is inextricable from our practice of making PIWs
and PJSSs. Divergences from this practice are local, as are the reasons for
those divergences. No single abstract principle of interpretation is needed,

²¹ Sometimes two speakers will disagree about the proper definition of the word without concluding
that they are actually using different words with different satisfaction conditions. This possibility is
crucial to the attempt to provide good definitions of words already in use. See Burge 1986a, and Ebbs
1997: chapter 8.
²² Although I think it is typical that speakers make PIWs for each other’s words, under certain
unusual circumstances we may come to the conclusion that we should not make PIWs for most of
a given English speaker’s words. Still, if we arrive at this conclusion, we will have worked our way
towards it by relying on PIWs for the words we use to pose questions to the speaker about what she
means by certain other words.
174   

since particular interpretations always presuppose some background of

momentarily unquestioned assumptions, and are therefore never generated
completely from scratch.²³

5.12. Frontiers of Translation

Davidson’s conception of the problem of interpretation goes hand in hand
with his assumption that we need a fully general philosophical theory of
how we interpret others. Such a theory must cover all cases, including cases
in which we know nothing about the language we are attempting to inter-
pret. These are cases in which we must engage in what Davidson calls radical
interpretation. Since he assumes that all interpretation must proceed by the
same principles that we use in radical interpretation, he thinks that all inter-
pretations are ultimately based on evidence that is available to an interpreter
who knows nothing about the language he is interpreting. Even when we
interpret our fellow English-speakers, Davidson believes, we employ the
same abstract principles of interpretation that govern radical interpretation.²⁴
I propose that we turn this reasoning on its head—assume not that
radical interpretation is the model for all interpretation, but that ordinary
interpretation, in which we take for granted a shared vocabulary for

²³ My approach also allows us to describe Al as having the sort of self-knowledge (or first-person
authority) that goes with minimal competence in the use of language, even though he does not know
that arthritis afflicts the joints only. Davidson’s conception of the problem of interpretation and principle
of charity prevents him from accepting this description of Al. As Davidson points out, his principle of
charity implies that ‘‘A belief is identified by its location in a pattern of beliefs; it is this pattern that
determines the subject matter of the belief, what the belief is about. . . . false beliefs tend to undermine the
identification of the subject matter; to undermine, therefore, the validity of the description of the beliefs
as being about that subject’’ (Davidson 1984: 168). To take Al to believe that one can’t have arthritis in
one’s thigh would ‘‘tend to undermine the validity of the description of the belief as being about that
subject,’’ arthritis. In short, as I stressed earlier, the principle of charity implies that each individual is the
ultimate authority on whether or not one of his words denotes a contextually salient object, such as a
boat sailing by, or a painful condition in one’s thigh. According to Davidson, to accept the commonsense
view that Al’s belief that he has arthritis in his thigh is false is in effect to deny that Al is an authority on
whether or not his word ‘arthritis’ denotes the painful condition in his thigh, and thereby to deny that
Al has first-person authority about what thoughts he expresses when he uses the word ‘arthritis’. Once
again, however, if we reject Davidson’s conception of the task and test of a theory of interpretation,
we are not committed to his principle of charity, so we are not committed to his assumption that each
individual is the ultimate authority on whether or not one of his words denotes a contextually salient
object. We are free to accept that Al has self-knowledge or first-person authority, even if he believes
falsely that he has arthritis in his thigh. I develop this consequence of my view below, in Chapter 9.
²⁴ Davidson stresses that ‘‘All understanding of the speech of another involves radical interpretation’’
(Davidson 1984: 125).
    175

raising questions about what other speakers mean, is the model for all
interpretation. My proposal is best understood in stages, beginning with
the familiar cases in which questions arise about how to interpret other
speakers’ words.
For instance, in the case involving John and Sally that I described above,
John’s question about Sally’s word ‘meat’ arises against the background of
PIWs for Sally’s words. John’s and Sally’s discussion of the SOED entry
for the word ‘meat’ was prompted by Sally’s protest that John ignored
her request that he not serve ‘meat’. This discussion led John and Sally to
conclude that an object x satisfies Sally’s word ‘meat’ if and only if x satisfies
John’s expression ‘meat or poultry’. Once they come to this agreement,
however, they could just as well begin to use each other’s words directly,
each now treating the word form m-e-a-t as ambiguous. But suppose they
keep track of the ambiguity by using subscripts. Then they will agree that
an object x satisfies ‘meatJ ’ if and only if x is meatJ , and that an object x
satisfies ‘meatS ’ if and only if x is either meatJ or poultry. (They will also
both accept the disquotational statement that an object x satisfies ‘meatS ’
denotes x if and only if x is meatS , but this would not help them to keep
track of the difference between ‘meatJ ’ and ‘meatS ’.)
If John and Sally begin to assimilate these regimented words into
their respective regimented languages and make PIWs for these words
in appropriate circumstances when they discuss food together, then their
explicit interpretations of the word ‘meat’ will fade into the background.
If and when this occurs, they will no longer question their PIWs for
these words. They will therefore no longer be involved in what I call
interpreting each other’s uses of these words. In this way it can happen that
an interpretation agreed on at one time may later become unproblematic,
and hence no longer be an interpretation, but simply part of the background
against which new questions about how to interpret other words can be
formulated and addressed.²⁵
This approach to interpretation says nothing about what Davidson calls
radical interpretation, which is supposed to be necessary when we seek to

²⁵ There is a parallel here with what Quine calls legislative postulation (Quine 1963: 394–5). But
Quine would regard our trust in our PIWs as a tacit endorsement of a homophonic translation
manual, and hence as just one of many acceptable ‘‘translations’’ of our fellow English-speakers’ words.
Moreover, his notion of legislative postulation is individualistic, having nothing in principle to do with
linguistic collaboration, understanding, or translation.
176   

understand the utterances of a language that we have never encountered

before, and for which there are no manuals of translation. If there were a
case of what Davidson calls radical interpretation, it would lie at the frontier
of entrenched translation practice, where hitherto unknown languages are
encountered. In my view, all interpretation proceeds against a background
of PIWs and PJSSs, and so interpretation cannot begin until some basic
ways of taking other speakers’ words are adopted and relied on. In short,
what Davidson calls radical interpretation is not what I call interpretation.
One might be willing to accept this terminological point, yet still insist
that we need a general method for pushing back the frontiers of entrenched
translation practice, so that we can get into the position in which we
make PIWs and PJSSs, and can therefore interpret puzzling utterances in
the way I described above. If we accept the demand for such a method,
then Davidson’s attempt to provide one will look attractive. Recall that
according to his method, a radical interpreter can identify the attitude of
holding a sentence true without knowing anything about how to translate
the sentence. Davidson’s method describes how we can use evidence of
when the native speakers ‘‘hold true’’ sentences of their language to test
particular proposed ‘‘interpretations’’ of this evidence. Since this evidence
must be available independent of any prior translation of the language,
our understanding of what it is to ‘‘hold true’’ a given sentence cannot be
explained disquotationally. Davidson’s method therefore commits him to
accepting a non-disquotational notion of truth. (Davidson himself stresses
this in Davidson 1990.) In contrast, I have been urging that we adopt
a disquotational Tarski-style method of defining truth for the pragmatic
reasons sketched above, we have no need for a non-disquotational notion
of truth. In particular, we have no need for a notion of ‘‘holding true’’ that
we can grasp independent of our practice of making PIWs and PJSSs for
each others’ words.
This conception of translation practice may seem to imply that we
cannot criticize particular translations. For it is tempting to think that we
can criticize a particular translation only if there are valid principles for
translation that are independent of our actual PIWs and PJSSs. That there are
such principles seems to be supported, for instance, by Quine’s influential
criticism of Levy-Bruhl’s doctrine that there are pre-logical peoples who
accept certain contradictions as true. To illustrate Levy-Bruhl’s doctrine,
Quine imagined that a group of speakers of a newly encountered language
    177

accept as true a certain sentence of the form ‘q ka bu q’, which is translated

into an English sentence of the form ‘p and not p’. Quine remarked that ‘‘If
any evidence can count against a lexicographer’s adoption of ‘and’ and ‘not’
as translations of ‘ka’ and ‘bu’, certainly the natives’ acceptance of ‘q ka bu
q’ counts overwhelmingly’’ (Quine 1963: 387). Quine’s criticism rests on
the obviousness of elementary logic. One might think that it depends on
the more general assumption that all translation is governed by a principle
of obviousness, according to which speakers of a language should not be
translated in such a way that some of the sentences they accept are translated
by English sentences that we regard as a obviously false. In a similar way,
one might think that any criticism of actual translation practice presupposes
that there are valid principles of translation that are independent of our
PIWs and PJSSs. This would be an objection to my proposal, since I accept
that we can criticize particular translations, yet I deny that there are valid
principles for translation that are independent of our PIWs and PJSSs.
The trouble with this objection is that for any general principle of
translation, there are bound to be cases in which our PIWs and PJSSs
conflict with the requirements of the principle. For instance, both Quine’s
principle of obviousness and Davidson’s principle of charity apparently
imply that we cannot learn from others that something we took to be
obviously true is in fact false. The principles are therefore vulnerable to
the criticism I made in this chapter of Davidson’s principle of charity.
Echoing Quine, we might say that if any evidence can count against these
principles, certainly their conflict with the phenomenon of learning from
others counts overwhelmingly.

5.13. The Method behind these Conclusions

I started this chapter by describing a typical case in which one speaker
learns from another by trusting his PIWs for her words and accepting what
he thereby takes her to say. I then explained why Davidson’s principle of
charity is incompatible with our ordinary observations about this typical
case. If Davidson is right, then we can’t learn from another by making
PIWs for her words and accepting what we thereby take her to say. The
problem is that according to Davidson’s understanding of interpretation,
each individual is the ultimate authority on whether or not one of his
178   

words is satisfied by a contextually salient object, such as a boat sailing by,

or a painful condition in one’s thigh.²⁶ But if we learn from others, then
we are not the ultimate authorities on whether or not one of our words
is satisfied by a contextually salient object. There is no way to avoid this
consequence without rejecting Davidson’s conception of the problem of
interpretation.
I argued that this conception is not mandatory—there is an alternative.
In particular, I proposed that we accept our PIWs and PJSSs as part of
the data that any account of interpretation must accommodate. I proposed
that we adopt the PJSS-based conception of words and trust our PIWs
and PJSSs unless we find concrete reasons in a particular context for
revising them. I then briefly sketched the consequences of this approach
for our understanding of learning from others. I argued that if we take this
alternative approach to describing our linguistic practices, we can accept
our ordinary observations about typical cases, such as Al’s, in which we
take ourselves to have learned from others. We have seen that Davidson’s
understanding of the word ‘word’ leads him to embrace his principle of
charity, which conflicts with the PIWs and PJSSs on which we rely when
we take ourselves to learn from others by accepting what they say. My
proposal that we adopt the PJSS-based conception of words is expressly
designed to explicate the word ‘word’ in a way that fits with our PIWs, and
thereby to clarify and preserve the phenomenon of learning from others. I
conclude that if we wish to make sense of those aspects of learning from
others that matter to us when we see ourselves as engaged in rational
inquiry, then we should reject Davidson’s conception of the problem of
interpretation, as well as the token-and-ex-use conception of words that
it presupposes, and combine the PJSS-based conception of words with a
Tarski-style disquotational account of satisfaction and truth.

²⁶ Recall that this does not imply that for Davidson our confident applications of our words make
true the assertions that we express by using them. See note 16 of this chapter.
 6
A Puzzle about Sameness
of Satisfaction across Time

6.1. An Intuition about Sameness of Satisfaction

across Time
I argued in the previous chapter that we should reject Davidson’s principle
of charity because it conflicts with the practical judgements of sameness
of satisfaction (PJSSs) on which we rely when we take ourselves to have
learned from others by accepting what they write or say. I also argued
that we can accommodate these PJSSs if we reject all token-and-ex-use
conceptions of words and adopt the PJSS-based conception of words
instead. I want now to examine the objection that it would be wrong
to adopt the PJSS-based conception of words because it conflicts with a
deeply entrenched intuition about sameness of satisfaction across time that
can only be accommodated by an account of sameness of satisfaction across
time that presupposes a token-and-ex-use conception of words.
The intuition is best introduced with an example. Suppose that in 1650
John Locke bought a ring from a jeweller who said, ‘‘This ring is gold,’’ and
today, three and a half centuries later, a chemist performs various tests on
the ring, and says, ‘‘This ring is not gold.’’ When told this story, we make
a complex PIW (practical identification of a word) for the jeweller’s word
‘gold’, trusting that there is a chain of simple PIWs that link our present use
of ‘gold’ back through time to the jeweller’s word ‘gold’, thereby taking
for granted that the jeweller’s word ‘gold’ and the chemist’s word ‘gold’ are
the same as our word ‘gold’, in the sense of ‘same word’ that licenses us to
make corresponding PJSSs for the jeweller’s word ‘gold’ and the chemist’s
word ‘gold’. These PJSSs together imply that
180      

(1) An object x satisfies the jeweller’s word ‘gold’ if and only if x satisfies
the chemist’s word ‘gold’.
The PJSS-based conception of words licenses us to accept the PJSSs that
imply (1) whether or not the facts about the ex-use of the jeweller’s word
‘gold’ in 1650 determined that x satisfies the jeweller’s word ‘gold’ if and
only if x is gold. This consequence of the PJSS-based conception of words
conflicts with the intuition I want to highlight. The intuition implies that
(2) If (1) then the ex-use of the jeweller’s ‘gold’ in 1650 determined that
x satisfies the jeweller’s word ‘gold’ if and only if x is gold.
From (1) and (2), it follows that
(3) The ex-use of the jeweller’s ‘gold’ in 1650 determined that x satisfies
the jeweller’s word ‘gold’ if and only if x is gold.
In short, if one accepts (2), one cannot accept (1) without also committing
oneself to (3). Those who feel they cannot give up (2) will therefore feel
compelled to reject the PJSS-based conception of words, which licenses us
to accept (1) even if we reject (3).
To state the intuition in a more general form, it helps first to recast
the gold example in terms not of one’s PJSSs across time, but of one’s
corresponding practical judgements of sameness of extension across time
(PJSEs). One’s PJSSs across time commit one to corresponding PJSEs
across time by a series of simple steps. First, one can use
(Sats) x satisfies my word ‘ ’ if and only if x is .¹
to specify the extension of one’s word ‘gold’—the set of objects that satisfy
one’s word ‘gold’—as follows:
x is a member of the extension of ‘gold’
iff x ∈ {x: x satisfies my word ‘gold’} [by definition]
iff x ∈ {x: x is gold} [by (Sats)]
iff x is gold.² [by concretion]³

¹ Recall that applications of (Sats) are short for corresponding disquotational definitions of satisfaction
of sentences by sequences of objects. See §3.3.
² ‘Gold’ is a mass term, true of each gold thing, and also of each aggregate of gold things. In contrast,
‘elm’, for example, is what I call a count term (also known as a count noun), true of each elm, but not
of aggregates of elms. Putting ‘elm’ in the blanks of (E) yields: ‘elm’ is true of x if and only if x is (an)
elm. From ‘‘elm’ is true of x if and only if x is (an) elm’, for example, I can infer ‘{x: ‘elm’ is true of
x} = {x: x is (an) elm}’.
³ The terminology is from Quine 1982: 134.
       181

Once one makes this simple derivation, one can see that when one makes a
PJSS across time for another speaker’s word ‘gold’, or for one’s own word
‘gold’ as one used it in the past, one thereby also commits oneself to a
corresponding PJSE across time for the same word. Hence a person who
accepts (1) commits himself to accepting
(1 ) The extension of the jeweller’s word ‘gold’ in 1650 is the same as
the extension of the chemist’s word ‘gold’ today.
The PJSS-based conception of words licenses us to accept (1 ) whether or
not the facts about the ex-use of the jeweller’s word ‘gold’ in 1650 and
the chemist’s word ‘gold’ today determine that (1 ) is true. In contrast, the
intuition I am trying to articulate implies that
(2 ) If (1 ) then facts about the ex-use of the jeweller’s ‘gold’ in 1650
determined that the extension of the jeweller’s word ‘gold’ was the
set of all and only gold things.
From (2 ) and (3 ), it follows that
(3 ) The ex-use use of the jeweller’s ‘gold’ in 1650 determined that the
jeweller’s ‘gold’ was the set of all and only gold things.
Again, if one accepts (2 ), one cannot accept (1 ) without also committing
oneself to (3 ). Those who feel they cannot give up (2 ) will therefore feel
compelled to reject the PJSS-based conception of words, which licenses us
to accept (1 ) even if we reject (3 ).
Behind (2 ) lies the intuition that
(M) For any two word tokens w and w such that the utterance or
inscription of w occurs at some time t before the utterance or
inscription of w , a PJSE across time for w and w is true only if for
some set E,
(a) E is the extension of both w and w , and
(b) the totality of facts about the ex-use of w at times prior to or
identical with t (hence prior to the utterance or inscription of
w ) determines that E is the extension of w.⁴

⁴ By assigning a set as the extension of a predicate, (M) presupposes that the predicate is determinately
true or false of any given object. Hartry Field, among others, rejects this presupposition. I consider
Field’s views in Chapter 7, where I present a revised formulation of (M) that allows for an assignment
of what I call a core extension to a predicate.
182      

Recall that the only restriction on what counts as a fact about the ex-use of
a word token w is that the fact can in principle be described without using
any sentence that expresses a PJSS or PJSE for w. The facts about the ex-use
of a word token w may include, for instance, facts about the dispositions
and functional states, if any, that are causally responsible for a speaker’s
application of w, facts about causal relations that exist between w, other
speakers, objects, and events in the social and physical environment in which
w is uttered, as well as irreducible facts about the intentions of the speaker
who uttered w or about what w means, and even the sorts of facts posited
by a theory that asserts that the ex-use of w determines that w has a Fregean
sense, and thereby determines w’s extension. In short, (M) encompasses all
substantive theories of, or conjectures about, what determines the extension
of a word token w at the time of its utterance or inscription.
The intuition captured by (M) challenges my proposal that we adopt the
PJSS-based conception of words. For the PJSS-based conception commits
us to clause (a) of (M), but not to clause (b) of (M). The PJSS-based
conception implies that a PJSE may be true whether or not the relevant
instantiation of clause (b) is true. We must therefore either reject (M) or
the PJSS-based conception of words.
Faced with this choice, one might be inclined to reject the PJSS-based
conception of words. Against this, in this chapter I shall present a thought
experiment that discredits (M). In this and the next two chapters, I shall
try to show that we are attracted to (M) only because we mistake certain
logical consequences of our PJSSs and PJSEs for commitments to (M). I
conclude in Chapter 8 that we have no grounds for accepting (M). Any
residual temptation we may feel to accept (M) should therefore not prevent
us from adopting the PJSS-based conception of words.

6.2. Methodological Analyticity

To appreciate the problems with (M) that I will raise below, it helps to see
first that most contemporary philosophers of language who are committed
to (M) nevertheless regard their PJSSs and PJSEs as more trustworthy than
particular versions of (M) that conflict with them. Consider, for instance,
the relationship between our PJSEs and the tempting thesis that some of
our sentences are ‘‘analytic’’, in the sense that we cannot abandon them
       183

without changing the subject. Standard versions of this thesis, including

the psychologistic versions of it held by such figures as John Locke and
David Hume, are versions of (M). The most radical version of the thesis is
that we make some of our sentences true by instituting a convention about
how they are to be evaluated.⁵ The main problem with this version of the
thesis is that it conflates truth and belief. As Frege observed in a different
connection, ‘‘Being true is different from being taken to be true, whether
by one or many or everybody, and in no case is to be reduced to it. There
is no contradiction in something’s being true which everybody takes to be
false’’ (Frege 1964: 13).
One way to support the thesis that some of our sentences are analytic
without conflating truth and belief is to derive the thesis from a description
of how we evaluate sentences. To see how this might be done, consider
W. V. Quine’s account of the deviant logician’s predicament. Against the
idea that deviant logicians may ‘‘reject the law of non-contradiction and
accept an occasional sentence and its negation both as true’’, Quine argues
as follows:
[They] think they’re talking about negation, ‘∼’, ‘not’; but surely the notation
ceased to be recognizable as negation when they took to regarding some conjunc-
tions of the form ‘p. ∼ p’ as true, and stopped regarding such sentences as implying
all others. Here, evidently, is the deviant logician’s predicament: when he tries to
deny the doctrine he only changes the subject. (Quine 1986: 81)

The moral is that even though truth is not up to us, for some words,
including ‘∼’, ‘not’, we can agree on criteria that settle whether or not a
speaker is using them to talk about the same subjects that we talk about
when we use them.
This is the truism behind what I call methodological analyticity—the
idea that even though truth is not up to us, there are sentences we

⁵ This version of the thesis goes hand in hand with what Paul Horwich calls ‘‘the strategy of implicit
definition’’, according to which ‘‘terms may be provided with their meanings by the assertion of
statements containing them’’ in such a way that some of the asserted statements could not be abandoned
without changing the extensions of the terms they contain. Horwich rejects this position, for reasons
he explains in chapter 6 of Horwich 1998b. I am sympathetic with Horwich’s objections, but I think
he does not expose the deepest problem with the strategy of implicit definition—that it ignores the
diachronic dimension of our pursuit of truth. My arguments in this section, as well as in §§6.7–6.11,
highlight the diachronic dimension of our pursuit of truth, and indicate how I would argue against the
stronger and even less plausible thesis that we make some of our sentences true by agreeing on how
they are to be evaluated.
184      

cannot reject without changing the subject. The least problematic version
of the idea, due to Rudolf Carnap, makes sense only for sentences of an
artificial language system (Carnap 1990). If we accept Quine’s textbook
explanations of negation and conjunction (symbolized here by ‘∼’ and
‘∧’, respectively), sentences of the form ‘∼(S ∧ ∼S)’ come close to being
‘analytic’ in Carnap’s strict sense of that troublesome word.
The sort of methodological analyticity I want to discuss is also supposed to
be a feature of natural language sentences, such as ‘Bachelors are unmarried
adult males’, and ‘Gold is a yellow metal’. The idea is that we tacitly agree
on criteria that settle whether or not a speaker is using ‘bachelor’ to talk
about bachelors, ‘adult’ to talk about adults, ‘gold’ to talk about gold, and
so on. We tacitly agree that no one can reject ‘Bachelors are unmarried
adult males’ or ‘Gold is a yellow metal’, for instance, without changing the
subject. Moreover, the criteria on which we tacitly agree are in principle
obvious to us without any special empirical investigations and without
awaiting the development of as yet unimagined scientific theories—we
can tell by reflecting on our own present usage of the terms whether or
not an explicit statement of the criteria is correct. If there are any natural
language sentences that are analytic in this sense, then they can play a
methodological role in our inquiries that is analogous to the more strictly
defined methodological role of analytic sentences in Carnap’s artificially
constructed language systems. That is why I call this sort of analyticity
methodological.
One might be inclined to dismiss methodological analyticity (even the
pure form of it that is restricted to artificial languages) with the claim
that for any sentence we accept, we can imagine that it’s false. This claim
may seem to follow immediately from Frege’s distinction between truth
and belief. But that distinction has no direct bearing on whether we can
imagine that a sentence we now accept is false. What would it be, for
instance, to imagine that a sentence of the form ‘∼(S ∧ ∼S)’ is false?
One might reply that even if we cannot imagine that a sentence of the
form ‘∼(S ∧ ∼S)’ is false, we can still employ the strategy just described to
undermine the claim that, for instance, ‘Gold is a yellow metal’ is analytic.
Thus consider the following objection by Saul Kripke:
Could we have discovered that gold was not in fact yellow? . . . Suppose there
were an optical illusion which made the substance appear to be yellow; but, in
       185

fact, once the peculiar properties of the atmosphere were removed, we would see
that it is actually blue. . . . Would there on this basis be an announcement in the
newspapers: ‘It has turned out that there is no gold. Gold does not exist. What
we took to be gold is not in fact gold?’ . . . It seems to me that there would be no
such announcement. On the contrary, what would be announced would be that
although it appeared that gold was yellow, in fact gold has turned out not to be
yellow, but blue. (Kripke 1980: 118)

This objection relies on the supposition that we could have discovered that
gold is not in fact yellow, but blue. A person who thinks that ‘Gold is a
yellow metal’ is analytic might not find that supposition intelligible. How
then might we try to persuade such a person that ‘Gold is a yellow metal’
is not analytic?
One way would be to remind him that we may at one time feel confident
that one could not reject a given statement without changing the subject,
but later realize that we were wrong. He might still insist that he is not
wrong about the analyticity of ‘Gold is a yellow metal’, even if he has
proved wrong about other cases of analyticity. Ultimately, the best way to
undermine his confidence is simply to point out that gold in its pure form
is not yellow, but white. He could dig in his heels and insist that ‘Gold
is a yellow metal’ is true. But to do so would be to reject the PJSSs and
PJSEs on which members of our own community relied when they took
themselves to have discovered that pure (unalloyed) gold is white.
To accept that we discovered that gold is white, we must trust our PJSSs
for the term ‘gold’ more than we trust our previous speculation that one
cannot reject the statement that gold is a yellow metal without changing
the subject. Trusting these judgements, we realize that we might be wrong
about gold—perhaps it’s yellow after all. But we take ourselves to have
discovered that gold is white, and we realize that to do so is to accept our
PJSSs and PJSEs across time for ‘gold’. Our present uses of ‘gold’ link
it to earlier uses of ‘gold’, and those earlier uses of ‘gold’ are linked to
even earlier uses of it. Taken together, these uses of ‘gold’ constitute a
trans-temporal chain of PIWs for ‘gold’ that licenses corresponding PJSSs
and PJSEs across time for ‘gold’. In a similar way, every inquiry brings with
it some chain or other of PJSSs and PJSEs across time.
One might think that this reasoning only shows that we can be radically
wrong about our own tacit criteria for applying our terms, not that
methodological analyticity is incorrect. As I defined it four paragraphs
186      

above, however, methodological analyticity implies that whether or not a

given statement is analytic is in principle obvious to us without any special
empirical investigations or the development of as yet unimagined scientific
theories—we can tell just by reflecting on our own current usage of a term
whether or not a given explicit statement of how it should be applied is
correct. Our discovery that gold is white shows that we can’t tell just by
reflecting on our own current usage of a term whether not a given explicit
statement about how it should be applied will survive a conflict with our
PJSSs. This discovery, and the ever-present possibility of others like it,
undermines methodological analyticity.

6.3. Causal-Historical Theories

The failure of methodological analyticity, combined with an unreflective
trust in our PJSSs and PJSEs and a tacit commitment to (M), led Saul
Kripke and Hilary Putnam to adopt a radically different picture of how
facts about ex-use determine the truth or falsity of our PJSSs and PJSEs.
In Naming and Necessity, Saul Kripke sketched a causal-historical account
of reference for proper names and the extensions of natural kind terms,
including ‘gold’. Hilary Putnam further developed the idea. Both Kripke’s
and Putnam’s causal-historical accounts of what determines the extensions
of proper names and natural kind terms can be seen as a response to the
failure of methodological analyticity—a response to the fact that beliefs
that we take at one time to be definitive of a given topic may later be
revised without changing the topic. In this section I will briefly explain how
Kripke’s and Putnam’s causal-historical accounts of what determines the
extensions of proper names and natural kind terms can be seen as attempts
to make sense of our PJSSs and PJSEs in light of their commitment
to (M).
Kripke’s version of the causal-historical theory of extension should be
distinguished from his thesis that names are rigid designators. Rigidity is a
modal notion defined as sameness of reference in all possible worlds, whereas
the causal-historical theory of extension is intended to explain sameness of
extension across time in the actual world. The motivating intuition behind
the causal-historical theory is the one I emphasized above—the intuition
that there must be something about the use of a term prior to a discovery
       187

that settles its extension. Kripke highlights features of our practice that
suggest that the extension of a natural kind term used by a given speaker is
not settled by the speaker’s beliefs about how to apply it. His account of
how the extension of a kind term is settled has the same structure as his
account of how the reference of a proper name is settled. The following
passage contains his most explicit statement of how the extension of a kind
term is settled:
In the case of proper names, the reference can be fixed in various ways. In an
initial baptism it is typically fixed by an ostension or a description. Otherwise,
the reference is usually determined by a chain, passing the name from link
to link. The same observations hold for such a general term as ‘gold’. If we
imagine a hypothetical (admittedly somewhat artificial) baptism of the substance,
we must imagine it picked out by some such ‘definition’ as, ‘Gold is the substance
instantiated by the items over there, or at any rate, by almost all of them’. (Kripke
1980: 135)

Thus Kripke suggests that the extension of ‘gold’ is fixed by a ‘‘baptism’’

that combines a ‘‘definition’’—‘‘Gold is the substance instantiated by the
items over there, or at any rate, by almost all of them’’—with pointing
gestures that identify items that exemplify the kind of things that we intend
our kind term ‘gold’ to denote. A speaker who was not present at this
‘‘baptism’’ may nevertheless be able to use ‘gold’ to denote gold things,
if she picked up her use of that term from other speakers who have links
with speakers who have links with speakers . . . who were present at the
‘‘baptism’’. The ‘‘links’’ between speakers that Kripke posits must track
what I call practical judgements of sameness of extension for ‘gold’, which
constitute the data Kripke is attempting to explain.
Kripke’s account of how the extension of a kind term is determined
is ambiguous between two different readings. On one reading, Kripke is
saying that a kind term such as ‘gold’ refers to a kind (of stuff ), and treats
kinds as items in the world that we can name. But it is clearer and less
problematic to recast Kripke’s account so that the extension of a kind term
is simply the set of objects of which the kind term is true—bits of gold,
for instance, in the case of ‘gold’.⁶ If we accept this modification, then we

⁶ I propose that we treat natural kind terms as predicates, not as names for kinds. I do not claim that
this way of thinking of natural kind terms captures the way Kripke himself thought about kind terms
when he wrote Naming and Necessity. The crucial consideration is that, as Scott Soames has recently
188      

can recast Kripke’s account as follows: the extension of the term ‘gold’ is
fixed by the following ‘‘ostensive definition’’:
(K) x is gold if and only if for most things y that I and other speakers in
my linguistic community have on other occasions called ‘gold’, x is
(a bit of ) the same substance as y.
The key point is that the question whether a given object x is (a bit of )
the same substance as y is not settled by our current beliefs about x or y.
Yet if we adopt the ostensive definition (K), we can then inquire into what
makes an object x (a bit of ) the same substance as some sample object y.
We may then come to accept that x is (a bit of ) the same substance as some
sample object y if and only if x and y both have the atomic number 79.
Looking back at our previous uses of the word ‘gold’, we can say that all
along it was true of all and only the objects with atomic number 79, even
if we did not know this when we first accepted (K).
Inspired by Kripke’s causal-historical theory of extension, Hilary Putnam
devised a thought experiment that highlights (among other things) the
relationship between our natural kind terms and our external environment.⁷
The details of Putnam’s Twin Earth thought experiment are now very well
known, but I will briefly review them here in preparation for further
applications below.
Suppose there is a planet called Twin Earth which is just like Earth
except that wherever there is water on Earth, there is another substance,
twin water, on Twin Earth. Twin water is indistinguishable from water
in ordinary circumstances, but the molecular structure of twin water is
different from the molecular structure of water. Putnam supposes that just
as we have discovered the molecular structure of water, our contemporaries
on Twin Earth have discovered the molecular structure of twin water. If
earthlings were able to visit Twin Earth, they might at first mistake twin
water for water. But after talking with the chemists on Twin Earth, they will
learn that twin water is not water, even though in ordinary circumstances

explained, ‘‘the natural kind terms with which Kripke is concerned include many expressions that
function primarily as predicates’’ (Soames 2002: 248). This fits with W. V. Quine’s observation that for
predicative uses of mass terms, among them natural kind terms such as ‘water’ and ‘gold’—predicative
uses such as ‘‘That puddle is water’’—‘‘we can view the mass terms . . . as general terms, reading ‘is
water’ . . . in effect as ‘is a bit of water’ ’’ (Quine 1960: 97–8).
⁷ In 1973, Putnam wrote, ‘‘Kripke’s work has come to me second hand; even so, I owe him a large
debt for suggesting the idea of causal chains as the mechanism of reference’’ (Putnam 1973: 198).
       189

it is indistinguishable from water. Twin Earthlings use a language called

Twin English that is almost indistinguishable from English. Just as we
apply the English word ‘water’ to water, they apply their Twin English
word ‘water’ to twin water. Putnam observes that in these circumstances,
we would say that an object x satisfies the English word ‘water’ if and
only if x is (a bit of ) water, not twin water, and an object x satisfies
the Twin English word ‘water’ if and only if x is (a bit of ) twin water,
not water. The extension of the English word ‘water’ is the set of things
(puddles, pools, raindrops, ice cubes, etc.) that are (portions of ) water, and
the extension of the Twin English word ‘water’ is the set of things (twin
puddles, twin pools, twin raindrops, twin ice cubes, etc.) that are (portions
of ) twin water.
Putnam then asks us to ‘‘roll the time back’’ to an earlier time, say
1650, when chemistry was not developed on either Earth or Twin Earth.
Even though the chemical structure of water and twin water was not
yet known, the Twin English word ‘water’ was applied to twin water,
and the English word ‘water’ was applied to water. We take for granted
in the English-speaking community that the extension of our (English)
word ‘water’ has not changed since 1650, and members of the Twin
English-speaking community take for granted that the extension of their
(Twin English) word ‘water’ has not changed since 1650. We (members
of the English-speaking community) take ourselves to have discovered
the molecular structure of water, so we naturally conclude that in 1650
the extension of the English word ‘water’ was the set of things (puddles,
pools, raindrops, ice cubes, etc.) that are (portions of ) water, even though
no member of our community had enough chemical knowledge in 1650
to distinguish water from qualitatively similar liquids, such as twin water.
Members of the Twin English-speaking community take themselves to
have discovered the molecular structure of twin water, so they naturally
conclude that in 1650 the extension of the Twin English word ‘water’ was
the set of things (puddles, pools, raindrops, ice cubes, etc.) that are (portions
of ) twin water, even though no member of their community had enough
chemical knowledge in 1650 to distinguish twin water from qualitatively
similar liquids, such as water. In short, members of both communities take
for granted that the extension of their term ‘water’ did not change since
1650. They trust their PJSEs across time, from 1650 to today, for their
word ‘water’.
190      

Putnam’s version of Kripke’s causal theory of the reference of natural

kind terms is meant to explain and justify these PJSEs across time. Putnam
presents his theory in the following passage:
Suppose I point to a glass of water and say ‘this liquid is called water’ . . . My
‘‘ostensive definition’’ of water has the following empirical presupposition: that
the body of liquid I am pointing to bears a certain sameness relation (say, x is the
same liquid as y, or x is the sameL as y) to most of the stuff I and other speakers in my
linguistic community have on other occasions called ‘water’. If this presupposition
is false because, say, I am without knowing it pointing to a glass of gin and not
a glass of water, then I do not intend my ostensive definition to be accepted.
Thus the ostensive definition conveys what might be called a defeasible necessary
and sufficient condition: the necessary and sufficient condition for being water
is bearing the relation sameL to the stuff in the glass; but this is the necessary
and sufficient condition only if the empirical presupposition is satisfied. (Putnam
1975: 225)

Putnam’s idea is that in 1650, English-speakers ostensively defined the

denotation of the English word ‘water’ as follows: ‘x is (a bit of ) water
if and only if for most things y that I and other speakers in my linguistic
community have on other occasions called ‘water’, x is the same liquid as
y’. This ostensive definition is designed to explain and justify the practical
judgements of sameness of denotation across time, from 1650 to today, of
the English word ‘water’. The key point is that whether or not x is the same
liquid as y ‘‘may take an indeterminate amount of scientific investigation
to determine’’ (Putnam 1975: 225). If members of the English-speaking
community had somehow been transported to Twin Earth in 1650, they
would have believed that bodies of twin water bear the same liquid relation
to bodies of water. This would have been a mistake even though they did
not at the time know enough chemistry to avoid or correct it.
Putnam thinks that a parallel definition explains and justifies the practical
judgements of sameness of extension across time—from 1650 to today—of
the Twin English word ‘water’. In 1650, Twin English-speakers ostensively
defined the extension of the Twin English word ‘water’ as follows: ‘x is (a
bit of ) water if and only if for most things y that I and other speakers in
my linguistic community have on other occasions called ‘water’, x is the
same liquid as y’. Since their linguistic community is different from ours,
and the bodies of liquid to which they apply the word ‘water’ are bodies
of twin water, not water, Putnam thinks that this ‘‘definition’’ determines
       191

in 1650 that their word water denotes (as we would say it) twin water.
Exactly parallel reasoning applies: if members of the Twin English-speaking
community had somehow been transported to Earth in 1650, they would
have believed that bodies of water bear the same liquid relation to bodies
of twin water. This would have been a mistake even though they did not
at the time know enough chemistry to avoid or correct it.
According to Putnam’s version of the causal-historical theory, for
instance, the extension of ‘gold’ in 1650 was fixed by an ‘‘ostensive
definition’’ such as the following:
(P) x is gold if and only if for most things y that I and other speakers in
my linguistic community have on other occasions called ‘gold’, x is
(a bit of ) the same metal as y.
This is slightly more explicit than (K), because the ‘same metal’ relation is
not as vague as the ‘same substance’ relation that (K) relies on. Nevertheless,
whether or not x is (a bit of ) the same metal as some sample y takes an
indeterminate amount of scientific investigation to discover. Putnam thinks
this partly explains how we can now say that the extension of ‘gold’ in
1650 was the set of things with atomic number 79, even though no one in
1650 had the theoretical knowledge even to state or understand this claim.

6.4. A Thought Experiment

I shall now argue that despite its appeal, the causal-historical theory of
reference fares no better than methodological analyticity as an explanation
or justification for our PJSSs and PJSEs across time. Moreover, the problem
I shall raise for the causal-historical theory cannot be solved by any version
of (M). I shall present a thought experiment that shows that even our most
deeply trusted PJSSs and PJSEs across time are incompatible with (M).
The historical background for the thought experiment is that platinum
was not discovered until the mid-eighteenth century, when chemists called
it ‘white gold’ because of its striking similarities to what they previously
called gold. Platinum has a higher melting point than gold. But platinum
and gold have a similar resistance to attack by acids.⁸ In 1650, a chemist

⁸ See Crosland 1962: 97. Crosland writes that platinum was compared with gold ‘‘because of several
similarities including its resistance to attack by acids’’, especially aqua regia.
192      

applying the best current acid tests to a sample of platinum might have
concluded that it should be called ‘gold’.⁹ We now know that platinum and
gold are different elements: platinum is the element with atomic number
78, and gold is the element with atomic number 79.
Against this background, my thought experiment proceeds as follows.
Suppose that there is a Twin Earth that is indistinguishable from Earth up
until 1650, when large deposits of platinum are uncovered in Twin South
Africa, and that once it is established by Twin Earth chemists that the
newly uncovered metal passes the best current acid tests for gold, members
of the Twin English-speaking community call it ‘gold’, treating it in the
same way we treat gold: the platinum is mined as gold, hammered (and
later melted) together with gold to produce coins and bars that are valued
by Twin Earthlings just as we value gold. Everyone on Twin Earth trusts
the Twin Earth chemists’ judgement that the newly uncovered metal is
properly called ‘gold’.
Suppose also that on Twin Earth chemistry develops in almost exactly
the same way in which it develops in our community, except that when
chemists in the Twin Earth community investigate what they call ‘gold’,
they conclude that there are two kinds of ‘gold’: one of these kinds of
‘gold’ is gold, which they know only as the element with atomic number
79, and the other is platinum, which they know only as the element with
atomic number 78. Long after their discovery, we visit their planet, and
learn their language. We conclude that their word ‘gold’ is true of an object
x just in case x is gold, the element with atomic number 79, or platinum,
the element with atomic number 78, or a mixture of the two.
Note that until 1650, the linguistic dispositions and mental states of
members of the two communities, as well as the relations the members
bear to their respective environments, are virtually the same.¹⁰ Moreover,
until 1650, the physical constitution of the things to which members of the

⁹ Crosland reports that in 1752, a Swedish chemist named Scheffer concluded that the close
similarity of (what we now call) platinum to gold justifies the claim that (what we now call) platinum
is white gold (Crosland 1962: 97). Crosland also points out that ‘‘the distinct nature of new substances
was not always easy to demonstrate by elementary analytical methods and the skeptics could always
maintain that any apparent discovery was really a substance previously known . . .’’ (Crosland 1962:
97–8), and that among alchemists in the mid-eighteenth century, platinum ‘‘was considered to be a
kind of gold’’ (ibid. 235, n. 33).
¹⁰ This judgement presupposes, as demanded by (M), that we describe these dispositions, states, and
relations independently of future developments in either linguistic community. In Chapter 8 I will
consider and reject a modified version of (M) according to which we must describe the community
       193

two communities are causally related is also the same. For we supposed that
Twin Earth is just like Earth with a slightly different future after platinum is
first uncovered in Twin South Africa in 1650. To see the possibility of this
Twin Earth scenario, it is enough to imagine a few accidental differences
between the two communities—such as a chance decision in the Twin
Earth community of a few individuals to dig for gold in the hills where
platinum lies hidden—that allow for the accidental uncovering of large
amounts of platinum in the Twin Earth community.¹¹
The crucial point is that just as members of our English-speaking
community take for granted that the extension of the English word ‘gold’
did not change as a result of the discovery that it is true of x if and only
if x is (a bit of ) the element with atomic number 79, so members of
the Twin English-speaking linguistic community take for granted that the
extension of their Twin English word ‘gold’ did not change as a result
of their discovery that it is true of x if and only if x is (a bit of ) the
element with atomic number 78 or x is (a bit of ) the element with atomic
number 79 (or a mixture of the two). Members of the two communities
assign different extensions to their respective tokens of g-o-l-d, even when
they are evaluating utterances made in 1650—before the uncovering of
platinum in the Twin Earth community—by speakers using sentences that
contain tokens of g-o-l-d.
To highlight this strange consequence of the thought experiment, let
us suppose that in 1650 John Locke and his twin on Twin Earth both

members’ dispositions, states, and relations in a way that includes future developments in their respective
linguistic communities.
¹¹ This thought experiment is similar to Mark Wilson’s Druid thought experiment: ‘‘A B-52
full of regular American types landed on their uncharted Island and the Druids exclaimed, ‘Lo, a
great silver bird falleth from the sky.’ . . . [After this event] . . . the extension of the predicate ‘is a
bird’ for the cosmopolitan Druidese is something like the set of flying devices (including animal
varieties). . . . [But] . . . If the hapless aviators had crashed in the jungle unseen and were discovered by
the Druids six months later as they camped discontentedly around the bomber’s hulk, their Druid
rescuers would have proclaimed, ‘Lo, a great silver house lieth in the jungle.’ . . . [In this alternative
linguistic community] airplanes are no longer [read: are not] held to be ‘birds’. . . . Which extension
should be assigned to ‘bird’ in cosmopolitan Druidese thus depends upon the history of the introduction
of B-52’s to the island . . .’’ (Wilson 1982: 549–50). See also Dennett 1987: 312. Despite the similarities
of Wilson’s and Dennett’s (much briefer) thought experiments, however, neither Wilson nor Dennett
emphasize what I regard as crucial to my thought experiment: there may be two isolated communities
whose ex-uses of a term are the same up until a given time, and then diverge later, while members of
both communities continue to make and to trust their PJSSs and PJSEs for previous uses of those terms
in their respective linguistic communities. I shall highlight this feature of my thought experiment in
this and the next two chapters.
194      

uttered the sentence ‘There are huge deposits of gold in those hills’,
with Locke indicating South African hills, and Twin Locke indicating the
corresponding Twin South African hills, both of which contain platinum
but no gold. We take Locke’s word ‘gold’ to be true of an object x
just in case x is gold, whereas members of the Twin Earth community
take Locke’s word ‘gold’ to be true of an object x just in case (as we
would say it) x is either gold or platinum (or a mixture of the two). We
conclude that Locke’s utterance is false, and our contemporaries in the
Twin Earth linguistic community conclude that Twin Locke’s utterance
is true. And yet by supposition, in 1650, before the uncovering of large
deposits of platinum on Twin Earth, Locke’s and Twin Locke’s linguistic
dispositions and mental states, as well as the non-semantic relations that
Locke and Twin Locke bear to their respective external environments and
the physical constitutions of the things to which Locke and Twin Locke
are causally related, are the same.¹²

6.5. The Standard Conception of the Options

for the Thought Experiment
The following diagram displays the standard conception of the options
for characterizing the extensions of tokens of g-o-l-d as used by members

¹² This way of illustrating the odd consequences of the first thought experiment is adapted from
Donnellan 1983: 103. Lance and O’Leary-Hawthorne 1997: 44–54 also present a thought experiment
in which a single sentence—‘There are witches’—is evaluated as true by one group of speakers and
false by another. But I find their witch thought experiment much less plausible than my gold–platinum
thought experiment, for two related reasons. First, it concerns hypothetical, deliberate translations of the
term ‘witch’, as tokens of that term were ex-used in Salem, Massachusetts, in the 1600s, into two different
languages used at some later time by linguistic communities that are unconnected by chains of PJSSs or
PJSEs across time to tokens of the Salemites’ term ‘witch’. (In my gold–platinum thought experiment,
in contrast, there are no deliberate translations of previous uses of ‘gold’ in either community; members
of each community simply trust their own chains of PJSSs and PJSEs across time for their respective
words.) Second, Lance and O’Leary-Hawthorne regard it as ‘‘clear’’ that a linguist from either of the
later communities they describe would translate the Salemites’ term ‘witch’ into her own language
‘‘homophonically’’ (Lance and O’Leary-Hawthorne 1997: 47). I don’t find this plausible, given the
supposition that the linguists’ uses of their term ‘witch’ are not connected to the Salemites’ tokens of
‘witch’ by chains of PJSSs or PJSEs across time. Finally, Lance and O’Leary-Hawthorne suggest that
‘‘either translation would be correct from the perspective in which they would be given’’ (Lance and
O’Leary-Hawthorne 1997: 51), thereby endorsing the idea that translation, hence also truth value and
extension, are in some way perspective relative. I investigate and reject a sophisticated version of this
kind of view in §8.8.
       195

of the two linguistic communities described in the first thought experi-

ment:¹³
E2
b
E1
a d

c
E3

Line ab represents the ex-use of tokens of g-o-l-d in our English-speaking

community, line ac represents the ex-use of tokens of g-o-l-d in the
Twin English-speaking community, and line ad, where lines ab and ac
overlap, represents the supposition that before the accidental uncovering
of large amounts of platinum on Twin Earth in 1650—an event rep-
resented by point d—members of the two communities have virtually
the same dispositions to ex-use of tokens of g-o-l-d, associate the same
mental states with their tokens of g-o-l-d, and bear the same relations
to the things and substances in their environment, provided that these
dispositions, states, and relations are described independently of future
developments in either linguistic community. ‘E1 ’, ‘E2 ’, and ‘E3 ’ represent
the extension of tokens of g-o-l-d at various points in the histories of the
two communities.
Note that ‘E1 ’ is the only symbol that represents the extension of tokens
of g-o-l-d during the period before the accidental uncovering of large
amounts of platinum on Twin Earth—the period when by supposition the
ex-use of tokens of g-o-l-d is the same in both communities. In this way
the standard commitment to (M) is built into the diagram.¹⁴ The rest of the
diagram is straightforward: the extension of tokens of g-o-l-d in the actual
English-speaking community after the branching is represented by ‘E2 ’, and
the extension of tokens of g-o-l-d in the counterfactual English-speaking
community after the branching is represented by ‘E3 ’.
Given this diagram and the natural assumption that E2 = E3 —the
extension of tokens of g-o-l-d is different in the two linguistic communities

¹³ This diagram (but not my thought experiment or my interpretation of the diagram) is adapted
from Larson and Segal 1995: 534.
¹⁴ Recall that I understand the ex-use of ‘gold’ in 1650 to encompass all non-semantic relations that
speakers in 1650 bore to speakers who used ‘gold’ before 1650, and the physical constitution of the
things to which these earlier speakers were causally related.
196      

after the branching—it appears that there are only three options for
characterizing the extension of tokens of g-o-l-d before the branching:
either (1) E1 = E2 and E1 = E3 , or (2) E1 = E3 and E1 = E2 , or (3)
E1 = E2 and E1 = E3 .¹⁵

6.6. A Preview of why Options (1) and (2)

are Unacceptable
Options (1) and (2) are both unacceptable for symmetrical reasons. To see
why, suppose first that E1 = E2 —the extension of the word type ‘gold’
in English before the branching is the same as the extension of word type
‘gold’ in English today. This accords with our practical judgement that
the extension of our word type ‘gold’ has remained the same despite our
discovery that gold is the element with atomic number 79. Since sameness
of extension is an equivalence relation, and by supposition E2 = E3 , we
must conclude that E1 = E3 , and hence we must accept option (1).
The trouble is that members of the Twin English-speaking community
take for granted that the extension of their word type ‘gold’ has not
changed either. If we accept option (1), we are committed to saying
they are wrong. But by supposition, before the branching, there is no
difference in external environment or in linguistic dispositions or mental
states, described independently of the branching futures. It seems that
nothing about the ex-use of tokens of g-o-l-d before the branching determines that
we are right and they are wrong.
It is difficult to accept this conclusion. We are strongly inclined to
think that there is some independent explanation or justification for our
conviction that our PJSEs across time for our tokens of g-o-l-d are correct
and their PJSEs across time for their tokens of g-o-l-d are incorrect. Should
we trust this inclination? To find out, we must look carefully at the most
promising strategies for explaining or justifying our PJSEs across time for
our tokens of g-o-l-d, beginning with the causal-historical theory that I
sketched above.

¹⁵ This abstract characterization of the options (but not my thought experiment or my interpretation
of the options) is also adapted from Larson and Segal 1995: 534.
       197

6.7. A Dilemma for the Causal-Historical Theory

One might think that one could appeal to the causal-historical theory
to explain or justify our PJSEs across time for our tokens of g-o-l-d,
and thereby to defend option (1). To explore this idea, suppose that in
1650 members of both linguistic communities affirmed the Kripke-style
‘‘ostensive definition’’ (K):
(K) x is gold if and only if for most things y that I and other speakers in
my linguistic community have on other occasions called ‘gold’, x is
(a bit of ) the same substance as y.
Does it follow that on both Earth and Twin Earth, it was already determined
in 1650 that the extension of tokens of g-o-l-d is the set of all and only (bits
of ) gold—i.e. (bits of ) the element with atomic number 79? The answer,
as I shall now try to show, is ‘‘No.’’
To see why, note first that to explain our practical judgement that
the extension of the English word ‘gold’ did not change since 1650 by
appealing to (K), we must assume that
(A) for all x and y, if x and y are gold, then x is (a bit of ) the same
substance as y
is true in English and Twin English. But even if (A) is true in Twin English,
and members of the Twin English-speaking community accept (A), it may
be that for some x and y, ‘x is (a bit of ) the same substance as y’ is true
in Twin English of the ordered pair (x, y), but x does not have the same
atomic number as y. One might try to rule this out by stipulating that
(B) for all x and y, if x is (a bit of ) the same substance as y, then x has
the same atomic number as y
is true in both in English and Twin English. One problem with this
strategy is that in 1650 no one was in a position to formulate (B), since
analytical chemistry had not yet been developed. A deeper problem is
that even if we suppose that (B) is true in English and Twin English,
and that members of both communities accept (B), we have no guarantee
that (A) is true in Twin English. One may ‘‘ostensively define’’ ‘gold’ by
198      

affirming (K) without thereby guaranteeing that (A) is true. Members of the
Twin English-speaking community described in the thought experiment
may conclude that despite their previous affirmation of (K), gold is not a
substance, and so (A) is false.
Alternatively, they may reject (B), and conclude that the ‘‘substance’’
that was picked out by their affirmation of (K) in 1650 was platigold, not the
element with atomic number 79 (which we call gold). They may take for
granted that tokens of their word ‘gold’ are satisfied by all and only bits of
some particular kind of ‘substance’, and conclude that since x satisfies their
word ‘gold’ x if and only if x is (a bit of ) the element with atomic number
79 or x is (a bit of ) the element with atomic number 78 (or a mixture of
the two), not all ‘substances’ map neatly onto the periodic table.
It is natural to suppose that just as members of both linguistic communities
take for granted that the extension of their term ‘gold’ has not changed
since some time before 1650, so members of both linguistic communities
take for granted that the extension of their expression ‘x is (a bit of )
the same substance as y’—the set of ordered pairs of which ‘x is (a bit
of ) the same substance as y’ is true in their language—has not changed
since some time before 1650. Under the circumstances described in the
previous paragraph, the extensions of ‘x is (a bit of ) the same substance as
y’ in English and Twin English are different yet equally compatible with the
affirmation of (K) prior to the branching in 1650.¹⁶
One might think it strange to suppose that the Twin English word
‘gold’ was true of samples of gold and platinum even before the accidental
uncovering of large amounts of platinum, when tokens of g-o-l-d were
actually applied only to samples of gold, and not to samples of platinum.
But this possibility does not seem so strange when one focuses on the
practical and social aspects of the use of a term. The sense of strangeness all
but disappears when we see that there are terms whose actual use resembles
the use of tokens of g-o-l-d in the Twin Earth community described
above. To take one famous (but superficially understood) example, our
word ‘jade’ is true of both jadeite and nephrite. The Chinese character yu

¹⁶ Under the circumstances described in the first thought experiment, the mental states, including
the recognitional capacities and linguistic dispositions, of the speakers in 1650, do not determine the
extension of ‘x is (a bit of ) the same metal as y’. For examples of theories that are undermined by this
observation, see Berger 2002, Brown 1998, Devitt and Sterelny 1987: 72–5, Horwich 1998b, Kitcher
1982: 341–2, and Kitcher 1993: 79–89.
       199

that we translate as ‘jade’ was actually applied only to nephrite until the
eighteenth century, when the Chinese first encountered jadeite and started
carving it. The mineralogical differences between nephrite and jadeite were
discovered in 1863, a century after the Chinese practice of applying yu to
both nephrite and jadeite became entrenched (Hansford 1968: 26–9). They
(and we) take for granted that the extension of the term yu did not change
when it was applied to jadeite. Similarly, Twin English-speakers take for
granted that the extension of their term ‘gold’ did not change when it was
applied to the element with atomic number 78.¹⁷
These considerations show that affirmations of (K) can settle the extension
of tokens of g-o-l-d only if we know that (K) could not turn out to be false.
As I emphasized earlier, however, we trust our PJSEs across time, so we
have no guarantee that any given sentence, including (K), is analytic, in
the methodological sense that we cannot revise it without changing the
subject.¹⁸ I conclude that affirmations of (K) in 1650 do not rule out either
the Earthlings’ or Twin Earthlings’ discoveries about the nature of the
objects that satisfy their tokens of g-o-l-d, and therefore do not justify the
PJSEs for tokens of g-o-l-d in either linguistic community.

¹⁷ My argument in this and the previous four paragraphs challenges the thesis that the extension of
a natural kind term such as ‘gold’ is determined by indexical applications of such descriptions as ‘gold
is whatever bears the same metal relation to the stuff which I and other speakers in my linguistic
community typically call ‘gold’.’ Hilary Putnam apparently endorses this thesis in, for instance, Putnam
1975: 225, and Putnam 1988: chapter 2. But Putnam simply takes for granted that the extension of
‘gold’ in previous centuries was the same as the extension of our word ‘gold’. In describing individuals
who at some previous time ‘‘mistakenly’’ apply the word ‘gold’ to an alloy, Putnam asserts that ‘‘what
they meant by ‘gold’ was what we mean by ‘gold’ ’’ (Putnam 1988: 37), but he offers no account,
independent of our actual judgements of sameness of extension across time, of when the extension
of a term used at one time is the same as the extension of a term used at some other time. If, as I’ll
argue later, our best grip on sameness of extension across time is given by our practical judgements
of sameness and difference of extension across time, then my first thought experiment shows that the
extension of ‘gold’ was not determined by the way it was ex-used in 1650.
¹⁸ Similarly, we have no guarantee that a ‘‘recognitional capacity’’ that we associate with a given
term determines the extension of the term, in this sense proposed by Jessica Brown in Brown 1998. On
any non-semantic account of a ‘‘recognitional capacity’’, the members of the Earth and Twin Earth
linguistic communities in my thought experiment associate the same ‘‘recognitional capacity’’ with
their tokens of g-o-l-d in 1650. Yet they later characterize the extension of their tokens of g-o-l-d
differently, and revise the ‘‘recognitional capacities’’ they respectively associate with their tokens of
g-o-l-d, without taking themselves to be changing the subject. If we take our practical judgements
of sameness of denotation across time as our best guide to when we have changed the subject and
when we haven’t, then my thought experiment undermines Brown’s theory of what determines the
extension of natural kind terms. For similar reasons, it also undermines Alan Berger’s theory that
the extension of certain terms, such as ‘gold’, is determined in part by what a baptizer of the term
‘‘focuses’’ on. For Berger’s theory of the reference-determining role of what he calls ‘‘focusing’’, see
Berger 2002.
200      

One natural response to this argument is that (K) is too vague to settle
the extension of ‘gold’. This might lead one to replace the vague word
‘substance’ with the more precise word ‘metal’, for example, and to
affirm (P).
(P) x is gold if and only if for most things y that I and other speakers in
my linguistic community have on other occasions called ‘gold’, x is
(a bit of ) the same metal as y.
Suppose that in 1650 members of both linguistic communities affirmed
(P). Does it follow that on both Earth and Twin Earth, it was already
determined in 1650 that the extension of ‘gold’ is the set of all and only
(bits of ) gold—i.e. (bits of ) the element with atomic number 79? The
answer, once again, is ‘‘No.’’
To explain our practical judgement that the extension of the English
word ‘gold’ did not change since 1650 by appealing to (P), we must assume
that
(C) for all x and y, if x and y are gold, then x is (a bit of ) the same metal
as y
is true in English and Twin English. But even if (C) is true in Twin English,
it may be that for some x and y, ‘x is (a bit of ) the same metal as y’ is true
in Twin English of the ordered pair (x, y), but x does not have the same
atomic number as y. One might try to rule this out by stipulating that
(D) for all x and y, if x is (a bit of ) the same metal as y, then x has the
same atomic number as y
is true in both English and Twin English. But if members of both linguistic
communities accept (D) they may reject (C). If for some reason they stick
with (C), then they may reject (D). Their trust in their PJSEs across time
for ‘gold’ may lead them to re-evaluate a number of beliefs they formerly
expressed by using their term ‘gold’. In such cases, none of their ante-
cedent commitments—even their supposed commitment to (P)—shows
that these re-evaluations are incorrect. This is just another application
of the lesson learned from the counter-examples to methodological
analyticity.
One could try to state an ‘‘ostensive definition’’ of ‘gold’ that is more
precise than (P). The trouble is that the more precise one’s ‘‘definition’’ of
       201

‘gold’ is, the more likely it is that it will later be rejected without changing
the extension of ‘gold’. The causal-historical theory of extension therefore
faces a dilemma: the more informative the supposed ostensive definitions
are, the more likely it is that they will later be revised without changing
the subject; but the less informative they are, the less likely it is that there is
only one way of correctly applying them. I conclude that even if, contrary
to what most philosophers believe, sentences such as (K) or (P) of §6.3 were
actually affirmed in 1650, these supposed ‘‘ostensive definitions’’ could not
rule out either the Earthlings’ or Twin Earthlings’ discoveries about what
their word form ‘gold’ denotes, and therefore could not explain or justify
our PJSEs for ‘gold’. Despite appearances, causal-historical theories do not
support option (1).

6.8. Dispositions
Perhaps we can support option (1) by supplementing the assumption
that members of both communities affirmed (K) or (P) before 1650 with
facts about their linguistic dispositions. Putnam uses this strategy when
he tries to defend his claim that our discovery that gold is the element
with atomic number 79 implies that some of Archimedes’ assertions about
gold—assertions made by using a Greek word that we now translate as
‘gold’—might have been false. Putnam explains his claim as follows:
[suppose] there were or are pieces of metal . . . which we can distinguish from
gold quite easily with modern techniques. Let X be such a piece of metal. Clearly
X does not lie in the extension of gold in standard English; my view is that
it did not lie in the extension of [the Attic Greek word we translate as ‘gold’]
either, although an ancient Greek would have mistaken X for gold . . . When
Archimedes asserted that something was gold . . . he was not just saying that it had
the superficial characteristics of gold . . . ; he was saying that it had the same general
hidden structure (the same ‘essence’, so to speak) as any normal piece of local gold.
Archimedes would have said that our hypothetical piece of metal X was gold, but
he would have been wrong. (Putnam 1975: 235–6)

Putnam is aware that it may seem as though we are imposing our current
standards onto our description of Archimedes’ hypothetical claim that X
was gold. To try to dispel this impression, Putnam relies on a counterfactual
202      

claim about how Archimedes would have reacted to our experiments and
theories if he had been exposed to them:
. . . there are a host of situations that we can describe (using the very theory that
tells us that X isn’t gold) in which X would have behaved quite unlike the
rest of the stuff Archimedes classified as gold. Perhaps X would have separated
into two different metals when melted or would have had different conductivity
properties, or would have vaporized at a different temperature, or whatever. If
we had performed the experiments with Archimedes watching . . . he would have
been able to check the empirical regularity that ‘X behaves differently from the
rest of the stuff I’d classify as [gold] in several respects’. Eventually he would
have concluded that ‘X may not be gold’. . . . If, now, we had gone on to inform
Archimedes that gold had such and such a molecular structure (except for X), and
that X behaved differently because it had a different molecular structure, is there
any doubt that he would have agreed with us that X isn’t gold? (Putnam 1975:
237–8)

This thought experiment is designed to convince us that an individual’s

‘‘ostensive definitions’’ of the denotations of his natural kind terms implicitly
express theoretical commitments that he would explicitly endorse if he
were presented with the relevant evidence and theories. Thus Putnam
suggests that ‘‘ostensive definitions’’ of the denotations of natural kind
terms, together with facts about what speakers would say if they were
presented with new theories or evidence, explain and justify many of our
practical judgements of sameness of denotation across time for natural kind
terms.
Applying this reasoning to the thought experiment above, one might
think that we can support option (1) by citing Locke’s and Twin Locke’s
affirmation of (P), together with facts about what they are disposed to say
if they are presented with the evidence we now have for the claim that x
is gold if and only if x is (a bit of ) the element with atomic number 79.
There are two serious problems with this strategy. The first problem is that
whether or not an individual would accept or reject certain sentences may
depend on the order in which he is presented with evidence that supports
those sentences.¹⁹ It is plausible to suppose that Locke and Twin Locke
would have affirmed the sentence ‘x is gold if and only if x is (a bit of )

¹⁹ This formulation of the challenge my thought experiment poses for dispositional theories is due to
Bill Robinson. The point was already implicit in Mark Wilson’s Druid thought experiment (see n. 11).
       203

the element with atomic number 79’ if they had been presented with the
same evidence that later English-speakers encountered, in the same order
in which they actually encountered it. But by hypothesis members of
the Twin English-speaking community described in the gold–platinum
thought experiment came to affirm the sentence ‘x is gold if and only if
either x is the element with atomic number 79 or x is the element with
atomic number 78 (or a mixture of the two)’ without thinking at any point
that they had changed the extension of their term ‘gold’. It is therefore
plausible to suppose that Locke and Twin Locke would have affirmed
the sentence ‘x is gold if and only if either x is the element with atomic
number 79 or x is the element with atomic number 78 (or a mixture of
the two)’ if they had been presented with the same evidence that later
Twin English-speakers encountered, in the same order in which they actually
encountered it. We have no independent grounds for saying that one of
these presentations of the evidence is correct and the other is incorrect, and
so an appeal to dispositions cannot show that our community’s PJSSs and
PJSEs for ‘gold’ are correct and theirs are incorrect.²⁰
The second and deeper problem with this strategy is that it is unclear
how a speaker’s dispositions could settle sameness of extension across
time. Our understanding of whether or not Locke and Twin Locke
had dispositions to agree or disagree with later investigators in the two
communities depends on our prior judgement of whether their terms have
the same extension as the similarly spelled terms used by later investigators in
the two communities. There is no obvious criterion in terms of dispositions
alone, apart from our PJSEs, for settling when two individuals are using a
term with the same extension. One difficulty for any proposed criteria is that

²⁰ The observations in this section suggest an argument against Paul Horwich’s use theory of
meaning. According to that theory, the meaning of a term is constituted by its possession of a certain
‘‘acceptance property’’ that can be specified independently of its meaning or denotation. Candidates
for such properties are facts about a speaker’s linguistic behaviour, in particular, facts about which
sentences the speaker is disposed to accept under various circumstances. For instance, according to
Horwich, ‘‘the acceptance property that governs a speaker’s overall use of ‘and’ is (roughly) his
tendency to accept ‘p and q’ if and only if he accepts both ‘p’ and ‘q’ ’’ (Horwich 1998b: 45). Moreover,
according to Horwich, two words express the same concept if they have the same basic acceptance
property (Horwich 1998b: 46), and any two words that express the same concept must have the same
denotation (Horwich 1998b: 69). Since Horwich can only appeal to linguistic dispositions, described
in non-semantic terms, he is apparently committed to saying that the term ‘gold’ expresses the same
concept, and therefore has the same denotation, in 1650 in both of the linguistic communities of
my thought experiment. This aspect of his use theory clearly conflicts with our confidence that the
extension of our term ‘gold’ today is the same as the extension of ‘gold’ in 1650.
204      

individuals sometimes stubbornly disagree with each other about issues that
are important to both of them. This suggests that Locke’s and Twin
Locke’s dispositions to accept or reject certain sentences, even if they were
well defined and perfectly objective, do not by themselves determine how
we should evaluate our PJSEs in the two linguistic communities.
Suppose we could somehow objectively determine that Locke and Twin
Locke are disposed to accept the sentence ‘x is gold if and only if x is (a
bit of ) the element with atomic number 79’ and disposed to reject the
sentence ‘x is gold if and only if either x is the element with atomic
number 79 or x is the element with atomic number 78 (or a mixture
of the two)’, even after they are presented with the evidence on which
members of the Twin English-speaking community after 1850 based their
conclusion that ‘gold’ is true of (bits of ) the elements with atomic number
79 or (bits of ) the elements with atomic number 78. If we accept the
PJSEs across time in both communities, we should conclude that Locke
is right, but Twin Locke is wrong. Hence the supposed dispositions
cannot by themselves show that we should not accept the PJSEs in both
communities.

6.9. Epistemic Possibilities and Primary Intensions

Frank Jackson and David Chalmers have recently developed a framework
for relating ostensive definitions of the sort exemplified by (K) and (P) to
individual speakers’ judgements about how they would apply kind terms
under various possible circumstances. They both present their view as a
justification of what I call our practical judgements of sameness of extension
(PJSEs) across time for natural kind terms, such as ‘water’ and ‘gold’. Their
view is more complicated than the two I have examined so far, but it fails
for similar reasons. Because of its greater complexity, it is more difficult to
see exactly where it goes wrong. In this section I will briefly sketch their
view, beginning with an account of the underlying motivation for it. In
the next section I will argue that their view cannot support option (1), and
is actually in direct conflict with it.
Jackson and Chalmers are not always explicit about the underlying
motivation for their view. I assume, however, that Jackson speaks for both
of them when he writes that
       205

Language . . . is a convention generated set of physical structures that has as a

principle function making it easy to articulate, and in consequence easy to
record, transmit in communication, debate the consequences of, and so on, how
someone . . . takes things to be. ( Jackson 1998b: 201–2)

This conception of language implies that when we rely on our PJSSs and
PJSEs, we take for granted that they enable us to report accurately how our
fellow English-speakers take things to be. In addition, on this view, how
our fellow English-speakers take things to be is independent of our PJSSs
and PJSEs. According to Jackson and Chalmers, a speaker’s beliefs about
how to apply the term settle how she takes things to be when she applies
it. As Jackson explains it, ‘‘. . . a name [or kind term] T used by S refers to
whatever has the properties that S associates with T ’’ ( Jackson 1998b: 203).
This is not just one plausible account among others, according to Jackson;
any alternative theory according to which the reference of a name [or kind
term] T used by S is not determined by the properties that S associates with
T leaves reference unexplained and mysterious:
. . . the crucial point here, and generally, it is that our classifications of things into
categories—grooming behavior, belief, pain, and so on and so forth—is not done
at random and is not a miracle. There are patterns underlying our conceptual
competence. They are often hard to find—we still do not know in full detail the
rules that capture the patterns underlying our classification of sentences into the
grammatical and the ungrammatical, or of inferential behavior into the rational
and the irrational—but they must be there to be found. We do not classify sentences as
grammatical, or inferential behavior as rational, by magic or at random! ( Jackson 1998b:
64, my emphasis)

In this passage Jackson emphasizes that our linguistic behaviour is guided

by our beliefs about how to apply our terms. In his view, language
itself is merely a conventional device for conveying beliefs, and hence
any adequate account of a speaker’s linguistic competence must explain
it without presupposing a public language, solely in terms of beliefs that
speakers associate with their words.
Jackson and Chalmers are aware that this conception of language seems
at first to be refuted by Putnam’s and Kripke’s arguments against the
description theory of reference. In response, Jackson and Chalmers offer a
new interpretation of what Kripke’s and Putnam’s arguments show. Jackson
and Chalmers first introduce a criterion for determining what properties
206      

a speaker S associates with a given term T. The key assumption behind

this criterion is that before one can even begin to make empirical inquiries about a
particular kind T, one must first be able to identify a particular item as a sample
of T, without presupposing any empirical beliefs. To discover how we identify
items to which a given term applies, they believe, we must ask ourselves
what we would say in various different epistemically possible circumstances.
When they characterize this epistemological task, Jackson and Chalmers
presuppose what has now become a widely accepted account of epistem-
ic possibility. Putnam’s method of describing Twin Earth scenarios (as
described, for instance, in §6.3 above) suggests that for each person we can
specify countless subjectively equivalent circumstances in which everything
seems the same, but the external environments are different from what they
are on Earth. Putnam assumed that Twin Earth actually exists. Nevertheless,
if I try to imagine that I am my twin on Twin Earth, I am in effect imagining
that I am in a different (but subjectively equivalent) possible world. The
difference between the actual world and this other possible world in this
case is simply the difference between occupying my current perspective,
and occupying the perspective of my twin on Twin Earth. More generally,
to specify an epistemic possibility we must specify a subjectively equivalent
world centred on a particular agent. I will call these agent-centred subjectively
equivalent possible worlds. On one standard view, epistemic possibilities
are explained in terms of agent-centred possible worlds.
This way of thinking about epistemic possibility yields a corresponding
understanding of a priority: a speaker’s belief is a priori if it is true in all
subjectively equivalent agent-centred worlds. According to this way of
thinking about epistemic possibility, a speaker’s belief is empirical if it is
not a priori.
David Chalmers (Chalmers 1996) defines the primary intension of a word
as a special sort of function from (agent-centred) worlds to extensions: in
a given (agent-centred) world w, the primary intension of a word picks
out what the extension of the word would be if w turned out to be
actual (Chalmers 1996: 57). To grasp the primary intension of ‘water’, for
instance, we must grasp a function that yields the set of all and only portions
of water as value if the actual (agent-centred) world has water in its rivers,
lakes, and oceans, but yields the set of all and only portions of twin water as
value if the actual (agent-centred) world has twin water in its rivers, lakes,
and oceans.
       207

What is distinctive of a primary intension, according to Chalmers, is that

our grasp of it is independent of any assumptions we make about which
of our agent-centred subjectively equivalent worlds we are actually in. He
argues that there must be a primary intension for any word that we can use
to express a discovery. If we are to express a discovery about water that is
based on our examination of a given sample of what we take to be water,
he reasons, we must be able to say why it counts as a sample of water by
appealing to rules that we can grasp without going through any empirical
investigation or presupposing any empirical beliefs (Chalmers 1996: 62).
This reasoning moves from a truism to a substantive epistemological
claim. The truism is that we take ourselves to express a discovery about
water by using the term ‘water’ only if we make a PJSE for ‘water’, and
thereby suppose that the extension of ‘water’ does not change as a result of
our supposed discovery. The substantive epistemological claim is that we
are entitled to take ourselves to express a discovery about water by using the
term ‘water’ only if we can justify our PJSEs for ‘water’ by appealing to rules
that we can grasp without going through any empirical investigation or
presupposing any empirical beliefs. This substantive epistemological claim
is a natural consequence of the conception of language that motivates the
Jackson–Chalmers account of primary intensions. The fundamental point
is that all linguistic relationships between speakers, both at a given time and
across time, must ultimately be grounded in the speakers’ beliefs about how
their terms are correctly applied.
A primary intension is well suited to this justificatory role, according
to Chalmers. ‘‘The intension specifies how reference depends on the way
the external world turns out, so does not itself depend on the way the
external world turns out’’ (Chalmers 1996: 57). By reasoning about ‘‘what
our words would refer to if the actual world turned out in various ways’’,
Chalmers thinks, we can simultaneously see that our words have primary
intensions and discover what they are.
According to Jackson and Chalmers, the existence of primary intensions
for our words makes it possible for us to identify items to which our words
apply, and to begin empirical inquiries into the nature of those items. They
grant that Kripke and Putnam describe one way in which these inquiries
can result in discoveries about the nature of those items. For instance, they
grant that once we can identify samples of water by means of our primary
intension for the word ‘water’ we can go on to discover that in fact in
208      

the actual world water is H2 O. Moreover, once one has made such a
discovery, Jackson and Chalmers agree, it is natural to say that if there is no
H2 O in a given world, then there is no water in that world. They explain
this by introducing the idea of secondary intensions for our terms. According
to Chalmers, for instance, ‘‘The secondary intension of ‘water’ picks out the
water in every counterfactual world; so if Kripke and Putnam are correct,
the secondary intension picks out H2 O in all worlds’’ (Chalmers 1996: 57).
More generally, the secondary intension of a term is a function that settles
the extension of that term in counterfactual worlds, given its extension in
the actual world. It is called secondary because it is defined in terms of the
primary intension of the term, which, according to Jackson and Chalmers,
must be fixed before an empirical inquiry into the extension of the term in
the actual world can even begin.
Jackson and Chalmers present this theory as an interpretation and
clarification of the epistemological status of ‘‘ostensive definitions’’ of the
sort exemplified by (K) and (P), and the relationship of these definitions to
individual speakers’ judgements about how they would apply kind terms
under various possible circumstances. For any given speaker S and natural
kind word T, an ‘‘ostensive definition’’ of the secondary intension expressed
by T is defined as whatever explains the underlying unity of all the objects
or items to which T actually applies, given its primary intension. S knows
the primary intension of T without any empirical inquiry—without
presupposing anything about which subjectively equivalent world S is
actually in. According to this view, a speaker may revise her understanding
of the primary intension of T , but only if she believes she has made a
mistake about the circumstances under which she would correctly apply T,
independent of any assumptions about which world is actual. Nevertheless,
according to Jackson and Chalmers, the primary intension and the secondary
intension of a given term T by definition yield exactly the same extension
for T in the actual world.

6.10. Problems with Primary Intensions

Can we appeal to Jackson’s and Chalmers’s theory of primary intensions to
defend option (1)? They claim that their view is consistent with Kripke’s
and Putnam’s causal-historical theory of extension, and that it yields the
       209

conclusion, for instance, that the actual extension of ‘water’, and hence also
(by definition) its secondary intension, is H2 O. But Kripke and Putnam
claim that the extension of the word type ‘gold’ in English is the set of
objects that are (bits of ) the element with atomic number 79, and that this
is an empirical discovery that did not change the extension of the word type
‘gold’ in English since at least as far back as 1650. To justify these claims,
hence to justify option (1), a person who accepts Jackson and Chalmers’s
theory of primary intensions would have to show that in 1650, the primary
intension of the English word ‘gold’, together with facts about the world in
1650, determined that the extension of the English word ‘gold’ was the set
of objects that are (bits of ) the element with atomic number 79, and not
the set of objects that are (bits of ) the element with either atomic number
78 or atomic number 79 (or mixtures of the two).
The first main problem with this strategy is that primary intensions are
apparently too vague to support (1). Since Locke and Twin Locke are
by hypothesis physical and behavioural twins, and future developments
in either linguistic community are not relevant to the primary intensions
they associate with the respective word types that they spell g-o-l-d, those
primary intensions must be identical. As I explained above, neither Locke’s
nor Twin Locke’s linguistic dispositions entail that platinum is not in the
extension of both of their respective tokens of g-o-l-d. If primary intensions
are settled by what a speaker would say under various circumstances, then,
the primary intension that Locke and Twin Locke associated with their
respective tokens of g-o-l-d does not support option (1).
Chalmers and Jackson might be willing to agree that the primary
intension that Locke and Twin Locke associated with their respective
tokens of g-o-l-d was too vague to support option (1). In a footnote,
Jackson asserts that ‘‘In the mouths and from the pens of the folk it is
indeterminate whether [water] is H2 O or the watery stuff on Twin Earth
that counts as water on Twin Earth’’ ( Jackson 1998a: 38 n. 12). If he
accepts this, it seems he would be more inclined to accept that the primary
intension that Locke and Twin Locke associated with their respective
tokens of g-o-l-d was too vague to support option (1). If Jackson makes
this move, and (1)–(3) exhaust the options for describing the thought
experiment, then Jackson must embrace option (3). I will examine this
kind of reaction to the gold–platinum thought experiment in the next
chapter.
210      

Another possible reply is that the primary intensions that Locke and
Twin Locke associated with their tokens of g-o-l-d in 1650 did in fact
determine that the extension of their tokens of g-o-l-d is the set of objects
that are bits of the element with atomic number 79, but that Locke and
Twin Locke may have been confused or unclear about this. According
to this reply, they could in principle have discovered the precise primary
intension of their tokens of g-o-l-d without going through any empirical
investigation, if they had attended carefully enough to what they would
say in various different epistemically possible circumstances.
One main problem with this reply, and with Jackson’s and Chalmers’s
account of primary intensions more generally, is that what we actually say
when we find ourselves in a previously imagined situation almost always
trumps our earlier speculations about what we would say if we were to find
ourselves in that situation. What we actually say when we find ourselves in
a previously imagined situation reflects our best current judgement of what
is true in that situation. When we are actually in the previously imagined
situation, our best judgement of what is true in that situation brings with it
PJSSs and PJSEs. If those PJSSs and PJSEs conflict with earlier speculations,
then so much the worse for those speculations. Recall, for instance, in the
seventeenth century scientists took for granted that gold is a yellow metal.
A person living at the time might well have believed that one cannot reject
the statement that gold is a yellow metal without changing the subject.
But we now know that in its pure form gold is white, not yellow. Later
scientists took themselves not to have changed the topic, but to have discovered
that gold is white. To accept this description of the case, we must trust the
later scientists’ complex PJSSs and PJSEs for the word ‘gold’ more than we
trust any previous judgement that one could not reject the statement that
gold is a yellow metal without changing the subject.
This example shows that statements we can’t imagine giving up without
changing the subject are not thereby guaranteed to be true. But one might
think that to accept Chalmers’s claim that some of our words have primary
intensions that we can know a priori, we need only suppose that some of
the statements that we can’t imagine giving up without changing the subject
actually are true. The trouble with this thought, however, is that Chalmers’s
primary intensions are supposed to ‘‘back a priori truths’’—statements that
are ‘‘true no matter how the actual world turns out’’ (Chalmers 1996: 59).
Hence to accept Chalmers’s claim that some of our words have primary
       211

intensions that we can know a priori, it is not enough to suppose that some
of the statements that we can’t imagine giving up without changing the
subject are true.²¹
Like Chalmers, Jackson tries to defend the inference from ‘we don’t
understand how we could give up statement S without changing the
subject’ to ‘we could not give up statement S without changing the
subject’. ‘‘[S]urely it is possible to change the subject,’’ Jackson reasons,
‘‘and how else could one do it other than by abandoning what is most
central to defining one’s subject? Would a better way of changing the
subject be to abandon what is less central?’’ ( Jackson 1998a: 38). The
mistake here is to suppose that our best current judgement about what
counts as changing the subject is immune to future revisions. It is a truism
that if we want to change the subject, we must rely on our understanding
of what is most central to defining it. But this truism does not establish that
our current understanding of what is most central to defining our subject
cannot be revised without changing the subject. This claim is discredited
by many actual cases in which we were once confident that we could not
revise a given statement without changing the subject, but discovered later
that we were wrong. For reasons I noted earlier, provided we regiment our
words and define truth for them disquotationally, each of our discoveries
brings with it some chain or other of PJSEs across time. In this sense, our
PJSEs are of a piece with our pursuit of truth.²²
One might agree that our PJSEs across time can in practice lead us
to revise our beliefs about the extension of a given term, but insist that
in such cases our PJSEs partly constitute our coming to see the a priori
conditions for applying our terms more clearly than we saw them before.
Both Chalmers and Jackson insist that we can be mistaken about a priori

²¹ One can define a function F from words and (agent-centred) worlds to extensions so that for any
ordered pair of words and agent-centred worlds, the value of F for that ordered pair is the extension
of the word as used by the agent in that agent-centred world. Suppose we know a priori that for each
pair of words and agents, there exists such a function. It does not follow, as Chalmers seems to assume,
that we can know a priori what the value of the function is for the world we are actually in. For a
similar criticism of a related position, see Stalnaker 1990: 131–45.
²² The link between our pursuit of truth and our PJSEs does not imply that the underlying traits of
the universe are somehow ultimately syntactical. My argument against Chalmers and Jackson suggests,
on the contrary, that ‘‘the syntactic construction of a quantity name may not reveal its actual ties to
other quantities adequately’’, as Mark Wilson has stressed. See Wilson 1993: 82. I agree with Wilson that
we may discover that many of our terms have what he calls honourable intensions. Unlike Chalmers’s
hypothetical primary intensions, honourable intensions are not known a priori.
212      

claims, so the fact that in practice we have revised many of our beliefs about
how to apply our terms does not show that those beliefs are not a priori.
The main problem with this reply is that it implies that in practice we
may be no more confident of a so-called a priori claim about how to
apply our terms than we are of many of our empirical beliefs. From an
epistemological perspective, then, on this view, a priori beliefs have no
distinctive role, no special methodological significance. The category of a
priori beliefs about extension has metaphysical significance, perhaps, but
no practical epistemological significance. Moreover, the emptiness of this
metaphysical–epistemological distinction becomes clear when one reflects
about particular examples of revisions in our beliefs about how to apply
our terms. It is plausible to suppose, for instance, that for centuries before
the discovery of the chemical composition of water, it was widely assumed
that the term ‘water’ applied only to liquids. But now that we have
discovered that the extension of ‘water’ is H2 O, we believe that ice is
in the extension of ‘water’, too. Trusting our PJSEs across time, we are
prompted, according to the present theory, to revise our prior, incorrect
beliefs about the preconditions for applying the term ‘water’. Here it
seems obvious that we revise those prior beliefs because of an empirical
discovery. The claim that the revision is nevertheless purely ‘‘a priori’’,
because it follows from one’s theory that it must be a priori if it concerns
our ‘‘primary’’ understanding of how to apply the term, is strained and
ultimately empty.²³
Someone might insist that the primary intension that Locke and Twin
Locke associated with their respective tokens of g-o-l-d in 1650 actually
determined that our PJSEs across time for tokens of g-o-l-d in our linguistic
community are correct, and the PJSEs across time for tokens of g-o-l-d in
the Twin Earth community are incorrect. But this claim ultimately rests
on the mysterious and empty assumption that even though the Locke and
Twin Locke were confused about the primary intension that they actually
associated with their own tokens of g-o-l-d, and even though as far as they
are understood, it might be applied either to gold or platinum, nevertheless

²³ In Russell 2008, Gillian Russell presents an account of ‘truth in virtue of reference determiner’
that is structurally similar to Chalmer’s and Jackson’s account of truth in virtue of primary intensions.
Unlike Chalmers and Jackson, however, Russell sees clearly that his account ‘‘has no hope of meeting
any kind of special epistemic condition’’ (Russell 2008: 67) and hence that we cannot in general know
a priori that a sentence is ‘true in virtue of reference determiner’.
       213

there was something about how they used tokens of g-o-l-d in 1650 that
determined that the extension of those tokens of g-o-l-d was the set of
objects that are bits of the element with atomic number 79. This is not
so much a theory in support of option (1), as a stubborn insistence that
option (1) is correct, combined with an uninformative account of why
this is so. In fact, as I noted earlier, Jackson himself would not attempt to
defend option (1) in this way. This is not surprising, given that the theory
of primary intensions stems from an account of language that strongly
suggests that most of our PJSEs are incorrect, whether or not we focus
on long spans of time across which we have certainly lost track of past
English-speakers’ beliefs about how to apply their terms. I conclude that
the theory of primary intensions proposed by Jackson and Chalmers does
not support option (1).

6.11. Implicit Conceptions

One might be convinced by these objections to Jackson’s and Chalmers’s
theory of primary intensions, yet still feel there is something right about
the idea that prior states of understanding determine the extensions of
our terms. Perhaps the main source of the problems with Jackson’s and
Chalmers’s theory is that they define a speaker S’s state of understanding
a term T at time t in terms of how S would apply tokens of T under
various different circumstances she can entertain hypothetically at t. Let’s
agree that this equation is incorrect for the reasons presented above, the
most important of which is that what a speaker accepts in a particular,
actual situation almost always trumps her prior speculations about what
we she would say in that such a situation if it were actual. Still, one might
think, this does not show that when a speaker does encounter a new
situation, her application of tokens of T is not determined in some way
by her prior understanding of how to apply tokens of T. To maintain
this position despite the objections to Jackson’s and Chalmers’s theory of
primary intensions, we need only reject the theoretical claim that such
prior states of understanding can always in principle be fully and accurately
articulated by the speakers who have them. In other words, to maintain
this position it may be enough to say that the mental states that determine
correct application are implicit.
214      

Christopher Peacocke has recently argued that in order to explain the

fact that speakers can arrive at new understandings of their own claims
by reasoning, we must posit implicit conceptions that guide a speaker’s
reasoning. Peacock focuses on cases in logic and mathematics, but his
reasoning, if it were correct, would also apply to the clarification of natural
kind terms. He seems to be guided by the thought that the only way to
explain how it is possible for a given speaker S to use tokens of a term T to
inquire into the truth of sentences in which such tokens occur is to theorize
that S associates with the tokens of T an implicit conception that determines
S’s evaluation of the truth of sentences in which such tokens occur.
Peacocke claims, for instance, that Newton and Leibniz had an implicit
conception of the limit of a series that guided their discovery of the calculus,
even though they could not provide an explicit account of the limit of a
series, and, in fact, no one at that time had the mathematical resources to
provide an explicit account of the limit of a series. Peacocke claims we can
explain Newton’s and Leibniz’s deep understanding of calculus only if we
attribute to them an implicit conception of the notion of the limit of a series
that predetermined a single, correct, and explicit account of that notion.
Peacocke wants to accommodate what he calls the Phenomenon of New
Principles—‘‘the rational, justified acceptance of new principles involving
a given concept, new in the sense that these principles do not follow
from those principles (if any) immediate acceptance of which is required
for possession of the concept’’ (Peacocke 1998: 65). This requires, he
argues, that implicit conceptions not be understood as ‘‘personal-level
conceptual-roles’’ (Peacocke 1998: 64).
Can an appeal to something like implicit conceptions help us to justify
or explain option (1)? The proposal would be that in 1650, prior to
the branching of the two linguistic communities described above, Locke
associated with his tokens of g-o-l-d an implicit conception that guided
him in his applications of the term, and determined that the extension of
the tokens is the set of objects that are bits of the element with atomic
number 79, not the set of objects that are either bits of the element with
atomic number 78 or bits of the element with atomic number 79 (or
mixtures of the two).
One problem with this suggestion is that if Locke associated a given
conception with his tokens of g-o-l-d in 1650, by hypothesis, Twin Locke
simultaneously associated a conception with his tokens of g-o-l-d in 1650.
       215

Since Peacocke wants his implicit conceptions to guide and predetermine

correct linguistic usage, he cannot appeal to later developments in the two
communities to distinguish between the conceptions that Locke and Twin
Locke associate with their respective tokens of g-o-l-d. By hypothesis,
however, if we disregard later developments in either community, we must
conclude that the implicit conception that Locke associated with his tokens
of g-o-l-d in 1650 is the same as the implicit conception that Twin Locke
associated with his tokens of g-o-l-d in 1650.
As I stressed above, however, it seems that nothing in the linguistic
behaviour or mental states of Locke and Twin Locke supports option
(1) or option (2). The bald assertion that in this case the implicit conception
associated by Locke and Twin Locke with their respective tokens of g-o-l-d
in 1650 supports option (1) leaves it mysterious how we can ever know
what the implicit conceptions associated by a speaker with his terms
actually are. To see the problem, note that when Peacocke claims that
the implicit conception that Locke associated with his tokens of g-o-l-d
in 1650 determined that option (1) is correct, Twin Peacocke on Twin
Earth simultaneously claims (when translated into English) that the implicit
conception that Twin Locke associated with his tokens of g-o-l-d in
1650 determined that option (2) is correct. Suppose Peacocke and Twin
Peacocke were to try to decide which of these two claims is correct.
What could possibly settle this? The thought experiment implies there is
nothing independent of their contested, hypothetical implicit conceptions
to ground (or undermine) either of their incompatible assertions.
Peacocke is sensitive to the possible charge that implicit conceptions
are mysterious and unknowable. He rejects this charge in the following
passage:
nothing in what I have said should encourage the view that implicit
conceptions . . . somehow transcend the knowable. There would be a commitment
to such transcendence if it were allowed as the possibility that there could be two
speakers whose rational judgements about particular applications of an expression,
and about principles involving it, are in actual and counterfactual circumstances
identical, and yet have differing implicit conceptions. Nothing I have said entails
that that is a possibility. (Peacocke 1998: 57)

It seems, however, as I have argued, that if Peacocke attempts to justify

option (1) by appeal to implicit conceptions, he will commit himself to the
216      

existence of mysterious, unknowable implicit conceptions. For his twin on

Twin Earth will presumably attempt to justify the PJSEs on Twin Earth in
the same way, thus generating the fruitless ‘‘dispute’’ that I just described.
Although Peacocke and his twin disagree, and are therefore not asserting
that Locke and Twin Locke had different implicit conceptions, they are
nevertheless unable to appeal to anything except their confidence in the
PJSEs across time in their respective communities to support their claims
about the implicit conceptions that Locke and Twin Locke associate with
their respective tokens of g-o-l-d.
If Peacocke retains his commitment to (M) yet does not insist on making
mysterious ungrounded claims of the sort just described, he must reject
both options (1) and (2), and thereby reject the PJSEs in both linguistic
communities. We thereby arrive, once again, at the conclusion that if we
accept (M) we must reject options (1) and (2). If we hang on to (M), the
only remaining option is (3), according to which E1 = E2 and E1 = E3 .
We must now consider whether it is more attractive to endorse option
(3) than to abandon (M) and the intuitions that support it.
 7
Sense and Partial Extension

7.1. Option (3)

We have now seen that if we accept

(M) For any two word tokens w and w such that the utterance or
inscription of w occurs at some time t before the utterance or
inscription of w , a practical judgement of sameness of extension
(PJSE) across time for w and w is true only if for some set E,
(a) E is the extension of both w and w , and
(b) the totality of facts about the ex-use of w at times prior to or
identical with t (hence prior to the utterance or inscription of
w ) determines that E is the extension of w

then we cannot accept either option (1), according to which E1 = E2 and

E1 = E3 , or option (2), according to which E1 = E2 and E1 = E3 , in the
gold–platinum thought experiment. We naturally find option (1) more
attractive than option (2), because it accords with our PJSEs across time for
‘gold’. But we have no basis independent of these judgements for claiming
that the extension of the English word ‘gold’ in 1650 was the same as the
extension of the English word ‘gold’ in our linguistic community today.
To insist on option (1) is to beg the question against option (2), which
members of the Twin English-speaking community today will no doubt
prefer, because it accords with their PJSEs across time for the Twin English
word ‘gold’.
Given (M), the only remaining option is (3), according to which E1 = E2
and E1 = E3 . To embrace the option (3) is to abandon our PJSEs across time
for ‘gold’. Moreover, it is not difficult to construct thought experiments
that challenge our PJSEs across time for other natural kind terms. To reject
the corresponding options (1) and (2) for all of these terms is to reject
218    

a vast number of our PJSEs across time, and thereby to undermine our
confidence in the way we actually conduct ordinary and scientific inquiries.
Those who accept (M) only because they think it might enable them
to explain and justify their PJSEs across time, including their PJSEs across
time for tokens of g-o-l-d, should therefore now be inclined to reject
(M) because it implies option (3). For all I have argued so far, however,
it may still seem reasonable to embrace option (3).¹ In this chapter I’ll
examine two different strategies for making sense of option (3), and explain
why one might be tempted to endorse these different strategies for making
sense of option (3), even though they commit us to rejecting many of our
PJSSs and PJSEs across time. I shall argue that although both strategies are
initially appealing, closer examinations reveal that neither one provides a
plausible alternative to trusting our PJSEs across time.

7.2. Dummett on Sameness of Satisfaction across

Time
Suppose you are studying a proof written in English that a plane intersects
a cylinder in an ellipse.² When you read the sentence ‘A plane intersects
a cylinder in an ellipse’, you take the author to be asserting that a plane
intersects a cylinder in an ellipse, and you take the word ‘ellipse’ that occurs
in the author’s sentence to be the same word ‘ellipse’ that occurs in your
present affirmations of
(e) x satisfies my word ‘ellipse’ if and only if x is an ellipse.

¹ Many philosophers who wish to explain and ground our PJSEs whenever possible would endorse
option (3). Mark Wilson, for example, writes that ‘‘in some common situations the proper ground for
determining whether a predicate is true of a particular individual becomes uncertain or ambiguous.
One kind of situation in which this can happen . . . is one in which the linguistic community is unaware
of the existence of kinds of objects to which a predicate might be thought to apply. It is frequently
indeterminate whether such unexpected objects should be assigned to the extension or to the counter-
extension of the predicate’’ (Wilson 1982: 549). Wilson’s intuition that ‘‘it is frequently indeterminate
whether such unexpected objects should be assigned to the extension or to the counter-extension of
the predicate’’ suggests that he is committed to (M). Moreover, a commitment to (M) would explain
Wilson’s proposed ‘‘adequacy condition’’ on assignments of extensions: ‘‘The evidence for assignment
of an extension to a predicate should be limited to such linguistic behavior as can be reasonably
extrapolated from the community’s contemporaneous practice and should not reflect accidental features
of the society’s later history’’ (Wilson 1982: 553).
² I get this example from Dummett 1978a: 300. See also Hacking 1985; Hilbert and Cohn-Vossen
1956: 7–8; and Spivak 1994: chapter 4, Appendix 2, Problem 2.
     219

You thereby also accept that x satisfies the word ‘ellipse’ that occurs in
the author’s sentence if and only if x is an ellipse. Conjoining these two
acceptances, you make a PJSS for the author’s word ‘ellipse’.
As you work through the proof, you occasionally glance back at earlier
parts of the proof that contain tokens of e-l-l-i-p-s-e, and you continue to
take those tokens of e-l-l-i-p-s-e to be tokens of the same word ‘ellipse’
that occurs in your own affirmations of (e). Given your acceptance of (e),
you accept that x satisfies the word ‘ellipse’, as it occurs earlier in the proof,
if and only if x is an ellipse, and thereby make a PJSS across time for the
author’s word ‘ellipse’.
Now suppose you are convinced by the proof. You take yourself to
have learned something new about ellipses. When you accept the proof,
you continue to trust your PJSSs across time for tokens of e-l-l-i-p-s-e.
Some of these practical judgements reach back to your own tokens of
e-l-l-i-p-s-e that existed before you worked through the proof. It is
tempting to assume that these practical judgements are correct or incorrect
according as your criteria for applying ‘ellipse’ before you worked through
the proof determined an extension for ‘ellipse’ that is identical to the
extension determined by the new criterion for applying ‘ellipse’ that you
accept after, and as a result of, working through the proof. According to
Michael Dummett, from whom I borrow this example,
The standard view of the effect of the proof [that a plane intersects a cylinder in
an ellipse] is that the new criterion which it provides enables us to recognize as
ellipses only figures which could already have been so recognized by the criteria
we already had: what the proof establishes is precisely that the new criterion,
where applicable, must always agree extensionally with that given by the original
definition of ‘ellipse’; that is why the proof persuades us to adopt the new test as a
criterion. (Dummett 1978a: 301)

This way of thinking about sameness of satisfaction across time is especially

gripping when we are trying to understand reasoning about abstract objects,
such as ellipses, that we feel we can only identify by means of definitions.³
But it is also natural to think that a new empirical claim made by using a
term T counts as a discovery only if the new criterion for applying T that
it introduces agrees extensionally with the relevant previous criteria for

³ Recall Meno’s paradox: we cannot inquire into the definition of virtue unless we can already define
it, for otherwise we cannot even identify the topic whose definition we seek.
220    

applying T. Consider, for instance, the original announcement (in English)

that gold is the element with atomic number 79. The English-speaking
chemists who made this announcement continued to take their previous
uses of tokens of g-o-l-d to be occurrences of the word type ‘gold’ that they
take themselves to be using to make their announcement. Now suppose
some of them affirmed
(g) x satisfies ‘gold’ if and only if x is (a bit of ) gold.
Then when they took their previous tokens of g-o-l-d to be tokens of
their word ‘gold’, they made a PJSS across time for their previous tokens of
g-o-l-d—they accepted both that x satisfies ‘gold’ if and only if x is (a bit of )
gold and that x satisfies their previous tokens of g-o-l-d if and only if x is (a
bit of ) gold. But, just as in the ellipse case, the standard view is that their trust
in this judgement would be misplaced unless the criteria for applying ‘gold’
to which they were committed before they made their supposed discovery
agree extensionally with the new criterion, according to which x satisfies the
word ‘gold’ if and only if x is (a bit of ) the element with atomic number 79.⁴
The standard view amounts to a version of (M). Taking this version of
(M) for granted, Dummett argues, in effect, that the ex-use of tokens of
g-o-l-d in 1650 on Earth and Twin Earth in the gold–platinum thought
experiment did not settle that x satisfies those tokens if and only if x is (a bit
of ) the element with atomic number 79, or that x satisfies those tokens if
and only if x is either (a bit of ) the element with atomic number 78 or a bit
of the element with atomic number 79, and hence that the PJSSs and PJSEs
across time for tokens of g-o-l-d made by members of the English-speaking
community on Earth and the Twin English-speaking community on Twin
Earth are false. To see why, one must examine Dummett’s accounts of
sense and implicit knowledge.

7.3. Dummett on Sense and Implicit Knowledge⁵

Dummett argues that a speaker’s uses of her words are guided by her
knowledge of her language, and that at the core of a speaker’s knowledge

⁴ This implies not that all their previous applications of ‘gold’ are correct according to the new
criterion, but that if one or more of their previous applications of ‘gold’ is incorrect according to the
new criterion, it is also incorrect according to their previous criteria for applying ‘gold’.
⁵ I am indebted to Imogen Dickie for very helpful criticisms of two earlier drafts of §§7.3–7.6
     221

of her language is her knowledge of the senses of the logically significant

expressions of her language, where ‘‘the sense of an expression—which is
part, and often the whole, of its meaning—is given to us as a means of
determining its reference (or its semantic value)’’ (Dummett 1991: 148). In
particular, the sense of a sentence is given to us as the grasp of the condition
that must be satisfied for the sentence to be true. The sense of a predicate is
given to us as the grasp of the contribution of the predicate to our grasp of
the senses of sentences in which it occurs. Dummett reasons as follows: ‘‘If
to grasp the sense of the predicate is to grasp something that determines its
reference, and if it is to lead us to apprehend the condition for a sentence
formed by means of the predicate to be true, then it must be given to us as
the grasp of the condition that must be satisfied, by any object, for it to be
mapped onto the value true or for it to be mapped onto the value false.’’⁶
The idea can be expressed without treating truth values as objects. For
instance, we can say that the sense of our word ‘gold’ is given to us as the
grasp of the conditions under which an object x satisfies our word ‘gold’.
If the sense of (tokens of ) a given predicate P at a given time t settles,
for every object x, the conditions under which a given object x satisfies
(tokens of ) P at t, then the sense determines the extension of (tokens of ) P
at t —the set of all and only the objects that satisfy (tokens of ) P at t. For
instance, if the sense of tokens of g-o-l-d used by English-speakers in 1650
settles, for every object x, the conditions under which a given object x
satisfies those tokens of g-o-l-d, then it settles the extension of those tokens
of g-o-l-d—the set of objects of which those tokens of g-o-l-d are true.
On Dummett’s interpretation of sense, (i) the sense of (tokens of ) a
predicate P at t is given to speakers at t as a grasp of the conditions under
which a given object x satisfies (their tokens of ) P, and (ii) the speakers’
grasp at t of these conditions must be among (or, at least, settled by) the
speakers’ mental states at t, hence part of the ex-use of (their tokens of ) P
at t.⁷ In short, the ex-use of (tokens of ) a predicate P at time t determines

⁶ Dummett 1991: 143. This interpretation of the sense of a predicate contrasts with P. T. Geach’s
view that the sense of a predicate is a function from the senses of names to the senses of sentences. See
Dummett 1991: 141–5.
⁷ Dummett emphasizes that ‘‘the function [of a theory of sense] is solely to present an analysis of the
complex skill which constitutes mastery of a language, to display, in terms of what he may be said to
know, just what it is that someone who possesses that mastery is able to do . . . [and] not . . . to describe
any inner psychological mechanisms which may account for his having those abilities’’ (Dummett
1993b: 37). In this passage Dumment may seem at first look to be rejecting the claim that facts about
222    

the sense of (tokens of ) P at t. As we saw in the previous paragraph, if the

sense of (tokens of ) P at t settles, for every object x, the conditions under
which a given object x satisfies (tokens of ) P, it determines the extension
of (tokens of ) P at t. We can conclude that if the sense of (tokens of ) P at
t settles, for every object x, the conditions under which a given object x
satisfies (tokens of ) P, then its ex-use at t determines the extension of (tokens
of ) P at t.
A speaker’s knowledge of the sense of a word is explicit only if she states
and deliberately consults criteria for applying it. But no speaker is ever in a
position to state non-circularly, for every word in her language, the criteria
for correctly applying it. At any given time, a speaker’s knowledge of the
criteria for correctly applying some of her words must be implicit.
One might object that if a speaker does not state and deliberately
consult criteria for applying her words, then it makes no sense to say, as
Dummett does, that her application of her own words is guided by criteria.
Dummett tries to answer this objection by tying implicit knowledge closely
to linguistic behaviour: a speaker’s implicit knowledge of how to ex-use
her words, he writes, ‘‘shows itself partly by . . . a readiness to acknowledge
as correct a formulation of that which is [implicitly] known when it
is a presented’’ (Dummett 1993b: 96). According to Dummett, we can
convince ourselves that a speaker’s implicit knowledge shows itself in this
way by reflecting on ordinary cases in which we come to accept explicit
formulations of rules as formulations of rules we have been following. He
reasons, for instance, as follows:
If a speaker always uses the pairs ‘I’/‘me’, ‘he’/‘him’, ‘she’/‘her’ and ‘who’/‘whom’
correctly, but, never having been taught the rudiments of formal grammar, has
never heard the words ‘nominative’ and ‘accusative’, can he be said to have an
implicit grasp of the concepts they express? A statement of the rule he tacitly
follows will involve an explicit formulation of those concepts and will necessarily
be somewhat lengthy. Still, we may credit the speaker with an implicit knowledge of that
rule, provided that, when he understands the statement of it, he acknowledges it as accurately
describing his existing practice. (Dummett 1991: 96, my emphasis)

the ex-uses of word tokens settle what senses those word tokens express, especially when such facts
are assumed to be specifiable in purely physical or psychological terms. What he is rejecting, however,
is that it is the function of a theory of sense to describe the particular physical and psychological facts
about ex-uses of word tokens that actually settle the senses that those word tokens express, or to
hypothesize more generally about how facts about ex-use settle the senses of word tokens.
     223

In this passage Dummett suggests the form of a sufficient condition for

a speaker to be credited with an implicit knowledge of a grammatical
rule—roughly, the speaker’s willingness to accept an explicit formula-
tion of it. But if a speaker’s implicit knowledge of a grammatical rule
can be manifested by the speaker’s willingness to accept an explicit for-
mulation of it, it should also be possible for a speaker’s knowledge of
a semantical rule to be manifested by the speaker’s willingness to accept
an explicit formulation of it. For instance, consider the following obser-
vation, which parallels Dummett’s observation about the grammatical
rule:
If a speaker always uses ‘gold’ in such a way that it is satisfied by x if and only if x
is (a bit of ) the element with atomic number 79, but, never having been explicitly
taught the rudiments of chemistry, has never heard the words ‘atomic number’,
we may still credit the speaker with an implicit knowledge of that rule, provided
that, when he understands the statement of it, he acknowledges it as accurately
describing his existing practice.

More generally, if we accept Dummett’s recipe for tying implicit knowledge

to an explicit formulation of it, we should also accept the following weak
constraint on our understanding of implicit knowledge of a semantical rule
of the form ‘‘an object x satisfies A’s (unambiguous) word w if and only if
x is F ’’.
Acceptance constraint: Speaker A has implicit knowledge that an object x
satisfies A’s (unambiguous) word w if and only if x is F
IF
if A were presented in appropriate circumstances with an explicit
statement that x satisfies w if and only if x is F, A understood that
statement, and A carefully considered whether to accept it, then A
would explicitly accept that x satisfies w if and only if x is F.
To apply this constraint, one would need to have some idea of when
circumstances are appropriate, what it is for a person to understand a statement
of the form ‘‘x satisfies w if and only if x is F ’’, and how long and in what
way she must consider such a statement for her acceptance of it to show
that she has implicit knowledge of the conditions under which an object x
satisfies her tokens of a given word. I shall not try to explicate these crucial
terms of the acceptance constraint. To understand the acceptance constraint,
224    

however, it helps to see that it as a partial interpretation of Dummett’s more

fundamental manifestability constraint on a theory of implicit knowledge:

Manifestability constraint: If a speaker’s use of a given word is guided by

her implicit knowledge of criteria for correctly applying it then there
is some practical ability by means of which her implicit knowledge of
those criteria for correctly applying it may be manifested.⁸

To apply this crucial constraint, we must have some idea of what it would
take for a speaker’s implicit knowledge of criteria for applying her terms
to be manifested. The acceptance constraint presents the schematic form
of one type of sufficient condition for a speaker’s implicit knowledge of
criteria for applying her terms to be manifested. More generally, Dummett
argues that a speaker’s implicit knowledge of criteria for applying a given
word is manifested by her practical ability to use whole sentences in which
that word occurs (Dummett 1993a: 38). It is a further question what sense
of ‘‘practical ability’’ Dummett has in mind. I shall not try to explicate
Dummett’s use of this term, or to formulate, on his behalf, necessary and
sufficient conditions for a speaker’s use of a given word to be guided by
her implicit knowledge of criteria for applying it in terms of her practical
ability to use whole sentences in which that word occurs. I shall assume for
now that we have some idea of how to apply the acceptance constraint,
seen as a partial interpretation of the manifestability constraint, and raise
difficulties for that assumption in §7.6.
In addition to the acceptance and manifestability constraints, there is a
constraint that implicit knowledge must satisfy if it is to be compatible
with (M). The roots of this third constraint are evident in Dummett’s
commentary on the proof that a plane intersects a cylinder in an ellipse.
Dummett’s commentary on that proof is designed in part to persuade us of
his view that an individual’s new applications of her words are guided by her
implicit knowledge, which must therefore exist prior to and independently of
those new applications. The criteria that she adopts as a result of accepting
a proof or discovery should determine satisfaction conditions for her words
that are extensionally equivalent to the satisfaction conditions determined
for those words by the implicit knowledge of how to apply them that she

⁸ Dummett’s commitment to the manifestability constraint is evident in most of his writings, but
see, in particular, Dummett 1993a: 37–8, 46, and 92.
     225

had prior to her acquaintance with the proof or discovery. But any two
criteria for applying a given word that agree extensionally with the criteria
for applying it that are determined by the implicit knowledge of how to
apply it that she had prior to her acquaintance with the proof or discovery
will agree extensionally with each other. In short, according to Dummett,
attributions of implicit knowledge to a speaker A must satisfy the
Extensional equivalence constraint: If w is not ambiguous for A, A has
implicit knowledge that x satisfies w if and only if x is F, and A has
implicit knowledge that x satisfies w if and only if x is G, then x is F if
and only if x is G.

7.4. Why Dummett’s Constraints Rule out Options

(1) and (2)
Dummett rejects Putnam’s claim that the extension of English-speakers’
tokens of g-o-l-d has not changed since the development of chemistry and
the discovery that gold is the element with atomic number 79. To discredit
Putnam’s claim, Dummett constructs the following thought experiment:
Suppose that, when chemical analysis became possible, it had been discovered
that there were two chemically distinct substances, one an element and the
other a compound, both satisfying both the ordinary criteria for being gold and
indistinguishable save by the most refined tests. Of those things ordinarily said to
be made of gold, some were composed of one substance, some of the other, and
a few of a mixture of the two. It is clear that the term ‘gold’ would then have
become useless for theoretical chemistry, so long as it continued to be applied
as before, e.g. in such a way that people now said that there were two kinds of
gold. Of course, an alternative would have been to reserve the term ‘gold’, in
its everyday application, for one or other of these two types of substance: since
that would entail considerable upheaval in social practice, in view of the symbolic
and economic significance of gold, it would be the more unlikely outcome of the
discovery. What is clear, at any rate, is that the word ‘gold’ did not, in advance of the
introduction of a theory and technique of chemical analysis, have a meaning which determined
the course to be followed. (Dummett 1978b: 428–9, my emphasis)

In this passage Dummett equates ‘meaning’ with ‘sense’; the italicized

conclusion of his thought experiment is that the word ‘gold’ did not, in
226    

advance of the introduction of a theory and technique of chemical analysis,

have a sense that determined the course to be followed. Dummett makes
this explicit as follows:

The thesis that the adoption of technical means for distinguishing gold from other
substances involves some alteration in the sense of ‘gold’ ought not to be resisted.
To resist it would be to hold it was determined in advance what we ought to
have done if what were ordinarily classified as gold had proved to comprise two
chemically distinct substances; and this is not the case. (Dummett 1978b: 429)

Dummett concludes that the sense of ‘gold’ did not ‘‘determine in advance’’
the course to be followed after the development of chemical analysis.
By this he means that in 1650, for instance, prior to the development
of chemical analysis, the practical ability to apply tokens of g-o-l-d was
not guided by implicit knowledge of either of the two rules for applying
‘gold’ that might be adopted after the development of chemical analysis.
His thought experiment is intended to show that these two criteria for
applying tokens of g-o-l-d determine extensions that are different from the
extension determined by the criteria for applying tokens of g-o-l-d that
English-speakers relied on in 1650. So the differences in criteria that guide
speakers’ use of tokens of g-o-l-d determine corresponding differences
in the extensions of tokens of g-o-l-d. By the extensional equivalence
constraint, then, we must conclude that in 1650, the sense of tokens of
g-o-l-d did not determine that tokens of g-o-l-d in 1650 had either of the
extensions that tokens of g-o-l-d come to have after the development of
chemistry in the two alternative futures that Dummett describes.
The same goes for the gold–platinum thought experiment. Recall that
members of both communities in the gold–platinum thought experiment
take themselves to be guided by their understanding of how to apply
their tokens of g-o-l-d—they each find their own applications of their
tokens of g-o-l-d natural and inevitable. After about 1850, when their
chemical discoveries become entrenched, they would not be persuaded
by members of the other community that they have made any mistakes
about the conditions under which an object x satisfies their tokens of
g-o-l-d. Moreover, by hypothesis in 1650 the samples to which they
actually applied their tokens of g-o-l-d, and their dispositions to apply
such tokens in various different situations, were in all relevant respects the
same. Hence, if the ex-use of tokens of g-o-l-d settles the senses of those
     227

tokens in 1650, as Dummett’s theory assumes, there are no grounds for

distinguishing between the senses that members of the two communities
associated with their tokens of g-o-l-d in 1650. Nevertheless, members of
the two linguistic communities reach different verdicts about the extensions
of their respective tokens of g-o-l-d.
One might wonder how they could reach these different verdicts, if
their ex-use of tokens of g-o-l-d in 1650 are by hypothesis the same. A
key consideration, emphasized above in §6.8, is that whether or not an
individual would accept or reject certain sentences may depend on the order
in which he is presented with evidence and theories that are relevant to
his evaluation of those sentences. It is plausible to suppose that Locke and
Twin Locke (and other members of their respective linguistic communities,
including jewellers and chemists) would have affirmed the sentence ‘x is
gold if and only if x is (a bit of ) the element with atomic number 79’ if
they had been presented with the same evidence and theories that English-
speakers have encountered since 1650, in the same order in which they
actually encountered them. But it is equally plausible to suppose that Locke
and Twin Locke (and others in their respective linguistic communities)
would have affirmed the sentence ‘x is gold if and only if x is either (a bit
of ) the element with atomic number 79 or (a bit of ) the element with
atomic number 78’ if they had been presented with the same evidence
and theories that Twin English-speakers have encountered since 1650,
in the same order in which they actually encountered them. We have no
independent grounds for saying that one of these presentations of evidence
and theories is correct and the other is incorrect.⁹

⁹ One might reply that even if Locke would accept the sentence ‘x is gold if and only if x is either (a
bit of ) the element with atomic number 79 or (a bit of ) the element with atomic number 78’ under
certain conditions, he could nevertheless appeal to his archetypical applications of g-o-l-d, which would
lead him to say, of a particular piece of platinum, ‘‘That is not gold.’’ If so, then this recognitional
capacity should lead him to reject the above sentence. The problem with this suggestion is that the order
in which Locke and Twin Locke are presented with evidence and theories also affects their archetypical
applications of g-o-l-d and the recognitional capacities they associate with g-o-l-d. It is plausible to
suppose that if they had been presented with the same evidence and theories that English-speakers have
encountered since 1650, in the same order, both Locke and Twin Locke would have denied that ‘gold’ is
true of what they recognize to be a piece of platinum. But it is equally plausible to suppose that if they
had been presented with the same evidence and theories that Twin English-speakers have encountered
since 1650, in the same order, they would have affirmed that ‘gold’ is true of that same piece of platinum,
which they would recognize to be a piece of what they call ‘gold’. Again, we have no independent
grounds for saying that one of these presentations of evidence and theory is correct and the other is
incorrect.
228    

If we focus just on the acceptance constraint, it may seem permissible

to attribute to Locke and Twin Locke both (i) the implicit knowledge
that x satisfies their tokens of g-o-l-d if and only if x is the element with
atomic number 79, and (ii) the implicit knowledge that x satisfies their
tokens of g-o-l-d if and only if x is the element with atomic number 78
or x is the element with atomic number 79. But these accounts of the
conditions under which an object x satisfies their tokens of g-o-l-d are not
coextensive. Hence the attributions of implicit knowledge just described do
not satisfy the extensional equivalence constraint. Dummett’s acceptance of
(M), his account of sense, and the constraints he lays down on attributions
of implicit knowledge together commit him to the conclusion that the
sense that Locke and Twin Locke associated with tokens of g-o-l-d in 1650
rules out options (1) and (2). If we accept Dummett’s premises, the only
remaining option is (3).

7.5. Dummett’s Version of Option (3)

But how shall we make sense of option (3) if we accept Dummett’s premises?
To answer this question, we must first look more closely at Dummett’s
understanding of the sense of the English word ‘gold’. Dummett observes
that
The criteria for the application of ‘gold’ used by ordinary speakers are sufficient
for ordinary purposes, but . . . such speakers are willing to yield to the criteria
employed by the experts, and unknown to themselves, in extraordinary cases.
(Dummett 1978b: 427)

But Dummett argues that the sense of ‘gold’ for ordinary speakers is not
different from the sense of ‘gold’ for the experts. How could this be?
Dummett’s answer is that
The meaning of the word ‘gold’, as a word of the English language, is fully
conveyed neither by a description of the criteria employed by the experts nor by
a description of those used by ordinary speakers; it involves both, and the grasp of
the relationship between them. (Dummett 1978b: 427)

If one accepts this view of the sense of ‘gold’, then to understand option
(3) one must characterize the sense that Locke and Twin Locke associated
     229

with their tokens of g-o-l-d in 1650 partly in terms of a criterion (a

rule, requirement, or test) that was actually taken by experts at that time
to settle whether or not an object x falls in the extension of tokens of
g-o-l-d at that time. It is not enough, however, to note that experts
in 1650 take some criterion to settle whether or not an object x falls
in the extension of tokens of g-o-l-d. If the criterion is to be part of
the sense of that Locke and Twin Locke associated with their tokens of
g-o-l-d in 1650, it must settle whether or not an object x falls in the
extension of those tokens of g-o-l-d in a way that cannot be extensionally
incorrect.
Dummett typically uses the word ‘criterion’ to mean a rule, requirement,
or test that settles whether or not an object x falls in the extension of
particular tokens of w in a way that cannot be extensionally incorrect. In
contrast, I use the word ‘criterion’ to mean a rule, requirement, or test
that speakers take to settle, for their purposes at a given time, whether or
not an object x falls in the extension of tokens of w at that time, but that
may be extensionally incorrect. To keep track of the distinction introduced
in the previous paragraph, I shall say that a criterion for using tokens of
w is weak if it may be extensionally incorrect, and strong if it cannot be
extensionally incorrect. Hence what Dummett calls a criterion is what I
shall call a strong criterion.
Suppose, provisionally, that there was a strong criterion for using tokens
of g-o-l-d in 1650, that it was explicitly endorsed by those who counted
in 1650 as experts in the proper use of tokens of g-o-l-d—jewellers and
chemists, for instance—and that it is part of the characterization of the
sense for tokens of g-o-l-d in 1650, a sense that settles, for every object x,
whether or not x is in the extension of tokens of g-o-l-d, as used by English-
speakers in 1650.¹⁰ Suppose, also, that the ex-use of English-speakers’
tokens of g-o-l-d in 1650 settled that those tokens expressed the sense just

¹⁰ Let us not worry about what makes it the case that Locke’s and Twin Locke’s tokens of g-o-l-d in
1650 have the same sense as the experts’ tokens of g-o-l-d in 1650. Dummett writes that ‘‘I shall not be
deterred, by the thought that he may not know what a gasket is, from telling someone that the gasket
in my car is leaking, any more than I am deterred by the fact that I myself do not know. The reason is
that I know that he will know how to find out what a gasket is if he needs to. The occurrence of this
word in our dialogue cannot be explained in terms of his idiolect or mine; it can only be explained
by reference to the English language, . . . . A language is not to be characterized as a set of overlapping
idiolects. Rather, an idiolect is constituted by the partial and imperfect grasp that a speaker has of a
language.’’ (Dummett 1991: 87) I shall not try to explain how this observation fits with Dummett’s
account of sense and implicit knowledge. The problem I raise below for Dummett-style versions of
230    

hypothesized, and that the corresponding extension of tokens of g-o-l-d

in 1650 is not identical with the extension of tokens of g-o-l-d today
in either the English-speaking community or the Twin English-speaking
community. Then it seems we can endorse what I’ll call Dummett’s version
of option (3), according to which
(i) there is a sense S1 of tokens of g-o-l-d in 1650 that is settled by the
ex-uses of tokens of g-o-l-d in 1650,
(ii) there is a sense S2 of tokens of g-o-l-d on Earth today that is settled
by the ex-uses of tokens of g-o-l-d on Earth today,
(iii) there is a sense S3 of tokens of g-o-l-d on Twin Earth today that is
settled by the ex-uses of tokens of g-o-l-d on Twin Earth today, and
(iv) each of S1 , S2 , and S3 is different from the others, and determines
an extension that is different from the extensions determined by the
others.
If we accept (i)–(iv), we will conclude that the ex-uses of tokens of g-o-l-d
in 1650 settle that those tokens have a determinate extension that is different
from the extensions that tokens of g-o-l-d later come to have in the two
linguistic communities of the gold–platinum thought experiment.¹¹

7.6. Two Problems for Dummett’s Version

of Option (3)
As I see it, there are two main problems with Dummett’s version of
option (3). The first is that it conflicts with PJSSs and PJSEs on which we
confidently rely. The second is that it leaves us without any clear practical
criterion for sorting ‘‘good’’ PJSSs and PJSEs from ‘‘bad’’ ones, and hence

option (3) suggests a related problem for Dummett’s view that a natural language is conceptually prior
to particular idiolects, but I shall not take the time to investigate that problem in this book.
¹¹ A minor variant of this interpretation is that according to Dummett the sense of tokens of g-o-l-d
was indeterminate in 1650 and therefore did not determine a classical extension for tokens of g-o-l-d in
1650. An apparent advantage of viewing the sense of tokens of g-o-l-d as indeterminate in 1650 is that
by doing so we would be able to regard later uses of tokens of g-o-l-d in both linguistic communities of
my gold–platinum thought experiment as rationally permitted by that indeterminate sense, and hence as
(in a way) rationally guided by speakers’ grasp of it. I shall not explore this interpretation of Dummett’s
account of the sense of ‘gold’ in 1650, however, because I think it faces essentially the same problems
as what I call Dummett’s version of option (3) and adds mostly unilluminating complications that make
those problems more difficult to discern.
     231

leaves us with no clear grasp of the distinction between strong and weak
criteria for using our words. I’ll discuss these problems in order, devoting a
subsection to each.

The first problem

Recall that according to Dummett’s acceptance constraint,
Speaker A has implicit knowledge that an object x satisfies A’s (unam-
biguous) word w if and only if x is F
IF
if A were presented in appropriate circumstances with an explicit
statement that x satisfies w if and only if x is F, A understood that
statement, and A carefully considered whether to accept it, then A
would explicitly accept that x satisfies w if and only if x is F.
As I noted above, this acceptance constraint presents the schematic form
of one type of sufficient condition for a speaker’s implicit knowledge of
criteria for applying her terms to be manifested. As I also noted above, the
key terms in the acceptability constraint are unclear, so it is unclear exactly
how we are to apply it. Even so, we know from the extensional equivalence
constraint that neither ‘x satisfies tokens of g-o-l-d in 1650 if and only if x
is (a bit of ) the element with atomic number 79’ nor ‘x satisfies tokens of
gold in 1650 if and only if x is either (a bit of ) the element with atomic
number 78 or (a bit of ) the element with atomic number 79’ characterize
the sense of tokens of g-o-l-d in 1650.
The extensional equivalence constraint therefore implies that while
speakers of both linguistic communities take themselves to be using their
respective tokens of g-o-l-d correctly as they develop their different
accounts of the extension of their tokens of g-o-l-d and rely on their PJSSs
and PJSEs across time for their own previous tokens of g-o-l-d and for
tokens of g-o-l-d previously produced by other speakers in their respective
linguistic communities, they gradually and unwittingly change the senses
of their tokens of g-o-l-d. Without noticing, they gradually cease to be
guided by the sense that guided speakers’ uses of tokens of g-o-l-d in 1650.
It is easy to see how ordinary users of tokens of g-o-l-d, who by assumption
don’t know the experts’ criteria for applying those tokens, might make
such a mistake. But in the circumstances described in the gold–platinum
thought experiment, the extensional equivalence constraint implies that
232    

even the experts gradually and unwittingly change the senses of their tokens
of g-o-l-d.
The first problem for Dummett is that the experts’ PJSSs and PJSEs
across time for tokens of g-o-l-d are among the PJSSs and PJSEs on which
we confidently rely in our inquiries. We take these PJSSs and PJSEs to be
guided by our understanding of the topics they concern. Dummett’s theory
of sense does not explain what this ‘‘guidance’’ or ‘‘understanding’’ consists
in, but instead presents a theory of sense that is designed to explain a different
sort of ‘‘guidance’’ and ‘‘understanding’’, one that conflicts with many of
the PJSSs and PJSEs across time on which we confidently rely. If we were
convinced that Dummett’s theory of sense explains a sort of ‘‘guidance’’ and
‘‘understanding’’ that matters to our inquiries, we would have to reject a vast
number of the PJSSs and PJSEs across time on which we confidently rely.

The second problem

As we saw in §7.3, according to Dummett’s theory the sense of an
expression at a given time is settled by the ex-uses of its tokens at that
time. One might therefore suggest that even if members of the two
linguistic communities gradually and unwittingly change the senses of their
tokens of g-o-l-d, members of the two linguistic communities today could
in principle discover and correct their alleged mistakes by reflecting more
carefully on the differences between their ex-uses of tokens of g-o-l-d
today and the ex-uses of tokens of g-o-l-d in 1650. Let us now try to
evaluate this suggestion.
Note first that we can take the suggestion seriously only to the extent that
we understand the assumption that (i)–(iv) of the previous section are true.
But do we understand how it could be that the ex-uses of tokens of g-o-l-d
at various times in the histories of the two linguistic communities in the
gold–platinum thought experiment settle the senses of tokens of g-o-l-d at
various times in such a way that (i)–(iv) are true? To address this question,
we must face a very difficult and obscure question that lies behind it: What,
if anything, settles whether or not a given criterion for using tokens of a
word w is a strong criterion for using those tokens, not a weak one?¹² This

¹² This is one version of the question of what it is to follow a rule. I discuss this question at length in
Ebbs 1997. For a sampling of other interpretations of the question and attempts to answer it, see Miller
and Wright 2002 and Wilson 2006.
     233

question is especially difficult for any theory which, like Dummett’s, implies
that a vast number of our PJSSs and PJSEs across time are mistaken. We
can begin to see why the question is difficult for such theories if we focus
not on deductions or discoveries, as Dummett usually does, but on mistakes.
In many cases, the best indication of whether or not a speaker has made
a mistake is his own linguistic behaviour—in particular, his willingness
to acknowledge that he made a mistake, in part by accepting an explicit
formulation of a criterion that he takes himself to have misapplied. For
instance, a speaker who at time t writes the phrase ‘for who the bell tolls’
may later come to acknowledge that relative to the grammatical rule that
he was implicitly committed to following at t, he should have written ‘for
whom the bell tolls’. As we saw above, Dummett himself emphasizes that
‘‘we may credit the speaker with an implicit knowledge of [such a] rule,
provided that, when he understands the statement of it, he acknowledges it
as accurately describing his existing practice’’ (Dummett 1991: 96). In such
a case we take the speaker’s acceptance of an explicit formulation of the
grammatical rule to express his commitment to it when he wrote ‘for who
the bell tolls’, hence to imply that he made a mistake when he wrote ‘for
who the bell tolls’.
The same kind of reasoning shows that in some cases a speaker’s explicit
formulation of a semantical rule licenses us to conclude that one of his
previous utterances was false. Suppose (first) that a speaker utters ‘‘That’s
gold,’’ while he was pointing at a piece of platinum, and (second) that a few
moments later, after some prompting, he says, ‘‘My word ‘gold’ is satisfied
by x if and only if x is (a bit of ) the element with atomic number 79,’’
thereby expressing a rule to which he takes himself to have been implicitly
committed when he said, ‘‘That’s gold,’’ while pointing at that piece of
platinum. Then he will rely on his PJSSs and PJSEs across time for the
token of g-o-l-d that occurs in his earlier utterance of ‘That’s gold’: if he
learns that the piece of platinum is not (a bit of ) the element with atomic
number 79, he will conclude that his utterance of ‘That’s gold’ was false.
This example shows that to take a speaker’s acceptance of an explicit for-
mulation of a semantical rule to express his own prior implicit commitment
to that rule, we must accept at least some of the speaker’s PJSSs and PJSEs
across time for word tokens he used in the past. And it seems that even
Dummett must concede that in some cases, at least, a speaker’s acceptance
of an explicit formulation of a semantical rule is indispensable to us, in the
234    

sense that without it we would lose our grip on which rules the speaker
was implicitly committed to at some earlier time. In such cases we would
lose our grip on which rules the speaker was implicitly committed to at
some earlier time if we did not rely in part on the speaker’s PJSSs and PJSEs
across time for some of his previous word tokens.
Against this, one might think that a speaker’s actual applications of
his sentences and words at a given time t always settle what criteria we
should use to evaluate those applications at t completely independently of
his own later acceptance of an explicit formulation of a semantical rule
to which he takes to have been implicitly committed at t. But this does
not fit well with the fact (stressed in Chapter 6) that in the midst of our
inquiries we trust our PJSSs and PJSEs across time more than we trust
any prior judgements that a particular criterion for applying our words
cannot be extensionally incorrect. If we are to base our understanding of
how a speaker’s commitment to a rule is manifested partly on our own
commonsense judgements about when a speaker is committed to a given
rule, then we must at least sometimes rely on a speaker’s PJSSs and PJSEs
across time.
This poses a second problem for the Dummettian claim that the experts’
PJSSs and PJSEs across time for tokens of g-o-l-d are mistaken. The
problem is that Dummett’s constraints on implicit knowledge and sense
don’t provide us with any criteria for judging in particular cases whether or
not our new applications of a word are properly guided by the (Dummettian)
sense of our own previous tokens of the word. The constraints therefore
leave us with no practical grip on when we can and when we cannot
rely on a speaker’s PJSSs and PJSEs across time. Dummett’s constraints
on implicit knowledge sense are supposed to motivate our acceptance of
(i)–(iv). On closer scrutiny, however, it is unclear how they could do so,
because they do not tell us exactly when we can and we cannot rely on
our PJSSs and PJSEs across time.
One might think it is easy to identify the Dummettian strong criterion
that guided English and Twin English-speakers’ applications of their tokens
of g-o-l-d in 1650—the criterion is ‘x satisfies a token of g-o-l-d in 1650 if
and only if x dissolves in aqua regia’. But this thought overlooks the crucial
distinction between taking a criterion such as ‘dissolves in aqua regia’ to be
strong—to settle the extension of tokens of g-o-l-d in 1650 in a way that
cannot be extensionally incorrect—and taking the criterion to be weak—to
     235

settle at best how such sentence tokens were actually evaluated in 1650,
without thereby guaranteeing that the criterion is extensionally correct. As
we saw in detail in Chapter 6, most chemists in 1650 were inclined to take
‘dissolves in aqua regia’ as a criterion for applying their tokens of g-o-l-d.
But this observation does not by itself settle whether the criterion was weak
or strong. And further details about their attitude to the criterion make it
difficult, at best, to defend the view that the criterion was strong. When
they confronted platinum for the first time, some were inclined to call it
‘gold’, but others hesitated, because they noticed, for instance, that samples
of platinum have a different melting point than paradigm samples of what
they called ‘gold’. The gold–platinum thought experiment is plausible
because there are different reasonable ways of weighing the various factors
that implicitly or explicitly inclined a person to apply tokens of g-o-l-d to
samples of platinum in 1650. The chemists and other speakers in the two
different linguistic communities in the gold–platinum thought experiment
apply their tokens of g-o-l-d differently, because, deliberately or not, they
weigh those factors differently. And so far we have found no good reason to
think that one way of weighing the factors is correct and the other incorrect.
Again, as I emphasized in Chapter 6, we trust our PJSSs and PJSEs across
time more than we trust any prior judgements that a particular criterion
for applying our words cannot be extensionally incorrect. The chemists
and other speakers in the two linguistic communities in my gold–platinum
thought experiment illustrate this general point. Trusting their own PJSSs
and PJSEs across time for their prior tokens of g-o-l-d, they take speakers
on Earth and on Twin Earth to have been committed to different criteria
for applying tokens of g-o-l-d in 1650. In short, we cannot discredit the
PJSSs and PJSEs across time for tokens of g-o-l-d that are accepted on Earth
or Twin Earth simply by noting that chemists in 1650 applied their tokens
of g-o-l-d to an object x if and only if x dissolves in aqua regia, since we
have good grounds for concluding that ‘dissolves in aqua regia’ was only a
weak, defeasible criterion for applying tokens of g-o-l-d in 1650.
To satisfy Dummett’s constraints on implicit knowledge and sense in a
way that addresses this problem, we would need to find compelling practical
grounds for concluding that a given speaker’s application of our word tokens
is guided by strong criteria and for distinguishing in practice between a
speaker’s strong and weak criteria for applying her word tokens. As the
gold–platinum thought experiment shows, however, speakers themselves are
236    

unable to distinguish between strong and weak criteria for applying their
word tokens in a way that meets the extensional equivalence requirement.
One might hope to save the extensional equivalence requirement by
second-guessing PJSSs and PJSEs that span very long periods of time. But
that would not be enough. For PJSSs and PJSEs that span very long periods
of time are sustained by chains of PJSSs and PJSEs that span very short
periods of time. Hence one would have to second-guess PJSSs and PJSEs
that span very short periods of time, including those we rely on in proofs,
such as the proof that a plane intersects a cylinder in an ellipse. As we
have seen, however, Dummett’s constraints leave us without any concrete
practical grounds for saying which, if any, of these PJSSs and PJSEs across
time are doubtful. His constraints therefore leave us in the dark about when
we should and when we should not trust our PJSSs and PJSEs—the very
judgements that his theory of sense is supposed to explain and justify.

7.7. Field on Partial Extension

Let us now examine a very different version of option (3), according to
which we should assign what Hartry Field (in Field 1973) calls partial
extensions to tokens of g-o-l-d in 1650. To see how, let’s suppose (first)
that the ex-use of ‘gold’ in 1650 did not rule out either of the extensions
that the members of the English-speaking community on Earth and the
Twin English-speaking community on Twin Earth would later assign to it,
(second) that both of these extension assignments are acceptable, and equally
so. These suppositions look attractive after our discussion of the difficulties
with Dummett’s version of (3). Assuming that both extension assignments
are acceptable, one might treat the set of gold things and the set of things
that are either gold or platinum as partial extensions of ‘gold’ in 1650. To
show how Field’s idea of a partial extension can be used to clarify option (3),
I shall also suppose that no other partial extension assignment is acceptable.
Nothing we have said so far supports this supposition. For now I accept it
without argument. I’ll evaluate it in §7.10 below.
Field’s idea of partial extension can be used to clarify the intuition that in
1650 some sentences containing occurrences of g-o-l-d are determinately
true, others are determinately false, and the rest are neither determinately
true nor determinately false. Field proposes that we explain what it is
     237

for a sentence to be determinately true or determinately false in terms of

structures that correspond to a sentence:
a structure for a sentence is a function that . . . maps each predicate [in the sentence]
into some set. The structure m corresponds to the sentence if . . . each predicate [in
the sentence] partially signifies the set that m assigns to it. Now, for each structure
m, we can apply the standard referential ( Tarski-type) semantics to determine
whether the sentence is m-true or m-false, i.e., true or false relative to m. ( To say
that the sentence is m-true is to say that it would be true if the . . . extensions of its
terms were as specified by m.) We can then say that a sentence is [determinately]
true (false) if it is m-true (m-false) for every structure m that corresponds to it.
Putting all these definitions together, we get definitions of [determinate] truth and
[determinate] falsity in terms of . . . partial signification. (Field 1973: 477)

We can apply this strategy to describe the indeterminacy in the extension

of tokens of g-o-l-d in 1650 as follows. What Field calls a structure for an
English or Twin English sentence used in 1650 is fixed by a translation of
that sentence into contemporary English. To simplify matters, let’s suppose
that the only English or Twin English predicate that is indeterminate in
1650 is ‘gold’. If we translate this predicate as ‘gold’ in contemporary
English, then we in effect assign the set of gold things as the partial
extension of ‘gold’ in 1650. But if we translate ‘gold’ as ‘either gold or
platinum’, we in effect assign the set of gold or platinum things to the
partial extension of ‘gold’ in 1650. Assume that all other English and Twin
English predicates that are ex-used in 1650 are translated into contemporary
English homophonically, thereby in effect assigning to those predicates, as
ex-used in 1650, the same extensions that the corresponding contemporary
English words have. Holding these homophonic translations fixed, one
structure for an English sentence used in 1650, structure m1 , is determined
by adding the homophonic translation of ‘gold’ into contemporary English,
and another structure for the sentence, structure m2 , is determined by
adding the translation of ‘gold’ as ‘either gold or platinum’. Finally, let’s
suppose that both m1 and m2 correspond to any English or Twin English
sentence containing tokens of g-o-l-d ex-used in 1650, in the sense that
they are each equally acceptable assignments of partial extensions to the
respective tokens of g-o-l-d in the sentences, and (again, without argument)
that no other partial extension assignments, and hence no other structures,
are acceptable.
238    

Now suppose that in 1650 Locke said, ‘‘This is gold,’’ pointing at his
gold ring. Since a gold ring is also either gold or platinum, the sentence
‘This is gold’, as used in 1650 by Locke as he pointed at his gold ring,
is both m1 -true and m2 -true, and hence by Field’s criterion determinately
true. Suppose that in 1650 Locke said, ‘‘This is gold,’’ pointing at a bar
of iron pyrites. Since a bar of iron pyrites is neither gold nor platinum,
the sentence ‘This is gold’, as used in 1650 by Locke as he pointed at the
bar of iron pyrites, is both m1 -false and m2 -false, hence by Field’s criterion
determinately false. Finally, consider again the case described above: in 1650,
Locke says, ‘‘There’s gold in those hills,’’ while pointing at hills in South
Africa that contain platinum but no gold. Under these circumstances, the
sentence ‘There’s gold in those hills’ is m1 -false and m2 -true, and hence by
Field’s criterion, neither determinately true nor determinately false. The
same points hold for Twin Locke’s utterances of these sentences. The only
difference is that the ring and the hills to which Twin Locke refers while
uttering those sentences are twins of the ring and the hills to which Locke
refers on Earth.

7.8. A Field-Style Argument against Options (1)

and (2)
Field’s strategy for assigning partial extensions to indeterminate predicates
does not tell us how to interpret the symbol E1 if we adopt option (3). I
propose that we use Field’s concepts of partial extension, structure, determinate
truth, and determinate falsity to define two different sets—T1 , the set of
objects of which tokens of g-o-l-d are determinately true in 1650, and
F1 , the set of objects of which tokens of g-o-l-d are determinately false
in 1650. T1 and F1 are well defined only if it is determinate which partial
extension assignments for ‘gold’ are acceptable. We supposed that the set of
gold things and the set of either gold or platinum things are the only acceptable
partial extension assignments for the English and Twin English word ‘gold’
as it was used in 1650. I’ll question this assumption below, after I explain
how one might combine it with Field’s methods to reject options (1) and
(2) for the gold–platinum thought experiment.
Let E1 = <T1 , F1 >. Under this interpretation, E1 is not an extension,
but it is like an extension in one crucial respect: it settles which objects
     239

tokens of g-o-l-d are determinately true of in 1650, and which objects

tokens of g-o-l-d are determinately false of in 1650. We can call E1 the
determinate core of an extension, or an extension core, for short. Similarly, let
E2 = <T2 , F2 >, where T2 is the set of objects of which tokens of g-o-l-d
are determinately true in English today, and F2 is the set of objects of
which tokens of g-o-l-d are determinately false in English today. Finally,
let E3 = <T3 , F3 >, where T3 is the set of objects of which tokens of
g-o-l-d are determinately true in Twin English today, and F3 is the set of
objects of which tokens of g-o-l-d are determinately false in Twin English
today. (Note that if tokens of g-o-l-d have an extension in English today,
the extension is T2 , and if tokens of g-o-l-d have an extension in Twin
English today, it is T3 .)
Under this interpretation of the diagram, E2 = E3 , which is to say that
<T2 , F2 > = <T3 , F3 >. Ordered pairs are identical just in case their first
members are identical and their second members are identical. In this case,
T2 = T3 , since ‘gold’ is determinately true of platinum things in Twin
English today, but ‘gold’ is not determinately true of platinum things in
English today; moreover, F2 = F3 , since ‘gold’ is determinately false of
platinum things in English today, but ‘gold’ is not determinately false of
platinum things in Twin English today.
From this interpretation of ‘E1 ’, ‘E2 ’, and ‘E3 ’, it follows that we
must reject both option (1), according to which E1 = E2 and E1 = E3 , and
option (2), according to which E1 = E2 and E1 = E3 . Here’s why. E1 = E2
amounts to <T1 , F1 > = <T2 , F2 >. But F1 = F2 . In particular, tokens of
g-o-l-d are determinately false of platinum things in English today, but
tokens of g-o-l-d are not determinately false of platinum things in English in
1650. Hence <T1 , F1 > = <T2 , F2 >, so (by definition) E1 = E2 . Similarly,
E1 = E3 amounts to <T1 , F1 > = <T3 , F3 >. But T1 = T2 . In particular,
in English in 1650 tokens of g-o-l-d are not determinately true of platinum
things, but in Twin English today, tokens of g-o-l-d are determinately
true of platinum things. Hence <T1 , F1 > = <T3 , F3 >, so (by definition)
E1 = E3 .
By conjoining these results we arrive at a version of option (3), according
to which E1 = E2 and E1 = E3 . Hence it seems we can use Field’s concepts
of partial extension, structure, determinate truth, and determinate falsity to define
extension cores <T1 , F1 >, <T2 , F2 >, <T3 , F3 >, whose assignment to E1 ,
E2 , and E3 , respectively, yields a version of option (3).
240    

7.9. A Field-Style Defence of Option (3)

Let us now see if we can defend option (3), interpreted in the way just
described. We were driven to reject options (1) and (2) in part by (M),
the entrenched metaphysical assumption that the ex-use of word tokens
determines their extension. Recall that in the diagram of §6.5, ‘E1 ’ is the
only symbol assigned to tokens of g-o-l-d during the period before the
accidental uncovering of large amounts of platinum in the Twin Earth
community—the period when by supposition the ex-use of tokens of
g-o-l-d is the same in English and Twin English. If ‘E1 ’ represents an
extension assignment, the standard presupposition that the ex-use of tokens
of g-o-l-d determines their extensions is built into the diagram. But in the
previous section we saw that we can interpret ‘E1 ’ as the extension core
<T1 , F1 >, where T1 is the set of objects of which tokens of g-o-l-d were
determinately true in 1650, and F1 is the set of objects of which tokens of
g-o-l-d were determinately false in 1650. If ‘E1 ’ is interpreted in this way,
then instead of the presupposition that the ex-use of tokens of g-o-l-d in 1650
determines their extensions, a slightly weaker metaphysical presupposition is
built into the diagram—the presupposition that the ex-use of tokens of g-o-l-d
in 1650 determines their extension cores—the (identical) sets of objects which tokens
of g-o-l-d, as ex-used in 1650, are determinately true of and the (identical) sets of
objects which tokens of g-o-l-d, as ex-used in 1650, are determinately false of.
More generally, if we adopt my proposed Field-style interpretation of the
standard diagram of options for the gold–platinum thought experiment, we
should replace (M) with the following metaphysical principle for predicates
(M ) For any two word-tokens w and w such that the utterance or
inscription of w occurs at some time t before the utterance or
inscription of w , a PJSE across time for w and w is true only if for
some set E,
(a) E is the extension core of both w and of w , and
(b) the totality of facts about the ex-use of w at times prior to or
identical with t (hence prior to the utterance or inscription of
w ) determines that E is the extension core of w.
This seems to be the strongest metaphysical principle that we can endorse if
we accept the standard diagram of options for the gold–platinum thought
     241

experiment.¹³ If we adopt (M ), then we must reject options (1) and (2) of
the gold–platinum thought experiment for the reasons sketched in the
previous section. In particular, if w is a token of g-o-l-d that Locke uttered,
and w is a token of g-o-l-d that I utter today when I make a PJSE for w,
then clause (a) fails, since there is no E that is the extension core of w and
the extension of w . Similarly, if w is a token of g-o-l-d that Twin Locke
uttered and w is a token of g-o-l-d that my twin on Twin Earth utters
today, then clause (a) fails, since there is no E that is the extension core of
w and the extension of w .
One might try to defend option (3), under the proposed Field-style
interpretation, by appeal to (M ). The gold–platinum thought experiment
might be taken to show that m1 and m2 (as defined in §7.7) are equally
acceptable structures for sentences used in 1650 that contain tokens of
g-o-l-d. It would then be tempting to conclude that any acceptable
interpretation of tokens of g-o-l-d in 1650, whether on Earth or Twin
Earth, must identify their core extensions—the sets of objects of which
those tokens were then determinately true and the sets of objects of which
they were then determinately false.

7.10. A Problem for the Field-Style Defence

of Option (3)
Although this Field-style defence of option (3) may seem more plausible
than a Dummett-style defence of option (3), it faces similar problems. To
see why, recall that T1 and F1 are well defined only if it is determinate which
structures correspond with sentence tokens that contained tokens of g-o-l-d

¹³ My goal in proposing (M ) is not to provide an accurate interpretation of Field’s position in

Field 1973, but to use his methods to articulate the metaphysical intuition behind option (3). It is not
clear that Field himself accepts (M ). He writes, for instance, that ‘‘[Newton’s term] ‘mass’ would have
undergone a simple refinement rather than a double refinement if everyone had followed Einstein’s
example in adhering to (4R) [which interprets ‘mass’ as ‘relativistic mass’]’’ (Field 1973: 479). This
suggests that, according to Field, what counts as an acceptable partial extension assignment for a term T
as used at some time t, and hence which objects are elements of what I call the core-extension of T at
t, depends not only on the ex-use of T up to and including t, but also on how members of the relevant
linguistic community later interpret T. If this is indeed what Field believes, then he is committed to
rejecting (M ). And if he rejects (M ), then the account of option (3) that I offer in this section applies
Field’s technical apparatus in ways that he himself would not apply it—namely, to try to articulate the
metaphysical intuition behind option (3).
242    

in 1650. To illustrate my proposed Field-style interpretation of option (3),

in the previous section I supposed that it is somehow settled which partial
extension assignments to tokens of g-o-l-d in 1650 are acceptable. I gave
no argument for this supposition, and noted that it does not follow from
my thought experiment. In the context of the proposed Field-style defence
of option (3), this supposition becomes especially important. The crucial
point is that the proposed Field-style defence of option (3) assumes that
the totality of facts about the ex-use of tokens of g-o-l-d at times prior
to and included in the year 1650 uniquely determines the core extension of
tokens of g-o-l-d in 1650—the set of objects of which those tokens were
determinately true and the set of objects of which they were determinately
false in 1650. Since core extension is defined in terms of acceptable partial
extensions, this assumption implies that the totality of facts about the ex-use
of tokens of g-o-l-d at times prior to and included in the year 1650
uniquely determines which partial extensions for tokens of g-o-l-d in 1650
are acceptable and which ones are not. If the totality of facts about the
ex-use of tokens of g-o-l-d at times prior to and included in the year 1650
leaves it indeterminate which partial extension assignments to those tokens
are acceptable, then, given (M ), the core extension of tokens of g-o-l-d in
1650 is undefined. Hence the proposed Field-style interpretation of option
(3) stands or falls with the assumption that the totality of facts about the
ex-use of tokens of g-o-l-d at times prior to and included in the year 1650
uniquely determines which partial extensions for tokens g-o-l-d in 1650
are acceptable and which ones are not.
The problem is that this assumption is no more plausible than the
assumption that the totality of facts about the ex-use of tokens of g-o-l-d at
times prior to and included in the year 1650 uniquely determines a classical
extension for tokens of g-o-l-d in 1650. If the gold–platinum thought
experiment discredits the latter assumption, similar thought experiments
will discredit the former assumption. To see why, it is enough to focus on
how the proposed Field-style defence of option (3) would define the first
component of the core extension of tokens of g-o-l-d in 1650, namely,
T1, the set of objects of which tokens of g-o-l-d are determinately true
in 1650. Suppose we initially conclude that according to Field’s theory of
partial extensions, the gold–platinum thought experiment implies that T1
is the set of all and only gold things—all and only bits of the element with
atomic number 79. To discredit this specification of T1 , it is enough to
     243

describe two linguistic communities whose ex-uses of tokens of g-o-l-d

are identical to the ex-uses of tokens of g-o-l-d on Earth until sometime in
the nineteenth century, when scientists in the two communities examine
paradigm instances to which they apply their tokens of g-o-l-d, but the
samples that are central and salient for the scientists are different in the two
communities, and so scientists in one of the communities conclude that
their tokens of g-o-l-d apply to any bit of the element with atomic number
79, whereas scientists in the other community conclude that their tokens
of g-o-l-d apply not to any bit of the element with atomic number 79,
but only to any bit of the naturally occurring isotope of the element with
atomic number 79.¹⁴ As before, we can stipulate that members of the two
communities encounter each other after scientists in the two communities
arrive at their respective analyses of the stuff to which they apply tokens
of g-o-l-d, and they see no conflict between their scientists’ conclusions.
Moreover, members of both communities trust their PJSEs across time for
tokens of g-o-l-d, and therefore assign different extensions to tokens of
g-o-l-d used in 1650. Faced with this scenario, it seems we must conclude
that both of these partial extension assignments are acceptable, and hence
that contrary to our initial supposition, T1 is not the set of all and only gold
things—all and only bits of the element with atomic number 79—but the
set of all and only bits of the naturally occurring isotope of the element with
atomic number 79.
But now we can easily describe a scenario that discredits this new
speculation about T1 . For instance, we can describe two communities
whose ex-uses of tokens of g-o-l-d are the same at least up to the end of
1650, but whose scientists later come to different conclusions about how
much of the naturally occurring isotope of the element with atomic number
79 counts as an object of which tokens of g-o-l-d are true. The scientists
in one of the two communities conclude that some clusters of atoms of the
naturally occurring isotope of the element with atomic number 79 are too
small to be in the extension of their tokens of g-o-l-d, whereas the scientists
in the other community conclude that no portion of such atoms is too small
to be in the extension of that community’s tokens of g-o-l-d—even a
single such atom is in the extension of that community’s tokens of g-o-l-d.

¹⁴ This is a variation on a thought experiment in Donnellan 1983. Donnellan describes a possible

future in which scientists take themselves to have discovered that their tokens of g-o-l-d are true of x
if and only if x is a bit of a particular isotope of the element with atomic number 79.
244    

Again, we may suppose that members of both communities trust their

PJSEs across time for tokens of g-o-l-d, and accordingly assign different
extensions to tokens of g-o-l-d used in 1650. Field’s theory then apparently
implies that T1 is not the set of all and only bits of the naturally occurring
isotope of the element with atomic number 79, but the set of all and only
large enough portions of the naturally occurring isotope of the element with
atomic number 79.
And now we can easily exploit the vagueness of ‘large enough’ to
construct a thought experiment that discredits this new speculation about
T1 . More generally, it is plausible to suppose that for any claim that the
totality of facts about the ex-use of tokens of a given orthographic word
type w at times prior to or identical with some given time t uniquely
determines that the core extension of tokens of w at t is identical with
some specified set S, we can construct a thought experiment that discredits
the claim. The problem is that given (M ), the core extension of tokens of
w at t is fixed only if the range of acceptable partial extension assignments
for tokens of w at t is fixed by the totality of facts about the ex-use of w
at times prior to or identical with t. But the range of acceptable partial
extension assignments for tokens of w at t is not fixed by the totality
of facts about the ex-use of w at times prior to or identical with t. To
suppose that the totality of facts about the ex-use of w at times prior
to or identical with t fixes a core extension for tokens of w at t one
must suppose that this totality of facts uniquely determines the boundary
between acceptable and unacceptable partial extension assignments, even
if we cannot always tell where that boundary lies. But the thought
experiments show that this supposition is no more plausible than the
assumption that the totality of facts about the ex-use of w at times prior
to or identical with t uniquely determines w’s (classical) extension. Hence
the Field-style version of option (3), whose main principle, (M ), rests
on this assumption, is not a stable and satisfying response to the thought
experiments.
Some philosophers assume that languages have built-in rules that set
constraints on acceptable partial extension assignments. For instance, the
super-valuation approach to vague predicates, such as ‘heap’, ‘bald’, and
‘thin’, uses meaning postulates to settle the boundary between accept-
able and unacceptable partial extension assignments. To rule out what
is called ‘‘higher order’’ vagueness—vagueness in the boundary between
     245

acceptable and unacceptable partial extension assignments to vague predic-

ates—proponents of the super-valuation approach simply adopt meaning
postulates that settle whether or not a partial extension assignment is
acceptable. According to the Field-style version of option (3) that we
are considering here, however, meaning postulates are irrelevant: the
ex-use of a predicate must by itself settle the boundary between accept-
able and unacceptable partial extension assignments.¹⁵ But we have now
seen that if the ex-use of a predicate does not uniquely determine its
classical extension, the ex-use of the predicate does not uniquely determ-
ine the boundary between acceptable and unacceptable partial extension
assignments, either.
Option (3) has content only if we suppose that totality of facts about
the ex-use of tokens of g-o-l-d at times prior to and included in the year
1650 settles something that we could plausibly regard as an interpretation
of E1 on the standard diagram for the gold–platinum thought experiment.
As we have seen, however, if we adopt Dummett’s approach, we are
left with no principled way to specify the implicit rules that guide a
speaker’s use of an expression, and thereby settle its supposed sense and
extension at a given time. Similarly, if we adopt the Field-style approach,
we are left with no principled way to settle which objects a particular
expression is determinately true of or determinately false of at a given
time. Like any version of option (3), Dummett’s version of option (3) and
the Field-style version of option (3) both imply that we must suspend
or reject our PJSEs across time for tokens of g-o-l-d. We could perhaps

¹⁵ One might be inclined to agree with Field that

the vagueness of our semantic terms serves no useful purposes, and as semantic theory develops we can
expect these terms to become more precise. On some ways of making them precise, [a claim near the
perceived borderline] will be absolutely true, on other ways of making them precise, it will not be
absolutely true but will have an extremely high degree of truth . . . The fact that there is no fact of the
matter as to whether [such a claim] is absolutely true, as we currently use ‘absolutely true’, is simply
due to the fact that our current use of ‘absolutely true’ is imprecise. (Field 1974: 227)
This kind of response does not address the problem raised in the text. According to the particular
application of Field’s approach that I am challenging, what a word token w ex-used at t is determinedly
true of is well defined only if

(∗ ) the set of acceptable partial extensions for w at t is uniquely determined by the totality of facts
about the ex-use of w at times prior to or identical with t.

But the thought experiments I sketched show that the truth value of (∗ ) cannot plausibly be settled
just by stipulating new, more precise meanings for ‘acceptable’, ‘partial extension’, and ‘uniquely
determined’.
246    

accept this extreme consequence if principles such as (M) or (M ) provided

us with plausible alternative criteria for sameness of extension or partial
extension across time. As we have seen, however, they do not. To adopt
option (3) is to lose one’s grip on the very idea of sameness of extension
across time.
 8
The Puzzle Diagnosed
and Dissolved

8.1. The Puzzle Reviewed

We saw in Chapter 6 that the token-and-ex-use model of words underlies
the apparently unobjectionable intuition that
(M) For any two word tokens w and w such that the utterance or
inscription of w occurs at some time t before the utterance or
inscription of w , a PJSE across time for w and w is true only if for
some set E,
(a) E is the extension of both w and w , and
(b) the totality of facts about the ex-use of w at times prior to or
identical with t (hence prior to the utterance or inscription of
w ) determines that E is the extension of w
where the only restriction on what counts as a fact about the ex-use of a
word token w is that the fact can in principle be described without using
any sentence that expresses a PIW, PJSS, or PJSE for w. Guided by the
token-and-ex-use the conception of words, we feel sure that (M) provides
the only possible framework for legitimating our PJSSs and PJSEs. But if
we accept (M) we are committed to the following diagram of options for
the gold–platinum thought experiment:
E2
b
E1
a d

c
E3
248     

Line ab represents the ex-use of tokens of g-o-l-d in our English-speaking

community, line ac represents the ex-use of tokens of g-o-l-d in the Twin
English-speaking community, and line ad, where lines ab and ac over-
lap, represents the supposition that before the accidental uncovering of
large amounts of platinum on Twin Earth in 1650—an event represented
by point d—members of the two communities have virtually the same
dispositions to ex-use tokens of g-o-l-d, associate the same mental states
with their tokens of g-o-l-d, and bear the same relations to the things
and substances in their environment, provided that these dispositions,
states, and relations are described independently of future developments
in either linguistic community. ‘E1 ’, ‘E2 ’, and ‘E3 ’ represent the exten-
sion of tokens of g-o-l-d at various points in the histories of the two
communities.
Recall that by hypothesis the two communities are isolated from each
other until after the development of analytical chemistry in the two
communities. Nevertheless, the standard diagram of options provides only
one symbol, ‘E1 ’, that represents the extension of tokens of g-o-l-d
during the period before the accidental uncovering of large amounts of
platinum on Twin Earth—the period when by supposition the ex-use of
tokens of g-o-l-d is the same in both communities. In this way the standard
commitment to (M), and hence to a token-and-ex-use conception of words,
is built into the diagram.¹ The rest of the diagram is straightforward: the
extension of tokens of g-o-l-d in the actual English-speaking community
after the branching is represented by ‘E2 ’, and the extension of tokens
of g-o-l-d in the counterfactual English-speaking community after the
branching is represented by ‘E3 ’.
Given this diagram and the natural assumption that E2 = E3 —the ex-
tension of tokens of g-o-l-d is different in the two linguistic communities
after the branching—it appears that there are only three options for
characterizing the extension of tokens of g-o-l-d before the branching:
either (1) E1 = E2 and E1 = E3 , or (2) E1 = E3 and E1 = E2 , or (3) E1 = E2
and E1 = E3 .² In Chapter 6 we saw that options (1) and (2) are arbitrary

¹ Recall that I understand the ex-use of tokens of g-o-l-d in 1650 to encompass all facts about tokens
of g-o-l-d in 1650 that can in principle be stated without using any sentence that expresses a PIW,
PJSS, or PJSE.
² As I noted in Chapter 6, this abstract characterization of the options (but not my thought
experiment or my interpretation of the options) is also adapted from Larson and Segal 1995: 534.
      249

and unacceptable for symmetrical reasons. And in Chapter 7 we saw that

to adopt any version of option (3) is to suspend or reject one’s PJSEs
and thereby to lose one’s grip on the very idea of sameness of extension
across time.
The problems with the options (1)–(3) stem from our commitment
to (M), and to token-and-ex-use conceptions of words more generally,
which sets the standard framework for our attempts to explain or justify
our PJSSs and PJSEs. Hence we face a puzzle: it seems that none of our
options—none of (1)–(3)—is acceptable. I see the puzzle as a reductio ad
absurdum of the token-and-ex-use model of words, which leads us to accept
(M) and the standard diagram of options.
In this chapter I shall elucidate the PJSS-based conception of words by
showing how it dissolves the apparent puzzle raised by the gold–platinum
thought experiment. I shall also investigate and reject the superficially
similar idea that the extension of tokens of a PJSS-based word type w at
time t can be settled by facts about how other tokens of w are ex-used at
some time after t.

8.2. The First Gold–Platinum Thought Experiment

and the Context Principle
Recall that at the heart of the PJSS-based conception of words lies the
context principle: never ask for the word type of a word token in isolation, but only
in the context of one’s own PIWs and PJSSs for that word token and other speakers’
PIWs and PJSSs for that word token, if such judgements exist. To accept the
context principle is to think of word types in such a way that one has
no conception of the word type of a given word token apart from one’s
own PIWs and PJSSs for that word token and other speakers’ PIWs and
PJSSs for that word token, if such judgements exist. I explained in §4.5
how we can explicate words in accordance with the context principle by
defining equivalence classes of word tokens, or marks. For present purposes,
however, we need not be explicit about what words are on the PJSS-based
conception; it is enough to note that the proper explication of words is
constrained by the context principle.
How will we identify and individuate the word types used by the two
linguistic communities of the first gold–platinum thought experiment if
250     

we adopt the context principle? To address this question we must first

highlight the relevant suppositions of the thought experiment:

(i) Members of both linguistic communities trust their PIWs, PJSSs,

and PJSEs across time for their respective tokens of g-o-l-d, and
continue to do so even after they learn of each other’s diverging
ex-uses of their tokens of those words.
(ii) The two linguistic communities are isolated from each other until
after the development of chemistry in each community.
(iii) Sometime after their contemporaneous development of chemistry,
members of the two communities encounter each other and come
to make PIWs, PJSSs, or PJSEs for some of each others’ word
tokens.
(iv) After members of the two communities encounter each other and
come to make PIWs, PJSSs, or PJSEs for some of each other’s word
tokens, they distinguish between the English word ‘gold’ and the
Twin English word ‘gold’, and take the two words to have different
extensions.

If we adopt the PJSS-based conception of words, then supposition (iii) is

crucial to the intelligibility of the gold–platinum thought experiment.³ The
crucial consideration is that if we (English-speakers) supposed, contrary to
supposition (iii), that the Twin Earth community were forever isolated
from us, then our understanding of the Twin English would be exhausted
by our stipulations about how members of the Twin English-speaking
community ex-use their word tokens. But the context principle implies
that we have no PJSS-based conception of word types used by members
of the Twin English-speaking community apart from the PIWs, PJSSs, and
PJSEs we come to make once we encounter them for the first time. In
particular, our grasp on the word types they used in 1650 is dependent on
PIWs, PJSSs, and PJSEs that link tokens of our word types to tokens of
their word types, both at a given time and across time. In short, the context
principle implies that our grasp on the word types used by members of the
Twin English-speaking linguistic community in 1650 is dependent on their
PIWs, PJSSs, and PJSEs across time.

³ I am indebted to Robert Gooding-Williams for helping me to see the importance of adding

supposition (iii) to my gold–platinum thought experiment.
      251

The principle also implies that our grasp on word types used by members
of our English-speaking linguistic community in 1650 is dependent on
our PJSSs and PJSEs across time, and those of other members of the
English-speaking community, through chains of such judgements that in
some cases extend from the present to at least as far back as 1650. Given the
context principle, supposition (i) licenses us to trust the PJSSs and PJSEs
of members of both linguistic communities for their respective tokens of
g-o-l-d, provided there are no special reasons for revising or rejecting
those PJSSs and PJSEs. Suppositions (ii)–(iv) together suggest (and I shall
hereafter assume) that there are no such reasons in the story for revising
or rejecting either of the two community’s PJSSs and PJSEs across time.
Hence given the context principle and suppositions (i)–(iv), we may trust
the PJSSs and PJSEs across time in the two linguistic communities.
For instance, in our English-speaking community today, speakers make
PJSSs and PJSEs across time for Locke’s tokens of g-o-l-d, and in the
Twin English-speaking community today, speakers make PJSSs and PJSEs
across time for Twin Locke’s tokens of g-o-l-d. We group these tokens
into different PJSS-based word types. Hence we (English-speakers) take for
granted that the extension of our PJSS-based word type ‘gold’ is the same
as the extension of Locke’s PJSS-based word type ‘gold’, and they (Twin
English-speakers) take for granted that the extension of their PJSS-based
word type ‘gold’ is the same as the extension of Twin Locke’s PJSS-based
word type ‘gold’. By the context principle and suppositions (i)–(iv) of
the thought experiment, we may simultaneously accept our PIWs, PJSSs,
and PJSEs for Locke’s PJSS-based word type ‘gold’ and the Twin-English
speakers’ PIWs, PJSSs, and PJSEs for Twin Locke’s PJSS-based word
type ‘gold’.
This will seem impossible to us if we accept the token-and-ex-use model
of word types, and principles such as (M) that follow from it. Since by
hypothesis Locke’s tokens of g-o-l-d and Twin Locke’s tokens of g-o-l-d
have exactly the same ex-use, the token-and-ex-use model of word types
implies that Locke’s tokens of g-o-l-d and Twin Locke’s tokens of g-o-l-d
are tokens of the same token-and-ex-use word type. As we have seen, this
consequence of the token-and-ex-use model of word types is built into
the standard diagram of options for describing the gold–platinum thought
experiment. If we adopt the context principle, however, we will not take
the facts about ex-use of tokens of g-o-l-d to settle questions about whether
252     

two word tokens of the same PJSS-based word type. We will instead base
our identifications of and distinctions between word types on our PIWs,
PJSSs, and PJSEs. And, as we have seen, our PIWs, PJSSs, and PJSEs in the
thought experiment lead to the conclusion that Locke’s tokens of g-o-l-d
and Twin Locke’s tokens of g-o-l-d are tokens of different PJSS-based word
types.
Hence if we adopt the PJSS-based conception of words, we can accept
a new diagram of options for the thought experiment:
E1
A B
b
a d
c
C D
E2

where line AB represents the PJSS-based word type ‘gold’ in the English-
speaking community, line CD represents the PJSS-based word type ‘gold’
in the Twin English-speaking community, E1 represents the extension of
the PJSS-based word type ‘gold’ in the English-speaking community, E2
represents the extension of the PJSS-based word type ‘gold’ in the Twin
English-speaking community, and, as before, line ab represents the ex-use
of tokens of g-o-l-d in our English-speaking community, line ac represents
the ex-use of tokens of g-o-l-d in the Twin English-speaking community.
This diagram of the options for the goal-platinum thought experiment
leaves open the possibility that we will accept the PIWs, PJSSs, and PJSEs
made by members of the English-speaking community and the PIWs, PJSSs,
and PJSEs made by members of the Twin English-speaking community.
In short, if we reject the token-and-ex-use model of word types and adopt
the PJSS-based conception of words, we will come to see that there is no
conflict between the two communities’ PIWs, PJSSs, and PJSEs.

8.3. The Second Gold–Platinum Thought

Experiment
To understand the new diagram of options for the gold–platinum thought
experiment, it helps to consider a superficially similar thought experiment
      253

in which tokens of g-o-l-d that come to have different ex-uses in two

isolated communities are all nevertheless tokens of the same PJSS-based
word type.⁴ Suppose that on Earth in 1651 the population of English-
speakers divides into two large groups that are from then on for many years
completely isolated from each other. In one of the groups the ex-use of
tokens of g-o-l-d is virtually the same as the ex-use of tokens of g-o-l-d
in the actual English-speaking community, and in the other the ex-use
of tokens of g-o-l-d is like the ex-use of tokens of g-o-l-d in the Twin
English-speaking community on the Twin Earth described in the first
gold–platinum thought experiment (described in §6.4). Suppose also that
just as on Earth and Twin Earth in the first thought experiment, members
of the two communities rely on their PIWs, PJSSs, and PJSEs across time
for their respective tokens of g-o-l-d. Finally, suppose that the isolation
between the two groups ends, and they learn that their PIWs, PJSSs, and
PJSEs across time for tokens of g-o-l-d extend back to tokens of g-o-l-d
used by members of the English-speaking community on Earth before
the split.
In this case members of the two groups are not likely to relinquish their
PIWs, PJSSs, or PJSEs across time for their respective tokens of g-o-l-d.
They will take themselves to disagree about the extension of a single PJSS-
based word type ‘gold’, and try to persuade each other of their views.
Even if they fail to come to an agreement about the extension of their
PJSS-based word type ‘gold’, they will not necessarily reject their PIWs,
PJSSs, or PJSEs across time for its tokens: members of each group may
stand by their judgements about the extension of their PJSS-based word
type ‘gold’ and reject those of the other group.
This thought experiment therefore contrasts with the first gold–platinum
thought experiment, in which by hypothesis the two linguistic communit-
ies are isolated until after their contemporaneous discovery of chemistry
and their subsequent accounts of the extensions of their respective tokens of
g-o-l-d. The context principle implies that in the first thought experiment
the tokens of g-o-l-d used by members of the two communities are tokens
of different PJSS-based word types. But in the second thought experiment
by hypothesis the divergent ex-uses of tokens of g-o-l-d are linked by chains
of PIWs, PJSSs, and PJSEs to tokens of a single unambiguous PJSS-based

⁴ This thought experiment is a development of one sketched in Ebbs 2000: n. 27.

254     

word type that was used prior to 1651, when the English-speaking com-
munity split into two isolated ones. Let’s suppose, in addition, that there
are no contextually salient reasons for revising those PIWs, PJSSs, or
PJSEs. Then in the second thought experiment, unlike in the first, the
context principle implies that tokens of g-o-l-d that are ex-used in the
two isolated communities are tokens of a single unambiguous PJSS-based
word type.
One might object that if we can make sense of the new diagram
of options for the first gold–platinum thought experiment, then we
should also be able to make sense of the possibility that in the second
gold–platinum thought experiment, in 1650, hence prior to the formation
of the two isolated communities, tokens of g-o-l-d had both of the two
extensions that English-speakers later come to attribute to it after the
original English-speaking community splits into two isolated ones. But this
objection overlooks the fact that the new diagram of options applies to
the first gold–platinum thought experiment only if we adopt the context
principle, which implies that the tokens of g-o-l-d on Earth in 1650 are
tokens of a PJSS-based word type that is different from the PJSS-based word
type of which the tokens of g-o-l-d on Twin Earth in 1650 are tokens.
But, for reasons I explained above, if we adopt the context principle, we
have no grounds for doubting that in the second thought experiment, the
tokens of g-o-l-d on Earth in 1650 are tokens of a single PJSS-based word
type. To suppose that in 1650, prior to the split, tokens of g-o-l-d had
both of the two extensions that English-speakers later come to attribute
to it would be to suppose, contrary to the context principle, that the
English-speakers’ PIWs, PJSSs, and PJSEs do not keep track of the word
types they use. Against this, I have urged that we adopt the context
principle, which implies that if members of both linguistic communities
are apprised of all of the relevant facts about their PIWs, PJSSs, and PJSEs,
both at a given time and across time, at least as far back as 1650, and they
continue to regard tokens of g-o-l-d as unambiguous in 1650, there are
no good grounds for questioning that tokens of g-o-l-d as unambiguous
in 1650.⁵

⁵ This leaves open the possibility that the speakers themselves conclude that their tokens of g-o-l-d
have different extensions, despite their initial confidence in their PJSSs and PJSEs, and hence that the
entrenched beliefs in the two communities are compatible. I take for granted that the members of
the two communities will resist this conclusion not because they presuppose a general metaphysical
      255

If we adopt the context principle, we will diagram the options for the
second gold–platinum thought experiment as follows:
E2
b
E1
a d

c
E3

where, just as in the diagram of the options for the first gold–platinum, line
ab represents the ex-use of tokens of g-o-l-d in one of the two branches of
the English-speaking community, line ac represents the ex-use of tokens
of g-o-l-d in the other branch of the English-speaking community, and
line ad, where lines ab and ac overlap, represents the ex-uses of tokens
of g-o-l-d before the original English-speaking community splits into two
isolated ones—an event represent by point d—and ‘E1 ’, ‘E2 ’, and ‘E3 ’
represent the extension of tokens of g-o-l-d at various points in the histories
of the two communities. In this diagram, unlike in the diagram of the
options for the first gold–platinum, however, both lines ab and ac represent
continuity of a single PJSS-based word type ‘gold’.
The lines and letters of the diagram of the options for the second
gold–platinum thought experiment and the standard diagram of options
are identical to the lines and letters of the diagram of the first gold–platinum
thought experiment. In both diagrams, there is only one symbol, ‘E1 ’, that
represents the extension of tokens of g-o-l-d during the period before
the events represented by the forking lines at their respective points
d. There is a fundamental difference between what the two diagrams
convey, however. The diagram of options for the first thought experiment
reflects the standard commitment to (M), and hence to a token-and-ex-
use conception of words, whereas the diagram of options for the second
thought experiment is a reflection of our adoption of the context principle,
which, as we have seen, is incompatible with (M).
In the second gold–platinum thought experiment, unlike in the first, the
context principle commits us to denying that E2 = E3 unless the speakers
themselves come to relinquish some of their PIWs, PJSSs, or PJSEs, a

principle such as (M), but because they trust their PJSSs and PJSEs unless they see some special reason
in a given context for not doing so.
256     

possibility we have hypothetically ruled out. As I have emphasized, in the

circumstances described in the second gold–platinum thought experiment
the context principle implies that there is a single unambiguous PJSS-
based word type ‘gold’ that members of both communities use. A single
PJSS-based word type has only one extension. Hence the context principle
implies that E1 = E2 and E1 = E3 . By the transitivity of identity, E1 = E2
and E1 = E3 together imply E2 = E3 . Hence the context principle implies
that in the second gold–platinum thought experiment, unlike in the first,
members of the two linguistic communities disagree with each other about
the extension of a single, unambiguous, PJSS-based word type ‘gold’.
In practice, however, they are likely to focus not on this metalinguistic
disagreement, but on the disagreement about whether gold is the element
with atomic number 79 that it mirrors.
We are now in a position to see that if we accept the token-and-ex-
use model of words, we are likely to mistake some of the commitments
that result from our PIWs, PJSSs, and PJSEs with commitments to (M).
From our perspective as participants in inquiries about gold, we cannot
help but find our PJSSs, and PJSEs for ‘gold’ factual and trustworthy.
As we have seen, these judgements themselves naturally incline us to
diagram the options for the second thought experiment in a way that
looks superficially like the standard diagram of options of the first thought
experiment. The token-and-ex-use conception of words implies that in
both the first and second gold–platinum thought experiments, all the
tokens of g-o-l-d used in 1650 have the same extension. Therefore, if we
tacitly accept the token-and-ex-use model of words, we will naturally but
unwittingly conflate the conclusion we draw from our PIWs, PJSSs, and
PJSEs in the second gold–platinum thought experiment—the conclusion
that all the tokens of g-o-l-d used in 1650 have the same extension—for
a metaphysical intuition that in the first gold–platinum thought experiment
all the tokens of g-o-l-d used in 1650 have the same extension. Once
we adopt the context principle, however, we become aware of the
conflation, and can therefore distinguish between the conclusion that in
the second gold–platinum thought experiment all the tokens of g-o-l-d
used in 1650 have the same extension, on the one hand, and the dubious
metaphysical intuition that in the first gold–platinum thought experiment
all the tokens of g-o-l-d used in 1650 have the same extension, on
the other.
      257

8.4. The Thesis that the Extension of a PJSS-Based

Word Type is Determined by Facts about
the Ex-Uses of Some of its Tokens
You might now be persuaded that we should reject the token-and-ex-use
model of words, and embrace instead the PJSS-based conception of words,
according to which the question whether or not two word tokens are
tokens of the same word type cannot be raised or settled independent of
our PIWs, PJSSs, and PJSEs for the tokens. You might still be tempted
to suppose, however, that the extension of a PJSS-based word type is
determined by facts about the ex-uses of some of its tokens. You might
suppose, in particular, that
(a) In the first gold–platinum thought experiment, the extension of the
English PJSS-based word type ‘gold’ is determined by the ex-use of
‘gold’ on Earth in 1750, after speakers on Earth doubted that they
should apply their word ‘gold’ to samples of (what we now call)
platinum, but before scientists on Earth agree on a chemical analysis
of the stuff they call ‘gold’.
and
(b) In the first gold–platinum thought experiment, the extension of
the Twin-English PJSS-based word type ‘gold’ is determined by the
ex-use of ‘gold’ on Twin Earth in 1750, after speakers on Twin
Earth began confidently applying their word ‘gold’ to samples of
(what we now call) platinum, but before scientists on Twin Earth
agree on a chemical analysis of the stuff they call ‘gold’.
You might have been inclined to accept that the extension of the
English PJSS-based word type ‘gold’ is different from the extension
of the Twin English PJSS-based word type ‘gold’ because they presup-
pose (a) and (b). If so, it may seem to you that my argument against
the token-and-ex-use model of words presupposes that the extension of
a PJSS-based word type is settled by facts about the ex-uses of some of its
tokens.
It is important to see, however, that the first gold–platinum thought
experiment does not commit us to (a) or (b). The thought experiment
258     

assumes that members of the English-speaking community find it natural

to conclude that
(∗ ) The extension of the English PJSS-based word type ‘gold’ is different
from the extension of the Twin English PJSS-based word type ‘gold’.
One might suppose that if (∗ ) is true, its truth must somehow be settled
by the ex-uses of tokens of the respective words in the two communities.
But this supposition is as dubious as the thesis that the ex-uses of word
tokens in 1650 determined the extension of those tokens. To see why,
it is enough to construct a thought experiment in which two linguistic
communities whose ex-uses of tokens of their respective PJSS-based word
types are identical to the ex-uses of tokens of the PJSS-based word type
‘gold’ on Earth until sometime in the nineteenth century, when they
examine paradigm instances to which they apply their word types, and
reach different conclusions partly because the samples that are central and
salient for the scientists are different in the two communities.⁶ As before,
we can stipulate that after they arrived at their respective analyses of the stuff
to which they apply tokens of their respective PJSS-based word types, the
members of the two communities encounter each other, and conclude that
there is no conflict between their conclusions. We may suppose, finally,
that members of both linguistic communities are right about the extensions
of their words.
A thought experiment of this sort could be used to show that if we adopt
the context principle and endorse the PJSS-based conception of words, we
should give up (a). A similar thought experiment could be used to discredit
(b). More generally, for any claim that some specified set of ex-uses of
tokens of a PJSS-based word type w determines that w’s extension is some
specified set we can construct a thought experiment that undermines the
claim. To resist the conclusions of the thought experiments simply because
they conflict with the metaphysical idea that some facts about ex-uses of
tokens of a PJSS-based word type w determine w’s extension, one would
have to question our PIWs and PJSSs for word tokens used by members
of the two communities, either before or after they encounter each other,
simply because these PIWs and PJSSs conflict with the metaphysical idea.
And this would be to reject the context principle.

⁶ I presented one such thought experiment in Ebbs 2000: §11.

      259

This reasoning does not establish that the context principle is incompat-
ible with the idea that some facts about ex-uses of tokens of a PJSS-based
word type w determine w’s extension, but it does suggest a weaker, meth-
odological conclusion that is directly relevant to the pragmatic approach to
truth and words that I favour: even if it is possible that some facts about
ex-uses of tokens of a PJSS-based word type w determine w’s extension,
this possibility makes no difference to our actual assessments of our PIWs,
PJSSs, and PJSEs, since by the context principle, we have no better grounds
for evaluating sameness and difference of extension than by relying on our
PIWs, PJSSs, and PJSEs, together with our investigations into the truth or
falsity of particular statements we express by using our words. I conclude
that we can adopt the context principle without committing ourselves to
the thesis that the extension of a PJSS-based word type is determined by
facts about the ex-uses of some of its tokens.
We may nevertheless still wonder whether the thesis that extension of
a PJSS-based word type is determined by facts about the ex-uses of some
of its tokens is compatible with the version of the Tarski–Quine thesis
that I have sketched. Either the thesis that the extension of a PJSS-based
word type is determined by facts about the ex-uses of some of its tokens
is compatible with the PJSS-based conception of words or it isn’t. If it is
compatible with the PJSS-based conception of words, then for reasons I
have explained, the context principle implies that it is irrelevant to truth and
satisfaction. If the thesis is incompatible with the PJSS-based conception of
words, then, since it presupposes the PJSS-based conception of words, it
is self-undermining. Either way, the thesis is not a threat to my version of
the Tarski–Quine thesis.
Still, the thesis is intriguing. If it is coherent, it may be true. And
if it is both coherent and true, then the extension of tokens of the
PJSS-based word type ‘gold’ uttered in 1650 may be determined by facts
about how other tokens of the PJSS-based word type ‘gold’ are ex-used
at some time after 1650. More generally, if the extension of a PJSS-
based word type is determined by facts about the ex-uses of some of its
tokens, then
(TE) For some PJSS-based word types w, the extension of tokens of w at a
given time t is or will be determined by facts about how other tokens of w
are ex-used after t.
260     

(Here ‘TE’ is short for ‘temporal externalism’, a phrase coined by Henry

Jackman, whose views I shall discuss below.) In the next three sections I
shall investigate two recent discussions of sameness of extension across time
that apparently support (TE). I shall try to show that these attempts fail
because (TE) both presupposes and is incompatible with the PJSS-based
conception of words.

8.5. George Wilson’s Riverdale Olympics Case

George Wilson describes a case in which a committee of high-school
coaches settles standards for qualifying to participate in a local athletic con-
test, the Riverdale Olympics. Athletes who want to qualify must perform
in April, but the precise standards for qualification are not decided by the
committee until May. Wilson stipulates that everyone who participates in
the try-outs agrees to this arrangement. In these circumstances, he points
out, it would be misleading to say that the standards for qualifying changed
from April to May, since all along it was understood that they would be
decided by the committee in May. Wilson thinks this is a case in which
what happens later fixes standards for an earlier time, yet does not change
those standards. Analogously, he reasons, it may be that our discoveries
about the extension of ‘gold’ fix the extension of ‘gold’ in 1650 without
changing its extension. Wilson expresses his conclusion as follows: ‘‘uses of
‘gold’ before t may be answerable to truth standards that came to be fixed
only after t’’ (Wilson 2000: 95).
This way of phrasing the thesis is ambiguous between a weaker, epi-
stemological version, and a stronger, metaphysical one. The ambiguity
is due to ambiguities in the words ‘use’ and ‘fixed’. To say that uses
of ‘gold’ before t may be answerable to truth standards that came to be
fixed only after t may be just to say that beliefs we express by (trans-
parently) using a PJSS-based word type w at some time after t settle how
we should (at that later time) evaluate beliefs that we or others expressed by
using w before t. I am committed to this epistemological thesis by my
adoption of the context principle. However, to say that uses of ‘gold’
before t may be answerable to truth standards that came to be fixed only
after t may be to say something stronger: that certain ex-uses of tokens
of a PJSS-based word type w that come to exist after t metaphysically
      261

determine w’s extension before t. Wilson apparently endorses this stronger

thesis as well.
Wilson endorses my recommendation that we accept the PIWs, PJSSs,
and PJSEs in both linguistic communities of the first gold–platinum thought
experiment.⁷ I argued above that we cannot accept these PIWs, PJSSs, and
PJSEs unless we adopt the PJSS-based conception of words. If this argument
is correct, then Wilson’s metaphysical thesis amounts to a version of (TE).
Let us now examine this thesis.
To begin with, we should acknowledge that Wilson’s Riverdale
Olympics case is similar in some respects to the first gold–platinum thought
experiment. For instance, we find it natural to say that in 1650 John
Locke did not know whether there was gold in the hills of South
Africa, in part because those hills had not yet been mined. Similarly, in
April of the year of the Riverdale Olympics, athletes who performed
in the try-outs do not know whether they qualify for the Riverdale
Olympics.
But the similarity is superficial: the reasons why Locke did not know in
1650 whether there was gold in the hills of South Africa are fundamentally
different from the reasons why athletes who performed in the try-outs do
not know in April whether they qualify for the Riverdale Olympics high
jump contest. To see why, it helps to look more closely at the details of
Wilson’s story. According to Wilson, prior to the meeting of the committee
of coaches in May, it is already settled that to qualify to participate in the
high jump contest, one must jump at least as high as 6 feet in the try-outs
in April. Moreover, it’s already settled that any jump in the try-outs of
over 6 ft 3 in qualifies. What the coaches must decide in May is where
within the range from 6 ft to 6 ft 3 in to set the standard for qualifying
for the high jump contest. In Wilson’s story, they finally settle on 6 ft
2 in. Now suppose one of the athletes, Fred, jumped 6 ft 1 in during the
try-outs in April. Suppose that right after he jumps, he sees that he jumped
6 ft 1 in; we then ask him, ‘‘Did you qualify?’’ and he replies, ‘‘That’s not
yet decided—the coaches will decide in May whether 6 ft 1 in is enough

⁷ More precisely, Wilson grants that my gold–platinum thought experiment shows that the use of
the term ‘gold’ up to 1650 does not determine that ‘gold’ was true of x iff x was (as we say) gold. He
does not challenge the PJSSs and PJSEs in either linguistic community, and takes them for granted in
his account of how later usage can settle the extensions of the two terms spelled g-o-l-d that are used
on Earth and Twin Earth, respectively, in 1650. See Wilson 2000: 90–1.
262     

to qualify for the high jump contest.’’ This remark is true, according to
Wilson’s story.⁸
In contrast, however, whether in 1650 there was gold in the hills of
South Africa is not settled by anyone’s decision about the standards for
applying the word ‘gold’. If we trust our PIWs, PJSSs, and PJSEs, we will
not regard Locke’s utterance of ‘There is gold in those hills’ as a prediction
about future language ex-use or about the beliefs of chemists—we will not
regard Locke’s predicate ‘is gold’ as equivalent to ‘will be called ‘gold’ in
the future by some experts in my linguistic community’, for example.⁹ We
will take for granted that Locke’s utterance of ‘There is gold in those hills’
was true or false in 1650, before later developments in either of the two
future groups of speakers that can trace their word ‘gold’ back to his in a
chain of PIWs, PJSSs, and PJSEs. By trusting our PIWs, PJSSs, and PJSEs
across time for ‘gold’, we commit ourselves to rejecting the idea that the
truth values of claims we make by using our PJSS-based word type ‘gold’
at a given time t are settled by facts about how tokens of that PJSS-based
word type are ex-used after t.
Note also that someone may coherently challenge the current view that
in 1650 there was no gold in the hills of South Africa. It makes sense,
even if it is false, to say, ‘‘Gold is not in fact the element with atomic
number 79, despite what the experts say now or will say in the future.’’ The
currently entrenched belief that gold is the element with atomic number
79 is not an analytic claim made true by stipulation; it is deeply held,
but possibly false. By contrast, in Wilson’s story the coaches could not be
wrong about the standards for qualifying for the high jump contest: Fred’s
jump of 6 ft 1 in could not qualify him for the high jump contest if the
coaches decide otherwise in May. Fred may feel that the coaches’ decision
is unfair, or that it should have been different. But the coaches cannot
be wrong about whether he qualifies: their decision settles whether Fred
qualifies.
The contrast disappears on the weaker, epistemological, reading of
Wilson’s claim that ‘‘uses of ‘gold’ before t may be answerable to truth

⁸ Wilson asserts that ‘‘It is not the case that facts about the coaches’ guidelines and about regular
season high jumps determined that a tryout jump qualifies the jumper for the Riverdale Olympics iff
that jump is 6 2 or higher’’ (Wilson 2000: 93).
⁹ Here I agree with Jessica Brown, who objects to Jackman’s view on the grounds that evidence
about the usage of a word w at time t would not generally be regarded as relevant to the truth of an
utterance made by using w at some time before t. See Brown 2000: 178–88.
      263

standards that came to be fixed only after t’’ (Wilson 2000: 95). For on this
weaker reading, we need not suppose that the analogue of the coaches’
decision for the case of ‘gold’ is a metaphysical determination of the extension
of ‘gold’ by its use after 1650. One need only maintain that sometime after
1650, speakers come to agree on a characterization of the extension of
‘gold’. Their agreement need not be thought to settle the extension of
‘gold’ in a metaphysical way, and therefore need not be vulnerable to the
criticism of the previous two paragraphs.
Nevertheless, Wilson takes the high jump case to establish (TE), accord-
ing to which for some PJSS-based word types w, the extension of tokens of w at
a given time t is or will be determined by facts about how other tokens of w are
ex-used after t. Wilson’s description of the role of the coaches in settling the
standards for qualifying for the high jump contest implies that if the coaches
had never met to settle the standards, it would have been indeterminate
whether or not a high jump of 6 ft 1 in meets the standards for qualifying
for the contest. At the same time, Wilson maintains that in the actual
world the coaches’ decision did not actually change standards for qualifying
in April. Hence he seems committed to the curious claim that given how
things actually developed, it was always the case in April that the qualifying
standards for the high jump were 6 ft 2 in, even though in the world in
which the coaches never decide the standard, it is indeterminate in April
what the qualifying standards for the high jump are.
Similarly, on the metaphysical reading, Wilson’s thesis that ‘‘uses of ‘gold’
before t may be answerable to truth standards that came to be fixed only
after t’’ (Wilson 2000: 95) implies that if the English-speaking community
had ceased to exist in 1650 (perhaps because a large asteroid destroyed
life on Earth in 1650), then the extension of the tokens of g-o-l-d would
have been indeterminate between (at least) the two extensions described
in the first thought experiment. At the same time, according to Wilson,
developments in the ex-use of tokens of g-o-l-d in the actual world did not
change the extensions of tokens of g-o-l-d in 1650. Hence he is committed
to saying both that in the actual world it was always the case that the
extension of the English word ‘gold’ in 1650 was the set of all and only
bits of the element with atomic number 79, just as it is now, and that
later uses of the word ‘gold’ fixed, or made it the case that, the extension
of the English word ‘gold’ in 1650 was the set of all and only bits of
the element with atomic number 79. But how could one determine (in
264     

the sense of ‘‘make it the case’’) that p is true if it is already determined

that p is true?

8.6. Henry Jackman’s Temporal Externalism

This puzzle is only apparent, according to Henry Jackman, whose view
of sameness of extension across time is similar to Wilson’s. Jackman
distinguishes between the question whether in 1650 the extension of ‘gold’
was the set of all and only bits of the element with atomic number 79,
on the one hand, and the question whether it was settled in 1650 that the
word ‘gold’ was the set of all and only bits of the element with atomic
number 79, on the other. According to Jackman, the ex-uses of the English
word ‘gold’ after 1650 settled that the extension of ‘gold’ in 1650 was
the set of all and only bits of the element with atomic number 79. In
the actual world in 1650 the extension of the English word ‘gold’ was
unsettled, but nevertheless the same as, the extension of the English word
‘gold’ today.
This distinction does not entirely remove the puzzle, however. For
Jackman’s theory implies that in the counterfactual world in which the
English-speaking community ceases to exist in July of 1650, the extension
of the English word ‘gold’ was both unsettled in July of 1650 and different
from its extension in the actual world in July of 1650—namely, the set of
all and only bits of the element with atomic number 79.¹⁰ To recover the
puzzle, we need only ask why the extension of the English word ‘gold’
in July of 1650 in this counterfactual world is different from its extension
in July of 1650 in the actual world. The answer implied by Jackman’s
theory is it is never settled in that counterfactual world that the extension
the English word ‘gold’ is the set of all and only bits of the element with
atomic number 79. It follows that if one settles, at some time t2 , what the
extension of a word w is at some previous time t1 one thereby makes it the
case that the extension of w at t1 is different from what it would have been
at t1 if it had not been settled at some time after t1 .

¹⁰ As Jackman himself notes, it follows from his theory that in Mark Wilson’s Druid thought
experiment, ‘‘If the Druids’ use of ‘ave’ died out before they encountered planes, the meaning of ‘ave’
would have been indeterminate between flying thing and bird’’ ( Jackman 1999: 168). He embraces
the consequence.
      265

Jackman explicitly compares this consequence of his theory with the

observation that the truth values of statements that contain what he
calls ‘‘temporally loaded expressions’’ depend on what happens in the
future ( Jackman 1999: 162). Suppose, for example, that Cheney shoots
Whittington at time t, and Whittington dies later from the gunshot wounds.
If Whittington had not died of the gunshot wounds that Cheney caused
by shooting him at t, then Cheney would not have killed Whittington by
shooting him at t. Looking back, we might loosely say that since Cheney
caused Whittington’s later death by shooting him at t, Cheney killed
Whittington at t. This apparently implies, paradoxically, that Cheney killed
Whittington before he died. Analogously, on Jackman’s view, in the actual
world the extension of the English word ‘gold’ was the set of all and only
bits of the element with atomic number 79 in 1650, before the extension
of the English word ‘gold’ in 1650 was settled, hence, paradoxically, before
it became true that the extension of the English word ‘gold’ in 1650 was the
set of all and only bits of the element with atomic number 79.
Jackman thinks that this paradox is no more problematic than the
misleading statement that Cheney killed Whittington before he died. In both
cases, he thinks, we can avoid genuine paradox by noting that some
words, such as ‘killed’, are ‘‘temporally loaded’’. The sentence ‘Cheney
killed Whittington’ is not true until Whittington dies. Strictly speaking the
sentence ‘Cheney killed Whittington’ was not true at t; the sentence became
true after t, when Whittington died from his wounds. Looking back, we
can see that at t, Cheney initiated the events that would eventuate in
Whittington’s death, hence that the sentence ‘Cheney killed Whittington’
was true at t. Similarly, Jackman supposes, the token of the sentence ‘There’s
gold in those hills’ that Locke uttered while pointing at hills in South Africa
is not true (or false) until after some members of his community settled the
meaning of Locke’s word ‘gold’ in such a way that the word is true (or
false) of some of the metal in those hills. In this case, according to Jackman,
time affects not just the truth value of the claim, but the meanings of the
words used to express it: Locke said something that was determinedly true
or false in 1650, even though what Locke said was not settled until some
time after 1650.
The superficial similarity between these two cases does not dispel the
air of paradox surrounding Jackman’s account of what Locke said in 1650.
The problem is that our understanding of what Locke said in 1650 is not
266     

like our understanding of what Cheney did at time t. We take for granted
that future events may change our understanding of what Locke said in 1650,
not its truth value. Yet Jackman claims that future events change the truth
value of what Locke said in 1650, in the modal sense that if we had settled
the meaning of his tokens of g-o-l-d differently at some time after 1650,
what Locke said in 1650, hence also the truth value of his utterance, would
have been different. Thus it seems we must conclude that what Locke said
in 1650 is not settled until the meaning of his tokens of g-o-l-d is settled
some time after 1650, and therefore that in 1650 it was not (yet) true that
Locke’s utterance was true or false.
This puzzling description of Locke’s utterance is an inevitable con-
sequence of Jackman’s assumption that if members of a linguistic community
arrive at an entrenched and stable set of beliefs that they express by using
a given term, say ‘gold’, then the combined ex-uses of that term in the
community constitute an ‘equilibrium’ for the term that settles (a) the
term’s meaning, and hence also (b) the contents of the entrenched beliefs
that members of the community express by using the term, in such a way
that (c) most of the entrenched beliefs that members of the community
express by using the term are true. For many of our terms, according to
Jackman, there is more than one accessible equilibrium—more than one
way in which the ex-use of the term could be extended and developed into
a coherent practice of ex-uses of the term that settle its meaning ( Jackman
1999: 160–1). In the first gold–platinum thought experiment (described in
§6.4), for instance, the ex-use of tokens of g-o-l-d in the two communities
in 1650 is identical, yet, after gradual and plausible extensions of that ex-use,
each community arrives at different entrenched beliefs expressed by using
tokens of g-o-l-d. In this and many other similar cases, Jackman supposes,
the two communities arrive at different but equally valid equilibria for their
respective tokens of g-o-l-d. Their different ex-uses thereby settle different
meanings for their respective tokens of g-o-l-d, and those meanings settle
that the entrenched beliefs they express by using their respective tokens of
g-o-l-d, though different, are mostly true.¹¹
Jackman combines this theory of meaning-settling equilibria with our
PIWs, PJSSs, and PJSEs, and infers that ‘‘the future behavior of an

¹¹ This description presupposes that members of each community sort its respective word tokens
into PJSS-based word types.
      267

individual or his society can affect the content of his [previous] thoughts
and utterances’’ ( Jackman 1999: 160). In 1650 on Earth, for instance,
when Locke said, ‘‘There is gold in those hills,’’ it was not yet settled
that an object x satisfies Locke’s word ‘gold’ if and only if x is a bit
of the element with atomic number 79. Jackman supposes that on Earth
today there is an equilibrium for the English word ‘gold’ that settles
that an object x satisfies ‘gold’ if and only if x is a bit of the element
with atomic number 79. We also trust our PIWs, PJSSs, and PJSEs for
Locke’s term ‘gold’. Looking back, we say that even in 1650, before
the meaning of Locke’s term ‘gold’ was settled, an object x satisfied
Locke’s word ‘gold’ if and only if x was a bit of the element with
atomic number 79. Members of the Twin English-speaking community
today will reach an analogous conclusion for Twin Locke’s term ‘gold’.
If we accept the PIWs, PJSSs, and PJSEs in both communities, then
if the equilibria reached in the two communities settle the meanings of
the respective terms used in those communities, as Jackman assumes, the
equilibria somehow also retroactively settle the meanings of Locke’s and
Twin Locke’s terms.¹²
Suppose provisionally that we can make sense of the idea that the ex-uses
of ‘gold’ today retroactively settle the meaning of Locke’s word ‘gold’ in
1650. Does this fit with our PIWs, PJSSs, and PJSEs for Locke’s word
‘gold’? I have stressed that to accept these PIWs, PJSSs, and PJSEs is to
suppose that in 1650 the beliefs that Locke expressed by using ‘gold’ were
not indeterminate, but true or false. Jackman tries to accommodate this
consequence of our PIWs, PJSSs, and PJSEs by distinguishing between
the question whether in 1650 an object x satisfies Locke’s word ‘gold’
if and only if x is a bit of the element with atomic number 79, on the
one hand, and the question whether it was settled in 1650 that an object
x satisfies Locke’s word ‘gold’ if and only if x is a bit of the element
with atomic number 79, on the other. He wants to say ‘‘Yes’’ to the first
question and ‘‘No’’ to the second question. This way of answering the
two questions does not directly conflict with our PIWs, PJSSs, or PJSEs in

¹² Jackman’s theory of meaning equilibria is not unlike Tim McCarthy’s neo-Davidsonian theory
of reference, which yields ‘‘a policy favoring the stability of the reference of natural kind terms over
time’’ (McCarthy 2002: 152), where the facts that are relevant to determining the reference of a term at
a given time t may include facts about how tokens of the same syntactic type are used in that linguistic
community at some after t. Although I shall focus just on Jackman’s theory, I think similar criticisms
would also discredit McCarthy’s.
268     

the first gold–platinum thought experiment, because we supposed in that

experiment that each community arrives at a single equilibrium for ‘gold’.
But Jackman concedes that a single linguistic community could split into
two communities, ‘‘with each sub-community developing accessible but
incompatible equilibria from the original community’s usage’’ ( Jackman
1999: 167–8). Let us now consider what his theory commits us to saying
about such a case.
Recall that in the second gold–platinum thought experiment (§8.3), a
single linguistic community splits into two isolated linguistic communities,
members of which arrive at different but equally entrenched beliefs about
the extension of their PJSS-based word type ‘gold’, while continuing to
take for granted their PIWs, PJSSs, and PJSEs for the PJSS-based word
type ‘gold’ used in the original community before the split. Jackman is
inclined to suppose that members of the two isolated communities in this
second gold–platinum thought experiment have developed ‘‘accessible but
incompatible equilibria from the original community’s usage’’ ( Jackman
1999: 167–8). Since the supposed equilibria in the two communities are
both valid, Jackman’s theory implies that in both linguistic communities
after the split the entrenched beliefs that members of the two communities
express by using sentences that contain the PJSS-based word type ‘gold’
are mostly true. As we saw above, however, we cannot say that prior
to the split, the word ‘gold’ had both the extensions that members of the
two later communities believe it to have. In such cases, I argued, we should
accept the PIWs, PJSSs, and PJSEs in both communities, and conclude that
members of the two communities disagree about the extension of a single
PJSS-based word type ‘gold’. But this description of the thought experiment
is not open to Jackman, given his supposition that each community has
reached a valid equilibrium for its tokens of g-o-l-d. His theory of
meaning-determining equilibria leads him to conclude that in the second
gold–platinum thought experiment and other similar cases, ‘‘the existence
of a rival community undermines each community’s entitlement to say what
their predecessors meant’’ ( Jackman 1999: 168). In particular, if the two
later communities in the second gold–platinum thought experiment reach
different but equally valid equilibria for their respective tokens of g-o-l-d,
as Jackman is inclined to suppose, then Jackman’s theory implies that
neither of the two communities is entitled to say what their predecessors
meant. Members of the two communities are therefore not entitled to
      269

suppose that they disagree about the extension of the PJSS-based word type
‘gold’ that was in use before the split.
In the second gold–platinum thought experiment there is a conflict
between our PIWs, PJSSs, and PJSEs, on the one hand, and Jackman’s
theory of how meaning is determined, on the other. Forced to choose
between our PIWs, PJSSs, and PJSEs and his theory of meaning, Jackman
chooses his theory of meaning.¹³ This shows that his theory of meaning
is incompatible with the context principle and the PJSS-based conception
of words. Yet Jackman rejects the token-and-ex-use conception of words
when he supposes that ex-uses at a given time can settle the extensions
of terms at an earlier time. The only way to justify such a retrospective
extension assignment, I submit, is to accept the context principle and PJSS-
based conception of words. Any other way of individuating words would
conflict with our own practical way of keeping track of words across time.
Hence Jackman’s position appears to be self-undermining. It both tacitly
presupposes the context principle and explicitly presupposes a theory of
meaning that violates the context principle by sometimes conflicting with,
and supposedly trumping, our PIWs, PJSSs, and PJSEs.

8.7. Counterfactuals about the Past and the Third

Gold–Platinum Thought Experiment
I argued above (§8.3) that if we adopt the context principle we will suppose
that in the second gold–platinum thought experiment, Locke’s utterance
of ‘There’s gold in those hills’ was already true or false in 1650. As we’ve
seen, this description of the second thought experiment conflicts with
Jackman’s theory of equilibria, according to which some ex-uses of terms
settle their meanings, in the metaphysical sense explained above. If we adopt

¹³ Henry Jackman suggested to me in a personal communication that he finds my second thought

experiment implausible, since it builds in the stipulation that members of the two later linguistic
communities will take themselves to disagree about gold when they discover that their PJSSs and PJSEs
across time for their tokens of g-o-l-d link both communities to the same tokens of g-o-l-d used in
1650, prior to their isolation from each other. He believes that even after a short discussion about their
differences, both sides would conclude that they mean something different by ‘gold’, and if they disagree
at all, it would be over which of the two means by ‘gold’ what was originally meant in 1650. To me,
however, this looks like a disguised expression of Jackman’s commitment to his theory of how meaning is
determined.
270     

the context principle, we must therefore resist the claim that the two later
communities in the second gold–platinum thought experiment each reach
a different meaning-settling equilibrium in the ex-use of their respective
terms. We must therefore distinguish between (1) the supposition that
members of the two communities described in the second gold–platinum
thought experiment express different entrenched beliefs by using their
respective tokens of g-o-l-d, on the one hand, and (2) the supposition that
these entrenched beliefs constitute two different meaning-settling equilibria
for those respective tokens of g-o-l-d, on the other. Jackman is tempted to
move from (1) to (2). But if we adopt the context principle, we commit
ourselves to resisting this temptation.
By adopting a theory of meaning that tempts us to move from (1) to
(2), we run the risk of conflating belief with truth. This is not to
say that Jackman’s theory implies there is no difference between belief
and truth, however. His theory leaves open the possibility that in any
given circumstance, a group of speakers can be wrong about wheth-
er they have arrived at a meaning-settling equilibrium for their words.
This leaves open the possibility that members of at least one of the
two linguistic communities in the second gold–platinum thought exper-
iment have not arrived at a meaning-settling equilibrium for their term
‘gold’, even if they think they have. In this way, one might think one
can reconcile Jackman’s theory of meaning-settling equilibria with the
supposition that there is a disagreement between the two linguistic com-
munities that they express by using a single unambiguous PJSS-based word
type that they trace back to the linguistic community from which they
both evolved.
There are two related problems with this attempted reconciliation.
First, to claim that only one of the two linguistic communities in the
second gold–platinum thought experiment has arrived at a meaning-
settling equilibrium for their PJSS-based word type ‘gold’ would be ad hoc.
For Jackman’s theory of meaning-settling equilibria is supposed to explain
why we think that in the first gold–platinum thought experiment, both
linguistic communities are right about what they respectively call ‘gold’:
the explanation is supposed to be that both linguistic communities have
arrived at a meaning-settling equilibrium for the PJSS-based word types
that they typically spell g-o-l-d. What difference is there between the
meaning-settling facts about ex-use in the first and second gold–platinum
      271

thought experiments? It seems that Jackman’s theory of meaning-settling

equilibria is unable to offer a principled answer to this question, and
it cannot be invoked in support of a claim that only one of the two
linguistic communities in the second gold–platinum thought experiment
has arrived at a meaning-settling equilibrium for their PJSS-based word
type ‘gold’.
A related problem is that to accept such a claim is in effect to concede
that our beliefs about whether a given pattern of ex-uses constitute a
meaning-settling equilibrium for one of our words will be no more secure
than our beliefs about the extensions of our terms. But, as we saw in §8.4,
our speculations about what determines extensions of our words are no
more secure than beliefs we express by using those words. Hence if we
adopt this suggestion, the claim that a range of ex-uses of tokens of a given
PJSS-based word type w is meaning-settling will be methodologically on a
par with the claim that the entrenched beliefs we express by using tokens
of w are true, hence equally open to revision. Why not then reject all
speculations about which ex-uses of a given term are meaning-settling, and
obey the context principle, as I recommend?
One tempting answer is that we cannot abandon such speculations,
because we need a theory of how ex-uses settle meanings and exten-
sions in order to describe certain counterfactual circumstances. Consider
again the counterfactual world in which the English-speaking community
ceases to exist in July 1650. One might suppose that we settle the
facts about the extensions of tokens of g-o-l-d in this world just by
stipulating that the ex-uses of tokens of g-o-l-d in that counterfactual
world are the same as the ex-uses of tokens of g-o-l-d in the actual
English-speaking community until July 1650, when the counterfactual
community ceases to exist. But if the extensions of those tokens are
settled by their ex-use, we reason, their extensions must be different
from the extensions of tokens of g-o-l-d in the actual English-speaking
community.
If we adopt the PJSS-based conception of words, however, we will
conclude that we cannot stipulate counterfactuals about extensions of
words by stipulating facts about the ex-uses of their tokens and asking
what extensions are determined by those facts. Instead, to make sense of
a counterfactual situation in which a group of speakers uses word tokens
of some PJSS-based word type, we must directly stipulate that the speakers
272     

use tokens of that PJSS-based word type in that counterfactual situation.¹⁴

To make sense of the counterfactual described in the previous paragraph,
for instance, we must stipulate that members of the counterfactual English-
speaking community in June 1650 use the same PJSS-based word type
‘gold’ that members of the actual English-speaking community use in June
1650. We can then specify the extension of their word ‘gold’ simply by
using it, hence without speculating about the relationship between ex-use
and extension.
This way of specifying of a counterfactual situation in which a group of
speakers uses word tokens of some PJSS-based word type is an application
of Kripke’s point that to make sense of trans-world identity, we need
(first) to identify what we wish to talk about, and (second) to say how
things might have been for that thing (Kripke 1980: 15–20). Starting with
this conception of trans-world identity, Kripke argues for a necessity of
origins thesis, according to which each thing has the properties of its origin
essentially. To find these properties, we must trace the thing back to the
earliest actual moment at which we could have identified it and said how
things might have been for it. PJSS-based word types don’t have origins
in the same sense in which physical objects do, but we obey something
like Kripke’s necessity of origins constraint when we stipulate how things
might have been for a particular PJSS-based word type if it is part of our
stipulation that the counterfactual world’s history overlaps with our world’s
history at all points prior to some given time. When we describe the world
in which our English-speaking community ceased to exist in July 1650,
for instance, we identify a PJSS-based word type ‘gold’ and say of it that
it might have been used by a community that was just like ours until July
1650.¹⁵
If the necessity of origins constraint applies to PJSS-based word types in
the way just described, however, and there are no further constraints on
how things might have been for particular PJSS word types, then we can

¹⁴ To allow for the possibility of such stipulations, some modification would be needed in my
Goodman-style explication of PJSS-based word types. We can perhaps still think of such word types as
sets of their tokens if we allow such sets to include possible as well as actual word tokens. An adequate
specification of such sets would require an extension of the context principle to encompass possible
PIWs and PJSSs, so long as these can be seen to bear the relation ‘is a PJSS-based replica of ’ to some
of our own actual PIWs and PJSSs.
¹⁵ John MacFarlane helped me to see that I need to be explicit about what kinds of constraints there
are on our descriptions of worlds in which speakers use some of our PJSS-based word types.
      273

stipulate that there is a counterfactual world w in which Locke exists, and

hence uses our PJSS-based word type ‘gold’, but future members of his
temporally extended linguistic community use tokens of g-o-l-d in just the
way that members of the linguistic community on Twin Earth in the first
gold–platinum thought experiment use their tokens of g-o-l-d.¹⁶ I shall
call this the third gold–platinum thought experiment.
One might be inclined to doubt that the world w just described is a
possible word. The trouble is that, by hypothesis, there is no non-semantic
difference between the ex-uses of tokens of g-o-l-d in world w and the ex-
uses of tokens of g-o-l-d on Twin Earth in the first gold–platinum thought
experiment. And we concluded that in the circumstances described in the
first gold–platinum thought experiment we have no grounds to criticize
or disagree with the verdict of scientists on Twin Earth that x satisfies their
PJSS-based word type ‘gold’ if and only if either x is a bit of the element
with atomic number 78 or x is a bit of the element with atomic number
79. But in the previous paragraph we stipulated that in world w members
of the English-speaking community use our PJSS-based word type ‘gold’.
Hence, if, as we suppose, it is not the case that x is gold if and only if
either x is a bit of the element with atomic number 78 or x is a bit of
the element with atomic number 79, then it follows from our description
of world w that members of the linguistic community on Earth in w are
wrong about gold. But how could our stipulation that in world w members
of the English-speaking community use our PJSS-based word type ‘gold’
make it the case that we are right and they are wrong about gold? Would
it perhaps be better to say that there is no possible world like w, hence that
our stipulations about how our PJSS-based word types might have been
used must satisfy further constraints, beyond the constraint suggested by
Kripke’s thesis of the necessity of origins?
These doubts about w rest on a mistake. It cannot be part of our
description of a counterfactual world such as w that we are actually right about
gold. Hence it does not follow from our description of w that members
of the linguistic community on Earth in w are wrong about gold. In this
respect, the relationship between the actual English-speaking community
today and the counterfactual English-speaking community today in the third

¹⁶ Henry Jackman pointed out to me that my view of PJSS-based word types apparently allows us to
describe this possible world, and raised the challenge I present in the next paragraph. It is my addition
to highlight the relationship of Jackman’s challenge to Kripke’s necessity of origins thesis.
274     

gold–platinum thought experiment is similar to the relationship between

the two later branches of our English-speaking community in the second
gold–platinum thought experiment, prior to their discovery that they
both use the same PJSS-based word type ‘gold’. Just as members of the
two later branches of our English-speaking community in the second
gold–platinum thought experiment use the same PJSS-based word type
‘gold’ to express their respective judgements about the make-up of gold,
so members of the actual English-speaking community today and members
of the counterfactual English-speaking community today in the third
gold–platinum thought experiment use the same PJSS-based word type
‘gold’ to express their respective judgements about the make-up of gold.
Naturally, we assume we are right about gold, but this does not follow
from our descriptions of either of these two possible worlds. What follows,
it seems, is just that the respective judgements made by using our PJSS-
based word type ‘gold’—namely, the judgement (made by members of
our actual English-speaking community today in the third gold–platinum
thought experiment, and by members of one of the two branches of
our English-speaking community in the second gold–platinum thought
experiment) that x is gold if and only if x is a bit of the element with atomic
number 79, and the judgement (made by members of the counterfactual
English-speaking community today in the third gold–platinum thought
experiment, and by members of the other branch of our English-speaking
community in the second gold–platinum thought experiment) that x is
gold if and only if x is either a bit of the element with atomic number 78
or a bit of the element with atomic number 79—cannot both be true.
But even this last statement of what follows is too strong. For just as we
left it open in the second gold–platinum thought experiment that members
of either of the two branches might later come to think some of their PJSSs
and PJSEs for their tokens of g-o-l-d were false, so we should leave it open
in the third gold–platinum thought experiment that members of either the
actual English-speaking community or the counterfactual English-speaking
community might come to think that some of their PJSSs and PJSEs for
their tokens of g-o-l-d were false. It is easy to allow for this possibility
in the counterfactual circumstances after 1650 in the third gold–platinum
thought experiment. All we need to do is to stipulate that members of
that counterfactual linguistic community make PIWs, PJSSs, and PJSEs
that together link their current utterances of sentences containing tokens of
      275

g-o-l-d back to their ancestors’ utterances of sentences containing tokens

of g-o-l-d in 1650, without also stipulating that those practical judgements
are true. It is less obvious how to accommodate the qualification in the
actual English-speaking community of the third gold–platinum thought
experiment. For if we come to think that some of our past PIWs, PJSSs,
and PJSEs for our tokens of g-o-l-d are false, we might conclude that Locke
did not actually use our current PJSS-based word type ‘gold’, despite our
initial assumption that he did. But if we were to come to this conclusion,
then due to the necessity of origins constraint discussed above, we could no
longer describe the counterfactual world as one in which Locke uses our
PJSS-based word type ‘gold’. Hence my original presentation of the third
gold–platinum thought experiment, in particular, the claim that we can
simply stipulate that w is a world in which Locke uses our PJSS-based word
type ‘gold’, needs to be qualified. We must now say that if Locke uses our
PJSS-based word type ‘gold’, as we actually suppose, then he might have
used that PJSS-based word type even in a world in which Earth ceased
to exist in 1650, or in a world in which the English-speaking community
develops after 1650 in exactly the same way that the linguistic community
on Twin Earth develops in my first gold–platinum thought experiment.
Let us understand the third gold–platinum thought experiment from now
on in this way.
Jackman’s theory of meaning-determining equilibria apparently commits
him to the conclusion that the tokens of g-o-l-d used by members of
the Twin English-speaking linguistic community on Twin Earth in the
first gold–platinum thought experiment and the tokens of g-o-l-d used
by members of the counterfactual English-speaking linguistic community
in the third gold–platinum thought experiment must be satisfied by all
and only the same objects, because the ex-uses of those tokens are by
hypothesis exactly the same. Hence his theory of meaning-determining
equilibria apparently commits him to rejecting the possibility that members
of the counterfactual English-speaking linguistic community in the third
gold–platinum thought experiment today use our PJSS-based word type
‘gold’, and that their judgements about gold disagree with ours. His theory
definitely commits him to rejecting the possibility that members of the
English-speaking community that ceases to exist in July 1650 use our PJSS-
based word type ‘gold’. By committing us to doubting and or rejecting
these possibilities on fully general theoretical grounds, Jackman’s theory of
276     

meaning-determining equilibria conflicts with the context principle and

PJSS-based conception of words.
As I emphasized above, however, Jackman rejects the token-and-ex-
use conception of words when he supposes that ex-uses at a given time
can settle the extensions of terms at earlier times. I’ve argued that the
only way to justify such a retrospective extension assignment is to accept
the context principle and PJSS-based conception of words. Any other
way of individuating words would conflict with our own practical way
of keeping track of words across time and with our strong sense that
statements expressed by using natural kind terms are true or false at the
time of utterance, even if it may take a great deal of further investigation
to discover their truth value. Hence we now have additional reasons
to conclude that Jackman’s position both tacitly presupposes the context
principle and explicitly presupposes a theory of meaning that violates the
context principle by sometimes conflicting with, and supposedly trumping,
our PIWs, PJSSs, and PJSEs.

8.8. Temporal Externalism and Relative Truth¹⁷

Let us now see if we can reconcile Jackman’s metaphysical speculations
about what determines the extensions of one’s terms with the context
principle and PJSS-based conception of words by relativizing our evaluations
of our utterances and beliefs about the extensions of our PJSS-based word
types to different contexts of assessment. To see how this might be motivated,
recall the problematic suggestion that only one of the two linguistic
communities in the second gold–platinum thought experiment has arrived
at a meaning-settling equilibrium for its PJSS-based word type ‘gold’. As I
noted above, by accepting this suggestion one might think one can reconcile
Jackman’s theory of meaning-settling equilibria with the supposition that
the two linguistic communities use a single unambiguous PJSS-based word
type ‘gold’ that they trace back to the linguistic community from which
they both evolved. But we noted that the suggestion faces two serious
problems. The first is that Jackman’s theory of meaning-settling equilibria

¹⁷ I am indebted to John MacFarlane for his advice about how to approach the main issues of this
section.
      277

is unable to offer a principled, independent reason for thinking that one

of the two linguistic communities in the second gold–platinum thought
experiment has arrived at a meaning-settling equilibrium for its PJSS-based
word type ‘gold’ and the other has not. The second is that the claim
that a range of ex-uses of tokens of a given PJSS-based word type is
meaning-settling will be methodologically on a par with the claim that
the entrenched beliefs we express by our corresponding transparent uses of
those tokens are true; both claims stand or fall together and are equally open
to revision. For this reason I suggested that we give up the metaphysical
speculations about what determines the extensions of our terms.
One might nevertheless find it repugnant to give up such metaphysical
speculations, and therefore seek some relief from the worry that they are ad
hoc. Let us focus first on the third gold–platinum thought experiment. It
would be odd just to put one’s foot down, and claim without independent
justification that the ex-uses of tokens of the PJSS-based word type ‘gold’
determine that x satisfies the PJSS-based word type ‘gold’ if and only if x is a
bit of the element with atomic number 79, and hence that, contrary to what
members of the counterfactual English-speaking community in the third
gold–platinum thought experiment believe, it is not the case that x satisfies
the PJSS-based word type ‘gold’ if and only if either x is a bit of the element
with atomic number 78 or x is a bit of the element with atomic number
79. For members of the counterfactual English-speaking community in the
third gold–platinum thought experiment are in an equally strong epistemic
position to claim that x satisfies the PJSS-based word type ‘gold’ if and only
if either x is a bit of the element with atomic number 78 or x is a bit of
the element with atomic number 79, and hence that, contrary to what we
(members of the English-speaking community) believe, it is not the case
that x satisfies the PJSS-based word type ‘gold’ if and only if x is a bit of
the element with atomic number 79.
Instead of claiming without independent justification that only one
of the two linguistic communities in the third gold–platinum thought
experiment is right and the other wrong, one might be tempted to think
that each of the two communities is right in its own way.¹⁸ John MacFarlane has

¹⁸ Mark Lance and John O’Leary-Hawthorne suggest that in their witch thought experiment,
‘‘either translation would be correct from the perspective in which they would be given’’ (Lance and
O’Leary-Hawthorne 1997: 51), thereby endorsing the idea that translation, hence also truth value and
extension, are in some way perspective relative. They are apparently committed to claiming that in all my
278     

recently developed a semantics according to which the truth values of some

utterances are defined only relative to contexts of assessment. Let us now see
if we can use MacFarlane’s semantics to make sense of the idea that in the
third gold–platinum thought experiment the two linguistic communities’
conflicting claims about the extension of their shared PJSS-based word type
‘gold’ are true relative to different contexts of assessment. We can then
ask if the same kind of approach can help us to make sense of the second
gold–platinum thought experiment.
To begin with, consider MacFarlane’s treatment of sentences about future
contingencies, such as ‘Whittington will die tomorrow from the gunshot
wounds that Cheney caused by shooting him today.’ Let us avoid the
temporal complications introduced by verbs such as ‘kill’—complications
that are important to Jackman’s defence of his view that on Earth in 1650
the extension of ‘gold’ was the set of all and only bits of the element with
atomic number 79, even though the extension of ‘gold’ was not yet settled
in 1650, but play no role in MacFarlane’s account of relative truth—and
focus instead on pure predictions. Consider, for instance, the sentence ‘It
will be sunny here tomorrow’, which I will write as
(1) Sunny (here, tomorrow)¹⁹
Suppose I now utter (1). Call this utterance of (1) u1 . How should I
evaluate u1 ? Let us assume that it is objectively indeterminate whether
it will be sunny here tomorrow. MacFarlane argues that to make sense
of this assumption, I must suppose that I am now actually in at least
two overlapping worlds that diverge tomorrow, and perhaps also diverge
sometime between now and tomorrow. To simplify, suppose I am now
actually in exactly two worlds that contain u1 —world, w1 , in which it
is sunny here tomorrow, and world w2 , in which it is not sunny here
tomorrow. If we evaluate u1 now, before w1 and w2 diverge, it seems
natural to conclude that u1 is neither true nor false, hence indeterminate.
But if we evaluate u1 tomorrow in w1 , we will conclude that u1 is true,

gold–platinum thought experiments, including the third one, members of each of the later communities
is right in its own perspective-relative way. As I shall explain in the text below, this vague idea can
be made precise by using MacFarlane’s semantics for relative truth. (For a brief description of the
main differences between my gold–platinum thought experiments and the Lance–O’Leary-Hawthorne
witch thought experiment, see Chapter 6, n. 12.)
¹⁹ This is (23) from MacFarlane 2008: 91, with the addition of the indexical ‘here’, which MacFarlane
leaves implicit.
      279

and if we evaluate u1 tomorrow in w2 , we will conclude that u1 is false.

More generally, as MacFarlane puts it, if we suppose there are branching
worlds in the sense just explained, it is natural to say both that ‘‘present
claims concerning the future can be shown to be untrue by a proof of
present unsettleness’’, and that ‘‘past claims concerning the present cannot
be shown to have been untrue by a proof of past unsettleness’’ (MacFarlane
2008: 90).
To highlight these observations about how we evaluate utterances of (1),
MacFarlane contrasts (1) with
(2) Settlednow : Sunny (here, tomorrow).²⁰
where
Settledt : s means that the truth value of
s is settled by how
things are at t. Suppose I now utter (2). Call this utterance of (2) u2 . How
should I evaluate u2 ? Since by assumption I am now actually in w1 and w2 ,
it is not now settled that it will be sunny here tomorrow. Hence unlike
u1 , which is now indeterminate in truth value, u2 is now false. And, since
u2 makes a claim about what is now settled, unlike u1 , u2 is false when
evaluated at all future times and contexts in either in w1 or w2 , the two
worlds that by hypothesis contain it.
MacFarlane explicates these intuitive evaluations by modifying and
extending a semantics in the style of Lewis 1980 and Kaplan 1989a that
relativizes truth to a point of evaluation—an ordered triple <C, w, a>, where
C is a context, understood as ‘‘a possible occasion on which a sentence
might be used’’, w is an index, which for now we can take to be a possible
world, and a is an assignment function from variables to objects in the domain
(MacFarlane 2008: 83). For instance, a semantic theory that relativizes
truth to points of evaluation might include the following two clauses:
(Q)
∀x is true at <C, w, a> if and only if  is true at every point
<C, w, a> such that a differs from a at most in the value it assigns
to x.
(N)
 is true at <C, w, a> if and only if  is true at every point
<C, w , a> such that w is accessible from w.
One way to relate the concept of the truth of a formula at a point
of evaluation to the concept of the truth of a closed sentence s (i.e.

²⁰ This is (24) from MacFarlane 2008: 91, with the addition of ‘here’.
280     

one that contains no free variables) as used in a particular context is as

follows:
(A) An occurrence of s in a context C is true if and only if s is true
at <C, wC , a>, where wC is the world of C and a is an arbitrary
assignment.²¹
In contrast, MacFarlane develops a framework for characterizing truth
relative to two contexts—a context of utterance and a context of assessment.
To capture our intuition that statements about future contingencies are
objectively indeterminate, he assumes a metaphysics of branching worlds,
as described above. This by itself requires some departure from (A), since
the definite description ‘the world of C’ in (A) fails to pick out any world
if there is more than one world that contains C, as in the example above,
in which a single occurrence (utterance) u1 of ‘Sunny (here, tomorrow)’ is
in both worlds w1 and w2 (where w1 = w2 ). In addition, as we have seen,
we find it natural to say that u1 is indeterminate as assessed now, before
w1 and w2 diverge, that u1 is true as assessed tomorrow in w1 , and that u1
is false as assessed tomorrow in w2 . MacFarlane explicates these and other
similar assessments of utterances by constructing a semantics that clarifies
what it means to say that an utterance is true-at-a-context-of assessment.
Suppose that C, C1 , and C2 are contexts. Let W(C) be the set of worlds
that overlap at C, and let W(C1 /C2 ) be W(C1 ), if W(C1 ) ∩ W(C2 ) = Ø,
and W(C1 ) ∩ W(C2 ), otherwise (MacFarlane 2008: 91). We can then
explain when an occurrence of a sentence at a context—the context
of utterance, here represented as CU —is true as assessed in a context of
assessment —here represented as CA —as follows (I shall call this ‘Assessment
Relative Truth’, hence the acronym ‘ART’):
(ART) An occurrence of a sentence s at CU is true as assessed at CA if
and only if for every w ∈ W(CU /CA ), s is true at <C U , w, a>.²²
To see how (ART) works, let us first apply it to utterance u1 dis-
cussed above. Let C1 be the context in which I uttered u1 , let C2 be the
context in which I assess u1 tomorrow in w1 , and let C3 be the context in
which I assess u1 tomorrow in w2 . To simplify, as before, assume that

²¹ Clauses (Q), (N), and (A) are, with minor notational changes, essentially the same as clauses (1),
(2), and (3), respectively, from MacFarlane 2008: 83.
²² This is clause (22) from MacFarlane 2008: 91.
      281

W(C1 ) = {w1 , w2 }. Then W(C1 /C1 ) = W(C1 ) = {w1 , w2 }, W(C1 /C2 ) =

W(C2 ) = {w1 }, and W(C1 /C3 ) = W(C3 ) = {w2 }. Applying (ART), we
find that
u1 is indeterminate as assessed at C1
u1 is true as assessed at C2
u1 is false as assessed at C3 .
Now let us apply (ART) to utterance u2 discussed above. To do this we
need an account of truth at a point of evaluation for sentences of the form

Settledt : s. McFarlane proposes the following:
(B)
Settledt : s is true at <C, w, a> if and only if s is true at all points
<C, w , a> such that w overlaps with w at t.²³
To apply (B) to the above-described utterance u2 of (2), namely, ‘Settlednow :
Sunny(here, tomorrow)’, we suppose that C is the context of utterance for
u2 . Hence, if w1 and w2 overlap at t = now, the occurrence of ‘Sunny(here,
tomorrow)’ that u2 contains is true at <C, w1 , a> and the occurrence of
‘Sunny(here, tomorrow)’ that u2 contains is not true at <C, w1 , a>, then
by (B) we may conclude that u2 is not true, i.e. false, at <C, w, a> for
every w. Applying (ART), then, we find that
u2 is false as assessed at C1 , C2 , and C3.
These results accord with our intuition that if we presuppose a metaphysics
of branching possible worlds, the truth value for u1 is not the same for all
contexts in which it is assessed, whereas the truth value of u2 is the same for
all contexts in which it is assessed. MacFarlane summarizes this observation
by saying that (1) is assessment-sensitive, whereas (2) is not (MacFarlane
2008: 92).
MacFarlane’s account of assessment-sensitivity can be seen as an explic-
ation of the idea that truth is relative, not absolute. Previous attempts to
explain relative truth can at best muster only very weak replies to the
objection that they conflate belief with truth. For instance, Jack Meiland
recommends that a relativist about truth reply to this objection by insisting
that, for instance, Jones’s belief that p is just a criterion for p’s being true-
for-Jones, not what p’s being true-for-Jones consists in (Meiland 1977: 580).

²³ This is clause (8) from McFarlane 2008: 84. As McFarlane notes, this clause yields a supervaluational
semantics of the sort presented in Thomason 1970.
282     

But the distinction between being a criterion for truth-for-Jones and being
what truth-for-Jones consists in is strained and legalistic; it is not much of
a defence against the charge that relativists about truth conflate truth with
belief. MacFarlane’s semantics promises to do better. And it clearly does do
better if we assume a metaphysics of branching worlds. As we saw when we
applied (ART) to an utterance of (1), if a single context of utterance is con-
tained within many worlds, MacFarlane’s notion of assessment-sensitivity
can be explicated without remainder in terms of contexts of utterance,
contexts of assessment, and truth at a point of evaluation, which is not itself
relativistic.
Let us now try to use MacFarlane’s method to make sense of the idea
that in the third gold–platinum thought experiment, the two linguistic
communities’ conflicting claims about the extension of their shared PJSS-
based word type ‘gold’ are true relative to different contexts of assessment.
To apply the method, let us add some detail to the third gold–platinum
thought experiment. Suppose, as we did in the first gold–platinum thought
experiment, that while pointing at some South African hills represented on
a map, Locke utters the sentence
(3) There is gold in those hills.
Call his utterance of (3) u3 . How should we evaluate u3 ? Let w3 be a
world we are now actually in, in which we judge that u3 is false, based
on our present account of the extension of our PJSS-based word type
‘gold’, and let w4 be the counterfactual world of the third gold–platinum
thought experiment—the world in which we now judge that u3 is false,
based on our present account in that world of the extension of our
PJSS-based word type ‘gold’. Suppose that w3 and w4 overlap at 1650,
hence both contain u3 . Let C4 be the context in which Locke uttered
u3 , let C5 be the context in which I evaluate u3 today in w3 , and let C6
be the context in which I evaluate u3 today in w4 . To simplify again,
assume that W(C4 ) = {w3 , w4 }. Then W(C4 /C4 ) = W(C4 ) = {w3 , w4 },
W(C4 /C5 ) = W(C5 ) = {w3 }, and W(C4 /C6 ) = W(C6 ) = {w4 }. Applying
(ART), we find that
u3 is indeterminate as assessed at C4
u3 is true as assessed at C5
u3 is false as assessed at C6 .
      283

A temporal externalist who is willing to apply (ART) can therefore make

sense of the idea that u3 is true as assessed by one of the linguistic
communities at C5 , and false as assessed by the other linguistic community
at C6 .
The resulting assessment-sensitivity of our evaluations of u3 contrasts
with our evaluations of utterances of sentences prefixed by ‘Settlednow :’
Thus suppose that while pointing at some South African hills represented
on a map, Locke utters the sentence
(4) Settlednow : there is gold in those hills.
Call this utterance of (4) u4 . By clause (B), u4 is false. Hence by (ART) it
is false for every context of utterance/context of evaluation pair. In short,
(3) is assessment-sensitive, whereas (4) is not.
Is a temporal externalist treatment of the third gold–platinum thought
experiment more plausible if it is combined with (ART) in the way just
sketched? We are still left with a hard question of why we should think
that if a community has entrenched, stable beliefs about the extension of
a PJSS-based word ‘gold’, then the community has reached a meaning-
determining equilibrium for that word. With (ART), however, one can
make the mystery symmetrical. If we apply (ART), we are not showing
bias or favouritism towards one of the two linguistic communities of the
third gold–platinum thought experiment over the other, since we get
different truth values—indeterminate, true, false, respectively—when we
assess the two communities’ evaluations of u3 from C4 , C5 , and C6 .
I think this is the best that can be said for combining temporal externalism
with (ART). And it comes at a cost—several new problems for temporal
externalism arise as a direct result of combining it with (ART). First,
when temporal externalism is combined with (ART), we lose our grip
on our initial intuition that members of the two linguistic communities
of the third gold–platinum thought experiment disagree about gold. The
problem is that by hypothesis each side to the dispute ascribes different
meanings and extensions to occurrences of the PJSS-based word type ‘gold’,
including the occurrences of that word in u3 . Each side takes the other to
mean what it means by ‘gold’, but the two sides do not agree about the
meaning or extension of ‘gold’, and that is what explains their different
assessments of u3 . They disagree about the extension of their common PJSS-
based word type ‘gold’, but their disagreement about its meaning prevents
284     

them from concluding that they disagree about gold itself. In contrast,
the different evaluations of u1 discussed above do not depend on different
assessments of the meanings of the words (‘Sunny’, ‘Here’, ‘Tomorrow’)
that occur in (1); and to the extent that the different evaluations of u1
reflect different assignments of extension to one of those words (‘Sunny’),
the differences in the assignments are due to differences in the worlds that
contain the different contexts from which u1 is assessed, not to differences
in the meaning-determining ex-uses of words, as in the temporal externalist’s
evaluations of u3 . In the application of (ART) to the temporal externalist’s
evaluations of u3 , if we have any grip at all on there being a common
subject matter about which both linguistic communities disagree, that grip
is secured by the existence of a single PJSS-based word type ‘gold’ that
members of both communities both take themselves to use, despite the
different meanings and extensions that they assign to it. (For this reason, I
shall not consider how one might apply (ART) to PJSSs themselves.) But
if we apply (ART) in the way just sketched, the existence of that shared
PJSS-based word type ‘gold’ does not bring with it either a shared meaning
or a shared extension for the word, and so there’s no longer any clear basis
for taking the two communities to disagree about gold.
We face a second problem if we try to use (ART) to explicate the
second thought experiment sketched above, in which two parts of a
given linguistic community become isolated from each other and develop
different entrenched beliefs about the extension of a single PJSS-based
word type ‘gold’. To explicate the different evaluations of u3 made by
members of the two linguistic communities in the second gold–platinum
thought experiment, one cannot appeal to branching possible worlds, since,
by hypothesis, both communities are in the same possible world. To apply
(ART) to the second gold–platinum thought experiment, then, we need
to find an appropriate interpretation of the variable in (ART) that ranges
over indices—the variable ‘w’. But I suspect that there is no way to do this
without conflating truth-at-a-context-of-assessment with belief. To back up
my suspicion, I would need to present and evaluate several interpretations
of ‘w’ in (ART). I don’t have the space here for such an investigation.
The basic reason for my suspicion, however, is easy to state: I don’t
see how the different assessments of u3 by members of the two linguistic
communities in the second gold–platinum thought experiment can be
explained by anything other than the different beliefs that members of those
      285

communities have about how to evaluate sentences that contain their PJSS
word type ‘gold’. I therefore find it difficult to see how a relativist about
truth who wishes to apply (ART) to u3 under the circumstances described
in the second gold–platinum thought experiment can avoid conflating
truth-at-a-context-of-assessment with belief.²⁴
If there’s no good answer to this second problem, as I suspect, then
to apply MacFarlane’s (ART) without conflating truth-at-a-context-of-
assessment with belief, we must presuppose a metaphysics of branching
worlds. If, in addition, we accept temporal externalism, according to
which the meaning and extension of a word depends on future ex-uses
of that word, we thereby commit ourselves to the conclusion that most
of our utterances are assessment-sensitive. There will be some exceptions,
including utterances of sentences prefixed by ‘Settlednow ’. But in general
we will not be able to avoid assessment-sensitivity even in sentences
such as (3) whose truth values we normally regard as determinate at
the time of utterance. In my view, this consequence is unacceptable,
since we easily avoid it by rejecting both absolute and relative forms
of temporal externalism and embracing the PJSS-based conception of
words.²⁵
I conclude that we cannot make temporal externalism more appealing
by combining it with a MacFarlane-style assessment-sensitive semantics. I
recommend that we eschew both absolute and relative forms of temporal
externalism and solve the puzzles raised by the gold–platinum thought
experiments by combining the Tarski–Quine thesis with the PJSS-based
conception of words.

²⁴ In Lasersohn 2005, Peter Lasersohn argues on empirical linguistic grounds that the truth value of
sentences containing what he calls predicates of personal taste, such as ‘fun’ or ‘tasty’, must be relativized
to individuals, whose judgements about what is fun or tasty differ. Lasersohn’s account of such relativity
is structurally parallel to MacFarlane’s (ART ). I am not fully persuaded by Lasersohn’s arguments, but
if they were successful, they would provide an interesting defence against the objection that Lasersohn’s
account conflates relativity of truth value with belief—the defence that English-speakers actually do
conflate relativity of truth value with belief, and hence any adequate theory of meaning must attribute
that conflation to them. Richard 2004 can be read in a similar way. My goal in this book is not to
provide a theory of meaning for English, however, but to find a truth predicate that enables us to clarify
and facilitate our inquiries. Given this goal, I see no advantage to combining temporal externalism with
(ART ) in a way that conflates truth-at-a-context-of-assessment with belief.
²⁵ In addition, as John MacFarlane pointed out to me, if most of our sentences are assessment-sensitive
for the reasons just sketched, then we cannot take for granted, as MacFarlane does in MacFarlane 2005:
338 n. 19, that the meta-language in which we state (ART ) is ‘‘devoid of assessment-sensitivity,’’ and
so cannot easily rule out the possibility that our attempts to state (ART ) are self-refuting.
286     

8.9. Immanent Realism

In the midst of our collaborative inquiries, we do not speculate about what,
if anything, determines the extensions of our words; instead, we take for
granted that our PIWs, PJSSs, and PJSEs are both factual and trustworthy,
and inquire directly into the questions that interest us. If we wish to try to
clarify and resolve a disagreement about whether there is gold in certain
specified hills in South Africa, for instance, we do not speculate about
what if anything determines the extensions of our term ‘gold’, but trust
our PIWs, PJSSs, and PJSEs for ‘gold’, and try to find out whether there is
gold in those hills in South Africa. We may come to agree on an answer,
and may even take it to be beyond doubt. But we do not and we need not
suppose that our answer metaphysically determines the extensions of our
terms. We treat any claim about what actually determines the extensions of
our terms as just another belief, more or less firmly accepted, and hence in
principle open to review and revision. After reviewing a given belief we
may conclude that it was false, taking for granted that our PIWs, PJSSs,
and PJSEs for the words we used to express the belief are trustworthy and
factual. As Quine emphasizes,
We should and do currently accept the firmest scientific conclusions as true, but
when one of these is dislodged by further research we do not say that it had been
true but became false. We say that to our surprise it was not true after all. Science
is seen as pursuing and discovering truth rather than as decreeing it. Such is the
idiom of realism, and it is integral to the semantics of the predicate ‘true’. (Quine
1995b: 67)

Just as our firm scientific conclusions may become dislodged by further

research, so any metaphysical speculations about what determines the
extensions of our terms may also be dislodged by further research. In
both thought experiments a crucial factor in our evaluation of our own
previous conclusions or speculations is our continuing practice of regarding
our PIWs, PJSSs, and PJSEs for the terms in question as both factual and
trustworthy.
As we saw in §§3.9–3.10, Quine himself fails to explicate ‘true’ in
a way that captures his insight about its semantics: his indeterminacy
thesis, which presupposes the token-and-ex-use model of words, implies
that our PIWs, PJSSs, and PJSEs are not factual. But if we reject the
      287

token-and-ex-use model of words, and combine the PJSS-based conception

of words with a Tarski-style disquotational truth predicate, we thereby
adopt a disquotational conception of truth that captures the realist semantics
of the predicate ‘true’. We are now finally in a position to see, in addition,
that the best way to make sense of the phenomenon of learning from
others and to solve the puzzles about sameness of extension across time
that I discussed in this and the previous two chapters is to adopt both the
Tarski–Quine thesis and the PJSS-based conception of words.
 9
Applications and Consequences

9.1. Introduction
As I have emphasized from the start of this book, my answers to the
pragmatic questions ‘‘Do we need a truth predicate?’’ and ‘‘If so, what sort
of truth predicate do we need?’’ are shaped by my assumption that we desire
to clarify and facilitate our inquiries by regimenting our sentences so that
we can express logical generalizations, such as ‘‘Every regimented sentence
of the form ‘S ∨ ∼S’ is true.’’ It seems that to express such generalizations
all we need is a Tarski-style disquotational truth predicate defined for
sentences of our own regimented language. If we suppose that there is no
more to truth than what is needed for expressing logical generalizations,
we may then be inclined to adopt the Tarski–Quine thesis that there is
no more to truth than what is captured by a Tarski-style disquotational truth
predicate defined for one’s own sentences. As normally understood, however, a
Tarski-style disquotational truth predicate tells us nothing about how to
apply logical generalizations to other speakers’ words or to our own words
as we used them in the past. But, as I argued in Chapter 3, we can clarify
and facilitate our inquiries only if we can express logical generalizations on
other speakers’ sentences and on our own sentences as we used them in
the past. Our desire to clarify and facilitate our inquiries should therefore
motivate us to adopt the intersubjectivity constraint: a Tarski-style disquotational
truth predicate defined for one’s own regimented sentences is satisfactory only if it is
supplemented by an account of why it is epistemically reasonable for one to regard
one’s practical judgements of sameness of satisfaction as both factual and trustworthy.
In Chapter 4 I argued (first) that if we presuppose the token-and-
ex-use conception of words, we will see no way to explain why it is
epistemically reasonable for a person to regard her practical judgements of
sameness of satisfaction as both factual and trustworthy without rejecting
    289

the Tarski–Quine thesis, but (second) that if we adopt the PJSS-based

conception of words, we can see that the Tarski–Quine thesis and the
intersubjectivity constraint are compatible. This may appear to leave it open
that we can satisfy the intersubjectivity constraint in a way that commits us to
rejecting the Tarski–Quine thesis. As I argued in Chapters 5–8, however,
this appearance is illusory: the only way to satisfy the intersubjectivity
constraint is to embrace both the Tarski–Quine thesis and the PJSS-based
conception of words.
My goal in this final chapter is to sketch several additional ways in
which adopting the PJSS-based conception of words transforms our under-
standing of topics that are central to the philosophy of language. I start
by explaining how the PJSS-based conception of words accommodates
the observations that motivate many philosophers to embrace the causal
theory of reference for predicates, and how one can regiment prop-
er names as descriptions without embracing a description theory that is
undermined by radical changes in belief. I recommend that we replace
the standard understanding of the division of linguistic labour with a
deflationary account that rests on our PIWs and PJSSs, and develop
notions of minimal competence and minimal self-knowledge that go
hand in hand with the PJSS-based conception of words. I also offer a
deflationary interpretation of the thought experiments that support anti-
individualism, and show that on the interpretation of anti-individualism
that I recommend, it does not posit or presuppose that facts about ex-use
determine norms for language use. Finally, I explain why our treatment
of logical laws as arbiters of disagreement is of a piece with our pursuit
of truth.

9.2. A Deflationary Alternative to the Causal Theory

of Reference: Predicates
If we adopt the PJSS-based conception of words, we thereby endorse
an alternative to the causal-historical theory of reference for predicates.
Recall (from §6.3) that Kripke and Putnam developed the causal-historical
theory of reference to explain and justify the PIWs, PJSSs, and PJSEs that
clash with methodological analyticity, the thesis that there are sentences
we cannot reject without changing the subject. The supposition that some
290   

sentences are methodologically analytic rests on the description theory

of reference for predicates, according to which an object x satisfies a
given predicate F as used by speaker A if and only if x satisfies certain
descriptive conditions that A associates with F. Kripke and Putnam used
historical examples and thought experiments to show that our PIWs, PJSSs,
and PJSEs sometimes conflict with methodological analyticity and the
description theory of reference for predicates. They tried to develop an
alternative theory of reference that explains and justifies our PIWs, PJSSs,
and PJSEs. The motivating idea behind their causal-historical theory of
extension is that there must be something about the ex-use of a term prior
to a discovery that settles its extension. But the gold–platinum thought
experiment of Chapter 6 shows that the causal-historical theory of the
extensions of our predicates does not fit with our PIWs, PJSSs, and PJSEs,
and hence cannot explain or justify them. I propose that instead of trying
to explain reference in terms of causal relations between words, speakers,
and things, we embrace the PJSS-based conception of words, and thereby
incorporate PIWs, PJSSs, and PJSEs into our understanding of words
themselves. Unlike the causal-historical theory, the PJSS-based conception
of words is constrained by the context principle and hence cannot fail to
fit with our PIWs, PJSSs, and PJSEs.
Kripke himself doubted that his remarks about reference could be
developed into a theory of necessary and sufficient conditions for a predicate
to be satisfied by a given object. He emphasized that ‘‘philosophical analyses
of some concept like reference, in completely different terms which make
no mention of reference, are very apt to fail’’ (Kripke 1980: 94). His central
goal was ‘‘to present a better picture without giving a set of necessary and
sufficient conditions for reference’’ (Kripke 1980: 94). We are now in a
position to see that any substantive specification of necessary and sufficient
conditions for reference will presuppose the token-and-ex-use conception
of words and hence violate the context principle. In place of Kripke’s vague
and general doubts about prospects for substantive theories of reference,
we now have a principled and systematic reason for abandoning the search
for such theories, given our desire to clarify and facilitate our inquiries
by formulating logical generalizations. At the same time, by adopting the
context principle we can respect the kinds of PIWs, PJSSs, and PJSEs that
Kripke highlighted in his persuasive arguments against description theories
of reference and methodological analyticity.
    291

Putnam also observed that our PIWs, PJSSs, and PJSEs conflict with
description theories of reference and methodological analyticity. Like Krip-
ke, Putnam tried to offer an alternative theory of reference that fits with our
PIWs, PJSSs, and PJSEs. But, as we have seen, the gold–platinum thought
experiment of Chapter 6 shows that Putnam’s causal-historical theory of
reference violates the context principle. Putnam’s Twin Earth thought
experiment, of which my gold–platinum thought experiment is a variation,
describes a Twin Earth on which speakers apply tokens of w-a-t-e-r
only to twin water, which has a molecular structure different from the
molecular structure of water. This feature of Putnam’s thought experiment
misleads him into adopting his contribution of the environment thesis, according
to which facts about the ex-uses of tokens of w-a-t-e-r on the two planets,
including facts about causal relations between those word tokens, speakers,
and things and kinds of stuff in the environment, determine the extensions
of the tokens. If this were correct, however, then his explanation of
reference should work for the gold–platinum thought experiment. But
it does not, for reasons I explained in detail in Chapter 6. Hence the
explanation fails even for Putnam’s own original thought experiment.
I recommend that we accept the PIWs, PJSSs, and PJSEs of the members
of the two communities in the gold–platinum thought experiment. For the
same reasons, I also recommend that we accept the PIWs, PJSSs, and PJSEs
of the members of the two communities in Putnam’s water-twin-water
thought experiment. In both cases we rely on the context principle and the
PJSS-based conception of words to extend our Tarski-style disquotational
understanding of satisfaction to encompass the predicates in question. The
context principle and the PJSS-based conception of words imply that
tokens of w-a-t-e-r have different extensions on Earth and on Twin Earth,
even in 1650, before analytical chemistry was developed in either planet,
despite the fact that, contrary to Putnam’s causal-historical theory, the facts
about the ex-uses of those tokens do not determine their extensions.

9.3. A Deflationary Alternative to the Causal Theory

of Reference: Proper Names
If we adopt the PJSS-based conception of words, we can also offer a
satisfying deflationary alternative to the causal theory of reference for proper
292   

names. Recall that, following Quine, I recommend that we regiment

proper names as definite descriptions. To do so is simply to settle the
logical form of the regimented sentences that are to replace ordinary
language sentences that contain proper names, not to offer substantive
necessary and sufficient conditions for being the referent of a name. Hence
the Quinean method of regimenting proper names does not amount
to a description theory of reference for proper names in the sense that
this is usually associated with Frege and Russell. Still, the method of
regimenting proper names cannot by itself accommodate the phenomena
that led Kripke and others to embrace the causal theory of reference for
proper names. To accommodate the phenomena, the Quinean method of
regimenting proper names must be joined with the PJSS-based conception
of words.
To see why, we must examine carefully the phenomena that many
have taken to undermine the description theory of reference for proper
names. Consider, for instance, Kripke’s celebrated Schmidt–Gödel thought
experiment:
Let’s suppose that someone says that Gödel is the man who discovered the
incompleteness of arithmetic . . . in the case of Gödel that’s practically the only
thing many people have heard about him—that he discovered the incompleteness
of arithmetic. Does it follow that whoever discovered the incompleteness of
arithmetic is the referent of ‘Gödel’? . . . Suppose that Gödel was not in fact the author
of the theorem. A man named ‘Schmidt’ whose body was found in Vienna under
mysterious circumstances many years ago, actually did the work in question. His
friend Gödel somehow got hold of the manuscript and it was thereafter attributed
to Gödel. On the view in question, then, when our ordinary man uses the
name Gödel, he really means to refer to Schmidt, because Schmidt is the unique
person satisfying the description, ‘the man who discovered the incompleteness of
arithmetic’. . . . So, since the man who discovered the incompleteness of arithmetic
is in fact Schmidt, we, when we talk about ‘Gödel’ are in fact always referring to
Schmidt. But it seems to me that we are not. (Kripke 1980: 83–4)

When we read the italicized sentence, we take Kripke to be inviting us

to suppose that Gödel was not in fact the author of the theorem. If we
regiment this sentence in the way Quine recommends, we take Kripke to
be saying that the x such that x = Gödel was not in fact the author of the
theorem. All along, though, we take for granted that an object x satisfies
‘= Gödel’ if and only if x = Gödel. To eliminate ‘Gödel’ as a singular term,
    293

we should regiment ‘= Gödel’ as an unstructured predicate, ‘Gödels’, and

accept that an object x satisfies ‘Gödels’ if and only if x Gödels. We should
do the same, also, for ‘Schmidt’. As a result of the regimentations, we will
introduced two new predicates (I shall call them name predicates), ‘Gödels’
and ‘Schmidts’, and suppose that no x both Gödels and Schmidts. When
we entertain the stipulations of Kripke’s story, we will also accept that the
person who Schmidts discovered the incompleteness of arithmetic. Hence we
reject the theory that the person who Gödels is the person who discovered
the incompleteness of arithmetic.
The crucial point is that we trust our PJSSs for Kripke’s words, and
accept his supposition, which is incompatible with the description theory.
Hence we must either reinterpret Kripke’s words, perhaps so that ‘Gödel’ is
translated by our word ‘Schmidt’, or agree with Kripke that the description
theory is incorrect. Many are likely to agree with Kripke.
According to Hartry Field, Kripke’s example tells us nothing substantive
about reference. Instead, it highlights the kinds of inferences we draw from
utterances that contain proper names. ‘‘What Kripke’s example . . . shows,’’
according to Field, ‘‘is that we should regard [the supposition that Kripke
expresses in the passage quoted above] as grounds for inferring ‘Gödel
did not [discover] the incompleteness theorem’ rather than as grounds
for inferring ‘Gödel was baptized as ‘Schmidt’ and never called himself
‘Gödel’ ’’ (Field 1994: 261). Field’s suggestion is problematic in several
ways. First, it is unclear why we should take the supposition that Kripke
expresses in the passage quoted above as grounds for inferring ‘Gödel did not
discover the incompleteness theorem.’ A committed description theorist
who believes that Gödel is the one who discovered the incompleteness
theorem may refuse to infer that ‘Gödel did not discover the incompleteness
theorem’, and insist, instead, either that Kripke’s word ‘Gödel’ should not
be translated as ‘Gödel’, or that Kripke’s supposition that Gödel did not
discover the incompleteness theorem is contradictory, and should therefore
be rejected. Field’s deflationary account of reference and truth is compatible
with a wide number of inferences; it does not imply that we should accept
the inference Field favours, and not the one that the description theorist
prefers. Hence Field’s suggestion that we regard Kripke’s observations
about reference as observations about inferential practice does not explain
why the description theorist’s response to Kripke’s thought experiment is
unacceptable.
294   

If we adopt the PJSS-based conception of words, however, we can

combine a deflationary account of satisfaction for the name predicates
‘Gödels’ and ‘Schmidts’ with our PIWs and PJSSs for the words of Kripke’s
thought experiment. We can resist the description theorist’s response by
pointing out that it amounts to a rejection of the context principle,
and with it of the PJSS-based conception of words. We have independ-
ent grounds for adopting the PJSS-based conception of words; the fact
that it fits with our judgements about Kripke’s thought experiment is
another good reason for adopting it. By supplementing Quine’s method
of paraphrasing proper names as descriptions with the PJSS-based con-
ception of words, we can simultaneously embrace the spirit of Kripke’s
thought experiments, hence reject robust description theories of prop-
er names, yet also reject Kripke’s causal theory of reference for proper
names.¹

9.4. What is Minimal Self-Knowledge?

If we reject the description theory of reference and embrace the PJSS-based
conception of words, as I recommend, then the beliefs one associates with
one’s words do not determine the extensions of one’s words. How then
can we make sense of minimal self-knowledge—the familiar fact that (in a
sense yet to be clarified) every speaker typically knows without empirical
investigation what thoughts his utterances express? To begin with, I propose
that we adopt the following constraint on the explication of the phrase
‘minimal self-knowledge’:
(C1) To take a speaker to use words of her language to express thoughts,
make claims, raise questions, and so on is also to take her to have
minimal self-knowledge of the thoughts she thereby expresses.
For instance, when Al says, ‘‘I have arthritis in my thigh,’’ in the situation
described in Chapter 5, we take Al to have said that he has arthritis in his
thigh. By (C1), to take Al to have said that he has arthritis in his thigh by
uttering the sentence ‘I have arthritis in my thigh’ is also to take Al to have

¹ Recall that if Quinean descriptions are prefixed by Burge’s @-operator (Burge 2005: 229–30),
they can also mirror the behaviour of rigid names in modal contexts. See Chapter 1, note 21.
    295

minimal self-knowledge of the thought that he thereby expresses—the

thought that he has arthritis in his thigh.
Unfortunately, however, (C1) conflicts with the tempting and deeply
entrenched assumption that
(C2) A speaker has minimal self-knowledge only if he has accurate
beliefs about the truth conditions of his utterances.
For instance, before Al talks to his doctor, he does not realize that his
utterance of ‘I have arthritis in my thigh’ could not be true, and so he does
not fully understand the truth conditions of his utterance. If we accept
(C2), we will conclude that Al does not know what thought his utterance
of ‘I have arthritis in my thigh’ expresses, and so he does not know without
empirical investigation what thought his utterance of ‘I have arthritis in my
thigh’ expresses. But when we conclude for this reason that Al does not
know what thought his utterance of ‘I have arthritis in my thigh’ expresses,
we still take Al’s utterance of ‘I have arthritis in my thigh’ to express the
thought that Al has arthritis in his thigh. And when we take his utterance
of ‘I have arthritis in my thigh’ in this way, according to (C1), we take Al
to know without empirical investigation what thought his utterance of ‘I
have arthritis in my thigh’ expresses. In this way, (C1) and (C2) conflict.
In the next few sections, I will explain how to resolve this conflict in
favour of (C1). I will first explain why so many philosophers are attracted
to (C2), and explore its consequences in more detail. I will then elucidate
linguistic competence and minimal self-knowledge in terms of our PJSSs
and explain how the resulting account of minimal self-knowledge differs
from accounts that entail (C2).

9.5. Minimal Self-Knowledge as Second Order

Most philosophers who accept (C2) find it so obvious that they are not
inclined to provide any explicit argument for it. To evaluate (C2), however,
one must first understand the source of its appeal. As I see it, (C2) is the
inevitable consequence of an attractive but flawed line of reasoning. The
reasoning begins with the harmless observation that
(1) A speaker has minimal self-knowledge only if he knows without
empirical investigation what thoughts his own utterances express.
296   

This apparently follows from the standard characterization of minimal

self-knowledge that I presented above. From (1) we can infer
(2) A speaker has minimal self-knowledge only if he knows what
thoughts his own utterances express.
It has long been standard in analytical philosophy to assume that
(3) The thought expressed by an utterance is individuated by the truth
conditions of the utterance.
This is best seen as a proposal for clarifying the ambiguous word ‘thought’
and, with it, minimal self-knowledge, which, given (3), can now be
described as the familiar fact that every speaker knows the truth conditions of
his utterances without investigation. Finally, it seems undeniable that
(4) A speaker knows the truth conditions of his utterances only if he has
accurate beliefs about the truth conditions of his utterances.
From (3) and (4) we can conclude
(5) A speaker knows what thoughts his utterances express only if he has
accurate beliefs about the truth conditions of his utterances.
And from (2) and (5) we can conclude
(C2) A speaker has minimal self-knowledge only if he has accurate
beliefs about the truth conditions of his utterances.
To accept this line of reasoning is to accept that minimal self-knowledge
is second order, in the sense that sentences that ascribe to a given speaker
A minimal self-knowledge of the thought expressed by A’s utterance of
sentence s have the form ‘‘A knows without empirical investigation that s
is true if and only if p,’’ where ‘p’ is replaced by a declarative sentence that
expresses the thought that A’s utterance of s expresses.
To avoid triviality, we cannot allow that to have minimal self-knowledge
it is enough to master the merely formal schema ‘‘s’ is true if and only if
s’. Donald Davidson seems dangerously close to trivializing self-knowledge
when he writes:
The speaker, after bending whatever knowledge and craft he can to the task of
saying what his words mean, cannot improve on the following sort of statement:
    297

‘‘My utterance of ‘Wagner died happy’ is true if and only if Wagner died happy.’’
(Davidson 2001a: 13)

For Davidson, however, even if each speaker’s epistemic access to this

biconditional is formal, her first-person use of it makes a substantive
second-order claim about the truth conditions of her sentence ‘Wagner
died happy’ that is guaranteed to be true by the principles that govern
interpretation. As we saw in Chapter 5, for Davidson all interpretation,
including our interpretation of a first-person use of this biconditional, is
governed by a principle of charity that guarantees that ‘‘nothing could count
as someone regularly misapplying her own words’’ (Davidson 2001b: 38).
The principle of charity requires that the interpretation of a speaker’s beliefs
about how to apply her words fit with the interpretation of her words in
such a way that she is not generally wrong about the truth conditions of
her sentences. Understood in this way, the speaker’s biconditional makes
a substantive claim about the truth conditions of her sentence. Moreover,
according to Davidson, substantive claims of this kind are in general
guaranteed to be true by the nature of interpretation, which is governed
by the principle of charity.
Ultimately the reason that we each have minimal self-knowledge, on
this view, is that interpretation is governed by the principle of charity. This
also explains why those who are attempting to interpret us do not know
without empirical investigation what the truth conditions of our utterances
are. Davidson’s second-order account of minimal self-knowledge therefore
accords with (C2), but only because it presupposes his principle of charity.

9.6. Basic Self-Knowledge and Containment

It is tempting to think that if we accept (C2), as many philosophers
do, then we can be sure without empirical investigation that a speaker’s
beliefs about the truth conditions of his sentences are accurate only if the
accuracy of these beliefs is guaranteed by an a priori principle that governs
interpretation, such as Davidson’s principle of charity. But Tyler Burge
has proposed an account of minimal self-knowledge that challenges this
reasoning. Like Davidson’s account of minimal self-knowledge, Burge’s
account is both second order and compatible with (C2). Unlike Davidson’s
298   

account, however, Burge’s account does not rely on a principle of charity.

Burge focuses on what he calls basic self-knowledge. An example is the
belief that I can express by using the sentence ‘I think (with this very
thought) that writing requires concentration.’ Such beliefs are second
order, in the sense that they are ‘‘about’’ first-order thoughts. My belief
that I think (with this very thought) that writing requires concentration is
‘‘about’’ the first-order thought that writing requires concentration, but it
is not metalinguistic. Burge’s explanation of why such beliefs are acceptable
is structural: ‘‘When one knows that one is thinking that p, one is not
taking one’s thought (or thinking) that p merely as an object. One is
thinking that p in the very event of thinking knowledgeably that one
is thinking it. It is thought and thought about in the same mental act’’
(Burge 1988: 654). In other words, ‘‘One knows one’s thought to be what
it is simply by thinking it while exercising second-order, self-ascriptive
powers’’ (Burge 1988: 656). This formulation makes it clear that according
to Burge, minimal self-knowledge is second order, but does not depend
for its justification on a principle of charity.
One problem with this account is that it presupposes that in general
speakers are in a position to think their first-order thoughts self-ascriptively.
It offers no explanation or justification of this presupposition. In the present
context, this is question-begging, since Burge applies his account of self-
knowledge to ‘‘explain’’ how speakers can know what they are thinking
even when they express their thoughts by using sentences whose truth
conditions they do not completely understand. For instance, according
to Burge, Al would be justified in accepting the belief he would express
by using the sentence ‘I think (with this very thought) that arthritis can
occur in one’s thigh.’ Hence Al knows that he is thinking that arthritis can
occur in one’s thigh. This routine application of Burge’s account of basic
self-knowledge violates Davidson’s principle of charity.
To reconcile his account with (C2), Burge might say that to have
accurate beliefs about the truth conditions of one’s utterances, in the sense
relevant to self-knowledge, is simply to be able to use one’s utterances to
express one’s second-order knowledge of the first-order thoughts (truth
conditions) that they express. In this sense, one can have an accurate
belief about the truth conditions of one’s utterances while still being quite
confused about the conditions under which they are true. For instance,
Al has accurate beliefs about the truth conditions of his sentence ‘Arthritis
    299

can occur in one’s thigh’ when he affirms his sentence ‘I think (with this
very thought) that arthritis can occur in one’s thigh’, which self-ascriptively
expresses the first-order thought expressed by the sentence ‘Arthritis can
occur in one’s thigh.’
According to Burge, all that is required for a speaker to have minimal
self-knowledge of the thought she expresses by using a given first-order
sentence s is that she has minimal linguistic competence in the use of
the second-order sentence ‘I am thinking (with this very thought) that s’,
where her minimal linguistic competence in the use of the second-order
sentence includes but does not augment her minimal linguistic competence
in the use of s. The standard way of thinking of minimal self-knowledge as
second-order, characterized by the argument (1)–(5) above, goes beyond
this formal requirement; it implies that minimal linguistic competence in
the use of a first-order sentence does not guarantee that one can use that
sentence to express one’s own thought or belief. According to this way
of thinking, a speaker who has the kind of linguistic competence in the
use of the first-order sentence s that is necessary for second-order minimal
self-knowledge of the thoughts and beliefs she expresses by using s also
has minimal linguistic competence in the use of s; but it is not in general
true that if a speaker has minimal linguistic competence in the use of a
different first-order sentence s, she thereby also has the kind of linguistic
competence in the use of s that is necessary for second-order minimal
self-knowledge of the thoughts and beliefs she expresses by using s. If she
does not have the kind of linguistic competence in the use of s that is
necessary for second-order minimal self-knowledge of the thoughts and
beliefs she expresses by using s, then she can ‘‘use’’ s only in an attenuated,
metalinguistic sense. Strictly speaking, she is not (transparently) using the
sentence to make claims or express her thoughts and beliefs, she is only
going through the motions of using it, and does not know what thoughts her
‘‘uses’’ of s express.
This reasoning suggests that when we ask ourselves whether we have
minimal self-knowledge of the thought or belief we express by using a
given sentence s, we are asking a substantive question that is not settled
by our minimal linguistic competence. We may take ourselves to be able
to use s to express our own thoughts and beliefs, when in fact we can at
best only describe our ‘‘use’’ of s metalinguistically, without knowing what
thoughts we express when we utter s.
300   

Several contemporary philosophers have described positions of this kind.

For instance, Gareth Evans agrees with Burge that linguistic interpretation
is not constrained by Davidson’s principle of charity, but insists, in effect,
that a speaker who has only minimal linguistic competence in the use
of certain words does not have minimal self-knowledge of the thoughts
that she expresses by using those words. Evans relies on the principle that
‘‘if a speaker uses a word with the manifest intention to participate in
such-and-such a practice, in which the word is used with such-and-such
semantic properties, then the word, as used by him, will possess just those
semantic properties’’ (Evans 1982: 387). Evans also proposes that we ‘‘think
of individuating the words of a language not only phonetically but also
by reference to the practices in which they are used. In these terms, the
requirement on a speaker using a proper name is not that he indicate which
object he intends to be (taken to be) referring to, but that he indicate which
name he intends to be (taken to be) using’’ (Evans 1982: 384). To this
extent, Evans’s approach to linguistic interpretation is compatible with our
PJSSs, and with Burge’s recommendations about how to interpret other
speakers. It therefore conflicts with Donald Davidson’s principle of charity,
for reasons I explained in Chapter 5.
According to Evans, however, further conditions must be met before a
speaker can be credited with entertaining the thoughts that her utterances
express. For any ordinary practice of using a proper name, there is a group of
speakers who have been introduced to the practice by their acquaintance with
the person to whom the name refers. For instance, there is a group of speak-
ers who have been introduced to the practice of using the name ‘Maurizio
Pollini’ by their acquaintance with Maurizio Pollini. Evans calls members of
this group the ‘‘producers’’ of the name-using practice (Evans 1982: 376). In
some name-using practices, there is also a group of speakers who participate
in the practice of using a particular name even though they are not acquain-
ted with the person to whom the name refers. There are many speakers
who participate in the practice of using the name ‘Maurizio Pollini’ even
though they are not acquainted with Maurizio Pollini. The speakers may be
introduced into the practice in a number of different ways. Perhaps some of
them are told that ‘‘Maurizio Pollini is the greatest living concert pianist’’,
for instance, while others simply hear on the radio, ‘‘That performance of
the Chopin Études, Opus 25, was by Maurizio Pollini.’’ Evans calls such
members of the name-using practice ‘‘consumers’’ (Evans 1982: 377).
    301

According to Evans, consumers of the name-using practice have no

role in connecting the name with the person to whom it refers. In
contrast, ‘‘It is the actual patterns of dealings the producers have had
with an individual—identified from time to time by the exercise of their
recognitional capacities in regard to that individual—which ties the name
to the individual’’ (Evans 1982: 382). In a similar way, he claims, there are
producers and consumers of practices of using natural-kind terms: ‘‘it is a
central feature of the practices associated with terms like ‘elm’, ‘diamond’,
‘leopard’, and like that there exist members—producers—who have . . . an
effective capacity to distinguish occasions when they are presented with
members of that kind, from occasions when they are presented with
members of any other kinds which are represented in any strength in the
environment they inhabit’’ (Evans 1982: 382).
On the basis of this distinction between producers and consumers of
practices of using proper names and natural-kind terms, Evans draws a
distinction between participating in a practice, and thereby ‘‘using’’ a given
name, on the one hand, and entertaining thoughts about the referent of the
name, on the other. He insists that ‘‘It is a perfectly intelligible possibility,
occasionally realized, that someone can use an expression to refer without
being himself in a position to understand the reference’’ (Evans 1982: 398).
One can participate in a practice of using a given name, such as ‘Maurizio
Pollini’, without being a producer of that practice, hence, according to
Evans, without being able to use that name to entertain thoughts about
Maurizio Pollini. According to this view, a speaker with only minimal
competence in the use of a proper name ‘N.N.’ does not understand what
is said by using that name in sentences, including sentences of the form
‘N.N. is F’. To say that he does not understand in this context, according
to Evans is to say that ‘‘he does not know the truth-conditions of, e.g.,
‘N.N. is F’ ’’ (Evans 1982: 403). Thus Evans believes that in order for the
speaker to understand what she says when she uses a sentence of the form
‘N.N. is F’, she must know the truth conditions of the thought that is
expressed in the practice by that sentence. Although Evans does not make
this explicit, he is clearly assuming that to know the truth conditions is to
have detailed and accurate beliefs about the truth conditions. But a speaker
who has only minimal linguistic competence in the use of ‘N.N.’ is merely
a consumer of the ‘N.N.’-using practice, and hence does not have detailed
and accurate beliefs about the truth conditions of thoughts she expresses by
302   

using that sentence. In this sense, such a speaker does not know the truth
conditions of those thoughts. In contrast, Evans believes, producers of the
‘N.N.’-using practice do have accurate beliefs about the truth conditions
of thoughts they express by using sentences in which the name occurs.
Hence producers of the ‘N.N.’-using practice are able to use the name
‘N.N.’ to refer to N.N., and consumers of the practice—those who have
only minimal competence in the use of ‘N.N.’—are not able to use the
name ‘N.N.’ to refer to N.N.
At the same time, however, Evans wants to grant that consumers of
the practice are in a position to ‘‘use’’ the name ‘N.N.’ But what can the
speaker ‘‘use’’ the name ‘N.N.’ for if not to refer to N.N.? Evans hints at
an answer when he notes that

There is . . . a powerful temptation to argue that, if someone is competent in the

use of a proper name (and hence able to function as a link in a chain by which
information about its referent can be transmitted), then his acquisition of that
competence must itself have put him in a position to entertain thoughts about its
referent. (Evans 1982: 403)

According to Evans, a person who counts as minimally competent in

the ‘‘use’’ of a word is not thereby able to use it transparently, in the
sense of ‘transparently’ explained in §4.2. Such a person may nevertheless
‘‘function as a link in a chain by which information about its referent can
be transmitted’’. But to function as a link in such a chain, the person must
‘‘use’’ the name in some sense, even if she does not entertain the thoughts
she expresses by using it in that way. Evans does not actually address the
question of how to describe the ‘‘use’’ of the name from the speaker’s
first-person perspective. She will probably take herself to be entertaining
the thoughts that she expresses by using it, even if, by Evans’s standards,
she is not actually able to do this. Such a speaker counts as ‘‘using’’ her
words in some sense, according to Evans, even though she does not use
them transparently.
Evans is driven to this conclusion by (C2), the assumption that minimal
self-knowledge requires second-order knowledge of the truth conditions
of the thoughts one’s utterances express, together with his observation
that ordinary linguistic interpretation is not governed by the principle
of charity. Hence Evans concludes that (C2) is incompatible with (C1).
Burge’s containment account of basic self-knowledge does not address
    303

this sort of position, and therefore does not show that (C1) and (C2) are
compatible.

9.7. Minimal Self-Knowledge as First Order

I propose that we reject (C2), and embrace (C1). In my view, to have
minimal self-knowledge of the thoughts one expresses by using one’s
words, it is enough to have minimal linguistic competence in the use of
those words. Moreover, in my view, to make PIWs and PJSSs for another
speaker’s words is to take her to be minimally competent in the use of those words.
Such competence requires more than simply writing or uttering sentences
in which the words occur. What more it requires can only be discovered
by examining our practices of taking speakers to be using words of English.
For instance, we know that we wouldn’t take a speaker to be using the
English word ‘apple’ competently if she applies it only to points of light
visible in the night sky. This does not mean that we English-speakers have
very strict requirements for competence in the use of the word ‘apple’.
A child who at first refuses to call a green apple an ‘apple’ might still be
taken to be able to use the word ‘apple’ to express beliefs and desires about
apples and to believe that the green apple is not an apple, provided that
she has some other beliefs about apples, including some true beliefs that
she expresses by using the sentence ‘That’s an apple.’ This does not imply
that to use a given word, we must make some true demonstrative claims
by using that word. Minimal competence in the use of some words can be
picked up very quickly, just on the basis of what the speaker was told, even
if what she was told is false. The requirements for minimal competence
in the use of words vary from word to word, and, in some cases, from
context to context. Given the context principle, we have no grip on these
requirements apart from our PIWs and PJSSs.
Even though there are no strict requirements for minimal competence
in the use of most words, to take someone to express a thought by using
a given word is also to take him to have some beliefs that he expresses
by using that word. Such beliefs do not determine the truth conditions of
the thoughts he uses that word to express, since the beliefs may be false or
misleading; instead, the beliefs are part of the speaker’s present understanding
of those truth conditions.
304   

Our judgements about minimal competence are intimately linked with

our judgements about when a speaker has minimal self-knowledge.² To
credit a speaker of a given natural language with minimal self-knowledge is
to take her to be able to use words of her own language to express thoughts,
make claims, raise questions, and so on. Any situation in which we are
willing to take another’s words at face value is thereby also one in which
we will credit her with having minimal self-knowledge.³ Hence, contrary
to (C2), a speaker may have minimal self-knowledge of the thoughts she
expresses by using her sentences even if her understanding of the truth
conditions of the sentences is partial, confused, or incorrect.
According to the account I recommend, minimal self-knowledge is
much more widespread than second-order knowledge of what one is
thinking. It is as widespread as the everyday use of language to express
thoughts, evaluate beliefs, raise questions, and so on, just as (C1) suggests.
The kind of self-knowledge embodied in these everyday uses of language is
not best viewed as a disposition to form justified second-order beliefs about
what one is thinking, either.⁴ To credit anyone with being able to form or
justify such second-order beliefs, we must presuppose that she already has
the kind of minimal self-knowledge that goes with linguistic competence.⁵
I conclude that minimal competence and minimal self-knowledge go
hand in hand—the same phenomena that lead us to say that Al believes
that he has arthritis in his thigh thereby also lead us to say that Al has
minimal self-knowledge of what thought his utterance of ‘I have arthritis
in my thigh’ expresses. To note that ordinary speakers have minimal
self-knowledge is not to make a substantive theoretical commitment that

² This does not imply that a speaker cannot use words that have the same spelling as words of a
public language so idiosyncratically that her words have meanings different from the meanings that
the identically spelled words have in the public language. Such uses would be judged incompetent as
uses of the identically spelled public-language words, and yet the idiosyncratic speaker may still express
thoughts by using her identically spelled words and have minimal self-knowledge of what thoughts
she expresses by using them. This happens much less frequently than most individualists believe,
however. And when it does happen, there are usually some words of the public language that she
uses competently, and that help other speakers to figure out what thoughts the speaker’s idiosyncratic
utterances express.
³ For a more thorough presentation of the points in this and the previous paragraph, see Ebbs 1996
and Ebbs 1997, §§100–23. For a parallel point about what it is to know the meanings of one’s own
words, see Putnam 1988: 32.
⁴ See McLaughlin and Tye 1998: 286–7 for a brief description (and apparent endorsement) of a
dispositional view of self-knowledge.
⁵ This is the kernel of truth behind Anthony Brueckner’s criticisms of Burge and Davidson in
Brueckner 1992, and the metalinguistic criticism of Burge sketched above.
    305

is independent of and stronger than our initial judgement that in the

circumstances described in Chapter 5, Al believes that he has arthritis in
his thigh; it is simply to acknowledge that in making PIWs and PJSSs for
another speaker’s words, we thereby also take her to know what she is
talking about in a minimal sense that goes with competence.
It is important to see that, according to my proposal, we do not take a
speaker to be minimally competent in the use of a word unless we take
her to be able to use the word, in the transparent sense of ‘‘use’’ that
contrasts with merely mentioning the word. On this account, there’s no
gap whatever between taking a speaker to be minimally competent in the
use of a given word, on the one hand, and taking her to have minimal
self-knowledge of the thoughts she expresses by using it, on the other.
Against this, one might think that to have minimal self-knowledge of
the thoughts one expresses by using certain words, one must know the
dictionary definition or conventional meaning of those words. A speaker
who has only minimal linguistic competence in the use of those words
might not know the dictionary definitions or conventional meanings of
those words and hence might have only a minimal understanding of the
thoughts she expresses by using those words.⁶ One might think that a
minimal understanding of the thoughts one expresses by using one’s words
is not enough for genuinely ‘‘entertaining’’ those thoughts—not enough
for having self-knowledge. This would be similar in some respects to
Evans’s position, yet apparently couched in terms of a first-order account
of self-knowledge.

⁶ Paul Horwich claims that

as long as [an] individual has acquired the word from the community and has a minimal understanding of
it, the communal language meaning may be correctly attributed to him—he means by it what everyone
else means. But he knows what it means only to the extent that whatever constitutes its meaning in his
idiolect resembles whatever constitutes its meaning in the communal language. (Horwich 1998b: 18)
Suppose provisionally that to make practical judgements of sameness of satisfaction for another speaker’s
words is also to take her words to have the same meanings as ours. Then the passage just quoted implies
that we take another speaker’s words to have the same meanings as our words if we take him to be
minimally competent in the use of those words. Horwich distinguishes between this claim and the claim
that a speaker knows the dictionary definitions or conventional meanings of those words. Horwich says
that ‘‘The degree to which an individual understands the word is constituted by the degree of similarity
between what it means in his idiolect and what it means in the communal language’’ (Horwich 1998b:
17–18). My account of minimal self-knowledge is not an account of knowledge of meaning. But
Horwich’s account might be understood in a way that conflicts with what I am saying about minimal
self-knowledge. For it might be thought that to have minimal self-knowledge of the thoughts one
expresses by using certain words, one needs more than minimal linguistic competence in the use of
those words. This is the objection I try to answer in the text.
306   

But if minimal linguistic competence is understood in the way that I

propose, to credit a speaker with minimal linguistic competence in the use
of certain words is simply to accept one’s PIWs and PJSSs for her utterances,
and thereby to take her to be able to use the words transparently. This is
then also sufficient for crediting her with minimal self-knowledge of the
thoughts her utterances of those words express. The fact that some speakers
know more about the consequences of their utterances and the conditions
under which they are true is irrelevant to the question whether speakers
who are minimally linguistically competent in the use of a given set of
words have minimal self-knowledge of which thoughts their utterances of
those words express. Even if other kinds of knowledge come in degrees,
minimal self-knowledge does not. One can define a more demanding kind
of self-knowledge than the kind of minimal self-knowledge that goes with
minimal linguistic competence. In the sense of ‘minimal self-knowledge’
that I aim to explicate, however, a speaker counts as having minimal self-
knowledge of the thoughts her utterances of a given set of words express if
she is minimally linguistically competent in the use of those words.
One might object that my account is inadequate because it does not
capture all of our intuitions about what it means to say that a given speaker
has minimal self-knowledge of what thought her utterances express. But this
objection rests on a misunderstanding. The phrase ‘minimal self-knowledge’
contains only ordinary words, yet there is no established ordinary use of it.
Although philosophers agree on a few core constraints on explications of the
phrase ‘minimal self-knowledge’, our intuitions about how to understand
it are not independent of our theoretical commitments, and hence do
not provide independent constraints on its proper explication. Many
philosophers are attracted to a substantive, second-order understanding of
minimal self-knowledge that is incompatible with (C1). I have proposed,
instead, a deflationary account of minimal self-knowledge that is compatible
with (C1). In my view, the deflationary account is the more fruitful one
because it fits and further illuminates the PJSS-based conception of words.

9.8. Minimal Self-Knowledge as Practical Knowledge

If we view minimal self-knowledge as first order in the way that I
recommend, we will see it as an aspect of our practical knowledge of how
    307

to use words, hence as a kind of knowledge how, not knowledge that, or

propositional knowledge. Jason Stanley and Timothy Williamson argue that
ascriptions of knowledge of how to do something are in fact ascriptions
of propositional knowledge, and take their conclusion to discredit any
view according to which minimal linguistic competence is a form of
non-propositional practical knowledge. Focusing on Michael Devitt’s view
that linguistic competence is a kind of knowledge how, they reason as
follows:
competence with the term t at least involves knowing how to use t with a
certain meaning, presumably the meaning that actually has. But if competence
with a term t involves knowing how to use t with the meaning it actually has,
the linguistic competence with the term does . . . yield propositional knowledge
about the meaning of that term. For, given a term t which has a certain meaning
m, x’s knowing how to use t with the meaning m amounts, for some contextually
relevant way w, to x’s knowing that w is a way for x to use t with the meaning m.
(Stanley and Williamson 2001: 444)

This reasoning is general and may therefore seem to apply to the account of
minimal competence and self-knowledge that I have proposed. To figure
out whether it does apply to my account, we must ask whether according
to my account there is an interpretation of ‘use’ under which
(∗ ) given a term t which has a certain meaning m, x’s knowing how to
use t with the meaning m amounts, for some contextually relevant
way w, to x’s knowing that w is a way for x to use t with the
meaning m
is true. The answer is ‘‘no.’’ To see why, recall that to mention a word,
such as ‘Boston’, one must use a name of that word. For instance, one
may use the word ‘ ‘Boston’ ’ to refer to the word ‘Boston’. In contrast,
I transparently use the word ‘Boston’ when I say, for instance, that Boston
is a populous city. But when I take a speaker to use the word ‘Boston’ in
this transparent way, I regard her as linguistically competent in the use of
‘Boston’. If we understand ‘‘use’’ as it occurs in (∗ ) in this way, it would
be legitimate to characterize a speaker’s practical knowledge of how to use
the word ‘Boston’ to refer to Boston, for instance, as her propositional
knowledge that using the word ‘Boston’ to refer to Boston is a way of using
the word ‘Boston’ to refer to Boston. But this characterization is circular: it
308   

presupposes what it is supposed to analyse—the knowledge of how to use

‘Boston’ transparently to refer to Boston. Hence the transparent sense of
‘use’ cannot be what Williamson and Stanley have in mind. The sense of
‘use’ that they have in mind must be ex-use —the kind of word use that can
be described independent of whether or not the individual is competent
to use a word transparently. Their arguments therefore presuppose the
doubtful thesis that a speaker’s practical knowledge of how to use a
language can be described exhaustively in terms of language ex-use.
Perhaps some accounts of minimal linguistic self-knowledge, such as
Devitt’s, presuppose this doubtful thesis, but mine does not. On the
contrary, my proposed elucidation of minimal self-knowledge in terms of
the PJSS-based conception of words presupposes the context principle,
according to which we have no understanding of the same-word relation
apart from our PIWs and PJSSs. The thesis that language use can be
described in terms of ex-use amounts to a rejection of the context principle
and the PJSS-based conception of words. If we adopt the context principle
and the PJSS-based conception of words, on the other hand, then to
attribute minimal self-knowledge to a speaker is not to attribute to her any
substantive metalinguistic knowledge of how to use her words. When an
English-speaker states that Boston is a populous city, for instance, she may
know that uttering the sentence ‘Boston is a populous city’ is a way of
saying that Boston is a populous city. But to attribute this disquotational
metalinguistic knowledge to her, we must make PIWs and PJSSs for her
words that presuppose that she can transparently use the sentence ‘Boston
is a populous city.’ We therefore cannot regard her knowledge of the
disquotational metalinguistic statement as a characterization or explanation
of her ability to use the words ‘Boston is a populous city’ transparently.
I conclude that both minimal linguistic competence and the minimal
self-knowledge that goes with it are forms of non-propositional practical
knowledge.

9.9. The Division of Epistemic Labour

The first-order, practical conception of minimal self-knowledge that I
described in the last two sections fits with (C1), the observation that to take
a speaker to use words of her language to express thoughts, make claims,
    309

raise questions, and so on is also to take her to have minimal self-knowledge

of the thoughts she thereby expresses. This observation is integral to our
understanding of the phenomenon of learning from others. As I argued in
Chapter 5, if we wish to save the phenomenon of learning from others, we
must adopt the PJSS-based conception of words. If we now add our new
explication of self-knowledge, we can also infer that to regard ourselves
or others as having learned that a particular sentence s that we previously
uttered is false we must see ourselves as having had minimal self-knowledge
of what thought we expressed by uttering s at some time before we learned
that it is false. Let us now consider how these observations bear on the
philosophical thesis that there is a division of linguistic labour.
Let S be a speaker whose transparent uses of a word w are only minimally
competent relative to what I will call the w-experts—those who know
most about the topic they identify by their transparent uses of w. According
to the standard account of the division of linguistic labour, it may be that
(a) S transparently uses word w at t.
(b) The extension of w as used by S at t is determined by facts about
how the w-experts in C ex-use w at or before t.
The thought experiments explored in Chapters 6–8 show that for many
words, among them the kinds of words that philosophers focus on when
they posit a division of linguistic labour, (b) is false. We must adopt the
PJSS-based conception of words and reject (b) if we want to preserve
and make sense of the ever-present possibility that even the most trusted
experts in our linguistic community are wrong about the extensions of our
terms. Hence there is no such thing as the division of linguistic labour, as
standardly understood. The hypothesis that there is a division of linguistic
labour is based in a misunderstanding of cases in which we trust or defer
to the judgement of another about some topic that we both identify by
transparently using the same word. Properly viewed, such cases illustrate
what I call the division of epistemic labour among speakers who express
thoughts about a given topic by transparently using what they take to be
tokens of a single PJSS-based word type.
The division of epistemic labour is more general than the phenomenon
of learning from others, since there is no guarantee that when we make
PIWs and PJSSs for another speaker’s words and trust what she tells us,
we are learning from her. The division of epistemic labour relies on trust,
310   

which doesn’t guarantee truth. Why then is it valuable to us? In the midst of
our inquiries we trust what others write or say. We are aware that no one’s
judgement is guaranteed to be true, so we also seek the benefit of criticism
and discussion from others who spend time investigating topics we have
spent a great deal of time investigating. This is a pragmatic, conditional
motivation for participating in the division of labour: given our interests in
inquiry, our limited time and resources, we want to participate in a practice
in which epistemic labour is shared.
Some speakers know more about some topics than others do. Unfortu-
nately, these are not always the speakers who were treated as epistemically
authoritative about those topics. Ideally, we would like our division of
epistemic labour to reflect differences in what speakers know. Whether or
not they reflect such differences, however, when we trust what others say,
we thereby regard them as authoritative. The phenomenon of deference of
one speaker’s usage to another that is typically associated with ‘‘the division
of linguistic labour’’ is best viewed not as a device that enables the reference
of the term to be ‘‘transmitted’’ from one speaker to another, but as a case
in which one speaker trusts another’s claims about a given topic, thereby
showing that she regards the other speaker as more knowledgeable than
she is about that topic.
According to my proposed account of self-knowledge, a speaker who
participates in the division of epistemic labour has minimal competence,
and thereby also minimal self-knowledge of the thoughts she expresses
by using sentences in which those terms occur. One might object to
this consequence of my account of the division of epistemic labour,
reasoning as follows: ‘‘You claim that your account of minimal self-
knowledge fits with our ordinary attributions of minimal self-knowledge,
and that these are elucidated by our judgements about when speakers
are minimally competent in the use of words of a shared language. Yet
many individuals who know very little about a given topic are able to
participate in the division of epistemic labour regarding that topic. They
are minimally competent, and hence, according to you, they have minimal
self-knowledge. The trouble is that in many of these cases it would be
appropriate to say, of such a speaker, when he uses the term that picks out
that topic, ‘Don’t listen to him—he doesn’t know what he’s talking about!’
In short, your account of minimal self-knowledge doesn’t fit with what we
are inclined to say about ignorant but minimally competent speakers.’’
    311

I agree that even if a speaker has minimal self-knowledge in the sense

I have described, it may be appropriate to say ‘‘Don’t listen to him—he
doesn’t know what he’s talking about!’’ But as I understand it, this is an
epistemic remark that could be paraphrased without loss with the words
‘‘Don’t trust what he says—he doesn’t know anything about that topic!’’⁷
If the speaker knows very little about cats, for instance, but she carries on
as if she knows a great deal about them, we might say ‘‘Don’t trust what
she says about cats—she doesn’t know anything about cats!’’ All of these
critical remarks take for granted that the speaker is competent in the use
of the relevant terms to make claims; the main thrust of the remark is that
we should not trust those claims. The objection goes further and suggests
that the remark conflicts with the assumption that she has minimal self-
knowledge. But a closer look at such cases dispels this appearance. Suppose
we confront the woman who knows very little about cats with a list of
her misunderstandings and errors, and we appeal to an authority she does
not doubt. Then she may well look back on her previous performances
and say something of the form, ‘‘I used to think cats are so and so, but
now I’m not so sure.’’ In other words, she takes responsibility for what
she thought previously, she does not in retrospect question whether she
even knew that she was entertaining thoughts about cats, or that she
believed that cats were so and so. In this sense she had minimal self-
knowledge even if it would have been appropriate to say ‘‘She doesn’t
know what she’s talking about.’’ Contrary to the objection, then, this
kind of remark typically expresses an epistemological criticism of a speaker’s
thoughts, one that takes for granted that she has minimal self-knowledge
of those thoughts, and hence can participate in the division of epistemic
labour.

9.10. Judging Minimal Linguistic Competence

across Time
My accounts of linguistic competence, minimal self-knowledge, and the
division of epistemic labour imply that we cannot accurately judge minimal

⁷ See Putnam 1975: 248 for a similar observation about the sense in which a person who thinks that
the Vietnam war was fought to help the South Vietnamese ‘‘doesn’t know what he is talking about’’.
312   

competence and minimal self-knowledge solely from our present epistemic

perspective. Consider, for instance, our current standards for regarding
someone as minimally competent in the use of the English word ‘Earth’.
If we encounter someone today who sincerely utters the sentence ‘Earth
is not round, but flat, and it is not a planet’, and this person appears to be
minimally competent in the use the words ‘flat’ and ‘planet’, we will likely
suspend our initial tendency to take the speaker to be minimally competent
in the use of our word ‘Earth’. In England many centuries ago, however,
the sentences ‘The Earth is round’, or ‘The Earth is a planet’ were widely
rejected. Yet our English word ‘Earth’ is in fact linked to those past uses
by chains of PIWs across time. Trusting these chains, we commit ourselves
to PIWs across time for tokens of E-a-r-t-h, as used by English-speakers
long ago. We thereby take these English-speakers to have been minimally
competent, relative to the standards of their time, in the use of our English
word ‘Earth’. Hence there is a conflict between what the historical chains
of judgements entail and what we are now inclined unreflectively to judge
about whether English-speakers long ago were competent in the English
word ‘Earth’.⁸
I propose that we trust our PIWs and PJSSs across time more than we trust
our current unreflective evaluations of the competence of speakers who
lived long ago. This fits with my earlier observation that we trust our PIWs
and PJSSs across time more than we trust our current speculations about
what counts as changing the topic and what counts as saying something
different about the same topic. Just as PIWs and PJSSs are central to our
grasp of the phenomenon of discovery, so they are central to our grasp of
minimal competence and minimal self-knowledge of speakers who lived
long ago. As the ‘Earth’ example shows, our current standards for minimal
competence and minimal self-knowledge may differ from the judgements
of minimal competence and minimal self-knowledge to which we are
committed when we trust our PIWs and PJSSs across time.
One might think that when PIWs and PJSSs across time conflict with
our current inclination to judge that speakers long ago were not competent
in the use of our words, it is just as reasonable to reject the chains of
PIWs, and the PJSSs they license, as it is to reject the inclination. There is
no neutral perspective from which to evaluate this concern. The force of

⁸ I owe this example to Erica Neely.

    313

any philosophical recommendation depends on one’s goals and interests.

Philosophers who do not care to save the phenomena of agreement and
disagreement between speakers, both at a time and across time, and of
sameness of satisfaction despite radical changes in theory and belief will
not be persuaded by the reasons I have offered for trusting our PIWs
and PJSSs across time more than our current inclination to judge that
speakers long ago were not competent in the use of our words. My
claim is conditional: if one wants to save the phenomena of agreement
and disagreement between speakers, both at a time and across time, and of
sameness of satisfaction despite radical changes in theory and belief, then one
should adopt the PJSS-based conception of words, and explicate minimal
competence and minimal self-knowledge in the way that I sketched
above.

9.11. Anti-Individualism, Externalism, and Linguistic

Communities
In Chapters 6 to 8 I used Twin-Earth thought experiments to show
that there is no way to satisfy the intersubjectivity constraint without
adopting the PJSS-based conception of words and the context principle.
I want now to explain how my conclusions bear on our understanding
of Hilary Putnam’s original Twin Earth thought experiments. Like the
thought experiments I presented in Chapters 6–8, Putnam’s Twin Earth
thought experiments (in Putnam 1975) highlight our PIWs, PJSSs, and
PJSEs. In the first of these we imagine that Oscar, an ordinary English-
speaker who is competent in the use of the English word ‘water’ but
does not accept (or reject) the sentence ‘Water is H2 O’, utters a sentence
containing the word ‘water’, for instance, the sentence ‘Water is a liquid
at room temperature.’⁹ Since Oscar is a competent English-speaker, other
English-speakers take his word ‘water’ to be the same as their word

⁹ It helps to imagine a context in which Oscar may actually say this. One possibility is that Oscar is
explaining to his son that ice is (solid) water, not just that water turns into ice when it freezes. In this
context, ‘Water is a liquid at room temperature’ may be the first of two sentences that Oscar utters,
the second one being ‘But ice is water, too—water that is at or below the freezing point.’ This is
compatible with our supposition that Oscar does not know that water is H2 O—Oscar may know that
ice is water that is at or below the freezing point even if he forgot, or never learned, that water is H2 O.
314   

‘water’, hence they take him to have said that water is a liquid at room
temperature. If, in addition, they think his utterance is sincere, they take him
to believe this.
In the next step of his thought experiment, Putnam stipulates that there
is a planet called Twin Earth which is just like Earth except that wherever
there is water on Earth there is twin water, a liquid with an underlying
chemical structure that is very different from the chemical structure of
water, on Twin Earth. He supposes that on Twin Earth there lives a
physical, phenomenological, and behavioural twin of Oscar, whom I’ll call
Twin Oscar. Twin Oscar is a normal speaker of Twin English, the Twin
Earth counterpart of English. When Twin Oscar utters the sentence ‘Water
is a liquid at room temperature’, his fellow Twin English-speakers take his
word ‘water’ to be the same as their word ‘water’, hence they take him to
have said (when translated into English) that twin water is a liquid at room
temperature. If, in addition, they think his utterance is sincere, they take him
to believe this.
Reconstructed in this way, Putnam’s first thought experiment establishes
that what a person believes and thinks does not supervene on the facts
about his linguistic dispositions, internal physical states, or phenomenal
experiences that can be described independently of his physical environment
and without using any sentences that express PIWs, PJSSs, or PJSEs for his
words. This thesis is what I call anti-individualism.¹⁰ We now have a more
general route to the same conclusion. For the context principle implies
that the semantic values of a person’s word tokens do not supervene on
facts about how he ex-uses them. Such facts include all facts about his
linguistic dispositions, internal physical states, or phenomenal experiences
that can be described independently of his physical environment and
without using any sentences that express PIWs, PJSSs, or PJSEs for his
words. Hence the context principle implies anti-individualism. If we adopt

¹⁰ According to the standard disciplinary history of anti-individualism, in Putnam (1975) Putnam

established that the references of a person’s words are not are settled by his linguistic dispositions,
internal physical states, or phenomenal experiences, described independently of his social and physical
environment. Burge (1979) is credited with making the corresponding case for beliefs and thoughts—the
case for the stronger thesis that I am calling anti-individualism. I have reconstructed Putnam’s reasoning
in a way that supports anti-individualism. Although I do not accept the standard interpretation of
Putnam’s reasoning in Putnam 1975, my goal here is not to present a historically accurate account of
what Putnam actually thought when he wrote Putnam 1975, but to highlight the methodology that in
my view explains what is persuasive about Putnam’s reasoning, whether or not he was clear about it.
    315

the context principle, as I recommend, we thereby also commit ourselves

to anti-individualism.
Putnam also claimed that even in 1750, before the scientists on Earth
or Twin Earth discovered the chemical properties of water and twin
water, respectively, a competent English-speaker who uttered the sentence
‘Water is a liquid at room temperature’ thereby expressed the thought
that water is a liquid at room temperature, while his Twin on Twin
Earth expressed the thought (translated into English) that twin water is
a liquid at room temperature. This anti-individualistic conclusion is also
a consequence of the context principle: it is because we trust our PIWs,
PJSSs, and PJSEs across time that we conclude that in both linguistic
communities, looking backward from today, the references of the words
for water and twin water did not change when the chemical properties of
the liquids to which they apply were discovered.¹¹
As we have seen, however, Putnam devised his causal-historical theory
of reference to explain and justify the PIWs, PJSSs, and PJSEs across time
that his thought experiments highlight. In doing so, he embraced external-
ism—the thesis that what a person believes and thinks supervenes on facts
about his linguistic dispositions, internal physical states, phenomenal exper-
iences together with all social or physical facts that one can describe without
using any sentences that express PIWs, PJSSs, or PJSEs for his words. The
facts relevant to externalism are all facts about language ex-use, as I defined
that notion. Hence externalism implies that what a person believes and
thinks supervenes on facts about language ex-use. In Chapters 6 to 8 I argued
that to satisfy the intersubjectivity constraint, and make sense of our PIWs,
PJSSs, or PJSEs, we must adopt the PJSS-based conception of words, which
is governed by the context principle. The context principle implies that
what a person believes and thinks does not supervene on facts about language
ex-use. Hence the context principle is incompatible with externalism.

¹¹ Many philosophers assume that in the early 1970s Putnam and Kripke accepted this aspect of the
Twin Earth thought experiment only because they believed they could explain it by constructing a
causal theory of reference. But both Putnam and Kripke were cautious about whether reference could
be given a non-circular explanation in causal terms. Moreover, no viable causal theory of reference
has yet been constructed, but the force of the thought experiment remains. In my view, the idea that
there is what Putnam called a ‘‘contribution of the environment’’ is rooted in our practice of making
and trusting PIWs, PJSSs, and PJSEs across time, and does not depend on the existence of a substantive
theory of reference that explains this practice. For more discussion of this point, but with different
terminology, see Ebbs 1997 and Ebbs 2000.
316   

Philosophers who embrace externalism sometimes distinguish between

two different ways in which facts about ex-use can determine semantic
values: (first) such facts can determine semantic values by linking word
tokens with objects and kinds, and (second) such facts can determine semantic
values by linking word tokens with ‘‘social norms’’ for language ex-use.
The causal-historical theory of reference applied to the English word ‘gold’
is usually taken to explain the first way. But there other words, such
as ‘arthritis’, that apparently do not group objects into natural kinds. An
externalist is therefore committed to coming up with some other account of
what settles the semantic value of such words as ‘arthritis’. One prominent
proposal (in Burge 1979 and 1986a) is that medical experts set the standards
for the ex-use of ‘arthritis’—the ‘‘social norms’’ for its ex-use—and thereby
settle its semantic value.¹²
All of these proposals presuppose that semantic values supervene on facts
about ex-use. But I have argued that semantic values do not supervene
on facts about ex-use. If my arguments succeed, then we should reject all
varieties of externalism, whether causal-historical or social, and embrace the
deflationary version of anti-individualism sketched above. When truth and
words are understood in the way that I recommend, our commitment to
anti-individualism is of a piece with our pursuit of truth, not independent
or additional to it.

9.12. Truth and Logic

The same holds for the laws of logic: when truth and words are understood
in the way that I recommend, our willingness to rely on the laws of logic
to facilitate and clarify our agreements and disagreements with others is of
a piece with our pursuit of truth, not independent or additional to it. This
consequence of my account of truth and words is crucial to my goal of
reconciling the Tarski–Quine thesis with the intersubjectivity constraint. I
motivated the intersubjectivity constraint by elaborating on Gottlob Frege’s
observation that we regard logical generalizations as arbiters of collaborative

¹² Goldberg 2007 presents and defends a similar view. For criticisms of inflationary versions of
anti-individualism like Burge’s and Goldberg’s, see Blackburn 1988 and Mercier 1993. I think the
problems that Blackburn and Mercier expose all stem from the fact that externalist views of meaning
and reference presuppose the token-and-ex-use model of words.
    317

inquiry. To explain why, I will first briefly review Frege’s explanation of

why the laws of logic are central to rational inquiry.
Consider the law of excluded middle, formulated as follows:
(L) Every sentence of the form ‘S ∨ ∼ S’ is true.¹³
In what sense is (L) a law? Frege distinguishes two senses of the word
‘law’: ‘‘in one sense a law asserts what is; in the other sense it prescribes
what ought to be’’ (Frege 1964: xv). Note first that (L) is a law in the
first sense, one that ‘‘asserts what is’’. Moreover, as Quine emphasizes, we
are interested in (L) not because of what it tells us about language, but
because of what it tells us about the world. We use a truth predicate to
express logical generalizations such as (L) not because these generalizations
are about language, but because we cannot actually write out all of the
sentences of a given logical form. Our affirmation of the law commits us
to affirming any and all of those sentences, and is therefore not primarily
an assertion about language, but about what is so. This does not distinguish
laws of logic from other laws. Any law, including laws of physics and
geometry, for instance, asserts what is so. Moreover, as Frege observes,
Any law asserting what is [so], can be conceived as prescribing that one ought to
think in conformity with it, and is thus in that sense a law of thought. This holds
for laws of geometry and physics no less than for laws of logic. (Frege 1964: xv)

Although laws that assert what is so needn’t and typically don’t also assert
that we ought to think in conformity with them, if we aim to judge in
accordance with what is so, we thereby also aim to judge in accordance
with all statements, including all laws, that assert what is so.
What distinguishes the laws of logic from laws of any other science is
that they hold for every subject: to judge in conformity with how things
are in any subject one must judge in conformity with the laws of logic. In
this sense, the laws of logic ‘‘are the most general laws, which prescribe
universally the way in which one ought to think if one is to think at all’’
(Frege 1964: xv). We must qualify this universal conception of logic if we
are to avoid the paradoxes that stem from unrestricted applications of a
truth predicate. But we can still say, in a Fregean spirit, that the laws of

¹³ Frege would not have formulated the law of excluded middle by using a truth predicate. (See
Chapter 2, n. 13, and Weiner 2005.) But his conception of logical laws as the most general laws can be
recast, with appropriate changes, for generalizations that we express by using a truth predicate.
318   

logic differ from other laws in their scope and application: they cut across
the subject matters of the special sciences. To reason about any subject at
all, one must reason in accordance with the laws of logic.
To express the laws of logic by using a Tarski-style disquotational
truth predicate, we need criteria for deciding, for any given logical form,
whether a given utterance is an utterance of a sentence of that form.
There are well-known syntactical rules for deciding whether one’s own
current utterances of a regimented sentence of one’s current language
is an utterance of a sentence of a given logical form. These syntactical
rules determine whether or not a sentence of our regimented language
results from admissible substitutions of regimented sentences or regimented
predicates for schematic sentence or term letters, respectively, in a given
logical schema. There are no syntactical rules for deciding whether another
person’s utterance or one of one’s own utterances in the past are utterances
of a sentence of a given logical form. But the PJSS-based word types
that we identify by syntactical criteria in our own current uses of our
regimented language are word types that others can use and that we may
have used in the past. By combining the syntactical criteria we can directly
apply to our own utterances with our PJSS-based conception of words, we
can determine whether another person’s utterance, or one of one’s own
utterances in the past, are utterances of a sentence of a given logical form.
When combined with the syntactical criteria, the PJSS-based conception
of words licenses us to apply logical generalizations such as (L) to other
speakers’ words and to our own words as we used them in the past
We may summarize these points as follows. According to the account of
truth and words that I recommend,
(1) We cannot commit ourselves to judging in accordance with the
truth without also committing ourselves to judging in accordance
with the laws of logic, conceived as generalizations that cut across all
subjects,
(2) The PJSS-based conception of words is integral to our own Tarski-
style disquotational definitions of truth, and
(3) The PJSS-based conception of words, together with the intrasub-
jective syntactical rules for deciding whether one’s own current
utterances of a regimented sentence of one’s current language is an
utterance of a sentence of a given logical form, enables us to express
    319

logical generalizations that we can apply both to other speakers’

words and to our own words as we used them in the past.
In short, according to the account of truth and words that I recommend,
our applications of logical generalizations to other speakers’ sentences and
to our own sentences as we used them in the past are of a piece with our
pursuit of truth.
References

Alward, P. 2005. ‘‘Between the Lines of Age: Reflections on the Metaphysics of

Words’’, Pacific Philosophical Quarterly 86: 172–87.
Austin, J. L. 1962. Sense and Sensibilia, ed. G. J. Warnock. Oxford: Oxford
University Press.
Barwise, J., and Cooper, R. 1981. ‘‘Generalized Quantifiers and Natural Lan-
guage’’, Linguistics and Philosophy 4: 159–219.
and Etchemendy, J. 2002. Language, Proof, and Logic. Stanford, Calif.: CSLI
Publications.
Bennett, J. 1988. ‘‘Quotation’’, Noûs 22/3: 399–418.
Berger, A. 2002. Terms and Truth: Reference Direct and Anaphoric. Cambridge, Mass.:
MIT Press.
Bezuidenhout, A., and Reimer, M., eds. 2004. Descriptions and Beyond. Oxford:
Oxford University Press.
Blackburn, T. 1988. ‘‘The Elusiveness of Reference’’, in P. A. French, T. E.
Uehling, Jr., and H. K. Wettstein, eds., Midwest Studies in Philosophy 12.
Minneapolis: University of Minnesota Press, 179–94.
Boghossian, P. 1989. ‘‘The Rule-Following Considerations’’, Mind 98: 507–50.
1990 ‘‘The Status of Content’’, The Philosophical Review 99: 157–84.
2000. ‘‘Knowledge of Logic’’, in P. Boghossian and C. Peacocke, eds., New
Essays on the A Priori. Oxford: Oxford University Press, 229–55.
2001. ‘‘How Are Objective Reasons Possible?’’, Philosophical Studies 106:
1–40.
Bonjour, L. 1998. In Defense of Pure Reason. Cambridge: Cambridge University
Press.
Boolos, G. 1975. ‘‘On Second-Order Logic’’, The Journal of Philosophy 73: 509–27.
1984. ‘‘To Be is to Be a Value of a Variable (or to Be Some Values of Some
Variables)’’, Journal of Philosophy 81/8: 430–49.
Burgess, J., and Jeffery, R. 2002. Computability and Logic, 4th edn. Cambridge:
Cambridge University Press.
Borg, E. 2004. Minimal Semantics. Oxford: Clarendon Press.
Brown, J. 1998. ‘‘Natural Kind Terms and Recognitional Capacities’’, Mind 107:
275–303.
2000. ‘‘Against Temporal Externalism’’, Analysis 60/2: 178–88.
  321

Brueckner, A. 1992. ‘‘Semantic Answers to Skepticism’’, Pacific Philosophical

Quarterly 73: 200–19.
Burge, T. 1979. ‘‘Individualism and the Mental’’, in P. A. French, T. E. Uehling,
and H. K. Wettstein, eds., Midwest Studies in Philosophy 4. Minneapolis: Uni-
versity of Minnesota Press, 73–122.
1986a. ‘‘Intellectual Norms and Foundations of Mind’’, The Journal of Philo-
sophy 83: 697–720.
1986b. ‘‘On Davidson’s ‘Saying That’ ’’, in E. Lepore, ed., Truth and Inter-
pretation: Perspectives on the Philosophy of Donald Davidson. Oxford: Blackwell.
1988. ‘‘Individualism and Self-Knowledge’’, The Journal of Philosophy, 649–63.
1993. ‘‘Content Preservation’’, The Philosophical Review 102: 457–88.
2005. Truth, Thought, Reason: Essays on Frege. Oxford: Clarendon Press.
Campbell, J. 2002. Reference and Consciousness. Oxford: Clarendon Press.
Cappelen, H. 1999. ‘‘Intentions in Words’’, Noûs 33/1: 92–102.
Carnap, R. 1942. Introduction to Semantics [Studies in Semantics, vol. i] Cambridge,
Mass.: Harvard University Press.
1956. Meaning and Necessity, 2nd edn. Chicago: University of Chicago Press.
1990. ‘‘Quine on Analyticity’’, in R. Creath, ed., Dear Carnap, Dear Van.
Berkeley and Los Angeles: University of California Press.
Cartwright, R. 1994. ‘‘Speaking of Everything’’, Noûs 28: 1–20.
Chalmers, D. 1996. The Conscious Mind. Oxford: Oxford University Press.
Chihara, C. 1979. ‘‘The Semanic Paradoxes: A Diagnostic Investigation’’, The
Philosophical Review 88/4: 590–618.
Coady, C. A. J. 1992. Testimony. Oxford: Clarendon Press.
Crosland, M. 1962. Historical Studies in the Language of Chemistry. London:
Heinemann Educational Books Ltd.
David, M. 1994. Correspondence and Disquotation. Oxford: Oxford University Press.
Davidson, D. 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press.
1986. ‘‘A Nice Derangement of Epitaphs’’, in E. LePore, ed., Truth and
Interpretation (Oxford: Blackwell).
1990. ‘‘The Structure and Content of Truth’’, Journal of Philosophy 87/6:
279–328.
1992. ‘‘The Second Person’’, in P. A. French, T. E. Uehling, Jr., and H. K.
Wettstein, eds., Midwest Studies in Philosophy, 18. Notre Dame: University of
Notre Dame Press, 255–67.
1994a: ‘‘What is Quine’s View of Truth?’’, in D. Davidson, Truth, Language,
and History, Oxford: Clarendon Press, 2005, pp. 81–5. First published in Inquiry,
37 (1994), 437–40.
322 

Davidson, D. 1994b. ‘‘Radical Interpretation Interpreted’’, Philosophical Perspectives

8: 121–8.
1996. ‘‘The Folly of Trying to Define Truth’’, in D. Davidson, Truth,
Language, and History. Oxford: Clarendon Press, 19–37.
2001a. ‘‘First Person Authority’’, in Subjective, Intersubjective, Objective. Oxford:
Oxford University Press, 3–14.
2001b. ‘‘Knowing One’s Own Mind’’, in Subjective, Intersubjective, Objective.
Oxford: Oxford University Press, 15–38.
Davies, M. 1981. Meaning, Quantification, Necessity: Themes in Philosophical Logic.
London: Routledge and Kegan Paul.
Dennett, D. 1987. ‘‘Evolution, Error, and Intentionality’’, in D. Dennett, The
Intentional Stance. Cambridge, Mass.: MIT Press, 287–321.
Devitt, M., and Sterelny, K. 1987. Language and Reality. Cambridge, Mass.: MIT
Press.
Dilworth, J. 2003. ‘‘A Refutation of Goodman’s Type-Token Theory of Nota-
tion’’, Dialectica 57/3: 330–6.
Donnellan, K. 1983. ‘‘Kripke and Putnam on Natural Kind Terms’’, in C. Ginet
and S. Shoemaker, ed., Knowledge and Mind. Oxford: Oxford University Press,
84–104.
Dummett, M. 1978a. ‘‘The Justification of Deduction’’, in M. Dummett, Truth
and Other Enigmas. Cambridge: Mass.: Harvard University Press, 290–318.
1978b. ‘‘The Social Character of Meaning’’, in Truth and Other Enigmas.
Cambridge, Mass.: Harvard University Press, 420–40.
1978c. ‘‘The Significance of Quine’s Indeterminacy Thesis’’, in Truth and
Other Enigmas. Cambridge, Mass.: Harvard University Press, 375–419.
1981. Frege: Philosophy of Language, 2nd edn. Cambridge, Mass.: Harvard
University Press.
1991. The Logical Basis of Metaphysics. Cambridge, Mass.: Harvard University
Press.
1993a: ‘‘What is a Theory of Meaning? (II)’’ repr. in The Seas of Language.
Oxford: Oxford University Press, 34–93.
1993b. ‘‘What do I Know when I Know a Language?’’ repr. in The Seas of
Language. Oxford: Oxford University Press, 94–105.
Ebbs, G. 1996. ‘‘Can We Take Our Words at Face Value?’’, Philosophy and
Phenomenological Research 56/3: 499–530.
1997. Rule-Following and Realism. Cambridge, Mass.: Harvard University
Press.
2000. ‘‘The Very Idea of Sameness of Extension across Time’’, American
Philosophical Quarterly 37/3: 245–68.
  323

2001. ‘‘Vagueness, Sharp Boundaries, and Supervenience Conditions’’,

Synthese 127: 303–23.
2002a. ‘‘Learning from Others’’, Noûs 36/4: 525–49.
2002b. ‘‘Truth and Trans-Theoretical Terms’’, in J. Conant and U. Zeglen,
eds., Hilary Putnam: Pragmatism and Realism. London: Routledge, 167–85.
2003. ‘‘Denotation and Discovery’’, in F. Schmitt, ed., Socializing Metaphysics:
The Nature of Social Reality. Totowa, NJ: Rowman and Littlefield, 247–68.
Eklund, M. 1996. ‘‘On How Logic Became First-Order’’, Nordic Journal of Philo-
sophical Logic, 1/2: 147–67.
Evans, G. 1982. Varieties of Reference. Oxford: Clarendon Press.
1985. ‘‘Pronouns, Quantifiers, and Relative Clauses (I)’’, repr. in G. Evans,
Collected Papers. Oxford: Clarendon Press.
and McDowell, J., eds. 1976. Truth and Meaning. Oxford: Clarendon Press.
Evnine, S. 1991. Donald Davidson. Stanford, Calif.: Stanford University Press, 1991.
Field, H. 1973. ‘‘Theory Change and Indeterminacy of Reference’’, The Journal of
Philosophy 70: 462–81.
1974. ‘‘Quine and the Correspondence Theory’’, The Philosophical Review 83:
200–28.
1992. ‘‘Critical Notice: Paul Horwich’s Truth’’, Philosophy of Science, 59:
321–30.
1994. ‘‘Deflationist Views of Meaning and Content’’. Mind 103: 249–84.
Fine, K. 1975. ‘‘Vagueness, Truth and Logic’’, Synthese 30: 265–300.
Fodor, J. 1983. Modularity of Mind. Cambridge, Mass.: MIT Press.
Forbes, G. 1986. ‘‘Truth, Correspondence and Redundancy’’, in G. Macdonald
and C. Wright (eds.), Fact, Science and Morality: Essays on A. J. Ayer’s Language,
Truth and Logic. Oxford: Blackwell, 27–54.
Frege, G. 1964. Introduction to The Basic Laws of Arithmetic, trans. and ed. M. Furth.
Los Angeles: University of California Press, 1964. First published in German in
1893.
1978. The Foundations of Arithmetic, 2nd rev. edn. German text with English
translation by J. L. Austin. Oxford: Blackwell. First published in German in
1884.
Fricker, E. 1995. ‘‘Telling and Trusting: Reductionism and Anti-reductionism in
the Epistemology of Testimony’’, Mind 104: 393–411.
Geach, P. T. 1962. Reference and Generality. Ithaca, NY: Cornell University
Press.
Glanzberg, M. 2005. ‘‘Minimalism, Deflationism, and Paradoxes’’, in J. C. Beall
and B. Armour-Garb, eds., Deflationism and Paradox. Oxford: Oxford University
Press, 107–32.
324 

Glanzberg, M. 2006. ‘‘Quantifiers’’, in E. Lepore and B. Smith, eds., The Oxford

Handbook of Philosophy of Language. Oxford: Clarendon Press, 794–821.
Goldberg, S. 2007. Anti-Individualism. Cambridge: Cambridge University Press.
Goldman, A. 1970. A Theory of Human Action. Princeton: Princeton University
Press.
Goodman, N. 1966. Structure of Appearance, 2nd edn. New York: Bobbs-Merrill.
1972. ‘‘Seven Strictures on Similarity’’, in N. Goodman, Problems and Projects.
Indianapolis: Bobbs-Merrill Company, Inc., 437–46.
1976. Languages of Art. Indianapolis: Hackett.
1983. Fact, Fiction, and Forecast, 4th edn. Cambridge, Mass.: Harvard Univer-
sity Press.
Grandy, R. 1973. ‘‘Reference, Meaning, and Belief ’’, The Journal of Philosophy 70:
439–52.
Grice, H. P. 1957. ‘‘Meaning’’, The Philosophical Review 66: 377–88.
Hacking, I. 1985. ‘‘Rules, Scepticism, Proof, and Wittgenstein’’, in I. Hacking,
ed., Exercises in Analysis. Cambridge: Cambridge University Press, 113–24.
Halbach, V. 1999. ‘‘Disquotation and Infinite Conjunction’’, Mind 108/429: 1–22.
Hansford, S. H. 1968. Chinese Carved Jades. London: Faber and Faber.
Heck, R. G., Jr., 2004. ‘‘Truth and Disquotation’’, Synthese 142: 317–52.
Heim, I., and Kratzer, A. 1998. Semantics in Generative Grammar. Oxford: Blackwell.
Hilbert, D., and Ackerman, W. 1950. Principles of Mathematical Logic, 2nd edn.
New York: Chelsea Publishing Co.
and Bernays, P. 1939. Grundlagen der Mathematick, vol. ii. Berlin: Springer
Verlag.
and Cohn-Vossen, S. 1956. Geometry and Imagination, trans. P. Nemenyi.
New York: Chelsea Publishing Co.
Hill, C. 2002. Thought and World: An Austere Portrayal of Truth, Reference, and
Semantic Correspondence. Cambridge: Cambridge University Press.
Horwich, P. 1995. ‘‘Meaning, Use, and Truth’’, Mind 104: 355–68.
1998a: Truth, 2nd edn. Oxford: Clarendon Press.
1998b: Meaning. Oxford: Clarendon Press.
Hugly, P., and Sayward, C. 1981. ‘‘Expressions and Tokens’’, Analysis 4/1: 181–7.
Jackman, H. 1999. ‘‘We Live Forward But Understand Backwards: Linguistic
Practices and Future Behavior’’, Pacific Philosophical Quarterly 80: 157–77.
Jackson, F. 1998a. From Metaphysics to Ethics: A Defense of Conceptual Analysis.
Oxford: Oxford University Press.
1998b. ‘‘Reference and Description Revisited’’, Noûs-Supplement: Philosoph-
ical Perspectives 12: 201–18.
  325

Kaplan, D. 1989a: ‘‘Demonstratives: An Essay on the Semantics, Logic, Metaphys-

ics, and Epistemology of Demonstratives and Other Indexicals’’, in J. Almog,
J. Perry, and H. Wettstein, eds., Themes from Kaplan. Oxford: Oxford University
Press, 481–566.
1989b. ‘‘Afterthoughts’’, in J. Almog, J. Perry, and H. Wettstein, eds., Themes
from Kaplan. Oxford: Oxford University Press, 565–614.
1990. ‘‘Words’’, Proceedings of the Aristotelian Society, suppl. vol. 64: 93–119.
Katz, J. 1998. Realistic Rationalism. Cambridge, Mass.: MIT Press.
Kitcher, P. 1982. ‘‘Genes’’, British Journal for the Philosophy of Science 33: 337–59.
1993. The Advancement of Science. Oxford: Oxford University Press.
Kleene, S. C. 1953. Introduction to Metamathematics. Amsterdam and New York:
Van Nostrand Reinhold.
Kraut, R. 1993. ‘‘Robust Deflationism’’, The Philosophical Review 102: 247–63.
Kripke, S. 1975. ‘‘Outline of a Theory of Truth’’, Journal of Philosophy 72:
690–716.
1976. ‘‘Is There a Problem about Substitutional Quantification?’’, in G. Evans
and J. McDowell, eds., Truth and Meaning: Essays in Semantics. Oxford: Claren-
don Press, 325–419.
1979. ‘‘A Puzzle about Belief ’’, in A. Margalit, ed., Meaning and Use.
Dordrecht: D. Reidel, 239–83.
1980. Naming and Necessity. Cambridge, Mass.: Harvard University Press.
1982. Wittgenstein on Rules and Private Language. Oxford: Blackwell.
Künne, W. 2003. Conceptions of Truth. Oxford: Oxford University Press.
Lance, M., and O’Leary-Hawthorne, J. 1997. The Grammar of Meaning: Normativity
and Semantic Discourse. Cambridge: Cambridge University Press.
Larson, R., and Segal, G. 1995. Knowledge of Meaning. Cambridge, Mass.: MIT
Press.
Lasersohn, P. 2005. ‘‘Context Dependence, Disagreement, and Predicates of
Personal Taste’’, Linguistics and Philosophy 28: 643–86.
Leeds, S. 1978. ‘‘Theories of Reference and Truth’’, Erkenntnis 13: 111–29.
Lepore, E., and Ludwig, K. 2005. Davidson: Truth, Language, and Reality. Oxford:
Oxford University Press.
2007. Donald Davidson’s Truth-Theoretic Semantics. Oxford: Clarendon
Press, 2007.
Lewis, D. 1968. ‘‘Counterpart Theory and Quantified Modal Logic’’, Journal of
Philosophy 65: 113–26.
1969. Convention. Cambridge, Mass.: Harvard University Press.
1974. ‘‘Radical Interpretation’’, Synthese 23: 331–44.
326 

Lewis, D. 1980. ‘‘Index, Context, and Content’’, in S. Klanger and S. Öhman,

eds., Philosophy and Grammar. Dordrecht: Reidel.
Loar, B. 1981. Mind and Meaning. Cambridge: Cambridge University Press.
McCarthy, T. 2002. Radical Interpretation and Indeterminacy. Oxford: Oxford Uni-
versity Press.
McCulloch, G. 1991. ‘‘Making Sense of Words’’, Analysis 51: 73–9.
MacFarlane, J. 2003. ‘‘Future Contingents and Relative Truth’’, The Philosophical
Quarterly 53: 321–36.
2005. ‘‘Making Sense of Relative Truth’’, Proceedings of the Aristotelian Society,
321–39.
2008. ‘‘Truth in the Garden of Forking Paths’’, in M. Kölbel and M. García-
Carpintero, eds., Relative Truth. Oxford: Oxford University Press, 81–102.
McGee, V. 2004. ‘‘Tarski’s Staggering Existential Assumptions’’, Synthese 142:
371–87.
McGinn, C. 1984. Wittgenstein on Meaning. Oxford: Blackwell.
McLaughlin, B., and Tye, M. 1998. ‘‘Externalism, Twin Earth, and Self-
Knowledge’’, in C. Wright, B. Smith, and C. Macdonald, eds. Knowing Our
Own Minds. Oxford: Oxford University Press, 285–320.
Martin, R. L., ed. 1984. Recent Essays on Truth and the Liar Paradox. Oxford:
Clarendon Press.
Meiland J. 1977. ‘‘Concepts of Relative Truth’’, Monist 60/4: 568–82.
Mercier, A. 1993. ‘‘Normativism and the Mental: A Problem of Language Indi-
viduation’’, Philosophical Studies 72/1: 71–88.
1999. ‘‘On Communication-Based De Re Thought, Commitments De Dicto,
and Word Individuation’’, in K. Murasugi and R. Stainton, eds., Philosophy and
Linguistics. Boulder, Colo.: Westview Press, 85–111.
Miller, A., and Wright, C., eds. 2002. Rule-Following and Meaning. Chesham:
Acumen Press.
Moore, G. 1980. ‘‘Beyond First-Order Logic, the Historical Interplay between
Logic and Set Theory’’, in History and Philosophy of Logic 1: 95–137.
1988. ‘‘The Emergence of First-Order Logic’’, in W. Aspray and P. Kitcher,
eds., History and Philosophy of Modern Mathematics, Minnesota Studies in
the Philosophy of Science, 11. Minneapolis: University of Minnesota Press,
95–135.
Neale, S. 1990. Descriptions. Cambridge, Mass.: MIT Press.
Parsons, T. 1987. ‘‘On the Consistency of the First-Order Portion of Frege’s
Logical System’’, Notre Dame Journal of Formal Logic 28: 161–8.
Patterson, D. 2002. ‘‘Theories of Truth and Convention T’’. Philosopher’s Imprint
2/5: 1–16.
  327

Peacocke, C. 1978. ‘‘Necessity and Truth Theories’’, Journal of Philosophical Logic

7/4: 473–500.
1998. ‘‘Implicit Conceptions, Understanding and Rationality’’, Philosophical
Issues, vol. ix: Concepts. Ridgeview: Atascadero, 43–88.
2000. ‘‘Explaining the A Priori: The Programme of Moderate Rationalism’’,
in P. Boghossian and C. Peacocke, eds., New Essays on the A Priori. Oxford:
Oxford University Press, 229–55.
Peirce, C. S. 1933. The Collected Papers of Charles Sanders Peirce, ed. C. Hartshorne
and P. Weiss. Cambridge, Mass.: Harvard University Press.
Putnam, H. 1973. ‘‘Explanation and Reference’’, in H. Putnam, Mind, Language,
and Reality: Philosophical Papers, vol. ii. Cambridge: Cambridge University Press,
1975, 196–214.
1975. ‘‘The Meaning of ‘Meaning’ ’’, in Hilary Putnam, Mind, Language, and
Reality: Philosophical Papers, vol. ii. Cambridge: Cambridge University Press,
215–71.
1983. ‘‘On Truth’’, in L. Cauman, I. Levi, C. Parsons, and R. Schwartz,
eds., How Many Questions? Essays in Honor of Sidney Morgenbesser. Indianapolis:
Hackett Publishing Co.
1988. Representation and Reality Cambridge, Mass.: MIT Press.
Quine, W. V. 1940. Mathematical Logic. Cambridge, Mass.: Harvard University
Press.
1960. Word and Object. Cambridge, Mass.: MIT Press.
1961a. ‘‘Notes on the Theory of Reference’’, in W. V. Quine, From a Logical
Point of View, rev. 2nd edn. Cambridge, Mass.: Harvard University Press, 130–8.
1961b. ‘‘Two Dogmas of Empiricism’’, in W. V. Quine, From a Logical Point
of View, rev. 2nd edn. Cambridge, Mass.: Harvard University Press, 20–46.
1963. ‘‘Carnap and Logical Truth’’, in P. Schilpp, ed., The Philosophy of
Rudolf Carnap: The Library of Living Philosophers, vol. xi. La Salle, Ill.: Open
Court, 385–406.
1976a. ‘‘Truth and Disquotation’’, in Ways of Paradox and Other Essays, rev.
and enlarged edn. Cambridge, Mass.: Harvard University Press, 308–21.
1976b. ‘‘Truth by Convention’’ (1935), in W. V. Quine, The Ways of Paradox,
rev. and enlarged edn. Cambridge, Mass.: Harvard University Press, 77–106.
1981. ‘‘What Price Bivalence?’’, in Theories and Things. Cambridge, Mass.:
Harvard University Press.
1982. Methods of Logic, 4th edn. Cambridge, Mass.: Harvard University Press.
1986. Philosophy of Logic, 2nd edn. Cambridge, Mass.: Harvard University
Press.
1987. Quiddities. Cambridge, Mass.: Harvard University Press.
328 

Quine, W. V. 1992. Pursuit of Truth, rev. edn. Cambridge, Mass.: Harvard

University Press.
1995a. ‘‘Interpretations of Sets of Conditions’’, in W. V. Quine, Selec-
ted Logic Papers, enlarged edn. Cambridge, Mass.: Harvard University Press,
205–11.
1995b: From Stimulus to Science. Cambridge, Mass.: Harvard University Press.
Rayo, A., and Uzquiano, G., eds., 2006. Absolute Generality. Oxford: Clarendon
Press.
Rescher, N. 1962. ‘‘Plurality-Quantification’’, Journal of Symbolic Logic 27: 373–4.
Resnik, M. 1997. Mathematics as a Science of Patterns. Oxford: Clarendon Press.
Rey, G. 1998. ‘‘A Naturalistic A Priori’’, Philosophical Studies 92: 25–43.
2006. ‘‘The Intentional Inexistence of Language—But Not Cars’’, in
R. Stainton, ed., Contemporary Debates in Cognitive Science. Oxford: Blackwell,
237–55.
Richard, M. 2004. ‘‘Contextualism and Relativism’’, Philosophical Studies 119:
215–42.
Russell, G. 2008. Truth in Virtue of Meaning. Oxford: Oxford University Press.
Schiffer, S. 1996. ‘‘Contextualist Solutions to Scepticism’’, Proceedings of the Aris-
totelian Society 1995/96: 317–33.
Shapiro, S. 1991. Foundations without Foundationalism: A Case for Second-order Logic.
Oxford: Clarendon Press.
Soames, S. 1984. ‘‘What is a Theory of Truth?’’ Journal of Philosophy 81/8: 411–29.
1998. ‘‘The Modal Argument: Wide Scope and Rigidified Descriptions’’,
Noûs 32: 1–22.
1999. Understanding Truth. Oxford: Oxford University Press.
2002. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity.
Oxford: Oxford University Press.
Sorenson, R. 2001. Vagueness and Contradiction. Oxford: Clarendon Press.
Spivak, M. 1994. Calculus, 3rd edn. Houston, Tex.: Publish or Perish, Inc.
Stalnaker, R. 1990. ‘‘Narrow Content’’, in C. Anderson and J. Owens, eds.,
Propositional Attitudes. Chicago: CSLI Lecture Notes.
Stanley, J., and Williamson, T. 2001. ‘‘Knowing How’’, Journal of Philosophy 98:
411–44.
Suppes, P. 1957. Introduction to Logic. New York: Van Nostrand Reinhold Co.
Tarski, A. 1936. ‘‘The Concept of Truth in Formalized Languages’’, in Logic,
Semantics, Meta-Mathematics: Papers from 1923 to 1938, trans. J. H. Woodger. 2nd
edn. Indianapolis: Hackett, 1983.
1944. ‘‘The Semantic Conception of Truth: And the Foundations of
Semantics’’, Philosophy and Phenomenological Research 4/3: 341–76.
  329

Tharp, L. 1975. ‘‘Which Logic is the Right Logic?’’ Synthese 3: 1–21.

Thomason, R. 1970. ‘‘Indeterminist Time and Truth-Value Gaps’’, Theoria 36:
264–81.
Travis, C. 1997. ‘‘Pragmatics’’, in B. Hale and C. Wright, eds., A Companion to the
Philosophy of Language. London: Blackwell, 87–107.
2000. Unshadowed Thought. Cambridge, Mass.: Harvard University Press.
Van Inwagen, P. 2002. ‘‘Generalizations of Homophonic T-sentences’’, in
R. Schantz, ed. What is Truth? Berlin: De Gruyter, 205–22.
Weiner, J. 2005. ‘‘Semantic Descent’’, in Mind, 114: 321–54.
Wetzel, L. 2006. ‘‘Types and Tokens’’, Stanford Encyclopedia of Philosophy,
http://plato.stanford.edu/entries/types-tokens/
Wiggins, D. 1980. ‘‘ ‘Most’ and ‘All’: Comments on a Familiar Programme, and on
the Logical Form of Quantified Sentences’’, in M. Platts, ed., Reference, Truth,
and Reality. London: Routledge and Kegan Paul, 318–46.
Williamson, T. 1994. Vagueness. London: Routledge.
2003. ‘‘Everything’’, Philosophical Perspectives 17: 415–65.
Wilson, G. 2000, ‘‘Satisfaction through the Ages’’, in A. Kanamori, ed., Proceedings
of the Twentieth World Congress of Philosophy, vol. vi: Analytic Philosophy and Logic.
Bowling Green: Philosophy Documentation Center, 89–97.
2006. ‘‘Rule-Following, Meaning, and Normativity’’, in E. Lepore and
B. Smith, eds., The Oxford Handbook of Philosophy of Language. Oxford: Claren-
don Press, 152–74.
Wilson, M. 1982, ‘‘Predicate Meets Property’’, Philosophical Review, 91/4: 549–89.
1993. ‘‘Honorable Intensions’’, in S. J. Wagner and R. Warner, eds.,
Naturalism: A Critical Appraisal. Notre Dame, Ind.: University of Notre Dame,
53–94.
Wittgenstein, L. 1969. On Certainty. Oxford: Blackwell Publishing.
2001. Philosophical Investigations, 3rd edn., trans. G. E. M. Anscombe. London:
Blackwell Publishing.
Wright, C. 1984. ‘‘Kripke’s Account of the Argument against Private Language’’,
Journal of Philosophy 81: 759–78.
This page intentionally left blank
Index

Acceptance constraint 223–4, 228, 231 Cappelen, Herman 124 n. 22, 125 n. 23,
Alward, Peter 124, 127 127 n. 26
agreement 90–2 Carnap, Rudolf on analyticity 184; on
ambiguity contextual 19–20; explication 66 n. 30; on interpretation
lexical 18–19; as homonomy 18. n 3 171; on sign designs and sign
analyticity 154–5, 170–1, 182–6 events 116 n. 13
anti-individualism 289, 313–16 causal-historical theory of
a priori 147 n. 1, 166, 206, 210–12, 297 reference 109–10, 186–91, 197–204
Assessment Relative Truth (ART) 280–5; Chalmers, David 204–13
see also MacFarlane, John changing the subject, versus
axioms of identity 48, 63; and discovery 210–11
validity 50 n. 11 Coady, C. A. J. 147 n. 1, n. 2, 169
n. 18
Barwise, Jon 35 compositionality 77 n. 40, 156–7, 165
belief sentences 36–8 n. 15
Berger, Alan 198 n. 16, 199 n. 18 context principle 290–1, 294, 303, 308,
Bezuidenhout, Anne 31 n. 23 313–15
binary quantification 34–6 Convention T 67–70; see also Tarski,
Blackburn, Thomas 316 n. 12 Alfred
Boghossian, Paul anti-reductionism about Cooper, Robin 35
meaning 113; on disquotational criterion and proof 19–20; and
deflationism about truth and rule-following 232–5, 238; weak
factuality 100 n. 9; on versus strong 229
meaning-constituting acceptance of
sentences 137 David, Marian on disquotationalism 80
Boolos, George 33–4, 50 n. 10, 54 n. 15 n. 45; on the problem of foreign
Borg, Emma 15 n. 1 intruders 81 n. 47; on using
Brown, Jessica against temporal substitutional quantification to define
externalism 262 n. 9; and truth 55 n. 18, 56
recognitional capacities 198 n. 16, 199 Davidson, Donald conception of the task
n. 18 and test of a theory of
Brueckner, Anthony 304 n. 5 interpretation 156–8, 165, 168;
Burge, Tyler arthritis criticism of Horwich’s minimal theory
thought experiment 148 n. 3, of truth 61 n. 25; criticism of Quine’s
@-operator 29 n. 21, 294 n. 1; proposal that we treat belief sentences
arthritis thought experiment 149; on as fused predicates 37–8; criticism of
basic self-knowledge 297–303; and Tarski–Quine thesis 3, 176;
history of anti-individualism 314 distinction between truth and
n. 10; on definitions of words already belief 167 n. 16; on defining truth for
in use 173 n. 21; and ‘‘social norms’’ utterances 79 n. 43; on explicable
for language use 316; on testimony error 161–3; ketch–yawl
and the principle of charity 147 example 161; on indexicals and
n. 1, n. 2 demonstratives 32; individualism of
Burgess, John 34 n. 25 his theory of meaning 169, 177–8; on
332 

Davidson, Donald (cont.) reference and conscious perceptual

intending to speak ‘‘correctly’’ 162; attention 31
paratactic account of belief Evnine, Simon 163 n. 14
sentences 37–8; principle of explanatory use (ex-use) see
charity 156–70; and radical token-and-explanatory (ex) use
interpretation 174–6; on relation explication account of 66–7; Davidson’s,
between transparent uses of of meaning 155, 167; of Kaplan’s
expressions and explanatory uses of common currency conception of
expressions 111; on quotation 140 words 122–4; of letters as classes of
n. 33; on self-knowledge 174, 296–7; marks 118 n. 16; of the PJSS-based
on using a Tarski-style theory of truth conception of words 128–33, 249,
to clarify meaning 77 n. 40, 109–10, 272 n. 14; MacFarlane’s, of the idea
154–67 that truth is relative 278–82; of
Davies, Martin 35, 39 ‘minimal self-knowledge’ 294–5, 306,
Dennett, Daniel 193 n. 11 309; of sentences as finite strings of
Devitt, Michael 198 n. 16, 307–8 classes of marks 119–20; of ‘true’
Dilworth, John 118 n. 16 79 n. 42, 84, 96–9, 101–2; of words as
disagreement 90–2 classes of marks 119, 122–4, 128,
discovery 92, 94–5, 102, 184–6, 188–9, 131–3
191–3, 196, 199, 201, 206–11, 214, extension core 239–41
219–20, 224–6, 232–3, 243, 260, extensional equivalence constraint 225–8,
269 n. 13, 276, 286 231–2
dispositions, linguistic 110, 113, 155, 171, externalism 315–16
182, 192–6, 197, 198 n. 16, 201–4,
209 facts 99
division of epistemic labour 229 n. 10, factuality 92, 98–104, 113–14, 134–8; see
308–11 also intersubjectivity constraint
division of linguistic labour 289, 300–3, Field, Hartry criticism of Horwich’s
309 Minimal Theory of truth 62 n. 26; on
Donnellan, Keith 194 n. 12 disquotational truth and foreign
Dummett, Michael acceptance languages 141 n. 34; on explaining the
constraint 223; on criteria for contribution of truth functional
applying words 229, 332–6; connectives to sentences in which
extensional equivalence they occur 76 n. 39; on Kripke’s
constraint 224–5; manifestability Schmidt–Gödel thought
constraint 224; on quantifying over experiment 293; on partial
everything 26 n. 15, 72 n. 35; on the extension 236–9, 241 n. 13, 242–6;
standard view of sameness of on using substitutional quantification
satisfaction across time 218–20; on to define truth 56 n. 20, 57; on the
sense and implicit knowledge 220–5 vagueness of semantic terms 245 n. 15
Fine, Kit 23 n. 9
Eklund, Matti 34 n. 24 Fodor, Jerry 84 n. 3
epistemically reasonable 105–6, 113, Frege, Gottlob on the distinction between
134–9; see also intersubjectivity belief and truth 183–4; on expressing
constraint logical generalizations without a truth
epistemic possibility 206 predicate 52 n. 13; on logical laws as
Evans, Gareth on minimal self-knowledge arbiters of collaborative inquiry 4,
and name-using practices 300–3, 305; 82–3, 102; on logical laws as
on pronouns 31 n. 23; on the role of a generalizations that hold for every
principle of charity in a non-reductive subject 317–18
theory of meaning 168; on singular Fricker, Elizabeth 147 n. 2
  333

Geach–Kaplan sentence 33–4 34–6, 40, 64, 78, 82–3, 102–3, 144,
Geach, P. T. 20, 30 288, 317; and methodological
Glanzberg, Michael 36, 65 n. 28, 68 n. 33 analyticity 184–5; and ordinary
Gödel numbering 69 language 17, 32–3; and primary
Goldberg, Sanford 316 n. intensions 206–7; and realism 102–4,
Goldman, Alvin 111 n. 7 286–7; and regimented
Goodman, Nelson 10 n. 7, 114–15, predicates 32–3, 77; and
117–18, 129 n. 28, 272 n. 14 rule-following 232–6; solitary 4, 83;
Grandy, Richard 170 n. 19 see also discovery, intersubjectivity
Grice, H. P. 109–10, 113 constraint, practical identifications of
generalizing on sentences 48 words, practical judgements of
sameness of extension, practical
Halbach, Volker 45 n. 6 judgements of sameness of satisfaction,
Hawthorne, John 194 n. 12, 277 n. 18 Tarksi–Quine thesis, temporal
Heim, Irene 36 n. 29 externalism
Horwich, Paul on analyticity and implicit intensions primary 204–13; secondary
definition 183 n. 5; on defining truth 208
for propositions 41, 79 n. 43; on interpretation Davidson’s theory of 156–8,
meaning and minimal 165, 168, 174–6; against a background
self-knowledge 305 n. 6; minimal of PIWs and PJSSs 171–6
theory of 58–63; on sorities intersubjectivity constraint compatibility of,
paradoxes and vagueness 23 n. 11; use with Tarski–Quine thesis 105–6,
theory of meaning 113, 198, 203 113–14, 133–40, 143; preliminary
formulation of 3–4, 82–4;
reformulation of 5, 95–6
indeterminacy and Dummett’s theory of
sense 230 n. 11; of extensions and
extension cores 237–46; and Jackman, Henry 264–76
realism 103–4, 286; of Jackson, Frank 204–13
translation 100–3, 113, 114 n. 9, Jeffrey, Richard 34 n. 25
155, 175 n. 25; see also temporal Jockusch, Carl 50 n. 10
externalism
identity axioms of 48, 63; and validity 50 Kitcher, Philip 198 n. 16
n. 11 knowing how 306–8
idiolect as related to a public language 229 Kratzer, Angelika 36 n. 29
n. 10; and Kaplan’s common currency Kraut, Robert 100 n. 9
conception of words 121–2; and Kripke, Saul causal-historical theory of
minimal self-knowledge 302 n. 4, reference 109–10; Putnam’s debt
305 n. 6; to 188 n. 7; on interpreting the
indispensability argument 40–8 ordinary use of ‘true’ 65 n. 29; on
implicit conceptions 213–16 letting a truth theory determine the
implicit knowledge 220, 222–6, 228–9, translation of the object language into
231, 233–5 the metalanguage 70 n. 34; on
inquiry, rational collaborative 4, 82–3, meaning scepticism 113; on natural
102, 104, 310; and context kind terms 109–10, 184–9, 289–91,
principle 130, 141, 256; and division 315 n. 11; Schmidt–Gödel thought
of epistemic labour 310; and experiment 292–4; simple
explication 66; and disquotational principle 90 n. 5; on
guidance/understanding 232–6; and substitutional quantification 56 n. 19;
implicit conceptions 213–16; and on transworld identity 272–3; theory
logical generalizations 1, 12, 14, of truth 67 n. 31
334 

Lance, Mark 194 n. 12, 277 n. 18 Neale, Stephen 28 n. 19, 31 n. 23, 35–6
Larson, Richard 39, 195 n. 13, 196 n. 15, necessity de re 38–9
248 n. 2 norms (social) for language use 315–16
Lasersohn, Peter 285 n. 24
learning from others 148–51, 158, 160, 167 ordinary language 32–3
n. 16, 169–70, 173, 177–8, 193–4, orthographic criteria for keeping track of
287, 309 our own regimented words 140;
Lepore, Ernie 31 n. 23, 36, 38 n. 30, 77 conception of words 107–8
n. 41, 155 n. 8
Lewis, David 39, 168, 170 n. 19
liar paradox 64–5, 67 partial extension 236–9, 241 n. 13, 242–6
linguistic competence see minimal linguistic Peacocke, Christopher 38–9, 138,
competence 213–16
Loar, Brian 40–8 Peirce, Charles Sanders 108 n. 3, 117 n. 14
Löwenheim–Hilbert–Bernays PJSS-based conception of words, and
theorem 50, 52 anti-individualism 313–16; and the
Ludwig, Kirk 31 n. 23, 36, 38 n. 30, 77 causal-historical of reference for
n. 41, 155 n. 8 predicates 289–91; and the
causal-historical theory of reference
for proper names 291–4; and the
McCarthy, Tim 267 n. 12 division of epistemic labor 308–11;
McCulloch, Gregory 123 n. 21 and externalism 315–16; and
MacFarlane, John 272 n. 15, 276–85 logic 316–19; and minimal
McGinn, Colin 113 self-knowledge 303–6; and practical
McLaughlin, Brian 304 n. 4 knowledge of how to use one’s own
manifestability constraint 224 words 306–8; preliminary
Meiland, Jack 281 characterization of 127–33; use of, to
Meno’s paradox 219 n. 3 satisfy the intersubjectivity constraint
Mercier, Adèle 127 n. 26, 316 n. 12 without rejecting the Tarksi–Quine
minimal linguistic competence 135, 205, thesis 133–40; use of, in second
295, 299–301, 303–8, 311 gold–platinum thought
minimal self-knowledge and basic experiment 269–75; use of, to solve
self-knowledge, according to the puzzles raised by the first
Burge 297–303; constraint on 294; as gold–platinum thought
first-order 303–6; as judged across experiment 249–56; use of, in third
time 311–13; as second order 295–7; gold–platinum thought
as practical knowledge 306–8 experiment 286–7; see also practical
Moore, Gregory 34 n. 24 judgements of sameness of
satisfaction
PJSS-based judgement of sameness of truth
names causal-historical theory of reference value 89–90, 92–3, 94–6, 130; see
for 186–7; of objects in a domain of also practical judgements of sameness
quantification 71; of logical constants of satisfaction
and predicates 71; proper, regimented PJSEs see practical judgements of sameness
as descriptions 17, 27–30, 32, 78, of extension
291–4; senses of 221 n. 6; sentences PJSSs see practical judgements of sameness
as, of truth values 53 n. 13, and of satisfaction
substitutional quantification 76; point of evaluation 279
structural-descriptive, of sentences 69 practical judgements of sameness of
natural kind terms see Kripke, Saul, extension (PJSEs) and the context
Putnam, Hilary, and Soames, Scott principle 249–56; and
  335

causal-historical theories 186–91, 197–204, 225–8, 289–91; criticisms of

197–201; and discoveries 186–91, Tarski’s method of defining truth 3,
247–62, 266–9, 272, 274–6, 286; and 47 n. 9, 81 n. 46, 83 n. 2; debt to
dispositions 201–4; and Kripke 188 n. 7; externalism 315; on
gold–platinum thought knowing the meanings of one’s own
experiments 191–4, 247–52, 252–6, words 305 n. 6; on not knowing what
269–76; and implicit one is talking about 311 n. 7; on
conceptions 213–16; and Twin Earth 188–91, 206, 313–16
methodological analyticity 181–6;
preliminary characterization quantification binary (most) 34–6; first
of 180–1; and primary order (objectual) 25–7, 33–9, 75–7;
intensions 204–13 second order 34, 53–5, 60–2;
practical judgements of sameness of substitutional 55–7, 76
satisfaction (PJSSs) across time 89; and quotation 140 n. 33
agreement and disagreement Quine, W. V. on artificial
90–2, 172–3; at a given time 89; and notations 14–15; on belief
the context principle 127–33, sentences 37; on bivalence 24 n. 13;
249–56; and discoveries 94–5, 179, on explication 66; on the clarity of
185, 247–62, 266–9, 272, 274–6, 286; disquotational truth 2; on facts 99; on
and explanatory (ex) use 111, 113–14; the deviant logician’s
factuality of 92, 98–104, 113–14, predicament 183; on the
134–8; and gold–platinum thought Geach–Kaplan sentence 33–4;
experiments 191–4, 247–52, 252–6, indispensability argument 40–8; on
269–76; and learning from letter and word classes 119–20; on the
others 93–4, 147–54, 159–61, obscurity of talk of meaning and
166–78; and the intersubjectivity synonymy 154–6; on mass terms and
constraint 95–7; preliminary natural kind terms 188 n. 6; on proper
characterization of 87–9; names as descriptions 28–9, 292; on
trustworthiness of 91–2, 97–8, satisfaction and truth for foreign
138–40 languages 141 n. 34; on the semantics
PIWs see practical identifications of words of the predicate ‘true’ 102–4, 286–7;
practical identifications of words (PIWs) on singular terms 27; on Tarski’s
across time 85; and agreement and method of defining truth 72 n. 36,
disagreement 90–2, 172–3; at a given 73 n. 38, 81 n. 48; theory of linguistic
time 85; and the context meaning 109–10; token and ex-use
principle 128–33, 249–56; and conception of words 112, 113; on
discoveries 94–5, 179, 185, 247–62, translation and indeterminacy 100–3,
266–9, 272, 274–6, 286; preliminary 113, 114 n. 9, 155, 175 n. 25; on
characterization of 84–7; and practical translating logical words 176–7; on
judgements of sameness of satisfaction the transparency of disquotational
(PJSSs) 87–9; and learning from truth 107 n. 2
others 93–4, 147–54, 159–61,
166–78
principle of charity 156–70; see also realism 102–4, 286–7
Davidson, Donald recognitional capacities 198 n. 16, 199
principle of humanity 168 n. 18
proof 220–25 reflective equilibrium 10, 129
Putnam, Hilary anti-individualism regimentation and ambiguity 17–20; and
313–16; causal-historical theory of explication 66–7; and indispensability
reference 109–10, 113, 186, 188–91, of ordinary language 32–3; limits of,
336 

regimentation and ambiguity (cont.) names as descriptions 28 n. 20;

into a first-order language 33–40; version of the liar paradox 64–5
pragmatic account of 14–17; Tarski–Quine thesis compatibility of, with
possibility of 20–2; of pronouns and intersubjectivity constraint 105–6,
demonstratives 30–2; of quantifier 113–14, 133–40, 143; statement of 3,
domains, of tense and discourse about 79; summary of motivation for 1–3,
time 25–7; regulative ideal of 20; and 78–9
vagueness 22–5 temporal externalism 260, 264–85
regimented language 17 testimony 147 nn. 1, 2
Reimer, Marga 31 n. 23 Tharp, Leslie 34 n. 26
relativity of adjectives 36 thought experiment arthritis 148–51; blue
replica, letter or word 117–20, 122–5, gold 184–5; druid 193 n. 11, 202
131–3, 140–1 n. 19; gold–platinum: first 191–4,
Rescher, Nicholas 35 247–52; second 252–6;
Resnik, Michael 141 n. 35 third 269–76;
Rey, Georges 115 n. 12, 120 n. 18, 137 Schmidt–Gödel 292–4;
Russell, Gillian 212 n. 23 witch 194 n. 12, 277 n. 18;
water-XYZ 188–91, 206, 313–16
Segal, Gabriel 39, 195 n. 13, 196 n. 15, 248 token-and-explanatory (ex) use
n. 2 characterization of 112–14; and
self-knowledge, minimal; see minimal Davidson’s theory of
self-knowledge interpretation 145–8, 154–65, 174;
sense 220–2, 224–37, 245 and Dummett’s theory of
Shapiro, Stewart 34, 53 sense 218–30; and Field’s theory of
Soames, Scott on ‘‘actually’’ as a rigidifying partial extension 236–41; and the first
operator 29 n. 21; on ambiguity as gold–platinum thought
homonomy 18 n. 3; on defining truth experiment 191–216, 249–52; and an
for propositions 41, 62, 79 n. 43; on intuition about sameness of extension
natural kind terms as predicates across time 179–82; and Jackman’s
187 n. 6; on substitutional meaning-constituting
quantification 55 equilibria 266–76; preliminary
Sorenson, Roy 23 n. 10, 24 n. 13 characterization of 112–14; and the
sorities paradox 23–4 second gold–platinum thought
Stanley, Jason 307–8 experiment 252–57; and question of
ST-schema 75–7 the incompatibility of intersubjectivity
ST-sentence 75–7 constraint with Tarski–Quine
Sterelny, Kim 198 n. 16 thesis 113–14, 133–40, 143; and
structural-descriptive name 69 temporal externalism 257–84
substitution of coextensive terms in opaque translation contrasted with practical
contexts 36–7; of regimented identifications of words 194 n. 12; in
predicates and sentences for schematic Convention T 68–70, 74; and Field’s
term and sentence letters 41, 45–6, 47 notion of partial extension 327; as
n. 8, 49–51, 55–7, 318; and perspective relative 277 n. 18;
substitutional quantification 55–7, Quine’s empirical theory of 100–3,
76 114 n. 9, 155, 175 n. 25
transparent use preliminary characterization
of 106–7; relationship to explanatory
Tarski, Alfred Convention T 67–70; (ex) use 111; and minimal linguistic
method of defining truth in terms of knowledge 299, 302, 305–9
satisfaction 70–4; on relativity of Travis, Charles cases described by, that
truth to a language 81; on treating illustrate contextual ambiguity 19; on
  337

occasion sensitivity 20 n. 6, 20 n. 7; quantification 34, 53–5, 60–2; and

claim of, that predicates don’t have sentences we have not translated 141,
extensions 20–1; on 155 n. 9; strong criterion for 229, 282;
‘‘understandings’’ of a predicate and ST-sentences 75–7; and
21–2 substitutional quantification 55–7, 76;
truth a priori 210; assessment and the Tarski–Quine thesis 1–3,
relative 280–5; and belief 102–3, 78–9, 82, 105–6, 113–14, 133–40,
166–7, 183–4, 270–1, 310; 143; and temporal
conditions 155–6, 295–8, 301–4; and externalism 259–60; and
Convention T 67–70; and Davidson’s token-and-explanatory (ex)
principle of charity 155–67; use 112–13, 133–4; and
Davidson’s theory of, for natural translation 69–70, 194 n. 12, 101,
languages 155–65; as defined for 277 n.; and T-sentences 45–8, 63–65;
propositions 41, 58–63, 79; degrees and validity 48–52; values as
of 245 n. 15; determinate 238–9; objects 221; values of vague
disquotational 47, 50, 52, 87, 92, 98, sentences 23–4; in virtue of
100–3, 107 n. 2, 127–8, 138–9, 144, meaning 154; in virtue of reference
171, 176, 178, 211; Field’s deflationary determiner 212 n. 23
account of 293; or falsity of our T-schema 63–5; see also ST-schema
PJSSs 186; and foreign languages 141, T-sentence 45–8, 63–5
155 n. 9; and Frege’s horizontal 52 Tye, Michael 304 n. 4
n. 13; and generalizing on types and tokens preliminary
sentences 40–52, 69, 96; and characterization of 114–18; and
indispensability argument 40–8; and Kaplan’s common currency
intersubjectivity constraint 3–4, conception of words 122–3; and the
82–5, 95–6, 105–6, 113–14, 133–40, PJSS-based conception of
143; and holding true 146, 157–65; words 127–31
and the liar paradox 64–5, 67; and
logic 14, 40–52, 69, 83, 107, 126,
use (transparent) 106–7
131, 142, 316–19; non-disquotational
notions of 77 n. 40, 176; and ordinary
uses of ‘true’ 47 n. 9, 65 n. 29, 67 validity 48–52
n. 31, 84, 104, 142; perspective vagueness 22–4
relative 194 n. 12, 277 n. 18; at a point
of evaluation 277–8; and PJSS-based Weiner, Joan 52 n. 13
conception of words 141–2, 316–19; Wetzel, Linda 119 n. 17, 121 n. 19
PJSS-based judgements of sameness of, Williamson, Timothy on knowing how as
value 89–90, 92–3, 94–6, 130; and propositional knowledge 307; on
PJSSs 211, 138; predicate 40–8, quantifying over everything 73 n. 35;
50–2, 54 n. 14, 57, 80, 83, 92, 96–8, on vagueness 23 n. 8
100–4, 106, 119, 127–8, 140, 142, Wilson, George 260–4
144–5, 155 n. 9, 288; pursuit Wilson, Mark 193 n. 11, 202 n. 19, 211
of 183 n. 5, 211, 289, 319; and n. 22, 218 n. 1
regimentation 14–17; and regimented Wittgenstein, Ludwig 53 n. 13, 113, 115
predicates 77; relative to an n. 11, 136 n. 31
interpretation 50; and realism 102–4, words token-and-explanatory model
286–7; and satisfaction 77, 108, 112, of 112–14; PJSS-based conception
120, 129, 133, 138–42, 145–6, 148, of 127–33; as sets of tokens
155, 173, 178; and second-order (marks) 119–20, 122–3, 131–3;
338 

words token-and-explanatory model (cont.) see also intersubjectivity constraint,

indispensability of, for mathematics words, practical judgements of
and logic 120; orthographic sameness of extension, practical
conception of 107–8, 114, 119, 125; judgements of sameness of satisfaction
practical identifications of 84–7; Wright, Crispin 113