Consciousness and Meaning Selected Essays

Consciousness and Meaning
Consciousness and
Meaning
Selected Essays
Brian Loar
EDITED BY
Katalin Balog and Stephanie Beardman
WITH INTRODUCTIONS BY
Stephen Schiffer and Katalin Balog
1
OUP CORRECTED PROOF FINAL, 12/12/2016, SPi
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the Universitys objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
in this volume Stephanie Beardman 2017
The moral rights of the authors have been asserted
First Edition published in 2017
Impression: 1
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, by licence 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 work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2016952448
ISBN 9780199673353
Printed in Great Britain by
Clays Ltd, St Ives plc
Preface
Ten years ago Brian Loar, my husband, arranged to publish a selection of his work
with Oxford University Press, but then illness prevented him from carrying out the
project himself. Before he died in 2014, Brian spoke of this collection with happy
anticipation, and though I regret that he did not live to see it materialize, I am much
gratied to see it brought forth now.
We dont know exactly which papers Brian would have chosen. In addition to
papers that are well known, we wanted to include signicant and difcult-to-locate
worksencouraged by the volume of emails Brian received from philosophers all
over the world requesting articles originally published in out of print collections that
are not accessible online. At the end of the book, we provide what we take to be a
complete list of Brians publications; circumstances and the passage of time dont
allow absolute certainty that everything is included, but we hope the list is useful.
The papers are arranged mostly in chronological order and introduced in two
parts. Stephen Schiffer presents Brians work in the philosophy of language; Katalin
Balog introduces Brians papers in the philosophy of mind. These two philosophical
subelds interlink, of course, but this is especially true in Brians work: He had a large
holistic conception of consciousness, thought, language, and meaning that drove him
to pursue questions and propose answers that rely on a substantively unied view of
mind and meaning. The psychology of communication was something to which he
was particularly attuned, and it is not surprising that his curiosity about the nature of
language and of thought motivated him throughout his life. There was no sharp
division between philosophy and his life. He was, as they say, a philosophers
philosopher: fellow philosophers read him to learn not just what he thinks, but
how he thinks.
Stephanie Beardman
New York
October 5, 2015
Contents
Acknowledgments ix
Part I. Philosophy of Language

Introduction to Part I 3
Stephen Schiffer
1. Reference and Propositional Attitudes 14
2. Two Theories of Meaning 29
3. The Semantics of Singular Terms 48
4. Must Beliefs Be Sentences? 69
5. Names in Thought 84
6. Truth beyond All Verication 97
7. The Supervenience of Social Meaning on Speakers Meaning 124
Part II. Philosophy of Mind

Introduction to Part II 137
Katalin Balog
8. Social Content and Psychological Content 153
9. Subjective Intentionality 165
10. Phenomenal States 195
11. Can We Explain Intentionality? 220
12. Elimination versus Non-reductive Physicalism 238
13. Reference from the First-person Perspective 256
14. Transparent Experience and the Availability of Qualia 273
15. Phenomenal Intentionality as the Basis of Mental Content 291
List of Works by Brian Loar 319

Bibliography 321
Index 329
Acknowledgments
This book like all things has come into existence as a result of many conditions. The
biggest factor, of course, has been the inspiration of Brians work. A number of
people have generously offered assistance and advice for which we are grateful.
Special thanks are due to our editor at Oxford University Press, Peter Momtchiloff,
who has been encouraging and helpful in every way. Lisa Miracchi and Zachary
Miller have assisted with many time-consuming editorial tasks. Barry Loewer and
Georges Rey spent long hours going through the manuscript; we are very grateful for
their insightful suggestions. Special thanks to Stephen Schiffer for agreeing to write
one of the introductions and for the many other ways in which he helped this book to
come together.
Katalin Balog
Stephanie Beardman
The editors and Oxford University Press would like to thank all the publishers for
permission to republish the following papers previously published elsewhere. None
but the most minor editorial alterations have been made to the published texts. Every
effort has been made to trace copyright holders, but if any have been inadvertently
overlooked the publisher will be pleased to make the necessary arrangement at the
rst opportunity.
Reference and Propositional Attitudes (1972). The Philosophical Review 81:
4362.
The Semantics of Singular Terms (1976). Philosophical Studies 30: 35377.
Must Beliefs Be Sentences? (1982). In P. Asquith and T. Nickles (eds.), Proceed-
ings of the 1982 Biennial Meeting of the Philosophy of Science Association, 2,
Philosophy of Science Association, 62743.
Names in Thought (1987). Philosophical Studies 51: 16985.
Subjective Intentionality (1987). Philosophical Topics 1: 89124.
Truth beyond All Verication (1987). In B. Taylor (ed.), Michael Dummett:
Contributions to Philosophy, Martinus Nijhoff, 81116.
Social Content and Psychological Content (1988). In R. H. Grimm and
D. D. Merrill (eds.), Contents of Thought: Proceedings of the 1985 Oberlin
Colloquium in Philosophy, University of Arizona Press, 99110.
Can We Explain Intentionality? (1991). In B. Loewer and G. Rey (eds.), Meaning
in Mind: Fodor and His Critics, Blackwell, 11936.
Reference from the First-Person Perspective (1995). Philosophical Issues: Con-
tent, 6, Ridgeview Publishing Company, 5372.
Phenomenal States: Second Version (1997). In N. Block, O. Flanagan, and
G. Guzeldier (eds.), The Nature of Consciousness: Philosophical Debates, MIT
Press, 597616.
The Supervenience of Social Meaning on Speakers Meaning (2001). In
G. Cosenza (ed.), Paul Grices Heritage, Brepols, 10113.
Phenomenal Intentionality as the Basis of Mental Content (2003). In M. Hahn
and B. Ramberg (eds.), Reections and Replies: Essays on the Philosophy of Tyler
Burge, MIT Press/Bradford Books, 22957.
PART I
Philosophy of Language
Introduction to Part I
Stephen Schiffer
I rst met Brian Loar when I happened to sit next to him at a philosophy talk in
Oxford in January 1964. Because I had been away from Oxford during Michaelmas
term 1963, when Brian entered Balliol, it was only during the conversation I had with
him that evening that I learned we were both Balliol research students who happened
to be living in college. I cant recall who the speaker was that evening, what his or
her talk was about, or the college in which the talk was held, but I have a vivid
recollection of meeting Brian. That may have been because immediately after intro-
ducing himself he asked me if I was a Quinean. Brian and I fast became very close
friends. This had as much to do with our distaste for Balliols dinners as it did with
our shared philosophical interests: several nights a week we would agree that the
thought of eating in Hall was intolerable and head off for the Taj in the Turl. Well,
there was also the fact that our philosophical conversations were enhanced by the
copious amounts of scotch and numerous unltered Players cigarettes that accom-
panied them. In what I think was the winter or spring of 1966, Brian and I attended a
seminar Paul Grice gave on topics he wanted to cover in the William James Lectures
he was committed to giving at Harvard in early 1967. After a few sessions of the
seminar, in which Brian and I routinely gave him a hard time, Grice asked us if we
would meet with him several mornings a week in his rooms at St Johns College to
discuss issues that arose in the seminar. Of course we agreed, and in those sherry-
drenched morning meetings began a close friendship that Brian and I enjoyed with
Paul Grice until his death in 1988.
Brian and I were already working collaboratively on the Gricean program when we
began meeting with Grice. I had recently begun work on my dissertation on Meaning1
under Peter Strawsons supervision, and Brian was just beginning to think about
doing a dissertation on Sentence Meaning. As we conceived the Gricean program, it
was a reductionist program in two parts. In the rst part, a notion of speaker-meaning
was dened, without presupposing any semantic notions, in terms of acting with the
1
A slightly revised version of the dissertation was published under the same title in 1972 by Oxford
University Press.
INTRODUCTION TO PART I
intention of affecting an audience in certain ways. In the second part, expression-

meaning, the semantic properties of linguistic items, was dened, again without
presupposing any semantic notions, in terms of the now dened notion of speaker-
meaning, together with certain ancillary notions, such as that of a conventional
regularity in behavior, which themselves were denable in terms of non-semantic
propositional attitudes. If the program succeeds, it will have reduced all questions
about linguistic representation to questions about mental representation. Three
aspects of the program were of especial interest to Brian, and knowing what they
are may provide some useful background to the papers collected in this volume.
Reduction. The Gricean program, as conceived by Brian and me, was a reductionist
program: it aimed to reduce semantic facts to propositional-attitude facts. That
aspect of the program was attractive to Brian, but it was attractive to him because
he saw it as essential to a larger reductionist aspiration of his. This was the reduction
of semantic and propositional-attitude facts ultimately to physical facts, perhaps via a
reduction to functional facts realized by physical facts; for Brian saw that as the only
way of making sense of the place of intentionalitythat is to say, of content-
involving factsin the natural order. Impressed by Hilary Putnams early work on
minds and machines, and then, a few years later, David Lewiss and Jerry Fodors
work on functionalism, Brian believed that psychological notions generally, but
especially propositional-attitude notions, could be dened in terms of functional
properties via their roles in psychological theories in which those notions functioned
as theoretical constructs, and it went without saying by everyone who was attracted
to functionalism that physical states and properties would be the ultimate realizers of
those functional properties. But while functionalism seemed a promising way to
reduce propositional-attitude and other psychological notions to non-psychological
notions, it didnt seem to hold much promise as a way to achieve the needed
reductionist account of semantic facts. Its precisely at this point that the attractive-
ness of the Griceans reduction of semantic notions to propositional-attitude notions
reveals itself to the aspiring physicalist: in showing that, and how, the intentionality
of language derives from the intentionality of thought, the physicalist is able to see
how she can get her desired reduction of all intentional notions by rst reducing
semantic notions to propositional-attitude notions, and then reducing propositional-
attitude notions to functional notions realized by physical states and properties. Back
in the day, Griceans were often asked what was the point of a theory that denes the
representational features of language in terms of the representational features of
thoughts when it has nothing to say about the latter. Brian Loar saw the point. Brians
romance with functionalism culminated in his profound and ingenious book Mind
and Meaning (1981), in which he worked out in great detail what I believe is the best
functionalist account of propositional attitudes. Alas, however, best doesnt mean
true, and once he had worked out what a functionalist account of propositional
attitudes needed to be, he was liberated to nd problems with it. Brian gave up on
functionalism in the mid-eighties and it was then that he began his work on
subjective intentionality, represented primarily in the chapters that Katalin Balog
will introduce.
Expression-meaning. The biggest challenge to the Gricean program was to show in
detail how expression-meaning, the semantic features of linguistic items, could be
explicated in terms of the Griceans account of speaker-meaning (together with
relevant ancillary notions). Brians greatest interest in the Gricean program was in
seeing how this challenge could be met, and when, upon achieving the B.Phil. degree
in spring 1965 he was awarded a prize research fellowship at Magdalen College, he set
out to meet that challenge in his doctoral dissertation on Sentence Meaning. The
account of sentence meaning I offered in Meaning was to a large extent based on an
idea of Brians.
Reference. The referential properties of expressions are among their most important
semantic properties; one cant be interested in expression-meaning without being
interested in reference. Brians interest in the problems of referencethe problems
posed by Frege, Russell, Strawson, Quine, and otherspredated his involvement with
the Gricean program, and several of those problems famously concern the behavior
of singular terms in the that-clauses of belief and other propositional-attitude
reports, and the sort of functionalism that interested Brian had important implica-
tions for the semantics of those reports. But the Gricean program had its own special
interest in reference. For just as there is a speaker-meaning/expression-meaning
distinction, there is also a speaker-reference/expression-reference distinction, and
this strongly suggests that for the Gricean expression-reference needs to be dened
in terms of speaker-reference, which in turn needs to be dened in terms of the
Griceans account of speaker-meaning. That is no easy feat to accomplish, and the
denitions that seem to be needed have no small degree of complexity, as is witnessed
by the one thing Brian and I ever coauthored: the recursive denition of a speakers
referring to something qua its being such and such that appeared in my article
Indexicals and the Theory of Reference (Schiffer, 1981a).
Chapter 1 Reference and Propositional Attitudes (1972). This was Brians rst
publication, and in it he offers a solution to a problem that was much exercising
philosophers at the time; it was intended to be a solution that a philosopher could accept
whether or not she was a physicalist or bought into the Gricean program. The problem
concerns belief reports that apparently contain referential occurrences of singular terms,
even though replacing the term with a coreferential term might well result in a belief
report whose truth-value differs from that of the rst report. To see this, suppose that at a
costume ball S gestures in the direction of a certain masked man and says to A,
(1) Michael thinks that that masked man is a diplomat,
and then for good measure adds, I know this because I overheard Michael say to
Jane, That masked man is a diplomat. Intuitively, in uttering (1) S referred to the
masked man with her utterance of that masked man and reported Michael as
thinking something about him, to wit, that he is a diplomat. We would have no
trouble believing that Ss utterance of (1) was true. But now suppose S then utters
(2) But Michael doesnt think that that guy who insulted Carla is a diplomat
although as it happens, that masked man is that guy who insulted Carla.
Here, too, it would seem that, intuitively, in uttering (2) S referred to the guy who
insulted Carla with her, Ss, utterance of that guy who insulted Carla, and here, too,
we would have no trouble accepting Ss utterance of (2) as true. But how can that be if
the occurrence of that masked man in (1) and the occurrence of that guy who
insulted Carla in (2) refer to the same person?
Nowadays philosophers are familiar with a few ways (1) and (2) can be true even
though the occurrences of the two singular terms refer to the same person, but that
wasnt so when Brian wrote his paper in what was probably no later than 1971. The
solution he offered was inspired by Quines example involving the pair of sentences
(3) Giorgione was so called because of his size,
which is true, and
(4) Barbarelli was so called because of his size,
which is false, notwithstanding that Giorgione in (3) and Barbarelli in (4) refer to
the same person. We dont think this is really much of a puzzle, because its pretty
clear what its solution is: as Quine points out, the difference in truth-value is
immediately explained when we see that (3) and (4) are equivalent, respectively, to
(5) Giorgione was called Giorgione because of his size [true]
(6) Barbarelli was called Barbarelli because of his size [false].
Brians proposal was that, just as Giorgione is revealed to make two contributions
to the truth conditions of (3), one referring to the painter, the other not referring to
him, so that masked man makes two contributions to the truth conditions of the
sentence to which the that-clause in (1) refers, so that the logical form of (1)
becomes
(7) B (Michael, x is that masked man and x is a diplomat, that masked man),
thus revealing that masked man in (1) to make two distinct contributions to the
truth conditions of (1): one of these is referential; the other is a contribution to the
satisfaction conditions of the whole or, more precisely, a partial determination of
what open sentence is the middle term of the triadic relation asserted by (1) (p. 21).
(At this interim stage of his analysis Brian, in deference to Quines formulation of the
problem, is provisionally assuming that representations of de re belief reports involve
a triadic relation among an agent, an n-place open sentence, and an n-ary sequence of
the things of which the agent is represented as believing the open sentence to be true.)
Chapter 2 Two Theories of Meaning (1976). This chapter reveals Brians

engagement with the Gricean program more than any other in this collection.
Donald Davidson spent the academic year 196970 in Oxford, in part to give the
John Locke Lectures, and his stay had a tremendous effect on such young Oxford
philosophers as Gareth Evans, J. A. Foster, John McDowell, Christopher Peacocke,
and David Wiggins, and even on more established philosophers such as Michael
Dummett and Peter Strawson. Davidson managed to convert a few of these philo-
sophers to his view that a meaning theory for a natural language should take the form
of a Tarski-style truth theory for that language, but what was more incredible was
that he managed to get all of them working on his semantic project. That engagement
was celebrated in Truth and Meaning: Essays in Semantics, edited by Gareth Evans
and John McDowell, in which almost all the papers engaged with issues dened by
Davidsons semantic project. Two Theories of Meaning was Brians contribution to
that collection. The two theories referred to in the papers title were Davidson truth-
theoretic semantics and the Griceans intention-based semantics. I wont rehearse the
objections Brian leveled against Davidsons program, but in my estimation they are
devastating. The part on Davidson is the negative part of the chapter, the positive
part builds towards a Gricean account of expression-meaning. The strategy Brian
used was borrowed from David Lewis.2 Let a language be any function from nite
sequences of sounds or marks or whatever onto intensions, things that might actually
be sentence meanings. If L is in this stipulated sense a language and L() = m, then we
may say that is a sentence of L and m is its intension, or meaning, in L. These
stipulative denitions arent the notions Brian is out to explain; they are merely
things to be used to get to the account of sentence meaning he seeks. This is the sense
of meaning in which meaning depends on use, and the Lewisian strategy is to dene a
relation Rthe actual language relationthat must obtain between a language L and
a population of speakers P in order for L to be the language of P. If R(L, P), then L is
the language used in P and every sentence of L means in P what it means in L. The
notion means m in P is the use-dependent notion of meaning that Brian sets out to
explicate. The account of the actual-language relation Brian gives is Gricean because
it assumes a Gricean account of speaker-meaning and a broadly Lewisian account of
convention and says that L is the language of P just in case there prevails in L a
convention conformity to which requires a member of P who utters a sentence of L in
circumstances of a certain kind to mean thereby a proposition that ts the meaning
that sentence has in L. Needless to say, its all in the details, and the account Brian
ends up with involves a subtle marriage of Grice and Chomsky; its probably as good
an account of sentence meaning as a Gricean can hope to achieve.
Chapter 3 The Semantics of Singular Terms (1976). Brian begins with a
beautifully concise statement of how for the Gricean semantical theory is part of
2
See, e.g., Lewis (1969) and (1975).
the psychology of communication (p. 49) and the sentences of a language form a
system of conventional devices for making known communicative intentions
(p. 48). On this view, a sentences meaning is something that constrains what a
speaker can mean in producing a literal and unembedded utterance of the sentence.
Within this Gricean framework, Brian then sets out a Fregean view of singular terms
wherein the function of a singular term is to introduce an individual concept into
what is meant or expressed on its particular uses (p. 49), and along with that a
description theory of names according to which the meaning of a name N is
equivalent to the meaning of the denite description the thing or person called N.
Brian develops the details of this account with his characteristic subtlety and ingenu-
ity. In defending his Fregean views, Brian shows that, contrary to what just about
everyone at the time believed, Kripke hadnt disproved the description theory of
names. Kripkes Feynman and Gdel counterexamples fail because in both cases he
overlooks the possibility that the reference of a name for a speaker may be xed by a
metalinguistic description of the form (roughly speaking) the person called N by
those from whom I acquired the name. Kripkes modal argument to show that
names arent rigid designators fails because it depends on the claim that, while Saul
Kripke might not have been Saul Kripke has no reading on which its true, it would
have a reading on which its true if the description theory were true and the sentence
were equivalent to, say, The author of Naming and Necessity might not have been the
author of Naming and Necessity. But Brian in effect points out that its consistent
with the description theory that names must take a wider scope than the modal
operator in such sentences so that, if Saul Kripke meant the same as the author of
Naming and Necessity, then Saul Kripke might not have been Saul Kripke could
only mean The author of Naming and Necessity is such that he might not have been
himself, which, of course, has no reading on which its true. One might also note that
in the course of developing his views Brian presents an account of self-ascriptive
belief (e.g., I believe that Im a paragon) that anticipates David Lewiss account of
de se belief (as Lewis himself acknowledges).3
Chapter 4 Must Beliefs Be Sentences? (1982). Here Brian compares and con-
trasts two kinds of functionalist accounts of propositional attitudes: the language of
thought hypothesis (LOT) and the propositional attitude based theory (PAT) (my
acronyms). LOT is a philosophical hypothesis that presupposes a scientic hypoth-
esis. The presupposed scientic hypothesis is that central among the causes of our
behavior are inner states with linguistic structure that play roughly the role we pre-
scientically ascribe to our beliefs and desires (p. 69). The philosophical hypothesis
is an explicative strategy for determining the contents of propositional attitudes. Let
believes* express that relation that a person x bears to a sentence s of her mentalese
just in case s is tokened in x as a belief. Then the idea is that we reach an account of
3
Lewis (1979).
what it is for a person to believe a proposition in two steps: in the rst step we dene
s means p in xs mentalese, and then in the second step we say that:
x believes p iff for some sentence s of xs mentalese, x believes* s and s means p in
xs mentalese.4
PAT is also an explicative strategy for determining the contents of propositional
attitudes. It explicates x believes p directly, rather than explicating it in terms of a
more fundamental ascription of content; so if there is a language of thought, its
sentences meanings would be explained in terms of the propositional attitudes they
somehow instantiate or realize (p. 70). Brian calls PAT a functionalist theory of
propositional attitudes, but that may be slightly misleading in that, unlike the most
familiar versions of functionalism, PAT doesnt dene believing as a functional
relation; that is to say, it doesnt say that x believes p just in case x is in a state that
has such-and-such functional role. PAT, like LOT, offers a two-component analysis of
content. In LOT believing* is dened as a functional relation between a person and a
sentence of mentalese, and then believing is dened in terms of believing* and
meaning-in-mentalese. PAT interprets propositional-attitude ascriptions as ascrib-
ing functional states with certain associated truth conditions (p. 70): the rst
component of content (narrow content) is that determined by the ascribed func-
tional, or conceptual, roles, and the second component of content (wide content) is
whatever determines the truth conditions the ascription associates with the func-
tional role it ascribes. While narrow content doesnt determine wide content, it does
make a context-free contribution to wide content (i.e., to truth conditions); specic-
ally, it determines what Brian calls general truth conditions, and which may be
explained in the following way. Suppose Lulu and Marie each has a belief she
would self-ascribe using I am witty. The specic truth conditions of Lulus belief is
that she is witty, and that of Maries belief is that she is witty; but there is a sense in
which their two beliefs have the same truth conditionsto wit, each is true iff the
believer is wittyand Brian calls these truth conditions general truth conditions.
According to PAT, functional role determines both general truth conditions and the
way in which specic truth conditions then depend on context. According to LOT,
he says, wide contenti.e., specic truth conditionsis to be determined by an ideal
indication theory, which seeks to rene the rough idea that p is the truth condition of
s for x iff ps being true would given optimal conditions cause x to believe* s.5
Brians comparison of LOT and PAT is done with an eye towards showing that
PAT is the better strategy for explicating the propositional contents of beliefs, but
I think Brian was already concerned about problems confronting the version of
functionalism defended in Mind and Meaning, and the chapter tellingly ends with
4
Ive changed the wording, but not the content, of Brians formulation a little.
5
Again, my wording departs slightly from Brians.
what may now be seen as a good segue to how his soon-to-be-discovered theory of
subjective intentionality enables a physicalistically acceptable account of wide content.
Chapter 5 Names in Thought (1986). Here we nd Brian weighing in on
Kripkes puzzling Pierre.6 Kripke argues that certain principles governing our de
dicto belief ascriptions lead us in the case of Pierre to a contradiction. In Paris Pierre
says Londres est jolie, and principles of disquotation and translation entitle us to say
Pierre believes that London is pretty. But Pierre moves to an ugly part of London
where he nds himself living among fairly uneducated people, and even after picking
up English from interacting with the people around him, he doesnt realize that
Londres and London refer to the same city, and, without changing his mind about
Londres est jolie, he now says London isnt pretty (and so of course is unwilling to
say London is pretty). His saying that conjoins with what Kripke calls the strength-
ened disquotation principle (A normal English speaker who is not reticent will be
disposed to sincere reective assent to p [iff] he believes that p) to entitle us to say
Pierre does not believe that London is pretty. Brian thinks that our ordinary belief
ascriptions dont lead to contradiction, and that the appearance of contradiction
disappears when we realize, rst, that we have standards of varying degrees of
permissiveness as to when we can say such things as Pierre believes that London is
pretty, and second, that Kripkes strengthened disquotation principle really ought to
read A normal English speaker who is not reticent will be disposed to sincere
reective assent to p or to something which p translates [iff] he believes that p,
which, unlike Kripkes principle, wont lead by contraposition to Pierre does not
believe that London is pretty. But, Brian realizes, his solution still leaves us with
another of Kripkes puzzles, viz., that our ordinary principles of belief ascription will
commit us both to
(1) Pierre believes that London is pretty
and to
(2) Pierre believes that London is not pretty,
notwithstanding that Pierre is a leading philosopher and logician who would never
let contradictory beliefs pass. But, Brian asks, does this show there is something
wrong? To which he answers, only if you assume it is the function of de dicto
ascriptions to capture how believers conceive things (p. 89). This brings us to Brians
real interest in Pierre. That interest is to discuss what Brian perceived as the
disconnect between the way names function in the that-clauses of our belief reports
and the way they function in thought.
As for the way names function in belief reports, Brians view is apparently the one
he ascribes to a character he calls the impure Millian. The impure Millian may say
6
Kripke (1979).
that that-clauses have two functions: they capture the singular proposition that
would be asserted, and they capture (or make a stab at capturing) directly or by proxy
how the belief would be expressed (p. 93). This brings us to the function of names in
thought:
Let us call the consistency in belief that intuitively we nd in Pierre his psychological
consistency. The joint acceptability of (1) and (2) then implies this: either
(a) psychological consistency is consistency in a kind of contentcall it psychological
contentand beliefs de dicto ascriptions do not (always) capture their psychological
content; or
(b) psychological consistency is not as such consistency in a kind of content; perhaps it is a
kind of syntactic consistency. (p. 93)
Brian says there is good reason to opt for (a), which is the view of psychological
content presupposed in Chapter 3, but he puts that view aside in order to consider
the view he is now inclined to favor, a view that resonates with the two-component
view of psychological content already mentioned in the preceding two chapters. This
is the view that psychological content might consist in conceptual role and not truth
conditions: Londres and London have different conceptual roles for Pierre, and it is
relative to them that Pierres beliefs are consistent (p. 94).
Chapter 6 Truth beyond All Verication (1987). In one sense of the philosoph-
ical term of art realism, realism is the view that the truth or falsity of a statement is
independent of our ability to verify or falsify it, and that, consequently, its possible
for a statement to be true or false even though it isnt possible for us either to verify or
to falsify it. Michael Dummett has famously argued that there is a problem as to how
we could understand statements if their truth or falsity transcended our ability to
verify or falsify them. The problem, as Dummett sees it, is that a persons capacity to
understand statements must be manifestable in her behavior, but that its very
unclear how a capacity to understand statements whose truth conditions transcend
their veriability conditions could be manifested in a persons behavior. Brian agrees
with Dummett that if a realist understanding of statements were not manifestable in
behavior, there would be something wrong with realism (p. 98). But Brian thinks
that a grasp of verication-independent truth conditions can be manifested in
behavior, and that is what this chapter aims to show. Dummett assumes that realism
requires a language to have a truth-conditional theory of meaning, which is to say, a
theory that compositionally determines the truth conditions of its statements on the
basis of the referential semantic values of its words as determined by the way speakers
of the language use those words. Brians strategy for undermining Dummetts
anti-realism appeals to the sort of two-component theory of meaning sketched in
the preceding chapter, but here his strategy is to show that realism can be supported
by the conceptual-role component of a two-component theory of meaning, where the
second, truth-theoretic, component need be nothing more than a disquotational
theory of truth according to which saying, e.g., that snow is white is true is simply
a metalinguistic way of saying that snow is white. The idea, developed by Brian in
some detail, is that (a) realism about the natural world is simply a consequence of
our theory of nature, and (b) the conceptual role or quasi-holistic theory of under-
standing explains perfectly well how such a realist conception of nature is possible
(p. 110). As for the requirement that our realist understanding of verication-
independent states of affairs be manifestable in behavior, Brian claims that we get
to satisfy that requirement for the simple reason that the relevant conceptual roles
are manifestable in our verbal behavior (p. 115). It is unclear what Dummett means
by his demand that our understanding of a notion be manifestable in our behavior,
and perhaps Brians deferential use of manifestable inherits that unclarity. But its
(a) and (b) that are doing the heavy lifting in Brians defense of our realist conception
of nature, and those two points strike me as being in excellent shape. As for (b), the
conceptual roles of our beliefs are obviously consistent with our scientic beliefs, so
the crucial point here must be (a). But isnt Brian obviously correct to take (a) to be
obviously correct? For its indeed obvious that the veriability of statements about
the physical world is dependent on natural contingencies in such a way that it is a
natural or scientic possibility that [those statements] be true even if not veriable
(p. 110). Just think of the all the contingent truths about how our visual systems work
in order for us to verify by sight that there is a tree in the quad.
Chapter 7 The Supervenience of Social Meaning on Speakers Meaning (2001).
Here we nd Brian evaluating with the benet of both hindsight and his distinction
between social content and psychological content the Gricean thesis that the
literal meaning of sentences in a social or public language, as used in communication
by a population of speakers, is derived from regularities in those speakers Gricean
communicative intentions in using that languagetogether perhaps with other non-
social psychological facts about individual speakers. Brian calls this a moderate
Gricean thesis, and he says that the key to it is that it is individualist about the basis
of social meaning (p. 125). By this he means that the contents of the intentions and
beliefs on which meaning is supposed to supervene are not even partly determined by
the meanings of ones expressions in ones public language of communication. Brian
points out that as originally conceived the Gricean project aimed at giving reductive
explications of speaker-meaning and expression-meaning. Brian doesnt hold out
much hope for the program so conceived, but in this chapter he proposes for
consideration a more modest Gricean positionnamely, that there is a theoretically
important way of understanding the Gricean project that doesnt depend on giving
explication in the form of sets of necessary and sufcient conditions. The relaxed
Gricean position holds that if we knew all the communicative intentions and other
propositional attitudes of members of P, as well as their correlations with utterances
of sentences of L [= the language of P] within P, and if we had time enough and
computational power for ideal reection, we could then directly infer a priori that L
is the language of P (p. 127). Thus, the question confronting this pared-down
Gricean position is: Does literal meaning asymmetrically supervene on speakers
intentions, and other psychological properties, including causal relations to external

factors that do not presuppose social-semantic factors? The answer, which Brian goes
on to develop in terms of his distinction between social content and psychological
content, is no, if speakers intentions and beliefs are individuated in a way that assigns
them truth conditions, but yes if they are individuated in terms of internally consti-
tuted properties. But then, as Brian goes on to demonstrate, the afrmative answer is
perhaps not terribly exciting (p. 131).
1
Reference and Propositional
Attitudes
HAVE I referred to Cynthia if I say that Herbert hopes that he will marry her?if, for
example, I say, Herbert hopes that he will marry Cynthia? It would appear so.
Frege nevertheless held that names and other singular terms do not have their
normal reference in such oblique contexts. Since Frege, weighty arguments have
been advanced and accepted, and the view is well entrenched. But it is a most
implausible claim, not easy to reconcile with certain obvious and ordinary facts.
Quine has come as close as anyone to recognizing this in print, but even he appears to
accept a variant of Freges theory. There is, though, in Quines qualied theory,1 a
foothold from which one might arrive at a more acceptable view. So I shall begin by
describing his theory.
I
Suppose that
(1) Ralph believes that a certain person is a spy.
This clearly supposes rather more than that Ralph believes that spies exist. It appears
to be a quantication into an oblique context, and hence by the usual conventions of
mixing quanticational notation and English it should be equivalent to
(2) (Ex) Ralph believes that x is a spy.
Quine points out that, despite its initial appeal, the signicance of this sentence is
doubtful. Here is a story he tells.
There is a certain man in a brown hat whom Ralph has glimpsed several times
under questionable circumstances on which we need not enter here; sufce it to
say that Ralph suspects he is a spy. Also there is a gray-haired man, vaguely
known to Ralph as rather a pillar of the community, whom Ralph is not aware of
having seen except once at the beach. Now Ralph does not know it, but the men
are one and the same.
Quine (1966), p.185.
1
Quine (1966).
REFERENCE AND PROPOSITIONAL ATTITUDES
Now let us suppose that Ralph, if required to give an opinion, would sincerely deny
that the man seen at the beach is a spy. We then have a case in which the man in the
brown hat is the man seen at the beach, and in which it is true that
(3) Ralph believes that the man in the brown hat is a spy
but in which it is false that
(4) Ralph believes that the man seen at the beach is a spy.
Quine draws from this failure of implication the conclusion that (3) does not assert a
relation between Ralph and some man. His assumption appears to be that if a relation
were thus truly asserted the same relation would be truly assertable between Ralph
and the man under any description. As a result, (2) is held by Quine to make no
sense, for it clearly requires that Ralph stand to someone in the putative relation.
Since the point about (3) is quite general, no relation is expressed by the context
(5) x believes that . . . y . . . .
If Quine is right, the man whose spyhood is in question does not receive reference in
(3) because of referential opacity. Thus Freges theory is upheld. To accommodate
this alleged fact, Quine proposes that we take (3) as asserting a relation between
Ralph and a sentence. So we have:
(6) B* (Ralph, the man in the brown hat is a spy).2
The irreferentiality of singular terms in belief contexts is thus to be explained by their
occurring in the formal mode.
But how are we then to construe the acceptable sentence (1)? Quines suggestion is
that, just as Ralph may stand in the required relation to a sentence, he may similarly
be related to an open sentence (predicate) and some item of which, as it were, he
believes the open sentence to be true. Thus it may be true that
(7) B (Ralph, x is a spy, the man in the brown hat).
The problematic term has here a normal referential occurrence, and together with the
identity (7) will imply
(8) B (Ralph, x is a spy, the man seen at the beach).
Both of these are to be taken as true in Quines story, though of course Ralph will not
know that his beliefs thus relate him to the man seen at the beach. (One may read (8)
as Ralph believes of the man seen at the beach that he is a spy.) Using this device we
may now represent the structure of (1), for we may generalize (7) to get:
(9) (Ey) B (Ralph, x is a spy, y).
2
The notation is not Quines. B* may be read believes true, and the subsequently introduced B may
be read believes true of.
This, then, is Quines version of (1).

An attractive feature of the analysis is that it enables us to represent concisely a
certain not uncommon way of being mistaken. In the case described, it may happen that
(10) B (Ralph, x is not a spy, the man seen at the beach).
The joint truth of (8) and (10) convicts Ralph, not of irrationality, but merely of
fallibility. For it does not follow that
(11) B (Ralph, x is a spy and x is not a spy, the man seen at the beach).
To make it plausible that this is a non sequitur and thus to defend both Ralph and Quine,
I suggest the following. If (8) does what we intuitively require, it will ascribe to Ralph that
state of mind he would be in if he were sincerely to assert, referring to the man in the
brown hat (who is the man seen at the beach), He is a spy. (10), in parallel fashion, will
ascribe to Ralph that state of mind required for sincerely asserting He is no spy, referring
to the same man. Even though Ralph has revised no judgments, he is not prepared
sincerely to assert of that man, He is a spy and yet no spy. This is not to suggest that belief
has a satisfying analysis in terms of sincere assertion, but merely that the connection
between the two makes it possible to point up features of this crucial notion belief of.
Before we proceed, there is a complication which needs to be dealt with. As we
have treated it so far, (4) is false in Quines story; but another way of reading it makes
it truenamely, as meaning something like Ralph believes of the man seen at the
beach that he is a spy. (We are at the beach and I say, Ralph thinks that that fellow is
a spy, proceeding with the story.) This is, in itself, no problem for Quine, since (8)
captures it. But (3) will then also have two readings, a strong interpretation (the one
we have given it so far) and a weak interpretation, on which it is rendered by (7). This
characterization (strong and weak) is given on the assumption that the former
implies the latter, but not conversely; that is, if Ralph believes that the man in the
brown hat is a spy, then Ralph believes of him that he is a spy. The result of this
assumption is that (3), on its strong interpretation, together with the identity, implies
(4) on its weak interpretation. The upshot is that there is a reading of (4) on which,
given the identity, it follows from both interpretations of (3). Although this requires
a qualication of the claim that substitution of the coreferential term in (3) fails
to preserve truth, the point remains that if the result of the substitutionnamely,
(4)is taken on its strong interpretation, truth is not preserved; indeed this failure to
be implied is the mark of the strong interpretation.
My assumption that the strong interpretation implies the weak has, in effect, been
maintained by Quine, for he holds with regard to a pair of sentences parallel to (3)
and (7) that the kind of exportation which leads from [3] to [7] should doubtless be
viewed in general as implicative.3 The device of considering would-be sincere
3
Quine (1966), p. 188.
assertions may make this clear. If (3) is true, Ralph is in that state in which he would
be if, referring to the man in the brown hat, he were sincerely to assert, The man in
the brown hat is a spy. In that case, he would be in a state appropriate to the sincere
assertion of He is a spy, referring to that man. The latter state, though, is what is
reported by (7).
The obtaining of this implication, however, leads to serious difculties with
Quines analysis, as follows. Recall that Quine holds that (3) (on our strong inter-
pretation) does not assert a relation between Ralph and some man; in fact, the
adequacy of (6) as a nal analysis depends on it. But this is hardly compatible with
(3)s implying (7), for the latter asserts some such relation. So, if the implication
holds, it would be natural to suppose that the man in the brown hat occurs with its
normal reference in (3), on its strong interpretation, since a statement of its truth
conditions would introduce, to speak roughly, the man in the brown hat, and not
merely the man in the brown hat.
But how is this possible? If the relevant singular term occurs with its normal refer-
ence on the strong interpretation of (3), the same should hold of (4). But then if (4),
on its strong interpretation, asserted a relation between Ralph and some man (in
consequence of the mans being referred to), it would seem that (4), on that
interpretation, should be true if that relation were truly asserted between Ralph
and that man under any other description of him. But that would be so only if (4),
on its strong interpretation, followed from (3) with the identity, and that, as we have
seen, fails to be so. So it would appear that certain natural assumptions lead us to a
contradiction.
I shall try to strengthen one of our incompatible assumptions, as a preliminary to
an attack on the other. Let us say that, on the weak interpretation of sentences like (3)
and (4), the relevant singular term occurs extensionally, and that, on the strong
interpretation, it occurs non-extensionally. Now there are independent reasons for
supposing that singular terms may occur in belief contexts non-extensionally, and yet
with their normal reference.
There is a tendency in some quarters to ignore pronouns and demonstratives when
considering the behavior of singular terms. This neglect, I think, is partially respon-
sible for Freges and Quines view of singular terms in belief contexts.
Suppose that we are at a costume ball, and I say, pointing to a certain man,
(12) Michael believes that that masked man is a diplomat.
Now in some circumstances it would be perfectly natural to take this on the strong
interpretation. That is, (12) may be asserted not merely on such grounds as that the
masked man is Ambassador Brown and that Michael believes that Ambassador
Brown is a diplomat. But even on this non-extensional occurrence of the term, it
seems plausible that I (and my words) referred to the masked man, and that, if you
like, I asserted a certain relation to hold between Michael and him. I offer the
following considerations in support of this.
(i) There seems, in this case, to be that need for specication from the context
which is characteristic of normally referential occurrences of demonstrative
phrases. Thus the speaker may be asked, To which masked man are you
referring? with the same appropriateness as if he had uttered, That masked
man is a diplomat. Notice that this question is totally inappropriate in at least
certain cases in which the term that masked man occurs in the formal mode.
Consider it as a response to: Jones wrote in his notebook the words That
masked man is a diplomat.
(ii) It seems not far-fetched that, in uttering this sentence, I asserted a certain
relation to hold between Michael and the masked man. For, in showing that
what I said is false, one might quite aptly point out that certain relations fail to
hold between the two, or that certain other relations do hold. Thus:
(a) Michael was not in his vicinity all evening. Perhaps you mean someone else.
(b) Michael was heard to say, pointing at the fellow, Hes obviously a gate-
crasher.
(c) The two were seen deep in conversation and the masked man, as we
know, speaks ungrammatically.
(d) The masked man is Congressman Smith, whom Michael would recog-
nize anywhere.
(iii) Consider the inference:
Whoever Michael believes is a diplomat is a diplomat. Michael believes that that
masked man is a diplomat. So, that masked man is a diplomat.
My intuitions have it that this is a perfectly acceptable inference. The most natural
way of construing its validity is to take it as having the form
xFx ! Gx; Fa; so; Ga:

Of course there may be more complex ways of representing its validity while denying
the relevant singular term in the minor premise a normal reference. But on the
obvious way of doing it the term has to be treated as having a normal reference.
(iv) The analysis of certain complex sentences would be simplied if we could

suppose such terms to have a normal reference. For example,
(13) Michael thinks that that masked man is a diplomat, but he obviously is not.
A very plausible thing to say about the role of the pronoun he is that it anaphorically
picks up the reference of that masked man. (Compare it with Jones wrote in his
notebook the words That masked man is a diplomat, but he obviously is not.)
These considerations do not prove the point, but they suggest, I think, that our
theories would be simpler and more realistic if we could allow a normal reference to
some singular terms which occur non-extensionally. But to do so requires the denial
of a principle which is so deeply entrenched that it would seem eccentric to deny it.
It is called by some Leibniz Law (the converse of one form of the identity of
indiscernibles), but I shall refer to it as the substitution principle. It may be
formulated like this:
If S0 arises from S by substitution of a singular term N0 for one or more occurrences of a
singular term N in S, and if N and N0 have the same reference in S and S0 , then S0 is true if and
only if S is true.
The application of the principle to Quines case is this. If, on the strong interpretation
of (4), the term the man seen at the beach occurred with its normal reference it
would have the same reference as the corresponding term in (3). Hence (4) should be
true if (3) is. But that is not so, hence the term does not occur with its normal
reference in (4).
I shall, in the next section, show how the substitution principle might be false, and
then, in the following section, how to construe non-extensional but referential
occurrences of singular terms in belief contexts.
II
Let us consider some general facts about the truth conditions of sentences. Utterances
of sentences containing genuine singular terms are true or false depending on
whether the items designated by the singular terms (on the occasion of utterance)
satisfy or fail to satisfy certain conditions associated with, or expressed by, the
sentence. Each such sentence, in other words, has satisfaction conditions such that
its utterance is true if, and only if, the appropriately ordered n-tuple of its references
satises those conditions. So, for example, an utterance of She shot the armadillo is
true if, and only if, the indicated female shot the indicated armadillo.
By a sentences frame I shall mean the result of deleting its singular terms, with
small letters inserted to mark the positions of deletion. So, the frame of She shot the
armadillo is x shot y. Now on the basis of the above fact about satisfaction
conditions, it is usually assumed that if two sentences have the same frame they
have the same satisfaction conditions. For normal well-behaved sentences this does
seem to be true: for example, for all sentences whose frame is x shot y. An utterance
of any such sentence will be true just in case the item designated by the term in the
x-position shot the item designated by the term in the y-position (shelving
irrelevant complexities introduced by the possible ambiguity of shot).
Now it is not difcult to see how this general assumption might be false and, most
importantly, false in cases which make the substitution principle false. That is, it is a
perfectly coherent possibility that there are pairs of sentences with the same frame
(where the respective singular terms are coreferential) but which do not have the
same satisfaction conditions. For suppose that singular terms in certain sentences
had a dual role of the following vaguely described kind. Not only do they designate,
or refer to, items which must satisfy the sentences satisfaction conditions if the
sentence is to be true, but they also contribute to the very conditions which must be
satised. In other words, still keeping in mind the distinction between the references
of an utterance and the satisfaction conditions of the sentence uttered, it may be the
case that there are sentences whose frames do not alone determine their satisfaction
conditions, but do so only in conjunction with what singular terms occur. It would, in
that case, be possible for two sentences with the same frame, and the same set of
references, to differ in satisfaction conditions and, hence, to differ in truth value.
Unfortunately there are few clear examples in English apart from those psycho-
logical contexts I wish ultimately to account for. But Quine, in an earlier paper,4
called attention to a type of sentence which will serve nicely as an example of my
generally described possibility. It is true that
(14) Giorgione was so called because of his size.
Despite the fact that Giorgione is clearly here referred to, truth is not preserved by the
substitution of another name of himfor example, Barbarellifor it is false that
(15) Barbarelli was so called because of his size.
The falsity of (15) is due to the fact that Barbarelli was not called Barbarelli because
of his size, though he was, of course, called Giorgione because of his size. This, at
least, is independent of how we refer to him. So the satisfaction conditions of (14) and
(15) do not depend uniquely on their frame (x was so called because of his size), but
also depend on what singular term occurs. The man Giorgione-Barbarelli satises the
conditions of (14) and fails to satisfy the conditions of (15). Hence the term
Giorgione in (14) is to be taken as making a twofold contribution to the sentence;
on the one hand, it has its expected role in locating the reference of the utterance and,
on the other hand, it contributes to the satisfaction conditions which the reference of
the utterance must satisfy if (14) is to be true.5 The substitution of coreferential
terms, then, may sometimes fail to preserve truth, for such substitution may give the
resulting sentence satisfaction conditions which differ from those of the original
sentence. Hence utterances of the two sentences may have different truth conditions;
in our examples, we may say that different properties are in the two cases predicated
of the same person, despite the identity of those sentences frames.
There is a Leibnizian principle salvageable from thisa principle which is fully
general, appropriately trivial, and which does not pretend to be an a priori constraint
on the behavior of singular terms. It is this.
4
Quine (1964), pp. 13940.
5
In so far as Giorgione makes this additional contribution to the satisfaction conditions of (14) it is to
be construed as having an autonymous occurrence. It is conceivable, however, that there should be peculiar
sentences in which the contribution which the singular term makes does not require it to be taken as having
an autonymous occurrence.
If the sequence x1, . . . , xi, . . . , xn satises the satisfaction conditions of a sentence S, and if xi = y,
then x1, . . . , y, . . . , xn satises the satisfaction conditions of S.
III
Recognizing the vague possibility that singular terms may make what I called a dual
contribution to the truth conditions of sentences does not of itself solve any
problems raised by singular terms in belief contexts. What is needed is some
specication of that additional contribution singular terms make to belief sentences
which accounts for the non-extensionality of their occurrence.
So let us look once more at (12). On the interpretation which gives the demon-
strative phrase an extensional occurrence, Michael and the masked man are asserted
to stand in a relation which we may represent as
(16) B (y, x is a diplomat, z).
Now I maintained that on the interpretation of (12) that gives the term a non-
extensional occurrence, Michael and the masked man are asserted to stand in a
certain relation. But this relation cannot be represented adequately by (16), since
something stronger is being assertedsomething whose truth depends upon
Michaels identifying the man as such and such and believing of him under that
description that he is a diplomat. I suggest that to believe such and such of
something under a certain description is to have a certain conjunctive belief with
regard to it. Hence, if (12) is true on the non-extensional interpretation, Michael
believes of a certain person both that he is that masked man (where this is meant to
capture those properties implied by the use of the phrase that masked man) and
that he is a diplomat. (12), then, asserts the following relation between Michael and
the masked man:
(17) B (y, x is that masked man and x is a diplomat, z).
And the full rendition of (12) is
(18) B (Michael, x is that masked man and x is a diplomat, that masked man).
The phrase that masked man makes two distinct contributions to the truth condi-
tions of (12), on its non-extensional interpretation. One of these is referential; the
other is a contribution to the satisfaction conditions of the whole or, more precisely, a
partial determination of what open sentence is the middle term of the triadic relation
asserted by (12). This twofold contribution is represented by the dual occurrence of
the phrase in (18).
That (18), in its fashion, is an adequate rendition of the non-extensional inter-
pretation of (12) seems plausible from the consideration that the two match each
other in truth conditions, falsity conditions, and truth-value-gap conditions. For if it
is true that Michael believes that that masked man is a diplomat (in the strong, or
non-extensional, sense) then, it would seem, it is true that Michael believes of that
masked man that he is that masked man and a diplomat; and conversely. The
equivalence between their falsity conditions and their truth-value-gap conditions is
similarly clear. If the reader does not believe in truth-value-gaps the equivalence
should for him be, if anything, plainer.
We can now see why (12), in its strong sense, fails to follow from, say,
(19) Michael believes that Ambassador Brown is a diplomat
together with the relevant identity. For this will have these two interpretations:
(20) B (Michael, x is a diplomat, Ambassador Brown)
and
(21) B (Michael, x is Ambassador Brown and x is a diplomat, Ambassador
Brown).
The substitution of coreferential terms which leads from (19) to (12) clearly may
change the satisfaction conditions of the sentence. For quite different relations may be
asserted in (12) and (19), as may be seen by comparing (18) with (20) and with (21).
There is a seemingly acceptable kind of statement of which Quines analysis fails to
make sense, but which our expanded version accommodates without difculty.
Consider:
(22) Ralph believes that a certain cabinet member is a spy.
This is not to be rendered as:
(23) B* (Ralph, Some cabinet member is a spy).
And it will often be taken to imply more than
(24) (Ey) (y is a cabinet member & B (Ralph, x is a spy, y)).
Ralph, we may suppose, believes it of the fellow under a certain description; that is,
(25) (Ey) (y is a cabinet member & B (Ralph, x is a cabinet member and x is a spy, y)).
The path to this result passed through the observation that the inference from (3) to
(7) appears to be valid. Can the validity of this inference now be represented? (3), on
a non-extensional interpretation, becomes
(26) B (Ralph, x is the man in the brown hat and x is a spy, the man in the
brown hat).
Does it follow that

(7) B (Ralph, x is a spy, the man in the brown hat)?
There is no generally valid inference from a quoted occurrence of a conjunction to

the result of substituting a quotation of one of its conjuncts. Moreover, it is extremely
implausible that there is a generally valid inference from x believes that p and
p entails q to x believes that q. Nevertheless, it does seem that in the specic case of
conjunction elimination (as it were) we do accept such inferences from beliefs to
beliefs. The only kind of proposed counterexample which I know of fails to be one.
It is this. Someone may believe a conjunction but also believe that the truth of one of
the conjuncts is a necessary condition of the truth of the other. But what this shows is
not that there is no valid inference from x believes that p and q to x believes that q,
but rather that there is no valid inference from x believes that p and q to x would
have believed that q even if he had not believed that p, which is obvious and
irrelevant. So I suggest that we take the validity of the inference from (26) to (7) as
imposing a condition on any adequate account of the relation expressed by B.
If the inference holds, (26) implies
(9) (Ey) B (Ralph, x is a spy, y).
That is, (3), even on its non-extensional interpretation, will imply that Ralph believes
that a certain person is a spy.6
6
In his paper, Quantifying In (Kaplan (1969)) David Kaplan claims that the inference which leads
from (3) to (7), and thereby to (9), is not generally valid. The counterexample which Kaplan offers is
roughly this:
Suppose that Ralph believes merely that there are such people as spies. He will, no doubt, think that there
is, among spies, one who has a certain unique featurefor example, being the smallest spy. It may then be
true that
(a) Ralph believes that the smallest spy is a spy.
But it will surely not follow that
(b) Ralph believes of a certain person that he is a spy.
That is, it will neither follow that
(c) B (Ralph, x is a spy, the smallest spy)
nor that
(d) (Ey) B (Ralph, x is a spy, y).
Now it is true that the inference is doubtful, but this merely shows that we must make some distinctions.
Let us take a more manageable example and consider its ambiguities.
(e) Ralph believes that the president of the Boardroom Sweepers Association is a spy.
There are at least three ways in which this sentence might be taken; they may be indicated roughly as
follows.
(e1) Ralph believes that whoever is president of the B.S.A. is a spy.
(e2) Ralph believes of the president of the B.S.A., under that description, that he is a spy.
(e3) Ralph believes of that person (the one I here refer to as the president of the B.S.A.) that he is a spy.
Kaplans counterexample (a) clearly needs to be interpreted in the way (e1) interprets (e). But that is not
the kind of non-extensional occurrence the singular term has in my example (12), or has on the most
natural interpretation of Quines (3). The distinction among (e1), (e2), and (e3) may be readily represented
in my adaptation of Quines theory. (The details are obvious. The interpretation of (e) which (e1) tries to get
at is represented as a relation between Ralph and a sentence, with the denite description having, perhaps,
a Russellian expansion within the quoted context.) [Footnote cont. on p. 24.]
The trouble with the peculiar sentence

(2) (Ex) (Ralph believes that x is a spy)
should now be clear. A term in the position of the variable may have more than a
referential function. Consequently, there will be no single relation asserted by all
sentences of the form y believes that x is a spy. One needs, or may need, more than
the reference of the term in the x-position to determine what is being asserted.
That a context may be quantied into merely means that the sentence which
contains the context entails an existential generalization over something referred to
by a term within that context. So, on the strong interpretation of (3)represented by
(26)there are two things which may be called its existential generalization with
regard to the term the man in the brown hatnamely, (9) and the stronger
(27) (Ey) B (Ralph, x is the man in the brown hat and x is a spy, y).
If what I have said about conjunction elimination is right, then, agreeably enough, (9)
follows from (27). That is, if there is something of which x believes such and such
under a certain description, then there is something of which x believes such and such.
IV
The account just given is subject to a serious, indeed crippling, objection, but only a
relatively slight modication is needed to meet it.
To give the logical form of a certain class of sentences is, at least partially, to
indicate how a recursively dened truth-predicate might apply to them (or, better
still, how a recursively dened satisfaction-relation might hold between them and
arbitrary n-tuples of references). For this to be possible, the language must have a
nite vocabulary or, at least, a basic vocabulary whose semantical contributions may
be determined by an effective procedure.7
This condition would appear to be violated on my account, at least if the account is
to accommodate belief sentences with indenitely many referential terms within the
[Footnote 6 cont.] Quines claim that (3) implies (7) is correct, on that non-extensional interpretation of
(3) which is parallel to the interpretation which (e2) gives to (e). In Kaplans example and in (e) on
interpretation (e1), the denite description does not really occur as a singular termthe utterer of the
whole has not thereby referred to something. Whereas in (3) (on its most natural reading) and in (e) on
interpretation (e2), the denite description occurs referentially, but non-extensionally.
I might add that Kaplans illuminating remarks on the causal and representational relations implied by
the notion belief of should be taken as relevant to the analysis of the relation B, and need not be
mentioned, as he thinks necessary, in giving the logical form of belief sentences.
7
That is, it is not necessary that the semantically primitive vocabulary be nite, provided that there is
an effective procedure for determining, e.g., the denotation of each such item of vocabulary. It should go
without saying that there must also be an effective procedure for specifying that vocabulary. This may
imply that, at a syntactical level, the vocabulary be constructed from a nite stock.
belief context. For the expression B, as used in my examples, expressed a triadic

relation. But a parallel analysis of
(28) William believes that Cynthia will marry Herbert
would presumably give us, on one reading,
(29) B (William, x will marry y, Cynthia, Herbert).
Here a new B is introduced, one which should express a tetradic relation. If, at each
step, we have to introduce a new ad hoc relation, there is no effective procedure for
determining the semantical contribution of the verb to believe. At least, my account
of such sentences fails to specify such a procedure.
To meet this difculty, I propose the simple maneuver of taking the belief of
relation to hold among persons, open sentences (that is, simple and complex
predicates), and sequences of items. Thus (7), for example, becomes
(7)0 B (Ralph, x is a spy, the man in the brown hat)
and (29) becomes
(29)0 B (William, x will marry y, Cynthia, Herbert).
Quantication remains easily representable. If William believes that Cynthia will
marry a certain person, then
(30) (Ez) B (William, x will marry y, Cynthia, z).
Of course, this is to be taken as implicitly generalizing over sequences, as well as over
people.
But what are we to do with those belief reports in which reference is made to no
particular object of belief ? The most obvious thing to say about them is that they
express relations between persons and sentences, but that leaves us with a dyadic
relation. Consider, for example,
(31) William believes that what goes up must come down.
If this is to be construed as
(32) B* (William, What goes up must come down)
we will then have on our hands both the triadic B and the dyadic B*. We might
stipulate that (31) expresses a relation among William, a sentence, and some favored
entity like the null sequence, but that would be a spurious economy. I am in favor of
keeping, at this stage, both the relational and the non-relational construals of belief,
since doing so captures a distinction which is already present in the notion. Some
belief reports do imply that the believer stands in certain relations to particular items,
and these relations are probably at least partially causal relations. Other belief reports
do not have such import, but seem to assert something of a rather more general
nature about the believer, as well as about his beliefs.
V
The simplicity of the analysis of belief contexts cannot, unfortunately, be exactly
mirrored in the analysis of other psychological contexts. Consider the sentence:
(33) Siegfried hopes that his father will be President.
This has an interpretation such that the occurrence of his father is non-extensional
and yet referential (as may be seen from the appropriateness of inserting a kind of
namely rider after his fatherfor example, you know, the politician I introduced
you to last night). On this non-extensional interpretation, (33) cannot adequately be
represented as
(34) H (Siegfried, x will be President, his father).
To parallel the representation of the corresponding belief sentence, (33) should
become
(35) H (Siegfried, x is my father and x will be President, his father).
There is an obvious but bad objection to (35) as a rendition of (33), and a slightly less
obvious but good objection. The bad one is this. It may appear that (35) implies that
Siegfried hopes that a certain person is his father, while (33) implies no such
insecurities on Siegfrieds part. But the fact is that (35) need not be supposed to
have this implication. For one may already believe, or know, that a certain predicate
is instanced in a certain thing, and hence not hope that it is. Nevertheless, one may
with complete consistency hope that it and another predicate are conjointly
instanced. This is one minor way in which hope differs from belief.
(35) is not, however, a good rendition of (33). Suppose that Siegfried is unsure
whether Sigmund is his father. If Sigmund becomes President, and subsequent
investigation shows that Siegfried is his son, Siegfried will be famous. Siegfried
knows this, and hopes for success. So, we have a case in which Siegfried hopes with
regard to Sigmund (who, we may now divulge, is his father) that he is his father and
will become President, and hence a case in which (35) is true. But it seems clear that
(33), in the sense intended, remains false. Siegfried lacks a cognitive attitude neces-
sary for the truth of (33). For that sentence implies, speaking roughly, that Siegfried
correctly identies a certain person as his father and hopes that he, under that
description, will become President. This we may represent by
(36) B (Siegfried, x is my father, his father) & H (Siegfried, x is my father and x
will be President, his father).
It may be that the rst conjunct is presupposed rather than asserted by (33) and, if
that is so, (36) may be modied according to ones favorite way of indicating
presuppositions. The difference in the analyses of hope and belief lies, at this level,
solely in the redundancy of the extra clause in the case of belief. It seems not at all
unlikely that in reporting certain noncognitive psychological states we advert to, or

presuppose, certain cognitive states.
VI
It has not been my intention to suggest that psychological states of the kind called
propositional are really to be construed as attitudes toward sentences or open
sentences. Indeed, employing this word may seriously interfere with a correct
understanding of how intermediate the position of the current analysis may turn
out to be.
To put the point very generally: in describing the truth conditions of complex
sentences in such a way as to subsume them under a truth-denition for the set of
sentences, or the language, to which they belong, it will be necessary at all levels of
description before the nal assignment of specic truth conditions to refer, metalin-
guistically, to some aspects of their sentential composition. Thus, Aram is Armenian
and Andrew is Scottish is true just in case Aram is Armenian is true and Andrew is
Scottish is true. Suppose that philosophers were at one time so misguided as to think
that this was as far as we could go in stating the truth conditions of conjunctions, and
hence formed the theory that conjunctions assert a relation between sentences (the
relation which holds between sentences when both are true). To carry the fantasy
further, one might imagine the objection to this theory that such sentences do not go
into French with a reference to English sentences. On this basis, the propositionalists
would maintain that conjunctions must be taken to assert a relation between
propositions. Both the sententialists and the propositionalists have here confused
the fact that a conjunction is true if and only if a certain relation holds between its
conjoined sentences or propositions with the non-fact that a conjunction asserts a
relation between sentences or propositions. Another example may be helpful. At a
certain level of semantic description it may be useful to say that Smith lacks
foresight is true if and only if Smith belongs to the extension of the predicate
lacks foresight. Thus the sentence is true if and only if Smith stands in a certain
relation to a certain linguistic entity. But this fact about the sentence does not imply
that its logical form is that of a dyadic relation; to predicate is not to assert a relation
between a thing and a property or a class or a predicate.
Now I am not suggesting, of course, that we can get any simple theory of belief
sentences which eliminates reference to linguistic entities in the way in which we can
go the whole hog with conjunction and predication. But a certain stage of analysis
may enable us to represent certain facts about the structure of a certain class of
sentences without thereby enabling us to represent all facts about their logical form.
So, though I have said that relational belief sentences assert triadic relations, one
might read this as a claim about belief sentences at a certain intermediate step in the
theory of their truth conditions.
Classical analyses of belief sentences may be thought of as proceeding in at least

three different directions from this stage. First of all, a modied kind of propositional
theory results from construing the relevant open sentences as designating intensions.8
This would preserve the triadic relational structure of such sentences, but here the
second relatum is an intension, and not an open sentence. Second, certain behaviorist
theories might be best construed as attempts to generate from the relation B and the
relevant open sentences complex dispositional relations between a believer and the
(sequence of) items to which his beliefs relate him. This would give belief sentences a
dyadic structure. Finally, there are those theories which suppose that beliefs are
relations between people and linguistic entities, and which thus take the present
level of description as the nal step in stating the logical form of such sentences.
The present point is that the referentiality of problematic singular terms in
belief sentences may be represented without our having a nished theory of their
logical form.
8
See Quines fanciful presentation of this possibility in Quantiers and Propositional Attitudes
(Quine 1966).
2
Two Theories of Meaning
The leading question in the general theory of meaning is what the form of a theory
of meaning for a particular language should be. What is the empirical status of
the semantical description of a language? How does it t into our other empirical
theories? What kind of apparatus is sufcient to do the job? There is no general
agreement about these matters.
The account which I am going to give locates semantical notions within the
general framework of propositional attitudes, and, hence, makes essential use of
intensional entities. But there is, these days, a not uncommon idea that empirical
semantics can be done within a wholly extensional framework, without intensional
entities. This is due, in large part, to Davidson, who has claimed1 that the
apparatus of (what we might call) extensional truth-condition semantics is suf-
cient for theories of meaning for natural languages. His suggestion may seem to
have the merits of a compromise: the attack led by Quine against intensions might
be granted success, without requiring the wholesale abandonment of intuitions
that there exist some kind of semantical facts. But it is my view that this com-
promise makes no sense. Semantics without intensions is Hamlet without the
Prince of Denmark.
The overall constraints are obscure. Part of what is at issue is whether certain
intuitive philosophical and pre-philosophical notions of meaning can be explicated,
or suitably replaced, in such a way that something of these notions is both preserved
and shown to have empirical application. So my arguments against Davidson, and for
my own account, will perforce rely on intuitions about these notions of meaning;
indeed, how could a theory which cut all ties with such intuitions properly be called a
theory of meaning?
I shall begin with some arguments against Davidsons theory, and then sketch a
quite different account. The upshot will be that semantics is part of propositional-
attitude psychology, and stands or falls with it. If propositional attitudes cannot be
accommodated in a scientic conception of reality then neither can semantics; but,
if they can, there is no need to cast about for anemic approximations to our
red-blooded intuitive semantical notions.
1
Davidson (1967).
TWO THEORIES OF MEANING
I
According to Davidson, an adequate theory of meaning for a particular language will
satisfy these conditions: rst, it will, in some appropriate sense, give the meaning of
each sentence of the language; second, it will show how the meaning of a sentence is a
function of its parts and structure; and, third, it will do these things in a testable way;
it will be suitably empirical. There is no arguing with these constraints; they are
partially denitive of the enterprise. That there is a further desideratum which
Davidson does not, and cannot, recognizenamely, that the semantical theory of a
particular language should be couched in such terms as to make it possible to relate
the theory to a broader psychological frameworkwill emerge.
The theory for a language L that satises these constraints, according to Davidson,
would be a Tarski-type truth theory for L, or rather a certain modication of a truth
theory that accommodates indexical sentences. A truth theory for L is a nite set of
conditions which, for each sentence S of L, implies an equivalence of the form S is
true iff . . . ; examples are Snow is white is true iff snow is white, and La neige est
blanche is true iff snow is white. If a truth theory for L implies only true equiva-
lences of that form, and does so in such a way as to exhibit how the truth conditions
of each sentence are a function of its parts and structure, then that is all we can
reasonably require of a theory of meaning for Lits implied equivalences will count
as giving the meaning of each sentence of L. Notice that it is not made an explicit
requirement for a truth theorys counting as a theory of meaning that the sentence on
the right-hand side of each such equivalence be a translation of the sentence
mentioned on the left. But the implication seems to be that if a truth theory for a
natural language meets the other conditions, the right-hand side will be at least an
approximation to a translation of the sentence mentioned on the left.
The initial appeal of Davidsons suggestion lies in the platitude that to give the
meaning of an indicative sentence is to give its truth conditions. The plausibility of
the platitude, however, may rest on an interpretation on which it is of no service here.
For to know the meaning of a sentence may indeed be to know under what
conditions it would be true in any possible state of affairs; it does not follow that
knowing Ss meaning is the same as knowing the material conditional S is true iff p.
Not that that is a point which anyone is likely to miss for long.
Perhaps the most striking and signicant feature of extensional truth-condition
semantics is its eschewing sentential meanings as entities. Davidsons avowed reason
for rejecting intensional entities is curious: my objection to meanings is . . . that they
have no demonstrated use.2 There are several issues here: rst he claims that
meanings, as entities, are not speciable independently of descriptions like The
meaning of Theaetetus ies. If that were so then meanings would indeed have
no use, since one could never informatively say what the meaning of Theaetetus ies
2
Davidson (1967), p. 307.
is. Second, he thinks that one cannot specify relations among such entities in such a
way as to be able informatively to say how the meaning of a sentence depends on
the meanings of its parts. And, of course, if expressions of the form The meaning
of . . . were the best we could do in specifying meanings, such generative relations
among meanings could not be informatively specied.
But the point is that meanings can be specied independently of such expressions
if, for example, you allow yourself enough possible world and set-theoretic apparatus.
And functions may be specied which map the meanings of parts on to the meanings
of wholes. In the time that has elapsed since Davidsons paper that has become
old hat.3
The real question about meanings is this: are they required for expressing gener-
alizations which are essential to the very nature of a semantic theory? If the answer to
that is yes, then the choice is simple: are we to allow that some semantic theory
or other might be true, or do we, out of physicalist asceticism, reject the possibility of
any such theory?
Let us get on to Davidsons positive account. Suppose one were to add to an
otherwise acceptable truth theory for English the equivalence Snow is white is true
iff grass is green. Would the resulting theory be a theory of meaning for English, and
would that equivalence therefore give the meaning of Snow is white on Davidsons
account? The answer is no: Davidsons claim is that such an equivalence gives the
meaning of the mentioned sentence only if it follows from a truth theory by virtue of
the more general contributions which its parts and structure make to the sentences in
which they occur. And it is difcult to see how there could be a truth theory for
English which satised that structural requirement (as I shall call it) and yet still
implied the equivalence. So the theory is not threatened by such simple wholesale
refutation.
On this point, Davidson says something rather puzzling. If a theory that satised
the requirement were to imply Snow is white is true iff grass is green, he would, he
claims, remain true-blue to his identication of meaning and extensional truth
conditions, for then there would not . . . be anything essential to the idea of meaning
that remained to be captured.4 This loyalty to theory is excessive. Knowing the truth
of that equivalence could not remotely be construed as understanding the sentence
Snow is white, and if a theory of meaning for L is not (partially) the theory of what
any speaker of L understands by its sentences, then it is hard to see what else it might
be. To urge the replacement of theories of meaning with truth theories, regardless
of the connection with understanding, would be to change the subject. And to
what avail?
The structural requirement does not eliminate all difculties of the kind that arise
from (as I would want to interpret it) trying to get intensional renements from
3 4
Cf. Lewis (1972). Davidson (1967), p. 312.
extensional apparatus. For an equivalent difculty arises at the level of satisfaction

conditions for lexical items. Probably, x is a camel has a nite extension. Hence,
armed with the complete catalogue of camels c1 . . . cn, we could provide extension-
ally adequate enumerative satisfaction conditions for x is a camel: it is satised by y
iff y = c1 or . . . or y = cn. This will have a systematically perverse effect on the truth
conditions of every sentence that contains camel. For each such sentence S, S is true
iff . . . would not intuitively give the meaning of S; for knowing the equivalence
would not be to understand S. Blithely accepting the result would again raise the
charge of changing the subject; as we shall see, there are essential connections
between the meaning of a sentence and certain propositional attitudes.
Of course if we were constructing a truth theory for English, we would not use the
enumeration, given our laziness and our ignorance. But that is beside the point; the
structural requirement (for a truth theory to be a theory of meaning) does not
eliminate enumerative satisfaction conditions. Apart from simply accepting the
resulta move whose disadvantage I have mentionedone might now try to impose
a further requirement on the appropriate class of truth theories.
One possibility is to try directly to eliminate from the acceptable truth theories
those troublesome admixtures of extralinguistic fact. Let us invent the cautious
semanticizer: he attends only to his knowledge of his language in constructing a
truth theory. So perhaps a truth theory for L is a theory of meaning for L only if it
would be constructed by the cautious semanticizer. (An implicit assumption of
something like this possibility may, for some, add to the plausibility of Davidsons
theory.) But, of course, the distinction between knowledge of language and know-
ledge of extralinguistic fact presupposes the very thing we are in search of: what it
is for a sentence of L to mean such and such. And Davidson, of course, must
reject the distinction, a fact that raises the further question: how can a theory that
fails to make such a distinction be a theory of meaning except by changing the
subject?
Another possible device for eliminating enumerative satisfaction conditions is to
require that for a truth theory to be a theory of meaning it should meet certain
standards of simplicity; and the condition x is a camel is satised by y iff y is a
camel is far simpler than the enumerative one. There are various possible difculties
here: the requirement may be too strong, and simplicity is a vague notion, but let
us waive them.
The account that is emerging, then, is this: S is true iff p gives the meaning of S in
L just in case it is implied by a truth theory for L which meets the structural and
simplicity requirements; any such truth theory can be counted a theory of meaning
for L. What could be more straightforward than that?
Unfortunately, these restrictions do not alchemize a truth theory into a theory of
meaning; to think they might is to fall prey to a simple confusion. To see this let us
experiment with Smith, who as yet knows nothing about L. Suppose that S is true iff p
follows from a truth theory for L which meets the requirements, and let us even
assume that this means that, by our intuitive notion of translation, the sentence in
the p-position is an acceptable translation of the sentence of L which is mentioned in

the S-position. Now let Smith be informed only that
(T) S is true iff p.
Does he thereby know what S means, or something approximating thereto? Of course
not, for although T, ex hypothesi, does follow from a suitable truth theory, that
information is not contained within T. All Smith now knows is that either S is true
and p or both S is not true and not-p. Virtually nothing about what S means follows
from that. The upshot is that, even when an equivalence of form T follows from a
suitable truth theory, it does not in itself give the meaning of S. That is too obvious.
What more does Smith have to know to know what S means in L? The preceding
discussion of requirements on a theory of meaning makes that clear. Smith has to
know at least that
(TT) some expression X denotes S, and X is true iff p is implied by a truth
theory for L which meets the structural and simplicity requirements and
which is true.
We might further require that Smith should know by virtue of which aspects of its
composition the equivalence follows from the truth theory.5
Now, TT is metametalinguistic with respect to L; it asserts a relation between the
sentence S of the object language and the sentence of the metalanguage which is in
the p-position. It would follow, then, that (1) to say what S means in L is to assert a
relation between S and a certain sentence in the language of some truth theory
for Lthat is, the sentence in the p-position in TT; and (2) a theory of meaning
for Li.e., a theory that implies, for each sentence S of L, something that says what
S meansis not simply a truth theory for L, but rather is metalinguistic with respect
to such a truth theory; it consists of the single assertion that a certain correct truth
theory for L meets the structural and simplicity requirements. This is certainly quite a
different account from what Davidsons seemed to be.
It is not too surprising that this should be the result. S is true iff p is far too
permissive to capture even approximately what S means. One advantage of entied
meanings is that they demarcate what a sentence means from everything else that is
true of it. In rejecting meanings, one creates the need for some other entity to do that
job: in Davidsons account, it is smuggled in surreptitiously as the sentence mentioned
in the p-position of TT, despite rst appearances of a de-entied theory of meaning.
An account in which TT is the basic form of meaning ascription is radically at
variance with the original Tarskian position that semantics treats of certain relations
between the linguistic and the extralinguistic; this seemed to be fundamental to
Davidsons suggestion. Let us remind ourselves why that is such an important
requirement for a semantic theory.
5
Davidson often implies that knowing this is part of knowing the meaning. This is not my view;
see p. 34.
First, if sentences stand in translational relations to other sentencesfor example,

the relation which Snow is white has to La neige est blanchethat would seem to
be so by virtue of properties that they have independently of the relation. The
meaning of a sentence in a particular population is some function of its psychological
and social role; its relations to other sentences in other populations are beside the
point. Whatever the relevant connections with the population of speakers are, they
are what constitute the semantics of a sentence.
Second, what is it that the monolingual speaker knows when he knows what a
certain sentence of his language means? Whatever it is, it is not in general that it
stands in a translational relation to some other sentence.
A second problem I nd in Davidsons theory concerns the basic, as it were,
pre-analytic notion of sentential meaningfulness that the theory is intended to
elucidate. A straightforward way of getting at the problem is to consider simple
signalsthat is, utterance-types that are conventionally meaningful in a group of
communicators but that do not have semantical structure. They show that meaning-
fulness is independent of membership in a language; most importantly, they have
their meaningfulness in common with sentences: a certain type of hand wave could
conventionally mean just what Here is what we have been looking for means.
The upshot of this is as follows. If you try to extend the notion of truth conditions
to simple signals, as if a simple signal belonged to a one-sentence language, you get
the grass is green problem with a vengeance. No consideration of structure or
simplicity could rule out arbitrary true equivalences S is true iff p as giving the
meaning of simple signal S. Consequently, if simple signals have meaning, there must
be some non-truth-conditional manner of ascribing their meanings to them. But,
since simple signals can mean the same as some sentences, that non-truth-condi-
tional style of ascribing meanings must apply to sentences as well, and so there has to
be more to a theory of meaning for L than a truth theory for L.
This argument raises the question as to what the connection between a sentences
meaning and its semantic structure is. Confusion is possible here. That the meaning
of a sentence is a function of its parts and structure does not entail that meaningful-
ness in general implies semantic structure. A way of viewing the matter is this. There
are certain communicational and expressive roles (which will be outlined later) to be
played by utterance-types: for any given role, either it may be directly assigned to a
certain utterance-type (a simple signal), or it may be assigned indirectly, by a system
of rules, to an utterance-type that is, therefore, semantically complex.
Confusion on this point may possibly be engendered by a certain ambiguity in the
notion of what it is to know a sentences meaning. A person may authoritatively be
informed as to what a certain sentence of an otherwise unknown language means,
without knowing any facts of its structurein that sense he knows its meaning, and that
is how I shall use the notion. Of course, one might want to say that such a person does
not fully know its meaning; but I would rather say what he does not know are further
facts about its semantical structure; facts as to how its language assigns it its meaning.
The third difculty with Davidsons theory is its inability to cope with certain
essential connections between the semantical properties of sentences and the prop-
ositional attitudes of language users. This has two distinct aspects.
First, the already discussed incapacity of a truth theory to discriminate certain
extensionally equivalent meanings is relevant in a new connection. Here is a fact:
(A) If a sincere, linguistically informed, attentive, sane English speaker who
wished to speak literally were to utter the sounds A zebra is a cordate,
then (generally) it would be the case that he believed that a zebra is a cordate.
This is not trivial, as may be seen by imagining (A) translated into Italian, preserving
the reference to the English sentence.
Now substitute for the one quoted occurrence of A zebra is a cordate in (A) a
quoted occurrence of A zebra is a renate. The resulting generalization will not be, as
it were, true to the same extent. It is slightly more likely that there will be counter-
examples to it than that there will be counterexamples to (A). The point is that these
two extensionally isomorphic sentences have different connections with propos-
itional attitudes, and the difference in the connection reects the intuitive difference
in their semantical properties. One might even say that it is because extensionally
equivalent propositional attitudes are not thereby identical propositional attitudes
that extensional semantical notions are not adequate for expressing all semantical
properties of sentences.
This becomes especially important when we consider the connections with prop-
ositional attitudes that are constitutive of a sentences being meaningful in a certain
groupconnections, as we shall see, with communicative intentions to produce
beliefs and actions in a hearer. Utterances of extensionally isomorphic sentences
may affect the beliefs of the hearer in quite distinct ways.
The second aspect of this problem is more general and more basic. Intensional
entities are needed to express generalizations that are essential to such questions as
what makes a particular language the language of a given population. As we shall see,
it is necessary to quantify over, as it were, the semantical content of sentences in
order to generalize their relation to the content of the propositional attitudes of
language users. As Davidson would be the rst to insist, the p position in an
equivalence of the form S is true iff p does not admit a genuine quanticational
variable. The point of extensional truth-condition semantics is to avoid postulating
meanings as entities; the effect of the austerity is to prevent semantics from being tied
correctly to the psychology of language use.
II
As you may anticipate, in the positive account of sentence meaning I am going to
offer, the semantical properties of sentences are a certain function of the propositional
attitudes of language users. It is a matter of some importance, therefore, whether
propositional attitudes are not best explicated as being certain relations to sentences
or other linguistic entities. If so, there would be implicitly a vicious circle in an expli-
cation of sentence meaning in terms of the propositional attitudes of language users.
There are only two such explications of propositional attitudes, as relations to
linguistic entities, that I know of which can be taken seriouslynamely, Carnaps,
and one which can be extrapolated from a theory of Davidsons, his analysis of
indirect discourse.
On Carnaps account,6 Jones believes that S asserts a relation between Jones and
the sentence S. But, as Church and others have pointed out, there must also be some
implicit reference to Ss meaning. How is that to be expressed except by making
propositional attitudes a relation between Jones, S, and a certain intensional entity:
(R(Jones, S, p))?7 But then the reference to the sentence becomes otiose, since the
form of words used is irrelevant to what is being asserted to be believed, once one has
got the proposition. Freges theory is much better: that S denotes the entity which is
in fact Ss meaning, without referring to S itself as having a certain meaning.
Davidsons theory of indirect discourse,8 extrapolated to propositional attitudes, is
much more subtle, and escapes, at least overtly, the objection to Carnap. According
to Davidson, an utterance of Galileo said that the Earth moves asserts a relation
between Galileo and a certain historical eventnamely, the utterance, by the speaker
of the whole, of the Earth moves as having a certain sense and reference. So its
logical form is S(Galileo, that). The relation thereby asserted is not, of course, the
direct discourse relation; it is this: x asserted something which makes him and the
utterer of y same-sayers. Same-saying is taken as unanalyzed.
The generalization, not made in that paper by Davidson, to other propositional
attitudes is obvious. The relation asserted by Galileo believed that the Earth moves,
between Galileo and the current utterance of the Earth moves, is: what-x-believes-is-the-
same-as-what-is-said-by-the-utterer-of-y, where that, again, is taken as unanalyzed.
Davidsons theory, thus extended, has great attractiveness. First, it eliminates, at
one stroke, those difculties about the logical form of propositional-attitude
sentences that arise from the variable number of referential positions in the that-
clause. Second, the analysis, at least ostensibly, precludes the need for supposing that
intensional entities are referred to in propositional-attitude assertions. For, one
might say, the utterer of the whole no more refers to the meaning of his utterance
of the Earth moves than he would have, had he just asserted the Earth moves.
In both cases he is producing a meaningful utterance; in neither case is he referring to
a meaning.
6
Carnap (1956), pp. 535.
7
It might seem that the implicit reference is to S as belonging to a certain language rather than as having
a certain meaning. That that wont do can be seen by considering cases in which S has more than one
meaning in its language, but where Jones believes that S is asserted, in its context, unambiguously.
8
Davidson (1969).
Now the main difculty I nd with Davidsons theoryor, rather, this extension of
itis that if my assertion of Galileo believed that the Earth moves is true, it is
impossible to see how this could be so unless something rather substantial was true of
Galileo quite independently of the existence of my utterance. The theory makes
beliefs irreducibly relations to the utterances of those who describe them. Arent there
actual beliefs that no one has ever uttered something equivalent to? If so, beliefs are
not in themselves relations to the utterances of their describers. How then can the
fundamental mode of ascribing beliefs be as related to utterances? Is there something
ineffable about their more intrinsic properties? Of course not. When I ascribe a belief
to someone, I am not asserting something which logically implies that I have spoken
and which therefore could not have been so had I kept quiet.
Here is a possible reply. The description of an objects length (it might be said)
essentially relates the measured object to some other objecta standard. But it does
not follow that had no such standard existed, the measured items would not have had
precise length. A meter-long object on Mars (the reply continues) could have been
that length had the standard meter bar never existed.
Now, the point is that what the last consideration really shows is that x is one
meter long is not to be analyzed as x is as long as the standard meter bar. Rather
what it means is x has that lengthi.e., the length that, as it happens, is the length of
the standard meter bar.9 The former analysis implies that had the bar not existed
nothing could have been one meter long. On the latter analysis, measurement asserts
a relation to a certain abstract entity.
The Davidson-style analysis of belief sentences cannot be conservatively repaired
by a similar move. If Galileo believed that the Earth moves is supposed to pick out
some particular independent fact about Galileo via helpful reference to my utterance,
then the analysis should be Galileo believed thati.e., that which happens to be
expressed by this utterance of mine. Now that asserts a relation between Galileo and
what is presumably an intensional entity.
The consequence is that, for both Carnaps analysis and the extension of Davidsons
analysis to propositional attitudes, suitable repairs introduce intensional entities, and
the reference to linguistic entities becomes superuous. Hence, a potential impedi-
ment to taking propositional-attitude notions as presupposed by semantical notions
is removed. What I want to show is that the theory of meaning is part of the theory of
mind, and not the other way around.
III
There is a tradition in the philosophy of language that would locate all facts about the
communicative intentions and beliefs of language users, and regularities concerning
9
This is essentially Saul Kripkes point; see Kripke (1972), p. 274.
them, in pragmatics and not semantics. Since the semantics of a language includes
facts about the meanings or senses of its terms and sentences (or, if you are an
extensionalist, about their extensions), it would follow that those semantical notions
are not to be construed as being about the communicative intentions and beliefs of
language users. So, the nature of the semantics-pragmatics distinction is no mere
terminological matter, but involves the question of the fundamental nature of
semantic concepts.
One enormously inuential view as to what semantics includes has been that of
Tarski, for whom semantical notions are denable entirely via abstract correlations
between terms and their denotata and via abstract Tarski-type truth denitions,
which employ only syntactical and logical notions together with a vocabulary
adequate for paraphrasing the object language.
There is an intensionalist counterpart to this formalist Tarskian conception of
semanticsone on which semantic concepts are denable entirely in terms of
abstract correlations between expressions and certain intensional entities.
Suppose, for example, that we were to dene a language as a function from
sentences to sentence-sized intensions (which we might identify with functions
from possible worlds to truth values). Would we, in specifying any such function,
thereby be dening a semantical notion? Is all there is to a certain sentences meaning
such and such that some such abstractly dened function maps it on to a certain
intension?
Verbal quibbling is a risk here, but some terminologies are more sensible than
others. What a sentence means is, of course, always relative to a language; and a
language may be identied with a certain abstractly dened function (although not,
in general, one which is quite as simple as the one just mentioned). But it does not
follow that all there is to semantical notions is captured by the notion of such abstract
correlations, with all facts about the psychology of their use being consigned to
pragmatics.
To see this, all that is needed is to see that such functions can be embodied in, or
abstractions from, facts of a kind that we would not at all be tempted to call linguistic
or semantical. As an example, suppose that on a certain planet in a certain possible
world cloud formations occurred in striking correlations with subnubilar phenomena
as follows. The set of physically possible cloud formations is denable by formation
rules over their spatial arrangement, by extrapolation from actually occurring cloud
formations.10 Most importantly, the correlation between the occurrence of cloud
formations and facts on the ground may be lawlikely generalized by employing a
function L from cloud formation types to sets of possible worlds, thus: if L(N) = the
intension I, then N occurs only if I is realized. So if I is the intension snow is white,
10
Perhaps the formation rules contain a transformational component, and cloud formations have
deep structures. Such a fantasy might dispel misconceptions about those over-romanticized entities, deep
structures.
and L(N) = I, then it is derivable from natural law, on that planet, that cloud
formation N occurs only if snow is white.
Is L in that context a language? Not until the dawning of nubisemantics.
Notions like language and means should always be thought of as intrinsically
relativized to a population of language users. The real semantical notions are not L is
a language, or S means M in L, but L is the language of population P and S means
M in the language of P. Clearly, those notions cannot be reduced formalistically to
logical and syntactical notions. Facts about the use of language in a population have
to be introducedand so psychological notions are needed in the analysis of
semantical concepts.11
The implausibility of formalist analyses of semantical concepts is more striking
if one considers non-indicative semantical features, and certain nonmodal features
like the second-person pronoun. That a sentence is imperative, or that you is a
second-person pronoun, are semantical matters. How is one adequately to dene
such notions except by reference to the standard use of those forms with certain
communicative intentions?
There is a distinction between semantics and pragmatics, and where the line gets
drawn is a hard question. Pragmatics is to be dened negatively, relative to the
denition of semantics; the pragmatics of the language of a population is all the facts
of a certain kind about language use in that population that are not semantical facts.
Some of these facts are general facts about the psychology of communication;
others are about the particular communicative practices of that population. The
key question, of course, is what belongs to semantics, and to that we should now turn.
IV
What facts about a sentence constitute its meaningthat is, its meaning in abstrac-
tion from any particular utterance, and in so far as it belongs to the language of a
particular population? If L is the language of P, then there are in P conventional
associations of utterance-types and types of communicative intention; L is a kind of
generalization of those associations. The language associates with the sentence a
range of possible communicative intentionsthose that a speaker may have in
uttering the sentence conventionally.
Before giving the account, I shall give the basic notions out of which it is
constructed; you can see that they are essentially psychological notions and logical
notions.
11
A certain abstractly dened function may be called a language, tout court, in the way in which a
certain abstract correlation of numbers with persons names may be called a telephone listing. They are
abstractions from communicational or electronic facts about a population. The substantial import of those
terms is a matter of what makes such an abstract entity a language of, or a telephone listing for, a given
population.
(1) There are various notions specically involving types of communicative

intentions:
(1a) Speakers or utterers meaning, introduced by Grice as the fundamental
notion in the theory of meaning.12 So, this account of mine is an essay in
what is often referred to as Grice's program. Both Grice13 and Schiffer14
have provided elaborate accounts of this basic notion; its core is the identi-
cation of a speakers meaning something with his intending, in a certain
way, to activate in his hearer a certain belief or action. The theoretical utility
of speakers meaning lies in its providing a concept of communication that
does not presuppose anything conventional or linguistic. Hence, linguistic
meaning can be explained in terms of conventions over such intentions.
The subsequent notions of referring and of illocutionary force are similarly
non-conventional in their denition.
(1b) Possibly a distinct notion of reference, to be dened in terms of a
speakers intention that his hearer should recognize, in certain ways, which
particulars are the subjects of his communication.
(1c) Notions of various illocutionary act-types, like requesting, asking, tell-
ing, objecting. Schiffer has provided an excellent general theory of their
denition, in terms of specications of concepts already present in the
denition of speakers meaning.15
(2) There is convention, to be dened as a regularity in action which is mutually
known to obtain, and which is conformed to because it is expected to be
conformed to. This is a simplication; for detailed accounts, see Lewis,
Convention,16 and Schiffer, Meaning, ch. 5.
(3) There are the notions of various types of abstract entity: property, relation,
and proposition, taken primitively; or possible worlds and sets, from which
various intensional notions can be dened.
(4) Possibly an epistemic notion of presupposition, which might be needed in the
assignment of a certain level of semantical properties.
A sharp distinction is to be made between what a speakers communicative inten-
tions are on a given utterancewhat he meansand what the overall semantical
properties of a sentence are in so far as it belongs to the language of a population
what it means. Facts of the latter kind are a function of regularities involving facts
of the former kind. A certain inconvenience arises from using means in two such
distinct senses, but, I think, the advantage of continuity with previous philosophical
usage overrides that inconvenience.
12 13 14
Grice (1957). Grice (1969). Schiffer (1972), chs. 13.
15 16
Schiffer (1972), ch. 4. Lewis (1969).
In the preceding section, I mentioned as a kind of language a function from

sentences to intensions.17 One might use such a function in the description of
the communicative practices of a group of speakers who utter only indicative, non-
indexical, unambiguous sentences: a speaker utters a sentence S of that language
with its conventional force only if he does so with the intentions that constitute
his meaning thereby that p, where p is the intension which that languagethat
functionassigns to S. So the intension, or sentential meaning, that such a language
assigns to a sentence in its way determines (given the kind of convention to be dened
shortly) what communicative intentions its utterer may have if the utterance is
conventional or literal.
Similarly, a language that has indexical, non-indicative, and ambiguous sentences
any natural languagewill be identied with a function that assigns to each sentence a
set (because of ambiguity) of certain entitiessentential meaningsthat are rather
more complex than just sentence-sized intensions, and which determine what range of
communicative intentions a literal or conventional utterer of the sentence may have.
The denition of what makes a language, however complex, the language of a
particular population will take this form:
L is the language of P iff
there exists in P the convention that, if L(S) = {M1 . . . Mn}, then one utters S, in
certain serious circumstances to be explicated, only if one's intentions in uttering S
are within the range determined by one of M1 . . . Mn.
Before specifying the relevant serious circumstances involved in such a convention,
and thereby completing the denition in that respect, let us rst see what kinds of
entities are to count as sentential meanings.
Consider a language with just indexical indicative sentences. A rst shot at saying
what its sentential meanings are like is this: the meaning of such a sentence is
something that determines what proposition a conventional speaker means given
the referential parameters of that particular utterance. I say referential parameters
here so as to avoid raising questions about whether indexicalspronouns, demon-
stratives, and denite descriptionsshould be regarded as introducing uniqueness
conditions or the actual referents themselves into the communicative content of an
utterance of an indexical sentence. So it might seem that the meaning of an indexical
sentence should be viewed as a function from the referential parameters of an
utterance to the proposition meant on that utterance. Now that that is not all that
is involved in the meaning of such a sentence may be seen by comparing She took
some cheese and He took some cheese. The functions that map arbitrary referential
parameters of utterances of these sentences on to the propositions that may be
17
The terminology may be confusing. Intensions are certain abstract entities; intention is being used,
not in the Brentano sense, but as derived from the ordinary usage of intends.
conventionally meant thereby are one and the same. Those sentences differ, rather, in
what their pronouns connote; and that is a matter of how the referent is conven-
tionally determined, and not of how, given the referent, the proposition that is
thereby meant is determined.
So we might represent the meaning of an indexical indicative sentence in two
stages: rst, there is the content function, which maps an utterances referential
parameters on to the proposition that is thereby meant; and, second, there is the
ordered n-tuple of what might be called the sentences referential qualiersroughly,
what is connoted by the sentences referring expressions. Referential qualiers are
sometimes themselves complex; if you represent, at their different levels of embed-
ding, the various properties and relations connoted by the referring expressions in
the cat which chased the mouse, which ate the cheese you get a complex entity. But
this is not the place for the details.
So, then, the communicative intentions of an utterer of such a sentence lie within
the range determined by one of its meaningsthe utterance ts that meaning
provided that (a) the proposition that he means is the one determined by the relevant
content function together with the referential parameters determined by his refer-
ential intentions, and (b) his referential intentionsthat is, roughly, his intentions
that his hearer recognize which particular items or uniqueness conditions gure in
the content of what he meansare in accordance with the referential qualiers.18
Non-indicative sentences may be accommodated by adding a third stage to
the specication of a sentences meaning. Some sentences are simply conventional
devices for meaning that such and such, and others for meaning that the hearer is
to do such and such. They are the simple indicatives and imperatives, and the
more specic illocutionary forces of their particular utterances are determined by
intentions that are to be recognized from generally non-conventional, or at least non-
linguistic, features of the context of utterance. But some sentences are conventional
devices for performing quite specic ranges of illocutionary actsthat is, for making
known those intentions that constitute an utterance an illocutionary act of a par-
ticular kind.19 Here we have interrogatives, sentences that begin I hereby . . . where
the verb is illocutionary, and so on.
The third element in the meaning of a sentence, then, is the type of illocutionary
act conventionally performed on its utterance or just, simply, the relation x means
that p, or x means that H is to do A. The other two elements remain as before, the
content function being now the general determinant, with the referential parameters,
of the content of the illocutionary act or of what is meant.
18
This does not imply that x counts as a referent of a conventional utterance of S only if the relevant
referential qualier of S is true of x. (See Donnellan (1966).) Referential intentions accord with a referential
qualier F provided that, roughly, the speaker intends the hearer to recognize x as his referent by virtue of
his and his hearers mutual belief that x is the F which is such and such.
19
For the general theory of illocutionary acts, and their denition in terms of kinds of intention, see
Schiffer (1972), ch. 4.
So a sentences meaning may be represented, in an abstract way, as an ordered

triple consisting of (a) something which determines whether the sentence is simple
indicative, simple imperative, or has a conventionally more specic illocutionary
force: this element can just be the appropriate act-type; (b) the content function; and
(c) the ordered n-tuple of the sentences referential qualiers, which may themselves
be complex embeddings of referential qualiers.
A particular utterance of a sentence ts one of its meanings provided that the
speakers referential intentions are appropriate to the referential qualier, the illo-
cutionary force is in the right range, and the content of the illocutionary act is the
value of the content function for the referential parameters of that utterance.
The question now arises as to what the overall convention is that makes L the
language of P. The most straightforward candidateone that would certainly establish
a sufciently strong correlation between sentence-types and kinds of intentionsis:
C: the convention that, if L(S) = {M1 . . . Mn}, then any member of P utters S only if
his utterance ts one of M1 . . . Mn.
But that convention is too strong. There are many circumstances in which sentences
are uttered without being accompanied by intentions of a kind conventionally
associated with them, and no confusion, obscurity, or frustration of the purposes
of language attends such utterances. Examples are: creative metaphors, hyperbole,
telling jokes, acting, testing microphones and typewriters, and, in general, utterances
such that the speaker knows that his hearer will not take his utterance literally.
One can imagine such non-literal utterances of sentences of L being very wide-
spread, without their existence interfering in the least with Ls being the language of P.
But a convention like C would not then obtain, since the appropriate underlying
regularity would not exist; hence C is not the right convention.
What, then, is the right correlation between utterances of sentences and the
communicative intentions of their utterers? It would seem to be something like:
when a speaker utters S in serious circumstances, his communicative intentions t
one of the meanings in L(S). But that, on the face of it, introduces a circularity, for
what are serious circumstances except those in which the speaker is going to be
taken literallythat is, in accordance with what his sentence means in P?
Compare these cases:
(a) Seeing a lachrymose acquaintance approaching, Jones says to Smith, Its about
to rain. Smith correctly takes Jones to mean, not that it is about to rain, but
that they are in for some weeping.
(b) The sky is solidly blue; Jones says Its about to rain. Smith does not infer that
Jones means that it is about to rain, but he is at a loss as to why Jones uttered
those words.
(c) The sky is cloudy, and Jones kicks a pebble. Smith does not infer that he means
thereby that its about to rain.
The rst two cases differ in a certain striking respect from the third. In them there
is no lack of a generally appropriate association between the words uttered and the
intentions that would be involved in meaning that its about to rainthere is an
association founded in past communicative practice. Rather, the hearer is in a
position to judge that it is improbable that the speaker means that it is about to
rain, in the rst case, because there is some other much better explanation of his
uttering those words, and, in the second case, because, given the circumstances, the
speaker would not want to be taken to mean that its about to rain by any utterance of
his. This would normally be mutually known by speaker and hearer.
In the third case, however, the improbability that Jones means, by kicking the
pebble, that it is about to rain rests on the lack of an appropriate association between
that action and that meaning. We might say, then, that in case (c) there is an
associative improbability that x means such and such by his utterance, whereas in
cases (a) and (b) there is a non-associative improbability that x means such and such
by his utterance.
If there is in circumstances C no non-associative improbability (excuse the nega-
tives, please; they are not cancellable) that xs communicative intentions in uttering
S are within the range of M, then, I shall say, xs utterance of S is free, in C, for M-ing.
Obviously, ones utterance may be free for M-ing, and yet there may still be an
associative improbability that one is thereby M-ing. So the convention employed in
the following is quite substantial, and sets up, non-trivially, that correlation that
exists between sentences of L and their literal meanings even when there exist
widespread non-literal utterances of sentences of L.

there is in P the convention that if L(S) = {M1 . . . Mn}, then any member of P utters
S in circumstances in which his utterance is free for Mi-ing only if his intentions in
uttering S are within the range determined by Mi.
The denition of sentence-meaning follows simply: S means Mi in P iff, for some L,

L(S) = {M1 . . . Mn} and L is the language of P.
The regularity that underlies the convention, then, is such that if x utters Its
about to rain when there exists, for his hearer, no non-associative improbability
that he means thereby that its about to rain, he has thereby meant that its
about to rain. This gives us an idea, at a rather simple level, of one central
factor in the inferences which speakers of a language make to each others comm-
unicative intentions.
It is interesting to see how the account accommodates sentences with more than
one meaning. If S means M and also M0 , then on a normal literal utterance of S the
speakers intentions conform to one of themsay Mand he can expect his hearer
to infer from the relevant contextual factors that that explanation of his utterance
overrides the alternative of his having uttered it with intentions conforming to M0 .
Hence, there is a non-associative improbability that he is M0 -ing, his utterance is not

then free for M0 -ing, and therefore the fact that he is not M0 -ing is not at variance
with the regularity.
V
A convention in P is a regularity in action among members of P that they mutually
know to obtain. Does that imply that, if L is the language of P, its members know the
rulesi.e., the syntax and semanticsof L? Is their knowledge of L under its
description as having such and such formation rules and mappings from sentences
to meanings?
There may be a Chomsky sense of knowledgehaving an internal representation
in which a speaker knows the rules of his language, but that is a psychological
hypothesis and, however reasonable it is, we do not want to build it into an
explication of what it is for L to be the language of P. Better that it should be offered,
at a later stage, as an explanation of how it is possible for a complex entity like English
to be the language of the population of English speakers.
David Lewis has suggested20 that the mutual knowledge of the relevant regularity
involves knowledge of all the sentences of L in sensu diviso and not in sensu
composito. Applied to my account, that is the difference between:
KD For each sentence S of L, if L(S) = M, then all members of P know (potentially
or implicitly) that, if anyone were to utter S in circumstances in which it is
free for M-ing, the utterer thereby would M.
KC All members of P know that for each sentence S of L if L(S) = M, then if
anyone, etc.
Adopting interpretation KD would certainly avoid having to attribute to speakers
knowledge of the rules of their language, but it has its own difculties. Do we really
want to say that KD holds for very complex sentences of English? Our interpretative
abilities with regard to sentences of our own language are quite limited; after so many
embeddings we fail to discern sense.
It is not wholly far-fetched to claim that the incomprehensibly complex English
sentences are not really part of our language at allafter all, we cannot understand
them! If English were thus reduced to a nite fragment of what it would be if our
brains were larger, there would be no problem for KD. But that is a wildly contro-
versial solutionand an avoidable one. The solution I prefer proceeds in two
stagesthe second of which generates notions that there are independent reasons
for wanting.
20
Lewis (1969), pp. 648, 1823.
The sentences of English we understand are enormous though nite in number.

We certainly have potential knowledge of a much greater number of sentences than
we have ever heard and, indeed, than have ever been uttered. Let us call the sentences
that a person understandsor could understand after some thoughtthe sentences
of his effective language. So now we can make the convention, and hence the
relevant mutual knowledge, relative to the effective language of the members of
the population,21 as follows.
L is the language of P only if
there is a large (enough) restriction L0 of L such that there is in P the convention
that if L0 (S) = {M1 . . . Mn}, then any member of P utters S in circumstances in
which his utterance is free for Mi-ing only if his intentions in uttering S are within
the range determined by Mi.
The knowledge of the regularity that underlies the convention has the form of KD
knowledge of the generalization in sensu diviso.
The preceding condition, though necessary, is obviously not sufcient; so we come
to the second stage of the solution. As it is stated, no language that arises from adding
arbitrary non-English sentences-cum-meanings to English is excluded by the condi-
tion. Call some such language English-plus. What, then, is the difference between
those sentences of English-plus, on the one hand, and those sentences of real English,
on the other, that are left when you subtract effective English?
The difference is that the incomprehensibly complex sentences of English have parts
and structural features that are realized in effective English as making certain contri-
butions to the meanings of its sentences. English is a projection from effective English.
But what selects the projectible pairings of sentential features and meaning contribu-
tions? Might there not be more than one simple way of extending effective English?
It is, I think, part of our pre-philosophical conception of language that speakers learn
their language and, much more importantly, maintain their linguistic competence, by
virtue of some kind of learned associations or other between sentential features (words
and types of construction) and meaning-constituents. This is not to retreat from my
earlier denial that knowledge of the rules or grammar is to be brought into the explication
of what it is for L to be the language of P. Something weaker is required: let us say that L is
grounded in P, with regard to its restriction L0 , just in case those correlations of sentential
features and meaning contributions that gure in the correct psychological explan-
ation22 of the continuing mastery of L0 (i.e., effective L) by members of P will generate,
when extended, the full language L, including its incomprehensibly complex sentences.
21
This, of course, will vary from group to group. One should think of the set of English speakers
dividing up into many sub-populations, each with their own language. English may be identied with the
intersection or union (depending on what we want such an entity for) of these various languages.
22
A system of internal representation, or of Chomsky-type knowledge of rules, would be one possible
such explanation.
This allows us to eliminate English-plus, and to keep real English in its innite
complexity as the language of English speakers, thus:
there is a large (enough) restriction L0 of L such that there is in P the convention
that if L0 (S) = {M1 . . . Mn} then any member of P utters S in circumstances in which
his utterance is free for Mi-ing only if his intentions in uttering S are within the
range determined by Mi, and L is grounded in P with regard to L0 .
David Lewis, supporting Quines position on the inscrutability of reference, has
denied that the language of a given population can be assigned a unique grammar,
except relative to some more or less arbitrarily chosen method of evaluating grammars.23
Now a grammar is just the system of correlations of sentential features and constitu-
ents of meaning that gure in the denition of groundedness.
It is, indeed, always possible that one and the same language may be abstractly
carved up in several equally simple waysinto different lexicons and compensatingly
different interpretations of sentential structures. But once we have decided on a
scheme of attributing propositional attitudes to a population24and that is presup-
posed, in Lewiss theory and in mine, in assigning a language to a populationthen
there is a non-arbitrary way of selecting one grammar. The grammar of the language
of a population is just the one that grounds that language in that population.
From this we get an account of word meaning: a words meaning in P is what gets
assigned to it by the lexicon of the grammar of P together with whatever general
type of contribution words of that kind make to the meaning of the sentences in
which they occur. Hence there is no need to postulate distinct conventions that
make words meaningful; the conventionality of word meaning is derivative from
the overall convention that makes L the language of P. What is the empirical status of
a semantic theory? The place of the theory of meaning among the rest of our
empirical theories can be represented in two stages.
First, there is the attribution of communicative intentions to a speaker on a particular
occasion. The question of its empirical basis is, of course, an instance of the broader
question concerning the empirical basis of propositional-attitude attribution in general.
Second, as the denition shows, the ascription of L to a given population is a kind
of generalization of facts about what speakers mean on particular occasions, taken
together with the assumption of potential mutual knowledge of the generalization in
sensu diviso, and the inuence of this knowledge on conformity. So problems about
the empirical status of semantics are likely to be instances of more general problems
about the status of theories of the mental life of individuals and groups.
23
Lewis (1969), pp. 197ff.
24
If there is an indeterminacy in the ascription of semantical notions, then it is derivative from an
indeterminacy, with respect to behavior and all physical fact and theory, of propositional-attitude
attribution.
3
The Semantics of Singular Terms
I
The sentences of a language are correlated in striking ways with certain propositional
attitudes of the speakers of the language. For there are regularities that associate
the syntactical features of sentencestheir structures and constituentswith aspects
of the communicative intentions of their conventional utterers. By communicative
intentions I mean certain intentions to produce beliefs or actions in ones hearer,
which are expected or intended to be out in the open.1 These regularities create
implicit expectations in hearers concerning the range of intentions with which any
given sentence of the language will be uttered. Speakers exploit these expectations
when they want to communicate, and consequently they conform to the regularity
because they are expected to. Now, a regularity in the actions of a group, which it
is mutually known obtains because members expect each other to conform, is a
convention.2 So there are conventions that correlate sentences with types of commu-
nicative intentions, and that do so according to systematic correlations of the syntactic
features of the sentences with aspects of the propositional content, and other features,
of the intentions.3 The sentences of a language form a system of conventional devices
for making known communicative intentions.
What is it to give a semantical theory for a language? If we were systematically to
pair the sentences of a language with the types of communicative intentions with
which they are conventionally correlated, that would include the essentials. We
would in effect have recursively dened a function from the sentences of the language
to certain entities which, in their way, encapsulate the constraints on what commu-
nicative intentions a conventional speaker might have in uttering each sentence.
These entities might appropriately be counted as the meanings for the sentence-types
to which they are assigned by that functionthat is, by that language.
To give the semantics of subsentential expressions of a certain kind qua expres-
sions of the language (for example, denite descriptions, pronouns, or names) is
to specify their contribution to the meaning of the sentences in which they occur, in
1
See Grice (1957), Grice (1969), and Schiffer (1972), chs. 13.
2
For explications of convention along these lines see Schiffer (1972), ch. 5, and Lewis (1969), chs. 12.
3
For the form of such conventions, which must be specied in such a way as to allow for non-literal
utterances, see Loar (1976) [Chapter 2 in this volume].
THE SEMANTICS OF SINGULAR TERMS
the foregoing sense. So, the basic question is this: What kind of communicative
intentions would a conventional utterer of a sentence like The cat belongs to
Georgina have? What restrictions do those words impose on what the conventional
speaker means?4
II
My answer is essentially Freges: the function of a singular term is to introduce an
individual concept into what is meant or expressed on its particular uses. Now
singular terms on the whole do not, qua expressions of the language, express
individual concepts in abstraction from particular utterances. The concepts that
they do literally express are normally non-individuating class concepts, which we
might call their referential qualiers; so, for example, being an oak, and being female,
are the referential qualiers of that oak and she. The task for a Fregean theory is
then to say what constraints are imposed by the referential qualier of a singular term
on what individual concepts may be meant on its particular uses.
Probably what occurred to many philosophers when they asked how Freges
theory, or Russells theory of descriptions, accommodates such sentences as the
wardrobe is walnut was that the propositions which its utterances express, or which
its utterers mean, have the form the unique such and such wardrobe is walnut. But the
requirement that the individual concept should be a logical restriction of the refer-
ential qualier is too strong, as is shown by that broad and central class of conven-
tional uses of denite descriptions, demonstrative phrases, and personal pronouns
that Donnellan has called referential.5 For one may utter the F is G perfectly literally,
but in such a way that the object to which one thereby refers could happen not to be
F, the consequence of which seems to be that being the F is not essential to what one
meantto what one primarily intended to communicate. The referential qualier
aids somehow in determining what is essential, without itself being so. Hence the
Fregean theory which I shall elaborate is more permissive than the classical version.
There is a certain non-logical constraint that the referential qualier imposes on what
individual concepts may be meant on the conventional utterances of a singular term.
Semantical theory is part of the psychology of communication, and I shall be
emphasizing the question what singular terms contribute to what speakers intend to
communicate. But there are, of course, substantial classical considerations as well,
not ultimately independent of the psychology of communication, but expressed
in the seemingly autonomous concepts of philosophical logic. The Fregean theory
4
I use means in two senses: rst, with regard to the speakers communicative intentions on a given
utterance; and secondly, with regard to the semantical properties of a sentence, qua sentence of a language,
in abstraction from particular utterances.
5
Donnellan (1966).
permits a natural treatment of: (a) how two utterances can make the same predica-
tion of the same object and yet differ in meaning; (b) why failure of reference does
not as such destroy sense or understanding; (c) why the nal clarication of what is
intended to be conveyed naturally takes the form of an individuating description; (d)
how negative singular existentials can be true; and (e) why the normal referent of a
singular term is, at least sometimes, irrelevant to its contribution to the truth-
conditions of opaque contexts.
Description theories, of which the Fregean theory is one kind, have in recent years
been sharply attacked along two different lines.6 First it is claimed that on many
actual or possible uses of singular terms, especially of proper names, there are no
individual concepts that pick out the referent and are appropriately related to the
utterance. My answer to that will be to specify the appropriate relation, and to argue
that every normal utterance of a singular term has that relation to some individual
concept, and that, moreover, in the suitable sense of refers, an object is referred to
only if it satises that individual concept.
The second, perhaps more important, line of attack has been to provide a powerful
alternative picture of the logical form of referential utterancesthat is (to speak
roughly) of utterances where it is natural to say that there is a denite, identied,
individual about which the speaker intends to communicate something, as contrasted
with utterances in which the communicative intention is to say that whoever or
whatever uniquely satises a certain condition is such and such. It is held that to
specify the semantical contentthe possible world truth conditionsof a referential
utterance one must specify the referent itself, and that no corresponding individual
concept is essential thereto. Hence the semantical content of a referential utterance of
t is G is, in an abstract way, identied with the ordered pair G, x, where G is the
predicated property and x the referent to t. But on other uses of singular termsthe
ones, roughly, that Donnellan calls attributiveno particular referent is essential to
content, for one means that the F, whatever it may be, is G. In assessing the possible
world truth conditions of attributive uses, unlike referential uses, no individual of this
world is carried through other worlds, since the F has different exemplications in
different worlds.
This radical two-use theory (as I shall call it) is normally accompanied by some sort
of causal theory of reference; the content of what is said, believed, or intended is
supposedly often a matter of extra-mental causal connections. Now the two-use
theory is, at least in a notional and general way, compatible with the account of the
nature of semantical theory I sketched at the outset. For it might be that on a
referential use of a singular term the speaker has a sort of irreducibly de re intention
about the referent, where this may be a matter of the intentions being caused in a
certain way by the referent.
6
See Kripke (1972), and Donnellan (1972).
I should like now to smooth the way for the Fregean theory by noting (what seem
to me to be) serious problems for the two-use theoryproblems that, of course, are
not unrelated to those matters in which the Fregean theory is most helpful. First, it
would seem that there might just as well be two kinds of expression corresponding
to the two uses of denite descriptionsthe one formed with an iota-operator,
and the other a sort of demonstrative phrase. One might accept that it is a kind
of synchronic accident that we have one kind of expression with two such distinct
functions, if the problem were conned to denite descriptions. But personal
pronouns would have to be assigned these two uses as well, for they have a sort of
attributive use as well as a referential use. For example, upon our discovery of the
murdered Smith I might say He is insane, meaning that Smiths murderer,
whoever that may be, is insane. So I mean a generalized proposition here, unlike
the more referential use of that sentence on which I am supposed to mean
something of the form G, x. Now it is not easy to accept that on the more
generalizing utterances of He is such and such the he has a different logico-
semantical role than it has on its more identifying utterances. On my account,
using he and the F to generalize, and using them, as it were, to identify, will
turn out to be implementations of a single convention.
Secondly, take any normal referential utterance and imagine a case as like it as
possible except that the speaker and hearer are deceived about the existence of the
supposed referent. There need be no element of content that the hearer fails to
understand. But on the two-use theory the non-existence of the referent should make
all the difference. Moreover, the utterance may still have that feature of specic
identication of the subject, which, on the two-use theory, is to be explained by the
referents being part of the content. The mere intention to refer does not explain the
distinctiveness of the subject in that sort of case.
Thirdly, the radical two-use theory implies that a sufcient condition of
understanding a referential utterance of t is G is merely correctly identifying
the referent of t and the property expressed by G. But that is not sufcient.
Suppose that Smith and Jones are unaware that the man being interviewed on
television is someone they see on the train every morning and about whom, in
that latter role, they have just been talking. Smith says He is a stockbroker,
intending to refer to the man on television; Jones takes Smith to be referring to
the man on the train. Now Jones, as it happens, has correctly identied Smiths
referent, since the man on television is the man on the train; but he has failed to
understand Smiths utterance. It would seem that, as Frege held, some manner of
presentation of the referent is, even on referential uses, essential to what is being
communicated.
Fourthly, the difference between paradigm referential and paradigm attributive
utterances is more a matter of degree than the radical two-use theory can make sense
of. Consider this range of cases. Case (1): we see a large empty shoe; S says about its
owner, whoever that may be, Hes rather big. This seems not relevantly different
from the attributive use of He is insane. Case (2): we see large footprints on the
beach; same utterance. Case (3): we see a mound of sand, which we assume covers a
man; same utterance. Case (4): like the last case, except that a leg is protruding. Case
(5): we see the man directly; S says, pointing, Hes rather big. This would seem to be
a paradigm referential utterance. On the radical two-use theory, somewhere along the
line truth conditions in this range of cases ip over from being a generalized
proposition to being a propositional complex. But at what point is that motivated?
Of no adjacent pair of these cases is it plausible to suppose that their contents are of
such radically different kinds.
III
The objects of my beliefs and utterances are often individuated for me only by virtue
of relations in which they uniquely stand to me. So, in a sense, my beliefs about them
are beliefs about myself; hence understanding the nature of the individuating con-
ceptions I have of other objects requires some understanding of the nature of my
beliefs about myself.
Having a belief about oneself clearly does not require having a purely qualitative
individuating conception of oneself: one need only imagine the early stages of
amnesia to see that. Nor is my believing myself to be such and such simply a matter
of having a de re belief about myself. Suppose that Watson sees himself through some
sort of mirror device and thinks he is seeing a burglar. He believes of Watson that he
is a burglar, but he does not believe himself to be a burglar.
I suggest taking self-ascriptive belief as unanalyzed. If Cynthia believes that she has
the u, then what is true is:
B* (Cynthia, y has the u).
She is related by the self-ascriptive belief relation to that propositional function. If
Cynthia believes that the tree she is looking at is a sycamore, then
B* (Cynthia, the tree which y is looking at is a sycamore).
Similarly with speakers meaning. If Cynthia says That is a sycamore, she may mean
that the tree she is pointing at is a sycamorethat is
M* (Cynthia, the tree which y is pointing at is a sycamore).
So when I say that an individual concept is essential to what is meant I intend that to
cover not only complete or saturated fully qualitative individual concepts, but also
individual concepts with a gap in them, one-place functions in intension like:
whatever is uniquely R to y. Such an individual concept may be essential to what
Cynthia means in the sense that
M* (Cynthia, whatever is uniquely R to y is G).
IV
The crux of the contemporary argument against the Fregean theory is the claim that
for any individuating description that the speaker might offer of the referent, it could
turn out that the description is false of the referent, and hence not essential to the
content of what is meant. But that is difcult to accept, given a realistic account of
what a normal utterer is intending to communicate on referential utterances. For,
among all those descriptions that the speaker might offer as picking out his referent,
there is always at least one that is connected in a certain special way with his
communicative intentions.
Not surprisingly, it is the attributive use that gives the clue to that special
connection. If one utters the F is G attributively, then in response to any discovery
of a mistake about the identity or existence of the F one would ideally judge:
Nevertheless, what I meant was that the F, whatever it may be, is G. This is not to
say that one would continue to maintain that it is true that the F is G, but merely that
that was what one had meant. On the attributive utterance of Smiths murderer is
insane, if the speaker had happened to believe that Jones was the murderer, and then
discovered that he wasnt, it would still be appropriate for him to express what he had
intended to communicate by uttering the same sentence. What makes the utterance
attributive is precisely that that would be the appropriate response to any discovery of
a mistake about the identity or existence of Smiths murderer; one would not need to
re-express in other terms what had been meant (although one might no longer wish
to assert it).
On an intuitively referential utterance of the F is G, this, so to speak, attributive
attitude might be maintained, not toward the F, but toward some other description,
in the following sense. Had this other descriptionthe H, saybeen uttered in
attempting to express what the speaker had meant by the utterance of the F is G, the
speaker would have judged, in response to every discovery of a mistake about the
identity or existence of the H, Still, what I meant was that the H is G. Had he chosen
to say the H is G in expressing what he had wanted to communicate by saying the F
is G, then in response to such mistakes he would not have had to choose another
description to re-express what he had meant.
Suppose, for example, that Smiths murderer is insane is uttered in reference to
the defendant in a trial. When it is discovered that the defendant is not Smiths
murderer, the speaker may choose other words to re-express what he had meant. If
he chose, say, the description the person one can see in that direction, it may then
turn out that in response to every discovery of a mistakeas a result, e.g., of
hallucination, of mistaking a hat-stand for a human, of mistaking that person for
the defendantan appropriate thing for him to say would be Nevertheless, what
I meant was that the person one can see in that direction is insane.
No one should want to deny that this may be true in some cases. But I shall make
the substantial and controversial further claim that on every normal referential
utterance of t is G there is some description the H such that the speaker could have
expressed what he had meant by saying the H is G, and that that utterance would
have been attributive.7 The truth of this counterfactual would indicate a certain
feature of the speakers original communicative intentions; what he had essentially
meant was that the Hwhatever it isis G. This is the core of the thesis that it is
the conventional function of singular terms to introduce individual concepts into
what is meant.8
Here are some considerations in favor of this claim. First, one might make a small
thought experiment: suppose I were to discover about a certain referential utterance
of mine, which had seemed perfectly normal, that my intended referent did not exist.
Would I not, after reection and perhaps some further thought experiments, have a
quite good grasp of what would have veried what I had intended to communicate?
Even so, it might be said, this does not positively support the case, for of course one
would know about a referential utterance of t is G that, had reference been
successful, the referents being G would have veried the utterance. But the point is
stronger. One could specify what would have veried ones utterancewhat one had
primarily intended to communicateindependently of the notion the referent of my
utterance of t. The alternative is disconcerting; one would have to think that, on a
certain proportion of normal referential utterances, if one discovered that ones
intended referent did not exist, one would not be able to say what one had intended
to communicate. But normally one would have no difculty in saying what one had
meant, within the bounds imposed in general by the vagueness of our propositional
attitude ascription.
Secondly, consider how we achieve an intuitive grasp on the referential cases: in
them, the individual concept that is connoted by the denite description is not
essential to what is meant. But whence this grasp on what is not essential to what
is meant, except from some awareness of what is essential? A possible reply is that
the recognition that a certain individual concept was not essential to what was
meant derives merely from ones understanding that ones utterance was referen-
tial, and hence that no individual concept was essential. But this does not t ones
grasp of ones own ability to sort out the essential from the inessential. The
characteristic experience after reection is: This is not what I meantfor what
I meant was that. It seems that our ability to judge that the referential qualier
was not essential to what was meant stems from being aware of something that
was essential.
7
This is inspired by an account that Christopher Peacocke gives of the connection between names and
denite descriptions, in the context of a rather different kind of overall theory, in an unpublished paper.
A related point has been made by Thomas Ricketts in an unpublished paper.
8
This is not to say that a singular term that is uttered referentially is used somehow inappropriately or
non-literally, as will become clearer in the nal section.
Thirdly, consider the following exchange:

(A) That tree in front of us is F.
(B) That is no tree; it is part of a stage set.
(A) What I meant was that the tree-like object there in front of us is F.
(B) Theres a mirror there; there is no tree-like object in front of us.
(A) What I meant was that the tree-like object which is visible in that direction is F.
(B) I have been deceiving you; you are hallucinating.
Several observations are in order here.

First, the plausibility of As choice of a new description at any step is somewhat
dependent on the predicate F. For example, if the predicate in the above case was is a
two-thousand-year-old redwood, his abandonment of tree as essential to what he
had meant would be far-fetched. This in general puts severe limitations on the
possible lengths of plausible sequences of the above kind.
Secondly, it is difcult to see how such sequences could go on indenitely without
the speakers substitutions appearing, fairly early on, to be far-fetched and insincere.
So I assume that such sequences must terminate in the sense that the speaker, after
ideal reection, can offer no further substitution.
Thirdly, here is the main point. Suppose that in the above case the sequence
terminates for A with As last mentioned reply. It would seem that A is then in one of
two positionsnamely, either (i) he says after Bs last remark something like:
Nevertheless, what I meant was that the tree-like object which is visible in that
direction is F which is what my account says should happen eventually, or (ii) he is
in a state which might best be expressed by his saying: Then I do not know what
I meanta state of confusion in which normal speakers would not nd themselves
on normal referential utterances.
The individual concepts that are essential to what we mean are often complex, and
implicit. That they are complex may be seen from the fact that in many cases one may
have the attributive attitude toward more than one individual concept. If, for example,
someone addresses us, and I then say He looked familiar, I may not be prepared then
to reject either the person who just appeared to speak to us, or the person we just saw
in such and such place, as essential to what I had meant. That in such a case it is their
conjunction that is the essential individual concept is made plausible by the fact that, if
through some bizarre contingency those individual concepts (toward both of which
one has the attributive attitude) are discovered not to be conjointly instanced, then one
will be inclined to think that there is nothing that clearly was ones referent.
This is often the case with regard to beliefs. There may be a certain complex of
individual concepts that is essential to how I locate, so to speak, a certain familiar person
in my picture of the world. And the discovery of certain kinds of breakdown in the
conjoint instancing of key individual concepts would naturally produce the feeling that
there really was no person who clearly was the object of at least some of those beliefs.
That this can be true also of our communicative intentions may be somewhat
obscured by the fact that, in our casual and unpremeditated utterances, a good deal of
the content is implicitthat is, not fully represented in our fully conscious thoughts
and intentions. This requires motivation. Consider rst the implicitness of what a
speaker believes about his and his hearers mutual beliefs about some familiar object
of conversation. That such implicit beliefs play a central role in the psychology of
interpersonal contact I take to be an uncontroversial assumption of any realistic
psychology. But the content of our intentions is shot through with the content
of our implicit expectations. This is especially operative in just those cases in
which a speaker reectively tries to isolate what was essential to what he had
intended to communicate; in effect he consults his background beliefs about
mutual beliefs, and decides which individual concept or concepts implicit in
those mutual beliefs were such that their recognition was essential to the success
of his communication.
V
A difference between the paradigm referential and attributive uses of denite
descriptions is that, on the attributive, the referential qualieri.e., what is literally
expressedis intrinsic to what is meant, while on the referential it is not. But
consider once again the two different uses of He is insaneto mean that whoever
murdered Smith is insane, and demonstratively to refer to some identied individual
and to say of him that he is insane. One is inclined to assimilate these uses of
pronouns respectively to the paradigm attributive and referential uses of denite
descriptions. But the distinction there has nothing to do with whether the pronoun
literally connotes something that is intrinsic to what is meant.
Let us call the former use of he, as well as the paradigm attributive use of denite
descriptions, generalizing uses of singular terms. And let us call the latter use of he,
as well as the referential use of denite descriptions, identifying uses of singular
terms. There are then two interesting distinctions between the paradigm referential
and attributive uses of denite descriptions: on the former, the use is identifying, and
it is (as we might call it) extrinsicthat is, the referential qualier is not intrinsic to
what is meant; whereas on the latter the use is generalizing, and it is intrinsicthat is,
the referential qualier is intrinsic, or essential, to what is meant.
The radical two-use theory attempts to explain the identifying uses of singular
termswhere one may think of the utterance as about some denite individualby
making the referent itself a constituent of what is meant. Against that I have been
maintaining that on all uses of singular terms an individual concept is intrinsic to
what is meant. How then do I account for the intuitive distinction between gener-
alizing and identifying uses of singular terms? The short answer is that, on identifying
uses of singular terms, the individual concept that is essential to what is meant is
identifying, in a sense I shall now explain.
Whether a denite description or individual concept identies an object for us

rather depends upon our classicatory interestson what we want to know. So,
relative to an interest in locating him, we learn who the secretary of the bridge club is
by learning that he is the person standing by the window. Relative to an interest in his
name and occupation, we learn who he is by learning that he is Oswald Culbertson,
author and ranking player. Relative to an interest in his performance, we may learn
who he is by learning that he is the one who bid and made seven spades missing two
aces. If we are looking through photographs of famous murderers, we may learn who
that ugly character is by learning that he is Smiths murderer. For any F, one can
specify a set of interests and circumstances relative to which learning that x is the F is
learning who or what x is. In that sense, any individual concept may be identifying.
But there are some classes of individual concepts whose members in general have a
more central identifying role for us than others. We may feel that we have learned
who Smiths murderer is, in a respect that is in general pragmatically important, if we
learn that he is: the person whom we see there, or the well-known criminal called
Sebastian Moran, or that person, unperceived and nameless as yet, who ts a certain
comprehensive dossier. In that same pragmatically important respect, we may feel
that we have not yet learned who Smiths murderer is if we learn only that he is: the
one who left the cigar butt, the person who gave a party for seventeen last week at the
Hunam Restaurant, or the rst person to swim across the Strait of Magellan. To be
sure, those descriptions would be identifying for us relative to certain other interests,
but they are not as such identifying relative to certain more central classicatory
interests that we generally have with regard to particulars. In all of these cases we may
want to know more; but in the former we feel that we have reached a stage of
identication that we have not reached in the latter.
Along these lines, I suggest that the following classes of individual concepts are
especially important:
(1) those of the form the such and such that we see (saw, perceived) to be such and
such, which we might call perceptual;
(2) those of the form, the person or object who or which approximately ts such
and such comprehensive dossier, which we might call comprehensive;
(3) those of the form, the such and such that is called N, or the such and such
referred to by x, which we might call reference-directed.
The next step is to take perceptual, reference-directed, and comprehensive individual
concepts as forming, perhaps together with a restricted few other classes, the class
of those individual concepts that are naturally regarded as identifying simpliciter because
they are identifying relative to those pragmatically important classicatory interests.
The idea, then, is that a use of a singular term is identifying just in case the
individual concept that is intrinsic to what is thereby meant is identifying in the latter
sense. The paradigm referential uses of denite descriptions are just those on which
the referential qualier is not intrinsic to what is meant, while the individual concept
that is intrinsic is an identifying one. Uses of demonstratives, personal pronouns, and

names are then intuitively assimilated to the referential uses of denite descriptions
when the corresponding individual concepts are identifying.
The referential-attributive distinction is therefore not the fundamental one. More
basic are the distinctions between extrinsic and intrinsic uses, and the distinction
between identifying and generalizing uses. While the former is not a matter of degree,
the latter iswhence the fact that there seems to be a spectrum of cases, from the
central attributive ones to the central referential ones, rather than a sharp distinction.
It is a matter of the relative identifyingness of the individual concepts involved; and
that perhaps depends upon something like the relative general pragmatic importance
of the different identifying classications. I have nothing more to say here on the
subject of identifyingness, but it does seem that it may involve matters of some
general philosophical interest.
VI
Now for a complication. Consider any embedding of a singular term within a larger
singular terme.g., the daughter of the reporter. The embedded singular termthe
reportermay be being used extrinsicallythat is, its referential qualier may not be
essential to the individual concept that it is the speakers more basic communicative
purpose to introduce by that terme.g., something like the person whom we see there.
That is, if it is pointed out to the speaker that that person is not a reporter, he may say
that he really meant to refer to the person whom we see there, and this latter concept
may survive the attributive test.
But it will often be the case that this individual concept, which it is the function of
the embedded singular term to introduce, is not intrinsic to what is meant on the
whole utterance, but, rather, is essential to a certain way of determining what is
intrinsic to what is meant on the whole utteranceas if it were part of the referential
qualier of the whole singular term.
So, for example, consider The daughter of the reporter is a geologist. If the whole
singular term is being used extrinsicallye.g., what is meant is something like the
such and such person to whom we were earlier introduced is a geologistthen
the individual concept that is introduced by the embedded term the reporterthe
person we see thereis not itself intrinsic to what is meant on the whole utterance.
Rather it functions as part of the concept daughter of the person we see there, which
we might call a derived referential qualier. Its function is to aid in determining (in
the same manner as the literal referential qualier of a simple singular term) the
individual concept that corresponds to the whole singular term and that is intrinsic to
what is meant by the whole utterance.
It is, then, the function of a singular term to introduce, on its extrinsic uses, an
individual concept either into what is meant by the whole utterance, if the singular
term is not embedded, or into the derived referential qualier of the singular term in
which it is immediately embedded. So, when elsewhere in this paper I say that it is the
function of a singular term to introduce an individual concept into what is meant,
that is to be understood as covering both of these. Giving a properly recursive
account, to cover unlimited embedding, requires merely some relatively simple
technical formulations.9
This account is no more complex than the facts. There would be a parallel demand
on the radical two-use theory; any developed version of it would have to account for
the following. It is only the referent of an unembedded singular term that might be
counted as intrinsic to truth conditions. The referents of embedded singular terms
would be essential, not to truth conditions, but to the determination of what is
essential to truth conditions. A singular term that occurs within a double embedding
would function to determine a referent that contributes to the determining of a
referent, which in its turn contributes to the determining of a referent that is intrinsic
to truth conditions. This is isomorphic to my account; the difference lies solely in the
question whether, on referential uses, it is an individual concept or the referent itself
that is introduced into the content of what is meant in the broad sense.
VII
What is to refer? Explaining the general semantical function of singular terms has not
required the notion of referring. Nor has explaining the referential-attributive dis-
tinction required it, which is perhaps not so surprising, in view of the fact that an
utterance may intuitively be of the referential kind even though reference fails.
Some may be prepared to use refers even of attributive uses, so that x has been
referred to merely if x instantiates some perhaps non-identifying individual concept
which is intrinsic to what is meant. But many philosophers have wanted to reserve
refers for some much more restricted relation than this merely denotational one.
The narrower usage has usually been prompted by the idea of a radical semantical
distinction between true reference and mere uniqueness generalization, and that is
quite against the spirit of my account. Nevertheless, I think that one may capture
within my account the intuitive notion with which those philosophers have, in effect,
been operating.
A rst approximation is this: x refers to y just in case y instantiates some identifying
individual concept which is intrinsic to what x means. This is not sufcient, for
reference has not occurred in the narrow sense unless the singular term admits, in a
9
The relevant semantical structure of a singular term that contains embeddings may be represented as a
tree, a nesting of ordered n-tuplesfor example, that which is R to the F and the G has as its meaning, at
this level of abstraction, the ordered pair R, F, G. The most embedded referential qualiers will in
particular utterances of that expression aid, together with contextual factors, in determining individual
concepts the H and the K. These are then part of the derived referential qualier R to the H and the K, which
in its turn aids in determining the individual concept introduced by the whole singular term.
certain way, of existential generalization.10 Let us capture this requirement by saying

that the singular term (or the individual concept) must occur with widest scope.11
I then propose the obvious: x refers to y just in case y instantiates some
identifying individual concept that is intrinsic to what x means, and which has
widest scope there.12
There is a case of Donnellans which would, if successful, show that those condi-
tions are not necessary for referring.13 A child says to his parents Tom is a nice man,
but, since the child saw various Toms in various similar circumstances, it happens
that there is no individual concept that his parents can infer as being what he means.
So, says Donnellan, they naturally take him to be referring to the Tom who was the
cause of his use of Tom.
For this to be a problem, there would have to be nothing relevantly in the childs
thoughts and intentions that individuates some Tom. But why then say that he has in
the appropriate sense referred to someone? The language we use to describe such
cases may be misleading. Suppose, for example, that Jones is trying to remember
whether he has invited someone besides A, B, and C to dinner; he has the feeling
that there may be one or two more. One might say to him: Whom might you have
in mind?, or Whom might you mean?, or even To whom might you be referring?
The point is that on other rather more central uses of these expressions, there is
in this sort of case no one whom he has in mind, or means, or refers to; he
cannot remember.
So we have to be careful not to conate that use of refers, and other such terms, on
which we describe a certain kind of cause of a vague or indeterminate utterance or
state of mind, and that use on which we describe the recoverable content or
something denotationally related to the content. There is always the danger of
forgetting that isolated intuitions about the appropriateness of using refers are not
a sufcient basis for semantical theory. The determination of what constitutes
referring, in the sense in which that term might be useful in semantical theory,
should result from the interaction of intuition with the theory in which that notion is
to play a part.
10
In the case of a singular term which occurs referringly but not wholly transparently, the existential
proposition that is entailed may not be expressed by the result simply of substituting a variable for the
singular term and prexing an existential quantier. So Giorgione was so called because of his size does
not entail (Ex)x was so called because of xs size, since that makes no sense. But it does entail (Ex)x was
called Giorgione because of xs size.
11
Or, in the case of referring but non-transparent occurrences, which are best unpacked as concealing a
double occurrence of the term (one transparent and the other non-referring), the requirement should be
that a certain one of the relevant occurrences of that term in the analysans occurs with widest scope.
12
The thought might occur that this is not sufcient, because of predicative occurrences of singular
termsi.e., as in x is the F. But if the F introduces an identifying individual concept, it is natural to read
its predicative occurrences as referring. A singular term occurs merely predicatively only if it introduces a
non-identifying individual concept.
13
Donnellan (1972), p. 364.
Are those conditions sufcient for referring? That is not strictly essential to my
principal thesis. For it might be that, while every conventional use of a singular term
introduces an individual concept into what is meant, something more than an
identifying individual concepts being intrinsic to what is meant and applying to x
is necessary for xs having been referred to. For example, some causal connection
might also be necessary.
Take this case, for example. Jones has never heard of the real Plato, but has had a
detailed dream about a great ancient philosopher called Plato by us. As it happens,
the dream did not have the real Plato among its causal antecedents, and, moreover,
the dream Platos biography differs substantially from our Platos biography. Jones
subsequently remembers the dream Plato as if real and (with impressive projection)
takes it for granted that Jenkins shares his beliefs. In a conversation with the latter (on
the question, say, of ancient anticipations of modern science) he says, Plato believed
in electrons. Suppose that the individual concept the pre-eminent ancient thinker
called, among us, Plato would alone pass the attributive test. Since I am counting
such individual concepts among the identifying ones, it follows from my account that
Jones has referred to the real Plato.
But, what Jones intended to communicate was false, and was false, moreover,
because of a fact about the real Platoso why not say that he managed to refer to the
real Plato? The trouble is that one does have a strong inclination to deny that Jones
referred to, or at least that Jones was referring to, the real Plato.
Now I might say, Very well, I grant you the need for adding some extra condition,
a causal one, say, to my explication of refers. Nevertheless my principal thesis
stands; it is the function of singular terms to introduce individual concepts into what
is meant, and it is a necessary condition of xs being referred to that x to satisfy some
such concept.
There is, however, a more fundamental issue at stake. The chief philosophical
interest of causal theories of referring consists in their purporting to be theories of
semantical content. The case under discussion has no tendency to promote causal
theories in that sense. The content of what Jones means is clear, and is fully
speciable independently of the causal relations in which his utterance stands. The
case would show at most that there is a partially causal sense of refers for which my
conditions are not sufcient; but perhaps that notion is not, interestingly, an
unmixed semantical notionthat is, a notion which is essential to, or denable
solely in terms that are essential to, the general account of the semantical function
of singular terms.
Is there then another, as it were, purely semantical sense of refers that my
denition captures and on which we can say that Jones has referred to the real
Plato? I do not mean to suggest that there are in ordinary-cum-philosophical usage
several rmly demarcated senses of refers. But my ear is not startled by the
suggestion that, given what Jones really meant, he succeeded in referring to something
quite different from what he had intended, under a certain description, to refer to.
It is compatible with a certain continuous philosophical usage to use refers in that

way. The semantical concept of referring is to be knocked down to the highest
bidderthat is, the best overall semantical theory.
VIII
There appears to be a conict between the radical two-use theory and my theory,
concerning the possible world truth conditions of a referential utterance of t is G.
On the former, there is a certain item x, the referent of t, such that the utterance is
true in a possible world w iff x is G in w. On my theory, however, there is some
individual concept F such that the utterance is true in w iff whatever is the F (in w) is
G in w. The two are not equivalent since, for non-essential F, x is not the F in every
possible world in which x exists.
This conict is merely apparent, however. For, on the usual form of the two-use
theory, the possible world truth conditions of an utterance are determined by what
statement is made (or by what is said); and supposedly on referential utterances
sameness of referent is a necessary condition of sameness of statement. Now,
whatever the merits of this claim about how we should individuate statements,14 it
is irrelevant to the question what is meant on referential utterances. The notion of the
possible world truth conditions of a referential utterance is then ambiguous as
between what is said and what is meant.
Moreover, the notion of what is said should be regarded as a derivative concept of
semantical theory. It is to be explicated partially in terms of the concept of what a
sentence means, which is a function of conventional regularities which correlate
sentential features with aspects of those communicative intentions that are consti-
tutive of what speakers mean. Consequently it is not to be employed in giving the
basic semantics of singular terms.
This brings us to the question of referential occurrences of singular terms within
modal contexts. Nothing in my account is incompatible with the idea that in
assessing the truth-value of Necessarily . . . t . . . or Possibly. . . t . . . , one sometimes,
so to speak, carries the real-world referent of t through all possible worlds. For
example, one might use That is necessarily a spatio-temporal object in such a way
that what is said or meant is true just in case the referent of that is a spatio-temporal
object in every possible world in which it exists. That happens if and only if that has
wider scope than necessarily. It does not follow that there is no individual concept
that is thereby introduced; rather, in the proposition that is meant, the individual
concept has wider scope than the modal operator: the F is such that necessarily it . . . .
14
If the motivation for the claim is the appropriateness of de re reports of what is said on referential
utterances, then it is doubtful that ordinary usage justies this sharp distinction between the kinds of
statements made on referential and on attributive utterances, since it permits de re reports of the latter
as well.
The behavior of referential occurrences of singular terms in modal contexts can

have a distorting effect on ones theory of their role in non-modal contexts, if one
reasons as follows. First, assume that Possibly t is G (or t is possibly G) is always to
be explicated, even when t occurs referentially, as having the form (t is G), that is,
as resulting from applying a sentential operator to a complete sentence. Secondly,
take S as true just in case there is some world at which the proposition which is
independently expressed by S is true. Thirdly, observe that if t occurs referentially
one is inclined to read Possibly t is G as true iff there is some world at which the
real-world referent of t is G. Fourthly, conclude that the proposition which is
independently expressed by t is G, with referential t, must somehow incorporate x
itself, the real-world referent of tthat is, must be the propositional complex G, x.
The point is that the initial assumption is gratuitous. For modal operators can be
quantied into, and it is just as good an hypothesis that some occurrences of
singular terms with a modal operator are to be read as quantiers that have wider
scope than the modal operator. Which is the most useful analysis depends upon an
independently motivated account of singular terms, in the context of a comprehen-
sive semantical theory.15
IX
What I regard as the correct account of proper names has in recent years been
unfairly rejectednamely, that the conventional meaning of a name N in a certain
group of speakers, that is, its meaning in abstraction from any particular utterance, is
the same as the conventional meaning of the denite description the thing or person
called N. Just as a literal use of the table does not imply that there is just one table in
the world, so it does not follow that a conventional use of Jones implies that there is
just one person called Jones. The denite description the table has as its referential
qualier the non-individuating property being a table; similarly, Jones has as its
referential qualier the non-individuating property being called Jones. More often
15
Throughout this paper, I am primarily concerned with that version of the two-use theory on which
the referential uses of singular terms (those that supposedly introduce a referent, rather than an individual
concept, into semantical content) are identifying. But there is a version on which that is not necessary;
indeed Kripkes theory would appear to be such. One might envisage two different ways of introducing a
proper name: (a) one announces When I say Altissimus is G I shall be saying that the highest molehill in
the world is G; (b) one announces The highest molehill in the world, whichever it is, is such that when
I say Altissimus is G I shall be saying that it is G. On the former, Altissimus is a non-rigid designator; on
the latter, it is supposedly a rigid designator, with the description xing the referent of the name in other
possible worlds by picking out the thing that satises it in this world. (I am indebted to Stephen Schiffer for
this perspicuous way of formulating Kripkes theory.)
Now I do not think that (b) has anything to do with the semantics of names or of any singular terms. If
someone were rst to announce (b) and then to assert Altissimus is in Ohio, I would have to interpret him
as having meant that the highest molehill in the world is in Ohioi.e., precisely what one would have
interpreted him as having meant if he had rst announced (a) and then said the same thing. This is not to
say that (a) has much to do with the semantics of names either; to which subject I now turn.
perhaps than not, the individual concept which is essential to what is meant on a use
of N has the form the F called N, and since such individual concepts are identifying,
these uses are intuitively assimilated in that respect to the referential uses of denite
descriptions.16
There has been a tendency to interpret description theories of names as claiming
that, insofar as N names x, there is some one individual concept that is meant on all
uses of N to refer to x. My thesis is far less dramatic: on each of Ns uses in referring to
x some individual concept or other that is satised by x is intrinsic to what is thereby
meant. There is, of course, a non-arbitrary relation between the name uttered and the
individual concept that is meant, but that is just an instance of the general relation
that holds between referential qualier and intrinsic individual concept. The point
here is that what is essential to what a speaker means in using N as the name of x
varies among speakers and occasions of utterance, according to what the speaker
assumes about the audiences beliefs, and within this assumption, according to what
the hearer would have to understand, independently of the existence of the referent,
for the utterance to have been a successful communication.
There are some separate issues here that need clarication.
(1) The referential qualier of N is, as I have said, being called N, and the occurrence
of called here may seem to introduce a circle. By x is called N I mean to cover these
relations: x is (commonly) referred to by the use of N, and x was dubbed N. Now
the concept of a name is clearly not presupposed in my account of refers, and since
to dub is to make known a decision or agreement to use a certain vocable to refer to a
certain thing, neither does the explication of x was dubbed N presuppose
the concept of a name. So, I do not understand why many think that this sort of
account threatens circularity. It is not as though one were trying to refer to
something by saying the referent of this very phrase; rather, on an intrinsic
use of N one means the F which has been referred to as, or has been dubbed, N;
and there is no more problem with that than there would be had one uttered those
very words.
Perhaps it is felt that since an initial dubbing may in some cases be what gives N
meaning, it is therefore absurd to suppose that the meaning that N is thereby given is,
as it were, the thing dubbed N. But, properly understood, this produces no absurdity.
We agree that N may be a way of referring to x, knowing that this will be possible
because on any particular utterance we may exploit the agreement and mean by
N that which is F and which we agreed we might refer to by uttering N.
16
There are also extrinsic uses of names, some of which are identifying and some generalizing. Of the
former we have this example: I point at someone leaning over a balcony and say Jenkins is off balance,
where it may not be essential to what I mean that the person there is called Jenkins. Of the generalizing
extrinsic uses there is this sort of example: We are examining a text attributed to Duns Scotus and I say
Scotus here reasons badly, meaning the author whoever it was.
(2) It follows from my account that, strictly speaking, to specify the conventional
meaning of Nwhat it contributes to the literal or conventional meaning of the
sentences in which it occursdoes not involve specifying anything of which N is the
name. But a strong central tradition in semantics has treated a name as having at least
as many distinct literal meanings in a group of speakers as there are bearers of the
name for that group. On my account, however, the fact that there are many Smiths
no more makes Smith literally ambiguous than the fact that there are many dogs
makes the dog literally ambiguous.
The notion of the literal meaning of a name is somewhat technical; it plays a
certain role in the account of the relation between the conventional regularities
governing expressions, and how these are exploited on particular conventional utter-
ances. There is indeed a conventional name-relation; for it is a conventional, and in its
way semantical, fact about a group of speakers that George Johnson names this
person and that person and that person. The point is that that conventional property
of N, that N names x, is derivative from, rst, the fact that N has the appropriate literal
meaning and, secondly, the fact that there is a practice of referring, or an agreement
that one may refer, to x by uttering N in accordance with its literal meaning.
So N is the name of x in a group G just in case it is mutual knowledge among the
members of G that it is their practice to (or they have agreed that they may) utter N
and thereby refer to x as being called N.
(3) It has been thought to be a problem for description theories of names that our
beliefs about a given historical gure may turn out mostly to be false. An example (of
Kripkes) is Jonah the prophet; it seems that little of what is said in the book Jonah is
true of anyone. Nevertheless, it seems that our use of that name does actually refer to
a certain Hebrew personage.
But this is a real problem only for description theories which hold that our use of
an historical name introduces substantial biographical facts. Recent discussion has
amply shown that even the most entrenched beliefs about an historical gure may
not be essential to what we intend to communicate when we use that persons name.
But that hardly shows that no individual concept is essential. If I am conversing with
someone whom I take to have ordinary information, and I say Jonah never really
visited Nineveh, I would be meaning that the pre-eminent referent of the Biblical use
of Jonah never visited Ninevehor even that that Biblical referent of Jonah who
was said to have been ingested by a large sh never visited Nineveh.
(4) A form of argument that gets used against description theories of names is this:
(a) If N were used to mean the F, then N might not have been the F would be false.
(b) But N might not have been the F is true.
(c) Therefore, N is not used to mean the F.
Of course, if this form of argument works, it works for every non-essential
individual concept the F: and obviously on my account the individual concepts
that are intrinsic to what is meant on the use of a name are non-essentiali.e.,
contingently applicable.
The argument is defective. Names are normally read as having wider scope than
modal operatorsand that is why premise (b) is true for the relevant Fs. For
example, Aristotle might not have been the pre-eminent ancient philosopher called
Aristotle is true because it asserts of Aristotle that he might not have had that
contingent property. But if the N-position in the sentence mentioned in premise (a)
is read similarly as having widest scope, then premise (a) is false, as may be seen
by substituting the F for N. That the pre-eminent ancient philosopher called
Aristotle means the pre-eminent ancient philosopher called Aristotle does not
entail that it is false that the pre-eminent ancient philosopher called Aristotle is
such that he might not have had that property.
Kripke anticipates that such a point might be made, and he replies: we use the
term Aristotle in such a way that, in thinking of a counterfactual situation in which
Aristotle didnt go into any of the elds and do any of the achievements we
commonly attribute to him, still we would say that was a situation in which Aristotle
did not do those things.17 Now we would indeed say that, and we would then be
giving Aristotle wider scope than the quantier over counterfactual situations. And
if we substitute the pre-eminent ancient philosopher called Aristotle and give it
widest scope we would be entitled to assert the result of that as well.18 Ordinary
particulars are the referents of our utterances always by virtue of their having
contingent individuating properties that are intrinsic to what we mean.
X
The theory so far has these components: (1) corresponding to every use of a singular
term, there is an individual concept which is intrinsic to what the speaker means on
that occasion; (2) sometimes the individual concept is not a logical restriction of the
referential qualierthat is, of the concept literally connoted by the singular term;
such uses are extrinsic, the others being intrinsic; (3) individual concepts range from
the identifying to the (more or less) non-identifying or generalizing; (4) the so-called
referential uses of denite descriptions are extrinsic and identifying; the attributive
uses are intrinsic and generalizing; (5) one refers to x by uttering a singular term just
in case x instantiates some identifying concept that is intrinsic to what one means
and which has widest scope; (6) a proper name N has as its referential qualier the
17
Kripke (1972), p. 279.
18
This issue is quite distinct from another central point of Kripkes with which I completely agree
namely, that in considering the truth conditions of Hawkins might not have been a taxi-driver, one need
not worry about what properties something has to have in another possible world to be Hawkins (or his
counterpart). Rather what it means is that there is some world to which our Hawkins stands in the relation
x is not a taxi-driver in w.
concept being called N. Like other singular terms, its uses may be extrinsic or
intrinsic, and identifying or generalizing.
One question remains: What constraint is imposed by the referential qualier of a
singular termby what it literally or conventionally means qua expression of
the languageon what individual concepts may be essential to what the utterer
of that singular term means? This is really the same question as: What conventional
regularities in the use of singular terms account for their semantical properties qua
expressions of English?
On a normal utterance of the F, the hearer infers what individual concept is
intrinsic to what is meantthe H, saypartially from the referential qualier F, and
partially from other facts which the hearer assumes have been taken by the speaker to
be mutually believed by them. That is pretty vague, but we can be more specic.
If on an utterance of the F is G the speaker means that the H is G, then normally
the speaker assumes that it is mutually believed by him and his hearer that the H is F.
For example, if by uttering the chair I mean the medium-sized object visible in that
direction, I have assumed that it is mutually believed by me and my hearer that the
medium-sized object visible in that direction is a chair. This mutual belief is not
necessarily epistemically presupposed; it may itself be brought about by my utterance,
as when I point at something hitherto unnoticed by my audience.
Now, the question is, how does this help the hearer to determine which individual
concept is essential to what the speaker means? It is useful, I think, rst to consider
unqualied indexicalsthat is, pronouns that do not have referential qualiers. Let
us pretend that it is unqualied.
How would we, as hearers, normally determine which individual concept was
intrinsic to what was meant on an utterance of It is F, in a context in which there is
no anaphoric reference to another term in the discourse? Virtually any fact about the
speakers current interests, motivations, concerns, preoccupations, habits of associ-
ation, perceptual circumstances, and beliefs in general, including his implicit
assumptions about his hearers beliefs about him and about virtually anything else,
may be relevant to the hearers inference to which individual concept is meant. There
are no special pragmatic rules for the interpretation of indexicals, apart from all those
psychological generalizations, and specic beliefs about the speaker, which we must
assume hearers somehow bring to bear in interpretation. From what he knows, the
hearer forms a hypothesis about what the speaker is most likely to have intended to
communicate.
Now the point about singular terms that are qualied indexicals is this: the
referential qualier restricts the range of individual concepts that might be meant;
and from this class the hearer infers which one is meant. The inference is now just
like the inference when the singular term is unqualied, except for the narrower
range of possibilities.
The principle of restriction we have in effect already seen. The individual concepts
that might be meant on an utterance of the F are not all logical restrictions of the
concept F; rather the H is among them just in case the speaker assumes that it is
mutually believed by him and his hearer that the H is F. In those cases in which the
H is a logical restriction of Fas Smiths murderer is of (the) murdererthe looser
restriction is obviously satised; thus attributive uses of singular terms satisfy the
condition just as much as referential uses do. It is in this that the ultimate univocity of
singular terms consists, in spite of the existence of two such seemingly distinct
conventional uses as the referential and the attributive. On an attributive, or intrinsic,
use, the individual concept that is intended to be recognized as intrinsic to what is
meant is just a logical restriction of the referential qualier.
So to specify the meaning of a sentence that contains a singular term requires
specifying its referential qualier. And what makes F the referential qualier of t is
just that an utterance of a sentence containing t is in accordance with the conventions
only if there is a certain individual concept the H such that (a) it is expected by the
speaker that (the hearer will believe that) it is mutually believed by speaker and
hearer that the H is F, and (b) the H is intrinsic to what is meant on that utterance.
Other possible conventions come to mind. Why not, for example, treat referential
and other extrinsic uses as extended or non-conventional uses? That would make it
possible to simplify the convention: basically, an utterance of the F is G would be
conventional only if the speaker means that the F which is H is G. But that is
unrealistic. Conventions are not to be introduced merely to simplify semantical
theory, to enable us to deal with the complexities of actual communicative practice
by casting them into the exterior darknessinto pragmatics. General semantical
conventions are regularities, and the referential uses of singular terms are no less
frequent than the attributive uses. We can easily imagine, moreover, that English
could semantically be just as it is if the referential uses of singular terms were
overwhelmingly predominanteven if referential misdescription were the rule. My
account of the convention allows for those possibilities. Referential uses of singular
terms are literal uses; they are fully in accordance with those regularities that establish
the conventional semantical properties of the language.19
19
From among the many people discussion with whom has helped me greatly on these matters, I want
to mention particularly John Bennett, Lee Bowie, Warren Ingber, Thomas Ricketts, and Stephen Schiffer.
4
Must Beliefs Be Sentences?
The language of thought hypothesis is this: central among the causes of our
behavior are inner states with linguistic structure that play roughly the role we
pre-scientically ascribe to our beliefs and desires. A philosophical thesis has been
proposed on the basis of that scientic hypothesisnamely, that the way to
explicate common-sense notions of the content of beliefs and desires, of their
intentionality, is in terms of the meaning of such internal sentences. I am going
to compare this explicative strategy with a certain functionalist theory of propos-
itional attitudes, on which propositional attitude ascriptions of the form x believes
(desires) that_____ are explicatively more fundamental than anything linguistic
and semantic. This is a modern-dress version of an old dispute; but the issue is not
between the naturalistic proponent of linguistic meaning and the anti-naturalistic
proponent of irreducible intentionality. The propositional attitude based theory
that I have in mind is as naturalistic as can be.
The issue in fact is more a family contention than a global struggle; for the two
accounts share important assumptions about the nature of belief. Both are mentalist,
by which I mean of course not that they are not physicalist, but that they count beliefs
etc. as internal states, and, moreover, in accounting for systematic features of
common-sense psychology and the individuation conditions of beliefs, both theories
invoke systematic internal roles for those states, characterized functionally.
This is not to say that, on these theories, meaning or content is entirely in the
head. Both can be components of two-level theories of content, on which the roles of
beliefs and desires in psychological explanation neither determine nor depend, in
general, on their truth and reference conditions. So on both theories there may be
two questions: what are the psychologically explanatory internal properties of beliefs
and desires, and what extrinsic relations to the world explain their reference and
truth conditions?1
We must sharply distinguish the language of thought hypothesis, which is a
scientic theory, from what I shall call the language of thought based theory, which
is a philosophical explicative strategy. The difference between the two explicative
1
Cf. Field (1978), Fodor (1980), Loar (1981), McGinn (1981), Schiffer (1981b). Harmans conceptual
role semantics, as presented, e.g., in (1982), is not a two-level theory. This is also made clear in Harmans
(1982a) comments on the present paper.
MUST BELIEFS BE SENTENCES ?
strategies to be discussed is not that one admits a language of thought while the other
admits propositional attitudes, but rather that they construct their explicative theories
of content on different bases. The language of thought based theory regards primary
ascriptions of meaning or content to be internal linguistic states. These can be repre-
sented as sentential attitudes, relations between persons and their own inner sentences,
for example, in the case of belief, as
(1) B*(x,s).
If propositional attitude ascriptions like Pierre believes that its raining are to be
accommodated, the theory counts them as derivative, thus:
(2) x believes that p
iff
for some internal sentence s, B*(x,s) and s means p.2
The propositional attitude based theory, on the other hand, takes propositional
attitude ascriptions as the primary ascribers of contentwhich is not to say that they
are not to be explicated, but that they are not to be explicated in terms of more
fundamental ascriptions of content, as in (2). The meaning of public language
sentences would, of course, be explained in terms of their correlations with propos-
itional attitudes. But what is more to the point, if there is a language of thought, its
sentences meanings would be explained in terms of the propositional attitudes they
somehow instantiate or realize. A rst approximation might be:
(3) The linguistic state (type) s, of person x, means p
iff
s is, in x, the belief (type) that p.3
Naturally, it is possible that these orders of explication are only notionally distinct,
that there is a theory of mental content in which they are merely variants, different
ways of organizing the same material. But I do not think so; at least on the
propositional attitude based theory that I shall sketch, the two approaches have
signicantly different presuppositions.
1 A Propositional Attitude Based Theory

The fundamental idea of the propositional attitude based theory that I shall now
outline is that propositional attitude ascriptions are to be interpreted as ascribing
functional states with certain associated truth conditions. Our ordinary ascriptions
2
Cf. Field (1978), Fodor (unpublished manuscript, Psychosemantics) [Fodor (1987), p. xiii, refers to
that manuscript as an earlier and substantially different ancestor of that book].
3
For a theory of this type, see Loar (1981).
are then supposed to be in varying degrees elliptical, both about the functional or
conceptual roles of a belief or desire and about its specic truth conditions. I am
going to suggest a somewhat articial style of representing in their fullest form both
aspects of a belief; of these idealized forms ordinary ascriptions are then to be
regarded as ellipses and transformations.
Suppose Lulu believes that she is witty. Some might take this to assert simply a
relation between Lulu and a singular proposition, that is, a proposition of which she
is a constituent, thus:
(4) B (Lulu, [Lulu is witty]).
That is better than taking the truth conditions to involve a description Lulu has of
herself; for the descriptions we have of ourselves seem ultimately irrelevant to what
makes our beliefs self-ascriptive.4 But I doubt that self-ascriptiveness is captured
simply by building the believer into the truth conditions. For might Lulu not also
believe that very proposition in circumstances in which, although she has a percep-
tual relation to herself, say via a deceptively arranged mirror, she fails to recognize
herself as herself? The difference between the two beliefs seems to lie in their roles in
Lulus thinking; the rst has a self-ascriptive role and the second does not. We may
capture this provisionally by a relation between Lulu and the English sentence that
we would ourselves use to express that self-ascriptive belief.
(5) BC (Lulu, I am witty).
Understand this as not implying that Lulu speaks English; somehow we capture the
conceptual role of her belief, which is to say, a certain functional state, by producing a
sentence of ours. How this might work I shall say below. The predicate that you get
by substituting a variable for Lulu in (5), unlike the corresponding predicate in (4),
captures what Lulu has in common with Marie if Marie self-ascriptively thinks she is
witty. This and the example of the mirror show that conceptual role (in the sense to
be explicated) and truth conditions are doubly independent: you can have two beliefs
with the same truth conditions and different conceptual roles, and you can have two
beliefs with the same conceptual role and different truth conditions.
Consider Putnams Twin-Earthers, who are conceptually like us, but whose beliefs
are sometimes held to have different truth conditions than ours, because Twin Earth
is endowed with different natural kinds than Earth. Using the above relation, we can
represent what their beliefs have in common with ours, for example:
(6) BC (Twin-Earthers, we bathe in water).
4
The point, often made, has two parts. Descriptions are not necessary to self-ascription; a person may
simply lack individuating (non-indexical) descriptions of herself. Descriptions are not sufcient; for any
non-indexical description, it always makes sense to wonder whether it applies to oneself. Invoking
descriptions that involve indexicals merely postpones the point.
There is no implication in this construction that they speak English, and there is no
implication that their belief is about water, that is H2O. The belief is characterized,
somehow, as having a certain conceptual role.
This example would show (if we take the rigid designation view of how to interpret
it) that at least sometimes you cant recover the truth conditions of a belief from its
conceptual role. For a full belief-ascription you need more than (6); that this may not
be true in (5) is simply an uninteresting consequence of the fact that when you
mention the holder of a self-ascriptive belief you thereby specify a constituent of its
truth conditions.
But while conceptual roles do not determine truth conditions, it seems plausible
that they impose systematic constraints on them. Consider the conceptual role of
and, on the one hand, and those of I and water on the other. If we identify a Twin-
Earthers belief as conjunctive in its conceptual or inferential role, we must then, it
appears, interpret its contribution to truth conditions as conjunctive, without check-
ing on further extrinsic relations between that belief and its context. This is not to say
that the inferential or conceptual role itself explains why we assign truth conditions at
all. The point is that, whatever the reason, we have a xed way of truth-conditionally
interpreting that inferential role. Let us say it makes a context-free contribution to
truth conditions. By contrast, those aspects of conceptual role that are associated with
I and water make context-dependent contributions to truth conditions. But even in
this latter case, how truth conditions depend on context is determined by the
conceptual roles involved. Thus, self-ascriptiveness and the present tense determine
certain functions from beliefs to references, whose values are the believer and the
time of the belief. Similarly, had we a good theory of natural kind terms that
corresponded to common intuitions about Twin-Earth cases, it would yield a func-
tion from beliefs to natural kinds depending on causal facts about the belief. We
might then say that those aspects of conceptual roles that make context-dependent
contributions to truth conditions determine reference functions, that is, functions
from beliefs to references that depend, in certain ways, on context.5
Now, there is a sense of truth conditions in which Lulus and Maries self-
ascriptive beliefs, the ones we ascribe using I am witty, have the same truth
conditions; for each is true iff the believer is witty. Call that their general truth
conditions, by contrast with their differing specic truth conditions, that is, the sets of
possible worlds in which they are true.6 General truth conditions are, as it were,
5
This differs from my treatment of Twin-Earth cases in Loar (1981). I have been much inuenced
by Stephen White (1982).
6
White (1982) invokes Kaplans (1989, originally presented in 1977) notions of character and content
to mark what is basically the distinction in the text. I have not used Kaplans terms for two reasons. First, on
Kaplans treatment, and in Whites application of it, there is the implication that character should cover all
relevant semantic properties apart from content, while on my account there are two factors over and above
contentnamely, the conceptual roles of I, water, and so on, and the reference-functions associated with
those conceptual roles. Secondly, Kaplans distinction concerns the semantic description of a language
relativized to a population, with the consequence that, while indexicals are assigned variable functions as
functions from contexts to specic truth conditions. We may count Twin-Earth

beliefs that are indexed by water as having the same general truth conditions as
our conceptually equivalent beliefs. The general truth conditions of a belief are
determined by two factors: (1) the context-free contributions of aspects of its
conceptual role such as conjunction, and (2) the reference-functions determined by
aspects of its conceptual roles such as those marked by I and water. In a sense, the
conceptual role of a belief determines its general truth conditions; and its specic truth
conditions (the set of possible worlds in which it is true) are determined by its general
truth conditions together with aspects of its context that are systematically selected
by aspects of its conceptual role.
The qualication (in a sense) is due to this. The conceptual role of a belief does
not explain why we ascribe truth conditions to it. That is, the internal functional
horizontal connections that beliefs have to each other and to perception, desire, and
action do not (except in the case of certain observational beliefs) apparently explain
why we assign to such states systematic relations to external states of affairs. This,
however, is compatible with the fact that a certain association of general truth
conditions with conceptual roles is so entrenched that it is denitive of true.
There is then the further question why we make such a fuss about true. It is at
this point in the theory of content, I suggest, that we should point to such factors as
the informative reliability of beliefs as, so to speak, barometers of the facts, under this
entrenched interpretation of conceptual roles. In other words, there are systematic
relations between internal functional states and extrinsic states of affairs, constituted
by the fact that the former, when produced in accordance with certain epistemic
procedures, tend to be reliable indicators of an impressively diverse range of
the latter. If there were something interestingly unique about this correlation, that
might explain why we have a single way of ascribing general truth conditions to
conceptual roles.7
Beliefs have content in two senses: they have conceptual roles and they have
specic truth conditions. Let me now indicate a way of building the latter into
idealized ascriptions of belief. Consider again
(6) BC (Twin-Earthers, we bathe in water).
Italicize in (6) those markers of aspects of conceptual roles that make context-
dependent contributions to truth conditions, thus we bathe in water. Then add
the n-tuple of their references in the Twin-Earth context, thus:
characters, proper names are assigned constant functions. My notion of general truth conditions, however,
does not apply to a language relativized to some population, but rather to conceptual roles in abstraction
from any given population. Consequently names, and natural kind terms, are not assigned constant
functions as their reference functions, but variable functions. There is evidently no conict between
Kaplans distinction and mine; indeed the latter is in a way a generalization of the former.
7
I have discussed the uniqueness of truth conditions, as assigned to conceptual roles, in ch. 8 of (1981).
(7) B (Twin-Earthers, we bathe in water, (Twin-Earthers, XYZ)).

Given that conceptual role determines general truth conditions in a xed way, which
we may regard then as implicit in (7), this full ascription of content determines a
proposition, that is, the specic truth conditions of that belief:
(8) Bt (Twin-Earthers, [Twin-Earthers bathe in XYZ]).8
The ascription (6), together with certain selected facts about the belief s context,
determines (8). The ascription (7) is, in effect, the conjunction of (6) and (8); it is the
idealized full specication of content from which ordinary ascriptions are, in varying
ways, elliptical.
Let us now turn to the question how constructions of the form BC capture
conceptual roles. We may rule out two answers. For obvious reasons, in a propos-
itional attitude based theory no semantic concepts can be presupposed (see footnote
6). In (5), the meaning in English of I am witty is not as such invoked. Secondly, (5)
does not mean something like: Lulu is in the same functional state we would be in
were we to utter I am witty. Sameness of functional state must be relative to a
specic functional system and we havent specied one yet. But, as I hope will become
clear, when we do specify the functional system that interprets BC ascriptions, the
notion of equivalence between functional states is quite secondary in the construction.
An analogy may be helpful. In the Fahrenheit and centigrade systems, numbers
index temperatures. They can do so because arithmetic relations correspond system-
atically to certain physical relations among temperature states. The correspondence
is secured by the relevant scientic laws, generalizations about boiling points, etc. The
Fahrenheit system employs indirect reference: by designating a number you can
thereby indirectly refer to the associated temperature state.
The parallel, then, is that English sentences index beliefs and desirescertain
functional states. They can do so because certain structural, indeed syntactic, relations
among the sentences correspond to certain functional relations among propositional
attitudes. The correspondence is secured by common-sense generalizations about
beliefs and desires, about their relations to each other, to perception and to action.
A functional or conceptual role is indirectly referred to in ascriptions of the form BC:
by designating a sentence, you thereby refer to the associated conceptual role.
8
Propositions arent necessary. Our favored assignment of general truth conditions to conceptual roles
involves a particular satisfaction relation for the matrices of the sentential indices of ascriptions of form
(7)in this case, x bathes in y. The relevant satisfaction relation, given the normal choice of indices, is
simply disquotational, that is, x bathes in y is satised by ha,bi iff a bathes in b. Truth is dened then for
beliefs as ascribed in (7) in such a way that a belief is true iff the relevant n-tuple of references hTwin-
Earthers, XYZi satises the relevant open sentence. On a xed denition of satises (7) contains all the
information that (8) contains, without invoking propositions.
As I say in the text, dening true for beliefs exploits an entrenched, conventional association of truth
conditions and conceptual roles. The real work for a theory of truth is then to say whether this
conventionally favored assignment corresponds to any objective relation between thoughts and things.
How might common-sense generalizations about propositional attitudes secure

this association? There are several points here. First, the common-sense generaliza-
tions are construed as a functional theory, a theory of persons at times that charac-
terizes counterfactual relations between certain input conditions and beliefs,
counterfactual constraints on beliefs (for example, that if you believe that p & q
you dont believe that not-p, or if you believe that x is north of y, you dont believe
that y is north of x), and counterfactual relations among beliefs, desires, and action.
Secondly, they constrain beliefs and desires under certain de dicto ascriptions.
Naturally, you can believe contradictory things about objects, properties, etc. under
different modes of presentation. So the sentential indices of attitudes are meant to
make explicit their modes of presentation, which are then primarily aspects of
conceptual role. Rationality, both theoretical and practical, governs beliefs according
to their conceptual roles as ascribed in the BC style.
Thirdly, this common-sense functional theory must somehow individuate the
conceptual roles of beliefs captured by BC (x,s), for that theory is all we have in
the interpretation of that construction. This might appear an impossible demand; for
the common-sense theory would have to apply to everyone to whom we ascribe
beliefs. Can there be generalizations that are strong enough to individuate beliefs and
are, nevertheless, interpersonally ascribable? This is the big question about the
propositional attitude based theory; but the problems are smaller than they might
at rst seem. First, the generalizations need apply, not to all creatures to whom in
principle beliefs can be ascribed, but only to those to whom we actually ascribe them
using the concepts we do. So the generalizations, and the functional system they
capture, may be culturally restricted. Secondly, this means we can incorporate a wide
range of recognitional abilities, those that are generally ascribable to members of our
society, in the perceptual input conditions. That would individuate a broad range of
perceptual beliefs. Thirdly, if every other belief were then related uniquely by these
generalizations to unique n-tuples of perceptual beliefs, all beliefs would thereby
be individuatedin terms of, that is, their conceptual roles, and not their (specic)
truth conditions. Let me remark here that a description-theoretic approach to non-
perceptual concepts like elm and molybdenum, which allows metalinguistic
descriptions such as is called molybdenum by the experts, may greatly improve
the prospects of the required individuation.
Fourthly, it may be that a natural kind term, as we use it in ascribing beliefs, is
associated, not with a single interpersonal condition, but with a set of conditions
one involving a recognitional ability, another a metalinguistic condition of the sort
just mentioned, and another a theoretical role, for example, the picture of electrons
as revolving around nuclei, or of gravity as a universal attractive force between
bodies, and so on. So the idea is not that all persons whose beliefs we would ascribe
using, de dicto, a term like water or elm, or electron, satisfy one such condition,
but rather a small disjunction of that sort. We then have to explain two things: why,
at the level of conceptual role, these three sorts of conceptual role are somehow
interchangeable as interpersonal constraints, and why one reference-function

for natural kind terms should normally apply equivalently to those three aspects
of conceptual role.
2 Comparison of the Two Theories at the

Functional Level
The language of thought theory asserts that beliefs and desires are internal states with
linguistic structure. Those states form a syntactic system whose members are com-
posed of elementswords, as it werethat recur throughout the system in various
recursively structured combinations. These states are functionally individuated. Even
if they are neurophysiological, what makes them a language is not specically
neurophysiological. How then does the language of thought based theory differ
from the functionalist propositional attitude based theory? Let me mention two
important differences.
First, as I am interpreting it, the language of thought theory requires something
stronger than the functionalist propositional attitude theory. The latter does not
imply that the rst-order states that realize the relevant functional roles must
themselves have an internal structure that resembles the syntax of a language. Two
beliefs may be conjunctions by having a certain aspect of their functional role in
commonthey exhibit a common pattern of relations to other beliefs. The rst-
order states that realize those beliefs need not have something structural in common,
no common word or syntactic form that means conjunction.
The propositional attitude theory functionally individuates beliefs as wholes; the
language of thought theory requires the functional individuation of constituents of
beliefs. In other words, the former says that a set of xs states at t are xs conjunctive
beliefs just in case they each have such and such functional role in x at t. The latter
theory says that a set of xs states at t are xs conjunctive beliefs only if there is a
constituent (or a small disjunction of constituents) that they share and that has (or
have) the functional role of and in x at t. One way of putting this is that on my
account states are conjunctive beliefs by having second-order properties in common,
while on the language thought theory they also have to have rst-order, structural,
properties in common.
Secondly, the two theories differ in their relation to common-sense psychology.
The functionalist propositional attitude theory requires something stronger than the
language of thought theorynamely, that those conceptual roles that individuate
beliefs should be interpersonally ascribable, among those persons, that is, to whom
we unproblematically ascribe specic beliefs. Of course, a language of thought theory
could also require this. But that would not vindicate the explicative claim (of the
language of thought based theory) that propositional attitude ascriptions presuppose
a language of thought. For those interpersonally ascribable functional roles would
sufce for the sentential index theory; and the rst-order structure condition would
then be irrelevant to the explication, and simply an additional, independent, hypothesis.
The language of thought based explicative theory, then, as opposed to the language of
thought hypothesis tout court, is motivated only if those functional roles that
individuate beliefs are not interpersonally ascribable.
But the language of thought theory is functionalistthat is, it presupposes that
sentences in the language of thought can be individuated functionally. But if not
common-sense psychology, which functional theories do the individuating? One
possibility is that, apart from minimal constraints that common-sense psychology
imposes on beliefs, etc., we have to wait on theoretical psychology to delineate a class
of functional organizations within which the inner sentences of each of us would be,
perhaps idiosyncratically, individuated. Another possibility is that we take the con-
ceptual role of an internal sentence of x to be individuated by its probabilistic
relations to other beliefs in x, by the whole set of subjective conditional probabilities
of which it is, for x, the consequent (Field 1977). If we then had a philosophical
theory of what all such probabilistic psychologies functionally have in common, that
would yield a theory of what it is to believe an internal sentence (to a given degree),
and therefore an account of the sentential attitudes B* (x,s), etc., that are needed in
the language of thought based theory.
The language of thought based theory, then, has at least one possible advantage
namely, not requiring interpersonally ascribable functional roles that individuate
beliefs. Are there other advantages? Let me now consider the various motivations its
proponents have advanced.
(i) A scientic motivation lies in computational models in psychology; Jerry

Fodor has argued extensively along these lines (1975, 1980). If our psychological
capabilities are explained by information processing, there must be states
with structure appropriate for the proof-theoretic operations that are
involved in such processes, that is, with linguistic structure. Now if humans
do work computationally, and if beliefs and desires are genuinely causally
involved in explanations of our behavior, the natural idea is that beliefs and
desires are themselves inputs and outputs of computational processes, and
hence that they have a language-like structure.
(ii) The language of thought hypothesis is suggested by the idea that learning
language consists in learning how to think, and acquiring conceptual
resources, rather than learning that sentences have meanings that one
already had the conceptual resources to capture (Harman 1973). One way
of making sense of this is that learning language involves, in some sense
to be explained, internalizing linguistic structures which thereby acquire
certain functional roles.
(iii) Another motivation is given by Gilbert Harman (1973). Beliefs have common
structures and structured relations to each other, for example, the relation
between the belief p & q and the belief that p, which are essential to their role
in psychological explanation. That structure is language-like; hence, the
argument goes, if it is realized, there is a language of thought.
(iv) Hartry Field suggests (1978) that the only naturalistic way to account for
intentionality is the language of thought hypothesis. That beliefs involve
relations to propositional contents is to be explained in the two-stage model
we saw earlier, in (2).
Now, the rst two motivations are indeed reasons for taking the language of
thought hypothesis seriously, but, as far as I can see, they do not add support to
the explicative theory. For a proponent of the propositional attitude based theory
might agree that the further hypothesis of rst-order structure is attractive because
computational theories of mental processes are attractive, or because, in its thinking
in English version, it would nicely account for certain intuitions about language
learning. It is perfectly consistent to maintain that the inner states postulated by such
theories would have content by virtue of being beliefs in the sense explicated by the
propositional attitude based theory (as in (3)).
As for the latter two motivations, my account shows, I think, that the language of
thought based theory is not the only strategy for accounting naturalistically for
structural relations among ordinary ascriptions of beliefs, for they may be accounted
for in terms of second-order, rather than rst-order structure. That is, we can
generalize over conjunctive beliefs via their sentential indices, and speak of all beliefs
of the form BC(x, s1 & s2). They have in common their conceptual relations to other
appropriately indexed beliefs, and not (necessarily) their rst-order structure. And,
of course, the account I sketched shows that the language of thought based theory is
not the only conceivable naturalistic treatment of intentionality. The upshot is that
I do not know of an advantage for the language of thought based theory, that is, the
philosophical theory, other than the negative one that it does not require interper-
sonally ascribable functional roles.
Have I been unfair to the language of thought based theory in construing it as
requiring a language-like rst-order structure, rather than merely the relevant
second-order functional structure? I have simply assumed that the language of
thought based theory means by language of thought whatever is meant by it in its
application to theoretical psychology. As I have pointed out, if computational
processes are to explain what they are supposed to, there must be a suitable under-
lying syntactic, formal structure that involves recurring words etc. As regards the
thinking in English hypothesis, moreover, surely it is explanatory only if there are
inner states that have syntactic structure in something like the same sense in which
spoken sentences have syntactic structure. So if, in the language of thought based
theory, that is, the philosophical theory, language of thought is interpreted as in its
scientic explanatory roles, the theory cant be taken to require only a weaker,
second-order, functional structure for beliefs and desires.
Different functionalist frameworks for explicating propositional attitudes indi-

viduate them in varying degrees of ne-grainedness. On the approach defended by
Robert Stalnaker (1976), logically equivalent beliefs count functionally as the same
belief. On my account this is not the case; two mathematically true beliefs, for
example, can have different conceptual roles. So beliefs are individuated according
to, as we might call them, ne-grained functional roles. The language of thought
theory requires even morenamely, that the underlying states that realize those
functional roles themselves have a ne-grained rst-order structure. My account,
then, resembles Stalnakers in not requiring linguistic structure; it resembles the
language of thought theory in generating ne-grained individuation conditions.
3 Truth Conditions
On the propositional attitude based theory, general truth conditions are associated
with conceptual roles, that is, with certain second-order functional properties. Those
rst-order states that have those conceptual roles, which may or may not be linguistic,
then have whatever specic truth conditions are determined by their context and
their general truth conditions.
The language of thought based theory dispenses with those mediating interper-
sonal conceptual roles; and so truth conditions must in some way be attached directly
to inner sentences. What is needed is an account of p is the truth condition of s that
depends on some contingent relation between inner sentences s and possible states of
affairs p. The only naturalistic theories that I know of that attempt to specify such a
direct relation share the following idea.
A mental state represents or indicates a possible state of affairs p provided that, had
p been the case, p would have caused that mental state given the obtaining of those
conditions under which our cognitive system ideally functions. This is an attractive
idea for certain observational beliefs; their contents are a matter of their discrimina-
tive capabilities. The idea is then extended to all beliefs:
(9) p is the truth condition of s for x
iff
if p were the case, then p would cause a token of s in x, given the optimal
application of such and such cognitive procedures in a properly functioning
environment.9
Call this the ideal indication theory.
Naturally, the class of these cognitive procedures would have to be captured
without using the concept of truth. Something about their teleological function might
be invoked, but there are difculties with this, as Jerry Fodor has pointed out. He
9
This is virtually equivalent to the account given in Fodor (unpublished manuscript, Psychosemantics)
[referred to in Fodor (1987), p. xiii].
suggests that we simply enumerate the cognitive procedures; but that apparently
would make it a mystery why truth conditions matter at all, that is, what the unied
concept is. This problem has a more general scope than the ideal indication theory:
on any explanation of truth conditions in terms of the reliability of beliefs, the actual
and possible belief-tokens that are collectively supposed to be reliable would have to
be restricted to those produced in certain ways.
I leave that as an open question. I wish to point out certain respects in which the
ideal indication theory seems too strong, while weakening it suitably in natural ways
leads back to the propositional attitude based theory. The ideal indication theory
requires potential omniscience: the diligent application of the cognitive procedures
ought to lead one to every true belief of which one is conceptually capable. There are
at least two problems with this, one about discovery and the other about conrm-
ation. First, on the face of it, some theories that we accept emerged as hypotheses
from the idiosyncratic abilities, and the good luck, of creative individuals. That
there is a method that is the natural endowment of each of us, and that could have
generated those hypotheses, is an interesting but far from obvious thesis. Secondly,
might there not be hypotheses whose truth conditions are quite unproblematic, but
which, because of non-accidental physical facts, we could never conrm or dis-
conrm? On the ideal indication theory that is impossible. These problems suggest,
not that we give up the idea that truth conditions somehow depend on our
epistemic sensitivity to the facts, but that we weaken the requirements. One appro-
priate way of doing this, moreover, is suggested by an independent consideration,
as follows.
The ideal indication theory requires that every inner sentence s is individually
sensitive to its truth condition p. But this apparently conicts with a rather basic idea
about truth conditions, namely, that the truth condition of a new sentence is
derivative from the independently constituted semantic properties of its constituents.
In other words, it seems irrelevant to a complex sentences truth conditions what
its individual responsiveness to the facts is, independently of its semantic position
in the language. This is implied by the very idea that a language is a system of
representation.
But the systematic nature of truth conditions, and the idea that they derive from
the sensitivities of inner sentences to outer states of affairs, are of course compatible.
Two ways of combining the two conditions naturally suggest themselves. One is
that a systematic truth theory for xs language of thought is correct if it maximizes,
for the language as a whole, some sort of epistemic responsiveness to the facts, call
it E. That is, a systematic truth theory T is correct only if there is no other such
theory which makes inner sentences more, say, reliably informative than does T. So
we have:
(10) s means p in xs language of thought L

iff
there is a systematic truth theory T of L such that T maximizes E (informative

reliability, ideal indication) for x and T implies that s is true iff p.
This is evidently weaker than (9) and does not require omniscience. But it still
apparently makes the truth conditions of complex sentences dependent on their
collective epistemic responsiveness, and one might be inclined to think such truth
conditions should be still more derivative from the semantic properties of simpler
sentences. That might be captured as follows:
(11) s means p in xs language of thought L

iff
there is a systematic truth theory T of L such that T extrapolates in some
suitable way the truth conditions of complex sentences from those of simpler
sentences, T maximizes E for x, for some suitable collection of the simpler
sentences of L, and T implies that s is true iff p.
Naturally, various questions arise about (10) and (11). An important one is whether
any truth theory could satisfy those conditions uniquely. There is a signicant
contrast, in this connection, between a language of thought based theory that
incorporates (10) or (11) and the propositional attitude based theory earlier sketched.
Suppose there is no uniquely best truth theory that satises (10) or (11). That
would not quite undercut semantic realism, the thesis that putative semantic facts
correspond to real relations between thoughts and the world, for the relevant facts
about reliability, relative to some non-unique but restricted class of truth theories,
would still be real enough. But the question would remain as to what explains the
assignment of truth conditions that we favor. Suppose there are interpersonally
ascribable conceptual roles of the sort the propositional attitude based theory
requires. Then, we may suppose, among those truth-theories for as internal states
that maximize E, one assigns them truth conditions that match (in a way to be
explained) the purely disquotational truth conditions of those sentences of ours that
have (or express) the same conceptual roles as as states. Suppose a is a Twin-Earther,
and that some state of as has the conceptual role that I bathe in water has for me.
That state of as is true iff a bathes in XYZ (the account would go) because my
sentence is true iff I bathe in water, and a and XYZ are the corresponding compo-
nents of as environment. The non-context-sensitive matrix (so to speak) is disquo-
tational, and (the suggestion goes) that combination of context and disquotationality
selects, from those truth theories which maximally satisfy the conditions of say (11),
the truth theory we favor. But without interpersonal conceptual roles, there would be
no such indirect disquotational explanation of what secures uniqueness of truth
conditions for beliefs in general, if, that is, the conditions of (10) and (11) do not
already do so. And without the uniqueness of those conditions, it is difcult to see
what theory of systematic truth conditions might be available to the language of
thought based theory.
Another way in which interpersonally ascribable conceptual roles may make a

difference in the theory of truth conditions is this. The constraints on truth condi-
tions that we have been considering are individualist: they involve a persons own
competence. An attractive idea is to weaken this to allow truth conditions to be
determined on a society-wide basis. The (counterfactual) reliability, or increasing
verisimilitude etc., of interpersonal agreements that result from the collective appli-
cation of certain cognitive procedures might seem a somewhat more realistic sup-
position than the corresponding thing for every individual. But if a persons beliefs
acquire their truth conditions thus, there must be some way of individuating beliefs
interpersonally, independently of their truth conditions. If public language sentences
were the issue there would be no problem: we could say that xs sentence horses
dont y has the truth conditions that that sentence has in xs society at large. But we
are talking about sentences in the language of thought. Given that neurophysiological
individuation conditions are unavailable, that seems to leave functional or conceptual
individuation. The upshot would be that if sentences in the language of thought have
socially based truth conditions, they would have to have interpersonally ascribable
conceptual roles.
In yet another respect, a language of thought based theory as supplemented by (9),
(10), or (11) appears too strong. It is apparently incompatible with realismthat is,
the independence of the facts from our epistemic capabilities, the logical possibility
that our beliefs could be radically false regardless of how careful we are, that, indeed,
we might be brains in vats. Drawing an anti-realist conclusion, from (10), say,
proceeds via a reductio (patterned after an argument of Putnams (1981)). Suppose
I am a brain in a vat. Then, no doubt, theres a possible truth-theoretic interpretation
of my beliefs that makes them about my sense-data and renders them largely true on
that interpretation. According to (10) that interpretation is then the correct one. So, if
I am a brain in a vat, my beliefs would be largely true. But most of my beliefs are
incompatible with the belief that I am a brain in a vat. So if I am a brain in a vat, the
belief that I am a brain in a vat would, according to (10), count as false. It seems to
follow from (10), then, that the logical possibilities implied by realism do not obtain.
If you do not like the conclusion, that is a reason for preferring the propositional
attitude based theory to any theory that builds reliability in that way into the meaning
of true. On the former, true is dened in terms of a specic entrenched association
of general truth conditions with conceptual roles. One may then go on to explain that
entrenched association by pointing to what we take to be facts about the reliability of
beliefs in the real world. That leaves it open that there are possible worlds in which
our beliefs are largely falsethat is, relative to that assignment of general truth
conditions that we count as denitive of true. But, if interpersonal conceptual
roles are not available, you cant dene true for the language of thought by invoking
some specic assignment of truth conditions to a set of states that are explicitly
characterized in some systematically structured way. For without interpersonal
conceptual roles, no such detailed characterizations of the language of thought are
now available. The only apparent alternative is to quantify over truth-theories, in the
manner of (10) and (11), building reliability, or some similar property, into the
meaning of true. But that apparently has the anti-realist consequence I have noted.
These considerations do not imply absolute advantages for the propositional
attitude based theory; for that there are interpersonally ascribable functional con-
straints that yield appropriate individuation conditions for conceptual roles is per-
haps not self-evident. Yet it would be nice to have a philosophical theory of belief that
does not entail the very adventurous language of thought hypothesis. But, if the
propositional attitude based theory cannot be vindicated, and the language of
thought based theorys account of truth conditions is too strong in any of the three
respects mentioned, then I do not know what to say about prospects for a substantial
philosophical theory of beliefs and their truth conditions.
5
Names in Thought
Saul Kripkes by now well-known example of Pierre, who believes that London is
pretty, was advanced to show that something is seriously wrong with our ordinary
practice of belief ascription. Kripke then used this to question a certain classic move
against his Millian theory of proper names. I want to take issue with Kripkes analysis
of Pierre; nothing particularly puzzling about belief ascription is forthcoming. Nor
then can Pierre serve to defend Millianism in the way Kripke intends, but, as I shall
try to show, a judicious Millianism does not require such a defense. Yet the case of
Pierre does show something important about belief ascription, about the relation
between de dicto ascriptions of beliefs and how believers conceive of things, some-
thing potentially extendable to aspects of mental content other than names.
1 A Puzzle about Belief

Here is the (all too familiar) case.1 Pierre, a monolingual Frenchman, hears of
London, or Londres as the French call it, and believes that it is pretty. He would
say Londres est jolie. This entitles us to say
(1) Pierre believes that London is pretty.
Later he moves to London, unaware that it is the famous Londres. He becomes
bilingual and would say London is not pretty. So we may also say
(2) Pierre believes that London is not pretty.
He has in no ordinary way changed his mind about Londres, and thus (1) remains
true.
Our ascription (2) is got by a disquotation principle: if a normal English speaker,
on reection, sincerely assents to p, then the speaker believes that p2 (p. 248). Our
(1) is got by the French version of the disquotation principle and the principle of
translation, which is that correct translation preserves truth. Kripke is surely right
that our ordinary principles imply that (1) and (2) are both true.
1
Kripke (1979). Page references in parentheses will be to this paper.
2
With Kripke, I shall ignore complications arising from indexicals.
NAMES IN THOUGHT
Notice that this pair is not merely true de re; that is, it is not just that Pierre has a
pair of contradictory beliefs of one and the same city. By conventional standards
these ascriptions are true de dicto (we shall return to the question what this means).
Nor is there any equivocation in our use of London in (1) and (2).
What then is puzzling? Kripke writes . . . we must say that Pierre has contradict-
ory beliefs. . . . But there seem to be insuperable difculties with this. . . . We may
suppose that Pierre is a leading philosopher and logician. He would never let
contradictory beliefs pass. . . . He cannot be convicted of inconsistency: to do so is
incorrect (p. 257).
If this is puzzling, at least there is so far no contradiction in our ascriptions. But
Kripke takes the case to yield a contradiction as well, via the strengthened disquota-
tion principle: A normal English speaker who is not reticent will be disposed to
sincere reective assent to p if (and only if) he believes that p (p. 249). In the
example, Pierre is not prepared to assent to London is pretty; so by contraposition
we have:
(3) Pierre does not believe that London is pretty.
Again there is no reason to say that any equivocation in London is involved.
If Kripke is right about our principles, they lead not only to a puzzle but also to a
contradiction.
These consequences are welcome to Kripke for an ulterior reason. He wishes to
defend a strong form of his Millian theory of proper names, in order to make a
weaker Millian thesis immune to a certain standard attack. That strong Millianism he
calls pure; it holds that proper names are Milliani.e., have no semantic function
other than to rigidly designate their referentseven in belief ascriptions. There are
actually two standard anti-pure-Millian arguments, which Kripke does not clearly
mark off, so let me do that now.
Suppose that
(4) Jones believes that Cicero is bald.
On an ordinary way of reading belief ascriptions it may, despite Tullys identity with
Cicero, also be false that
(5) Jones believes that Tully is bald,
and true that
(6) Jones believes that Tully is not bald,
and of course true that
(7) Jones does not believe that Tully is bald.
The rst anti-Millian argument is this: if names were purely Millian, a substitution
principle would apply and there would be no reading on which (5) would not be
NAMES IN THOUGHT
truedespite (6). And the second argument is this: if names were purely Millian, we
would have the contradiction (5) + (7).
The parallel with Pierre is that apparently unacceptable pairs similar to (5) + (6)
and (5) + (7) can be got without pure Millianism, namely (1) + (2) and (1) + (3);
so why blame pure Millianism for the former results? The source may lie deeper.
2 Is Anything Here Puzzling?

Finding a contradiction in the case of Pierre is rather odd, that is, insisting that
(3) Pierre does not believe that London is pretty
is implied by principles we actually use. In fact, the strengthened disquotation
principle appears to me simply misformulated.
There is a contradiction in our ordinary principles only if they lead to (1) and (3)
without equivocation about the kind of information conveyed by that-clauses. There
is no contradiction simply if there is a way of insisting on (3) based on Pierres refusal
to assent to the actual words London is pretty, and also a reading of (1) that is more
permissive and requires merely that Pierre have uttered something that London is
pretty translates.
Evidently we accept the more permissive standard in asserting
(1) Pierre believes that London is pretty,
for that depends on translation. But the strengthened disquotation principle that
leads to (3) ignores that fact. If we want to formulate a principle that does not
overlook our having counted (1) as true, then it should go something like this:
A normal English speaker who is not reticent will be disposed to sincere reective
assent to p or to something that p translates if (and only if) he believes p.3 That is
plausible and it does not generate (3). For Pierre is disposed to assert a translation of
London is pretty, thus satisfying the disjunctive consequent of the principle; and
so (3) does not follow by contraposition as it does from Kripkes version. Pierre
is simply a counterexample to Kripkes strengthened disquotation principle. His
version sounds all right until one is shown, by Pierre, that bilingual English speakers
may coherently assent to a sentence in another language whose translation into
English they do not assent to. Surely nothing in our practice implies that we operate
with Kripkes principle rather than the suggested one.
3
One might think that translation is not a suitably basic semantic notion, and that what is needed is
something like this: A normal speaker is disposed to assert, in some language, a sentence that expresses p if
(and only if) he believes p. But this quite arbitrarily and counterintuitively rules out the possibility of the
impure Millian theory of Section 4. For it requires that we say, in the case of Jones, that if (4) is true (7)
cannot be true. The proposed emended strengthened disquotation principle does not have that result.
I explain in Section 3 why ordinary translation has a crucial role to play in de dicto that-clauses, and why
Tully does not count in that-clauses as a translation of Cicero.
NAMES IN THOUGHT
When Pierre is unwilling to assent to London is pretty, one may casually assert
(3). Suppose Jean-Claude, bilingual and more knowledgeable than Pierre, remarks
but what you say is not true: he does believe that London is pretty, for he assents to
Londres est jolie. At this point either one says I am using (3) in such a way
that unless Pierre will assent to London is pretty I will count (3) true and thereby
rejects (1), or one accepts correction from Jean-Claude, agreeing that translation is
sufcient to justify (1), and thereby rejects (3). The last thing one says is but the
conventions of English require me to assert both (3) and (1) without equivocation.
I'm sorry, Jean-Claude, but there it isIm forced to contradict myself. There is no
motivation for that.
Suppose Kripkes strengthened disquotation principle had been formulated in a
different context: we were not looking for problems but were simply trying to
formulate our practice, including our practice, such as it is, of ascribing beliefs to
speakers of English on the basis of their sincere assertions in other languages. And
suppose someone then invented the case of Pierre as an objection to that version of
the strengthened disquotation principle: you have ignored the possibility that
N believes that p may be true even if N speaks English and will not sincerely assent
to p, if N sincerely assents to something in another language which p translates.
Surely emending the principle would then have been a matter of coursedon't
forget those odd translation cases.
What would count as evidence that Kripkes strengthened disquotation principle
(given that translation justies (1)) reects our practice better than the emended
version? Presumably only this: when presented with Pierre in a real-world situation,
English speakers would feel constrained either to avoid that-clauses or to contradict
themselves unequivocally, a quite improbable state of affairs.
Adapting another example of Kripkes (used by him for a different point) might
seem to make his case better (p. 268). In Hebrew, Ashkenaz and Germaniah
are both names for Germany; they each translate into English as Germany. Now,
let . . . stand for the Hebrew rendition of is cold, and suppose a Hebrew speaker
would assent to Ashkenaz . . . and dissent from Germaniah . . . So it can be asserted
in Hebrew that
(9) She believes that Germany is cold
using Ashkenaz . . . And it may also be asserted in Hebrew that
(10) She does not believe that Germany is cold
using Germaniah . . . Ordinary translation into English yields (9) and (10). Now in
forming the Hebrew version of (10) all that was needed was the Hebrew version of
the emended strengthened disquotation principle. Hence it seems the emendation
would not after all avert contradiction.
Does this show that something is wrong with our ordinary principles of belief
ascription? I doubt it. The more plausible explanation is that the employed procedure
NAMES IN THOUGHT
of translation is not ours. There is no contradiction in the Hebrew belief ascriptions;

why tolerate one in the English translation? (It would be odd to say we are con-
strained to translate here in a way that generates a contradiction in English when
there is none in the original.) If you want translation always to preserve consistency
in whole discourses, translate distinct names by distinct names. This matters espe-
cially in translating contexts in which the choice of names is sensitive to which names
were involved in the situation being reported, an utterance or belief, say. If the
Hebrew versions of (9) and (10) had occurred in a novel, an English translator
would not just baldly produce (9) and (10).
That no straightforward English translation in this case yields an unproblematic
pair is a reection of the fact that translation is cramped by the resources of the home
language. But precisely the same problem arises for direct quotation. Usually in
translating novels one translates spoken or written discourse. So suppose in a Hebrew
novel it is said that the heroine writes down Ashkenaz . . . and fails to write down
Germaniah . . . The translation would be contradictory were Germany used for both.
But this does not show that something is amiss with our practice of direct quotation:
the contradiction would arise from inadequate translation. Of course extending the
point about direct quotation to belief ascription depends on the latters being sensitive
in some suitably analogous way to the names involved in the beliefs reported. I shall
return to how this may be so even though belief ascription is oblique and not direct.
That the contradiction depends on translation can be seen from another example
of Kripkes which involves no translation. Suppose Peter has heard of Paderewski the
great pianist and Paderewski the postWorld War I prime minister of Poland,
wrongly assuming that they are not the same. From Peters English assertions we
may conclude both that
(11) Peter believes that Paderewski had musical talent
and that
(12) Peter believes that Paderewski had no musical talent.
Thus we get Kripkes rst result. But neither version of the strengthened disquotation
principle yields
(13) Peter does not believe that Paderewski had musical talent.
(Nor does Kripke say so.)
Thus a contradiction arises only if either (i) we apply Kripkes principle in a case
where we have got the rst halfx believes that pby translation and then get the
second half by ignoring that the rst half can be true even though x is not disposed to
assert p itself, or (ii) we translate, as in the Ashkenaz example, without due
sensitivity to the names involved.
We are left with Kripkes rst puzzle: even though Pierre is not being illogical, our
principles imply (1) and (2), that Pierre has beliefs which are inconsistent relative to
NAMES IN THOUGHT
their de dicto ascriptions. Does this show there is something wrong? Well, the
surprise which the result engenders does probably show that we have been making
some wrong assumption about the function of de dicto belief ascriptions. Pierre is
indisputably consistent in how he conceives things. If you assume it is the function of
de dicto ascriptions to capture how believers conceive things, then Pierre is a problem.
The moral is that the assumption is false: some de dicto belief ascriptions, involving
proper names, do not reect in any exact way how believers conceive things, and
therefore do not purport to capture that in respect of which Pierre is indisputably
consistent. This is no defect; such that-clauses have a different function, one that
tolerates the joint truth of (1) and (2). I shall return to the question of what this
function is, and its relation to how Pierre conceives things and his consistency therein.
Kripke rightly insists that one does not answer his problem by offering a rede-
scription of Pierre on which no puzzle arises. But that is not my point. The
contradiction does not arise from any principle there is reason to suppose we operate
with. And as for the joint truth of (1) and (2), one has to ask: just what is supposed to
be paradoxical herewhy not simply accept that both are true? The answer But
Pierre is being consistent brings into the open a presupposition of puzzlement
here, namely, the incorrect assumption that a de dicto that-clause should capture
how a person conceives things, i.e., capture that factor in respect of which Pierre is
being consistent.
3 Jones and Pierre

Pure Millianism asserts that there is no reading of
(5) Jones believes that Tully is bald
on which it is not entailed by
(4) Jones believes that Cicero is bald
together with Tully = Cicero, for pure Millianism says that names are Millian on
every occurrence. But that means that pure Millianism entails that either there is no
reading of
(7) Jones does not believe that Tully is bald
on which it is true in the case of Jones, or a contradiction is true.
That then is one major difference between Jones and Pierre. Our ordinary prin-
ciples do not entail the disjunction in the case of Pierre. They do not exclude a strict
way of reading (1) on which (3) may be counted as true; but then they do not force us
also to assert (1) on that reading. And if we assert (1) on a more permissive sort of de
dicto reading, no well-formulated strengthened disquotation principle forces us also
to assert (3).
NAMES IN THOUGHT
But if we allow that there is a de dicto reading of (5)thus yielding (7)and hold
for pure Millianism we have no choice but to contradict ourselves. And so the simple
fact is that pure Millianism must arbitrarily exclude that de dicto reading of (5) that
makes it false and (7) true (hardly a surprise). The case of Pierre, in which no
unequivocal contradiction is implied, does not get pure Millianism out of this jam.
To argue for pure Millianism on the grounds that despite appearances there is no
de dicto reading of (5) distinct from its de re reading would make the case of Pierre
completely irrelevant, and we should have to have a different discussion.
Let us turn to the other argument against pure Millianism. In the case in question,
(4) Jones believes that Cicero is bald
is true while (5) is false on one of its ordinary readings. But pure Millianism implies
that there is no reading on which (4) is true and (5) is false. Prima facie this implies
that pure Millianism is false.
Kripkes suggestion that the case of Pierre somehow deects this consequence is
not very plausible. The argument is not merely that pure Millianism would lead to an
inconsistency in Joness beliefs, viz. (5) and
(6) Jones believes that Tully is not bald.
The argument is more basic: there is a perfectly good reading of (5) on which it is
straightforwardly false despite the truth of (4). And that is that.
Still, we do well also to note that the resulting inconsistency in Joness beliefs(5) and
(6)would be quite different from the merely apparent inconsistency in (1) and (2),
which result from translation and disquotation, and from (11) and (12), which in the
case of Peter result from disquotation alone. You cannot get (5) by disquotation;
Jones does not believe that there is another Tully. Nor would Jones say Tullius est
chauve.
Are we being arbitrary in not counting Cicero as a translation of Tully, thus
getting (5) from (4) as we got (1) from the French report of Pierre? Should we do so?
There is no should here. The question is whether we have a coherent practice, in
which we accept (1) by ordinary translation, and yet reject the substitution principle
and deny (5). I shall now try to say what that coherent practice is.
I have supposed that (1) and (2) may unparadoxically be true together in the case
of Pierre, on a de dicto reading, that is, a reading of (1) that requires more than that
London and Londres are coreferential, yet a reading on which we do not count (5)
and (6) as conjointly true. There must then be a relevant difference between Cicero/Tully
and Londres/London.
Kripke writes that in the Pierre case the situation of the subject is essentially the
same as that of Jones with respect to Cicero and Tully (p. 253). Now, it is true that
Joness and Pierres attitudes toward the relevant sentences are parallel. The sentences
Londres est jolie and London is not pretty are in different languages, unlike Cicero
is bald and Tully is not bald, but this is irrelevant to genuine similarities in the
NAMES IN THOUGHT
cognitive roles of those sentences in Pierre and Jones. In each case the two are
assented to without illogicality, even though the predicates are cognitively contra-
dictory and the names coreferential.
Our topic, however, is belief ascriptions and not simply the sentences our subjects
would assent to and dissent from. It would be too quick to move from the parallel just
noted to the conclusion that (5) and (6) are parallel to (1) and (2). The fact is that
conventional translation generates (1) from Pierre croit que Londres est jolie and
does not generate (5) from (4), and that is a relevant difference.
Let us recall the situation with (1) and (2). I wish to say there is a reading that is not
de re but on which they are unparadoxically both true. Of course that kind of de dicto
reading is not so strict that (1) requires that Pierre assent to London is pretty. So it
must be a reading that is permissive enough to be secured by conventional transla-
tion, and its avor may be gathered from this. Suppose Pierre would assent to
Cicron est chauve, so that the French sentence Pierre croit que Cicron est chauve
is true de dicto, while he would not assent to Tullius est chauve. We then have by
translation Pierre believes that Cicero is bald, on a de dicto reading, while Pierre
believes that Tully is bald is not true on a de dicto reading. It is in that sense that (1) is
true on more than a de re reading: (1) correctly translates a sentence that is true on a
de dicto reading.
But is this translational difference between Londres/London and Cicero/Tully
still not arbitrary? In fact, no. We are supposing that Londres does not occur in
English that-clauses, that it does not move from the sentences that serve as the basis
of our de dicto reports into the reports themselves. That is why translation matters:
it is a relation between expressions of the belief-ascribing language and expressions
that are being treated as unavailable but as having the former as proxies in the belief-
ascribing language.
This is why certain de dicto belief ascriptions that use London can be made true
by sincere utterances involving London and those involving Londres, while de dicto
belief ascriptions using Tully are not made true by sincere ascriptions using Cicero.
I am trying to describe and not to legislate. It is natural to understand (4) and (5) in a
way that permits the former to be true with the latter false. That is what I am calling
the de dicto reading. But it does not require direct quotation; for we allow transla-
tions to qualify, in a way in which merely coextensive reports do not. If Jones
sincerely asserted Cicron est chauve, there would be a way of understanding (4)
and (5) on which the former but not the latter is thereby true. But if Londres were as
readily available in English as Tully, then (1) would no more be, as it is, a sensitive de
dicto report of Pierres disposition to assert Londres est jolie than (5) is of Joness
disposition to assert Cicero is bald. Because Londres is not available we allow a
proxy, and so conventional translation makes a difference.
The basis of this is hardly profound. We have a practice of de dicto belief
ascription, that is, of ascribing beliefs in a way sensitive to how believers would
express things. But we are not fanatics; among other adjustments, we allow
NAMES IN THOUGHT
translation and try to be as sensitive as possible to how beliefs would be expressed

given that we (most of us) do not employ foreign names if we have English proxies
for them. This is just a matter of our practice, nothing more. And the whole point of
this practice entails resisting the substitution principle.
4 Pure Millianism
One wonders whether there is a good reason for a Millian to be pure, that is, to hold
the substitution principle. I see our de dicto practice as innocuous, but is there
perhaps a theoretical reason to revise it?
There may be a perceived threat to impure Millianism, that is, to the position that
outside epistemic and quoted contexts the sole semantic function of proper names is
to rigidly designate what they designate. I can think of two arguments from the denial
of pure Millianism to the denial of impure Millianism which could implicitly be
operative here.
(1) De dicto that-clauses do not have the function of reporting the words a person
would use in expressing a thought. For we translate a foreigners words in
reporting his beliefs. Nor, given the denial of pure Millianism, is it their function
merely to capture the extensional properties of a thought. So (the argument goes)
terms in that-clauses must have further non-quotational and non-extensional
semantic properties. But how do terms acquire these further semantic properties?
They do not pop into existence within that-clauses, and so they must be present in
some form outside that-clauses. This leads to the Fregean idea that names have
semantic properties over and above their referents in non-epistemic contexts; and
thus the denial of pure Millianism leads to the denial of impure Millianism.
(2) It is the function of a belief ascription to capture the content of what would be
asserted by an expression of the reported thought. But on an impure Millian theory
what is asserted by Cicero is bald is identical with what is asserted by Tully is
bald. But then (the argument goes) the substitution principle would have to be
valid. Thus impure Millianism requires pure Millianism, and denying the latter
would again apparently lead to some sort of Fregean theory.
But both arguments are misguided. Here is an impure Millian reply to (1). What
makes a belief ascription de dicto is that it is sensitive to how the belief would
be expressed. But this is consistent with belief ascriptions not being quotations,
for they involve the production of proxies. It may still be the case that the only
further semantical properties captured by belief ascriptions are extensional. That
is, de dicto that-clauses may capture beliefs references together with how they
would be expressed either directly or by proxies, and that is that. There is no
need for Fregean senses outside epistemic contexts because there is no need for
them within.
NAMES IN THOUGHT
That in belief ascription we translate and do not quote could suggest that it is the
function of that-clauses to capture the common meaning of their embedded sentences
and other sentences, in other words that that-clauses capture synonymy classes. But,
as various writers have noted, that-clauses have a reading such that substituting
synonyms is not truth-preserving: Henry may believe that Waldo is an undertaker
without believing that he is a mortician. Intuitively that is as acceptable as the non-
substitutivity of Tully for Cicero; the inclination to treat them differently has
probably been due not to ordinary language but to the inuence of Fregean theory.
And this suggests that non-substitutivity derives from that-clauses sensitivity (not to
Fregean senses but) to how beliefs would be expressed, as the suggested impure
Millian theory has it.
An impure Millian reply to (2) may be equally straightforward. There is nothing
incoherent in this conjunction: (a) Jones believes that Cicero is bald has a reading
on which substitution fails; (b) on that reading the that-clause captures what would
be asserted by Cicero is bald; (c) that is identical with what would be asserted
by Tully is bald. They are consistent because (b) does not say that the reading
captures only that. The impure Millian may say that that-clauses have two func-
tions: they capture the singular proposition that would be asserted, and they
capture (or make a stab at capturing) directly or by proxy how the belief would
be expressed.
5 Psychological Content and de Dicto That-clauses

Let us call the consistency in belief that intuitively we nd in Pierre his psychological
consistency. The joint acceptability of (1) and (2) then implies this: either
(a) psychological consistency is consistency in a kind of contentcall it psycho-
logical contentand beliefs de dicto ascriptions do not (always) capture their
psychological content;
or
(b) psychological consistency is not as such consistency in a kind of content;
perhaps it is a kind of syntactic consistency.
Now I believe there is good reason to opt for (a), but I shall not argue for it here.
Pierres consistency is a consistency in how he conceives or represents things as
being, and that, it seems to me, is not a matter of which syntactic objects constitute
his thoughts. There are at least two frameworks within which to understand psycho-
logical content. On the one hand, it might consist in a kind of truth conditions. Then
those aspects of Pierres thoughts to which his words Londres and London
correspond would be non-rigid, in the sense that they determine distinct uniqueness
conditions, derived in each case from the totality of what Pierre believes about the
city in question, perhaps along the lines David Lewis has suggested (1981). On the
NAMES IN THOUGHT
other hand, psychological content might consist in conceptual role and not truth
conditions: Londres and London have different conceptual roles for Pierre, and it is
relative to them that Pierres beliefs are consistent (I rather favor this).
The point of course is not that our London, as we use it in (1) and (2), is
equivocal. It is perfectly consistent with both ways of theorizing psychological
content even that London occurs in those belief ascriptions as a univocal rigid
designator, in the manner of the impure Millian position above. The point is that
the psychological content of beliefs may not as such be reected in their de dicto
ascriptions. The latter have another function, and although it is not my topic here it
may help to gesture toward it.
A de dicto ascription captures, normally, a combination of how the belief would be
expressed and what we might call its social truth conditions. This means, as we should
by now expect, that any truth conditions that might gure in psychological content
will diverge from the truth conditions captured by that-clausesa difference as it
were between personal and social truth conditions. It was the expectation that
these should coincide that led to the feeling of paradox about (1) and (2); but as
soon as we see that there are two elds of description herethat which is relevant to
accounting psychologically for Pierre, his consistency etc., and that which is relevant
to capturing the expressive and social dimensions of Pierres beliefthe feeling of
paradox subsides.
The contradictoriness of what (1) and (2) ascribe is simply a possibility that our
practice of de dicto ascription allows, even though it does not turn up often.
It happens because a belief ascriptions being de dicto is a matter of its capturing
relatively supercial facts about the beliefhow it would be expressedand not
(exactly) cognitive facts about it. If one wishes to be as close in ones reports as
possible to how Pierre or Peter would express their beliefs and not to how they
conceive thingsif, that is, one is simply engaging in ordinary de dicto ascription
then one will produce ascriptions that make those beliefs contradictory. But the
decision to look toward how the belief would be expressed was precisely a decision
not to focus on those aspects of Pierres and Peters beliefs that constitute their real
consistency. This is no aw in our complex practice.
6 Millianism about Names in Thought

As I have noted, there is an impure Millian position on which names in de dicto belief
ascriptions have the dual function of rigidly designating an object and adverting to
how the belief would be expressed. This position is compatible with holding that the
psychological content of, as it were, names in thought is non-rigid or descriptive.4
4
For a position on which names are non-rigid in thought and Millian in the public language,
see Schiffer (1981a).
NAMES IN THOUGHT
If psychological consistency is a matter of consistency in the content of thoughts,

then we cannot take names in thought to be simply Millian, in the sense that what
they designate exhausts their contribution to content. For then Pierres beliefs
would not be psychologically consistent. But if Pierres consistency is a syntactic
matter, there is no reason in what we have seen not to take a Millian view of names
in thought.
What should we say if we regard thoughts as having psychological content that is
constituted by their conceptual roles? Londres and London would then differ for
Pierre in their semantic properties, if semantic covers that sort of content. And then
we would have to count the occurrence of names in thought as not Millian. But was it
the spirit of the Millian thesis to deny that co-designative names can differ in their
conceptual roles? Presumably not. So there is no clear answer to the question Given a
conceptual role theory of content, are names in thought Millian?
But the conceptual role theory certainly leaves it open that names in thought are
rigid designators. That is compatible with taking them to have those further seman-
tical properties, their conceptual-role-contents, and thus not to be Millian in the
strictest sense. If there is a substantive question here it is whether Lewiss view (1981)
can be made out, that is, whether names in thought can be systematically assigned
general non-rigid truth conditions in some way that is essential to psychological
explanation. (It should be clear that it is not implied by such a view that names in
that-clauses have such conditions as their meanings.)
7 The Lessons of Pierre

Perhaps the most direct lesson is that (some) de dicto ascriptions do not merely assert
a relation between a person and a proposition. The failure of the substitution
principleno novelty, thisevidently implies that Jones believes that Cicero is
bald does not assert merely a relation expressed by believes between Jones and
that singular proposition which is true at any possible world in which our Cicero is
bald. But the case of Pierre shows further that it is not the function of the likes of that
London is pretty to denote a non-singular proposition the F is pretty. For the only
plausible motivation for such a description-theoretic account of names in that-
clauses is that a non-singular proposition captures the psychological content of the
ascribed belief (and that that is a descriptive condition). But the univocality of
London in (1) and (2) would then entail that Pierre is being psychologically
inconsistent, which he is not. Naturally this does not show that Pierres beliefs do
not involve, in some underlying way, relations to a pair of propositions the F is
pretty and the G is not pretty; but given that London is univocal, it is not the
function of those de dicto ascriptions to capture such facts.
The case of Pierre has repercussions concerning the nature of psychological
explanation. It is standardly assumed that psychological content, as that is individu-
ated in its role in psychological explanation, is captured by de dicto that-clauses.
NAMES IN THOUGHT
(This is not unnatural, for the failure of substitution in the case of Jones can
seem to be explained by a difference in how he conceives the object of his beliefs.)
This assumption is used to draw substantive conclusions. So we nd arguments such
as this: a blind person and a sighted person may both believe that cats have tails
(de dicto univocal reading); consequently the sighted persons ability to recognize cats
cannot be essential to the psychological content of his beliefs about catscannot be
constitutive of his concept of cats.5 And arguments such as this: a person who
misunderstands a term like arthritisperhaps he thinks it can occur outside the
jointsand a person who understands it perfectly may nevertheless both believe that
x has arthritis; consequently the idiosyncrasies of how a person understands a term
cannot be constitutive of the psychological content of his beliefs, which is determined
rather by social facts.6
I reply to these attacks on functionalist, individualist, theories of psychological
content elsewhere.7 The case of Pierre suggests a further extensive series of cases not
involving proper names, which show that these arguments correct premisesthat
such and such de dicto ascriptions are truedo not imply their conclusions. It is
already obvious from the limited example of Pierre that psychological content has a
life of its own independent of that-clauses, and that holds across the board, for
general terms and not merely for proper names. This deeply affects the topic of
psychological explanation.8
5
Cf. Stich (1983), pp. 6668 and Schiffer (1987), ch. 2.
6
Here of course I refer to Tyler Burge (1979).
7
Loar (1998a) and (1988b) [Chapter 8 in this volume].
8
For advice on an earlier draft, thanks are due to Stephen Schiffer, Hartry Field, and Richard Warner.
6
Truth beyond All Verication
I
In perhaps the most fundamental sense of realism, a realist about certain statements
holds their truth or falsity to be independent of our ability to verify or to falsify them.
This does not imply that we are not in fact in a position to verify or to falsify them,
but that it could happen that they were true or false even though we were not in that
position. Thus the idealist thesis that reality is entirely mental, non-material, is not in
itself incompatible with realism. Consider Berkeleys theory that the truth about
ordinary objects is a matter of perceptions in the mind of God; if it is also held that
what occurs in Gods mind is not dependent on our ability to verify it, then the theory
is realist in the relevant sense. Is this not an eccentric use of realism? Not at all, for it
directly reects certain central concerns in epistemology and in the theory of concept
formation. For Berkeley (in another frame of mind), Kant, the vericationists, and
recently Michael Dummetts anti-realist, two questions about realism are thought to be
unanswerable: if the reality about which apparently we think and speak were consti-
tuted independently of its epistemic accessibility to us, then (1) how could we know
about it? and (2) how could we have a bona de conception even of its possibility? (It is
the latter question with which Dummett is primarily concerned.) Granted that a non-
mental material world was standardly a component of realism, its interest in epistem-
ology and in the theory of concept formation lay chiey in the further thesis of such a
worlds being constituted independently of its veriability by us.1
This is largely Michael Dummetts conception of realism,2 and it is Dummetts
challenge to realism, conceived thus, that I wish to discuss and reply to here. Realism
1
As regards concept formation, this applies fairly, I think, to Hume, Kant, and the vericationists, but
perhaps not obviously to Berkeley. Humes discussion of our conception of the independence of bodies (Treatise
(173940), Bk. I, Pt. IV, Sec. II) seems applicable to any putative conception of independently existing things,
whether mental or non-mental. For Kant the fundamental issue is the verication-transcendence of any sorts of
facts; whether they are mental or non-mental would seem irrelevant. But for Berkeley, the central concept
formational problem for materialist realism (apart from the inconceivability of primary qualities without
secondary qualities) is that an idea can be like nothing but an idea. So perhaps Berkeley could say that the
verication-transcendence (for us) of Gods ideas does not preclude our having a conception of them via their
resemblance to our ideas. Evidently this is not an answer that would satisfy a vericationist.
2
Cf. Dummett (1978), especially the Preface and these essays: Realism, The Philosophical Basis of
Intuitionistic Logic, The Reality of the Past; Dummett (1976); Dummett (1977), pp. 18, 36089;
Dummett (1982).
TRUTH BEYOND ALL VERIFICATION
is of course to be understood as about some specic class of statements, and my

concern is realism about the natural world. This allows that some apparently
descriptive statements about the natural world, for example about secondary qual-
ities, may require an anti-realist interpretation. The applicability of my defense of
realism about the natural world to other classes of statement will vary; it will apply to
statements about the distant past but not directly to mathematics.
Dummett poses the question as one in the theory of meaning, as about our
capacity to understand statements whose truth transcends their veriability. The
same question can be posed in terms of non-linguistic mental content, about our
capacity to form conceptions of states of affairs that transcend our capacity to verify
them. Realism is about understanding and concept formation, and not merely about
truth, because there is no point in holding that truth is independent of verication
unless it is clear that an adequate theory of concept formation would account for
genuine conceptions of unveriable states of affairs. Dummetts challenge has two
stages. Can a capacity to understand statements whose truth conditions transcend
their veriability conditions be manifested in a persons behavior? If such under-
standing is not manifestable in behavior, can we reasonably take the framework in
which we communicate to be coherently realist? Dummett is negative at both stages.
With regard to the rst, he argues that whatever aspects of our conceptions are
manifested in behavior simply belong to their assertion or verication conditions.
No grasp of truth conditions that transcend such verication conditions does or can
then reveal itself. Here, I believe, the realist has an effective reply. But with Dummetts
second negative position I am completely in agreement. If a realist understanding of
statements were not manifestable in behavior, there would be something wrong with
realism. This depends on neither behaviorism nor vericationism; if ascriptions of
understanding or conceptualization were not conrmable on the basis of behavior,
there would be no reason to take them seriously.
After clarifying what I regard as the central issues, I shall lay out three possible
lines of realist reply. Two of these I shall reject, but the third reply I shall defend at
length. It clashes outright with a fundamental assumption of Dummettsnamely,
that realism requires a truth-conditional theory of meaning, that is, of understanding.
I shall argue instead that realism can be supported by a non-truth-conditional theory
that equates meaning with holistic conceptual roles, and I shall respond to some of
Dummetts negative reections on holism.
It is a complex question how the concept of truth is related to the realist/anti-
realist dispute. On one view the anti-realist eschews the notion of truth altogether,
substituting assertability or verication conditions for truth conditions. This could
seem prima facie motivated if, as Dummett maintains, only assertability conditions
are manifested in behavior. But Dummett has also suggested that the anti-realist
could accept a notion of truth, one constrained by assertability or veriability, with
no sense given to truth conditions that transcend verication conditions. It is not
necessary to discuss these two anti-realist positions separately. Their common part is
this: there is no bona de concept of truth on which the truth (conditions) of

statements can transcend their assertability or veriability (conditions). The question
is whether a realist demonstration of the contrary can be made out.
Dummett takes the realist to be committed to some substantive notion of truth,
over and above the purely disquotational true, that is, the intralinguistic true such
that s is true is a metalinguistic variant of s. But I do not think that realism, as
characterized so far, needs more than the disquotational true. This may sound odd,
preposterous even, if one takes realism to require some chunky notion of truth that
involves objectively determinate semantic correspondence relations between words
and facts. A central point of this paper is to give a different picture. Suppose I say it
could happen that the star Sirius has ve planets even though we could never be in a
position to verify that Sirius had ve planets. Apparently that makes me a realist
about that possible state of affairs, but I have used no concept of the truth of the
statement that Sirius has ve planets. Suppose that I now wish to assert, of all states of
affairs of a given sort, what I asserted about Siriuss having ve planets. The
disquotational true is our device for generalizing such assertions: for all s of such
and such class, it could happen that s were true although s were unveriable. In a
certain sense, of course, this does commit the realist to a substantive notion of truth,
for it implies that truth can transcend veriability. But that does not prevent it from
being the disquotational true. That that is sufcient for at least one thesis that
Dummett has disputed is shown by this: his anti-realist cannot allow, for disquota-
tional true, that it is possible for s to be true but unveriable. I shall return to the
question whether there are further reasons for which realism might require a more
semantically substantive concept of truth or reference.
Let us keep in mind that the realist must indeed produce something in the theory
of meaningviz., an account of how one is able to understand s is true but
unveriable. But that by no means implies that the realist must produce a theory
of meaning in which there is a notion of truth over and above the disquotational
true. It seems sometimes to be thought that were the disquotational true the only
legitimate conception of truth, that would in itself have anti-realist consequences, as
though it kept reality at the level, so to speak, of thoughts and sentences. But the
right-hand side of is true iff is not in general about thoughts or sentences. The
disquotational true may be seen as a trim way of speaking of the obtaining of states
of affairs: to say that snow is white is true is to say in effect that the state of affairs
that snow is white obtains. Realism is just the thesis that certain states of affairs
could obtain without being veriable; the disquotational truth predicate enables us
to achieve the effect of that without quantifying over such entities as possible states
of affairs.
It is not necessary for an anti-realist to maintain that s is true is equivalent to it is
veriable that s, for it may not be held that the latter is sufcient for the former.
A criterial theory, for example, might allow that defeating conditions could always
emerge, however much evidence we have for s. But I suppose that an anti-realist
about a class of statements s will hold that s is true entails s is veriable, at least in
the sense that there is no understanding of possible on which it can coherently be
maintained that it is possible that s is true and s is unveriable. Of course there is not
a unique notion of veriability. For Dummett, understanding s consists in mastery of
verication conditions such that, were s true, I would now be in a position to realize
those conditions. But there is also the more permissive Peircean notion of veriability
in the long run of ideal scientic inquiry. Naturally the realist denies that truth entails
even Peircean veriability.
Realism is a modal claim: it does not say that we are actually not in a position to
verify such and such true statements, but that it could be the case that we were not in
such a position, even given ideal inquiry. This places the realist position I shall defend
in a somewhat oblique relation to what Dummett counts as realism. His discussion
focuses on statements that are undecidable, relative, that is, to normal unskeptical
standards: undecidable statements in mathematics, statements about the past for
which no present evidence pro or con exists, statements about others mental states
for which no behavioral evidence pro or con exists. The modal conception covers
realism about such statements; for that certain statements are true or false although
unveriable or unfalsiable implies that they could be true but unveriable. But the
modal formulation permits a realist/anti-realist disagreement also about statements
that are by normal standards veriable or falsiable, for example, statements about
the present primary qualities of observable physical particulars. Dummett apparently
regards such statements as immune to realist/anti-realist disagreement, but I do
not. The point of course is not epistemological: it is not that such statements are not
conclusively veriable and therefore like the statements Dummett regards as
subject to realist/anti-realist contention. There is a more fundamental ground
for such a contention, concerning whether it could (logically could, not epistemo-
logically could) happen that they were true even though not, as in fact they are,
veriable by us.
The core of the realist/anti-realist dispute in the modern tradition is the modal
question; for it concerns the current properties of observable physical objects.
Berkeley, Hume, Reid, Kant, the phenomenalists, and modern realists have all been
concerned with our ability to conceive (and also to have knowledge) of states of
affairs that are verication-transcendent in the modal sense, that is, whose obtaining
is constituted independently of their epistemic accessibility to us. So the classical
concept-formational concern has extended beyond statements that are by ordinary
lights undecidable. But not only is this the dominant classical conception, Dummetts
anti-realists motivations are such that the scope of his concern should coincide with
the classical concern. For the anti-realists argument is that, because understanding
consists in mastering verication conditions, we have no bona de understanding of
truth-independent-of-veriability. When a philosopher then suggests the modal
thesis that it could be the case that s were true without being veriable, Dummetts
anti-realist should, given that claim about understanding, deny that. Thus he would
be committed to a dispute even about some decidable statements, and thus to the
modal version of anti-realism.
If there is to be a genuine dispute as to whether statements about the currently
observable properties of particulars could be true but not veriable, then veriable
cannot mean simply logically possible to verify; for the realist non-skeptic does not
deny that if such a statement were true it would be logically possible to verify it. How
strong then does veriable have to be to generate anti-realism? For the intuitionist
in mathematics, that a statement is true implies that we have an actual proof of it. The
corresponding anti-realist position about an empirical statementthat it is true only
if veriedsounds too preposterous to consider. Any weakening of that position,
motivated by analogies with intuitionism, would have to maintain that a statement is
true at least only if it is currently accessible to verication in some real sensethat is,
on the basis of evidence that actually exists and that we in some real sense can
acquire. The realist is responsible for answering anti-realists other than Dummett,
and so other interpretations of veriable have to be kept in mind. The Peircean anti-
realist holds that ss truth requires its veriability in the long run of scientic inquiry.
That again cannot mean the logical possibility of verication, which would not make
Peirceanism interesting nor have the anti-skeptical consequences it is sometimes
intended to have. The most obvious interpretation is in terms of a subjunctive
conditional: if the human race survived and were able to carry out certain investiga-
tions, the upshot would be a verication of s.
There is another aspect in which my denition of realism is not Dummetts
ofcial one. He has often characterized realism about certain statements as a com-
mitment to bivalence about them: each such statement is either true or false. The
connection between this criterion and the one employed so far is clear enough:
if true implies veriable, then if neither s nor ss negation is veriable, s cannot be
true-or-false. If that were the only ground on which bivalence might be denied, then
the realist would be committed to bivalence. Now there is a tendency in Dummetts
recent work to give the bivalence criterion a life of its own as a necessary condition of
realism, independently of its connection with the issue of verication-transcendent
truth.3 But bivalence is neither sufcient nor necessary for realism in the sense
discussed so far. That it is not sufcient is shown by this. Certain statements about
observable particulars are in the ordinary sense decidable, and so it is agreed that
bivalence applies to them. But that does not secure a fully realist interpretation.
Realism implies that, even though such statements are in fact decidable, it could
happen that they were true even though unveriable by us. The anti-realist must
deny this further possibility, and thus bivalence is not sufcient for realism. That it is
not necessary I shall simply suggest now, deferring discussion until later. There can
be grounds for denying bivalencee.g., vaguenessthat have nothing to do with an
3
Cf. Dummett (1982).
anti-realism that asserts that truth requires veriability. Realism is committed only to
the thesis that there is no connection between truth and veriability that sufces for a
rejection of bivalence.
II
There is in Dummetts writings the persistent equation of realism with the thesis that
the correct theory of meaning is a truth-conditional semantics. A theory of meaning
for Dummett is a theory of what ones understanding a language consists in. For the
realist, our understanding of a statement consists in our grasp of its truth conditions,
which determinately either obtain or fail to obtain, but which cannot be recognized
by us in all cases as obtaining whenever they do; for the anti-realist, our understanding
consists in knowing what recognizable circumstances determine it as true or false.4
For the anti-realist any involvement of truth conditions in understanding is
accounted for by the involvement of verication conditions. For the realist, on the
other hand, because truth conditions transcend verication conditions, a theory of
understanding must be a truth-conditional semantics in some non-derivative sense.
Let us call a use theory of meaning any theory on which understanding consists in
the mastery of various procedures (e.g., of verication or proof), inference patterns,
and so on, that are speciable independently of a theory of truth conditions. This is to
be contrasted with a theory on which understanding consists in a mastery of truth
conditions that cannot be reduced to mastery of the sorts of things involved in a use
theory. Vericationism is a use theory, but not the only possibility; I shall be
sketching a holistic theory that differs importantly. Dummett holds that a realist
cannot subscribe to a use theory of understanding. The reasoning, I believe, is simply
this. Use-theoretic conditions are assertability conditions. But realism holds that
truth conditions transcend assertability conditions; and so a use theory cannot
account for our mastery of such verication-transcendent truth conditions. Now
I believe that this is a mistake.
There appear to be three possible realist lines of reply to Dummett.
(1) The realist agrees with Dummett that a use theory cannot account for mastery
of verication-transcendent truth conditions, but holds that certain internal or
purely mental facts, which are not explainable by a use theory, constitute ones
thoughts or sentences as having realist truth conditions. This realist might
accept the manifestability condition on a certain weak interpretation (see
below, p. 104), but basically must reject it as Dummett intends it.
(2) The realist agrees with Dummett that a use theory cannot account for
verication-transcendent truth conditions, accepts the manifestability condi-
tion (in a broad sense in which truth conditions must be discoverable from
4
Dummett (1978), p. 23.
external facts about a persons behavior and environment), and holds that
realism is vindicated by the fact that verication-transcendent truth condi-
tions are constituted, at least in part, by external facts that transcend anything
a use theory might appeal to.
(3) The realist rejects Dummetts assumption that a use theory cannot account for
verication-transcendent truth conditions in the sense required by realism.
I assume that on any use theory, the manifestability condition is automatically
satised; even the internal uses of sentences in reasoning are manifestable
in verbal behavior. It is this third reply that I shall defend, and I reject the
rst two.
(That these three possibilities are exhaustive is shown in what follows. The three
variable components of these realist replies are about (i) whether realist truth
conditions are manifestable (in some strong sense) in behavior; (ii) whether the
determining factors of those aspects of meaning or content to which the realist
must appeal are internal; (iii) whether a use theory of meaning can constitute an
adequate defense of realism. This gives us an exhaustive array of the combinations
(see Table 6.1). Combination A is theory (1) of the text, if manifestability has
the strong sense of displayability in behavior; this manifestability is ruled out, as
Dummett asserts, if the internal meaning factors transcend assertability conditions
and conceptual roles. Combination B is theory (2) of the text. Combinations C and
D are covered by theory (3) of the text; in this paper, nothing hinges on the difference
between internal uses and external uses. Combination E is ruled out if manifestability
is taken in the strong form mentioned under combination A. Combination F is ruled
out on the assumption that all external non-use determinants of truth conditions
are manifested in the sense that matters. Combinations G and H are ruled out on
the assumption mentioned, that use theories automatically satisfy the manifestability
condition.)
Perhaps it will be helpful, in understanding what (1) might come to, to have in
mind in a general way how (2) might be implemented. The idea in (2) is that the truth
conditions of our thoughts and statements are determined by their causal-referential
properties and other aspects of their extrinsic contingent relations to the world, for
example their reliability as indicators of the facts. Now Dummetts conception of the
A B C D E F G H
Truth conditions manifestable No Yes Yes Yes Yes No No No
Use theory adequate No No Yes Yes No No Yes Yes
Meaning factors internal Yes No No Yes Yes No No Yes
Table 6.1. Candidate realisms

realists commitment to a truth-conditional semantics is, I think, of something that

both cannot be elucidated in a use theory and is somehow internal to whatever
constitutes a persons understanding of his language; he usually takes it for granted
that (1) is the only possible realist position.
I quite agree that, were that so, realism would be in deep trouble. For if no use
theory could account for mastery of truth conditions, then presumably such internal
mastery would simply be unexplainable in further terms. (Functionalist and concep-
tual role theories are use theories.) Consider how things would stand for such a realist
on each, separately, of two assumptions, namely (a) that our conceptions are not in
language, and, on the other hand, (b) that they consist in understanding language.
In case (a) the position would presumably involve some sort of irreducible grasping
of propositions or Brentanian intentionality. If this were the best a realist could do,
that would seem to me a reductio of realism. On condition (b) the situation is no
better. The realist must in that case say that understanding consists in knowledge of
truth conditions. But that cannot mean propositional or, as Dummett puts it,
explicit knowledge of truth conditions, for that presupposes understanding the
metalanguage in which that knowledge is couched, and what would we then say
about that understanding? Dummett has the realist maintain that understanding
consists in implicit knowledge of truth conditions. But it is difcult to see what that
could mean except some sort of use theory. A notional suggestion is that under-
standing consists in some otherwise unexplained relation among a person, a sen-
tence, a propositionx understands s in such a way that p is its truth condition
but that is merely a variant on the irreducible grasping of propositions.
As regards manifestability, surely Dummett is right that what would be displayed
(as it were) in behavior are the assertability conditions (or conceptual roles) associ-
ated with such irreducible states of conception or understanding, and therefore
according to (1) not their truth conditions. Manifestability might be maintained in
a weak sense: inferability as the best explanation. But I, for one, do not see how such
irreducible states of conception or understanding could possibly be the best explanation
of a persons verbal or other behavior.
Dummett sometimes characterizes the realist as for want of something better
explaining our grasp of verication-transcendent states of affairs thus: the realist
holds that we give sense to those sentences of our language which are not effectively
decidable by appealing tacitly to means of determining their truth-values which we
do not ourselves possess, but which we can conceive of by analogy with those
who do.5 (My emphasis.) Is this an alternative to (1)(3)? Perhaps it is, but I must
say that then I do not understand it even prima facie. The problem is that any such
superhuman verication procedure, such as counting an innite set, is such that its
implementation by a superhuman creature would be as verication-transcendent for
5
Dummett (1978), p. 23.
us as the state of affairs whose verication procedure it is supposed to be. So the

conception of such a procedure should be as problematic as the conception of the
corresponding state of affairs. Thus we could not use such a conception to acquire a
conception of verication-transcendence. If the core of the idea is, however, simply
analogy, then verication procedures are beside the point, which then would be that
we understand verication-transcendent states of affairs by analogy with veriable
states of affairs. Now this does, perhaps, have an interpretation on which it makes
sense: we understand the former by virtue of certain conceptual connections they
have with the latter. In that case, the suggestion can be accommodated under (3); but
then the general theory of meaning that supports it is not truth-conditional.
III
On the second line of reply to Dummetts anti-realist, the realist agrees that realism
requires a theory of meaning to be a truth-conditional semantics, and that the truth
conditions of ones statements must be manifestable, if not precisely in behavior, then
in behavior and its further relations to the world. This realist may grant Dummett
that all that is revealed by a persons behavior in the context of its immediate
perceptible environment is the verication or assertability conditions of her state-
ments. But this realist maintains that there are extrinsic facts about statements, over
and above such behavioral facts, whereby they have truth conditions that transcend
their veriability conditions. It will help if we give this reply a setting within a certain
broad framework for the theory of meaning.
A two-component theory of meaning distinguishes a level of content or meaning
that is constituted by the conceptual roles of our thoughts or sentences, and another
aspect of content or meaning, viz., truth conditions, determined by the extrinsic
relations of our thoughts or sentences to external things, properties, and states of
affairs. The conceptual roles of sentences include recognitional abilities with which
they are associated, inferential dispositions, and the background of theories whose
acceptance affects those recognitional abilities and inferential dispositions. The
verication conditions of a sentence would of course be part of its conceptual role
in this sense; in general, the notion of a use theory of meaning is intended to cover a
conceptual role semantics.
The conceptual roles of sentences or thoughts fail in general to determine their
references and truth conditions, in two ways. First, there are indexical and Twin-
Earth phenomena. Two thoughts of mine to the effect that I have the u, entertained
at different times, may have the same conceptual role, but will have different
temporal references and therefore different truth conditions. Our doubles on Twin
Earth may have just the conception of water that we have, but while our concept
refers to H2O, theirs may refer to some other substance. In this sense in which
conceptual roles fail to determine truth conditions, they are not matched one-one.
But there is a further sense which is more important in the current context. Even if
we ignore indexicality and Twin-Earth phenomena, even if somehow the conceptual

roles and ne-grained truth conditions of our thoughts were matched one-one (as
they may be for certain classes of thoughts), that thoughts have conceptual roles does
not account for, does not explain, their having truth conditions. Truth conditions
involve external relations between thoughts and the world that transcend the facts
that constitute their conceptual roles, including of course their verication condi-
tions. If one is operating with an unexplained conception of Brentanian intention-
ality, of grasping possible states of affairs, then the very individuation of a thought
involves its truth conditionsthey are, so to speak, intrinsic to it. But once one sees
the key mental and behavioral properties of a thought in terms of its conceptual role
then a substantive question arises about truth conditions. This is evidently related to
Dummetts question how truth conditions are to be accounted for if the theory of
understanding is based on verication conditions.
On the two-component model, the theory of understanding or conception is only
part of the theory of meaning; it concerns how thoughts or sentential acceptances are
related to each other, to perception, and to action. But the theory of truth conditions
and reference is concerned with patterns of relations between sentential acceptances
or thoughts, as individuated by the theory of understanding or conception, and
external states of affairs, relations that have to do with the success of speakers or
thinkers in representing those states of affairs. Those representational capacities are
constituted by patterns of correlations between thoughts or sentences and extrinsic
states of affairsthink of how the number of rings in a tree represents its age.
A truth-conditional semantics then systematizes and extrapolates those representa-
tional patterns; and, it is plausible to add, the associated reference scheme captures
causal relations between words and objects or properties, etc., which are partially
explanatory of the representational success of the sentences or thoughts.
Because truth conditions and reference are determined by something over and
above verication conditions, it could seem that Dummetts problem for the realist
is solved. Truth conditions legitimately transcend verication conditions and are
manifestable in behavior in the broad sense of being empirically accountable. We
may then interpret the utterances of others as being about states of affairs that
transcend their verication conditions.6
The two-component theory of meaning is, I believe, correct; and as a theory of
meaning, it may be shared by proponents both of this reply and of the third reply. But
its current applicationthe second realist line of replydoes not constitute a
satisfactory answer to Dummetts anti-realist, for two reasons. First, the proposal
avoids the issue of what specically our conception or understanding of realist states
of affairs consists in; it simply is silent on the matter of a realist theory of under-
standing, and Dummetts problem about verication-transcendent truth conditions
6
Two defenses of realism of this type are: Hilary Putnam (1978); and Colin McGinn (1982).
is posed as a question about understanding. Dummetts challenge is to say how we

can conceive of verication-transcendent states of affairs.
Secondly, there seems to be nothing in the idea of extrinsically determined truth
conditions that a exible anti-realist could not consistently accept. For may he not
say that the relevant causal and representational relations between our thoughts or
statements and other states of affairs, like all other aspects of nature, cannot obtain
unveriably? Let semantic realism be the thesis that our ordinary ascriptions of truth
conditions and references to the sentences and beliefs of others correspond to
objectively determinate properties and relations that constitute those truth condi-
tions and references. I take this to be the denial of Quines thesis of indeterminacy,
that there is no objectively best way to ascribe semantic properties to the statements
of others. Now the two-component theory of meaning is semantic realist. The point
then is that semantic realism is not sufcient for realism, for the anti-realist may
simply absorb the theory, maintaining that its semantic relations are subject to the
standard requirements of veriability.
That a two-component theory might have appeared to be an adequate response to
Dummett is in part a consequence of what Dummett has to say about the relation
between meaning and truth. He equates the theory of meaning with the theory of
understanding, and supposes that any legitimate notion of truth conditions must
somehow get a hold within that context. But if we reject the Brentanian idea that
determinate non-trivial truth conditions are intrinsic to our conceptions, Dummetts
view leaves room apparently for nothing but the disquotational true. So one might
then think that the anti-realism in Dummetts framework consists in its failure to
accommodate substantive truth conditions that may serve as the basis for a corres-
pondence between statements and external states of affairs that transcend the
verication conditions of those statements. Dummett encourages this by character-
izing realism as a commitment to a truth-conditional semantics that goes beyond a
verication-based theory of meaning. But a theory of extrinsic truth conditions has
no tendency to explain how there could be unveriable facts, or how we can have a
genuine conception of such facts. We might see this reply to Dummett as resting
on an equivocation; for truth conditions transcend verication conditions can mean
(a) there are substantive truth conditions that are determined by factors over
and above those conditions whose mastery constitutes understanding, or (b)
truthi.e., disquotational truthdoes not imply veriability. The latter and not
the former is realism.
But semantic realism, even if not sufcient, could be thought an essential compo-
nent of realism, on the grounds that realism requires some sort of correspondence
theory of truth, and therefore requires semantic realism.7 But I believe semantic
realism is not necessary for realism. Suppose one accepts physicalism, and is persuaded
7
In Loar (1980) I took the line that a two-component theory, with an extrinsic ascription of truth
conditions, is an essential component of a realist answer to Dummett, one which also must include
by Quinean arguments of the indeterminacy of our ascriptions of semantic properties.

One might nevertheless hold that our conception of the physical world is of a
collection of states of affairs whose obtaining is independent of the possibility of
our verifying them. Anti-semantic-realism is simply an eliminativist view about a
certain class of putative properties and relations, and of no more signicance for
realism about the natural world than eliminativism about, say, mentalistic properties
such as phenomenal qualities. It does not imply vericationist anti-realism.
The idea of a correspondence theory of truth has been various. In one strong form,
it requires an ontology of facts. But apart from that it seems to have incorporated two
elements: the independence of the facts that make our statements true or false from
their epistemic relations to us; and the existence of semantically determinate corres-
pondence relations between our statements and the world. The correspondence
theory thus conceived is a complex conception, the conjunction of realism and
semantic realism.
Dummett writes: Realism is a semantic thesis, a thesis about what, in general,
renders a statement in the given class true when it is true.8 In a sense that is so; if one
reads true disquotationally, realism is the thesis that our statements are made true
by states of affairs that are independent of their veriability. But my point is that that
is not true if it implies that realism requires semantic realism. In various places
Dummett has suggested that the realist is committed to the notion of reference
playing a certain role in the theory of meaning: The fundamental thesis of
realism . . . is that we really do succeed in referring to external objects, existing
independently of our knowledge of them.9 But again, the existence of determinate
objective reference relations is not to the point. Our disquotational referential
terminology enables us to say: My utterance of that rock refers to that rock,
which exists independently of my knowledge of it.10
For all this, it is possibly not easy to dispel the idea that there is an intimate
connection between realism and semantic realism. Part of the reason lies in our
relations to our own language or concepts. How, if we conceived the facts as being
out there independently of their veriability by us, could we avoid holding that there
something about understanding. And in other places I have incorrectly taken semantic realism to be
essential to realism.
8 9
Dummett (1982), p. 55. Dummett (1982), p. 104.
10
Reference and realism are connected by Dummett in another way. Consider statements that
apparently quantify over directions, and a theory that holds such statements to be reductively translatable
into statements that quantify, not over directions, but over straight lines. This theory is anti-realist about
directions. Dummetts diagnosis of what makes it anti-realist is that, in the determination of the truth-
values of statements about directions the reference of direction plays no role, because their truth-values are
determined indirectly, via the truth-values of their translations. Dummett accepts that this reductionist
thesis can be coupled with bivalence about such sentences, and so accepts that bivalence is not sufcient for
realism. He calls this reductionist thesis a comparatively mild species of anti-realism. But notice that it is
not as such anti-realist in our sense at all, for it does not imply that such statements cant be true and
unveriable.
are real relations between our statements and those facts whereby the former can be
made true or false by the latter? Let me make two points. First, the disquotational
reference relations and truth predicates that one applies to ones own sentences do
involve genuine relations between words and things; that that is so does not require
semantic realism. Paris denotes Paris associates a name with a city. To say that
denotes is used merely disquotationally is, in effect, to say that for a denition to be
adequate it must merely pair certain expressions with certain things. Such a pairing
can be achieved enumeratively for names; as regards that pairing of sentences with
possible states of affairs (so to speak) which ones disquotational true captures, a
recursive denition would be required to explicitly capture that truth predicate,
although one has effortlessly mastered it without benet of Tarski. The point is
that such a denition would associate expressions with, as one must see them, non-
linguistic matters. So, from ones rst-person perspective on ones own language, one
does not have to suppose semantic realism to be true in order to associate, as one sees
it, aspects of ones language with aspects of the world. Of course this ability cannot be
constitutive of ones understanding ones language, for it presupposes understanding
a metalanguage. But if a theory of understanding independently accounts for ones
ability to have thoughts of the form it could be the case that s were true (disquota-
tional sense) even though s were unveriable then one does not need semantic
realism in order to regard ones thoughts as referring (disquotational sense) to things
and states of affairs that could exist independently of ones ability to verify them.
Secondly, the denial of semantic realism could seem to raise this question: if there
are no non-trivial reference relations between my thoughts and the facts, how, if
those facts are constituted independently of my ability to verify them, could it be
other than an accident that my beliefs had any tendency to be true? This question is
suggested by Putnams view that non-trivial semantic relations are vindicated by
their role in the explanation of the success of science or, as we might also say, of the
reliability of our beliefs as indicators of the facts. If semantic realism were false, then
it might seem that the only way to make it not an accident that our beliefs tend
toward the truth would be to regard the facts as, so to speak, constituted by their
veriability. But this goes too far. A Quinean denial of determinate reference
relations, of there being an assignment of references that is objectively best, does
not entail that there are no systematic causal relations between our beliefs or
statements, on the one hand, and things, properties, and so on, on the other, that
explain how it is possible for my beliefs to tend toward the truth (disquotational
sense). Clearly, if the conjunction of realism and anti-semantic-realism had the
consequence that truth is an accident, that would be a serious problem. But I do
not think there is a good argument to that effect.11
11
This is not to deny that there may be further problems for realism concerning what makes it not an
accident that beliefs arrived at via a certain preferred methodology tend toward the truth. The point is that
denying semantic realism does not exacerbate such problems.
IV
I now turn to the third strategy for replying to the anti-realistnamely, a certain use
theory of understanding or conception on which our grasp of verication-transcend-
ent states of affairs is accounted for in such a way that it is unproblematically
manifestable in behavior. (This strategy is developed in this and the following
section.) The theory of understanding is a conceptual role theory, which I have
mentioned as one stage of a two-component theory of meaning: our concepts are
determined by their roles in our verications, reasonings, and general theoretical
behavior. The account differs from a vericationist theory, and has important
features in common with holistic theories of meaning, which Dummett has, for
various reasons to be discussed, rejected. It has, it seems, sometimes been thought
that holistic theories of understanding and conception are essentially anti-realist, but
that, I believe, is profoundly mistaken.
The idea is going to be that (a) realism about the natural world is simply a
consequence of our theory of nature, and (b) the conceptual role or quasi-holistic
theory of understanding explains perfectly well how such a realist conception of
nature is possible. The question whether realism, in any transcendental or meta-
physical sense, can simply be internal to our theory of nature I shall subsequently
take up. The relation of the argument to realism about mathematics I shall also
touch on.
The rst stage of the argument is independent of the theory of understanding, and
consists essentially in consolidating a certain well-known prima facie objection to
anti-realism. Dummett dubbed a version that Putnam proposed the idealistic fallacy
argument, and this seems to me an appropriate name for the version I shall give. The
core of the argument, as I see it, is that for the relevant range of statements s, the
veriability of s, whether Peircean or individual, is dependent on natural contingen-
cies in such a way that it is a natural or scientic possibility that s be true even if not
veriable. Thus realism does not depend upon mere logical possibilities of the
deceiving God variety; the anti-realist could always reject them as simply incoherent.
The idea is rather to specify a sort of possibilitypossibility according to the laws of
naturethat a naturalistically inclined anti-realist must nd unexceptionable, but
that shows it to be prima facie comprehensible that s be true but not veriable. The
conceptual role theory of understanding can then certify that prima facie compre-
hensibility as the real thing.
The prima facie argument against anti-realism will not be unfamiliar; it is part of
what many see as (philosophical) common sense. If s is about the primary qualities of
objects in the garden, in Antarctica, or in the Andromeda galaxy, then ss veriability
by us depends upon various contingent circumstances not entailed by anything that
we could regard as laws of nature. Those circumstances are naturally contingent: our
neural pathways are arranged in certain ways, the regions of space through which the
relevant light or sound must travel lack distorting properties, and so on. It would not
surprise us if clever scientists were to describe ways the natural world might have
been, compatible with the laws of nature, in which such and such facts obtained while
those circumstances on which those facts veriability by us depends did not obtain.
I have the impression that it is sometimes thought that Peircean veriability
circumvents this, that our scientic methodology, ideally applied, must enable us
to see through whatever barriers to knowledge nature might put in the way. But that
misses the point. If a certain fact is veriable in the long run of scientic research,
then that depends on the sort of contingencies I have mentioned. Knowledge-
yielding relations between natural facts and the collective states of a society, however
ideally arrived at, are as contingent as they are between natural facts and the beliefs of
individuals. How could they be otherwise?
The upshot then is that our general conception of the contingencies on which our
knowledge depends implies that it is possible, for the relevant range of s, that s be true
but not veriable. The point is not merely that the conjunction is, so to speak,
syntactically consistent with the laws of nature. Our being in a position to make
certain verications can be seen as depending on two factors: the laws of nature and
the particular facts or initial conditions. Our scientic theories imply that the latter
can be disposed differently than they are, so that we would not then be in that
position. Thus our scientic conceptions not merely permit, but also imply, the
possibility that s be true but not veriable. Naturally the point is not that it is possible
that we should verify later that s is true but not now veriable. For Peircean
veriability would be immune to that: we cannot Peirceanly verify that something
is true but was not Peirceanly veriable, i.e., veriable in the ideal long run. The point
of course is rather that our (to such and such extent veried) theory of nature implies
that it is possible that s be true but not even Peirceanly veriable.
Realists will not think, of course, that one has to rely on theoretical possibilities in
order to nd counterexamples to anti-realism. Suppose that there in fact exists no
evidence for or against the statement that Nebuchadnezzar had a wart on his chin.
A realists conviction that either he did or he didnt will appear to the realist to be a
refutation of anti-realism. But Dummetts question about realism is not whether such
a conviction is natural or even inevitable, but rather whether the realist is entitled to it.
Now by pointing out that the realist is also committed to certain modal possibilities
with respect to statements that are in fact veriable, my intention is to indicate that
there is something systematic and general (something that is not possessed merely by
statements for which evidence does not in fact exist) in our conceptions of nature,
which is essentially realist and is implied by our scientic theories. If it can then be
shown that this systematic aspect of our conceptions is compatible with an adequate
theory of understanding, then the real-world counterexamples will be vindicated. But
that will depend on the more general modal point.
Various objections to the claim that the possibility of verication-transcendent
truth follows from our theory of nature may come to mind. One might grant the
empirical point, but hold that it is irrelevant to realism, that it establishes merely
empirical realism; to this I shall return. Another objection might be that the
prima facie possibility is not validated by an adequate theory of understanding.
Before facing that (in section V), I shall now discuss a certain direct argument
to the effect that what I describe is no possibility, that there is a fallacy in the
empirical argument.
The possibilities that I have mentioned call to mind the brain in the vat, who is
supposed to be radically deceived about the world. This unfortunate creature plays a
role in discussions not only of skepticism but also of realism, for its very conceiv-
ability apparently demonstrates that veriability is one thing and truth another. Now
there is a line of attack against the possibility of such a brain in a vat which
might appear to call into question the form of the idealistic fallacy argument that
I have presented.
Whether a person xs beliefs could be systematically false depends on what
constitutes the truth conditions and reference conditions of xs beliefs. But that, the
reply goes, depends on certain relations between xs psychological states and the rest
of the world. Truth conditions are determined, not merely by what is in the head, but
also by causal relations to things and properties, and, more generally, by facts about
xs capabilities as a gatherer of information. A principle of charity governs ascriptions
of truth conditions. But such considerations apparently preclude the possibility of
such a deceived brain; for the references and truth conditions of its thoughts would
then be, so to speak, internal to the circumstances that constitute its imaginings.
As Hilary Putnam might say, the concept tree would denote trees-in-the-image; it is
information about such things that the brain is adept at gathering. Consequently no
such wedge as those so-called natural possibilities, the reply goes, can be driven
between the truth (conditions) of xs thoughts and their veriability.
In response to this, let us rst notice that the deceived brain in the vat, and
analogously isolated individuals and societies, are simply irrelevant to the central
question whether truth implies veriability. The natural possibilities mentioned were
designed to show that s is true does not imply s is veriable by us, where true is
purely disquotational. In this form of the idealistic fallacy argument, the envisaged
possibilities need not involve anyones thoughts; for a world in which there are no
thinkers would be quite to the point. Consequently what constitutes the truth
conditions of thoughts is irrelevant. What is relevant, rather, is whether it is a natural
possibility that the Moon has ice on its surface even though that is not veriable
by us. The question is whether certain possible states of affairs could obtain even
though we could not verify that; it is irrelevant whether certain true or false thoughts
might be veriable or falsiable.
But does this not change the subject? Dummetts question appears to be whether a
statements or thoughts truth could transcend its veriability; and that apparently is
about what constitutes a statements truth conditions. Now that is indeed Dummetts
question, but it conates two distinct issues. On the one hand, there is the question
as to what constitutes the truth conditions of a persons statements or thoughts.
This has to do with issues connected with semantic realism. On the other hand, there
is the issue of realism, whether s can be the case even though we are not in a position
to carry out the appropriate verications. This involves, in its general form, only
disquotational truth, and the applicability of that concept raises no questions about
what constitutes the truth conditions of thoughts or statements. It is true that a
question in the theory of meaning is raised: whether s is true but not veriable by me
can be validated as comprehensible by an adequate theory of understanding. But that
raises no issue about what constitutes a thoughts truth conditions; the true occurs,
disquotationally, within the scope of a possible thought, and not as a predicate that is
applied to some thought.
I have been supposing that the anti-realist theories in contention have the form
s is true implies s is veriable. It seems, however, plausible that the realist
concerning certain states of affairs s must also assert that no non-trivial epistemic
conditions are logically sufcient for the truth of s.12 But the realist may then appear
to be committed to the possibility of some sort of brain in the vat hypothesis. For it
would have to be possible that one has veried that s even though s is not true, and
that seems to mean, possible for one to have a veried thought that s even though s is
false. So if s = the external facts are thus and so, and s* = my brain is lodged in a
body related to those facts in such and such a way, then the realist about s and s*
seems to be committed to the brain in the vat hypothesis as described. But that
apparently leaves the realist open to the objection that the truth conditions of
thoughts depend upon their extrinsic relations to the world in such a way that,
given the principle of charity, such radical deception is impossible.
But let us keep in mind what is anti-realist about anti-realism. For certain states of
affairs, their obtaining or not obtaining is seen by the realist as constituted inde-
pendently of their actual and possible relations to our epistemic or experiential states.
But if what is counted as an epistemic state is not carefully specied, anti-realism is
not captured. Consider we know that s implies s is true, which evidently
commits no one to anti-realism. Clearly if realism and anti-realism are to be
genuinely debatable positions, then when it is said or denied that certain epistemic
conditions are sufcient for truth, the relevant conditions cant be like we know that
s or if someone were to believe that s she would thereby have a true belief.
Now consider the anti-realist thesis that it is veriable by us that s is strongly
sufcient for s is true; and suppose that some variant of the principle of charity is
invoked, say, that we cannot generally justify that it is veriable that s unless s is true.
(This might be on the grounds that it is veriable that s iff were we to implement
certain ideal procedures we would believe that s, and that charity requires ascribing
12
This does not mean, obviously, that the realist must assert that no such conditions are epistemically
sufcient for knowing that s is true. It would take a substantial epistemological argument to show that the
former implies the latterthat realism implies skepticismone that, contrary to the anti-realist tradition,
I do not think is forthcoming.
truth conditions to such beliefs so as to maximize truth.) Now the point is that there
is no more reason to call the conjunction of those two theses (that veriability is
sufcient for truth and the principle of charity) anti-realist than there is the prop-
osition that we know that s implies s is true. One has simply decided not to call a
range of situation-types veriabilities that snot to ascribe them content in that
wayunless in general s is true. I could certainly accept a convention to do that
without ceasing to have a realist view of the world. Genuine anti-realism has to be
substantive, has to involve psychological states that do not have the relevant truths
built into them; but then the principle of charity cannot defend anti-realism against
the idealistic fallacy argument.
V
The objective (the third realist reply) is a theory of understanding that accommodates
statements whose truth can transcend their veriability and that shows how our
mastery of them is manifestable in behavior. Such a theory is to be a use theory, that
is, will explain understanding as the mastery of certain ways of operating verbally or
conceptually, rather than in unexplicated terms of grasping truth conditions or
propositions. Let me assume that thought is in language, to avoid the tediousness
of having to formulate things also to cover the other possibility, where one would
speak of conceptualization and thoughts rather than understanding and sentences.
On a vericationist use theory such as Dummetts, understanding a sentence
consists in mastering verication conditions that are not holistically constituted,
that is, that can be specied independently of the verication conditions of other
sentences, except of course for their constituents. The alternatives to vericationism,
among use theories, are what we might call holistic conceptual role theories.
A conceptual role theory holds that the meaning or content (or one component
thereof) of a sentence or thought consists in certain dispositionsits potential role in
reasoning (in a network of such roles), in perception, in action guidance. In the
philosophy of science, holism has sometimes meant the thesis that a sentences
or terms meaning, for a person x, depends on its place in the overall theory or
conjunction of theories that x actually accepts. But holism would, I suppose, be
allowed to cover also the potential responsiveness of a sentence to certain new
evidence, how its acceptance would affect other beliefs, and so on; for otherwise
holism would have nothing to say about the meanings of sentences that are frameable
in xs language but not actually accepted by x. On the other hand, any plausible
account of those dispositions that constitute the conceptual roles of xs beliefs
would regard at least certain features of the theories that x actually accepts to be a
determining factor. So in characterizing conceptual roles, I shall speak of both
inferential dispositions and the acceptance of theories.
It should now be evident how the idealistic fallacy argument ts together with a
holistic account of understanding: our mastery of that schematic theory of nature to
which the argument appeals constitutes our conception of a verication-independent

world. Understanding is in general constituted by mastery of a network of conceptual
connections; those that make up our theory of nature imply that truth can transcend
veriability. This account of concept formation doubtless will appear to some to
make realism too easy, to have all the advantages of theft over honest toil. Such an
impression could be fully neutralized only by a full-scale motivation of a holistic
theory of understanding, which I shall not undertake; but I shall offer a defense of it
against certain criticisms. Realists have found the having of concepts of certain
debated categories as not especially difcult. Dummett is doubtless correct in
diagnosing one implicit realist picture as the unexplained grasping of propositions,
of irreducible intentionality. But the view that the debated concepts are easy to
have does not depend on propositional graspings; a holistic conception of meaning
and concept formation would also account for a realists lack of worry about
concept formation.
So there is a short answer to Dummetts request for a realist theory of under-
standing that satises the requirement of manifestability, namely: (i) our theory of
nature implies the verication-independence of facts about the natural world, (ii) the
holistic conceptual role theory of understanding makes our mastery of that theory
sufcient for our having a bona de conception of verication-independent states of
affairs, and (iii) that realist conception of ours is manifestable in behavior for the
simple reason that the relevant conceptual roles are manifestable in our verbal
behavior. Dummett agrees that realism is, as it were, the entrenched, unreective,
natural view, but he suggests that it may rest on an unacceptable conception of truth
conditions. But he also writes: As long as any feature of existing linguistic practice
has not been shown to be incoherent, it remains in possession, and we shall
automatically opt for any meaning theory that accords with it in preference to one
that does not.13 My suggestion is that the holistic conceptual role account of
understanding and conception is that theory.
That, as I said, is the short answer. But more has to be said about the relation
between conceptual roles and assertability in order to avert a charge of paradox. The
conceptual role of a statement is a matter of its inferential properties and respon-
siveness to perception. That could seem to imply that the meaning of a statement
consists in a certain set of assertability conditions. But how then could ss truth be
understood as independent of ss assertability, given that assertability conditions
constitute ss meaning? Indeed it is not unlikely that something like this lies behind
Dummetts supposition that a realist theory of understanding must rely on a know-
ledge of truth conditions (or propositional graspings) that cannot be explicated in a
use theory of meaning. Must a use theory not be a theory of assertability and
therefore preclude conceiving of ss truth as independent of ss assertability?
13
Dummett (1976) 8:6.
The holistic theory might seem to lead not to realism but rather to a non-vericationist
holistic anti-realism. Indeed an anti-realist might think to take advantage of the
holistic conceptual role theory, in extending conceptualization beyond perceptual
and enumerative inductive constraints, without thereby compromising the denial
that truth can transcend veriability, provided, of course, that a liberal and holistic
notion of the latter is adopted. (So it may seem that accepting abstract ideas in the
sense in which Berkeley rejected them does not commit one to realism.)14
I shall argue that a holistic conceptual role theory does not equate meaning with
assertability conditions. Let us divide statements into two classes: those whose
assertability conditions must be exclusively holistic, and those that can be regarded
also as having non-holistic, perceptual, assertability conditions. As for statements of
the rst class, a theoretical statements inferential connections with perceptual
statements do not thereby endow it with its own non-holistic assertability conditions.
The question is whether such statements meanings, i.e., their conceptual roles, can
somehow be identied with their holistic assertability conditions. There are complex
questions here about epistemology; let me make a bold assertion. That ss conceptual
role determines holistic assertability conditionsthat is, the enormous array of all
combinations of observational and theoretical statements that would support
sseems to presuppose an a priori holistic conrmation theory, that is, one which
would, without further empirical assumptions about the reliability of our belief-
forming procedures, assign to a conceptual role such an array of assertability
conditions. Now it seems highly unlikely that a sufciently detailed such a priori
conrmation theory exists. But, if not, a conceptual role neither determines nor is
determined by a set of assertability conditions.15
So what about statements that apparently have non-holistic assertability condi-
tions as part of their conceptual roles? My concept cat, say, is in part determined by
my ability to recognize cats. But my general understanding of nature does not allow
my recognitional ability to remain a brute fact; it must somehow be tted into the
framework of nature, and so I see my ability as depending upon contingent relations
between cats and me. When I contemplate a state of affairs involving a cat and the
breakdown of those relations, that recognitional ability which determines (in part)
14
Scientic realism is the view that certain statements about unobservables are literally true, let us say
roughly. On the face of it, this is compatible with anti-realism in our sense, for it seems that a scientic
realist could go on to say that any such true statement must, at least in some holistic sense, be veriable. But
I wonder whether this could be a stable position, given any normal scientic realist view about the
contingency of these relations between unobservables and us that make it possible for us to verify
statements about those unobservables.
15
Indeed, we can turn any such anti-realist argument that presupposes an a priori conrmation theory
on its head. If there were such a theory, it would imply that meaning (conceptual role) is exhausted by
assertability conditions. But that cannot be so, given the coherence of the conceptual role of s is true but
unassertable. Therefore, no such a priori conrmation theory is correct. It seems to me that all theories on
which conrmation relations are entirely a priori (e.g., foundationalism and coherentism) have anti-realist
implications: how could s be necessarily good evidence for s0 if that is not somehow reected in the
conditions under which s0 is true?
the assertability conditions of theres a cat may still come into play, for I can think: a
state of affairs that in real world circumstances I could verify at a glancee.g., a cats
being on a matcould obtain even though not veriable by me (because I dont exist
or there is an undetectable distortion somewhere between my thoughts and the cat).
Perhaps a distinction of Kripkes will help, between what xes the referent of a term
(e.g., heat) in the actual world and what the term requires of its referent in arbitrary
possible worlds. I may conceive of a state of affairs in terms of how it would affect me
in real-world circumstances, but also conceive its existence independently of its
affecting me in that way. Its real-world properties follow that state of affairs into
other possible worlds not by being true of it there, but by picking out there the state
of affairs that here affects me thus and so.16 Is my conception of a cat, then, simply
of whatever it is that triggers my recognitional capacities, of the x that underlies my
cat-perceptions? Evidently not, for the conceptual roles of my thoughts about cats are
rich and complex, tting them into a spatiotemporal framework, giving them a
variety of causal powers, of compositional structures, and so on. The virtue of the
conceptual role theory of meaning is that it makes sense of this traditional realist
idea, makes it more than blustering in the face of Berkeley. The fundamental point is
this: I can conceive of ss being true independently of my being able to recognize it as
true, even though that recognitional ability is part of ss conceptual role, because
I have mastered a theory of nature in which that recognitional ability is only con-
tingently connected with the truth of s. The further aspects of ss conceptual role that
make this modal thought of mine thinkable are what constitutes for me the tran-
scendence of ss (disquotational) truth conditions over ss veriability conditions.
On the conceptual role theory of meaning, the distinction between decidable and
undecidable statements, which on Dummetts account is central, does not have much
signicance in connection with how (disquotational) truth conditions are deter-
mined. Let me briey consider both statements about the distant past about which
no present evidence exists, and universal generalizations. The conceptual role of
statements about the past is presumably determined in part by the role of memory,
but it is also determined, as it were independently, in a way that is not tied to their
assertability. I mean their place in a theoretical framework that contains, among all
the rest of the apparatus of our theory of nature, the relation earlier-than, which is
(roughly) the ancestral of that temporal relation whose obtaining is observable.
So my understanding s was the case is a function of my understanding s is now the
case together with my mastery of the relation earlier than. My further capacity to
understand s was the case is true but no evidence now exists for it derives from the
fact that my theory of nature implies that my ability to verify some past facts depends
on quite contingent evidential relations between those facts and me, relations that
16
Something close to this point was suggested (in discussion) by Lloyd Humberstone, and supported
(in correspondence) by Derek Bostock, as a way in which a sort of vericationist theory of meaning could
be squared with the idealistic fallacy argument. I am here turning it to the realists advantage.
can and do break down. What I know of the workings of nature makes it perfectly
comprehensible that traces of the past vanish, and that explains my ability to think
it is possible that s was the case but is now unveriable.
As for universal generalizations, how does their role in reasoning give us a
conception of their truth as constituted independently of their being conclusively
veried? I believe there is no special problem here. Our ability to understand
statements of the form (x) . . . x . . . is constituted by their having a certain role in
our reasoning, and it is coupled with our mastery of the disquotational true. So we
are capable of thinking (x) . . . x . . . is true. What then makes it possible for us
coherently to think (x) . . . x . . . is true but not veriable? First, if what is meant is
conclusively veriable and the generalization is suitably unrestricted, then our
understanding here evidently consists both in our accepting that the requirements
of conclusive veriability are not satised in this case and in our ability to accept such
a generalization as true on less than conclusive grounds, or to think that it is probably
true, and so on. These are ordinary features of the way we think, which the
conceptual role theory simply accepts. The anti-realist thesis that s is true but
unveriable is incoherent in general, has been rejected on independent grounds,
and so the theorist of meaning can apparently appeal to no general principle to
support the rejection in this case of those features of our thinking as incoherent.
Secondly, if what is meant by veriable is conrmable to some degree, the
veriability of a true universal generalization could conceivably be blocked by natural
contingencies, as with all other statements. What the realist says about that is by
now evident.
The topic has been realism about the natural world, but Dummett has of course
also been concerned with realism about mathematics. It is evident that this defense of
realism about the natural world cannot be extended to mathematics: for we have no
conception of contingent relations between truth and provability as we have with
regard to truth and veriability. So the independence of mathematical facts from the
existence of proofs cannot be tted into the metaphysical picture of contingent causal
relations between facts and their veriability. This may be good reason in itself for
anti-realism about mathematics, on the familiar grounds that we then have no
conception of knowledge-yielding relations between us and an independent math-
ematical realm. But even if so, intuitionism is not the only anti-realist theory
conceivable. Various forms of nominalism and ctionalism are options, and on
them the choice between intuitionist and classical mathematics is a distinct issue.
Even on a ctionalist view, of course, some account of our understanding of
mathematical statements is required, and so a preference for classical mathematics
would have to be backed by a suitable theory of understanding. Given anti-realism,
excluded middle could not be defended on the grounds that mathematical facts
transcend their provability. But the pro-classical anti-realist will wonder why any
special defense of excluded middle, etc. is required. Dummett argues that an adequate
theory of understanding for mathematics, one which is non-holistic, leads to
eschewing excluded middle and to a retrenchment to intuitionist mathematics.

Dummett recognizes that the proponent of classical mathematics may appeal to
holism, to the idea that our understanding is simply a mastery of a system of rules,
e.g., classical mathematics; and it does seem that that is where the issue must lie. In
the following section I discuss Dummetts reasons for rejecting holistic theories of
understanding, but not from any special interest in defending classical mathematics
realism about the natural world is still the focus.
VI
The theory of understanding that I have sketched is holistic because on it a statements
meaning is not some conditione.g., a truth condition or assertability conditionthat
can be specied independently of the statements position within a network of state-
ments. But holism carries other implications that should be neutralized.
The holistic theory of understanding differs from the holistic epistemological
position often called the Quine-Duhem thesis. This thesis is that we are never in
an epistemological position to evaluate non-arbitrarily a single statement, but must
rather always regard ourselves as evaluating the conjunction of a given statement
with the rest of our assumptions and theories; if the upshot is negative, our choosing
to reject that statement, while pragmatically perhaps sensible, must be epistemolog-
ically arbitrary. But the holistic theory of understanding says nothing about justi-
cation. It says that meaningwhat one understandsconsists in a position in a
network of recognitional and inferential dispositions, the latter being in part deter-
mined by the theories one accepts. I take this to be compatible with an epistemo-
logical theory that holds that conrmation can presuppose beliefs that are not
currently being reevaluated.
Holism is sometimes used to imply a completely egalitarian treatment of a
persons inferential dispositions in constituting the meaning of his statements, the
consequence of which is that each persons meanings would be idiosyncratic. But
I mean a network of shareable conceptual roles and assume that it is possible to
isolate, within the total set of a persons recognitional and inferential dispositions and
theories, a skeletal framework of beliefs and recognitional and inferential dispositions
that are specially constitutive of our shared concepts. It is not intended that these
cannot change over time, or that similarity of conceptual roles is not a matter of
degree; but conceptual role holism does not conict with the existence of commu-
nicable meanings.
I shall now turn to some of Dummetts reasons for resisting holism.
(1) Dummett advances an argument about the meaning of logical connectives17
which, if generalized, would show that holistic theories of meaning are not adequate.
17
Dummett (1976), 7:204.
It is that we cannot regard a grasp of the deductive rules that govern a connective as
adequate for a grasp of its meaning, because that would imply that any unfamiliar
logical system, characterized proof-theoretically (perhaps one of the many-valued
logics), could thereby be understood by us. From the falsity of the latter, Dummett
concludes that something more is necessary for understanding a statement that
contains a given connective (apart of course from the other elements of the state-
ment), namely, a grasp of the way its semantic value is determined in accordance
with its composition, where this yields a knowledge of its assertability conditions.
Now the holist may grant Dummett that learning a new formal system does not
thereby yield a grasp of meanings. For what we are capable of understanding depends
on how we actually think, on our actual inferential and recognitional capacities,
depends, in other words, on translatability into the entrenched conceptual roles of
our statements or thoughts. Understanding is not a matter of reexive, propos-
itional, knowledge of rules; it is a matter of the actual network of dispositions that
link thoughts to each other, to perception, and to action. Learning a new formal game
simply does not amount to acquiring the dispositions that are essential to thinking in
terms of the new rules. Moreover, holism is quite compatible with holding that a
system of dispositions, characterized formally, amounts to a system of thought only if
it satises certain criteria of coherence, and even, perhaps, only if it is possible on
extrinsic grounds to assign truth conditions to the relevant statements or thoughts
in such a way that there is a systematic match between truth-conditional structure
and inferential structure. But holism is in itself a theory of understanding and not a
comprehensive theory of representation. Let me emphasize that this does not imply
that if Alpha Centaurians satised the relevant criteria for having a system of thought
we could understand them; for it could still be that their conceptual roles did not
match ours sufciently for translation.
(2) Dummett holds that understanding what is communicated by a statement
requires that that which determines its meaning must be surveyable; our implicit
grasp of its meaning must have a focus, consist in a compact piece of knowledge.18
Holism of course makes the meaning of a statement consist in something that is not
surveyable in Dummetts sense. Is this a problem? Let me rst observe that under-
standing a language may involve two capacitiesnamely, the ability to think in it,
and the ability to understand the utterances of others. On conceptual role holism, the
rst ability does not consist in having knowledge, even implicit knowledge; it consists
rather in being disposed to reason in a certain way, to form certain beliefs in response
to certain perceptual stimuli, and so on. This ability consists in a network of such
dispositions; it does not require knowledge of the network, of something unsurvey-
able. But nor does it require knowledge of something surveyable; understanding is
18
Dummett (1976), 7:8.
knowing how and not knowing that. Secondly, it is true that in understanding
anothers utterance one has a compact piece of knowledge: one knows that the
person means or says that such and such. That judgment itself has a certain
conceptual role, which of course does not require knowledge of the totality of ones
own or the other persons conceptual roles. The correctness of my judgment about
what another person says or means depends in part on a certain conceptual role of
his matching one of mine.
(3) Holism makes it impossible, according to Dummett, to take a sentences sense as
systematically derivable from the sense of constituents.19 Holism may indeed seem
to have odd consequences for the relation between the meaning of a sentence and its
parts; the theory makes a sentences meaning depend, apparently not on the meaning
of its constituents, but on its place in the conceptual network. But the consequences
are not, I think, really counterintuitive. We may distinguish: (i) a sentences meaning
is a function of the meaning of its parts; and (ii) the correct theory of understanding
is a componential semantics, that is, a systematic description in which the meanings
of components are specied (other than by translation, of course) independently of
the network, and the meanings of sentences are then generated from those meanings.
Holism of course denies (ii), but (i) is harmless: a sentences parts have meaning, that
is, they make regular contributions to holistic conceptual roles; and a sentences
meaning is a systematic function of what its parts are. Now an apparently entrenched
belief among many theorists of meaning is that an adequate explanation of language
learning would require a componential semantics, on something like the grounds
that language learning is acquiring the ability to project new meanings from what one
has learned. But a theory of what understanding consists in, in particular a holistic
theory of understanding, is not a psycholinguistic theory; it does not speculate about
the mechanisms whereby we learn language. Somehow one learns new words, by
simple ostension or by whatever it takes. How that ostension (say) gives one those
abilities is an empirical question; that it happens is undeniable, that we have those
abilities is undeniable, and the holistic theory simply asserts that understanding
consists in those abilities.
(4) Dummett has characterized holism as the doctrine that the application of the
predicate true to a sentence cannot be explained in terms of the composition of the
sentence.20 Consider again both the disquotational true and the interpersonal
true in xs language. Nothing in a holistic theory of ones understanding of
Paris is crowded precludes ones judging that Paris is crowded is true iff the
referent of Paris is in the extension of crowded, or making the corresponding
componential judgment about the extrinsic truth conditions of that sentence in
another persons language.
19 20
Dummett (1976), 6:34. Dummett (1976), 6:26.
VII
The independence of truth from veriability resides in natural contingenciesfacts
that in some sense could have been otherwise. Can such a fundamental metaphysical
doctrine as realism rest on an empirical and naturalistic basis? The thought may
occur that what I have described is in fact some sort of empirical realism, or, in
Putnams phrase, internal realism, and that the truly problematic metaphysical or
transcendental realism has not been captured. It could seem that what has been left
out is precisely what Dummett has regarded the realist as committed to, namely a
conception of truth on which truth is determined by substantive correspondence
relations to an independently existing reality, that determine bivalence, and in the
grasp of which our understanding consists.
Empirical realism has often meant simply the assertion that physical objects exist,
in certain spatial relations to us. Kant was an empirical realist in this sense, as was
Berkeley; and of course Peircean anti-realism is empirical realist in this sense. But
realism as I have characterized it is a distinct thesis, and concerns not existence but
independence, that is, asserts that the relevant existential facts are independent of
their veriability by us. That thesis, however empirical, is evidently incompatible
with Peircean anti-realism and, despite disclaimers I have heard, with Kantian
transcendental idealism. (If the latter could accommodate independence, what
could its content be? What would be its answer to skepticism?) The point is that
the relevant independence is entailed by our theory of nature; and it is difcult to
envisage what could possibly be meant by a further metaphysical or transcendental
independence that had not thereby been secured. What is naturally possible is
logically and metaphysically possible. So on its most substantive interpretation,
metaphysical realism is entailed by an empirical theory.
As for correspondence relations, we have already seen the irrelevance of seman-
tic realism, which concerns as limited an aspect of reality as, say, realism about
phenomenal qualities. If Quine were right and there were no objectively determinate
semantic facts, everything else would remain in place, the independence of the facts
unthreatened. Suppose semantic realism is false. Then our choice of a truth-theoretic
interpretation for French, say, is to some extent arbitrary or conventional. Neverthe-
less, given our usual choice of truth theory, whether an utterance of il pleut is true
depends on facts that could obtain independently of their veriability by us. The
sense of correspondence, that, it seems to me, has been central in intuitions about
what constitutes realism has required, not that the correspondence relations be
objectively determinate in the sense which Quinean indeterminacy denies, but rather
that there be independently constituted facts so that, relative to a given truth-
theoretic interpretation, one can speak of one set of facts (thoughts or utterances)
as corresponding to a distinct set of facts. It is the notion of independence that is
crucial, and that, I say, is secured by our theory of nature quite independently of
semantic realism.
That a theory of meaninge.g., conceptual role holismis a use theory and not a
truth-conditional theory has been widely understood as sufcient for such a theorys
being anti-realist. But that understanding consists in inferential dispositions and not
in a grasp of truth conditions simply does not imply that whether electrons have
negative charge depends on our ability to verify it. Realism is not itself a theory of
understanding; but it requires a theory of understanding because if our conceptions are
genuinely realist, a theory of conception ought to account for that. Conceptual role
holism accounts for our realist conceptions in terms of their realist conceptual roles.
Finally, let me return to Dummetts equation of realism with a commitment to
bivalence. I earlier pointed out that bivalence is not sufcient for realism, given that
there can be a realist/anti-realist dispute about statements that are de facto decidable.
But nor is bivalence necessary for realismthat is, for verication-transcendence.
Vague statements can be neither true nor false. Now suppose that every vague
predicate can be made precise, so that bivalence would result for every revised
statement. There then would be no reason to take an anti-realist view of the vague
statements; for if their precise revisions satisfy the realist thesis of this paper, the
states of affairs that the vague statements vaguely describe obtain independently of
their veriability. On the other hand, suppose the relevant vocabulary cannot be
made precise. Still, the statements couched in its terms will have clear cases of
application, that is, there are possible worlds in which they are clearly true or clearly
false, even though there are possible worlds in which they are neither. Now nothing
in the vagueness of these predicates precludes (1) that in the clear cases, the fact of the
matter which makes the statement true or false is constituted independently of its
veriability; or (2) that in an indeterminate case, when the predicate and its comple-
ment both fail to apply, there is another predicate from the same class of predicates
(think of color predicates) which although itself vague does clearly apply in that case,
where again the relevant fact is independent of its veriability. The failure of bivalence
for vague predicates, then, does not imply that the facts they are about are constituted
by their veriability; bivalence is not a necessary condition of realism.
Dummetts equation of realism with commitments to both verication transcend-
ence and bivalence does accord with a familiar modern picture, a picture that usually
also includes semantic realism: our thoughts and statements are made true or false by
objective reference relations to a verication-independent world that is carved up
into facts that are individuated isomorphically to the statements. But this logical
atomist realism can be resolved into its independent components: rst, semantic
realism; secondly, the logical atomism of determinate facts that correspond exactly to
our statements; and, thirdly, what I have been calling realism, the thesis that the facts
transcend their veriability.21
21
Thanks are due to Stephen Schiffer, Hartry Field, Richard Warner, and Christopher Peacocke for
their very helpful advice and comments.
7
The Supervenience of Social
Meaning on Speakers Meaning
The structure of Paul Grices theory of meaning (1957, 1968, 1969) may be sketched
simplistically as follows: (i) among our semantic notions that of utterers meaning is
more basic than that of the meaning of expression-types in a language; and
(ii) utterers (or speakers) meaning is to be explained in terms of certain Gricean
intentions. There has been much debate on the best explication of those Gricean
intentions. But lets say that they are, minimally, intentions to produce effects, e.g.,
beliefs, in some audience by such and such intended means and with a certain
intended openness. What is crucial is that Gricean intentions can be specied
without mention of any other semantic propertiesunless, that is, you count the
content of propositional attitudes as a semantic property.
There are two ways to interpret the view that speakers meaning is basic, one quite
strong and the other moderate but still interesting enough to be challenged. Accord-
ing to the strong interpretation, anything that can be counted as natural language
meaning can be explained in terms of speakers intentions: this would mean that all
linguistic meaning is by its very nature a function of the use of language in commu-
nication. That would have been a not implausible thesis in the vaguely behavioristic
1950s and 60s; for the idea that language might be used to think in was not then
popular among analytic philosophers. They may have been inclined to point out that
having words running through your head is not exactly the same as using language
to think with, and that, on the face of it, language occurs primarily in communication.
Cambridge philosophy earlier in the century may have been more congenial to the
idea of using language to think in. Russells theories of meaning, and Wittgensteins
in the Tractatus, seem best understood as theories of meaning for language as a
medium of thought.
The strong thesis is then open to the reasonable objection that, intuitively, natural
language meaning could be constituted independently of the use of language in
communication. There are such phenomena as working out ideas on a word
processor, thinking out loud in sentences, engaging in inner speech: can we not
conceive of settings in which they stand on their own, independently of communi-
cation? And, more basically, there is the possibility that we normally think, at least
in part, in a natural language, that our beliefs and intentions depend on having
THE SUPERVENIENCE OF SOCIAL MEANING ON SPEAKER S MEANING
internalized some natural language, perhaps even in a way that is not typically
phenomenologically available. I doubt that it can be argued persuasively on philo-
sophical grounds that this makes no sense. It does not seem to be an obvious
conceptual truth that such private meaningful uses of language would presuppose,
even indirectly, communicative intentions or linguistic manifestations thereof. The
strong Gricean thesis is not something I am inclined to defend.
The Gricean thesis also has a less ambitious interpretation that is still quite
substantive: the literal meaning of sentences in a social or public language, as used
in communication by a population of speakers, is derived from regularities in those
speakers Gricean communicative intentions in using that languagetogether per-
haps with other non-social psychological facts about individual speakers. This I take
to be compatible with those intentions themselves being in some natural language.
For we can distinguish the social-semantic properties of sentences etc. from whatever
semantic properties they acquire from their role in thought. If semantic properties
of both sorts are real they are of course deeply causally connected. But even if we
think in a natural language, social-semantic properties of that language are not
thereby determined, not even by the aggregation of speakers individual thoughts.
For social-semantic properties are not realized unless language is used in certain
interpersonal ways.
The key to this moderate but still strong enough Gricean thesis is that it is
individualist about the basis of social meaning. The idea is that the social phenom-
enon completely depends on the intentions of individual speakers and possibly other
psychological facts, and that this holds asymmetrically, that is, that there is no
essential additional dependence of speakers intentions on social meaning. Individu-
alist contrasts here with social rather than with externalist. Individuals intentions
can of course be about social meaning. But individualism implies that those inten-
tions are not in any essential way constituted by social meaningexcept perhaps in
the way a thought about water is, according to some externalists, constituted by its
relation to water. But that would make some thoughts essentially social only to the
extent that it makes other thoughts essentially wet. I take the Gricean thesis to be
quite compatible with that. I will return to a more threatening way in which speakers
intentions may depend on social meaning.
This all presupposes that there is such a thing as social or conventional literal
meaning. I nd that difcult to doubt, despite skepticism from various quarters.
Conventional literal meaning appears to have a non-trivial role in our understanding
other speakers; without conventional regularities of meaning, ordinary linguistic
communication would not be possible in real world circumstances. The point is
evidently not that linguistic conventions are metaphysically necessary for linguistic
communication. God could program you to interpret a person as speaking a certain
(abstractly constituted) language, and thattogether with your knowledge of the
speakers psychology and general knowledgecould be all you need in order to
understand her systematically. Or you might just be awfully good at coming up with
passing theories of other peoples personal languages in the manner of Davidson,

without making assumptions about what conventionally means what. But appealing
to such abilities is, in my view, rather implausible. Everyone knows more or less what
people in general mean by certain expressions, or what those expressions mean in the
community, and that knowledge is hardly idle in how we go about interpreting
others. To this extent, then, the Gricean position I have in mind is not radically
individualist. It is, though, individualist in its priorities; for it holds that the concept
of social meaning is conceptually derivative from the concept of individual meaning.1
The Gricean position might seem unrealistic in the following respect. The
emphasis in the literature has been on speakers intentions described in a way that
suggests that the intended effects, e.g., certain beliefs, are conceived quite language-
independently. But with many linguistic communications, it is rather implausible
that an ordinary hearer conceives the proposition communicated except in linguistic
terms: think, for example, of technical information about how an internal combus-
tion engine works. Despite certain impressions about the Gricean program, this
phenomenon is perfectly compatible with it. The sort of case envisaged could be an
argument that certain beliefsof a wide range perhapsare best thought of as being
in a natural language. That of course is quite compatible with the moderate Gricean
position we are considering.
1 Explication vs. Conceptual Supervenience

The Gricean project has often been interpreted as aiming at explication and reduction.
A characteristic contribution to the project, according to that view, would consist in
formulating necessary and sufcient conditions:
S means such and such by utterance U =def S produces U with the following
audience-directed intentions . . .
A language L is the language of population P =def Such and such regularities of
speakers meaning exist in P.
Given the goal of reduction, those explicative conditions would have to be non-
circular: what is necessary and sufcient for a speakers meaning something, or for
Ls being the language of P, must be characterizable without presupposing social-
linguistic meaning or the like.
But the basic illumination shed by Grices account of speakers meaning does not
depend on such precise conceptual explication. Paul Grice was amused byand liked
to recountadvice he was once offered: Beware the counterexamples of your
friends! This turns out to have been good advice. The Gricean program, as I see
1
As with most social facts, this one is fuzzy. There are, doubtless, borderline cases of a sentences
meaning literally a certain thing; but this hardly counts against the reality of literal meaning. We should
note that literal meaning is normally, or often, drawn on even when people speak non-literally.
it, does not require getting the details exactly right, because it does not require that
there be exact details to get right. We can describe an alternative, a sort of minimal
Griceanism.
The central issue is whether socially constituted meaning can plausibly be seen
as asymmetrically explainable in terms of individualist psychological factors. So
a clearly interesting minimal Gricean thesis is that facts about communicative inten-
tions are conceptually more basic than facts about public-language meaning, i.e.,
literal meaning as determined by social use. Facts about speakers intentions together
with certain other psychological factors asymmetrically determine social meaning;
the latter conceptually supervenes, asymmetrically, on the former. By conceptually
supervenes I mean roughly this. Suppose that the population P uses the language L.
If we knew all the communicative intentions and other propositional attitudes of
members of P, as well as their correlations with utterances of sentences of L within P,
and if we had time enough and computational power for ideal reection, we could
then directly infer a priori that L is the language of P.
This thesis does not insist on a well-dened notion of utterers or speakers
meaning; and I very much doubt that there is such a notion. What is important, I will
propose, is that whatever the relevant speakers intentions are they constitute an
adequate basis for asymmetric supervenience. In the actual world, certain normal
speaker-intentions are, according to this minimal Griceanism, sufcient for ground-
ing a language. Or rather, this is so given a certain other crucial factor, viz., the
internalizing of a grammar, to which we turn shortly.
Strawson (1964) and Schiffer (1972) fashioned illuminating counterexamples to
the sufciency of Grices original conditions, cases that involved peculiar and
uncommon sorts of complex intentions. Their examples are certainly interesting;
for we would like to be able to explain why in those cases we dont nd what we
would think of, intuitively, as speakers meaning. The strength of our intuitions
about those cases could well seem to support the idea that there is a core notion of
speakers meaning for which a set of necessary and sufcient conditions ought to be
forthcoming. And Schiffer did produce interesting and ingeniously constructed
conditions that apparently accommodated those cases. But, as Schiffer (1987) has
subsequently pointed out, not only are those conditions apparently open to coun-
terexamples, but they also seem quite overintellectualized. The latter point reects a
dilemma: either Gricean conditions are too simple to withstand intuitive counter-
examples, or they are so sophisticated that they at least appear to require of ordinary
speakers far more complexity in their thoughts than they plausibly have.2
Minimal Griceanism evades such issues: it is real-world speakers intentions and
their ilk that matter. Consider, for example, the following familiar issue in the
Gricean literature. Strawson (1964) and then Schiffer (1972) described cases of
2
For a suggestion how to lessen the intellectualizing, see Loar (1981), ch. 10.
deceitful other-directed intentions that fulll certain simple Gricean conditions but
intuitively are not sufcient for a speakers meaning something on some common-
sense understanding of meaning. The minimal thesis simply ignores those compli-
cated examples, and points out that it is a fact that people are rarely deceitful in their
communicative intentions, in the manner of Strawsons and Schiffers examples, even
when they lie. It is hardly a matter of concern that real-world speakers do not have
complex intentions whose point is to rule out such deceitful intentions, if their actual
intentions provide an adequate supervenience base for social-linguistic meaning.
This hardly means that Gricean intentions are irrelevant. On the contrary. We can
take speakers meaning to cover communicative intentions of a certain open-ended
class. Those that involve the Gricean mechanismintentions to produce a response
by means of the recognition of that intentionare central. Real-world speakers utter
sentences that mean p (speaking roughly) with diverse and not very complex
intentions: intentions roughly of letting it be known that p, making it clear that one
thinks that p (both of which imply the Gricean mechanism), reminding that p, calling
attention to p. These are the intentions that form the real world supervenience basis
for social-linguistic meaning. Suppose they are suitably correlated with sentences so
as to capture whatever mapping is present in the abstractly dened language L (given
the internalization of a grammar to be explained). The minimal Gricean thesis allows
that they may not all satisfy some speciable set of necessary and sufcient reductive
conditions that intuitively capture the essence of speakers meaning.3 Patterns of
linguistic utterance with more straightforward intentions will sufce for grounding
social-linguistic meaning, on the minimal Gricean view. If we were given the
paradigm real world Gricean intentions with which people speak English (individu-
ated phono-syntactically), whatever those actual intentions may be, then, the pro-
posal is, we would not be able to make conceptual sense of denying that it is English
(individuated semantically) that they are speaking.
Granted that this will seem an incomplete and unsatisfying claim to some analytic
philosophers who will want to know how to extend the notion of speakers meaning
to possible worlds that are like ours. In what would the similarity consist? That is a
fair enough question. But answering it, or showing that the request for an answer is
based on an illusion, is not essential to the present point. If certain other-directed
more or less Gricean intentions turn out, non-vacuously, to provide an asymmetrical
supervenience base for social-linguistic meaning then the minimal Gricean theory
would be vindicated. The crucial question remains: can that supervenience in fact be
asymmetrical?
3
Some speakers intentions that are intuitively sufcient for speakers meaning but that do not
apparently require Gricean intentions depend on language. A good example is going on record that p, etc.
This arises in a pure form when someone asserts that p even though there is no chance of his Gricean
audience-directed intentions being fullled. This sort of example was held to be a counterexample to
standard explications of speakers meaning. On the present proposal it is irrelevant: such cases simply are
not in the supervenience base.
It should be clear from this formulation that the minimal Gricean need not
support a strong notion of convention of the Lewis (1969)/Schiffer (1972) sort. The
supervenience base will include dispositional expectations that others would mean
certain things by certain expressions and sentences, that others would have the same
expectations of what others would mean, etc. But we must not take it for granted that
people have ordinary knowledge of the regularities that govern utterances of sen-
tences, not so as to presuppose that people know the grammar in the ordinary
unChomskyan sense. More about grammar shortly, but rst a word about the nature
of sentence meaning.
What factors are present in the literal meaning of a sentence? A safe answer is, at
least, this: the literal meaning of S consists in the references of the Ss constituent
terms taken literally, and the way they constrain, together with grammar, the truth-
conditions of literal utterances. What constrain signies here is the hard question.
Trivially, if a speaker means that p on a literal utterance of S, then the literal meaning
of S (whatever it is) will, together with facts about context, determine that the speaker
means that p. The work will be to say just how those semantic facts and context t
together. The picture of literal meaning as a well-dened character, in Kaplans sense,
is probably too simple to yield what we need. Evidently what is said here is too vague
to answer questions about the boundaries between semantics and pragmatics. But the
main point is that (as I am inclined to think) what is essential to sentence meaning
are not so much Fregean factors4 as reference and functions (of some sort) from
contexts to references.
2 Internalized Grammar
It is often assumed that a language embraces an innity of sentences, generated by
indenitely iterable operations. But if that is so, then the actual utterances of
sentences of a language, paired with the intentions with which they are uttered,
will not determine which language is realized in a population. It might seem that
dispositions to usage could adequately ll out the picture. But there is no obvious fact
of the matter about the truth of counterfactuals that involve arbitrarily complex
sentences. Perhaps the only way to specify adequately the antecedents of counter-
factuals about what intentions we would have in uttering very complex sentences
would be to stipulate that our computational powers were enhanced in a way that
4
It is of course possible to distinguish, at least in principle, understanding an expression from getting
the reference right. But it does not follow that expressions have well-dened Fregean senses. A child may
exhibit a correct understanding of cat by manifesting an ability to visually identify instances; an adult by
getting certain zoological facts about cats right, despite lacking the recognitional ability. It seems to me that
neither the recognitional conception nor the zoological conception should be counted as essential to the
sense of cat. Taking a set of such conceptions as the sense probably wont work, since it may be somewhat
open-ended what we would count as sufcient for understanding cat.
secured that our dispositions matched the innite grammar in question, i.e., in a
vacuously circular way.
Let a grammar be an assignment of references to the expressions of a language that
also, on that basis, secures an assignment of truth conditions to the languages
sentences. Several writers have suggested that our ordinary intuitions about mastery
of a language can be met only if we suppose that a grammar of the language is
somehow internalized.5 Its not obvious that this achieves the desired effect. For there
may not be a fact of the matter as to whether we actually internalize an innite
grammar; in particular it is not obvious that a suitably substantive account can be
given of a competence/performance distinction. But there are other reasons to favor
the idea of internalizing a grammar,6 which are compatible with the internalization of
a nite if very substantial grammar.
3 Grammar and Convention

There may be a tension between (a) the underlying psychological-syntactic under-
pinnings of language and (b) the conventional status of linguistic meaning. Conven-
tions that correlate sentences with (sets of ) speakers intentions could well seem a
thin basis for the complexities of the language we socially share.
Suppose we grant the Chomskyan thesis that we are endowed with an internal
representation (consciously inaccessible) of the grammar of our language. Is it
compatible with the spirit of minimal Griceanism to count Chomskyan facts (if
there are such) as being in the supervenience base? Well, clearly it is, if Chomskyan
internalization is itself an individualist fact; and there seems no reason to doubt that.
What about the conventionality of a social language? We can hardly maintain, given
the Chomskyan theory, that the details of the grammar are up to us. It is prima facie
compatible with the Chomskyan position that it is up to us what we mean by our
words, including our logical and other structural words. But the structures them-
selves, once learned, have a powerful autonomy; it is for the most part not open to us
to adjust the grammar of our working language. Still, grammatical regularities in
communicative use are necessary for sentences having the social meaning they have;
and that, together with the straightforward conventionality of word meaning,
perhaps warrants counting the social language as, on the whole, conventional. But
regardless of what we count as conventional, given the individualism of the super-
venience base, minimal Griceanism is quite compatible with these Chomskyan
assumptions.
5
See Chomsky (1980), Loar (1981), Schiffer (1993). Schiffer suggests that a certain sort of translation
algorithm is a better candidate for internalization than a grammar as dened in the text. I quite accept this
possibility.
6
See Loar (1981), ch. 10.
A full assignment of meanings for a natural language will then, according to the
present Gricean thesis, supervene on (i) internal realizations of a grammar of that
language, and (ii) regularities in ordinary speakers intentions with which sentences
are uttered, and (iii) some suitably explanatory interaction between (i) and (ii.) Keep
in mind that those states of ours that realize our beliefs and intentions may them-
selves be linguistic in structure.
Having formulated the minimal Gricean position, we now face the main question.
Does literal meaning asymmetrically supervene on speakers intentions, and other
psychological properties, including causal relations to external factors that do not
presuppose social-semantic factors? The answer will be that it depends on how
speakers intentions and other propositional attitudes are individuated. If they are
individuated in a way that presupposes predicate-reference and truth conditions,
the answer is No. If they are individuated in terms of internally constituted
properties, the answer is Yes. But the latter afrmative is perhaps not terribly
exciting, as we will see.
Until the late 1970s it was taken for granted, a bit carelessly as it may now appear,
that propositional attitudes supervene on non-socially constituted states of individ-
uals (including perhaps their non-social relations to their environment). Beliefs etc.
are of course extrinsically and socially shaped. But, it was assumed, they are no more
socially constituted than, say, measles. Gradually this position became doubtable: for
it was difcult to resist the force of Hilary Putnams division of semantic labor (1975)
and its application by Tyler Burge (1979) to the propositional content of psycho-
logical states.7 Burge argued persuasively that the truth conditions of X believes that
S, for a wide range of instances, are socially sensitive in the following way. Suppose
Hubert believes that he has arthritis in his thigh. He may well accept correction and
grant that his belief was false, even though, as he had conceived arthritis, it includes
just the sort of inammatory condition that in fact affects his thigh. So, it seems, the
truth conditions of Huberts belief depend in part on how other members of his
community use arthritis. What he believes is not a completely individualist fact
about him, but depends also on how the rest of the population use the language. The
same point of course holds of Huberts communicative intentions: his communica-
tive intention, that his audience believe that he has arthritis in his thigh, will, given
the Burge point, itself be individuated in part socially.
The meaning of sentences that contain terms like arthritis will then, it seems,
not asymmetrically depend on what individual speakers intend to communicate.
7
Somewhat less difcult to resist was Kripkes Wittgenstein, who was made to argue, in effect, that all
mental content must be grounded in social systems of rules for applying words and concepts. The
normativity of meaning and content is far too indeterminate a notion to be taken as a serious threat
to the Gricean program. On the most plausible interpretation of the normativity of satisfaction and truth
conditions, it is quite compatible with naturalistic individualism. If normativity is based primarily in truth
conditions, then, it seems to me, we can explain the normativity of meaning in naturalistic terms. And we
wont even have to infer a (categorical) ought from an is.
The social meaning of such a term will not be as it were the sum of purely individua-
list facts about the communicative intentions of users of the term. But then the
asymmetrical dependence claim, the rst minimal Gricean thesis, will be false.
We must emphasize that the point is basically about reference. The references of
communicative intentions that involve socially deferential concepts are determined
by social relations. And literal meaning, as I have suggested above, is fundamentally
a matter of reference and truth conditions. So our supervenience thesis is in effect a
thesis about the dependence of social reference on individual reference. But it says
more than simply that public language reference in general supervenes on individual
references in general. (This may turn out to be more or less true.) The thesis rather is
that the reference of a given term t depends on the specic communicative intentions
of users of t. That is the crux of the difculty that the Burge phenomenon raises.
If speakers intentions are interpreted as involving reference and truth conditions,
then I know of only two ways to eliminate the incompatibility of Griceanism with
Burges thesis. One is simply to give up the asymmetrical supervenience of the
minimal Gricean thesis. That is possible without giving up the supervenience of
public language meaning on speakers meaning; but the dependence of social mean-
ing on what individuals mean will not then be asymmetrical. A person uses the
sentence I have arthritis in my thigh in accordance with its literal meaning if he
intends (say) to inform someone that he has arthritis in his thigh. That sentence will
mean what it does in P only if all or most members of P utter sentences containing
arthritis with appropriate, socially anchored, communicative intentions. This clearly
ts a Gricean explanatory pattern. It might be said that my minimal Gricean thesis
was more a reection of the metaphysical work I hoped the Gricean program
would accomplish than of the program itself. Perhaps. But I am sure that before
the Burge phenomenon was widely appreciated, the Gricean program was a form of
individualism, and did implicitly assume the minimal Gricean thesis. Not that this
bit of history matters philosophically; but the metaphysical promise of Griceanism
was attractive.
Another way to eliminate the incompatibility is by deating Burges thesis. This is
a familiar individualist strategy. Consider the thought (belief, communicative inten-
tion) expressed by I have arthritis in my thigh. Corresponding to the word arthritis,
the thought contains the personal descriptive concept the so and so ailment that the
experts call arthritis. Suppose we can generalize this to cover all those terms that
raise Burge-type concerns. Then the references of the basic concepts that occur in
personal communicative intentions will all be determined non-socially.
This still requires a technical adjustment. We are assuming that arthritis as it
occurs in the public language is a rigid designator, while the concept that occurs in
corresponding communicative intentions is a denite description. The solution of
course is to rigidify that description, and to then make the public language term
arthritis a rigid designator of whatever the shared descriptive concept picks out in
the actual world.
I am uneasy about this solution as a whole. As with all description-theoretic ways

of avoiding uncomfortable apparent consequences of the new theory of reference,
one wonders whether there will be enough basic concepts left over so that all of their
references are (in the present case) personally and not socially determined.
If this uneasiness is taken seriously, the individualist who had hopes of Griceanism
is in an unsatisfactory position. It would be nice, metaphysically, if literal meaning
behaved according to the rst minimal Gricean thesis. For then we would have a
straight account of how the supervenience obtains and why it is asymmetrical, that is,
why communicative intentions are metaphysically more basic than sentence meanings.
And now, it seems, we have no such straight account.
But there is another thesis of the asymmetrical supervenience of social meaning on
speakers intentions that appears to me quite plausible. Suppose that propositional
attitudes can be individuated internally and intentionally but in such a way that
reference is not thereby determined. Twin-Earth intuitions are preserved: but, unlike
the description-theoretic solution, we do not assume that any property-references are
essential to individuating propositional attitudes intrinsically. A brain in a vat may
lack all our references, even to spatial properties; but that brain may nevertheless
have the same Gricean intentions as us, if the latter can be individuated in some
intentional way that preserves how the intention conceives its object.8 Now, let us
recall that the social element in individuating propositional attitudes, using that-
clauses, stems from the social determination of reference: a that-clause that correctly
describes a belief in effect captures its references and truth conditions. On the
presently suggested view, intentions can be individuated independently of reference
and hence of social factors. This internalism is evidently compatible with individu-
alism. Suppose that such an internalist individuation of propositional attitudes
can be made out. Then we can pose a revised Gricean thesis of the asymmetrical
supervenience of social meaning on speakers intentions, etc. All facts about
social meaning asymmetrically supervene, conceptually and metaphysically, on
the following: speakers intentions as internally individuated, and all non-socially
constituted causal and other relations that obtain between individuals and every-
thing else, including other people, external objects and properties. Of course this
supervenience thesis becomes interesting only to the extent that we can characterize,
within this framework, the distinctive role of Gricean intentions in grounding
social meaning. As far as I can see, it is reasonable to take seriously the possibility
of such an account.
The nice thing about the Gricean thesis as usually formulated was that one could
expect some correspondence of constructions of the form S means that p in
population P and constructions of the form Members of P utter or would utter
S with such and such intentions to mean that p.
8
For an account of this sort, see Loar (2003) [Chapter 15 in this volume].
But that p of course captures the references of the intentions, and it is the
reference of propositional attitudes that, given Burges argument, is socially deter-
mined in a crucial range of cases. On the revised supervenience thesis we cannot
preserve that nice correspondence, and the Gricean thesis, however likely to be true,
ceases to be as directly satisfying as it might have seemed. Nevertheless, the super-
venience base would still contain speakers intentions in some essential way. They are
still the key, ultimately, to assigning meanings to sentences, modulo their external
relations to objects, properties, and other speakers. The resulting Gricean thesis is not
as pretty as one would like. But perhaps that does not matter so much, if Grice turns
out to have been right about the central and foundational role of speakers intentions
in the determination of social meaning.
PART II
Philosophy of Mind
Introduction to Part II
Katalin Balog
Brian Loar came to Rutgers University in 1994, which was when I was just beginning
work on my dissertation on the conceivability arguments against physicalism. Word
among graduate students of Brian being a difcult, exciting, and deep philosopher
preceded his arrival; and I soon discovered that all of this was true to a very high
degree. It was around this time that his last, great papers on consciousness and
phenomenal intentionality were taking shape. I was at rst reticent in my interactions
with Brian, mostly due to the demanding nature of his writing and thought; but
I soon found a most charming and welcoming mentor and friend in him. I learned
from him a new way to look at a whole array of philosophical problems; it also seems
to me I learned a new way to look at myself.
Brian was the kind of philosopher that always goes for the most fundamental,
deepest issue. He rarely wrote papers on small, technical questions. His sweep was
grand, his views connecting questions about the nature of phenomenality, intention-
ality, and the mind-world, and language-world interface, the nature of introspection
and its relation to perception, skepticism and our access to the external world,
reference, narrow content, physicalism, reduction, the relationship between the
scientic and the lived world, between subjectivity and objectivity.
His major contributions in the philosophy of mind include his groundbreaking
approach to the mind-body problem (in Phenomenal States, Chapter 10 in this
volume), which originated a research program called the phenomenal concept
strategy;1 and his idea that the concept of reference is an essentially rst-person
concept and that there is a kind of intentionality (subjective, or phenomenal inten-
tionality) in addition to referential intentionality (Chapters 8, 9, 11, 13, 14, 15 in this
volume), themes taken up by various philosophers in the phenomenal intentionality
research program.2
1
See, e.g., Perry (2001), Tye (2003), Carruthers (2004), Aydede and Gzeldere (2005), Hill and
McLaughlin (1999), Papineau (2002, 2006), Balog (1999, 2012a, 2012b), Block (2007), David Chalmers
(2003), and Levin (2007).
2
See, e.g., Horgan and Tienson (2002), Horgan, Tienson, and Graham (2004), Siewert (1998), and
Kriegel (2011, 2013).
INTRODUCTION TO PART II
The central concern in Brians philosophy was how to understand subjectivity in a

purely physical universe. Brian was committed to the reality and reliability of the
subjective perspective; and he found that subjective phenomena like intentionality
and consciousness are, in a certain sense, ineliminable and irreducible to objective
ones. At the same time he believed that intentionality and consciousness are
grounded in the physical. One of his great contributions was showing how to
reconcile these two positions by being a conceptual and explanatory anti-reductionist
about both consciousness and intentionality but a metaphysical reductionist none-
theless. He had a deep commitment both to physicalism and to the reality and
signicance of the subjective point of view.
In the following I provide some background and discuss the main topics of the
papers included in this volume. I try to point out the many interconnections between
them; Brians work in the philosophy of mind was remarkably synoptic. In these
papers he proposed a compelling and unied account of subjectivity, consciousness,
intentionality, and their place in the physical universe.
Chapter 10 Phenomenal States (1997). I will mostly go in chronological order,
charting the development of Brians ideas in the philosophy of mind; but I start with
this seminal paper, which he rst published in 1990 and then in a revised, second
version in 1997, both because of its inuence on subsequent discussions of the mind-
body problem and because its central idea is illuminating when applied to ones
reading of his prior and subsequent papers on intentionality.
In this paper Brian addresses anti-physicalist arguments proposed in the last
few decades. Anti-physicalists (Kripke (1972), Nagel (1974), Jackson (1982),
Chalmers (1996, 2009), etc.) argue that there is an epistemic/conceptual/explanatory
gap between phenomenal and physical descriptions of the world, and from that they
deduce that phenomenal states and qualities cant be identical to physical states and
properties. Brian agrees about the existence of these gaps but denies that the falsity of
physicalism follows from it.
Physicalists have come up with various different strategies to counter these
arguments.3 The one offered in this paper, the phenomenal concept strategy, is
based on the idea that these conceptual, epistemic, and explanatory gaps can be
explained by appeal to the nature of phenomenal concepts rather than the nature of
non-physical phenomenal properties. Phenomenal concepts, on this proposal,
involve unique cognitive mechanisms, but none that could not be fully physically
implemented. Brian proposes that this approach can undermine a central assumption
of the anti-physicalist arguments that he calls the Semantic Premise. He formulates the
Semantic Premise thus: A statement of property identity that links conceptually
3
E.g., eliminativism (Dennett (1991)); analytic functionalism (Lewis (1966)); analytic representation-
alism (Jackson (2003)); the denial that the phenomenal case involves gaps that other special science cases
dont (Block and Stalnaker (1999)).
independent concepts is true only if at least one concept picks out the property it
refers to by connoting a contingent property of that property.
This premisewhich ts bona de a posteriori metaphysical necessities (like
Water = H2O) introduced by Kripke (1972), but not any putative phenomenal-
physical identitiesis the target of Brians ingenious move. According to Brian,
phenomenal concepts are concepts that a person applies directly to her qualia.
They are tokened when, for example, a person sips a red wine and notices rst
sensations of brightness and then a sensation of fruitiness. Brian says that phenom-
enal concepts belong to a wide class of concepts he calls recognitional concepts.
Recognitional concepts are those that enable their possessors to perceptually recog-
nize instances of the concept under certain circumstances. Thus recognitional con-
cepts are connected, via their inferential roles, with basic perceptual concepts,
sensory inputs, images, etc. Brian says that recognitional concepts have the general
structure is of that kind where the demonstrative purports to refer to a kind as
exemplied through a perception or image of an instance of the concept. Here is an
illustration. Jerry sees a platypus for the rst time in the zoo, and forms the concept
ANIMAL OF THAT KIND,4 where the demonstrative is focused on its reference by
his perception. Possessing this concept, he is able to recognize other instances.
Here is the key idea. Phenomenal concepts are a special kind of recognitional
concept. Their basic application is to ones own phenomenal states as they occur, for
example, ITCH AGAIN. Of course, we also apply phenomenal concepts in mem-
ory and in reasoning and to other people. Unlike other recognitional concepts, a
phenomenal concept does not refer via a contingent mode of presentation. Instead, it
is applied directly to an internal state. Brian suggests that a phenomenal concept has a
mode of presentation that is essential to its referent. What this means is that when
tokening a phenomenal concept, the reference is presented via a token of that very
referent. The mode of presentation exemplies the very property referred to. For
example, when tokening PAIN, the mode of presentation is the very painfulness of
the token of pain to which the concept is applied. Thus the mode of presentation of a
phenomenal concept is essential to its referent.
If this is the case, we can see how the Semantic Premise might be false.
If phenomenal concepts are direct recognitional concepts, they do not have contin-
gent modes of presentation. Nor do their physical counterparts in any putative
phenomenal-physical identity. But there is nothing in Brians account of direct
recognitional phenomenal concepts that rules out that these concepts, as well as
their referents, are, or are realized by, purely physical states. PAIN and C-FIBER
FIRING both can refer to the same physical state (pain) but this state is being
presented once in the theoretical/descriptive mode (C-FIBER FIRING), and once
directly, quotationally (PAIN). Thus we can see why the Semantic Premise, though
4
I will indicate concepts by capital letters.
initially plausible, need not be true, indeed, why it is false on an internally consistent,
physicalist assumption. Moreover, his account shows why the conceptual, explanatory,
and epistemic gaps between the phenomenal and physicalfar from being problematic
for physicalismare to be expected, given his account of phenomenal concepts.
Brian in this article shows a new way to be physicalist: fully realist about phenom-
enology and subjective experience, fully aware of the ineliminable signicance and
meaning that only the subjective perspective provides for human beings; yet also fully
committed to physicalism and physics as the ultimate account of the fundamental
ontology of the universe even as admitting the lack of explanatory reduction for
phenomenal states.
Chapter 8 Social Content and Psychological Content (1988) is his inuential
opening salvo in a series of papers dealing primarily with the nature of intentionality
and its relationship to phenomenality (Chapters 8, 9, 11, 13, 14, 15 in this volume).
Ever since Putnams (1975) and Burges (1979) seminal papers the idea that the
content of many ordinary concepts is partly dependent on factors external to the
head has gained signicant traction. It has also seemed reasonable, however, that
from the point of view of psychological explanation all that matters is how things are
in the head. Some philosophers proposed that, even if ordinary content goes beyond
what is in the head, what is in the head still determines another kind of content
so-called narrow content. In this paperwritten at the same time or earlier than
Subjective Intentionality (Chapter 9 in this volume), despite the later publication
dateBrian enters this debate by drawing a distinction between social content,
which is externally-socially determined and essentially tied to language and commu-
nication; and psychological content, which is central for psychological explanation,
is determined by how one privately conceives things, and is narrow.
While it has been widely held that sameness of de dicto or oblique ascription of
belief implies sameness of psychological contentthat is, sameness in how one
privately conceives of thingsBrian argues that psychological content is not in
general identical with what is captured by oblique that-clauses, and that psycho-
logical content is not especially elusive for that. He employs Kripkes (1979) Pierre
example to show that sameness in social content (content expressed in oblique that-
clauses) doesnt imply sameness in psychological content; and Burges (1979) arthritis
and Putnams (1975) Twin-Earth examples, as well as some variations on them, to
show that difference in social content doesnt imply a difference in psychological
content. Psychological content, Brian claims, is tied to perceptual phenomenology
and private conceptual role, so it is narrow.
The important point for Brian is that two thoughts that are identical in their truth-
conditions but play a different role in psychological explanation do not just differ
functionally, but also in their content. In other words, Brian proposes that what is in
the head, and is thus involved in psychological explanation, is itself a kind of
(narrow) content that is different from ordinary content, and it has to do with how
one conceives of the world. As he says, It is difcult to see how one can consider how
one oneself conceives things without that in some sense involving what ones
thoughts are about (p. 162). These ways of conceiving should be considered
content by what he calls the Principle of the Transparency of Content: whatever
appears, from an unconfused perspective, as content, is content.
He briey mentions two objections to narrow content that he grapples with in
subsequent papers as well. One is the Argument from That-clauses and it goes like
this. There could be narrow content only if purely internal factors could determine
truth conditions; after all, any content must specify the ways the world must be for
that conception to be accurate. Truth conditions in turn can be captured by that-
clauses; but that-clauses capturing the content of my thought that water is delicious
and the parallel thought of my twin on Twin Earth where the relevant liquid is not
H2O but XYZ express different truth conditions even though they are internally
exactly alike. It follows that, since purely internal factors do not determine truth-
conditions, they cannot determine genuine content either.
Though Brian agrees that there are no that-clauses that capture the personal,
psychological truth conditions of our thoughtshe takes Putnam and Burge to have
shown that that-clauses are always sensitive to social and external factorshe offers,
without fully endorsing, the idea that narrow content determinesnot truth-
conditions butrealization conditions that pick out worlds corresponding to ones
conception of things. These worlds are centered worlds; they correspond to how the
thought conceives the world as being, given the context specied at the center (to take
into account the indexical element of thought). He calls these context-indeterminate
realization conditions; they are shared by intrinsic duplicates and are sensitive to
Burgean, Putnamian, and Kripkean deviations between truth-conditional content
and psychological content.5 I will take up discussing this proposal in my review of his
next paper, Subjective Intentionality.
The other objection is that narrow content is unspecied; that proponents should
supply a better explication of the notion. In this paper Brian seems to describe narrow
content as a matter of conceptual role and perhaps perceptual phenomenology. He
claries his concept of narrow content in his subsequent papers, especially in
Chapters 14 and 15 in this volume, as he keeps trying to hone in on the idea of it.
Chapter 9 Subjective Intentionality (1987). In this rich and complicated paper,
Brian continues his defense of narrow content against various objections. In response
to the Argument from That-clauses (also familiar from his previous paper), he
considersand ultimately rejectsone proposal to specify narrow content on
which narrow content is truth-conditional in some way other than ordinary content.
The proposal is that narrow content is a mapping from contexts to broad contents.6
5
The idea of realization conditions owes some similarity to Dennetts discussion of notional worlds
(1982).
6
Context is best understood here as a centered world. For similar ideas see, e.g., Kaplan (1989),
Stalnaker (1978), White (1982), Fodor (1987, 1998), and Brians earlier (1982) paper, Must Beliefs Be
So, for example, if Twin Oscars concept TWATER shares the same narrow content
with Oscars concept WATER, it is in virtue of the fact that their respective concepts
would pick out the same referent in each different context (where these contexts are
understood as the context that determines reference)in spite of the fact that in their
respective environments Oscar refers to water and Twin Oscar to XYZ. Brian rejects
this account of narrow content on the grounds that the relevant function from
context to broad content would be the same for the thought that gin is delicious
and the thought that vodka is delicious; because this function is determined in part
by a subfunction that maps a context onto whatever liquid is at the origin of a certain
kind of causal path leading to the thought (p. 171).
That concepts can be associated with mappings from contexts (centered worlds) to
contents is a pretty straightforward idea given that meaning must be determined by
factors involving the thinkers head and factors outside the thinkers head. What is
controversial is the nature and importance of these mappings; in particular, whether
the concepts in the head (the narrow aspect of concepts) that mapping accounts
identify are the right kind of items to gure in psychological explanation, and equally
importantly, to have a kind of content.
On some accounts of concepts, arguably they are not (at least for the question of
content). For example, on Fodors (1987, 1998) theory, concepts are individuated by
syntax and orthography (or whatever corresponds to orthography in the language of
thought) and by reference. Reference is externally determined by what the concept
asymmetrically depends on at a world. Only the syntax and orthography are internal.
Inferential role, in particular, plays no role in individuating a concept although as
part of the context at a world it may play anot constitutive, but merely causalrole
in partly determining asymmetric dependence. So on Fodors theory, COW is
associated with a function that maps a world w onto whatever COW asymmetrically
depends on (if anything) at w. On this view, narrow concepts are not very
interestingmappings notwithstanding, there doesnt seem any particular reason
to attribute to them their own kind of content.
On the other hand, take Frank Jacksons (1998) account. According to it, the
relevant internal aspect of a concept has to do with its inferential role, which can be
captured by descriptions associated with the concept. For example, WATER has an
internal aspect involving narrowly determined descriptions, very roughly, the
description watery stuff, that is, clear, odorless, etc. . . . liquid around here, which,
given facts about the actual world, determines that its reference is H2O. Accordingly,
Jackson thinks that the relevant mappings specify what someone understands when
they grasp a concept and that these narrow concepts are involved in rationalizing
explanations of both thought and action.
Sentences? (Chapter 4 of this volume). For later developments see, e.g., Chalmers (1996, 2006), Chalmers
and Jackson (2001), Jackson (1998).
Brian wanted narrow content to both gure in psychological explanation and

involve its own kind of content. However, he despaired of the idea that there can be
narrow descriptions specifying how one privately conceives the world; as we have
seen, he was convinced by Putnam and Burge of the sensitivity of most that-clauses to
social and external factors. This might be behind his rejection of the whole mapping
conception of narrow content. To return to his example of the concept GIN and the
concept VODKA, Brian suggests that they would be associated with the same
mapping as in each case the relevant function . . . maps a context onto whatever
liquid is at the origin of a certain kind of causal path leading to the thought (p. 171).
It is not clear, however, what specic account of narrow concepts Brian presupposes
in his critique that makes him assume that the relevant mappings for GIN and
VODKA would be determined by a description such as whatever liquid is at the
origin of a certain kind of causal path leading to the thought. This assumption would
seem to go well beyond what content externalism requires.
As it turns out, Brians own proposal is not so far from the mapping conception. It
is that narrow content corresponds to the set of worlds that are as the thinker
personally conceives things as being (p. 172), given the context specied at the center.
He calls these, as in his paper Social Content and Psychological Content (Chapter 8
in this volume), context-indeterminate realization conditions. But this account can be
equally formulated as a mapping from centered worlds to content; except that
whereas the previous notion portrays narrow content as a function from context to
ordinary content, Brians realization conditions seem to determine a mapping from
contexts to something like would-be contents, that is, contents that a thought would
have if ones conceptions were accurate. So, for example, on the previous conception
of narrow content, Berts concept ARTHRITIS maps the actual world to the disease
known as arthritis, whereas on Brians account, it maps the actual world to a disease
both of the joints and muscles. It would be rewarding to fully tease out the relation-
ship between the two accounts; here it will sufce to say that it seems that Brians
notion is the more individualistic and psychological. Brian himself did not say much
by way of specifying these worlds, apart from gesturing at the way subjects conceive
the world as being. But this is how he saw things must go with narrow content;
attempts to specify it from an objective point of view are doomed; narrow content
can only be grasped, if at all, from the subjective perspective.
The other objection he considers to the notion of narrow content he calls the
Argument from Unmotivation. It is a kind of skepticism about whether represen-
tational content of any outward-directed kind can be determined by internal
physical-functional factors (p. 172). No objective, third-person information about
how the brain works can, by itself, justify the ascription of content, it seems, without
appeal to some external relation between the brain and the environment. It follows,
the argument goes, that one cannot be both a physicalist and an advocate of narrow
content. Brian spends the better part of the paper to show why this line of thought is
misguided; that the existence of subjective intentionality, or narrow content, is
compatible with physicalism in spite of the fact that narrow content as such is
impossible to discern from the objective point of view.
The key strategy Brian employs in defense of this claim anticipates the one he later
elaborated in much more detail in his 1990/1997 paper Phenomenal States
(Chapter 10 in this volume) in defense of the thesis that the existence of phenomenal
consciousness is compatible with physicalism even though consciousness as such is
impossible to discern from the objective point of view. Here, as in that paper, he
claims that there could be two cognitively independent concepts of the same
physical-functional phenomenon; even when this violates what he calls in the later
paper the Semantic Premise. The Semantic Premise, recall, is the thesis that an
identity statement that links conceptually independent concepts is true only if at
least one concept picks out its referent by connoting a contingent property of that
referent. The idea in his (1990/1997) paper is that phenomenal concepts involve a
subjective perspectivethey depend on the ability to recognize or discern certain
states in the having of them. In this way, they pick out their referents directly (so
without connoting a contingent property, as do theoretical-physical concepts).
This ensures their cognitive independence from physical-functional concepts but is
perfectly compatible with physicalism.
There is a parallel to this in the case of our subjective conceptions of intentionality.
According to Brian, in the subjective mode we conceive of the meaning of our
thoughts directly, without conceiving of any external relation to them as such; rather
we conceive of their meaning by displaying them, so to speak, in our minds.
A reexive thought (a thought about the meaning of a thought) incorporates in a
certain waynot unlike the case of our subjective conceptions of phenomenal
statesthe thought itself. The directness of our subjective, as Brian puts it,
display-conceptions of our own thoughts explains their distinctness from any
objective physical-functional conceptsthe ones appealed to in the Argument
from Unmotivation. Nevertheless, as in the phenomenal case, all this is perfectly
compatible with physicalism.
However ingenious and important this idea isand it will yield rich philosophical
return in subsequent papersit doesnt, as Brian seems fully aware, quite provide
the answer to the original question, which had to do specically with the existence
of narrow content. The Argument from Unmotivation poses a challenge to the idea
that any purely physical-functional goings-on that narrowly characterize the head
can constitute content that always seems of an outward-directed kind. That
meaning can be conceived subjectively and directly (that is, not as an external
relation) doesnt mean that what is conceived is purely subjective, narrow inten-
tionalitythat seems like a non sequitur. There appears to be a crucial difference
between the phenomenal and intentional case: whereas phenomenal concepts pick
out something narrow, something in the head, subjective intentional concepts
might pick out, in a direct way, an external relation between our concepts and
their objects.
Brian notices but is unhappy with this interpretation of the situation. What he
needs to show to answer the Argument from Unmotivation is that subjective
reection hones in on the narrow content of our thought, and it does so in such a
direct way that it is impossible to discern this from the objective perspective. But the
concept of intentionality could be essentially subjective even if there is no narrow
content. The reason he musters for reection discerning narrow content is not
entirely persuasive. He says [f]or a person may think his thought is about an external
object . . . when no suitable object exists. That seems as much an instance of the
phenomenon as when a suitable object does exist (p. 177). But it is quite a leap from
the observation that one can apparently think of non-existent objects to the conclu-
sion that there is meaning that is constituted by wholly internal factors. And even if
reection on thoughts about non-existent objects showed that one can reect on
narrow contentsince Brian probably does not want to deny that it is possible to
reect subjectively on ordinary (wide) meaninghe needs to explain differences
between a wide and a narrow reection on content. He will ll in those gaps in his
argument in his later papers Transparent Experience and the Availability of Qualia
and Phenomenal Intentionality as the Basis of Mental Content (Chapters 14 and 15
in this volume).
Chapter 11 Can We Explain Intentionality? (1991). Here Brian offers a critique
of Jerry Fodors asymmetric dependence theory of intentional content. Fodors
account is a version of the idea that meaning is information, that a predicate refers
to a property of which its tokens carry information. Such accounts form a crucial part
of putative reductive, naturalistic explanations of belief, that is, reductive explan-
ations of what persons believe in terms of physical-functional facts. The question of
intentionality was one of the most hotly debated topics in philosophy in the 1980s.
According to the dominant understanding of what naturalizing requires,7 semantic
concepts need to be functionally explicated so as to t intentionality into the natural,
physical order. This paper stakes out, in the context of this program, Brians unique
position on intentionality. His paper has a dual agenda: to criticize Fodors particular
naturalistic account, and through it, the assumption underlying much of the natur-
alistic program that semantic notions are characterizable in functional terms; and at
the same time argue that the failure of these naturalistic accounts doesnt mean the
failure of naturalism itself, in the broader, metaphysical sense.
Fodors asymmetric dependence account is meant to provide a naturalistic solu-
tion to the problem of error; how it is that meaning is robust in that it can survive
false applications. According to Fodor,
A predicate F means a property P if things that are P cause F and any such causal relation
between some other property and F is asymmetrically dependent on the former relation. (1987,
pp. 11920.)
7
See, e.g., Fodor (1987), Dretske (1981, 1988), and Stalnaker (1984).
So, according to this theory, what explains that your concept SHEEP can be
mistakenly applied to a goat is that your false application depends on the independent
causal relation of sheep to SHEEP but not vice versa.
Brian has various objections to this account for socially deferential terms. He thinks
that the account has the best chance of being correct in the case of demonstrative,
recognitional concepts. But even in this case, he argues, it doesnt work; it doesnt
provide either necessary or sufcient conditions for meaning. The reason lies in the
perspectival nature of recognitional concepts, in the fact that some perspectives of use
are reference determining, pace Fodor. As Brian puts it, some perspective is part of
the sense of the concept, of how it conceives its reference (p. 234). Appealing to
perspective in explaining falsity without asymmetric dependence, and asymmetric
dependence without falsity, Brian is trying to make a case for the more general thesis
that intentional notions are ineliminable in any armchair explication of reference.
The general understanding at the time, and to a large degree ever since, has been
that consciousness is the really hard part of the mind-body problem and intention-
ality is easier to naturalize. Brian disagrees: he argues that they are both equally hard,
in fact impossible, to naturalize in the conceptual (if not the ontological) sense, in the
sense of providing a philosophical explication in naturalistically kosher terms.
Building on one of the themes of his paper Subjective Intentionality (Chapter 9
in this volume), in which he suggested that there are essentially subjective, reective
notions of meaning we can form in introspection that are conceptually independent
of any objective, physical-functional notion, he now suggests that our core notions of
meaningquite like our core notions of consciousnessthemselves are ineliminably
subjective; that we cannot form an adequate objective, third-person conception of
meaning at all. He takes up the task of explicating that core notion of meaning in his
paper Reference from the First-Person Perspective (Chapter 13 in this volume).
Chapter 12 Elimination versus Non-reductive Physicalism (1993). In this
paper Brian explores the metaphysics of mind. Finding both dualism and elimina-
tivism about the mental unattractive, he wonders whether a non-reductive, as
opposed to reductive, physicalism is a viable alternative. He concludes that it is
not. Non-reductive physicalism endorses ontological physicalism (the thesis that all
fundamental entities and properties are physical) but denies metaphysical physical-
ism, which Brian understood to be the thesis that mental properties are reducible to
physical or functional properties. He further characterizes what an interesting non-
reductive thesis would have to look like by stipulating that such an account has to
1) be realist about both mental and physical truth; 2) posit no independent causal
powers for the mental; and 3) account for how objective mental resemblances can be
naturalized. He discusses Stephen Schiffers nominalist account, sentential dualism.
Schiffer is trying to reconcile the apparent irreducibility of the mental with the
implausibility of massive causal overdetermination of behavior by mental and phys-
ical properties. On his view, there are no mental properties, but there are true mental
ascriptions. Mental ascriptions are not reducible to physical statements; nevertheless,
physicalism can be true in the ontological sense. Brian argues that, if Schiffers view is
not to be a form of eliminativismthat is, if it is to be interestingit runs into the
problem of causal overdetermination.
Chapter 13 Reference from the First-Person Perspective (1995). The last three
papers of this volume, of which this is the rst, contain Brians most developed views
on the nature and concept of intentionality and its relationship to phenomenality.
Despite the gap in their publication dates, these papers were all written in close
succession (the last two papers were already circulating in the late 1990s). The term
phenomenal intentionality, or subjective intentionality as he calls it in his eponym-
ous paper Subjective Intentionality (Chapter 9 in this volume), labels two distinct
though related theses. One is a thesis about the subjective nature of our core concepts
of meaning. This thesis, suggested at the end of Can We Explain Intentionality?
(Chapter 11 in this volume) is the topic of this paper. The other is a thesis about the
existence of a kind of (narrow) intentionality, distinct from referential intentionality,
determined by phenomenology and conceptual role alone. That is the topic of the last
two papers in this volume (Chapters 14 and 15).
While this paper has received somewhat less attention than his other phenomenal
intentionality papers, it contains important and interesting ideas about the concept
of meaning and the nature of reference. In it, by way of providing a philosophical
analysis of the concept of meaning, Brian actually proposes an argument for the
determinacy of reference.
According to Quines (1960) inscrutability thesis, there is no fact of the matter as
to whether the term rabbit refers to rabbits, undetached rabbit parts, un-
disconnected rabbit stages, etc. Brian considers this thesis rst from the objective
point of view; from an understanding of reference as a certain sort of causal relation
connecting tokenings of the concept to instances of the referent. He points out that
the causal relations that seem to be candidates dont uniquely determine reference.
Take a visual demonstrative concept like THAT TREE, and a relation O that is the
relation of belonging to a causal chain that prompts the tokening of THAT
TREE. O doesnt single out a tree as opposed to a retinal image, a tree surface,
undetached tree parts, etc. What seems to secure unique reference is the implicit
qualifying concept associated with the demonstrative, in this case the concept
TREE. But of course the problem of inscrutability arises similarly with kind concepts
like TREE; whose determinacy in turn depends on the determinacy of singular
demonstratives. Given that the reference of most other types of concept depends in
some way, according to Brian, on demonstrative reference, there seems to be no way
out of a vicious circle of interpretation.
Brian considers the suggestion that, in the absence of any objective factor that
singles out relation O as restricted to, say, trees, as opposed to tree surfaces, as the
relevant reference relation for THAT TREE, reference is scrutable only by convention.
He thinks that while this might strike one as plausible for third-person ascriptions of
reference, it is bizarre when considered in the case of our own thoughts. It is
intuitively obvious that, in the rst-person case, I know that my concept THAT TREE
refers to that tree and not, for example, a collection of undetached parts of that tree.
Brian of course doesnt think that this in itself is decisive. He points out that this
asymmetry between the third-person and rst-person perspectives is quite compat-
ible withactually, has an explanation in terms ofthe disquotational-deationary
theory of reference and truth. But what Brian is interested in is not the refutation of
the deationist interpretation of the third-person/rst-person asymmetry. He is
rather interested in presenting an alternative account, in terms of a non-deationary
theory of reference. He proposes an account of referencevia an account of the core,
subjective concept of referencethat bears out the rst-person intuition of deter-
minate reference as correct. He acknowledges the force of the Quinean argument,
without conceding its point.
Brian explains the core concept of reference in terms of subjective reection on
what he calls disquotational pairs, that is, term-object pairs. He asks us to reect on
these by entertaining the concept *THAT TREE*,8 and THAT TREE; *THAT
HAND*, and THAT HAND, etc., where the second member of each pair is a
visual demonstrative concept. Now as you entertain these concepts, the concept
and object will appear as linked in a certain, phenomenologically salient way. And
here is the central, crucially important idea of the paper: according to Brian, we think
about reference, in the canonical case, as the relation holding between pairs linked in
that way. The concept of reference conceived in this way will have, as Brian puts it,
an appropriate combination of subjective and objective factors: the pairs are
those presented as noted, and the relation we conceive as objectively holding of
those pairs (p. 266).
Brian suggests that linked in that way applies across concept categories, so that
pairs across concept categories will have an intuitive phenomenological similarity;
pairs such as *THAT TREE*, and THAT TREE; *PLATO* and PLATO; *COPPER*
and COPPER will appear similar in a certain way. Consequently, linked in that way
is not tied to particular instances. Rather, it depends on past and potential discrim-
inations of the way disquotational pairs appear to be connected.9 As he puts it, this
concept is in its way demonstrativethat relation, and . . . in a certain way the
disquotational, or mention-use, conguration constitutes that demonstrative
concepts dening perspective (p. 266).
The question is: how does providing a phenomenological conceptual counterpart
to causal-informational accounts of referencethat is, an account that has an
appropriate combination of subjective and objective factorshelp resolve the
8
I use * to indicate whatever operation in the mind plays the role quotation plays in language.
9
It is important to note that according to Brian, this appearance is purely phenomenal and is not
mediated by the concept of reference, which would make the account circular and useless for the purposes
of Brians argument.
issue of the determinacy of reference?10 Brians answer is that it does via showing two
things. First, that, on the assumption of referential determinacy, we can explain how
it can be that our subjective concept of referencenon-illusorilytracks which of
various external relations qualify as reference for our concepts (so that the concept
THAT TREE indeed refers to that tree and not, say, that tree surface). And second,
that, at the same time, it is not possible to understand how this can be from the
objective, third-person perspectiveeven if reference is indeed determinate. The
reason we cannot disambiguate reference from the third-person point of view is
that, understood properly, reference has a subjective, phenomenal component that
just like phenomenal experience itselfis not explicable from the objective perspective.
So, in a way, Brians strategy is the inverse of the deationary strategy: whereas the
former explains the rst-person appearance of determinacy as an illusion, Brian
explains the third-person appearance of indeterminacy as an illusion.
This, of course, should not for a moment be taken as a denial that reference consist
in perfectly naturalistic, physical-functional relations between concepts and their
referents. It is rather an instance of Brians general outlooklabeled by Chalmers
(2002) in the context of the metaphysics of consciousness type B materialism
according to which one should take subjective phenomena and the subjective per-
spective fully seriously and then explain the failure of various third-person strategies
to understand the phenomenon via appealing to the conceptual isolation of rst-
person subjective understanding.
Chapter 14 Transparent Experience and the Availability of Qualia (2002).
In this paper, Brian attempts to combine the representationalist insight about the
transparency of normal visual experience with the qualiphile commitment to intro-
spectable intrinsic qualia. He rejects both what he calls the standard view of raw
qualia according to which qualia are not essentially representational, but rather, like
paint on canvas, are individuated independently of their representational properties,
and what he calls the standard representationalist attack on qualia, according to
which no matter how well you try to introspect your normal visual experience, all you
will notice is the apparent objects and properties that your visual experience presents.
The upshot of the paper is that non-relational, yet intentional, qualia, are needed to
explain certain intuitions about phenomenal sameness; and representationalism
doesnt have the resources to do that.
Against representationalism, Brian argues that the Inverted Earth11 thought-
experiment shows that it is possible to have qualitatively different color experiences
representing the same surface properties of objects and so it cannot be that all we
10
Searle (1991, 1992) makes a similar move by appealing to phenomenal consciousness to account for
the determinacy of reference that he, as Brian, nds so compelling from the rst-person case. But he
doesnt explain the discrepancy between the subjective and objective perspectives on indeterminacy by an
appeal to the difference in subjective and objective conceptions of reference. For similar ideas, see also
Kriegel (2014).
11
Block (1990).
discern when we attend to what it is like to have a certain experience are the
represented objects and their properties. At the same time, he also points out that
this thought-experiment doesnt give us a grip on raw, unrepresentational color
qualia; the qualitative features we are conceiving of are best regarded as essentially
property-directed qualia. Property directedness, according to Brian, is part of the
phenomenology of perceptual experiences and it is what explains the transparency
of normal experience. What two qualitatively identical color experiences, for
example, have in common is their property directedness with a certain qualitative
character; this property directedness, however, might be involved in presenting
different properties or no properties at all. Brian maintains that introspecting qualia
is not easy or natural; we can do it only after abstracting away in thought of the object
of the experience.
Because we can make sense of these property-directed experiences occurring in the
absence of their normal references, we can make sense of property directedness
independently of any referential properties. In other words, Brian argues that per-
ceptual experience is essentially intentionalbut not essentially representational. In
his paper Phenomenal Intentionality as the Basis of Mental Content (Chapter 15 in
this volume), he extends this idea to thought by arguing for a kind of intentionality
that is independent of representational content.
Chapter 15 Phenomenal Intentionality as the Basis of Mental Content (2003).
In this, one of his last published papers, Brian brings it all together. He mounts a
grand defense for the idea that in addition to referential content, thought has a
content that is narrow, and non-referential. And even more signicantly, he nally
provides an answer to the question that has been to some degree up in the air in
previous papers: in what does narrow content lie? Building on his theme of the
narrow intentionality of perceptual experience (Chapter 14 in this volume), he
locates the origins of narrow intentionality for thought in phenomenal features of
perceptual experience.
As for his defense of narrow content, he introduces, in addition to briey rehearsing,
or amending earlier ones, an argument from the brain-in-a-vat scenario. Thought
experiments like the inverted spectrum, or Inverted Earth scenarios, make sense of
introspectable, narrow qualia. They show that content can vary while ones mental
states remain qualitatively the same. But they are compatible with content being
essentially referential content. To make sense of introspectable, narrow content, we
need, according to Brian, the more potent medicine of brains-in-a-vat.12 Because,
according to Brian, we can coherently conceive of an envatted-brain twin that shares,
from the rst-person point of view, ones way of perceiving and thinking about the
world, we can conceive of narrow, non-referential intentionality. The picture that
emerges is one that leaves the view that reference is partly externally determined in
12
For a similar argument see Horgan, Tienson, and Graham (2004).
place while denying that all content is essentially referential. Of course, Brian doesnt
take intuitions about envatted brains as proof that there is such a thing as internal
intentionality. But he thinks they make at least a case for the coherence of the
idea. Giving a physicalistically respectable account of what narrow content consists
in goes the rest of the way. In the rest of the paper, he presents this account
in elaborate detail.
The key idea is that the narrow content of perceptions and thoughts consists in
their directedness, in their purporting to refer. This directedness is a phenomenal
feature that is shared between me and my envatted-brain twin. Brian explains the
intrinsic directionality of all thought, step by step, from the phenomenal direction-
ality of perceptual experience, elaborated in more detail in Chapter 14. As he puts it,
one cannot phenomenologically separate the purely qualitative aspects of perceptual
experience from its purporting to pick out objects and their properties (p. 302). The
internal aspect of perceptual experience is, in his memorable line, the paint that
points (p. 314). This phenomenal directionality, which belongs primarily to per-
ceptual states, underlies the phenomenal directionality of singular perceptual
demonstrativeslike THAT LEMONthat incorporate the perception. Type
demonstrative recognitional conceptslike YELLOW, LEMON, etc.pick out a
kind by virtue of past perceptions and a disposition to pick out future instances
from its dening perceptual perspective. They are not descriptions embedding
singular perceptual demonstratives; one doesnt need to have particular instances
in mind to have the recognitional concept. Their directionality then doesnt come
from any singular demonstrative concept but from the directionality built into their
perspective. The directionality of non-perceptual concepts is more mediated; they
acquire their directionality via their conceptual connections with perceptual and
other concepts. The subjective intentional properties of non-perceptual concepts are
always a matter of, as it were, looking sideways via their connections with perceptual
concepts (p. 310). In the end, all thought has its intrinsic directionality derived from
its various links to perceptual experience.
In his previous papers (Chapters 8 and 9), Brian hypothesized that narrow,
phenomenal intentionality determines what he called realization conditions. That
view is close to the mapping conception of narrow content we discussed earlier; on
this view narrow content is something that, given a particular environment, deter-
mines a particular broad content. But while some philosophers interpreted Brians
views in this paper as a variant of the mapping conception,13 Brian simply doesnt
return to that idea here. It is probably not that he thought the idea that such a
mapping exists is implausible. It is more likely he ignored it because he thought
realization conditions could not be specied as such. For one thing, he rejected the
idea that narrow content can be descriptively characterized. In the paper he provides
13
Burge (2003), p. 448; Lycan (2008), }13.
a critique of the internalist-descriptivist strategy according to which all our concepts

can be considered as descriptions involving narrow concepts that include logical
concepts, sensory concepts, the concepts of cause, physical object, and various spatial
relations. The problem with this view, according to Brian, is that spatial concepts and
even the concept of physical object have a demonstrative element; their reference is
externally determined. More generally, he thought that our grasp of narrow content
being essentially subjective, even if there is an objective specication of it (as he
thought there is), we would have no use for it as we wouldnt see that it is an objective
equivalent to the subjectively grasped phenomenon.
Concluding Remarks
After all the discussion of subjective intentionality and narrow content, one might
ask what the philosophical signicance of the issue of narrow content is. An obvious
answer, and one that Brian is explicit about, is that narrow content gures in the best,
most plausible understanding of psychological explanation of behavior. But in a
broader sense his interest seems driven by his general philosophical outlook. As he
put it (Chapter 15 in this volume):
Conceptions of mental content in the analytic tradition have tended to be phenomenologically
impoverished, largely because of the emphasis on language and reference. And when we turn
to the phenomenology, as I will try to show, we do get a grip on internal intentionality.
(pp. 2923)
Brian thought the subjective perspective is essential for self-understanding; and

consequently he wanted to explore the mind as fully as possible from the subjective
point of view. He found (Chapters 14 and 15 in this volume) that the subjective
perspective revealed that there is a purely internal aspect to both qualia and inten-
tionality. He further found (Chapters 11 and 13 in this volume) that even the idea of
reference cant be properly understood from the objective perspective. At the same
time, he was equally driven to show that one doesnt have to turn ones back on
subjectivity and phenomenology, implausibly sanitizing ones outlook on what is
real, to make the world safe for physicalism. Did he succeed in his program? In the
case of phenomenal experience, I think the answer is afrmative. The case of
intentionality is more difcult; it is harder to rightly understand both the obstacles
to naturalization, and the ways around them.
8
Social Content and
Psychological Content
By psychological content I shall mean whatever individuates beliefs and other

propositional attitudes in common-sense psychological explanation, so that they
explanatorily interact with each other and with other factors such as perception in
familiar ways. In discussions of what appropriately individuates propositional attitudes
there occurs the following kind of argument. Some thesis about psychological content
is proposed, about, say, what constitutes it in general or constitutes some aspect of it.
The reply is made that the thesis fails to capture the correct individuation conditions
because correct ascriptions of attitudes using that-clauses count them the same when
the proposed thesis distinguishes them or vice versa. Here are some examples.
1. There is the idea that the perceptual ability to discriminate objects of a given
kind, cats say, may constitute a concept of that kind of object. Think of concept here
as meaning a certain abstraction from the individuation conditions of certain beliefs
and other attitudes, about cats, say. Then the capacity to have certain beliefs about
cats, beliefs that have certain psychological contents, would involve the ability
perceptually to discriminate cats. Variants of this idea have seemed natural to
empiricists and recently to certain functionalists: what better criterion for having
certain ordinary concepts than the ability systematically to pick out their instances?
But there is a problem when we consider how apparently we individuate beliefs, as
Stephen Stich points out. Suppose I tell a blind person and a sighted person that
there is a cat in the next room, and they believe my report. It seems natural to say that
they both come to have the same belief: the belief that there is a cat in the next room.
Our intuitions remain the same if we change the example by replacing the merely
blind subject with a person like Helen Keller whose perceptual deciencies are
staggering.1 On these grounds Stich regards as false any theory that implies that
the sighted and the blind persons beliefs have distinct contents.
The premise appears correct, for believes that there is a cat in the next room does
seem univocally assertible of both the sighted person and Helen Keller. For if each
were sincerely to assert there is a cat in the next room their words conventionally
1
Stich (1983), pp. 667.
SOCIAL CONTENT AND PSYCHOLOGICAL CONTENT
would mean the same; and so by ordinary criteria the belief ascription would be true
of each on an oblique or de dicto univocal reading. (The sameness of belief ascription
is not then merely a function of a common de re reference to the kind cat.)
2. There is the very general thesis that the psychological content of a persons
attitudes consists in their conceptual or cognitive-functional roles, thus presupposing
that our common-sense system of psychological explanation individuates attitudes
along non-social, individualistic lines. Tyler Burge has mounted an imposing
counterargument.2 He has two objections: that the individualist conceptual role
theory counts beliefs as different which common sense counts the same; and that
the conceptual role theory counts beliefs as the same which common sense counts
different. Both objections rest on the following well-known example of Burges.
Suppose that a person who is otherwise a normal English speaker believes that he
has arthritis in his thigh, and that he also has many true beliefs about arthritis, for
example, that he has it in his wrists and ankles. When a doctor tells him that arthritis
cannot occur in the thigh (arthritis means an ailment of the joints) he is surprised
but takes the doctors word for it. Now consider that earlier belief, which he would
have expressed as I have arthritis in my ankles. On the conceptual role theory, that
belief should count as distinct from the doctors belief that his patient has arthritis in
his ankles. For the two have, or had, crucially different ideas about what arthritis
means, and consequently the two beliefs have (should on a conceptual role theory be
counted as having) crucially different conceptual links to other beliefs. But, as Burge
argues, common sense ascribes the same belief to both: the belief that the patient has
arthritis in his ankles. Thus sameness of conceptual role is not necessary for sameness
of psychological content.3 Now it seems clear that Burge is right that the belief
ascription applies both to the doctor and to the patient in his uncorrected state on
a univocal reading, one that is, moreover, oblique or de dicto.
That sameness of conceptual role is not sufcient for sameness of content is argued
as follows. Suppose the patient (whom well now call Bert) had lived in a world much
2
Burge (1979), pp. 73122.
3
It is fair to say that many have taken this to be the message of Burges paper, that is, that beliefs are not
individuated in common-sense explanation by their conceptual roles. Of course Burges direct point is
about the presuppositions of ascriptions of beliefs. But because of the widespread supposition that the
signicance of this point lies in the stronger point, I am taking the stronger point to be the consequential
burden of Burges paper. He writes: It is expressions at oblique occurrences within content clauses that
primarily do the job of providing the content of mental states or events, and in characterizing the person
(Burge (1979), p. 76). And again he writes of the idea that the distinctively mental aspects can be
understood fundamentally in terms of the individuals abilities, dispositions, states, and so forth, con-
sidered in isolation from his social surroundings as follows: our argument undermines this latter
suggestion. Social context infects even the distinctively mental features of mentalistic attributions. No
mans intentional mental phenomena are insular (Burge (1979), p. 87). More recently, Burge has argued
for an apparently less stringent position, namely, that even though there might be a level of scientic
psychology which is individualist, (a) there are important examples of scientic psychology that cannot be
construed individualistically, (b) such non-individualist explanation is legitimate as regards scientic
methodology, and (c) common-sense psychological explanation is non-individualist. In this paper I am
concerned to argue that Burges observations about belief ascriptions may be accepted while denying (c).
like this one, but one in which doctors apply arthritis not to a disease specically of
joints but to a broader class of rheumatoid ailments including one that can occur in
thighs. Suppose that Berts history in that world had from an individualistic point of
view been identical with what it is in the real world, and that therefore before visiting
the doctor Bert had had a belief he would have expressed as I have arthritis in my
thigh. As we see things, would that have been a belief that he had arthritis in his
thigh? Burge says no, and again I believe he is right. In that world Bert may have had
a belief that he had tharthritis (as we may choose to say) in his thigh; for there
arthritis does not mean what it means among us. So in the actual and the counter-
factual situations, the individualistic facts about the conceptual roles of Berts beliefs
are the same, but distinct belief ascriptions are true of him. Burge draws the strong
conclusion that the content of a persons beliefs depends in part on social facts that
are independent of his cognitive make-up, social facts of which he may not be aware.
Stichs point and the rst of Burges points have a common structure: it is said to
be false that discriminative abilities are partially constitutive of psychological con-
tents, and that individual conceptual roles are constitutive of psychological contents,
because those theses would count beliefs as distinct to which the same de dicto or
oblique ascription univocally applies.
3. Metalinguistic contents have often been invoked for beliefs that normally would
not be expressed metalinguistically. Thus suppose a person asserts there are elms in
Spain but knows of no non-metalinguistic distinguishing features of elms among
trees or of Spain among countries. It could be said that the content of this persons
belief involves the conception of elms as those trees which among us are called
elms and of Spain as that country which among us is called Spain.
One objection (there are others I shall not discuss) is the following. Suppose an
Italian who would assert Ci sono olmi in Spagna is in the same situation as our
English speaker; he does not know elms (olmi) from other trees or Spain (Spagna)
from other countries. To be consistent we then say that his beliefs involve metalin-
guistic reference to those Italian nouns. But of course if everything else is normal we
should rather say that the two speakers believe the same thing, namely that there are
elms in Spain. This holds of them de dicto and univocally. They have the same belief,
while the metalinguistic analysis counts them as having distinct beliefs. Once again,
I nd the premise about the oblique or de dicto ascription correct, not open to an
unforced denial; it would be misguided to insist that we must throw out the univocal
that-clause and substitute a pair of metalinguistic that-clauses in order to describe
those beliefs correctly.
The question then to be addressed is the relation between de dicto or oblique
ascriptions of beliefs and their psychological contents, between such ascriptions and
their individuation in common-sense psychological explanation. I shall argue that
psychological content is not in general identical with what is captured by oblique
that-clauses, that common-sense constraints on individuation induce only a loose t
between contents and that-clauses, and that this does not make contents ineffable or
even especially elusive. Let me emphasize that the topic is not some theoretical
renement of common-sense psychology, but ordinary everyday psychological
explanation.
Behind the three argumentsagainst recognitional concept theories, conceptual
role theories, and metalinguistic concept analyseslies something like the following
assumption:
(*) Sameness of de dicto or oblique ascription implies sameness of psychological
content.
Perhaps there are facts about the occurrence of indexicals and demonstratives that
would generally be perceived nowadays as counterexamples to (*) in its unqualied
form. But it is pretty evident, I think, that the above arguments presuppose a version
of (*) restricted to general terms (including also proper names; cf. Spain). So we
have something like:
(A) Sameness of the de dicto or oblique occurrence of a general term in two belief
ascriptions implies, if everything else is the same, sameness of the psycho-
logical content of the two beliefs thus ascribed.
My reply to the above arguments involves denying (A): sameness of the general terms
in a pair of belief ascriptions does not (even though all else is equal) ensure that the
ascribed beliefs are individuated as the same belief in common-sense psychological
explanation.
I shall also argue the falsity of the converse of (A), viz.
(B) Differences in de dicto or oblique ascription imply differences in psycho-
logical content.
This is important again in connection with Burges anti-individualism, for it seems to
be required for his argument that sameness in the conceptual roles of thoughts is not
sufcient for their sameness in psychological content.
A variant of a well-known example of Kripkes may serve to introduce the reason
for rejecting (A). In the original example4 Pierre grew up monolingual in France,
where he had heard of a pretty city called Londres; he was moreover disposed to
assert Londres est jolie. Subsequently he was taken to live in London, not knowing it
to be the Londres he had heard of; the part he lived in was unattractive and he was
disposed to assert London is not pretty. Our ordinary principles of belief ascription
lead us then to say, as Kripke points out, both that Pierre believes that London is
pretty and that Pierre believes that London is not pretty. These ascriptions are true on
an oblique reading5 and London is univocal as we use it.
4
Kripke (1979).
5
This assumes that when we ascribe beliefs obliquely to speakers of other languages, the correct way to
do so is (roughly) to translate how they would be expressed.
Now Pierre might have been more fortunate; he might have been taken to an
attractive part of London and thus been happy to assert London is pretty, still
unaware that this was the Londres he had heard of. The upshot is interesting: Pierre
believes that London is pretty is true by virtue of the earlier facts about Pierre, and it
is true by virtue of the later facts. And its double truth is on a univocal oblique
reading. The point does not depend on translation; parallel cases arise in which
someone mistakenly thinks a name names two things, and ascribes the same predi-
cate twice.
But how many beliefs does Pierre have? In other words, how many belief-types are
involved, as that is individuated by common-sense psychology? Clearly there are two
beliefs, and they are as distinct as my beliefs that Paris is pretty and that Rio is pretty.
Those beliefs would interact differently with other beliefs in ordinary psychological
explanation. Perhaps in France Pierre came to believe that were he ever to live in
Londres he would live in the same city as Oscar Wilde, and he retains this belief. But
he does not draw from the conjunction of this belief and his later beliefs the
conclusion that he now lives in a pretty city also inhabited by Oscar Wilde; and
this is not because he has not bothered to put them together.
These beliefs not only are individuated by common-sense psychology as distinct in
their psychological roles; it also seems quite appropriate to regard them as distinct in
content. The differences in their interactive properties ow from differences in how
Pierre conceives things, in how he takes the world to be, in what he regards the facts
as beingthat is, differences in some semantic or intentional dimension. And yet one
and the same oblique belief description is true of Pierre univocally by virtue of these
beliefs that are distinct in their psychological content.
Let us now look at some beliefs involving general terms. Suppose that Paul, an
English speaker, has been raised by a French nanny in a sheltered way. She speaks
English with Paul, but amuses herself by referring to the cats around them as chats
(she says shahs, pronouncing the s) and never as cats. Paul acquires thereby a
perfectly good recognitional acquaintance with cats and many beliefs about them, but
he does not know that in English they are properly called cats. Suppose he forms the
belief he would express as All chats have tails; it seems we are then justied in
asserting that Paul believes that all cats have tails, on an oblique reading.6 As it
happens, he occasionally sees his parents, who speak of animals called cats. Because
no cats are ever present, nor any pictures of cats, Paul does not realize that cats are his
familiar chats. Now Pauls parents tell him various things about cats, in particular
that they all have tails. On this basis it is again true of Paul that he believes that all cats
have tails. And it seems clear that Paul has two beliefs, with distinct psychological
contents. For they interact potentially with other beliefs in different ways despite
their common univocal ascription.
6
How else to represent the belief obliquely than by translating chats as cats? Keep in mind that the
point does not depend on translation.
Had Pauls parents told him of Manx cats, it would have been true that Paul
believes that all cats have tails and believes that not all cats have tails, on oblique
unequivocal readings. But we should say that those beliefs are not inconsistent in
their psychological contents,7 and this means that these oblique ascriptions do not
individuate Pauls beliefs in a way that reects their psychological relations.
Stephen Stichs reply to the general thesis that recognitional abilities may be crucial
to the individuation of beliefs was that Helen Keller and a sighted person both may
believe that there is a cat in the next room. Now suppose that Paul had had a slightly
different phonological history. His nanny used the English cat, but somehow Paul
got the idea that there are two different meanings of cat, each referring to a distinct
kind of animal (cf. crab), the kind he recognizes at a glance and the kind his parents
speak of. This idea is so entrenched that when his nanny and his parents, on one of
their rare joint appearances, both say theres a cat in the next room, Paul believes
that there are two animals in the next room and is interested to see nally one of the
unfamiliar cats. Now Helen Keller has conveniently dropped by, and she overhears
Pauls parents remark. It seems that she thereby acquires only one belief that theres
a cat in the next room, but that Paul has two such beliefs, distinct as types in their
psychological individuation. Helen Kellers belief is then identical in type with at most
one of Pauls beliefs. And so it is left open that the content of the other belief is
constituted in part by Pauls ability to recognize cats. Naturally (A) is thereby falsied.
Now consider again Tyler Burges rst thesis, that two beliefs may differ in their
conceptual roles (by virtue of different understandings of some concept) and never-
theless have the same content. Suppose that when Paul leaves home he lives in France
for a while, learns about a rheumatoid ailment called arthrite, and comes to believe
that he has it both in his thigh and in his ankles. He would be surprised to learn that
you cant have arthrite in your thigh. As it happens Paul has a perfectly good
understanding of the English arthritis, which he does not realize is renderable in
French as arthrite (perhaps he never sees them written down). He is unfortunately
given to hypochondria, and comes to believe that he has two problems with his
ankles, in his words arthrite and arthritis. It seems that believes that he has
arthritis in his ankles is doubly but univocally true of Paul, by virtue of beliefs
with distinct psychological contents. Had he been less inclined to hypochondria his
English belief could have instead been that he does not have arthritis in his ankles.
Now that belief would clearly have been psychologically consistent with his French
belief that he has arthritis in his ankles, but not with his actual English belief. The
latter two therefore must be distinct in psychological contentunless, that is, you
want to deny that the relevant sort of consistency is consistency in content.8
7
In Loar (1987a) [Chapter 5 in this volume], I discuss the signicance of the phenomenon in
connection with Saul Kripkes A Puzzle about Belief (1979).
8
Perhaps it is some sort of formal consistency. But then beliefs would not be individuated in common-
sense psychological explanation by their content, which seems implausible.
So Burges observation that believes that Bert has arthritis in his ankles is true of
the doctor and Bert on an oblique univocal reading, which I have agreed is correct,
does not imply that their beliefs have the same content as that is individuated in
common-sense psychological explanation.
I shall not go into the third argument, the one against metalinguistic analyses. But
it should by now be clear that it does not follow from the fact that believes that there
are elms in Spain is univocally true of those English and Italian speakers that their
beliefs are not metalinguistic with regard to their respective languages and therefore
distinct in their psychological contents.
It may be useful to distinguish two theses in what I have been arguing, namely, a
thesis about how beliefs are individuated in common-sense psychological explan-
ation, and a thesis about content, the former being more minimal than the latter.
Common-sense psychological explanation appeals to various elementary struc-
tures in the relations among beliefs, wants, and so on. There are motivational
structures: xs believing something, xs believing something else to the effect that
given the rst thing doing A would have a certain result, and xs desire for that result
may explain xs doing A. There are inferential structures: xs believing something and
xs believing something else to the effect that the rst thing is sufcient for a certain
further thing may explain xs believing that further thing. There are structures of
irrationality: xs believing something, xs desire for a certain thing, and xs belief to
the effect that the rst thing could rule out the second may conjointly explain xs
compartmentalizing or suppressing the rst belief. And so on.9
These structures apply to beliefs and desires only as they are appropriately indi-
viduated. The simple cases I have been discussing can be spun out in obvious ways to
show that the appropriate individuation conditions are not captured by oblique
readings of ordinary belief ascriptions. For example, imagine Pauls English belief
that he has arthritis in his ankles interacting with a French belief of his that if he has
arthritis in his ankles he should apply heat: not much happens as a result. The correct
individuation transcends, in some crucial respects at least, what ordinary ascriptions
capture. And I am speaking always of common-sense explanation.
As for psychological content, if it is not captured by that-clauses, what constitutes
it? Are we entitled to regard my alleged underlying psychological individuation as
determining a kind of content? I shall return to this question.
Let us take up the second strong Burgean thesis. Suppose arthritis had meant
tharthritis: even if Berts non-socially-described ruminations remained the same it
would not have been true that Bert believed that he had arthritis in his ankles
and thigh. Therefore sameness of individualist conceptual role is not sufcient for
sameness of psychological content.
9
The circumlocutory wording is meant to avoid propositional variables that appear to presuppose that-
clauses. For a way of understanding these structures of rationality and irrationality, see sections 4 and 6 of
my accompanying paper A New Kind of Content (1988a).
Now Burges premise, that our old belief ascription would not then be true of Bert,
is correct; and it is an important discovery that belief ascriptions are thus sensitive to
social facts which may not be reected in believers own versions of things. But the
further thesis, that content as it is individuated in psychological explanation depends
on independent social factors, is I think not correct.
The anti-individualist conclusion depends on (B), that differences in oblique
ascription imply differences in psychological content. But the intuitions that in the
cases of Paul and Pierre led us to reject (A) ought also to bring us to reject (B). We
should hold that despite their different ascriptions Berts belief that he has arthritis in
his ankles and his belief that he has tharthritis in his ankles have the same psycho-
logical content, because they have the same potential for explanatory interaction
with other beliefs; what intuitively appeared to determine that potential in the case
of Paul and Pierre was how they, as it were, personally conceived things. But let me
give some new arguments directed specically against (B) and the Burgean thesis
that sameness of individualist conceptual role is not sufcient for sameness of
psychological content.
That (B) is false is already accepted by whoever takes a certain widespread view of
Twin-Earth cases. Although those Twin-Earthling thoughts that they express using
water are, as the story goes,10 like ours in their personal conceptual roles, we cannot
ascribe to them the thought that, say, they bathe in water. Twin-Earthlings have
referential contact not with H2O but with a chemically distinct if phenomenally
indistinguishable substance, and so we cannot translate their water into English
as water and hence cannot assert of them anything of the form believes that . . .
water . . . . Conceding these facts about belief ascriptions, many have found it intui-
tive, indeed have taken it to be the point of Twin-Earth cases, that Twin-Earthlings
thoughts have the same content as ours as that is individuated in psychological
explanation, the same narrow content.
Such intuitions appear to be vindicated by two rather different thought
experiments.
1. Suppose Bert is a full member of two English-speaking communities that
differ linguistically in small ways of which he is unaware. The rst is ours, where
arthritis means arthritis; but in the second arthritis means tharthritis. Let the
individual facts about Bert be as in Burges case. How are we to describe him? If
there is no reason to choose just one of the languages as his language, then
apparently the best thing for us to say is that Bert believes that he has arthritis in
his ankles and believes that he has tharthritis in his ankles. But in explaining Bert
psychologically the natural thing to say is that he has just one belief, one way of
conceiving what is wrong with his ankles. Similarly, we may imagine a commuter
10
Putnam (1975).
between Earth and Twin Earth who is biworldly in his language without knowing
of the systematic referential differences between English and Twin English.
He would assert Water quenches thirst. Again it seems that two belief ascriptions
are in order, but that they should be seen as merely different extrinsic descriptions
of what is, as regards psychological explanation, the commuters one way of
conceiving things.
2. Here is a different thought experiment. One is given a diary and told that it is by
either an Earthling or a Twin-Earthling but not which. An entry says: No swimming
today; we think the water is too rough. This reports a psychological explanation, one
that loses nothing from our ignorance of the diarys provenance, that is, from our
ignorance of whether it would be correct, in reporting that thought obliquely in a
that-clause, to use water or twin-water. It is not that we switch rapidly back and
forth between two explanations, one in terms of water and the other in terms
of twin-water; all we have to have been told is that the diary was written in one of
a class of worlds that resemble Earth in the relevant respects. Or, again, suppose
that I do not know whether in Berts linguistic community arthritis means arthritis
or tharthritis, but that I know all the relevant individualist facts about Bert. I read in
his diary: I fear I have arthritis, and so today I have made an appointment with a
specialist. It is difcult to accept that we do not fully understand the psychological
explanation given here, despite our not being in a position to produce the correct
that-clause. We understand the diarists explanations because we know how they
conceive things.
What is there to be said against these intuitions in favor of narrow content? Two
objections could be thought to have force. The rst is that so-called narrow content
cannot capture an intentional property; for the two beliefs in the Burge case and those
in the Twin-Earth case do not share truth conditions. Content should mean inten-
tionality, and intentionality is a certain directedness of thoughts onto things, prop-
erties, states of affairs, in short, truth conditions and the components of truth
conditions. The second objection is that there is no appropriate way to specify the
common content in those pairs of beliefs; and thus the notion of narrow content is
just hand-waving.
I shall not say in response to either objection that there are that-clauses which do
not contain water or arthritis and which capture the common content of those
pairs of beliefs. I am quite prepared to concede that that-clauses are so generally shot
through with social and causal presuppositions that narrow content cannot in
general be captured thus.
There is a kind of reply to the objection concerning intentionality and truth
conditions that I believe is important but shall not develop at length here. Put
sketchily the idea is this: the conceptual roles of thoughts are distinct from their
truth conditions, and in more than one sense do not determine truth conditions (except
perhaps for certain demonstrative judgments involving perceptual discriminative
concepts).11 But common-sense psychological explanation of the sort we have

intuitively appealed to in discussing Paul and Pierre individuates attitudes according
to their conceptual roles, the specication of truth conditions having some further
functionon which more anon. How can there be content without truth conditions?
This is, I think, not merely a terminological question, for it involves intuitions about
the apparent intentionality of ones own thoughts as judged from a rst-person
perspective. The point is that the conceptual roles of ones thoughts determine how
one conceives things, and it is difcult to see how one can consider how one oneself
conceives things without that in some sense involving what ones thoughts are
about. And that is appropriately called content, by a Principle of the Transparency
of Content: if something from an unconfused perspective appears to be content then
it is a kind of content. As I say, this is sketchy, but I mention it to register that the
defender of narrow content has more than one line of defense: it would not matter
that narrow content does not determine truth conditions if there is a kind of content
that does not involve truth conditions.
Furthermore the demand for a narrow or individualist account of intentionality, in
the sense of the outward directedness of thoughts onto states of affairs, is not
unanswerable. Consider Berts diary again. I do not know whether Berts use of
arthritis involves a misconception because I do not know the social facts about it.
But I do know this: how the world would be if Berts conceptions are or were not
misconceptions. How Bert thinks of thingsas that is described from an individu-
alist perspectiveappears to determine a set of possible worlds, namely, those in
which Berts thoughts are or would be true if they are or were not misconceptions.
Call that set of worlds the realization conditions of Berts beliefs. If my ability to
explain Bert psychologically presupposes a grasp of something intentional, some-
thing like truth conditions, then it would seem that my grasp of the realization
conditions of Berts beliefs is sufcient. The diary thought experiment supports this.
Realization conditions are of course not truth conditions in our ofcial sense. The
truth conditions of a belief depend on some that-clause which correctly ascribes it,
and as Burge has shown that is not in general determined by individualist facts. Thus
I am not saying that the real truth conditions of thoughts are their realization
conditions. I am not proposing a redenition of anything. Rather I am saying that
if psychological explanation involves a mapping of thoughts onto possible states of
affairs, then realization conditions are there for the taking. There is nothing recondite
in the idea; our common-sense understanding of others delivers the realization
conditions of their beliefs without our having a name for them.12
11
See Loar (1982) [Chapter 4 in this volume].
12
This is perhaps similar to what David Lewis proposes in Lewis (1981).
As several people have pointed out to me, my context-independent realization conditions are quite
similar, including their detachment from that-clauses, to Daniel Dennetts notional worlds (see Dennett
(1982)). I discuss the relation between my account, as that is elaborated in section 6 of A New Kind of
Content, and Dennetts theory in footnote 6 of that paper.
An adjustment to this suggestion is needed. Suppose I nd a diary with the entry:

Hot and sunny today; phoned Maria to invite her to the beach. Now, the date has
been torn off the page. Still I appear to understand the diarists explanation of his/her
phoning Maria, despite not knowing the truth conditions (in one sense) of the
thought expressed by hot and sunny today. Is there not, however, a sense in
which I do know the truth conditions? Suppose on Tuesday one thinks it is hot
and sunny today, and on Friday one thinks it is hot and sunny today. They have the
same truth conditions in the sense of conditions of truth in abstraction from context.
Call them context-indeterminate, by contrast with the context-determinate truth
conditions that determine sets of possible worlds. Then if understanding the psycho-
logical explanation given in the diary requires in some sense knowing truth condi-
tions, they need merely be context-indeterminate.
To put this together with the former point, we may say this: if psychological
explanation involves intentionality, then context-indeterminate realization conditions
are all the intentionality required.13
There is still the objection that we cannot in general specify the narrow content of
thoughts. Now if this means merely that narrow content is not in general captured by
ordinary that-clauses, it is difcult to see why it is an objection. We have perfectly
sound intuitions about when to distinguish the beliefs of Paul, Pierre, and Bert,
despite the fact that that-clauses do not make the right distinctions. If we then lack
specications of narrow content, in the sense in which we have specications of wide,
social content, that must mean that psychological explanation does not require such
specications. We get along perfectly well without them; we individuate beliefs and
understand their realization conditions without an ofcial system of generating such
specications. Narrow contents are not ineffable; we get at them in context via
various devices. (a) We use that-clauses with one eye on the background facts:
different narrow contents are implied by Paul believes that cats have tails in the
chat context and in the cat context. (b) We report a persons words, or approxi-
mations thereto, together with other utterances that help us to interpret his words:
Bert says I have arthritis in my ankles, but he also says I have arthritis in my thigh.
13
Jerry Fodor has recently proposed that narrow content be construed in terms of functions from
contexts to sets of possible worlds (in a paper given at UCLA, spring 1985). And in correspondence he has
suggested that this would preclude the need for the realization conditions I propose here. But I do not think
that Kaplanesque characters will in fact do the job of capturing the narrow content of general terms, or not
unless I am missing something. The reason is that, if you treat natural kind terms as if they are pure
indexicals whose semantic values are determined by context, then water and alcohol would count as
having the same narrow contentviz., that function which maps a natural kind term onto a natural kind in
accordance with certain causal facts in its history of use. A Kaplanesque character, a function of that kind,
may well individuate the narrow content of a very special feature of thought such as the rst-person
pronoun, for it can be argued that self-ascription is the only aspect of narrow content whose reference-
function always maps the belief onto the believer. But in the narrow individuation of beliefs involving
water and alcohol we want their contributions to be different, in accordance with their conceptual roles.
That combination of indexicality and substantive conceptual content is what context-indeterminate
realization conditions are supposed to capture.
The second helps one to understand the narrow content expressed by the rst. (c) We
ascribe narrow content by producing words that have the same narrow content
for us. Imagine a Twin-Earthling whose language is Twin German; it may help to
render one of his beliefs as I bathe in water even though there is no water there and
those are not the words he would utter. Not that we have much opportunity for
interpreting Twin-Earthlings, but interpretation by approximately matching narrow
contents is one of our fundamental techniques in psychological explanation.
It now seems to me somewhat extraordinary that we should have thought that
psychological states are captured by a neat set of content-specications. But what
then are that-clauses for? Of course they play a central role in psychological explan-
ation, given suitable background information; but we have been misconceiving that
role in thinking that they dene precisely the individuation conditions of psycho-
logical states. That-clauses on their oblique readings are sensitive, either directly or
indirectly via translation, to how beliefs would linguistically be expressed, and that is,
as the examples of Paul and Pierre show, only loosely related to psychological
content. Now, as Burges cases show, that-clauses capture how a belief would be
expressed by exhibiting something that is equivalent in social content (as we might
say) to what the subject would utter, given his deference to the usage of his linguistic
community. This enables that-clauses to capture certain extra-psychological relations
of propositional attitudes to independent states of affairs, what we may think of as
their socially determined truth conditions. The fundamental usefulness of this is that
we may then describe people as conveyors of more or less determinate information,
which remains constant even when the psychological contents of their states vary.
That-clauses enable us to impose a grid of socially regularized information on the
vagaries of individual psychology. Presumably the system of propositional attitude
ascription is part of a larger framework of restraints, even, on the centrifugal
tendencies of the thoughts of each of us.
9
Subjective Intentionality
1 Internal Mental Content

To say that mental content or intentionality is in the head is to say this: internal
properties of the mind constitute how a person represents the world to himself.1 Not
they constitute the means by which he represents the world to himself, but they
constitute the way the world is according to how he represents it to himself, how he
conceives the facts, how he conceives the world as being.
This thesis is, I think, correct on an interpretation, one that I hope will be strong
enough to raise the hackles of anti-internalists, those numerous claustrophobes who in
recent years have undertaken to externalize the mind. To be clear about where the
disagreement will not lie, let me bow to naturalism; this is no last-ditch Brentanianism.
Internalism can be more cautious than what concerns me here and still contro-
versial, for example in this form: theoretical psychological explanation ought to
appeal wholly to internal states.2 Or in this form: common-sense psychological
explanation individuates thoughts in terms of internal features, possibly functional
or conceptual roles.3 Both of these weaker internalist claims can be noncommittal
about my current concern, whether representational contentintentionality4is in
the head.
2 Two Objections
There are at least two important reasons for denying that representational content is
in the head. First, there is the Argument from That-clauses, as follows. The strong
thesis implies that internal factors determine truth conditions; for understanding how
1
This paper is based on a talk (Understanding Others) given at the Pacic Division APA meetings in
March 1987. I am grateful for remarks by the commentators, Charles Chastain and John Wallace, and also
by Anthony Brueckner, Barry Loewer, and Stephen Schiffer.
2 3
Stich (1983). A possibility suggested in Loar (1988b) [Chapter 8 in this book].
4
I once (unfortunately) used intentional to mean not representational but non-extensional (Loar
(1981), p. 4). Suppose you have a theory on which the functional individuation of thoughts accounts for the
non-extensionality of our ascriptions of them, and on which the relevant functional individuation is
secured by internal properties. That would imply that internal properties determine intentionality. But,
evidently, what I am concerned with here is a distinct thesis.
SUBJECTIVE INTENTIONALITY
another person conceives the facts involves understanding how the world must be for
that person to have correctly conceived the facts, and what could that mean but
understanding the truth conditions of his thoughts? But whatever determines it, the
truth condition of a thought is captured by some that-clause. Now the well-known
Twin-Earth and Burgean arguments (Burge (1979), (1982)), that internal factors do
not determine content, stem from the observation that two thoughts can be alike in
their internal properties even though no that-clause captures their common truth
condition. It would be incorrect to describe a Twin-Earther as believing that gin is
delicious given the deciency of ethanol on his planet, despite the identity of his
internal states with those of an Earthling with that belief; and similarly incorrect to
ascribe to someone who says I have hives the belief that he has hives if those to
whose usage he defers mean some other disease by hives, even though his (incor-
rect) conception of the disease they mean is just like our (correct) conception of
hives. To give the content of those beliefs using that-clauses we require neologisms:
twin-gin, chives. It is argued that no other that-clausese.g., metalinguistic ones
do justice to the truth conditions of such beliefs about gin, twin-gin, hives, and chives.
Two thoughts may thus be alike in internal features but not in truth conditionsand
therefore, the Argument from That-clauses concludes, not in content, not in how
they represent things as being.5
Then there is what may be called the Argument from Unmotivation, vaguer but (as
I see it) more important than the Argument from That-clauses. Internal properties
whether biochemical, neurophysiological, psychofunctional, or common-sense
functionalcannot motivate the ascription of, or explain, or imply, externally directed
truth conditions. How could a description of another persons merely biochemical
or neurophysiological states (think of what such a description might look like), or of
internal causal-functional relations among those states, motivate the leap to truth
conditions involving trees or animals or planets, or even external bodies or external
space? The merely internal properties and interconnections of brain states, however
exhaustively described, will never tell us that they are to be interpreted with such and
such truth conditions concerning external possible states of affairs.6
The Argument from Unmotivation is not this: indenitely many ways of assigning
truth conditions to internal states are equally appropriate on internal grounds. That
does seem a compelling objection to a purely internalist conception of content; but
the two objections are distinct. Consider some rst cousins of internalism, viz.,
5
Two renements. i) Distinguish anti-individualism from externalism. The latter, at least notionally,
leaves open the possibility that content depends in part on distal causal relations which are not socially
determined; ii) By externalism I mean the denial not only of internalism but also of localism. This
externalist argument would show also that content is not a function of internal facts coupled with facts
about local perceptual input.
6
It should be clear that the point does not assume a controversial semantic theory, e.g., that names are
not descriptive or that predicates are rigid designators; the most description-oriented theory of meaning is
apparently vulnerable. No purely internal properties motivate assigning even descriptive content about
external states of affairs.
localist theories of content. Make the content of observational thoughts depend on

proximal or local input (there goes a cat-like object; this one is larger than that
one), and make the contents of the rest depend on their cognitive relations to
observational thoughts. The indeterminacy, it might be held, can be eliminated by
Ramsifying: non-observational mental terms are cognitively tied to each other and to
observational terms so that collectively they constitute a Theory, which then spreads
satisfaction conditions from observational to non-observational terms. Thus the
mental term x is a cat has as its satisfaction conditions something like this: x is a
member of a kind that stands in such and such relations to things, situations, etc.,
that are F, G, and H (here we have the interpreted observational terms) and in so and
so relations to other kinds (here you bring in other non-observational terms, which
in turn have connections with observational terms, etc.). Naturally, metalinguistic
properties and relations would play a role; a person may conceive of albatrosses as:
those birds called albatross. Now, put aside the question whether we can select from
a persons total Theory a set of constitutive connections that would make non-
observational concepts interpersonally shareable and therefore unproblematically
ascribable. There is an even more fundamental problem with localist theories of
representational content, namely, Unmotivation.
On the Ramseyan model truth conditions can spread from observational thoughts
to the rest only if mental logical connectives and forms have suitable truth-conditional
propertiesonly if a certain mental connective has the truth-table for or, a certain
mental analogue of concatenation means predication, a certain structure means
existential quantication. Whence such semantic properties? You cannot just say:
these features of thought behave like disjunction, negation, existential quantication,
predication. It is still a substantive question what the point would be of assigning
semantic interpretations on that basis. (Can there not be purely syntactic systems?)
Given a constraint that certain inferences be deemed truth-preserving, we could say:
the tendency to infer p$q from p alone and from q alone and so on, gives $ the truth-
table for or. But what internal facts would warrant such a constraint, failing a
general motivation for assigning truth-values to non-observational thoughts on
internal grounds? Consider things more holistically. Grant that observational
thoughts have truth conditions; what in the internal inferential relations among
observational and non-observational thoughts would warrant assigning truth condi-
tions to the latter collectively? Given only the internal facts, there seems no objective
reason not to count the non-observational thoughts as merely some among myriad
contentless states with causal relations to the observational thoughts. If such internal
states play a certain role in psychological explanation, then would that not show
merely that psychological explanation can, to that extent, abstract from truth condi-
tions in favor of functional properties?7
7
This point is familiar from a certain way of motivating a two-factor theory of content (cf. Loar
(1981)). Such a theory may go on to say that thoughts nevertheless have a kind of internal or narrow
The Argument from That-clauses leaves it open whether only certain ingredients
of mental content, e.g., the satisfaction conditions of predicates, require external
determination. But the Argument from Unmotivation extends to logico-semantic
properties, and to the overall point of ascribing truth conditions. It is for this sort of
reason that systematic external relations between thoughts and the world, e.g.,
reliability, and not merely, say, causal relations between predicates and natural
kinds, have been thought by some8 to be essential to explaining mental content.
3 The Argument from That-clauses

I have recently discussed this argument elsewhere.9 There are two important issues
about the relation of that-clauses and representational content: whether sameness in a
pair of de dicto that-clause ascriptions (or in their general terms) is sufcient for
sameness in (the relevant aspect of) the representational content of the beliefs they
ascribe; and whether difference in that-clause ascription is sufcient for difference in
representational content. The Argument from That-clauses assumes the latter. But
the former also can play a role in anti-internalist argument. Tyler Burge, for example,
has pointed out (1979) that two thinkers who differ crucially in how they conceive of
arthritis may nevertheless both believe, say, that one of them has arthritis in his
ankles. The that-clause is true de dicto and univocally of both, and so it would seem
the two thoughts have the same representational content. That common property is
explained, not by their thoughts sharing an individualistic property, e.g., a concep-
tual role, but by something socialthe social meaning of a jointly assented to
sentence in a shared language.10 And so it would seem that representational content
is determined in part by something outside the head.
content, constituted not by truth conditions but by functional or conceptual roles. (See Loar (1982)
[Chapter 4 in this book], and Block (1986).) Now, of course, such a functionalist conception of content
does not as such purport to get around the Argument from Unmotivation. For the latter means by content
something representational or intentional, that is, something about how a thought conceives things as
being, and on the face of it that must involve something that determines what would make the thought
correct, truth conditions or something like truth conditions. The two-factor reply to the externalist
apparently uses content in a special sense, one perhaps that can be motivated by its role in the explanation
of behavior via the structures of common-sense propositional attitude psychology, but a special sense
nevertheless. Such a two-factor theory is of course incompatible with certain strong anti-internalist claims,
as noted above, but it does not bear on what is of concern here, that is, whether representational content is
in the head.
8
E.g., Field (1978), Fodor (1987), Stampe (1977), Dretske (1981), Stalnaker (1987), and Loar (1981).
Evidently Davidsons principle of charity (1974) might be viewed in the same light.
9
Loar (1988b) [Chapter 8 in this volume].
10
Suppose one of them, who in fact misconceives what hives means, had belonged to a community
whose meaning for hives was what he (in the actual world) misconceives it as meaning. Then it would
have been true not that he believes that he has hives in his ankles but that he believes that he has (say)
chives in his ankles. This shows that in the case described, the sameness of that-clauses does depend on an
external factor, and not on some further common internal factor.
Burge is right, I believe, about the social sensitivity of that-clauses, and that is very
important. But in fact, our common-sense conception of how thinkers represent
things to themselves does not correspond exactly to the that-clauses correctly
ascribed to them, and this makes a difference to the anti-internalist argument just
sketched. Suppose Peter, having an odd education, comes to think cat ambiguously
names two kinds of animalsone whose members he sees daily about the place, and
another that he learns of by description (but that in fact are just cats). Suppose he
believes that alls have tails and also that all . . . s have tails. For each that-clause,
there is an equally good case for inserting cat with its ordinary meaning, so that one
univocal de dicto ascription is twice true.11 But it would seem that those two beliefs
have in some sense different representational contents. For do they not represent the
world to Peter as containing two distinct facts? Then sameness in de dicto ascription
does not imply sameness in mental content, if the latter means how the thinker
conceives things.12 Thus the foregoing point, that something external to the mind
(something social) in part determines a common that-clause for two beliefs, does not
imply that something external to the mind in part contributes to their representa-
tional content in the relevant sense. (The that-clause, we might say, captures the
common social content of the two thoughts.)
Now it may be said that in this example the two beliefs are distinct merely in their
functional roles, and not in any kind of representational content.13 But this is not an
intuitive diagnosis. It is natural to say that Peter represents the world as containing
two distinct facts. And in our understanding the world as Peter conceives things
Peters notional world, to use Dennetts term14we populate it with a pair of facts
(which we think the actual world does not distinguish). That does not seem like
ascribing to Peter merely distinct functional states. And so, in the arthritis example,
the fact that a single de dicto that-clause describes the two thoughts leaves intact the
intuitive view that they represent the world differently, that in an important sense
they have distinct representational contents.
Of course this does not directly counter the more basic Twin-Earth/Burgean point,
which depends on the assumption that (truth-conditional) differences in de dicto
ascription imply differences in mental representational content. But the case of Peter
and the cats should at least soften up anti-internalist intuition here as well. It seems a
plausible conjecture that, when we come to understand the individuation involved in
Peters two ways of conceiving things, it will turn out to be something an Earthling
11
Some readers may think the case demands our inventing different terms to capture Peters concep-
tions of cats. But this misses the point. To be consistent that would require inventing a vast language of
neologisms to capture ordinary beliefs, for ordinary peoples conceptions often vary as much as Peters two
cat conceptions. That we feel no constraint to do so (unless in the grips of a philosophical theory) appears
to show that that-clauses are not designed to capture personal conceptions; they are designed rather to capture
the common social information shared by beliefs that may involve different personal conceptions.
12
This is obvious if one has in mind demonstratives in that-clauses. But my point here concerns general
terms.
13 14
Suggested by Jerry Fodor in a recent letter. See footnote 17.
and a Twin-Earther can share. For our understanding Peters distinct conceptions
does not appear to depend on any factor in which Earth and Twin Earth or Burges
two linguistic communities differ.
There are more direct replies to the assumption that a difference in de dicto that-clauses
means a difference in representational content. Consider this thought-experiment. Sup-
pose we nd a diary whose provenance is uncertain (Earth?, Twin Earth?, triplet-Earth?),
and which says Weve run out of gin, so Albert has defected. Suppose we accept
that the diary either is Earthly or hails from a world/planet resembling Earth as Twin
Earth does. It is fairly clear that we understand the psychological explanation here,
despite our ignorance of the diarys origin and therefore of which that-clauses are
appropriate; we understand how the author of the diary, and Albert, conceive things.
Now like the former, this case, too, can support a weak and a strong conclusion
about that-clauses. The weaker, which does not invoke content, is this: differences in
that-clauses do not imply differences in thought-type, as individuated in common-
sense psychological explanation. This means we might diagnose that commonality
which permits us to interpret the diary as (say) merely functional; and that would still
tell against attempts (not unknown) to use Burgean and Twin-Earth cases against
internalism of the weaker (non-representational) sort described at the outset. But the
stronger conclusion, involving content, seems to me amply warranted: in our ordin-
ary understanding of the diary we in fact do not lter out a purely functional level of
individuation, even though that is theoretically possible. We somehow understand
the world as it is according to the diary in spite of ignorance of that-clauses.15 To sum
up, then, the unavailability of a common that-clause does not imply that represen-
tational content is not shared by a pair of Earthling and Twin-Earthling thoughts.
Earlier we saw a step from internal properties determine representational content
to internal properties determine truth conditions. Can we be true to the spirit of this
if internal properties do not, in these cases, determine that-clauses? It seems difcult
to deny that the truth conditions of a thought are captured by some that-clause.
There are three possible moves here.
(1) While an Earthly thought that gin is delicious and a Twin-Earthly thought that
twin-gin is delicious differ in truth conditions thus specied, the two thoughts also
share a distinct common that-clause, which captures their common representational
content and hence truth conditions in a distinct sense. There are social truth
conditions, relative to the language the thinker defers to, and personal truth condi-
tions, determined by his personal conceptions; two that-clauses capture different
aspects of each thoughts properties. For example, a that-clause of the latter type
might be metalinguistic: Albert thinks we have run out of the stuff they call gin
15
Evidently it wont do to say that we have a set of that-clauses, among which we switch back and forth
in understanding the diary. We need have in our repertory no set of terms denoting the myriad possible
ginlike liquids.
around here. But let us not linger over this; it is doubtful that that-clauses capturing
personal truth conditions are generally available (see Loar (1988a)). Let me grant that
that-clauses are generally sensitive to social factors, including social relations to
natural and other kinds.
(2) It may be granted that internal factors do not determine truth conditions of the
kind captured by that-clauses (roughly sets of possible worlds), but claimed that
nevertheless they determine truth conditions of a different kind, which would equally
vindicate internal intentionality. Compare those thoughts that Manfred and Engel-
bert express by saying I like my name. In a sense they have the same truth
conditions, i.e., the same conditions of truth in abstraction from context. Similarly,
it might be said, Georges belief that gin is delicious and twin-Georges belief that
twin-gin is delicious share truth conditions in that sense: their beliefs share a
context-indeterminate (as we might say) truth condition, i.e., a function that maps
the contexts they respectively inhabit onto the sets of possible world that constitute the
context-determinate truth conditions of their beliefs.16
But this fails to capture what it should. The relevant function from contexts to sets
of possible worlds would be the same, presumably, for the belief that gin is delicious
and the belief that vodka is delicious; for the idea appears to be that function is
determined in part by a subfunction that maps a context onto whatever liquid is at
the origin of a certain kind of causal path leading to the thought. But intuitively
George and twin-George share a content that they may not share with Dmitri and
twin-Dmitri, who think vodka and twin-vodka are delicious; the former pair of
thoughts represent the world differently from the latter. This is not to deny that
context-indeterminate content must be part of the story of internal content, of what
George and twin-George share, but it is not the whole story.
It is not much help to say that something about conceptual or functional role has
to be added, to distinguish the gin/twin-gin beliefs from the vodka/twin-vodka
beliefs. For I am supposing the difference to be a difference in representational
content, and facts about conceptual role are prima facie not facts about representa-
tional content. In any event, such an addition would make the current discussion
otiose: if conceptual role can amount to internal representational content, functions
from contexts to determinate truth conditions would be extrinsic to internal content.
(3) Representational content means something like truth conditions, but it does
not have to mean truth conditions in the ofcial sense. Let us grant that the latter, like
the that-clauses which capture them, are sensitive to social factors, including social
16
My terminology. The idea is Kaplanesque, of course. But it differs from David Kaplans actual theory
by assigning variable functions to proper and common names, rather than the constant functions he would
assign, given his quite different objective, which is to characterize a language relative to its actual speakers.
Proposals of this sort have been made by Stephen White (1982), Loar (1982) [Chapter 4 in this volume],
and by Jerry Fodor (1987).
relations to natural and other kinds. Still, when we read the diary we learn, in some
sense, how its author conceives things as being, and therefore something of how the
world must be for it to realize his conception. We understand, lets say, the realiza-
tion conditions of his statement; these are distinct from truth conditions in the
ofcial sense, i.e., no class of that-clauses is designed to capture just such realization
conditions. So a given thought may have two quite different representational dimen-
sions: the set of worlds in which it is truei.e., relative to its social meaning; and the
set of worlds that are as the thinker personally conceives things as being. The latter
does not purport to be captured by that-clauses, and any claim that this dimension of
a thought is determined by internal factors is thus immune to the Argument from
That-clauses. (Actually, we should understand realization conditions as context-
indeterminate, as abstracting from the actual references of indexicals, etc., and so
not exactly as sets of possible worlds. The second of the three moves under discussion
then, while not sufcient, is part of the picturecontext-indeterminate realization
conditions.)17
Apart from its vagueness, there are two obvious problems with this suggestion. The
rst is how such realization conditions can be central to understanding and to
psychological explanation given that there is no standard way of specifying them.
Evidently no construction other than that-clauses has that function. The second is
simply the Argument from Unmotivation, nothing in which depends on that-clauses;
it denies that representational content of any outward-directed kind can be determined
by internal factors.18 As it happens, the answer I shall give to the more fundamental
problem of unmotivation also yields a natural answer to the problem of specication.
4 The Non-external Intentionality

of My Own Thoughts
A natural view of ones own thoughts stands in sharp contrast with the conclusion of
the externalist argument. From a pre-critical perspective, knowledge of the references
of my own thoughts is privileged in a certain way, and that perspective involves no
17
This notion of context-indeterminate realization conditions is introduced in Loar (1988b) [Chapter 8
in this volume]. As several people have pointed out to me, it is similar to Dennetts conception of notional
worlds (Dennett (1982)). Because (it seems) Dennett introduces the latter in an instrumentalist vein in
accounting for behavior in a way that takes environmental factors into account, he does not have to face the
question whether internal factors determine representational content; his project is different from mine.
Still, it would not be inappropriate to take my question here to be: can internal factors determine notional
worlds?
18
The question of realization conditions (individualist truth conditions that are not in general
captured by that-clauses) is distinct from that of internal content. One can at least notionally discern a
view (individualist-externalist) on which such realization conditions have an important role to play in
understanding others, but are explained in part by external relations, the common elements of the
reliability of Earthlings and Twin-Earthers. Naturally the question how these realization conditions are
specied remains. For a discussion of such causalized realization conditions, see Loar (1988a).
apparent conceptions of external reference relations. This is a basic conception of

mental representation, and, understood correctly, it makes sense of representational
content is in the head in a way compatible with the Argument from Unmotivation.
But it will take a bit of explaining.
I am now attending to my thought that Freud lived in Vienna. I register what the
thought is aboutFreud, Vienna, the ones inhabiting the other; I note the thoughts
references and truth conditions. I may be wrong about the non-semantic question
whether Freud actually exists (timeless). But it is difcult to see how I might be wrong
in my purely semantic judgment that this thought is about Freud if Freud exists.
What reference relation am I conceiving of? Is it disquotational? Intuitively not;
on the face of it, the about is not relativized to a language. The natural, pre-critical
view is that a Chinese speaker can have thoughts about Freud in the sense I know my
thought to be about Freud. Do I conceive this reference relation as externally
determined? Suppose, for the moment, that my thought about Freud is not under a
denite description (e.g., the famous psychologist called Sigmund Freud in my
social group). There seem to be two candidates for an externally determined refer-
ence relationa reductionist causal-historical relation and an unreduced social-
semantic relation.
It is implausible that in judging that my thought is about Freud I judge that it has a
given causal-historical relation to Freuda relation no one has yet managed to
characterize. Likewise, we should note, for the reductionist idea that truth conditions
consist in reliability or information-theoretic properties: my knowledge that my
thought is true iff Freud lived in Vienna does not depend on knowledge that it has
a certain role in some system of reliable indicators. On the natural view, indeed,
judging that my own thoughts are reliable indicatorsi.e., endorsing those
thoughts on reectionpresupposes knowing what states of affairs they purport to
be about.
Perhaps I conceive the relation as an unreduced social-semantic relation, thus:
a) my thought refers to whomever Sigmund Freud refers to in this social group, and
b) Sigmund Freud refers to Freud in this social group. The question is what
judgment b) amounts to. Can I have privileged knowledge of b), as I apparently
have of the references of my own thoughts? If I specify the social group I belong to
independently of myself, I have no privileged knowledge that one of its terms refers to
a given object. My knowledge of semantic facts about a nearby social group must
apparently be as unprivileged in principle as my knowledge of the references of a
name in Italian. But how then can adding that I belong to the group (defer to it) give
me privileged knowledge that its references are such and such? There is an answer
and it is paradoxical. I can have privileged knowledge that their use of Freud refers
to Freud only because, given my membership in the group (my deference), my use is
the same as theirs and my use refers to Freud.
This, it will justly seem, puts the cart before the horse. But it is unavoidable if one
has privileged knowledge of the reference of ones thought and if that is knowledge of
a social-semantic relation. The natural view is that one has some kind of privileged
semantic self-knowledge (try to doubt that a relevant thought of yours is about Freud
if Freud exists). But then you cannot hold, without a bizarre consequence, that what
you know thereby is a social fact.
A reply may go like this. My conception of Freud is descriptive I conceive of
(say): the famous psychologist referred to as Sigmund Freud in my social group.
Then my knowledge that my thought refers to Freud is simply knowledge that my
thought refers to the famous psychologist referred to as Sigmund Freud in this social
group, whoever he is. This involves knowing that my thought refers to the relatum of
a social-semantic relation; but there is no mystery in the seemingly privileged status
of my knowledge. It is no more mysterious than the similar status of my knowledge
that my thought that my great-great-great-grandfather must have lived in Ohio is
about my great-great-great-grandfather, whoever he was. Now this reply merely
postpones recognizing the basic point (i.e., that one can have knowledge of the
references of ones own thoughts that is not ostensibly knowledge of externally
determined relations), postpones it in the following sense.19
The description-theoretic interpretation would exclude Freud from (as we might
call them) the basic references of my thought, references that are not via descriptions.
But you cant do that indenitely. Certain demonstratives, predicates, and logical
connectives must be basic, in the sense of not implicating denite descriptions; they
correspond to Russells constituents (although Russell notoriously had an overly
narrow view of those constituents). The point about Freud, on a non-descriptive
interpretation, is transferable to predicates, demonstratives, logical connectives: one
apparently has privileged knowledge of the basic references of ones thoughts (subject
when appropriate to the existence proviso), and ones reexive semantic conceptions
here are intuitively neither of external contingent relations (reductionist causal-
historical or unreduced social-semantic), nor of merely intralinguistic disquota-
tional properties.
Consider perceptual and memory demonstratives, thoughts of the form that
object is such and such that are backed by a perception or a memory. I judge that
the animal is an elephant. I then reexively note that my thought refers to that animal
if that animal exists. Is there an external causal relation R such that a judgment that
my thought is R to that animal (if it exists) is identical with my judgment that my
thought is about that animal (if it exists)? A familiar point about perception is that
the modern causal view differs radically from the Greek view. In registering that my
thought is about that animal (if it exists) must I be taking a theoretical position on
that differenceeven before I read philosophy, when I was asked And what are you
daydreaming about now? The natural view again is that this reexive semantic
judgment has a status different from that of any judgment about a causal relation.
19
Actually the point can be reformulated even in connection with thoughts involving such
descriptions.
Indeed, it seems I can raise the causal question about my thought by rst identifying
its reference: This thought is about that animal (if it exists). Is that animal then a
causal factor in the thoughts formation? However inevitable we nd the afrmative,
it seems natural to say that the referential judgment is from a rst-person perspective
independent of thoughts about causal reference relations. Moreover, the natural and
pre-critical view is unquestionably that the reference relation of which I have such
privileged knowledge in my own case can obtain also between the thoughts of
Chinese speakers and that animal. (See the appendix at the end of this chapter for
points about demonstratives and disquotationality.)
It should be noted that the semantic propositions I judge true reexively here are
purely extensional. I judge that my thought refers to that animal if that animal exists.
And if I also judge that that animal is the worlds smartest elephant, then I should
conclude that my original thought was, as it happens, about the worlds smartest
elephant.
Like some proper names, some predicates in thought may implicate metalinguistic
descriptions. Thus, a thought that parsnips are nourishing may be a thought that the
vegetables that are called parsnips in my social group are nourishing. But this
evidently is not true of all predicates in my thoughts, e.g., of the basic predicates
that metalinguistic description is itself couched in. Suppose my thought that most
cats have tails does not rest on a metalinguistic description of cats. Then my judging
that thought to be about cats does not, for reasons by now familiar, involve concep-
tions of either externally determined or disquotational properties. And let us not
forget logical connectives. My knowledge that a thought of mine is (semantically)
disjunctive, say, is not knowledge of a social-semantic relation, still less of the
thoughts role in some system of externally reliable indicators. Nor intuitively is it
knowledge of an intralinguistic property, for the usual reason: the natural view is
that it may share that property with thoughts of speakers of other languages.
My thoughts appear self-interpreting. And what I judge about my own thoughts is
intuitively conceivable of the thoughts of others; I want to think of you as having self-
interpreting thoughts as well. It seems I then have a conception of (what can only be
called) representational properties of others thoughts that is not of externally
determined properties. If our intuitive conceptions here are not anti-naturalist, we
have a conict with the Argument from Unmotivation, that no internal physical-
functional properties of the brain can amount to representational properties. I, for
one, nd that Argument persuasive. If I wish to accommodate both intuitive
perspectivesthe subjective/projective perspective and the naturalistic third-person
perspectivemust I lapse into dualism?
5 Displaying Thoughts to Oneself

When one attends reexively to ones thought that Freud lived in Vienna, it is
difcult not to see it as pointing outwards, towards Freud and Vienna and the
ones inhabiting the other. My thoughts appear transparent, as if I look through them
towards their objects.
Attending to a thought introspectively is quite different from attending to a
sentence one has written down. When you display a sentence you do not thereby
use it. But when you display a thought in order to attend to it, you do use it, for you
think it, not necessarily by asserting it but at least by entertaining it. I now decide to
think about a certain thought, that Freud lived in Vienna. In the normal way of doing
this, I not only think about that thought, but I also think the thought and thereby
think about Freud, Vienna, and so on. This, I want to say, is the key to our privileged
knowledge of the non-externally determined references of our own thoughts.
Let us distinguish another phenomenon. In a third-person report of an utterance,
we may refer to the things, properties, etc., that it is about. So if you say that Rudolf
mentioned that Freud lived in Vienna you speak not only about Rudolf s utterance
but also about Freud et al. But my point about the reective awareness of a thought is
not merely that it involves a conception of the thought that refers to the very things
the thought refers to. That is neither here nor there. The point rather is that, in
summoning up a thought of ones own in order to refer to it, one produces the
thought, thinks it, and thereby thinks about its objects.
When I reect on a thought, from my perspective there is bound to be an intimate
connection between that mental state and certain external objects and states of
affairs, the ones, as I see it, about which I am thinking. My thought that Freud
lived in Vienna appears transparent or intrinsically directed at objects because from
my subjective perspective nothing would count as bringing it onto the scene without
also bringing on Freud and Vienna. Thus what makes certain thoughts about Freud,
from my perspective, is simply that they have this in common: if the thought heaves
into sight, so inevitably does Freud.
6 An Objective, Non-intentional
View of the Phenomenon
I now make two claims. One is that internal, or better perhaps, non-external
intentionality, which is not as such discernible from an objective point of view, arises
from a subjective/projective perspective alone. The other is that those features of our
thoughts that give rise to the subjective conception of internal intentionality can (in
principle) be completely captured from an objective perspective that registers only
non-intentional features.
Consider phenomenal states. The usual argument against their being physical is
that, because complete knowledge of the physical-functional facts about another
creature is not thereby knowledge of its phenomenal states, phenomenal facts are
not physical-functional facts. But the argument is awed. It assumes that if phenom-
enal and physical-functional conceptions and concepts are cognitively distinct and
connectable only a posteriori, then phenomenal facts and properties are not physical-
functional. But there are models of the relation between concepts and the properties
they stand for on which it makes perfectly good sense for there to be a pair of
cognitively independent concepts that stand for exactly the same property (and not
because one of them connotes a higher-order reference-xer). One of these concepts
may, in a certain important sense, involve a subjective perspectivesay, because it
involves the ability to recognize certain states in the having of them. So, that a
phenomenal concept involves such a subjective perspective, and that a physical-
functional concept involves an objective perspective (one not presupposing the
ability to be in the states conceived), is a matter of the independence of those
concepts psychological or cognitive roles, and does not imply a distinctness in
what they are about.
Something like this, I wish to say, is involved in the distinctness of our objective
physical-functional conceptsthe ones appealed to in the Argument from
Unmotivationand our display-conceptions of our own thoughts.
I nd intentionality in my own thoughts. They are in a sense self-interpreting:
normal reection on ones own current thoughts yields what appears from this
perspective as a relation between thoughts and objects, properties, and so on. But
in noting these relations I adopt a perspective distinct from the objective perspec-
tive of the Argument from Unmotivation. Is there a parallel with phenomenal
conceptionsthat is, might our subjective conceptions of internal intentionality
discriminate purely physical-functional properties, even though those subjective
conceptions are cognitively inequivalent to any objective, physical-functional, or
external-relational, conceptions?
If the phenomenon to be accounted for has the form this thought is about that
external object, etc., the parallel would not be very close: we would have to say that
there are objective relations between thoughts and objects of which we have sub-
jective conceptions. But I do not think that is the right parallel. For a person may
think his thought is about an external object, that it has a certain object as a basic
reference (that man in the bushes), when no suitable object exists. That seems as
much an instance of the phenomenon as when a suitable object does exist.
This suggests that what needs explaining is not, as it were, intentionality itself so
much as the discerning of intentionality in ones own states. Or rather this: ones
thoughts are such that if one were to reect on them in the familiar way, they would
(as it were) reveal themselves as about this and that. It is true that one does regard
ones thoughts as having intentional objects apart from ones reexive attention. But,
I want to suggest, the intentionality we nd there is in fact a product of the reexive
perspective, or rather of its possibility. The solubility of sugar in water derives from
its own properties only given those of water. Similarly, the intentionality of object-
level thoughts derives from their own properties only given their systematic relations
to reexive thoughts and their properties. Those (objectively non-intentional) prop-
erties of object-level thoughts that contribute to explaining why upon reection they
reveal themselves as about this and about that can be counted as the basis of
subjective intentionality. One cannot identify subjective intentionality with such
properties (of object-level thoughts) any more than one can with full felicity identify
sugars solubility in water with its internal structure. Subjective intentionality is the
disposition of thoughts to reveal themselves as intentional upon reection.
So if the parallel with phenomenal properties is to be pursued, we ought to be
seeking, as the objective fact corresponding to subjectively-appearing intentionality,
some systematic relation between object-level thoughts and the corresponding
reexive thoughts. It is not a philosophical enterprise to say what in detail these
objective properties of the mind are, any more than it is a philosophical enterprise to
say what phenomenal states are in objective detail. We may nevertheless hope to say
something about this structure that at least makes the appearance of subjective
intentionality unsurprising. The basic idea is that, when I display a thought in
order to think about it, my conception incorporates certain aspects of the displayed
thought; it is a display-conception of that thought. From an objective or external
point of view, it is this incorporation, as it were of subject matter, that is
explanatorynot of course of intentionality, but of the reexive discerning of
intentionality.
In what follows I point towards an explanation. It must be emphasized that the
project is not to explain that-clause ascriptions in internalist terms, which I take to be
impossible. I shall use the term conceptual role, and a certain notation, impression-
istically and inexactly. Anyone who is radically externalist, in denying not only that
representational content is in the head but also that there is a non-external level of
psychological explanation and individuation, will not accept my principal presup-
position, so let me make it explicit. It is that we can take for granted an understanding
of internal states of conceiving things and structured relations among such conceivings.
If one is not tempted by an anti-naturalist or eliminativist view, one will then take
those subjectively noted features and structures to be identical with objective,
physical-functional, internal properties of the mind. And it is those internal struc-
tures that the following is an attempt to indicate from an objective perspective.
The notion of conceptual role is notoriously underspecied and controversial. But
let me use it in the following imprecise way. Consider the thought of mine that
I would express saying Freud lived in Vienna. Say that, relative to me, it has the
*Freud lived in Vienna* conceptual role, and that it is a *Freud lived in Vienna*-
thought.20 This thought ascription attempts to point towards an objective feature of
that thought. It is not a presupposition of the notation that the thought is in English.
And of course the ascription does not itself refer to Freud. Consider all other
thoughts I would express by sentences that use Freud non-quotationally, or that
in the same intuitive sense involve the relevant conception of Freud. Assuming that
20
If this puts one in mind of certain usages of Wilfred Sellars, that would not be inappropriate.
that commonality has an objective basis, let us call them all *Freud*-thoughts.
To classify a thought of mine as a *Freud*-thought then is meant to point, again
imprecisely, towards its internal objective propertiese.g., its conceptual roleand
not to presuppose anything about its referential properties.
Let us now try to redescribe the phenomenon that from a subjective perspective
one sees as internal intentionality. A reective thought that involves a display
conception of a *Freud*-thought is itself a *Freud*-thought; a thought involving a
display conception of my thought that Freud lived in Vienna is itself a *Freud lived
in Vienna*-thought. We may put it rebarbatively thus: a **Freud*-thought*-thought
is a *Freud*-thought. This is by contrast with thoughts involving quotation concep-
tions of the sentence Freud lived in Vienna; they are not thereby *Freud lived in
Vienna*-thoughts. Display conceptions of a thought, to return to the earlier way of
speaking, use that thought, while display conceptions of a sentence do not use the
thought it expresses.
In sketching the phenomenology of internal intentionality I said that the trans-
parency of my thought that Freud lived in Vienna arises because from my subjective
perspective nothing counts as displaying it that does not also involve bringing onto
the subjective scene Freud and Vienna themselves. So I am now suggesting that those
phenomenological appearances have an explanation that somehow realizes this
structure: **Freud*thought*-thoughts are *Freud*-thoughts. From a subjective per-
spective, that amounts to this: one must see attending to certain thoughts (which in
fact are *Freud*-thoughts) as attending also to Freud. From an objective perspective,
it means that reexive thoughts about *Freud*-thoughts, i.e., those involving display
conceptions of the latter, have that property, that conceptual role, etc., which is
shared by all and only *Freud*-thoughts. (Let me emphasize that this notation points
to conceptual roles only relative to a given person.)
So I am advancing a conjecture: the subjective discernment of internal intention-
ality is a psycho-functional phenomenon that is objectively describable in non-
intentional terms; and whatever the correct objective account is, the structure just
sketched will be an essential part of it. What then is the answer to the initial question,
whether representational content is in the head? It could well seem that I am
answering a different one. I turn later to this, but the application of these ideas to
our understanding other persons, to which I now turn, ought to make it at least
plausible that I am not changing the subject.
7 Understanding Others
We ascribe social content to the thoughts of others by using that-clauses. But we
conceive of the personal content of their thoughts by conceiving how they conceive
things. Of course we lack forms of words conventionally designed for reporting
personal conceivings, but that is unimportant. It would hardly be economical to
have quasi-that-clauses for capturing the full range of idiosyncratic personal
conceptions.21 And in any event conceiving how another person conceives things is
projective and not descriptive. I understand your thought by thinking (i.e., entertaining)
a thought so as to be in a position to judge that you have a thought like that one.
I maneuver myself experimentally into a certain conceptual situation, I try out a
thought, I project it onto you, and if in fact you have a thought like that one,
I understand how you think.
Projective understanding of course involves conceptions of the representational
content of others thoughts, and is not just a matter of conceiving, say, their
conceptual or functional roles. But this internal intentionality that we ascribe to
others is, as it were, what we nd in their thoughts when we try them out for
ourselves, or try out thoughts like them in certain respects. Just as I nd in my
own thoughts intentionality that ostensibly does not involve external causal relations,
etc., so in my natural pre-critical frame of mind I will ascribe that sort of intention-
ality to your thoughts.
This is not to say that whenever I ascribe a thought I go through such a piece of
projection. I am often content with a somewhat vague conception of how another
conceives things, content even with the that-clauses ascribable to him. They are
enough to go on for many purposes of psychological explanation and description, for
they are normally closely if not exactly in line with how others personally conceive
things.22 But although I do not always stop for the details, having a grip in general on
how others think depends on being able at least roughly to conceive, when the need
arises, how they personally conceive things. It is our potential for this that allows us
implicitly to understand others, even if we only sometimes exercise that capacity
in detail.
To see how the ability to entertain projectible thoughts is crucial to psychological
explanation, consider this commonplace of recent theorizing: structural relations
among the contents of thoughts are essential to psychological explanation, the usual
candidates involving, in at least minimal forms, consistency, inference, practical
reasoning, and so on. It has sometimes been observed (correctly) that these structures
are sensitive, not (only) to the ofcial truth conditions of thoughts, but to their
modes of presentation. (The latter usually have been proposed in connection with
demonstratives and proper names, but ought to cover all elements of thought.) Now,
it seems to me, conceiving of the modes of presentation of a thought is just
conceiving of how that thought personally conceives things; and this, I say, is a
matter of projecting thoughts of our own. (Fregean senses, in their role as cognitive
aspects of thoughts, can then be identied with what one projects.) But then our
21
Cf. Loar (1988a).
22
It is pretty clear that which that-clauses are true of x is not a function solely of the social meaning of
the words x would utter; they also depend in part on the degree to which his personal conceptions
approximate to the social meanings of the words he utters. There is a certain amount of indeterminacy
here, but I believe that informative things can be said about the relation between xs personal conceptions
and the that-clauses that are ascribable of x, given their social contents in the group to whose usage x defers.
intuitive conceptions of psychological structures such as consistency, modus ponens

inference, practical syllogism, and so on, are relative to our projective conceptions of
others thoughts.
Of course, I am often not in a position to conjure up a conception of your thoughts
detailed enough to include all psychologically active aspects. Suppose I take you to
have a pair of contradictory thoughts (which you manage by compartmentalizing),
involving a memory demonstrative, say, and am not able to conceive how you
conceive the remembered object. In that case I existentially quantify over modes of
presentation of a certain kind, those resembling thoughts I can conceive except for
details of the demonstrative conception. So I project a generically conceived class of
contradictory thought-pairsI project the kind rather than the specics. This of
course is meant to apply to psychological explanation in general: even when I do not
know the details, understanding another in terms of such and such psychological
structures still involves projection, in varying degrees generic.
The details that I do know of your conceptions may be signicantly unlike mine,
e.g., if you think you have arthritis in your thigh. Here I can experiment, consider-
ing how it would be to conceive things thus, and that is no problem. When
I experimentally entertain such a thought and display it for inspection, once again
I nd in it a certain intentionality; the thought is about my thigh and about such and
such physical condition. And again I project that thought onto you, with the
intentional dimension it reveals from this perspective.
Suppose you take Shangri-La to be a real place, a land in the Himalayas where
people stay young. How shall I understand you? I do not think Shangri-La exists. But,
still, I can conceive of your thought as having an intentional object if not a real one, to
put it Brentano-wise. This does not seem to t my account of how internal inten-
tionality arises. For that, I must be able to try out the thought that people stay young
in Shangri-La, displaying it and judging it reectively to be about Shangri-La. But
I cannot do that if I do not think there is a Shangri-La for it to be about. The solution
is not hard to nd. Adopting your point of view, I make-believe Shangri-La exists, and
in that state I try out the thought, judging it in a make-believe way to be about that
place. This gives me a way of understanding you. The intentionality I make-believe is
to be found in my thought I conceive of you as nding in yours less ctively. What
about indexicals? I understand Norman Mailers thought that he is tough. I try out
that thought in my own case and I succeed in entertaining it. But in projecting its
intentionality onto Norman Mailer I cannot project the reference of my thought.
Is there a substitute? Yes; I can make a perspective switch, and nd reference from a
new point of view. This is not make-believe in the earlier sense, because in Norman
Mailers thoughts I is not a vacuous term. But it is another sort of make-believe.
I make-believe I occupy a different perspective, entertain from it the thought that
I am tough, and think reectively under a display conception of that thought (now
this is from the make-believe Norman Mailer perspective) it refers to me. It seems
a psychological fact that I can make-believe Shangri-La exists, and take up
imaginatively anothers rst-person perspective, and in both cases think thoughts in

which I nd intentional features that I cannot nd in my non-ctive thoughts.
This brings us to Twin-Earthers, whose thoughts and language are like ours, except
that their word water denotes a substance that is in appearance like water but is not
water; their water-thoughts are objectively about that substance. Can they never-
theless be said to resemble us in how they represent things to themselves? The idea
that we understand them perspectivally is available here as well. I put myself in the
position of a Twin-Earther, and think what it would be like for him to think
reectively about the thought he would express as theres water in that lake.
I judge, from what I contrive of his perspective, this thought refers to water and to
that lake; I project that thought onto him. But I do not project the objective reference
of my non-ctive water, any more than I project the reference of my non-ctive I.
Once again, it seems a simple psychological fact that I am able to do this. There is
nothing magical in any of these psychological capacities.
We come to brains in vats. Attempts to show they are impossible have involved
externalist argument: no coherent account can be given of the contents of their
thoughts, because no externalist account can be given. Now I do not wish to be glib
here; brains in vats are puzzling. It will nevertheless be evident from the foregoing
that I do not nd the usual externalist arguments compelling. If I can take seriously
the idea that the contents of my thoughts are internal propertiesobjectively matters
of non-intentional factors while phenomenologically intentionalwhy not the same
for brains in vats? Does something bar me from conceiving them as resembling me in
respect of this state and that, and then conceiving the intentionality of their states
projectively, by adopting a posture of make-believe and a transferred perspective?
I do not think I have seen arguments that show I cannot do that. But if there are
objective internal bases for my own states and for their subjectively discerned
intentionality, and if there is no good philosophical reason not to identify these
subjective phenomena with their objective internal bases, then is there not reason to
say that brains in vats may objectively resemble us in the very respect that externa-
lizers of the mind have denied?
In outlining what the account accommodates, let us not forget Fregean senses.
As it happens they are conserved not only in their cognitive role, but also in their
reference-determining function. The internalist Fregean conception is that, while the
internally determined contents of a thought do not imply the existence of the
thoughts referents, they determine something such that, if certain objects exist,
they are thereby the referents of that thought, with no further externally determined
relation required to hold between that thought and those objects. The present
account allows retaining something of this. If the referent of a thought is judged
from the subjective perspective, then, as I earlier pointed out, the natural form of
what one knows in abstraction from external relations is: this thought is about Freud
given that Freud exists. But then all that is required, from this perspective, for an
object x to be the referent of that thought is that x = Freud; and that does not involve
any further causal, etc., relation of x to the thought. So how a thought conceives
things then involves modes of presentation as reference-determinersfrom the
subjective perspective, of course.
The distinction between sense and reference is preserved. I can judge that this
thought is about Hesperus and that that thought is about Phosphorus not knowing
they are the same; and when I learn that they are the same I then judge that those two
thoughts, though distinct in their senses, are alike in their reference. That is,
I continue after discovering the identity to regard the thoughts as differing in their
reference-determining features (without having to take a description-theoretic view),
because from a higher epistemic perspective I conceive the abstract possibility that
my judgment of identity is mistaken.
8 The Objective Basis of Projective Intentionality

To say that internal intentionality is an artifact of the subjective perspective on ones
own thoughts, projected onto others, could suggest an anti-realist or instrumentalist
notion of content in the head, as if all there is to it is projecting. But in fact
projections can be objectively correct or incorrect, or at least I shall try to make a
case for that. Given physicalism, the correctness of the projective ascription of a
thought to x must depend on xs internal physical-functional properties; and two
major questions arise about that. The rst is whether each projected thought may
have a physical-functional basis B in the projector y such that, if y projects that
thought onto x and x is in state B, then x has that thought. Of course, our conceptions
of our own and others thoughts do not involve physical-functional individuating
conceptions. When I conceive of a thought of mine as about Freuds living in Vienna,
I do not thereby conceive of it as having specic internal physical-functional prop-
erties. How then might such properties be necessary and sufcient for the correctness
of projections onto x? If one wants to regard projective internal intentionality as the
key to understanding others, might one after all have to be either instrumentalist or
anti-naturalist?
The second question is what ordinary behavioral evidence there might be for the
correctness of a projective ascription of a thought to x, if internal physical-functional
properties of x are necessary and sufcient for the correctness of that projection.
It would be awkward if the ordinary non-scientic data did not sufce for assessing
the correctness of projective ascriptions, given, that is, the claim that there is an
objective basis for that correctness.
A similar question to the rst has arisen about phenomenal qualities: given that a
judgment that x is in a given phenomenal state is never implied by any judgment
about physical-functional properties (conceived thus), how can phenomenal states be
physical states? The answer is as follows. Granted that there is no cognitive implica-
tion of that form; nevertheless, it is left open that when I conceive of a phenomenal
state of mineit is like thatthe property I thereby refer to or demonstrate is for all
I know some physical-functional property. The standard anti-physicalist argument

mistakenly denies this on the grounds that such phenomenal quality conceptions are
not cognitively implied byare connected only a posteriori withspecic physical-
functional characterizations. But certain concepts (e.g., recognitional concepts) can
be cognitively independent of physical-functional concepts but still refer to purely
physical-functional properties (without connoting higher-order non-physical aspects
of those properties).23
When one ascribes phenomenal states to another person, one judges: x is in a state
of this remembered, imagined, or currently experienced kind. Such a third-person
judgment is objectively correct provided that x is in whatever physical-functional
state ones phenomenal conception discriminates in the home case. (This follows if
you accept that the only available reference for state of this kind is some physical-
functional state, and accept that the introspective referential impulse should be taken
at face value.) Thus subjectively based concepts can have objective, physical-
functional conditions of correctness. When we project phenomenal qualities we are
not merely projecting.
The same can be true of the subjective discerning of intentionality. When
I produce a display conception of a thought that Freud lived in Vienna, no
physical-functional characterization of internal states cognitively implies having
a thought like this one. I have conjectured that the subjective discernment of that
thoughts references is explainable in terms of non-intentional properties of that
thought, as constrained by the *Freud*-role structures. These properties would
account for why I do not view it as (and now I speak from the subjective perspective)
a thought that Adler, say, or Anna Freuds father, lived in Vienna. By the analogy
with phenomenal quality concepts, there is no philosophical barrier to this, i.e., to its
being the case that when I individuate thoughts in my subjective reections I thereby
discriminate, although this is not how I conceive it, what are in fact physical-
functional properties of those thoughts. But in that case, when I project a thought
by judging: x has a thought of this type (i.e., as conceived in subjective intentional
terms), my projection is correct only if x is in the relevant internal physical-
functional state, whatever it may be.
Now although I shall say something about these physical-functional properties, it
must be emphasized that what has been said so far requires no philosophical account
of them. There is no a priori reason why a theoretical psychology, or even a more
physiological level of theory, should not uncover a basis for such thought individu-
ation, and therefore an objective basis for projective correctness, even if philosophical
23
This argument has been developed in Loar (1990) [Chapter 10 in this volume]. We should note that
phenomenal conceptions of structural relations among ones phenomenal states may impose considerable
constraints on which physical-functional properties they can be identical with, relations, e.g., of determin-
able to determinate, of mutual exclusion, of sameness of sensory modality, and so on. Thus, in a sense,
ones system of phenomenal concepts gives some access to abstract properties of the system of physical
states, if there is one, that constitutes their references.
reection itself yielded little. But in fact, something can be said about these proper-
ties, something that helps to answer the question about evidence for projective
ascriptions.
One understands others by approximation, but in what respects? Conceptual role
is relevant. When I fashion a thought and project it, I do not, as we have observed,
thereby conceive of you as having a thought with a certain abstractly spelled
out conceptual role. Nevertheless, my subjective judgments of differences in my
thoughtsincluding those fashioned experimentally in order to project themin
fact are sensitive to differences in their conceptual roles. And this has implications for
the correctness of my projections.
I do not mean by conceptual role the fully holistic notion. If I come to think Freud
never smoked cigars, this will not affect my ability to retain the mentioned thought
that Freud lived in Vienna. But if I come to think that the Freudian corpus was the
work of a committee, none of them named Sigmund Freud, that no one of that name
ever saw patients, and that certain familiar photographs are of an unemployed actor
named Sigmund Freud, then I will not be able to have my old thought, except of
course by ctively adopting the perspective of my former beliefs. To put the point in
the third person, if I learn that x thinks Freud never smoked cigars, that will not affect
my ability to project onto x my thought that Freud lived in Vienna. But if I learn that
x subscribes to the committee theory, etc., then I will not be able projectively to
ascribe to x that thought.
If I come to think chicken can be roasted by putting it out in the sun in Death
Valley, this need not affect my ability to retain my current thought that, say, poached
chicken is tenderer than roasted chicken. But if I come to think that you can roast
something by putting it in a freezer (knowing well enough the usual properties of
freezers), then I will have lost my old concept of roasting, and ceased to have that old
thought about chicken, even though I now think poached chicken is tenderer than
roasted chicken. Such judgments about individuating thoughts are not, I take it,
unintuitive.24
But they are of course radically at variance with contemporary skepticism about
the distinction between constitutive and non-constitutive conceptual connections.
Let me make a few brief points about this issue. 1) There need be no sharp distinction
between the two sorts of conceptual connection, just clear cases of each. 2) The
distinction is not as such a modal distinction; it is rather about psychological
24
Many who have not accepted Tyler Burges anti-individualist argument have of course shared this
intuition. But it must be emphasized that the basic point here does not conict with Burges point about
that-clauses, although it does conict with the anti-individualist consequences that are supposed to ow
from that point. It is compatible with what I say here (although it may be doubted on other grounds) that
you can continue to say Jones thinks that poached chicken is tenderer than roasted chicken, even if Jones
thinks you can roast things by putting them in the freezer as well as in the oven. Given the points made
earlier, it does not follow from the continued appropriateness of this that-clause that Jones continues to
have a thought with the same psychological content, i.e., as we projectively conceive it.
individuation. 3) The usual Quinean arguments are to the effect that the distinction is
not determined by purely behavioral criteria. But what philosophical argument
shows that such a distinction cannot be a component of an explanatory psychological
theory? There is no prima facie incoherence in the idea that thinking requires a
framework of conceptual connections, which provide stability over belief change and
in fashioning novel hypotheses, and which in part explain our reexive individuation
of our own thoughts. 4) It is respectable to take seriously, as a clue to explanatory
models in psychological theory, intuitive subjective conceptions of the structural
features of ones own thoughts. If a distinction between constitutive and non-
constitutive conceptual roles shows up there, as it surely does, then it is reasonable
to suppose that a correct scientic account of thoughts might well preserve that
structure. It should go without saying that there are no guarantees here.
So while when we reect on the intentional properties of a thought we do not
explicitly conceive it as having this or that constitutive conceptual role, still we have
an implicit grasp of aspects of such a role. Whether we can also articulate this is not
essential; what matters is the implicit grasp. Thus it is not implied that one can
abstract ones whole conceptual framework, with the conceptual roles of all ones
thoughts laid out for inspection. The grasp of these connections is piecemeal, resting
on exercising concepts rather than describing them purely functionally. But if such
connections are constitutive of psychological content, then any physical realization of
the total set of thoughts of a person at a time must satisfy corresponding functional
conditions, those that could in principle be abstracted.
To return to the second question, having evidence for the objective basis of
projective ascriptions to x depends then in part on having evidence that x is suitably
conceptually organized. My ordinary understanding of x proceeds on the assumption
that x resembles me conceptually (i.e., constitutively) except for the deviations
I explicitly note. There seems to be no special epistemological problem about having
behavioral evidence that the conceptual roles of xs thoughts are like mine in general
and unlike mine in given ways. (Once again, I do not abstract conceptual roles but
exercise concepts in implicitly grasping their interconnections.) But if one were then
to take a more theoretical point of view, does this ordinary evidence not become
evidence that x is functionally organized in accordance with those conceptual roles?25
Thoughts, conceived subjectively, are not individuated solely in terms of concep-
tual relations, for modes of presentation can involve experiential statesconsider a
thought involving a perceptual-demonstrative: that dog is a husky. If there is a further
evidential problem here, it concerns projecting a phenomenal (imaginative, etc.)
25
This is of course controversial, and a complex issue. Anti-mentalists may accept the importance of
conceptual roles to content, while denying that they imply anything about the functional organization of
internal physical states. And some philosophers who are perfectly happy about the relevance of whats
inside the head to psychological explanation are skeptical about there being any level of actual functional
organization that corresponds sufciently to common-sense psychology. But I have not been persuaded by
either.
statethat is, being justied in supposing that x is in a physical state, or kind thereof,
that in oneself is identical with a phenomenal state or kind appropriate for serving as
a perceptual-demonstrative mode of presentation. One hopes there is no problem
about my having ordinary evidence for that, for if there were, our ordinary assump-
tions about others experiences in general would not be well backed.26 Thus, the idea
that projections can be objectively correct does not apparently raise new epistemo-
logical problems.
But suppose the best epistemological view is that the ordinary common-sense
evidence (for the relevant functional and phenomenal conditions) is inadequate.
Still, there is no denying our inclination to project thoughts onto others. And it
could turn out, after all, that the best theoretical evidence is that physically we are in
fact relevantly functionally, etc., organized. Would that not then objectively vindicate
our ordinary inclination to project thoughts and experiences onto others? But if so, we
may as well regard our projective ascriptions as now having objective conditions of
correctness, i.e., regardless of whether our projective inclinations outrun our current
epistemological justication for taking such objective conditions to be satised.
9 Has the Argument Been to the Point?

The initial question was whether content is in the headi.e., whether internal
properties determine how a person represents the world to himself, not in the
sense of the means by which this is done, but in the sense of how the world is
according to that persons conceptions. My answer has been: in a sense, yes, despite
the Argument from Unmotivation. That argument is blocked, or rather reduced in
scope, as follows. Grant that a complete internal physical-functional description of x
never cognitively implies that a given assignment of possible worlds, or realization
conditions, etc., to xs thoughts is correct, that is, grant that associations of such
internal properties and intentional properties must always be a posteriori. It does not
follow that internal physical-functional properties do not determine the correctness
of projective ascriptions of intentionality. The Argument from Unmotivation does
show something, that no objectively (non-projectively) intentional properties are
determined by the internal physical-functional states of x. For to assert that there
are properties that are internal, objectively intentional, and compatible with physic-
alism, presumably means just that their intentionality is expressible incognitively
equivalent to something inphysicalist terms. How else should one construe objectively
intentional?
But the question remains whether the projective account really supports some-
thing that externalists have been denying. In assessing this, it helps to consider the
26
This is not the usual Problem of Other Minds. If there is a problem here, it is whether ordinary
evidence is sufcient for the proposition that others resemble us in the internal physical states we
discriminate introspectively.
philosophical concerns that give a point to the question whether representational

content is in the head.
First, something like the following anti-internalist argument has been implicit in
much of the discussion. 1) The psychologically active property of a thought (i.e.,
whatever individuates it in psychological explanation) is a matter of how the thought
represents things. 2) But no internal properties determine how a thought represents
things. Hence 3) psychological explanation appeals to more than internal properties
of thoughts. If the anti-internalist nds this argument congenial (as many I think do),
that helps to interpret at least one anti-internalist claim, viz., premise 2. Such an anti-
internalists idea of how thoughts represent thingsthe contents of thoughtsis
the idea of those representational properties of thoughts that are essential to psycho-
logical explanation. But my argument has in effect been that representational prop-
erties that are essential to psychological explanation need be representational or
intentional only from a subjective/projective perspective. Hence, my argument is
right to the point of at least some anti-internalist arguments, for it shows how to deny
2) as it occurs in the context of such an argument. (If it is stipulated that content or
how a thought represents things means an objectively intentional property, then of
course I deny premise 1).)
Secondly, there are familiar anti-Cartesian concerns, which extend beyond psycho-
logical explanation into epistemology (anti-skepticism) and metaphysics (anti-
realism). It is thought that externalizing the mind will eliminate that divide between
our conceptions of the external world and the external world itself which for many
philosophers has implied skepticism. Now I rather think that subjective/projective
internal intentionality ought to be regarded, by anyone susceptible to the usual
argument, as implying skepticism for just the usual reasons. (Naturally this requires
discussion.) But then, if internal representational content is whatever property of the
mind supposedly raises the classical problems, my account is again to the point.
I have been defending what such externalizers have been denying.
Thirdly, some anti-internalists have simply found internalist or Brentanian intro-
spective conceptions of the representational properties of the mind confused and
easy to parody (cf. Platonic graspings). Doubtless some mentalists have been
confused. I have tried to provide a way of understanding the basic Brentanian
intuition that makes it acceptableby contrast of course with the anti-physicalist
or, a distinct matter, anti-functionalist conclusions sometimes drawn from that
intuition.
I envisage dissatisfaction with this account of internal intentionality on many
grounds. Here is one that might not be wholly explicit but operative nevertheless.
One way to think of theoretical psychology is as an extension of common-sense
psychology, one that adds complex novel structures that dwarf the original, but an
extension nonetheless. An implication for some has been that theoretical psychology
must then appeal to content. If that-clauses or other explicit content ascriptions play
a key role in common-sense explanation, then theoretical explanations will also use
that-clauses, etc. But on my account explicit ascriptions of contentthat-clauses or

whateverare not central to common-sense psychological explanation, however
practically indispensable they are in communicating evidence (social content) for
what is central. But that, it may seem, counterintuitively rules out the idea that
theoretical psychology is an extension of common sense; for surely projection plays
no role in the former.
The last observation is correct. But the moral to be drawn is that theoretical
psychology need not appeal to representational content (ineliminably) in order to
be an extension of common-sense psychology in the appropriate sense. The con-
tinuity should be envisaged as a continuity of functional structures rather than of
relations of content. (Of course these remarks are bald assertion, but they outline a
position that, I believe, can be defended in detail.)
10 Objective Reference and Subjective Intentionality

Let me now point to the vast question of connections between subjective intention-
ality and objective semantic properties, i.e., causal reference relations and socially
determined truth conditions. My thoughts refer to gin, to me, to this hand, to Freud.
These apparently are objective external facts and not artifacts of introspection. What
is the relation between them and subjective intentionality? It would be highly
unsatisfactory if there were simply two topics, two independent senses of this
thought is about x. And it could seem that that is inevitable. Consider twin-George,
on whose thoughts I may adopt two perspectives. On the one hand, in conceiving
how he conceives things, I project a thought, which I would express by saying there is
water in that lake. In judging twin-Georges thought to be like that, I conceive it from
a subjective perspective, neutral as to whether his neighborhood contains real water
or good old XYZ; and I make a referential judgment from that perspectivethis
thought refers to that stuffto which causal relations are irrelevant. On the other
hand, I can take the external point of view and say that twin-Georges thought refers
to twin-water, invoking at least implicitly a causal, or social-causal, reference relation.
Are these referential judgments simply about distinct topics?
Unlikely, given that in the home case one must judge the reference relations to
coincide. With the deliverances of the subjective perspective in mind, nothing
prevents me from shifting to an external objective perspective on my own thoughts.
And I notice a striking pattern. Those subjectively apprehended relations have salient
correlations with certain (vaguely conceived) objective external relations. Not just
one relation. The relation between memory-demonstrative thoughts and their refer-
ences differs from the one between perceptual-demonstrative thoughts and their
references. And still other externally determined relations are relevant to proper
names and to kind terms. The pattern is this: for each conceptually unied class of
terms (thought-elements) in ones repertory, there is an externally determined
relation that holds between any term of that class and whatever object, property,
set of objects, etc., is from ones subjective perspective the reference of that term.
So, for perceptual demonstratives, if I judge from the subjective perspective that a
thought of mine is about that lake, that star, or that person, I note from the objective
perspective that my thought stands to that object in a distinctive externally deter-
mined relation, even if I cannot specify it precisely.
It may be wondered how such coincidences can unify subjective and objective
reference. I will now sketch a strategy that takes the subjective as primary: an
objective relation or property is semantic by virtue of its direct or indirect connec-
tions with subjective intentionality. It would follow, of course, that a relations
(propertys) being semantic is not an objective feature of it, even though such a
relation (property) is itself objective and connects us in real enough ways with the
rest of the world.
A question that naturally arises in connection with (e.g.) causal theories of
reference is: what makes a given causal relation semantic? What makes certain
causal relations amount to objective reference? One answer might be that that is
just what reference is, but I cannot imagine this satisfying anyone. Another is that
they amount to reference because they have a certain intimate connection with
objective truth conditions. This could be merely another form of the rst answer:
when asked what makes certain external correlations, which incorporate causal
reference relations, amount to truth conditions the answer might be that that is just
what truth conditions are. Again, hardly satisfying. A more sophisticated idea is
that both reference and truth are concepts in some large theory relating thoughts to
the world, a theory that explains objective individual and social properties distinct-
ively enough that a special category resultsthe semantic. I have worries about the
idea that true and refers are implicitly dened by any purely objective theory, but
this is not the time to vent them. Let me suggest a different approach. (Perhaps
what I am about to say can be seen as suggesting a theory-theory strongly anchored
to a subjective element.)
Our primitive notion of aboutness is subjective, and this is the foundation of the
semantic. But, as noted, we inevitably also take an objective view of our own
thoughts; and the coincidence between the subjective reference of a demonstrative
thought, and the object related to it by a salient causal relation, appears not accidental.
But why? There is no question of an objective explanation of why the relations
coincide, given that one of them arises only from the subjective perspective.
Perhaps despite appearances, I shall not be suggesting a genetic account of how a
persons semantic conceptions arise; rather, the idea will be, ones thoughts on these
matters naturally cohere in certain ways, and that coherence explains what is
distinctive of semantical relations and properties.
Let us start with objective semantic properties and relations, and work backwards.
These, it seems to me, are two-tiered: there are social-semantic and individual
semantic relations and properties. The scope of the social-semantic is broad, not
just a matter of the use of language in communication. The objectively ascribable
truth conditions of thoughts (if there are such properties) are socially determined.27
And the same goes for objective referential properties of those predicates and proper
names in a persons thoughts which are taken over in the normal way from the public
language; if such referential properties involve causal relations to particular objects
and to objective resemblances, then those relations are social-causal.28
It would of course be absurd to suggest that such social-causal reference relations
count as semantic for one because they coincide with subjective intentionality in
how they relate ones own thoughts proper names and predicates to objects and
objective resemblances. Whatever objective relation exists between Sigmund Freud
and Sigmund Freud, one will have to acknowledge that it is semantical by virtue of
its social features, and not because it coincides with ones own subjectively inten-
tional name relation. A related point is this. It does not seem possible to specify
causal relations between my use of Aristotle and Aristotle without using social-
semantic terms. So one cannot say that there is an otherwise non-semantic causal
relation here that counts for me as semantic because of its coincidence with
subjective intentionality.
Still, suppose that social-semantic relations themselves depend, somehow, on
externally determined individual, non-socially mediated, reference relations, say
between perceptual or memory demonstratives and objects. Suppose also that this
dependence is plausibly regarded as what makes those social relations semantic.
Suppose nally that those objective individual reference relations can be regarded as,
while objectively non-semantic, semantic by virtue of their coincidence with subject-
ive intentionality. Then the topic of the semantic is after all unied.
Social-semantic facts may supervene29 on non-social-semantic facts, where the
latter include facts about objective individual demonstrative reference (who points to
what), the totality of individual conceptual roles (who thinks how about what),
including individuals conceptions of the division of semantic labor (who defers to
whom about what). Of course you cannot take a particular social-semantic fact, say
that gold denotes a certain stuff, and specify a totality of non-social-semantic facts
on which that fact supervenes. You quickly run into other social-semantic facts that
seem indispensable, and hence the supervenience must be global. But the idea is that
the fact that a society points at and conceives the world thus and so supervenes on
facts about how individuals in complex combinations refer demonstratively in their
27
What that-clauses are true of a person who is normally deferential to the meanings of her linguistic
group depends not just on how she conceives things (although of course this constrains things in basic
ways) but also on the social meaning of the sentences she would utter in expressing her thoughts. What
about the totally undeferential person (mythical though she be)? I doubt that objective truth conditions
could be assigned.
28
Can one invent predicates and link them personally to objective resemblances without the aid of
social-causal reference relations involved in other predicates in ones thoughts?
29
The notion of supervenience perhaps ought always to be seen as shorthand for something else, for
without backing it smacks of metaphysical magic. It is nevertheless a useful notion in gesturing towards
dependency relations.
thoughts to particular objects and personally conceive the world, including their
personal conceptions of their own communicative intentions and of the social-
semantic facts. Is it implied that the extremely complex structure of this super-
venience can be spelled out a priori? I do not think so; it can be left to empirical
investigation just how social meanings supervene on individualist facts. Perhaps all
that is a priori is that there exists a supervenience, in the weak sense that a given
totality of non-social-semantic facts in all worlds determines the same totality of
social-semantic facts.
Now we have the question of externally determined individual demonstrative
reference, of what makes causal relations of perception and memory semantic.
Unlike the case of socially mediated proper names and predicates, here there is no
problem with the idea that what makes those causal relations semantic is their
coincidence with subjective intentionality. The causal perceptual relation between
thoughts T of mine and objects x is referential because it holds between T and x when
and only when T involves a conceptual element of the relevant kind (a perceptual
demonstrative, as that kind is reexively identied) and is, as I see it, thereby about x.
But why should the causal relation inherit semantic status from the coincidence?
Think about thoughts as wholes. When I reect on a thought I conceive of a (possibly
non-actual) state of affairs such that, as I see it, if it obtained the world would be as
that thought personally conceives it to be. So, from the subjective perspective, there is
an intuitive mapping of thoughts onto (sometimes non-actual) states of affairs, one
that is determined, as I see it, by intrinsic properties of my thoughts; call the mapping M.
(I am avoiding truth conditions because of its social-semantic connotations.)
Now I notice something extraordinary about M; as I see them, my thoughts past
and present are highly reliable relative to M, high enough to require explanation.
Then it becomes clear to me that those causal relations play a substantive role in the
explanation of my thoughts M-reliability; that is, as I subjectively conceive things,
they transmit information about the objects my thoughts are demonstratively about.
In this way the coincidence of those external relations with subjective reference
makes the external relations reference by as it were non-trivial and intimate
association.
(A subjective account of semantic properties has nothing to do with epistemological
foundationalism. We are not inside battling our way out; rather we have a vast array of
beliefs, indeed knowledge, about the external world, and a question has arisen about
the relation of subjective intentionality to the external world as we know it.)
Now of course social-semantic relations between my thoughts and objects also
transmit information: the social name-relation between my thoughts and objects
plays a key role in the M-reliability of my thoughts about named objects. You might
say that those relations are doubly semantic; they are socially semantic, and they also,
as one sees it subjectively, transmit information about what certain of my thoughts
are about. The basic point here is that, regardless of what holds of social-semantic
relations, certain individualist causal relations, involving perception and memory, do
have semantic status of the latter sort and thereby serve as part of the individualist
basis for semantic relations and properties of the former sort.
Given the special status of perceptual and memory relations in my own case, they
naturally take on a semantic aspect in my ascription of them to others. And similarly,
to return to the matter at last, for those relations that obtain between my thoughts
and gin, water, arthritis, etc. They give me an external semantic way of thinking about
twin-Georges thoughts, one that complements my projective way of thinking, by the
chain of connections I have just sketched. There are not really two different subjects
after all: subjective intentional relations inevitably get linked with external relations
in ways that make the latter derivatively semantic.
If there existed a satisfying account of objective conditions that yield reference and
truth conditions, i.e., that made certain external relations semantic, then the present
account might appear otiose. But in fact it does not appear to me that, say, purely
information-theoretic and causal-theoretic accounts are likely to be successful.30
Two-factor theories have been a party to the widespread scorn for the natural
intuition of internal intentionality; and they have thereby made it appear that the
relevant internal and external features of thoughtstheir conceptual roles and their
extrinsic semantic propertiesare merely adventitiously linked. The present
account, I believe, changes this situation for anyone with two-factor inclinations,
and introduces unity into the semantic.
Of course, the upshot is that the world contains no relations that are semantic just
by virtue of their objective roles. But this in no way threatens the fact that there are
plenty of nice objective relations between thoughts and things, perhaps even all that
are needed for a robust non-skeptical realism.
Appendix: Disquotationality and

Subjective Intentionality
Can subjective intentionality be explained in terms of disquotationality? Here are two
reasons for thinking not.
1) Disquotationality is in part a formal property, and judgments of subjective
intentionality are not about formally characterized states; hence, they are not dis-
quotational. A relation R is disquotational only if we give a special semantic status to
sentences of the form w is R to w. In any instance of this, the reference on the left
is to something under its formal properties. But in central cases of reective aware-
ness of the reference of ones thoughts, ones conceptions of those thoughts are not
formal or even non-intentional. I can say that I am now thinking about a certain
30
For powerful arguments to this effect, see Schiffer (1987).
treethat my thought refers to that treewithout having to hold that that thought
involves certain wordsthat tree.
2) There is no general disquotational notion of demonstrative reference, while there

is an intuitive notion of subjective intentionality for demonstratives.
For non-demonstrative termsgeneral terms and proper nameswe can dene a
general disquotational reference relation, using substitutional quantication.
For all t, t refers to t(s) [if it/they exist[s]].
But how is one to do this for demonstratives? Consider
For all F, this F refers to this F [if it exists].
This makes no sense, any more than For all F, this F is green makes sense.
There are, in fact, non-general disquotational contexts for demonstratives which
are guaranteed to be correct, for example,
The second occurrence of refers to , given thatexist(s).
The second occurrence of this hand in this sentence refers to this hand, given that
this hand exists seems safely true. But apparently no generalization of it makes
sensethe already mentioned lack of coherence is present here with a vengeance:
For all F, the second occurrence of this F in this sentence refers to this F if it exists.
Given that the general notion of subjective intentionality applies as evenly to
demonstrative aspects of thoughts as to those involving proper names, subjective
intentionality is not explainable in terms of disquotationality.
10
Phenomenal States
On a natural view of ourselves, we introspectively discriminate our own experiences

and thereby form conceptions of their qualities, both salient and subtle. These
discriminations are of various degrees of generality, from small differences in tactual
and color experience to broad differences of sensory modality, for example, those
among smell, hearing, and pain. What we apparently discern are ways experiences
differ and resemble each other with respect to what it is like to have them. Following
common usage, I will call these experiential resemblances phenomenal qualities, and
the conceptions we have of them, phenomenal concepts. Phenomenal concepts are
formed from ones own case. They are type-demonstratives that derive their refer-
ence from a rst-person perspective: that type of sensation, that feature of visual
experience. And so third-person ascriptions of phenomenal qualities are projective
ascriptions of what one has grasped in ones own case: she has an experience of
that type.
Phenomenal quality can have a different sense, namely, how the object of a
perceptual experience appears. In this sense, a phenomenal quality is ascribed to an
object and not directly to an experience. Some have argued that all we discern
phenomenologically are phenomenal qualities in this sense; they deny that experi-
ences themselves have introspectible qualities that are not ascribed primarily to their
objects (Harman (1990); Block (1990)). I will not pursue the issue here, but will
assume a certain view of it. For the present objective is to engage antiphysicalist
arguments and entrenched intuitions to the effect that conscious mental qualities
cannot be identical with ordinary physical properties, or at least that it is problematic
to suppose that they are so. Antiphysicalists typically suppose that such mental
properties are not relationalthat is, that they present themselves as not intrinsically
involving relations to things outside the mind. They may allow that, say, visual
experiences are in some sense intrinsically representational. That is hard to deny
because, as regards ordinary visual experiences, we cannot apparently conceive them
phenomenally in a way that abstracts from their purporting to represent things in a
certain way. The antiphysicalist intuition is compatible with visual experiences
having (some sort of) internally determined intentional structure, so that it is an
introspectible and nonrelational feature of a visual experience that it represents
things visually as being thus and so. Antiphysicalists suppose that we have concep-
tions of how visual experiences differ and resemble each other with respect to what it
PHENOMENAL STATES
is like to have those experiences. These conceptions then are of qualities of experi-
ences, whatever allowances one may also make for the apparent qualities of the
intrinsic objects of those experiences. I will assume that the antiphysicalists phe-
nomenological and internalist intuitions are correct. The idea is to engage them over
the central point, that is, whether those aspects of the mental that we both count as
phenomenologically compelling raise substantive difculties for the thesis that phe-
nomenal qualities (thus understood) are physical properties of the brain that lie
within the scope of current science.
We have to distinguish between concepts and properties, and this chapter turns on
that distinction. Antiphysicalist arguments and intuitions take off from a sound
intuition about concepts. Phenomenal concepts are conceptually irreducible in this
sense: they neither a priori imply, nor are implied by, physical-functional concepts.
Although that is denied by analytical functionalists (Levin (1983), (1986)), many
other physicalists, including me, nd it intuitively appealing. The antiphysicalist
takes this conceptual intuition a good deal further, to the conclusion that phenom-
enal qualities are themselves irreducible, are not physical-functional properties, at
least not of the ordinary sort. The upshot is a range of antireductionist views: that
consciousness and phenomenal qualities are unreal because irreducible;1 that they
are irreducibly nonphysical-functional facts;2 that they are forever mysterious, or
pose an intellectual problem different from other empirical problems, or require new
conceptions of the physical.3
It is my view that we can have it both ways. We may take the phenomenological
intuition at face value, accepting introspective concepts and their conceptual irredu-
cibility, and at the same time take phenomenal qualities to be identical with physical-
functional properties of the sort envisaged by contemporary brain science. As I see it,
there is no persuasive philosophically articulated argument to the contrary.
This is not to deny the power of raw metaphysical intuition. Thoughtful people
compare phenomenal qualities and kinds of physical-functional property, say the
activation of neural assemblies. It appears to them to be an evident and unmediated
truth, independent of further premises, that phenomenal qualities cannot be identical
with properties of those types or perhaps of any physical-functional type. This
intuition is so compelling that it is tempting to regard antiphysicalist arguments as
rationalizations of an intuition whose independent force masks their tendentious-
ness. It is the point of this chapter to consider the arguments. But I will also present a
positive account of the relation between phenomenal concepts and physical proper-
ties that may provide some relief, or at least some distance, from the illusory
metaphysical intuition.
In recent years the central problem with physicalism has been thought by many
to be the explanatory gap. This is the idea that we cannot explain, in terms of
1 2
Cf. Rey (1995) and Dennett (1991). Jackson (1982), (1986).
3
Nagel (1974), (1986); McGinn (1993).
PHENOMENAL STATES
physical-functional properties, what makes a certain experience feel like this, in the
way we can explain what makes a certain substance a liquid, say. It is concluded that
physicalism is defective in some respect, that there cannot be a (proper) reduction of
the mental to the physical. Before we consider this explanatory gap, we must rst
examine, in some detail, a more basic antiphysicalist line of reasoning that goes back
to Leibniz and beyond, a leading version of which is now called the knowledge
argument. Answering this argument will generate a framework in which to address
antiphysicalist concerns in general.
1 The Knowledge Argument and its Semantic Premise

The knowledge argument is straightforward on the face of it. Consider any phenomenal
quality and any physical property however complex. We can know that a person has
the physical property without knowing that she experiences the phenomenal quality.
And no amount of a priori reasoning or construction can bridge this conceptual gap.
That is the intuitive premise. The conclusion is drawn that the phenomenal quality
cannot be identical with the physical property. The argument is equivalent to this:
since physical and phenomenal conceptions can be connected only a posteriori,
physical properties must be distinct from phenomenal properties.
The best-known and liveliest version of the knowledge argument is Frank Jacksons,
which features the physiologically omniscient Mary, who has never seen color and so
does not know what it is like for us to see red, despite her knowing all the physical-
functional facts about us.4 She later sees colors, and thus learns what it has been like
all along for us to see red. She learns a new fact about us. Jackson concludes that this
fact is not among the physical facts, since Mary already knew them. It is not difcult
to see that this argument depends on a more or less technical premise.
In my view, the physicalist should accept Jacksons intuitive description of Mary:
she fails to know that we have certain color experiences even though she knows all
relevant physical facts about us. And when she acquires color experience, she does
learn something new about usif you like, learns a new fact or truth. But this is to be
granted, of course, only on an opaque reading of Mary learns that we have such and
such color experiences, and on corresponding readings of learns a new fact or truth
about us. For as regards the transparent versions of those ascriptions of what Mary
did not know and then learned, they would beg the question, amounting to this: as
for the property of having such and such color experiences, Mary did not know, but
then learned, of that property that we have it. Physicalists reject this, for according to
us those experiential properties are physical properties, and Mary already knew of
all our physical properties that we have themunder their physical descriptions.
4
Jackson (1982, 1986).
PHENOMENAL STATES
What she lacked and then acquired, rather, was knowledge of certain such properties
couched in experiential terms.
Drawing metaphysical conclusions from opaque contexts is risky. And in fact
inferences of Jacksons form, without additional premises, are open to straightfor-
ward counterexamples of a familiar sort. Let me describe two cases.
(1) Max learns that the bottle before him contains CH3CH2OH. But he does not
know that the bottle contains alcohol. This holds on an opaque reading: he
would not assert that theres stuff called alcohol in the bottle, or that the bottle
contains the intoxicating component of beer and wine. Let sheltered Max even
lack the ordinary concept alcohol. After he acquires that ordinary concept, he
learns something newthat the bottle contained alcohol. If the knowledge
argument has a generally valid form, we could then infer from Maxs epistemic
situation that alcohol is not identical with CH3CH2OH. Evidently this does
not follow.
(2) Margot learns about the element Au and reads that people decorate them-
selves with alloys of Au. But she has never seen gold and cannot visually
identify it: she lacks an adequate visual conception. She later is shown some
gold and forms a visual conception of it, that stuff, and she acquires a new
piece of informationindividuated opaquelyto the effect that those previ-
ously read about embellishments are made of that stuff. Again, if the know-
ledge argument were unrestrictedly valid, it would follow that that stuff is not
identical with Au. This case differs from the case of Max by involving not a
descriptive mode of presentation but (as we might say) a perceptual mode of
presentation.
It is not difcult to nd a difference between both these cases and the case of Mary.
Max lacks knowledge of the bottles contents under a contingent description of it
ingredient of wine and beer that makes you intoxicated. What Margot lacks is a
certain visual conception of Au, which is to say gold. This typically would not be a
descriptive conception; it would not self-consciously take the form the stuff that
occasions this type of visual experience. Still on the face of it such a concept
implicates a visual-experience type. For it picks out the kind it picks out by virtue
of that kinds occasioning experiences of that type. And that is a crucial contingency
in how the concept that Margot lacks is related to its reference. I hope I will be
understood, then, if I say that the visual take on Au that Margot lacks would have
conceived Au under a contingent mode of presentation.
This brings us back to Mary, whose acquired conception of what it is like to see red
does not conceive it under a contingent mode of presentation. She is not conceiving
of a property that presents itself contingently thus: it is like such and such to
experience P. Being experienced like that is essential to the property Mary conceives.
She conceives it directly. When Mary later acquires new information about us
(construed opaquely), the novelty of this information cannot be explainedas in
PHENOMENAL STATES
the case of Margotas her acquiring a new contingent mode of presentation of

something she has otherwise known of all along. She has a direct grasp of the
property involved in the new information; she conceives of it somehow, but not
under a contingent mode of presentation. Proponents of the knowledge argument
will say that is why it is valid on an opaque reading: there is no contingency in Marys
conception of the new phenomenal information that explains it as a novel take on old
facts. She learns new facts simpliciter and not new conceptions of old facts.
Notice how close this comes to Saul Kripkes well-known antiphysicalist argument
(1972a). Kripke assumes that a phenomenal concept such as pain cannot be a priori
linked with a physical concept such as that of the stimulation of C-bers. The case of
Mary is a vivid way of making the same point. Kripke points out that property
identities can be true even if not a priori, for example, heat = such and such
molecular property. It seems fair to represent the next step in his argument as
follows. Heat has a contingent higher-order mode of presentation that connotes
the property feeling like this. That is what accounts for the a posteriori status of the
identity. But, as Kripke points out, this cannot be how pain works: the phenomenal
concept pain does not pick out its referent via a contingent mode of presentation; it
conceives pain directly and essentially. Kripke concludes that pain is not identical
with a physical property.
The two arguments then turn on the same implicit assumption. The only way to
account for the a posteriori status of a true property identity is this: one of the terms
expresses a contingent mode of presentation. This ought to be given a place of
prominence.
(Semantic premise) A statement of property identity that links conceptually
independent concepts is true only if at least one concept picks out the property
it refers to by connoting a contingent property of that property.
The knowledge argument and Kripkes argument then depend on two assump-
tions: the conceptual independence of phenomenal concepts and physical-functional
concepts, which I accept, and the semantic premise, which I deny.
The antiphysicalist intuition that links concept-individuation and property-
individuation (more closely than is in my view correct) is perhaps this. Phenomenal
concepts and theoretical expressions of physical properties both conceive their
references essentially. But if two concepts conceive a given property essentially,
neither mediated by contingent modes of presentation, one ought to be able to see
a prioriat least after optimal reectionthat they pick out the same property. Such
concepts connections cannot be a posteriori; that they pick out the same property
would have to be transparent.
But as against this, if a phenomenal concept can pick out a physical property
directly or essentially, not via a contingent mode of presentation, and yet be concep-
tually independent of all physical-functional concepts, so that Marys history is
coherent, then Jacksons and Kripkes arguments are ineffectual. We could have
PHENOMENAL STATES
two conceptually independent conceptions of a property, neither of which connote

contingent modes of presentation, such that substituting one for the other in an
opaquely interpreted epistemic context does not preserve truth. Even granting that
our conception of phenomenal qualities is direct, physicalism would not entail that
knowing the physical-functional facts implies knowing, on an opaque construal, the
phenomenal facts; and so the failure of this implication would be quite compatible
with physicalism. The next few sections give an account of phenomenal concepts and
properties that would justify this claim.
2 Recognitional Concepts
Phenomenal concepts belong to a wide class of concepts that I will call recognitional
concepts. They have the form x is one of that kind; they are type-demonstratives.
These type-demonstratives are grounded in dispositions to classify, by way of
perceptual discriminations, certain objects, events, situations. Suppose you go into
the California desert and spot a succulent never seen before. You become adept at
recognizing instances, and gain a recognitional command of their kind, without a
name for it; you are disposed to identify positive and negative instances and thereby
pick out a kind. These dispositions are typically linked with capacities to form
images, whose conceptual role seems to be to focus thoughts about an identiable
kind in the absence of currently perceived instances. An image is presumably of a
given kind by virtue of both past recognitions and current dispositions.
Recognitional concepts are generally formed against a further conceptual back-
ground. In identifying a thing as of a recognized kind, we almost always presuppose a
more general type to which the kind belongs: four-legged animal, plant, physical
thing, perceptible event. A recognitional concept will then have the form physical
thing of that (perceived) kind or internal state of that kind, and so forth.5
Here are some basic features of recognitional concepts that it will help to have in
mind in connection with the account of phenomenal concepts that follows.
1. You can understand porcelain from a technical description and only later
learn visually, tactually, and aurally to recognize instances. By contrast, in the
phenomenon I mean, the concept is recognitional at its core; the original
concept is recognitional.
2. A recognitional concept need involve no reference to a past instance, or have
the form is of the same type as that (remembered) one. You can forget
particular instances and still judge another one of those.
5
How such background concepts themselves arise is not my topic; but we might think of them variously
as deriving from more general recognitional capacities, or as functions of complex inferential roles, or as
socially deferential; or they may be components of innate structures. Background concepts are not always
presupposed. Someone may be extremely good at telling stars from other objects (e.g., lightning bugs,
airplanes, comets, planets) without having any real idea of what they are.
PHENOMENAL STATES
3. Recognitional abilities depend on no consciously accessible analysis into com-

ponent features; they can be irreducibly gestalt.
4. Recognitional concepts are perspectival. Suppose you see certain creatures up
close and form a recognitional conceptthose creatures1; and suppose you see
others at a distance, not being able to tell that they are of the same kind (even
when they are), and form another recognitional conceptthose creatures2.
These concepts will be a priori independent. Now the respect in which they
differ is perspectival, in some intuitive sense. A recognitional concept is in part
individuated by its constitutive perspective. Here is the important point: a
recognitional concept can be ascribed outside its constitutive perspective;
that thing (seen at distance) is one of those creatures1 (seen up close) makes
perfectly good sense. This plays a key role below in the account of third-person
ascriptions of phenomenal concepts.
(This casual invoking of reference-determining dispositions will be a red ag for
many who are aware of the vexing foundations of the theory of reference. Problems
about referential scrutability, rule-following, naturalizing intentionalityhowever
one wishes to put itare as frustrating as any in contemporary philosophy. I do
not propose to address them here. The idea rather is to appeal to unanalyzed
common sense concerning a natural group of concepts and apparent conceptual
abilities. The apparent irreducibility of phenomenal qualities itself arises from appeal
to intuitions independent of the theory of reference; and it seems reasonable that we
should, in resolving that issue, appeal to notions that arise at the same intuitive level.
That we appear to have recognitional concepts and identifying dispositions that are
more or less determinate in their reference is hard to deny. My conception of those
hedges (seen around the neighborhood) may unambiguously pick out a variety of
eugenia. An example closer to the present topic is this. We can imagine an experi-
ment in which the experimenter tries to determine which internal property is the
focus of her subjects identications: again, . . . there it is again. There seems no
commonsensical implausibilityputting aside foundational worries about the
inscrutability of referencein the idea that there is a best possible answer to the
experimenters question, in the scientic long run.6)
3 Phenomenal Concepts as Recognitional Concepts

Here is the view to be defended. Phenomenal concepts are recognitional concepts
that pick out certain internal properties; these are physical-functional properties of
the brain. They are the concepts we deploy in our phenomenological reections; and
there is no good philosophical reason to deny that, odd though it may sound, the
6
For more on recognitional concepts and on the determinacy of reference, see Loar (1991a) [Chapter 11
in this volume], (1991b), (1995) [Chapter 13 in this volume].
PHENOMENAL STATES
properties these conceptions phenomenologically reveal are physical-functional

propertiesbut not of course under physical-functional descriptions. Granted that
brain research might discover that (what we take to be) our phenomenal concepts do
not in fact discriminate unied physical-functional properties. Failing that, it is quite
coherent for a physicalist to take the phenomenology at face value: the property of its
being like this to have a certain experience is nothing over and above a certain
physical-functional property of the brain.
Phenomenal concepts are conceptually independent of physical-functional
descriptions, and yet pairs of such concepts may converge on, pick out, the same
properties. Rebutting the semantic premise of the knowledge argument requires
making sense of the idea that phenomenal concepts conceive physical-functional
properties directly, that is, not by way of contingent modes of presentation. The
objective is to show that the knowledge argument fails for the same reason in the case
of Mary as in the case of Max: both arguments require substitution in opaque
contexts of terms that are conceptually independent. In the case of Max, the
conceptual independence appears to derive from alcohols connoting a contingent
mode of presentation that is metaphysically independent of the property referred to
by the chemical concept. In the case of Mary it has a different source.
What then accounts for the conceptual independence of phenomenal and
physical-functional concepts? The simple answer is that recognitional concepts and
theoretical concepts are in general conceptually independent. It is true that recogni-
tional concepts other than phenomenal concepts connote contingent modes of
presentation that are metaphysically independent of the natural kinds they pick
out, and hence independent of the kind referred to by the theoretical term of the
pair. But we need not count this metaphysical independence as essential to
the conceptual independence of coreferring recognitional and theoretical concepts.
Concepts of the two sorts have quite different conceptual roles. It is hardly surprising
that a recognitional conception of a physical property should discriminate it without
analyzing it in scientic terms. Nor should it be surprising that, if there are recogni-
tional concepts that pick out physical properties not via contingent modes of
presentation, they do not discriminate their references by analyzing them (even
implicitly) in scientic terms. Basic recognitional abilities do not depend on or get
triggered by conscious scientic analysis. If phenomenal concepts reect basic
recognitions of internal physical-functional states, they should be conceptually inde-
pendent of theoretical physical-functional descriptions. That is what you expect quite
apart from issues concerning physicalism.
An antireductionist may reply that the physicalist view depends on an ad hoc
assumption and that it is tendentious to suppose that phenomenal concepts differ
from all other recognitional concepts in not having contingent modes of presentation.
But this is not fair. Even on the antiphysicalist view, phenomenal concepts are
recognitional concepts, and we have direct recognitional conceptions of phenom-
enal qualities, that is, conceptions unmediated by contingent modes of presentation.
PHENOMENAL STATES
Evidently it would be absurd to insist that the antiphysicalist hold that we conceive of
a phenomenal quality of one kind via a phenomenal mode of presentation of a
distinct kind. And why should the physicalist not agree that phenomenal recogni-
tional concepts are structured in whatever simple way the antiphysicalist requires?
That is after all the intuitive situation, and the physicalist simply claims that the
intuitive facts about phenomenal qualities are compatible with physicalism. The
physicalist makes the additional claim that the phenomenal quality thus directly
conceived is a physical-functional property. On both metaphysical views, phenom-
enal concepts differ from other recognitional concepts; phenomenal concepts are a
peculiar sort of recognitional concept on any account, and that can hardly count
against physicalism. The two views agree about conceptual structure and disagree
about the nature of phenomenal qualities. To insist that physicalism implies,
absurdly, that phenomenal concepts could pick out physical properties only via
metaphysically distinct phenomenal modes of presentation is unmotivated. There
is, though, still more to be said about whether phenomenal concepts should be
regarded as having modes of presentation of some sort, and we continue the account
in section 5.
Suppose this account of how phenomenal concepts refer is true. Here is a semantic
consequence. The physicalist thesis implies that the judgments the state a feels like
that and the state a has physical-functional property P can have the same truth
condition even though their joint truth or falsity can be known only a posteriori.
I mean, same condition of truth in a possible world. For truth conditions are
determined in part by the possible world satisfaction conditions of predicates; and
if a phenomenal predicate directly refers to a physical property, that property
constitutes its satisfaction condition.
On this account, a phenomenal concept rigidly designates the property it picks out.
But then it rigidly designates the same property that some theoretical physical
concept rigidly designates. This could seem problematic, for if a concept rigidly
designates a property not via a contingent mode of presentation, must that concept
not capture the essence of the designated property? And if two concepts capture the
essence of the same property, must we not be able to know this a priori? These are
equivocating uses of capture the essence of. On one use, it expresses a referential
notion that comes to no more than directly rigidly designate. On the other, it means
something like be conceptually interderivable with some theoretical predicate that
reveals the internal structure of the designated property. But the rst does not imply
the second. What is correct in the observation about rigid designation has no
tendency to imply that the two concepts must be a priori interderivable.
4 The Concept Phenomenal Concept

Not all self-directed recognitional concepts are phenomenal concepts, as may be seen
in these two cases.
PHENOMENAL STATES
(1) Cramps have a characteristic feel, but they are not feelings. Cramps are certain
muscle contractions, while feelings of cramp are, if physical, brain states.
(Witness phantom-limb sufferers.) One has a recognitional concept that
picks out certain muscle contractions in the having of them. This is not a
phenomenal concept, for it does not purport to pick out a phenomenal quality.
But of course, in exercising this concept, one often conceives its reference by
way of a phenomenal mode of presentation, a cramp feeling or a cramp-feeling
image.
(2) A more fanciful self-directed nonphenomenal concept can be conceived.
To begin with, consider blindsight. Some cortically damaged people are phenom-
enally blind in restricted retinal regions; and yet when a vertical or horizontal
line (say) is presented to those regions, they can, when prompted, guess what
is there with a somewhat high degree of correctness. We can extend the
example by imagining a blindsight that is exercised spontaneously and accur-
ately. At this point we shift the focus to internal properties and conceive of a
self-directed recognitional ability, which is like the previous ability in being
phenomenally blank and spontaneous but which discriminates an internal
property of ones own. If this recognitional ability were suitably governed by
the concept that state, the resulting concept would be a self-directed recogni-
tional concept that is phenomenally blank.
The two examples show that phenomenal concept cannot mean self-directed
recognitional concept. This is compatible with my proposal. For it implies neither
(a) that we can reductively explicate the concept phenomenal quality as property
picked out by a self-directed discriminative ability, or (b) that we can reductively
explicate the concept phenomenal concept as self-directed recognitional concept.
Phenomenal concepts are certain self-directed recognitional concepts. Our higher-
order concept phenomenal concept cannot be reductively explicated, any more
than can our concept phenomenal quality. The higher-order concept phenomenal
concept is as irreducibly demonstrative as phenomenal concepts themselves.
5 Phenomenal Modes of Presentation

Self-directed recognitional concepts of the blindsight type might appear to raise a
problem for the claim that phenomenal concepts pick out physical-functional prop-
erties directly. Here is a way to put the point.
The difference between a self-directed blindsight recognitional concept and a phenomenal
concept appears to be that the latter involves a phenomenal mode of presentation while the
former conceives its referent in some other, odd, way. So, if the phenomenal concept is taken to
discriminate some physical property, it then does so via a phenomenal mode of presentation.
But that conicts with your assertion that phenomenal concepts refer directly, with no
contingent mode of presentation. A similar point arises concerning recognitional concepts of
PHENOMENAL STATES
cramps and of cramp feelings. Both concepts must presumably have modes of presentation.
It is far-fetched to suppose that one of them has and the other lacks a mode of presentation; the
phenomenal concept does not pick out a physical state nakedly. The cramp concept connotes
a mode of presentation of the form the physical state that causes such and such phenomenal
state. If we attempt to capture the phenomenal concept analogously, its mode of presentation
would have the form the state that has such and such phenomenal aspect. But then, contrary
to what the physicalist must say, phenomenal concepts point to physical states only by way of
phenomenal modes of presentation.
What might an antiphysicalist say about these various self-directed recognitional

concepts? Let me make a good-faith attempt to present a reasonable version.
(1) A cramp concept picks out a muscular property indirectly, by way of a causal
chain that is mediated by the phenomenal quality associated with the concept.
In addition to this mode of presentation typethe phenomenal qualitywe
can also note the role of, as we might say, token modes of presentation. One
and the same cramp concept (type) can on different occasions be focused
differently: by an actual cramp feeling, by a cramp-feeling image, or by an
imageless inclination to identify cramp feelings when they occur (with a
cramp-feeling image on the tip of ones imagination).
(2) We turn from cramp concepts to cramp-feeling concepts. These do not refer
(i.e., to cramp feelings) by way of contingent modes of presentation. But they
can mimic the working of cramp concepts as regards token modes of pres-
entation. If one can focus attention on the bodily property of cramp by way of
a token cramp feeling, surely one can focus attention on the phenomenal
quality cramp feeling by way of a token cramp feeling. The same goes for
cramp-feeling images and those gossamer identifying inclinations. Should
antiphysicalists say that cramp-feeling concepts have noncontingent modes
of presentation? We might say that a phenomenal concept has as its mode of
presentation the very phenomenal quality that it picks out. We might also say
that phenomenal concepts have token modes of presentation that are non-
contingently tied to the phenomenal qualities to which those concepts point:
particular cramp feelings and images can focus ones conception of the
phenomenal quality of cramp feeling.
(3) As for self-directed blindsight concepts, the antiphysicalist then ought to say,
they differ from phenomenal concepts in the obvious way, whether one puts it
by saying that they lack the noncontingent phenomenal modes of presentation
(types) that phenomenal qualities have, or that they lack their phenomenal
token modes of presentation.
The main point is by now more than obvious. Whatever the antiphysicalist has
said about these cases the physicalist may say as well. The idea that one picks out the
phenomenal quality of cramp feeling by way of a particular feeling of cramp
(or image, etc.) is hardly incompatible with holding that that phenomenal quality
PHENOMENAL STATES
is a physical property. The contrast between phenomenal concepts and self-directed

blindsight concepts and cramp concepts nds physicalist and antiphysicalist equally
able to say something sensible.
A phenomenal concept exercised in the absence of the phenomenal quality it
stands for often involves not merely a recognitional disposition but also an image.
And so, as a psychological state in its own right, a phenomenal conceptgiven its
intimate connection with imagingbears a phenomenological afnity to a phenom-
enal state that neither state bears to the entertaining of a physical-theoretical concept.
When we then bring phenomenal and physical-theoretical concepts together in our
philosophical ruminations, those cognitive states are phenomenologically so different
that the illusion may be created that their references must be different. It is as though
antiphysicalist intuitions rest on a resemblance theory of mental representation, as
though we conclude from the lack of resemblance in our phenomenal and physical-
functional conceptions a lack of sameness in the properties to which they refer.
6 Third-person Ascriptions
Ascriptions of phenomenal qualities to others ostensibly refer to properties that
others may have independently of our ascribing them:7 we have realist conceptions
of the phenomenal states of others. But at the same time they are projections from
ones own case; they have the form x has a state of this sort, where the demonstrative
gets its reference from an actual or possible state of ones own.
Can phenomenal concepts as we predicate them of others be identied with the
recognitional concepts we have characterized? A question naturally arises how
essentially self-directed recognitional concepts can be applied in cases where it
makes no sense to say that one can directly apply these concepts. This is a question
that exercised Wittgensteinians.
As we have already pointed out, recognitional concepts are perspectival, in the
sense that their reference is determined from a certain constitutive perspective
(depending on the concept). The above concept those creatures1 (seen up close)
picks out a creature-kind that one discriminates on close sightings. But nothing
prevents ascribing the recognitional concept one of those creatures1 to something
observed from a different perspective, seen in the distance or heard in the dark. We
have to distinguish the perspective from which reference is determined and the far
broader range of contexts in which the referentially xed concept can be ascribed.
The former perspective hardly restricts the latter contexts. This holds also for
phenomenal concepts. We acquire them from a rst-person perspective, by discrim-
inating a property in the having of it. Assuming that we successfully pick out a more
7
The earlier version of this chapter [Loar (1990)] made heavy weather of third-person ascription of
phenomenal concepts. General considerations about the perspectival nature of recognitional concepts
permit a far neater account, which I here present.
PHENOMENAL STATES
or less determinate physical property, the extraperspectival ascription she is in a state

of this kind makes complete sense. And so it is not easy to see that Wittgensteinians
succeeded in raising a philosophical problem that survives the observation that
we can discriminate physical properties and so x the reference of phenomenal
concepts from a rst-person perspective, and then go on to ascribe those concepts
third-personally.
There is though a more up-to-date worry about the interpersonal ascribability of
rst-person concepts, however physical we suppose their references to be. Evidently
there will be vagueness, and indeterminacy, concerning whether another person
whose neural assemblies will presumably always differ from mine in various
respectshas a certain physical property that I discriminate phenomenally. And
this on the face of it poses a problem, which may be framed as follows:
The question whether another persons phenomenal states resemble yours can hardly consist
in their neural assemblies resembling yours. Any physical similarity you choose will be
arbitrarily related to a given phenomenal similarity. Suppose there is a small physical difference
between a neural state of yours and another persons state. What makes it the case that this
small neural difference constitutes a small phenomenal difference or a large one or no
phenomenal difference at all? It appears that there cannot be a fact of the matter.
But this objection appears to me to overlook a crucial element of the physicalist view
we have presentedthat phenomenal concepts are (type-)demonstrative concepts
that pick out physical properties and relations. A rst step in answering it is to
consider the connection between interpersonal and intrapersonal phenomenal simi-
larity. It appears that ones phenomenological conception of how others phenom-
enal states resemble ones own has to be drawn from ones idea of how ones own
phenomenal states resemble each other. A persons quality space of interpersonal
similarity must derive from her quality space of intrapersonal similarity. How else is
one to get a conceptual grip on interpersonal phenomenal similarity? This seems
inevitable on any accountphysicalist or antiphysicaliston which phenomenal
concepts are formed from ones own case.
But conceptions of phenomenal similarity relations are as much type-
demonstrative concepts as those of phenomenal qualities. All one can apparently
mean by that spectrum of phenomenal similarity is that ordering among my
phenomenal states. Physicalism implies that if such a type-demonstrative refers, it
picks out a physical ordering. And there is no obvious philosophical difculty (if we
put aside skepticism in the theory of reference) in the idea that discriminations of
resemblances and differences among ones own phenomenal properties pick out
reasonably well-dened physical relations.
Now I have to confess some uneasiness about extending this to interpersonal
similarity without qualication; but the implications of the foregoing remarks are
clear enough. If they are correct, whatever physical ordering relations are picked out
by ones personal notions of phenomenal similarity must also constitute (what one
PHENOMENAL STATES
thinks of as) interpersonal phenomenal similarity. It is easy to see that there still is
room here for further trouble. But the difculty the objection raises seems consider-
ably diminished if one insists on the demonstrative nature of all phenomenal
concepts, however relational and of whatever order. For the objection then becomes,
Suppose there is a small physical difference between a neural state of yours and
another persons state. What makes it the case that this small neural difference
constitutes a small difference of that type, or a large one, or no difference of that
type at all? If that type picks out a physical relation, then the question answers itself,
and there seems no gloomy philosophical threat of phenomenal incommensurability.
Naturally there is the risk that physical investigation will not deliver the right
physical properties and relations. Even if the risk is increased by bringing in inter-
personal similarities, the nature of the risk is the same as in ones own case: the
phenomenal might turn out to be not adequately embodied.
It goes without saying that one can coherently conceive that another person has P,
conceived in physical-functional terms, and doubt that she has any given phenom-
enal quality; that has been central to this chapter. But one cannot coherently wonder
whether another person in a P state has a state with this phenomenal quality if one
acknowledges that ones concept this quality refers to the property the concept
discriminates in oneself (what else?) and that moreover it discriminates P.
Why then is there an apparent problem of other minds? It is as if one wishes to do
to others as one does to oneselfnamely, apply phenomenal concepts directly, apply
phenomenal recognitional capacities to others from a rst-person perspective. The
impossibility of this can present itself as an epistemological barrier, as something that
makes it impossible to know certain facts. Doubtless more can be said in explanation
of the naturalness of the conation of the innocuous conceptual fact with a severe
epistemological disability. It is not easy to shake the grip of that conation or
therefore easy to dispel the problem of other minds. The cognitive remedy, the
fortication against the illusion, is the idea of recognitional concepts that can be
ascribed beyond their constitutive perspective, coupled with the reection that there
is no reason to doubt that it is physical-functional properties that those recognitional
concepts discriminate.
7 Knowing How versus Knowing That

Consider a different physicalist reply, to an antiphysicalist argument posed in this
form: knowledge of physical-functional facts does not yield knowledge of the
phenomenal facts; therefore phenomenal facts are not physical-functional. Lawrence
Nemirow and David Lewis have replied in effect that the premise is true only if you
equivocate on knowledge.8 The rst occurrence means theoretical knowledge, the
8
Nemirow (1980); Lewis (1983c).
PHENOMENAL STATES
second the ability to discriminate introspectively or to imagine certain properties.

But theoretical knowledge of physical-functional properties that are identical with
phenomenal qualities does not yield the other sort of knowledge of the same
properties, that is, the ability to discriminate them in introspection or to imagine
them. There are two epistemic relations to one class of properties.
Now this suggests something signicantly different from my account. On the
Nemirow-Lewis proposal, the only knowledge that such and such is knowledge
couched in physical-functional terms, while what corresponds to (what we have been
calling) phenomenal concepts is knowing how to identify or to imagine certain states.
What I have proposed is evidently different. Knowing that a state feels a certain way
is having distinctive information about it, couched in phenomenal conceptions.
There is of course a central role for recognitional abilities, but that is in the
constitution of phenomenal concepts. Antiphysicalists are right to count phenom-
enal knowledge as the possession of distinctive information, for it involves genuinely
predicative components of judgment, whose association with physical-functional
concepts is straightforwardly a posteriori.
Physicalists are forced into the Nemirow-Lewis reply if they individuate pieces of
knowledge or cognitive information in terms of possible-world truth conditions, that
is, hold that knowing that p and knowing that q ascribe distinct pieces of know-
ledge just in case that-p and that-q denote distinct sets of possible worlds. Then
knowing that xs phenomenal qualities are such and such will be distinct from
knowing that xs physical properties are so and so only if the former qualities are
distinct from the latter properties. So then a physicalist who counts the basic
antiphysicalist premise as true on some interpretation must deny either that know-
ledge, cognitive information, is individuated in terms of possible-world truth
conditions or deny that knowing the phenomenal facts (in the sense that makes
the basic antiphysicalist premise true) is knowing that such and such or having
distinctive information about it. Nemirow and Lewis deny the latter. Of course I deny
the former; there are ample independent reasons to deny it, and it seems otherwise
unmotivated to deny the latter.
There are straightforward reasons to prefer the phenomenal concept view.
1. A person can have thoughts not only of the form coconuts have this taste but
also of the form if coconuts did not have this taste, then Q. You may get away
with saying that the former expresses (not a genuine judgment but) the mere
possession of recognitional know-how. But there is no comparable way to
account for the embedded occurrence of coconuts have this taste; it occurs
as a predicate with a distinctive content.
2. We entertain thoughts about the phenomenal states of other peopleshe has a
state of that type; this clearly calls for a predicative concept. It does of course
involve a recognitional ability, but one that contributes to the formation of a
distinctive concept.
PHENOMENAL STATES
3. For many conceptions of phenomenal qualities, there is no candidate for an

independently mastered term that one then learns how to apply: thinking of a
peculiar way my left knee feels when I run (a conception that occurs predica-
tively in various judgments) is not knowing how to apply an independently
understood term. I suppose a functionalist might say that, in such cases, one
implicitly individuates the state in terms of some functional description that is
fashioned on the spot, but this appears psychologically implausible.
8 The Explanatory Gap

Can we explain how a certain phenomenal property might be identical with a certain
physical-functional property? The answer is no, and then again, yes.
First, the no. When we explain, say, liquidity in physical-functional terms, the
explanation is in crucial part a priori. You may nd this surprising; but what we in
effect do is analyze liquidity (or more precisely those aspects of liquidity that we
count as explained9) in terms of a functional description, and then show that the
physical theory of water implies, a priori, that the functional description is realized.
But given the conceptual independence of phenomenal concepts and physical-
functional concepts, we cannot have such an a priori explanation of phenomenal
qualities in physical-functional terms.
Does this matter? The explanatory gap, as it appears to me, is an epistemic or
conceptual phenomenon, without metaphysical consequences,10 and it is predictable
from the physicalist account we have proposed. But this may seem somewhat glib.
As Georges Rey points out (Rey (1995)), the mere fact of conceptual inequivalence
for recognitional type-demonstratives and descriptive terms does not generate an
explanatory gap. Many examples would make the point. We do not nd a troubling
explanatory gap in judgments of the form that stuff is CH3CH2OH, even though this
does not hold a priori.
Now what is it that needs accounting for? This seems to me to be it: how identity
statements that connect phenomenal concepts and physical-functional concepts can
be true despite our sense that, if true, they ought to be explanatory and yet are not.
We can explain how such identity statements fail to be both explanatory (conceptual
independence) and true; but this does not account for the thought that something
that ought to be there is missing. We have to explain away the intuition that such
identity statements ought to be explanatory.
9
This leaves open the possibility of Twin-Earth cases in which the apparently dening properties of
liquiditythose that are functionally explainedare kept constant across worlds even though the underlying
kind changes. The dening properties then turn out to be merely reference xing.
10
For illuminating accounts of the explanatory gap and its signicance see Levine (1983), (1993).
Levines diagnosis of the signicance of the explanatory gap is different from mine.
PHENOMENAL STATES
There must be something special about phenomenal concepts that creates the
expectation and the consequent puzzle. We have already seen a signicant differ-
ence between phenomenal concepts and all other phenomenally mediated recogni-
tional concepts. Might this make the difference here as well? That is what I will try
to show.
Perhaps this is why we think that true phenomenal-physical identity judgments
ought to be explanatory. It is natural to regard our conceptions of phenomenal
qualities as conceiving them as they are in themselves, that is, to suppose we have
a direct grasp of their essence. So in this respect there is a parallel with liquidity: the
phenomenal concept and the concept liquid both pick out properties directly, that
is, not via contingent modes of presentation. And of course the physical-functional
theoretical term of the identity, couched in fundamental theoretical terms, also
reveals the essence of the property it picks out. Since both conceptions reveal this
essence, then, if the psychophysical identity judgment is true, the sameness of that
property, it might seem, ought to be evident from those conceptions, as in the
liquidity case. The physical-functional concept structurally analyzes the property,
and so we expect it to explain, asymmetrically, the phenomenal quality, much as
physics explains liquidity, on the basis of an a priori analysis. The fact that this is not
so makes it then difcult to understand how there can be just one property here.
If this is what makes the explanatory gap troubling, then the idea that phenomenal
concepts are recognitional concepts of a certain sort does account for the explanatory
gap in a way compatible with physicalism. Phenomenal concepts, as we have seen, do
not conceive their reference via contingent modes of presentation. And so they can
be counted as conceiving phenomenal qualities directly. Calling this a grasp of
essence seems to me all right, for phenomenal concepts do not conceive their
references by way of their accidental properties. But this is quite a different grasp
of essence than we have in the term liquid: for that term (or what there is in it that
we count as functionally explained) is conceptually equivalent to some functional
description that is entailed by the theoretical term of the identity.
The problem of the explanatory gap stems then from an illusion. What generates
the problem is not appreciating that there can be two conceptually independent
direct grasps of a single essence, that is, grasping it demonstratively by experiencing
it, and grasping it in theoretical terms. The illusion is of expected transparency: a
direct grasp of a property ought to reveal how it is internally constituted, and if it is
not revealed as physically constituted, then it is not so. The mistake is the thought
that a direct grasp of essence ought to be a transparent grasp, and it is a natural
enough expectation.
The explanatory gap has led many philosophers of mind seriously astray into
mistaken arguments for epiphenomenalism, for mystery, for eliminativism. At the
root of almost all weird positions in the philosophy of mind lies this rather elemen-
tary and unremarkable conceptual fact, blown up into a metaphysical problem that
appears to require an extreme solution. But it is a mistake to think that, if physicalism
PHENOMENAL STATES
is true, consciousness as we conceive it at rst hand needs explaining in the way that
liquidity as we ordinarily conceive it gets explained.
There is another interpretation of can we understand how physicalism might be
true?, for which the answer is clearly yes. For we can explain, and indeed we have
explained, how a given phenomenal concept can manage to pick out a particular
physical-functional property without remainder: the concept discriminates the prop-
erty but not via a contingent mode of presentation. This in its way closes the
explanatory gap between the phenomenal and the physical. We understand how
such and such phenomenal quality could pick out physical property P, even though
such and such phenomenal quality = P does not provide an (a priori) explanation in
physical terms of why a given phenomenal quality feels as it does. Since the former,
when generalized, would entail that physicalism about phenomenal qualities is true,
and since we understand both of these things, we thereby understand how physical-
ism can be true.
9 Subjective Concepts and Subjective Properties

You can ascribe an objective propertyone completely expressible in the objective
terms of natural scienceunder a subjective conception: xs state has this quality.
Thomas Nagel writes that mental facts are accessible only from one point of view.11
This does reect something about phenomenal concepts; they are in some intuitive
sense from a point of view and moreover subjective. Phenomenal concepts are
subjective because they are essentially self-directed, involve capacities to discrim-
inate certain states in the having of them, and also involve imaginative capacities
anchored in such recognitional capacities. If that is it, then Nagel takes a correct
observation about concepts and draws a wrong conclusion about facts and proper-
ties. For concepts can in that sense be from a point of view and subjective, and still
introduce properties that are exhaustively captured in objective science.
But we can go further. Let us grant even that the property of experiencing such and
such is aptly counted as subjective, as intrinsically involving a point of view. Why
should this subjectivity not itself be identical with a physical-functional property, and
therefore completely objectively conceivable under its physical-functional descrip-
tion? There is no contradiction in supposing that a property that is subjectivein the
sense of being individuated in a way that invokes a relation to a mindis also
conceivable under an objective mode of presentation. There is no incoherence in
the thought that the subjectivity of a phenomenal quality is identical with an
objective physical-functional aspect of that property.
Does a fully objective description of reality not still leave something out, viz., the
subjective conceptions? This is a play on leave something out. A complete objective
11
Nagel (1974).
PHENOMENAL STATES
description leaves out subjective conceptions, not because it cannot fully characterize
the properties they discriminate or fully account for the concepts themselves as
psychological states but simply because it does not employ them.
10 Phenomenal Structure, and Exotic Others

Some functionalists might think this account ignores a major feature of our conceptions
of the mental, namely, their systematic structure. We have conceptions of different
sensory modalities, and of intramodality comparisons along various spectra, of pitch,
timbre, hue, brightness, shape, size, texture, acidity, acridity, and so on. These could be
seen as subsidiary functional organizations within a theory of the mental. Antiphysicalists
may share something of the point, wanting to speak of phenomenological structures.
My account could seem to imply that phenomenal concepts are atomistic, unstructured,
unsystematic, for are these recognitional dispositions not in principle independent of
each other?
We have phenomenal recognitional concepts of various degrees of generality.
Some are of highly determinate qualities, and others are of phenomenal determin-
ables: crimson, dark red, red, warm colored, colored, visual. The last is the recogni-
tional conception of a whole sensory modality. And there is the most general of all,
the recognitional concept phenomenal (state, quality), the highest ranking phenom-
enal determinable. (This is a recognitional concept. One discriminates phenomenal
states from nonphenomenal states, feeling a twinge from having a bruise, hearing a
chirp from jerking a knee, and that highly general discriminative capacity is the basis
of the concept of a phenomenal quality.)
There are also relational concepts: quality x is a determinate of quality y; quality x
is more like quality y than like quality z; quality x is of a different modality from
quality y. These are also recognitional concepts: dispositions to classify together, on
phenomenal grounds, certain pairs and triples of phenomenal qualities. Combining
them yields complex conceptions of abstract phenomenal structures, for example, of
a structured sensory modality. Ones general conception of such a structure is in
effect ones ability to exercise in concert a group of such general phenomenal
concepts.
Now it is important that our conceptions of such phenomenal structures, while
abstract, are yet phenomenal conceptions. No purely functional conception of a
complex structure, however isomorphic to a phenomenal-structure conception it
may be, will be cognitively equivalent to it; purely functional conceptions ignore that
the structures are of phenomenal similarity relations, of phenomenal determinate-
ness, and so on.
But given the falsity of the semantic minor premise, that is no impediment to
holding that those abstract phenomenal conceptions can have purely functional or
physical-functional structures as their references. For such structures may well be
what these abstract phenomenal recognitional capacities in fact discriminate. Indeed
PHENOMENAL STATES
we may go on to say that, if our phenomenal conceptions are to be fully vindicated by

brain science, then the brain must have a certain functional structure; any possible
totality of (as it were) semantic values for our phenomenal conceptions must have
certain functional structures. This perhaps explains the strong intuition of some
common-sense functionalists that phenomenal concepts are functional concepts,
without our having to accept that counterintuitive view.
Can your projection analysis accommodate the thought that a bat has highly
specic, determinate, phenomenal states that are not like anything I can experience
or imagine? It seems to me that your program will require you to bring in the bats
own recognitional-imaginative capacities, such as they are.12
When one thinks about a bats sonar phenomenal states, one thinks about them as
phenomenal, that is, as having in common with my phenomenal states what
I discriminate them all as having in common, and that may be something physical-
functional. One also thinks of them as of a distinctive phenomenal kind or modality,
different from ones own states, of roughly that order of determinateness at which
ones visual states are marked off from ones auditory states. One has such a general
concept from ones own case, and one can project it. Again, that conceptdistinctive
phenomenal modalitymay denote a physical-functional property, of sets of phe-
nomenal states. And one thinks of the bats sonar states as exhibiting phenomenal
variation of different degrees of specicity. These conceptions of general phenomenal
structure, determinable-determinate relations, resemblance relations, and so on, we
have, as I have said, from our own case.
Now nothing in the foregoing requires that a necessary condition of having certain
phenomenal qualities is having the capacity to discriminate them. (See, however, the
discussion of transparency in section 11.) We ascribe to bats not phenomenal
concepts but phenomenal states; and we do that by projection, in the manner
characterized above. Other-directed phenomenal conceptions are of others states,
and not as such of their conceptions.
Nagel proposes that we can achieve objectivity about the mental by abstracting
from subjective conceptions of our own psychology, fashioning objective mental
conceptions that are neither physical nor functional.13 This would enable us to
conceive abstractly of mental lives of which we have no subjective, projective,
understanding whatever. Now that is evidentially at odds with my proposal. It
appears to me that all mental concepts that are not functional concepts (where
the latter include concepts of theoretical psychology) are subjective-projective
concepts, however general and abstract they may be. The reason is simple: as far
as I can determine, I have no objective nonfunctional mental concepts. If I try to
conceive an alien mind in nonfunctional mental terms, I rely on concepts like
sensory modality and other general conceptions of phenomenological structure
12 13
Thomas Nagel, in a note commenting on an earlier draft. Nagel (1974).
PHENOMENAL STATES
of the sort mentioned above, and I understand them from my own case. They are
abstract conceptions; but, it appears to me, they are still recognitional concepts
and hence as subjective as the highly specic phenomenal concept of having an
itch in the left ankle.
11 Transparency
The following could appear possible on my account: another person is in the state
that in me amounts to feeling such and such but sincerely denies feeling anything
relevant. It apparently has been left open that others have phenomenal states that are
not introspectible at will, for no requirement of transparency has been mentioned.
Then the property that is the referent of my concept of feeling like that could, even if
it occurs transparently in me, occur non-transparently in you. But (the objection
continues) denying transparency is tantamount to allowing unconscious experiences;
and it would not be unreasonable to say that the topic of phenomenal states is the
topic of certain conscious states.
There really is no issue here. Suppose that any phenomenal quality must be
essentially transparent, and that no property I correctly identify as phenomenal
can be realized in another non-transparently. If cognitive integration is essential to
the intuitive property of transparency, so be it; there is no reason to think that such
integration itself is not a physical-functional property, as it were implicated by each
phenomenal property.
But it is not obvious that phenomenal properties must be transparent in such a
reexive cognitive sense. What about infants and bats? There has always been a
philosophical puzzle about how subtracting reexive cognitive awareness from
phenomenal or conscious states leaves something that is still phenomenal or con-
scious. But that puzzle is independent of the present account. All that is implied here
is that if I have a conception of a phenomenal quality that is shared by me and an
infant, my conception of it involves a recognitional concept, and there is no reason
why that phenomenal quality itself should not be a physical-functional property.
Whatever indenable, elusive aspect of phenomenal qualities might constitute their
being conscioustransparent in some appropriately minimal sensewithout requir-
ing reexive conceptualizability, there would be no reason to doubt it is a physical-
functional property.
12 Incorrigibility
Physicalism, it may be said, cannot acknowledge the incorrigibility of phenomenal
judgments of the form it feels like that. For surely there is no guarantee that a
capacity for recognizing a given physical property does not at times misre; and
perhaps even more to the point, there can be no guarantee that to a given recogni-
tional disposition there corresponds a repeatable physical property. Perhaps an
PHENOMENAL STATES
antiphysicalist will grant that certain kinds of mistake about phenomenal qualities
are possible;14 but the antiphysicalist will insist that we cannot be wrong in thinking
that there are phenomenal qualities.
Now suppose it turns out that no system of physical-functional properties corres-
ponds to the system of our phenomenal concepts. Would a physicalist not then have
to say there are no phenomenal qualities? And is the fact that physicalism leaves this
open not a serious problem?
But that very possibility ought to make us dubious about the incorrigibility of
the judgment that there are real phenomenal repeatables. What reason have we to
think that our phenomenal judgments discriminate real properties? Memory, one
might say, cannot be that mistaken: we can hardly deny that present inner states
resemble past states in ways we would recognize again. Despite this conviction,
however, if no system of physical-functional properties corresponded to ones
putative phenomenal discriminations, an alternative to nonphysical qualities
would be this: memory radically deceives us into thinking we discriminate internal
features and nonrandomly classify our own states. Strong evidence that no suitable
physical-functional properties exist might amaze and stagger one. It would then
have emerged that we are subject to a powerful illusion, a cognitive rather than a
phenomenal illusion; we would be judging falsely that we thereby discriminate
real properties.
It does seem likely that we genuinely discriminate internal physical-functional
states in introspection.15 But with that said, positing non-physical properties to
forestall the possibility of radical error, however theoretically adventurous (even
reckless) this may be, would in something like a moral sense still be rather faint-
hearted. The whole point about the phenomenal is how it appears. And that means
there is no introspective guarantee of anything beyond mere appearance, even of
discriminations of genuine repeatables. The dualist balks at the implications and
invents a realm of properties to ensure that the appearances are facts, but this does
not respect the truly phenomenal nature of what is revealed by introspection at its
least theoretical.
I have to grant that, if it were to turn out that no brain properties are suitably
correlated with our ascriptions of phenomenal qualities, one might well feel some
justication in questioning physicalism. But that does not imply that one now has
such a justication. There is no good reason for prophylactic dualism.
14
See Warner (1986).
15
When I see a ripe lemon in daylight and attend to my visual experience, I form the memory belief that
what I introspect is what I introspected (phenomenologically inclined as I am) the last time I saw a ripe
lemon in daylight. It seems a reasonable empirical inference that probably ripe lemons in such circum-
stances cause in me states that my memory accurately records as the same. But this inference is, I take it,
not reasonable on introspective grounds alone; it presupposes much about how the world works.
PHENOMENAL STATES
13 Functionalism
There are two functionalist theses: that all concepts of mental states are functional
concepts, and that all mental properties are functional properties. The rst I rejected
in accepting the antiphysicalist intuition. I agree with the antiphysicalist that
phenomenal concepts cannot be captured in purely functional terms. But nothing
in philosophy prevents phenomenal properties from being functional properties.
There are two possibilities: they are common-sense functional properties or they
are psychofunctional, and I take the latter to be the interesting one.16 Might the
phenomenal quality of seeing red be identical with a property captured by a detailed
psychological theory? This would be so if the repeatable that triggers ones phenom-
enal concept seeing red has psychofunctional rather than say biochemical identity
conditions. That this is possible has been denied by antifunctionalist physicalists on
the grounds of inverted qualia and absent qualia possibilities, but I do not nd these
arguments persuasive.
The inverted qualia argument is commonly advanced against identifying phenom-
enal qualities with common-sense functional properties and also against the psycho-
functional identication. The position I espouse is agnostic: for all philosophers
know, phenomenal qualities are psychofunctional, neurofunctional, or some other
ne-grained functional properties. The opposing argument is that it is possible that
the functional role that seeing red has in me is had in you by, as I would think of it,
seeing green. If this is, as they say, metaphysically possible, then of course phenom-
enal qualities are not functional properties.
But it seems the only argument for the possibility is the coherent conceivability of
inverted qualia. One cannot presuppose that inverted qualia are nomologically
possible. There seems to be no philosophical reason to assert that, apart from the
coherent conceivability of inverted qualia. If there is empirical reason to assert that
nomological possibility, then of course we should retreat from agnosticism. The
present point is that nothing about the idea of inverted qualia provides philosophical
reason to reject functionalism about qualia. For that would require another version of
the antiphysicalist argument: it is conceivable that any given functional state can
16
It is empirically unlikely that phenomenal qualities are identical with common-sense functional
properties. Here is one way to see this. We know sensations can be produced by nonstandard means, that
is, by poking around in the brain; but this of course is no part of the common-sense functional role of the
property of seeing red. Now suppose this property is produced in me by a brain probe. What constitutes its
being a sensation of red? If it is its common-sense functional role, then that property would be the
sensation of red by virtue of (something like) its normally having such and such causes and effects (it
doesnt have them here). But this makes sense only if the property in question is itself a distinct lower-order
property about which it is contingently true that normally it has such and such causes and effects although
it lacks them here. That lower-order property would then be a far better candidate (than the common-sense
functional property) for being the property ones phenomenal conception discriminates. For this reason,
such brain probes turn out to be strong and perhaps even conclusive evidence that phenomenal qualities,
the ones we discriminate in applying phenomenal concepts, are not identical with common-sense
functional properties. There are other ways of reaching the same conclusion.
PHENOMENAL STATES
occur without the seeing of green and with the seeing of red, say; therefore the
psychofunctional role and the phenomenal quality involve distinct properties.
Clearly one cannot accept this argument against functionalism without also accept-
ing the analogous argument against physicalism itself; the philosophical antifunc-
tionalist argument requires a premise that implies antiphysicalism.
There is a well-known absent qualia argument against functionalism by Ned Block
(1978). Suppose the Chinese nation were organized so as to realize the psychofunc-
tional organization of a person seeing green. Evidently the Chinese nation would not
collectively be seeing green or having any other sensation. Any psychofunctional
property could in this way be realized without a given phenomenal quality and hence
cannot be identical with one. Now this argument might appear dialectically more
telling than the inverted qualia argument, for it apparently rests on more than a
conceptual possibility. It seems a plain truth that the Chinese people would not
thereby be having a collective sensation. Surely it is barmy to be agnostic about that.
Block suggests a principle. If a doctrine has an absurd conclusion which there is no
independent reason to believe, and if there is no way of explaining away the absurdity
or showing it to be misleading or irrelevant, and if there is no good reason to believe
the doctrine that leads to the absurdity in the rst place, then dont accept the
doctrine.17
While we doubtless nd an absurdity in ascribing phenomenal qualities to the
Chinese nation as a whole, the matter is not so simple. It is hard to see how such a
judgment of absurdity can be justied except by our having some intuitive knowledge
of the nature of phenomenal qualities whereby we can say that the Chinese nation
cannot have them collectively. Have I a special insight into my physical states
whereby I can say: the repeatable that I reidentify whenever I attend to my seeing
green is not a functional property? One feels skeptical that introspection can yield
such knowledge. If the argument is not they do not collectively have, by virtue of
their functional organization, however ne-grained, what I have when this occurs,
then what is it? Is a further philosophical argument in the ofng? It is difcult to see
whence chest-beating to the contrary derives its credibility. Perhaps a dualist con-
ception of Platonic insight into mental essences might help. But, on a naturalist view
of human nature, one ought to nd it puzzling that we have such a rst-person
insight into the nature of our mental properties. Perhaps there is reason to suppose
that what one introspects and reidenties is a categorical and not a dispositional
property. That has an intuitive ring to it, but it is not that easy to produce a decent
argument for it. We are left with this question: how might we know short of detailed
brain research that what we reidentify in ourselves when we see green is not a ne-
grained functional property? But if we cannot know this by sheer insight into the
essence of our own properties, or by philosophical argument, then we cannot know
17
Block (1978).
PHENOMENAL STATES
that the Chinese nation lacks what we have. Our ignorance concerns the nature of
our own properties, and that ignorance would appear to prevent drawing substantive
conclusions from thought experiments of this type.
There is no question that ordinary intuition counts strongly against applying
phenomenal concepts to things that are not single organisms, and one cannot deny
that the reply just given makes one uncomfortable, at the very least. And yet the
alternative appears to be Platonism about mental essences, and that sits awkwardly
with naturalism. It is possible that phenomenal qualities are biochemical properties:
and yet again it is difcult to see that philosophers know anything that implies that
they are not ne-grained functional, or neurofunctional, properties.18
18
(Original version) For pointing out a substantial error in an ancestor of the paper, I am indebted to
George Myro, whose correction put me on the right track as I now see it. I have learned much from
conversations on phenomenal qualities with Janet Levin and Richard Warner. Stephen Schiffer made
several valuable suggestions about the structure of the paper and got me to clarify certain arguments. I am
also grateful for comments on the mentioned ancestor to Kent Bach, Hartry Field, Andreas Kemmerling,
Dugald Owen, Thomas Ricketts, Hans Sluga, Stephen Stich, and Bruce Vermazen.
(Revised version) Many thanks to Ned Block for raising questions about modes of presentation and the
blindsight case, to Georges Rey for making me see that more needed to be said about the explanatory gap,
and to Kent Bach for helpful remarks on a number of points.
11
Can We Explain Intentionality?
I
Jerry Fodors treatment of intentionality in chapter 4 of Psychosemantics (1987) is
wonderfully direct, so much so that the rst time through you can hardly believe the
effrontery of it. Even many causal theorists have come to think that what determines
the reference of predicates cannot be simple. But Fodor puts aside holism, anti-
individualism, and all the rest to propose an exceedingly straightforward theory of
meaning (some would say reference), one that is atomistic in the extreme, reductionist
with no hedging at all, and completely unfazed by anti-individualism. Not that he
explicitly undertakes to explain all intentionality, for he formally proposes a sufcient
rather than a necessary or general condition for predicate reference, happy if he shows
that intentionality can at least sometimes be secured by his naturalistic conditions. But
you do get the impression that he is inclined to think the account points to a general
solution. This stimulating and provocative theory, which Fodor so adeptly applies to
difcult cases, is admirable in its not mincing about but striding boldly through the
land minesatomism, reductionism, individualismbracing stuff.
The theory is a variant on the idea that meaning is information, that a predicate
means the property to which its ascriptions are responsive. The question arises how
falsity is possible on any such theory. Meaning, as Fodor puts it, is robust; a predicate
Fs meaning P can survive indenitely many ascriptions of F to non-Ps,1 and thus
F does not mean just any property to which it is responsive. His solution is this. If you
falsely call some goats sheep, the causal relation of those goats to your utterance
depends in part on the semantic fact that the word sheep means sheep; had it meant
anchovy, or bottle of beer, or nothing at all the goats would not have occasioned you to
utter it. Now, Fodor says, keep that structure; but instead of saying that false ascriptions
of sheep to goats depend on the independent semantic relation of sheep to sheep, say
they depend on an independent causal relation of sheep to sheep. Then you have a
sufcient condition of meaning: sheep cause ascriptions of sheep, and if any non-sheep
1
Fodor rejects a certain natural class of solutions, to the effect that a terms meaning consists in what it
discriminates in special meaning-determining situations, that meaning consists in information modulo
such situations, while in other situations truth or falsity depends on meaning thus determined. Fodor
objects that there are no such meaning-determining situations; in all situations it will be possible to assert
sheep of a goat, and nevertheless mean sheep and not sheep-or-goat (or whatever). More on this below.
CAN WE EXPLAIN INTENTIONALITY ?
cause them, this latter relation depends on the former causal relation; and this does not
also hold the other way around. So a predicate F means a property P if things that are P
cause F and any such causal relation between some other property and F is asymmet-
rically dependent on the former relation. In a recent paper2 Fodor modies this to
allow for Fs meaning an uninstantiated property (cf. unicorn), by weakening the
causal requirement: things that are P need not actually have caused F but there must be
a nomic relation (some counterfactual causal relation) between P and F, on which
false ascriptions would asymmetrically depend.3
Fodor is, I believe, right to hold that discriminative causal relations, socially
unmediated, can determine meaning or reference. But this seems to me to obtain
not for sheep, horse, and the like as they normally occur in our thoughts, but rather
for predicative concepts of a distinctive and not widely acknowledged variety. Even
for these, however, we will not nd the fully naturalistic treatment Fodor seeks,
for no causal condition can explicate reference except in conjunction with further
intentional conditions. This leads me to reject the strict atomism Fodor aims at,
although something more or less atomistic about the reference of such predicates
will survive.
II
Begin with this sample question: which predicates or predicative concepts might
Fodors condition t? Many predicates in my thoughts do not apparently have
meaning or reference in Fodors way. Their reference derives, not from my ascriptions
of them, but rather from the language I speak, from the social-semantic relations I draw
on. This fact stems from my semantic deference, my accepting that thoughts that
ascribe such a predicate can be false because I misconceived its meaning as that is
determined socially. I mistook the meaning of maple, often ascribing it to the sweet
gums I saw, taking it to mean them. Those ascriptions in no way depended on cases in
2
[The reference in the original publication was to a paper in press, probably intending to refer to Fodor
(1990b).]
3
If ordered to say sheep by a maniac holding a gun to my head, I would comply; but then Fodors
condition does not secure that sheep means sheep for me, for not every utterance of sheep by me would
either be caused by a sheep or be asymmetrically dependent on such a causal connection (my complying
would not hang on the meaning of sheep). Such a counterfactual may well be true of virtually every term
of mine, so that Fodors condition would be satised by none of them; but a sufcient condition of meaning
that applies to none of my words is not very interesting.
There is this answer. It is in the spirit of the theory to require that F have the cognitive-functional role
characteristic of assenting predication, assuming that that is accountable for in wholly non-semantic terms.
More specically, to require that the relevant nomic relation hold between P and ascriptions of F that have
that functional role and that any ascriptions with that functional role should depend asymmetrically on
that nomic relation. Perhaps we should not object to appealing to non-semantic functional conditions at
this level of generality. But the more specic the functional role, the more dubious it is that a non-semantic
or non-intentional specication of that role is waiting in the wings.
which maples caused me to judge maple; there were no such cases. Not that it was not
in my power to use maple in my thoughts to refer to sweet gums. Social meanings do
not deprive me of semantic autonomy when I insist on it. But for many predicates
I waive autonomy for the sake of drawing on social reference, and thereby risk a certain
kind of falsehood. I pass by the question why socially deferential concepts are desir-
able,4 and simply take for granted that they are central in our thinking.5
Socially deferential concepts make vivid the need to be specic about which
concepts are candidates for tting Fodors condition: they must acquire reference
as a function of non-socially mediated discriminations of an individuals ascriptions.
4
I suggest reasons in Personal References (1991b). I also argue there that the social determination of
the reference of such concepts cannot be explained away by construing them as metalinguistic.
5
Given the agility with which Fodor adapts his theory to various cases (both in (1987) and in (1990b), it
is worth asking whether it can also be made to accommodate semantic deference. Suppose it is said that,
while maple has never in fact been caused by maples, there still is a suitable nomic relation between them
on which my ascription of maple to sweet gums has depended. For had my usage been adjusted in
response to correction, maples would have cause judgments of maple, and had that socially mediated
causal counterfactual not obtained (i.e., to the effect that had I corrected my usage maples would have
caused maple), I would not have ascribed maple to sweet gums, for I would not have ascribed it to
anything.
There are two problems with this. The rst is that my false ascriptions did not in fact depend on such a
nomic relation between me and maples. Even if we grant that were maple not to have meant maple it
would not have meant anything and in consequence would not have been uttered by me at all, still my
actual usage depended on the social fact and not on the irrelevant further matter of whether such a
counterfactual causal relation also obtained between me and maples. Many contingencies could have
broken that counterfactual connection, without disturbing the social facts or my actual ascriptions of
maple.
The second problem is that the relevant counterfactual, to the effect that maples would have
prompted judgments of maple had I accepted correction, relies implicitly on the social meaning of
maple, for accepting correction means changing to a usage that conforms to the social meaning. It is
hard to see how that counterfactual causal connection between maples and my uttering maple could be
expressed without appealing in the antecedent to something semantic. There are many different causal
counterfactuals that relate my utterance of maple to objects of all sorts, depending on what adjusting
condition is envisaged in the antecedent. And it is difcult to see how one could say that only the
socially mediated counterfactual corresponds to the real nomic relation between maples and maple.
Fodor directly faces up to objections of this sort in other cases, maintaining that in pointing to such a nomic
relation the semantic property is appealed to eliminably, and that the nomic relation is in itself nonsemantic.
Now I do not doubt that in some sense the underlying causal mechanisms involved in adjusting ones usage in
response to correction could ultimately be accounted for in nonsemantic terms. But it would not follow that
the semantic property is there appealed to eliminably in the sense Fodor apparently requires. His project is not
just to assert that semantic properties need not be mentioned in fundamental causal explanations, but rather
to naturalize intentionality. And by this Fodor appears to mean something like providing an armchair
condition that sufces for intentionality, or, better, providing an armchair condition that relies on further
semantic notions only if (a) they are ancillary, that is, do not do the main explanatory work, and (b) it is not
implausible to suppose that they are in turn susceptible of armchair explication. Now, given that we are taking
the armchair naturalizing of semantic notions as the main point here, it is far from obvious that social-
semantic notions are in fact thus susceptible. But, more importantly in the current case, it seems that for
socially deferential concepts the central referential mechanism is precisely social, so that the notion of
correction in the above counterfactual is hardly ancillary. It presupposes what is essential to the reference
of my socially deferential concepts. We cannot naturalize intentionality in the armchair by presupposing
social-semantic reference.
A natural idea is that they are a kind of demonstrative, of the form thing of that sort
(which I see here and there). While Fodor speaks not of demonstratives but of
common nouns like horse and even proton (which he takes on with admirable
sang-froid), perhaps those words may sometimes serve to express demonstrative
conceptions. But consider what he says about proton. Suppose a person ascribes it in
response to visible traces of protons, depending on theoretical beliefs to the effect that
these traces are caused by protons in such-and-such physical ways. According to
Fodor, proton can come to refer to protons by virtue of such a causal connection (he
turns aside the objection that this presupposes the intentionality of the theoretical
beliefs).6 But as we described the case, proton does not express a demonstrative that
acquires its reference from those discriminations, as becomes clear by comparing
another case. You observe the same visible phenomenon and with no theory of protons
in mind you form the conception that sort of thing, the conception of that, whatever it is.
If you succeed in referring to an underlying property or kind, the referential link is in a
general way demonstrative.7 But that is not what happens in Fodors case, where
proton is already embedded in a background physical theory and has a cognitive
role independent of and presupposed by its employment in the current discrimin-
ation of traces, which apparently makes it not a demonstrative whose reference
derives from that discrimination.8 So to take Fodor as accounting just for demon-
strative concepts does not t what he says about this case.9 But they may yet be the
best candidates for his theory and he probably should not want it to explain
reference for the proton in that case.10 Indeed, comparing the two proton
examples strongly suggests that it is demonstrative concepts that individualist
information-based theories, Fodors included, should focus on, a more modest
objective than usual.
It is clear enough to me that most of the predicative concepts I express using
common nouns, water, horse, and even red, are not demonstratives whose
reference in my thoughts to external kinds is determined by my discriminations.11
But it also seems to me that we do have such demonstrative concepts, recognitional
6
Fodor (1987), pp. 1212.
7
Demonstratives can be indirect, as when you point toward a bulge in the curtain and say thats Rover.
8
This does not rule it out that proton is a demonstrative that gets its reference independently,
although of course that would be unusual.
9
Unless I misunderstand his discussion and he does mean the case to be taken in the straight
demonstrative way.
10
Suppose your theory that those traces are caused by protons is false; they are caused by positrons.
This is possible because, as mentioned in the text, the conceptual role of your proton presupposes that it
has reference independent of your current discriminations. Then, if Fodor is successful in bending the
example in the text to his condition, it would follow that in the present example proton refers to positrons.
11
I am puzzled by the facile vericationism that underlies so much anti-realist and realist discussion,
and also many discussions of Kripkes Wittgenstein. My dispositions in verifying the applicability of a
predicate can hardly determine reference for my socially deferential concepts, which is to say, the bulk of
the concepts I express using English predicates.
concepts as we might call them.12 If you go to Mississippi for the rst time, you may
learn to discriminate new natural kinds, initially thinking of them as this kind of
bird, that sort of vine; and you may invent names to express those demonstrative
concepts. When you then learn a standard name, say for the kudzu vine, it could have
a double role in your thoughts, in its marking your recognitional concept and in
meaning kudzu properly so called, concepts you implicitly hold coextensive. And
nothing prevents complete expropriation of kudzu, so that it expresses only a
personal recognitional concept (vine of that sort) for you, regarding others using
it to refer to the same kind as merely incidental. This would not be the usual thing,
but it is up to you.13
More will be said in characterizing personal recognitional concepts and answering
doubts even that there are such things, but we perhaps have enough to go on (pro
tem). If any of our concepts t Fodors account of reference, such recognitional
concepts are prime candidates, and the question is whether Fodors condition is
sufcient for reference for them. And although he has not proposed a generally
necessary condition of reference, if his condition were not satised by these concepts
(because, e.g., the asymmetrical dependence condition is too strong), then, given no
other plausible candidates, we might conclude that Fodors theory does not explain
intentionality for any concepts of ours.
III
Problems for Sufciency: Deferential Concepts
The existence of socially deferential concepts raises the following difculty for any
information-based theory of reference that purports to give a sufcient condition free
of intentional notions. Suppose my concept horse is deferential, that I implicitly take
its reference to be determined by the language I speak, that I intend to refer in my
thinking to whatever horse refers to in that language. But suppose I take horse to
apply both to the usual creatures and to (what are in fact) gazelles, that is, I take
its extension to be, as others would say, horses or gazelles, so that my ascribing
horse to gazelles does not asymmetrically depend on my ascribing it to horses.
Then if Fodors condition (or certain other information-theoretic conditions)
were sufcient for reference it would follow that my concept horse in fact refers
to both horses and gazelles. But given the deferential status of my concept that
is not so.
The most straightforward strengthening would require that a concept be undefer-
ential and demonstrative in the manner of recognitional concepts, that it have, as we
12
I have given a somewhat detailed treatment of recognitional concepts in Personal References
(Loar (1991b)).
13
A sign of this would be your not being disposed to accept semantic correction.
might say, an undeferential conceptual role. The question is how to capture that. One
might attempt it negatively: the concept is not deferential, not one whose ascriptions
one would withdraw on the grounds that (i.e., because one believed that) one had
mistaken its meaning. Or positively: the concept is of things of that perceived or
remembered type, i.e., has that intuitively indicated intentional property. In neither
case is it obvious how to de-intentionalize these characterizations of undeferential
concepts, and it seems we must require a concept to be undeferential if an individu-
alist information-based condition is to determine its reference.14
Fodor has a general line of reply to any complaint that some intentional property is
presupposed: look to the nomic relation (roughly the counterfactual) between predi-
cate and referred-to property, the one that false ascriptions asymmetrically depend
on, and do not worry about whether the language in which we are inclined to
describe the connection between predicate and property happens to be intentional;
what matters is the nomic relation. But to apply this to the present question requires
showing that ascribing the deferential horse to gazelles asymmetrically depends on
the concepts nomic relation to horses. Suppose it is said that were ones mistake
corrected one would no longer call gazelles horse (except when mistaking them for
horses), and hence that ones ascriptions are asymmetrically dependent on the nomic
relation of horse to horses. But this would not be correct. Ascriptions of horse to
gazelles would be asymmetrically dependent on the ascriptions of horse to horses,
and this given the satisfaction of a certain condition, a condition that involves
apparently ineliminable intentionality (accepting correction, etc.). It does not follow
that ascribing horse to gazelles in fact asymmetrically depends on ascribing horse
to horses. In the case described, ascriptions of horse to horses and to gazelles are
symmetrically dependent, even though, because of semantic deference, horse is not
true of gazelles.
Questions about Sufciency: Guiding Conceptions

Consider Armand, a novice birdwatcher who learns that the birds he has been
identifying under a certain recognitional concept are of two not especially closely
related kinds. There are three potential upshots. (i) He would regard his past usage as
having lacked a determinate reference (I meant to be referring to a single kind of
bird). (ii) He would now regard his past usage as referring to what has turned out to
be a disjunctive kind. (iii) He would regard the information as irrelevant to his
14
Suppose there were undeferential concepts that were not recognitional concepts but theoretical
concepts, i.e., they get their reference somehow from their role in a systematic personal theory of things.
They could be ascribed on the basis of perception, but they would not get their reference from such
ascriptions. (A personal analogue of how I took proton to behave in Fodors example.) But that means that
such a concept could in an information-theoretic or a Fodorian way discriminate a property P, even though
P is not its reference. This generates a counterexample to sufciency that does not depend on the existence
of deferential concepts. This seems to show that a sufcient condition requires a positive characterization
of recognitional concepts, and as I indicate in the text it is not easy to see how that can be done except in
intentional terms, in armchair theorizing, that is.
reference, which was meant to be determined by a resemblance in conguration, one

that further kinds may well share, and which we may suppose in fact to be objective
even if supercial. (Of course it is also possible that he did not conceive his reference
in any such specic way.)
A natural account of these cases is that, because Armand has in each conceived his
reference differentlyas a natural kind (a species or something on that order), as a
kind with a principled but not necessarily monolithic basis, or as a congurational
kindhis reference in each differs or may differ from the others. Given the facts, it
seems reasonable to say that in case (i) Armands conception has failed of reference,
that it has picked out no property, that, as Armand would judge, there is no such
thing as that kind of bird; that in (ii) it referred to the disjunctive kind; and that in
(iii) it referred to the congurational kind. (The last two differ in reference if
unobserved instances of the disjunctive kind differ in appearance from the observed
ones or if other unobserved kinds have their appearance.)
Fodors condition apparently does not distinguish these cases, but implies that
reference in each is to the disjunctive kind (or to the congurational kind?), despite
the difference in guiding conceptions. Suppose it is insisted that in case (i) Armand
has simply misconceived his reference, which is, say, the disjunctive kind. To
insist on this seems theory-driven, as compared with agreeing with Armand that
there is no such thing as that kind of bird. And why should ones general
conception of the reference of a recognitional predicate not constrain it?15 But
then Fodors condition requires an intentional supplement, the satisfaction of
guiding conceptions. One virtually never refers demonstratively, whether singu-
larly or predicatively, without some at least general conception that constrains
ones reference, and this is a problem for any causal condition that purports to
sufce for reference.
A possible reply to this objection is that the Fodor condition, understood properly,
does not after all imply that Armands predicative concept in case (i) refers to the
disjunctive property, and that in fact it refers to no property, implying this entirely on
causal grounds and without appeal to the intentional features of guiding conceptions.
It was always counterfactually true that, were Armand to discover the real state of
things, he would withdraw that recognitional concept from active predicative use;
and so, it might be said, that relevant nomic relation does not relate it to any
15
If a singular memory demonstrative, that person, derives equally from perceptions of two persons
whom one has conated, the natural call seems to be that the demonstrative fails of reference. On the other
hand, if one lived in a world of twins and so had implicitly been aware that two people might be involved,
ones guiding conception might implicitly have been that person or persons, which in some ways would be
analogous to the disjunctive-kind guiding conception. A proposed sufcient condition of reference for
singular demonstratives that failed to take such guiding conceptions into account would be open to
counterexample. The causal condition on its own is not enough.
But of course, not everything that looks like a guiding conception is a constraint on reference; some
predicates associated with a demonstrative may well be better construed as registering judgments about its
references rather than as constraints on it.
property. The problem of course is that this counterfactual depends on an intentional

condition, viz., Armands discovering the real state of things. To point beyond that
intentional counterfactual to the relevant nomic relation is to rely on a ction. There
is no more reason to regard a counterfactual that depends on Armands forming a
correct belief than one that depends on his forming an incorrect belief (which is a
quite ordinary possibility) as indicating the real nomic relation.
There is another problem. Suppose Armand would never have noticed rhododen-
drons had he seen only small ones, but that certain grand ones around the neigh-
borhood catch his eye, so that he forms the conception plant of that kind. Only then
does he notice the occasional small one, and they bring him to judge plant of that
kind only because the large ones do; still he would produce that concept in response
to the large ones even were he not to notice the small ones. This is not a diachronic
asymmetry; it is true of Armand at t that were large rhododendrons not causes of his
concept at t then small rhododendrons would not be causes of it at t, while the
converse does not hold. So if Fodors condition were sufcient Armands concept
would refer to the property of being a large rhododendron. (The general idea of this
comes from Steven Wagner.)16
It might be replied that the objection depends on an unsympathetic reading of the
counterfactual, which should be understood as: were it the case in the circumstances
that small ones failed to elicit the concept then it would be the case that large ones
did. Since that is true, the reply goes, the extension is the whole set of rhododendrons
by Fodors criterion. But this is not obvious. Were small ones not to have elicited the
concept, this could in the circumstances have been because Armand failed to
recognize them as of the same kind as the large ones that do elicit the concept. So
the counterfactual asymmetry would obtain and Fodors condition would incorrectly
imply that Armands concept refers only to large rhododendrons.
Now this dialogue could continue, but trading counterfactuals seems pointless. It is
difcult to accept that the concepts referring to all rhododendrons rather than to
large ones depends on a counterfactual or nomic relation that we have no clear rules
for selecting, rather than on the fact that Armand conceives his references as plant of
that kind, knowing full well that they come in all sizes. The natural account of things
makes Armands reference depend in part on that further intentional fact.
Another case. Suppose Armand, never having seen pomegranates, is presented
with papier mch ones but told they are real. He forms the conception fruit of that
kind. Shown real pomegranates he then easily distinguishes them from the others;
but he takes the real ones to be papier mch. If Fodors condition were sufcient for
reference, Armands concept fruit of that kind would be true of the papier mch
fakes, and that seems wrong, for the simple reason that the fakes do not satisfy his
16
Letter to Jerry Fodor.
guiding conception.17 Once again it is doubtful that this can be blocked by appeal to
counterfactuals free of intentional provisos. The intentional properties of guiding
conceptions play a role in determining the reference of recognitional concepts, a role
that, I surmise, cannot be eliminated in armchair theorizing. (Which is not to belittle
armchair theorizing; it takes us some distance in these matters.)
Questions about Sufciency: References outside the Head

Fodors condition appears to imply that some concepts refer to states of ones visual-
cognitive system when in fact they refer to external objects. He considers this and
denies it, but let me make the case. One is tempted to say that the fact that sheep
trigger a recognitional conception of sheep depends asymmetrically on certain retinal
states triggering the concept, and that would imply that the concept refers to those
states. Fodor would reply that no retinal-state-kind that is not open-endedly dis-
junctive will have this feature; for any concept of sheep may be sensitive to appear-
ances that are far from models of sheepliness, e.g., when the bush shakes or when you
glimpse a shepherd wielding his crook. No retinal state worth counting as a kind is
common to all sheep detections.
I see three problems with this reply. For one thing, its presupposition that all sheep
discriminations are equally reference-determining is questionable; the reference of an
ordinary recognitional concept of sheep does not appear to be determined as much
by shaking bushes as by paradigm sightings of sheep. As in the proton case, correct
ascriptions can depend on a concepts background role, its reference determined
independently of those ascriptions; this can be true just as much of demonstrative
concepts as of theoretical concepts. If a demonstrative concept refers to so-and-sos,
that stems, it seems to me, only from its sensitivity to so-and-sos in certain reference-
determining elicitings. This is rmly denied by Fodor, and I will return to the point;
but if correct it means that ascriptions of the concept to sheep may well asymmet-
rically depend on the concepts triggering by a fairly unied property of the visual
system, one common to reference-determining sightings of sheep. Then the question
is why the recognitional concept does not refer to that property.
A second point is this. It could happen that a person has a recognitional concept
of, say, tennis balls, but lacks beliefs that enable him to detect tennis balls other than
stationary ones in full view. (Suppose any attempts to do so turn out false.) Then,
quite apart from the foregoing point, there may be a unied property of the visual
system, one that typically arises from straight-on views of unmoving tennis balls, on
which his ascription of the concept to tennis balls asymmetrically depends, and
which Fodors theory then counts as the reference.
17
My point does not require the stronger claim that this concept refers to the real pomegranates; given
the oddness of the case one may not know what to say about that. The point is that it does not refer to the
fake ones.
Third, even if we were to grant that any true ascription of a recognitional concept
is as relevant to its reference as any other, who is to say there is no unied neural
property deep in the visual-cognitive processing system that is the immediate cause
of all and only ascriptions by a particular person of a particular recognitional concept
of sheep? But if there were such a property Fodors condition would count it as the
referent.
One suspects that any theory that takes reference to consist in a causal relation
unconstrained by guiding conceptions is vulnerable to the objection. And this
suggests that referring to external properties depends on, in some sense, conceiving
them as external. It should be clear that this is no objection to the idea that the
reference of a recognitional concept is determined causally, as long as that is
constrained by the concepts guiding conception, which is an intentional property.18
The last claim may seem rebuttable, by a competing explanation of reference to
external objects: the external property counts as the referent because the very point of
ascribing reference to others is to mark their propensity to register features of the
passing scene. Now this seems to me not to get to the heart of the matter. Even when
an external property is being tracked it is not always the reference: nothing prevents
me from having a recognitional conception of a kind of visual sensation, which
happens to be caused, say, by stationary tennis balls. No point in replying that this
responsiveness to tennis balls asymmetrically depends on responsiveness to tennis
ball sensations, for that is true also when tennis balls are the reference. The difference
in the two cases lies apparently in how the reference is conceived.
Ones interest in how others track external properties, moreover, presumably
depends on ones own interest in external objects. The claim that a non-intentional
condition sufces for the reference of ones own concepts is hardly vindicated if one
has to add provided ones interest in external properties is taken into account. That
seems just a backhanded way of admitting the ineliminable role of the guiding
conception external property.
18
Can there be reference without a guiding conception? Suppose I am strapped down and cannot move
my head in a strange environment, not able to tell whether what appear to be spots uttering before my
eyes are specks of dust in the air, shadows on the wall, muscae volitantes, or strange events deep in my
brain. If I am, in fact, tracking specks of dust it seems my concept things of that sort refers to them; but if
there are no suitable external things, then my concept perhaps refers to internal events. I am not sure that
this example shows reference without a guiding conception. Perhaps one has implicitly a sort of condi-
tional guiding conception: uttering things out there if there are any or otherwise internal occurrences.
As regards what I say in the next paragraph in the text, it is true that in such a case, unless our interest
focuses on internal states, we seem to want to choose the external property if there is one. But this may
simply reect the fact that we are most basically interested in external causes of our states if there are any
(see the conclusion of the next paragraph). This would mean that external object or property is a sort of
default guiding conception.
IV
Simple Predicates and Atomic Reference
Keep in mind that Fodors condition is not meant to explain reference for complex
expressions, and it may appear that, by xing on recognitional concepts that incorp-
orate guiding conceptions which have independently determined reference, one has
departed from the scope of this concern. But it has to be observed that, if such
concepts lie outside the scope of an individualist information-based theory of
predicate reference, it is doubtful that such a theory applies to concepts of ours.
Perhaps there are odd cases,19 but they will hardly be central. Moreover, a concepts
incorporating a guiding conception is no bar to its having information-based
reference. Consider singular demonstratives: they typically are qualied at least in a
general way, so that the reference of that moving object there must be a physical
thing (you are not referring to a glint of light on the lenses of your eld glasses); but of
course a causal condition determines which physical thing is referred to. So with the
reference of recognitional concepts to properties.
It is natural to think that, however dependent one may be on concepts whose
reference is socially determined, ones ability to think about the world requires at
least having the capacity to form ones own recognitional concepts.20 And it may be
natural further to think that that capacity is in some way foundational for other
referential capabilities. But this should not imply strong atomism, that the reference
of foundational information-based predicates must depend on no further intentional
backing. A recognitional concept is atomic in a weak sense; its reference is not
determined by the reference of semantically independent constituents, and it does
depend on the concepts distinctive external relations. To deny that causal theories of
reference give us strongly atomic concepts of course contradicts their promise to
provide a basis for reference in completely naturalistic terms.
V
Reference-determining Circumstances
As noted, Fodors considered view is that reference requires no causal relation to
actual so-and-sos; it sufces that unicorn stand in a suitable nomic relation to the
property unicornhood.21 But qualifying the straight causal condition for this reason
19
See footnote 18.
20
The qualication is due to this. As I try to make clear in Loar (1991a) [Chapter 11 in this volume], a
recognitional ability attached to a socially deferential concept may not in itself have the role of a separate
concept. But it presumably always could, and it is difcult to see how ordinary thinking about the world
could go on without such a capacity.
21
[The reference in the original publication was to a paper in press, probably intending to refer to
Fodor (1990b).]
seems to me misguided. Is there such a property as unicornhood? There is the

property of looking like a horse with a horn, but then if that is what unicorn
means, presumably it expresses a description, and such descriptive concepts are
not candidates for Fodors theory. If unicorn is noncomposite and purports to
refer as it were to a real essence, then (as my intuitions have it, at least) it lacks
reference;22 if composite, its reference is not determined directly by an information-
theoretic mechanism; and thus either way an information-based theory should
ignore it.
A natural view is that a recognitional concept picks out the kind of thing that in
some way actually occasions its ascription.23 But suppose certain ascriptions of F are
occasioned by things of a certain kind only given a background belief that involves F.
It seems the intuitive view, of the role of such a belief in those ascriptions, presup-
poses that F has independent reference (as appeared to be the case with proton).
And then, of course, those ascriptions are not determinative of Fs reference.24 This is
not to say that reference-determining ascriptions of a concept must not depend on
background beliefs, but rather that the ascription must not be mediated by a belief in
the manner mentioned. The role of that belief in these ascriptions presupposes that
the predicate has independent reference does not mean merely that were these
ascriptions not to have occurred the predicate would still have referred to whatever
it refers to. That, presumably, holds of recognitional concepts for which later
ascriptions can be as relevant to their reference as earlier ones, even though the
earlier ones were sufcient to determine what is in fact their reference. This can
happen because those later ascriptions are in some sense as stipulative as the earlier
ones, because if they had been occasioned by a property different from the earlier
ones (think of Armands birdwatching) the concept might have, depending on the
guiding conception, failed of reference, or referred to a disjunctive property. But on
the natural understanding of how proton occurred in the background belief, had the
particle traces and therefore those ascriptions of proton been caused by positrons,
proton would still have referred to protons.
22
Consider this: Unicornhood is the genetic property responsible for such an appearance in a possible
world closer to the actual world than any other world in which a genetic property is responsible for that
appearance. But it might then follow (who knows) that unicornhood is the property of being a kind of
mutant donkey. And it strikes me that there would be as little reason to count that as the reference of
unicorn as there is to count a fat man in a red suit at the North Pole as Santa Claus.
23
A painter may imagine a shade of blue she has never seen, and then be sure when she encounters a
new blue that it was the one she imagined. Perhaps in certain circumstances we would regard her as having
genuinely conceived that shade, that property. But this may be a very special exception to the condition in
the text; for perhaps it applies only to property concepts that t phenomenally into a spectrum of concepts
whose references are independently determined by causation by actual instances. Unicorn seems
different.
24
Analogy: I see a man wearing a kilt and carrying a bagpipe, and then see him several times
subsequently in trousers and without the bagpipe, and each time think its that guy, where the reference
goes back to the earlier sighting and is in no way determined by the later ascriptions.
We already know that not all of a persons predicates refer to properties that cause
his ascriptions of them and thus that not all predicates are recognitional concepts.
The point aimed at here, though, is that what was true of proton can be true of
recognitional concepts themselves. Suppose such a concept arises from my seeing
crows (those creatures), and that subsequently I form the belief that cawings tend to
be caused by those creatures. When later I hear cawings in the garden they cause me,
given that belief, to judge one of them. But these latter ascriptions need be in no way
determinative of the reference of my conception; if, unknown to me, the caws had
issued from my neighbors mouth, that will not have undercut the reference of the
recognitional concept (one of them) to crows. Some ascriptions of a recognitional
concept are naturally viewed as reference-determining or stipulative and others not.
But this conicts with a primary motivation for Fodors asymmetrical depend-
ence explanation of falsity. Straight information-theoretic accounts of reference
must explain falsity, according to Fodor, by distinguishing type 1 situations in
which ascriptions of a predicate are determinative of its referent and type 2
situations in which they are not and in which falsity can occur. Fodor rejects this,
on the grounds that meaning can be robust throughout situations of all kinds; there is
no kind of situation (specied non-question-beggingly) in which Fs ascription to
non-Ps ipso facto interferes with Fs continuing straightforwardly to mean P.
One will be misled on this point if one ignores the difference between deferential
and recognitional concepts. Doubtless deferential predicates are robust across situ-
ations in Fodors sense; but then the reference of my deferential sheep is not
determined by my ascriptions of it, and its general robustness in my ascriptions
stems from that fact.25 But as regards personal recognitional concepts, intuition
seems to count against across-the-board robustness. If I see novel creatures up
close, some llamas, say, and form a new demonstrative conception of those crea-
tures, that seems connected with a special perspective on them. For compare these
cases. (i) I ascribe one of those creatures to all and only llamas in front of my nose,
but then overcondently ascribe it to creatures across the eld that happen to be
sheep. (ii) As in the rst case, I ascribe it to all llamas in front of my nose but also start
applying it to sheep seen in just those circumstances. In case (i) the across-the-eld
25
A qualication: if I think sheep (in both spoken and written forms) means sound made by canaries
and other small birds then, even though it has a deferential conceptual role for me, sheep does not refer to
sheep in my thoughts. This contrasts with what I said about the above case in which a person takes it to
refer to sheep and gazelles (as we would say). If you accept my intuitions about both cases, they may suggest
something like this rule: if xs concept F is deferential and F refers to Ps in the language to which x defers,
then F refers to Ps in xs thoughts unless x has too radically misconceived the reference of F. From this we
should expect broad borderline areas of unclarity as to whether a predicate has in xs thoughts the reference
it has in the language to which x defers. Moreover, we might expect that what counts as too radical a
misconception varies across types of concept. Thus a technical concept may refer in xs thoughts to its
social reference despite a fairly substantial misconception of that reference by x, whereas less technical
everyday concepts may tolerate a lesser degree of misconception, that is without nding a difference in
reference. If this is a fair intuition, I believe one can give a satisfactory account of it.
ascriptions seem perfectly consistent with taking the concept to refer to llamas and
therefore with taking those ascriptions to be false. In case (ii), one is inclined not to
say that the concept refers (just) to llamas. And this suggests that, for recognitional
concepts, some perspectives are reference-determining, so that their reference is not
robust through all situations.
Now Fodor may wish to say that his theory also accounts for the intuition, while
explaining it differently: the across-the-eld ascriptions asymmetrically depend on
the up-close ascriptions, and the ascriptions to sheep and to llamas are neither
asymmetrically dependent on the other. Let me note that I propose the example
not as an objection to the asymmetric dependence condition but rather to show that
the distinction between reference-determining and non-reference-determining per-
spectives is an intuitive one. But, of course, it is also of interest to consider whether
the asymmetric dependence mechanism does in fact explain that intuition. The
difculties raised so far have concerned the sufciency of Fodors condition. And
that leaves open the possibility that, while not sufcient, it plays a role in explaining
falsity for recognitional concepts, in the following departure from an echt Fodorian
formulation. Suppose Q captures the guiding conception of a recognitional concept F:
then we might say that F refers to Ps provided that Ps are Q, Ps cause F, and if any
other Qs cause F that is asymmetrically dependent on Ps causing F. (Then false
ascriptions of F would be either to non-Qs or to non-Ps whose causing F is
asymmetrically dependent on Ps causing F.)
Now I do not think this is the best explanation of falsity for recognitional concepts.
The giant rhododendron example was earlier used to raise a problem for sufciency;
but it raises one also for necessity. In that example Armands concept referred to
rhododendrons in general, while its ascription to non-giant ones was asymmetrically
dependent on its ascription to giant ones. But then the condition is too strong to
account for the reference of recognitional concepts in general, and that calls into
question its explaining reference and falsity in any case.
Let us return to the main point, the viability of a perspectival constraint. In my
view it is quite a good explanation of what is going on in the llama case to say that it is
how one conceives creatures of that kind which makes the perspective in question
determinative of its reference and makes its causal relations apart from that perspec-
tive not determinative of its reference. This is a somewhat weaker claim than the one
Fodor rejected, which was that there is a certain kind of situation that for all concepts
is reference-determining (although the weaker claim is doubtless no more palatable
to him). On my view each recognitional concept has a sort of built-in perspective.
Suppose that in Kenya you see animals with a distinctive distant appearance wan-
dering across the valley, and you cannot tell how they would look up close. You form
the conception creatures of that kind. You also see nearby gazelles of a certain
species, and form the conception creatures of this kind. If you then come to think
they are the same, it yet seems natural to say that the reference of the rst conception
is determined by the distant sightings and not by the proximal sightings, and vice
versa for the reference of the second. This accords with cases that similarly involve
pairs of singular perceptual demonstratives and/or memory demonstratives.
The point has two interesting implications. First, it would undercut a motivation
for the asymmetric dependence condition if that stems from the thought that one
cannot explain falsity by distinguishing reference-determining and non-reference-
determining circumstances. (A recognitional conception of those creatures seen
across the valley is falsely ascribed to creatures seen up close if they are not of the
same kind as the creatures seen across the valley. More on this shortly.) And second,
it points to another way unanalyzed intentionality crops up. The perspectival slant of
a recognitional concept seems part of the concept itself; it points to its reference from
an at least implicitly conceived perspective. This suggests a somewhat Fregean view
of demonstratives; for both singular demonstratives and recognitional concepts, a
demonstrative mode of presentation is an intentional feature and it is perspectival.
Some perspective is part of the sense of the concept, of how it conceives its reference.
VI
False Ascriptions of Recognitional Concepts
So far we have seen the following points about falsity. First, granted that deferential
concepts such as ones ordinary concept horse are robust, that is, can be false through all
situations, this may not apply to recognitional concepts. Second, the asymmetric depend-
ence condition does not explain falsity for recognitional concepts. False ascriptions of the
rhododendron concept in the above example are not identical with those that are
asymmetrically dependent on ascriptions that determine the concepts reference; for its
ascription to small rhododendrons can be asymmetrically dependent on ascriptions to
giant ones without being false. Third, the perspectivalness of recognitional concepts
allows for the falsity of certain ascriptions, to certain objects outside the relevant
perspective. (In many cases false ascriptions are indeed asymmetrically dependent
on reference-determining ascriptions. This I surmise accounts for what is intuitive in
Fodors condition, even though if I am right that condition is not the essential explanation.)
Consider again the question whether a recognitional concept can be robust even in
situations I would count as reference-determining for it. A negative answer was
suggested in connection with the llama example above, but there is more to be said.
Suppose a recognitional conception deriving from llama-sightings is subsequently
ascribed to a few alpacas from the same perspective. It seems that these ascriptions
could well count as false, that the recently observed creatures are not of the demon-
strated kind. And does this not show that a recognitional concept is robust even
through a situation that according to me is reference-determining for it?
Let us distinguish two closely related sorts of demonstrative predicative concepts:
recognitional concepts, and what we may call pure type-memory demonstratives.
A concept of the former kind typically points backwards and forwards, by involving
both the memory of a perceived kind and a stipulative disposition to include new
things in its extension.26 This happens when there is no reason to distinguish the role
of current dispositions and past identications in stipulatively xing the concepts
reference. (Of course this makes the reference conditions of such concepts potentially
unstable; for such a concept to make a contribution to truth conditions, things
beyond the thinkers control have to be just right. As with singular demonstratives,
nothing in the concept itself guarantees that it contributes anything to, as it were, the
possible world truth conditions of the thoughts in which it occurs.) Here is an
example of a pure type-memory demonstrative: you remember a kind of bird from
childhood but count no current ascriptions of your conception as stipulative of what
you mean by that kind of bird, so that the demonstrative points exclusively in
memory. This is a fossilized recognitional concept, and the possibility of a false
current ascription of it requires no special explanation.
Now the question is whether a recognitional concept can be falsely ascribed from
any perspective whatever. My answer is, despite what has been said so far, yes in a
manner of speaking. Suppose you have a recognitional conception of a kind of bird, a
stipulatively active concept and not a fossilized pure type-memory demonstrative, and
suppose that from the relevant perspective you nod one of those at a bird of a different
kind. It may yet be useful to treat this concept as a pure type-memory demonstrative
and thus to count the present ascription as false. (Suppose you conceive your concepts
intended reference as a unied kind. We might then say false in order to register that
your current ascription deviates from a hitherto unied string of identications. But
this will make sense for only the odd exception; larger deviations lead to referential
breakdowns of the sort already encountered in the rst Armand case.) So we may have
it both ways: a recognitional concepts ascription can be from a reference-determining
perspective, and yet be treated for certain purposes as false, if we bracket the concepts
ongoing stipulativeness and attend solely to what kind it memorializes. This gives the
appearance of robustness throughout all perspectives, but it is compatible with a
recognitional concepts being perspectival, that is, being such that only ascriptions
made from a certain perspective are reference-determining.27
26
See Loar (1991b) for more on the distinction between pure type-memory demonstratives and
ordinary recognitional concepts.
27
A question remains. Suppose that, were I removed to a distant planet, I would ascribe to some non-
gazelles (same appearance, different genetics) a concept that apparently refers only to gazelles. Should we then
say that the concept refers to a disjunctive kind, or even that it fails of reference, as in the rst Armand case?
What Fodor says about Twin-Earth cases implies that his theory counts only gazelles as the reference, on
the grounds that the nomic relation of that concept to any non-gazelles depends asymmetrically on its
relation to gazelles, and this because of the discriminations one would make were one to discover how to
tell those other creatures from gazelles. But as I indicated earlier, this is unsatisfactory, for, in determining
these nomic relations, nothing that is not question-begging gives a privileged role to counterfactuals about
how one would discriminate things given correct information, over related counterfactuals given incorrect
information.
No point in invoking the perspectival condition, for my distant discriminations of those non-gazelles
would be from the same psychological perspective. We need a contextual factor, and the question is
In calling attention to recognitional concepts as the most promising predicative

candidate for an information-based account of reference, and in explaining falsity
without asymmetric dependence, I have made essential use of a number of inten-
tional notions. These I take to be ineliminable in any armchair explication of
reference, and that seems in direct conict with Jerry Fodors objective of naturaliz-
ing intentionally, at least if naturalizing means philosophical explication.
But a lesson learned from concepts of phenomenal qualities is that it does not
follow, from the unavailability of a naturalistic philosophical explication of a notion,
that the notion does not fully have application by virtue of entirely natural features of
its reference: nothing in the unanalyzability of phenomenal concepts implies that
phenomenal qualities are not physical-functional properties.28 Similarly with certain
notions of intentionality, as I argue in another paper.29 The intentional notions
invoked in the present paper, of specic guiding conceptions, of conceiving a
reference as an external object, and so on, should be taken as reexive or subjective
notions, arising not within an objective or impersonal framework of description and
explanation, but from subjective reection on how our own thoughts appear to us.
The third-person use of such intentional notions can be regarded as a sort of projec-
tion. The lack of an explication, as I argue there, is compatible with such subjective
intentional (second-order) conceptions discriminating thoroughly physical-functional
properties.30
The resulting idea of what determines the reference of recognitional concepts, as
regards armchair explication, is: a mixture of subjective and objective factors. The
former include conceiving of those things as external objects, conceiving of those
things as viewed from this perspective, and so on. The latter include those specied
causal/counterfactual relations that make a certain property the one being discrim-
inated and hence referred to. This combination of subjective and objective factors is
unsurprising, once one gives up the idea that naturalism requires that intentionality
be explicated in objective and externalist terms. Intuitively, the reference of a
demonstrative is determined both by ones conception and by an extrinsic relation;
and how a demonstrative, whether singular or predicative, conceives its reference
determines, at least from a subjective perspective, which objects or properties are
whether we have to fall back on relativizing: the reference of concept C is P relative to context K. There is a
difference between that and holding that the concepts reference to gazelles is determined by its context.
(Compare saying the reference of my utterance of I is relative to context (false) with the reference of my
utterance of I is determined by context (true).) My inclination is to say that the concept may well refer
just to gazelles despite the counterfactual in question. And I suggest that this calls for yet a further
intentional element: I conceive the reference of the recognitional concept as (say) a natural kind in my
present context (vaguely drawn).
28 29
See Loar (1990) [Chapter 10 in this volume]. Loar (1987b) [Chapter 9 in this volume].
30
As if our second-order conceptions of our rst-order conceptions are recognitional concepts, the
former discriminating those internal properties that constitute the latter. How this might account for the
appearance of intentionality is discussed in Loar (1987b) [Chapter 9 in this volume].
candidates for reference, and which causal or discriminative relation selects the
reference from among them.
These abbreviated comments about subjective intentionality are hardly satisfac-
tory in themselves. And they are not essential to the basic points of this paper.
Suppose it is correct that no purely causal or counterfactual or nomic conditions can
explain the reference of recognitional concepts, that is, without the help of inten-
tional notions. And suppose those notions of intentional properties are not them-
selves explicable in naturalistic terms, and that no extrinsic naturalistic constraint
secures that they pick out a determinate reference relation. Then we should have to
conclude that the reference of recognitional concepts cannot be wholly a matter of
objective naturalistic factors.
12
Elimination versus Non-reductive
Physicalism
There is reason to be pessimistic about the physicalist reduction of the mental,

especially as regards propositional attitudes. But the straightforward alternatives
are unappealing, namely, dualism and the eliminativist proposition that mental
ascriptions are never strictly true. The question is whether some form of non-
reductive physicalism is a more promising alternative, that is, commitment both to
the comprehensively physical nature of things and despite non-reduction to the
reality of mental facts. I am going to argue that current versions do not stand up.
But the point is not that reduction or elimination must then hold (putting aside
dualism as incredible). It is quite consistent with my argument here that we reach an
impasse, that standard presuppositions permit no viable position. At least there is no
easy non-reductive way out, despite the claims of two distinctive theses: Stephen
Schiffers to the effect that nominalism about mental properties permits physicalism
without reduction; and the widely prevalent doctrine of non-reductive supervenience.
Non-reductive physicalist views reject an atomist picture of how physical facts make
mental ascriptions true, namely by correspondence: the mental ascription must
correspond to a physical fact in such a way that the ascribed mental predicate stands
for a physical-property constituent of the truth-making fact. Schiffer rejects this by
rejecting properties, and supervenience views reject it by making the relation between
ascription and the physical facts more oblique. The argument of this paper is in effect
that the correspondence picture is not so easy to get rid of.
1 Nominalism and Ontological Physicalism

In Remnants of Meaning and in a recent paper1 Schiffer denies that rejecting
reduction forces a choice between a non-natural view of thoughts and their elimin-
ation. His argument is direct and ingenious. Mental properties are irreducible. On the
face of it, however, mental properties are causes of behavior. But if mental properties
are not reducible to physical or functional properties, then we have an unbelievable
1
Schiffer (1987), ch. 6, and Schiffer (1990).
ELIMINATION VERSUS NON - REDUCTIVE PHYSICALISM
causal overdetermination of behavior by mental properties and physical properties.

So mental properties must be reducible, and hence, given the failure of reduction,
both reducible and irreducible. Now that implies that there are no such properties as
mental properties. But this does not mean that there are no true ascriptions of mental
states. For Schiffer denies that the truth of statements of any sort, mental or physical,
requires the existence of properties; he subscribes to thoroughgoing nominalism.
Schiffer calls his anti-reductionism sentential dualism: there are true mental ascriptions
and true physical statements, and the former are not reducible to the latter. But this
dualism is innocuous. For it is quite compatible with (what he calls) ontological
physicalism: all actual entitiesparticulars, properties, relationsare physical. Given
nominalism, ontological physicalism is true because all particulars are physical.
Schiffer allows that we may speak of mental properties and facts, in what he calls a
pleonastic sense of property and fact. Why deny the trivial inference from Alfredo
believes in ghosts to Alfredo has the property of believing in ghosts? What makes this
ontologically neutral is that, on one understanding, that property is a sort of ghost
itself, not introduced by ontologically committed objectual quantication. Schiffer
speaks of pleonastic properties as not language-independent, as though they are
introduced by merely substitutional quantication. He hesitates about this, however,
and so let us say merely that pleonastic properties are introduced by non-objectual
quantication. The important point is that, according to Schiffer, no awkward causal
overdetermination arises from merely pleonastic mental properties (a central point, to
which we will return). Nominalism then makes sentential dualism, anti-reductionism,
harmless by making it compatible with ontological physicalism.
Schiffers reason for rejecting ontological dualismin particular, the existence of
non-physical non-pleonastic mental propertiesis that it implies an unbelievable
overdetermination of behavior by both physical and non-physical causes. As to why
his own sentential dualism does not imply such an overdetermination, he says
roughly this. We may suppose that the conceptual roles of my mental ascriptions to
x are related non-accidentally to xs underlying physical workings in such a way that
predictions of xs behavior based on my mental ascriptions mesh with the output of
xs physical workings. Given this, he says, there is nothing like causal overdetermination,
but rather a nice t between a predictive technique and a system of physical
causation. Nothing in that picture, even with the failure of reduction, warrants
overturning our entrenched view that mental ascriptions are or can be literally true.
Before we examine Schiffers proposal it will be useful to have in mind some
conditions that any interesting non-reductionist physicalism ought to meet. I take
the following to be fair.
1. Mental truths should be not second string, not assigned, for example, an instru-
mentalist status while physical truths are taken more substantively. It is a kind of
eliminativism to assign mental ascriptions an etiolated truth basis, making them
true in a way secondary to the truth of physical ascriptions. A risk for non-reductive
physicalism is that the non-competing or non-dualist role it assigns to mental

ascriptions is so thin that we have a sort of genteel eliminativism.
2. If a non-reductionist physicalism is to be interesting, both mental and physical
truths should be counted as true in a realist sense, that is, independently of
ascribers verications, and not consisting just in successful systematizing.
Evidently if physical truths are taken in a realist sense and mental truths
some other way, the latter are second string; and the reverse asymmetry
would not be physicalist. Suppose we have across-the-board anti-realism.
Then, faced with the failure of reduction, but given the ideal instrumental
success of mental explanation, we might simply acknowledge two instrumen-
tally successful systemsthe physical sciences and ordinary psychological
explanationwith each constituting facts at its own level of description.
Naturally anti-realism does not rule out motivations for seeking unied science,
and if the mental does not t in with the rest of science that would count as a
signicant discovery. But suppose the mental does not t in, that there is no
reduction. Given anti-realism it is difcult to see this as threatening a prob-
lematic metaphysical dualism that requires metaphysical countermeasures, say
nominalism or non-reductive supervenience. Across-the-board anti-realism
could even be seen as a non-reductive dissolution of the mind-body problem,
removing a picture that is problematically dualist only given realism. But it
would not be a physicalist solution, and it would make physicalist proposals
such as nominalist ontological physicalism and non-reductive supervenience
beside the point. So a realist sense of both mental and physical truths is essential
to motivating non-reductive physicalism.
3. A non-reductive position is genuinely physicalist only if mental properties are
allowed no independent causal powers; otherwise we have a kind of causal
dualism. A sort of minimal physicalism may be characterized in a non-arbitrary,
non-question-begging sense thus: no non-physical properties or resemblances
make a causal difference. The matter is complicated by epiphenomenalism:
epiphenomenalist mental resemblances are intuitively as incompatible with a
well-motivated physicalism as causally potent irreducible mental resemblances.
(Why should impotence make you metaphysically unobjectionable?) Of course
the condition of no independent causal powers is intended merely as a necessary
condition of a successful non-reductionist physicalism. But it can be strengthened.
Count a property or resemblance as epiphenomenal in an objectionable sense if,
while it does not make a causal difference, there is nothing incoherent in the idea
of its doing so, or, a somewhat different idea, it could (does in another possible
world) make a causal difference. So unreduced mental resemblances meet our
strengthened condition only if it makes no sense to regard them as having
independent causal powers. The delicate question then is how and whether
nominalism, or non-reductive supervenience, permits this condition to be met
without making mental resemblances second-string affairs.
4. Objective mental resemblancesthe realist satisfaction conditions of mental

predicatesmust be naturalized, that is, shown to belong to the natural order,
and a non-reductive sense must be given to this. For the conjunction of
non-reduction with realism about those satisfaction conditions is prima facie
tantamount to classical property dualism, and the burden apparently lies with
the non-reductive physicalist to counter this perception and to naturalize
the mental.
Condition (4) may well seem necessary and sufcient for blocking problematic
dualism, so that (3) need not be mentioned as a distinct constraint, on the grounds
that if the satisfaction conditions of mental predicates are part of the natural order
their causal powers are natural and hence not independent. But it is worth stating (3)
separately, for these reasons. (a) There may be a temptation to assert supervenience,
thus purporting to meet condition (4), but then to tell a story about the explanatory
role of the mental (see section 3 on autonomous explanation) that despite the
assertion of supervenience seems to endow mental properties with independent
causal explanatory powers. (b) Suppose one is skeptical of supervenience because
of its reliance on modal notions. Then denying problematic dualism may consist in
denying that mental properties have or could have independent causal powers.
(c) Schiffer addresses what is in effect condition (3) as the main issue, putting the
naturalizing of mental predicates somewhat on the margin of discussion (its most
salient form being non-reductive supervenience).
As we noted earlier, the principal motivation for Schiffers nominalist ontological
physicalist position is that unreduced mental properties, in the Platonistic sense,
generate an unbelievable causal overdetermination of behavior. These properties are
objects . . . that have causal powers, but ones that are utterly superuous, adding
nothing to the causal powers of physical properties that sufce to govern physical
events.2 The rst question I wish to pose about Schiffers position is whether
nominalismthe elimination of Platonistic mental propertiesblocks causal over-
determination in the sense that violates condition (3), given that there is causal
overdetermination without nominalism.
Objective resemblances, or realist satisfaction conditions, apparently are easily
generated from Schiffers pleonastic properties. For any pleonastic mental property
M and any person x, we can say that it is verication-independent whether x is M,
that is, that xs being M is not constituted by its veriability by us. And if M is
univocally true of x and of y, we then ought to be able to speak, again in pleonastic
mode, of its being verication-independent that a certain objective mental resem-
blance, being M, holds between x and y. So if we have pleonastic properties for the
taking, we also have a non-Platonistic sense of objective or realist resemblances.
2
Schiffer (1990), p. 158.
There are then two senses of realism about properties: (a) quantication into
predicate positions makes sense objectually; and (b) the question whether a given
object has a certain perhaps pleonastic properly is verication-independently true or
false. In denying realism about properties in the Platonistic sense Schiffer is not on
the surface of his words denying realism in the latter sense.
Now it could be said that realism about properties in the second sense amounts to
Platonism. But if nominalism is not compatible with objective mental resemblances a
non-reductive physicalist position that depends on it fails to meet condition (1). If on
the other hand they are compatible, overdetermination looms. Do irreducible object-
ive mental resemblances (pleonastic sense) make an independent causal difference? It
can hardly be said that it is only on a Platonistic view of properties that the question
of making an independent causal difference arises. For what does that leave intact of
the causal role of physical resemblances? If we say the difference is that in the case of
the physical there are non-pleonastic physical properties that do the causing, then we
not only depart from Schiffers fully nominalist account, but also give mental
ascriptions a second-string status, with the noted obliquely eliminative upshot.
If mental resemblances are there independent of our ascriptions and verications,
then, however pleonastic our conception of them, they are out there, they are not
physical, and they causally matter. Prima facie this looks like overdetermination in the
problematic sense. This is not to deny that the appearance can be deected, perhaps by
some account of supervenient causation, on which more below. The point is simply
that if there is a prima facie problem of overdetermination without nominalism, then
there is one with it, provided of course that nominalism is compatible with a realist
view of mental resemblances and does not cast them in a second-string role.
But there is another way in which nominalism may be intended by Schiffer to
block overdetermination. At points he adopts somewhat deationary language about
mental explanation, that is by comparison with physical explanation, and he takes
nominalism to motivate that deationary language. In a nominalistic world, Schiffer
writes, there are no nonpleonastic propositional-attitude properties but only our
cognitive and linguistic propositional-attitude practices. He goes on to say and they
[those practices] are anything but superuous. Just think of all the ways wed be worse
off without them.3 To repeat my earlier characterization of his view, the conceptual
roles of my mental ascriptions to x are related non-accidentally to xs underlying
physical workings in such a way that predictions of xs behavior based on my mental
ascriptions mesh with the output of xs physical workings. Nominalism motivates this,
it seems, because if mental predicates do not stand for Platonistic properties then all
that remains is their conceptual role. But that strongly suggests that mental causal
explanation is entirely a matter of the conceptual role of the ascribers predications,
which on the face of it is not compatible with realism about the mental.
3
Schiffer (1990), p. 158.
We have on our hands not just the conceptual roles of mental predicates, but also
verication-independent mental resemblances (pleonastic sense). Even if the
required quantication is merely non-objectual, the verication-independence
of mental resemblances gives them a solid standing in fact apart from the concep-
tual roles of their ascriptions. And the question is why that does not mean:
(a) verication-independent mental resemblances (satisfaction conditions, pleon-
astic properties); (b) verication-independent physical resemblances; (c) given
common sense, causal roles for each; and (d) given non-reduction, independent
causal roles for mental resemblances, that is, short of some adjustment apart from
nominalism. This is the prima facie situation.
2 Non-reductive Supervenience
A non-reductionist physicalist view cannot avoid endorsing the supervenience of the
mental on the physical, even given nominalism. This is physicalism at its minimum:
with or without reduction, mental resemblances cannot vary independently of
physical facts. But ontological physicalism and nominalism do not entail super-
venience, and so it must be an added component of Schiffers non-reductive phys-
icalism. How else to naturalize the mental? The question then is whether nominalism
is essential, why non-reductive supervenience is not all the physicalism anyone
needs. A possible answer, constructed from separate things Schiffer says, is this.
He endorses supervenience but he rejects as obscurantist non-reductive supervenience
between non-pleonastic properties. He accepts supervenience for pleonastic mental
facts (somewhat tentatively): in every world in which the physical facts are as they
actually are, so are the pleonastic mental facts. This avoids obscurantism because on a
pleonastic view supervenience involves just the conceptual roles of mental ascriptions
(the very move that was intended to deect problematic causal overdetermination).
Given nominalism, supervenience is not a relation between language-independent
properties, but is a conceptual relation. So, on this construction, the function of
nominalism in non-reductive physicalism is to make sense of supervenience.
Now I do not think nominalism succeeds in this, nor, more generally, that non-
reductive supervenience is a promising account of the mental. Let us begin with
supervenience between full-blooded non-pleonastic properties and then extend the
point to pleonastic properties. Suppose it is held that x and y cannot differ in their
aesthetic properties without differing in their natural properties, even though
aesthetic properties are not natural properties. This of course is silent about how
narrowly individuated are the subvening natural properties: if x and y differ in their
aesthetic properties must they differ in their sensible properties? Or, at the opposite
extreme, might any Cambridge property of x and y be relevant? No matter. If x has
aesthetic property F it has some natural property or set of natural properties G such
that whatever is G in any possible world is also F. This is not a determinable-
determinate relation, for determinables of natural properties are themselves natural
properties. Supervening properties ride on the backs of natural properties; and one
straightforward picture is that supervening properties are higher-order properties.4
The aesthetic property F of individuals is the having of some natural property G that,
in every possible world, has the non-natural aesthetic property of properties F*.
Elegance is non-reductively supervenient if it consists in having natural properties
that are (essentially and non-naturally) elegant*, for then a things elegance is
metaphysically determined by, but not identical with, its natural properties.
Let this serve as the ofcial model. In its terms we can see that non-reductive
supervenience purports to combine factors that drive in opposite directions;
intuitions swing from one unacceptable extreme to another. Here is one. Invoking
supervenience is meant intuitively to naturalize mental properties, that is, to show
that non-reduction ts a naturalistic view of the world. But if the higher-order
property picture is apt, supervenience no more intuitively brings mental properties
into the natural order than it does Moores ethical properties. The physical deter-
mines the mental is misleading. This sense of determines is weak enough that even
on Moores account the natural determines the ethical; for if elegance* is an essential
property of a natural property, then necessarily whatever has that natural property is
elegant. But Moores view of aesthetic or ethical properties is determinedly not
naturalistic, and does not purport to bring them into the natural order. Why should
intuition be different regarding the mental? Now to the other extreme. In reaction it
may be said that Moores diagnosis was confused. A relevant higher-order ethical or
mental property is of the essence of the subvening natural or physical property, for the
latter has the former in all possible worlds, and thus the ethical or mental property is
naturalized, indeed is a natural or physical property. Now this seems both correct and
self-defeating. Correct, for how to make sense of a propertys being of the essence of
physical properties and yet not physicalthat is, not even in the broad sense in which
temporal, causal, and numerical properties are physical? Self-defeating, because the
non-reductive point is that an ethical or mental property is not a natural or physical
property. (No doubt some diehard will distinguish physical1 and physical2 properties.)
The upshot as it appears to me is this: to the extent that non-reductionmental
properties are not physical propertiesis the leading idea, the non-naturalizing
Moore picture dominates, and to the extent that mental properties are seen to ow
from the essence of the physical, non-reduction is contradicted.
The difculty is compounded. A straight thesis of non-reductive supervenience
leaves it open that mental properties supervene also on non-physical properties (rst-
order non-physical properties, unrealized in the actual world). Now, that is quite
different from functional properties being realized by non-physical states. For
functional properties involve (in addition to physical input or output) only topic-
neutral categorical factorstime, causation, numberthat are counted as physical,
4
As Hilary Putnam suggested to me in conversation.
and if mental properties involve merely such factors the question of non-reductive
supervenience does not arise. So we have this choice: either accept that mental states
are of the essence both of some physical and of some non-physical states and yet are
not topic-neutral in any way we understand; or impose the constraint that mental
states supervene only on physical states. The former I take not only to be mystifying
but also to reinforce considerably the difculty of seeing how supervenience natur-
alizes the mental; and the latter reinforces the puzzle how mental properties can be of
the essence of physical properties without being physical properties.
The question is whether nominalism helps with these problems. Does the trouble
stem from taking properties too seriously? As earlier noted, a realist perspective on
pleonastic properties makes them language-independent in the sense that their
realization is verication- or theory-independent. A light construal of predicate
quantication should not undercut a realist commitment to what predicates ascribe,
for otherwise we have reason to reject nominalism.5 Nor should pleonastic properties
resist higher-order predication; that would again be reason to reject nominalism, for
we want to make sense of He practices the third oldest profession and the like, not to
speak of higher-order functional properties. So nominalism does not block those
puzzling consequences of non-reductive metaphysical supervenience. And they are
not an artifact of the critics arbitrarily imposing modal formulations; non-reductive
supervenience is a modal thesis without which there is no apparent distinction of
non-reductive realism from dualism, no way for non-reduction to leave mental
properties part of the natural order.
Schiffers proposal to make supervenience just a matter of conceptual role then
goes beyond nominalism. Indeed it can be seen as inconsistent with realism about
mental properties even if nominalism is not. If a mental property M (pleonastic
sense) consists in having a physical property that verication-independently is M*,
that is not just a matter of conceptual roles. On the other hand, an anti-realist or
eliminative position can of course accommodate mere conceptual supervenience: if
mental ascriptions have only a systematizing role, supervenience could well be just a
constraint on assertability. Counting supervenience as merely conceptual does
demystify it, but at the cost of anti-realism.
3 Supervenient Explanation
On a non-reductionist physicalist theory mental properties ought not to be inde-
pendent causal powers. Nominalism does not secure that this is so, and presumably
we have to look to supervenience. But the matter is delicate. It would be a poor
defense of the mental against elimination that implied that mental properties do not
explain behavior. If they are to be counted as real they must presumably be capable of
5
Or we have across-the-board anti-realism, which as I noted earlier seems to preclude interesting
worries about dualism.
making a real difference: if they made no difference to behavior and were not even
the causes of apparent introspections of mental properties, our devotion to them
would seem more theology than common sense. We had better not go too far in
insisting on no independent causal powers. On the other hand, if they are super-
venient on the physical properties that completely causally account for behavior but
they are not identical with such properties, in what sense do they make a difference?
I assume that explanation by functional properties is, in its way, genuine explan-
ation, and that if mental properties were functional properties the problem would be
solved. Failing functional reduction, it may still be true that our best ascriptions of
mental predicates are correlated person-specically with physical properties, different
properties for different people. Now, the idea may be, unreduced mental predicates
can be taken to explain a persons behavior in the sense of indicating that in that
person there are corresponding physical (say) properties that explain his behavior,
properties that are structurally and causally related to each other and to behavior in
ways parallel to the structural and explanatory relations among mental predicates
and behavioral descriptions. If mental predicates express supervenient properties,
the idea then is that they explain by pointing to the existence of such underlying
explanations.6
A systematic development of the idea might have the effect simply of relativizing
the reference of mental predicates to individuals: M refers to physical property P in x,
to Q in y, etc. But that on its own would not apparently imply objective respects in
which the physical states of different people resemble each other. And so this sort of
explanation would not show that mental predicates stand for real unreduced
supervenient properties. Suppose it is said, We have to suppose that there are such
common properties in order to capture objective psychological generalizations
among people. This is fair. But these common properties must not only generalize;
they must in each persons case explain behavior, and the current proposal is that
they do so just by indicating physical properties that explain. I suggest that this
conjunctionmental predicates express properties that generalize across people and
they point person-specically to physical explanationsmakes sense only if mental
predicates express specic functional properties.
It is somewhat perplexing that one would care enough about the reality of mental
properties to nd the disjunction of reduction and elimination unacceptable, and on
that account accept the not overly clear notion of unreduced supervenient properties,
and yet go on to propose that the explanatory role of mental properties is the
foregoing totally passive one. If mental properties explain merely by pointing to what
really explains, why resist the eliminativist idea that they are just manners of
speakingthat what points are the predicates without the properties? Puzzling.
6
This perhaps bears some resemblance to Frank Jackson and Philip Pettits proposal, although some of
their examples suggest explanation in terms of functional properties (see Jackson and Pettit (1990)).
It seems that mental properties must have some more substantive role if allegiance to
them as objective resemblances is rational given the failure of reduction.
The question of the explanatory role of supervenient properties overlaps a related
question about propositional attitude explanation. Consider a thought about a
distant object, say Genghis Khan. And suppose that the thought itself, the current
event (token), is a physical event in the brain, which causes a certain action. How is
the wide content of the thought (= that brain state) related to this causal fact?
Presumably the content does not explain the action in a way to which the brain
states causal relations are irrelevant; the thoughts content does not, say, merely
rationalize the action. If the content explains the action it explains why this thought
caused this action; that is, it is an explanatorily or causally relevant property of the
thought. But how can this be? How can a historical property of an event explain
its current causal role?7 One expects its causally active properties to be its current
non-historical properties.
The question arises regardless of whether wide content is reducible, and it is not
specically about non-reductive supervenience. But a certain current answer to the
question, proposed by several writers, seems also relevant to supervenience and is
worth looking at in that connection.
The idea is to explain explanation counterfactually. The wide, or supervenient,
properties of our current brain states make a counterfactual difference to what their
effects are, and this is said to amount to an explanatory difference.8 Half of the
analysis is this: if the thought had not had that wide content, had it not been about
Genghis Khan, it would not have caused that action, say the utterance of Genghis
Khan was a great horseman. The other half introduces counterfactual sufciency: if a
thought with that wide property were in those circumstances to occur, it would cause
that action. This roughly is Lepore and Loewers suggestion.9 (The truth of the
antecedent and the consequent should not conjointly entail this latter counterfactual,
as happens on some accounts.) Schiffer puts a different second condition on the
explanatory role of wide contents, to secure sufciency: their ascriptions should
systematically predict actions.10
The relation between counterfactuals and causal explanation is notoriously con-
troversial, but I do not see these conditions as yielding explanation of any sort. They
may imply that wide contents make a difference, but, as Kent Bach put it to me, the
difference they make may be merely a correlational difference. Now, it will be
7
It is worth noting that the pointing account of supervenient explanation does not answer this
question, not if the supervenient property explains by pointing to a physical property on which it
supervenes. For if a mental property is wide, it supervenes on a physical property only if that property is
itself wide, and the question remains how the latter explains. But if wide content explains by pointing to
narrow properties, then the wide contents supervenience on the physical apparently plays no explanatory
role at all.
8
See Lepore and Loewer (1989), Schiffer (1991), Horgan (1989).
9 10
Lepore and Loewer (1989). Schiffer (1991).
conceded on all sides that a thought causes what it causes primarily because of its
internal physical features. Suppose there are person-specic systematic compos-
itional associations of wide content with neural features, and that, in a given case,
if no state with certain neural features had occurred no other state whose neural
features were correlated with the same wide content would have occurred. It appears
obvious that this mere correlational fact does not imply that either the states neural
features or its causal consequences are explained by its wide content. But the
correlational fact does imply in a sense that if that state had not had that wide
content, it would not have been true that it had those neural features, perhaps not
even true that it occurred, and thus not true that it had those behavioral effects. This
resembles backtracking counterfactuals. If proponents of a counterfactual account
of explanation required counterfactuals that are not merely correlational or back-
tracking, why would that not exclude the counterfactuals that relate wide content and
behavior? At the very least it begs the question. The point applies also to the
counterfactual sufciency of wide content for behavior, and also to its predictive
role. Mere counterfactuals do not explain, unless one stipulates that they are not
merely correlational, which begs the question.
If these counterfactuals do not mean that wide content explains a thoughts
behavioral effects, then presumably similar counterfactuals, for similar reasons, do
not show that unreduced supervenient intentional properties are explanatory, even if
the putative supervenient properties involve, say, only some sort of non-wide content.
We appear to have seen that if counterfactuals connect content (wide or narrow) or
other supervenient properties to behavior, merely by virtue of some projectible
mapping of contents or other supervenient properties to neural states that causally
explain the behavior, then those counterfactuals do not thereby yield an explanation
of behavior in terms of the correlated content or supervenient properties. We might
note that it would not in itself be fatal to wide properties reality that they do not
explain behavior, for that is compatible with their being reduced in some other
connection. But one may well wonder about unreduced supervenient properties,
how real they can be if they do not in some real way explain.
It may seem that the argument proves too much. Surely we want statements such
as He said it because he thinks that Genghis Khan was a great horseman to be
assertable. Granted. But this does not mean that we thereby assert that wide contents
in some sense explain actions. In general, the property an explanation explicitly
mentions may be extrinsic to the implied explanation: The man became ill because
he drank some green stuff. Ascriptions of such extrinsic properties can be explana-
tory because they implicitly indicate (in varying degrees of deniteness depending on
contextual background information) further properties that are intrinsically causally
explanatory. And we can in fact see propositional attitude explanation as extrinsic, as
purporting to point contextually to internal intentional mental factors of the belief
aspects of how a thought conceives things, modes of presentationthat are straight-
forwardly causally explanatory. There is no guarantee of course that such internal
properties are reducible to physical or functional properties. But reasons for being
skeptical that they are reducible appear to me far less impressive than reasons to be
skeptical of the reduction of wide content properties.11
Suppose then that internal intentional properties are reducible, and that propos-
itional attitude properties have only an extrinsic role in explanation. We may then
take a relaxed view of the status of that-clause ascriptions if they resist physicalist
reduction. If they explain extrinsically by pointing to real internal intentional mental
factors, it does not matter to the reality of intentional explanation whether propos-
itional attitude ascriptions themselves are literally true or are second-string,12 for the
internal intentional factors do the real explaining and their prospects for reduction
have (in my view) not seriously been challenged.13 So an argument for non-reductive
supervenience would then fail, if it holds that unless propositional attitude ascrip-
tions are literally true there are no correct common-sense mental explanations.
On the other hand, suppose that internal content cannot be made sense of, or that
if it can it is not reducible. Whether we then would have reason to regard wide or
narrow intentional properties as really out there, supervening on the physical, is the
next topic.
4 A Priori Reasons for Supervenience

There are subject matters in which supervenience without reduction appears well
motivated. Consider the totality of social facts, about language, conventions, law,
money. Let the non-social facts include facts about individual psychology, physi-
ology, and causal relations that are speciable independently of social facts.
Conceiving pairs of situations in which the social facts but no non-social facts differ
is not easy. Apparently it is not conceivable that dog means something other than
dog even though our society is in all non-social respects as it actually is, with people
saying dog in response to dogs in the same non-socially-described circumstances as
we do, with sensory and functional internal states associated with dog the same as
ours, and so on. Then apparently it is a priori that social facts supervene on non-
social facts. And it would be brave to assert that there exists a reduction of social facts
to non-social facts, that socially constituted resemblances are identical with physical
or physical-functional resemblances.
11
For the latter, see Schiffer (1987).
12
Though of course it will matter to the reality of truth conditions and reference, not to be lightly
abandoned.
13
A bold remark. But the objection to the reduction of propositional attitude ascriptions that I nd
most compelling is simply that all promising proposals appear refutable. I do not think the same is true of
demonstrative-introspective reductions of internal mental factors (see section 5). And I do not nd
compelling a priori objections to the reduction of intentionality that are based say on the alleged normative
status of content.
The moral is not that metaphysical supervenience without reduction makes sense.
For that ignores the live philosophical option of taking social facts to be second-string
facts. One interpretation of the case is that here again we have mere conceptual
supervenience, the sort of thing a non-cognitivist about moral properties might hold,
that there is a supervenience of acceptable ascriptions of moral or social predicates on
the natural or non-social facts.14 Metaphysical questions about supervenience and
ndings of obscurantism are avoided because that proposal would be deationary
about moral or social facts. Granted that it is counterintuitive to relegate social facts
to second-string status. Still, we have not seen reason to suppose that supervenience
rescues the social from this fate, for we have not seen reason to suppose that there is a
coherent notion of metaphysical supervenience without reduction. It is hardly
obvious that one saves common sense by appeal to such a relation.
The point can be made independently of the coherence of non-reductive meta-
physical supervenience. Consider a latter-day behaviorist position, one that is non-
reductive, that is, does not count each mental predicate as standing for a distinct
behavioral-environmental property. It holds merely that the totality of the behavioral-
environmental facts determines all mental facts, and in familiar behaviorist style
holds this a priori. This means rejecting the conceivability of inverted spectra, of
radically dissimulable thoughts and feelings, etc. The question is the resultant status
of mental facts, whether they survive with a rst-string status. Is the a priori super-
venience of mental facts on the behavioral-environmental facts more than notionally
different from an anti-realist position on which mental ascriptions merely concep-
tually supervene on the physical facts and have no claim to truth beyond their
systematizing the behavioral-environmental facts? It is hard to see how there can
be a real difference. And if the a priori supervenience of the social on the non-social is
parallel, if that is the strongest assertable dependence of the social on the non-social,
then we ought to regard the truth of statements ascribing social properties as
consisting in their systematizing the non-social facts.
It might appear that this is all to the good, that not only does it elucidate a reason
for holding certain statements to supervene non-reductively on a certain totality of
facts (it being the point of mental and social concepts to organize those facts), but it
also makes metaphysical supervenience unmysterious. If the social facts are consti-
tuted by their successfully systematizing the non-social facts, we then have super-
venience without reduction. And this is not anti-realist in a vericationist sense; for
the fact that certain statements systematize certain verication-independent facts
may itself be verication-independent.
There are two sorts of systematizing. If statements of a certain class have non-
systematizing truth conditions (say, statements about conscious thoughts and feelings),
an eliminativist or skeptical attitude toward them is compatible with acknowledging
14
Of course the point of moral discourse and of systematizing social discourse is substantially
different.
their systematizing capabilities, in an instrumentalist spirit. Successful systematizing in

this case does not secure truth, and so the position is eliminativist, assigning a
straightforwardly second-string status to the truth of such statements. But as regards
statements whose semantic point is systematizing, things are not so clear. Given their
marked contrast, though, with statements that have non-systematizing truth condi-
tions (say from chemistry), regarding their truth as second-string, in this case consti-
tutionally so, appears reasonable. If asked whether a social or mental predicate (given a
priori supervenience) stands for an objective resemblance, I would not know what to
say, except perhaps that the question is idle: social and mental ascriptions would
supervene conceptually on the non-social and the behavioral-environmental and
nothing seems to be gained by saying that that is anything more than mere conceptual
supervenience.
5 A Posteriori Reasons for Supervenience

But is it a priori that mental facts supervene on physical facts? It takes a large dose of
either functionalism or Wittgensteinianism to accept that for sensations. Nothing
seems easier than to conceive the phenomenal facts varying while the physical-
functional facts are the same. (Not that this epistemic point implies anything
interesting about the nature of sensations.) But the concern of many proponents of
non-reductive supervenience is with propositional attitudes, with content-bearing
states, and not sensations. A possible view is that there is an a priori supervenience of
propositional-attitude ascriptions on behavior and external relations. But should we
ignore conscious thoughts? These are content-bearing states with prima facie inten-
tional properties that can be conceived in the Cartesian way as varying even while the
physical facts stay the same. This is bound to be conceptually coherent if internal
conscious factors play an essential role at some level of thought individuation.
Dualism could not have been so successfully entertained if the intentional in general
supervenes a priori on behavior and external relations.
If that is granted, then there have to be a posteriori reasons for holding that there
are mental properties that non-reductively supervene on the physical. It is instructive
to compare a posteriori reasons for reduction. Simplify for the moment, by supposing
that non-reductive supervenience is an incoherent idea, so that physicalism is
necessarily reductive. Now a distaste for dualism and eliminativism would provide
merely a reason to seek reduction and not a reason to hold it can be found. There are,
of course, indirect empirical reasons for physicalism and against dualism. But given
physicalism (and the provisionally assumed incoherence of non-reductive super-
venience) the only empirical reasons for rejecting elimination would be empirical
reasons for reduction. If you dig your feet in and say there could be no good empirical
reason for the eliminative position, then presumably you will allow that there could
be empirical reasons to give up physicalism. But if you say there could be no
empirical reason to give up either, and assert reduction despite the non-existence
of empirical reasons, then (given that supervenience is not a priori) I doubt you hold
a rational position.
Reverse the roles of non-reductive supervenience and reduction. Suppose we reject
both reduction and a priori supervenience. If there is no empirical reason to assert
that there are unreduced supervenient mental properties, then continuing both to
accept physicalism and to reject elimination is not rational. If the reason for denying
that there could be empirical evidence for elimination is that normal human behavior
makes it inconceivable that humans lack mental states, the position reverts to a priori
supervenience. If on the other hand one rejects elimination on introspective grounds,
then the absence of further empirical grounds for taking the introspected properties
to be supervenient is a reason to entertain dualism.
Thus asserting non-reductive mental-physical supervenience requires non-trivial a
posteriori reasons, this by contrast with the usual reliance on right thinking; you
cannot simply assert that non-a priori supervenience holds because you want to
endorse physicalism and to reject elimination. Now it is useful to consider what those
reasons might be in the light of potential empirical reasons for asserting reduction.
Speaking very roughly, there are two kinds of a posteriori reason for which one might
assert reductionas we might call them demonstrative-introspective and functional-
explanatory. Regarding the rst, suppose we accept introspective phenomenal
concepts as legitimate despite Wittgensteinian objections, and take them to be
recognitional concepts, so that conceptions of toothache have the form feeling of
that sorttype-demonstratives, as we might say. If it is discovered that our ascrip-
tions of such concepts correspond projectibly to properties of the brain, then we may
regard them as discriminating those properties, as pointing demonstratively, and
hence as referring to them.15 Would there be empirical reason to take them to
discriminate physical properties directly, rather than indirectly via the discrimination
of non-physical phenomenal qualities with which those physical properties are
correlated? The answer is yes, in this sense: there would be no empirical reason to
think there are any such non-physical qualities and plenty of empirical reason to
regard all our properties as physical (putting aside dubious conceptual arguments
for dualism).
Suppose instead that phenomenal concepts do not discriminate physical-
functional properties. Then they discriminate unreduced supervenient properties,
or non-supervenient (dualist) properties, or no objective repeatables at all. Might
there be empirical reasons to assert supervenience? In the reductive case we envis-
aged that phenomenal concepts discriminate physical properties to which we have
independent scientic access. But there is no independent access to unreduced
supervenient properties, and so the epistemological situation would be importantly
different, presumably requiring a blind inference-to-the-best-explanation to
15
For an elaboration of this idea see }}2, 3, and 5 of Loar (1990) [Chapter 10 in this volume].
unreduced supervenient properties, perhaps along the following lines. Something

must explain our inclination to use a given range of concepts as if in discrimination
of a real repeatable; the best explanation is that there are genuine repeatables there.
And independent empirical reasons for physicalism warrant holding that this some-
thing is supervenient rather than non-supervenient (i.e., dualist), namely, precisely
those independent reasons that the reductionist relies on in the above argument
against mediating dualist properties.
Consider another possible empirical reason for reduction, functional-explanatory
we can call it, and then its non-reductive-supervenience counterpart. Suppose the
systematic success of causal explanations in terms of mental states is explained by a
system of physical properties that yield explanations of the same behavior, system-
atically parallel to the mental explanationsthat is, some one-to-one assignment of
mental to physical properties (or to small sets thereof) preserves causal explanatory
adequacy. Mental properties are then identied with the corresponding physical
properties, and we have physical reduction. This matching of explanatory structures
does not appear to me to require laws, even ceteris paribus laws, and so would differ
from classical reduction.16 What is essential is a systematic matching of causal
explanations (which do not presuppose laws couched in mental terms). Think of
good mental causal explanations as systematically reecting underlying physical
causal explanations.17 In conrming this reductive situation we again have inde-
pendent scientic access to the physical properties with which we identify the mental
properties that systematically reect them.
But suppose instead that no such systematic matching of mental and physical
properties turns up. To justify unreduced supervenient properties we are again led to
a blind inference to the best explanation, this time an inference from the success of
these explanatory patterns to the causal role of properties to which we have no
independent access. This, let us suppose, is preferred to the hypothesis of dualist
properties on the same grounds as above, namely, independent empirical reasons for
physicalism.
Thus we have two very similar arguments, each offering itself as an a posteriori
reason for non-reductive supervenience, corresponding to the two a posteriori
reasons that could exist (but that we are supposing do not exist) for physical
reduction. They assume that either our inclinations to make phenomenal discrim-
inations, or the success of third-person mental explanation, constitutes a prima facie
16
For a version that invokes laws, see Field (1991). For a telling attack on ceteris paribus laws see
Schiffer (1991).
17
If you think of mental explanations in terms of that-clauses, this conception of reduction may
appear implausible. But if you think of mental explanations in terms of projectively ascribed modes of
presentation or ways of conceiving things, then it is not at all implausible in my view. I know of no current
good objections to the idea that introspective ways of individuating thoughts systematically capture
physical-functional properties of the brain. On this way of conceiving our conceptions of mental content,
the two styles of reduction may well converge.
reason to infer, as their best explanation, the reality of mental properties. That propos-
ition is then coupled with independently empirically established non-reduction and
physicalism to give us a complex a posteriori argument for unreduced supervenient
mental properties. But there is an obvious soft spot in the argument, for adding
physicalism and non-reduction to the prima facie assumption of best explanation may
well undercut its prima facie plausibility. Granted that explanatory success, and inclin-
ations to make phenomenal discriminations, each provide prima facie abductive reason
to suppose that mental properties are real. Unless, however, we have an independent
reason to take unreduced supervenient properties seriously, physicalism conjoined with
non-reduction could instead give us strong reason to suppose that the success of mental
explanation can be explained away in physical terms.
It may be suggested that there is independent reason to take unreduced super-
venient properties seriously in the success of special sciences, say biology, thus
strengthening the above argument. But there is a well-known, and indeed quite
reasonable, alternative attitude to biology: if no physical-functional properties corres-
pond reductively to biological predicates, the latter lack reference and are of instrumental
import only, and we can expect biology to revise itself accordingly. Of course this makes
sense only if those biological predicates have a revisable role. But suppose they are so
persuasively projectible that no one who both has a feel for science and is not a village
eliminativist would regard them as not standing for real properties and relations. This is
how Jerry Fodor sees the special sciencesthey contain ceteris paribus laws that any
non-extremist must see as both unreducible and true.18
But is this an argument for unreduced supervenient properties? Let us divide the
discussion into two cases, assuming with the argument that reduction fails: there is an
explanation of the projectible systematizing success of those laws in physical terms,
or there is no such explanation. In the former case, we have then explained the
apparent truth of those laws without the hypothesis that they refer to unreduced
objective properties. Fodor says that without special science laws we lose general-
izations, that is, over creatures that physically differ but of whom the laws appar-
ently hold. This could suggest that the laws predicates stand for functional properties
that are multiply realized, but that would mean functional reduction. If the general
application of special science laws has a physical explanation that does not yield
physical or functional reduction, then there is no inference-to-the-best-explanation
to unreduced supervening properties: the best explanation of the systematizing and
generalizing success of those laws lies in the straight physical facts.
Suppose that the physical facts imply the projectibility of special science laws, but
that they do not explain them in some pragmatic or epistemic sense of explain
they dont explain them for us. That a computer could derive the projectibility of
those laws from physical fact and theory, it may be said, does not mean that there is a
18
See, e.g., Fodor (1989).
physical explanation of those laws success. The underlying physical facts may be too
unwieldy, too diffusely related to special science laws to explain, for us, what the latter
explain. If anything is to explain what they purport to explain, the special sciences
themselves then have to be counted true. Now this appears a somewhat anti-realist
understanding of inference to the best explanation. It is hard to see how, on a realist
view, the limitations of our knowledge and intelligence could be an empirical reason
to hold that there are, out there, certain biological or other properties. And, as
I earlier suggested, an anti-realist understanding of these matters considerably lessens
the interest of non-reductive supervenience.
The remaining possibility is this: the special sciences success is not explained by
the physical facts even in the objective non-pragmatic sense. Now evidently this is
not the situation proponents of non-reductive supervenience suppose we are in. For
how are we to make sense of the special science predicates supervening on the
physical facts if the physical facts do not imply the projectible success of special
science predictions? We have then, presumably, either unexplained success or a
dualist explanation of it. With or without a physical explanation, one is left thinking,
there cannot be a special science vindication of unreduced supervenient properties.
The eliminative alternative to unreduced supervenient properties means that very
fundamental common sense could be overthrown. But surely we cannot bring
ourselves to doubt that our fellows have thoughts, feelings, and experiences! Now
this rhetorical point ought not to be just another way of asserting the a priori
supervenience of mental states on behavior and external relations, for here we are
assuming that supervenient mental properties need an inductive explanatory infer-
ence from the straight physical facts. But what then is the point? If common sense is
not behaviorist, it is not easy to see how it can have an unrevisable commitment to
supervenient mental properties. It might appear that unrevisability derives from
introspective certainty that there are mental properties, and that this leaves it open
whether they are supervenient or dualist properties. But it is difcult to see how this
makes sense. If the physical facts explain (without vindicating) our introspective
property-identifying tendencies, there is no empirical reason then to suppose we are
discriminating unreduced supervenient properties. On the other hand if the physical
facts do not explain (in the non-pragmatic sense) those tendencies, it seems that what
we introspect cannot be properties that supervene on the physical facts. So if
introspection were unrevisable, and if we accepted that the physical facts do not
explain our introspective inclinations, we would be committed to dualist properties.
Again unreduced supervenient properties lack empirical support on any supposition.
They appear to be irrelevant.
13
Reference from the First-person
Perspective
In this paper I wish to address two questions about reference that are among the most
fundamental issues in the theory of meaning and intentionality. They are 1) what
makes different external (e.g., causal) relations count as semantic relations, count as
reference; and 2) whether reference is objectively indeterminate or inscrutable and
whether a positive answer subverts our common-sense conceptions of semantic facts.
It appears to me that the two matters are deeply connected, and I will propose a more
or less simple idea that permits a unifying answer.
1 What Makes a Causal Relation

a Semantic Relation?
First issue. Suppose we accept that reference is determined by causal and other
externally constituted relations. There are various ways in which an individuals
thoughts and concepts can refer to, be about, pick out, a given thing or property,
etc. And these ways appear to be correlated with what are intuitively psychologically
distinct kinds of concept. Here are some examples (see Table 13.1).
Concept category External reference relation
Visual demonstratives An optical relation

that tree (which one sees)
Visual-memory demonstratives A memory-cum-optical relation
that tree (which one remembers seeing)
Recognitional-type-demonstrative A discriminative ability-cum-memory relation
that kind of thing (see section 8)
Socially deferential names A social-causal-historical relation to a particular
Julius Caesar, Richard Wagner
Socially deferential kind terms A social-cum-? relation to a kind or property
aluminum, arthritis
Table 13.1. Concepts and reference relations

REFERENCE FROM THE FIRST - PERSON PERSPECTIVE
It is widely supposed that what makes these relations reference relations is their role
in determining the truth conditions of the thoughts in which the corresponding
concepts occur. But it is possible to be unhappy with this. It is difcult to believe
that there is a naturalistic explanation of non-deationary, robust, truth conditions
that does not rely on reference relations that are (partially) independently consti-
tuted. It appears to me that the widespread view that the notion of truth conditions is
somehow prior to, or more basic than, the notion of reference is not well motivated.
Granted that, if there were no such things as thoughts and utterances with truth
conditions, the reference relation would have no instances. But it does not follow
that the notion of reference is asymmetrically derivative from the notion of truth
conditions. As I will try to show, there is an answer to the question of what those
relations have in common that is of quite a different sort, and that (it seems to me)
throws light not just on the notion of reference but also on the semantic notion of
truth conditions.
2 The Objective Indeterminacy of Reference

The causal-external relations that vaguely come to mind do not uniquely determine
reference. Suppose O is the relation of belonging to a causal chain that initiates the
occurrence of a visual demonstrative concept, e.g., that tree. O does not single out a
tree, by contrast with a retinal image, a stream of photons, a tree surface, a tree trunk.
This underdetermination is masked from uncritical common sense; for demonstra-
tives have implicit qualifying concepts, e.g., tree (or perhaps detached object). The
demonstrative thought conceives its referent as a tree, and the referent is the tree to
which the concept bears the relation O, and not the tree surface, etc. Commonsensi-
cally, a visual demonstrative concepts referent is the object that falls under the
qualier and to which the concept stands in that optical relation.
But of course the demonstrative concept refers to a tree only given that tree refers
to the kind tree and not the kind tree surface. And the question of inscrutability arises
also about the determinacy of kind-reference. Consider Eve, who fashions an idea of
trees in general from her pattern-recognitional groupings of objects in Eden.
Whether her concept thing of that kind refers to the kind tree or the kind tree
surface depends on whether Eve counts detached objects or surfaces as things of that
kind. But this in turn depends on her singular demonstrative references. For what
grounds her ascriptions of tree or thing of that kind to a certain group of objects are
her judgments this is one, that is one too, and heres another . . . If we wish to assign
reference to Eves concept tree, we have to rely on the reference of her singular
demonstratives.
We are then caught in a circle of interpretation. (This is not, essentially, an
epistemological circle.) The determinacy of singular demonstrative reference
depends on the determinacy of the reference of the kind-concepts that serve as
qualiers of the singular demonstratives, and vice versa. To take the idea a step
further, notice that, if the reference of certain socially deferential names and kind-
terms depends on the demonstrative references of individuals who establish usage,
the reference of socially deferential names and terms is similarly indeterminate. And
resolving the indeterminacy of reference for them means breaking out of the more
basic circle of interpretation.
It is instructive to consider a certain stipulative resolution of reference indeter-
minacy for visual demonstrative concepts. (I make these demonstrative concepts
central in this paper not only to have a central example, but also because, it seems to
me, a theory of reference for demonstratives must be regarded as quite fundamental
in the theory of reference.) The stipulation of a reference scheme, from among
various other reference schemes that determine the desired truth conditions, might
begin like this. Suppose that D is an externally determined relation that maps
qualifying conceptspredicates such as tree, detached object, object surface
onto certain sets, e.g., the set of trees, of detached objects, of object surfaces. We then
stipulate that D counts as denotation (in the reference scheme we are constructing).
Now the reference of a visual demonstrative depends not just on its qualier but also
on the unrestricted optical relation O: the reference of a demonstrative concept with
qualier Q is determined by restricting O to objects in D(Q). We might think of O as
the reference-anchoring relation for visual demonstratives.
So given that stipulation of D as denotation, the optical relation mentioned at the
outset is the restricted optical relation O*:
(where x is a concept-token and y an object) O* xy iff Oxy and, if Q is xs qualier,
y belongs to D(Q).
O* determines reference for visual demonstrative concepts. The effect of this is
equivalent to having stipulated that O*the restricted optical relationcounts as
reference for visual demonstrative concepts.
There are various ways to think about indeterminacy, but the foregoing structure
gives us a fairly straightforward way. The question can be put like this: what makes
the relation O* a reference relationrather than some competing relation O** that is
dened from O together with a different function Ref0 ? Or, more or less equivalently,
what makes the relation D rather than some other relation D0 a denotation relation?
The stipulative answer says: these are the relations we count as reference or as
denotation. The point of picking this relation rather than that relation (actually, this
overall reference scheme rather than that one) is that we ought to x on one, to
account for the compositionality of our independently motivated assignments of
truth conditions. Or, at least, that is the stipulative account I have in mind.1
But there are reasonable worries about the point of such a stipulative solution.
Here are two that come to mind. (i) Why do we nd the causal relation O important
1
This bears some resemblance to Hartry Fields conventionalist solution to referential inscrutability, in
Field (1975).
enough that we have bothered to promote one of its restrictionsO*as (so to

speak) the designated representative of the set of O-restrictions? This is in effect to
ask why we want a notion of reference at all. Let me consider two answers. (ia) The
rst, and more familiar, is that substantive reference relations are needed to explain
the compositional determination of truth conditions (where the latter are themselves
held not to be conventional or stipulated facts). This has been supposed by philo-
sophers who have held reference to be inscrutable, or to be (what is not so different)
scrutable by convention.2 But if reference is merely stipulated, it is obscure how
it can explain something with a more secure standing, viz., truth-conditional
compositionality. Perhaps it will be said that what does the explaining is collectively
the set of all reference schemes that generate those truth conditions (where each
restriction of O belongs to one such reference scheme). But this undercuts the
idea that one of those reference schemesand its constituent O*is to be awarded
a special status. We are left with the rather disappointing idea that one of those
schemes is designated to represent the explanatory role of the collective.
The more appealing idea is that some factor in the determination of reference is
independent of its role in determining truth conditions. Truth-conditional compo-
sitionality will then depend asymmetrically on reference in this strong sense: it is a
fact that needs explaining only if constituted (in part) by relations whose semantic
status is independent of that constitution.3 But it is far from evident that a compos-
itional explanation of language-understanding must be truth-conditional. Other
explanations, which involve translation rather than truth theory, appeal to rules
for mapping others sentences onto our sentencesthat are in their way compos-
itional but not truth-theoretic. For an argument of this sort see Schiffer (1987),
chapter 7.
(ib) A second reason for wanting a notion of reference might be this: those causal-
informational relations that we count as reference relations play a crucial role in
explaining the reliability of beliefs and utterances (relative to a standard assignment
of truth conditions).4 Yes. But while this may eliminate some truth-conditionally
adequate reference schemes, it does not eliminate gavagai-type alternative reference
schemes.5 And as regards that restricted class of schemes, something similar to the
point above, about compositionality, applies: explaining reliability in terms of
informational-causal relations does not require stipulating a special role for one
among the adequate (gavagai-type) schemes.
2
Field (1975), Davidson (1979).
3
Understanding the novel utterances of others is widely held to require compositional explanation; and
it may be said that this requires assuming that some single reference scheme plays an implicit conceptual or
computational role in that explanation.
4
For this point see Loar (1981). Truth conditions are construed loose-grainedly, so that different overall
reference schemes are consistent with a single assignment of truth conditions.
5
In Loar (1981), I acknowledge this point, although I go on to attempt to minimize the signicance of
the gavagai alternatives. In the present paper I take this more seriously, in part because of the point about
the rst person that follows.
Indeed a more fundamental point threatens any such motivation for a stipulative
determination of reference. Reliability is relative to a specic assignment of (loose-
grained) truth conditions. But it is not implausible that quite different such truth
condition assignments permit similar degrees of reliability. The question then of
course is whatapart from an independent and non-arbitrary selection of reference
schemesmakes a given assignment of truth conditions privileged. In the face of a
striking absence of candidates, we have a strong incentive to nd such a non-
arbitrary factor in reference determination.
(ii) Even if we accepted that third-person reference ascriptions are determinate
only by selective stipulation, that would hardly t rst-person ascriptions. When
I judge that a current visual demonstrative thought of my own is about a certain tree,
other interpretations are not optionalthere is no intuitive scope for arbitrary
selection. And this suggests that the rst-person perspective on reference is special.
When we consider the references of our own thoughts, it appears bizarre to suggest
they are established as such by stipulation.
3 Disquotation as a Constraint on Reference

Certain deationary accounts of reference give a special role to the rst person.6 On
such a theory to say that
George Sand names George Sand, or
Tree denotes trees, or
The next occurrence of this hand refers to this hand,
is to say no more than is captured by the disquotational structure of those ascriptions.
And disquotation can be granted a special status in determining reference only for
the terms of ones current language.7 (It is conceptually open whether, on another
persons use, George Sand names George Sand.)
There is something bizarre, as we have remarked, about nding indeterminacy in
the rst person. However disconcerting and disruptive of ordinary assumptions the
idea of indeterminacy in the third person is, we at least nd intuitive room for
entertaining the possibility when it is forced on our attention. And this asymmetry
between the two perspectives has of course a nice explanation on a disquotational-
deationary theory of reference and truth. On a deationary account, third-person
ascriptions of reference of the form
N as she uses it refers to t
are explained thus:
6
See Field (1994).
7
As Field puts it, disquotational truth and reference are dened only for expressions that one
understands.
N as she uses it is translated by t as I use it and t refers (disquotationally) to t.

There is room for indeterminacy in third-person ascriptions of reference if there is
room for indeterminacy in term-term translation.
Deationism may then count the intuitive asymmetry as support. It is not on our
present agenda, however, to consider the merits of deationism. We are working
within the non-deationist assumption that reference is determined by various
contingent external relations of the kinds mentioned above. Now, the thesis that
rst-person disquotational ascriptions have a special status is distinct from the
deationary thesis that refers is a pure disquotational notion. And it is quite
compatible with the non-deationist assumption that disquotational ascriptions of
reference have a privileged status,8 and that such ascriptionsor something akin
play a special and non-arbitrary role in determining which of various external
relations qualify as reference. Still, elaborating this role is not as straightforward as
one might casually suppose.
Here is a rst try. Deationary theories may be interpreted as saying that the set of
term-object pairs that instantiate reference for my language can be laid out, for each
concept category, as follows (to take proper names as examples):
George Sand, George Sand, Cary Grant, Cary Grant, Lisbon, Lisbon, etc.
Suppose that, for the proper names in my language, some robust relation R holds of
all and only such pairs, open-endedly. The proposal then is that R counts as a
semantic relation: it determines reference; and likewise for each relation that satises
the disquotational condition for each other concept category. Suppose that, in our
(pre-philosophical) counterfactual judgments about R and about reference, we can
nd no counterexample to the thesis that
for any person P, if N is used as a proper name by P, then N bears R to x iff N refers
to x in Ps language/thoughts,
and similarly for any relations that match the disquotational conguration for
other concept categories. We have then pointed to R by way of its holding of
(as we might say) disquotational pairs, and then have veried that, given ones
ordinary notion of reference as applied to the languages and thoughts of others,
R coincides with reference for the proper names of others. (This of course allows
that R holds for names used by others but not by me.) Similarly for other concept
categories.
Might we not then reasonably claim to have explained what makes a robust
external relation a semantic relationmakes it referenceby essential appeal to the
8
For an extended discussion of puzzles about this idea, see Loar (1994).
disquotational pattern? And would we not then have explained our intuitive con-
ception of the determinacy of robust reference relations? The ideas attraction lies in
this. The disquotationality of the home case seems an unchallengeable constraint on
reference, indeed constitutive in part of the very idea of reference. If the disquota-
tional generation of term-object pairs sufces, given external contingencies, to pick
out relations that t intuitions about reference in third-person cases, we have found a
single featurehome case disquotationalitythat is necessary and sufcient for
otherwise disparate external relations to count as semantic.9 And this feature satises
our further desideratum, a constraint on reference that is independent of the role of
reference in determining truth conditions.
A reference relation is a certain relation that applies to disquotational pairs. That is
the idea; but does it make sense? There is an elementary problem. No disquotational
pattern is to be found among pairs of terms and objects. You do not produce the
woman George Sand by peeling away quotes from George Sand. No disquotational
principle unies any set of concept-object pairs, or explains what later pairs have in
common with earlier pairs. I cannot then point to a relation between terms and
objects by virtue of its holding of certain disquotational term-object pairs, it seems;
for there are no such pairs.
The disquotational pattern is of course found among term-term or concept-
concept pairs: George Sand, George Sand, Lisbon, Lisbon,. . . . Now
this is quite an important fact, for while it does not directly constrain reference
relations, there is no reason why it should not explain what makes a relation-concept
a semantic concept. We will return to this. Our present question, however, concerns
relations and not relation-concepts. And those term-term pairs cannot count as
determining reference relations indirectly. Call a term-object pair N, x disquota-
tional just in case it is picked out by a pair of (ones own) terms N, N. Suppose
one then says that R is a reference relation if R holds of disquotational pairs.
(Somehow open-endedly?) But that of course wont advance our project, for it
appeals to a semantic relation (picked out by) whose determinacy, and constitution
as semantic, are what we seek to explain.10
Our problem in explaining how disquotation can be a constraint on reference is
then this. A disquotational constraint on term-object pairs or term-object relations
makes no sense. And while a disquotational constraint on term-term pairs and term-
term relations does make sense, it cannot be counted as constraining term-object
reference relations without presupposing semantic relations.
9
There is the awkward possibility that other people have concept categories different from ours. We
would not then have explained why any given relation between terms of those categories and things/properties
should count as reference.
10
Compare: O* is a reference relation, because O* and not O** is the one I pick out in saying refers.
4 Disquotational Term-object Pairs from the Subjective

Perspective. The Interplay of Subjective and Objective
Perspectives
There is a way around the problem, but it involves a shift of perspective. It does
appear to me that the idea of disquotational term-object pairs is not completely
hopeless. But we have to look at the matter phenomenologically, with an eye to how
things appear in the rst person.
Visual demonstratives again. Produce, as realistically as you can, pairs of concepts
in the following pattern,
(V) the next occurrence of that tree, that tree

the next occurrence of that hand, that hand
the next occurrence of that chair, that chair
...,
where the second member of each pair is a visual demonstrative concept. Now when
I say produce those pairs, I mean exercise those concepts. Use them to think with, and
dont (merely) mention them. What is the upshot? From your perspectivethat is,
the perspective of the user of those concept pairspairs of visual demonstrative
concepts and certain objects appear on the scene. (From the third-person or objective
perspectivewhich one also can take toward oneselfpairs of concepts are being
exercised.) How are those concepts and objects apparently linked? This is a delicate
matter. I am tempted to say this: they are linked in that characteristic way (which as it
happens is consequent upon exercises of those concept pairs above). If, from our
rst-person perspective, we count concepts and objects as linked in a disquotational
way, we make some sort of sense. That idea is basic in the present account. Before
returning to the question what sort of sense we might thereby make, let us be clear
how this would serve our project.
We seek a constraint on concept-object relations that makes them count as
semantic relations, as reference. The rst step is that certain concept-object pairs
may be counted as suitably disquotational from a phenomenological, subjective, or
rst-person perspective. So we lay on the table, for further investigation, as many
such concept-object pairs as we can. Now heres a key point. Nothing prevents our
shifting our perspective back to the objective perspective at this point, keeping those
pairs in mind while considering them afresh from that third-person perspective.
Does some well-carved external, objective, relation hold of them, in some projectible
way? Presumably it is natural to think there is such a relation, the one we have been
calling O*, even though we do not know its scientic details. But we apprehend that
relation as having a special status. It is as if O* is latched onto by the exercise of those
concepts; foras the phenomenology has itthat conceptual exercise generates
those concept-object pairs. It moreover appears to do so open-endedly, so that the
projectibility of that concept-object generation matches the relations open-ended

application. That relation can then hardly be seen, from the phenomenological
perspective, except as having an intimate connection with those concepts. That is
why O* counts as semantic, as a reference relation. A certain primitive idea of what a
thought is about, it is natural to say, derives from concepts appearing to bring objects
onto the subjective scene, and the corresponding optical relation O* is privileged by
its intimate connection with this phenomenon.11
Back to the plausibility of construing the felt link between those concepts and objects
phenomenologicallyrelated in that familiar way. The suspicion is that the link is just
the old visual demonstrative reference relation, that causal relation, somewhat
obscured by the vagueness of the phenomenological attitude. But the suspicion is,
I think, not borne out. In your conception of those concept-object pairs, you use
no semantic or intentional notions; you use the concept-concept pairs (quotation
concept and visual demonstrative concept) without semantic commentary. Having
put the concept-object pairs on the table, you may then of course go on to think, self-
consciously: I have used that concept-token that tree to refer to that tree. But this does
not appear to be the inevitable way in which one intuitively discerns a link between the
concept and the tree. For it seems quite natural that when such a concept-object pair is
thus displayedwithout explicit semantic commentaryit will appear to have a
characteristic conguration, a distinctive feel, as if the object is peculiarly salient in
the company of that concept. This apparent link between concept and object is hardly
magic. We might think of it as a straight projection of the intimate link between
disquotation concept and visual demonstrative concept onto the corresponding
concept-object pair, the latter being brought onto the scene by the exercise of the
former concepts. The appearance of such a common linkage in those concept-object
pairs seems quite expectable. And the phenomenon is registered without the use of
explicitly semantic concepts: the apprehension of the objects salience in the presence
of the concept is far more primitive and elementary than characterizing their relation
in causal-referential or intentional terms.
Describing the special concept-object link as apparent is apt, for it is an illusion:
no objective relation x is salient in the presence of concept Cbetween object and
concept is created by the mere exercise of the two concepts. (But recognizing this
does not destabilize the phenomenology; the illusion is robust.) What matters is that
the name-object pairs we lay on the table are subjectively privileged, their status
explainable as a projection of the exercise of disquotationally related concepts. This
of course permits proceeding to the further stage of investigation, when we shift
11
This primitive aboutness relation appears to be, in its way, disquotational. Should we not then count
reference as disquotational, at bottom so to speak?
This would be too quick. There are good reasons to count reference as constituted by externally
determined relations. Not the least of them is the phenomenon of counterfactual reference shift (as we
might call it).
perspectives, and see that those concepts and objects are objectively related after all,
independently of the phenomenology.
5 Appearances
In the foregoing phenomenological description we appealed to appearances. This
raises the question whether we might be presupposing referential determinacy in the
context it appears that. . . . That that could be a problem may be seen by considering
the following, different, proposal.
It appears to me that, by the following occurrence of that tree, I pick out that tree (rather than
that tree surface, etc.). It is of course an illusion that some privileged objective relation holds
between that term and that tree and not between that term and that tree surface. But still we
may say that as the result of the appearance of that connection, the former term-object pair has
a special status that the latter does not.
Now, interpreted as an objective statement about appearances, this of course does not
work. The construction it appears to me that . . . that tree . . . is a de re propositional
attitude ascription, and the question of inscrutability arises as much for such construc-
tions as for ascriptions of reference. We cannot assume that that appearance-
conguration objectively relates me and the tree and not also me and the tree surface.
But the phenomenological remarks do not make essential use of the construction
it appears that . . . , do not depend on (as it were) the objective ascription of
appearances. We capture appearances also by conveying what a certain experience
is like (e.g., by saying it is as if . . . ), and the appearance language in which we convey
this does not require objective construal. In phenomenological description anything
goes, and literality is mostly a barrier.
It is not part of the proposal that those pairs are linked by a certain objective fact
about how things appear. No special objective relation, of any sort, is being asserted
to single out those pairs. When we shift from the phenomenological perspective, the
pairs that we have placed on the table are not special, from the new objective
perspective. Our objective investigation can be seen as having a signicant semantic
upshot only from the subjective perspective. And so, success in advancing this
present proposal requires the readers phenomenological cooperation. The apparent
concept-object linkages are (I venture to suggest) something recognized in ones own
experience, when one exercises the corresponding concept-concept pairs.12
12
This presupposes that we can understand or conceive our conceptual and experiential situations
independently of that-clauses; evidently I am committed to that. For an account of such conceptions see
my How to Conceive Mental Content [this is almost certainly an ancestor of Loar (2003), chapter 15 in
the present volume].
6 What Makes a Relation a Reference Relation?

There are two ways to take what we have said so far, namely, as a proposal about what
makes a relation like O* a reference relation, and as a proposal about what makes a
relation-concept a reference-concept. As a proposal of the rst kind we have this:
O* is a reference relation because it holds of pairs linked in that way. Here we have to
suppose that in that way applies across concept categories. This means that pairs
such as the following have an intuitive phenomenological similarity: the next
occurrence of that tree, that tree, Plato, Plato, copper, copper, and so on.
Evidently this makes sense only when one exercises those concepts for oneself. Given
that the similarity among those pairs is not objectiveis merely subjectivethen so
is what makes a relation a reference relation. But we have a reasonable understanding
of what that is.
7 What Makes a Relation-concept

a Reference-concept?
We have a conception of the interesting relation that holds between visual-
demonstrative concepts and objects, and it has the form the relation that holds of
this pairthe next occurrence of that tree, that treeand that pairthe next
occurrence of that hand, that handand this other pair. . . . That conception
incorporates concept pairs that exhibit a disquotational pattern. Call them mention-
use pairs of concepts. Here we have an appropriate combination of subjective and
objective factors: the pairs are those presented as noted, and the relation we conceive as
objectively holding of those pairs. And what makes this common-sense conception a
reference-relation-conception is the role those disquotational concept pairs play in that
complex descriptive concept.
But this is somewhat disappointing. I appear to have a more general conception of
the optical reference relation than such a description, which mentions specic
demonstrative concepts and objects. My conception does not change from day to
day according to the sightings I keep in mind. It is rather an open-ended conception,
not tied to particular references. What I want to propose is that this concept is in its
way demonstrativethat relationand that in a certain way the disquotational,
or mention-use, conguration constitutes that demonstrative concepts dening
perspective.
8 Recognitional Concepts and Perspectives

We take a digression. Some of our concepts are, as we might say, recognitional type-
demonstratives, or more briey recognitional concepts. Here is an example. In East
Africa you see certain four-legged creatures up close. Not knowing their name, you
yet quickly learn to identify others of that kind, as you think of it (a kind of gazelle
in fact). It is not that you then reidentify the kind by remembering the earlier
individuals one by one. You may not keep the individuals in mind, having formed a
generalized type-memory-demonstrative, as well as an inclination to identify further
things as of the same kind. Suppose you also then spot some creatures wandering
on the far slope across the valley, with distinctive gait and distant appearance. You
form another demonstrative conception, of creatures of that kind. Then you raise
the question whether this kind = that kind. Let us say that they are of the same kind,
although for you it is an open question. You have two independent recognitional
concepts with the same reference. They differ perspectivally, by which I mean,
not in the external relations that hold between them and the two groups of
creatures, but in some psychological or subjective perspectival aspect. So there
are type-demonstrative recognitional concepts that are individuated at least in part
perspectivally.13
9 Conceiving Relations from the

Mention-use Perspective
This notion of recognitional concepts can account (if we take an abstract enough
view of them) for the above-mentioned demonstrative conception of the visual-
demonstrative relation, as well as for demonstrative conceptions of those relations
that constitute reference for other concept categories. We can account, moreover,
for what makes those relation-conceptions semantic or intentional conceptions.
Proposal:
Ones demonstrative conception of the visual demonstrative relation is a higher-order recogni-
tional concept, which purports to pick out a certain external relation from the mention-use
perspective. (Higher-order because concepts are in the domain of the conceived relation.)
My conception of the visual demonstrative relation appeals to no particular array of

mention-use pairs. It does not change over time, as it would if it involved reference to
the particular mention-use pairs one might have in mind at the moment. Compare
the above recognitional concepts of African animals; they need involve no memories
of particular animals. Those concepts type-demonstrative status depends on past and
potential discriminations of animals of a certain kind; they point in a certain
characteristic way to certain kinds. But they do not point descriptively via specic
reference to particulars.
Similarly for ones demonstrative conception of the visual demonstrative relation.
From the perspective from which we exercise such conceptions of concept-object
pairs in the mention-use conguration, the pairs themselves are phenomenologically
presented as from a distinctive perspective. It is as if there is a rule that generates,
13
For a more extended discussion of recognitional concepts, see Loar (1990) [Chapter 10 in this
volume].
open-endedly, the concept-object pairs themselveseven though we know reect-

ively that the rule governs not the pairs themselves but our conceptions of them. This
merely subjective perspective on concept-object pairs is phenomenologically stable,
and from it we are able to conceive that objective external relation (as we judge it),
which holds of this pair and that pair etc. It is in this sense that we may count our
conception of the visual demonstrative relation as a recognitional concept that is
formed from an essentially rst-person perspective. I retain, from day to day, the
same demonstrative conception of the visual demonstrative relation, even though the
instances I have in mind change. This parallels the fact that I retain the same
recognitional concept of a certain unnamed kind of ower that I see around town,
despite the change in its instances.
We want the relation thus demonstrated to be the one we referred to earlier as O*,
for otherwise we have not advanced. But this raises the question why we should
regard that recognitional concept, formed from the mention-use perspective on
visual demonstrative concepts, as discriminating O* rather than say O itself. (Recall
that O holds not just between the visual demonstrative concept this tree and a
certain tree, but also between that concept and a certain tree surface.) Granted that
the pairs one picks out are all instances of the restricted relation O*; they are also all
instances of the broader relation O.
The wrong answer is this: if we have a conception of a concept-object pair that is
not a disquotational or mention-use conception of that pair, then our recognitional
concept picks out relation R only if R does not hold of that pair. That excludes too
much: for there are many ways to conceive of concept-object pairs that are in the
extension of O*. We are not restricted to conceiving of them disquotationally, in the
mention-use conguration.
Back to the phenomenological concept-object link. Our recognitional concept that
relation is sensitive to pairs that present themselves as linked thus. So, if there is no
way of conceiving of a concept-object pair that engenders that apparent linkage, then
that pair is not in the extension of the relation discriminated as that relation. Indeed,
this implies that the relation picked out from the mention-use perspective must be a
many-one relation, a function. Then we can see that the pair that tree, that tree
surface is actively ruled out. The reason is fairly obvious. The tree surface is not
identical with the tree. But our conception from the mention-use perspective that
relation is a conception of a function, and given that function has in its extension the
pair that tree, that tree; the pair that tree, that tree surface is not in its
extension. In other words: if the apparent rule picks out c, a and one judges that
a is not identical with b, then the rule rejects c, b.14 A more direct route to the same
conclusion is this: it is clear from the pattern of our discriminations that we conceive
of the demonstrated relation as a function, and O* is a function and O is not.
14
Note well that this does not imply that, if the rule picks out c, a and fails to pick out c, bi.e., thus
conceivedthen the rule actively rejects c, b. For one may not know whether a = b.
What then makes a relation-conception a semantic conception? The recognitional

concept that relation is a semantic-relation-concept just in case it is related as
described to the mention-use perspective. I will consider below the signicance of
this subjective or rst-person nature of semantic concepts.
10 Factoring O*
As we saw earlier, we cannot simply declare that O* determines reference for visual-
demonstrative concepts and be done with it. We have shown why O* and not some
other relation determines reference for visual demonstrative concepts. But we have
not explained the structure of O*s reference determination. And this matters.
A deationist may take a hard line, asserting that all that matters in determining
reference, for expressions simple and complex, is the disquotational or mention-use
conguration, and that the standard assumption that semantic structure plays an
explanatory role in the determination of reference is a mistake.15
On a non-deationary view, though, it appears undeniable that the reference of
visual demonstratives is determined by two factors: the reference of the qualifying
concept, and the reference-anchoring relation O. But this is all right. Once we have
motivated counting O* as reference, a semantic factoring of O* is straightforwardly
motivated. For two simple facts dovetail. a) It is a fact from the third-person
perspective that in any large array of positive and negative instances this pattern
obtains: whenever x and y stand in the relation O* they stand in the relation O, and
the restriction to O* ts exactly a certain straightforward mapping D of concept-
types to sets of objects. b) From the rst-person perspective, all instances of the
mapping D are validated in the mention-use conguration. We have, that is:
tree denotes trees,
physical object denotes physical objects,
surface denotes surfaces,
and so on throughout all qualifying concepts. Thus there is a convergence: D emerges
from a straightforward factoring of O*s instances, and D also emerges from adopting
the mention-use perspective on the referential qualiers of visual demonstratives.
This would appear to vindicate the non-deationists regarding the reference of
visual demonstratives in two ways: as determined by O* (picked out from the
15
Regarding semantic structure, a deationist could say this. Explaining the generative or open-ended
nature of our capacities to produce sentences in thought and in communication does require some sort of
compositional structure. But that structure explains conceptual roles, or dispositions of use, and not
reference or truth conditions. (Cf. Schiffer on compositionality, in (1987), ch. 7.)
The reference of syntactically complex expressions does of course exhibit certain semantic structure. But
that is just a matter of patterns in reference that are consequential upon the primary determination of
reference, which on a deationist theory is disquotational.
mention-use perspective), and as determined compositionally by the joint action of

D and O, the reference of the qualier and the visual demonstrative relation. There is
no conict between the two perspectives on reference.
11 Reference for Concepts of Other Categories

Let us consider the socially deferential name relation, and how it is conceived
demonstratively. This is the relation that holds, for instance, of my proper name
George Sand and the woman George Sand. Speaking of my conception of that
relation as a recognitional concept could be puzzling. The relevant concept-object
pairs can be widely separated in time. And the relation is complex and difcult to
specify, a compound of my relation to a certain group of speakers and their relation
by way of a causal-historical chain as they sayto George Sand. I cannot take all that
in so to speak at a glance. But still there is a sense in which I have a recognitional
concept of that relation, and it is rather modest. Once again, there is a mention-use
conguration, in this case involving proper names in thought. And this generates an
open-ended way of conceiving of concept-object pairs: Cary Grant, Cary Grant,
Lisbon, Lisbon, and so on indenitely. From the point of view from which one
conceives those pairs in that way, one can think of that externally determined
relation, with the open-ended concept-pair-generating disposition guiding the
demonstrative. If some externally determined relation is in fact thereby singled
outas we might note from a combined rst-person/third-person perspective
that relation is then the referent of that relation. It is in this sense that we have a
recognitional concept of the socially deferential name relation.16
16
Evidently we conceive of the socially deferential name relation in somewhat more analytical detail.
It is the product of two contingent relations, one that holds between me and other people, and the other
that holds between them and some person, place, building, object. (We can envisage a further empirical
investigation into the details of these relations.) It could seem that, if I can analyze the socially deferential
name relation into those two components, and if each of those components is conceived in a way that is not
essentially rst person, then my conception of that relation would, unlike my conception of the visual-
memory relation, not be essentially from a rst-person perspective. And that would contradict the account
below of the unity of semantic concepts. But it is not obvious that we have such a third-person analysis. For
the fact that one conceives of the socially deferential name relation as the product of two such relations
hardly implies that one can conceive those relations independently of ones conception of the composite
relation. In fact we have a way of conceiving the composite relationviz., from the open-ended mention-
use perspective on socially deferential proper names. Perhaps the rst of the component two relationsthe
one that holds between me and some group to whose usage I defercan be seen as a sort of abstraction
from the demonstratively conceived composite relation. And as for the secondthe relation that holds
between a population, a name, and the names referentit is hardly obvious that we have a conception of
this relation that does not depend on a conception of individuals using names socially deferentially.
(The apparent circularity is intrinsic to our conception of social meaning; this is why social meaning
appears not to be explicable in terms of individual meaning.) In any event, it is apparently an elementary
fact that we have a rst-person, mention-use, conception of the socially deferential relation. Suppose we
also have an independent third-person conception of that relation. Then we can say that the two
conceptions converge in what they pick out.
12 Objective vs. Subjective Determinacy

of Reference
Finally we turn to the bearing this account has on the determinacy of reference and
truth conditions. It appears quite likely to me that there is no resolution of the
indeterminacy problem in objective terms, and hence that the Quine view is in a
certain straightforward way correct. But it seems to me also that the objective
indeterminacy of semantic properties does not have the eliminative, subversive
consequences it is usually seen to have. For, our ordinary conceptions of semantic
relations are in part subjective; that is, it is essential to those conceptions that
they conceive the relations they pick out from the mention-use conguration.
From the subjective perspective, reference is determinate, while realized by external
objective relations that are themselves, as Quine points out, not unique in their
objective roles.
The import of these remarks will be clearer if we return to the comparison of our
familiar O* and another relation O**. The latter maps that tree onto not a tree but a
tree surface. The correlation of the conceptual role of that tree with the class of trees
as opposed to the class of tree surfaces will from a third-person perspective appear
arbitrary, to whoever nds the basic Quinean inscrutability argument plausible. And
then of course no objective requirements on reference select between O* and O**.
Evidently this is where the subjective perspective matters; for it is O* and not the
relation O** that is discriminated from the mention-use perspective on visual
demonstrative concepts. And of course we must understand discriminated itself
subjectively.
Earlier we remarked that, from the mention-use perspective on proper names, it is
as if there is a rule that generates an open-ended series of term-object pairs. And this
could easily be taken to mean that the intuitive determinacy of reference is based on
an illusion. It is a short step from this to an eliminative view of semantic relations,
that is, to conceding to Quine the main point.
The issue is delicate. There is a straightforward sense in which, it appears to me,
Quine ought to be conceded one of the main points, namely that there is no
principled objective way of marking off O* from O**. That is eliminative of a
philosophical supposition that reference is objectively determinate. But is it elimina-
tive of the intuitive conceptions that one brings to ones initial philosophical reec-
tions about the nature of semantic facts? This is far from obvious. In reecting on the
import of what I have been suggesting, I nd that my basic intuitions appear to be left
in place. There is indeed an objective relation between my visual demonstrative
concept that tree and trees that does not obtain between that concept and tree
surfacesnamely, the relation O*. The question is whether the present account of
what makes O* a semantic relation is subversive or eliminative. I cannot see that the
answer proposed in any obvious way overthrows my intuitive perspective on the
nature of thought and the intentionality of thoughts. On the contrary. It seems rather
intuitively clear that the rst-person perspective on the references of my own

thoughts is somehow constitutive of my conception of reference, of intentionality,
of aboutness. The present proposal explains how this can be so even though
reference is externally determined.17
17
Thanks are due to Paul Horwich and David Sosa for their penetrating and helpful remarks on this
paper at the Lisbon meetings. Thanks also for comments by Jaegwon Kim and Paul Boghossian, and for
very useful extended comments by Barry Loewer, which got me to reformulate a number of points.
14
Transparent Experience and
the Availability of Qualia
Two strong intuitions about visual experience seem to conict radically. One is that
visual experiences have discernible qualitative features, often called qualia. They are
aspects of what it is like to have particular visual experiences, subjective or felt aspects
of experiences. They present themselves as intrinsic and non-relational properties of
visual experience, and they come in great detail. Almost all qualiphiles think of qualia
as introspectible. The competing intuition is that visual experience is transparent:
when you attend to a visual experience as it is going on, you will notice its objects, i.e.,
the things you see or apparently see, including their apparent properties and rela-
tions, and you will notice your (diaphanous) visual relation to those external objects
and properties;1 and, representationists say, that is all.
I endorse the idea that normal visual experience is transparent, both object-
transparent and property-transparent. But I want also to say that there are visual
qualia, and that we can directly discern them. This pairing of views is not usual, but
I hope it will become plausible. Not to be too paradoxical at the outset, I can say that
the resolution will be that we can have two perspectives on our own experiences: in
one mode of attention, visual experience is phenomenally transparent, while in
another visual qualia are discernible.
1 The Standard View of Qualia

I take the standard view to be this: normal visual experiences of the surface property of
being red, which may be a primary or secondary quality, have a distinct intrinsic and
introspectible property that we may call red*. This property is the subjective feel of
those visual experiences, what it is like to have them. Red* is a paradigm visual quale;
and according to most proponents of qualia we can discern it by reecting on our
experience, and thereby be aware of it as a purely qualitative property of experience
and not a property of the ordinary objects of experience. A similar intuition, though
1
Confusingly, these properties are also called qualia by some philosophers (e.g., Dretske, Lycan).
But here the term is reserved for qualitative features of experience itself.
TRANSPARENT EXPERIENCE AND THE AVAILABILITY OF QUALIA
not so initially obvious, reveals shape qualia: while angularity is a feature of things out
there in space, angularity* and its countless forms are visual qualia.
According to the standard view qualia are not in themselves, not intrinsically,
representational or intentional. A way of putting this is to say they are aspects of
sensation and not, all on their own at least, perceptual properties; you have to add
something to qualia to make them into perceptual properties. Like paint on canvas,
qualia are individuated, on this view, independently of their representational
or referential properties, andagain like paintindividuated independently even
of their purporting to represent, independently of their having intentional properties
even in Brentanos sense. We can say that what the standard view defends are
raw qualia. The standard view seems to me implausible, and I will propose an
alternative.
2 Against Qualia
Qualia are not universally loved. They have been seen by many physicalists as a
reactionary woolly-minded doctrine that would impede a fully naturalistic account of
the mental. Others nd them undesirable because of their contribution to Cartesian
internalism, which is supposed to lead to bad thingsskepticism, or disconnection
from the world. These naturalist and anti-internalist complaints about qualia are of
course rather different; but qualia-opponents of both sorts might well endorse the
same remedy.
The project of getting rid of qualia has to some appeared to require philosophical
work. You must argue carefully that the idea of qualia is a mistake: perhaps the idea
of qualia is initially seductive but turns out on investigation to be incoherent (cf. the
private language argument). Quite a few philosophers have recently suggested,
however, that getting rid of qualia takes virtually no work at all: when you get right
down to it and have a good look, qualia dont even seem to be there.
Philosophers who point out that visual experience is transparent, in the above
sense, typically regard this as incompatible with the reality of qualia. Those incom-
patibilists have been called representationists and intentionalists; they hold that the
phenomenology of a visual state can deliver only how the visual experience represents
external things as being. (Cf. Harman (1990); Tye (1995); Dretske (1996); Lycan
(1996).) The position to be rebutted then is this, that however scrupulously you
attend to your normal visual experience, you will not discern anything like visual
qualia; all you will notice is that your visual experience presents certain (apparent)
objects and their (apparent) properties. I note that my current visual state presents
that desk and that piece of paper, their colors, shapes, and spatial relations. I can
attend to nothing else in my experience than my visual relation to some apparently
perceived space, its occupants, and their properties, including sometimes exotic,
merely intentional, objects. Whoever claims to be able to spot qualia is making
things up, or (as Austin reportedly jested about Ayers claim to be aware of sense
data) lying through his teeth.2 My argument will be that this incompatibilist view is
not correct. There is the phenomenon of transparency, but it is compatible with there
being visual qualia as well.
3 Some Inconclusive Points on the Side of Qualia

I should say what would count as success in a defense of qualia. I take it to be a
conceptual matter; and I would be happy simply to show that qualia make sense, that
they are conceivable, and that we apparently know how to discern them in experi-
ence. It would also be nice to be able to refute philosophical arguments to the effect
that qualia are illusory; and I do think it is possible to answer the standard such
arguments. But in this chapter my objective is simply that qualia should make prima
facie sense, and, more specically, that this can be defended against representationist
representations to the contrary.
Some have appealed to the conceivability of inverted spectra in defense of visual
qualia on the standard view of them. I imagine that you see a cucumber as having the
color that I think of as red. Does the conceivability of inverted spectra give us qualia?
The issue is phenomenological. And what I imagine when I think of you and me as
seeing the cucumbers color differently in fact invites a representationist interpret-
ation, i.e., that what I imagine is that you and I see the cucumber as having different
features.3 That is about apparent properties of objects and not about qualia in our
present sense, not about aspects of visual sensation.
Sydney Shoemaker has proposed an elegant defense of qualia that respects the
representationist interpretation of the inverted-spectrum intuition. We imagine
different people seeing a certain object as having differentas it were personal
color properties. For Shoemaker (1994) our conception of qualia in the standard
sense derives from the conceivability of different personal color properties. Those
properties are secondary qualities, dispositions to cause qualia. We perceive those
properties as non-dispositional, rather in the way we perceive heaviness as a non-
dispositional property of objects, even though on conceptual reection it is clear that
it is dispositional. Those personal secondary qualities are what we imagine as
inverted. We do not discern qualia themselves directly: they are inferred from their
making sense of that attractive construal of inverted spectra.
Shoemaker is right in holding that the conceivability of inverted spectra does not
entail introspectible qualia. But his defense of qualia abandons what seems to me
essential to qualia, that they are phenomenally introspectible. My concept of a quale
is the concept of a property that presents itself as a non-relational feature of
2
I owe this anecdote to Laurent Stern.
3
If the proponent of qualia has an independent argument for the conceivability of qualia, then of course
he will also have a direct account of inverted spectra. The point here is that inverted spectra on their own
are not a compelling argument for qualia.
experience. This is not stipulation. I do not know what to make of the idea of a
phenomenal quality that cannot be directly attended to. I have no grip on how to
conceive such a property, for a quale is a way it is like to be in a certain state. (Some
nd it intelligible that a certain conscious quale might be instanced unconsciously.
Even if this makes sense, ones understanding of such a property would be by way of
its conscious accessibility. One would think: that could occur unconsciouslywhere
that is a concept that presents a property via how it is consciously experienced.4)
Let us return to the issue of whether visual qualia might present themselves as non-
intentional, raw qualitative features of visual sensations. It has been suggested that
there are clear examples of raw visual qualia at least in marginal visual experiences.
Ned Block once proposed phosphenes, which appear when you press ngers against
your closed eyelids. And I have been told that recent equipment used for optical
diagnosis can splinter and isolate features of visual experience in such a way that a
person is inclined to count them as purely sensational, undirected, non-intentional
features. That was my informants inclination,5 and I am not skeptical of the report.
As for phosphenes, I do not myself experience those features as non-intentional: to
me they appear to present luminous happenings in strange spaces. But the main
point is that it is unclear that our central concern is affected by such examples. What
might the discernibility of exotic raw qualia imply about normal visual experience?
Clearly not that we can directly discern raw visual qualia in the ordinary case. They
might lend weight to Shoemakers proposal: the occurrence of isolated raw visual
qualia might suggest that similar raw qualia are hidden components of normal visual
experience. But even then it would give us merely a generic conception of certain
properties hidden in ordinary visual experience without specic conceptions of them.
It is not clear that this either makes sense or would explain the point of asserting the
reality of qualia if it did. The simplest and surest way to make sense of qualia is to give
them a form whereby they are discernible in all their specicity.
Here is an example of Christopher Peacockes (1983), offered in defense of qualia.
You see two trees along a road that stretches away from you, one tree 100 and the
other 200 yards away. The trees appear the same size. But the nearer tree occupies
more of your visual eld than the more distant tree. (I should say that this way of
putting things may suggest sense data rather than qualia.) Peacocke takes this to
indicate that there is more to notice in visual experience than its objects, and in
particular that one notices aspects of the visual experience itself. Now I am inclined to
say he is right about this; one does seem here to notice aspects of how the visual
4
Another worry about Shoemakers account of qualia is that there seems to be a difference between
heaviness and personal color properties. In the former case, we have a direct awareness of a qualitative
state; we can conceive of what it is like to pick up something heavy. But according to Shoemaker we do not
have a direct awareness of color qualia. In the former case there is a reason to identify the property of
heaviness with a disposition to produce in one a certain qualitative state, and there is no similar direct
reason to identify personal color properties with dispositions to cause color qualia.
5
Thanks to Professor William Craig.
experience represents its objects. But it is not obvious that the example shows this on
its own. Quite reasonable representationist replies have been made by Hill (in terms
of appearances) and by Lycan (peculiar intentional objects),6 and as against them
I nd Peacockes phenomenon an inconclusive argument for qualia. I am pessimistic
about defending qualia by trying to defeat representationists at their game. Perhaps
no visual phenomenon will all on its own block a determined representationist
interpretation.7 It would be nice to have a way of discerning qualia that makes
representationist construals irrelevant.
4 Phenomenal Sameness
According to the representationist, the complex property of an ordinary visual
experience that exhausts its phenomenal character consists in the subjects visual
relation to certain apparent external objects and their apparent properties and
relations. But a veridical visual experience and a visual hallucination can have exactly
the same phenomenal character, on an intuitive understanding of the latter. Can the
representationist accommodate this phenomenal sameness?
The representationist might say that there really is no sameness here, that there is
no introspectible property or even apparent property that the veridical and hallu-
cinatory experiences have in common. There is, I think, an intuitive point behind
this, and I will return to it at the end. But there is also a strong and compelling
intuition that two such experiences are not merely indistinguishable but also share a
positive phenomenal property; and one question is whether representationism can
properly describe that property.
5 Three Accounts of Hallucinations without Qualia

(a) Merely Intentional Objects
Some representationists, for instance Gilbert Harman (1990), construe the phenom-
enal similarity of a veridical and a hallucinatory experience by invoking intentional
objects. What the two experiences have in common are intentional objects to which
they ascribe the same properties and relations. The hallucination has what I will call a
merely intentional objectit is merely an object of experience with no existence of its
own; perhaps its a Meinong object. I do not deny that the notion of a merely
intentional object is phenomenologically apt, or that hallucinations have intentional
objects can be a helpful manner of speaking or convey something intuitive. The
question is whether there is an inclusive notion of having an intentional object that
6
Hill (1991), pp. 1979; Lycan (1996), pp. 89 ff.
7
Visual blur is a case in point. It seems reasonable to say that a blurred viewing of a scene might be
visually equivalent to a clear view of a scene that is in itself blurry.
can be understood literally and that accounts for phenomenal sameness by virtue of
its applying univocally to both veridical experiences and hallucinations. It would
have to invoke a single relation that holds between visual experiences and both
ordinary objects of vision and merely intentional objects. It may be tempting to
think there is such a relation. Ordinary reective phenomenology can seem to deliver
a relation of presenting, instanced by both <this visual state, the tennis ball> and
<that visual state, the merely intentional object>. Think of it demonstratively: this
(diaphanous) presenting-relation holds between visual states and both ordinary
objects and merely-intentional objects.
But, however intuitive the idea, it is I am sure illusory. For consider this. Had the
veridical visual experience and the tennis ball not stood in a certain externally
determined causal relationa certain optical relationthe experience would not
have been of that ball. And then the intuitive diaphanous relation would not have
held between experience and that tennis ball. But whatever that causal relation is, we
can be sure it has no merely intentional objects in its range. And if there is no
common relation, we can hardly thereby have explained what phenomenal sameness
consists in. A common-sense view of the hallucination is that there is nothing there at
all, that a visual experience that lacks a real object has no object of any sort.
Phenomenal sameness must then be independent of relations to objects both ordin-
ary and abstract.8
(b) Property Complexes

A more attractive explanation of phenomenal sameness is this: a visual experience,
hallucinatory or veridical, represents a complex of external properties and relations
instantiated in certain patterns. These properties are ordinary shapes, colors, spatial
relations, and the like. That is what a veridical visual experience and its hallucinatory
counterpart have in common: they represent the same property complex. Perhaps
this is what some mean by the intentional content of a visual perception. (A related
idea is that the visual experience has the sense or character of a denite description,
which contains a self-referring indexical, etc.)
8
Further possibilities come to mind: (a) There are two relations in the veridical case that extensionally
coincideone causal and the other more ethereal, whose range includes both ordinary and Meinong
objects. Believe it if you can. (b) As in (a) but where the additional relation is deationary. It will be a
struggle to say how such a deationary reference relation can have both ordinary and abstract objects in its
range (given that the states in its domain are all visual experiences). Moreover, as is well known,
deationary reference relations are not counterfactually sensitive in the right way. (c) There is a relation
that somehow supervenes both on that optical relation and on some relation that holds between visual
experiences and mere intentional objects. This is a tall order. Producing the two subvenient relations (the
ordinary causal one, and some relation to abstract objects) will not, it seems to me, explain phenomenal
sameness. If the supervenient relation is to help, it would have to explain phenomenal sameness intuitively,
even if the two subvenient relations do not do so. (Keep in mind that it is essential to the transparency
thesis that the veridical experience should not have both a real intentional object and a Meinong object.)
I do not think such a supervenient relation can be made sense of.
The objection to the property-placing account of phenomenal sameness is this: as

we will see, we apparently have no difculty conceiving of phenomenal sameness
across property-shifts and even of, as it were, property-hallucinations. This I will argue
below. If that is so, in what would phenomenal sameness then consist? Not in
represented external properties and relations. Non-qualitative resources here could
well appear to run out, if my later points about property representation are correct.
(c) Appearances
A natural proposal is that what the two experiences have in common is making true a
certain single proposition of the form it appears that p. But suppose p stands for a
Russellian proposition, containing properties, relations, connectives, quantiers. The
appearance presumably has to be anchored in the properties and relations that
the visual experience actually represents. And then the proposal seems equivalent
to the property complex analysis. It might be said that appearances express Fregean
propositions or senses. But what are they? Suppose they are non-psychological
entities, conditions that the world may or may not satisfy. It is not easy to
understand this unless Fregean propositions entail relations to externally constituted
properties. But then if we can argue (as I propose to) that all such relations to external
properties can shift even though phenomenal sameness is held constant, Fregean
senses will not help. On the other hand, if one means by Fregean sensesin an
unhistorical use sometimes encounteredcertain psychological factors, then we
need to know more. A natural way to think of such factors is as involving in part
the very experiential factors that the representationist denies. I doubt that the
appearance proposal can get very far.
6 How to Spot Qualia

What should we look for? The question concerns normal full-blown visual experi-
ences. As regards them, I cannot make phenomenal sense of visual qualia that leave
everything intentional behind. I can make sense of features of experience occurring
in the absence of any or all of their normal references, whether external objects,
properties, or relations. But this does not mean that we can abstract qualia from the
various ways in which ordinary visual experiences purport-to-represent external
objects, properties, and relations. The central idea will be that visual qualia are
intrinsically directed. They are qualitative features of experience that present them-
selves on reection as purporting-to-refer; but they are conceivable quite independ-
ently of all referential properties, including their representing basic spatial properties.
I will try to make sense of this phenomenologically. The technique of qualia spotting
is fairly simple. One attends to or imagines a visual experience, and conceives of it as
lacking some or all of its actual references, whether objects, properties, or relations,
and then attends to what phenomenally remains.
7 Object-directedness
Imagine a psychology experiment in which you are visually presented rst with a
lemon and then an indistinguishable lemon-hallucination. Not only can you not
distinguish these experiences, but you perhaps have a strong inclination to think they
exactly resemble each other in a certain visually detailed way. A way of putting this is
representational: the two experiences present the real lemon and a merely intentional
object as exactly similar, and that is what makes the experiences indistinguishable.
At the same time, one has a good sense of reality, and so wants to hold that the
merely intentional lemon is nothing at all, and so not something that can resemble
something else. This is reected in how one engages the two experiences. To begin
with, one adopts the perspective of transparent reection. This is the perspective on
visual experiences that initially supports representationism. From this perspective, if
you have full information and a lively sense of reality, you will think different things
about the two experiences. Of the veridical experience you will think that lemon is
real, and of the hallucination that isnt real; theres no such thing as that. The
similarity does not consist in the two experiences objects, and yet they share
something object-wise. One cannot nd this shared property from the perspective
of transparent reection, given what one knows of the facts.
At this point, you can take a less engaged perspective and abstract from the real
object of the veridical experience. The two experiences purport-to-refercall this
their object-directedness. Judging that a hallucination has the property of object-
directedness is identical with judging that it has-a-merely-intentional-object, where
the latter is existentially noncommittal. It is what survives when one takes a skeptical
view of the merely-intentional objectthat is nothing; there is no such object as
that. That object-directedness is a non-relational feature that the hallucination
shares with the veridical lemon sighting. And given this features intimate connection
with the inclination to posit merely intentional objects, as well as with our ordinary
transparent experience of real objects, it seems appropriate to call this common
phenomenal property intentional. This way of attending to experiencediscerning
directednessI will call the perspective of oblique reection, the oblique perspective.
We abstract from the objects of experience, and attend to how the two visual
experiences present their apparent objects.
Consider judging this experience is of that merely intentional lemon-like object.
Even though this is about a visual experience one believes to be hallucinatory, it is in
its way a transparent reective judgment. But it takes little to change that transparent
perspective to an oblique perspective, in fact two steps: (1) One judges that that
merely intentional object is nothingthere is no such relation as the apparently
transparent relation between this experience and that merely intentional object. (2)
One then understands that the experience has the property of its being exactly like
thisand here the phenomenology is the same, or almost the same given loss of
innocence. The property of purporting-to-refer is now understood as a non-relational
property of the visual experience with virtually the same phenomenology as the
seemingly understood property of the experiences standing in the transparent
relation to a merely intentional object. Discard the latter, retain the reective
phenomenology as much as you can, and you have an oblique take on the experi-
ences object-directedness.
It is not as though there is one instance of directedness for each visual experience.
Any normal experience is multiply directed, to various large and small would-be
particulars, parts, and empty spaces. So each such experience has an indenite
number of directed featuresthose that incline one to say that ball, that sharp
edge, and so on.
The directedness of a visual experience supports a strong disposition to refer
demonstratively. But that is not what directedness consists in; for directedness has
an occurrent phenomenology, one that makes the idea of merely intentional objects
so compelling. If I am in the grip of a hallucination, it hardly does justice to the
experience to note that I am disposed to assert that lemon is yellow.
One may be tempted to suppose that demonstrative concepts play a role. The
object-directedness of a hallucination could suggest the involvement of a concept of
the form that object, dressed in visual clothes. This could seem to explain how a
visual experience can be phenomenally directed without a real object or an inten-
tional object. But that idea goes beyond what I nd in the phenomenology of visual
experience. My use of phenomenal directedness is not intended to entail the
involvement of concepts. I take the interface of perceptual organization and proper
conceptualization to be a theoretical and not a phenomenological matter. It may
emerge that the phenomenology of visual experience is built on a partially conceptual
foundation. But the idea of intentional qualia or phenomenal directedness is
intended to be neutral.
In what then does our concept of directedness consist? We cannot dene it, for it
is a phenomenological concept. We have a feel for what seeing a lemon shares with
an indistinguishable hallucination. We can reect on the two experiences in
imagination and discern what they have in common. The proponent of intentional
qualia rejects the ontology of merely intentional objects, which are in any case
useless in explaining phenomenal sameness. We discern a non-relational phenom-
enal quality of visual experience that captures that aspect of experience which
makes it tempting to take merely intentional objects seriously. From this oblique
perspective we step back, withdrawing from the object-involvement of transparent
reection. The non-relational intentional phenomenal quality is found in both
veridical and hallucinatory cases. We can identify similar object-directedness
throughout all normal visual experiences and in hallucinations phenomenally
identical to them. We thereby grasp the general concept of object-directedness.
Such higher-order phenomenological concepts are, I want to say, recognitional con-
cepts, type-demonstratives that pick out repeatable phenomena, in this case, intentional
features of experience.
8 Color-directedness
Visual experience is transparent not only to objects but also to properties and
relationscolors, shapes, relations of size, etc. If you attend to two indistinguishable
visual experiences of a red thing, it is natural to judge that they present the same color
property. That is a standard representationist intuition, and it seems right. The point
to be made is that we can also take a different perspective on color experience, one
that discloses a phenomenal sameness in two experiences that the representationist
should regard as representing different properties. From this perspective what the
two experiences have in common is a certain phenomenal property-directedness,
which is a qualitative intentional feature of experiences, independent of their actual
property references. Property-directedness comes in a vast variety of qualitative
avors and modes. We will consider two broad categories of such phenomenal
qualities of experience, in the present section qualities of color experience, and, in
the following, qualities of shape experience.
Representationists deny that surface colors are identical with dispositions to cause
qualia, for they deny that there are introspectible qualia. On their view, colors may be
intrinsic properties of surfaces or perhaps dispositions to produce in us something
other than qualia. Suppose, with some representationists, that red designates a
physical surface property that is quale-independent, and that certain visual experi-
ences transparently represent red, that surface property. The qualiphile holds that
this is compatible with the conceivability of color qualia. To see this, we consider a
simplied version of Ned Blocks Inverted Earth, which is a Twin-Earth case for color
experiences (Block (1990)). We can conceive of an Inverted-Earthians having visual
experiences that are phenomenally the same as our experiences of red even though
her red refers to a surface property other than red. The reduction of redness to
physical surface properties is beside the point. For the point stands even if we
suppose that surface colors are irreducible non-relational properties of surfaces
over and above their basic physical properties. What the Inverted Earth thought
experiment requires is that we can conceive that those color properties systematically
differ on Earth and Inverted Earth, even though our and our twins visual experiences
are phenomenally the same. I take it that we can conceive of Gods arranging that.
Block takes Inverted Earth to be an argument for the conceivability of color
qualiaand so it is, I think. But I do not see it as giving us a grip on raw color
qualia. What is true is that we can conceive color-related qualitative features of visual
experience that are independent of the surface properties of objects, whatever they
may be. Those features of experience, however, are best regarded (not as raw qualia
but) as property-directed qualia.
I reect on the two experiences, one as it occurs and the other in imagination, with
full knowledge of the external facts. As regards my own experience I judge that it
represents its object as red. And I imagine that my twins experience on Inverted
Earth does not represent its object as being red. At the same time, I consistently
imagine the two experiences as phenomenally the same. How shall we conceive what
they have in common? Two representationist conceptions of the common property
come to mind.
(a) Consider Shoemaker-properties again, in this case shared secondary qualities
of the objects of the two experiences. Shoemaker preserves the phenomenology by
taking such properties to present themselves as categorial properties, that being
corrected when we think about them analytically. What makes adopting this pro-
posal interesting in the present context is that it counts what the two experiences
have in common as a shared property of their objects. As we saw above, however, that
requires hidden raw qualia, and they are difcult to conceive.
(b) It may seem that merely intentional objects crop up not only in hallucin-
ations, but also in the phenomenal sameness of the color experiences we are
imagining. Perhaps what the two visual experiences have in common is a merely
intentional object, of a sort corresponding to properties rather than particulars.
That intentional objectthat would-be colorpresents itself as a property of
surfaces. Can we make sense of this idea? Such quasi-color-properties would
be abstract objects that are unanchored in real resemblance. We may suppose
that it is as if there were such merely intentional objects; but it is another thing
to endorse them. I take our topic to be what the two experiences actually have in
common and not what it is as if they have in common. Rather than endorse such
merely intentional objects, it is better to regard what the two experiences have
in common as having a hope of psychological reality, a property of conscious
experience that is a candidate for being a real resemblance. If raw qualia were
phenomenologically available they would qualify; they are in a general way the sort
of thing we are pursuing.
In the case of the lemon hallucination, I proposed trading having a merely
intentional object for the non-relational being object-directed. So with color qualia.
I suggest that what the two color experiences have in common is a property-
directedness with a certain qualitative character. This qualitative state can in different
contexts present what are in fact different surface properties, perhaps even different
objective colors. And it can occur even in the absence of an object, in a sort of color
hallucination (cf. section 9).
Phenomenal-directedness is phenomenologically closely related to having merely
intentional property objects. If there could be such entities as merely intentional
color-objects, an experience with such an intentional object and an experience with
the corresponding object-directedness would be phenomenally indistinguishable.
Still these are different ideas. If one conceives a visual experience as having a color-
like merely intentional object, ones reection on that experience is in its way from a
transparent perspective. The state of affairs to which one attends is a would-be
relation between the experience and a pseudo-property. One could then judge:
there is no such entity as that merely-intentional-property, and yet this experience
and the experience of my twin are still like that. One may regard the two experiences
as sharing a property-directed quale, a way of presenting surface colors. That is how
it goes for any sort of phenomenal intentionality. One may get a grip on it by
attending as if to a merely intentional object, judging that there is no such entity
there, and then noting that one can conceive of a non-relational phenomenal quality
of the experience thus: it is like that, it has that directed color quale. These locutions
should be understood with the phenomenon in mind; they are not technical terms,
but stand for how one conceives the experience when one rejects the reality of the
would-be color-property. Again, I say that we conceive of that intentional feature
from an oblique perspective because the feature is non-relationalone reects on
the experiences intentional and phenomenal features and not on its objects; and one
is from that perspective not immersed in the experience in the manner of transparent
reection.
A certain question about the relation of color qualia and the visual experience of
shapes naturally arises at this point. In reecting on ones color experience one
normally will also be visually presented with shapes; and so far we have not
considered how we might reect on shape experiences obliquely. In our thought
experiment about Inverted Earth, we will apparently have to engage in a bi-perspectival
frame of reection, attending transparently to our shared visual experiences relation
to shapes and attending obliquely to the shared color experience. How might we
conceive that?
Consider the following simple limiting case, in which shape plays a vanishingly
small role. Imagine looking at a uniformly colored wall that lls your eld of vision,
and also imagine your twins having a phenomenally indistinguishable experience, in
the presence of a wall with a different surface property. Conceive the two experiences
as sharing a merely intentional quasi-color-property. This apparently captures the
shared phenomenal color-quality of the two experiences from a transparent perspec-
tive. Then imagine judging there is no such thing as that property, and also
judging that the two experiences are phenomenally indistinguishable color-wise.
In conceiving what it is like to have each of the two experiences, conceive of it
non-relationally: it is like this. Now on the face of it what they have in common is a
property-directed qualitative state. As we suggested above, this color*-property-
directedness is just that non-relational feature of the experience that replaces the
intuition of a shared merely intentional quasi-color-property. When one then
regards the two experiences as presenting the wall in the same way, one abstracts
from a conception of them as experiences of a certain color-related property.
When one reects on a more complex and normal visual experience of shapes, the
imaginative project becomes more demanding, attending obliquely to color-
experience and transparently to the experience of shapes. This may in the abstract
seem to require two modes of attention that do not harmonize. But perhaps this
formula will help: the two visual experiences represent a given shape; they moreover
present that shape in a certain color-wise way, which we can conceive obliquely, as if
noting the mediation of a color lteralthough this is not a perfect analogy. If we

shift back and forth determinedly, the two properties of the experienceone rela-
tional and the other notphenomenally coincide. But next on the agenda we have a
more radical thought experiment, which should eliminate the strain of coordinating
different reective perspectives on color and shape aspects of experience.
9 Isolated Brains
Consider how visual experiences represent spatial properties. Can we hold a visual
experience constant and vary the properties that spatial* phenomenal features pick
out? We require a more radical imaginative possibility than Inverted Earth. The
banal but useful thought experiment involves an isolated brain, a phenomenal
duplicate of oneself isolated from past and present ordinary contacts with our
world. I will be content if you grant at least a supercial coherence to the thought
that my isolated twin-in-a-vat has visual experiences exactly like mine. It seems to me
that this is fully coherent, and, more directly to the present point, that such an
isolated brains experiences would not refer to, represent, spatial properties and
relations. About reference I am inclined to be completely externalist. The isolated
brain does not stand in the right relations to spatial properties for his perceptual
states to refer to them. In fact I do not think we are forced to suppose that the spatial*
features of the isolated brains visual experience pick out any properties or relations
whatever. If that is right the isolated brain is subject to what, from the outside, may be
conceived as a radical property hallucination.
We hold something qualitative constant as we imaginatively shift or remove spatial
references, i.e., the rich phenomenal qualities of visual experience. Shape-presenting*
qualia might stand for properties other than shapes, or no properties at all. Those
phenomenal qualities have a sort of directedness, like color qualia; they are intrin-
sically intentional. As I conceive the isolated brains mental life, it iseven if devoid
of referenceas intuitively replete with intentionality as my own. Suppose its visual
states fail to refer not only to spatial properties and relations but to any properties or
relations. As in the familiar case of object-hallucination, speaking of merely inten-
tional objects here conveys something intuitive, where by objects we mean ersatz
spatial properties and relations. But again, taking those intentional objects seriously
seems unrealistic; however intuitive they may be, they are just manners of speaking.
Better to conceive the spatial* features of the isolated brains visual experiences as
having property-directedness and (for spatial* features that in ordinary vision
represent relations such as betweenness) relation-directedness.
How then do we conceive these complex features? We again note that merely
intentional objects are inessential to the phenomenology: we make sense of the
isolated brains spatial* experiences being phenomenally the same as ours when
we judge that there are no such entities as those merely intentional property and
relation-like objects. We can think: an isolated brain could have an experience just
like this, with this spatial* intentional phenomenal quality, that intentional quality . . .
We have to keep in mind that our reective recognitional conceptions of these
qualities are formed at varying degrees of specicity and generality. We are able to
conceive many rather different spatial* features of visual experience as all being
curvy* qualia, and many more as all being spatial* qualia. These pattern recogni-
tional abilities have complex interrelations, up, down, and sideways, and we dont
know how we do it. That is how a quality space is, a brute psychological phenom-
enon. For the isolated brain and his normal twin, what it is like to have their
experiences is the same; the intentionalized qualia of their visual experiences are
the same. We may judge: this experience could be just like this even if it were not
an experience of those particular objects and did not present its particular objects
as having those properties, including those spatial properties. This is what the
experience of my isolated twin would be like. Again, this perspective on visual
experience is from what I am calling the oblique perspective. The non-relational
features one holds constant from this perspective deserve to be called both inten-
tional and qualia.
10 Qualia: Inferred or Presented?

The question arises whether intentional qualia depend more on theory than on
phenomenology, a possibility that could undercut the idea that such states present
themselves as features of experience. The thought may go like this.
The idea of intentional qualia depends on our denying that the visual states of isolated brains
represent spatial properties and relations. That denial depends on theory, namely, radical
externalism about reference. But then intentional qualia are creatures of theory. And if so, how
can we conceive them phenomenologically?9
The idea perhaps is this: a philosopher who does not hold an externalist theory of
reference could not think he discerns these intentionalized qualia in his visual
experience. As an internalist about reference, he holds that the references of his
visual statescolor properties, shapes, spatial relationscannot shift across environ-
ments. His account of phenomenal sameness then would presumably be the
property-complex account. Now this philosopher is going to have to say, I think,
that my account of the phenomenology is just mistaken. He will perhaps understand
why I think I can discern intentional qualia, certain non-relational features of
experience. But he will count this as an illusion, stemming from my accepting
externalism about reference and thento secure consistencyinventing exotic
intentional qualia to play the role of being whatever it is that I imagine as constant
across the supposed shifts in reference. The fact is that there are no such shifts of
9
Thanks to Barry Smith, and to Scott Sturgeon, for raising versions of this question.
reference, he says, and, once we acknowledge this, those exotic properties will
disappear.
Now suppose one adopts agnosticism about reference-externalism. One will think
that something phenomenologically available can be held constant through the
various environments, including the isolated brains environment. If we are neutral
about externalism we are agnostic about the nature of that factor. If we accept
externalism about perceptual reference, we will perhaps regard those features of
experience as intentional qualia. And if we accept internalism about the property-
references of, say, spatial features of experience, we will presumably construe those
features as intentional property complexes, in the sense of 5(b).
The challenge posed by the opening question can be met. Theory does have a
bearing, it is true. But theory does not create the phenomenology. From a neutral
position there is a certain phenomenology of perceptual experience. What is missing
from the neutral position is a conception of the nature of what is thereby presented. If
ones intuitions are externalist, one should regard the factor that is held constant
across the various environments as a non-relational phenomenal feature of experi-
ence. And if one shares my skepticism about the availability of raw qualia, and nds
the idea of phenomenal intentionality coherent, one should regard the factor that is
held constant as consisting in intentional qualia.
The objection we began with proposed that, because of the essential role of
externalism about reference, intentional qualia are theoretical features of experience
and cannot be discerned directly. We now see what is wrong and what is right in that.
Those features are discerned directly, in all their detail, regardless of what theory we
accept. What adopting an externalist theory of reference engenders (putting aside the
previous paragraph) is an understanding of what those features are, that is, inten-
tional qualia.
11 Intentional Qualia and Concepts

On the present conception, intentional qualia are not (merely) features of visual
sensation, whatever that might mean. We might call them percepts, not in a
theoretical sense, but simply as labeling what is phenomenally available. And the
idea does not exclude the possibility that, in some theoretical sense of concept,
intentional qualia involve concepts. Indeed I have been told that thinking of inten-
tional qualia as conceptualized qualia makes the idea of directedness clearer. But it is
easy to be pulled in the opposite direction. For there is also the thought that percepts
are inputs to certain concepts, those concepts being constituted by subsuming
percepts, whatever that might come to. In fact I nd this idea quite attractive.
As a piece of phenomenology, intentional qualia may help to explain why some
nd the idea of the narrow content of thoughts plausible. Narrow content must
somehow be intuitively intentional. Narrow intentionality cannot be truth
conditions; for they presuppose reference to properties, which is a matter of
externally determined relations. (Even if truth conditions are characters or the like,
they presuppose property-references.) So the idea of thoughts having narrow con-
tents may appear to make sense because (1) intentional qualia are phenomenologic-
ally real, (2) they are partially constitutive of perception-based concepts, and
(3) concepts that are not directly perceptual are intentional via conceptual connec-
tions (Loar (2003) [Chapter 15 in this volume]).
12 Three Frames of Mind: Transparent

and Oblique Perspectives
(a) Unreective Transparency
This is how ordinary in-the-world unselfconscious experience is. One sees the object
directly, phenomenologically speaking, i.e., one doesnt see the object by attending to
the visual experience, as one might see the man by seeing the coat. But we can step
back and judge that this ordinary transparency is the upshot ofas we might say
exercising or undergoing visual percepts. When I unreectively see a tennis ball,
I undergo a corresponding percept. By undergoing that percept I have not thereby
attended to it.
The following two frames of mind are the main point; they involve different ways
of framing attention to a visual experience. I assume that attention involves some-
thing like concepts. The difference between the two frames of mind stems from the
concepts employed in the two ways of attending.
(b) Transparent Reection

This is the reective posture that the representationist regards as the only one
possible. One reects on a visual experience, and notes that it stands in a diaph-
anous visual relation to a certain external state of affairs that consists in the
instantiation of external properties and relations. One exercises perceptual con-
cepts (which we may think of as incorporating visual percepts), attending to
external objects and their properties and relations, as well as to ones visual relation
to those objects and properties, and not attending to the experiences non-
relational phenomenal features.
So we understand reective transparency not only as a matter of attending to an
experience and its transparent visual relation to its objects, but also as involving
certain visual perceptual concepts. These perceptual concepts are intimately con-
nected with visual percepts, by virtue of the latters being as it were embedded
lling the empty slotin the former, concepts of the form that object, that color, and
that shape. The rst is what we can call a singular perceptual demonstrative concept.
The latter two we can call visual recognitional concepts, or recognitional type-
demonstratives (Loar (1991b)).
The structure of a simple reectively transparent attention may then be represented

like this:
[Attention: concept of this visual experience + concept is of + visual demonstrative
concept (that object, that property).]
NB: The concepts mentioned following attention are complements of the attitude of
attending: they are the concepts employed in the attending and not the objects of that
attending.
(c) Oblique Reection
There is another manner of framing attention to a visual experience. It is the upshot,
e.g., of the holding-constant-while-imaginatively-shifting-references routine, in
which ones attention is turned reectively on what is held constant.
[Attention: concept of this visual experience + concept has + recognitional concept
of a certain qualitative perceptual property] namely the perceptual property that
this experience has in common with those that are phenomenally exactly similar,
including those of the isolated brain. (Again the terms in italics pick out comple-
ments of attention and not its objects.)
It appears to me evident that we are capable of this way of attending to a visual
experience. Discerning qualia is one of the two ways of framing attention to ones
visual experience; it is compatible with what weve been calling the transparency of
experience. The latter is not the only option, but is simply what you nd when you
adopt that attitude (of reective transparency) which is the dominant untutored
manner of reecting on ones visual experience. The argument of this chapter has
been that the less obvious attitude of oblique reection, and the intentionalized qualia
thereby noted, are also fully available.
13 Final Observations
We have characterized phenomenal sameness as it arises from the oblique perspective.
Many philosophers though have seemed to take it for granted that phenomenal
sameness is found from (what I have been calling) the transparent perspective, the
perspective from which one attends to the objects of visual experiences whether they
be real or apparent. But this transparent account of phenomenal sameness requires
objects that are hard to take seriously. These are of two sorts, what Ive been calling
merely intentional objects, and sense data. A prevalent assumption of the sense-data
tradition required abandoning ordinary transparency in the veridical case; but that is
phenomenologically unpersuasive, for phenomenal transparency in the veridical case
is difcult to deny. The substitution of merely intentional objects avoids that
implausibility, but still posits strange objects to account for hallucinations; but such
objects offend against common sense, which says there is nothing there at all. It is an
illusion that phenomenal sameness of veridical and hallucinatory experiences can be

discerned from the transparent perspective; in the case of a hallucination one
should simply judge that there is nothing there. If that is so phenomenal sameness
can be intelligibly discerned only from the oblique perspective; and it consists in
resemblance in intentional qualia.
If a philosopher agrees about merely intentional objects and sense data, and yet
cannot nd in his experience what I call oblique reection and intentional qualia, he
will have to say that veridical and hallucinatory experiences have nothing in common
that can be pointed to in experience. Two experiences would be indistinguishable
from the rst-person perspective even though one experience presents an actual state
of affairs and the other presents nothing at all. In denying that there is a positive
phenomenal sameness here, this is related to disjunctivism (cf. Snowdon (1981);
McDowell (1986)). If it is right, and all that is available is the transparent perspective,
it would be reasonable to treat the two experiences quite differently.10 But the sense
of positive phenomenal sameness is compelling, will not go away, and begs for a
coherent rendition. That is provided by the idea of intentional qualia, and the oblique
perspective that reveals them.
Should the reality of intentional qualia raise fears of our being isolated from the
world? Skepticism itself is the product of epistemological views, and the right way
with it is to attack them. There is no point in eviscerating the mind to x epistem-
ologists mistakes. As for our sense of connection to the world, it is the upshot of a bit
of phenomenology, that is, the transparent perspective on experience. Is this sense of
connection compromised by the availability of the oblique perspective? That is a nice
complex question. Worries about the divergence of the lived world and sciences
world arise, in part, from the availability of a phenomenal realm that has its own
integrity. Whether such worries can be assuaged is still an open question. But it is
rather implausible to answer it by denying the availability of phenomenal sameness.11
10
But suppose the opaque perspective were not available. Then perhaps a coherent conception of
isolated brains would not be available. And in that case the reason given in the text for rejecting the
property-complex account of phenomenal sameness would not be available. Ned Block has suggested to me
that McDowells disjunctive view seems to overlook the view that what the two experiences have in
common is that they represent the same property complex. It might be replied that property complexes
are not particular states of affairs, and that the disjunctive view can be taken to deny merely that
hallucinations involve relations to phenomenally available states of affairs.
11
Thanks to Ned Block, Michael Martin, Nenad Miscevic, and Gabriel Segal for very useful remarks on
earlier drafts.
15
Phenomenal Intentionality as
the Basis of Mental Content
The mental or psychological content of a thought is a matter of how it conceives

things; that is what we hope to grasp, at least approximately and in part, when we try
to understand another person. We want to know not merely what her thoughts
represent as it were impersonally, but also how they represent things to her. A persons
thoughts represent things to herconceive thingsin many ways: perceptually,
memory-wise, descriptively, by naming, by analogy, by intuitive sorting, theoretic-
ally, abstractly, implicitly, and explicitly. These various manners of conceiving
have something in common: They have intentional properties, and they have them
essentially. The conceiving cannot be pulled away from the intentional properties,
in our ordinary reexive understanding of them. But this creates a problem. It is
not unnatural to suppose that conceivings are in the head. So if the intentional
properties of conceivings are essential to them, intentionality must be in the head as
well. The problem is that there are fairly compelling externalist reasons to the
opposite conclusion. Yet it seems to me that there must be something right about
the internalist thesis and the intuition that backs it, something quite basic to our
understanding of the mental. This is what I will try to make coherent.
Mental content has often been supposed to be what oblique that-clauses capture.
That would lead directly to a considerable difculty in the idea of internal intention-
ality. For that-clauses capture references, and the references of our outward-directed
thoughts areaccording to the most believable theories1determined by external
relations. There is a quick way to deal with this difculty, and I hereby adopt it.
Mental content is in fact individuated independently of that-clauses. This seems to
me to follow from the semantic behavior of that-clauses together with a basic
constraint on mental content or ways of conceivingwhat is often called Freges
constraint. Something counts as a judgments content only if we cannot make sense
of a persons judging both it and its negationunless she in some way compartmen-
talizes those judgments. You can make sense of a persons judging that Paderewski
plays well and at the same time judging that Paderewski does not play well even
1
Classic texts are Kripke (1972a), Putnam (1975), Burge (1979), (1982).
PHENOMENAL INTENTIONALITY AS THE BASIS OF MENTAL CONTENT
though the two beliefs have the same reference and draw on the same public name.
Nothing semantic distinguishes the ordinary meanings of those that-clauses except
the negation. Given Freges constraint, this means that mental content is individu-
ated more ne-grainedly than the interpersonally shared oblique content of certain
that-clauses. I rather think the phenomenon is all-pervasive, that for virtually any
that-clause a similar underspecication of content can be shown.2 A closely related
point is this. Consider any perceptually nuanced conception of mine. I can invent a
neologism to express that conception, and use it in self-ascribing that-clauses. But the
that-clauses are then secondary: What matters is my reexive grasp of the perceptual
concept, its psychological content. That-clauses as they are standardly used appar-
ently capture too little information, even on oblique interpretations, and that infor-
mation is not of the right sort: That-clauses are more about socially shared concepts
and their referents than about the various perceptually based and other ways in
which thoughts conceive their referents. They are not especially psychologically
informative.
If mental content is accessible and is not literally expressed by that-clauses, how
does it get conveyed? Typically in the gaps between the words. Suppose you say that
Guido thinks that the woman over there resembles Greta Garbo, and you say this
while he has the woman in full view. I understand you to mean that Guidos thought
picks her out visually. That visual mode of presentation is a constituent of (what
I mean by) the mental content of his thought. That we might invent a word to capture
just that highly specic visual mode of presentation, and insert it in a that-clause, is
not interestingnor is it even particularly interesting that we can say the person he
is looking at. Guidos thought involves, among other factors, a visual mode of
presentation, and we conceive it independently of what is mentioned in that-clauses.
But this is neutral between internalist and externalist views of mental content. For it
is compatible with neo-Fregeanism,3 the idea that for example perceptual modes of
presentation are to be individuated object-dependently and property-dependently.
The present point, though, is simply to put space between mental content and that-
clauses. Our conceptions of mental content have a life of their own apart from
that-clausesthere are for example perceptually based demonstrative concepts, as
we intuitively understand when we think about Guido.4 Conceptions of mental content
in the analytic tradition have tended to be phenomenologically impoverished, largely
2
As Kent Bach puts it: every case is a Paderewski case. Loar (1988a), (1988b) [Chapter 8 in this
volume], (1987b) [Chapter 9 in this volume].
3
See Evans (1982); McDowell (1986).
4
Your ascription of Guidos belief does not even implicitly convey an exact conception of his highly
specic visual conception; we rarely have an exact conception of anothers mode of presentation. What is
conveyed rather is a type of mode of presentation to which Guidos precise visual concept belongs. It is the
specic modes of presentation that individuate beliefs; but we typically merely gesture in their direction.
The theory that types of modes of presentation are implied contextually in the gaps between the worlds in
that-clauses (so to speak) was rst introduced by Stephen Schiffer (1977). He has subsequently argued
against his own theory (Schiffer 1992), but I am not persuaded.
because of the emphasis on language and reference. And when we turn to the
phenomenology, as I will try to show, we do get a grip on internal intentionality.
A compelling intuition about mental life sees it as a stream of conscious thoughts,
feelings, and perceptions. This is not all or even perhaps the larger part of the mind.
But it is central to our founding conception of the mental. When we conceive these
various conscious states, moreover, we conceive them as intentional. The stream of
conscious thoughts, memories, and perceptions seems to have a life of its own that is
constituted independently of its external environment. This is intuitively supported
by an obvious thought experiment. Apparently I can imagine what it is like to be an
isolated brain that is a physical duplicate of my own brain. What I imagine includes
not just that brains nonintentional phenomenal states, its utters and pains, but also
states and events that correspond to my own outward-directed thoughts and percep-
tions. I imagine my isolated twins states and events as subjectively representing
things in the same manner as those thoughts and perceptions of mine. The intuition
supports the view that my own mental streams intentional featureseven those of
its outward-directed thoughtsare constituted independently of my actual situation
in the world. (Note well that I have said intentional features and not references.)
This is not to say that the seeming imagining of the isolated brains intentional states
proves there is such a thing as internal intentionality. But it surely makes one wonder
if we can make sense of the idea, make a case for its coherence.
The reader will reasonably want to know what is meant by intentional if not
referential. Let me say for the time being that the internal intentionality of percep-
tions and thoughts consists in their apparent directedness, in their purporting
subjectively to refer in various complex ways. This is, according to what follows, an
ineliminably phenomenal feature that is shared by my and my isolated twins states
as I imagine them.
Why care if a phenomenological conception of internal intentionality can be made
sense of? It is there for the noticing; and we have a wrong philosophical view of our
intuitive conception of the mind if we persuade ourselves in the abstract that internal
intentionality cannot be there. Does this matter to common-sense psychological
explanation? Yes of course. There have been strenuous efforts to explain how causal
and social relations to distal objects can be essential to psychological explanations of
behavior; and the resulting theories are, in my view, more than a little strained.
A consequence of making sense of internal intentionality is to vindicate a classical
internalist view of common-sense psychological explanation, or at least to make it
coherent. Still the main question is more basic than the explanation of behavior.
It concerns whether mental properties as they are in themselves have merely contingent
connections with behavior and environment. That is hardly a small matter if we are
interested in what we are: We have inner mental lives.
The intuitive idea that the intentionality of outward-directed thoughts can be
internally determined has run into serious trouble, for a certain externalist con-
ception of intentionality has considerable intuitive force. Tyler Burge, as much as
any other philosopher, has made a powerful case for externalism about mental
content.5 This he has done by arguing thatto put it in a way more abstract than
histhe semantic resources of the analytic tradition, whereby intentionality con-
sists in the truth conditions and satisfaction conditions of thoughts, cannot support
internalism about intentionality. I agree with this. But I draw a different conclusion:
What matters to intentional internalism does not depend on those classical truth-
conditional factors. Something theoretically novel (though familiar in experience)
needs acknowledging. My homage to Burge in this volume [the volume in honor of
Burge in which this paper originally appeared] will be expressed by my being driven
to extremes.6
While the internalist intuition appears to me correct, the core of current externalist
theory also appears correct. So the core of externalist theory must be compatible with
intentionalitys being an internally constituted feature of mental states. Externalists
are right about the reference and truth conditions of thoughts. But despite vivid
appearances to the contrary, intentionality does not presuppose reference and it is
not externally determined. That is the idea I will try to make sense of.
1 Externalism about Intentionality

Externalists about intentional mental content regard the opposing position as con-
ceptually incoherent.7 Here is the externalist reasoning, as I try to put it straightfor-
wardly to myself. (It is worth noticing that this externalist line of reasoning does not
presuppose that mental content is as that-clauses capture it. The earlier point in
denying this was not to confute externalism directly but to open up mental content as
it were to phenomenological access.) The externalists rst premise is that thoughts
can be intentional or directed, can purport to refer, only by presupposing actual
5
See, for example, Burge (1979), (1982), (1986).
6
I engaged some of Tyler Burges arguments in an earlier paper (Loar (1988b)) [Chapter 8 in this
volume]. There I made four proposals: (i) Burges arguments depend on the supposition that the
psychological content of the predicative aspects of thoughts is identical with what obliquely interpreted
that-clauses capture. (ii) That-clauses do not capture psychological contentany that-clause can apply by
virtue of different psychological contents. (iii) Thoughts with the same psychological contents will in
different contexts require different that-clauses to express them. (iv) Psychological content can be
understood in terms of realization conditions.
It has for a long time seemed to me that the fourth proposal is incorrect, because it relies on semantic
resources (basically possible worlds) equivalent to those mentioned in the text as referential. As will
become clear below, any such proposal is vulnerable to further externalist argument about reference
putting aside issues about that-clauses. Another paper of mine (Loar (1987b)) [Chapter 9 in this volume]
proposed a notion of subjective intentionality that is nonreferential and hence different from that of
realization conditions. My current account of subjective intentionality is different again from the 1987
account, which was not phenomenal in the same way.
7
That is, the externalists who pose the most serious threats to the internalist conception of intention-
ality. One can notionally conceive of an externalist position that holds that internalism is coherently
conceivable but that, as a matter of fact, intentionality consists in externally determined properties.
references. The externalist may grant that a thought can purport to refer to something
external even if it does not succeed in referring, even if there is so to speak nothing
there. Even so it must represent what it purports to refer to as such and such, as
having some property F. And so it must succeed in referring to that property. The
second premise is that such reference is constituted by externally determined
causal, social, and so onrelations. If the premises were both correct, no sense
could be made of the isolated brains having the same intentional states as me;
intentional mental content could not be internally determined.
Almost everyone agrees that any singular concept may fail to refer and still have
the intentional content it would have if it had succeeded. This holds in the most
obvious way of ordinary denite descriptions. The oldest dolphin in Andorra
purports to refer, and for all I know it fails. It also holds of perceptual-
demonstrative concepts: I may exercise a visual demonstrative conceptthat
horrifying animaland yet be hallucinating. Despite their failing to refer, both
concepts intuitively have full intentional content: Each purports to point, quite
specically, to an object, even though the world does not put an object in its way.
But the presupposition thesis is satised because, on the face of it, the intentionality
of singular concepts depends on what those general concepts refer to: dolphin and
animal refer to kinds or properties, and according to the externalist this consti-
tutes, at least in part, the singular concepts intentionality. The externalist may
of course allow that a property-concept can itself fail to refer, even while fully
purporting to pick out a property. But this must be grounded in further concepts
that actually refer and that hence stand in externally determined relations to
externally constituted properties. Another way to put the externalists (often impli-
cit) point about intentionality: Thoughts cannot purport to refer unless they
impose success-conditions, or satisfaction-conditions; and these depend, however
indirectly, on reference to objective properties. Externalism about intentionality
assumes, on this account, that intentionality presupposes reference.
The externalists second premise is that referring to and connoting external
properties consist in externally determined relations between concepts and proper-
ties, at least for concepts that purport to be outward directed. Externalist positions
about reference may diverge: Some regard all basic reference relations as non-socially
mediated causal relations to things, and others, Tyler Burge famously, as including
social relations to the usage of others. I accept Burges view to this extent: We cannot
realistically deny the role of social relations in the mediation of much ordinary
reference. How much farther we should go is not clear. Suppose ones concept
animal derives its reference socially, from biologists conceptions. It is not clear
whether their more basic concepts might ultimately be determined purely personally
by nonsocial perceptual and conceptual-role relations. But this issue is beside the
present point.
In agreeing with externalism about reference I accept this: Basic property-
references and property-connotations are constituted by relations that, at least in
part, are externally determined, whether socially or not. In the standard debate, this
concedes substantial ground to externalism about intentional content. But it does not
imply it; for it does not imply that intentionality presupposes reference.
On the externalist view of mental content, another brains perceptual states and
thoughts can be intentionally equivalent to my externally directed perceptual states
and thoughts only if we share at least some of my actual external references. The
equivalence would require an overlap in property-reference. If the isolated brains
perceptual states and other concepts do not pick out some of the same properties as
my externally directed perceptual states and thoughts, then none of its mental states
can have the same intentional content as my outward-directed mental states. This is
the externalist premise to which we must reply, for we agree that property-reference
consists in externally determined relations.
To sum up the externalist line of reasoning: (i) mental content is intentional;
(ii) intentionality presupposes reference; (iii) reference, for outward-directed thoughts,
consists in externally determined relations (especially to kinds and properties);
therefore (iv) my outwardly directed thoughts do not have internally determined
mental content or intentionalitythat is, the mental content of my outward-directed
concepts cannot be shared with an isolated brain. I must emphasize the distinction
between externalism about reference and externalism about intentional mental
content: I accept the former and deny the latter.
2 Conceptual Roles and Mental Content

What we seek is a conception of mental content that is available commonsensically
and that is internally determined. This could seem to be easily delivered by an
established idea, namely, that we implicitly individuate thoughts in terms of their
conceptual roles. A thoughts conceptual role consists in its inferential and probabil-
istic connections with other thoughts, desires, and so on, and with perceptions.
Conceptual role theory standardly avoids appeal to intentionality. That can seem
to be a considerable advantage, for it permits an attractive clarity: The horizontal and
vertical aspects of mental content are factored out, that is, internal-explanatory
conceptual role and external-referential intentionality. Externalism about intention-
ality is acknowledged, but internalism about mental explanation is nevertheless
defended.
Now I am quite sure that conceptual role is central in individuating thought-
contents. How a thought conceives things must consist in part in its conceptual
commitments. That is essential to what I will propose. But conceptual role on its own
seems to me inadequate to explain our ordinary understanding. Conceptual roles are
too blank to constitute internal mental content as we conceive it. Thinking is
something livelythere is something that it is like to engage in it. So phenomeno-
logical reection on thinking hardly conceives its properties in purely dispositional
terms. But perhaps we might add phenomenal states to conceptual rolesand would
this not give us the internal liveliness? We might then think of perceptual states and
other phenomenal states as among the realizers of conceptual roles, or somehow
intimately connected with such realizers. The liveliness of thinking in general would
stem from perceptual states, linguistic states, various forms of imagery, with concep-
tualizing supplied by their connections within an interlocking network of conceptual
roles. And who knows what innate conceptual structures there might also be into
which perceptual states could nicely t.8
Although the picture thus vaguely put is doubtless on the right track, it does not
seem to me to promise an internalist conception of mental content. For we appar-
ently lack appropriate nonintentional conceptions of perceptual states. We can
hardly peel the phenomenal aspects of vision away from its intentionality; we just
do not have nonintentional conceptions of visual elds or the like. Or try as I might
I cannot muster such conceptions. Visual perception is phenomenologically focused
on objects, spaces, and their properties; there are no pure visual sensations that might
add nonintentional life to conceptual roles. If the externalist is right about intention-
ality, a phenomenal elaboration of a conceptual role theory will not yield ordinary,
intuitive conceptions of internal mental content.
The externalist might in any event complain that the project would be futile even if
we had purely phenomenal and nonintentional conceptions of perceptual states.
Internal goings-on would not on their own constitute a mental life: For they would,
phenomenologically, not look out to external space. It would be in McDowells
dramatic phrase all darkness within.
I can envisage a spirited defense by the conceptual role theorist. Both of the
foregoing objections ignore the availability of a deationary notion of intentionality,
that is, of reference and truth conditions. It may well be that we cannot conceive
visual qualia nonintentionally. But this could have the following explanation.
We cannot conceive ordinary visual experience unconceptualizedthat is, unless it is minim-
ally conceptualized by object-concepts, spatial-concepts, and so on. This conceptualization
may be understood in terms of something like conceptual role or conceptual structures as long
as we also grant the conceptual role theorist what he is classically entitled to, namely, a
disquotational or deationary notion of reference.9 To conceive a way of conceptualizing
visual experience simulationally will then employ object-conceptions, and so on. And when we
reect on them we can hardly avoid, as it were, disquoting. They are our concepts and we can
hardly think of them in nonsimulational objective terms.
This explanation is tantamount to proposing a deationary internalist theory of

intentionality, which it may be argued quite adequately accounts for the phenomenology.
Let us have a look at such a theory.
8 9
See Spelke (1995). Field (1994).
3 An Internalist Theory of Intentionality Rejected

What we might call the standard internalist conception of intentional mental content
denies that the reference of basic concepts is constituted by externally determined
relations (the second externalist premise). At the same time, it does not question that
intentionality presupposes reference (the rst externalist premise). On this standard
internalism, certain basic concepts express or pick out properties and relations inde-
pendently of causal relations between the mind and those properties. This might for
example be held of our concepts of shape and spatial relations, that is, that they express
spatial properties and relations without externally determined referential mediation.10
And a similar view may be taken of concepts of categorial properties such as causation
and physical-objecthood. Externalists are dismissive, and there are two rather different
objections. The rst is the thought that the externalist thought experiments of Kripke,
Putnam, and Burge11 show that reference is environmentally and hence externally
determined. The second objection is more basic, and harder to get a grip on: Such
internalists are accused of holding implicitly a magical theory of reference (as Putnam
called it), that the mind somehow grasps externally constituted properties independ-
ently of natural relations to those properties. Let us see how a reasonable internalist
about reference might respond to each objection.
As a preliminary we classify concepts according to the apparent role that wide
contexts play in determining their references. By reference I mean not only the
reference of singular terms and predicates, but also the truth-conditional contribu-
tions of logical connectives, and more generally of all concepts. So we have:
A. Wide concepts, whose reference is determined by externally determined rela-
tions to external contexts. These include singular indexicals and demonstratives, and
kind terms such as water, tiger, arthritis. Whether a concept belongs to this type is
decided by thought experiments that shift contexts between Earth and Twin Earth,
between arthritic twins, and the like.
B. Narrow concepts, for which reference is context independent, that is, independ-
ent of contexts that transcend internal conceptual role and the like. Paradigms are
the logical connectives. If a connective has the conceptual role of and or all it eo
ipso expresses conjunction or the universal quantier. There are no Twin-Earth
reference-shifts for logical connectives. Presumably mathematical and modal con-
cepts belong here as well. (It will be convenient, if somewhat inelegant, to include in
group B indexicals that pick out internal states: this sensation, this thought; for
their referring arguably does not consist in externally determined relations.)
10
It is not uncommon for philosophers simply to conate concepts and properties.
11
Kripke (1972), Putnam (1975), Burge (1979).
C. Debatable cases, predicates that do not uncontroversially or determinately

belong to class A or to class B. These are the concepts already mentioned: cause,
physical object, spatial concepts, and perhaps others. As I have found among
philosophers in conversation, intuitions about such concepts are far more open
than they are about water and the like.
A conservative internalist strategywhich I will describe but not endorse
proceeds as follows. Assign concepts in class C to class B. If for example a concept
has the internal mental role (conceptual, imagistic) of our spatial concept between,
then count it as by that very fact referring to spatial betweenness, regardless of
context. On this proposal even isolated brains can think about space. Next, count
such concepts as descriptive primitives. Then count kind- and property-concepts in
class A as abbreviations for descriptions whose logical and nonlogical primitives are
the B and C concepts. Perhaps with a few additions, logical, categorial, causal, spatial,
and sensory concepts would then give us a basic working stock of primitives. This is a
familiar description-theoretic internalist idea. Next, take water to mean something
like the so and so stuff that causally grounds our use of water. The reference of the
concept water then varies with the reference of our. Reference-shifts and reference-
constancies of familiar sorts are accommodated.12 If the description-theorist runs
into circularity problems with terms of class Aexplaining parent via child,
Cicero via Catiline, and so onthere are generalized descriptive techniques for
dealing with that.
This assumes that we legitimately assign concepts of class C to class B and not to
class A. But might there not be analogues of Twin-Earth thought experiments for
those basic concepts? Can isolated brains conceive of spatial relations? The internalist
I have in mind says that we are conating cases that can easily be kept apart. The
Kripke, Putnam, Burge thought experiments tell us something substantial concern-
ing our intuitions about concepts in group A, but those intuitions need not extend to
concepts of group C: the internal properties of our spatial experience determine
which spatial properties and relations our thoughts are about.13
It would be wrong to charge our conservative internalist with magic. Perhaps some
unreective philosophers, and some dualists, do have a magical conception of
reference, implicitly taking the mind somehow to reach out and grasp, non-naturally,
externally constituted properties. But there is a less dramatic way to deny that
reference is essentially causally-socially determined. No magic underlies the intuition
that and stands for conjunction without mediation by a causal or other contingent
natural relation. That standing-for relation can be explained prosaically. Adopt a
12
It is not to the point to note the infelicity of substituting a metalinguistic description for occurrences
of water in that-clauses. For as noted above, the current topic is not that-clause ascriptions, but the modes
of presentation that can make true such ascriptions without being explicitly captured by them. (Doubtless
the metalinguistic ascription is semantically inequivalent to the unmetalinguistic original.)
13
Cf. McGinns weak externalism in McGinn (1989).
deationary conception of the reference of and and of all expressions in class B. All
there is to the reference of and is captured in this schema: P and Q is true iff
P and Q. Suppose a connective of anothers language has the same conceptual role as
a connective of our language. Then assign to their connective the deationary truth-
conditional interpretation of ours.14 This (projectivist) way of putting things is
equivalent to the (nonprojectivist) idea that, for group B concepts, conceptual role
determines reference, that is, contribution to truth conditions, without the mediation
of further contingent relations. Our notion of reference, we might suppose, simply
takes the reference of such logical concepts to be thus minimally determined. The
internalist theory of reference, then, as I am characterizing it, takes concepts of class
B (including C) to refer in a minimal or deationary sense.
So denying externalism about basic predicate reference does not thereby commit
internalists to magic. For in a plain sense internal conceptual role determines reference
for concepts of class B cum C without mediation. What does determines mean here?
There is a conventionalist element in deationary theories; on them, reference is not
substantively determined. It is as if we conventionally assign certain references to
certain basic conceptual roles.15 A projective-deationist theory of reference captures
that conventional assignment without its appearing arbitrary.
This account seems to work well enough for or and all. Does it work for spatial
concepts? Keep in mind that magic is beside the point. The question yet remains
whether reference to spatial properties is like reference to truth-functions or like
reference to water. I do not nd it so plausible to count isolated brains as capable of
concepts that pick out spatial relations, and here is why. My own spatial concepts
appear to have a crucially demonstrative element, pointing visually and tactually to
certain relations and properties, at least vaguely. I of course cannot dene straight
line by pointing. But this does not mean that what determines spatial reference is not
in part demonstrative. By pointing to the sorts of relation and properties that are to
count as curviness, betweenness, and so on, spatial perception apparently gives
worldly content to otherwise purely abstract concepts. Without such diffuse pointing
in visual and tactile experience, spatial concepts would, it seems to me, be empty. The
internalist about reference may say: But brains in vats have visual and tactual
experiences, and purport thereby to refer demonstratively. Quite so. But the percep-
tual factors of spatial concepts imply something about how their reference is deter-
mined: If those concepts are in part both visual and demonstrative, then their
references will have to be determined in part in the manner of visual demonstrative
concepts. If spatial concepts depend on concepts of the form a relation of that
general sort, where that points visually, then the reference of that depends on
14
Field (1994).
15
This is quite close to the theory I proposed in Loar (1981). There it was expressed without appeal to
deationism in the home case. Conceptual roles for primitive concepts determined truth-conditional
contributions. The spirit of this part of that theory was certainly conventionalist.
seeing, and seeing is an externally determined relation.16 It follows that spatial

concepts are wide concepts: They belong to class A, where all outwardly directed
perceptual-demonstrative concepts belong. But if spatial concepts are not in class B
then the internalist about reference really has no hope. No description-theoretic
explaining away of the apparent external determination of reference for concepts
of class A will be viable.
4 Phenomenal Intentionality
How are phenomenal aspects of perceptual states related to their intentional prop-
erties? Several views are current. At one extreme there are pure qualia views. The
qualitative aspects of (say) visual experience are in themselves nonintentional; those
sensational aspects of visual experience are intrinsically as nonrepresentational as the
blotched paint on a stucco wall. This is familiar from certain ways of construing color
perception: The surface property of redness causes experiential red*-ness. External-
ists comfortable with qualia might then regard perception as structured thus: A given
visual experience has the property of red*-ness, and that property, although not itself
intrinsically intentional, is a component of a perceptual representation whose inten-
tionality lies in a causal relation to redness.17
At another extreme, phenomenal aspects of experience are held to be an illusion.
Representationism18 holds that the only phenomenal qualities we can discern are the
properties perceptions represent their (purported) objects as having: There is only
redness and no red*-ness. Externalist representationism holds that visual represen-
tation is a matter of externally determined (e.g., causal) reference relations. The view
is apparently widely held, and, interestingly enough, often on phenomenological
grounds.19 These two very different externalist views of the relation between phenomenal
qualities and intentionality provide useful contrasts with the internalist view of
phenomenal intentionality that I nd intuitive. On the one hand I rather think that
we have a coherent conception of the felt aspects of perceptual experience; on the
other hand I do not think these aspects are purely qualitative, that is, in themselves
nonintentional.
Let us begin with the latter point, for it is an important source of the representa-
tionists intuition. The idea of nonintentional visual qualia appears (to me) unmoti-
vated. We cannot phenomenologically separate the pure visual experience from its
16
This is not to deny that spatial concepts may depend on transmodal perceptual concepts. Transmodal
does not imply nonperceptual. The transmodal spatial conceptual capacities of blind people will get their
content, presumably, from tactile, auditory, and proprioceptive perceptual states. To the extent that sighted
people can understand the mental contents of blind people, to that extent presumably they conceive of tactile,
proprioceptive, etc. modes of presentation.
17
Some proponents of this view regard the purely qualitative aspects of sensation as experienced only
obliquely or even as inferred. See Hill (1991); Shoemaker (1994).
18 19
To use Ned Blocks name for the position. Harman (1990), Lycan (1996).
purporting to pick out objects and their properties. It may seem that this makes sense
for certain afterimages, phosphenes, and the like; even that strikes me as dubious, but
I will not discuss it here. What seems to me obvious is that ordinary visual experience
admits of no phenomenological bracketing of intentional properties: We simply
cannot attend to the pure visual eld and its nonintentional components. In some
sense ordinary visual experience comes phenomenologically interpreted. But this
does not imply representationismalthough it seems often supposed to do so. It is
compatible with, and in my view best explained by, a certain internalist view of
intentionality that relies on the idea of phenomenal aspects of experience, in a broad
sense. Let me rst sketch the basic idea and then consider the representationists
denial of qualitative aspects of experience.
What I will call phenomenal intentionality is a phenomenologically accessible
feature of virtually all perceptual experience and of perceptually based concepts, for
example visual demonstrative concepts. The following will I hope convey the gist.
Suppose some indistinguishable lemons are one after the other brought to my visual
attention. The lighting, the position of my eyes and so on, are held constant. I am
asked to think something about each lemon in turn, say, thats yellow. Afterwards
I am told that some of the apparent lemons were hallucinations (that is what the
wires were for). I am asked whether, despite this, my successive visual demonstrative
thoughts all visually presented their objects in the same way. Surely a natural reply is
yes, in a rather intuitive sense.
This presents itself as sameness in an intentional feature. For those demonstrative
concepts (both the ones that succeed in referring and the ones that do not) all
purport to pick out some object visually. You cannot capture this common feature
by generalizing over objects: There is some object that the demonstrative concept
visually presents. And surely the content of those thoughts is not itself existentially
quantied: I am seeing some lemon or other and it is yellow. The thoughts in
question are demonstrative and they are not self-consciously reexive. An apt way to
put those concepts common feature seems to be this: Those visual demonstrative
concepts, and the perceptions that underlie them, are all singularly visually directed.
This is a nonrelational phenomenal feature, by which I mean something rather
strong: We are aware of internally determined phenomenal features of visual experi-
ence, of their manifold felt aspects, and among those featuresthough not separable
in imaginationis the directedness just mentioned.
The feature presumably belongs primarily to a visual perception, and derivatively
to a visual demonstrative concept that incorporates the perception. I will speak
loosely of its being a feature sometimes of the perception and sometimes of the
concept. Why call it intentional? I do this in the hope of engaging archaic intuitions.
A natural way to capture the phenomenon is this: the visual perception purports to
refer, it is directed, it points. When we considered whether conceptual-role
properties, individuated syntactically, leave out something importantly representa-
tional about thoughts, we could surely have noted the relevance of phenomenal
directedness. Does the idea of phenomenal directedness commit me to there being

some marka little arrowin the visual eld? Well, the visual eld would have to be
packed with arrows, since virtually every one of its parts is directed on some bit of the
passing scene.20 When I say that directedness is phenomenal I mean merely that
I can identify it in experience. I apparently can tell that hallucinatory experiences
have a purporting to refer property that is also present when visual experiences pick
out real objects in the normal way.
Even if singular directedness is an internally constituted property of perceptual
concepts, it does not on its own vindicate an internalist view of perceptual inten-
tionality. For the externalist will surely object that a visual perception that fails of
reference will nevertheless purport to refer to its (nonexistent) reference as having
some property F. I earlier noted that this requires (in the simplest case) reference to
the property F, which is in general externally determined. So even if a singular
demonstrative has a phenomenal directedness that is independent of the demonstra-
tives reference, internalism about intentionality is hardly thereby made coherent.
The point is clearly correct.
The present idea though is not that the singular phenomenal directedness of a
visual demonstrative concept is sufcient for internalism about that concepts
intentionality. But it is a key step in constructing the notion of internal intention-
ality. The apparent intentional properties of a singular visual demonstrative
concept are not exhausted by the references of its constituent kind terms (its as F
contents). The latter do not account for the intentionality of the visual demonstra-
tive as a whole, for that apparently is an intentional property over and above the
referential properties of the constituent qualifying concepts. So there is an inten-
tional property, (singular) directedness, that does not consist in (singular) reference
and is not explained merely by the reference of kind terms. Perhaps it will be said
that this directedness is just a matter of conceptual role. That can be said, but it
hardly neutralizes the phenomenology. Who knows whether what appears as
directedness consists in some underlying factor that might aptly be called concep-
tual role? We apparently do not conceive it in terms of its place in some system of
conceptual roles. We have a phenomenological take on it, and that is what I call
attention to. Once phenomenal directedness is admitted, it is difcult not to admit
also, as we will see in section 10, something analogous for a crucial group of kind
concepts. I will argue that this is all we need as a satisfactory basis for internally
constituted intentionality for thought in general.
20
The visual experience supports a complex intentional structure. If you shift attention to another
aspect of the visual experience, that reveals a distinct intentional directedness. Evidently there is a compact
and quite complex set of such intentional directednesses within most visual experiences.
5 The How and the What of Intentionality:

Mere Intentional Objects Considered
As I see it, phenomenal intentionality is a matter of how ones perceptions and
thoughts represent things if they succeed, rather than of what is thereby represented.
The representationist says the latter. It may be replied that the how-what distinc-
tion is bogus. For to say how a visual perception purports to present something is to
say what it presents it as. Even if no appropriate physical object is there, my
perception presents something as a snake. And that presupposes reference to a
property, at least the property of being a physical object of a certain approximate
size and shape. Right. But this misses my point. As we noted above, directedness is
needed in addition to predication to explain the how of an empty visual demonstra-
tive concept. The nonrelational intentionality of predicative concepts is yet to be
discussed. But for now let us be content to say that the directedness of a visual
perception is an aspect of how the perception (and the demonstrative concept that
incorporates it) presents things. It is not a matter of the perceptions presenting
something as F, but rather of its style or manner or mode of presentation.
But the representationist has a reply. Representationists typically count a percep-
tion that fails to refer as yet having an intentional objecta mere intentional object so
to speak. This suits the what conception of intentionality: The intentional properties
of even an empty thought or perception would consist in what it represents (as being
such and such), namely, its intentional object.21 But this is a peculiar way for the
representationist to put things. The representationist is a referentialist. In the lemon
hallucination the only references to be noted are properties, that is, the references of
the perceptual demonstratives as it were predicative factors. There is no further
reference: The perception purports to refer and fails. And failing to refer is not a form
of reference, however apt talk of representation may appear. Speaking of mere
intentional objects is all right in its way; but it should not mislead one into claiming
to have characterized phenomenal intentionality in purely referential terms.
An intentional how a perception presents things cannot easily be avoided.
To pursue the point, here is a simple fact: In the veridical case there are not two
intentional objects, the mere intentional object and the real object. What the hallucin-
atory perception has in common with the veridical perception is, then, not a mere
intentional object. The two perceptions have something intentional in common, and it
is that common feature which concerns me. The representationist may say that
although they do not share having a mere intentional object, they do, however, share
the property of having an intentional object. But it is hard to take this seriously. The
veridical perception has an intentional object in a transparently relational sense:
It refers. It could only be a fanciful Meinongianism that construes having a mere
intentional object relationally. But suppose we went Meinongian. It is still far-fetched
21
Harman (1990).
to suppose that we then end up with something that the veridical and hallucinatory
perceptions have in common, for having an intentional object would have to stand for
two very different relations. The simple fact is this. What the two cases have in common
is something phenomenological. We could call it having-an-intentional-object, with
the hyphens marking a nonrelational reading. But it is less misleading to use an overtly
nonrelational form, for example, directedness. And this clearly concerns the manner in
which a perception or visual demonstrative concept presents things rather than what is
represented. Why call directedness intentional if it is nonrelational, if it is about the
how rather than the what of perception? What else to call it? It seems to be the primitive
basis of our intuitions of the phenomenal aboutness of perception.
The determined referentialist may pursue a different strategy. For there is still the
language of as if, the language of appearances. We can describe the lemon hallu-
cination by saying that it is as if I am seeing a lemon, or it appears to me that I am
seeing a lemon. If one holds an externalist view of reference one will also hold an
externalist view of such appearances: For the function of a that-clause (as in appears
that) is to capture the references of the state or property thereby ascribed. If
appearance-properties are then captured in that form, intentional qualia will turn
out to involve relations to external objects and properties, and so cannot be regarded
as entirely internal properties.
Now this strategy rather overlooks the representationists commitment to
phenomenology. Recall how Smart attempted to capture the experience of a yellowish-
orange afterimage: Something is going on in me that is like what is going on in
me when I am seeing an orange. The problem with this analysis of sensory experience
is that it is phenomenologically blank: it does not imply that there is anything in
particular that it is like to experience a yellowish-orange afterimage. The language of
appearance is, unlike Smarts locution, at least mental in its implications. But to say
that it appears that . . . , or that it is as if . . . , is not to say how it phenomenally appears.
And the point of the lemon case was that there is something phenomenologically in
common among the various visual experiences, and that it included something that is
phenomenologically intentional. Mere talk of appearance may point to that extrin-
sically, but it does not capture what it is like. This does not mean that appears has no
phenomenological role. For one can say to oneself it appears thus and point in
memory or imagination or present experience to a phenomenal type. But that of
course does not help the externalist.
6 Is There Phenomenal Paint?

According to Gilbert Harman,22 when you turn your attention to your perceptual
experiences, all you can discern phenomenologically are properties of the (apparent)
22
Harman (1990).
object of the experience, that is, shape, color, and so on. This is in strong contrast
with how pictures appear: We can attend to the paint in a picture, but, according to
Harman, there is no phenomenal paint of which we are introspectively aware.
Phenomenal paint fails to appear not only in unreective experience but even on
phenomenological reection. Of course he does not deny that some perceptual
experiences can lack real objects and yet have a fully phenomenal presence. But he
is content to appeal to mere intentional objects, intentional objects that do not exist.
An experiences qualities in such a case consist in the properties it attributes to the
object that isnt there. The structure remains the same: We are aware not of
the experiences phenomenal qualities (for there are no such properties) but of the
properties of the apparent object of experience.
The argument of the last section, concerning the need for a nonrelational com-
monality between veridical and nonveridical perceptions, does not, apparently, touch
Harmans point. For it does not imply that the phenomenology delivers any highly
specic qualitative aspects of experience, as opposed to highly specic properties of
the apparent objects of experience. Perceptual experience is phenomenally transparent:
We seem to be directly aware of properties of objects rather than properties of
experience itself. And it may seem that carping about mere intentional objects will
not neutralize that observation. So the question I wish now to raise is this: Is there
phenomenal paint? The concept of directedness purports to be of a phenomenal
property in the sense of a property of experience rather than a property of an
apparent object of experience. At the same time, the directedness of perception is
not separable in imagination from the more specic phenomenal aspects of percep-
tion. It seems then that if there is phenomenal directedness there must be phenom-
enal paint.
Is there not a phenomenal difference between visual and tactual perceptions of
shapes, a difference in the felt qualities or qualia of vision and touch, which is to say, a
difference between visual and tactual paint? Consider the obvious phenomenological
differences between seeing and touching a quarter. The representationists reply, as
I understand it, is this. What we are inclined to think of as specically visual and
tactual differences in how we perceive a quarter are in fact differences in its perceived
qualities over and above its shape and size, differences between its color and
luminosity, and its texture and solidity. So we are not forced by the phenomenal
differences between sight and touch to admit differences in qualitative aspects of
experience, that is, differences in how experiences represent rather than what they
apparently represent.
Bill Lycan, whose position is strongly representationist, has developed an inter-
esting strategy for defusing apparent cases in which the representational content of
two perceptions is shared while their phenomenal manner of presentation differs.
The idea is to take the alleged difference in qualitative manners of presentation to be
in fact differences in apparent properties of intentional objects. Lycans account
requires nding multiple levels of intentional objects in problematic experiences.
I will describe an experience and consider how Lycans strategy avoids qualia in
accounting for it. Keep one eye open and use your ngers to stretch it in different
directions. You see some apples in a bowl, say, in blurry distortion. Surely the
perception does not represent the apples as themselves blurry. Rather, the proponent
of qualia will say, the perception blurrily represents the apples. The blur is a
qualitative aspect of the visual experience. Lycan says not so: There could be a
scene that is objectively blurry in that way, and that (nonexistent) scene is a sort
of secondary intentional object of the blurry visual experience. There are two levels of
intentional objects here: the ordinary apples in the bowl, and the (nonexistent)
objectively blurry-apples-in-a-bowl.
I am willing to grant that there could be an objective scene that looks to the normal
eye just like that.23 But it seems to me that Lycans is a forced and ultimately wrong
account of visual blur. The blur is an aspect of how the perception represents its
objects, certain normal apples; it is not in its normal role a perceived property of
some abnormal apples. The question is whether that can be argued more or less
conclusively on its own terms, or whether a larger argument for visual qualia is
needed in order to give the qualitative account of visual blur and the like its proper
force. The latter is in fact what I am inclined to think. But let me rst present an
analogy from ordinary depiction to nudge intuitions in the right direction. For there
is an intuitive distinction between what is depicted in a picture and how it is depicted,
where by the how I do not mean the surface of the picture but something
intentional.
7 Aesthetic Interlude: The How and

the What of Pictures
Consider representational paintings that are not photographically realist, for
example, one of Picassos portraits of Marie-Thrse Walter. It represents its subject
distortedly, if quite gracefully. Marie-Thrse is captured with rounded swooping
lines and bright colors, and fragmentedlyher head, say, has one half in prole and
the other half full-face. Doubtless there could be a real three-dimensional scene that
looks just like this picture. But we do not see the Picasso portrait as representing a
Martian: It seems unmotivated to say that it represents Marie-Thrse by way of
representing a Martian. The picture does not, at least as I am inclined to see it,
represent any object as having that distorted shape. Rather it gets you to see its object
in a distorted way (and part of the visual pleasure lies in deciphering the picture,
following here and there and back again how it represents its object). The distortedness
23
I also grant that there could be an external arrangement of lights that look, undistorted, just like
phosphenes, pace Block (1996).
is not a matter of intentional content but of intentional style, not a matter of what is
represented but of how it is represented.
There are Picasso pictures that do seem to represent a distorted objecta gure on
a beach, made out of bony pieces, like a surrealist sculpture of a seated bather. Two
points about these pictures are it seems to me instructive. First, if the picture is of a
bony sculpture, then that sculpture is itself represented in a realistic way, and that is
itself a manner of representation, an intentional styleone that we usually do not
attend to, for it is, until noticed, diaphanous. The diaphanousness of a realistic
portrait should not blind us to pictorial realisms being an intentional style. Second,
the realistically depicted surrealist sculpture itself represents something distortedly;
and the sculpture does not represent a further surreal or distorted object, and so on.
It represents a woman, surreally. You cannot get rid of the manner, the how. I am not
speaking of the physical paint, but of something perceptual, an intentional way in
whichas we visually engage itthe picture presents its objects. (The Picasso picture
is even more interesting than I have made it. We can move back and forth between
the pictures realistically presenting a bony surrealist sculpture, and its surrealistically
depicting a woman on a beach.)
The same holds for visual representations. It is difcult to see how you can get rid
of the how, or the manner, of perceptual representation. That manner is as accessible
as the how of pictures; and it is intentional. This seems to me to be a coherent view of
blurriness and the like. The question is now whether we can turn that coherence into
something stronger.
8 Inverted Spectra and Inverted Worlds

Harmans and Lycans representationism is externalist. Those aspects of visual
experiences of which we are phenomenologically aware are their ordinary referential
properties; they involve externally determined relations to external objects and
properties. There is a familiar, quite elementary reason for rejecting this view,
which I nd persuasive. We can coherently, and easily, conceive of subjectively
different color experiences that are of the same objective properties of objects. (The
idea of inverted spectra is one way of conceiving this.) We can also conceive of a
single color experience that is, in different circumstances, of different objective
properties. This is persuasively shown by Blocks Inverted Earth, a color version of
Twin Earth.24 The arguments of representationists against these possibilities do not,
I think, depend on phenomenological intuition so much as on externalist theory. The
phenomenology of imaginationand that is after all our current eld of playseems
squarely on the side of the qualia-exponent. And all the subjectivist apparently needs
here is that the phenomenology is coherent, which it appears to be. We must add this:
24
Block (1990).
These color qualia are not pure qualia, for they are phenomenally intentionalthey
phenomenally represent (what we conceive of as) object-surfaces as having certain
properties. How to interpret this remains to be explained, as it will be below; but that
it makes sense seems clear to me from the basic thought experiments. There are color
qualia, and they are intrinsically intentional.
9 Isolated Brains
Externalists, as we have seen, often have no trouble regarding visual demonstrative
concepts that fail of reference (in a hallucination, say: that hand reaching out from
the wall) as having genuine intentional content.25 But according to these externalists,
that intentional content is essentially anchored in the properties that the perception
represents the (merely intentional) object as having. Putting colors aside, these
properties will include physical-object-types, spatial relations, and so on. (The visual
hallucination represents the hand as an object with protruding appendages thrusting
in my direction.) According to the externalist, my perceptual state represents a
merely intentional object as spatially located only if that perceptual state stands to
certain externally constituted properties in externally determined reference-relations.
Evidently standard Inverted Spectra and Inverted Earth thought experiments do
not count against this point. For they hold constant the basic physical properties that
visual perceptions represent. So not surprisingly, we require a more radical concep-
tual possibility than those if we are to establish intentional internalism. It must show
that we can hold constant phenomenologically accessible intentional visual qualia
while varying all the properties that they represent things as having.
Brains in vats to the rescue. One of the interesting facts about the current debate
about representationism is how tame the thought experiments are. If the game is
phenomenology, then we really ought to exploit all possibilities that are phenom-
enologically conceivable and prima facie coherent. So, once again: I could have a
mental twin whose brain is a molecule for molecule duplicate of me; and I can
conceive that twin as having the same visual experiences that I have, even though its
brain is isolated from all the normal causal relations to the world that give my visual
experiences their actual references. The point is that when I imagine how the brains
visual experiences represent their (merely intentional) objects, I apparently imagine
those experiences as in some sense intentional, despite the brains difference from me
in all its references. Is this coherent? Discussions of representationism and qualia
avoid this thought experiment, it seems to me, because defenders of qualia think they
dont need it. This is because they are concerned with qualia and not intentionality;
they want merely to show that color qualia make sense. Even if visual qualia are
25
Burge (1986) and Rcanati (1993), as against Evans (1982), and McDowell (1986).
phenomenally intentional this will not in itself support a purely internalist concep-
tion of visual experience.
What is wrong with the idea that my twin-in-a-vat can have visual experiences
intentionally equivalent to mine? There seems no phenomenological incoherence in
the idea. There has been thought to be a conceptual incoherence, however. For if the
brains visual experiences are intentionally the same as mine, then according to the
referentialist about intentionality they must share references with mine, which
according to externalism about reference is impossible. But of course the argument
is fragile, for it ignores the coherence of nonreferential or phenomenal conceptions of
intentionality. When I imagine the brains visual states and at the same time conceive
of them as having no references in common with mine, what am I conceiving?
Here we return to the how versus the what. What I hold constant in imagining
the brains visual experiences is how it conceives things. That is, I can coherently
imagine a complete sharing of my experiences phenomenological details conjoined
with a complete unsharing of its references, at least with regard to my outward-
directed states.
So we need an analogue of the phenomenal directedness of singular perceptual
demonstratives for the other representational factors in perception, that is, the factors
that represent external properties and relations. Suppose we can extend the idea of
directedness to those aspects of visual experience that purport to represent spatial
properties, and so on. Then my twins visual experiences and mine share that
directedness; but mine refer to spatial properties (metaphysically rather than phe-
nomenally spatial so to say) while his do not. If all this can be made out, we are aware
of directed qualia, qualia that internally purport to refer not only to objects but to
basic properties. My twin and I conceive things thoroughly in the same intentional
manner. This is what we hold constant across twin brains, namely, highly specic
forms of property-directedness. There are no shared properties of intentional objects.
To return to the promised relevance of conceptual roles, we hold constant not only
intentionalized phenomenal experience but also conceptual roles. Internal intention-
ality is to be located primitively in perceptually based concepts. It will be derivatively
located in nonperceptual concepts via their conceptual connections with perceptual
concepts. The subjective intentional properties of nonperceptual concepts are always
a matter of, as it were, looking sideways via their connections with perceptual
concepts. The earlier complaint about the intuitive deciencies of our conceptions
of narrow conceptual rolesas purely syntactic, as not capturing how thoughts
subjectively represent the worldis I think answered on this picture.
10 Recognitional Concepts
The idea of directedness can be extended to demonstrative concepts that purport to
pick out, perceptually, kinds and properties rather than individuals, what we can
call recognitional concepts. They are an important if somewhat elusive variety of
kind-concept or property-concept. They appear to me to have an intuitively evident

sort of internal intentionality, which may be thought of, by analogy with object-
independent, as kind-independent or property-independent intentionality. As we
have seen, committed externalists about intentional content concede that individual
demonstrative concepts have an object-independent conceptual or psychological
integrity. But they will draw the line at property-concepts, kind-concepts, relation-
concepts, insisting that their psychological or intentional individuation must incorporate
the kinds, properties, or relations for which they stand.26 I am quite committed
to externalism about the reference of these property-concepts: That a given recogni-
tional concept refers to a given kind is of course a matter of some external, for
example, causal, relation between the thinker and the kind. What I wish to deny is
that either the external relation or the kind itself is part of the intentional individu-
ation of the recognitional concept.
Recognitional concepts are personal, and they are perspectival. Their reference is
determined by non-socially mediated actual and dispositional relations between the
thinker and the kind. And these concepts are individuated, in part, by the perspective
from which they are conceivedfor example, a perceptual perspective. So a visual
demonstrative kind-concept may pick out a kind by virtue of past perceptions of its
instances and a disposition to pick out further instances from its dening visual
perspective. It is important to be clear about the following point. Visual recognitional
concepts are not descriptions of the form the kind to which this thing, that thing, and
so on belong, that is, they are not descriptions that embed singular visual demon-
strative concepts. Suppose I have a solid recognitional conception of a species of elm,
without knowing its name. I need have in mind no particular elms, nor any group of
them: My conception is of them in general, but from a certain perspective, from
which I can take one of them in at a glance, say, while being able to see its bark and its
branches. So we are not proposing a descriptive account of the kind-independent
intentionality of recognitional concepts.
How then shall we intuitively conceive the directedness of a recognitional
conceptits purporting to pick out a certain kind of tree? Think of exercising the
concept in imagination without applying it perceptually. For example, one wonders
whether there are trees of that kind in Philadelphia; here one points in imaginative
memory to a kind. What is useful about these cases, where the concept is not applied
directly in perception, is that it makes it easy to isolate the purported reference to a
kind (that kind of tree) from corresponding purported references to individuals
(that tree).
The question is whether we have here an object-independent intentional property.
It wont surprise you that it seems clear to me that we have. Conceiving of a given
26
See Burge (1986) and Rcanati (1993).
visual kind-demonstratives failing of reference is not hard: One can be wrong in

thinking that, from a certain perceptual perspective, one has picked out, or is able to
pick out, a kind. But ones recognitional concept may nevertheless have been as
coherent as any, and perceptually focused as if on a kind. So the analogy with the
lemon hallucination seems fair. We also would like to establish a further analogy with
the lemon case, that is, that recognitional type-demonstratives hold constant across
reference change. And so I have to inict on you a familiar waving of intuitions, but
with a new emphasis.
Imagine, then, some worlds like ours, as supercially similar as you like, but
populated with different underlying kinds. Could that same recognitional
conceptthe visually embedded concept that kindoccur in all those worlds even
though it picks out different kinds in each? This strikes me as straightforwardly
evident. And, as in the lemon case, what intuitively we hold constant across these
worldscertain conceptual and experiential factorsare not easily equated with
some combination of functional or syntactic and purely sensational properties.
A kind of intentional directedness is again present, associated with the kind-
demonstrative. It is analogous to the directedness of singular demonstrativesand
doubtless derives from itdespite the difference between singular demonstrative
pointing and pointing to a kind via a recognitional capacity. So it seems quite easy to
conceive of a recognitional concept kind-independently and yet intentionally, as
purporting to refer, pattern recognitionally, to a kind.
Let us be clear about what the point has been so far. It is not that we nally have
shown that recognitional concepts have purely internally determined intentional
properties. That cannot be so, for we havent yet dealt with those further general
concepts that are presupposed by, say, an ordinary visual recognitional concept (e.g.,
of elms), especially the spatial concepts that are entwined with visual experiences, as
well as the general concept of three-dimensional objects persisting through motion
and change. What rather it seems to me we have shown is that recognitional concepts
have kind-independent intentionality in this sense: Even though a recognitional
kind-demonstrative fails to pick out a kind, it nevertheless has (and now we speak
phenomenologically) an overall intentional content that is organized around the
concepts visual kind-directedness, that is, its purporting as a whole to pick out,
visually, a kind, a property, and so on.
A recognitional concept purports to refer in two ways: (a) It purports to refer by
way of an imaginative capacity. This can only be conceived intentionally, for one
has little grip on purely sensory visual imagination. This imaginative capacity
somehow involves as it were schematic singular visual demonstratives. The generality
of the directedness of a recognitional concept, and the schematic form of imagined
occurrences of individual visual demonstratives, are somehow closely connected.
(b) It purports to refer by way of a disposition to respond to singular demonstrative
visual experiences, where, as before, these are conceived intentionally and
object-independently.
11 General Concepts of Physical Objects

and Spatial Relations
Now we turn to spatial and other concepts, which I argued belong among wide
(class A) concepts, whose reference is externally determined; these concepts include
certain general concepts of approximate spatial relations, shapes, and the like, and a
certain conception of a three-dimensional object as it persists over time. If we can
extend to these concepts something analogous to the treatment of recognitional
concepts, we will have rather a strong reason to think that concepts whose references
are externally determined can in general be individuated by internal intentional
properties. Class C concepts will moreover play a crucial structural role in explaining
the internal intentionality of the huge remaining class of wide terms and predicates.
Certain basic physical and spatial concepts do not have and do not need socially
deferential roles. Of course isosceles may well, for many of us, have as socially deferential
a role as arthritis. But having the concept of an isosceles triangle would not be
possible unless we independently had recognitionalvisual and tactualconceptions
of more basic spatial relations. That at least is how it strikes me. Of course our
recognitional concepts are not very precise; we hardly acquire the concept of a (perfect)
right angle from perception alone. And yet concepts of more or less angular and curvy
boundaries, of spatial betweenness, of relative distancethat is, the raw material of
further precision in spatial conceptsare plausibly regarded as recognitional concepts.
To say that they are recognitional concepts is not to deny them structural interrela-
tions. Its not a topic we can pursue here, but there seems to me no fundamental
difculty in the idea of structural interrelations among recognitional concepts. Quality
spaces, after all, are structured. We also appear to have a recognitional concept of
physical object in general. This does not mean that we have an image of a physical
object in general; but there is nevertheless a perceptual ability to group together three-
dimensional objects of all shapes and sizes. To say perceptual leaves it open that some
recognitional concepts are transmodal, that is, apply on the basis of both visual and
tactual information. But I take transmodal concepts themselves to be perceptual
concepts and not (as it were) pure categorial concepts, that is, not amodal.
Do these general recognitional concepts have the kind-independent phenomenal
directedness we claimed for less general recognitional concepts? Here I must again
appeal to the fully intentional mental life of my twin-in-a-vat. All externalists abandon
me at this point, however indulgently they have followed so far. I hope though that now
we have not merely an intuition, but something approaching a principled account of it.
Given externalism about the reference of these concepts, none of that twin brains
concepts pick out physical-objecthood or spatial properties. The internalist claim is
that my twins concepts are exactly similar intentionally to my recognitional concepts
of the various spatial properties and relations. For they conceive the properties and
relations to which they purport to refer in precisely the same way as my concepts
do, via the same highly specic visual and tactual experiences and guided by the
directedness of that property, that relation, and so on. The twin brain has a fully
phenomenally intentional visual eld. Given that the special directedness of recogni-
tional concepts, including spatial and basic-object concepts, derives from the singular
directedness of perceptual experience, it makes perfectly good sense to regard the
intentionality of the brains general (i.e., nonsingular) concepts to be identical with
mine. We need not decide whether the twin brains spatial concepts refer to some
nonstandard properties and relationsfor example, properties of the visual system
itselfor fail of reference entirely. It is not clear that this is an interesting question. But
if we can make sense of intentional properties that persist through shifting kind-
references and the failure of kind-reference, then I cannot see why that should not
also apply to spatial recognitional concepts.
You may object. The sense we made of intentional directedness in connection
with less general recognitional concepts depends on qualifying concepts that them-
selves are somehow intentional. That was intuitively crucial in supporting the
intuition that the recognitional conception of elms had its own kind-independent
intentionality. Well, yes. But that does not mean that we then depended on the
intentional properties of those basic qualifying concepts being externally determined.
It appears to me that it is quite coherent to ascribe object-independent directedness
to recognitional concepts all at once, including basic spatial concepts and so on.
12 The Paint that Points

Before turning to nonperceptual concepts and the question of their intentionality, let us
look back to the representationist-qualiphile dispute. I have agreed with the represen-
tationist that visual experience is intrinsically intentional but denied that this requires
an externalist treatment. I have also argued that the notion of mere intentional object,
which the representationist requires if he is to be true to the phenomenology, is
dubiously compatible with externalism about intentionality. Moreover, I argued that
what appear to be coherent phenomenological intuitions support the qualiphiles thesis
that we have intuitive conceptions of the qualitative aspects of experience, although we
have no way of separating the qualia from the intentionality. But this is all right, given
that the phenomenological intentionality of perception is to be explained via directed-
ness, which is itself a phenomenal notion. Now if by paint one means something that
we can conceive independently of its intentionalitylike the paint on a canvasthen,
at least in vision, there is no (pervasive) paint. But if paint means qualitative aspects of
experience that are separable from referential properties, then there is such a thing as
phenomenal paint. And it points.
13 Personal Systematic Concepts

Presumably there are concepts that are neither recognitional concepts nor socially
deferential concepts nor logical concepts. Calling them theoretical makes familiar
sense from philosophy of language, but it is perhaps somewhat overblown, and what
I mean is not all that grand. So let me call them personal systematic concepts.
To begin with, heres what I mean by personal. Suppose that Fiona thinks that
one way of becoming a mother is adopting a child and caring for it. When we tell her
that mother means a biological relation, she replies, determinedly, When I say
mother it means what I mean and not what someone else means. What construal
shall we give of Fionas undeferential concept? We might try a description, or a
cluster of descriptions. But that would, at best, be a matter of local convenience, and
not a strategy for cashing out her personal theoretical concepts en masse. The reason
is circularity. It is doubtful that we could explain those concepts using ordinary
descriptions or description-clusters that appeal only to recognitional and logical
concepts. We have to invoke other concepts that are in the same boat, concepts
such as female, child, raising, and so on. Getting its content from having a role in a
network of conceptual connections with similar concepts is what makes Fionas
concept systematic, or if you prefer theoretical.
Now consider her personal systematic concepts as a whole. They are bound to be
multifariously linked with recognitional concepts, including the general concepts of
physical object and spatial relations. Recognitional concepts that pick out children,
that pick out the subjective psychological state of attention, that pick out attentive
behavior, that pick out feelings, that pick out kinds of physical activity will also play
essential roles in giving content to Fionas systematic concepts.
We come to the question: How are we to conceive of the internally determined
intentionality of personal systematic concepts? What I want to suggest is that their
intentional properties are dispositional. We do not take in the intentional properties
of a systematic concept all at once. We do so rather by nding our way about among
a systematic concepts lateral interconceptual connections. You may ask how the
conceptual role of a concept can amount to an intentional property. We are used to
thinking of conceptual role as syntactic role (as is often said). But what we uncover
is hardly just the concepts syntactic or functional or inferential connections. For one
constantly engages, at every turn, perceptual recognitional conceptions that have
their own independent directedness. The phenomenological world-directedness of a
personal theoretical concept, I want to propose here, derives from its intimate
conceptual connections with perceptual intentionality.
So the idea that every concept can be revealed in an introspective glance, or even
in an introspective stare, is not essential to the defense of internal, phenomeno-
logical, intentionality. This is not simply to assert that the conceptual roles of
concepts are crucial to their individuation; that does seem to me beyond doubt.
The point is not so much about individuation as about intentionality. The intuitive
world-directedness of a conceptthat phenomenological propertyneed not consist
in its having its own perceptual focus, as do perceptual demonstratives. Its inten-
tionality may come rather from the accessibility of conceptual repositionings and
sidelong glances.
14 Socially Deferential Concepts

Socially deferential concepts include most of the proper names in ones repertory,
and the extensive group of kind terms to which Burge has called attention. Socially
deferential concepts are of course not perceptually based in the manner of recogni-
tional concepts, for recognitional abilities do not x their reference. And so percep-
tual focus does not give us an intentional property of such concepts as a whole.
Socially deferential concepts have about them something more discursive and
linguistic: They involve conceptions of other speakers and of the shared language.
What I propose is that socially deferential concepts belong among the personal
systematic concepts. This is perhaps perplexing: How can socially deferential concepts
be personal concepts? There are two ways in which a concept can be said to be personal.
The rst concerns how its reference is determined, that is, whether it is socially
deferential or not. The secondas used in the phrase personal systematic concept
concerns how the concept is individuated, which is to say by the systematic role that the
concept has in ones own thoughts. And when I say that the socially deferential concepts
belong among the personal concepts, I mean simply that those conceptsincluding
their internal intentional propertiesare determined in that way. At the same time,
their reference is determined socially. The link between the two is this: That a concept of
mine is socially deferential depends entirely on its systematic role in my thoughts. If it
has the socially deferential role in my thoughts, then its reference is determined socially.
If not, then its reference is determined otherwise, perhaps in the manner of a recogni-
tional concept, or in the manner of personal systematic concepts that are not social.
As a last note on social deference, we might observe that my twin-in-a-vat of
course also has socially deferential concepts, but only in the sense that he has
concepts that are equivalent in their internal intentionality to my socially deferential
concepts. Might they have reference? If they have reference, it is not via the expected
routes; my twin-in-a-vat has no concepts that refer via the usage of other people.
Perhaps its concepts refer to some states of its own? I doubt that our concept of
reference applies here; better to say that, like most of the rest of my twin-in-a-vats
concepts, his socially deferential concepts fail of reference.
15 Directedness and Reference

How are internal intentionality and reference connected? Intentional directedness
is an object-independent property, and it does not involve relations to objects.
Reference comprises various causal and other relations to objects, and it is absurd
to think of those reference relations as somehow instantiated without objects.
My answer may not be fully digestible without more explaining and consequent
ruminating, but here goes. Directedness is an object-independent property. But it is
intimately involved with what is often called the diaphanousness of perception.
Directedness is diaphanousness without an actual object. Earlier I pointed to the
inadequacies of mere intentional objects in furthering the representationist project.

But that leaves intact the usefulness of intentional objects in phenomenological
description; and, in a phenomenological vein, we might say that directedness is
diaphanousness toward intentional objects, whether mere or real. Now imagine
having one of the lemon experiences without knowing whether it is veridical. You are
strongly tempted to say that object. Your perceptual processing presses mightily on
your belief inclinations, so strongly that you seem both to commit yourself, by using a
demonstrative, and to take it back at the same time: that object may or may not
exist. The phenomenology gives you the feel of a sort of ontologically neutral object
that could have the property of existing or not-existing; and directedness is phenom-
enologically very like a relation to that neutral object, which could turn out to be real.
Suppose you then discover that it is real. At this point the question arises of the
objects actual nonphenomenological relation to your perception. It turns out that a
certain optical-causal relation holds in all such cases. The ghostly internal relation
gets embodied in something nonmental and out there.
The point of this fanciful description is to explain the relationship between
directedness and reference. But the explanation of course is phenomenological-
psychological; it is from a combined rst-person/third-person perspective that di-
rectedness is intimately connected with reference.27
16 Concluding Remarks
The lemon demonstratives had this property in common: They purport-to-pick-
out-an-object. This was said in a phenomenological vein. We are, it seems to me, as
entitled to speak of phenomenological intentionality as we are of the felt qualities of a
sensation. And the Cartesian intuition that is rejected by externalists about content is
after all primarily a phenomenological intuition. We might reject that intuition by
rejecting phenomenological or subjective conceptions in the philosophy of mind. But
the only way to reject phenomenological intentionality selectively is to show that there is
after all no such apparent phenomenon, or that the idea is incoherent. It is hard to see
that externalist arguments are of the right sort to show that it is incoherent.
If there is no reason to deny phenomenal directedness and no reason to regard this
phenomenal feature as object-dependent, then there is no warrant for the externalist
idea that internalism about mental content somehow leaves mental content blind,
or that then it is all darkness within. In fact it is odd of the externalist to see his
theory as providing interior illumination. The metaphor seems to ow in the
opposite direction: If the only intentional content is externally determined then it
is all darkness within. Still the thought naturally arises, how could something in
the brain account for intentional directedness? But just this question arises about
phenomenal features in general, and here I am content to put it aside.
27
See Loar (1995) [Chapter 13 in this volume].
List of Works by Brian Loar
Meaning, Speech Acts, and Imperative Inference (1965). University of Oxford, Faculty of Literae
Humaniores Thesis (B.Phil), i + 105.
Sentence Meaning (1972). University of Oxford, Faculty of Literae Humaniores Thesis
(D.Phil.), xv + 150.
Reference and Propositional Attitudes (1972). The Philosophical Review 81: 4362.
The Semantics of Singular Terms (1976). Philosophical Studies 30: 35377.
Two Theories of Meaning (1976). In G. Evans and J. McDowell (eds.), Truth and Meaning,
Oxford University Press.
Names and Descriptions: Reply to Devitt (1980). Philosophical Studies 38: 859.
Ramseys Theory of Belief and Truth (1980). In D. H. Mellor (ed.), Prospects for Pragmatism,
Cambridge University Press.
Syntax, Functional Semantics, and Referential Semantics (1980). Behavioral and Brain
Sciences 3, 1: 8990.
Mind and Meaning (1981). Cambridge University Press, xi + 268. Reprinted 1989.
Conceptual Role and Truth Conditions (1982). Notre Dame Journal of Formal Logic 23, 3:
27283.
Must Beliefs Be Sentences? (1982). In P. Asquith and T. Nickles (eds.), Proceedings of the
1982 Biennial Meeting of the Philosophy of Science Association 49, 2: 62743.
Reply to Fodor and Harman (1982). In P. Asquith and T. Nickles (eds.), Proceedings of the
1982 Biennial Meeting of the Philosophy of Science Association 49, 2: 6626.
Review of D. Holdcrofts Words and Deeds (1982). The Philosophical Review 91: 3036.
Review of J. R. Searles Expression and Meaning: Studies in the Theory of Speech Acts (1982).
The Philosophical Review 91, 3: 48893.
Review of Gareth Evanss The Varieties of Reference (1984). Philosophical Books 25, 1: 4651.
Review of Saul Kripkes Wittgenstein on Rules and Private Language (1985). Nous 19: 27380.
Names in Thought (1987). Philosophical Studies 51: 16985.
Subjective Intentionality (1987). Philosophical Topics 1: 89124.
Truth beyond All Verication (1987). In B. Taylor (ed.), Michael Dummett: Contributions to
Philosophy, Martinus Nijhoff Publishers.
A New Kind of Content (1988). In R. H. Grimm and D. D. Merrill (eds.), Contents of Thought:
Proceedings of the 1985 Oberlin Colloquium in Philosophy, University of Arizona Press.
Social Content and Psychological Content (1988). In R. H. Grimm and D. D. Merrill (eds.),
Contents of Thought: Proceedings of the 1985 Oberlin Colloquium in Philosophy, University
of Arizona Press. Reprinted in D. M. Rosenthal (ed.), The Nature of Mind, Oxford Univer-
sity Press, 1991. Reprinted in A. Pessin and S. Goldberg (eds.), The Twin-Earth Chronicles:
Twenty Years of Reection on Hilary Putnams the Meaning of Meaning, M. E. Sharpe,
1996 (reprinted Routledge 2015).
Phenomenal States (1990). Philosophical Perspectives 4: 81108.
Can We Explain Intentionality? (1991). In B. Loewer and G. Rey (eds.), Meaning in Mind:
Fodor and His Critics, Blackwell.
LIST OF WORKS BY BRIAN LOAR
Personal References (1991). In E. Villanueva (ed.), Information, Semantics and Epistemology,

Blackwell.
Elimination versus Non-reductive Physicalism (1992). In D. Charles and K. Lennon (eds.),
Reductionism, Explanation and Realism, Clarendon Press.
Can We Conrm Supervenient Properties? (1993). In E. Villanueva (ed.), Philosophical
Issues: Naturalism and Normativity 4: 7492.
Functionalism Can Explain Self-Ascription (1993). Behavioral and Brain Sciences 16,
1: 5860.
Self-Interpretation and the Constitution of Reference (1994). Philosophical Perspectives
8: 5174.
Meaning (1995). In R. Audi (ed.), The Cambridge Dictionary of Philosophy, Cambridge
University Press.
Reference from the First-Person Perspective (1995). Philosophical Issues: Content 6: 5372.
Comments on John Campbell, Molyneuxs Question (1996). Philosophical Issues, 7: 31924.
Phenomenal States: Second Version (1997). In N. Block, O. Flanagan, and G. Guzeldier
(eds.), The Nature of Consciousness: Philosophical Debates, MIT Press. Reprinted in
D. J. Chalmers (ed.), Philosophy of Mind: Classical and Contemporary Readings, Oxford
University Press, 2002. Reprinted in Y. Nagasawa, P. Ludlow, and D. Stoljar (eds.), Theres
Something about Mary: Essays on Phenomenal Consciousness and Frank Jacksons Know-
ledge Argument, MIT Press, 2004.
Is There a Good Epistemological Argument against Concept-Externalism? (1998). Philo-
sophical Issues 9: 21317.
Review of David Chalmerss The Conscious Mind (1999). Philosophy and Phenomenological
Research 59, 2: 46572.
Should the Explanatory Gap Perplex Us? (2000). Proceedings of the World Congress of
Philosophy: Metaphysics 2: 99104.
The Supervenience of Social Meaning on Speakers Meaning (2001). In G. Cosenza (ed.),
Paul Grices Heritage, Brepols.
Phenomenal Intentionality as the Basis of Mental Content (2003). In M. Hahn and
B. Ramberg (eds.), Reections and Replies: Essays on the Philosophy of Tyler Burge, MIT
Press/Bradford Books.
Qualia, Properties, Modality (2003). Philosophical Issues 13: Philosophy of Mind 1: 11329.
Transparent Experience and the Availability of Qualia (2003). In Q. Smith and A. Jokic
(eds.), Consciousness: New Philosophical Perspectives, Oxford University Press.
Language, Thought, and Meaning (2006). In M. Devitt and R. Handley (eds.), The Blackwell
Guide to the Philosophy of Language, Blackwell.
Thinking about Qualia (2007). In M. ORourke and C. G. Washington (eds.), Situating
Semantics: Essays on the Philosophy of John Perry, MIT Press.
Bibliography
Aydede, M. and Gzeldere, G. (2005). Cognitive Architecture, Concepts, and Introspection:

An Information-Theoretic Solution to the Problem of Phenomenal Consciousness, Nos 39, 2:
197255.
Balog, Katalin (1999). Conceivability, Possibility, and the Mind-Body Problem, The Philo-
sophical Review 108, 4: 497528.
Balog, Katalin (2012a). Acquaintance and the Mind-Body Problem, in C. Hill and
S. Gozzano (eds.), The Mental, the Physical, Cambridge: Cambridge University Press.
Balog, Katalin (2012b). In Defense of the Phenomenal Concept Strategy, Philosophy and
Phenomenological Research 84, 1: 123.
Block, Ned (1978). Troubles with Functionalism, in C. Wade Savage (ed.), Perception and
Cognition: Issues in the Foundations of Psychology, Minnesota Studies in the Philosophy of
Science, Minneapolis: University of Minnesota Press.
Block, Ned (1986). Advertisement for a Semantics for Psychology, Midwest Studies in
Philosophy 10: 61578.
Block, Ned (1990). Inverted Earth, Philosophical Perspectives 4: Action Theory and Philoso-
phy of Mind, 5179. Reprinted in W. Lycan (ed.), Mind and Cognition (2nd ed.), Blackwell,
1999.
Block, Ned (1996). Mental Paint and Mental Latex, Philosophical Issues 7, Perception, 117.
Block, Ned (2007). Max Blacks Objection to Mind-Body Identity, in Torin Alter and
Sven Walter (eds.), Phenomenal Concepts and Phenomenal Knowledge, Oxford: Oxford
University Press.
Block, N. and Stalnaker, R. (1999). Conceptual Analysis, Dualism, and the Explanatory Gap,
Philosophical Review 108: 146.
Burge, Tyler (1979). Individualism and the Mental, Midwest Studies in Philosophy 4:
73121. Reprinted in David M. Rosenthal (ed.), The Nature of Mind, New York: Oxford
University Press, 1991.
Burge, Tyler (1982). Other Bodies, in A. Woodeld (ed.), Thought and Object, Oxford:
Burge, Tyler (1986). Cartesian Error and the Objectivity of Perception, in John McDowell
and Philip Pettit (eds.), Subject, Thought, and Context, New York: Oxford University Press.
Burge, Tyler (2003). Phenomenality and Reference: Reply to Loar, in Martin Hahn and Bjrn
Ramberg (eds.), Reections and Replies: Essays on the Philosophy of Tyler Burge, Cambridge,
MA: MIT Press.
Carnap, Rudolf (1956). Meaning and Necessity. Chicago: University of Chicago Press, 2nd ed.
Carruthers, Peter (2004). Phenomenal Concepts and Higher-Order Experiences, Philosophy
and Phenomenological Research 68, 2: 31636.
Chalmers, David (1996). The Conscious Mind. New York: Oxford University Press.
BIBLIOGRAPHY
Chalmers, David (2002). Consciousness and its Place in Nature, in D. Chalmers (ed.),
Philosophy of Mind: Classical and Contemporary Readings, Oxford: Oxford University
Press. Repr. in S. Stich and F. Wareld (eds.), Blackwell Guide to the Philosophy of Mind,
Oxford: Blackwell, 2003.
Chalmers, David (2003). The Content and Epistemology of Phenomenal Belief, in Q. Smith
and A. Jokic (eds.), Consciousness: New Philosophical Perspectives, Oxford: Oxford
University Press.
Chalmers, David (2006). The Foundations of Two-Dimensional Semantics, in M. Garcia-
Carpintero and J. Macia (eds.), Two-Dimensional Semantics: Foundations and Applications,
Oxford: Oxford University Press.
Chalmers, David (2007). Phenomenal Concepts and the Explanatory Gap, in T. Alter and
S. Walter (eds.), Phenomenal Concepts and Phenomenal Knowledge: New Essays on Con-
sciousness and Physicalism, Oxford: Oxford University Press.
Chalmers, David (2009). The Two-Dimensional Argument against Materialism, in
B. McLaughlin (ed.), The Oxford Handbook of the Philosophy of Mind, Oxford: Oxford
University Press.
Chalmers, D. and F. Jackson (2001). Conceptual Analysis and Reductive Explanation,
Philosophical Review 110: 31561.
Chomsky, Noam (1980). Rules and Representations. New York: Columbia University Press.
Davidson, Donald (1967). Truth and Meaning, Synthese 17: 30423.
Davidson, Donald (1969). On Saying That, in Donald Davidson and Jaako Hintikka (eds.),
Words and Objections, Dordrecht: Reidel.
Davidson, Donald (1974). Belief and the Basis of Meaning, in Inquiries into Truth and
Interpretation, Oxford: Oxford University Press.
Davidson, Donald (1979). The Inscrutability of Reference, The Southwestern Journal of
Philosophy 10: 719.
Dennett, Daniel (1982). Beyond Belief, in Andrew Woodeld (ed.), Thought and Object,
Dennett, Daniel (1991). Consciousness Explained. Boston: Little Brown.
Devitt, Michael (1990). Meanings Just Aint in the Head, in George Boolos (ed.), Meaning
and Method, Cambridge: Cambridge University Press.
Donnellan, Keith S. (1966). Reference and Denite Descriptions, Philosophical Review 75:
281304.
Donnellan, Keith S. (1972). Proper Names and Identifying Descriptions, in Donald Davidson
and Gilbert Harman (eds.), Semantics of Natural Language, Dordrecht: Reidel.
Dretske, Fred (1981). Knowledge and the Flow of Information. Cambridge, MA: MIT Press.
Dretske, Fred (1988). Explaining Behavior. Cambridge, MA: MIT Press.
Dretske, Fred (1996). Phenomenal Externalism, Philosophical Issues 7, Perception, 14358.
Dummett, Michael (1976). William James Lectures [published as The Logical Basis of
Metaphysics, Cambridge, MA: Harvard University Press, 1991].
Dummett, Michael (1977). Elements of Intuitionism. Oxford: Oxford University Press.
Dummett, Michael (1978). Truth and Other Enigmas. Cambridge, MA: Harvard University
Press.
Dummett, Michael (1982). Realism, Synthese 52: 55112. Reprinted in M. Dummett, The
Seas of Language, Oxford: Oxford University Press, 1993.
BIBLIOGRAPHY
Evans, Gareth (1982). The Varieties of Reference. New York: Oxford University Press.
Evans, G. and McDowell, J. (eds.) (1976). Truth and Meaning: Essays in Semantics. Oxford:
Field, Hartry (1975). Conventionalism and Instrumentalism in Semantics, Nous 9: 375406.
Field, Hartry (1977). Logic, Meaning and Conceptual Role, Journal of Philosophy 74:
379409.
Field, Hartry (1978). Mental Representation, Erkenntnis 13: 961.
Field, Hartry (1991). Physicalism, in J. Earman (ed.), Empiricism, Explanation, and the
Philosophy of Science, Berkeley: University of California Press.
Field, Hartry (1994). Deationist Views of Meaning and Content, Mind 103, 411: 24985.
Fodor, Jerry (1975). The Language of Thought. Cambridge, MA: Harvard University Press.
Fodor, Jerry (1980). Methodological Solipsism, in Richard Boyd, Philip Gasper, and
J. D. Trout (eds.), The Philosophy of Science (1991), Cambridge, MA: MIT Press.
Fodor, Jerry (1987). Psychosemantics. Cambridge, MA: MIT Press.
Fodor, Jerry (1989). Making Mind Matter More, Philosophical Topics, 17: 5979.
Fodor, Jerry (1990a). A Theory of Content and Other Essays. Cambridge, MA: MIT/Bradford.
Fodor, Jerry (1990b). A Theory of Content II: The Theory, in A Theory of Content and Other
Essays, Cambridge, MA: MIT/Bradford.
Fodor, Jerry (1991). A Modal Argument for Narrow Content, Journal of Philosophy 88: 526.
Fodor, Jerry (1998). Concepts: Where Cognitive Science Went Wrong. Oxford, New York:
Clarendon Press.
Grice, H. P. (1957). Meaning, Philosophical Review 66: 37788.
Grice, H. P. (1968). Utterers-meaning, sentence-meaning, and word-meaning, Foundations
of Language 4: 22542.
Grice, H. P. (1969). Utterers Meaning and Intentions,Philosophical Review 78: 14777.
Harman, Gilbert (1973). Thought. Princeton, NJ: Princeton University Press.
Harman, Gilbert (1982). Conceptual Role Semantics, Notre Dame Journal of Formal Logic
23: 24256.
Harman, Gilbert (1982a). Beliefs and Concepts: Comments on Brian Loar, Must Beliefs Be
Sentences? in P. Asquith and T. Nickles (eds.), Proceedings of the 1982 Biennial Meeting of
the Philosophy of Science Association 49, 2: 65461.
Harman, Gilbert (1990). The Intrinsic Quality of Experience, Philosophical Perspectives 4:
Action Theory and Philosophy of Mind, 3152.
Harman, Gilbert (1996). Explaining Objective Color in Terms of Subjective Reactions, in
E. Villanueva (ed.), Philosophical Issues 7, Perception.
Hill, Christopher (1991). Sensations: A Defense of Type of Materialism. Cambridge: Cambridge
University Press.
Hill, C. and McLaughlin, B. (1999). There Are Fewer Things Than Are Dreamt of in
Chalmers Philosophy, Philosophy and Phenomenological Research 59: 44554.
Horgan, T. (1989). Mental Quasation, Philosophical Perspectives 3: 4774.
Horgan, T. and Graham G. (2009). Phenomenal Intentionality and Content Determinacy, in
R. Schantz (ed.), Prospects for Meaning, Amsterdam: de Gruyter.
Horgan, T. and Tienson, J. (2002). Phenomenology of Intentionality and the Intentionality of
Phenomenology, in D. J. Chalmers (ed.), Philosophy of Mind: Classical and Contemporary
Readings, Oxford: Oxford University Press.
BIBLIOGRAPHY
Horgan, T., Tienson, J., and Graham, G. (2004). Phenomenal Intentionality and the Brain in a
Vat, in R. Schantz (ed.), Externalist Challenge: New Studies on Cognition and Intentionality,
Amsterdam: de Gruyter.
Horgan, T., Tienson, J., and Graham, G. (2006). Internal-World Skepticism and Mental Self-
Presentation, in U. Kriegel and K. W. Williford (eds.), Self-Representational Approaches to
Consciousness, Cambridge, MA: MIT Press.
Hume, David (173940). A Treatise of Human Nature.
Jackson, Frank (1982). Epiphenomenal Qualia, Philosophical Quarterly 32: 12736.
Jackson, Frank (1986). What Mary Didnt Know, Journal of Philosophy 83: 2915.
Jackson, Frank (1998). From Metaphysics to Ethics: A Defense of Conceptual Analysis. Oxford:
Jackson, Frank (2003). Mind and Illusion, in A. OHear (ed.), Minds and Persons,
Cambridge: Cambridge University Press.
Jackson, Frank and Pettit, Philip (1990). Program Explanation: A General Perspective,
Analysis 50, 2: 10717.
Kaplan, David (1969). Quantifying in, in D. Davidson and J. Hintikka (eds.), Words and
Objections: Essays on the Work of W. V. Quine, Dordrecht: Reidel.
Kaplan, David (1989). Demonstratives, in J. Almog, J. Perry, H. Wettstein (eds.), Themes
from Kaplan, Oxford: Oxford University Press.
Kriegel, Uriah (2011). Sources of Intentionality. Oxford: Oxford University Press.
Kriegel, Uriah (2013). The Phenomenal Intentionality Research Programme, in Phenomenal
Intentionality, Oxford: Oxford University Press.
Kriegel, Uriah (2014). Two Notions of Mental Representation, in Uriah Kriegel (ed.),
Current Controversies in Philosophy of Mind, New York: Routledge.
Kripke, Saul (1972a). Naming and Necessity. Cambridge, MA: Harvard University Press.
Kripke, Saul (1972b). Naming and Necessity, in Donald Davidson and Gilbert Harman
(eds.), Semantics of Natural Language, Dordrecht: Reidel.
Kripke, Saul (1979). A Puzzle about Belief, in A. Margalit (ed.), Meaning and Use, Dordrecht
and Boston: Reidel. Reprinted in Philosophical Troubles: Collected Papers, I, Oxford: Oxford
University Press, 2011.
Lepore, Ernie and Loewer, Barry (1989). More on Making Mind Matter, Philosophical Topics
17, 1: 17591.
Levin, Janet (1985). Functionalism and the Argument from Conceivability, Canadian Jour-
nal of Philosophy Supplementary Volume 15: 85104.
Levin, Janet (1986). Could Love Be Like a Heatwave? Philosophical Studies 49: 24561.
Levin, Janet (2007). What Is a Phenomenal Concept? in T. Alter and S. Walter (eds.),
Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and
Physicalism, Oxford: Oxford University Press.
Levine, Joseph (1983). Materialism and Qualia: The Explanatory Gap, Pacic Philosophical
Quarterly 64: 35461.
Levine, Joseph (1993). On Leaving Out What It Is Like, in M. Davies and G. Humphreys
(eds.), Consciousness, Oxford: Blackwell.
Lewis, David (1966). An Argument for the Identity Theory, Journal of Philosophy 63,
1: 1725.
Lewis, David (1969). Convention. Cambridge, MA: Harvard University Press.
BIBLIOGRAPHY
Lewis, David (1972). General Semantics, in Donald Davidson and Gilbert Harman (eds.),
Semantics of Natural Language, Dordrecht: Reidel.
Lewis, David (1975). Languages and Language, in Gunderson, K. (ed.), Minnesota Studies in
the Philosophy of Science, Minneapolis: University of Minnesota Press.
Lewis, David (1979). Attitudes De Dicto and De Se, Philosophical Review 88: 51314.
Lewis, David (1981). What Puzzling Pierre Does Not Believe, Australasian Journal of
Philosophy 59: 2839.
Lewis, David (1983a). Languages and Language, in Philosophical Papers, vol. 1, Oxford:
Lewis, David (1983b). Philosophical Papers, vol. 1. Oxford: Oxford University Press.
Lewis, David (1983c). Mad Pain and Martian Pain, in Philosophical Papers, vol. 1, Oxford:
Loar, Brian (1976). Two Theories of Meaning, in G. Evans and J. McDowell (eds.), Truth and
Meaning, Oxford: Oxford University Press.
Loar, Brian (1980). Metaphysical Realism and Semantic Realism, paper read at the Pacic APA.
Loar, Brian (1981). Mind and Meaning. Cambridge: Cambridge University Press.
Loar, Brian (1982). Must Beliefs Be Sentences? in P. Asquith and T. Nickles (eds.), Proceed-
ings of the 1982 Biennial Meeting of the Philosophy of Science Association 49, 2: 62743.
Loar, Brian (1987a). Names in Thought, Philosophical Studies 51: 16985.
Loar, Brian (1987b). Subjective Intentionality, Philosophical Topics 1: 89124.
Loar, Brian (1988a). A New Kind of Content, in R. H. Grimm and D. D. Merrill (eds.),
Contents of Thought: Proceedings of the 1985 Oberlin Colloquium in Philosophy, Tucson:
University of Arizona Press.
Loar, Brian (1988b). Social Content and Psychological Content, in R. H. Grimm and
D. D. Merrill (eds.), Contents of Thought: Proceedings of the 1985 Oberlin Colloquium in
Philosophy, Tucson: University of Arizona Press.
Loar, Brian (1990). Phenomenal States, Philosophical Perspectives 4, 81108.
Loar, Brian (1991a). Can We Explain Intentionality? in B. Loewer and G. Rey (eds.),
Meaning in Mind: Fodor and His Critics, Oxford: Blackwell.
Loar, Brian (1991b). Personal References, in E. Villanueva (ed.), Information, Semantics and
Epistemology, Oxford: Blackwell.
Loar, Brian (1994). Self-Interpretation and the Constitution of Reference, Philosophical
Perspectives 8: 5174.
Loar, Brian (1995). Reference from the First-Person Perspective, in Philosophical Issues 6,
Content, 5372.
Loar, Brian (1997). Phenomenal States: Second Version, in N. Block, O. Flanagan, and
G. Guzeldere (eds.) The Nature of Consciousness: Philosophical Debates, MIT Press.
Loar, Brian (2003). Phenomenal Intentionality as the Basis of Mental Content, in M. Hahn
and B. Ramberg (eds.), Reections and Replies: Essays on the Philosophy of Tyler Burge,
Cambridge, MA: MIT Press/Bradford Books.
Lycan, William (1996). Layered Perceptual Representation, in E. Villanueva (ed.), Philosoph-
ical Issues 7, Perception, 81100.
Lycan, William (2008). Phenomenal Intentionalities, American Philosophical Quarterly 45:
23352.
Margalit, A. (ed.) (1979). Meaning and Use. Dordrecht: Reidel.
BIBLIOGRAPHY
McDowell, John (1986). Singular Thought and the Extent of Inner Space, in J. McDowell and
P. Pettit (eds.), Subject, Thought, and Context, Oxford: Oxford University Press.
McGinn, Colin (1981). The Structure of Content, in A. Woodeld (ed.), Thought and Object,
McGinn, Colin (1982). Realist Semantics and Content-Ascription, Synthese 52: 11334.
McGinn, Colin (1989). Mental Content. Oxford: Blackwell.
McGinn, Colin (1993). Consciousness and Cosmology: Hyperdualism Ventilated, in
M. Davies, and G. Humphreys (eds.), Consciousness, Oxford: Blackwell.
Nagel, Thomas (1974). What Is It Like to Be a Bat? Philosophical Review 83: 43550.
Nagel, Thomas (1986). The View from Nowhere. Oxford: Oxford University Press.
Nemirow, Lawrence (1980). Review of Nagels Mortal Questions, Philosophical Review 89:
4737.
Papineau, David (2002). Thinking about Consciousness. Oxford: Oxford University Press.
Papineau, David (2006). Phenomenal and Perceptual Concepts, in T. Alter and S. Walter
(eds.), Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and
Physicalism, Oxford: Oxford University Press.
Peacocke, Christopher (1983). Sense and Content. Oxford: Oxford University Press.
Perry, John (2001). Knowledge, Possibility and Consciousness. Cambridge, MA: MIT Press.
Putnam, Hilary (1975). The Meaning of Meaning, Minnesota Studies in the Philosophy
of Science 7: 13193. Reprinted in Mind, Language, and Reality: Philosophical Papers, vol. 2,
Cambridge: Cambridge University Press, 1975.
Putnam, Hilary (1978). Reference and Understanding, in his Meaning and the Moral
Sciences, London: Routledge & Kegan Paul.
Putnam, Hilary (1981). Reason, Truth, and History. Cambridge: Cambridge University Press.
Quine, W. V. O. (1960). Word and Object. Cambridge, MA: MIT Press.
Quine, W. V. O. (1964). From a Logical Point of View. Cambridge: Cambridge University
Press, 2nd ed.
Quine, W. V. O. (1966). Quantiers and Propositional Attitudes, in W. V. O. Quine, The
Ways of Paradox, New York: Random House.
Rcanati, Franois (1993). Direct Reference. Oxford: Blackwell.
Rey, Georges (1995). Towards a Projectivist Account of Conscious Experience, in T. Metzinger
(ed.), Conscious Experience, Paderborn: Ferdinand Schningh and Exeter: Imprint Academic.
Rey, Georges (1997). Contemporary Philosophy of Mind. Oxford: Blackwell.
Schiffer, Stephen (1972). Meaning. Oxford: Oxford University Press.
Schiffer, Stephen (1977). Naming and Knowing, Midwest Studies in Philosophy 2: 2841.
Schiffer, Stephen (1981a). Indexicals and the Theory of Reference, Synthese 49: 43100.
Schiffer, Stephen (1981b). Truth and the Theory of Content, in H. Parret and J. Bouveresse
(eds.), Meaning and Understanding, Berlin: Walter de Gruyter.
Schiffer, Stephen (1987). Remnants of Meaning. Cambridge, MA: MIT Press.
Schiffer, Stephen (1990). Physicalism, in Philosophical Perspectives 4: 15385.
Schiffer, Stephen (1991). Ceteris Paribus Laws, Mind 100: 117.
Schiffer, Stephen (1992). Belief Ascription, Journal of Philosophy 89: 499521.
Schiffer, Stephen (1993). Actual Language Relations, Philosophical Perspectives 7: 23158.
Searle, John (1991). Consciousness, Unconsciousness, and Intentionality, Philosophical
Issues 1: 4566.
BIBLIOGRAPHY
Searle, John (1992). Rediscovery of Mind. Cambridge MA: MIT Press.

Shani, I. (2008). Against Consciousness Chauvinism, Monist 91, 2: 294323.
Shoemaker, Sydney (1994). Phenomenal Character, Nous 28: 2138.
Siewert, Charles (1998). The Signicance of Consciousness. Princeton, NJ: Princeton University
Press.
Snowdon, P. F. (1981). Perception, Vision and Causation, Proceedings of the Aristotelian
Society, 81: 12150.
Spelke, Elizabeth (1995). Object Perception, in Alvin Goldman (ed.), Readings in Philosophy
and Cognitive Science, Cambridge, MA: MIT/Bradford.
Stalnaker, Robert (1976). Propositions, in A. Mackay and D. Merrill (eds.), Issues in the
Philosophy of Language, New Haven, CT: Yale University Press.
Stalnaker, Robert (1978). Assertion, Syntax and Semantics, 9: 31532. Reprinted in Context
and Content, Oxford: Oxford University Press, 1999.
Stalnaker, Robert (1984). Inquiry. Cambridge, MA: MIT Press.
Stampe, Dennis (1977). Toward a Causal Theory of Linguistic Representation, Midwest
Studies in Philosophy 2, 1: 4263.
Stich, Stephen (1983). From Folk Psychology to Cognitive Science. Cambridge, MA: MIT/
Bradford.
Stoljar, Daniel (2005). Physicalism and Phenomenal Concepts, Mind & Language 20, 5:
46994.
Strawson, Galen (2008). Real Intentionality 3: Why Intentionality Entails Consciousness, in
Strawson, G., Real Materialism and Other Essays, Oxford: Oxford University Press.
Strawson, P. F. (1964). Intention and Convention in Speech Acts, Philosophical Review 73:
43960.
Tye, Michael (1995). Ten Problems of Consciousness. Cambridge, MA: MIT Press.
Tye, Michael (2003). A Theory of Phenomenal Concepts, Philosophy 53: 91105.
Warner, Richard (1986). A Challenge to Physicalism, Australasian Journal of Philosophy 64:
24965.
White, Stephen (1982). Partial Character and the Language of Thought, Pacic Philosophical
Quarterly 63: 34765.
Index
Austin, J.L. 274 Horgan, Terence 137 n., 150 n.

Aydede, Murat 137 n. Humberstone, Lloyd 117 n.
Ayer, A.J. 274 Hume, David 97 n., 100
Bach, Kent 247, 292 n. Jackson, Frank 138, 141 n., 142, 196 n., 1979,
Balog, Katalin 137 n. 246 n.
Berkeley, George 97, 100, 116, 117, 122
Block, Ned 137 n., 138 n., 149 n., 167 n., 195, Kant, Immanuel 97, 100, 122
218, 276, 282, 290 n., 301 n., 307 n., 308 Kaplan, David 23 n., 72 n., 129, 141 n., 171 n.
Bostock, Derek 117 n. Kriegel, Uriah 137 n., 149 n.
Brentano, Franz 41 n., 104, 1067, 165, 181, Kripke, Saul 8, 10, 37 n., 50 n., 63 n., 656,
188, 274 8491, 117, 131 n., 138, 139, 1401, 156,
Burge, Tyler 96 n., 1312, 134, 1401, 143, 151 n., 158 n., 199, 223 n., 291 n., 2989
15464, 166, 16870, 185 n., 291 n., 2934, 294
n., 295, 2989, 309 n., 311 n., 316 Leibniz, Gottfried Wilhelm 1920, 197
Lepore, Ernest 247
Carnap, Rudolf 367 Levin, Janet 137 n., 196
Carruthers, Peter 137 n. Levine, Joseph 210 n.
Chalmers, David 137 n., 138, 141 n., 149 Lewis, David 4, 7, 8, 31 n., 40, 45, 47, 48 n., 93,
Chomsky, Noam 7, 456, 12930 95, 129, 138 n., 162 n., 2089
Church, Alonzo 36 Loewer, Barry 247
Craig, William 276 n. Lycan, William 151 n., 273 n., 274, 277, 301 n.,
3067, 308
Davidson, Donald 7, 2937, 126, 168 n., 259 n.
Dennett, Daniel 138 n., 141 n., 162 n., 169, McDowell, John 7, 290, 292 n., 297, 309 n.
172 n., 196 n. McGinn, Colin 69 n., 106 n., 196 n., 299 n.
Descartes, Ren 188, 251, 274, 317 McLaughlin, Brian 137 n.
Donnellan, Keith 42 n., 4950, 60 Meinong, Alexius 277, 278 n., 304
Dretske, Fred 145 n., 168 n., 273 n., 274 Mill, John Stuart 10, 8495
Duhem, Pierre 119 Moore, G.E. 244
Dummett, Michael 7, 1112, 97123
Nagel, Thomas 138, 196 n., 212, 214
Evans, Gareth 7, 292 n., 309 n. Nemirow, Lawrence 2089
Field, Hartry 69 n., 70 n., 77, 78, 168 n., 253 n., Papineau, David 137 n.
258 n., 259 n., 260 n., 297n., 300 n. Peacocke, Christopher 7, 54 n., 2767
Fodor, Jerry 4, 69 n., 70 n., 77, 7980, 141 n., Peirce, Charles Sanders 1001, 11011, 122
142, 1456, 163 n., 168 n., 169 n., 171 n., Perry, John 137 n.
22036, 254 Pettit, Philip 246 n.
Foster, J.A. 7 Picasso, Pablo 3078
Frege, Gottlob 5, 8, 1415, 17, 36, 4951, 53, Plato 188, 21819, 2412
923, 129, 180, 182, 234, 279, 2912 Putnam, Hilary 4, 71, 82, 106 n., 109, 110, 112,
122, 131, 1401, 143, 160, 244 n., 291 n., 2989
Graham, George 137 n., 150 n.
Grice, Paul 3, 45, 78, 12, 40, 48 n., 12434 Quine, W.V.O. 3, 5, 6, 1420, 22, 23, 28 n., 29,
Gzeldere, Gven 137 n. 47, 1079, 119, 122, 1478, 186, 271
Harman, Gilbert 69 n., 77, 195, 274, 277, 301 n., Ramsey, Frank 167
304 n., 3056, 308 Rcanati, Franois 309 n., 311 n.
Hill, Christopher 137 n., 277, 301 n. Reid, Thomas 100
INDEX
Rey, Georges 196 n., 210 Stampe, Dennis 168 n.

Ricketts, Thomas 54 n. Stern, Laurent 275 n.
Russell, Bertrand 5, 49, 124, 174, 279 Stich, Stephen 96 n., 153, 158, 165 n.
Strawson, Peter 3, 5, 7, 1278
Schiffer, Stephen 3, 5, 40, 42 n., 48 n., 63 n., Sturgeon, Scott 286 n.
69 n., 94 n., 96 n., 1279, 130 n., 1467,
193 n., 23845, 247, 249 n., 253 n., 259, Tarski, Alfred 7, 30, 33, 38, 109
269 n., 292 n. Tienson, John 137 n., 150 n.
Searle, John 149 n. Tye, Michael 137 n., 274
Sellars, Wilfred 178 n.
Shoemaker, Sydney 275, 276 n., 283, 301 n. Wagner, Steven 227
Siewert, Charles 137 n. Warner, Richard 216 n.
Smart, J.J.C. 305 White, Stephen L. 72 n., 141 n.,
Smith, Barry 286 n. 171 n.
Snowdon, Paul F. 290 Wiggins, David 7
Spelke, Elizabeth 297 Wittgenstein, Ludwig 124, 131 n., 2067,
Stalnaker, Robert 79, 138 n., 141 n., 145 n., 168 n. 223 n., 2512

Consciousness and Meaning Selected Essays

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Consciousness and Meaning Selected Essays

Hochgeladen von

Copyright:

Verfügbare Formate

Consciousness and Meaning

Katalin Balog and Stephanie Beardman

Stephen Schiffer and Katalin Balog

Part I. Philosophy of Language

Part II. Philosophy of Mind

List of Works by Brian Loar 319

intention of affecting an audience in certain ways. In the second part, expression-

Chapter 2 Two Theories of Meaning (1976). This chapter reveals Brians

intentions, and other psychological properties, including causal relations to external

This, then, is Quines version of (1).

xFx ! Gx; Fa; so; Ga:

(iv) The analysis of certain complex sentences would be simplied if we could

(24) (Ey) (y is a cabinet member & B (Ralph, x is a spy, y)).

Does it follow that

There is no generally valid inference from a quoted occurrence of a conjunction to

The trouble with the peculiar sentence

belief context. For the expression B, as used in my examples, expressed a triadic

unlikely that in reporting certain noncognitive psychological states we advert to, or

Classical analyses of belief sentences may be thought of as proceeding in at least

extensional apparatus. For an equivalent difculty arises at the level of satisfaction

the p-position is an acceptable translation of the sentence of L which is mentioned in

First, if sentences stand in translational relations to other sentencesfor example,

(1) There are various notions specically involving types of communicative

In the preceding section, I mentioned as a kind of language a function from

So a sentences meaning may be represented, in an abstract way, as an ordered

L is the language of P iff

The denition of sentence-meaning follows simply: S means Mi in P iff, for some L,

Hence, there is a non-associative improbability that he is M0 -ing, his utterance is not

The sentences of English we understand are enormous though nite in number.

Thirdly, consider the following exchange:

Several observations are in order here.

Whether a denite description or individual concept identies an object for us

that is intrinsic is an identifying one. Uses of demonstratives, personal pronouns, and

certain way, of existential generalization.10 Let us capture this requirement by saying

It is compatible with a certain continuous philosophical usage to use refers in that

The behavior of referential occurrences of singular terms in modal contexts can

1 A Propositional Attitude Based Theory

functions from contexts to specic truth conditions. We may count Twin-Earth

(7) B (Twin-Earthers, we bathe in water, (Twin-Earthers, XYZ)).

How might common-sense generalizations about propositional attitudes secure

interchangeable as interpersonal constraints, and why one reference-function

2 Comparison of the Two Theories at the

(i) A scientic motivation lies in computational models in psychology; Jerry

Different functionalist frameworks for explicating propositional attitudes indi-

(10) s means p in xs language of thought L

there is a systematic truth theory T of L such that T maximizes E (informative

(11) s means p in xs language of thought L

Another way in which interpersonally ascribable conceptual roles may make a

1 A Puzzle about Belief

2 Is Anything Here Puzzling?

of translation is not ours. There is no contradiction in the Hebrew belief ascriptions;

3 Jones and Pierre

translation and try to be as sensitive as possible to how beliefs would be expressed

5 Psychological Content and de Dicto That-clauses

6 Millianism about Names in Thought

If psychological consistency is a matter of consistency in the content of thoughts,

7 The Lessons of Pierre

is of course to be understood as about some specic class of statements, and my

this: there is no bona de concept of truth on which the truth (conditions) of

Table 6.1. Candidate realisms

realists commitment to a truth-conditional semantics is, I think, of something that

us as the state of affairs whose verication procedure it is supposed to be. So the

we ignore indexicality and Twin-Earth phenomena, even if somehow the conceptual

is posed as a question about understanding. Dummetts challenge is to say how we