Introduction

Introduction There's something odd about the history of cognitive
theories. On the one hand, practically all of them, from Descartes forward,
have been thoroughly committed to mental representations as explanatory
constructs. But, on the other hand, a continuing critical tradition in both
philosophy and psychology argues that the mental representation construct
is inherently defective and cannot be made scientifically respectable. This
has been going on for a long time.1 It's a bit as though physics had developed
in parallel with a line of criticism which claimed that the notion of a particle
is incoherent and must be dispensed with. Surely, one would think, some sort
of resolution should eventually be achieved: either the criticisms should be
shown to be misdirected, or we should give up the construct criticized. One
would think, too, that there ought to be some way of telling whether one's
theoretical commitments are incoherent, and that three hundred years or
so ought to be long enough to find out.
Anyhow, the sky is falling again. We have a cognitive science whose
main tenet is that the mind is a device for the manipulation of representations.
But we also have a line of philosophical criticism that goes like this: Nothing
is a representation except insofar as it has representational content, and the
notion content of a mental representation is in jeopardy. In particular, there's
a new argument that is taken to show that, even if there are mental representations,
and even if mental representations have contents, still the content of
a mental representation is not a function of psychological variables as cognitive
scientists understand such variables. So, to that extent, the notion content of
a mental representation is not available as an explanatory construct in theories
of the sort that cognitive scientists have hoped to develop.
Now, the arguments currently fluttering the dovecotes are actually rather

indirectly related to this unsettling conclusion,

COGNITIVE SCIENCE 99
indirectly related to this unsettling conclusion. In fact, these arguments arose,
in the first instance, out of discussions of lexical meaning.2 Working out just
how the issues about lexical meaning connect with the issues about mental
representation is actually not easy much of this paper will be devoted to doing
so. For starters, though, here's the original argument without comment or
elaboration.
Hilary Putnam (see especially [9]) imagines a place that's just like here
except for certain peculiarities of microchemistry. Call this place "Twin Earth"
("Earth2" for short). On Earth2 they speak a language that is just like English
in respect of its phonological, morphological, and syntactic properties. They
call this language by a word which they pronounce /English/ and which they
write "English" (but which we will write "English2" in aid of notational
perspicuousness). Since English2 is phono-morpho-syntactically just like English,
it contains a word (which we will write as "water2") that is pronounced
/water/.
The microchemical difference between Earth and Earth2 is that, although
they, like us, have a transparent fluid that they drink, sail on, wash thencars
with, and refer to by the vocable /water/, and although that fluid passes
all the, as it were, phenomenological tests for water (it has specific gravity 1,
it freezes at zero C, and so forth), still the stuff that looks like water on Earth2
is, in point of chemical fact, made of XYZ {Φ H2O).
Putnam's intuitions about what's going on on Earth2 run like this:
a. What English2 speakers refer to by using the word "water2" is not
water.
b. English2 expressions like "water2 is wet" have different truth conditions
from the homophonic expressions of English. In particular,
unlike the English homophone, the truth of "water2 is wet" does not
depend upon the wetness of water. Hence,
c. "water2" is not the same word as "water"; the two words differ in
their semantic properties.
It follows trivially that English Φ English2 ("water" occurs in one language
but not in the other). Since, however, we have assumed that the difference
between XYZ and H2O is the only (relevant) difference between Earth and
Earth2, it also follows that people who are as similar as you like in their
physical constitution (people who are, as we shall say, molecularly identical*)
may nevertheless speak different languages. Notice that it has not been shown
to follow (yet) that molecularly identical people may mean different things
by what they say; the grounds for that latter inference remain to be explored.
So much for Earth2. Its philosophical chemistry is, I suppose, now sufficiently
well publicized that we needn't develop the example in further detail.
However, before we get to our main topic, which is what Putnam's case is
supposed to show about the notion content of a mental representation, it is
desirable to understand something of what cognitive scientists have wanted
that notion for; what role appeals to content are supposed to play in the
sorts of explanations that cognitive science seeks to provide. Hence the following
section.

Why Mental Representations Have To Have Contents,

100 J. A. FODOR
Why Mental Representations Have To Have Contents Quantification overtheoretical
commitment to—mental representations is what cognitive science
has in common with the Classical tradition in epistemology as it developed in
both Rationalist and Empiricist versions. In particular, according to both
Classical and current theories, behavior is typically the effect of mental
processes, mental processes are typically causal sequences of mental operations,
and mental operations have mental representations as their domains. This
general picture, which I have elsewhere (see [3]) called the Representational
Theory of Mind (RTM) is presumably well known and I shall spare the reader
further exposition.
However, current cognitive science4 differs in important respects from
earlier versions of RTM. In particular, the contemporary movement is explicit
in endorsing the claim that mental processes are computational: mental
operations apply to mental representations in virtue of "formal" or "syntactic"
(or, anyhow, nonsemantic) properties of the representations.5 So, to the
extent that one assumes that the content of a mental representation is some
sort of construct out of its semantic properties, it follows that mental operations
are defined without reference to the content of the representations they
apply to. (I am putting this very loosely; the question of the relation between
the semantic properties of a representation and its content will loom large later
on and I do not want to prejudice the issue at this stage.)
Now suppose (as above) that behavior is to be explained by exhibiting
its contingency upon the mental processes that cause it. And suppose that
every mental process is a sequence of mental operations and that mental
operations apply to mental representations in virtue of the form of the
representations. Then it looks as though it does not matter whether the notion
of the content of a mental representation is in jeopardy since it looks as though
that notion is never going to be required in order to give the explanations
that cognitive scientists want to give. The idea that mental operations are
formal can thus be taken to imply that the content of a mental representation
is a dispensible construct, at least for the purpose of cognitive science. Indeed,
some philosophers have read the moral in just this way: see, for example,
[12], [13], and some of the remarks in [10].
I chink, however, that this line of argument is not well-advised. Roughly,
the point is this: we want not just to be able to characterize the causal chains
upon which behavior is contingent, but also to state such true and counterfactual
supporting generalizations about the etiology of behavior as there are
to be stated. But, to put the point in a nutshell, it looks as though such
generalizations typically hold in virtue of intentional properties of the
behaviors that they subsume, and it looks as though we shall need to advert
to the content of mental representations as part of our account of the intentionality
of behavior.
Notice that just about all of the familiar, rough-and-ready examples
of generalizations about the way in which behavior is contingent on mental
states and processes appear to make essential reference to intentional properties
of the behaviors they apply to. Consider, for an instance, that paradigm of
spruced-up common-sense psychology, the practical syllogism. It says something
about how someone will act (or, if action is thwarted, what the agent
will try to do; or, at a minimum, about the properties that the agent would

The generalization that the practical syllogism,

COGNITIVE SCIENCE 101
prefer his behavior to exhibit) given that he has certain specified beliefs and
desires. The generalization that the practical syllogism (and, mutatis mutandis,
other decision theories) articulates thus applies to behavior under intentional
description; and so, in similar ways, do the rest of our folk psychology and
practically all of our cognitive science.
Now, RTM proposes a two-step reduction of the intentionality of
behavior to the content of mental representations. Obviously, the details
are much in dispute. But, very schematically, the idea is that (step 1) for
behavior to have such and such an intentional property involves its being
caused by a mental state having the corresponding propositional content;
and (step 2) to have a mental state with the propositional content that P is
to be related, in a certain way, to a mental representation which expresses
the proposition that P. Attempts to bring it about that P are thus explained
by reference to intentions to bring it about that P, and intentions to bring it
about that P are in turn explained by reference to mental representations
which, in effect, mean that P. What is essential to this pattern of explanation
is that the first step accounts for the intentionality of behavior by reference
to the content of a (causally efficacious) propositional attitude, and the
second step accounts for the content of the propositional attitude by reference
to the content of a mental representation.
It is worth emphasizing that, in canonical psychological explanations
of the sort that RTM contemplates, the required specifications of propositional
attitudes are characteristically de dicto rather than de re. (By a de dicto
specification of a propositional attitude, I mean approximately one in which
substitution of coreferring expressions does not, in general, preserve truth
unless the expressions are synonymous.) The point here is that de re specifications
of propositional attitudes are generally too weak to support explanations
of behavior when the latter is intentionally characterized. So, de re,
Oedipus' desire to marry Jocasta = his desire to marry his mother = his desire
to marry the tallest woman in Greece (assuming that Jocasta was the tallest
woman in Greece at the time when Oedipus desired to marry her). But it
is only the first of these specifications of what Oedipus desired (or maybe
the first two, depending on how you feel about depth psychology) that
figures in canonical explanations of the behavior that Oedipus tried/intended
to produce.
What all this comes down to, then, is that we need the notion of the
content of a mental representation to reconstruct the notion of the content
of a de dicto propositional attitude6; and we need the notion of a de dicto
propositional attitude in order to reconstruct the notion of the intentionality
of behavior; and we need the notion of the intentionality of behavior in order
to state a variety of psychological generalizations which appear to be (more
or less) counterfactual supporting and true, and which subsume behavior in
virtue of its satisfaction of intentional descriptions.
This does not, however, quite settle the issue; it still is not out of the
question that some purely syntactic (formal) specification of mental representations
might do the job. For example, it is conceivable that there should
be some formal property (call it "£/") that mental representations have iff
they express the property of being a unicorn, and some (different) formal

102 J. A. FODOR

102 J. A. FODOR
property (call it "W") that mental representations have iff they express the
property of being a witch. So, then, instead of explaining the difference
between Seymor's witch hunting and his unicorn hunting by reference to
differences between the contents of the causally implicated mental representations,
we could explain it by reference to the difference between U and W.
And, of course, if mental operations are indeed computational, there is going
to have to be some formal difference between the mental representation(s)
that mediate unicorn hunting and the one(s) that mediate witch hunting,
assuming that the difference between hunting unicorns and hunting witches
could show up (even counterfactually) in distinct behaviors. For, as we have
seen, it is the burden of the computational theory of mental operations that
only formal differences between mental representations can issue in distinct
behaviors; mental representations that differ only in their semantic properties
must ipso facto be identical in their causal roles.
There is, however, really no reason at all to suppose that there are formal
doppelgangers of each feature of the contents of mental representations that
we need to advert to in our accounts of the intentional properties of behavior.
Positing such "type-to-type" correspondences between formal and semantic
properties of mental representations involves a much stronger assumption
than that each causally efficacious difference in content must correspond
to some formal difference or other. Clearly, we have no right to build our
theories of mind upon this stronger assumption since it seems entirely possible
that formally quite different mental representations should be, as it were,
synonymous. (Dennett has emphasized, correctly in my view, that we shall
have to take this possibility especially seriously if we want ascriptions of
propositional attitudes to be comparable across individuals; a fortiori if we
want them to be comparable across species.) Moreover, one might feel, even
if there were such coextensions between features of content and features
of form, it would nevertheless be semantic facts, and not the syntactic ones,
that really account for intentionality. If, for example, there is something
about a mental representation that makes the behavior it causes unicorn
hunting rather than witch hunting, surely it is not something about the
shape of the representation; it is something that has to do with what the
representation is (or purports to be) about.
So, I think that there really are only two options given the general
framework of RTM. Either we assume that there are no explanatorily indispensible
intentional properties of behavior (specifically, that there are no
counterfactual supporting generalizations that subsume behavior in virtue
of its intentional properties) or we assume that the notion of the content
of a mental representation is ineliminable at least insofar as macrolevel
psychological theorizing is concerned. I simply cannot take the first of these
options seriously since we have—or so it seems to me—no notion of behavioral
systematicity at all except the one that makes behavior systematic under
intentional description. So I shall simply take it for granted that you cannot
save the cognitive science program by going syntactic. Either mental representations
are going to honest-to-God represent, or we are going to have to
find an alternative to RTM.

A Puzzle about Earth2 Beliefs,

COGNITIVE SCIENCE 103
A Puzzle about Earth2 Beliefs The upshot of the discussion so far is that
we need for purposes of theorizing about the intentional properties of behavior
the two coordinate constructs: de dicto specification of a propositional
attitude and content of a mental representation. If, then, we want to think
about the implications of the Earth2 examples for RTM, we need to ask
what they show about the de dicto propositional attitudes of people who talk
English2- We now turn to that question.
The first thing to say is that Putnam gives us very little help here. His
discussion is framed almost solely in terms of semantic and lexicographic
issues: e.g., in terms of such questions as "what does 'waterj mean!"1 Now,
one might suppose at first-blush that to settle questions about what "water2"
means is to settle the corresponding questions about the de dicto propositional
attitudes of speakers of English2. The reason one might suppose this is that
it is very natural to assume a "Gricean" account of the relation between the
meanings of linguistic forms and the de dicto propositional attitudes of
speaker/hearers of the language that contains the forms. The idea would be
that meanings are, as it were, logical constructs out of the de dicto propositional
attitudes of speaker/hearers (e.g., out of their de dicto havings and
recognizings of communicative intentions). There might be some uncertainty
about just which logical construction out of propositional attitudes meanings
are; but however that goes the view would be that to fix the ascription of
meanings to verbal forms is to presuppose, more or less uniquely, an ascription
of corresponding de dicto propositional attitudes to speaker/hearers.8
I am very much in sympathy with this sort of view, and it may be that
it can be reconciled with Putnam's intuitions about what "water2" means.
The present point, however, is that if one does accept Putnam's intuitions,
one cannot simply take the existence of a Gricean reduction of meanings
to propositional attitudes for granted.9 We will see more of this later (indeed,
it will become a major theme) but here is one relevant consideration: Putnam
wants to make the extension of a term one of the parameters of its meaning
so that, presumably, "water2" means XYZ together with some other stuff.
And Putnam also wants to argue that speaker/hearers need not know what the
terms they use refer to. It is, indeed, the conjunction of these two doctrines
that Putnam expresses by the slogan "meanings aren't in the head". However,
I suppose we can assume that people normally do know their own de dicto
intentions; that de dicto propositional attitudes are "in the head" even if
meanings are not. Surely this assumption is part and parcel of the sort of
reduction of meanings to propositional attitudes that Grice proposes. But if
this is right, it is hard to see how the view that "water2" means (something
like) XYZ could be squared with the idea that a word means what it does
because speaker/hearers have the de dicto propositional attitudes that they
have.
All this is pretty tentative. It would, for example, be possible to give
up the idea that people know their own de dicto propositional attitudes. You
might then manage a Pickwickean-Gricean reduction of meanings to communicative
intentions; Pickwickean because the communicative intentions to
which meanings are reduced would, in a relevant sense, not themselves be
psychological states. (And, of course, you would need to find some way of

Interpreting the de re/de dido distinction that does not appeal to RTM,

104 J. A. FODOR
interpreting the de re/de dido distinction that does not appeal to RTM; one
that does not assume that de dicto specifications exhibit the way the content
of a propositional attitude is mentally represented.)
Anyhow, for the moment I am not arguing that Putnam's views about
what "water2" means cannot be reconciled with Grice's views about how
meanings are related to mental states. I am arguing only that you should
not just assume, on Gricean grounds, that accepting Putnam's intuitions about
the meanings of English2 expressions closes the issue about how the communicative
intentions of English2 speakers ought to be described. To put
the point more succinctly: so far we have arguments that molecularly identical
people can speak different languages, but we have no argument for the
conclusion that would make a difference to cognitive scientists; viz., that
molecularly identical people can differ in their de dicto propositional attitudes.
One way to get the latter conclusion is to assume Grice's principle that
difference in de dicto propositional attitudes of speaker/hearers can be
inferred from differences in the meanings of the linguistic forms they use.
But this inference is not available to Putnam. Grice's principle cannot be
assumed by a theorist who holds that meanings are not in the head not at least
without some further tinkering.
So here is the question that I am claiming Putnam's discussion leaves
open and the answer to which I am claiming is essential to understanding
the implications of Putnam's examples for the cognitive science project. What
communicative intentions do speakers of English2 use such verbal forms
as "water2 is wet" to express? Or, to put much the same question slightly
differently: What de dicto belief is a speaker of English2 claiming to have
when he says that he believes that water2 is wet? Or, to put the question
slightly differently again: What statement is "water2 is wet" standardly
used to make in English2? (Perhaps these three questions are not in fact
equivalent; it hardly matters for the discussion that follows since it does
seem clear that if we knew how to answer any one of them we would be
well on our way to answering the rest.) I will now run through some answers
that are wrong in edifying ways. In the long run, I shall be claiming that there
is no way of answering these questions compatible with preserving Putnam's
intuitions about what "water2" means; hence that Putnam's intuitions must
be misled.
First gambit: "Water2 is wet" is used to express the de dicto belief that
water2 is wet. Reply, this proposal is unhelpful since it is part and parcel
of our quandary that we do not know which belief the belief that water2 is
wet is. One way to see the difficulty is to notice that since "water2" is not,
strictly speaking, an expression of English, the formula "'water 2 is wet'
expresses the belief that water2 is wet" is not, strictly speaking, well formed.
(Compare " Ί a plume est sur la table' is true iff la plume is on the table".)
What we need, of course, is a convention for understanding "water2" when
it occurs used (as opposed to mentioned) in English sentences. The present
proposal is thus unstable since what it claims about the de dicto propositional
attitudes of English2 speakers will depend entirely upon which such convention
we adopt. If, for example, we decide that "water2" translates as "water",
then the proposal reduces to " 'water2 is wet' expresses the de dicto belief that

water is wet". If, by contrast we translate "water2" as "XYZ",

COGNITIVE SCIENCE 105
water is wet". If, by contrast we translate "water2" as "XYZ", then the proposal
reduces to " 'water 2 is wet' expresses the de dicto belief that XYZ is wet";
and so forth for other possible construals of "water2". These various alternatives
will be discussed severally further on.
These considerations may suggest that there is some problem about
construing the communicative intentions of English2 speakers vis avis " water 2
is wet" that does not arise when we try to explicate our communicative
intentions with respect to the homonophonic expressions of English. And
it is true, of course, that " 'water is wet' expresses the belief that water is
wet", though perhaps uninformative, is at least well formed. However, the
appearance of asymmetry is surely spurious. When we do the combinatorial
part of semantics, we have some justification for simply assuming that the
vocabulary of the object language is available in the metalanguage of choice;
we thus take the semantics of "good" and "actress" for granted and show
how the semantics of "good actress" arises therefrom (see [2]). But, of course,
we cannot do that when we are embarked on Putnam's project, which is
precisely that of lexicographic analysis. If there is an issue about what "water2"
means, and if it is question-begging to answer that it means water 2i then
surely there must be the same issue about what "water" means and it must
be equally question-begging to answer that it means water. Maybe what this
shows is just that lexicography is a mug's game; indeed, I strongly suspect
that that is true. However, that line is unavailable to Putnam, whose discussion
is explicitly "almost entirely about the meaning of words rather than about
the meaning of sentences" ([9], p. 216). It must, in short, be obvious that
if Putnam's examples make the relation between meaning and de dicto
propositional attitudes problematic for English2, they must also do so for
English. The home language cannot be viewed as privileged in this sort of
study.
Second gambit: "Water2 is wet" is used to express the de dicto belief
that water is wet. Reply: I take it that we can set this proposal to one side;
not because it is obviously false (on the contrary, I shall eventually argue
that, for all Putnam has shown, it may well be true) but rather because if
it is true, then there is no Twin Earth problem for us to solve. As we saw
above, if Putnam's example is a problem for cognitive science, that is because
it seems to show that molecularly identical people can have de dicto propositional
attitudes that differ in content. What invites this conclusion is the
"Gricean" assumption that linguistic forms which differ in meaning must
ipso facto differ in the propositional attitudes they are used to express.
By contrast, the present proposal is that whatever may be the case with the
meanings of "water is wet" and "water2 is wet", they are used to express the
same propositional attitude: viz., that water is wet. On this account, the only
moral to be drawn from Putnam's examples would be the irrelevance of the
semantics of natural language expressions to the individuation of the propositional
attitudes of speaker/hearers. (This is a moral which I shall eventually
endorse, though on a very narrow construal. I propose to take the sting out
of it by suggesting (a) that the content of a linguistic expression should be
distinguished from such of its semantic properties as its truth conditions;

Third gambit: "Water2 is wet" is used to express,

106 J. A. FODOR
and (b) that content is—though truth conditions are not—a construct out of
the communicative intentions of speaker/hearers.)
Third gambit: "Water2 is wet" is used to express the de dicto belief
that what is called "water2" around here is wet ("around here" being used
to index Earth2). Reply: I put this one in because some things that Putnam
says about the indexicality of kind terms may suggest it. It is, however, highly
implausible and I very much doubt that Putnam actually has this solution in
mind. For one thing, it is too metalinguistic sounding; whatever belief "water2
is wet" is used to express is surely one that animals, prelinguistic children and
(nb) people who have never heard of the word "water2" can share; and none
of this would be so if one accepted the present proposal about what "water2
is wet" is used to say. To put this point quite generally: the belief that
"water2 is wet" expresses must turn out, on any acceptable analysis, to be
identical with the belief that water2 is wet (whatever belief that turns out to
be; see above). But, surely, it must be possible to have the belief that water2
is wet without having any metalinguistic beliefs at all. The belief that water2
is wet is a belief about water2, not a belief about language.
Such considerations suggest that the indexicality story does not provide
necessary conditions for being the belief that "water2 is wet" expresses. What
is more important, however, it does not provide sufficient conditions either.
Consider the parallel proposal that the (English) formula "tigers have stripes"
expresses the belief that what are called "tigers" around here have stripes.
Well, if this is true it must follow that having the belief that what are called
"tigers" around here have stripes is sufficient for having the belief that tigers
have stripes; for surely, that tigers have stripes is the belief that "tigers have
stripes" is used to express. But that's no good for the following reason: you
could have the belief that what are called "tigers" around here have stripes
and not have the belief that tigers have stripes if, for example, you happen
to think that what are called "tigers" around here are pyjamas. In that case,
you would have one false belief (viz., that pyjamas are called "tigers") and
one true one (viz., that pyjamas have stripes), the conjunction of which is,
patently, not equivalent to the belief that "tigers have stripes" is used to
express.
Similarly, with bells on, for such proposals as that "water2 is wet"
expresses the de dicto belief that this stuff is wet ("this stuff being used
to index some water2), You could believe that this stuff is wet while believing
that this stuff is, say, tomato juice. In that case, believing that this stuff is
wet would not be believing that water2 is wet, even though this stuff is, as
a matter of fact, both wet and water2.10
Fourth gambit: "Water2 is wet" is used to express the de dicto belief
that XYZ is wet. Reply: I think that this is what a lot of philosophers would
say who share Putnam's intuitions about how lexicography should be pursued.
For example, Burge [ 1 ] has recently accepted the corresponding solution for
a class of examples which, as he remarks, are in important respects quite similar
to Putnam's but do not involve terms for natural kinds. Nevertheless, it seems
clear to me that quite familiar considerations preclude taking this line. Since
the reasons for denying that "water2 is wet" expresses the de dicto belief that
XYZ is wet are equally reasons for not accepting Burge's account of the

COGNITIVE SCIENCE 107

COGNITIVE SCIENCE 107
examples he investigates, it may be worth a digression to run through one
of Burge's cases.
Burge asks us to accept, for purposes of argument, the following assumptions:
a. The fact that contracts need not be written (verbal contracts bind)
is constitutive of our concept of contract.
b. There is an English speaker (call him Jones) whose views about contracts
are much like ours except that he is misinformed about (what we
would call) "verbal contracts". In particular, Jones believes that contracts
must be written, hence that soi-disant verbal contracts are not
binding.
Burge's intuition is that we ought to say in this case that Jones has the
same concept of contract that we do, notwithstanding that he (Jones) denies
a truth that is, by assumption, constitutive of our concept of contract. Burge
also takes it (if I read him correctly) that when Jones utters (in otherwise
normal circumstances) "Smith just signed a contract", his utterance should
be taken to express the belief that Smith just signed a contract; i.e., to
express, inter alia, the very concept of which valid though not written is, by
assumption, constitutive. Burge's point is, approximately, that what concept
is expressed by what you utter is determined not (or not just) by what is "in
your head", but also by what concept is expressed by the homophonic
utterances of other speakers of your language. And, of course, it is a truism
that, for paradigmatic English speakers, "contract" expresses the concept
contract.
Burge is, in my view, putting more weight on the notion same language
than that notion will bear. As the linguists are forever reminding us, language
and language community, when not merely mystical concepts, are largely
geopolitical ones having much to do with who has got the gunboats. But
let us grant Burge same language and, while we are at it, let us grant him the
notion of a truth constitutive of a concept, Still, it seems to me, we cannot
grant Burge his intuitions about what belief Jones uses "Smith just signed
a contract" to express. For, surely, Jones expresses the same concept by
"contract" when he says that as when he says, for example, "I deny that
verbal contracts bind". But if the concept of contract expressed in this
latter case is our concept of contract (and if, by assumption, being binding
when verbal is constitutive of our concept of contract) then the belief that
Jones is expressing when he denies that verbal contracts bind is explicitly selfcontradictory.
Specifically, the belief expressed is that what is binding when
verbal is not binding when verbal. Notice, moreover, that we have to read
this belief de dicto it is not just that Jones believes of something which is as
a matter of fact so and so that it is not so and so (cf. Russell's "I thought
your yacht was longer than it is"). If it means anything to say that Jones
has our concept of contract, it must mean that we should construe his
utterances of "contract" in the same way we would construe our own. If,
however, we do translate that way, we get self-contradictions whenever Jones
says of verbal contracts what, by Burge's own assumption, Jones believes to
be true of them: viz., that there aren't any. I take it, however, that there is

What is common to Burge's example and Putnam's,

108 J. A. FODOR
a principle of charity which operates to prohibit accusing one's fellows of
inconsistency in this flagrant and inflammatory way.
What is common to Burge's example and Putnam's is that in each case
something that is taken to be part of the meaning of an expression that
speakers use is nevertheless assumed to be something that the speakers need
not grasp. In Burge's example, it is a necessary truth constitutive of the
meaning; in Putnam's example, it is the extension. My point is that that is all
right so far, but you get into trouble with the principle of charity if you
also make the assumption that, in effect, what a verbal form means is interchangeable
with the concept that it expresses: in the present case, that
"contract" expresses the concept contract or that, mutatis mutandis, "water2"
expresses the concept XYZ. For then all sorts of innocently false statements
("verbal contracts do not bind"; "water2 is not XYZ") are going to be taken
to render self-contradictory beliefs, and this the principle of charity forbids.
I have the impression that there are a lot of philosophers who think
it is all right to say that "water2" means XYZ or, to continue with Burge's
example, that Jones uses "contract" to express the concept contract. It may,
therefore, be worth making explicit the ingredients of the present bind and
considering in some detail the various options Burge has for getting out of it.
Burge's difficulties arise from the interaction of the following five
assumptions:
1. Verbal contracts bind is constitutive of the meaning of "contract".
2. The meaning of a word is a construct out of the concept it expresses;
in particular, words that express the same concept are synonymous.
3. When Jones says "contract" he expresses the same concept that we do
when we say "contract".
4. The principle of charity.
5. The intersubstitutability of synonymous expressions in de dicto belief
(or "says that") contexts.
Of these assumptions, Burge is more or less explicitly committed to 1 and 3,
and something like 2 would seem to be required if the relation expressing
the same concept as is to be distinguished from weaker forms of semantic
equivalence that words may enter into.11 Assumptions 4 and 5 strike me
as pretty plausible, but Burge might want to try ditching one or both. Neither
possibility seems very attractive, however, as we are about to see.
Suppose you give up the principle that you can substitute synonyms
for synonyms in de dicto specifications of beliefs. That would suffice to
block such inferences as from "Jones believes that (says that) verbal contracts
do not bind" to "Jones believes that (says that) what binds when not verbal
does not bind when not verbal"; similarly, mutatis mutandis, it would serve
to block the inference from "Jones2 doubts that water2 is XYZ" to "Jones2
doubts that XYZ is XYZ". The trouble is, however, that if you do block these
inferences, it is hard to see what is left of assumption 1 (or, mutatis mutandis,
of the claim that "water2" means XYZ.) Claims about lexical meaning seem to
turn very largely on the issue of what substitutes for what in de dicto contexts.
Moreover, it will not do for Burge to just say that "contract" and "what
is binding though verbal", although they express the same concept,12 are

Unsurprisingly, precisely the same sort of trouble arises for beliefs about water2

COGNITIVE SCIENCE 109
nevertheless not substitutable in de die to contexts; he will have to give some
account of why they are not. The trouble is, however, that the only conceivable
reason why one should not be able to make such substitutions is
that beliefs about contracts and beliefs about what is binding though verbal
are not identical beliefs.13 But one wants to ask how beliefs about contracts
could be distinct from beliefs about what is binding though verbal if, as Burge
assures us, "contract" and "what is binding though verbal" express the same
concept.
Unsurprisingly, precisely the same sort of trouble arises for beliefs about
water2 and beliefs about XYZ. If you cannot substitute "water2" for "XYZ"
salve veritate in de dicto belief contexts even though, by assumption, "water2"
means XYZ, that must be because beliefs about water2 and beliefs about XYZ
are ipso facto different beliefs. But if we now add the (presumably untendentious)
premise that "water2 is wet" expresses a belief about water2, we get
a contradiction of the proposal we have been investigating: viz., that "water2
is wet" expresses a de dicto belief about the wetness of XYZ. In particular,
we have (1) beliefs about water2 Φ beliefs about XYZ (in order to account
for the unsubstitutability of "water2" for "XYZ" in de dicto belief contexts);
(2) "water2 is wet" expresses a belief about water2 (by assumption); hence
(3) "water2 is wet" does not express a belief about XYZ.
Where we are is: there is a prima facie clash between the principle of
charity and Burge's assumption 3, but the way out is not to jettison the
substitution salve veritate of synonyms for synonyms in de dicto contexts.
An obvious alternative, however, would be to give up the principle of charity
in these cases. So let us look at that.
This is not a decision to be taken lightly; charity should not be confused
with mere politeness. The point is—to switch the discussion back to Putnam's
example—it would be unreasonable for us to take English 2 speakers to be
expressing self-contradictions when they utter things like "water2 is not XYZ".
There might, for example, perfectly well be a point in the development of
their chemistry when water2 is not XYZ is the rational thing to believe given
the evidence. It can, no doubt, be rational to entertain a belief that is
necessarily false; but it is hard to see how one could rationally entertain a
belief with the de dicto content P and ~P. Could the weight of the evidence
favor a contradiction?
It may be felt, however, that this sort of argument is too good. For,
if the principle of charity precludes taking "water2 is wet" to express the
belief that XYZ is wet (on pain of attributing too many inconsistent beliefs
to English2 speakers) does it not also prohibit taking "water2" to mean XYZ
(on pain of attributing too many inconsistent sayings to English2 speakers)?
If "water2" means something like XYZ, then it looks as though the form of
words "water2 is not XYZ" is going to be something like analytically false.
I am perfectly prepared to accept this argument since I am not wedded
to Putnam's intuitions about the meaning of "water2". If, however, you do
not like it, there is a way of avoiding it. You can argue that, given Putnam's
assumptions, it is not obvious that the principle of charity should be applied
to what we say since what we say need not be, in any very direct way, an
expression of what we de dicto believe.

If meaning is not in the head, then we are, in a certain sense,

110 J. A. FODOR
If meaning is not in the head, then we are, in a certain sense, not
responsible for what what we say means. In particular, we are not responsible
for the consistency of what we say in the way that we are, I suppose,
responsible for the consistency of our de dicto beliefs. Suppose that "water2"
means XYZ. Then, presumably, there is some sense in which the form of
words "water2 is not XYZ" is self-contradictory in virtue of the meanings of
its constituent expressions. But it does not follow that " water 2 is not XYZ"
expresses a self-contradictory de dicto belief; no doubt one contradicts something
when one uses that form of words, but one need not be said to contradict
oneself. That would follow only on the assumption that you can infer, Gricewise,
from the meaning of what someone says to the content of the propositional
attitudes he entertains. But, as we have repeatedly had cause to remark,
given Putnam's lexicographic views, that assumption cannot be taken as selfevident.
On the contrary, the principle of charity forbids us to make it in this
case. To put the point in a nutshell, if meaning is not in the head, then talking a
language you know is a lot like talking a language you do not know; in neither
case is there a direct inference from what you utter to what you believe.
I am not, of course, recommending Putnam's lexicographic intuitions;
I am only saying that he has the technical option of holding onto the principle
of charity for beliefs and to "water2 means XYZ" by, in effect, giving up
Grice's principle and refusing to permit direct inferences from what people
say to what they de dicto believe. On the other hand, while this position is
coherent it is surely unattractive since the question what it is that someone
believes when he believes that water2 is wet is still unanswered. And we are
running out of candidates.
Other Options From here on, the argument will go like this. I am going
to accept the intuition that Putnam's account of the meaning of "water2"
is primarily intended to explain: viz., that utterances of "water2 is wet" on
Earth2 have different truth conditions from the homophonic utterances on
Earth. But I am going to claim that this difference in truth conditions has
nothing to do with the meaning of "water2" (or of "water", or of "H2O",
or of "XYZ",. . . etc.). Indeed, I shall claim that it has nothing to do with
any fact of lexicography. Here is how I propose to show this: I shall assume,
contrary to the spirit of Putnam's proposals, that the belief that "water2 is
wet" expresses is something of the order of: the transparent, drinkable . . .
stuff people sail on is wet. (I shall refer to this as the "phenomenological
belief.) And I shall argue that, even on that assumption, you would expect
tokens of "water2 is wet" to have different truth conditions from tokens of
the corresponding English expression. Notice that I am not claiming that
"water2 is wet" does express the phenomenological belief; I propose to
remain totally agnostic on that issue. My argument will be just that the right
explanation of the facts about truth conditions survives that assumption;
hence that, so far as those facts are concerned, we can hold that "water2 is
wet" expresses the phenomenological belief if we're inclined to do so.14
To begin with: If "water2 is wet" expresses the phenomenological
belief, so too, presumably, does "water is wet". So far, then, the present
analysis raises no problems for cognitive science since we no longer have

wetness of H2O? Or, if you do not like,

it that molecularly identical people can differ in de dicto propositional
attitudes. But how do you capture the intuition that "water2 is wet" is true
in virtue of the wetness of XYZ whereas "water is wet" is true in virtue of
wetness of H2O? Or, if you do not like "true in virtue o f talk, we can put
the question this way: If "water is wet" and "water2 is wet" express the same
de dicto belief, how do you account for the intuition that "water is wet iff
water2 is wet" is contingent?
This is, of course, where the indexical analysis did its thing; to say that
"water2" is used to pick out stuff of the same kind as certain (ostensively
specified) stuff "around here" is to guarantee that "water2 is wet" and
"water is wet" are evaluated with respect to local samples of the wet, transparent,
potable stuff that people sail on; hence utterances of the phonological
form /water is wet/ get evaluated with respect to XYZ when they occur on
Earth2 but with respect to H2O when they occur on Earth. We have, however,
given up the indexical analysis on internal grounds and we are heuristically
committed to: "water is wet" and "water2 is wet" both express the phenomenological
belief. Now what?
I propose to resolve the difficulty by distinguishing between the content
of a belief and its truth conditions. By the content of a belief I mean approximately
what we would specify if we were asked to write down its logical form,
with constants for the predicate terms. So, the content of the phenomenological
belief is something like: (x) (x is drinkable, transparent, sailable-on,. . .
etc., only if x is wet).15 I assume that the contents of beliefs and the contents
of sentences are connected by the (Gricean) principle that sentences share the
contents of the beliefs they are used to express.
The important claim is this: you cannot go directly from the content
of a belief/sentence to its truth conditions (to the conditions for its evaluation);
you need at least to specify the universe of discourse for the bound
variables. Moreover, I want to suggest, there is a sort of Principle of Reasonableness
that operates in deciding how the universe of discourse of bound
variables is to be assigned, and the effect of this principle is to determine that
the evaluation of universally quantified standing sentences is relevantly local.
Specifically, it ensures that such sentences are evaluated in much the way they
would be if they contained demonstratives. The difference between the truth
conditions of "water is wet" and of "water2 is wet" are thus the consequence
of the application of the Principle of Reasonableness to the two cases, or so
I am about to claim.
A rough formulation of the Principle of Reasonableness might go: do not
be bloody-minded in deciding what universe of discourse sentences and beliefs
will be evaluated with respect to. I shall refine this principle, slightly, a little
farther on. For the moment, consider an example of its operation in respect
to quantification over times. Marco Polo wrote ([7], p. 62): "Kesmur is a
province distant from Bascia seven days' journey". The first point to notice
is that this looks to be a universal standing sentence with bound variables
ranging over times; something along the lines of: for all pairs (t,tf), if t is the
time of the start of a journey from Kesmur to Bascia and t' is the time of the
end of that journey, then t1 = (f + 7 days). And similarly the other way around
for journeys from Bascia to Kesmur. The second point to notice is that there

The third point to notice,

112 J. A. FODOR
are no (explicit) indexicals, no demonstratives, none of that stuff. The third
point to notice, however, is that tokens of what Marco Polo wrote are not
evaluated with respect to literally all pairs of times. For example, it does not
make what Marco Polo said false that we can now do the trip in an hour and
a half by jet, or in seven seconds by manned satellite. Nor, for that matter,
does it tell against the truth of what Marco Polo wrote that, in the fifth century
B.C. (before they had turbocharged camels or whatever) it might have been
a forty days journey; or, indeed, that the journey might then have been
impossible.
What we do, in our relentless pursuit of unbloody-mindedness, is this:
we evaluate what Marco Polo wrote as though it had contained an indexical;
as though he had written something like "Kesmur is a province distant from
Bascia seven days journey by the means of transport now available". If, in the
course of making the truth conditions of tokens explicit, we now eternalize
this latter sentence by replacing indexicals with names of what they index,
wet get something like " 'Kesmur is a province distant from Bascia seven days'
journey' is true (as tokened by Marco Polo) iff Kesmur is a province distant
from Bascia seven days by the means of transport available circa 1275". Parity
of analysis would, of course, apply to a tokening of the same sentence by,
say, Alan Shepard; so that, for Shepard's token, the truth rule would run
" 'Kesmur is a province distant from Bascia seven days' journey' is true iff
Kesmur is a province distant from Bascia seven days' journey by means of
transport available circa 1980". The very same standing sentence would thus
be true when Marco Polo tokened it and false when Alan Shepard did. "What
we have discovered", you might be inclined to say, "is that 'journey' is
implicitly indexical". But, of course, we have discovered no such thing. All
that has happened is that we have conscientiously avoided bloody-mindedness
in deciding how implicit quantified variables shall be evaluated; in particular,
we have evaluated them with respect to the universe of discourse over which
the speaker may be presumed to have intended them to range. Thus does the
application of the Principle of Reasonableness in determining the universe
of discourse of implicit quantified variables sometimes contrive to simulate
the operation of implicit indexicals, thereby misleading unwary philosophers.
Just as we evaluate implicit quantifiers with respect to the relevantly
local times, so too we evaluate them with respect to the relevantly local
places', and that, I claim, is what makes kind terms seem indexical. Putnam's
indexical story, you will remember, went something like: "water is wet"
is used to express the belief that stuff of the same kind as this (indexing water)
is wet. The present view, by contrast, is that when somebody on Earth believes
that water is wet, he holds a universally quantified belief with approximately
the content: all the potable, transparent sailable-on, . . .etc., kind of stuff is
wet.16 And when somebody on Earth2 believes that water2 is wet, what he
holds is a belief with that very same content. So, according to this story, the
content of the belief that water is wet = the content of the belief that water2
is wet after all. But establishing the identity of the contents of these beliefs
is not yet establishing how their tokenings shall be evaluated (or, equivalently,
what the truth conditions on the tokens are). To do the latter, we apply the
Principle of Reasonableness in the form: evaluate universally quantified beliefs

About Φ-stuff with respect to relevantly local samples

about Φ-stuff with respect to relevantly local samples of Φ-stuff specifically,
evaluate universally quantified beliefs about potable, transparent, sailable-on,
. . . etc., stuff with respect to relevantly local samples of stuff that is potable,
transparent, sailable-on, etc. Since, as it turns out, the stuff that satisfies
that description on Earth is of a different kind from the stuff that satisfies
that description on Earth2, beliefs about potable, transparent, . . . etc., stuff
get evaluated in different ways in the two places. And since, to run it into
the ground, the stuff that satisfies that description on Earth is H2O and the
stuff that satisfies that description on Earth2 is XYZ, it turns out that Earthwise
tokens of the phenomenological belief are true iff H2O is wet while
Earth2-wise tokens of that belief are true iff XYZ is wet. Which is, as the
patient reader may recall, just where we wanted to get to.
I have imposed the principle that universally quantified beliefs about
Φ-stuff should be evaluated with respect to "relevantly local" samples of
Φ-stuff, but I have not said what relevant localness comes to. If I had to make
a stab at it, I would guess that relevant localness is fundamentally an etiological
notion so that what the Principle of Reasonableness is telling us to do, in this
case, is to evaluate beliefs about Φ-stuff with respect to the kind of Φ-stuff
that gave rise to them. Since, it turns out, the stuff that gives rise to the
phenomenological belief on Earth is of a different kind from the stuff that
gives rise to it on Earth2, the phenomenological belief gets evaluated differently
in the two places. This, however, is very tentative, and I should want to keep
the issue of analyzing the notion of relevant localness clear of the issue whether
we are enjoined to evaluate universal beliefs with respect to relevantly local
phenomena. The latter question seems to me a good deal less murky than the
former.17
One more word about being reasonable. It is not only informal discourse
that demands circumspection in the evaluation of quantified variables. Consider
the quantifiers that bind variables in lawlike statements. The fact is that we
evaluate them too with respect to our "local" bits of the universe; roughly,
with respect to those regions of space-time for which isotropy can reasonably
be assumed.18 It would be worse than nit-picking, it would, in fact, be bloodyminded,
to object to the periodic table of elements, or to the germ theory of
disease, . . . etc., on the grounds that, for all we know, they do not hold prior
to the initial bang or on the other side of black holes. Nor is this exercise of
reasonableness merely "optional" (to use a term that Rorty coined for a
related issue). We could not say just how our nomologically bound variables
should be evaluated even if we wanted to, since we do not know with what
generality the laws of even our most basic sciences hold. Scientists are just like
us: they get to use bound variables even though they cannot in the usual case
produce a theory that will pick out the universe of discourse over which the
variables range.
Stopping Having come all this way, a brief retrospective may make
clear the structure of the argument. What there seems to be no way of doing
is to preserve simultaneously:
a. Putnam's intuitions about "water2"
b. the Gricean reduction of meanings to pro positional attitudes

The principle of charity

114 J. A. FODOR
c. the principle of charity
d. the de re/de die to distinction.
Roughly, to preserve the first two we must hold that /water is wet/ expresses
different beliefs on Earth and Earth2. But the different beliefs that commend
themselves are XYZ is wet and H2O is wet, and these beliefs can be ascribed
only de re if the principle of charity is to be respected. To put it slightly
differently, the de re/de dicto distinction seems to be the distinction between
what is in the head and what is not; so we cannot both say (with Burge and
Putnam) that meanings are social and say (with Grice) that meanings are logical
constructs out of de dicto propositional attitudes.
Giving up the Gricean reduction would get us out of this, of course,
but only by blunting Putnam's polemical darts. For, if meanings are not
constructs out of de dicto propositional attitudes, it is perfectly possible
that nonsynonymous formulas (such as "water is wet" and "water2 is wet",
assuming Putnam's lexicography) may function to express the same de dicto
propositional attitudes. But if this is possible, then even if Putnam is right
about what "water" and uwater2" mean, he has provided no argument that
molecularly identical people can differ in what they de dicto believe. If, in
short, he gives up Grice's principle, there would seem to be nothing in Putnam's
lexicography that is relevant to the cognitive science project.
Alternatively, we could imagine giving up the distinction between de re
and de dicto propositional attitudes. But that really would make the sky fall
down. For one thing, believing that water is wet looks to be a different state
of mind from believing that H2O is, and we need the de re/de dicto distinction
to say what the difference amounts to. Moreover, as I remarked above, it
looks as though we are going to need the notion content of a de dicto belief
if we are to have any psychology of the contingency of behavior upon mental
states at all. What would it be like to give that up?
Anyhow, it is just as well that we do not have to. What we can do instead
is account for the fact that /water is wet/ has different truth conditions here
and on Earth2 by appealing to the operation of what are essentially pragmatic
rather than semantic considerations: viz., by appealing to the Principle of
Reasonableness rather than the putative nonsynonymy of "water" and
"water2". This leaves the lexicographic issues wide open; and if it turns out
that such issues have no principled resolutions, maybe that is all right too.
Here are some questions that people have asked me about this paper:
1. "Aren't you saying that 'water' isn't a rigid designator for H2O?"
Answer: no, I am not saying that. What I do claim is that, if the evidence
that "water" rigidly designates H2O is the difference in truth conditions
between "water is wet" and "water2 is wet", then we have no evidence
for the rigid designation claim. For, as we have seen, that difference in truth
conditions is compatible with "water" and "water2" both expressing the
phenomenological concept.
On the other hand, the fact that "water is not H2O" is not of the de dicto
form P and ~P is consistent with "water" being a rigid designator for H2O.
What would make them inconsistent is the principle that if two terms rigidly
designate the same thing then they have the same meaning. But we have

COGNITIVE SCIENCE

COGNITIVE SCIENCE 115
independent reason for doubting that this is so; cf. "equilateral triangle" and
"equiangular triangle", both of which designate equilateral triangles in all
possible worlds.
A fortiori, I have no argument (and no grudge) against the claim that
"water is H2O" is metaphysically necessary, since I take it that the claim is
to be defended, if at all, on grounds independent of lexicography or semantics.
2. "Yes, but: is meaning in the head?" Answer: yes and no. What
determines behavior (things like what I have called belief contents) is in the
head, but you cannot get truth conditions directly out of what determines
behavior. You need pragmatic principles to tell you such things as how to
evaluate bound variables. On the other hand:
a. Barring explicit indexicals,19 none of what is outside the head but
relevant to the determination of truth conditions has been shown
to be specific to particular lexical items; so it is unclear that Earth2
examples have any implications for lexicography, or for what you
have to learn to learn a word, or for what you have to know to know
a word, . . . etc.
b. Though what is in the head does not determine extension by itself, it
does determine extension in the context of appropriate, general
pragmatic principles. So, contrary to what Putnam often suggests,
there is no reason why the usual semantics of truth and reference
should not apply when we map from belief contents onto the world;
and, if there ever was any reason to trust evidence about extensions
in establishing inferences about intensions, these reasons survive
Putnam's examples. Only, when you make such inferences, keep the
pragmatic effects in mind.
c. We have not had to tell a story about "linguistic division of labor".
No doubt communication is a cooperative enterprise and if we insist
upon being unreasonable with one another, the thing is not going to
work. But nothing about the way that "water" and "water2" operate
suggests that their meanings have somehow been delivered into the
hands of the experts. From what I have seen of how experts use
language, I would be disinclined to trust them with it.
d. Individualism is all right if the notion of de dicto beliefs is all right;
and the notion of de dicto belief is all right if and only if we can
indeed explain the behavior of organisms by reference to the contents
of their propositional attitudes. That has been clear for a long time
and despite the present excitements, nothing much would seem to
have changed around here.
You can come back, Chicken Little; I expect that everything is going
to be all right.

J. A. FODOR

116 J. A. FODOR
2. Though several recent papers have sought to make the putative morals for cognitive
science explicit;see [12], [13], [8], and [1].
3. Of course they are not really, since, as Geach [4] points out, there are molecules of
H2O in us but not in them. I take it that this detail will be seen not to prejudice the
spirit of Putnam's example.
4. From here on, "cognitive science" usually denotes not a problem domain but a body
of doctrine; specifically, a theory whose main tenets are RTM together with the claim
that mental processes are computational. The locution "cognitive science" should
thus be construed on the model of, say, "Christian Science" rather than of, say,
"cognitive psychology".
5. It is unclear to what extent earlier versions of RTM accepted this doctrine implicitly.
There was, for example, a traditional disagreement over whether association "by
similarity" could be a fundamental psychological mechanism, along with association
by spatio-temporal contiguity. It is possible to view this debate as in part about
whether mental operations are computational. Since the sort of similarity at issue
was similarity of content, to acknowledge association by similarity as irreducible
would have been to violate the condition that mental operations apply in virtue of
the form of representations in their domains. Contiguity, by contrast, counts as a
formal property within the meaning of the act.
6. I shall often use "de dicto propositional attitude" as short for "propositional attitude
under de dicto specification". No propositional attitude can be de dicto per se, or
de re stricto dictu. Of course.
7. In a recent paper [8], however, Putnam opts for a "verificationist" notion of concept
such that "water" and "water2", though nonsynonymous, are nevertheless used to
express the same concept so long as speakers are ignorant of the relevant microchemical
facts. The effect is thus to detach the question "what does the word W mean?" from the
question "what concept is the word W used to express?" and, more generally, to detach
semantic questions from questions about the propositional attitudes of speaker/hearers.
I think that there is something right about this since, as will appear, I think that the
content of a belief does not determine its truth conditions (or those of the forms of
words used to express it). My reasons for thinking this are, however, a good deal less
dramatic than Putnam's. Though it will turn out that beliefs of identical content may
differ in their truth conditions, the considerations that lead to this conclusion are
philosophically innocuous. They do not, in particular, lend any comfort to verificationist
s.
8. I propose to use such expressions as "Grice's principle", "Gricean theory", "Gricean
reduction", and so forth to refer to any of a wide variety of doctrines which have in
common the idea that meanings are somehow constructs out of propositional attitudes.
I do not mean to assume the particular account of that reduction which Grice sets out
in [6]; nor, of course, is Grice to be blamed for the use I have made of his leading
doctrine.
9. This is pointed out by Burge in [1], although the morals he draws are quite different
from the ones that I shall endorse.
10. Professor Michael Lipton has suggested to me the following snazzy argument, which
makes trouble for all forms of the indexical proposal. What we are trying to do is pick

COGNITIVE SCIENCE

out the de dicto belief that "water2 is wet" is used to express. But you cannot do this
with a formula of the form "the belief that this stuff i s . . . " because, since indexicals
always occur transparently in descriptions of propositional attitudes, no such formula
can, even in principle, specify a belief de dicto. What I suspect that this argument
shows is that there can be no word that is defined in terms of an indexical (which is
not, of course, to say that there can be no indexical words).
11. Notice that 2 is not a version of Grice's principle, for although it connects semantic
properties with properties of concepts, the latter are not being assumed by Burge
to be mental in anything like the sense that Gricean reductions require.
12. As the reader will have divined, I am pretending that verbal contracts bind exhausts
the concept of contract instead of just constituting part of it as per assumption 1.
It simplifies the discussion and changes nothing to do so.
13. More precisely, I take the following principle to be valid: Let a and b be distinct
expressions, and let believes that. .. a . . . and believes that... b . . . be formulas
that specify beliefs de dicto. Then, if a and b are synonyms (express the same concept)
either D [(x) (x believes t h a t . . . a ...) = (x believes t h a t . . . b . . .)] or believes that...
a . . . and believes that. .. b . . . designate distinct beliefs.
14. I have no story at all to tell about Burge's examples insofar as they do not involve kind
terms, since I find that I do not share many of the intuitions that motivate Burge's
solutions. A natural thing to do in the case of "contract" would be to take it to express
some such concept as legally binding agreement, so that the belief that contracts must
be written would be consistent but false. For what it is worth, The American College
Dictionary says that a contract is "an agreement enforceable by law", thereby leaving
it as a legal (rather than a conceptual) issue whether verbal contracts bind. This seems to
be entirely plausible, for it seems to me not incoherent to wonder whether, for
example, verbal contracts are binding in France.
15. For convenience, I am assuming that "water is wet" expresses an (implicit) universal
generalization. But the analysis I shall propose has an obvious extension to the
assumption that "water" is a singular term. Indeed, the principles involved in the
analysis appear to be quite general in their application. See note 17 below.
16. The "the" is there to indicate that the belief that water is wet purports to be about a
single kind of stuff, and the "kind" is there to indicate that the belief that water
is wet purports to be about a kind. In these respects the present analysis shares
Putnam's assumption that "water" is (or, anyhow, purports to be) a kind term.
17. Analogous considerations apply to ensure that the evaluation of existentially quantified
variables should also be local, a point that has been widely noticed. For example,
nobody would evaluate "there are cookies to eat" with respect to cookies in China.
Similarly with respect to singular terms, whose referential ambiguity would otherwise
make life miserable. Contemporary tokens of "John is in the shower" are not, of
course, evaluated with reference to John the Baptist.
18. Isotropy does not, of course, supply an independent characterization of the relevant
universe of discourse since isotropic regions of space-time just are those that are homogeneous
with respect to the sorts of descriptions that laws deploy.
19. This is not a fudge since, on anybody's story, explicit indexicals can presumably be
distinguished from words like "water" on grounds independent of their indexicality.
For example, the explicit indexicals belong to "closed class" vocabulary.