Meaning and Truth-Conditions

Richard G Heck Jr
Department of Philosophy, Brown University

1 Opening example, discovering what sorts of semantic values need to be assigned to

novel semantic primitives, such as mass terms, and explaining the semantic
Donald Davidson (1984g) famously proposed that what we want from a significance of novel modes of composition, such as [V PV [CP S]].1
theory of meaning we may get from a theory of truth. What we want from It is unfortunate that a proposal with such great advantages faces such
a theory of meaning for a natural language, that being the case at issue, grave difficulties. Indeed, it is probably fair to say that most philosophers
includes at least an account of the phenomena that give rise to the principle of language, and perhaps even most working semanticists, regard truth-
of compositionality: We want an explanation of how the meaning of a theoretic semantics as a non-starter, for the simple reason that it seems to
complex expression, such as a sentence, depends upon and is determined by have the wrong subject-matter. We will not get what we want from a theory
the meanings of its syntactic parts. One central advantage of Davidson’s of meaning from any theory that does not specify meanings: No theory that
proposal, as he saw it, is that we know from the work of Tarski (1958) how does not specify what the sentence ‘snow is white’ means, for example, can
to construct theories of truth for languages with certain sorts of structures, possibly explain how this sentence’s meaning what it does is determined by
and such a theory does help us to understand how at least some of the its parts’ meaning what they do. But a theory of truth for English does not
semantic properties of complex expressions are determined by those of their specify what even one sentence of English means, since any such theory is
parts. Davidson’s practice suggests he may have hoped that all sentences consistent with infinitely many different suppositions concerning what any
of natural language could be assigned logical forms of the variety familiar sentence of English means. For example, a theory of truth that proves this
from mathematical logic. If so, then semantics would all but reduce to T-sentence2 for ‘snow is white’:
syntax: The only outstanding problem would be to uncover the ‘logical 1 I do not mean to suggest that these are the only issues worth thinking about. And in
forms’ of the sentences of, say, English. This hope does not now seem actual research, semantics and syntax interact with one another, and work on semantic issues
reasonable. But the methods available for investigating logical form—that can inform work in syntax. It is, in fact, a nice question whether syntax is ‘autonomous’ in
is, for investigating syntax—have become so powerful that it is not at all the sense my remarks somewhat sloppily suggest it is.
2 By a ‘T-sentence’ here I mean any sentence of the form: S is true iff p. In particular,
unreasonable to suppose that the syntactic part of the problem is solvable, at
no relation between the sentence mentioned on the left-hand side and that used on the
least in principle, and Tarski’s work can be adapted to many of the resulting right-hand side is required for a sentence of that form to be a T-sentence. By a ‘correct’
structures. The work that is left for the semanticist proper then becomes T-sentence, I mean one that is true: No stronger relation is required for a T-sentence to be
to solve the problems syntax and Tarski do not solve: These include, for correct.

1
1 Opening 2

(1) ‘snow is white’ is true iff snow is white One way of motivating the restriction to canonical proof is psychological:
We think of a competent speaker as tacitly knowing a theory of truth for
is consistent with that sentence’s meaning anything, so long as it means her language, and the conception of canonical proof is then conceived as
something true. A monolingual speaker of German who believed that ‘snow an empirical hypothesis concerning how her ‘language faculty’ goes about
is white’ was true iff snow was white could rationally believe as well that churning out T-sentences. Richard Larson and Gabriel Segal (1995, §2.2.1)
‘snow is white’ meant that grass was green: If ‘snow is white’ did mean present a form of this proposal that modifies the underlying notion of a
that grass is green, then, since snow is white iff grass is green, it would ‘theory’, restricting the sorts of inferences that are permitted so that only
still be true iff snow was white.3 Since this sort of objection was first canonical proofs are possible. I am sympathetic to that idea,4 but there
forcefully raised by John Foster (1976), it has come to be known as the seems to me also to be a way of motivating the restriction without appealing
‘Foster problem’. to psychology. The meaning of a sentence is supposed not just to depend
It is important not to confuse this problem with a closely related one, upon but to be wholly determined by the significance of its constituents
which I shall call the ‘many T-sentences problem’. One often finds people and how they are combined. Canonical derivations, as I have described
speaking of ‘the T-sentence’ that a given theory of truth proves for a given them, are ones that appeal only to information about the significance of
object-language sentence. Obviously, however, there isn’t any such thing as the constituents and the modes of construction. A proof of (2) will appeal
the T-sentence any such theory proves. Quite generally, if a theory proves to that information, but not only to it, and so will of course include more
p ≡ q and also proves r, then it proves p ≡ q ∧ r. So any theory that proves information than what ‘snow is white’ means.
(1) also proves: In any event, the many T-sentences problem is not the one I want to
discuss here. My goal here is to defend a form of Davidson’s proposal
(2) ‘snow is white’ is true iff snow is white and r
from the Foster problem. What most fundamentally distinguishes my view
for any statement r that it also proves. This will include all logical truths from Davidson’s is that I accept a ‘cognitive conception of understanding’:
and much else besides, though exactly what it includes will depend upon According to me, our actual ability to interpret the utterances of other
the details of the theory of truth. speakers depends upon our consciously5 knowing what their utterances,
The standard response to this problem is, in effect, to define what ‘the as made on that occasion, mean. I shall not argue for this conception of
T-sentence the theory proves’ is supposed to mean, and the now standard understanding here,6 but I shall not simply assume it, either. What I am
way to do that is to invoke a conception of ‘canonical proof’: When we going to argue is that no solution to the Foster problem is available unless
speak of ‘the’ T-sentence a theory proves for some given sentence S, we we adopt the cognitive conception and so that truth-theoretic semantics is
mean the T-sentence it proves in a certain special way, namely, by taking committed to it. In particular, I shall argue in section 2 that Davidson’s own
S apart, piece by piece, then applying the axioms that characterize the proposed solution rests, and rests essentially, upon his metaphysics of the
meanings of S’s ultimate constituents, and then finally putting these together mind, and I, at least, find that metaphysics unpalatable. In section 3, I shall
in accordance with the principles that characterize the significance of the consider an attempt to resolve the Foster problem based upon a suggestion
various constructions that appear in S. It has sometimes been objected, 4 See, in particular, the discussion in my “What Is Compositionality?” (Heck, 2004).
however, that, while this may do the trick, formally speaking, the proposal 5 Let me emphasize that element of the view. The claim is not that speakers must tacitly
is ad hoc and unmotivated (Soames, 1992, p. 34, n. 11). know facts about what their words mean, but that they must consciously know facts about
what uttered sentences mean.
3 This way of developing the point is due to Scott Soames (1992). 6 See my paper “Reason and Language” (Heck, 2005a) for the argument.
2 Radical Interpretation 3

made in a related but somewhat different context by John McDowell. I shall for example, one need only replace it with
argue that it too fails unless we embed it within the cognitive conception,
(6) ‘is white’ is true of x iff x is white and grass is green,
towards which it therefore points, once properly understood. In section 4,
I shall explain and further motivate the cognitive conception and then, in which is true if (5) is (Foster, 1976, p. 13).
section 5, drawing on work of James Higginbotham and Ian Rumfitt, argue In later work, Davidson insists that T-sentences that purport to be inter-
that the Foster problem can be solved from within the theoretical framework pretive “are empirical generalizations about speakers, and so must not only
thereby established. Section 6 addresses an objection due to Scott Soames, be true but also lawlike” (1984g, p. 26, fn. 11).8 This suggestion, too,
namely, that the view I am defending implausibly ascribes a concept of truth however, is insufficient. Soames (1988, pp. 187-8) observed that, if we
to all competent speakers. The final section mentions two issues that still replace (5) with
need addressing.
(7) ‘is white’ is true of x iff x is white and arithmetic is incomplete,

2 Radical Interpretation as a Solution to the Foster then the resulting theory delivers
Problem (8) ‘snow is white” is true iff snow is white and arithmetic is incomplete,

Davidson addresses versions of the Foster problem in several places. His which is an empirical law if (1) is. As Soames notes, the same example
first suggestion was that a T-sentence cannot, in the relevant sense, ‘give disposes with Foster’s own suggestion that we need only strengthen the
the meaning’ of the sentence it mentions on its left-hand side unless it is a biconditional so that it requires the two sides to be necessarily equivalent:
consequence of a compositional truth-theory for the language to which the Clearly, (8) will then be true if (1) is.9
sentence belongs (Davidson, 1984g, pp. 25-6). That plausibly accounts for Davidson’s solution to the Foster problem must therefore be taken to lie
such cases as7 elsewhere, namely, where it is usually taken to lie, in his discussion of radi-
cal interpretation. Davidson claims, at the end of “Radical Interpretation”,
(3) ‘snow is white’ is true iff grass is green that
since, as Davidson notes (1984g, p. 26, fn. 10), it is hard to see how this A T-sentence of an empirical theory of truth can be used to inter-
sentence could be a (canonical) consequence of a correct compositional pret a sentence. . . provided we also know the theory that entails
truth-theory. By itself, however, this amendment is insufficient. As Foster it, and know that it is a theory that meets [certain] formal and
observed, one can easily to reformulate any compositional theory of truth empirical criteria. (Davidson, 1984c, p. 139)
that yields the desirable (1) so that it instead yields:
The main formal criterion is just that the theory of truth in question should
(4) ‘snow is white’ is true iff snow is white and grass is green. be compositional. The empirical criterion is that the theory should be
verifiable by a radical interpreter.
If the theory makes use of the axiom
8 The footnote from which this quotation is taken was added in 1982, the paper itself

(5) ‘is white’ is true of x iff x is white, having been published in 1967. See also (Davidson, 1984d, p. 174).
9 I myself think that there is an even stronger objection to be made to Foster’s proposal,
7Obviously, truth here needs to be relativized to a language, but I shall omit that variable namely, that (1) itself is not true if “iff” is read as requiring necessary equivalence. But some
to avoid cluttering the exposition. philosophers think it is, if the relativization to a language is made explicit.
2 Radical Interpretation 4

There are different ways we might understand this proposal. In the constitute a workable alternative to Tarski’s appeal to translation—then
passage just quoted, Davidson says that one can use a theory of truth to “each T-sentence will in fact yield an acceptable interpretation” (Davidson,
interpret a language if one has certain sorts of knowledge about that theory. 1984c, p. 139). But Davidson does not actually given any argument that the
This proposal is oddly meta-linguistic, however, and that fact has troubled ‘holistic constraint’ is adequate to its intended purpose.
many commentators.10 There is further discussion of the matter in “Belief and the Basis of
But there is a better way to understand Davidson’s suggestion, so that Meaning”, where Davidson draws well-known analogies with problems in
what is required is not that one should know something about a theory of decision theory. But there, too, the really pressing issue, the status of (RI),
truth but, rather, that one’s knowledge of the theory should have a certain is only taken up in the last couple pages (Davidson, 1984a, pp. 152–4). And,
provenance. On this interpretation, Davidson is committing himself to the so far as I can tell, Davidson says essentially nothing in defense of the view
following thesis:11 that, if a theory of truth optimally fits the evidence available to a radical
interpreter, then it can be used to interpret. He does not,in particular, give
(RI) Someone who knew a truth-theory for a language L, and who knew any argument that might even seem to show that a theory of truth issuing in
that theory on the basis of the sort of evidence that would be available such T-sentences as (8) cannot meet the ‘holistic constraint’.12
to a radical interpreter, could use T-sentences (canonically) derived My old colleague Burton Dreben is reputed to have said that what a
from that theory to interpret speakers of L. philosopher doesn’t argue for is more interesting than what s’he does argue
for. And in this case, at least, he’s right. I think it just seemed obvious
So far as I know, however, Davidson offers no serious argument for (RI)
to Davidson that any theory of truth that was confirmable by a radical
in “Radical Interpretation” or, for that matter, anywhere else. Davidson
interpreter would have to be usable for interpretation, and the interesting
is of course aware of the need to say something in defense of (RI). But
question is why this seemed so obvious to him. To answer this question, we
the discussion of this issue in “Radical Interpretation” occupies just the
need to examine more closely the thought experiment in which Davidson is
last page and a half Davidson (1984c, pp. 138–9), and Davdison does little
asking us to engage.
more there than to assert that (RI) is true: “The present idea is that what
Tarski assumed outright for each T-sentence”—namely, that the sentence Radical interpretation is modeled on Quine’s notion of radical translation
used on its right-hand side translates the sentence mentioned on its left-hand (Quine, 1960, ch. 2).13 We are to imagine someone traveling abroad to a
side—“can be indirectly elicited by a holistic constraint”, namely, “that country whose language is utterly alien to her and attempting to construct
the totality of T-sentences should. . . optimally fit evidence” available to and confirm a theory of truth for that language on the basis of her obser-
a radical interpreter. If so—that is, if the holistic constraint really does vations of native speakers. But why should this thought experiment be
of any interest to a philosopher?14 Davidson is, of course, free to engage
10 Foster (1976, §3) was perhaps the first to be troubled by it. Soames (1992, pp. 25–28)
12 And again, in “Reply to Foster”, Davidson writes: “By a course of reasoning, I have
raises several problems of detail for how it is supposed to work.
11 I think that (RI) is the most charitable way to read Davidson and that it sidesteps a lot tried to show that if the [formal and empirical] constraints are met by a theory, then the
of the issues about formulation that bother Foster (1976, §3). Davidson himself remarks, T-sentences that flow from them will in fact” be interpretive Davidson (1984d, pp. 172–3).
in “Reply to Foster”, that he had “imagined the theory [of truth] to be known by someone But I must confess to having no idea what “course of reasoning” Davidson has in mind here.
13 I sometimes think Davidson thought this was all that needed saying here. We’ve all
who had constructed it from the evidence. . . ” and that this is the key to his answer to the
Foster problem (Davidson, 1984d, p. 173). But Davidson himself does not formulate (RI). read our Quine, right?
Nonetheless, I shall not pause to consider the question whether Davidson really does accept 14 The importance of this question was made clear to me by Michael Dummett’s (1978b)

(RI). I shall simply assume that he does. insistence upon a similar question regarding radical translation, and my remarks here are
2 Radical Interpretation 5

in whatever speculation he wants, and if he wants to know what sorts of speaks English, I would appeal to a great deal of evidence that Davidson
theories of truth could be confirmed by a radical interpreter, he is free to would deny a radical interpreter. One might respond that this evidence could
ask that question. But plainly, Davidson thinks the answer to this question itself be called into question and that, ultimately, my evidence is only what
matters: He thinks one can draw important conclusions from it. Davidson would be available to a radical interpreter. But the occurrence of the word
seems to think, for example, that one could establish a thesis with the same ‘ultimately’ here is a sure sign of a kind of foundationalism that has gone
import as Quine’s thesis of the indeterminacy of translation by showing that badly out of style and is anyway not the sort of epistemology one typically
the evidence that is available to a radical interpreter will not, in general, associates with Davidson.15
support a unique theory of truth for a given language (Davidson, 1984c, More importantly, the epistemological construal of the thought experi-
p. 139). But that makes it all the more important to ask why this thought ment does not fit the use Davidson makes of it. As noted, Davidson agrees
experiment should be supposed to illuminate questions about meaning. with Quine that meaning is, in a certain sense, indeterminate, and his reason
One answer might begin with Davidson’s claim that “[a]ll understanding is that the evidence that is available to a radical interpreter does not suffice
of the speech of others involves radical interpretation” (Davidson, 1984c, to restrict the interpretive theories of truth to one. No such conclusion can
p. 125). Surely, however, we do not actually confront other speakers from possibly be forthcoming on the epistemological interpretation. All we could
a standpoint of total ignorance about what their words mean and what conclude would be that the evidence available to a radical interpreter is
their beliefs are: There is a clear sense, then, in which we are not all inadequate to distinguish among the competing theories. In so far as the
radical interpreters. Then again, in the very first paragraph of “Radical radical interpreter modeled the epistemic situation of ordinary speakers,
Interpretation”, Davidson explicitly disclaims any interest in the psychology one could then conclude that no ordinary speaker could justify one such
of language-use, so he cannot have intended radical interpretation to model interpretation over another. But it simply would not follow that there was
the psychology of ordinary communication. Perhaps, then, the clue to no fact of the matter which interpretation was correct.
Davidson’s intentions is the sentence immediately preceding the one just It seems to me, therefore, that the questions that are bothering Davidson
quoted: “Speakers of the same language can go on the assumption that for are not epistemological but metaphysical. They are the questions Sir Peter
them the same expressions are to be interpreted in the same way, but this Strawson asks at the beginning of his Inaugural Lecture (Strawson, 1971):
does not indicate what justifies the assumption” (Davidson, 1984c, p. 125, In virtue of what are sentences meaningful? In virtue of what does a given
my emphasis). Maybe what radical interpretation is supposed to model, sentence have the very meaning it has? Davidson argues in “Theories
then, is not the psychology of ordinary communication but its epistemology. of Meaning and Learnable Languages” (Davidson, 1984e) that an answer
Although there is, as I’ve noted, some evidence for this reading, it is to these questions must take due notice of the fact that the meaning of
not, in the end, satisfying. There are two main difficulties. First, the a sentence depends upon the meanings of its parts; he argues in “Truth
epistemology to which Davidson would be committing himself is deeply
problematic. It is all well and good to raise the question why we are entitled 15 Tyler Burge (1999) makes some remarks along similar lines, though his are much

to assume, as we typically do, that what sounds like the language we know better developed. My remarks here also have important points of contact with some ideas
expressed by McDowell in his paper “Anti-realism and the Epistemology of Understanding”
really is the language we know, but it simply does not follow that the
(McDowell, 1998a). McDowell would not apply these points to Davidson, however, as he
epistemic situation of the ordinary speaker is comparable to that of a radical seems to read Davidson somewhat differently from how I do. McDowell’s target is Dummett.
interpreter. If challenged to defend the assumption that, say, Ned Block For what it’s worth, however, my own view is that Dummett is not really an appropriate
target after 1978—after, that is, the emergence of the views presented in “What Do I Know
inspired by Dummett’s. When I Know a Language?” (Dummett, 1993d) and the other papers of that period.
2 Radical Interpretation 6

and Meaning” (Davidson, 1984g) that only a theory of truth can give a he characterizes it.16 Why should the conditions under which a sentence
substantive account of this dependence. For that reason, he thinks, a theory is ‘held true’ by speakers of the language be included among the facts
of meaning must take the form of a theory of truth. If so, then we may be about how it is used? Why should semantic facts, such as facts about what
tempted to reformulate the initial question what it is for a sentence to mean expressions of the language are used to say, be excluded? Why should
what it does as: In virtue of what is a theory of truth correct or incorrect? “detailed descriptions of the speaker’s beliefs and intentions” (Davidson,
As Foster was the first to emphasize, however, if our interest is in questions 1984c, p. 134) be excluded?17
about meaning, that is not quite right: A theory of truth can be correct (in Let me say as clearly as possible what question is at issue here: Why
the sense that all its theorems are true) yet not throw much light at all on should we think that, if two sentences have different meanings, this differ-
questions about meaning. So the question that needs asking, as Davidson ence must manifest itself in some difference in the conditions under which
sees things, is rather: In virtue of what is a theory of truth interpretive? those sentences are held true by speakers?18 Why, in particular, should we
Davidson’s answer to this question, contained in the notion of radical
16 The dialectic here is thus essentially the same as that in Chomsky’s paper “Quine’s
interpretation, invokes a form of what I have elsewhere called the Use-
Empirical Assumptions” (Chomsky, 1969). Absent some reasonably precise characterization
Meaning Thesis (Heck, 2005b): The meaning of a sentence is determined of the ‘behavioral dispositions’ on which facts about language are supposed to supervene,
by how it is used. Of course, until we are told what ‘meaning’ and ‘use’ are, Quine’s position has no empirical content. But the precise characterization Quine gives,
the Thesis can be no more than programmatic. It is clear enough, however, in terms of the responses the speaker would give in “a questionnaire circumstance, the
how Davidson understands these notions: Meaning is to be understood circumstance of being offered a sentence for assent or dissent or indecision or bizarreness
reaction” (Quine, 1972, p. 445), simply raises the question why we should think that facts
in terms of truth-theories; use is to be understood in terms of Davidson’s about language actually do supervene on such a narrow base. This is what Chomsky has
notion of ‘holding true’ (or, more generally, in terms of the evidence that in mind when he remarks that “. . . we can, of course, raise the conditional probability of
would be available to a radical interpreter). So, in Davidson’s hands, the any sentence as high as we like. . . relative to ‘situations’ specified on ad hoc, invented
Use-Meaning Thesis takes the following form: Whether a theory of truth for grounds. . . ” (Chomsky, 1969, p. 57). Quine is thus quite wrong to compain that his reply to
Chomsky’s criticism of his use of ‘behavioral dispositions’ has been ignored (Quine, 1972,
a given language is interpretive is determined by the circumstances under p. 445). It had been anticipated, and dismissed as unmotivated.
which sentences of that language are held true by its speakers. 17 I have argued elsewhere (Heck, 2005b) that the answer to the second question—why

semantic facts should be excluded—is just that their inclusion would trivialize the problem.
Why does it seem obvious to Davidson that use (in his sense) determines
With respect to the first, David Lewis (1985c) quarreled with Davidson’s inclusion of
meaning (in his sense)? Davidson is moved, I think, by much the same sort holding true. For all its differences, however, Lewis’s treatment of radical interpretation is
of thought that has attracted so many other philosophers to the Use-Meaning still vulnerable to the objections I am bringing against Davidson. That is because Lewis’s
Thesis: If there is no difference in how two expressions are used, then there view depends as heavily upon his metaphysics of mind—his analytical functionalism—as
can be no difference in what they mean; nothing that does not somehow Davidson’s does upon his.
18 This is the same question, by the way, that Dummett presses against Davidson, though
surface in the use of an expression can be part of its meaning. One might from a different persepective. The so-called ‘manifestation constraint’ simply is the principle
wish to quarrel with that idea, but one might well just concede it until more we are discussing. And Dummett’s argument for anti-realism is then that, if we require all
has been said about what use and meaning are supposed to be, since it differences in meaning to be manifested by a difference in what is held true under which
is otherwise without content: As McDowell (1998a) notes, if it is a fact conditions, then there will be no way to defend the attribution of classical truth-conditions
to, e.g., mathematical language. This point is especially worth remembering when reading
about how ‘snow is white’ is used that it is used to say that snow is white, Dummett’s writings from this period, especially the two papers titled “What Is a Theory
then we can all quickly concede the Use-Meaning Thesis. The interesting of Meaning?” (Dummett, 1993e,f). I do not think these papers can properly be understood
question is why Davidson thinks use can, and should, be characterized as except as part of an on-going conversation with Davidson. In particular, the reason one
2 Radical Interpretation 7

think that there could not be differences in meaning that make no difference interpretation. (Davidson, 1984c, p. 134)
to which sentences are held true when but that do make a difference to “the
complex and delicately discriminated intentions with which the sentence[s The difficulty is that, although this remark seems fair enough if it is intended
are] typically uttered” (Davidson, 1984c, p. 127)? The question is particu- as one about the epistemic predicament of a radical interpreter, the question
larly pressing in light of the variant of the Foster problem due to Soames. why we should care about the epistemic predicament of a radical interpreter
How can Davidson’s radical interpreter can distinguish the case in which is still unanswered. If the radical interpreter were intended to model the
‘Snow is white’ means that snow is white from the case in which it means epistemological predicament of an ordinary speaker, then that would provide
instead that snow is white and arithmetic is incomplete? The sentence could a reason: Perhaps ordinary judgements about meaning would be vulnerable
of course mean either thing. But since it is a necessary truth that arithmetic to skeptical attack if meaning were not determined by the evidence available
is incomplete, it is not clear how such a difference in meaning could make to a radical interpreter. Sir Michael Dummett (1993a, p. 177) is clearly
any difference to the conditions under which ‘Snow is white’ is held true. moved by just such a worry, but I have already registered my doubts about
The point, of course, is that Davidson understands ‘the conditions under the epistemology that lies behind it.
which a sentence is held true’ to be external conditions. It is clear enough The correct interpretation, once again, is to take Davidson to be making
what kind of difference such a difference in meaning might make to the not an epistemological claim but a metaphysical one. On this interpretation,
internal conditions under which the sentence was held true. As things are, Davidson is denying the radical interpreter prior knowledge of what her
many speakers who know nothing of logic hold ‘Snow is white’ true, and if subject believes and intends not because one can only know what someone
that sentence meant that snow is white and arithmetic is incomplete, one believes and intends when one is able to interpret her words, but because
would not expect that to be the case. Such differences of meaning are the facts about what someone believes and intends are not independent of
thus easily enough distinguished if we are allowed to characterize use in facts about what her words mean. The radical interpreter’s epistemological
terms of the mental states of speakers. But Davidson explicitly denies such predicament, on this interpretation, is thus not a model of ordinary speak-
information to the radical interpreter.19 ers’ epistemological situation but rather a reflection of the metaphysics of
Davidson’s reply would surely be not so much to concede these points as intentionality.
to insist upon them. On his view, the radical interpreter is not just trying Some might suggest that we ought simply to have begun with the interpre-
to interpret her subjects’ words but also to interpret her subjects’ minds. tation at which we have now arrived, and I certainly am not claiming to have
Davidson denies the radical interpreter access to prior information about discovered some hidden commitment of Davidson’s view. As is well-known,
her subjects’ minds precisely because he thinks that there is only the one, the metaphysical view this interpretation ascribes to Davidson—that mean-
unified interpretive project: ing, belief, and intention are constitutively intertwined—is one to which he
explicitly and enthusiastically subscribed (Davidson, 1984f). What I am
The evidence [available to a radical interpreter] cannot consist doing here is simply emphasizing that Davidson’s metaphysics of meaning—
in detailed descriptions of the speaker’s beliefs and intentions, his view that a whether a theory of truth is interpretive is determined by the
since attributions of attitudes, at least where subtlety is required, circumstances under which sentences are held true—necessarily depends
demand a theory that must rest on much the same evidence as upon his metaphysics of intentionality. It is not just that Davidson happens
searches in vain for any real defense of the manifestation constraint is that Dummett feels
to hold both views.
no need to argue for it: Davidson accepts it wholeheartedly. The question we have been discussing is whether every difference of
19 Similar points are made by James Higginbotham (1991, pp. 277-8). meaning must manifest itself in a difference in the circumstances under
3 Frege on Meaning 8

which sentences are held true. I have argued that, in so far as Davidson interpreter remained the same. The question for Davidson, then, is what
offers us reasons to believe that this is so, those reasons necessarily include reason he can give us to suppose that no such possibility can be realized. His
the claim that meaning, belief, and intention are constitutively intertwined. answer would be to insist that radical interpretation is simultaneously both
But if we conjoin this thesis with the view it is being used to establish, then of a speaker’s words and of her mind. And Davidson’s point, once again,
we thereby commit ourselves to a metaphysics of the mind that seems to me, would not just be an epistemological one: It is not just that it is hard to tell
at least, to be extremely unattractive. Radical interpretation, in Davidson’s when one is in such a situation. Davidson’s claim, rather, is that mentality
sense, is possible only if meaning supervenes on the sorts of facts that are supervenes on the kind of evidence available to a radical interpreter—on
available to the radical interpreter. But if meaning, belief, and intention are what is held true when—just as meaning does, so that there could not be
constitutively intertwined, then belief and intention must also supervene on such a situation: Mental states cannot vary while the facts about what is
the facts available to a radical interpreter: That is to say, what mental states held true when remain the same.21
one has, and with what contents, must be determined by (i) the conditions Davidson, then, may well have had a solution to the Foster problem that
under which one would hold various sentences true and (ii) the facts about worked for him: If mentality supervenes on the kind of evidence that is
which sentences one actually does hold true. This view is but half a step available to a radical interpreter, as he thought it did, then, plausibly enough,
beyond behaviorism, and, in my opinion, it deserves the same fate. But a theory of truth known to be true on the basis of such evidence can always
it is undoubtedly a view Davidson held (1984f, esp. p. 162): It is just an be used to interpret. But the price of using radical interpretation to solve
‘interpretive’ view of the mind—what one’s mental states are is a matter of the Foster problem is a commitment to Davidson’s metaphysics of mind:
how one would best be interpreted—coupled with a particular view about A theory of truth known to be correct on the basis of the kind of evidence
what the basis of interpretation is. that is available to a radical interpreter can be guaranteed to be interpretive
The argument here can perhaps be made a little clearer if we state it only if mentality supervenes on what is held true when. The mental doesn’t
in the form of a challenge to Davidson. Davidson is claiming that what supervene on what is held true when, however, at least not according to me,
a sentence means is determined by the circumstances under which it and and so Davidson’s solution is not one that will work for me.
related sentences are held true. But suppose there were two situations in
which the facts available to a radical interpreter—the facts about what is held 3 Frege on Meaning and Truth-conditions
true when—were the same, but in which the facts about speakers’ mental
states—and in particular the facts about the communicative intentions with Davidson’s suggestion that a theory of truth may serve as a theory of
which they uttered various sentences—were different. On Grice’s view meaning was arguably anticipated by Frege. One might, therefore, seek a
(1989b), meaning is determined not (just) by what is held true when but solution to the Foster problem in Frege’s writings.
by facts about speakers’ communicative intentions; on the view I hold Part I of Frege’s Grundgesetze der Arithmetik (1962)22 is devoted to the
(Heck, 2005a), meaning is determined by what speakers believe about the
21 In principle, Davidson could insist that, although mental states can vary while the facts
meanings of their words.20 We would both be open to the possibility that
about what is held true when remain the same, they can vary only within very narrow limits,
the meanings of sentences could vary while the facts accessible to a radical so that such variations would not affect the meanings of sentences. But it is hard to see how
one could argue for that position, and it amounts, in any event, to assuming that mentality
20 Of course, neither Grice nor I would hold that every difference in the psychological almost supervenes on what is held true when.
states of speakers entails a difference of meaning: Supervenience is a one-many, not a 22 The English is primarily taken from Montgomery Furth’s English translation (Frege,

one-one, relation. 1964), with some minor changes. All further references will be made by section number and
3 Frege on Meaning 9

exposition of the formal system in which he proposes to work. As part names, whether simple or themselves composite, of which the
of that exposition, of course, Frege must explain his formal language. In name of a truth-value consists, contribute to the expression of the
Part I.1, he specifies the denotations of the language’s primitive symbols. thought, and this contribution of the individual component is its
Concerning the identity-sign, for example, he writes: sense. (Frege, 1962, §32; emphasis in original)
(9) ‘Γ = ∆’ shall denote the True if Γ is the same as ∆; in all other cases it That is to say, Frege takes the stipulations he has made regarding his primi-
shall denote the False. (Frege, 1962, §7) tive signs to determine the truth-condition of each sentence of his formal
language, and we may take a sentence’s truth-condition to determine, or
Frege takes this clause to specify which function the identity-sign denotes: It
perhaps just to be, its meaning. So, for example, the stipluations determine
denotes the function whose value for first argument Γ and second argument
that the formula ‘0 = 1’ denotes the True if, and only if, zero is the same
∆ is the True if Γ is the same as ∆ and is the False otherwise. There is a
as one (these being, on Frege’s view, the value-ranges of certain functions).
similar clause for each of the other primitive symbols.23 Frege then argues,
The sense of this formula is thus the thought that zero is the same as one.24
in part I.2.i, that these stipulations are sufficient to assign denotations to all
There is something obviously right about Frege’s remark. Formal lan-
well-formed expressions of his formal language. The core of that argument
guages are introduced by philosophical logicians all the time. And when
is contained in sections 30 and 31: In the former, Frege argues that any
they are, an informal theory of reference is what is typically offered by way
complex expression correctly formed from denoting expressions denotes; in
of explanation of the expressions of that formal language. Such explana-
the latter, he argues that the simple expressions of the language denote. It
tions are typically intended to convey not just what the denotations of those
follows that “the proper names [which, infamously, include the sentences],
expressions are but also what they mean, and the explanations often succeed:
and names of first-level functions, that we can form. . . out of our simple
When I have properly appreciated such an explanation, I usually take myself
names. . . always have a denotation” (Frege, 1962, §31).
to understand the formal language in question, that is, to know what its
Frege then writes:
expressions mean. So I am, and am meant to be, reading the meanings of
. . . [N]ot only a denotation, but also a sense, attaches to all names the expressions of the langauge off a theory of reference for it, just as Frege
correctly formed from our signs. Every such name of a truth-value suggests his readers should.
[that is, every sentence] expresses a sense, a thought. Namely, Now, according to Frege, sense determines reference, but reference cer-
by our stipulations it is determined under what condition the tainly does not determine sense (Frege, 1984, p. 159, op. 27), and the same
name denotes the True. The sense of this name—the thought it plausibly holds for any reasonable notion of meaning that might replace
expresses—is the thought that this condition is fulfilled. . . . The the notion of sense. If so, however, it can easily seem mysterious how
a theory specifying what the references of certain expressions are could
are to volume I. (The numbering of the sections is reset in volume II.)
23 The stipulation concerning the horizontal is in §5; negation, §6; the first-order universal possibly determine the meanings of those expressions: If reference does not
quantifer, §8; the definite article, §11; the conditional, §12; and the second-order univeral determine meaning, then specifiying the reference of an expression simply
quantifer, §20. The stipulation governing the smooth breathing, from which names of cannot suffice to specify its meaning. That line of thought, however, misses
value-ranges are formed, is given in §9 and is, unsurprisingly, not only different from the a crucial aspect of the case we are discussing, one nicely brought out by
other stipulations Frege makes but deeply problematic. I shall ignore the difficulties it poses
here, as I have discussed them elsewhere (Heck, 1998, 1999, 2007). Those same papers also 24 One might object that it ought to be the thought that: the condition that zero is the same
contain a defense of the interpretation of Part I that I am presenting, which is not universally as one is fulfilled. I take it, however, that Frege speaks of a condition’s being fulfilled only
accepted. because semantic ascent is required to state the general principle he is trying to state.
3 Frege on Meaning 10

Dummett: symbol is to have is Frege’s using a certain sentence to specify the reference
of that symbol. It is not what Frege says, then, that shows what sense ‘=’ is
When we are concerned. . . with laying down. . . what the inter- to have, but his saying what he says in a particular way or, perhaps, under
pretation of a newly introduced symbol is to be, the particular particular circumstances. Showing, in Dummett’s sense, is thus akin to
specification of the reference may be taken as conferring simulata- implicating. (It may even be a kind of implicating.) Only speech acts, we
neously a definite corresponding sense. . . . We may adapt here, in may thus conclude, show anything, in the relevant sense.
expounding a doctrine of Frege’s, the famous distinction between This point does not affect Dummett’s vindication of Frege, but it does
saying and showing that Wittgenstein used in the Tractatus: the affect any attempt to extend it to the semantics of natural language, where
specification of the reference says what the reference is to be, and, a similar problem arises. The problem I have in mind arises for anyone
by saying it in a particular manner, shows what the sense is to be. who accepts the apparently obligatory distinction between meaning and
(Dummett, 1991, p. 149) reference, but it arises particularly clearly for those who accept a distinction
between sense and reference for proper names. McDowell (1998d) proposed
One’s first reaction might be to wonder just how helpful any explanation can some time ago that this form of the problem can be solved if we appeal to
be that invokes the distinction between saying and showing. But Dummett’s the idea of Frege’s I have just been expounding.25 I shall therefore explain
idea here is reasonable enough. In saying what the reference is, Frege has the matter first for the case of proper names.
had to choose among different ways in which the reference itself might Consider the name ‘Hesperus’. One could correctly say what its reference
correctly be specified. So, for example, there are infinitely many ways Frege is in a variety of ways: One could say that it denoted Hesperus, or Venus, or
might correctly have specified the reference of the function-symbol ‘=’ in Phosphorous. To say what its reference is, then, one must choose one such
his formal language. Actually to specify what its reference is, Frege has way, and which way one chooses will, under the right circumstances, show
had to choose one such way, and the choice he has made may be taken what one takes the sense of the name to be. One might therefore suggest
to indicate—that is, to show—what sense he intends the symbol to have. that a semantic theory for English need say nothing about the meaning
Similar remarks apply to Frege’s other primitive expressions, and to other of a proper name other than what its reference is, even if one accepts the
formal languages: How its inventor chooses to state what the references of distinction between sense and reference for proper names: The theory will
a new language’s expressions are may reasonably be taken to indicate how show what the sense of a name is even if it all it says is what its reference is.
s’he intends those expressions to be understood. It is clear that, if this solution works at all, then there is a similar solution
It is important to note that Dummett’s remark explicitly concerns a case in to the Foster problem. Suppose we have a truth-theory for English that
which one is stipulating the meaning of a previously uninterpreted symbol. (canonically) proves the T-sentence
Dummett is not, I think, as clear about this point as he should be. In
particular, the last line of the passage just quoted is importantly ambiguous. (10) S is true iff p.
Dummett speaks of “the specification” of a symbol’s reference as showing
what the sense is to be, but a specification can be an act of specifying as It is true enough that, as far as the correctness of the truth-theory is con-
well as a sentence used to make a specification. The sentence Frege uses to cerned, one can replace ‘p’ with any other sentence that has the same
specify the reference of the identity-sign does not ‘indicate’ anything about 25 Following McDowell, Evans claims quite generally that Dummett’s remarks “contain
what the sense of this symbol is: Qua sentence, it says what the reference the answer to those who argue. . . that Davidsonian theories of meaning are inadequate
of the symbol ‘=’ is to be, and that is all. What indicates what sense the because they do not state the meanings of sentences” (Evans, 1985, p. 295, n. 7).
3 Frege on Meaning 11

truth-value. But to say what the truth-value of a sentence is, one must that is all; only speech acts indicate. If so, however, then, if what one wants
choose among the infinitely many different ways of specifying that truth- a semantic theory for English to do is to specify the meanings of its various
value, and which way one chooses will show what meaning one takes the sentences, then a semantic theory for English cannot consist simply of a
sentence S to have. So, for example, in propounding a truth-theory that theory of reference for English, not if one accepts the distinction between
canonically proves (1) rather than (3) or (8), we thereby show what meaning meaning and reference. The point is clearest in the case of proper names:
we take the sentence ‘snow is white’ to have, even if the theory itself says It is not enough that, by propounding this theory of reference instead of
nothing that distinguishes it from any other true sentence. that one, I might thereby show what I take the sense of ‘Hesperus’ to be.
Dummett rejects this view: Although he thinks that a stipulation of what One wants the theory to explain the phenomena, not the act of its being
a new expression’s reference is to be can convey what sense it is to have, he propounded by a theorist.
insists that One way to appreciate this problem is to consider how a theory of truth,
understood as McDowell suggests, might be used to explain the composi-
. . . a meaning-theory is required to do more than merely show tionality of meaning. For example, how does such a theory explain how the
(to someone who understands the meta-language in which it is meaning of ‘Hesperus is a planet’ depends upon and is determined by the
formulated) what the senses of the words of the object-language meanings of its parts? The theory derives a T-sentence like
are. (Dummett, 1991, p. 149)
(11) ‘Hesperus is a planet’ is true iff Hesperus is a planet
But Dummett does not, so far as I can tell, offer any reason for this claim, and
so it might simply be rejected.26 One might suggest that the parenthetical from clauses like
remark is the key to his thinking: The theory will convey what the the sense
of the name is only to someone who already grasps that sense by associating (12) ‘Hesperus’ refers to Hesperus
it with the relevant word of the meta-language. But the theory will convey (13) ‘x is a planet’ is true of x iff x is a planet,
which object the name denotes only to someone already able to refer to that
object by means of the relevant word of the meta-language. It is not clear and, more generally, derives facts about the references (or semantic values)
what the distinction between these two cases is supposed to be. of sentences from facts about the references of their syntactic parts. In
The proper complaint is that theories do not indicate things. The fact that that way, the theory explains how the reference of a complex expression is
a particular theorist is proposing one theory of truth instead of some other determined by the references of its parts. But the theory seems to offer no
may indicate what senses that particular theorist takes expressions in the explanation at all of how the sense of a sentence is determined by the senses
relevant language to have. But it is, as we have seen, no part of what the of its parts. Certainly, in propounding a theory that has a consequence like
theory actually says that a particular name has a particular sense. And the (11), one indicates what one takes the meaning of ‘Hesperus is a planet’ to
clauses of the theory, qua sentences, do not indicate anything about what be; and since the theory also includes (12) and (13), one similarly indicates,
the senses of expressions are: They say what those expressions denote, and by propounding the theory, what one takes the senses of ‘Hesperus’ and
26
‘is a planet’ to be. Since the theory derives (11) from (12) and (13), then,
If Dummett does offer a reason, it is equivalent to the insistence that a theory of meaning
must, in his sense, be ‘full-blooded’ rather than ‘modest’ (Dummett, 1993e). But it seems
it derives a theorem whose (canonical) provability in the theory one is
likely that this issue and the one we are discussing here are effectively the same, and propounding indicates what one takes the meaning of ‘Hesperus is a planet’
McDowell, in any event, would reject this requirement (McDowell, 1998b). to be from axioms whose presence in the theory one is propounding indicate
4 Understanding and Knowledge 12

what one takes the senses of ‘Hesperus’ and ‘is a planet’ to be. But it simply interpretive theory “absurd” . Davdison nonetheless insists “that any theory
does not follow that one has thereby explained why ‘Hesperus is a planet’ that passes the tests will serve to yield interpretations” (1984c, p. 139). This
has the sense it has in terms of the fact that ‘Hesperus’ and ‘is a planet’ have remark is more puzzling, I think, than it is usually taken to be: In what sense
the senses they do. is a truth-theory supposed to “yield interpretations”? What the theory yields
The problem can be seen from this diagram: would seem to be theorems in the form of T-sentences. Perhaps the right
?
question to ask is therefore a different one: What does it mean to say that a
sense(A) sense(B) ⇒ sense(C) theory of truth can be used to interpret a language, or a speaker? Suppose
↑ ↑ ↑ Fred utters a sentence, and I wish to interpret this utterance. Suppose further
re f (A) re f (B) ⇒ re f (C) that I happen to know that Fred’s utterance is true iff Bedrock is cold. What
The double arrow ‘⇒’ symbolizes the relevant relation between the am I supposed to do with this information?
properties of the parts and the properties of the whole; the single arrow ‘↑’ One proposal would be that the radical interpreter is supposed to use her
may be taken to mean implies, indicates, or shows, as one likes. No matter theory of truth explicitly to assign meanings to utterances. The radical inter-
which of these readings one gives it, there is simply no reason to suppose preter would thus need to make something like the following inference:28
the there should be a double arrow in the top line, at least for all that has S is true iff p
so far been said. Perhaps one could argue that for the special case in which
‘↑’ means shows, the double arrow will always be present, but that very That fact is a consequence of a theory of truth I know to be correct
much needs to be argued. A good argument, it seems to me, would be one on the basis of the sort of evidence available to me qua radical
that replaced the metaphor of showing with something much more precise: interpreter.
It would need to be said just what relation holds between a specification ∴ S means that p
of reference and a specification of sense. I conjecture that, if this could
be done, then what replaced the metaphor of showing would be precise If interpreting a speaker requires one to assign meaning to her utterances, in
enough to be embedded in a theory of meaning, and there would no longer to sentences. One might want to question that assumption. It is not, in fact, obvious that
be any need for the sort of solution to the Foster problem we have been Davidson himself would have endorsed it. Quine (1960) famously denied that the criteria
considering. for correct translation restrict the range of admissible translation manuals even to those
that are materially equivalent. So far as I know, Davidson never says whether he thinks the
truth-value of a sentence is uniquely determined by his criteria.
4 Understanding and Knowledge of Meaning 28 In the spirit of the interpretation of Davidson offered in section 2, one could take the

inference simply to be from ‘S is true iff p’ to ‘S means that p’. The inference is not,
In our earlier discussion of radical interpretation, the central question was of course, always truth-preserving, but one might insist that it will be truth-preserving if
the premise is canonically provable in a theory of truth known on the basis of the sort
whether the ‘formal and empirical criteria’ Davidson specifies are sufficient of evidence available to a radical interpreter. The radical interpreter will not go wrong,
to restrict the range of acceptable truth-theories to those that are ‘interpre- therefore, if she makes the inference only under these circumstances. Another way to put
tive’.27 Now, Davidson endorses Quine’s thesis of the indeterminacy of this point is to say that the second premise here need not be known but need only be true.
translation: He calls the suggestion that his criteria determine a unique So we would be thinking of the epistemology of our knowledge of meaning as being, in this
respect, externalist. Though this point is important in other contexts, it is point is not terribly
27 As most commentators usually do, I tacitly conceded, for the sake of argument, that the important in the present context. I shall therefore speak as Davidson himself does, as if the
critieria admit only correct truth-theories, that is, theories that assign the correct truth-values interpreter knows that her theory of truth satisfies the emprical and formal conditions.
4 Understanding and Knowledge 13

this sense, however, then the theory of truth is simply an auxiliary device the right-hand side of the T-sentences the theory (canonically) proves will
used to generate assignments of meaning (compare Soames, 1992, REF). It translate the sentences mentioned on their left-hand sides. But suppose a
does not seem likely that Davidson wanted the notion of meaning to emerge radical interpreter were nonetheless irresponsibly to settle for one of the
so crucially here at the end of the story. theories the criteria admit and set about using it to interpret the speakers of
XXXXX the language. Then it is tempting to say that, just as Frege’s choosing to
A different suggestion is that the T-sentence itself can be used to interpret specify the reference of a term in a particular way shows us what sense he
the speaker. The difficulty is to say how, and part of the problem here is takes it to have, so the radical interpreter’s using her chosen theory amounts
that the notion of interpretation itself is not particularly clear: What exactly to her choosing to regard sentences of the target language as having certain
does Davidson mean by ‘interpreting’ an utterance, a speaker, or a language, meanings rather than others. That is to say, the same line of thought we
whichever the right target might be? I really don’t know, so I propose to considered in the last section seems to lead to the conclusion that a radical
reformulate the question. I take it that what Davidson has in mind is that interpreter who chose to interpret using a theory that proved (1) rather
someone who knew a theory of truth for a language, and knew that it met than one that proved (8) would thereby have chosen to regard utterances
the formal and empirical criteria he specifies, could use that theory to speak of the sentence ‘snow is white’ as meaning that snow is white rather than
and understand that language. So the question I am going to discuss is: that snow is white and arithmetic is incomplete. But of course several
How could one use a theory of truth, whatever its pedigree, to speak and questions remain: In what sense does using a particular theory of truth to
understand the language it concerns? That, of course, depends upon what interpret a sentence involve assigning a particular meaning to that sentence?
one means by ‘speaking and understanding the language’. As is common, I Is there some analogue for ordinary speakers of our irresponsible radical
am going to concentrate here on what I take to be one fundamental aspect interpreter’s decision to use one theory of truth rather than another?
of speaking and understanding a language, namely, using the language to Before I continue, I need to say something explicit about the overall
communicate, that is, to exchange information with other speakers: To be structure of the considerations to follow. As I said earlier, I accept a cogni-
able to speak a language at least involves being able to use it to acquire tive conception of (occurrent) understanding: Understanding an utterance,
beliefs from other speakers and to convey one’s own beliefs to them.29 according to me, is a propositional attitude Heck (2005a). That view leaves
Our question, then, may be put as follows: In what sense, if any, can a open the question what the content of that attitude is, however, and that is
theory of truth for a given language be used by a radical interpreter (or an the question I mean now to discuss. How, though, should we evaluate the
ordinary speaker) to allow her to communicate using that language? The various proposals that might be made here?
examples due to Foster’s and Soames’s may seem to suggest that the answer The following conditional should be uncontroversial:
should be: It can’t. But there is reason for hope. I argued earlier that Foster
and Soames are right, as regards Davidson’s actual position, that we have (1) If B utters a sentence S, which in that context means that p,
no good reason to believe that a theory of truth that is known to be correct and if A understands B’s utterance and meets whatever conditions
on the basis of the sort of evidence available to a radical interpreter will a correct theory of testimony would require us to include at this
always satisfy Tarski’s Convention T—that is, that the sentences used on point, then A is in a position to come to know, or at least justifiably
to believe, on a certain distinctive kind of ground, that p.
29 Ian Rumfitt emphasizes in Rumfitt (1995) that there is no single reason for which we
utter sentences, a point some of my own previous work—e.g., Heck (1995)—neglects. The The conditional is uncontroversial because trivial. One might well say
present discussion is intended, however, to be compatible with Rumfitt’s observations. that the correct theory of testimony just is whichever theory makes this
4 Understanding and Knowledge 14

conditional correct. It is controversial, of course, which theory that is, but know its truth-condition.
we may ignore that issue here: We are not discussing the epistemology of Suppose that Fred asserts the sentence ‘Bedrock is cold’. Suppose further
testimony. What we are discussing is what constitutes understanding an that Barney meets whatever conditions a good theory of testimony would
utterance. And what I am suggesting is that, just as satisfying (1) is very say he needed to meet: Let’s just assume he has excellent reason to believe
nearly what it is for a theory of testimony to be correct, so satisfying (1) that Fred is sincere and excellent reason to regard Fred as relaible with
is very nearly what it is for a theory of understanding to be correct: To regard to what he has said; any reasonable theory would regard that as
understand an utterance is to be in a position—modulo one’s satisfaction sufficient. Then, if Barney understands Fred’s utterance, he will be in a
of the sorts of conditions a correct theory of testimony would specify—to position to form a justified belief, indeed, a belief that might constitute
acquire a belief for which one would then have a certain distinctive sort knowledge, namely, the belief that Bedrock is cold. Now suppose that
of justification. If so, then theories of understanding may be tested by Barney knows
determining whether they satisfy (1). Particular examples, of course, may
depend upon controversial claims about what the correct theory of testimony (2) ‘Bedrock is cold’ is true iff Bedrock is cold.
is. But not all examples need so depend. Particular examples may be neutral
on what precise conditions a correct theory of testimony would require to Intuitively, knowing (2) seems insufficient for Barney to garner knowledge,
be satisfied: All (reasonable) theories may agree that the relevant conditions from Fred’s utterance, that Bedrock is cold: The truth of (2) is, after all,
are satisfied in that case. compatible with the hypothesis that ‘Bedrock is cold’ means that Bedrock
There are some substantive assumptions about testimony to which I is broke, so long as Bedrock is broke iff Bedrock is cold. But if what
shall be appealing, but I take these to be relatively uncontroversial. I shall the sentence ‘Bedrock is cold’ actually means is that Bedrock is broke,
assume, for example, that, to acquire knowledge by testimony, one must then it is not plausible that Barney could, in reaction to Fred’s utterance,
be appropriately sensitive—whatever exactly that may mean—to whether come to know that Bedrock is cold, at least not in the distinctive way in
one’s informant is sincere and reliable. I shall also assume, slightly more which one comes to know something by being told. One might want to
controversially, that the distinctive sort of justification testimony makes object that if Barney has good reason to regard Fred as sincere and reliable,
available has a certain kind of structure. In particular, there is a way in that constitutes good reason to suppose that Fred’s utterance is true. That,
which the justification one might have had for the belief that p can be together with (2), which we are assuming Barney knows, then gives him
undermined that leaves one with justification for the weaker claim that the good reason to suppose that Bedrock is cold. But although Barney could
speaker believes that p. In particular, one might be presented with strong come justifiably to believe, in this way, that Bedrock is cold, his belief would
evidence that the speaker is not reliable with respect to the question whether not be justified in the distinctive way in which beliefs acquired through
p: That would deprive one of justification for the belief that p but not for testimony are justified. The reason is that, absent knowledge of what Fred’s
the belief that the speaker believes that p.30 utterance actually meant, Barney’s belief cannot be appropriately sensitive
to the question whether Fred is reliable. To evalute Fred’s reliability, Barney
So let us now consider this proposal: To understand an utterance is to
must know what belief Fred is expressing. Barney may regard Fred as
30
authoritative with regard to the weather, say, but not with regard to municipal
This last assumption entails that one can only have testimonial justification that p if one
also has testimonial justification that one’s informant believes that p. That could be, and has
finance, and these attitudes may be justified. So if Barney takes Fred
been, questioned, but it is weak enough, and intuitively plausible enough, that I am willing sincerely to have said something about the weather, he may be prepared to
to rest my case upon it. believe it, and be justified in doing so, so that he might acquire knowledge
4 Understanding and Knowledge 15

about the weather from Fred. But Barney may not be prepared to believe what one takes them to mean oneself, and perhaps such a default assumption
what Fred says about municipal finance, and he would not be justified in is in some sense necessary. Nonetheless, when I am responding to the words
doing so if he did, whence he cannot acquire knowledge about municipal of another, the crucial question I must answer is what she means by her
finance from Fred. That ‘Bedrock is cold’ is true iff Bedrock is cold does words.32
not, however, imply anything about what belief Fred is expressing. I take it, then, that the following inference would be a reasonable one for
That is the fundamental point: Absent knowledge of what Fred’s ut- me to make:
terance means, Barney is in no position to determine what belief Fred is
expressing.31 The point is worth making directly. Suppose Barney under- (3) The speaker, N, has uttered the sentence S.
stands Fred’s utterance and justifiably regards it as sincere. Then even if (4) N takes her utterance of S, in the present context, to mean that
he is not sure about Fred’s reliability, he may conclude, and even come to p.
know, that Fred believes that Bedrock is cold. But if all Barney knows about (5) N’s utterance of S is sincere.
Fred’s utterance is that it is true iff Bedrock is cold—if he does not know
what it means—then he can draw no conclusion whatsoever about Fred’s (6) Hence, N believes that p.
beliefs. And so Barney’s knowing a correct T-sentence for the sentence
What I am now going to argue is that the role played in such an inference
Fred uttered cannot constitute his understanding it.
by (4) may be played instead by something like:
The natural counter-proposal is that understanding consists not in knowl-
edge of an utterance’s truth-condition but in knowledge of what it means (7) N takes her utterance of S, in the present context, to be true iff
(Richard, 1992). Perhaps surprisingly, however, even knowledge of what an p.
utterance means is insufficient to support attributions of belief on the basis
of what someone has said. Consider again Fred’s utterance of ‘Bedrock The motivation for the suggestion can be explained as follows. We began
is cold’ and suppose that Barney knows that ‘Bedrock is cold’ means that with the proposal that the work being done by (16) could be done instead by
Bedrock is cold. Fred may not: Maybe he thinks it means that Bedrock
(8) N’s utterance of S is true, in the present context, iff p.
is broke. Barney may even have reason to doubt that Fred thinks it means
that Bedrock is cold. If so, then Barney cannot justifiably conclude that In moving from (4) to (8), we have made two changes: We have replaced the
Fred believes that Bedrock is cold. What Barney needs to know if he is to reference to the utterance’s truth-condition with a reference to its meaning,
attribute a belief to Fred is not just what Fred’s words mean but what they and we have replaced an absolute judgement about what the meaning of
mean to Fred. The point should be obvious once stated: Communication the utterance is with a judgement about what the speaker takes it to mean.
with another speaker—in so far as it involves attributing beliefs to her and My suggestion is that the latter change is the one that really matters and
acquiring beliefs from what she says—depends not upon what her words that it is, in fact, adequate by itself to resolve the problems faced by the
mean nor even upon what one takes her words to mean oneself: It depends initial proposal. Actually, that’s not quite right: (7) will not quite do. But
upon what she means by her words and upon what one takes her to mean by something much like it will do. Before I explain how it needs to be amended,
her words. Now typically, what one takes someone to mean by her words is though, let me emphasize two features of the proposal I am making.
31 Thanks to Crispin Wright for helping me understand the structure of my thought on 32 In fact, something similar is true even when I am speaking: The crucial question for me
these matters. is how my words will be understood. See Heck (2006) for more on this matter.
5 Understanding and Truth-Conditions 16

First, one might wonder whether it is enough to defend Davidson’s conditions we consciously know are the upshot of subpersonal processes
suggestion that a theory of truth may serve as a theory of meaning to that draw upon the information contained in the axioms of a theory of
defend the view that the knowledge of meaning on which ordinary linguistic truth and, indeed, that these processes implement an algorithm for deriving
action depends consists in knowledge of truth-conditions. Questions about T-sentences from tacitly known axioms of the theory of truth. I think
meaning, one might say, are one thing, and questions about what, if anything, something along those lines is correct Heck (2004), but I will not defend
ordinary speakers know, in virtue of which they are able to communicate, the claim here.
are another Soames (see, for example, 1988). But what interesting questions It is unclear to what extent this view differs from Davidson’s. On the one
would remain unanswered if it could be established—as I think it can be— hand, Davidson insists, in “Radical Interpretation”, that the question that
that a speaker’s ability to communicate depends upon, and is explained by, interests the semanticist is hypothetical: What could we know that would
her knowledge of truth-conditions? I do not know of any. In any event, the allow us to interpret the words of other speakers? Regarding this question,
view I am defending accepts a strong form of Dummett’s oft-cited insistence he writes that it
that “a theory of meaning is a theory of understanding” Dummett (1993e, p.
3): I take the primary notion to be how a speaker understands an uttered is not the same as the question what we do know that enables us to
sentence, not what a word means in a language. One can reconstruct the interpret the words of others. For there may easily be something
latter notion if one likes in terms of the former notion, but I do myself think we could know and don’t, knowledge of which would suffice for
it is of much theoretical interest.33 interpretation, while on the other hand it is not altogether obvious
that there is anything we actually know which plays an essential
Second, my view, as I’ve stated it, does not require that there be any
role in interpretation. Davidson (1984c, p. 125, emphasis in
sense in which speakers know theories of truth for their languages. If
original)
understanding an utterance involves knowing a T-sentence for it, and if
knowledge of that T-sentence is to be among one’s reasons for reacting It is tempting to read this passage as rejecting the view I am defending. If,
to the utterance in a certain sort of way, then one’s knowledge of the T- however, what Davidson means by ‘know’ here is consciously know, and if
sentence must be conscious, in the sense that it is available for the rational what he has in mind is not just knowledge of T-sentences but knowledge of
control of action. It is agreed on all sides, of course, that ordinary speakers a theory of truth, then there need be no disagreement between us: Davidson
do not have conscious knowledge, in this sense, of semantic theories for could consistently agree that understanding another speaker requires con-
their languages. So if speakers do, in some sense, know semantic theories scious knowledge of what she takes the truth-conditions of her utterances to
for their languages, that knowledge is tacit, and tacit knowledge is, almost be while remaining agnostic about the empirical question what the source
by definition, not available for the rational control of action.34 That said, of such knowledge is.
however, the question nonetheless begs to be asked what the source of
our conscious knowledge of truth-conditions is, if we really do have such
knowledge. It is a natural, though not inevitable, proposal that the truth- 5 Understanding and Truth-Conditions

33
The view I am defending, then, is that, in so far as we rely, in linguistic
This view is standard-issue among linguists but relatively uncommon among philoso-
phers. See Heck (2006) for a defense of it.
communication, upon knowledge of meaning, knowledge of meaning may
34 If speakers did consciously know semantic theories for their languages, then semantics be taken to consist in knowledge of truth-conditions. More precisely, I am
would be easy: Just ask some native speakers what they know. claiming that the role that is played by (4) in this reasoning
5 Understanding and Truth-Conditions 17

(3) The speaker, N, has uttered the sentence S. And even if he wasn’t paying attention in science class, surely Fred must
(4) N takes her utterance of S, in the present context, to mean that know
p. (11) ‘Bedrock is cold’ is true iff Bedrock is cold and water is wet.
(5) N’s utterance of S is sincere. There are plenty of similar cases.35
(6) Hence, N believes that p. Higginbotham has suggested that, though the Foster problem shows that
“meaning does not reduce to reference, . . . knowledge of meaning reduces to
can instead be played by something similar to the norms of knowledge of reference” Higginbotham (1991, p. 274). These
“norms” concern what a competent speaker of the language is supposed to
(7) N takes her utterance of S, in the present context, to be true iff
know in so far as she is competent. So, very roughly:
p.
To know what a sentence S means is: (i) to know that S is true iff
I use the non-committal term ‘takes’ here because I mean to leave it open p and (ii) to know that knowing that S is true iff p is what, and all,
what the relevant psychological state is. I have said already that it must be competence requires a speaker to know about it.
conscious, in the sense that it is available for the control of rational action,
but nothing in what follows will depend upon whether it is knowledge, For example, to know what ‘Bedrock is cold’ means is to know (2) and to
belief, or something else still. know that (2) is what, and all, a competent speaker is supposed to know
Embedding Davidson’s proposal within the cognitive conception already about it. If so, then since one is not, in general, expected to know (9), (9)
gives us some headway with the Foster problem. On this view, after all, what does not, in any sense, ‘give the meaning’ of ‘Bedrock is cold’.
one takes an utterance to mean is, in effect, a matter of what one believes It is unclear, however, what Higginbotham means by what one is expected
the speaker takes its truth-condition to be, so not just any T-sentence one to know ‘in so far as one is a competent speaker’. Is one expected to know
knows for the utterance will equally serve to interpret it. One might well
(12) “All bacherlors are married” is true iff all bachelors are
know oneself, for example, that
female
(9) ‘Bedrock is cold’ is true iff Bedrock is cold and arithmetic is in so far as one is competent? On some views, the answer would arguably
incomplete, be that one is. “All bachelors are unmarried”, the thought might be, is
analytically false: Anyone who understands it must know that it is false,
but if one does not think that Fred knows (9), one will not use one’s knowl-
and so perhaps one must know that it is false if one is competent. But “All
edge of (9) to interpret his utterances of ‘Bedrock is cold’. But, of course,
bachelors are female” is also analytically false, so maybe one is expected to
that’s far from sufficient progress. Perhaps one met Fred at a logic con-
know (12) simply in so far as one is a competent speaker. Perhaps not, but
ference and so presume that he does know (9). And even if Fred isn’t a
it is not obvious there are no similar counterexamples.
logician, maybe he was paying attention in his basic science courses, in
But even if Higginbotham’s proposal is extensionally adequate, it is not
which case he presumably knows
clear how much explanatory value it has. Most competent speakers of
(10) ‘Bedrock is cold’ is true iff Bedrock is cold and water is 35 Note that since (11) is presumably common knowledge if 2 is, it will not suffice to
H2 O. strengthen (7) by inserting a reference to common knowledge.
5 Understanding and Truth-Conditions 18

English do know (11), and they expect other competent speakers to know would deploy that belief in interpreting his utterance, it would be irrational
it, too. It is natural to say that they do not know (11) in the same way for him to utter “Bedrock is cold” in an effort to get Barney to believe, in
they know (2); that their knowledge of (11) depends somehow upon their response to his utterance, that Bedrock is cold and water is wet. Of course,
knowledge of (2); and that their knowledge of (2) is ‘linguistic’ in a way Fred might suppose that, if he can get Barney to believe that Bedrock is
that their knowledge of (11) is not. We might summarize these intuitions by cold, he can thereby get him to believe that Bedrock is cold and water is wet.
saying that competent speakers of English know (2), but not (11), simply as But Fred’s reasons for uttering the sentence, in this case, are just different
competent speakers of English (Higginbotham, 1991, p.274). But talk of from the sorts of reasons he has if he assumes that Barney will deploy (2) in
what one knows simply as a speaker is obscure, and one would like to have interpreting his speech.
these intuitions explained, not just labeled. Such an explanation is what I The lesson that is supposed to emerge from this fantasy is that one way
shall now attempt. to distinguish interpretive T-sentences, such as (2), from non-interpretive
Utterances are acts. As such, they are typically performed for reasons. T-sentences, such as (11), is simply by reference to the roles they play
Fred’s uttering ‘Bedrock is cold’ is an act, and one typical reason for uttering in the production and comprehension of speech: A speaker treats a given
it might be to communicate, to Barney, that Bedrock is cold. But Fred’s T-sentence as ‘giving the meaning’ of a sentence if she deploys that T-
utterance of this sentence is rational, as a means to this end, only if he is sentence in a certain distinctive way, both when speaking herself and when
assuming certain facts about Barney, for example, that Barney will take interpreting the speech of others. For example, we may reconstruct Fred’s
him at his word. Of special interest to us, however, is the need for Fred reasons for uttering ‘Bedrock is cold’ roughly as follows:
to assume that Barney will understand him as having said that Bedrock is
cold. If Barney instead took him to have said that Bedrock is broke, Fred’s (13) I want Barney to know that Bedrock is cold.
attempt to communicate to Barney that Bedrock is cold would fail. So (2) “Bedrock is cold” is true iff Bedrock is cold.
if, as I am assuming, Barney’s understanding Fred’s utterance consists in
(14) Barney knows (2), and he will deploy that knowledge in
his knowing something about it, we may now ask: What is it that Fred is
interpreting my speech.
assuming Barney knows?
(15) If I utter “Bedrock is cold”, in the present context, Barney
Earlier, I suggested that, if an irresponsible radical interpreter were simply
will know (or believe) that I have spoken the literal truth.
to settle upon a theory of truth and set about using it to interpret the natives,
her doing so would amount to her assigning certain meanings to their (16) If I utter “Bedrock is cold” in the present context, Barney will
utterances. Suppose Fred regards Barney as similarly irresponsible: Despite (be in a position to) come to know that Bedrock is cold. (From
the fact that Barney knows that Fred knows many correct T-sentences for (14) and (15).)
‘Bedrock is cold’, Barney has nonetheless decided to use a particular one of (17) I shall utter Bedrock is cold. (From (13) and (16).)
them to interpret Fred’s utterances. If Fred believed that the one on which
Barney had settled was (2), then Fred’s uttering ‘Bedrock is cold’ in an That (2) ‘gives the meaning’ of “Bedrock is cold”, for Fred, simply consists
effort to communicate to Barney that Bedrock is cold would make perfect in his using it in the way here illustrated, and in the following sort of way
sense: If Barney takes Fred at his word, that is precisely the belief he will when interpreting Barney’s utterances:
acquire. It may well be that Fred also expects that Barney believes (11),
at least implicitly. But unless Fred were prepared to suppose that Barney (18) Barney has uttered “Bedrock is cold”.
5 Understanding and Truth-Conditions 19

(2) “Bedrock is cold” is true iff Bedrock is cold. Unfortunately, I can’t seem to do that quite yet.36 So let me instead suggest
(19) Barney knows (2), and he expects me to deploy that knowl- a way of approaching the issue from a slightly different direction.
edge in interpreting his utterance. As mentioned earlier, there is a strong intuition that knowledge of (9) and
(11) depend in some way upon knowledge of (2), that is, that (2) is, in some
(20) Barney is speaking the literal truth. sense, one’s most basic piece of semantic knowledge regarding the sentence
(21) Bedrock is cold. (From (18), (20), and (19).) “Bedrock is cold”. So one might hope to characterize linguistic knowledge
in terms of this notion of a speaker’s most basic semantic knowledge. One
One could obviously go on in this way for some time, exploring the role way to do so would begin with the thought that our most basic linguistic
knowledge of (2), and mutual expectations about its deployment, play in knowledge has a distinctive source.
the many different forms of reasoning in which speakers engage. In any event, the question begs to be asked what the source of our con-
Note that Fred’s expectation that Barney will deploy the belief that his scious linguistic knowledge is, be it knowledge of truth-conditions or knowl-
utterance of “Bedrock is cold” is true iff Bedrock is cold is an expectation edge of meaning. The sorts of phenomena that are used to motivate the
that Barney will suppose that Fred himself has deployed the same belief principle of compositionality suggest, though they do not of course imply,
in speaking. So Fred expects Barney to expect him to have deployed this that, as Dummett puts it, “a speaker of a language derives his understanding
belief in speaking, and we could continue the iterations. The expectations of any sentence of that language from his knowledge of the meanings of the
of the participants in a given communicative exchange are thus, at least words” that compose it Dummett (1993f, p. 36). Of course, such a deriva-
when all goes well, mutual, not just in the sense that they are shared but tion is not conscious, so the proposal would have to be that our concsious
in the stronger sense that they are intertwined: Each conversational par- linguistic knowledge is the product of subpersonal computational processes
ticipant’s expectations concern, inter alia, the expectations of the others. that draw upon the information contained in tacitly known theories of truth.
As Higginbotham notes regarding a similar feature of his own proposal If so, then the knowledge we deploy in speech and interpretation is, in the
(Higginbotham, 1991, p. 274), this fact reflects the same sort of phenomena normal case, delivered by the language faculty. It is, of course, an empirical
that lead to Grice’s emphasis on the overtness of communicative intentions question what knowledge is so delivered, in any particular case. But it is
(Grice, 1989a, p. XXX) and to Lewis’s emphasis on common knowledge not unreasonable to suppose that (2) is so delivered and (11) is not.
(Lewis, 1986, p. XXX). McDowell emphasizes, throughout his writings on such topics, that we do
The suggestion, then, is that linguistic knowledge is knowledge that is not simply hear the words other people utter and then add an interpretation
used in a certain distinctive way: It is the knowledge one deploys in speech in thought McDowell (1998a, pp. 331ff). Rather, in the normal case, our
and interpretation. The notion of how one deploys information in speech appreciation of what someone has said is a result of our perceiving not just
and interpretation is thus the crucial one. I take the notion to be explained the words she has uttered but also, in the strictest sense, perceiving what
implicitly by the role it plays in the sorts of reasoning we have just been she has said. If so, then one’s interpretation of speech is not always a matter
discussing: A speaker has deployed the belief that S is true iff p in speaking
36
if she uses it in the way Fred uses his knowledge of (2) in deciding to utter Crispin Wright suggested that, for the case of the speaker, one might appeal to some
“Bedrock is cold”; an interpreter deploys her belief that S is true iff p in notion of the intention that is most immediately controlling the act of uttering the sentence.
Once spelled out, that is probably close to a proposal of Rumfitt’s. Adapted it to the present
interpreting an utterance if she uses it in the way Barney uses his knowledge framework, it is that the T-sentences that constitute understanding are those that occur in
of (2) in interpreting Fred’s utterance of “Bedrock is cold”. But of course what he calls the ‘lowest practical syllogism’ that occurs in practical reasoning of the sort
it would hardly be unreasonable to ask that the implicit be made explicit. we’ve been discussing Rumfitt (1995, p. 842).
6 Truth and Semantic Knowledge 20

of what information one chooses to deploy but is typically a matter of what 6 Truth and Semantic Knowledge
information just does get deployed in one’s perception of speech. One
should not, however, suppose that such an observation shows that higher The foregoing constitutes a principled, if still incomplete, solution to the
cognitive states, like beliefs, have no central role to play in speech and Foster problem: We do have reason to believe that a speaker who con-
interpretation.37 If one happens to know that Jones uses a particular word— sciously knew the truth-conditions of sentences of a given language (as
say, ‘livid’—differently from how one uses it oneself, then, although one uttered, or potentially uttered, in given contexts), and who was prepared to
may still hear Jones as saying that Smith was pale when he utters “Smith deploy that knowledge in speaking and interpeting others, would understand,
was livid”, one may choose not to deploy that information in interpreting and be able to use, that language, if the truth-conditions she assigned were
him. Rather, since one knows, via some other route, that, when Jones utters the same as those assigned by other speakers.
“Smith was livid”, he is not himself deploying a belief that it is true iff The arguments so far given do not, however, show that all competent hu-
Smith was pale but rather a belief that it is true iff Smith was flushed, one man speakers do have this sort of knowledge. And Soames, for one, doubts
deploys that information instead. Of course, I usually just proceed on the that they do. In particular, he thinks we have no reason to suppose that all
assumption that Jones’s words mean to him what they mean to me. But so competent speakers possess a meta-linguistic concept of truth. Unless some-
to proceed is, in effect, to take appearances at face value and so to act on the one does grasp such a concept of truth, she will be in no position to have
perceptually justified belief that Jones’s utterance is true iff Smith was pale. conscious beliefs about the truth-conditions of her sentences: One cannot
believe that “snow is white” is true iff snow is white (or grass is green, or
Ultimately, or so I suspect, an adequate answer to the question what anything else) unless one grasps a meta-linguistic concept of truth, one that
linguistic knowledge is will have to make reference both to facts about applies to utterances, sentences as (potentially) uttered in contexts, or some
how that knowledge is put to use and to facts about its normal source. such thing. There is, of course, no reason to suppose that every competent
Unfortunately, I must confess that I am unable at present to say anything speaker must understand a word synonymous with “true”: Indeed, it is
that is likely to satisfy someone skeptical that an adequate answer can be likely that not every competent speaker does. So the strongest defensible
sewn from the materials we have been discussing. In my defense, I plead claim is that competent speakers implicitly grasp a meta-linguistic concept
that the question is famously difficult one. The proposal I am making is at of truth.
least not already known to fail. Soames doubts that claim can be defended, however:

Must [young children] possess a metalinguistic concept of truth,

even though they have no word for it? I don’t see why. Perhaps
37 McDowell sometimes gives the impression that, epistemologically, it is of no particular it will be suggested that a person who lacked such a concept
interest that we do not just perceive what someone has said but also perceive her words.
Tyler Burge has argued for a similar view, namely, that one’s justification for the claim that
couldn’t be a language user, since to use language one must realize
someone has said that p need not involve any claim about what words she has uttered Burge that assertive utterances aim at truth. . . . But this suggestion is
(1993). But the possibility of the sort of re-interpretation I consider here depends upon our confused. The child will get along fine so long as he knows that
perceiving someone’s words, and we do not perceive someone’s words and what she has ‘Momma is working’ is to be assertively uttered only if Momma
said as one can perceive an apple and an orange: We perceive what has been said in the
words somehow, and so our perception of speech is itself ‘semantic’, in the sense that in it
is working; ‘Daddy is asleep’ is to be assertively uttered only if
meaning binds to sound. Precisely what that might mean is a very nice question. See Heck Daddy is asleep; and so on. . . . [I]t is not necessary to have [a
(2005a) for more on this issue. metalinguistic concept of truth] in order to learn and understand a
6 Truth and Semantic Knowledge 21

language. (Soames, 1988, p. 189) could perhaps avoid this conclusion by abandoning the assumption that a
regularity is a convention only if parties to the convention have certain sorts
So the idea would seem to be that, rather than having to know the truth- of intentions and so forth. To do so would, however, in effect be to abandon
conditions of a given sentence, all a speaker need know about a sentence, what, in Heck (2005a), I call the ’propositional rationality of speech’, that
to understand it, is when it can appropriately be uttered. In particular, to is, the claim that speech is intentional under descriptions of the form saying
understand the sentence “Momma is working”, it is enough to know what that p. That claim is not optional.
I shall call the semantic injunction for it: that it is to be uttered only if Let me return, then, to the question whether every competent speaker must
Momma is working. Since no meta-linguistic concept of truth is involved implicitly grasp a meta-linguistic concept of truth. I simply do not find it as
here, competent speakers need not have any such concept. peculiar as Soames apparently does to attribute such a grasp even to young
It is worth noting, before we continue, that Soames’s view is not an children. Every competent speaker must have a conception of an utterance’s
alternative to but a version of the cognitive conception of (occurrent) un- being correct or incorrect. Without such a conception, one could hardly be
derstanding. Like my view, his is committed to the claim that a speaker’s rational in one’s use of language: One would be unable even to conceive of
understanding of a sentence consists in knowledge or beliefs that she has the possibility of an utterance’s being incorrectly made; one therefore could
about it. On Soames’s view, these beliefs are of the form: One is to utter S not so much as make sense of the idea of evaluating what was said in order
only if p, whereas on my view they are of the form: S is true iff p. If Soames to decide whether one should believe it. Now, in a sense, of course, there
is right, then, we need not abandon the view that to understand a sentence S are many ways in which an utterance can be incorrect: There is a very broad
is to know something about it nor even that it is to know what S means: We sense of incorrectness, for example, in which an utterance is incorrect if it is
should say, rather, that to know what S means is to know the condition under rude, inappropriate, or irrelevant. But it is hardly unreasonable to suppose
which it may be uttered, either by oneself or by someone else. Knowledge that there is a more fundamental notion of an utterance’s being objectively
of meaning, on this view, is thus knowledge of semantic injunctions, rather correct that underlies the undifferentiated conception of correctness,38 and
than knowledge of truth-conditions. And so, as said, Soames and I disagree even quite young children seem to understand that, although Daddy said that
not about whether competence with a language is constituted by linguistic he has only one cookie, he may in fact have more. Of course, these claims
knowledge but about what the content of that knowledge is. could be questioned. But it is not hopelessly ad hoc to suggest that our
Something similar is true of David Lewis’s view that the semantic proper- implicit grasp of the concept of truth lies in the conception of an utterance’s
ties of a language are fixed by conventions regarding its use (Lewis, 1985a, objective correctness that competence presupposes.
1986). For it to be a convention in P that S is uttered only when q, members
38
of P must intend to utter S only when q, expect one another to do so, know I am inclined to think, myself, that rational thought demands deployment of a corre-
that others expect them to do so, and have such intentioBedrock is coldns sponding conception of a belief’s being objectively correct. I shall not try to justify this
metaphysical presumption here. For an attempt to do so, see Dummett (1993c). It is perhaps
for certain sorts of reasons. These conditions concern what members of worth nothing that the objectivity required here is very weak: It is not, in particular, distinc-
P intend and what they expect. These intentions and expectations need tive of realism, in Dummett’s sense. All that is required is that there should be a difference
not, of course, always or even often be occurrent. Nonetheless, on Lewis’s between what is true and what is appropriate, relevant, or justified, as opposed to justifiable
view, the ability to use a language in accord with the relevant conventions in principle. Dummett argues, in quite general terms, that at least this much objectivity
is required of any notion of truth, be it realist or anti-realist. See Dummett (1978a, pp.
involves having certain sorts of psychological attitudes that are conscious in XXX-XX) for the argument. A Dummettian anti-realist thus accepts that truth is objective in
the sense that they are available for the control of rational action, such states the weak sense that is relevant here. (Indeed, what I am arguing is, as Dummett would put it,
including, for example, the standing intention to utter S only when p. One precisely that truth must be the central notion of the theory of meaning in the ‘weak’ sense.)
6 Truth and Semantic Knowledge 22

Now, it is clear enough that no (meta-linguistic) notion of truth is needed (20*) In uttering “Bedrock is cold”, Fred has satisfied (what he
to state semantic injunctions: Knowledge of semantic injunctions does not takes to be) the semantic injunction associated with that utterance,
demand grasp of a concept of truth, but only of such concepts as that of an viz., (2*).
utterance and of normative notions that are central to practical reason in
general. It is for this reason that Soames regards possession of the concept Now (21) follows, for Fred has satisfied the injunction if, and only if,
of truth as unnecessary for competence. But there is a subtle illusion here: Bedrock is cold.
Even if the statement of a semantic injunction—one should utter S only if A speaker thus cannot make use of semantic injunctions in interpreting the
p—does not require use of a notion of truth, its application does, and there speech of others unless she has and deploys the concept of an utterance that
is reason to suppose that the statement of the injunction actually involves satisfies the appropriate semantic injunction. But such a concept just is a
such a notion, as well. The notion in question is, once again, that of an concept of an utterance’s being objectively correct. Even though Bedrock is
utterance’s being objectively correct. indeed cold, Fred’s uttering “Bedrock is cold” may have failed to satisfy all
Consider again Fred’s utterance that Bedrock is cold. How is Barney sorts of other conditions, conventional and otherwise, regarding when one
to reason to the conclusion that Bedrock is cold if Soames is right about is to utter it: Fred’s utterance may have been rude, irrelevant, or unjustified:
what Barney, qua competent speaker, knows about Fred’s utterance? If The semantic injunction is not that one is to utter S if and only if p but only
we replace (2), from our earlier discussion of such cases, with a semantic that one is to utter it only if p. That one has satisfied the semantic injunction
injunction, then we have the following: for the sentence one has uttered implies that one’s utterance is correct only
in a certain limited sense, namely, that it was objectively correct.
(2*) In the present context, one is to utter “Bedrock is cold” only Indeed, the concept of an utterance that satisfies its associated semantic
if Bedrock is cold. injunction simply is a concept of (literal) truth for utterances. Let us make
the following definition:
(18) Fred has uttered “Bedrock is cold”.
(19*) Fred knows (2*) and deployed this information is speaking. (22) N speaks the literal truth in uttering S ≡ d f N utters S and in
doing so satisfies the semantic injunction for S.
(20) Fred is speaking the literal truth.
(21) Bedrock is cold. Now suppose the semantic injunction for S is to utter it only if p. Then if N
does utter S, he will satisfy the injunction iff p. Hence:
But now (21) does not follow, since (2*) says nothing about truth. We need
(23) If the semantic injunction for S is that it should be uttered
to replace (??) with something like:39
only if p, then, if N utters S, N thereby speaks the literal truth iff
39 Exactly how (??*)—and the rest of this reasoning—should be formulated is presumably p.
open to discussion, and one might worry that the discussion to follow turns on special
features of how I have formulated it here. But it does not, as a little experimentation should Given the semantic injunction for a given sentence, we can thus derive a
show. For example, one might want to replace (??*) and (19*) with something like: Fred T-sentence for utterances of it from (22).
intends in uttering “Bedrock is cold” to utter it only if Bedrock is cold. Then (??*) would
become something like: Fred’s satisfied the relevant intention in uttering “Bedrock is cold”.
In a sense, then, semantic injunctions are equivalent to statements of
And now an utterance can be regarded as true if, and only if, the utterer satisfies the relevant truth-conditions: As we have just seen, statements of truth-conditions are
linguisitc intention in uttering it. easily derived from semantic injunctions. Conversely, semantic injunctions
6 Truth and Semantic Knowledge 23

are derivable from statements of truth-conditions. If S is true iff p, and cconventions that give our words meaning? I see no plausibility whatsoever
if one intends to speak the literal truth when uttering S, then one should to that claim, and the alternative mentioned for the case of irony does not
utter it only if p. One might suggest, then, that there is not much at issue exist here: It is no convention that Smith’s utterance means what it does in
between Soames and myself: We could debate whether semantic injunctions these circumstances.41
or truth-conditions are more fundamental, but what really hangs on who At best, then, there is conditional convention of literal truthfulness: If
wins this debate? In fact, however, Soames’s view faces a serious problem, one intends to speak the literal truth, then one should utter “Bedrock is cold”
one that is revealed by the need, two sentences back, to restrict attention to only if Bedrock is cold. One might not think the difference would matter.
utterances intended to be literally true. Such ‘literal injunctions’, the thought might be, state the literal meanings of
Understanding cannot consist in knowledge of semantic injunctions, the sentences they concern, and what matters here is that speakers should
because there are no semantic injunctions for English sentences. It just isn’t know the literal meanings of their sentences. But that is mistaken. The
a convention of English that one is to utter “Bedrock is cold” assertively literal injunction for “Bedrock is cold” is a hypothetical imperative: It tells
only if Bedrock is cold: There is no convention of truthfulness, in Lewis’s one when the sentence may be uttered if one intends to speak the literal
sense.40 One can speak in such a way that one’s audience will recognize truth. It does not tell one anything about utterances of sentences that are not
one’s communicative intentions only by recognizing one to have intended intended as utterances of literal truths. To know the literal injunction for
to speak literally falsely: Irony and hyperbole are like that. Perhaps more “Bedrock is cold” is thus at best to know what it means when it is uttered
commonly, one can utter a sentence intending to communicate something with the intent to speak the literal truth. But that is not to know what its
other than what one thereby literally says without caring whether one’s literal meaning is, since a sentence literally means what it does whether its
utterance is literally true: Metaphor is perhaps like that. Now, in so far utterer intends to speak the literal truth or not.
as irony, hyperbole, and metaphor are themselves conventional forms of It is tempting to reply that, while of course not all utterances are intended
communication, one might plausibly argue that the semantic injunctions for to be literally true, the account we are considering just involves a harmless
the associated sentences shouldn’t apply to them but rather other, related idealization. Of course, one might say, something needs to be said about
semantic injunctions: For example, one should utter “John is a fine friend” speech that is not intended to be literally true, but that account can be built
ironically only if John is not a fine friend(Lewis, 1985a, pp. XXX-XX). on top of an account of speech that is intended to be literally true. The fact
But the phenomenon to which I am drawing attention is not limited to such that literal meaning is more fundamental than non-literal meaning makes
cases, and the familiar example of the letter of recommendation can be made it legitimate to focus, in the first instance, on literal speech. But this reply
to illustrate this point. When Prof. Smith writes only that Jones has always equivocates. As just noted, the idealization consists in focusing on speech
been on time for tutorials, thereby implicating and so communicating that that is intended to be literally true and temporarily setting aside speech
Jones is a poor candidate for graduate school, it matters not at all whether that is not. To justify the idealization, one therefore needs to claim that
what Smith says is literally true. Perhaps not even that much is true of
41 Examples of so-called “loose talk” may also be relevant here: Utterances of “It is three
Jones, but Smith figured he had to say something nice and that was the
o’clock” often do not communicate that it is exactly three o’clock; but it is no violaton of
closest he could get. He doesn’t intend to communicate anything about
linguistic convention to have uttered a sentence that is literally false in order to communicate
Jones’s punctuality, and he knows his audience won’t care to know about something that is true. It is controversial, in fact, whether loose talk should be explained in
it. Are we really to believe that such an utterance violated the linguistic terms of implicature or rather in terms of pragmatic processes that affect what is said. See
[REF] for discussion. My own view, however, is that implicature has not yet been shown
40 Similar remarks would apply, mutatis mutandis, to any alleged convention of trust. inadequate to such cases.
6 Truth and Semantic Knowledge 24

literal speech is more fundamental than non-literal speech. And while I am speakers strive to utter sentences that are literally true in whatever language
certainly prepared to agree that literal meaning is more fundamental than they speak. Without that presumption, we cannot infer from the fact that
any kind of non-literal meaning—be it the sort that derives from implicature, speakers regard it as permissible to utter S only if p that S means that p in
metaphor, irony, or what have you—it simply does not follow that literal their language. And, it seems to me, this presumption should be rejected. To
speech is more fundamental than non-literal speech. I see no reason to reject it, however, one need not reject the core intuition that communication
believe this further claim. is possible only if one is entitled to rely upon one’s fellows. One need only
Lewis’s suggestion that there is a convention of truthfulness to which deny—to use Grice’s terminology—that the core intuition concerns what
language-users are party is a development of an intuition that many philoso- is said as opposed to what is communicated or meant. That is: Grant that
phers share, namely, the intuition that, in some sense, language could not be there is some sort of requirement that people speak truly. Now does the
used for communication unless one were entitled to rely upon one’s fellows. requirement concern what they say? or what they mean? The answer seems
It is important to understand that Lewis does not think it is a convention obvious: It concerns what is meant. There is no obvious reason that one
that one is to speak truly. Among the many objections Lewis considers in must, in general, be able to trust what one’s fellows strictly and literally say.
“Languages and Language” is that, by his own lights, part of what makes a All that is necessary is that one should, in general, be able to trust what they
regularity conventional is the existence of alternatives, and there is simply mean. Of course, what one says is often part of what one means, and so one
no alternative to speakers’ generally being truthful: If they weren’t, their would expect speakers to be literally truthful much of the time, but there is
words couldn’t have any meaning at all (Lewis, 1985a, p. XXX). Lewis no general requirement that speakers should be literally truthful all or even
grants the argument in its entirety. But what he is claiming is conventional most of the time.
is not speakers’ being truthful simpliciter: It is their being truthful in a
particular language L . It is not a convention that we should utter sentences The issue we have been discussing concerns explanatory priority: The
that are true in whatever language we happen to speak: That, Lewis thinks, question is whether literal injunctions or statements of truth-conditions are
has to be so if our language is to be meaningful at all.42 But if L is the more fundamental. Now, admittedly, I have not proven that one cannot
language we happen to speak, then, Lewis insists, it is a convention that regard literal injunctions as basic and somehow explain how they determine
we should utter a sentence assertively only if it is true in L : There is an the literal meaning even of utterances that are not intended as literal truths.
alternative to that possibility, since we might have spoken L 0 instead. But I do hope to have done enough to make it seem unlikley that this can
Lewis’s strategy is, in effect, to use the non-conventional regularity he be done. The prospects seem bright only so long as one neglects certain
believes must obtain if langauge is to be meaningful at all as a kind of bound- distinctions, such as the distinction between what is said and what is meant,
ary condition and to solve for an unknown, namely, the language we happen and the distinction between literal meaning and literal speech. Moreover,
to speak. We know that speakers must strive to utter truths in whatever the converse derivation, of literal injunctions from truth-conditions, just
language they speak: If we can figure out under what circumstances they seems like the right story: On that story, the literal injunction for a given
regard it as permissible to utter a sentence S, then we can figure out when utterance is a consequence of what the uttered sentence literally means and
they regard S as being true and so what they take it to mean and hence what conventions governing speech that is intended to be literally true; corre-
language they speak.43 Note, however, that the strategy presupposes that sponding injunctions for speech that is not (simply) intended to be literally
true are in turn consequences of what the literal meaning of the uttered
42 Note that Lewis’s view actually entails this claim. CONTINUE. sentence is and the facts, whatever they might be, about how non-literal
43 Davidson as well. Quine? Lots and lots of people, to be sure. meaning is determined by literal meaning and the environment in which the
7 Conclusion 25

utterance is made.44 compositionality is not, as it were, entirely a tacit phenomenon.46

If we can, then, we should distinguish facts about literal meaning from
facts about the norms that govern literal speech: It would be best if we could References
say what a sentence literally means without adverting to speech intended to
be literally true—or, for that matter, to any other specialized form of speech. Burge, T. (1993). ‘Content preservation’, Philosophical Review 102: 457–
That, of course, is precisely what the view I am defending allows us to do.45 88.

(1999). ‘Comprehension and interpretation’, in L. Hahn (ed.), The

7 Conclusion Philosophy of Donald Davidson. Chicago, Open Court, 229–250.

Chomsky, N. (1969). ‘Quine’s empirical assumptions’, in D. Davidson and

The view defended here has concerned only the semantic properties of sen- J. Hintikka (eds.), Words and Objections: Essays on the Philosophy of
tences. As noted at the outset, however, part of what makes truth-theoretic W.V. Quine. Dordrecht, Reidel, 53–68.
semantics attractive is that it promises to make a substantive explanation of
how the meanings of parts contribute to the meanings of wholes possible. Davidson, D. (1984a). ‘Belief and the basis of meaning’, in Davidson
For all that has been said here, however, natural languages might as well 1984b, 141–54.
not be compositional. As I said earlier, however, it would certainly be in
the spirit of this view to hold that what our conscious knowledge of the (1984b). Inquiries Into Truth and Interpretation. Oxford, Clarendon
truth-conditions of utterances is the product of our tacit knowledge of a Press.
truth-theory. But what is tacit is, by definition, unconscious: Our knowledge
(1984c). ‘Radical interpretation’, in Davidson 1984b, 125–139.
of the semantic properties of whole sentences could be generated from tacit
knowledge of the semantic properties of their parts even if we did not con- (1984d). ‘Reply to Foster’, in Davidson 1984b, 171–179.
sciously recognize sentences as having parts at all. But we do consciously
recognize sentences as having parts: Our knowledge regarding the meaning (1984e). ‘Theories of meaning and learnable languages’, in Davidson
of a sentence thus includes but is not limited to knowledge that it is true iff 1984b, 1–15.
p; somehow it also includes an appreciation of the sentence’s articulation
into parts and knowledge of what those parts mean. On the other hand, (1984f). ‘Thought and talk’, in Davidson 1984b, 155–70.
however, it cannot of course include knowledge of the sentence’s syntactic 46 My work on these issues has benefitted enormously from conversations over the years
structure, in anything like the sense in which a linguist would think of it. with Jim Higginbotham, Michael Rescorla, and Jason Stanley. Material from this paper was
And so one question that is left open here is how we should conceive our presented during a series of seminars I gave at the University of St Andrews in January and
February 2004 as a British Academy Visiting Professor and then again in a graduate seminar
conscious knowledge of meaning so as to take due notice of the fact that
at Harvard University in Spring 2004. Thanks to all who attended and participated for their
comments and encouragement, especially Jake Beck, Tony Corsentino, Bernard Nickel,
44 [[FIXME]] What I do NOT mean by that. Daniel Nolan, Agustin Rayo, and Crispin Wright. Thanks to the British Academy and to
45 Obviously, the truth-conditional aspect of my view is not what is responsible for this Arché, the AHRC Research Centre for the Philosophy of Logic, Language, Mathematics
benefit. What is responsible for it is, rather, the idea that competence rests upon knowledge and Mind, for their support, which is much appreciated, and to Crispin for arranging the
of meaning. visit.
7 Conclusion 26

(1984g). ‘Truth and meaning’, in Davidson 1984b, 17–36. (1964). The Basic Laws of Arithmetic: Exposition of the System, ed.
and tr. by M. Furth. Berkeley CA, University of California Press.
Dummett, M. (1978a). ‘The justification of deduction’, in Dummett 1978c,
290–318. (1970). Translations from the Philosophical Writings of Gottlob Frege,
Geach, P. and Black, M., eds. Oxford, Blackwell.
(1978b). ‘The significance of Quine’s indeterminacy thesis’, in Dum-
mett 1978c, 375–419. (1984). ‘On sense and meaning’, tr. by M. Black, in B. McGuiness
(ed.), Collected Papers on Mathematics, Logic, and Philosophy. Oxford,
(1978c). Truth and Other Enigmas. London, Duckworth.
Basil Blackwell, 157–77. Also in Frege 1997, 151–71.
(1991). The Logical Basis of Metaphysics. Cambridge MA, Harvard
(1997). The Frege Reader, Beaney, M., ed. Oxford, Blackwell.
University Press.
Grice, H. P. (1989a). ‘Meaning’, in Grice 1989b, 213–223.
(1993a). ‘Language and communication’, in Dummett 1993b, 166–
187. Grice, H. (1989b). Studies in the Ways of Words. Cambridge MA, Harvard
University Press.
(1993b). The Seas of Language. Oxford, Clarendon Press.
Heck, R. G. (1995). ‘The sense of communication’, Mind 104: 79–106.
(1993c). ‘The source of the concept of truth’, in Dummett 1993b,
188–201. (1998). ‘Grundgesetze der Arithmetik I §§29–32’, Notre Dame Journal
(1993d). ‘What do I know when I know a language?’, in Dummett of Formal Logic 38: 437–74.
1993b, 94–105. (1999). ‘Grundgesetze der Arithmetik I §10’, Philosophia Mathemat-
(1993e). ‘What is a theory of meaning? (I)’, in Dummett 1993b, 1–33. ica 7: 258–92.

(1993f). ‘What is a theory of meaning? (II)’, in Dummett 1993b, (2004). ‘What is compositionality?’, Manuscript.
34–93. (2005a). ‘Reason and language’, in C. MacDonald and G. MacDonald
Evans, G. (1985). ‘Understanding demonstratives’, in Collected Papers. (eds.), McDowell and His Critics. Oxford, Blackwells, 22–45.
Oxford, Clarendon Press, 291–321. (2005b). ‘Use and meaning’, in L. Hahn (ed.), The Philosophy of
Foster, J. (1976). ‘Meaning and truth-theory’, in G. Evans and J. McDow- Michael Dummett. Chicago, Open Court.
ell (eds.), Truth and Meaning: Essays in Semantics. Oxford, Oxford (2006). ‘Idiolects’, in A. Byrne and J. J. Thomson (eds.), Content and
University Press, 1–32. Modality. Oxford, Oxford University Press, 61–92.
Frege, G. (1962). Grundgesetze der Arithmetik. Hildesheim, Georg Olms (2007). ‘Frege and semantics’, Grazer Philosophische Studien 75:
Verlagsbuchhandlung. Translations of Part I are in Frege 1964, and 27–63.
of portions of Part III, in Frege 1970. Extracts appear in Frege 1997,
194–223. Higginbotham, J. (1991). ‘Truth and understanding’, Iyyun 40: 271–288.
7 Conclusion 27

Larson, R. and Segal, G. (1995). Knowledge of Meaning. Cambridge MA, (1992). ‘Truth, meaning, and understanding’, Philosophical Studies
MIT Press. 65: 17–35.

Lewis, D. (1985a). ‘Languages and language’, in Lewis 1985b. Strawson, P. F. (1971). ‘Meaning and truth’, in Logico-Linguistic Papers.
London, Methuen, 170–89.
(1985b). Philosophical Papers Volume 1. Oxford, Oxford University
Press. Tarski, A. (1958). ‘The concept of truth in formalized languages’, in
J. Corcoran (ed.), Logic, Semantics, and Metamathematics. Indianapolis,
(1985c). ‘Radical interpretation’, in Lewis 1985b, 108–18. Hackett, 152–278.

(1986). Convention. Cambridge MA, Harvard University Press.

McDowell, J. (1998a). ‘Anti-realism and the epistemology of understand-

ing’, in McDowell 1998c, 314–343.

(1998b). ‘In defense of modesty’, in McDowell 1998c, 87–107.

(1998c). Meaning, Knowledge, and Reality. Cambridge MA, Harvard

University Press.

(1998d). ‘On the sense and reference of a proper name’, in McDowell

1998c.

Quine, W. V. O. (1960). Word and Object. Cambridge MA, MIT Press.

(1972). ‘Methodological reflections on current linguistic theory’,

in D. Davidson and G. Harman (eds.), Semantics of Natural Language.
Dordrecht, Reidel, 442–454.

Richard, M. (1992). ‘Semantic competence and disquotational knowledge’,

Philosophical Studies 65: 37–52.

Rumfitt, I. (1995). ‘Truth-conditions and communication’, Mind 104: 827–

62.

Soames, S. (1988). ‘Semantics and semantic competence’, in S. Schiffer and

S. Stelle (eds.), Cognition and Representation. Boulder CO, Westview
Press, 185–207.