MIDWEST STUDIES IN PHILOSOPHY, XVI (1991)

What Metaphors Do Not Mean*

JOSEF STERN

T h a t aesthetics has become such a lively, integral part of the contemporary

philosophical scene can be traced to at least two factors. First, aestheticians have
succeeded in establishing strong connections between their particular concerns
and mainstream currents in philosophy-in epistemology. metaphysics, and
especially the philosophy of language. Second, they have nurtured a healthy
self-awareness in themselves that the problems of aesthetics are often too frag-
ile and subtle to be forced into the mold of one or another off-the-rack theory
imported from a core area of philosophy.' On the other hand, it is also becoming
increasingly better recognized that other areas of philosophy as well as aes-
thetics stand to gain from their interaction. Problems which, because of their
special character, were often relegated in the past to aesthetics-where they
were left either to lie fallow or to be cultivated within its narrow boundanes-
are now being rediscovered and newly appreciated for what they can tell us
about our general theories of language, knowledge, perception, and the like.
Those who come to these problems from this last perspective may have little
intrinsic interest in aesthetics. However, like exotic phenomena in the sciences,
these special problems take on a derivative but no less significant value because
of their implications for more central questions.
There is hardly a better example of this development than the problem
of metaphor which has been subject to profuse attention by philosophers of
all fields over the past dozen years. While aestheticians and literary theorists
have addressed the topic almost continuously since Aristotle (and continue
productively to do so), in recent years the topic has also attracted numerous
philosophers of language, epistemologists, and philosophically minded linguists
and cognitive scientists. In this essay, I shall examine one of these recent ap-
proaches to metaphor, both to show how current work in the theory of reference
and philosophy of language can inform our understanding of this traditional
problem in aesthetics and, as important, to demonstrate how sensitivity to an
apparently special and remote problem in aesthetics can have considerable
13
14 JOSEF STERN

implications for our general conception of semantic knowledge and the theory
of meaning.
My text for this lesson will be Donald Davidsons influential account
of metaphor, although I shall limit my discussion here to one small part of
his story. In his one essay specifically on metaphor, What Metaphors Mean,
Davidson makes two negative claims and one positive proposal.2 On the neg-
ative side, he denies (1) that metaphors have a metaphorical meaning in
addition to their literal meaning and, apart from the error of treating it as
meaning, he also denies (2) that a metaphor has a definite cognitive con-
tent that its author wishes to convey and that the interpreter must grasp if he
is to get the message (262). (3) On the positive side, he proposes, though
only in the sketchiest form, that metaphor is an imaginative use of language
whose intended effect is to make us notice likenesses. Of these theses, liter-
ally three-quarters of Davidsons main arguments in WMM are in support
of (2). In this essay, however, I shall put them aside and assume for the
sake of argument that metaphors can and often do convey or communicate
(to use neutral terms) some kind of cognitive ~ o n t e n t .Here
~ I shall address
only (I), Davidsons critique of metaphorical meaning-the thesis that the
content communicated by a metaphor is, or should be located in, its mean-
ing-and his complementary proposal (3) that metaphor is rather a matter of
use. In WMM Davidson explicitly presents both this critique and the positive
proposal only in a few bare sentences. However, his argument, which articu-
lates a widespread attitude toward the possibility of a semantics of metaphor,
should be understood and elaborated through his other writings on radical
interpretation and, in particular, one of his most recent essays, A Nice De-
rangement of epitaph^."^ So, while my aim here is not Davidsonian exegesis,
I will freely weave among these essays to construct the strongest case that
can be made.
Davidsons account is especially well suited to illustrate my opening
observation about aesthetics because Davidson brings to his discussion of
metaphor a well-developed theory of meaning which has a clear role in shaping
its contours; at the same time he shows considerable sensitivity to the special
nuances of the problem. However, I also have an ulterior motive for this ex-
tended critique of Davidson which is to motivate a rather different account
of metaphor-one that employs metaphorical meanings-that I have begun to
develop elsewhere.5 Because my own account and Davidsons start from such
similar positions, it is especially instructive to see where and why we diverge.
We both hold that the notion of truth (or reference, of which truth would be a
special case) should occupy a central place in a theory of meaning, and we both
also draw a sharp distinction between what words mean and what they are used
to do. We also agree that, in some sense yet to be explicated. essential to the
very notion of metaphor is that the metaphorical depends on the literal. And,
finally, we both begin from the observation that metaphorical interpretation is
highly contextdependent. However, whereas Davidson, along with most other
philosophers, draws from this the conclusion that metaphor must be a matter
 WHAT METAPHORS DO NOT MEAN 15

of use and not meaning, a subject for pragmatics rather than semantics, I argue
that we can give a semantics for metaphor precisely by embracing it as a kind
of context-dependent expression on the order of the demonstratives and index-
icals. In this essay, I shall not attempt to work out my alternative theory in
any detail, but at various points in my running critique and in the last section
I shall take the opportunity to indicate how the argument might lead to rather
different conclusions than Davidsons.
The structure of my argument will be as follows. In Section I, I begin
by explaining what exactly is at stake in Davidsons proposal that metaphor
should be explained as a use of words which possess only their literal meaning,
an explanation that rejects all notions of metaphorical meaning, including
a variant of speakers meaning for metaphor. In Section 11, I then turn to
Davidsons conception of literal meaning which plays perhaps the central role
in his account of metaphor as use and, in Section 111, to his argument why
what a metaphor communicates should not be considered a kind of metaphor-
ical meaning analogous to literal meaning. In the next two sections, I shall
then examine two different ways in which Davidson proposes to explain how
words are used to achieve their distinctively metaphorical effect through, or
depending on, their literal meaning. In Section IV, I explore the idea put
forth in WMM that utterance of the metaphor cuuses the effects distinctive of
metaphor, and in Section V, his suggestion in NDE that a metaphor succeeds in
communicating in the same way that a referential definite description succeeds
in referring, despite the literal falsity of each. With both of these proposals,
but especially the latter, I shall try to show how their specific weaknesses
reflect more general difficulties with the underlying conception of semantics
Davidson employs in his account of metaphor. In the concluding Section VI,
I shall briefly indicate the morals I would draw from these difficulties for the
alternative conception of semantics and meaning I employ in my own account
of metaphor.

Davidsons starting place is the distinction between what words mean and
what they are used to do (WMM 247). Writers on metaphor, he charges,
typically confuse the two: they take what metaphors are used to do-make us
recognize hitherto unnoticed aspects of things or surprising analogies and
similarities-and read these contents into the metaphor itself (WMM 261),
thereby converting them into their meaning. This is wrong because posit[ing]
metaphorical or figurative meanings, or special kinds of poetic or metaphorical
truth in order to explain how words work in metaphor is
like explaining why a pill puts you to sleep by saying it has a dormative
power. Literal meaning and literal truth conditions can be assigned to words
and sentences apart from particular contexts of use. This is why adverting to
them has genuine explanatory power. (WMM 247)
16 JOSEF STERN

That is, the appeal to metaphorical or figurative meanings is vacuous and,

therefore, bad explanation-for a reason somehow connected to their context-
dependence.
On the other hand, if we treat a metaphor as a kind of use of words,
Davidson says that we can explain how it brings offits particular effect in
terms of the imaginative employment of words and sentences [which] depends
entirely on the ordinary meanings of those words and hence on the ordinary
meanings of the sentences they comprise (WMM 245). In other words, here
Davidson seems to be saying that to the extent to which metaphorical interpre-
tation depends on meaning zt all, it depends entirely on the literal meaning of
the words used; however, like dream interpretation to which he also compares
metaphorical interpretation, the act of interpretation is itself a work of the
imagination, an act hardly guided by rules (ibid.).
Now, many philosophers have reacted to Davidsons bald denial of meta-
phorical meaning, and his complementary claim that a sentence used
metaphorically only means what its words literally mean, with a mixture of
bafflement and disbelief. After all, could Davidson himself really believe that
Romeo meuns and only means-in any sense of the term means-the literal
meaning of Juliet is the sun? There are two responses to this reaction, one
short, the other long. The short response is deflationary. Despite his deliberately
provocative way of putting it, Davidson is simply drawing a corollary of the
familiar distinction between what words mean (the subject matter of seman-
tics) and what they are used to do (the subject matter of pragmatics). Once we
conservatively restrict our use of the term meaning to the former and locate
metaphor in the latter, then what metaphors communicate, propositional or
not, will ips0 fact0 not be meaning.
Davidsons point is not, however, merely terminological, and that brings
us to the longer response. One could reply that we might still count the ef-
fect of a metaphor as a kind of meaning while maintaining a clear distinction
between semantics and pragmatics, by distinguishing between semantic or
word-meaning or sentence-meaning, on the one hand, and utterance-meaning
or speakers meaning, on the other, a distinction familiar enough from the
work of Grice and Searle. Davidson, however, rules this way out. In the
course of denying that metaphors convey any propositional content (beyond
their literal meaning), he asserts parenthetically: nor does [the] maker [of a
metaphor] say anything, in using the metaphor, beyond the literal (WMM
247). This parenthetic remark is not peculiar to Davidsons discussion of
metaphor. For in general he expresses doubt about the possibility of codi-
fying the .abilities and skills involved in so-called speakers or utterance
meaning in the form of principles (a la Searle) or maxims (a la Grice),
pre-established or conventional rules which would be either specific to lan-
guage or linguistic activities like conversation. Instead the kinds of inferences
and reasoning these activities employ require only, he says, the cleverness,
intuition, luck, and skill which are necessary for any rationd activity or
for devising a new theory in any field.6 What words are used to do should
 WHAT METAPHORS DO NOT MEAN 17

never, then, be considered meaning, even qualified by the terms speaker or

utterance.
This is not the place for a full-dress evaluation of the question whether
conversational implicatures and speakers meaning should be considered vari-
eties of meaning. However, I would like to offer some support for Davidsons
view that metaphor in particular, insofar as it is a matter of use, should also not
be subsumed under a category of speakers, or utterance, meaning. Consider
briefly Searles treatment of metaphor along just these lines.
Searle begins from the premise that our ability to understand a metaphor,
although not part of our semantic competence (which governs sentence mean-
ing), is nonetheless systematic rather than random or ad hoc. Therefore, it
must be governed by principles (of speakers utterance meaning) which
explain how speakers can say metaphorically S is P and mean S is R,
where P plainly does not mean R.* Searle further emphasizes that there is no
single principle that serves this purpose for all metaphors; in fact he proposes
eight different principles in his essay, with the confident proviso that even
they are not exhaustive. However, all these principles have the common func-
tion that they call R to mind given an utterance of P. Now, the question I
want to raise is whether these principles, exhaustive or not, justify thinking
of R as a meaning of the metaphorical use of P.
Two things are evident when we examine Searles eight principles. First,
the eight principles nicely show that the contents or features [ R ] that can serve
as possible metaphorical interpretationsof expressions [PIare as heterogeneous
a class as one might ever imagine. The Rs can range over features that are
either definitionally, or necessarily, (Principle 1) or contingently (Principle 2)
true of the Ps.But in general they need not be actually true or even believed to
be true of the P s. An R feature may only be culturally or naturally associated
with [PI in our minds (Principle 4) or the condition under which it holds may
somehow be like that of R (Principle 5). In other words, there is no one
interesting principle (e.g., resemblance) that might describe the contents of all
metaphorical interpretations; the best we can do is the uninteresting description
that the one, the expression P, calls the other [the feature R ] to mind, and
in order to make this relation more interesting, we can describe the range of
Rs, case by case, in more detail.
For this last purpose, Searles eight principles provide us with a helpful,
if rough. catalogue of what can serve as an interpretation of a metaphorically
used expression, the range of possible values of R . However-and this is the
second point-for the very same reason Searles eight principles provide no ex-
planation of how P imparts R . Indeed the descriptive strength of the principles
is their greatest explanatory weakness. For the very variety of the principles is
such that, taken conjointly, they place no restrictions on the class of possible
features that can enter into a given metaphorical interpretation. What one prin-
ciple rules out, another rules in. Furthermore, it is completely obscure what it
is for one thing to call another to mind. We simply do not know what kind
of psychological ability or complex of abilities this is; surely it is no better
18 JOSEF STERN

understood than the phenomenon of metaphor which it is meant to explain.

Searle himself comes close to acknowledging the explanatory poverty of his
account when he says that the question, How do metaphors work? is a bit
like the question, Howdoes one thing remind us of another thing? However,
whereas Searle goes on to say that the two questions are alike in that there is
no single answer to either (my emphasis), it would be more correct to say, as
philosophers since Hume have recognized, that what they have in common is
that we know no answer to either of them. Finally, it should be obvious that,
even with an explanation of this psychological phenomenon, not everything
that something calls to mind, or reminds us of, is something it means. Like-
wise for metaphor: even allowing that it involves calling a feature to mind, that
is surely not sufficient to count the feature a part of its meaning.
In sum, if we locate metaphor in the domain of use, Davidson is abso-
lutely right in my opinion that there is no more explanatory power to be gained
by thinking of the features communicated by a metaphor as its meuning-even
as its utterance or speakers meaning-than as effects which its use causes or
makes us see. In other words, why use meaning when all we really mean
is use?9
Davidson, of course, draws a stronger conclusion than the above condi-
tional. Because he also holds that metaphor should be entirely located in the
domain of use, he detaches the consequent and asserts that metaphors have no
meaning other than their literal meaning. However, one philosophers modus
ponens is anothers modus tollens. The conclusion I would draw from the
same argument is that if metaphors do have a meaning other than their lit-
eral one, then that meaning must be something different from the features they
are used to communicate on particular occasions of use. But what might such
metaphorical meaning be?
Rather than immediately answer that question, let me raise another: What
conditions should we require of any candidate in order to be countenanced as
meaning? Two necessary conditions-which prima facie exclude metaphorical
meaning-are already suggested in Davidsons remarks. First, like a theoretical
entity in a scientific theory, meanings are posits for which there is no justi-
fication unless they explain some regularity in our use of language. Second,
since it is words, or types, that have meaning, it must be possible to assign
the candidate independently of, or invariantly across, particular contexts of use,
i.e., to all tokens of one type.10 Now, the ordinary notion of literal meaning
does seem to pass these two tests: it attaches to types, and it is in virtue of the
fact that every token of a given type presumably has the same literal meaning
that we are able to explain why they all do some things the same-e.g.,
why they all assert the same content or have the same truth conditions-and
other things differently4.g.. their illocutionary or perlocutionary effects. For
analogous reasons, metaphorical meaning would seem to fail the test: it is
not a feature of the word that the word has prior to and independent of the
context of use and there are no analogous regularities it captures because each
metaphorical utterance says something different from every other one.
 WHAT METAPHORS DO NOT MEAN 19

Now, with this much by way of motivation of the point of Davidsons

distinction between what words mean and what they are used to do, the
question I shall explore for the remainder of this essay is whether Davidsons
way of drawing the distinction can carry the explanatory weight he intends it to
bear. In the next section I begin by examining his notion of what words mean,
or literal meaning, in light of his general conception of the theory of meaning,

I1

Davidsons thesis that a theory of meaning for a natural language is (in whole
or part) a theory of truth after the style of a Tarski truth definition is too well
known to need introduction. For our purposes, however, not this claim but
his more general conception of linguistic interpretation is more relevant. This
conception begins from the assumption that a theory of language is a theory of
linguistic acts, i.e., utterances, which should be explained as a species of rational
acts performed for ultimately non-linguistic purposes.*1 But not only must all
language be used for some such function. Although Davidson never explicitly
states this, he also works with the assumption that there is one function-
namely, communication-which is involved in all uses of language and in
whose absence a purported instance of language would only questionably be
a use of language. It is not clear whether Davidson holds this because he
believes that language is in its essence a tool of communication or because he is
concerned with language only insofar as it is an instrument of communication
but, in either case, this assumption lies behind most of Davidsons further
claims.
In particular, because the basic function of language is communication,
a minimum of two are necessary to perform a linguistic act: a speaker S and
hearer H or, as Davidson prefers, an interpreter I. When communication suc-
ceeds, Ss utterance is interpreted as he intends; i.e., I s understanding of the
utterance corresponds to S s intended understanding of his utterance. Hence,
nothing should be lost, Davidson argues, if we shift the explicit object of our
theorizing away from the speaker S to the interpreter I.I2 Although I s theory is
itself a theory of the linguistic behavior of S, our explicit theory is not directly
about Ss utterances but about what I must know (or his ability) that enables
him to interpret S as he (S) intends.*3
It lies beyond the scope of this essay to describe the many subtle and de-
tailed strokes with which Davidson fills in this picture of the theory of meaning
as a theory of the interpretation of communicative utterances. However, four
specific consequences that are especially pertinent to his characterization of
literal meaning should be mentioned. First, all and only communicative ut-
terances are the objects of such a theory of interpretation. That is, because
utterances of ungrammatical as well as grammatical sentences communicate
information, the theory must account for them all, including (as we discover in
NDE) malapropisms, slips of the tongue, and half-finishedsentence fragments.14
However, any utterance which does not have the function of communication-
20 JOSEF STERN

even if it is grammatically well formed-will not fall within the scope of such
a theory; indeed such an utterance may not truly be language for Davidson.
Second, because communication requires that S and I share 1 s interpre-
tation of S s words. any admissible theory of interpretation must be such that
it could plausibly be shared whenever there is communication.
Third, because communication occurs when I understands what S intends
to say, a theory of interpretation should be adequate (NDE444) to the kind
of interpretation required for understanding. These adequacy requirements are
of two sorts: substantive and formal. First, as a substantive condition which
follows from the holistic nature of linguistic understanding, Davidson requires
that the theory provide interpretations of aZl utterances of the speaker or com-
munity in question. Second, as a formal condition, he requires that the theory
represent the interpreters ability to interpret, or understand, a potentially in-
finite number of novel sentences and expressions in a systematic form that
acknowledges his finite capacity. In particular, Davidson proposes to meet this
condition by requiring that the interpretation of each utterance be a function of
the interpretations of its simple components drawn from a finite stock of basic
vocabulary, composed into more complex expressions by a finite stock of rules
of composition. Now, it is at this point that Davidson makes his well-known
claim that a Tarskian theory of truth is adequate to interpretation of this kind.
Details aside, I would emphasize two general points for our present purposes.
(i) The fundamental notion here is understanding, and corresponding to it, in-
terpretation, and it is only because the truth conditions yielded by a Tarski-like
theory serve as a measure of understanding that they then come to have the
central explanatory role in working out the theory.I5 (5) It is a single type of
semantic interpretation, e.g., truth conditions, which is taken to correspond to
the required kind of understanding, or interpretation, for all utterances in the
language. That is, Davidson leaves no allowance for partial degrees of under-
standing, or interpretation. This has the enormous advantage of simplicity, but
it is also a source of problems.
Fourth-and this is the conclusion for which Davidson argues in NDE-
any theory of interpretation that satisfies the previous conditions should not
attribute to the interpreter and speaker a common language in order to explain
how they communicate, at least if a language is anything like what many
philosophers and linguists have supposed, namely, a system of learned con-
ventions or regularities available in advance and independently of the occasion
of utterance. For any such theory must be plausibly shared by S and I when
they communicate, and no such plausibly shared and descriptively adequate
theory of this kind is, Davidson argues. what any philosopher or linguist would
consider a language. As an alternative, Davidson proposes a complex account
of communication in terms of speakers and interpreters intentions and mutual
beliefs. Each S and I brings to each occasion of utterance his own prior
theory of interpretation. S brings a theory of how he intends his words to be
interpreted, I brings a theory of how he believes S intends his (S s) words to
be interpreted.16 Unfortunately, however, S S and 1 s prior theories do not in
 WHAT METAPHORS DO NOT MEAN 21

general fully match as they initially stand. Therefore, S and I revise, or mutu-
ally adjust, their respective prior theories in order to achieve the fully shared
understanding, or interpretation, of the utterance that constitutes communica-
tion. The resulting theories that make communication between S and I possible,
Davidson calls their passing theories. But passing theories, Davidson empha-
sizes, are much too utterance-, occasion-, interpreter-, and speaker-specific to
eount as languages. Indeed each passing theory of interpretation is a theory
only of a feature of words and sentences as uttered by a particular speaker on
a particular occasion, i.e., of tokens (utterances) rather than types (sentences
or words).17 Hence, what accounts for communication is the sharing of these
utterance-specific passing theories rather than a shared language. And because
languages are posited only, or primarily, in order to explain how communi-
cation by speech is possible, Davidson concludes, there is no such thing as
a language, not if a language is anything like what many philosophers and
linguists have supposed (NDE, 446).
Now, a number of philosophers have found Davidsons conclusion
downright astonishing, an iconoclastic reversal undoing many of his best-
known claims. Ian Hacking, for example, asks: Is there no longer language
for there to be philosophy of?...[IfJ True-in-L is at the heart of Davidsons
philosophy, [wlhat is left, is there no such thing as an L?Js My own view
is that Davidsons claim is of a piece with the rest of his philosophy of lan-
guage, indeed a natural conclusion for him to draw on the basis of his other
claims about radical interpretation. However, in order to make clear what he is
claiming, it is important to distinguish language as a domain of investigation
from language as an explanatory notion. Davidsons provocative conclusion
is not meant to deny the existence of language or languages like English and
L as domains of investigation, as objects about which we can theorize. He
only wants to deny that there is a (philosophically interesting) notion of lan-
guage which can serve as a means of explanation, as a notion with explanatory
power. And as a denial of language as a notion with explanatory power, his
negative conclusion is a straightforward corollary of the rest of his theory of
meaning. For if a theory of interpretation describes the speakers and inter-
preters shared knowledge or ability that is adequate to [the] interpretation
(NDE 444) of each and every communicative utterance; and an interpretation
of an utterance corresponds to its full understanding; and its full understand-
ing consists in knowledge of its truth-conditions; then any such theory must
inevitably be sensitive to its context for each utterance. That is, each inter-
pretive theory must be attuned to the individual utterance of an individual
speaker addressed to an individual interpreter as interpreted in an individual
context. And there is no reason to think that such an interpretive ability is
governed by rules in the strict, nomic sense required by a language, rather than
by rules of thumb-rough maxims and methodological generalities supple-
mented by a healthy dose of wit, luck, and wisdom (NDE 446). So, if one
focuses only on this kind of interpretation, then, given Davidsons premises,
there seems to me to be nothing at all astonishing about his conclusion that
22 JOSEF STERN

it is not in virtue of knowing a language that we are able to arrive at such

interpretations.
However, what is striking and perhaps surprising-though not necessarily
astonishing-is that, even after having rejected notions like language and after
having endorsed his heavily context-oriented theory of interpretation, Davidson
nonetheless insists on drawing a sharp distinction between what words mean, or
their literal meaning, and what they are used to do. Why? What function does
this distinction serve in Davidsons language-liberated theory of interpretation?
And, furthermore, how in fact is Davidson able to explicate this distinction
without falling back, at least tacitly, on what he thinks are bogus explanatory
notions like that of language?
First, the point of the distinction. Lets assume that .a theory of intefpre-
tation is a theory of what (the interpreter believes that) the speaker intends his
words to mean. The first methodological difficulty an interpreter encountering
a speaker will face is that he is not given words to interpret but utterances, and
the meaning of an utterance does not deliver itself already factored into the
respective contributions of the words themselves as distinct from their use to
mean much else. Moreover, any utterance (like any human act) can be inter-
preted, or (as Davidson would say) described, as intended for indefinitely many
ends. If we can interpret the utterance yoreid sheleg as an act of saying that
it is snowing, then we can also interpret it as an act advising someone to wear
his boots and overcoat, or warning him to drive carefully, or proposing that we
go skiing. In order for any sort of theorizing about language to be possible, the
first task for the interpreter must be, then, to distinguish among, or systematize,
these various kinds of intentional descriptions of an utterance-beginning with
the distinction between the meanings of the words themselves and their use.
Or to put the point a bit differently: No theory of interpretation can be a theory
of everything a concrete utterance means. However, if the theory is only
of the meanings of the words used (as distinct from what they can be used to
do), then the interpreter already needs a theory in order to identify, or abstract
out, those meanings from all the further intentions of the concrete utterance
he is given in experience. This is the point of the distinction between what
words mean and what they are used to do. Yet how can Davidson draw this
distinction without letting in the very pseudo-explanatory terms like language
which he wants to avoid?
Suppose that an utterance is a kind of rational action whose interpretations
are its speakers various communicative and non-communicative intentions and
purposes. Davidson proposes that one way to identify the meanings of the
words is to order these interpretations or intentional descriptions of the utterance
according to their relations to each other as means to ends.9 Thus my utterance
of yoreid sheleg to my son is a means to saying that it is snowing, the saying
of which is a means to encouraging him to believe that we will go sledding
tomorrow, which is a means to keeping him from nagging me all night, which
is a means to preserving my sanity. The first intention in order in this series-to
say such-and-such-Davidson labels the first meaning of the utterance, which
 WHAT METAPHORS DO NOT MEAN 23

in turn is used to achieve the second and later intentions in the ordering (all of
which I shall include under the heading of secondary intentions). This notion
of first meaning, he now proposes, can serve as an explication of, or at least as
a preliminary stab at characterizing (NDE 434), the ordinary notion of literal
meaning, what words mean as opposed to what they are used to do:m
Now, the first meaning of an utterance may indeed be the speakers first
means toward achieving his ultimate (non-linguistic) purpose or intention in
performing the utterance. But why, because and only because it is first in the
order of means, should it be singled out from the other intentions as a special
kind of meaning-namely, the literal?
Note for a start that Davidsons denial of the explanatory role of language
in communication rules out the possibility of connecting first meaning and
literal, or word, meaning via the notion of linguistic meaning. What makes
first meaning literal meaning cannot be-as its etymology meaning according
to the letter might suggest-that it is the first meaning of an utterance because
it is directly encoded in the form of the words as determined by the rules
of language (and not merely conveyed by their use). According to Davidson,
there are no such rules of language.
On Davidsons own picture of communication, it may also seem unclear
how first meaning can do the work of literal meaning. For the explanatory power
of literal meaning is precisely due to its context-independence (Literal meaning
and literal truth-conditions can be assigned to words and sentences apart from
particular contexts of use ... [WMM 247]), while first meaning, as we have
seen, is context-dependent to the extreme, a feature of words and sentences as
uttered by a particular speaker on a particular occasion. This characteristic of
first meaning is, furthermore, no mere artifactual detail of Davidsons theory.
Because the theory of interpretation must yield interpretations that correspond
to the understanding of utterances-an understanding that consists in knowledge
of truth conditions-first meaning must be sensitive to all kinds of contextual
parameters for each utterance.
I will leave it to Davidson exegetes to patch up the apparent inconsis-
tency between these two passages. Davidsons actual views apart, we should
distinguish two kinds of context-dependence in these cases. The first, which I
shall call pre-semantic context-dependence, is the context-dependence of our
assignments of meanings (whatever ones theory of meaning ultimately is; for
now, following Davidson, assume it is truth conditions) to sounds or shapes (or
perhaps words, considered just as syntactic entities). We are all familiar with
the task, on hearing some concrete sound pattern, of determining the semantic
description it should be assigned or with which it should be typed. I hear the
sound pattern ei: even knowing that the speaker is speaking English, I must
decide whether what I heard was the first person indexical I or the common
noun eye or the affirmative aye or the groan ai. In making this judgment,
it is obvious that we rely on all sorts of contextual cues-the appropriateness of
the alternative types within the immediate sentence and then within the larger
discourse, our beliefs about the speaker and his intentions. and so on. Now,
24 JOSEF STERN

on Davidsons account of communication, this task is more general and more

radical. On every occasion of utterance, the interpreter is faced with the task
of assigning each sound pattern not only one of an alternative number of types
within a language but rather a language, or a theory of first meaning, as a
whole. The interpreter does not, of coune, approach this task on each occasion
from scratch; he brings to it his prior theory. However, it is at best a matter of
chance, or else providence, that his prior theory turns out to fit the speakers
intention. Each new, previously unknown proper or common name, as well as
deviations from previous word patterns and malapropisms, is a potential source
of misfit that requires revision leading to a different passing theory. Moreover,
in each new context, or even at each turn of the discourse, this task begins
anew; thus a passing theory is not specific even to speaker, like an idiolect,
but to a speaker at a moment, an idiolect-slice. Of course, in constructing his
theory which makes these assignments of interpretations to sounds, the inter-
preter will utilize every available contextual cue or piece of information. So,
in this pre-semantic sense, each assignment of a first meaning to an utterance
is highly context-sensitive and utterance-specific.
This pre-semantic context dependence which applies to all first mean-
ings of utterances must be distinguished from a second sort of post-semantic
context-dependence. Having assigned an utterance a first meaning, that utter-
ance may then be used for an indefinite number of extra-linguistic ulterior
purposes or intentions. Which of these further intentions is communicated will,
of course, depend on the context-the speakers and interpreters mutual be-
liefs, intentions, and expectations. Yet, whichever of these secondary intentions
is attributed to the speaker, the first meaning of his utterance remains the same;
the same communicative interpretation serves indifferently as the first means
for all these ulterior purposes. In that sense, we might say that the first meaning
of an utterance is the meaning which it has on all of its uses to communicate
further non-linguistic intentions or, more accurately, regardless of how it is
so used.21
Given these two different ways in which context bears on the interpreta-
tion of an utterance, it should now be clear that first meaning is pre-semantically
context-dependent but post-semantically context-independent. Context func-
tions in assigning a first meaning to a given utterance but, once assigned,
that first meaning remains invariant regardless of the further extra-linguistic
purposes or intentions for which it is used. In this sense, the first meaning
of an utterance is autonomous of its extra-linguistic purposes. And for this
reason, in turn, first meaning is an interesting candidate for literal meaning
which, in one sense of the term, is what the words themselves mean no mat-
ter how they are used. That is, a criterion of literal meaning is post-semantic
context-independence.
In sum, even if we do not accept all the details of Davidsons critique
of language as an explanatory notion or his positive account of communi-
cation in terms of prior and passing theories, what he does clearly show is
that notions like language, on the one hand, and full (including pre-semantic)
 WHAT METAPHORS DO NOT MEAN 25

context-independence,on the other, are unnecessary in order to settle on a sig-

nificant notion of literal meaning, what words mean as opposed to what they are
used to d0.n However, what Davidsons own account of literal meaning does
assume is that a theory of interpretation is a theory of utterances whose func-
tion is communication and whose aim is understanding, understanding which,
in turn, is cashed out in terms of knowledge of truth-conditions. The question
we must now address is whether this notion of literal, or word, meaning-
based on the idea that a theory of meaning is a theory of interpretation, or
understandingdan bear the weight of Davidsons employment of it, first, to
exclude metaphorical meaning and, second, to explain how utterances are used
metaphorically.

Davidsons objection to the very idea of metaphorical meaning is that, un-

like literal meaning, or what a word means, the metaphorical interpretation,
or intention, of an utterance is never prior to and independent of its con-
text of use. In light of what we have now said about context-dependence,
this objection cannot be that the metaphorical interpretation is pre-semantically
context-dependent because so is all first or literal meaning. Rather it must be
that the metaphorical interpretation is also posf-semantically context-dependent.
That is, what an utterance communicates qua metaphor is not autonomous of
its extra-linguistic secondary intention or ulterior purpose and, therefore, what
it communicates as a metaphor cannot be a kind of meaning.
But what is the ulterior purpose, or secondary intention, of a metaphor?
As we said earlier, Davidson charges that others start from the trite and true
observation that a metaphor makes us attend to some likeness (WMM 247)
and then falsely read the likeness into the utterance as its meaning. For David-
son, on the other hand, this effect-i.e., the resemblance or feature we are made
to notice-simply is the ulterior purpose of the utterance. That is, metaphor is
just one among many uses of language to obtain special effects-like assertion,
hinting, lying, promising or criticking (WMM 259). But just as no one would
suggest that these other uses introduce meanings in addition to the literal/first
meanings of their words, so there is no additional kind of meaning we should
associate with the metaphorical utterance beyond the 1iteraVfirst meaning of
its words. Whatever explanatory work might be done by such a metaphori-
cal meaning is already done by the combination of the literal meanings of
the words uttered and their extra-linguistic secondary intention as metaphor to
make us notice a certain likeness.
Take Romeos utterance of Juliet is the sun. Romeo wants to tell us
that and why he loves, even worships, Juliet. He does this by telling us how
she nourishes him with her caring attention, how he cannot live without her.
But this he does, not by expressing those thoughts literally (either because he
doesnt have or know the right words, or because he thinks it will be more
effective this other way) but by making us notice them, or by inviting us to
26 JOSEF STERN
discover them, in turn, by likening Juliet to the sun. And this he does, not by
literally comparing Juliet to the sun or by visually displaying Juliet and the
actual sun, but by uttering the (false) sentence Juliet is the sun. Of course,
in order for this utterance to make us consider features which Juliet and the
sun share, Juliet must mean Juliet and the sun must mean the sun-i.e., the
utterance must have its literal/first meaning. Q.E.D., Davidson would argue,
we have fully explained Romeos utterance while taking into account only the
literal/first meanings of the words uttered and the special metaphorical ulterior
purpose or intention of the speaker in performing the utterance, namely, to
bring features of likeness to our attention. Given this explanation, it would be
vacuous to posit in addition yet another kind of metaphorical meaning.
Now, this argument is certainly right up to a point. If the (first) meaning
of an utterance must be autonomous of its extra-linguistic purpose or use, and
if the purpose of a metaphor (or, more precisely, the metaphorical purpose of
an utterance) is to make us notice a likeness, then the meaning of the metaphor
cannot itself consist of one of those features of likeness. This Davidsons argu-
ment convincingly shows. But, as we hinted earlier, that still leaves open the
possibility that the metaphorical meaning might consist of something else. Let
me put this argument in somewhat different terms, terms in which Davidson
himself would never put it. Suppose metaphors are truth-valued, or express
propositions or content, and, in particular, that Romeos utterance u of Juliet
is the sun is true iff Juliet is P, where P is a feature u makes us notice in
virtue of which Juliet resembles the sun. What Davidsons argument shows
is that P cannot be part of the meaning of u - e v e n though it is a part of its
truth condition or propositional content. That is, he shows that the meaning of
a metaphor must always be distinguished from its truth conditions or proposi-
tional content. However, the argument obviously does not show that no other
condition or rule could be the metaphorical meaning of the utterance.
Metaphors are, moreover, not unique insofar as their meaning (if they
have one) must be distinguished from their truth conditions or content. Despite
Davidsons tendency to conflate, or interchange, literal meaning and literal
truth conditions (as in the passage quoted in Section I), there is a whole
class of expressions (commonly taken to be literally used) of just this type
whose meaning and truth conditions must be distinguished, although they are
systematically related. This is the class of demonstratives and indexicals whose
meaning (as I mentioned at the start of this essay) I have elsewhere proposed
to take as a model, or paradigm, for the meaning of a metaphor.24 Thus a
demonstrative (word-type) like I has a literal meaning it possesses prior to
and independent of any context in which it (or one of its tokens) is used,
but that meaning is never itself part of its truth-condition in any context, a
truth-condition which, furthermore, will vary systematically from context to
context depending on who is speaking. Therefore, given only knowledge of the
meaning of the indexical, we are still not in a position to predict (or explain)
its content or truth condition in any context-despite the systematic connection
between the two. Now, i f we could find regularities and systematic variations
 WHAT METAPHORS DO NOT MEAN 27

in the context-dependence of metaphors like those of I,, we might also, then,

motivate an analogous distinction within their interpretation between their
meaning and what they say (truth conditions or contents on each occasion).
So, to return to our previous question, What might a metaphorical meaning
be? As we have already seen, it cannot be the feature or content expressed
by the metaphor in any particular context. Instead I propose taking it to be
the conditions that determine the particular features that can enter into what
is said by, the content, or truth conditions of metaphorical utterances in their
respective contexts. And because these conditions would remain constant from
context to context, they would also constitute a kind of metaphorical meaning
that belongs to semantics rather than pragmatics.
In order to get a better sense of this alternative notion of meaning exem-
plified by the meaning of demonstratives, I want to look more closely now at
the least explicit part of the story I rehearsed a few paragraphs back, Davidsons
account of how we can explain how a metaphor works simply in terms of its
first/literal meaning and its metaphorical purpose. The specific part I shall focus
on is that in which the utterance of the (literally meant) sentence Juliet is the
sun is said to make us notice some (interesting or pertinent) feature whereby
Juliet resembles the sun, the feature in virtue of which the metaphor can be
said to have worked. In language more familiar in the metaphor literature,
this part of Davidsons account corresponds to the claim that the metaphorical
depends on the literal, a formula which for Davidson would be more accu-
rately stated as the claim that the metaphorical use of a sentence depends (to
the extent to which it depends on meaning at all) on its literal meaning. In
his various writings, Davidson provides two models for this dependence.= In
WMM, he suggests that metaphor is a matter of using literal words to cause a
certain effect that consists in our noticing the likenesses that make the utterance
work (or successful) as a metaphor. In NDE, he suggests that the way in which
we manage to say something true by metaphorically using a sentence whose
literal meaning is false (or, more precisely, whose literal meaning is not that
which in the context of utterance is relevant to judging whether the utterance
is true) is directly analogous to Donnellans explanation of how we can refer
to an individual with a definite description which does not (literally) denote
him. In the next two sections I shall discuss these two proposals.

IV
Davidson introduces his causal account of metaphor in the course of com-
paring metaphors to jokes and dreams, i.e., to kinds of works performed or
produced with language or imagery.*6 Putting aside dreams for now, the com-
parison to the joke-as well as to the riddle-is a deep one which goes back
as far as Aristotle and deserves to be explored in its own right. The one ques-
tion that concerns us here is whether metaphors are like jokes specifically with
respect to the issue of meaning vs. use. On one negative point, the answer is
clearly Yes. Jokes make you laugh, metaphors make you see likenesses. But no
28 JOSEF STERN

one would posit a joker meaning in jokes (in addition to their literal mean-
ing) to explain why they make you laugh, similarly, we should not be tempted
to posit a metaphorical meaning in metaphors to explain their effect. Rather
joke or dream or metaphor can, like a picture or a bump on the head, make us
appreciate some fact-but not by standing for, or expressing, the fact (WMM
262). That is, at least like a bump, if a metaphor makes us appreciate some fact,
it is by causing us to recognize it rather than by expressing it as its meaning
or as its propositional content. Just as an apple dropped on his head reputedly
caused Newton to think of the First Law of Gravity, so Romeos (literally
interpreted) utterance of Juliet is the sun-a spatio-temporal event-causes
him and us to see that Juliet is like the sun in that we cannot live without
either of them. And just as the apple neither itself meant nor served as a reason
for believing the First Law of Gravity, so Romeos utterance neither expresses
that resemblance as its meaning nor furnishes a reason for believing that Juliet
resembles the sun in that respect.
Davidson uses a slew of expressions in WMM to refer to the relation
between the utterance of the metaphor and its effect-e.g., makes us see,
alerts, inspires, leads us to notice, prompts, draws our attention to,
provokes. On the one hand, the differences among these may make one
wonder whether there is any one relation at work here. On the other, the
epistemic character of all these expressions makes it clear that Davidsons
causal account should be distinguished from another tradition in the literature
on metaphor that also gives central place to its causal power. This is the view,
first suggested by Max Black and subsequently developed by many others, that
(at least vital) metaphors create similarities rather than express or formulate
some similarity antecedently existing.*zs Now, when Black originally proposed
this, his formulation left room for a weaker epistemic reading of the claim-
that metaphors create similarities only in the sense that they reveal hitherto
unnoticed similarities. In subsequent writing. however, he has emphasized that
what he really did mean was the stronger, and more controversial, ontological
thesis that metaphors actually cause (to exist) the aspects of reality they also
enable us to see. Be that as it may, the relation Davidson has in mind is, in
contrast, entirely epistemic: the utterance of the metaphor makes or causes us to
notice or recognize the resemblances that enter into its effect, or interpretation;
it reveals rather than constitutes them.
However, if this is so, then a question of the opposite kind arises: Has it
in fact been shown that the sense in which a metaphor reveals, or makes us
recognize, a resemblance is causal at all? There is at least one other possible
explanation which I would propose: the sense in which the utterance of a
metaphor makes its interpreter recognize or notice a specific similarity is
analogous to the sense in which a speaker makes a presupposition when
he asserts a proposition that requires that presupposition in order to render it
appropriate in its context (e.g., when I tell you that John was recently divorced,
I make the presupposition, known or unknown to you previously, that he was
married until then).*9 Similarly, we make the interpreter of an utterance u
 WHAT METAPHORS DO NOT MEAN 29

recognize the similarities or features F that are required for an appropriate

metaphorical interpretation of u when we utter u in a context in which it is taken
to be interpreted metaphorically.30 But this sense in which an assertion makes
a presupposition, or the utterance of the metaphor makes us recognize a
similarity, is not causal, at least if a causal relation requires that the cause
be distinct from its effect. Here the similarity we are made to notice, like a
presupposition made by an utterance, is not the content of a distinct mental act
(presupposing, noticing) which could be separated by an act or will from the
overt linguistic behavior, the ~tterance.~ In short, then, while Davidson does
show that the metaphorically related resemblance or feature is not the meaning
of the metaphor but what the utterance makes us see. he has yet to show that
-
here the relevant sense of make is a causal nation at all.
There is another significant difference between Davidsons account and
most other theories of metaphor like Blacks, a difference which is also rel-
evant to Davidsons idea that a metaphor is to be explained in terms of the
effects it causes. Most philosophers like Black are concerned with metaphors
only insofar as they occur in full statements or utterances. However, within
the sentence uttered (say, Juliet is the sun) they distinguish (to use Blacks
terminology) the focus (is the sun), i.e., the constituent expression(s) used or
interpreted metaphorically in the sentence, and the frame (Juliet), i.e., the re-
mainder of the sentence which is used or interpreted literally. That is, it is really
a sub-sentential expression (simple or complex) that undergoes the metaphori-
cal interpretation even though it has that interpretation only when it occurs in
the larger frame or context, which is minimally a full sentence. Davidson, on
the other hand, can draw no such distinction between focus (the constituent ex-
pression that undergoes the metaphorical interpretation) and frame (the context
of the interpretation). Because he acknowledges no metaphorical meaning that
would attach to one rather than another constituent of the sentence or utter-
ance, there is no prima facie reason to make any such sub-sentential expression
the metaphor. All there is is whatever makes us notice the metaphorical
feature or resemblance-and that is the utterance in its totality; obviously no
component like Juliet or the sun taken individually could cause us to see a
resemblance between the two of them. So, for Davidson the minimal linguis-
tic unit which constitutes the metaphor- what does the alerting, inspiring,
prompting, etc.-is always the full sentence or string uttered in context.
There is, then, more to Davidsons comparison of metaphors to pictures
and jokes than perhaps first meets the eye. Like a joke whose effect is a function
of its entire telling, or story, and cannot be reduced to, or identified with, any
of its isolable parts, like its punchline, it is the total utterance in its context
which achieves the metaphorical effect for Davidson. And like pictures which
do not, in any well-defined sense, have a syntax?* metaphors for Davidson
have no internal metaphorical structure beyond their literal structure and, as
metaphors, they also make no use of their literal syntactic or semantic structure
in achieving their effect. To be sure, Davidson repeatedly says that the sentence
being used metaphorically has its literal meaning, but all he in fact turns out
30 JOSEF STERN

to mean by the literal meaning of the sentence are the separate and separable
literal meanings and referents of its individual terms. AS with T.S. Eliots
poem me Hippopotamus, yet another work to which Davidson compares
metaphor, the way in which metaphors alert us to aspects of the world by
inviting us to make comparisons is simply by the simultaneous presentation
or display-the brute juxtaposition, as it were-of the words of the utterance
with their respective literal meanings and referents-regardless, say, of their
subject/predicate structure. So, while Davidson claims that the metaphorical
depends on the literal, much of what is usually included in our notion of the
literal meaning of a sentence does no work in, and is even excluded from,
Davidsons account of this dependence-relation.
Furthermore, this limited characterization of literal meaning raises a de-
scriptive problem for Davidsons account. For if it is simply the literal mean-
ings, or referents, of the words used which make us notice the metaphorical
resemblance, it is not clear why sentences which differ only in their respective
subject-predicate structures, such as A man is a wolf and A wolf is a man (or
the ungrammatical strings A man-a wolf or A wolf-a man) should have sys-
tematically different metaphorical effects; ignoring their structural differences,
they all invite us to make comparisons between the same things.33
But apart from this descriptive problem, there is a deeper difficulty with
Davidsons causal account. Even if we agree with Davidson that there is a
causal relation at work in metaphor, the question remains: What kind of causal
explanation can be given of how the utterance of the literal sentence is related
to the specific metaphorically related feature we subsequently notice or see?
And, in particular, will such a causal explanation explain how the metaphorical
effect (or interpretation) of the utterance depends on its literal meaning? What
I shall now argue is that to the extent to which Davidson tells us what such a
causal explanation of the metaphorical effect would be, there is no reason to
think that it would (and perhaps good reason to think that it would not) include
an account of this metaphoricaliliteral dependence. But any account that leaves
out this essential feature of metaphor, I would argue (and Davidson seems to
concur), is inadequate as an explanation of metaphor.%
Before showing how this objection applies to Davidsons rather subtle
account, I want to show how the same difficulty arises on a more extreme
version of his position which has been put forth by Richard Rorty who invokes
Davidsons theory of metaphor as part of a larger project of his own whose aim
is to displace the cognitive from the dominating position in science, language,
art, and ethics that the tradition has assigned it.35
Rorty appeals to Davidson because, unlike most everyone else in the tra-
dition who explains how metaphors are indispensable36for scientific, moral,
and intellectual progress by forcing them into a cognitivist mold in which they
express a special kind of meaning or a sui generis content, Davidson offers
him a fully naturalistic yet noncognitive alternative: he lets us see metaphors
on the model of unfamiliar events in the natural world-causes of changing
beliefs and desires-rather than on the model of representations of unfamiliar
 WHAT METAPHORS DO NOT MEAN 31

worlds.37 In order to expose utterances of metaphors as nothing more than

non-cognitive causes-mere stimuli, mere evocations (291)-Rorty, how-
ever, goes considerably further than does Davidson himself: he divests them
of their status, not only as bearers of special metaphorical meaning, but also
as uses of language and even as intentional, rational actions. Adapting a fig-
ure of Quines, he begins by identifying the realm of meaning, the proper
domain of semantics, as a relatively small cleared area4efined by the
regular, predictable, linguistic behavior which constitutes the literal use of
language-within the jungle of use. Hence, metaphor which is, Rorty claims,
unpredictable and irregular by its very nature falls outside the realm of meaning
and within the surrounding jungle. Now, up to this point, Rortys exposition
is more or less faithful to Davidson; however, their respective jungles of use
are quite different locales in which to place metaphor and it is instructive to
see how they differ.
Rortys jungle of use includes, to begin with, not only utterances of
words, but also unfamiliar bird-calls, thunderclaps, and indeed the whole spec-
trum of sounds found in nature. What all these sounds have in common-and
apparently makes them a jungle-is that they are all just noise: sounds for
which we have no prior theory with which we can make sense of them but
which, because of their very strangeness and anomalousness, have considerable
effect on us. Metaphors, then, are also just a kind of more or less exotic noise,
although they stand a tat closer than, say, bird-calls to the cleared space of
literal meaning because diachronically they can d i e - o r , as Rorty puts it more
dramatically, be killed off-and made part of stale, familiar, unparadoxical,
and platitudinous literal language.38 However, as long as they are genuinely
alive, metaphors are just jungle noises which, given their irregularity and un-
predictability, cannot be understood or interpreted except in the way that
we come to understand anomalous natural phenomena, namely, by revising
our antecedent theories to fit them.39 In a similar way, Rorty acknowledges
that metaphors may be responsible for a lot of cognition, but only in the
same non-intentional causal sense of responsible in which the same can be
said about anomalous non-linguistic phenomena like platypuses and pulsars.
In short, our ability to understand a metaphor does not fall under the rubric of
mastery of language (whatever ones theory of linguistic competence may be)
and how the metaphor causes its effects also does not involve any of its prop-
erties as language or even as a species of intentional action. Like a thunderclap
or birdsong, the utterance of a metaphor is nothing but a non-intentional event
for all of its causal functioning.
Davidsons jungle of use is quite different. As we saw earlier, he also
begins with use, i.e., with actual utterances or linguistic acts, whose pre-
theoretical meaning might also be described as a jungle but only in the
sense that in that state it is dense and undifferentiated and, only after con-
siderable theory-laden chopping, pruning, and trimming, can be systematically
differentiated into the meanings of its words (their literal or first meaning) and
the various secondary meanings which constitute all the other things the words
32 JOSEF STERN

can be used to do. Furthermore, for Davidson, unlike Rorty, what remains when
we have cleared away the space of literal meaning, however unruly, wild, and
mysterious that surrounding jungle may be, is still intentional, rational linguis-
tic action. Some of the secondary extra-linguistic purposes of these other uses
of words may be causal effects of their utterance-as with metaphor-but the
utterances are none the worse as rational, intentional human acts because they
function as causes. For Davidson, indeed, it is never noise, i.e., non-linguistic,
non-intentional events, that falls in our domain of investigation, but intentional
human action of which linguistic action is one case. And it is only within that
corpus that we distinguish what the words used themselves mean, the subject
of semantics, and the many things they are used to do or mean, including (as
in metaphor) those things that their use causes us to learn or recognize. -
In sum, there is all the difference between Rortys and Davidsons respec-
tive positions. Yet, despite their differences, Rortys presentation drives home
more clearly than Davidsons own presentation of his position one important
consequence of his causal explanation of metaphor. Although Davidson is
right that the fact that the utterance of a metaphor does its work causally is not
incompatible with it being an intentional (linguistic) action, what will actually
figure in a causal explanation of how the metaphor works will contain little
if anything of its character as a use of language (or as an intentional act). To
the extent to which its causal explanation and the underlying causal laws in its
explanation all treat the metaphor simply as a physical event, Rorty is there-
fore correct to view metaphors just as noise, i.e., non-intentional, non-linguistic
events. Of course, metaphors are not just noise. Their interpretation depends
critically on the literal meanings of the words that are used metaphorically-a
dependence that is not simply causally necessary-and metaphors are governed
by many of the structural constraints, both syntactic and semantic, that govern
so-called literal language (as I shall illustrate in Section VI). Hence, Rorty is
wrong to describe metaphors solely as noise even if it is also possible to ex-
plain them fully as noise. However, what is right about his description is that
a Davidsonian account cannot claim both to explain how a metaphor works by
means of a causal story and that what the metaphor causes depends entirely on
the ordinary [i.e., literal] meanings of those words. For the causal explanation
of a metaphor will treat it just as a non-intentional, non-linguistic event-as
a meaningless noise.
Recall that, on Davidsons own account of causal explanation, for one
event to be said to cause another it is only necessary that the events be nomo-
logically related under some description of each.40 The true singular causal
statement need not itself instantiate the law or refer to the events in question
under their nomologically appropriate description. Hence, we can agree with
Davidson that the metaphor-utterance event causes the resemblance-noticing
event (even allowing for the anomalism of the mental) insofar as there are
grounds to believe that there exists some nomological relation between the two
under some description of each-presumably physical (or neurophysiological)
descriptions of the events.
 WHAT METAPHORS DO NOT MEAN 33

But this, as I have already said, is not enough for Davidson. For his
claim is not simply that there is a causal relation between the metaphorical
utterance and what it makes us notice-under some appropriate description
of each. Davidsons thesis is that we can explain how the causal effect of
the metaphor is brought off by the imaginative employment of words and
sentences [which] depends entirely on the ordinary meanings of those words
and hence on the ordinary meanings of the sentences they comprise (WMM
247, my emphasis). Therefore, to produce such an explanation, its vocabulary
must itself make reference to words, use, and literal (or first) meaning. Of
course, Davidson need not himself actually produce such a causal explanation
of metaphor, but he also does not give us any reason to believe that any such
law is in the offing.
Let me emphasize that I am not objecting to the ultimate possibility
of causal explanations of metaphor or, for that matter, of any linguistic phe-
nomenon in terms of the physical or neurophysiological properties of utterances
and their effects. Such an explanation would, in principle, treat the utterance
of a metaphor and its effect no differently than the auditory perception of a
thunderclap or a birdsong and their effects on us. What I am objecting to is the
claim that we can explain how a metaphor works both as a use of language
which depends entirely on the literal meanings of its words and in terms of its
causal relation to the effect of its utterance. We may be able to make the sin-
gular causal judgment that Romeos utterance caused us to see how we cannot
live without Juliet just as we cannot live without the sun, but as soon as we try
to provide the explicit explanation of that relation, we lose the substance of
Davidsons claim that it is as a use of language that metaphor does its work.
Of course, at this point Davidson might reply that we cannot really explain
metaphor under its description as a use of language which depends entirely
on its literal meaning; we can tell a story about each such case (as we did in
Section 111) but that should not be confused with a nomic explanation. On the
contrary, under such a description, metaphor is explanatorily anomalous. But if
this is what Davidson really holds, he should say so; for his disagreement with
those who hold that we can explain metaphor by metaphorical meanings is not,
then, with the explanation of how metaphor works its wonders (WMM 247,
my emphasis) but rather with the possibility of an explanation of metaphor as
a linguistic act, period. In order to do what he claims to do, Davidson must
produce a causal explanation of metaphor as a kind of use of the sentence with
its literal meaning in which the explanans itself makes use of literal meanings,
words, and so on. But then metaphors would not just be noises or bumps on
the head-unless those also bear meanings in the

V
The second kind of use of language in terms of which Davidson attempts to
explicate how the metaphorical depends on the literal is exemplified by Donnel-
Ians idea of a referential definite de~cription.~*
In this use of a description-in
34 JOSEF STERN

contrast to its attributive use-the speaker, Davidson claims, succeeds in com-

municating his intended meaning (or reference) despite the fact that it differs
from the literal meaning (or reference) of the words used, which is nonetheless
the meaning with which they are in fact used on that occasion. Hence, refer-
ential and attributive descriptions are examples for Davidson of two kinds of
uses of language which differ in what they ~ay~3-and therefore in their re-
spective truth-values-without their (first) meanings differing. This difference
in use Davidson contrasts with the case of malapropisms (and the introduction
of new proper and common names) where the meaning of the word used itself
changes in its context of utterance. And metaphor, Davidson adds in passing
(NDE 440),should be analyzed like a referential definite description, as a use of
language, rather than on the model of a malapropism which involves a chahge
of meaning.&
Although there are various objections one might raise both against David-
sons presentation of Donnellans idea of a referential description and against
his claim that malapropisms involve a change of meaning, I shall grant him
both of these for the sake of argumentP5 What I want to challenge is his further
claim that metaphor should be explained in the same way as a referential def-
inite description, i.e., that it is just a use of language which has, or retains,
its (prior) literal meaning while it is used otherwise.
Davidson illustrates the referential use of a definite description by the fol-
lowing case, adapted from Donnellan. A speaker Jones has in mind a particular
person-call him Max-whom he believes to have murdered Smith and whom
he therefore believes is uniquely designated by the description, Smiths mur-
derer. Seeing Maxs behavior in the courtroom dock, Jones exclaims: Smiths
murderer is insane. Unfortunately, however, Joness belief that Max is Smiths
murderer is (unbeknownst to him) false. Nonetheless, where Jones uses the de-
scription Smiths murderer referentially-i.e., simply as a means or tool to
refer to, or single out, Max and not with the intention to describe the indi-
vidual to whom it refers-Donnellan argues and Davidson agrees that he says
something true with the sentence Smiths murderer is insane provided Max
is insane.
Here there is certainly no reason to think that the referential definite de-
scription has undergone a change of meaning just because Jones is able to
say something true about Max by uttering Smiths murderer is insane even
though Max does not satisfy the content of the description. On the contrary,
it is because Jones (falsely, as it turns out) believes, or represents himselfas
believing, that Max is Smiths murderer, in the standard, prior literal meaning
ofthe words Smiths murderer, that he is able to use that description to re-
fer successfully to Max. And it is either because his interpreter also (falsely)
believes that or because he recognizes that Jones (falsely) believes that of
Max (in the standard, prior literal meaning of the words) that he interprets
Joness utterance of Smiths murderer is insane to be saying something true
about Max. Were we to inform Jones that Max is not in fact Smiths murderer
(in the standard, prior literal meaning of the words), he would not respond
 WHAT METAPHORS DO NOT MEAN 35

that we have failed to understand the meaning of his use of that descrip-
tion on that occasion, or that he intends for it to have a new passing first
meaning in that context which does in fact uniquely designate Max. Instead
he would say-given his intention to be using the description referentially-
that whats important is that he meant to say something true about Max
even though he was wrong in believing that he was Smiths murderer (in
its passing literal meaning) and, therefore, wrong (or at least misleading) to
use that description. In this case, then, Davidson seems to be absolutely right
that the reference is none the less achieved by way of the normal mean-
ings of the words and the words therefore must have their usual reference
(NDE 43940).
With a malapropism, in contrast, Davidson claims that the word itself
acquires as its (passing first) meaning what the speaker intends to say in its
context of utterance. For example, when Yogi Berra reputedly thanked the
crowd in Yankee Stadium on Yogi Berra Day for making this day necessary,
possible is what the word necessary meant on that occasion and not merely
what Yogi meant by using that word (in its ordinary literal meaning).a Here is
how Davidson would describe this situation. Yogis interpreter brings a prior
theory of Yogis linguistic behavior to the occasion of utterance, a theory in
which necessary means necessary. Hearing the absurdity or inappropriateness
of Yogis utterance with that prior first meaning, he knows that Yogi must
have intended to say something else. At this point his interpreter has at least
two options. Either he can keep its prior first meaning as its passing first
meaning and add an ulterior purpose for the utterance-say, take Yogi to be
speaking ironically or comically-or he can adjust his prior theory of Yogis
first meanings in order to give the word necessary as its passing first meaning
what he believes Yogi intends to be saying on that occasion, namely, possible.
Now, Davidson does not tell us how Yogis interpreter chooses to interpret his
utterance in one rather than another of these ways. However, if he decides that
what Yogi intends to be saying is that he wants to thank everyone for making
this day possible, then, Davidson would argue, he must be taking possible as
the passing first meaning of necessary and not simply as something imparted
by Yogis use of necessary in its prior first meaning. For, unlike the case of
the referential description where there are grounds for saying that the speaker
understands his words in their prior first meaning (namely, the fact that he
believes, albeit falsely, that Max is Smiths murderer, in the prior first meaning
of these words), here there is no such reason to think that Yogi understands by
necessary necessary, that he intends to say that he wants to thank everyone
for making the day necessary in its prior first meaning. On the contrary, we
would explain why he said what he did by saying that he intended to thank
everyone for making the day possible-and he believed, or represented himself
as believing, that on this occasion necessaryjust (first) means possible. To be
sure, Yogis interpreter can figure this out only because he knows that in Yogis
prior theory necessary means necessary and possible means possible, and
he conjectures that Yogi is confusing the two (or that the two come to be
36 JOSEF STERN

confused in his mental lexicon because of their semantic proximity to each

other).47 However, the fact that the interpreter knows the one interpretation by
way of the other does not show that the latter belongs to his passing theory
and should be prior to Yogis intended interpretation of the word in the passing
theory itself. For the passing theory is a theory of what the speaker intends on
the occasion of utterance. And Yogi had no such intention-to mean necessary
by necessary-on that occasion.
In sum, lets grant Davidson both these views: that referential definite
descriptions are an instance of speakers using their words to mean or do
things other than what the words themselves mean and that malapropisms are
an instance of words with one prior first meaning acquiring a new passing
first meaning on their occasion of utterance. Now, how should we describe
metaphor?
Although Davidson gives no explicit argument for his assertion that the
intended interpretation of a metaphor belongs, along with referential definite
descriptions, in the realm of use rather than, with malapropisms, in the realm
of meaning, presumably his argument would go like this: What a metaphor
makes us notice depends on its ordinary [=literal] meanings* (WMM 247);
therefore, an adequate account of metaphor must allow that the primary or
original [=literal] meanings of words remain active in their metaphorical set-
ting (WMM 249). Hence, a metaphor like Juliet is the sun must have
its ordinary, primary, or original meaning-i.e., its prior first meaning-as its
passing first meaning on its occasion of utterance. This the speaker, in turn,
uses in order to make us notice hitherto unnoticed resemblances or features,
his ulterior purpose for his metaphorical utterance.
But is the way in which the metaphor depends on its literal meaning
really like the way in which the referential definite description depends on
its prior first meaning? Exactly how do the primary or original meanings
of words remain active in their metaphorical setting [my emphasis]? And,
in particular, is it so clear that the dependence of the metaphorical on the
literal is a dependence within the passing theory of the utterance between
its secondary intention (the feature of resemblance it makes us notice) and its
first meaning, rather than a dependence that holds behveen the first meaning
of the metaphorical utterance in its passing theory and its first meaning in its
prior theory?
The answer to all these questions, I think, is that they have no satis-
factory answer with the conceptual resources of Davidsons theory. On the
one hand, with metaphorical utterances, unlike referential definite descriptions,
there are no analogous grounds for holding that their literal, or prior first, mean-
ing remains their first meaning in their passing theory. With referential definite
descriptions, recall that Jones must intend his words Smiths murderer to be
understood according to their literal, or prior first, meaning because he believes
that Max is Smiths murderer (in their ordinary literal meaning), a belief we
attribute to him in order to explain why he uses the (improper) description
Smiths murderer to refer to Max. But with a metaphor, there is no analogous
 WHAT METAPHORS DO NOT MEAN 37

reason for claiming that the speaker intends his words to be understood, or
interpreted, according to their literal, or prior first, meaning. Romeo, for exam-
ple, surely does not believe, or represent himself as believing, the absurd or at
least patently false proposition that Juliet is the sun-expressed by the literal
meanings of the words-when he utters Juliet is the sun. And because there
is no reason to hold that he believes that proposition, there are also no grounds
for claiming that he himself understands his words, or intends for his interpreter
to understand his words, with that prior first meaning on that occasion. Hence,
there are no grounds for saying that their prior first meaning should also be
considered their passing first meaning on occasions on which words are used,
or interpreted, metaph~rically.~g
On the other hand, there are also good reasons not to consign the lit-
eral meaning of words used metaphorically entirely to the prior theory of the
utterance. To do so would leave us without a clear candidate for the passing
first meaning of the metaphorical utterance inasmuch as the features which the
metaphor makes us notice cannot themselves be the first meaning of the utter-
ance (as we argued in Section 111). Moreover, as Davidson correctly observes,
the literal, or prior first, meaning of the words does, and must, remain active
in the metaphorical setting. That is, whatever we ultimately determine to be
the precise relation between the literal and the metaphorical, knowledge of the
literal (first) meaning of the words used is necessary on the occasion of utter-
ance itserf in order to determine their metaphorical interpretation, or what the
metaphor makes us notice. And the relevant literal meaning is not what the
literal meaning of the word was on some prior occasion of utterance, but what
it actually is on the very same occasion on which it is interpreted metaphor-
ically. Therefore, it would not be sufficient to explicate the way in which the
metaphorical depends on the literal simply as a kind of diachronic, or causal,
dependence that holds berween the first meaning of the metaphorical utterance
in its passing theory and what its first meaning had been in the prior theory.
Within the passing theory there must also be room for the literal, or prior first,
meaning of the utterance. The problem with Davidsons account is making
room for this in the passing theory itself. For if interpretations must capture
the speakers and interpreters understanding of the utterance, then where the
literal is made the first meaning of the passing theory of the metaphorical utter-
ance, we are forced to say that the speaker himself understands his utterance,
and intends his interpreter to understand his utterance, according to its literal
meaning. And that conclusion, I have argued, is false or, at least, groundless
in the case of metaphor.
Put a bit differently, I am arguing that the words as used, or interpreted,
metaphorically do have their literal, or first, meaning as part of their pass-
ing and not merely their prior theory of interpretation-because their literal
meaning is necessary for determining their metaphorical interpretation. But the
appropriate sense of having their literal meaning is not one that is or can be
expressed with the conceptual apparatus of Davidsons theory. For the words
do not have their literal meaning in the sense that they are, or are intended to
38 JOSEF STERN

be, understood literally. Rather we need another sense of having their literal
meaning which is different from that which corresponds to our understanding
of the utterance and, in turn, is cashed out by truth conditions. In the next and
last section of this essay, I will turn to what that other sense might be, but it
should be clear that Davidson himself cannot introduce any such alternative
without seriously revising his central views about language as an instrument
of communication whose interpretation is what we understand by an utterance,
namely, its truth conditions. Therefore, what we need is really an alternative
theory of meaning.

VI

As I have hinted more than once already, for this alternative conception of
meaning we should look in the direction of the demonstratives and indexicals.
One crucial difference between the meaning of demonstratives and indexicals
and the notion of meaning on which Davidsons theory focuses are their respec-
tive levels of interpretation. Davidson, largely because of his conception of
language as a means of communication, takes meaning to be the understanding
shared by speaker and interpreter when they communicate which, in turn, he
cashes out in terns of truth conditions or content. Now the truth conditions, or
contents, of most words remain invariant from utterance to utterance; therefore,
if one concentrates on such eternal, i.e., context-independent, expressions, it
is understandable why he would take their meaning to be an object of under-
standing. However, demonstratives and indexicals are an exception to this rule.
Take Louis XIVs utterance Lktat cest moi which is true if and only if the
state is Louis XIV.49 This truth condition should be the first meaning of the
utterance according to Davidsons account: in order to communicate both the
speaker and interpreter must take it to be the speakers intention, and this com-
municative intention of the utterance remains the same regardless of whether
Louis utters the sentence in order to announce whom he believes he really is
or to warn the masses not to mess with him or to impress them with his wit
or to say something memorable for the history books, i.e., regardless of his
further secondary intentions.
Yet, this truth condition obviously does not remain the same for all ut-
terances of the sentence Lbtat cest moi. If Richard Nixon were to utter that
same sentence, its truth condition, and first meaning, would be that the state is
Richard Nixon. Now, focusing on the demonstrative moi or I as it occurs
in this sentence, how can Davidson describe its context-dependence within his
theory?
The context-dependence of I is neither pre-semantic nor post-semantic.
It is not post-semantic because what varies with context in this case is not an
extra-linguistic intention or ulterior purpose which is performed by the utter-
ance, but rather the communicative intention of the utterance, its truth condition
or content, what must be understood by both speaker and interpreter for there
to be communication. Nor is the context-dependence of I pre-semantic, the
 WHAT METAPHORS DO NOT MEAN 39

way in which (first) meaning or a lexical description or a type is assigned to a

(semantically uninterpreted) sound, shape, or word depending on various con-
textual cues and presuppositions; here the sound ai has already been assigned
the lexical type of the first person demonstrative I and it is only because it
has been assigned that type that it is context-dependent in this additional way.
In addition to pre- and post-semantic context-dependence, let us now
call this third kind of context-dependence of demonstratives semantic context-
dependence. Here the context functions not to pair a meaning or type with
a shape but, where the type associated with the meaning described above is
already assigned to the shape, to determine a truth condition on a particu-
lar occasion of utterance. As we have just seen, semantic context-dependence
cannot be reduced to either pre- or post-semantic context-dependence. And,
inasmuch as Davidsons understanding-oriented notion of interpretation and,
in particular, of first meaning is itself cashed out in terms of truth conditions,
the rule associated with I also cannot itself be either first meaning or a gen-
eralization of this notion. For suppose that we were to try to abstract out its
invariant or constant meaning from among the first meanings of all utterances
of tokens of 1. Such invariantfirst meaning might, then, be taken to be a
better candidate for an explication of literal meaning even if the first meaning
of any particular utterance will not do. Nonetheless it will still not be good
enough, for the rule of meaning for I (that the truth condition or content of
each of its tokens is its actual speaker) will not be included in its invariant first
meaning. That rule only constrains the possible first meanings of any utterance
of I but never is itself part of its first meaning on any particular occasion, let
alone, part of its invariant first meaning in all of its utterances. Yet this rule
is exactly what we would take as the meaning-indeed, literal meaning-of
the word I, what the speaker knows in virtue of his linguistic competence. In
Davidsons scheme, however, such a rule just does not fit in; at best it might
be considered zero meaning.
Demonstratives, then, point us to a general conceptual difficulty with
Davidsons account of meaning. For insofar as the meaning of a demonstrative
diverges systematically from its truth conditions, there is reason to think that
the proper level at which to aim ones theory of meaning should not be the
level of understanding furnished by truth conditions which is necessary for
communication but at a more abstract level of knowledge.
This more abstract level of knowledge is not, however, unrelated to the
kind of understanding required for communication even though it cannot be
identified with it. Indeed one important fact which any theory of this more
abstract knowledge should explain is what we might call the autonomy of
meaning: the fact that what expressions (literally) meanis independent, or
autonomous, of what individual speakers intend to communicate on given oc-
casions. Not only dont speakers intentions determine what their expressions
(literally) mean; more important, which intentions can be communicated by
which specific words is determined, or constrained, by their (literal) meaning.
Take I once again. Suppose I, Josef Stern, believe (falsely) that I am Louis
40 JOSEF STERN

XIV. Suppose that I intend to refer to Louis XIV by my utterance of moi in

Lttat cest moi. That is, I intend to say that the state is Louis XIV. Suppose I
give my interpreter sufficient cues that this is my intention-that I believe that
I am Louis XIV and that I intend to be referring to Louis with my uses of the
first-person demonstrative. And suppose that in point of fact Louis XIV was
the state (whatever that would be). Nonetheless, regardless of my intention and
regardless of the fact that my interpreter knows my intention, my utterance of
the sentence Lttat cest moi will be false. For the meaning of the indexical
moi, or I, willy-nilly determines that my utterance on that occasion says
that the state is Josef Stern, not Louis XIV-and Josef Stern, as I well know,
is no state (whatever that would be).
Indexicals are one case, in the realm of lexical meaning, where the-au-
tonomy of meaning reveals itself. Another important but non-lexical class of
cases of autonomous meaning involves the meaning camed by grammatical
configurations.5 Consider, for example, the fact that (i)
(i) He saw John in the mirror
cannot be understood as meaning (ii)
(ii) John saw himself in the mirror.
For reasons of space, I shall not review the detailed explanation for this fact
about meaning, but the general idea is that what prevents the anaphoric pronoun
in (i) from taking the noun John as its antecedent is a formal condition on
an abstract representation of the string. However, this condition, like the rule
of meaning for I,, constrains the interpretation of the anaphoric construction
without itself being part of that interpretation. The condition does not state what
the anaphoric interpretation is or rnusr be but what it cannor be; it only excludes
certain interpretations. Thus, this notion of meaning, like that of demonstratives,
does not enter into the content of what speakers understand by an utterance
but at most circumscribes the bounds of their understanding by ruling out the
impossible interpretations.
Now, these examples illustrate the fact that one function of meaning
is to constrain speakers intentions to say certain things with certain linguistic
expressions or structures. In order to explain this fact I want to argue in turn that
a notion of meaning one level more abstract than truth conditions is necessary.
But why cant we account for this fact in terms of truth conditions, the kind
of understanding involved in communication? Why not say, as a Davidsonian
explanation might propose, that speakers intend to say only what they believe
their interpreters will understand or only what they expect others will reasonably
be able to interpret them as saying? Just as we do not usually intend to do
things we believe nature will prevent us from doing, so a speakers knowledge
of how other people can be expected to behave as interpreters-in light of his
knowledge of their previous linguistic experience-constrains his intentions in
choosing particular forms to communicate specific contents.52
The difficulty in applying this kind of explanation to the facts we have
described, the kinds of facts which exemplify the autonomy of meaning, is
 WHAT METAPHORS DO NOT MEAN 41

that it is simply implausible given the kinds of rules which seem necessary to
account for those facts. Those rules, in short, are much more general, fixed, and
abstract than the sort of context-specific expectations and beliefs that govern the
assignment of first meanings according to Davidsons picture of interpretation.
Take the demonstrative I again. Neither is its content, or truth-conditional
factor, in any particular context itself its meaning nor is its meaning a direct
generalization, in any obvious way, from its contents in particular contexts.
Rather its meaning is an instance of a still more general schema that holds for
all expressions in the class of demonstratives (or, slightly more specifically,
indexicals, e.g., 1, now, here, actually) in all contexts in which they are
used. Furthermore, the meaning expresses a condition which applies not to the
concrete utterance but to a much more abstract representation: a-representation
of the utterance, or sentence uttered, which both abstracts away from partic-
ulars of its contents in actual contexts and which contains abstract features
not explicitly represented in the concrete expression. Thus the meaning of a
demonstrative like I must mark the fact that its interpretation (content) is
always fixed relative to its context of utterance, regardless of its scope in the
sentence and regardless of whether the matrix sentence also contains another
expression, or operator, which makes the truth (as opposed to content) of the
sentence dependent on circumstances other than that of its context of utter-
ance. For example, if I say I might have been assassinated my utterance is
true in my context of utterance if there is some possible circumstance (relative
to the circumstance of the context of utterance) in which I, Josef Stern, the
individual actually speaking in the context, is assassinated; it is not enough
in order for it to be true that there be some possible circumstance (relative
to the context) in which whoever might be speaking in that circumstance is
assassinated. Finally, it should be noted again that this condition on the in-
terpretation of the demonstrative does not itself yield its interpretation in any
context but only constrains its possible interpretations by excluding certain al-
ternatives. So, however speakers come to learn these rules or conditions of the
meaning of I, if the autonomy of meaning is to be explained ii la Davidson
in terms of a speakers expectations about his interpreter, those expectations
will necessarily require the attribution of a knowledge of rules, or conditions,
of corresponding generality, systematicity, and abstractness. But knowledge of
this kind would be much more fixed, invariant, and inflexible than Davidsons
highly context-specific characterization of passing theories would suggest.53
What moral do I wish to draw from this story? Davidsons argument
moves from the premise that a theory of linguistic or interpretive competence
must be a theory of speakers and interpreters shared understanding of all
communicative utterances to the conclusion that such a theory cannot be a
theory of a common language, or grammar, in virtue of which they are able
to communicate. However, the same argument might equally well be taken to
show that if a speakers linguistic competence does consist in knowledge of a
language, or grammar, then a theory of that competence will not be a theory of a
speakers and interpreters ability to interpret, or understand, all communicative
42 JOSEF STERN

utterances, i.e., a theory which accounts for the truth conditions of all utterances
in their respective contexts. This is not to deny that understanding an utterance
may be a matter of knowing its truth conditions. Rather I am challenging
the assumption that a theory of an interpreters linguistic competence should
directly and fully explain his understanding of utterances. Instead, I would
argue, ones linguistic competence is only one factor that contributes to such
understanding. That is, a theory of a speakers knowledge of language is never a
theory of understanding tout court, but only of one kind of knowledge specific to
the linguistic properties of utterances that only partially determines everything
that goes into ~nderstanding.5~
With this background, let me finally return to metaphor. Here, too, I want
to propose that a theory of a speakers semantic competence in metaphor, a
theory of metaphorical meaning, should not be assumed to be a theory of his
ability to use, or to communicate with, metaphors, a theory of what an in-
terpreter understands when he understands a metaphor, a theory which would
itself specify for each utterance of a metaphor the feature or resemblance it ex-
presses on that occasion. Rather, from among the various competences, skills,
and faculties that conjointly account for this complex ability (including the
interpreters extra-linguistic knowledge of the relevant contextual parameter),
a semantic theory of metaphorical interpretation should address only the in-
terpreters knowledge of meaning specific to metaphorical interpretation. To be
sure, much of the evidence for any such theory will consist of actual utterances
with their full metaphorical interpretations in their respective contexts, but
the semantic theory should account not for that evidence per se-the utterances
with their contents in their respective contexts-but only for the one kind of
knowledge underlying the speakers ability to assign such interpretations that
is specific to their linguistic, or semantic, properties. It follows, then, that the
semantic theory of metaphor will leave untouched a whole range of aspects of
metaphor that depend on a larger model of speech performance or use-e.g.,
criteria of appropriateness, the psychological processing of metaphors, and the
various rhetorical and non-cognitive effects of metaphors that bear on their
success and belong to their meaning in a broad sense of the term. But all this
and more: for if the ability to interpret a metaphor even in the limited sense of
assigning it a propositional content, or truth condition, in a context involves the
contributions of multiple abilities, skills, and presuppositions which are non-
or extra-linguistic, such as the ability to judge similarities or recognize salient
features or make presuppositions of various kinds, then the semantic the-
ory by itself will never suffice to determine even the interpretation, i.e., truth
condition or propositional content, of a metaphorical utterance.55
What work, then, does the meaning of a metaphor perform? And with
what aspects of metaphorical interpretation should a semantic theory of met-
aphorical meaning be concerned? In order to factor out the component of the
speakers interpretive ability that belongs specifically to his semantic com-
petence in metaphor, we might begin by identifying those facts about the
interpretation of a metaphor that would prima facie be a function specifically
 WHAT METAPHORS DO NOT MEAN 43

of its meaning. In particular, if meaning in general, as we have been arguing,

serves to provide formal, or structural, consfrainrs on interpretations-on what
speakers can intend to say given particular forms of expression-then for evi-
dence of metaphorical meaning, we should look for signs of formal constraints
on possible metaphorical interpretations. In other words, if there exists an in-
teresting notion of metaphorical meaning, then, as with demonstratives and
our earlier example of anaphora, there should exist a level of interpretation
for metaphor which is also autonomous of a speakers intentions and serves
to constrain how those intentions-or, in the case of a metaphor, the features
of likeness we are made to see by its utterance-enter into its interpreta-
tion, i.e., content or truth condition. As with demonstratives and the anaphoric
constructions, the autonomous aspects of its interpretation woufd be one kind
of data about metaphor that positing a specific level of metaphorical meaning
would serve to explain.
To give you a glimpse of these facts, let me conclude this essay with
two brief examples of constraints on metaphorical interpretation, examples that
correspond to the two examples we have already given of the autonomy of
meaning of literal, or non-metaphorical, use of language.
1. It has been frequently observed that in cases of verbphrase anaphora in
which the antecedent verbphrase allows multiple interpretations, the antecedent
verbphrase and the anaphor must always be given the same interpretation. For
example, may can be interpreted either with the sense of permission or with
that of possibility in
(1) John may leave tomorrow.
But in (2) and (3)
(2) John may leave tomorrow, and the same is true of Harry.
(3) John may leave tomorrow, and Harry, roo.
both (italicized) constituents must be given the same interpretation, either that
of possibility or that of permission. Now, this fact would seem to be a clear
instance of the autonomy of meaning: the speakers communicative intentions
are constrained by a structural condition on the copying of the interpretation
of the antecedent verbphrase onto the anaphor.56 And because this constraint
has obvious consequences for the truth-value of the sentence, it is a matter of
semantic significance or meaning. There is, furthermore, no interesting way in
which this condition on verbphrase anaphora can be explained as a condition
on use, or defined directly in terms of an interpreters intentions or expectations
h la Davidson. Whatever I intend to say, and whatever I can do to make my
audience interpret me as I intend, I simply cannot mean by (2) or (3) that John
has permission to leave tomorrow, and it is possible for Harry to leave then. (2)
and (3) are each always only two-ways ambiguous while, if it were a matter of
an individuals intentions and mutual expectations, in at least in some contexts
it ought to be possible for them each to be four-ways ambiguous. Hence, this is
a constraint on interpretation that would appear to call for semantic structure.
44 JOSEF STERN

Now, similar facts obtain with metaphor. Contrast the interpretation (i.e.,
content or truth condition) of the sun in the literal sentence
(4) The largest blob of gases in the solar system is the sun.
with its metaphorical interpretation in
( 5 ) Achilles is the sun
or in
(6) Juliet is the sun
Each of these, I will assume, has a different metaphorical interpretation, al-
though exactly how they differ will not matter for our purposes. But consider
now the following (semantically) ill-formed sentences which are examples,
again, of verbphrase anaphora:
(7)*The largest blob of gases in the solar system is the sun, and Juliet/
Achilles is, too.
(8) */? Juliet is the sun, and Achilles is, too.
In each of these sentences, what is ill-formed would seem to be the result
of violating the same kind of interpretive-constraint at work in (2)-(3). That is,
because the interpretation of the antecedent is copied onto the anaphor, both
the antecedent and the anaphor must have the same interpretation (whatever
that interpretation is). Hence, in (7)where the interpretation of the antecedent
is the literal meaning of the sun and the interpretation of the anaphor would
seem to be metaphorical (on pain of absurdity), we have one violation. In
(8) (which is slightly more acceptable to informants), we have two different
metaphorical interpretations of the sun, also violating the constraint.57 Both
interpretations are, then, ill-formed-although, as is often the case with such
figures, we tend to impose an interpretation on the strings despite the violation.
However, it is precisely the feeling of play or pun that accompanies such
imposed interpretations that reveals the underlying semantical ill-fomedness
of the strings.
These are examples in which certain aspects of metaphorical interpre-
tation show themselves to be autonomous of-precisely because they serve
as constraints on-interpreters intentions. And as with the earlier examples,
it is difficult to see how we might account for these constraints in terms of
use or mutual beliefs and expectations. What is needed is rather a structural
condition-the same condition that applies to verbphrase anaphora in general-
in which case we must attribute to metaphor the semantic structure, or meaning,
necessary for the requisite condition to apply.
2. As we also saw earlier, when demonstratives occur within the surface
scope of a modal operator, their interpretation is nonetheless determined with
respect to their context of utterance and never with respect to the circumstance
with respect to which the truth of the sentence is determined. Similar facts
obtain with metaphor: the interpretation, i.e.. truth condition or content, of the
metaphor is always fixed relative to its context of utterance, never relative to
 WHAT METAPHORS DO NOT MEAN 45

the circumstance on which the truth of the sentence is dependent. Suppose that
Pans disagrees with Romeos utterance of Juliet is the sun but concedes that
(9) Juliet might have been the sun.
(9) will be true just in case there is some possible circumstance w in
which Juliet has the particular set of properties P which is the content of is the
sun interpreted metaphorically in its actual context of utterance cm i.e., Juliet
must fall in the extension of w of the relevant property P, but the relevant
property P is fixed relative to the context of utterance c, not reIative to the
counterfactual circumstance w . It is not sufficient for (9) to be true that Juliet
possess whatever property happens in w to be the content of is the sun were
it interpreted metaphorically there. Now, as in the case of demonstratives, this
condition does not tell us what the interpretation of the metaphor is; it only tells
us what it cannot be. And, again, as in the earlier examples, it is hard to see how
we might explain the presence of this constraint in terms of use, intentions,
or mutual expectations. Again, this kind of constraint calls for metaphorical
meaning.58
To conclude: according to the alternative conception of semantics and
meaning I have sketched in this last section, a theory of metaphorical meaning
will always be much less than the complete story even of a single metaphorical
interpretation (truth condition or propositional content) in a context. It will seek
to describe and explain only the speakers knowledge underlying his ability to
interpret a metaphor insofar as that knowledge is proper to metaphor and prop-
erly linguistic.59 However, I hope that the restricted character of this notion of
meaning will prove to be a strength rather than a deficiency of my account. For
it is only by abstracting away from the other contributing factors that jointly
determine full metaphorical interpretations-contents or truth conditions-that
it becomes possible to discern their underlying meaning or semantic structure.
And this is something which past theories of metaphor have never been able
to do precisely because they have focused all their attention-unsuccessfully,
as writers like Davidson convincingly s h o w - o n rules or meaning that
directly yield nothing less than a full understanding of a metaphor in its
context.60
NOTES
* I want to thank Jay Atlas, Jonathan Berg, Avishai Margalit, James Higginbotham,
and Ellen Spolsky for their comments on an earlier version of this material, as well as the
audiences at departmental colloquia at the Hebrew University of Jerusalem and Ben-Gurion
University of the Negev. I also wish to acknowledge the support of the American Council
of Learned Societies during 1988-89 when the original version of this essay was written.
1. For further reflectionson this subject, see Ted Cohen. Aesthetics. Social Research
47 (Winter 1980): 600-11.
2. This essay (henceforth: WMM) first appeared in a special issue on metaphor in
Critical Inquiry 5 (1978): 31-47, and has been subsequently reprinted in, among other
places, Donald Davidson, Inquiries into Truth and Inferpretation (Oxford, 1984), 245-64.
All page references are to the latter. In addition to the three main claims mentioned in the
text, much-imdeed, the overwhelming proportion--of WMM is concerned with criticisms,
46 JOSEF STERN

large and small, of assorted previous accounts of metaphor. Since these matters do not bear
on my present topic, I forgo discussing them here.
3. I address at length these other arguments--each of which touches. I believe, on a
sensitive spot for any full-blown account of metaphor-in a work in progrcss Metaphor in
Contexr (of which the present essay is a part). It should be noted that Davidson himself
repeatedly falls back into talk of metaphor as if it does convey cognitive content; see, e.g..
WMM, 257, 261, 263. As we all know, skepticism is hard to live by.
4. Ln (among other places) Truth and Interpretation: Perspectives on the Philosophy of
Donald Duvidson. edited by Ernest LePore (Oxford, 1986). 433-46. All references to this
essay (henceforth: NDE) are to this publication.
5. See my Metaphor and Grammatical Deviance, Nous 17. no. 4 (November 1983):
577-99; Metaphor as Demonstrative, Journal of Philosophy 80, no. 12 (December 1985):
677-710; and Metaphor without Mainsprings: A Rejoinder to Elgin and Scheffler. Journal
of Philosophy 85, no. 8 (August 1988): 427-38.
6. Davidson. Communication and Convention. in his Inquiries into Truth and Inter-
pretation, 279. See also Davidsons prefatory comment to this essay: The principles of such
inventive accommodation [i.e., the principles of conversation] are not themselves reducible
to theory, involving as they do nothing less than all our skills at theory construction (In-
quiries, xix).
7. John Searle, Metaphor, in Metaphor and Thought. edited by Andrew Ortony (Cam-
bridge, 1979). 92-123.
8. bid., 113.
9. Two brief points of clarification: First, my endorsement of Davidson in this paragraph
is solely conditional on locating metaphor entirely in use, an assumption with which I shall
take issue later in this essay. As I also stated earlier, I am concerned here only with the issue
of metaphorical meaning; hence, even if what a metaphor communicates is an effect of its
use-rather than a kind of meaning, where that term is meant to carry explanatory import-
that effect may be propositional and thus (in the ordinary sense) a meaning-like entity.
Second, I often write elliptically (and potentially misleadingly) of the features communicated
by a metaphor as effects which its use causes or makes us see, when strictly speaking
what the utterance of a mctaphor causes is not a feature but an event like the interpreters
recognizing or noticing such a feature. However, despite the difference in ontology, nothing
in my argument seems to me to be affected by my loose talk. (For both of these points, I
am indebted to comments by Jonathan Berg.)
10. Compare Davidsons objections to a figurative meaning for similes:
The point of the concept of linguistic meaning is to explain what can be done with
words. But the supposed figurative meaning of a simile explains nothing; it is not a
feature of the word that the word has prior to and independent of the context of use.
and it rests upon no linguistic customs except those that govern ordinary meaning.
(WMM 255)
11. Linguistic utterances always have an ulterior purpose....There is perhaps some el-
ement of stipulation here, but I would not call it a linguistic act if one spoke words
merely to hear the sounds, or to put someone to sleep; an action counts as linguistic only
if literal meaning is relevant. But where meaning is relevant, there is always an ulterior
purpose ....What matters is whether an activity is interestingly considered linguistic when
meanings are not intended to be put to use (Communication and Convention, 273-74).
12. 1.e.. nothing should be lost concerning what is known by S and I that enables
them to communicate, and thus interpret S s behavior. This assumes. however, that we can
abstract away from all differences between how they respectively produce and comprehend
their common interpretation of the utterance.
13. It should be emphasized that, while Davidson often describes the theorists explicit
theory as a theory of 1s linguistic, or interpretive, competence, he denies any commit-
ment to its psychological reality. In this respect, his program is sharply different from
 WHAT METAPHORS DO NOT MEAN 47

Chomskys. even though both of their explicit theories take the f o m of recursive systems
of rules. Note, moreover, that Davidson also refuses to endorse the physiological reality
of his theory. that it makes any claims about the details of the inner workings of some
part of the brain (NDE438). Thus his general motivation seems to stem, not specifically
from anti-Cartesianism, but from a view that, because all our evidence is behavior, namely,
utterances, we are never justified in doing more than describing the interpreters behavioral
repertoire. Likewise, it is no doubt part of Davidsons motive for shifting away from the
speaker to the interpreter to make the theory of meaning more public. to set meaning from a
more external and behavioral, and less mentalistic and private, perspective. But like w i n e ,
then, Davidson can at most claim that his theory of interpretationfits the interpreters compe-
tence, not that it corresponds to something that guides it. The sense in whichbitis nonetheless
a theory (inasmuch as it does not attempt any sort of theoretical, i.e.. non-observationd,
explanation) is simply formal: the description has a recursive (inductively specifiable) smc-
ture which explicitly describes only a finite fragment of the interpreters competence but is
meant to apply to the potentially infinite totality of utterances in his language.
14. Contrast this view with the idealization involved in the conception of linguistic
competence found in most current theoretical linguistics, for example, Chomskys opening
paragraphs of Aspects of the Theory of Synfax (Cambridge, Mass.. 1965).
15. That understanding is the fundamental notion for Davidson tends, I think, to be
obscured by his emphasis, for a variety of reasons, on interpretation. However, when, among
other places, he must explain the standards of interpretation. it is understanding to which he
turns; see, for example. his comment, while explicating his use of the Principle of Charity,
that the aim of interpretation is not agreement but understanding (Inquiries. xvii). I shall
not try to substantiate this claim further here, but, if it is right. then the Davidson-Dummett
dispute over whether a theory of meaning ought to take the form of a theory of truth or
of warranted assertability ought to be viewed as an infernal dispute within one camp for
whom meaning is understanding-in contrast to those in other camps, such as Chomsky,
for whom it is not.
16. More specifically, 1 s theory is of what he expects S will intend to say with his (Ss)
particular words on this occasion, given 1s prior knowledge of S and of the first meanings
S has attached to his (Ss) words in the past. Ss theory, on the other hand, consists of his
intentions that particular words of his will be interpreted by I as saying such-and-such, given
his beliefs and expectations about 1 s ability to interpret him (S)as saying those things with
those words. So, if S believes that I will not be able to interpret him as saying such-and-such
with certain w&, he will not intend for those words to be interpreted in that way.
17. Davidson adds that the conditions of utterance (speaker, audience, and occasion)
must also be nomal or standard. but he explicitly refuses to explain these terms, so it
is not clear what constraints they impose.
18. Ian Hacking, The Parody of Conversation. in Truth andlnterpretation: Perspectives
on the Phifosophy of Donald Davidron edited by E. LePore (London, 1986). 447.
19. Davidson asserts. without argument, that the intentions with which an act is per-
formed are usually unambiguously ordered by the relation of means to ends (435). but this
is far from obvious. In any case, any given unambiguous ordering would hold only for a
particular utterance; there is no a priori reason to think that the first meaning of one token
of a type should necessarily be the first meaning of all tokens of that type.
20. Note that Davidson introduces first meaning as a preliminary stab (NDE 434, my
emphasis) at literal meaning, but there is nothing preliminary about the way in which he
continues to use it as its explication throughout the remainder of NDE; see, e.g., Every
...
deviation from ordinary usage, so long as it is agreed on for the moment is in the passing
theory as a feature of what the words mean on that occasion. Such meanings, transient
though they may be, are literal; they are what I have called first meanings (NDE 442).
21. There may be exceptions to this generalization. Knowledge of the secondary intention
of an utterance (e.g.. knowing that it is a promise or threat) may affect which first meaning
is assigned to some word: therefore, the assignment of first meaning may also be sensitive to
48 JOSEF STERN
the post-mantic contextual features that bear on determination of the utterances secondary
intention.
22. This notion of autonomy is essentially that of Davidsons principle of the autonomy
of linguistic meming which he illustrates by the fact that a sentence with a given linguistic
meaning can be used to serve almost any extra-linguistic purpose (Thought and Talk,
Inquiries, 164) Elsewhere, however, his formulation of the principle varies somewhat; cf.
Communication and Convention, 274. All these notions of autonomy of meaning should,
however, be distinguished from the notion I discuss in Section VI.
23. One virtue of this explication of literal meaning is that, inasmuch as it acknowledges
its pre-semantic context-dependence, it avoids the attacks of the various deconstructionist
philosophers and literary theorist who deny the literal/non-literal distinction on the grounds
that all meaning is relative to a culture or various kinds of beliefs and presuppositions. True
as this observation surely is, all it shows is that all meaning is pm-semantically context-
dependent; the literal/non-literal distinction, which hinges entirely on post-semantic context-
in/dependence, is an entirely separate matter. Similar remarks apply to John Searles recent
argument that literal meaning is not context-independent, in Literal Meaning, Erkenntnis
13 (1978): 207-24; all his examples are clearly cases of pre-semantic contextdependence.
24. For the semantic account of demonstratives and indexicals I am assuming here, see
David Kaplan. Demonstratives. (at last published!) in Themesfrom Kuplan. edited by
Joseph Almog. John Perry, and Howard Wettstein (Oxford, 1989). In what follows I use the
term demonstrativeto cover both the proper demonstratives (This, That) and indexicals
(I, now, etc.).
25. Note that I take this formula-that the metaphorical depends on the literal40 be
describing a dependence that holds in the metaphorical interpretation of an utterance. One
might, however, explicate it as referring not to a synchronic relation in language but to a
diachronic one-that there first had to be literal language before there could be metaphor-r
as a claim not about each individual metaphor but about the language as a whole at a time.
26. See WMM 261, 262.
27. On this connection, see Ted Cohen, Jokes in Pleasure, Preference and Value. edited
by E. Schaper (Cambridge, 1983) and Metaphor and the Cultivation of Intimacy, Critical
Inquiry 5 (1978).
28. Max Black, Metaphor. in his Models and Metaphors (Ithaca, N.Y.. 1962). 2547,
esp. 37. Cf. also Nelson Goodman, Lunguuges of Art (Indianapolis, 1968), 78. To anticipate a
possible misunderstanding: while Blacks account places great weight on the creative, hence,
causal, power of a metaphor, I am not suggesting that his is a causal theory of metaphor.
29. See Robert Stainaker, Presuppositions. Journal of Philosophical Lagic 2 (1973):
447-57.
30. The claim needs qualificationbecause there is typically no unique similarity or feature
that would provide an appropriate metaphorical interpretation for an utterance. In Metaphor
in Contexr, chap. 5. I attempt to work out the requisite details. On the question when an
utterance is taken. or identified, to be metaphorical, see my Metaphor and Grammatical
Deviance.
3 1. See further Stalnaker, Presuppositions, 45 1.
32. On this parallel, see now Richard Moran, Seeing and Believing: Metaphor, Image,
and Force, Criricul Inquiry 16 (Autumn 1989): 87-1 12. I also explore this analogy further
in Metaphor in Context, chap. 7.
33. This objection is not the objection often raised against resemblance or comparison
theories of metaphor that resemblance is a symmetrical relation while what metaphors ex-
press is not: for I would agree with Amos Tversky that resemblance is itself assyrnetrical;
see his Features of Similarity. Psychological Review 84, no. 3 (July 1977): 322-52. My
point here is, apart from the nature of resemblance, that on Davidsons account that makes
use only of the literal referents of the component terms. there should be no metaphorical
difference among the different strings cited in the text inasmuch as they all refer to the
same individuals. At this point, Davidson might of course appeal to context to distinguish
 WHAT METAPHORS DO NOT MEAN 49

among the different metaphors; but, without further explanation as to how the context so
functions. it should be clear that such an appeal is simply a way of defening unsolved
problems or depositing them in a theoreticians wastepaper basket.
34. See WMM 249.
35. Richard Rorty, Unfamiliar Noises I: Hesse and Davidson on Metaphor, Proceed-
ings ofthe Aristotelian Society, supp. vol. 61 (1987): 283-96, esp. 293. Rortys larger project,
it should be noted, is not Davidsons; for reasons of space, here I cannot address this di-
mension of Rortys essay.
36. Rorty, Unfamiliar Noises I, 283, 296. Davidson also emphasizes that his denial
of cognitive content to metaphor should not be taken to mean (as it did for many earlier
positivists) that it is confusing, merely emotive, unsuited to serious, scientific, or philosophic
discourse (WMM 246). But his praise of metaphor never reaches the exalted heights of
Rortys; at best for Davidson, metaphor is a legitimate device not only in literature but in
science, philosophy, and the law (ibid., my emphasis). .
37. Rorty. Unfamiliar Noises I, 284. The view Rorty contrasts here with Davidsons is
that of Mary Hesse; see her paper (in the joint symposium with Rorty), Unfamiliar Noises
11: Tropical T a k The Myth of the Literal, Proceedings of the Aristotelian Society, supp.
vol. 61 (1987): 297-31 1 (and references therein to her earlier papers on metaphor).
38. Rorty. Unfamiliar Noises 1, 295.
39. Note that this sense in which Rorty claims that metaphors cannot be understood
or interpreted, namely, that they cannot be brought under an antecedent scheme (ibid.,
290). is much more general than the notion of understanding specific to language, i.e., what
it is to understand what a word means. Indeed Rortys alternative notion of understanding
(which fits metaphor) is ironically much closer to the way in which Davidson describes
the general process by which speakers and interpreters arrive at a mutual understanding,
or interpretation, of their utterances by revising their prior theories to make them suitable
as passing theories.
40. See Donald Davidson, Causal Relations, reprinted in his Essays on Actions and
Events (Oxford, 1980), 149-62.
41. Although it is never very clear how much Davidson wants us to draw out of the
various things to which he compares a metaphor (e.g., pictures. jokes, poems), an objection
similar to that raised in the text should apply against the view that metaphorical interpretation
should be understood on the model of dream interpretation. as Davidson seems to suggest in
his opening sentence of WMM: Metaphor is the dreamwork of language (WMM 245). For
reasons of space, I cannot explore this intriguing analogy in depth; suffice it to say that, while
dream interpretation (say, according to Freuds theory) makes much use of causal relations,
the literal meanings of the words (used to report the dream) play little role in its explanation.
42. Keith Donnellan, Reference and Definite Descriptions, Philosophical Review, 75
(1966): 281-304. Despite my objections to Davidsons use of Donnellans referential de-
scriptions, there is a deeper connection here than meets the eye, and elsewhere, in my own
analysis of metaphor as a kind of demonstrative, I have drawn on some other elements that
lie beneath the surface of the comparison.
43. Davidson uses this phrase as a primitive notion without explication.
44. In this passage Davidson lumps together irony with metaphor, apparently assuming
the commonly held view that all such figures belong to one class with one explanation. In
Metaphor in Context. I argue, however, that irony and metaphor are two different kinds of
non-literal interpretation that should not be treated on a par. Hence, I leave it out of my
discussion here.
45. To begin with, Davidsons focus (admittedly inspired by some of Donnellans original
examples) on cases where the referential description does not semantically designate the
intended refmnt i s misleading; whether or not a description is referential is independent
of whether the description is or is not satisfied by its referent. The description is, rather,
referential just in case there is a specific individual for whom the speaker uses the description
simply as a means of reference. In the normal case, moreover, the description serves as such
50 JOSEF STERN
a means because the individual is in fact semantically designated by ic and where the
individual does not satisfy it, the speaker must (except in certain special contexts) at least
believe that he satisfies i t 4 n pain of being misleading. It is also misleading to contrast,
as Davidson does, saying something true with the sentence without at least raising the
question whether the speaker has thereby made a statement or expressed a proposition that
is true. Although thm has been much discussion in the literature over whether Donnellans
distinction is semantic or pragmatic, I would argue that it is both: i.e.. intuitions of both
kinds are at work in Donnellans description of his distinction. The semantic dimension
of the distinction cornsponds to the difference between singular and general propositions.
The pragmatic distinction is between teleological acts (which refer to a predetermined
referent) and blind (to refer to whoever happens to meet the description, known or not)
uses of descriptions.
46. I owe the example to Sidney Morgenbesser who raised it to illustrate a rather different
point. Not everyone might agree that this is an example of a malaprppism rather than a kind
of semantic error, though it should count as a malapropism on Davidsons characterization. In
any case. the same moral can be drawn if the example is not strictly speaking a malapropism.
For discussion of the problem of distinguishing malapropism from other kinds of semantic
speech errors, see David Fay and Anne Cutler, Malapropisms and the Structure of the
Mental Lexicon, Linguistic Inquiry 8. no. 3 (Summer 1977): 505-20.
47. See Fay and Cutler, Malapropisms and the Shucture of the Mental Levicon.
48. Of course, not all metaphors may be absurd or even false when their words are taken
literally-e.g., Noman is an island. However, even where the utterance, taken literally,
would be true in which case we can presume that the speaker may in fact tacitly believe
what it literally says, it would be philosophically gratuitous to ascribe such belief because
that is how he must understand their utterances as metaphors.
49. He? I assume that all demonstratives are directly referential terms, i.e., that their
propositional content (in a context) is the individual or thing to which they respectively refer
rather than a Fregean Sense or some other intensional entity. For argument, see Kaplans
Demonstratives.
50. This notion of the autonomy of meaning should be distinguished from Davidsons
principle of the same name (see above note 21) which, on none of its formulations, expresses
the idea that linguistic meaning constrains speakers communicative intentions.
51. For an especially clear exposition of cumnt linguistic thinking about these con-
structions, see James Higginbotham. Knowledge of Reference, in Reflectionson Chomrky.
edited by Alexander George (London, 1989). 159-65. Although the first version of this
section of this essay was Written before Higginbothams essay came to my attention, the
present formulation of its argument owes much to his discussion of the same issues.
52. Cf. Higginbotham, Knowledge of Reference, 162.
53. Because the conditions in question primarily exclude interpretations the sentence
does not and never could have, and these excluded interpretations are presumably never part
of the linguistic data to which speakers are exposed, the conditions cannot be inductively
learned from confirming instances. Therefore, in the absence of negative evidence (as
this problem is known in the psycholinguistics literature), the question naturally arises, and
especially for non-rationalist accounts like Davidsons: How do speakers come to know
these conditions? For further discussion, see Higginbotham. Knowledge of Reference,
162ff, and N. Hornstein, La& us Grammar (Cambridge, Mass.. 1984). chap. 1.
54. The contrapositive nasOning I have just rehearsed should be reminiscent of one of
Chomskys arguments for the so-called competence-performance distinction: i.e., because
acceptable use of language is a function of multiple psychological, physiological, and social
abilities and capacities, most of which would not be d d b e d as linguistic on any reason-
able characterization, a theory of linguistic competence proper will itself never account for
acceptable use but will explicate only what the speaker knows about the specifically linguis-
tic properties that contribute to his complex ability that issues in acceptable speech. For a
similar conclusion based on rather different arguments, see now Ellen Spolsky. The Limits
 WHAT METAPHORS DO NOT M E A N 51

of Literal Meaning, New Literary History 19 (1987-88): 419-40, and Ellen Schauber and
Ellen Spolsky, The Bounds of Interpretation: Linguistic Theory and Literary Text (Stanford,
1986).
In David Kaplans semantic theory for demonstratives, this knowledge would correspond
to the speakers knowledge of the character of an expression rather than knowledge of its
content in a context, where the character ofa expression is a rule. or function, that determines
for each context of utterance the content of the expression in that context. In order to know
the actual content of the expression in a context, it is necessary in addition that the interpreter
have extra-linguistic knowledge of the relevant contextual parameter. For further details, see
Kaptan. Demonstratives. It should also be noted that all expressions in the language are
assigned characters by the semantic theory, although only those of the demonstratives are
non-stable, i.e., determine different contents in different contexts. In effect, then, only for the
demonstratives, or for other kinds of context-dependent expressions to which the semantics
might be extended, does the character-content distinction make a difference.
55. In Metaphor as Demonstrative (see above n. 5), I proposed that we identify the
speakers semantical knowledge underlying his ability to interpret an expression metaphor-
ically with his knowledge of its metaphorical character. i.e., with the character of the
expression as it is interpreted metaphorically. Like the character of a demonstrative or, more
generally, any expression (e.g., descriptions) interpreted demonstratively, the character of
an expression interpreted metaphorically is a non-stable rule, or function, from its context
to its content in that context. In the case of metaphor, the relevant context, or contextual
parameter, ~ I Ecertain subsets of (extralinguistic) presuppositions associated with the ex-
pression being interpreted metaphorically. while its content in a context is usually (at least
for predicates) a set of properties. For further discussion, see Metaphor as Demonstrative,
7W10. and Metaphor in Context. chap. 5 .
56. Here I follow Edwin S.Williamss analysis of verbphrase anaphora which involves an
interpretive (or copying) rule of Discourse Grammar rather than a deletion rule of Sentence
Grammw, see his Discourse and Logical Form, Linguistic Inquiry 8, no. I (Winter 1977):
101-39. Cf. also Noam Chomsky, Lunguage and Mind, enlarged ed. (New York, 1968).
33-35; and Ruth Kempson. Semantic Theory (London, 1977). 128-32.
57. According to the semantic theory sketched in Metaphor as Demonstrative, the
violations in (7)and (8) are. in fact, formally different. In (7) the antecedent and anaphor
will be structurally described (or intcrpreted at the level of Logical Form. in Williamss
theory) with different interpretive forms or. in my terminology. different characters. So, if
the metaphorical character of is the sun is the character of the metaphorical expression
Mthat[is the sun] (see below), then (7)will be represented at the level of Logical Form. or
at the level corresponding to the speakers proper semantic competence, by the character of
(7*)*The largest blob of gases in the solar system is the sun, and Juliet/Achilles is
Mthat[the sun].
Since is the sun and MThat[is the sun] area expressions with evidently different
characters, the violation on copying in (7)will be already at the level of character. In (8).
however, the antecedent and anaphor. both being metaphors (Le.. both having metaphorical
expression types). will have the same character.
(8*)*/?Juliet is Mthat[the sun]. and Achilles is Mthat[the sun].
Hence, their violation of the copying constraint must necessarily occur at the level of
content. Now. the different levels or stages at which the copying constraint is violated
do in fact affect our intuitions of unacceptability. It is possible to cancel the implication
of univocality at the level of content, as in (8). but not at the level of character, as in (7).
Contrast (7**)and 8**):
(7**)*The largest blob of gases in the solar system is the sun. and Juliet/Achilles
is, too-but not in the same sense/way.
(8**) Juliet is the sun. and Achilles is. t-but not in the same sense/way.
52 JOSEF STERN

58. It should be noted that on Davidsons account both (6) and (9) are absurdly false
according to what they literally mean-which is, of course, all that they mean for Davidson.
Hence, insofar as they have the same truth-value, it is not clear how he can even describe
Romeos and Pariss disagreement in the situation imagined.
59. Although a full explanation would require a more detailed and technical presentation
of the specific semantic theory of metaphor I proposed in Metaphor as Demonstrative,
since I have criticized Davidson on this score, I owe the reader at least an indication here
of how I would go about explicating how a metaphorical interpretation of an expression
depends on its literal meaning. In the above essay, I proposed lexically representing the
metaphorical interpretation of an expression 0 by the metaphorical expression Mthat[0]
which was intended to parallel Kaplans Dthat-descriptions of the form DthatI01 which
represent the demonstrative interpretations of arbitrary definite descriptions. Hence, at the
level of interpretation corresponding to the interpreters semantic knowledge of metaphor,
is the sun interpreted metaphorically in Juliet is the sun (in the context described in
Shakespeares play) would be represented by the metaphorical expression Mthat[is the
sun], whose character would be a function from the metaphorically relevant contextual
presuppositions associated with the sun (interpreted literally) to some property (of Juliet).
Here the literal meaning, or character, of the expression the sun is critical in determining
the relevant set of associated presuppositions, even though the criterion of relevance may
itself vary from context to context. Moreover, here the appropriate sense of meaning (of
is the sun) is character rather than content because of demonstratives used metaphorically
whose character rather than content is obviously what is the relevant (e.g., it in cummingss
a salesman is an it that stinks to please.). However, all this is at the level of the character
of the metaphorical expression rather than at the level of its content, or truth condition, in
a context. In short, then, on my account, the metaphorical depends on the literal in that
the metaphorical character of an expression 0 interpreted metaphorically (i.e.. the character
of the metaphorid expression Mthatia]) is a function of, or depends on. the character
of 0 (interpreted literally).
60. Apart from accounting for the autonomous aspects of this metaphorical inter-
pretation, elsewhere I have presented other reasons for positing a level of metaphorical
interpretation like that of character. See Metaphor as Demonstrative, 681-90. 709-10,
Metaphor without Mainsprings, 43638, and Metaphor in Conrexr. chap. 7.