Metaphor, Literal, Literalism: Josef Stern

Metaphor, Literal, Literalism
JOSEF STERN
Abstract: This paper examines the place of metaphorical interpretation in the current
Contextualist-Literalist controversy over the role of context in the determination of
truth-conditions in general. Although there has been considerable discussion of
‘non-literal’ language by both sides of this dispute, the language analyzed involves
either so-called implicit indexicality, loose or loosened use, enriched interpretations, or
semantic transfer, not metaphor itself. In the first half of the paper, I critically evaluate
Recanati’s (2004) recent Contextualist account and show that it cannot account for the
metaphorical-literal dependence characteristic of metaphor. I then turn to Carston’s
(2002), and Bezuidenhout’s (2001) Contextualist accounts and show that they place no
constraints on metaphorical interpretations. In the second half of the paper I sketch a
Literalist theory of metaphor elaborated in Stern (2000) and respond to two kinds of
Contextualist criticisms of that account by Camp (2005) and Stanley (2005).
1. Metaphor Among the Non-literals
Since the 70’s it has been the prevailing view among philosophers, linguists, and
cognitive scientists that metaphor falls in pragmatics, not semantics. Once one
distinguishes between sentence-meaning, or what a sentence says, and speaker’s- or
utterance-meaning, or what a sentence can be used by its speaker to mean, it is
only natural, for at least three reasons, to categorize metaphor as a kind of use
rather than linguistic meaning. (1) The interpretations of metaphors—i.e.,
expressions used, or interpreted, metaphorically—typically vary from utterance to
utterance and therefore cannot, unlike the semantic notion of literal meaning, be
individuated by expression-types. (2) Metaphorical interpretations depend on their
speakers’ intentions, or what the speakers mean their hearers to recognize as their
intention in using the words they do. (3) The truth-conditions or propositional
contents of metaphors draw on extra-linguistic associations and presuppositions of
varied kinds. Some of these are false but common knowledge, others depend on
skills and abilities—like the ability to perceive similarities or what is exemplified or
sampled by some object—that are not specific to language. All these factors point
one in the direction of pragmatics, use, or general theories of symbols.
For comments and criticism, I wish to thank participants in the Mind & Language conference on
Metaphor at which a draft of this paper was presented, and especially Robyn Carston, Bob
Gibbs, Sam Guttenplan, Roger White, and Deirdre Wilson; Nat Hansen, Yitzhaq Benbaji, and
Elizabeth Camp; and two anonymous referees.
Address for correspondence: Department of Philosophy, University of Chicago, 1115 E. 58th

Street, Chicago, IL 60637.
Email: j-stern@uchicago.edu
Mind & Language, Vol. 21 No. 3 June 2006, pp. 243–279.
© 2006 The Author
Journal compilation © 2006 Blackwell Publishing Ltd
244 J. Stern
However, all the classic pragmatic or use-oriented characterizations of

metaphor—Grice (1975), Searle (1993), Davidson (1984), Cohen (1975), and
Sadock (1993)—assume that there is a clear, sharp notion of sentence-meaning, or
what is said by a sentence, that contrasts with what it can be used to mean.
Sentence-meaning is the propositional content, or truth-conditions, of the sentence
uttered, determined by its conventional or linguistic meaning (plus assignment of
values to demonstratives, indexicals, and names) for a representative cross-section
of utterances in natural language; it is what hearers intuitively understand by the
utterance, its so-called literal meaning, and it is the kind of meaning that constitutes
the domain of semantics. Pragmatics, then, takes up all the further uses in context
of these sentences with their sentence-meaning.
In recent years, the tide has shifted. The ubiquity of non-literal language has
come to be much more widely appreciated, leaving it less certain how much
natural language is actually literal. Many now recognize how little conventional
meaning determines the intuitive truth-conditions, or contents, for most utterances,
and how much more is contributed by context and speakers’ intentions, and in
ways that seem to resist systemization and formalization. This suggests for some
theorists that the paradigm for the analysis of intuitive truth-conditions, content,
and what is said should be non-literal language. In that case, even the determination
of truth-conditions or content will fall in pragmatics, not semantics.
What are the implications of this change of view for our understanding of
metaphor? At a first glance, one would think that this surge of interest by
philosophers, linguists, and cognitive scientists in the non-literal would be good
news for metaphor. After a long history of relative neglect—Lycan (1999, p. 205)
calls metaphor ‘the dark side’ of the philosophy of language and Blackburn says
that its quality of discourse is not ‘one of analytic philosophy’s brighter
achievements’ (1984, p. 180)—the agenda of ‘serious’ philosophy of language
finally seems to be coming around to metaphor. But on a closer look the
situation is less rosy. For the most part, the non-literal language or meaning on
which the contemporary theorists focus consists either in conversational
implicatures and indirect speech acts or in utterances whose intuitive,
communicated truth-conditions contain ‘unarticulated constituents’, factors that
are not expressed by concrete, phonetically realized linguistic items in their
sentential surface structures and instead appear to be directly contributed by
context. Examples of unarticulated constituents are the intuitive propositional
elements in (1)–(3) corresponding to the parenthetic expressions in their asterisked
counterparts (1*)–(3*).
(1) I had breakfast.
meaning,
(1*) I had breakfast (today/on some contextually indicated day).

(2) It’s raining.
© 2006 The Author
Metaphor, Literal, Literalism 245
meaning,
(2*) It’s raining (here/at some contextually indicated location l).

(3) You are not going to die.
(as spoken by a mother to her son who just cut himself ) meaning,
(3*) You are not going to die (now/in the immediate future).
Everyone agrees that the intuitively understood truth-conditions, or content, of

these utterances—what they intuitively say—is richer than anything determined
by the conventional meaning explicitly represented in the actual sentences uttered.
Therefore, if we count the latter as the literal meaning of the utterance, the
contextually enriched meaning should count as non-literal meaning. But this raises
the question whether the context autonomously—independently of linguistic
constraints—makes this enrichment to what is intuitively said or whether the
contextual input is filtered through a more abstract level of linguistic representation
that underlies the surface structure of the concrete sentence uttered. In the current
controversy, Contextualists take the former position, Literalists or Indexicalists, the
latter. A parallel controversy involves contextual operations called ‘loosening’ and
‘semantic transfer’ (which corresponds to the classical figure of metonymy).
Metaphor, finally, comes in but only at the end of the story, and it is explained or
analyzed in terms of one of these more basic kinds of operations—enrichment,
loosening, or semantic transfer. This attempt to reduce metaphor to one of the
other modes of non-literal meaning makes good methodological sense. But the
fact is that ‘the paradigm case of non-literal meaning’, as Recanati (2004, p. 76)
calls metaphor, gets relatively short shrift in this story.
In this paper I shall discuss the consequences for metaphor of the current
Contextualist-Literalist debate over the effect of context on truth-conditions and
over the status of literal and non-literal meaning. I begin with Francois Recanati’s
(2004) recent variant of Contextualism in order to sharpen the problem that
metaphor poses for these accounts. I then turn to a close look at the process of
loosening, which Contextualists including Recanati, Sperber and Wilson (1985/6;
1995), Carston (2002), and Bezuidenhout (2001) propose for metaphor. In the
second half of the paper, I first sketch a Literalist theory of metaphor, based on
Stern (2000), and then turn to Contextualist criticisms of that model. I leave it to
the reader to reach a verdict on this match.
2. The Problem of Metaphor for Contextualism
One claim that present-day Contextualists and Literalists—unlike many writers in

the past, including Davidson (1984), Rorty (1987), Black (1962, 1993), and Cavell
(1967)—do not dispute is that metaphors express truths. Everyone also agrees that
© 2006 The Author
246 J. Stern
these truths are no different in kind than those expressed by literal language and
that the content of an expression which in fact is used metaphorically differs from
what it would have expressed had it been used literally. The controversy begins
only when we ask: What determines the truth-conditions of the utterance when it
is used metaphorically? For the Literalist, metaphorical interpretations cannot be
individuated via, or be determined simply by, the linguistic meanings of the
concrete words so used (plus assignment of references to demonstratives and
names) because the same expression, with the same linguistic meaning, interpreted
metaphorically in different contexts can have different truth-conditions. Nor can
each metaphorical interpretation in each context count as a different meaning for
the expression because, given the indefinite number of different, novel
interpretations that can be generated in different contexts, we would thereby rob
the notion of meaning of all explanatory power. Furthermore, the multiple
metaphorical interpretations of one expression do not fit any of the standard
typologies of ambiguity or polysemy (Stern, 2000). The Literalist, then, has two
remaining alternatives. Either he can posit a richer underlying linguistic
representation whose meaning will determine the truth-conditions of each
metaphorical utterance in context or he can restrict semantics to literal meaning
and assign metaphorical interpretation to pragmatics. I shall advance the first
alternative in section 4; the second alternative has been the preferred solution for
most Literalists, including those who adopt the first alternative for other instances
of non-literal interpretation (see Stanley, 2004).
For Contextualists like Recanati, the problem of metaphor is entirely different.
He begins from the vantage point that no ‘intuitive’ truth-conditions, nothing
intuitively said by the utterance of any sentence, is determined solely by its linguistic
meaning independently of context and the speaker’s intentions and beliefs. A
so-called ‘minimal’ literal proposition expressed solely in virtue of a sentence’s
conventional linguistic meaning is either the semanticist’s fantasy (because even
processes like reference assignment for demonstratives and names must appeal to
speakers’ intentions) or is explanatorily vacuous. Instead the primary propositions
or truth-conditions understood by speakers—where understanding requires
awareness or consciousness of what is said (Recanati’s ‘Availability constraint’)—are
already infected by pragmatic processes like enrichment, loosening, and semantic
transfer. But if this is so, and if the literal is still taken to be the linguistically
determined, what then remains of the intuitive distinction between the literal and
non-literal? Is all meaning, everything intuitively said, non-literal? At this point,
Recanati could simply abandon the literal/non-literal distinction to folk-linguistics
but he doesn’t. Instead he defines two notions of the non-literal (and three notions
of the literal). The problem raised by metaphor is that it falls under neither of these
notions of the non-literal.
To show this, let’s briefly review Recanati’s typology of literal and non-literal
meaning.
1. T(ype)-literal meaning is the conventional meaning of simple expression-types,
assigned by rules of the language and inherited by each of its tokens.
© 2006 The Author
2a. M(inimal)-literal meaning is the meaning of a token that is either identical to the
T-literal meaning of its type or that differs from it only in a way it signals and governs
(e.g. referential ‘saturation’ of a value to an indexical).
2b. M(inimal)-non-literal meaning is the meaning of an utterance that is a product

of primary but optional pragmatic processes like free enrichment (where each (a)
sentence M-non-literally means the (b) sentence):
(4a) They got married and had children.

(4b) They got married and then had children.
loosening:
(5a) This dish is raw.

(5b) This dish is not fully cooked; and
and semantic transfer:
(6a) The ham sandwich is getting restless.

(6b) The ham-sandwich-orderer is getting restless.
These processes go beyond the government of T-literal meaning, and (i) operate
only on sub-propositional linguistic units, not on prior propositions; (ii) are not
necessary for the sentence to express a (minimal) proposition; and (iii) are blind,
mechanical, sub-personal, and not consciously available or accessible to speakers
and interpreters. (I return to these conditions.) It follows that the M-non-literal
outputs of these optional pragmatic processes are meanings that are not inferred
from a prior M-literal meaning (which would presuppose its conscious availability);
instead they are the primary meanings intuitively said by their utterance.
Nonetheless, as Recanati admits, each of (4)–(6), though M-non-literal, is not
what we intuitively consider ‘non-literal’.1 Why?
3a,b. P(rimary)-literal and -non-literal meaning: This last pair of meanings derives
from the pre-theoretical idea of primary and secondary meanings: a secondary
meaning is one inferred from or conveyed by a (logically) prior meaning through
a pragmatic context-dependent process of inference, e.g. conversational implicatures,
indirect speech acts, and ironic interpretations. In this sense of secondary meaning,
interpretations that are not pragmatically inferred from still prior interpretations of
the same utterance count as primary. Now, because a secondary meaning is
intuitively non-literal, Recanati reasons that primary meanings are literal, hence, he
calls the latter p(rimary)-literal while secondary meanings are p-non-literal. In what
follows I use ‘secondary meaning’ to refer to p-non-literal meanings.
1
In fact, (6) is an instance of the classic figure metonymy and is intuitively non-literal.
© 2006 The Author
248 J. Stern
Notwithstanding Recanati’s new labels, these various ways of drawing the literal/
non-literal distinction are familiar. T- and M-literality re-work the old idea that the
literal is the encoded lexical meaning of an expression. P-literality is a variant on the
idea that the literal is the interpretation of an utterance regardless of, and invariant
across, its further uses or additional inferences drawn from it (Davidson, 1986).
However, Recanati’s reconstruction of these notions raises one crucial question:
Why are M-non-literal meanings (e.g. (4) and (5), generated by free enrichment
and loosening) not intuitively non-literal in contrast to secondary meanings that
are?2 Recanati’s answer is phenomenological: secondary meaning is ‘felt’ to be
‘special’; i.e. ‘the language users must be aware that the conveyed meaning exceeds
the conventional significance of the words’ (2004, p. 75). Recanati calls this feeling
or awareness ‘transparency’ or the ‘external duality’ of the utterance. In contrast,
M-non-literality, and the processes of enrichment or loosening that generate
M-non-literal interpretations, are blind, sub-personal, and not consciously available
to interpreters. We are aware only of the M-non-literal proposition that is the output
of the pragmatic processes, not of a prior M-literal proposition on which those
processes operate. Hence, there is no felt duality, and no intuitive non-literality.
One might raise various questions about Recanati’s phenomenological criterion
for non-literality. But let me put them aside to turn to metaphor. Metaphorical
interpretations/meanings are intuitively non-literal if anything is. But they do not
fit into either of Recanati’s categories of the non-literal. Metaphorical interpretations
are not secondary meanings because they are not inferred from another (primary,
literal) proposition, unlike a conversational implicature or indirect speech act.
(I return to Recanati’s apparent argument for this claim in section 2.) But a
metaphorical interpretation is also not a typical instance of M-non-literal meaning,
generated by a primary pragmatic process (like loosening or enrichment), because,
unlike M-non-literal meaning, metaphor is intuitively non-literal. So, metaphor is
prima facie neither M-non-literal nor P-non-literal (a secondary meaning). This is
the problem metaphor raises for Recanati’s Contextualism.3
Recanati’s solution is to explain the intuitive non-literality of metaphorical
interpretations by analogy to his phenomenological explanation of the intuitive
non-literality of secondary meanings. He proposes that metaphors have an internal
duality felt in the tension or conflict between the M-literal predicate (the literal
vehicle ‘is a bulldozer’) and what it is about (its ‘target’), e.g. the referent of the
subject term ‘Robert’ in (7):
(7) Robert is a bulldozer.
2
One might balk at classifying such inferred meanings as either literal or non-literal. It might
be more accurate to say that if implicatures and indirect speech acts must be so classified, they
are intuitively non-literal rather than literal.
3
The problem is not unique to Recanati, although other Contextualists like Sperber and
Wilson, 1985/6 and Carston, 2002 play down the intuitive figurativity of metaphor by placing
it on a continuum with loose use.
© 2006 The Author
The felt duality is the falsity, necessary falsity, categorical violation, or semantic
anomalousness perceived between the application of the M-literal vehicle to its
subject, and this ‘underlying conflict’ or felt ‘discrepancy’ gives rise to the figurativity,
or intuitive non-literality, of the M-non-literal metaphorical interpretation.4
Let me mention a number of difficulties with this solution. First, as Carston,
2002 has observed, literal categorical falsity or semantic anomalousness is at odds
with the standard contextualist explanation of metaphorical interpretation (which
Recanati seems to endorse) in terms of the primary pragmatic process of ‘loosening’.
On this account, the interpreter first sorts through the totality of features—both
lexical/analytic and encyclopedic (i.e. extra-linguistic common knowledge or
communally presupposed beliefs)—to which she has access (by decoding the lexical
entry of the literal vehicle) and then drops those features not intended to apply to
the subject. This operation broadens the extension of the loosened predicate, but
it works best, and possibly only, for metaphors whose literal vehicle is possibly true
of its subject, e.g.,
(8) Kripke is an alchemist.
We arrive at the metaphorical interpretation of (8)—say, that Kripke can turn the
most ordinary or common ideas and intuitions into philosophical gold—through
loosening because the features in the lexical entry for ‘alchemist’ (including the
presuppositions and encyclopedic knowledge we associate with alchemists) are also
literally applicable to Kripke (although in fact literally false of him). With literal
category mistakes or necessity falsehoods, like (7) whose metaphorical interpretation
is that Robert is obstinate, persistent, and insensitive, loosening cannot select the
appropriate psychological features since they are not of the set of features that
constitute the lexical meaning, including encyclopedic knowledge, associated with
the literal vehicle ‘bulldozer’. Carston raises this as a general problem for
contextualist explanations that employ loosening. (We will return to this problem
in section 3.) But it applies mutatis mutandis to Recanati. The more he relies on
‘internal duality’ in order to explain the intuitive non-literalness of metaphor, the
less he can explain the content of the metaphorical interpretation in terms of
loosening. The metaphorical interpretations generated most plausibly by loosening
are precisely those without the kind of internal duality that would account for
their intuitive non-literality.
Second, the appeal to underlying tension between the literal vehicle and its
target is both an old story and one that is now known to be an old wives’ tale. One
of the most widely appreciated lessons for metaphor learned over the last twenty
years is that in general metaphors are not, and certainly do not need to be, were
they taken literally, grammatically deviant, semantically anomalous, category
mistakes, or even just plain false. We now recognize the prevalence of twice-true
4
By ‘literal vehicle for a metaphor’ I mean the expression, e.g. ‘is a bulldozer’ in (7), as it
would be (semantically) interpreted literally.
© 2006 The Author
250 J. Stern
(Cohen, 1976) and twice-apt (Hills, 1997) metaphors and, in general, the
explanatory vacuity of what we might call the ‘literal deviance’ thesis (Stern, 1983;
cf. also White, 2001). There need be nothing deviant, irregular or unacceptable—
syntactically, semantically, or pragmatically—about the literal meaning of a sentence
used metaphorically. Although there is surely some reason why we identify an
utterance as a metaphor—one main function the literal deviance condition has
traditionally played—it has turned out to be a totally red herring as an explanation
of metaphorical interpretation.5
Third, the very idea of literally deviant meaning underlying metaphor assumes
a picture of non-deviant conventional, antecedently-fixed meaning also at odds
with the Contextualist’s fluid, context-sensitive, and ‘modulated’ (Recanati, 2004,
p. 131) conception of meaning, tailored from use to use to fit its circumstances and
the needs of communication. The more flexible, adjustable, and mutable the
Contextualist makes meaning and interpretation, the less place there is for the rigid
meaning structures that would generate the tension Recanati wishes to exploit to
account for the felt non-literality of metaphor. He can’t throw out the cake and
eat it, too (Cf. White, 2001 on Davidson).
Yet, even if ‘felt transparency’ is the wrong place to look for an explanation of
the intuitive non-literality of metaphor, Recanati is on one right track. No account
of metaphor will be adequate without explaining the fact that something about the
meaning of the literal vehicle remains active in metaphorical interpretation
(Davidson, 1984). In more familiar terminology, any account must explain the fact
that, and how, the metaphorical depends on the literal. It is not easy to say what
this truistic dependence consists in. For example, if the metaphorical depends on
the literal, you would think that the metaphorically interpreted, or used, utterance
still ‘has’ its literal meaning. But what is the relevant sense of ‘have’? In sections 3
and 4, I will explore this dependence in more depth, but let me end this section
with three examples in which active metaphorical/literal dependence is manifest.
First, a metaphorical meaning of an expression is not cognitively independent of
its literal meaning in the way in which the multiple meanings of ambiguous
expressions like ‘bank’ are independent or the meaning of an idiom like ‘red
herring’ is independent of the meanings of ‘red’ and ‘herring’. I can know one of
these without knowing the other.6 But I cannot know, or understand, what is
metaphorically said by (7) or (8) without knowing the literal meaning of ‘bulldozer’
or ‘alchemist’ (and without holding a variety of beliefs about their literal extension).
This is one way the metaphorical meaning of an utterance depends on its literal
meaning that is not true of other meanings.
5
This is not to deny that there are grammatically deviant and literally false metaphors. But its
significance remains one of the still outstanding open questions in the field.
6
A red herring is a fallacy of irrelevance, so-called because of the reputed practice of hunters
who used smoked herrings, which are red, to throw bloodhounds off the scent. It is not
difficult to imagine a current speaker who knows that a red herring is an irrelevant argument
even if she has never learned that a herring is a fish or (much less plausibly) that red is a color.
© 2006 The Author
Second, there is meaning or informational content carried by a metaphor that is

not part of its truth-conditional content but is a function of its literal meaning.
Consider, for example, the explanation an anorexic Marie gave to her therapist of
why she stopped eating (Danto, 1978; Stern, 2000). She said that her mother
forbade her to continue seeing her boyfriend, to which she reacted by saying:
(9) I won’t swallow that.
The truth-conditions of (9) as uttered by Marie referring to her mother’s

interdiction are something like:
(10) Marie won’t obey her mother’s interdiction.
But (10) cannot exhaust all the information conveyed by (9) without dropping the
explanatorily relevant information—the coherent if irrational connection between her
utterance and her behavior—carried by the specificially metaphorical mode of presentation
of those truth-conditions that is captured in the literal vehicle ‘swallow’. This is
information conveyed by the metaphor which is not part of its truth-conditions in its
context of utterance—and that information seems closely connected to the literal meaning
of the word being interpreted metaphorically. The challenge for the theorist is how to
incorporate that literal-dependence in the metaphor to capture that information.7
Third, the literal meaning of the vehicle remains active in a metaphorical
interpretation insofar as it directs its extension to other metaphors. For example,
Recanati, 2004 proposes that if we want to make the relatively dead metaphor:
(11) The ATM swallowed my credit card.
‘figurative, we have only to belabor the metaphor to make it conspicuous, as in
(12) The ATM swallowed my credit card, chewed it up, and spit it out’ (p. 77).
I disagree that there is any belaboring this metaphor. (12) is an interesting complex
metaphor precisely because it switches around the normal order of eating, which
is to spit something out before chewing it, but in any case to chew before
swallowing. Rather (12) relies on a schema, or network of metaphors, to which
‘swallow’ belongs.8 As Goodman, 1976 was perhaps the first to appreciate,
metaphors move, or are transferred, in families, not as loners.9 By extending
7
To say that this explanatory information is not part of the truth-conditions of (9) is not to say
that it cannot be expressed propositionally, only that it is not part of the truth-conditions of
the given utterance in its context.
8
Note that the occurrence of the metaphor ‘swallow’ in (11) differs from its occurrence in (9)
at least insofar as the two belong to disjoint schemata.
9
This theme has been elaborated by Lakoff (1993) and his co-authors.
© 2006 The Author
252 J. Stern
metaphors, by drawing out their metaphorical schematic relatives we manifest our

grasp of a metaphor. Perhaps this is better evidence of our grasp of the metaphor
than by giving a literal paraphrase. But for present purposes the important point is
that all these schematic relations are individuated in terms of the literal meaning of
the vehicle used metaphorically. If it did not have, or lost, its literal meaning, we
would be at a loss to account for the schematic connections between the
metaphorically interpreted expressions in (12).
In sum, any explanation of metaphor must take into account its literal meaning
in order to capture these ways in which metaphor actively depends on the literal.
The problem for Contextualists like Recanati is that they do not have the resources
to represent this dependence.
3. Metaphor, Loose Use, and Overgeneration
Contextualists can take one of two avenues to explain metaphor: treat it either as
an inferred secondary meaning or as a pragmatically processed M-non-literal
meaning. Recanati blocks the first route on the phenomenological grounds
mentioned in section 2. But his evidence simply asks us to recognize a ‘feeling’.
There is, however, an argument we might exploit in this context that shows that
metaphor cannot be a secondary meaning if being a secondary meaning presupposes
that the utterance always has a prior literal (sentential) meaning from which the
metaphorical interpretation is inferred or implicated (Cf. Margalit and Goldblum,
1994; White, 1996, and Levinson, 2001—all of whom directed this objection
against Davidson, 1984). On this argument, each of the individual words in an
utterance like (11) may have, or retain, its literal meaning, while the sentence does
not, at least if the literal meaning of a sentence is something like its truth-conditions
or a proposition (or something that determines these). The argument is that we do
not understand, or know, what must be the case for the sentence (11) to be true,
namely, the conditions in which an ATM (literally) swallows credit cards. There
is, of course, a sense in which we know that the sentence is false, but knowledge
of truth-conditions requires that we also know what the world would have to be
like for the sentence to obtain—and that we do not know. But if we do not
understand its literal truth-conditions, the purported p-literal meaning of the
sentence, we could not have inferred its metaphorical meaning from it. Hence,
there is good reason to reject the secondary meaning route but it is not
phenomenological, like Recanati’s Availability condition, nor even pragmatic, but
straightforwardly semantic.10
Let me add that this argument also counts against Griceans who argue that
metaphor is something meant—i.e. implicated—by the utterance as distinct from
what the sentence (literally, in the strict sense) says. If the argument is on the right
10
This argument assumes that some but not necessarily all metaphors are deviant or false in
order to refute the claim that all metaphorical interpretations are secondary meanings.
© 2006 The Author
track, it is not clear that something is always (literally) said by the sentence apart
from the metaphorically expressed proposition.11 And if nothing is said, then it is
also not clear how something else can be meant. Furthermore, the argument also
shows that the sense in which ‘the metaphorical depends on the literal’ cannot be
that the metaphorical interpretation (meaning, proposition) of an utterance U of a
sentence S depends on the proposition expressed by the sentence S were it interpreted
literally. Even if each of the constituent expressions in S retains or has (in some
sense) its literal meaning, there may be no compositionally determined literal
proposition, or truth-condition, for the whole sentence—at least one that we
understand.
Let me return now to the Contextualist explanation of metaphor: if it is not
secondary meaning, could it be a case of M-non-literal meaning generated by a
primary pragmatic process like loosening? A word first about this pragmatic process.
The operation of loosening is employed by Recanati (2004), Carston (2002), and
Bezuidenhout (2001) among others, to explain first and foremost the ‘loose use’ of
language, a kind of non-literal talk first studied by Sperber and Wilson (S&W)
(1985–6) such as:
(13) I live in Paris (uttered by someone who lives one block outside the city
limits).
(14) There is a rectangle of lawn in front of the house (where the opposite
sides are not equal in length and the corner angles not each 90
degrees).
(15) The dorm floor is silent (despite some background noise).
(16) = (5) The dish is raw (not fully cooked).
S&W went on to claim that metaphor differs from loose use only in degree and
therefore should be given the same explanation, but they did not themselves
employ the pragmatic process of loosening to explain either phenomenon.12 Only
more recent Relevance-theorists, following Carston (1996; 2002), have appealed
to the actual process of loosening in order to develop a ‘unified’ approach that
treats enriching (or narrowing the truth-conditions) of utterances (such as (1) and
(4)) and loosening (or broadening their truth-conditions) in one uniform way.
With both, a non-linguistically controlled pragmatic operation (free enrichment or
loosening) is applied to the linguistically encoded proposition or concept to generate
the communicated proposition or concept with its intuitive truth-conditions.
11
For a recent attempt to defend the Gricean assumption that something is said by the
categorical falsehoods like (11), see Camp, 2004.
12
S&W (1985–6; 1986/95, 224–237) appeal to notions of interpretive representation and
propositional resemblance to explain both loose use and metaphor. Apart from the fact that
they assume that one thing can represent another simply in virtue of resembling it despite
strong counter-arguments that resemblance is neither necessary nor sufficient for representation
(see Goodman, 1976), Carston (2002, pp. 320–349) has objected to their asymmetric
treatment of loose use and enriched (narrowed) use.
© 2006 The Author
254 J. Stern
Specifically, loosening takes the encoded concept expressed by the uttered

expression and replaces it with an ad hoc concept, with a broader extension; the
resulting proposition is what is intuitively understood by hearers and was presumably
what the speaker intended to communicate.
But what exactly does the operation of loosening involve? Carston speaks of
‘relaxing’ or ‘weakening’ the lexically encoded meaning (1996; 2002, p. 336). Recanati
(2004) says that there is loosening ‘whenever a condition of application packed into
the concept literally expressed by a predicate is contextually dropped’ (p. 26). This
characterization is similar to the idea of feature-cancellation, proposed years ago to
explain metaphor (L. J. Cohen and Margalit, 1972; L. J. Cohen, 1993; and Matthews,
1971), and it works fine for a sub-set of loose uses, e.g. for mathematically ideal
predicates like ‘rectangle’ in (14); or graduated predicates like ‘raw’ in (16), meaning
not fully cooked (instead of fully non-cooked), or ‘silent’ in (15), meaning not noisy
above a contextual threshold (instead of totally absent all sound).13 These are cases of
what we might call ‘pure’ loose use, and the loosening involved is nothing more than
canceling, or relaxing, a degree of satisfaction required for literal use of the predicate.
However, Carston also uses ‘loose use’ to refer to interpretations that are not
simply the result of feature-cancellation or standard-relaxation. The ad hoc concept
raw* implicates, not only that something is insufficiently cooked but also that it
is inedible and disgusting and the speaker wants it replaced by another better done
dish of the same type. Let’s call this a case of ‘impure’ loose use whose ad hoc
concept raw* results, not just from feature cancellation but by including a variety
of ‘contextual implications’. The rub is that we are told nothing about how these
implications are derived from context in a way that comports with loosening.
Even on a liberal interpretation of the encyclopedic information encoded in a
lexical entry, it is implausible that, say, the features of arousing disgust or being
something the speaker wants replaced are included in the encyclopedic knowledge
associated with the literal sense of ‘raw,’ as Carston’s theory predicts. But if these
features are not part of the literal lexical entry of ‘raw’ to begin with, how can they
come to be expressed through a process like loosening?14
Carson’s use of loosening already, then, has problems in explaining loose use, but they
get worse when loosening is appropriated to account for metaphors such as (11) or:
(18) (= (7)) Robert is a bulldozer (meaning: aggressive, insensitive, persistent,

obstinate, intolerant of others’ ideas).
(19) Little Harriet is a piglet (meaning: filthy yet endearing).
Since these predicates are not graduated, their metaphorical interpretations also
cannot be generated by loosening taken simply as relaxation of a standard or
dropping a condition. In the case of the category-crossing (18), the various
13
See Carston, 2002, pp. 367–8, n. 3 who first acknowledges the similarity of her account to
Cohen but then identifies differences that render the similarity superficial, pp. 372, n. 12.
14
Cf. Stern, 2000, pp. 240–242 on Cohen’s feature-cancellation account of metaphor.
© 2006 The Author
psychological properties in the ad hoc concept also cannot be drawn from the literal
lexical entry of ‘is a bulldozer’, including its encyclopedic features. In his analysis
of (11), Recanati (2004) refers to a second stage in the primary pragmatic process
of loosening in which the interpreter ‘unreflectively constructs the sense in which
the ATM can be said to “swallow” the card by adjusting the meaning of the word
to the situation talked about’ (p. 76; cf. also 26). But he tells us nothing more about
this second stage of ‘adjustment’ and ‘construction’. Neither do S&W (1986), who
characterize a parallel second stage as selection from among strong and weak
implicatures, nor Carston (1997) who writes that ‘certain elements in the
encyclopedic entry of BULLDOZER [in (18)] which are highly accessible in the
particular context are carried over into the construction of the new concept which, as a
constituent of the explicature, warrants the particular implicatures derived’ (2002,
p. 350), without telling us what it is to be ‘highly accessible’ or how the elements
will be ‘carried over’ or the new concept ‘constructed’.
The problem here is not simply lack of explicitness; the real difficulty is that there
is no way to individuate pragmatically derived ad hoc concepts unless every feature
difference constitutes a different concept. But in that case, as we shall next see, the
pragmatic operation of loosening over-generates metaphorical interpretations,
differences of interpretation that are not reflected in our intuitive judgments.15
Furthermore, Carston (2002, p. 321) (appealing to Fodor, 1998) treats each
pragmatically derived ad hoc (as well as lexically encoded) concept as an atomic
concept. But that makes it difficult to account for metaphorical/ literal dependence
and the way in which the literal remains ‘active’ in metaphorical interpretation.
Although the label for the pragmatically processed ad hoc concept that is the
propositional constituent for ‘bulldozer’ in (18) is written ‘bulldozer*’, ‘bulldozer’ no
more occurs in ‘bulldozer*’ than ‘doze’ occurs in ‘bulldozer’. Nor can she reconstruct
the active metaphorical-literal dependence in terms of the ‘on-line processing’ history
of the metaphorically-loosened concept, or its derivation from a literal encoded
concept, so long as she holds, as does Recanati (2004), that primary pragmatic
processing is sub-personal, unconscious, and inaccessible to the speaker-interpreter.
To illustrate the problem of overgeneration consider Bezuidenhout’s (2001)
analysis of the metaphorical interpretation of (20):
(20) Quine demolished Carnap’s argument.
which, she argues, is not simply the result of feature-cancellation (e.g. of the
patient theme of the verb, say, [+physical) but also imports all sorts of extra-
linguistic presuppositions:
Although the verb ‘to destroy’ [sic] is not subcategorized for an obligatory
following prepositional phrase, talk of demolishing things will bring to
15
Cf. Stern, 2000, pp. 13, 33ff.; for a similar objection, see Stanley, 2002.
© 2006 The Author
256 J. Stern
mind various sorts of instruments and implements used in demolition jobs.

We may think of wrecking balls and dynamite and huge earth moving
equipment. What specifically we think of will depend on the size of the
thing to be torn down. A dam wall or a huge building will probably call
for large amounts of dynamite, but if it is simply a rickety old wooden
outhouse that needs to be torn down, then probably a single person
wielding an axe can accomplish the task … . Carnap was a good philosopher
whose arguments were generally well constructed. So refuting his arguments
would be no easy matter. Presumably Quine was up to the task, being
himself no slouch. But Quine and Carnap were also pretty well matched in
intellect. For Quine to demolish Anne’s arguments would be an easy
matter, requiring no heavy-duty logical apparatus. In the case of Carnap,
Quine might have to muster argumentative techniques of a more
sophisticated sort (Bezuidenhout, 2001, 174-5).
From this Bezuidenhout concludes that ‘by relying on such non-linguistic

assumptions, we arrive at an ad hoc concept demolish*’ which is more much than
the result of feature-cancellation.16
It would be more accurate to say that we arrive at any number of different
ad hoc concepts among which we cannot clearly distinguish one from another;
hence, our intuitive judgments about the truth of utterances expressing them will
be equally unclear. Although ‘demolish’ does not obligatorily demand a thematic
role for an instrument, Bezuidenhout says that it can ‘bring to mind’ wrecking
balls, dynamite, and axes, and which one will depend on the identities of the
demolisher and demolished. True, but it also brings to mind potential dangers to
bystanders and the environment, nostalgia for the destroyed past, a lot of dust and
dirt, hard-hats and muscles, the personality-types of demolition workers, and so
on. Which features, among these innumerable things brought to mind, belong to
the interpretation, ad hoc concept, or truth-conditions of (20)? Not clear. Suppose,
as Bezuidenhout suggests, that the ad hoc concept demolish* in (20) includes the
feature of having mustered sophisticated techniques to refute Carnap, like the
argument of ‘Truth by Convention’ employing Huttington’s postulates for
16
Bezuidenhout cites this example as an objection to a distinction drawn in Stern (2000, p. 27)
between extended metaphors like (20) whose interpretation is claimed to result just from
feature-cancellation (which is context-independent insofar as the features that constitute its
interpretation are not drawn from contextual presuppositions) and transferred metaphors like:
(21) Juliet is the sun.
whose interpretations are context-dependent in that they draw on extra-linguistic
presuppositions and skills for the features that figure in their content. That distinction does
not exclude Bezuidenhout’s interpretation of (20) since it would not count as an extended
metaphor by its definition of the term, but her interpretation is also no evidence against the
distinction. To counter the distinction, Bezuidenhout would have to show that there is no
possible (pure) extended metaphorical interpretations of sentences like (20).
© 2006 The Author
Euclidean geometry and Lukasiewitcz’ three postulates for the propositional

calculus. In a counterfactual circumstance in which Quine were to ‘demolish’
Carnap by one simple counterexample, would the utterance of (20) under
Bezuidenhout’s intuitive metaphorical interpretation be true or false? If every
difference of feature counts as a different ad hoc concept, hence, content, the answer
should be False. But then we end up with too many concepts. And if not every
difference counts, we have yet to be told a criterion to distinguish those that
matter.
One way to distinguish among contents (including the Contextualist’s ad hoc
concepts) is through the possibility of copying or deletion involved in verb
phrase anaphora or ellipsis. The general semantic constraint governing verb
phrase anaphoric relations whose antecedent admits multiple interpretations is
that the anaphor must always be given the same interpretation as the antecedent.
In Stern, 2000 I employ this test in order to motivate underlying abstract
semantic structure that would distinguish different semantic types (characters, in
Kaplan’s (1989) terminology) and contents for literal and metaphorical
interpretations of (what would superficially appear to be) one expression. I argue
that we find similar unacceptability judgments with mixed literal/metaphorical
anaphoric relations, and even with anaphoric relations between different
metaphorical interpretations (contents) of one expression (type), as we find with
mixed literal/literal anaphoric relations. This suggests that the same semantic
constraint governs both. But that means that, in mixed literal/metaphorical
utterances, what is superficially one expression with different (literal or
metaphorical) uses is in fact two expressions of different types. Suppose ‘is a fish’
has its literal interpretation in:
(22) The shark in Jaws is a fish.
and different metaphorical interpretations in (23) and (24):
(23) Mark Spitz is a fish (e.g. at home in water).

(24) Richard Nixon is a fish (e.g. slippery, hard to catch, and smelly).
Now consider the following examples of verb phrase anaphora:
(25) *The shark in Jaws is a fish, and Mark Spitz/Richard Nixon is, too.
(26) */? Nixon is a fish, and Spitz is, too.
Both (25) and (26) seem to violate the semantic constraint that requires that the
antecedent and anaphor have the same interpretation when the one is copied onto
the other. In (25) where the interpretation of the antecedent is literal and the
anaphor metaphorical (on pain of categorical absurdity), the violation is of type/
character. In (26) (which is more acceptable to informants), we have two different
metaphorical contents (concepts) for one metaphorical expression type (with one
© 2006 The Author
258 J. Stern
character).17 Whether the violation is of character/type or of content/concept

seems to affect our intuitions of unacceptability: it is possible to cancel the implication
of univocality at the level of content, as in (26.1), but not at the level of character/
type, as in (25.1), suggesting that (26) is closer to a pragmatic violation.
(25.1) *The shark in Jaws is a fish, and Mark Spitz/Richard Nixon is, too—
but not in the same sense/way.
(26.1) Nixon is a fish, and Spitz is, too—but not in the same sense/way.
Note that, even while (25) and (26) are ill-formed, we try 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 gives away their underlying
semantical ill-formedness.18
Against this argument, Bezuidenhout (2001, pp. 171, 181) objects that there is a
parallel pattern of infelicitous crossed interpretations employing literal and loose
use. Insofar as those cases should be explained pragmatically (and, on her and
Carston’s view, metaphor is a variant of loose use), there is no need to go semantic
17
In the terminology of Stern, 2000, the type/character of the metaphorical anaphor in (25) is
‘lexicalized’ by employing the ‘metaphorical expression’ ‘Mthat[“is a fish”]’ for the occurrence
of ‘is a fish’ interpreted metaphorically. Hence, the different types, with different characters,
of the antecedent (‘is a fish’) and (italicized) purported anaphor, which is made explicit in
logical form as:
(25.2) * The shark in Jaws is a fish, and Spitz/Nixon Mthat[‘is a fish’].
In (26) the antecedent and anaphor share the same character/type but have different contents
(relative to different sets of contextual presuppositions:
(26.2) *Nixon Mthat[‘is a fish’], and Spitz Mthat[‘is a fish’].
So, in (25) the violation is at the level of character; in (26) at the level of content.
18
Note that the indefinite description ‘a fish’ in (25)–(26) shows that Camp (2005, p. 721) is
wrong to object that the semantic ill-formedness of the examples used in Stern, 2000 is due
to the fact that ‘a single definite description [“is the sun”] [is used] to refer to two distinct
objects in the same context’, and that the same intuitions do not follow with indefinite
descriptions. A similar response applies to Bezuidenhout (2001, 180). Camp also objects that
infelicities like (25) and (26) also arise with examples like
(25.2) The shark in Jaws is a fish. And Mark Spitz is a fish.
(26.2) Nixon is a fish. And Spitz is a fish.
which cross sentence boundaries and therefore need not obey a syntactic constraint forcing a
single reading for both tokens of “is a fish”’ (p. 722). Oddly, my own intuitions find (25.2)
more acceptable than (26.2), possibly because the two articulated terms involve different
types/characters. But I agree with Camp that the shift in content in (26.2), which would turn
on shifting contextual presuppositions, makes its unacceptability much closer to a pragmatic
infelicity; cf. Stern, 2000 for similar remarks. Finally, as Camp notes, there can be non-
semantic sources for the infelicity of many metaphorical utterances like ‘Rosaline is the milky
mysterious moon. And a bursting atomic bomb is the sun’ whose juxtaposed metaphors rest
on different background assumptions. But these examples have no logical bearing on my
argument for underlying semantic structure.
© 2006 The Author
to explain (25) and (26) nor need we posit underlying semantic structure. In reply,
the first question we need to ask is whether there is in fact a parallel pattern of
ill-formed crossed literal/loose interpretations employing the ad hoc concepts
generated by the Contextualist process of loosening. Consider the literal use of ‘is
raw’ in (27), its loose use in (28), and the mixed sentence (29):
(27) That slab of fresh slaughtered meat, dripping blood, is raw (totally
uncooked).
(28) The steak the waiter brought is raw (undercooked).
(29) ?That slab of fresh slaughtered meat, dripping blood, is raw, and the
steak the waiter brought is, too.
My own intuitions about (29) are equivocal. If ‘is raw’ is given a pure loose use (i.e.
only the degree of satisfaction of the predicate is relaxed), (29) sounds acceptable.
If it is not a pure loose use, then I tend to read into the antecedent some of the
encyclopedic features that are conveyed by the anaphor, again making (29) more
though not fully acceptable. Note, too, that we can form comparatives between
literal and loose uses, e.g.:
(30) The fresh slaughtered meat, dripping blood, is more raw than my well-
done steak.
which would also suggest a single concept.

Nonetheless, let’s suppose for the sake of argument that (29), with its mixed
literal/loose interpretations, is unacceptable. Suppose also that Quine moonlighted
as a hard-hat construction worker to supplement his paltry philosopher’s salary.
On Bezuidenhout’s explanation, the literal (31) and metaphorical (32) should also
be analyzed as instances of strict and loose (or loosened) use, respectively.
(31) Quine demolished Carnap’s old house.

(32) Quine demolished Carnap’s argument.
From which it should follow that (33) is unacceptable:
(33) *Quine demolished Carnap’s old house and his argument, too.
But suppose that Quine had to use all the sophisticated tools of ‘Truth by Convention’
to demolish Carnap, but only a quick observation to point out a blatant non sequitur
in Anne’s argument. Given Bezuidenhout’s apparent criteria for individuating
concepts, the ad hoc concept in (32) also ought to differ from that in:
(34) Quine demolished Anne’s argument.
Yet, what she ought to count, then, as the mixed metaphorical:

© 2006 The Author
260 J. Stern
(35) Quine demolished Carnap’s argument, and Anne’s too,
sounds fine. Or suppose little Anne, like David smiting Goliath with his slingshot,
refuted Carnap with a powerful, elegant little objection:
(36) Quine demolished Carnap’s argument, and Anne did too.
On Bezuidenhout’s Contextualist criteria for distinguishing ad hoc concepts, (36)

contains two different metaphorically expressed ad hoc concepts; yet the anaphora
is unobjectionable. Thus, not every difference of ad hoc concept (individuated
according to Contextualist criteria) manifests itself in our intuitive judgments
concerning possible VP anaphora. On my view, (33) does indeed cross two different
types with different characters, hence, it is unacceptable. (35), on the other hand,
involves one metaphorical character and, since the instrument thematic role is not
obligatory, the differences between the instruments of demolition does not yield an
intuitive difference in truth-conditions or content, hence, it is acceptable.
The moral I draw from these examples is that the Contextualist explanation of
metaphor based on loosening overgenerates ad hoc concepts and therefore
metaphorical interpretations. If anaphoric relations hold only between occurrences
of one semantic type, or expressions with the same character, and if loosened use of
an expression (type) is still a use of the same expression (type), then mixed literal/
loosened (hence, metaphoric) anaphoric relations ought to obtain. Furthermore,
unlike Bezuidenhout’s examples, only differences in the values assigned to obligatory
thematic roles of a metaphorically interpreted expression seem to distinguish among
contents or concepts. Certain aspects of metaphorical interpretation thereby show
themselves to be autonomous of—precisely because they serve as constraints on—
interpreters’ metaphorically relevant intentions, beliefs, and mental associations. It is
not at all clear how general formal conditions of these kinds can be formulated in
terms of use, intentions, or mutual beliefs and expectations.19 The phenomena thus
suggest that there is semantic structure underlying metaphorical interpretation of
the kind that interacts in general with syntax and logical form, e.g. with the
conditions that govern verb phrase anaphora. But if so, contrary to Contexualism,
we must attribute underlying, phonetically unrealized semantic structure to
metaphors in order for the requisite conditions to apply.
To conclude this section, I want to propose that loose use does not run parallel
to metaphor but perpendicular to it. (13)–(16) are all examples of sentences that
19
Camp (2005) claims that it is possible to explain the ‘infelicity’ of (25) and (26) pragmatically
in terms of what hearers are led to expect of the anaphor on hearing the antecedent and in
terms of relevant resemblances. Her vague albeit brief suggestions do not, however, begin to
do justice to the generality of the formal constraints in question that abstract away from the
substance of the expectations and resemblances. And the more general and formal one makes
such pragmatic explanations the less distinguishable will they be from semantic conditions.
On Camp’s comparison of her pragmatic explanation to my Grice-like rules of ‘interpretation’
(Inconsistency and Redundancy), see Stern, 2000, pp. 141–3.
© 2006 The Author
admit both strict and loose use of an expression interpreted literally. But there are
also strict (literal) and loose uses of the same expression when it is interpreted
metaphorically. For example, ‘is raw’ interpreted metaphorically in:
(37) The student’s thoughts in the first draft of his paper were raw,
can be used either strictly (i.e. not worked out at all) or loosely (i.e. less than fully
worked out). Moreover, the mixed literal/metaphorical (38) and (39) are both
unacceptable even if we understand each antecedent-anaphor pair either exclusively
strictly or exclusively loosely.
(38) *That slab of meat dripping blood is raw, and so are Jack’s thoughts.
(39) *The steak Jack ordered is raw, and so are the thoughts in his final draft.
In short, the strict (literal)/loose and literal/metaphorical distinctions do not

coincide. And in neither (38) nor (39) can we cancel the violation by adding ‘but
not in the same sense’ which suggests that the difference in literal/metaphorical
interpretation of the expressions is at the level of their types (characters) rather than
simply in their concepts (contents), as the Contextualist explanation in terms of
loose use would imply.
4. A Literalist Theory of Metaphor
In sections 2 and 3 I raised two main objections to Contextualist accounts of

metaphor: (1) they cannot account for the intuitive non-literalness that distinguishes
metaphor from other pragmatically processed (M-non-literal) interpretations and
(2) their explanations over-generate metaphorical interpretations characterized as
atomic ad hoc concepts. Furthermore, Contextualists do not seem to have the
resources to explain how metaphor ‘depends’ on the literal. For the same reason,
they also cannot distinguish between information conveyed by the metaphorical
mode of expression that properly belongs to the truth-conditions of an utterance
in a context, and information that does not (see the second and third cases at the
end of section 2). Unless they treat metaphor as a secondary meaning (which
requires recognizing a prior literal meaning), there is no place for information
conveyed by literal meaning in their typology.
Two motivations for Literalism are to capture this extra-propositional literal
information carried by metaphorical utterances and to express constraints on
metaphorical interpretation that block over-generation. Its point is not to insist, come
what may, that linguistic meaning determines, or exhausts, propositional content or
what is intuitively said by an utterance (Recanati, 2004, p. 160ff.). Rather it aims to
show how underlying linguistic structure both constrains the expression of speakers’
intentions and enables the expression of more information than that conveyed in the
truth-conditions of what is explicitly articulated in the actual sentence uttered.
© 2006 The Author
262 J. Stern
The language resources, or types of linguistic representations, that will suffice

for this job are a matter for empirical investigation. Meaning, or interpretation, is
necessarily the meaning, or interpretation, of a symbol or representational form,
and semantics is concerned specifically with speakers’ knowledge of the principles
of meaning or interpretation of linguistic forms in virtue of which they are able to
communicate. Either these representations consist exclusively of linguistic elements
or they are hybrids of linguistic and contextual elements. Some of them are the
concrete events of human speech, sentences uttered or structures close to the
surface of utterances of which speakers are conscious; others contain linguistic
structures and syntactic information that are considerably richer and more abstract.
Without documenting the point, empirical linguistic research over the previous
half-century has shown the existence of such abstract linguistic structure to be
necessary (and highly productive) in order to explain speakers’ rich understanding
of speech. I shall therefore assume as a general principle that something like the
underlying linguistic level of logical form, which includes phonetically unrealized
variables, structures, and thematic information, is necessary for adequate semantic
interpretation and is accessible, even if not consciously, to speakers and
interpreters.20
This general conception of the role of linguistic structure in interpretation
oriented my semantic account of metaphor in Stern, 2000. Its central idea is that
metaphorical interpretation can be treated as a kind of context-dependent
interpretation on the classical model of demonstratives and indexicals. By analogy
to David Kaplan’s (1989) Dthat-descriptions that ‘lexically’ represent the
demonstrative interpretations of definite descriptions, I proposed that an
expression ␾ interpreted metaphorically should be analyzed as the ‘metaphorical
expression’ ‘Mthat[␾]’. ‘Mthat’ is an operator at the level of logical form which,
when prefixed to a (literal) expression ␾, yields a context-sensitive expression
‘Mthat[␾]’ whose tokens in each context c express a set of properties presupposed
in c to be m(etaphorically)-associated with the expression ␾, such that the
proposition <… {Mthat[␾]}(c)…. > is either true or false at a circumstance.21
The set of properties presupposed to be m-associated with ␾ are so associated in
a variety of ways: either as a stereotypical or normal property of ␾’s, through
20
It is not obvious that Contextualists share this assumption. Recanati’s conception of
Minimalism invariably takes the relevant linguistic meaning that determines his minimal
propositions to be that read off the surface structures of sentences uttered in concrete speech,
ignoring the role of abstract understood elements at a ‘deeper’ or more abstract level of
linguistic representation. This assumption is especially evident in his discussion of Travis’
examples (p. 149, n. 46); cf. Stern, 2003.
21
According to Kaplan’s (1989) semantics for demonstratives, for each expression Ø, {Ø}
denotes the character of Ø; {Ø}(c) denotes the content of Ø in the context c; and {Ø}(c)(w)
the extension (referent, truth-value) of Ø uttered (and interpreted) in c and evaluated at the
world w. Thus, if Ø is a predicate, then {Ø}(c) denotes a property and {Ø}(c)(w(c)) denotes
the extension of Ø uttered in the context c at (the world of) that context.
© 2006 The Author
exemplification by ␾’s, or in more idiosyncratic ways. For example, (21) (repeated

here):
(21) Juliet is the sun,
in which ‘is the sun’ is interpreted metaphorically, will now be represented at the
level of its logical form as containing the corresponding metaphorical expression:
(21.1) Juliet Mthat[‘is the sun’].
The metaphorical expression ‘Mthat[␾]’ has a non-constant character, i.e. a

meaning or rule (or function) that in different contexts yields different contents,
i.e. different truth-conditional factors (for predicates, properties). Thus, knowing
the character of a metaphor is like mastering a general skill of interpretation. It is
a rule or function mastered by speakers that enables them, for each expression ␾
that can be interpreted metaphorically in a sentence S, to map the properties
presupposed to be metaphorically associated with ␾ in c into a subset of properties
that determines the truth-value of utterances of S in c. Exactly how that rule is
applied to each context (including identification of the relevant m-associated
properties) is not specified in the semantics but in the pragmatics. However, we
can distinguish two sets of presuppositions at work in each context of metaphorical
interpretation. The first set of p(roductive)-presuppositions is associated with the
vehicle interpreted metaphorically; these include presuppositions about the
m-associated properties, and it is from these presuppositions that all the properties
that are metaphorically expressed by Mthat[␾] are drawn. The second set of
f(ilter)-presuppositions is associated with the linguistic and extra-linguistic
environment of the metaphorical expression; its function is to filter out properties
generated by the p-presuppositions that cannot be appropriately said in the
context.22
With this formal apparatus, we can also now begin to explain how the
metaphorical ‘depends’ on the literal and why it remains ‘active’ in the metaphorical
interpretation—although this explanation will also reveal a limitation of our
semantic theory. As we have just said, the m-associated presuppositions on which
the character, or meaning, of a metaphorical expression Mthat[␾] operates are
individuated according to the expression ␾ (the literal vehicle), not according to the
extension (referent) or intension (property) of ␾.23 So, one set of presupposed
properties will be m-associated with the (literal) term ‘dog’, another with (the
22
See Reinhart, 1970. These two kinds of interpretive presuppositions should be distinguished
from those in light of which we judge whether what is said by the metaphor is an appropriate
thing to assert/question/request, etc. in that context.
23
Stern, 2000, pp. 108, 221, 293-4. Although the m-associated presuppositions for
exemplificational metaphors involve properties presupposed to be exemplified by the referent
of the literal vehicle, which property is exemplified also depends on how the referent is
(qualitatively) presented by the literal vehicle (Stern, 2000, p. 155).
© 2006 The Author
264 J. Stern
co-referential and even co-intensional) ‘canine’; one with ‘mongrel’, another with
‘cur’; one with ‘sweat’, another with ‘perspiration’; one with ‘Tully’, another with
‘Cicero’ in metaphors like:
(40) Bush is no Cicero;

(41) Bush is a Tully,
thereby yielding their different metaphorical interpretations. This is the root idea
behind our explanation of metaphorical-literal dependence.
Now, when we turn to articulate this idea in terms of our semantic theory,
the natural move is to look to the character of the (literal) vehicle ␾ which is
interpreted metaphorically to work out the relativity of the m-associated
presuppositions to the expression (with its literal meaning). And prima facie
this is the right direction in which to turn. Neither the content nor extension
of Mthat[␾] is compositionally determined by the content or extension of ␾,
but the character of Mthat[␾] is individuated by the character of ␾. Two
metaphorical expressions Mthat[␾] and Mthat[∂] have different characters iff ␾
and ∂ have different characters. Moreover, this is the case despite the fact that
the character of Mthat[␾] is semantically simple insofar as it is parametric (like
an indexical, e.g. ‘I’); Mthat[␾] ‘depends’ on (the character of the literal vehicle)
␾ because the set of m-associated presuppositions to which it is parametrically
sensitive is itself individuated by (the character of) ␾. So, the formal notion of
character prima facie does the needed work to express the expression-relativity
of the relevant presuppositions on which the metaphorical interpretation
depends.
However, this explanation assumes that the notion of character is fine-grained
enough to capture all differences among expressions, or their linguistic meanings,
that are necessary to distinguish among different sets of m-associated presuppositions
related to those expressions. This is not always true. The strengths of the notion
of character are its power to capture and constrain context-dependence and
that it is much closer than truth-conditional content to linguistic meaning,
hence, to the expression itself. But its weakness is that it does not generalize
well to context-independent expressions. Recall that all expressions, not only
demonstratives and indexicals, are assigned characters (as well as contents and
referents). But context-independent, or eternal, expressions like ‘dog’ have
constant (or stable) characters that determine the same content in all contexts.
Thus any two eternal expressions that have the same content in all contexts will
also have equivalent characters. And to make matters even worse: if the content
of a directly referential term like a proper name just is its referent (or, for a simple
predicate, a property), then a constant character determines the same content and
the same referent in all contexts—and all three notions, character, content, and
referent, in Kaplan’s words, collapse onto one another (Kaplan, 1989, p. 562;
Stern, 2000, p. 293). It follows that pairs of eternal, simple, co-directly referring
proper names (‘Tully’ and ‘Cicero) or general expressions (‘cur’ and ‘mongrel’)
© 2006 The Author
will have the same referent and content in all contexts and, hence, the same
character. But this leads to a serious problem for our semantics. For if the character
of the metaphorical expression Mthat[␾] is individuated by the character of ␾,
and if ␾ is an eternal expression that co-directly refers with another eternal
expression ␮, then ␾ and ␮ also have the same content in all contexts, hence, the
same (constant) character. Hence, Mthat[␾] and Mthat[␮] will also have the same
character. But insofar as each of ‘cur’ and ‘mongrel,’ or ‘Cicero’ and ‘Tully’, have
different meanings manifest in their associations to different sets of presupposed
m-associated properties, the identical characters of ‘Mthat[␾]’ and Mthat[␮] will be
inadequate to distinguish between them, to pick out the different presupposed
sets of properties.24
In sum, there is a tension here between our philosophical claim that the relevant
set of m-associated presuppositions is individuated by (the linguistic meaning of)
the literal vehicle, the expression or word, not by the property it expresses or by its
extension, and our formal semantics that employs a notion of character that is not
fine-grained enough to capture the relevant notion of meaning to distinguish
among all (eternal) expressions. One might draw different morals from, or propose
different solutions to, this problem. One alternative is to assign different characters
to any two different expressions insofar as there is some contextual parameter
(namely, presuppositions) to which they are differently sensitive (at least in the
context of the Mthat-operator). But this would yield properties (as their contents)
that are more fine-grained than the possible worlds analysis that underlies Kaplan’s
semantics. A second alternative, drawn by Hills, 2002, concludes on the basis of
this problem that the analogy between demonstratives and metaphors breaks down.
This moral seems to me too drastic if only because demonstratives or, more
precisely, Kaplan’s dthat-descriptions suffer from the same problem. ‘Dthat[“the
oldest cur in the world on Dec. 12, 2005”]’ and ‘Dthat[“the oldest mongrel in the
world on Dec. 12, 2005”]’ have the same content and character; but they have
different meanings which will affect their behavior, and inter-substitutivity, among
other places, in attitudinal contexts. Instead, the moral I draw is that these problems
reflect the limitations of formalization in semantics to express intuitive philosophical
distinctions. The situation is not unique to metaphor. We cannot capture the very
real philosophical difference between rigid designation and direct reference in
possible worlds semantics, or the difference between singular propositions and
those that contain rigid conceptual complexes. Similar tensions also arise in our
analyses of quotation, direct and indirect (Stern, 2000, p. 103). In all these cases
either we can revise the philosophical claim in order to accommodate our formal
resources or we can hold onto the philosophical thesis, try to make it maximally
explicit and precise using available formal apparatus, but, where we cannot formally
24
This problem is forcefully presented in Hills, 2002 but it was already raised in Stern, 2000,
p. 293; my response here elaborates the line taken there.
© 2006 The Author
266 J. Stern
cash out a philosophically valuable intuition, we should acknowledge the limits of

formalization.
In any case, there are also several other important differences between metaphor
and demonstratives/indexicals and their respective semantic accounts. (i) Kaplan’s
logic of demonstratives was constructed with singular demonstrative terms in mind
(not, say, for predicate demonstratives), hence, the values of his demonstratives/
indexicals are individuals. The primary semantic unit of metaphor, on the other
hand, is the predicate, hence, the basic values are properties. (Cf. Nunberg, 1995.
Metaphorical singular terms are in turn analyzed in terms of predicates; see Stern,
2000, pp. 225–9.) (ii) Kaplan’s dthat-descriptions directly refer to the things denoted
by their embedded descriptions in their contexts; the properties directly expressed
by metaphorical expressions are not determined denotationally but instead, like the
referents of indexicals, parametrically. (iii) The contexts of Kaplan’s demonstratives
and indexicals are tuples of speakers, times, locations, and worlds; those of metaphor,
also include sets of presuppositions (i.e. propositions) or presupposed properties.
These differences notwithstanding, the analogy between metaphors and
indexicals/demonstratives rests on three parallels:
(1) For both demonstratives/indexicals and metaphors we must distinguish the
pre-semantic, semantic, and post-semantic roles of the context. In its first pre-
semantic role, context functions to assign sound events to expression-types, to
determine that an expression is to be interpreted one way or another, with one or
another particular type of meaning (character).25 For metaphor, in other words,
context in its pre-semantic role serves to determine that a given utterance is to be
interpreted metaphorically rather than literally, not what its interpretation is. In its
second semantic role, context and specifically the contextual presuppositions
function as the argument for the character of an expression, yielding its truth-
conditional content in the context. In its third, post-semantic role, context,
especially the facts or circumstances of the context (though also presuppositions
about those facts) enable us to evaluate whether the content is right or wrong to
assert in those circumstances. In most cases, this judgment of the metaphor depends,
at least in part, on its truth-value. But aptness, aesthetic value, and other criteria
for success may also enter at this stage.
25
Bezoudenhout, 2001 objects that since Ø and Mthat[Ø] are different expression types, I must
deny that ‘one and same expression type can be both literally and metaphorically interpreted’
(p. 178, my emphasis). True, but one and the same utterance can be assigned to both literal
and metaphorical types. Bezoudenhout also charges that to posit distinct literal and
metaphorical types is to make an ‘invidious distinction between literal and metaphorical
interpretation that does not seem to exist’ (p. 179). Now, I do make a distinction between
metaphorical and literal interpretations—so does everyone but deconstructionists—but there
is nothing ‘invidious’ about it. Literal and metaphorical expression-types are treated equally;
indeed since each has its own type, there is no discrimination against one or the other.
Moreover, contrary to what Bezuidenhout (p. 179) claims, ‘Mthat’ introduces, no ‘interpretive
mechanism’ that is not already necessary for demonstrative interpretation (e.g. ‘Dthat’); the
only cost is one additional entry in the lexicon.
© 2006 The Author
(2) The characters of indexicals (more so than demonstratives) and metaphors

function, not to self-sufficiently determine the content of their utterance, but both
to direct and to constrain the role of the (semantic) context insofar as it contributes
to content.26 For both indexicals/demonstratives and metaphors, it is therefore
crucial to distinguish the stage of interpretation (at which character + context
operate) from the stage of evaluation (at which content + circumstance operate).
We must distinguish between the content actually metaphorically/demonstratively
expressed by an utterance in its context and the content that it would metaphorically/
demonstratively express in some other context; and we must distinguish between
the truth-value of the proposition actually metaphorically/demonstratively expressed
were it evaluated at a counterfactual circumstance and the truth-value of the
propositional content it would have metaphorically/demonstratively expressed at
that counterfactual circumstance.
(3) Both indexicals/demonstratives and metaphors obey the Actual Context
Constraint (ACC). That is, the contents of both indexicals/demonstratives and
metaphors are always fixed relative to their context of utterance, even when
embedded in modal and attitudinal contexts.
In sum, this account of metaphor, like Literalist explanations of other phenomena,
is based on the assumption that extra-linguistic contextual effects on truth-
conditional content should be traced to an element in logical form or the syntactic
structure of the linguistic representation underlying the utterance.27 In the
remaining two sections, I address two Contextualist objections against this Literalist
account of metaphor.
26
On meaning, or character, as directions, see Strawson, 1950.
27
However, unlike some Literalist accounts, e.g., Stanley, 2000, the element in logical form
that I posit for metaphorical interpretation—the Mthat-expression—is a narrow, though
hybrid, demonstrative-indexical, not simply a variable. Stanley’s reason for taking the
underlying elements to be variables is in large measure because his arguments turn on the
possibility of quantificationally binding the underlying element, and only variables, not
narrow indexicals, are bindable. It is not clear whether such considerations would apply to
metaphor. However, I would also argue that even in Stanley’s examples, in order to capture
the explanatory and causal role of the underlying element, it is necessary to represent the
underlying element as an indexical, not simply as a variable. If my utterance of:
(2) It is raining,
is to explain why I am taking out my umbrella and raincoat, it is not enough to claim that
underlying it is:
(2’) It is raining at l;
what is needed is:
(2’’) It is raining here,
for reasons given in Perry, 1979. Stanley’s own reasons for insisting that narrow indexicals
are not bindable are also not clear. It is true that the overt indexicals in natural language do
not have bindable inflections or variants, but we use ‘he’, ‘then’, and ‘there’ with the meaning
of ‘I’, ‘now’, and ‘here’ in bound constructions and, with a little philosophical imagination,
we can invent formal devices for this purpose, e.g. Casteneda’s (1967) ‘he*’.
© 2006 The Author
268 J. Stern
5. Other Figures
The general form of the first Contextualist objection runs as follows: ‘The Literalist
claims that a certain linguistic argument or evidence L implies that metaphor is (at
least in part) semantic. But if so, the same argument or evidence L should entail
that another figure F also falls in semantics. However, even the Literalist claims
that F is pragmatic. Therefore L is not reason or evidence to take metaphor to be
semantic’.
For example, Camp (2005) argues that ‘it is difficult, if not impossible, to hear
just one conjunct of:
(42) John is a real genius, and Sam is too.
as ironic. But Stern dismisses a semantic analysis of irony out of hand’ (see Stern,
2000, pp. 232–3); therefore, she concludes that I ‘need an independent motivation
for the claim that [examples like (25) and (26) above] demand a distinctive, semantic
explanation’. Similarly, Stanley (2005) argues that if examples like (25) and (26)
support a semantic analysis of metaphor, similar reasoning should apply to:
(43) John: Bill is a fine friend.

Sally: Sue is too.
where John’s utterance is intended ironically, in which case ‘Sally’s utterance must
be understood ironically as well. But this does not show that irony is semantic [in
support of which he also cites Stern ibid.]… So, such ellipsis facts do not demonstrate
that a phenomenon is semantic’.
As Camp notes, examples of VP ellipsis like (25)–(26) function as evidence of
formal constraints on possible metaphorical interpretations that go unexplained by
use-theorists like Davidson (1986) who sees no interpretative regularities for
metaphor even to explain (Stern, 2000, pp. 69–71). As we mentioned in section 3,
one kind of constraint on VP anaphora requires sameness of character (violated in
cases of mixed literal/metaphorical anaphora like (25)), a second requires sameness
of content (violated in cases of mixed metaphorical/metaphorical content anaphora
like (26)). Only where the violation is of character do we strictly speaking see a
semantic constraint at work; violations of content like (26) are more pragmatic and
cancelable.
In contrast, irony is not semantic by anyone’s account. How, then, should it be
explained? In Stern, 2000 (pp. 232–8), I distinguish two general classes of figures:
M-figures (including metaphor, simile, metonymy) and I-figures (including irony,
meiosis, hyperbole, under/overstatement). Both classes are context-dependent.
But M-figures are operations on sentence-types that assign them non-constant
characters that, in turn, yield (for each of their tokens) contents in contexts while
I-figures are operations on contents (in contexts) to yield different contents (in
contexts). So, in a case of mixed literal/ironic anaphora, e.g., (42), the antecedent
© 2006 The Author
and (copied) anaphor will nonetheless have the same type (character) but different
contents—similar to the mixed metaphorical/metaphorical (content) (26). And in
both (26) and (42), it is possible to cancel the implication of univocality, as in
(26.1) and:
(42.1) John is a real genius, and Sam is too—but not in the same sense/way.
With mixed literal/metaphorical anaphora, however, the antecedent and anaphor

have different characters or types, a semantic difference that cannot be cancelled.
Therefore, Camp’s and Stanley’s arguments from irony do not suffice to show that
the ellipsis facts are not evidence that there is semantic structure underlying
metaphor.
Stanley (2005) also offers a second set of objections from another figure, ‘deferred
reference’, to show that metaphor is not semantic, e.g.,
(44) The ham sandwich is getting annoyed.
whose subject description refers to the person who ordered the ham sandwich. He
argues that, despite the fact that the deferred referent is the value of the antecedent
in anaphoric relations like:
(45) The ham sandwich wants his bill now.
and despite the fact that it is impossible to get mixed readings of VP anaphora as in:
(46) Bill served a ham sandwich, and John did too.
where one occurrence refers to the orderer and the other to the sandwich itself—
despite all this evidence, deferred reference is really not semantic. Why? His
argument goes like this: If this evidence shows that deferred reference is semantic,
it will also suffice to show that metaphor is semantic. And indeed he gives two
examples parallel to (45) and (46) using metaphor that are perfectly acceptable.
However, ‘On a standard view of metaphor, metaphor is not semantic’. So, if the
metaphorical meaning or referent is not part of the semantic content of its sentences,
the evidence from deferred reference ‘should not lead us to believe that deferred
reference is semantic’.
The classical name for deferred reference is metonymy, the figure in which a
term for one entity stands for an associated entity, such as:
(47) The pen is mightier than the sword.
which says that publishing is more effective than military force, or:
(48) Bush bombed Saddam.

© 2006 The Author
270 J. Stern
which says that the armed forces controlled by Bush bombed the country controlled
by Saddam Hussein.28 So, we might paraphrase Stanley’s argument as follows: If
evidence E shows that metonymy is semantic, then the same evidence should
suffice to show that metaphor is semantic. But ‘on a standard view’ metaphor is
not semantic. Therefore, metonymy is not semantic.
Stanley is right that metaphor and metonymy stand and fall together: both are
M-figures. But one philosopher’s modus tollens is another’s modus ponens. If it is
indeed the case, as the initial evidence suggests, that metonymy is semantic, that
would be further evidence that metaphor is semantic. Since Stanley gives no
evidence for the ‘standard view’ of metaphor, what are his arguments for denying
that deferred reference, or metonymy, is semantic, despite the prima facie evidence
that it is? (In what follows I will follow Stanley’s usage of ‘deferred reference’.)
Apart from the argument from irony already canvassed, he gives two main
‘theoretical’ arguments.29
First, one might think that deferred reference is semantic because the literal
meaning of the vehicle ‘the ham sandwich’ ‘provides a guide to its deferred
meaning’, the person who ordered the ham sandwich. However, Stanley objects,
the literal meaning of ‘x has a nice handwriting’ in a reference letter also ‘guides’
us to the implicature that x is a bad philosopher. In both cases, the literal meaning
is employed. But that does not show that it linguistically controls the deferred meaning
and, in order for the deferred meaning to count as semantic, its content must be
constrained by the literal meaning: in particular, the deferred referent must be
something of which the literally meant expression is true.
As I said in section 4, I agree that the function of linguistic meaning is to
constrain the speaker’s intentions, i.e. to demarcate the range of possible contents
that can be communicated by a given linguistic form. The issue over which Stanley
and I disagree is whether deferred reference is or is not linguistically constrained.
Surely mere ‘guidance by literal meaning’, stated so vaguely, is not sufficient to
render an interpretive phenomenon semantic. But Stanley’s example, Grice’s ‘x
has nice handwriting’ in a letter of reference, is irrelevant to metaphor, since its
implicature (that x is a bad philosopher) is a secondary meaning, calculated by an
inference from a prior proposition whose communication flouts a maxim. For
metaphors (and metonyms), as we saw in section 3, there is in general no prior
literal proposition that is the semantic content of the utterance, what it says, from
which an implicature to what the speaker meant (its metaphorical or deferred
28
Recanati, 2004, p. 26 calls this figure by the name of ‘semantic transfer’. Related figures to
metonymy are metalepsis in which we refer to something by means of a term for something
else to which it is remotely related, either causally or spatially, and synechdoche, which is
based on part/whole, individual member/class and material/thing relations.
29
Stanley also argues that, if an adequate theory of some phenomenon e.g. deferred reference
(or metaphor), turns out to be completely unconstrained by conventional meaning and
restricted only by pragmatics, then the phenomenon is not semantic; as proof, he cites Sag
(1981). Whether or not Sag’s is an ‘adequate’ theory, there may, of course, be others in the
literature Stanley does not discuss which involve semantic constraints.
© 2006 The Author
meaning) could be inferred or calculated. Stanley needs a better example of

non-semantic guidance to compare to metaphor.
Furthermore, despite Stanley’s assertion to the contrary, deferred reference (like
metaphor) is semantically constrained. Note, first, that deferred reference is
context-dependent. Depending on the context in which (44) is uttered, ‘the ham
sandwich’ might refer to the person who made or sold or ate the ham sandwich, not
just the one who ordered it. So, as a first stab, the definite description:
(49) The ham sandwich.
in (44) should be represented at the level of logical form by:
(49.1) (The x) (x R the ham sandwich).
where the value of the relation R can vary from context to context, e.g. taking
Ordering in one context, Making in another. To capture its context-
dependency, we might also mimic our analysis of metaphor, inventing an
operator ‘Def’ that maps the literal referent onto the contextually-determined
deferred referent.30
(49.2) (The x) (x Def[‘is a ham sandwich’].
Yet while the deferred meaning/referent may vary with context, once we fix its
interpretation in a context, it remains fixed. Not only can’t we get mixed literal/
metonymic interpretations of (46) where the antecedent is the sandwich and the
anaphor its orderer, we also can’t get mixed deferred/deferred interpretations
where the antecedent is interpreted as an orderer and the anaphor as a maker.
Likewise, ‘Two ham sandwiches are getting annoyed’ (Stanley, n. 4) cannot mean
that one ham sandwich orderer and one maker are getting annoyed; the two must
either be both orderers or both makers.31 If deferred reference was simply pragmatic,
not semantic, it is a good question why these mixed readings are impossible, no
matter what the particular mixed contents happen to be.
Stanley also claims that the literal meaning of ‘ham sandwich’ does not constrain
its deferred meaning/referent because the semantic content of ‘ham sandwich’
30
This is only a first approximation to a semantic analysis of deferred reference/ metonymy as
an operator on the model of ‘Mthat.’. Here I take the primary unit of metonymy, like
metaphor, to be a predicate, although the former is typically a singular term. In a full analysis,
as with nominative metaphor, we would presumably need to embed the description within
a Dthat-description in order to capture its rigidity; see Stern, 2000, pp. 225-9. It is also an
open question whether the Def operator takes the referent of the description as its argument,
or its character (as with ‘Mthat’.). My main interest here is to secure rather than save the
phenomena and to show that there are prima facie constraints on deferred reference/
metonymy.
31
As this example shows, metonymy is not limited to singular terms.
© 2006 The Author
272 J. Stern
need not be true of the deferred referent. True, but it does not follow that the
literal meaning of the phrase does not constrain its deferred referent. Suppose that
a certain restaurant’s specialty is ham sandwiches but it also has a clientele that
observes the Jewish dietary laws (which prohibit ham). So, the restaurant serves
them kosher vegetarian ham-substitute sandwiches that look and taste like the real
thing. To avoid confusing them, we can imagine a rule in practice in the restaurant
that says:
(50) Ham sandwiches sit on the right side of the restaurant, kosher
sandwiches on the left.
It is their consumers, not the sandwiches, who sit on the different sides, but what
is crucial to determining the identities of the deferred referents—and constrains
their respective semantic contents—is precisely the semantic content of the literal
terms. One might, in other words, say that deferred meaning, or metonymy, no
less than metaphor, depends on the literal meaning of its vehicle.
Stanley’s second objection is that any expression can be used metaphorically and
any term can be used for deferred reference. Therefore, neither has to do with the
‘semantics of any particular construction’. Instead, both metaphor and deferred
reference are uses of linguistic constructions with one semantic meaning to
communicate something different.
Yes and No. Not every expression can be used metaphorically to express just
anything and not every term can be used with deferred reference to refer to
anything—and this even apart from pragmatic constraints. As we have now seen,
there are semantic constraints on both metaphor and deferred reference imposed
by the meaning of the literal vehicle. As for the construction-specificity of metaphor
and deferred reference: Apart from the configurational constraints (exemplified by
(46)) on deferred co-reference, the objection assumes that the only constructions
to be taken into account are those phonetically realized in the surface structures of
sentences uttered, i.e. articulated constituents. But if our analysis of metaphor and
metonymy is on the right track, there are abstract constructions at the level of
logical form specific to both metaphorical and metonymical interpretation, no
different in principle from the abstract constructions posited by Literalists (e.g.
Stanley, 2000, 2002; Stanley and Szabo, 2000) to constrain other contextual effects
on semantic content.
In sum, a close examination of deferred reference confirms rather than refutes
the hypothesis that metaphorical interpretation falls in part under semantics.
6. The Actual Context Constraint
Even a Contextualist like Recanati (2004, p. 159) concedes that there are some
expressions in natural language whose referential value is context-dependent and
determined semantically, or linguistically, rather than pragmatically. Let’s grant
© 2006 The Author
him that this class of expressions is smaller, even much smaller, than many assume,
that it includes only the pure indexicals (‘I’, ‘now’, ‘here’), not even demonstratives
(‘he’, ‘that’) whose value depends on the speaker’s referential intentions, and let’s
also grant that this linguistic process (‘referential saturation’) is not representative
of any general way in which context contributes to or shapes content or
truth-conditions. With all of these qualifications in hand, let’s focus on one feature
of this special case of saturation, what I earlier called the Actual Context Constraint
(ACC): namely, that the referential value or content of a pure indexical is always
fixed relative to features of its actual context of utterance, and contributes an
invariant element to the proposition expressed, regardless of the circumstance at
which that proposition is evaluated, hence, even when the indexical is embedded
in a modal context.32
If metaphors also obey the ACC, that would be powerful evidence in favor
of a Literalist theory of metaphor according to which, underlying metaphorical
interpretation, there is an abstract indexical-like linguistic form that imposes this
constraint. In Stern (2000) I argued that we do see the ACC in action in
examples like:
(51) Juliet might be the sun.
which is true in the circumstance of its context of utterance just in case there is
some possible circumstance in which she has whatever feature is actually
metaphorically expressed by ‘is the sun’ in its speaker’s or interpreter’s context,
relative to our actual m-associated presuppositions related to ‘is the sun’ (and
‘Juliet’). It is not enough for the actual truth of (51) that there be some possible
circumstance in which whatever would be expressed metaphorically by ‘is the sun’
in that possible circumstance is true of Juliet in that circumstance.
Camp (2005) has recently challenged both the ACC applied to metaphor and
the assumption that we can,
distinguish suppositions that specify the circumstances of evaluation from

‘actual’ contextual presuppositions. Without a way to tease these two sets
of suppositions apart, though, there is no force to Stern’s claim that ‘like
the character of an indexical, the character of the metaphor constrains its
content by limiting its contextual source to actual presuppositions—even
when we evaluate its truth in non-actual circumstances’ (Stern, 2000,
p. 197) (p. 726).
32
On the ACC, see Kaplan, 1989 for whom it is a direct consequence of the directly referential
status of demonstratives and indexicals. Kaplan dubs any operator that would shift the
saturation of an indexical to some other non-actual context a ‘monster’, i.e. a device utterly
alien to, indeed impossible in, natural language. For a recent challenge to the ACC, see
Schlenker, 2003. I intend to discuss his critique elsewhere but his objections, as far as I can
see, do not bear on the present argument.
© 2006 The Author
274 J. Stern
Her counterexamples are subjunctive conditionals in which the antecedent specifies

counterfactual presuppositions (hence, a non-actual context) relative to which a
metaphor in the consequent is interpreted. For example, suppose Paris replies to
Romeo’s (21) with (52):
(52) (i) As matters stand, Juliet’s no sun.

(ii) If she had married me, though, she might have been the sun.
(iii) She could have been the leading lady of Verona, and all the other
ladies would have deferred to her.
Camp supposes that the content contributed by ‘is the sun’ (in (ii)) is very different
than it is for Romeo’s utterance (21).33 Romeo treats Juliet’s virtues, like the sun’s
light, as ‘natural and original’; according to Camp, ‘Paris’s utterance attributes the
possibility of a highly artificial importance to her, roughly of being the center of
Veronese social life’. She proposes that Paris’s ‘imagined-Juliet raises different
resemblances with the sun to salience, and thereby determines a different
metaphorical content, than actual-Juliet does’ (p. 725)—contrary to the ACC. She
also gives a second counterexample in which Benvolio responds with
(53) If we were living after a nuclear apocalypse, then Juliet could have
been the sun. After all, she’s pale, cold, and distant, and that’s how the
sun would be then. But I can say she sure isn’t the Tuscan sun.
Again, Benvolio’s utterance contributes a content metaphorically expressed

by ‘is the sun’ fixed relative to the counterfactual presuppositions, or context,
specified in the antecedent of the conditional, not relative to the presuppositions
m-associated with ‘is the sun’ in its actual context, hence, contrary to the ACC.
In reply, I would argue that neither (52) nor (53) violates the ACC (even if
Camp is right about their interpretations) because the counterfactual subjunctive
antecedent of a conditional alters the prior context of interpretation. That is, with
all such conditionals, first we ‘actualize’—add to the initial actual context set of
presuppositions—the subjunctively expressed antecedent of the conditional and,
then, cued by its subjunctive mood which is a conventional device (in English) for
indicating that prior presuppositions to the contrary are being (temporarily)
suspended, simultaneously drop conflicting prior contextual presuppositions. (For
33
I have added the small roman numerals for expository purposes. Camp does not say how she
interprets (52i). Presumably Paris is not simply denying that Juliet is literally the sun. Is he then
denying that she is the sun, under its metaphorical interpretation relative to Romeo’s context?
Call that interpretation of the metaphor P. In that case, ‘As matters stand’ would seem to be
referring to the circumstances of evaluation, not the context of interpretation. So (48i) states
that in the present circumstances of evaluation, Juliet is not P—contrary to Romeo’s assertion,
both agreeing on the content of (21) in that context. As a follow-up to (52i), the antecedent of
the conditional in (ii) would be best understood then, contrary to Camp’s analysis, as supposing
a counterfactual circumstance of evaluation, not a counterfactual context of interpretation.
© 2006 The Author
more details, see Stern, 2000, pp. 207–8, following Stalnaker, 1968, 1972.) Relative
to the altered—what is then the actual—context of utterance, the metaphorical
interpretation of the consequent of the conditional obeys the ACC.
But Camp’s two counterexamples (52) and (53) also differ from each other
in a way that illustrates how one might solve the more general problem she
raises as to how to ‘tease apart’ those presuppositions that specify the
circumstances of evaluation from those that constitute the ‘actual’ context of
interpretation. Note that the antecedent of (52) supposes a counterfactual claim
associated, not with the literal vehicle for the metaphor, ‘the sun’, but rather
with the subject ‘Juliet’, part of the linguistic environment of the metaphor. In
(53), the counterfactual presuppositions are ones associated with ‘the sun’. This
difference parallels the distinction we drew in section 4 between two sets of
presuppositions in the interpretive process: the p(roductive)-presuppositions
associated with the literal vehicle for the metaphor and the f(ilter)-presuppositons
associated with its environment. The p-presuppositions supply the pool of
properties or associations that are candidates for interpretations of the metaphor.
However, since those presuppositions over-generate, the f-presuppositions filter
out inappropriate candidates; the f-presuppositions themselves do not introduce
new properties.
Now, the antecedent of (52) ascribes counterfactual properties to Juliet, so it is
related to the environment of the metaphor and, if it adds a supposition to the
interpretation of the metaphor, it ought to fall among the f-presuppositions that
promote and demote candidate properties but do not themselves add new ones. It
is very hard to see, then, how the counterfactual supposition of (52) could yield
the significantly different metaphorical interpretation of ‘is the sun’ purported by
Camp.34 What seems more plausible is that the antecedent of (52) is not functioning
at all as part of the interpretive context that generates interpretations, but rather as
part of the evaluative context in which we determine whether the content
expressed by the metaphor is true or false. Given a more or less shared interpretation
P of ‘is the sun’ (something like being unparalleled among her peers and being the
center to whom everyone else defers), the disagreement between Romeo and
Paris is over the question whether, as Romeo says, the proposition that Juliet is P
is actually true or whether it would only be true in the counterfactual circumstance
Paris entertains. On the other hand, the antecedent of (53), which supposes
34
In order to get the very different properties Camp wants to express of Juliet, what she needs
is really a different metaphor with different associations, e.g. this reply by Paris:
As matters stand, Juliet’s no sun. If she had married me, though, and become the leading
lady of Verona, all the other ladies deferring to her, she would have been the Tiffany lamp
to outshine all others..
Here Paris first denies the truth of the metaphor (21) with the content it expresses in Romeo’s
context. Then he proposes a different metaphor, with a different content (say, the kinds of
artificial social properties Camp suggests), which relative to the counterfactual circumstances
stated in its antecedent, is true of Juliet.
© 2006 The Author
276 J. Stern
counterfactual properties associated with the metaphorical vehicle ‘the sun’, adds
new p-presuppositions to the initial context, thus enabling a very different
interpretation for the metaphor relative to the altered actual context.
In sum, (52) and (53) do not show that the ACC does not hold for metaphor.
And by paying close attention to the distinctions between p- and f-presuppositons
and between interpretive and evaluative presuppositions, we can begin to sort out
what belongs to the context of interpretation and what belongs to that of
evaluation.35
I will conclude with two additional alleged counterexamples to the ACC, which
is equivalent to the claim that metaphors cannot be bound by an operator:
(54) In no remotely conceivable way am I a rose!36
This sentence is ambiguous. On one reading it says that there is no possible

(= conceivable) way in which the world might be in which I am a rose, where ‘am
a rose’ is interpreted metaphorically in its actual context of utterance c, relative to
the presupposed m-associated properties for ‘is a rose’ that obtain in its actual
context c.
(54.1) —(᭚ w)({I}(c)(w) ⑀ {Mthat[‘is a rose’]}(c)(w)
Here the metaphorical interpretation is fixed according to the ACC and is not
bound by an operator. On a second reading, (54) means that there is no possible set
of presuppositions, that is, there is no possible context, in which ‘I am a rose’ could
be interpreted to express a truth in the circumstance of that same context. That is,
(54.2) —(᭚ c)({I}(c)(w(c)) ⑀ {Mthat[‘is a rose’]}(c)(w(c)))
Again, no violation of the ACC.

A second interesting example is
(55) Juliet and Achilles are each diamonds in their own different ways.
which Camp (2005) says should be unacceptable if the metaphorical interpretation

of ‘diamond’ is determined semantically ‘because the very same occurrence of
‘diamond’ must be assigned multiple contents simultaneously’ (p. 721). For reasons
of space, I cannot give a full analysis of the logical form of (55) which would
involve syntactic details (about ‘each’ and binding) beyond the scope of this paper.
But an informal account will suffice to show why Camp’s conclusion does not
35
Camp (2005) also criticizes my analysis of metaphors in belief- and attitudinal-contexts which
conform to the ACC. Reasons of space prohibit a full response to her criticism but a full
discussion would have to take into account Schlenker, 2003; see n. 24 above)
36
I owe this example to David Hills.
© 2006 The Author
follow. Suppose that ‘being a diamond in a way’ (interpreted metaphorically) is to

be (diamond-like) precious or (diamond-like) unscratchable or (diamond-like)
sparkling and so on, where each of these is interpreted metaphorically to express
some way Wi. Let c be the context of utterance of (55), namely a context set of
p-presuppositions m-associated with ‘is a diamond’ and f-presuppositions related
respectively to ‘Juliet’ and ‘Achilles’. What (55) says is that:
(55.1) (᭚ Wi) (᭚ Wj) (Wi = {Mthat[‘is a diamond’]}(c) &
Wj = {Mthat[‘is a diamond’]}(c) & Wi ¹ Wj & Wij & Wja)
On this analysis, (55) is interpreted relative to one, the actual, context set of
presuppositions, and the different f-presuppositions related to ‘Juliet’ and ‘Achilles’
account for the different ways in which each of them is (metaphorically) a diamond.
Thus, one, we avoid violating the ACC and, two, we avoid having to assign
simultaneously multiple contents to the one occurrence of the metaphor because
no particular content is assigned. We merely assert that there exist the two different
ways that are the content of ‘is a diamond’ when metaphorically ascribed to Juliet
and Achilles, respectively. The only remaining question is how a Contextualist
could possibly analyze (55) without the resources of the character of the underlying
metaphorical expression ‘Mthat[␾]’ that is so essential to the Literalist explanation.
Department of Philosophy
University of Chicago
References
Bezuidenhout, A. 2001: Metaphor and what is said: a defense of a direct expression

view of metaphor. In P. A. French & H. K. Wettstein (eds), Midwest Studies in
Philosophy (Vol. 25): Figurative Language. Boston, MA and Oxford, UK: Blackwell,
pp. 156–186.
Black, M. 1962: Metaphor. In M. Black. Models and Metaphors. Ithaca, N.Y.: Cornell
University Press, pp. 25–47.
Black, M. 1993: More About Metaphor. In A. Ortony (ed.), Metaphor and Thought.
Cambridge: Cambridge University Press, pp. 19–41.
Blackburn, S. 1984: Spreading the Word. Oxford: Clarendon Press.
Camp, E. 2004: The generality constraint and categorial restrictions. The Philosophical
Quarterly, 54, 209–231.
Camp, E. 2005: Metaphors and demonstratives: Josef Stern’s Metaphor in Context. Noûs
39, 4, 715–31.
Carston, R. 1996: Enrichment and loosening: complementary processes in deriving
the proposition expressed? UCL Working Papers in Linguistics, 8, 205–32.
Carston, R. 2002: Thoughts and Utterances: The Pragmatics of Explicit Communication.
London: Blackwell.
© 2006 The Author
278 J. Stern
Casteneda, H-N. 1967: Indicators and quasi-indicators. American Philosophical Quarterly, 4:2.
Cavell, S. 1967: Aesthetic problems of modern philosophy. In M. Black (ed.), Philosophy
in America. Ithaca: Cornell University Press, pp. 74–97.
Cohen, L. J. 1993: The semantics of metaphor. In A. Ortony (ed.), Metaphor and
Thought. 2nd edn. Cambridge: Cambridge University Press, pp. 58–70.
Cohen, L. J. and Margalit, A. 1972: The role of inductive reasoning in the interpretation
of metaphor. In D. Davidson and G. Harman (eds), Semantics of Natural Language.
Dordrecht: Reidel, pp. 721–762.
Cohen, T. 1975: Figurative language and figurative acts. Journal of Philosophy, 72, 669–690.
Cohen, T. 1976: Notes on metaphor. Journal of Aesthetics and Art Criticism, 34, 249–259.
Danto, A. 1978: Freudian explanations and the language of the unconscious. In J. H.
Smith (ed.), Psychoanalysis and Language. New Haven: Yale University Press,
pp. 325–353.
Davidson, D. 1984: What metaphors mean. In D. Davidson, Inquiries into Truth and
Interpretation. Oxford: Oxford University Press, pp. 245–264.
Davidson, D. 1986: A nice derangement of epitaphs. In E. LePore (ed.), Truth and
Interpretation: Perspectives on the Philosophy of Donald Davidson. Oxford: Basil
Blackwell Ltd, pp. 433–446.
Fodor, J. 1998: Concepts: Where Cognitive Science Went Wrong. Oxford: Clarendon.
Goodman, N. 1976: Languages of Art. 2nd edn. Indianapolis: Hackett.
Grice, H. P. 1975: Logic and conversation. In D. Davidson and G. Harman (eds), The
Logic of Grammar. Encino, CA: Dickenson, pp. 64–75.
Hills, D. 1997: Aptness and Truth in Verbal Metaphor. Philosophical Topics, 25, 1, 117–154.
Hills, D. 2002: Review of Josef Stern, Metaphor in Context. Philosophical Review, 111,
23, 473–78.
Kaplan, D. 1989: Demonstratives. In J. Almog, J. Perry, and H. Wettstein (eds), Themes
from Kaplan, Oxford: Oxford University Press, pp. 481–563.
Levinson, J. 2001: Who’s afraid of a paraphrase? Theoria, 67, 7–23.
Lycan, W. G 1999: Philosophy of Language: A Contemporary Introduction. London:
Routledge.
Matthews, R. 1971: Concerning a ‘linguistic theory’ of metaphor. Foundations of
Language, 7, 413–425.
Nunberg, G. 1995: Transfers of meaning. Journal of Semantics, 12, 109–132.
Recanati, F. 2004: Literal Meaning. Cambridge: Cambridge University Press.
Reinhart, T. 1970: On understanding poetic metaphor. Poetics, 5, 383–402.
Rorty, R. 1987: Unfamiliar noises I: Hesse and Davidson on metaphor. Proceedings of
the Aristotelian Society, supp. volume 61, 283–296.
Schlenker, P. 2003: A Plea for Monsters. Linguistics and Philosophy, 26, 29–120.
Searle, J. 1978: Literal meaning. Erkenntnis, 13, 207–224.
Searle, J. 1993: Metaphor. In A. Ortony (ed.), Metaphor and Thought. 2nd edn.
Cambridge: Cambridge University Press. pp. 83–111.
Sperber, D. and Wilson, D. 1985/86: Loose talk. Proceedings of the Aristotelian Society,
86, 153–171.
© 2006 The Author

Sperber, D. and Wilson, D. 1995: Relevance: Communication and Cognition. 2nd edn.
Oxford: Basil Blackwell.
Stalnaker, R. 1968: A theory of conditionals. In N. Rescher (ed.), Studies in Logical
Theory. Oxford: Blackwell, pp. 98–112.
Stalnaker, R. 1972: Pragmatics. In D. Davidson and G. Harman (eds), Semantics of
Natural Language. Dordrecht: Reidel, 380–397.
Stanley, J. 2000: Context and logical form. Linguistics and Philosophy, 23, 391–434.
Stanley, J. 2002: Making it articulated. Mind & Language, 17, 1&2, 149–168.
Stanley, J. 2005: Semantics in context. In G. Preyer and G. Peters, (eds), Contextualism
in Philosophy: Knowledge, Meaning, and Truth. Oxford: Oxford University Press,
pp. 221–54.
Stanley, J. and King, J. 2005: Semantics, pragmatics, and the role of semantic content.
In Z. Szabo (ed.), Semantics vs. Pragmatics. Oxford: Oxford University Press,
pp. 111–164.
Stanley, J. and Szabo, Z. 2000: On quantifer domain restriction. Mind & Language,
15, 219–261.
Stern, J. 1983: Metaphor and grammatical deviance. Noûs, 17, 577–99.
Stern, J. 2000: Metaphor in Context. Cambridge, MA: MIT Press.
Stern, J. 2003: Review of Charles Travis, Unshadowed Thought. Mind, Oct., 805–811.
Strawson, P. F. 1950: On referring. Mind, 59, 320–344.
Travis, C. 2000: Unshadowed Thought. Cambridge, MA, Harvard University Press.
White, R. M. 1996: The Structure of Metaphor: the Way the Language of Metaphor Works.
Oxford: Basil Blackwell.
White, R. M. 2001: Literal meaning and ‘figurative meaning’. Theoria, 67, 24–59.
© 2006 The Author


Metaphor, Literal, Literalism: Josef Stern

Uploaded by

Metaphor, Literal, Literalism: Josef Stern

Uploaded by

Metaphor, Literal, Literalism

1. Metaphor Among the Non-literals

Address for correspondence: Department of Philosophy, University of Chicago, 1115 E. 58th

However, all the classic pragmatic or use-oriented characterizations of

(1) I had breakfast.

(1*) I had breakfast (today/on some contextually indicated day).

(2*) It’s raining (here/at some contextually indicated location l).

Everyone agrees that the intuitively understood truth-conditions, or content, of

2. The Problem of Metaphor for Contextualism

One claim that present-day Contextualists and Literalists—unlike many writers in

2b. M(inimal)-non-literal meaning is the meaning of an utterance that is a product

(4a) They got married and had children.

(5a) This dish is raw.

and semantic transfer:

(6a) The ham sandwich is getting restless.

(7) Robert is a bulldozer.

(8) Kripke is an alchemist.

Second, there is meaning or informational content carried by a metaphor that is

(9) I won’t swallow that.

The truth-conditions of (9) as uttered by Marie referring to her mother’s

(10) Marie won’t obey her mother’s interdiction.

(11) The ATM swallowed my credit card.

‘ﬁgurative, we have only to belabor the metaphor to make it conspicuous, as in

metaphors, by drawing out their metaphorical schematic relatives we manifest our

3. Metaphor, Loose Use, and Overgeneration

Speciﬁcally, loosening takes the encoded concept expressed by the uttered

(18) (= (7)) Robert is a bulldozer (meaning: aggressive, insensitive, persistent,

(20) Quine demolished Carnap’s argument.

mind various sorts of instruments and implements used in demolition jobs.

From this Bezuidenhout concludes that ‘by relying on such non-linguistic

Euclidean geometry and Lukasiewitcz’ three postulates for the propositional

(22) The shark in Jaws is a ﬁsh.

and different metaphorical interpretations in (23) and (24):

(23) Mark Spitz is a ﬁsh (e.g. at home in water).

Now consider the following examples of verb phrase anaphora:

character).17 Whether the violation is of character/type or of content/concept

which would also suggest a single concept.

(31) Quine demolished Carnap’s old house.

From which it should follow that (33) is unacceptable:

(34) Quine demolished Anne’s argument.

Yet, what she ought to count, then, as the mixed metaphorical:

(35) Quine demolished Carnap’s argument, and Anne’s too,

(36) Quine demolished Carnap’s argument, and Anne did too.

On Bezuidenhout’s Contextualist criteria for distinguishing ad hoc concepts, (36)

In short, the strict (literal)/loose and literal/metaphorical distinctions do not

4. A Literalist Theory of Metaphor

In sections 2 and 3 I raised two main objections to Contextualist accounts of

The language resources, or types of linguistic representations, that will sufﬁce

exempliﬁcation by ␾’s, or in more idiosyncratic ways. For example, (21) (repeated

(21) Juliet is the sun,

(21.1) Juliet Mthat[‘is the sun’].

The metaphorical expression ‘Mthat[␾]’ has a non-constant character, i.e. a

(40) Bush is no Cicero;

cash out a philosophically valuable intuition, we should acknowledge the limits of

(2) The characters of indexicals (more so than demonstratives) and metaphors

(42) John is a real genius, and Sam is too.

(43) John: Bill is a ﬁne friend.

With mixed literal/metaphorical anaphora, however, the antecedent and anaphor

(44) The ham sandwich is getting annoyed.

(45) The ham sandwich wants his bill now.

(46) Bill served a ham sandwich, and John did too.

(47) The pen is mightier than the sword.

(48) Bush bombed Saddam.

meaning) could be inferred or calculated. Stanley needs a better example of

(49) The ham sandwich.

in (44) should be represented at the level of logical form by:

(49.1) (The x) (x R the ham sandwich).

(49.2) (The x) (x Def[‘is a ham sandwich’].

6. The Actual Context Constraint

(51) Juliet might be the sun.

distinguish suppositions that specify the circumstances of evaluation from

Her counterexamples are subjunctive conditionals in which the antecedent speciﬁes

(52) (i) As matters stand, Juliet’s no sun.

Again, Benvolio’s utterance contributes a content metaphorically expressed

(54) In no remotely conceivable way am I a rose!36

This sentence is ambiguous. On one reading it says that there is no possible