Journal of Pragmatics 34 (2002) 969991

www.elsevier.com/locate/pragma

What is said
Patrick Hawley*
Department of Linguistics and Philosophy, MIT, 77 Massachusetts Ave.,
Cambridge, MA 02139, USA
Received 20 November 2000; received in revised form 14 February 2002

Abstract
A misunderstanding of Grices distinction between saying and implicating leads at least one
theorist to misconstrue the pragmatics/semantics distinction. I clarify the Gricean picture,
hoping to shed light on debates about the relationship between pragmatics and semantics.
This paper begins with a presentation of Grices theoretical distinction between saying and
implicating, emphasizing its grounding in the intuitive distinction between conveying something
literally and directly, and merely suggesting or hinting it. I point out that someperhaps most
followers of Grice believe that, in some way, what is implicated depends on what is said. F.
Recanati (Recanati, Francois, 1993. Direct Reference: From Language to Thought. Blackwell, Oxford, UK.) is one example. The thought seems to be that the hearer in a conversation
needs to use what is said in a calculation to determine what is implicated. After speculating
that Griceans who accept the dependency claim are unduly focused on literal uses of language, I argue that the dependency claim is mistaken. Rejecting dependency undermines
Recanatis argument for a particular way of separating semantics from pragmatics. I conclude
by exploring how what is said and what is implicated may be construed without dependency.
# 2002 Elsevier Science B.V. All rights reserved.
Keywords: Conversational implicature; Grice; Recanati; Semantics/pragmatics distinction; Semantic
underdetermination; Availability principle

1.
Suppose that I am sitting in a room with a friend. The sun is setting, and the room
is growing darker. I want my friend to turn on the light. I could just say so directly.
Friend, I might say, I want you to turn on the light. Or I could be more indirect, relying on hints or suggestions. I may say My, its getting dark in here, or
exclaim I cant see you anymore! Are you still there?
* Corresponding author.
E-mail address: phawley@mit.edu (P. Hawley).
0378-2166/02/$ - see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S0378-2166(02)00043-7

970

P. Hawley / Journal of Pragmatics 34 (2002) 969991

This dierencethe dierence between conveying something literally and directly,

and merely suggesting, hinting or implying itis clear and intuitive. And the dierence is not simply a dierence between rudeness and politeness. When I say something literally and directly, what I mean seems closely tied to the meanings of the
words I use and how they are put together. When I say something literally and
directly, someone who understands my words, understands me. But when I am
hinting or suggesting, my words alone wont tell you what I mean. If I say to my
friend My, its getting dark in here, my words have nothing to do with electric
light switches or my wants. My friend could understand my words and not realize
that I want him to turn on the light. My friend could understand my words but fail
to understand me.
Paul Grice developed an inuential account of this intuitive dierence. (Grice,
1989a) According to Grice, there is a distinction between saying and implicating.
When a speaker makes an utterance, what is said is what is directly or literally conveyed, and what is implicated is what is suggested, hinted or implied. Recall Grices
oft-repeated example of a philosophy professor writing a job recommendation letter
on behalf of one of his students. The professor writes The candidate is prompt and
has excellent penmanship, and nothing more. The professor does not mention the
candidates philosophical abilities. Clearly, the professor is trying to convey something other than he literally said, namely that the candidate is bad at philosophy.
The professor said that the candidate is prompt and has excellent penmanship, and
implicated that the candidate is bad at philosophy. In the previous example, saying
that its dark in the room implicates that I want my friend to turn on the light.
Although Grice does not draw the distinction with precision, its intuitive appeal is
evident.
To make the distinction a little clearer, we should consider what Grice means by
the words saying and implicating. Grice uses the word implicate as a term of art,
standing in for words like hint or suggest or imply. The things I implicate are
called implicatures. Saying (in Grices sense) is supposed to be used in much the
same way as it ordinarily is. After all, the distinction is drawn intuitively; if we
intuitively think that the professor says that the candidate is bad at philosophy, then
what is said is that the candidate is bad at philosophy. What is said is also supposed
to be closely related to the conventional meaning of the words. . . uttered. (Grice,
1989c: 25) As I mentioned earlier, when someone says something literally and
directly, what they mean seems closely tied to the words they use and how they are
put together. Note, however, that saying, like implicating, is a technical term for
Grice. Grice admits that saying in his favored sense is in some degree articial, slightly diering from ordinary uses of the word say. (Grice, 1989f: 118) The
dierences are for greater theoretical utility. (Grice, 1989f: 121) Although Grices
say is used, generally, as we intuitively think it ought to be used, Grice apparently
permits that in judging what is said, we may sometimes set aside intuitions about
what someone says in favor of what would be theoretically useful.
Beyond the intuitive appeal, Grice makes the distinction between what is said and
what is implicated compelling by sketching an account of one kind of implicature
conversational implicature. Conversational implicatures are present, according to

P. Hawley / Journal of Pragmatics 34 (2002) 969991

971

Grice, because hearers presume that speakers are rational and cooperative. In order
to maintain this presumption, hearers may draw conclusions about what a speaker is
implicating. As Grice puts it: what is [conversationally] implicated is what is
required that one assume a speaker to think in order to preserve the assumption that
he is observing the Cooperative Principle (and perhaps some conversational maxims
as well). (Grice, 1989e: 86) The Cooperative Principle is to make your conversational contribution such as is required, at the stage at which it occurs, by the
accepted purpose or direction of the talk exchange in which you are engaged.
(Grice, 1989c: 26) Grice suggests a list of maxims that are supposed to constrain
conversation in accord with the Cooperative Principle.1
For example, in the recommendation letter case, the professor would not be a
cooperative conversational participant unless he were trying to convey something
about the candidates philosophical abilities. Since he does not mention them, then,
assuming that he is trying to be cooperative, he must be suggesting something. The
letters recipient, recognizing this, supposes that the professor is hinting that the
candidate is bad at philosophy.
One goal of this paper is to clarify the relationship between saying and implicating.
What is implicated apparently depends on what is said in the following way: what is
implicated is calculated by the hearer from what is said along with other features of
the context of utterance; as Grice puts it, what is implicated is built on what is
said. (Grice, 1989b: 49) This apparent dependence is, I will argue, only apparent.
Grice adds a further condition on conversational implicature. The speaker must
believe that (potential) hearers are in a position to recognize the implicature:
the presence of a conversational implicature must be capable of being worked out;
for even if it can in fact be intuitively grasped, unless the intuition is replaceable by
an argument, the implicature (if present at all) will not count as a conversational
implicature; it will be a conventional implicature. (Grice, 1989c: 31)
In other words, the speaker must believe that there is a rational path available to a
hearer to recognize the presence of a conversational implicature. Presumably, Grice
adds this condition because he thinks that conversations proceed rationally (because
he thinks that communication is a form of rational behavior).
The Gricean picture suggests a tidy division of labor. Semantics accounts for what
is said; general pragmatic principles account for conversational implicatures.
Separating semantics (a specic theory, tied to features of a particular language)
1

In Logic and Conversation, Grice denes conversational implicature thus: A man who, by (in, when)
saying (or making as if to say) that p has implicated that q, may be said to have conversationally implicated that q, provided that (1) he is presumed to be observing the conversational maxims, or at least the
Cooperative Principle; (2) the supposition that he is aware that, or thinks that, q is required in order to
make his saying or making as if to say p (or doing so in those terms) consistent with this presumption; and
(3) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the
competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (2) is
required. (Grice, 1989c: 3031). Note that in the Gricean framework, implicatures may either be conventional
or nonconventional. Conversational implicatures are a species of nonconventional implicature. In this
paper, I will not discuss so-called conventional implicatures. I only discuss conversational implicatures.

972

P. Hawley / Journal of Pragmatics 34 (2002) 969991

from pragmatics (a language independent general theory) has the appeal of making
the entire theoretical project seem more tractable. We have the appealing prospect of
a clean and simple semantics distinct from a general pragmatics.
Note that apart from capturing the intuitive distinction between conveying something directly and suggesting or hinting it, and apart from giving us an appealing,
simple separation between semantics and pragmatics, the Gricean story is compelling for another reason. The saying/implicating distinction may encapsulate the
tempting idea that there is something special and important about direct and literal
uses of language.
I will call this point of view the priority of the literal. This is not a specic thesis;
rather, it is a vague, general point of view.2 The believer in the priority of the literal
thinks that the literal meaning of words plays an essential role in making it possible
for someone to mean something by an utterance. The Gricean sympathetic to the
priority of the literal holds that the literal meaning of words plays an essential role
both in making saying possible and in making implicating possible. The Gricean
may (rather naturally) adopt the priority of the literal by holding that (1) the literal
meaning of words plays an essential role in making saying possible, and (2) saying
makes implicating possible. So, a Gricean sympathetic to the priority of the literal
may make a claim about the relationship between implicating and saying, namely
that, in some way, implicating depends on saying.
The priority of the literal is tempting for the Gricean because it meshes with two
observations. First, to understand what someone is saying, you start, it seems, by
understanding what their words literally mean. Second, to understand what someone is hinting or suggesting, it seems that you need to rst understand what they are
literally and directly saying; understanding what is implicated seems to require
understanding what is said.
The claim of dependency of implicating on saying may be worked out in more
than one way. For example, the Gricean might suppose that:
(A)understanding the explicature of an utterance is necessary for understanding the
conversational implicatures of that utterance.
(Explicature is a technical term that nicely parallels implicature. I use it in the
following way: the explicature of an utterance is what is said by the person in making that utterance. Note that others (like Carston (1988) and Sperber and Wilson,
(1995)) use this term dierently.) Claim (A) gives priority to what is said in terms of
understanding: the hearer must understand the direct and literal content of an
utterance before they can understand other things conveyed by an utterance. One
can imagine many other ways a Gricean could claim dependency between implicating and saying.
2
In the Retrospective Epilogue to Studies in the Way of Words, Grice suggests such a point of view
when he wonders whether there is any kind, type, mode, or region of signication which has special
claims to centrality, and so might oer itself as a core around which more peripheral cases of signication
might cluster, perhaps in a dependent posture. (Grice, 1989d: 359).

P. Hawley / Journal of Pragmatics 34 (2002) 969991

973

I am now in a position to be more specic about what I am trying to accomplish.

My general goal is to articulate and defend a view which rejects the priority of the
literal, while preserving a useful distinction between pragmatics and semantics. I will
not complete this task here. The more restricted goal of this paper is to show that
the Gricean may accept:
(1)

There is a useful and intuitive distinction between what is said and what is
implicated.

but reject:
(2)

What is implicated depends on what is said.

Since the Gricean already accepts (1), to justify this view, it will suce to justify
rejecting (2) and show that rejecting (2) is compatible with accepting (1). In this
paper, I will argue for the rejection of (2) by trying to show that obvious ways of
spelling out a proposed dependency will not work. Claim (A), for example, will not
work. Then, to show that rejecting (2) is compatible with accepting (1), I will sketch
a picture that rejects a dependency of what is implicated on what is said while
maintaining a distinction between them. Finally, I will discuss how the results of this
paper relate to debates about the semantics/pragmatics interface.
The views I am arguing against are not made of straw. Many Griceans indeed
hold that, in some way, what is implicated depends on what is said. In fact, says
Stephen Levinson, the Received View of Griceans is that what is said is the input to
the pragmatic reasoning responsible for output of implicatures: what is implicated is
calculated on the basis of what is said (together with aspects of how it was said, in
the case of Manner implicatures). (Levinson, 2000: 171)
Francois Recanati, for example, is a theorist who thinks that what is implicated
depends on what is said.3 In his book Direct Reference, Recanati (1993) maintains
something like claim (A), and uses this claim in an argument for a principle to distinguish what is said from what is implicated. If I am successful, I will have shown
that Recanati is mistaken in holding the (A)-ish claim, and I will have shown that we
should not accept the argument based on that claim.
3
Here are other examples: . . .conversational implicatures are only indirectly associated with the linguistic content of utterances. They are derived from the content of the sentences used. . . (Sadock, 1991:
366) [conversational] implicatures are not semantic inferences, but rather inferences based on both the
content of what has been said and some specic assumptions about the cooperative nature of ordinary
verbal interaction. (Levinson, 1983: 104) A conversational implicature. . . is an inference that derives
from what has been said in context taken together with some general background maxims of conversation. . . (Levinson, 1995: 92) Communication can be achieved by two dierent means: by encoding
or decoding messages or by providing evidence for intended inference about the communicators informative intention. Verbal communication, we argue, exploits both types of process. The linguistic meaning
of an utterance, recovered by specialized decoding processes, serves as the input to unspecialized central
inferential processes by which the speakers intentions are recognized. (Sperber and Wilson, 1987: 786).
One author tries to be more careful: An implicatum is completely separate from what is said and is
inferred from it (more precisely, from the saying of it). (Bach, 1994: 140).

974

P. Hawley / Journal of Pragmatics 34 (2002) 969991

This result bears on the distinction between pragmatics and semantics, since
Recanati uses his principle to argue for a controversial way of separating pragmatics
from semantics. He defends the thesis of semantic underdetermination, the claim
that pragmatic eects play a central role in determining what is said. This is stronger
than the claim, obvious to many, that what is said is not determined by semantic
properties alone (given a context). For many agree that word disambiguation, and
the xing of the references of terms, are to be described by general, language independent principles. However, Recanati and others (Carston, 1988) argue that,
beyond disambiguation and reference xing, the correct distinction between saying
and implicating forces us to conclude that pragmatic eects play a key, important
role in determining what is said. The possibility of semantic underdetermination
may be worrisome. If what is said is determined in large measure by the same
pragmatic means as what is implicated, then the distinction between semantics and
pragmatics is far more hazy than it initially appeared. Indeed, we may well wonder
whether the distinction between semantics and pragmatics collapses. However, if we
do not accept Recanatis criterion distinguishing what is said from what is implicated, then we should not accept his argument for the thesis of semantic underdetermination.
If Recanatis argument fails, that does not, of course, show that other arguments
in favor of the thesis of semantic underdetermination fail; thus my conclusion may
seem rather limited. However, my claim about the relation between what is implicated and what is said has broader interest. We can see this by considering Levinsons recent discussion of the semantics/pragmatics interface. (Levinson, 2000:
Chapter 3) Levinson describes the Gricean Received View as follows. According to
the Received View, semantics is prior to pragmatics: the output of semantic processes serves as input to pragmatic processes. (For example, conversational implicatures are calculated from what is said.) However, holders of the Received View
also acknowledge that the output of pragmatic processes can serve as input to
semantic processes. (For example, disambiguation, reference xing and indexical
resolution can sometimes occur by pragmatic processes.) Levinson calls this Grices
circle. This is indeed a circle, he argues; post-semantic pragmatic processes and presemantic pragmatic processes are sometimes the same kind of process. For example,
he argues, reference can be determined by the same kind of process that determines
conversational implicature. Levinson takes Grices circle as a paradoxical chickenand-egg problem which brings out questions about the pragmatics/semantics
interface. (Levinson, 2000: 172) The upshot of Levinsons discussion is that because
of Grices circle (along with other reasons) we should reject the Received View.
Levinson concludes that semantics is not prior to pragmatics; semantic and pragmatic processes are distinct but interwoven.
I would quarrel with Levinsons formulation of the problem. The problem of
semantic underdetermination is, in my opinion, not best seen as a puzzling circle,
but as a worry that a clear theoretical distinction between semantics and pragmatics
is blurred or even threatened. Still, accepting Levinsons framework for a moment,
we can see that the claim I am makingthat what is implicated is not dependent on
what is saidpoints to one way out of Grices circle. Pragmatic processes need

P. Hawley / Journal of Pragmatics 34 (2002) 969991

975

not be seen as dependent on semantic processes which are in turn dependent on

pragmatic processes and so on. Instead, there is a way out of the circle: pragmatic
processes may function without semantic input. Thus, I, like Levinson, reject the
Received View that semantics is prior to pragmatics. Our reasons, however, dier
substantially. I hope at some future time to more fully confront Levinsons rich and
complicated discussion, but perhaps these brief remarks are enough to make clear
that my conclusion here is not limited to Recanatis argument for semantic underdetermination. It bears more generally on discussions about the relation between
semantics and pragmatics.
I will now try to clarify the relationship between implicating and saying. I claim
that what is implicated is independent of what is said; so it is a mistake to think that
what is implicated depends on what is said. My claim is not that in no case does
what is implicated depend on what is said. For I think that it is obviously true that
in some cases what is implicated depends on what is said. Rather, my claim is that in
some cases what is conversationally implicated does not depend on what is said. To
justify this claim, that in some cases what is conversationally implicated does not
depend on what is said, I will need rst to be more explicit about what sort of
dependency I am denying, and then to produce some cases in which the distinction
between what is said and what is implicated is clear, and in which what is conversationally implicated does not depend on what is said.
Note that to assert that what is implicated is independent of what is said is not to
assert that implicating is independent of saying. Certainly, a person may implicate
something by saying something. When present, an implicature is determined by an
act of utterance.4 Additionally, and independently, an act of utterance determines
what, if anything, is said. Grice, rather obscurely, appears to make this very point:
the implicature is not carried by what is said but only by the saying of what is
said, or by putting it that way. (Grice, 1989: 39)
While Grice is not clear what it is for an act of saying to determine (to carry) an
implicature, he is clear that what is implicated is not determined (carried) by what
is said. Investigating the metaphysics of utterance actsespecially the distinction
between the act of saying something, and what is saidwould be helpful here. For
merely pointing out that what is said and what is implicated are both determined by
an utterance act does not yet reveal the relation between what is implicated and
what is said. We need more information about the connection between what is
implicated and an utterance act, and between what is said and an utterance act, in
4
I mean utterance in Grices articially wide sense: any case of doing x or producing x by the
performance of which U meant that so-and-so (Grice, 1989f: 118) But note that in this essay I am not
assuming Grices analysis of U meant that x in terms of reciprocal intentions. However, I am assuming
that the examples I present are all intuitively clear cases of a speaker meaning something. For present
purposes I leave open the question how to analyze speaker meaning. (It may turn out that, upon accepting
a particular account of speaker meaning, one or two of the cases I present turn out not to be cases of a
speaker meaning something. If so, my argument would not be undermined because the remaining cases
would be enough to support my conclusions. I present a battery of cases; the loss of one or two will not
matter.)

976

P. Hawley / Journal of Pragmatics 34 (2002) 969991

order to conclude something about the relation between what is implicated and what
is said. Analyzing the metaphysics of acts could help in developing a conceptual
argument concluding that what is implicated does not always depend on what is
said. Although a better understanding of the metaphysics of acts would likely be
helpful, I argue in this paper by counterexample; I will produce some cases in which,
I claim, what is conversationally implicated does not depend on what is said.
My claim is that what is implicated by an act of utterance is, in some cases, not
dependent on what is said by that act of utterance. This may seem obvious. However, as I have mentioned, many Griceans disagree, assumingsometimes tacitly
that all conversational implicatures depend on what is said. And, I believe, this
mistake leads them astray. If so, correcting this mistake matters, even if I am just
explicating Grices position. For example, Recanati explicitly states that conversational implicatures always depend on what is said:
Grice said that the presence of an implicature must be capable of being worked
out (Grice 1989c: 31) For an implicature to be worked out, two conditions
must be satised: (i) both what is said and what is implied must be grasped (ii)
the inferential connection between them must also be grasped. . . (Recanati,
1993: 245)5
According to Recanati, the hearer always grasps an inferential connection from
what is said to what is implicated. Recanati is apparently committed to (A), the
dependency claim I mentioned earlier:
(A)understanding the explicature of an utterance is necessary for understanding the
conversational implicatures of that utterance.
It is easy to come away with this picture after reading Grice, who sometimes
sounds as if he believes that what is conversationally implicated is always calculated
from what is said by means of simple inferences. At one point, Grice remarks that
what is implicated is built on what is said. (Grice, 1989b: 49)
(A) is too strong. Suppose John says Its hot in here, suggesting that he wants
someone to turn on the air conditioner. Alice hears him, understands the explicature,
5
Recanati thinks that both speaker and hearer must be capable of working out the implicatures. The
passage continues:. . .

Many followers of Grice have (wrongly) interpreted this as requiring that the theorist be capable of
working out whatever conversational implicature is posited to explain a given semantic phenomenon;
but Grice clearly had in mind the participants in the talk-exchange themselves: it is the speaker and
hearer who must be capable of working out the implicatures, and this entails that they have conscious
access both to what is said and what is implicated.
I am not sure why Recanati thinks the speaker needs to be capable of working out the implicatures because we
may want the speaker to be able to unintentionally implicate something. (Although this runs against Grices
view of speakers meaning as reducible to speakers intentions, I do not see why we need to hold Grices
position on meaning in order to accept his picture of conversational implicature.) In the text, I only take the
hearer into account in arguing against proposals for necessary conditions for the presence of a conversational
implicature. I do not think this matters because if the condition based on the hearer alone is not necessary,
then the (more restrictive) related condition based on both hearer and speaker is not necessary.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

977

and calculates the implicature on the basis of the explicature and other features of
the context. Alice gets up and turns on the air conditioner. Ben is there too, and
doesnt catch what John says, but seeing John sweating and Alice moving toward
the air conditioner also concludes that John wants someone to turn on the air conditioner. Ben understands the conversational implicature. However, Ben does not
understand the explicature.6
Recall that Grice says that
the presence of a conversational implicature must be capable of being worked
out; for even if it can in fact be intuitively grasped, unless the intuition is
replaceable by an argument, the implicature (if present at all) will not count as a
conversational implicature; it will be a conventional implicature. (Grice,
1989a: 31)
This suggests that what makes for a conversational implicature is, in part, that
there be a computational path from what is said to the implicature, not whether this
path is taken. So this leads to another proposal for dependency between what is
implicated and what is said:
(B) an utterance cannot express a conversational implicature unless it expresses an
explicature, and there is a reasonable path of inference, requiring the
explicature, for a hearer to calculate theimplicature.
Despite what he says, Recanati may hold (B) rather than the implausible (A). If
so, I still think he is mistaken; here are four examples to show that claim (B) is
incorrect.
1. An explicature may be present but not needed to calculate the conversational
implicature:
Suppose Fred and Ethel are having a discussion in the presence of a 5 year old
child. [cf Grice (1989c: 36)] Fred and Ethel are merrily talking, when suddenly,
Ethel lowers her voice to a whisper and says that the child is naughty. Fred does
not hear what Ethel says. Still, he reasons that Ethel has implicated that she
does not want the child to understand because Ethel has violated the conversational maxim of Manner, which says to avoid obscurity of expression.
The explicature is not needed in this reasoning; Ethel could have said almost
anything and the same conversational implicature would have been generated.
6
Perhaps someone may respond that whatever Ben understands, it is not a conversational implicature
because he has not reached his understanding in the right sort of wayhe has understood something with
the same content as the conversational implicature, but he has not understood it qua conversational
implicature. If the objection is that for Ben to understand it qua conversational implicature, Ben must
infer it on the basis of the explicature, then the objection begs the question because it assumes (A). But
even if the objector can hold on to (A) in the face of this example, the proposed counterexamples to (B)
(coming soon) also serve as counterexamples to (A).

978

P. Hawley / Journal of Pragmatics 34 (2002) 969991

(Another example: suppose you are lost in a place where English is rarely
understood. You walk up to a stranger and ask Do you speak English? He
replies Its nice weather were having today, isnt it? Plausibly, the stranger
has conversationally implicated that he speaks English.)
2. Its possible to generate a conversational implicature even though no explicature is expressed:
(a telephone conversation)
Sally:
What did you end up doing last night?
Jack:
Oh, we went to that new movie. Why didnt you come? Did you have a
ght with your husband again?
(pause. Sally says nothing.)
Jack:
Well, I hope you work things out somehow.
This conversation seems perfectly natural. Here is a plausible analysis: in remaining silent, Sally conversationally implicates that she does not want to talk about
what happened the night before, and perhaps also conversationally implicates that
she has indeed had a ght with her husband. However, Sally has clearly said nothingher act of remaining silent expresses no explicature.7
3. A fragment of an explicature is enough to generate a conversational implicature:
Nathan and June are discussing the weather in places they have been. Nathan
starts to say In Amarillo, Texas, its hot and dusty. However, a car alarm
goes o as Nathan speaks, and Nathan only says In Amarillo, Texas. . . Plausibly, Nathans utterance does not express an explicature, but still conversationally implicates that Nathan has been in Amarillo.
4. An explicature may be unrelated to the conversational implicature:
Roy is sitting in his oce staring at the wall. Emilia walks in and starts asking
Roy questions about a lecture they both attended. After answering each question
briey and truthfully, Roy closes his mouth and looks at Emilia. Emilia gets the
message that Roy is not interested in conversing at that moment. Emilia goes
away. Plausibly, Roy would be failing to be a good conversational participant
unless he is conversationally implicating that he is not interested in being a
conversational participant at the moment. Roys comments on the lecture are
unrelated to this conversational implicature.
In each of these examples, I claim, a conversational implicature is carried by an
act of utterance with no inferential path to the conversational implicature requiring
an explicature. If so, then claim (B) is incorrect.
7
Grice gives a related example in which an utterance of Mrs. X is an old bag is followed by a
moment of stunned silence, and then a shift in topic. (Grice, 1989c: 35) Grice says there is a conversational
implicature that a social gae has been committed.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

979

There are two main considerations in favor of the claim that, in these examples,
there are indeed such conversational implicatures. First, these cases bear all the
marks of conversational implicature. As I have explained, conversational implicature is a particular kind of indirect communication: the hearer infers a conversational implicature on the basis of the assumption of the Cooperative Principle; and it
is an inference that the speaker can expect the hearer to work out. Consider, for
example, the third case. There, June concludes that Nathan has been in Amarillo to
preserve the assumption that Nathan, being a cooperative conversationalist, was
about to say something relevant. And Nathan expected June to draw this conclusion
on the basis of the utterence he was trying make. Note that each of the other purported examples also satisfy the description I have given of a conversational implicature. Someone might complain that, because I have not applied the standard tests
for conversational implicatures, I have failed to demonstrate conclusively that these
are indeed conversational implicatures. This complaint is misplaced. For if I am
correct that what is implicated does not depend on what is said, then we need to
reconsider whether the standard tests are correct. We cannot simply apply the tests
blindly.8
The second main consideration in favor of concluding that these are indeed conversational implicatures is that there is no theoretically interesting distinction
between the examples I have given, and standard cases of conversational implicature. For example, compare the letter of recommendation case (often held up as a
8
For readers who wish to have this last point explained in detail: Levinson (1983: Chapter 3) lists four
main properties of conversational implicature: Non-Detachability, Cancellability, Calculability, NonConventionality. Levinson claims that possession of the properties is necessary and sucient for something to be a conversational implicature. Note rst that if Recanati is correct, the conditions are not
jointly sucient to distinguish what is said from what is implicated. For, according to Recanati, elements
of what is said may have the same properties as conversational implicatures. So, in the context of this
essay, we cannot simply assume that testing for these four properties will isolate a conversational implicature. Furthermore, if I am correct, Non-Detachability is not a property shared by all conversational
implicatures. (A Non-Detachable implicature is a conversational implicature that is preserved when what
is said is the same, but said in a dierent way.) If a conversational implicature conveyed by an utterance U
depends not on what is said, but on some other feature of U, then it may, perhaps, not be conveyed by a
dierent utterance, U0 saying the same thing. For U0 may not share the relevant features of U. NonDetachability is better seen as a derived property. A conversational implicature is Non-Detachable when
(and because) it has a certain relation to what is said. If a conversational implicature depends on what is
said by an utterance U, and on no other features of U, then that implicature is Non-Detachable. Perhaps
Non-Detachability has been seen as a property of conversational implicatures because theorists have
focused on implicatures that depend only on what is said. (It has been suggested that Non-Detachability
admits exceptions: some implicatures generated by violations of the Maxims of Manner are detachable.
(Levinson, 2000: 15) If I am correct, these exceptions are, instead, symptoms of the dependency of these
implicatures on features of the utterance apart from what is said.). For what it is worth, the other standard properties are possessed by the examples. 1. Cancellability: an inference to a Cancellable implicature
can be defeated without contradiction by adding further premises. I will not laboriously work through all
the cases, but here is one example: when the car alarm stops, Nathan could add, . . . although I have
never been to Amarillo. 2. Calculability is a property I have already discussed and the examples I
have given are all clearly Calculable. 3. The last main property of conversational implicature is NonConventionality. But since Cancellability indicates Non-Conventionality, we may conclude that the
examples display this property too.

980

P. Hawley / Journal of Pragmatics 34 (2002) 969991

paradigm of conversational implicature) to the case where Sally remains silent.

Notice that the essential premise needed by the letters recipient to conclude that the
professor is implicating that the candidate is bad at philosophy is not what the
professor said, but what he did not say.9 The recipient needs to know that the professor remained silent about the candidates philosophical abilities. This striking
parallel between the two cases makes it dicult to see how someone could maintain
that one is a case of conversational implicature, and one is not. We can make this
point even more vivid by considering the following variant of the letter of recommendation case. Instead of sending a letter saying that the candidate has excellent
penmanship, the professor sends a blank sheet of paper. Note the parallel reasoning:
Original Case:
1. The professor wrote the candidates penmanship is excellent and nothing
more.
2. So, the professor said nothing about the candidates philosophical abilities.
3. So, the professor is suggesting that the candidate is no good at philosophy.
New Case:
1. The professor sent a blank piece of paper and nothing more.
2. So, the professor said nothing about the candidates philosophical abilities.
3. So, the professor is suggesting that the candidate is no good at philosophy.
The step from 1 to 2 is an entailment. 2 follows from 1, given that someone who
sends no more than a blank piece of paper says nothing. The other inferences (2 to 3,
and 2 to 3) are identical. But it is the inference from 2 to 3 that makes the Original
Case a case of conversational implicature. For 3 is the conclusion drawn on the
assumption that the professor is being cooperative. So if the Original Case is a case
of conversational implicature, then surely the New Case is too. But in the New Case,
like the case of Sally, what is implicated does not depend on what is said, because
nothing is said.
Someone might raise the following objection. If we accept that examples 14 display conversational implicatures, then we must admit the following unpleasant
conclusion: gestures and facial expressions can generate conversational implicatures.
In reply: the unpleasant conclusion does not follow, if implicating is restricted to
linguistic acts. For gestures and facial expressions are not linguistic acts, while the
examples I have given are linguistic acts. Thus, my examples show nothing about
non-linguistic acts. Perhaps the objector could press harder, and try to argue that
the examples I have given are not linguistic acts. I am not sure what to make of this
objection without a clear statement of the distinction between linguistic and nonlinguistic acts. Such an approach might work against the second examplesilence

Gabriel Uzquiano pointed this out to me.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

981

may be a conversational act but not a linguistic act, or remaining silent may not be
an act. However, examples 1, 3 and 4 certainly appear to be linguistic acts.
Moreover, even if we must agree that gestures and facial expressions can generate
conversational implicatures, I fail to see why this result is unpleasant. The theorist of
communication should, indeed, welcome a broadening of the Gricean theory. If the
same theory can explain features of both linguistic communication and non-linguistic communication, that would be an interesting discovery about the Gricean
theory. Also it would reveal that, in thinking about conversational implicatures we
rather narrowly restricted our attention to linguistic acts.10
I will stop here, and assume that Ive shown that claim (B) is incorrect.
Beyond (A) and (B), there are other plausible claims of dependency available. For
example,
(C) an explanation of the expression (or understanding) of theexplicature of an
utterance is (or forms a part of) an explanationof the expression (or
under-standing) of the conversationalimplicatures of an utterance.
(D) if it were not possible to express an explicature, then it would not be possible
to express a conversational implicature.
Although the investigation is of interest, I will not consider further dependency
claims in this essay.
Where are we now? So far, I believe, I have shown that, at least on the construals
(A) and (B), what is implicated does not always depend on what is said, and I take it
that this is enough to show that Recanati is mistaken when he claims that, to
understand what is implicated, the hearer must grasp both what is said and what is
implicated and the inferential connections between them. This matters for Recanati
because he puts the dependency claim to philosophical work in an argument in support
of a criterion for distinguishing what is said from what is implicated. This criterion, the
10

Here are two nal objections: (1) Someone might claim that example 2 (the silence case) is not a
counterexample to (B) because there is an explicature expressed in this case, namely an empty explicature,
and, says the objector, there is a reasonable path of inference requiring the empty explicature to calculate
the implicature. But even if we concede that there are empty explicatures, what is needed in the calculation
is not the explicature itself, but rather, the fact that it is empty. Still, the objector says, this is indeed just
saying that the explicature is needed in the calculation in some sense. The correct response is, I believe, to
notice that the objector is eectively claiming that there are two kinds of dependency between an implicature and something else. An implicature can depend on what is said, or an implicature can depend on
the fact that nothing is said. But this is just to concede that sometimes what is implicated does not depend
on what is said. (2) An anonymous reviewer objects that the version of case 1 in which the stranger conveys that he speaks English may be a case of conventional implicature, not conversational implicature.
The objection is incorrect. If the implicature is conventional then it would always be carried by an utterance of that sentence (when what is said is that its nice weather were having today). But the implicature
is not conventional; an utterance of Its nice weather were having today, isnt it? need not convey that
implicature. Again suppose you are lost in a place where English is rarely understood. You walk up to a
stranger and ask Do you speak English? The stranger replies in his own language, making it quite clear
that he does not understand you. Luckily, you have a phrasebook, which provides parallel phrases in
English and in the strangers language. You give the book to the stranger, who ips a few pages and then
says, in an odd, but perfectly understandable accent: Its nice weather were having today, isnt it? The
stranger says that the weather is nice, but does not conversationally implicate that he speaks English.

982

P. Hawley / Journal of Pragmatics 34 (2002) 969991

Availability Principle, states that we ought to respect our pretheoretical intuitions

about whether something is said, or only hinted or suggested. In Recanatis words:
Availability Principle: in deciding whether a pragmatically determined aspect of
an utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter. (Recanati, 1993: 248)
The Availability Principle is important because it is used to justify the claim of
semantic underdetermination (that there are signicant pragmatic elements in what is
said). Recanati does not present the Availability Principle as a criterion that will decide
every case; sometimes intuition does not tell us clearly which way to go in separating
what is said from what is implicated. Instead, Recanati maintains that when a theorist
claims that something is said where pretheoretical intuition has it that it is not said, by
the Availability Principle, we ought to reject the theorists claim. For example,
according to some theorists, whenever Janice utters John has three children,
Janice says that John has at least three children, and implicates that John has no
more than three children.11 I think we can agree that there is no intuition supporting
this analysis, and that there is in fact an intuition that, on some occasions, Janice says
that John has exactly three children. According to Recanati, since intuition conicts
with the analysis, we should, by the Availability Principle, reject the analysis.
Recanati says that his view of the relationship between what is implicated and
what is said entails the Availability Principle: . . . the speaker and hearer. . must be
capable of working out the implicatures, and this entails that they have conscious
access both to what is said and what is implicated. Recanatis idea seems to be that
since, according to Grice, conversational implicatures must be calculable, there is a
requirement on conversational implicatures: if a hearer understands a conversational implicature then it is possible for the hearer to (consciously) grasp both what
is said and the inferential connection between what is said and what is implicated. If
it is possible for them to grasp the inferential connection between what is said and
what is implicated, then they can consciously separate what is said from what is
implicated. If speaker and hearer can consciously separate what is said from what is
implicated, we have reason to trust their intuitions about what is said, and so we should
accept the Availability Principle. This reasoning is rather unclear but if the examples I
have just presented are correct, it cannot even get o the ground. Pointing out, as I did
earlier, that in some cases what is said is not required (or is even irrelevant) in calculating
what is implicated shows that Recanatis justication of the Availability Principle
will not work. For if, in some cases, what is said is not required to calculate what is
implicated, then there is no reason to agree that an inferential connection between
what is said and what is implicated can always be consciously grasped.
11

Some may suppose that this analysis is supported by a cancellability test: if we can cancel the
suggestion that John has no more than three children (John has three children. He may even have four.
) then what is cancelled must be an implicature. However, what is under discussion here is the way to
distinguish what is said from what is implicated. If Recanati is right, then cancellability does not distinguish what is said from what is implicated. So, in evaluating the Availability Principle, we cannot appeal
to a cancellability test without independent support for a such a test.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

983

Perhaps Recanati would respond by weakening the Availability Principle to apply

only in cases where dependency holds. So weakened, the Availability Principle
would not hold generally for all cases of conversational implicature. This weakened
position may be unattractive to those now convinced that cases of conversational
implicature where dependency holds are not interestingly dierent from cases of
conversational implicature where dependency fails to hold. Still, apparently, the
weakened position is tenable. Someone may believe that cases of conversational
implicature where dependency holds are interestingly dierent because the Availability Principle applies in these cases. So let us suppose that Recanati is only trying
to defend a Weak Availability Principle which only applies when what is implicated
depends on what is said. In this case, further argument is needed to undermine his
reasoning; pointing out that some conversational implicatures fail to depend on
what is said is now beside the point.
Here we need to ask what sort of dependency is required to support the Weak
Availability Principle. Now, claim (A) helps bolster something like Recanatis reasoning in favor of the Weak Availability Principle. For if understanding what is said
is necessary for understanding what is implicated, and if understanding is conscious,
then, more or less plausibly, what is implicated must be consciously calculated from
what is said. Unfortunately for the defender of the Weak Availability Principle,
there is probably no case that satises claim (A). Recalling the earlier example of the
air-conditioner, we may press an argument that in any case of conversational
implicature, we can imagine a hearer who misunderstands what is said, and then, by
luck or mistake, understands what is implicated. It thus seems always possible for a
hearer to understand a conversational implicature without understanding what is
said. So it seems always possible for a hearer to understand what is implicated
without what is said being consciously accessible to them.
Unlike (A), dependency claim (B) is plausible in some cases. However, (B)-type
dependency will not support an argument for the Weak Availability Principle. (B)
only requires that there be a reasonable path from what is said to what is implicated,
not that that path be followed. (B) makes no claim about the psychological processes actually involved in utterance understanding. Thus it seems puzzling how one
could conclude that what is implicated must be consciously separable from what is
said when a (B)-like dependency holds.
Although I have only briey considered (A) and (B), the problem is clear. We need
a claim about psychological processing to argue for the Weak Availability Principle.
(A)-dependency can be construed psychologically, but (A)-dependency is implausible. (B)-dependency, although plausible, is metaphysical rather than psychological.
I believe that there is a general argument lurking here against the drawing of psychological conclusions from metaphysical theses about communication (and this is
something I plan to pursue in future work). For now, I only claim that, barring a
plausible, psychological construal of dependency, neither the Availability Principle
nor the Weak Availability Principle have been justied.
I have accomplished two things so far. First, I have taken steps to clarify the
relationship between saying and implicating. I have given examples that I think
show that the Gricean should not hold that what is implicated depends on what is

984

P. Hawley / Journal of Pragmatics 34 (2002) 969991

said, when dependency is spelled out in the most obvious ways, (A) and (B).
Although this alone does not justify concluding that the Gricean should deny that
what is implicated depends on what is said, it at least partly justies this conclusion.
Second, I have traced a connection between the question of dependency and the
issue of how to draw a line between semantics and pragmatics. For Recanati, a
supposed dependency between what is said and what is implicated leads to the thesis
of semantic underdetermination, by way of the Availability Principle. Rejecting
dependency (in the form of (A) and (B)) leaves the Availability Principle in need of
justication, and the thesis of semantic underdetermination in doubt. Since the
Availability Principle was supposed to help separate what is said from what is
implicated, we might worry, as a result, that rejecting dependency leaves us unable
to draw an interesting and useful line between what is said and what is implicated.
This suspicion is groundless, as I will now try to show.

2.
I shall now oer a speculative proposal to distinguish what is said from what is
implicated without dependency, and argue that the distinction so drawn is useful
and interesting. I will distinguish what is said from what is implicated by giving a
criterion for saying that will reduce the theoretical burden taken up by Grices (and
Recanatis) concept of what is said; my notion of what is said will capture some, but
not all, of the features of Grices (or Recanatis) notion of what is said. For Grice,12
what is said is both
(X)

the truth-conditional content of an utterance,

and
(Y)

closely tied to the conventional meaning of the words uttered.

Recanati keeps (X) but argues that (Y) should be relaxed. I propose to keep (X),
but set aside (Y) in some cases. I will rst compare my concept of what is said to
Grices and Recanatis. I will then motivate my analysis of what is said by giving
some examples to suggest that sometimes the truth conditional content of an utterance is unrelated to the conventional meaning of the words used. Then, I will sketch
my analysis of what is said. Finally, I will give reason to think that this version of
the saying/implicating distinction is theoretically useful and interesting.
If Recanati is right, then (X) and (Y) are in tension. Recanati argues that (Y)
should be relaxed; the Availability Principle (coupled with some other premises)
leads him to conclude that sometimes what is said is not so closely related to the
conventional meaning of the words uttered. This is his thesis of semantic under12
The view I am attributing to Grice in this section is an interpretation. I am not at all sure if the view
I am here calling Grices is indeed Grices.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

985

determination. Recall Janices utterance of John has three children. On some

occasions, intuitively, Janice says that John has exactly three children, not that John
has at least three children. If the conventional meaning of the words uttered in a
context only determines the proposition that John has at least three children, and
not the proposition that John has exactly three children, then the conventional
meaning of the words uttered does not fully determine what is said. Recanati thinks
that examples like this help to show that there are two levels of pragmatics. The rst
level lls the (purported) gap between semantics and what is said. The second level
explains how what is said determines conversational implicatures.
We may contrast Recanati with Grice. Intuitive judgments of what is said lead
Recanati to conclude that (Y) should be relaxed. Grice, on the other hand, is willing
to give up some intuitive judgments about what is said for the sake of theoretical
simplicity. Grice, in the interests of simplicity, might hold that when someone utters
John has three children, he says that John has at least three children, and (sometimes) implicates that John has at most three children. This goes against an intuitive
judgment. Grice retains both (X) and (Y) at the expense of giving up some intuitive
judgments.
Like Recanati and Grice, I wish to maintain that what is said is the truth-conditional
content of an utterance. Like Recanati but unlike Grice, I think that in some cases
what is said may be rather distantly related to the conventional meaning of the
words uttered; in some cases (Y) should be relaxed. Unlike both Recanati and Grice,
I think that in some cases what is said may bear no relation to the conventional
meaning of the words uttered; in some cases (Y) should be suspended. In other
words, I hold that the truth-conditional content of an utterance is, on some occasions, unrelated or only distantly related to the semantic features of the words
uttered. While I do not have an argument to show that this possibility occurs, the
following examples are suggestive.
1. Slips of the tongue: Suppose I utter the sentence Im going to Wordsworth
today to buy a bunch of new looks. There is some intuition that I said that
Im going to Wordsworth today to buy a bunch of new books; I can speak
truly about books without using the word books. I used the word look, but
what I said does not depend on the meaning of look.13
2. Malapropism (cf Davidson, 1986): a malapropism is an often humorous
substitution of one word for another: What a nice derangement of acorns
you have on your web page! I exclaim. Thanks, but these arent icons,
theyre JPEGS, my friend replies. Since this seems like a coherent conversation (at least among computer lovers), there is reason to think that I
said just what I would have said if instead I had exclaimed What a nice
arrangement of icons you have on your web page!.
13
I think its also true that I said Im going to Wordsworth today to buy a bunch of new looks.
Carston (1988) (crediting Kempson) calls this the quotational sense of say. Im not sure we need to posit
dierent senses of say here, but I do maintain that there is an intuition that it is true that I said something about books, not about looks.

986

P. Hawley / Journal of Pragmatics 34 (2002) 969991

3. Metaphor: if I say the sea is glass on a calm day, it seems that what I said
may well be true even though the sea is not literally made of glass. The same
sentence can be used non-metaphorically to say something false. This suggests that what is said when I speak metaphorically can be far removed from
the ordinary, conventional meaning of the words I use and how they are put
together.14
4. (cf Grice, 1989d: 181) Im talking with you about a sleazy looking guy sitting
across the room. We both know that he works as a management consultant
and has not uttered a word about religion in his life. I utter Hes just an
evangelist meaning that he is a sanctimonious money-grubber. I can speak
truly about the sleazy guy although he has never discussed religion; what I
said has nothing to do with religion.15
5. Suppose, for ninety-eight days, after dinner, Jennifer says that the dinner
tasted good. However, each day she uses dierent words (The meal was
delicious, the dinner was tasty and so on.) On the ninety-ninth day, in the
same tone of voice as before, she announces the zutness was ertly. We can
imagine that she said the same thing she said all along, although zutness and
ertly do not mean anything in English. Suppose, on the hundredth day, she
announces the french horn was happy. Could we not imagine that she said
the same thing she said all along, without supposing that she has false beliefs
about the words she used?
In each of these cases, there is an implicature account available: what I am taking
to be said is actually only conversationally implicated. For example, someone may
claim that, in the case of the metaphorical use of the sea is glass, what I said is
false, although I have conversationally implicated something true. (Grice analyzes
some cases of metaphor this way in Logic and Conversation.) According to such
an account, what is said remains closely tied to the conventional meaning of the
words used. Or, we may oer an account similar to Recanatis, in which what is said
in the metaphor case is, as I claim it to be, true, yet is derived by both semantic and
pragmatic means from the conventional meaning of the words used. Or, we may
claim that glass is ambiguous between a literal and a metaphorical sense. I am not
oering knockdown reasons that such accounts are mistaken; these examples are
only suggestive. I only claim that there is intuition to support the position that
sometimes what is said is not only distant from conventional meaning, it is unrelated
to it; in some cases the truth-conditional content is not derived from the conventional meaning of the words used; in some cases (Y) should be set aside.

14
Compare N. Goodmans reply to D. Davidsons claim (Davidson, 1984) that most metaphors are
trivially and literally false: The lake is a sapphire is. . literally false but metaphorically true, while
Muddy Pond is a sapphire is both literally and metaphorically false. (Goodman, 1979: 126) (Metaphorically true for Goodman means that the sentence taken metaphorically is true.)
15
In later work, Grice (1989d: 181) separates dictiveness from formality. Dictiveness is saying (or
what is said) and formality is conventionality (or what is conventional). He gives much the same example,
and says that it shows dictiveness without formality (that is saying without conventionality).

P. Hawley / Journal of Pragmatics 34 (2002) 969991

987

I will now give an analysis of saying that will permit (Y) to sometimes be set aside.
I will have shown that my proposal should be preferred to Grices and Recanatis if I
can show that it accounts for data as simply as Recanati or Grice, yet accommodates relevant intuitions that they fail to account for. The rst step in showing
this is to develop the proposal. To the extent that the proposal is sketchy the argument is incomplete. (When the proposal is fully developed, I would go on to argue
that Recanatis view is too complicated because his pragmatics level 1 duplicates
much of pragmatics level 2. I would also argue that Grices view seems to treat
intuitions arbitrarily. These intuitions are the main reason for accepting the saying/
implicating distinction. Yet sometimes strong intuitions are set aside, with no reason
apart from theoretical economy. I will not spell out these arguments here.)
Suppose S performs an act of utterance U. My suggestion is that S says that p is
true when the following three conditions are met:
1. (association) p is a proposition associated with U
2. (explicitness) S bears an explicit attitude to p
3. (prominence) p is as prominent as any other proposition satisfying 1 and 2
Before working to spell out these conditions, let me explain what they accomplish.
Condition 1 requires that what is said be appropriately related to the act of utterance, ruling out the possibility that S can say something that is unconnected to the
act S performs. What counts as appropriately related is a question I will leave open.
At least, someone who likes Grices story about conversational implicature would
count the conversational implicatures conveyed by an utterance to be appropriately
related to that utterance. Condition 1 will rule out some propositions as not said.
For instance, suppose that a sneeze is an act. We may think that the act of sneezing
determines no proposition, so no propositions meet condition 1; when I sneeze I do
not say anything.
Condition 2 is supposed to ensure that S is related to p in an appropriate way. It is
supposed to rule out possibilities like the analysis in which on all occasions when
Janice utters John has three children she says that John has at least three children.
For on some occasions Janice does not bear the right attitude to the proposition that
John has at least three children- she is not aware of it, has never considered it. (What
I have in mind here is a tacit/explicit distinction, but I do not mean to claim that
bearing an explicit attitude toward a proposition requires being aware of it. I am
supposing that lack of awareness is evidence of the absence of an explicit attitude.)
Condition 2 also rules out the possibility that in uttering the Wordsworth sentence, I
was saying anything about looks. In some cases, condition 2 may rule out all propositions closely tied to the conventional meaning of the words uttered. If so, it does
the work of suspending (Y).
Condition 3 is the reformulation of (Y). In normal cases, the most prominent
proposition will just be the proposition that is most closely tied to the conventional
meaning of the words used. So, in normal cases, what is said will be closely tied to the
conventional meaning of words. In other cases, the proposition most closely tied to
conventional meaning will be ruled out by Condition 2, and some other proposition

988

P. Hawley / Journal of Pragmatics 34 (2002) 969991

will count as what is said. This is what will happen in the metaphor case, and the
other suggestive cases mentioned earlier.16
On the proposed picture, no dependency has been presumed between what is said
and what is implicatedthe proposed picture accommodates each possibility for
dependency. In a particular case, it may be that what is implicated depends on what
is said, or there is no dependency between what is said and what is implicated, or
what is said depends on what is implicated. This last possibility may be surprising.
But consider the lover of conversational implicature who thinks that some of the
propositions satisfying Condition 1 are the conversational implicatures of the utterance. The proposed picture holds open the possibility that what is said is sometimes
a conversational implicature: the most prominent proposition satisfying both Conditions 1 and 2 may be a conversational implicature. So, in a sense, sometimes what
is said may depend on what is implicatedin the sense that sometimes what is said
is a conversational implicature.
The proposed account of saying is supposed to help point to a psychological description of the processes involved in making and understanding utterances. The psychological account will hopefully explain what the intuitions of directness versus indirectness
are intuitions of. I will now explain the psychological relevance of the three conditions.
When a speaker prepares to make an utterance, and then does utter something,
certain psychological processes occur. A description of these processes will make
reference to the attitudes that the speaker takes to various propositions. For
instance, suppose the speaker thinks that his friends hat is ugly, and wants him to
take it o, but not wanting to oend his friend, he decides to only ask if he could
take a closer look at it. Or suppose that the speaker decides to tease his friend about
the hat. He thinks the hat is ugly, and knows that the friend is unusually sensitive
about his appearance. So he decides to unctuously say what an extraordinarily
beautiful hat! realizing that the friend will wonder why he said that, and get nervous. Each of these stories of the speakers reasoning process describes the speaker
as believing some things, desiring some things, rejecting other things. The point of
Condition 2 is that the speaker cannot say something unless he reasons about it
before speaking. Someone who is reading a newspaper aloud without understanding
is not saying anything; he is simply mouthing the words. Someone who makes a slip
of the tongue hasnt said what the mistaken sentence says. A description of the
psychological process leading the speaker to speak on some occasion will describe
the speaker as holding attitudes to certain propositions. So by explicit in Condition 2, I mean explicitly involved in the reasoning leading to an utterance.
An explanation of prominence will help explain intuitions about saying. At present, I have but sketchy ideas about prominence. Prominence is a relation between
propositions, with respect to an utterance. Of two propositions, the proposition that
is more prominent (with respect to an utterance) is the proposition that is easier to
16
One might think that Condition 3 could, on its own, cause the suspension of (Y). However, it ought
to be a constraint on measures of prominence that propositions more closely tied to conventional meaning
are more prominent than other propositions. (If more than one proposition has the highest prominence,
then S has said more than one thing by the same utterance.)

P. Hawley / Journal of Pragmatics 34 (2002) 969991

989

derive17 by a hearer of the utterance. This explanation of prominence will help

explain intuitions about saying, if when people perform derivations, particular psychological processes occur. And, since hearers can misderive things, or perform
longer derivations than necessary, some intuitions about saying can count as mistaken, and intuitions between people may dier.
The associated propositions admitted by Condition 1 may include what is normally thought of as the content of an utterance, as well as presuppositions and
implicatures, and perhaps even some propositions present in the current context. For
someone who accepts Grices account of meaning, x is associated with U if S meant x
by his utterance U, where this is analyzed in Grices way in terms of intentions.18
Since, at the moment, I am unable to provide more detail about the proposed
picture, I will conclude by comparing it to Recanatis view. One dierence between
Recanatis view and the proposed picture is that Recanati presumes that in all cases
what is said is derived from the conventional meanings of words (along with other
things). Semantics along with pragmatics level 1 explains how, given a context, what
is said is derived from the conventional meaning of the words uttered and how they
are put together. Recanatis view demands a derivation starting from conventional
meaning to what is said, for each case of saying; whenever a theoretically unexplained case of saying is found, there is pressure to either change the semantics, or
expand the derivational possibilities of pragmatics level 1. On the proposed picture,
17

Perhaps prominence may be dened as derivational dependence: if all (reasonable) derivations of

proposition q (from an utterance in a context) require proposition p then p is more prominent than q. In
the letter of recommendation case, for instance, suppose that deriving the proposition that the candidate
is bad at philosophy requires the proposition that the candidates penmanship is excellent. This would
mean that the proposition that the candidate is bad at philosophy is less prominent then the proposition
that the candidates penmanship is excellent. So the speaker does not say that the candidate is bad at
philosophy. (Hence, to some degree, we can incorporate Recanatis thought that inferential dependency
of what is implicated on what is said underlies intuitions about the distinction between saying and
implicating.) However, derivational dependence is not enough to characterize prominence because there
are many propositions that are derivationally independent from each other. (Two propositions are derivationally independent if neither is derivationally dependent on the other.) For example, in the examples
in section I, what is implicated is derivationally independent of what is said, so prominence characterized
as derivational dependence is not enough to distinguish what is said from what is implicated for these
examples. A better measure of prominence is derivational length: if the shortest derivation of p (from an
utterance in a context) is shorter than the shortest derivation of q, then p is more prominent than q. This
may not be enough. It may be that when measuring prominence we will need to rank the sources of
derivations, so that, for instance, propositions that can be derived from the uttered sentence without
reference to the context will count as more prominent than propositions derived from other things. Such
a ranking will make propositions more closely associated with conventional meaning more prominent.
18
See Grice (1989a) for Grices analysis. (Note that I am not endorsing Grices account of meaning in
this essay; I am just using it as an example.) Someone adopting Grices account of meaning in spelling
out Condition 1 may not need Condition 2 because everything that S meant seems to meet the explicitness condition. However, other accounts of what propositions are associated with an utterance will need
Condition 2, so I keep Condition 2 separate. For example, someone may think that some implicatures
are unintended: suppose the hearer can infer that the speaker is tired by their rate of speech and tone of
voice. Someone could think that the proposition that the speaker is tired is associated with the utterance,
but fails to be in the running for what is said because it fails Condition 2. For such a theorist, Condition
2 is not redundant.

990

P. Hawley / Journal of Pragmatics 34 (2002) 969991

there is no such pressure because no derivation starting from conventional meaning

is required. For instance, we might in some cases conclude that what is said is a
conversational implicature. If so, there is no pressure to modify the semantics. For
example, there is no pressure to think that looks sometimes means what books does.
I consider this a virtue of the proposed picture, given the suggestive examples above.
Readers may be puzzled. While in the rst part of this essay I rejected Recanatis
argument that we should always take intuitive judgments about what is said seriously, in the second part I have taken intuitive judgments about what is said seriously. There is no contradiction heremerely an attempt to disentangle issues: the
saying/implicating distinction, the semantics/pragmatics interface, the priority of the
literal. In the rst part, I tried to show that accepting the saying/implicating distinction does not require us to take intuitive judgments about what is said seriously
in all cases. In the second part, I tried to show that taking intuitive judgments about
what is said seriously in some suggestive cases reveals that the saying/implicating
distinction is independent of the priority of the literal.
Readers may be disappointed by the proposed picture because it fails to support a
clean separation between semantics and pragmatics. The hope of a simple semantics
distinct from pragmatics was one of the tempting aspects of Grices story. Since, on
the proposed picture, what is said may contain no semantically determined elements,
the proposed distinction between what is said and what is implicated fails to cleanly
separate semantics from pragmatics. Although it may be disappointing to nd that
things are not as simple as we had hoped, we should not conclude that a distinction
between what is said and what is implicated is unimportant or uninteresting. We
should only conclude that the proposed distinction between what is said and what is
implicated does not by itself separate semantics from pragmatics. The proposed
analysis of what is said loosely circumscribes semantics. This is a rst step; other
distinctions and theories are needed to nish the job of isolating semantics. Additionally, in setting aside dependency of what is implicated on what is said, a picture
emerges in which pragmatics is no longer seen as dependent on semantics.
In this paper, I have argued that what is implicated does not depend on what is
said by trying to show that obvious ways of spelling out a proposed dependency
wont work. I then sketched a picture that maintains a distinction between saying
and implicating while presuming no dependency of what is implicated on what is
said. My general goal, to articulate and defend a view which rejects the priority of
the literal, while preserving a useful distinction between pragmatics and semantics, is
still remote. For I have not shown that we ought to reject the priority of the literal; I
have not shown that it is not the case that literal meaning makes it possible for
someone to mean something by an utterance. However, by clarifying the relationship between saying and implicating, and presenting a sketch of saying, I believe that
I have opened a path to such a view.19
19
Many thanks to those who read earlier versions especially: Rajesh Bhatt, Matti Eklund, Adam Elga,
Liz Harman, Miguel Hernando, Olafur Jonsson, Ishani Maitra, Sarah McGrath, Anthony Newman,
Bernhard Nickel, Agustn Rayo, Robert Stalnaker, Robert Streier, Roger White, and Gabriel Uzquiano.
Special thanks to Michael Glanzberg (who read more than his share of drafts) and Robert Stalnaker, for
helpful criticism.

P. Hawley / Journal of Pragmatics 34 (2002) 969991

991

References
Bach, Kent, 1994. Conversational impliciture. Mind and Language 9, 124162.
Carston, Robyn, 1988. Implicature, explicature, and truth-theoretic semantics. In: Kempson, R.M. (Ed.),
Mental Representations: The Interface Between Language and Reality, Cambridge University Press,
Cambridge, pp. 155181.
Davidson, Donald, 1984. What metaphors mean. In: Inquiries into Truth and Interpretation. Clarendon
Press, Oxford, UK, pp. 245264.
Davidson, Donald, 1986. A nice derangement of epitaphs. In: Lepore, Ernest (Ed.), Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson. Blackwell, Oxford, UK, pp. 433446.
Goodman, Nelson, 1979. Metaphor as moonlighting, Critical Inquiry, 6(1): 125130.
Grice, Paul, 1989a. Studies in the Way of Words. Harvard University Press, Cambridge, MA.
Grice, Paul, 1989b. Further notes on logic and conversation. In: Grice (1989a), pp. 4157.
Grice, Paul, 1989c. Logic and conversation. In: Grice (1989a), pp. 2240.
Grice, Paul, 1989d. Retrospective epilogue. In: Grice (1989a), pp. 339385.
Grice, Paul, 1989e. Utterers meaning and intentions. In: Grice (1989a), pp. 86116.
Grice, Paul, 1989f. Utterers meaning, sentence-meaning, and word-meaning. In: Grice (1989a), pp. 117137.
Levinson, Stephen C., 1983. Pragmatics. Cambridge University Press, Cambridge, UK.
Levinson, Stephen C., 1995. Three levels of meaning. In: Palmer, F. R. Ed., Grammar and Meaning:
Essays in Honour of Sir John Lyons, Cambridge University Press, Cambridge, UK, pp. 90115.
Levinson, Stephen C., 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. MIT Press, Cambridge, MA.
Recanati, Francois, 1993. Direct Reference: From Language to Thought. Blackwell, Oxford, UK.
Sadock, Jerrold M., 1991. On testing for conversational implicature. In: Davis, Steven (Ed.), Pragmatics,
A Reader, Oxford University Press, Oxford, UK, pp. 365376.
Sperber, Dan and Wilson, Deirdre, 1987. Presumptions of relevance. Behavioral and Brain Sciences 10, 736
754.
Sperber, Dan and Wilson, Deirdre, 1995. Relevance: Communication and Cognition, 2nd Edition. Blackwell, Oxford, UK.