Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

Grice (1967) famously distinguished between what a speaker said by an utterance, on the one hand, and what she implicated by saying what she said, on the other. This was widely taken, at the time, as providing a clear-cut distinction between semantics and pragmatics. Semantics would deal with ‘what is said’, also known as ‘the proposition expressed’ by the utterance (or sentence in context), its ‘propositional content’.Footnote 1 (See Kaplan (1989) for a classic statement). Pragmatics would be concerned with implicatures and any other aspect of meaning not pertaining to the truth-conditional content of the utterance. As Gazdar (1979) famously put it

$$ {\text{PRAGMATICS }} = {\text{ MEANING }} - {\text{TRUTH}} - {\text{CONDITIONS}} . $$

But things were soon shown to be not so simple. Figuring out the reference of many natural language expressions seems to require reasoning the same sorts of pragmatic reasoning involved in the generation and understanding of Gricean implicatures. We have pragmatics intruding on the realm of semantics.Footnote 2

Grice’s initial impact on pragmatic theory was due to the wide range of applications for his theory, but as with all philosophical theories, considerable critical attention was paid to fundamental ideas and distinctions. Interest first focused on his distinctions between conventional and conversational implicatures, and particularized and generalized ones. With time, attention turned to the concept of ‘what is said’ and the extent to which this is affected by “pragmatic intrusion,” that is, the extent to which understanding what is said by the speaker requires the intervention of pragmatic processes.

The issue of the amount of pragmatic intrusion into the semantic content of the utterance is at the heart of the debate held by literalists (or minimalists) and contextualists and a number of positions in between (indexicalists, situationalists, and radical and moderate of all sorts). Despite their differences, they seem subject to what Levinson (2000) dubbed ‘Grice’s circle’:

Grice’s account makes implicature dependent on a prior determination of ‘the said.’ The said in turn depends on disambiguation, indexical resolution, reference fixing, not to mention ellipsis unpacking and generality narrowing. But each of these processes, which are prerequisites to determining the proposition expressed, may themselves depend crucially on processes that look indistinguishable from implicatures. Thus what is said seems both to determine and to be determined by implicature (Levinson 2000: 186).

Thus, within the Gricean picture, the processes of intention-recognition that invoke the Cooperative Principle and the conversational maxims seem to be needed to determine the proposition expressed by the utterance; but, at the same time, the proposition expressed seems to be required for these processes to get started. Many working practitioners accept that semantics and pragmatics are irremediably entangled in the determination of utterance content. There is no clear-cut delimitation between semantics and pragmatics. They ignore the Gordian knot, rather than trying to untie it.Footnote 3

We have a bit of terminology we find useful in thinking about this. What is said, or the proposition expressed, plays a central role in the classic picture. We call pragmatics in the service of figuring what is said, “near-side pragmatics” (see Korta and Perry (2006b)). This is the pragmatic reasoning that seems to be intruding into something that is none of its business on the classical conception. Pragmatic reasoning that starts with what is said, and seeks to discover what one is doing in saying or by saying it—what illocutionary and perlocutionary speech acts were performed, we call “far-side pragmatics”. So the question is, how does near-side pragmatics make sense?

Some tried a somewhat different approach: to assume that, all in all, semantics does not yield fully truth-conditional content. The “output” of semantics is the “input” to pragmatics, but this output is not a fully determined proposition, and not what is said. Instead of talking about ‘pragmatic intrusion’, we should accept that what is said is systematically a pragmatically determined content and not the input to pragmatics; the input to pragmatics is just what semantics gives us. This approach has been seen as undermining truth-conditional semantics, as cutting through the knot rather than untying it, to pursue one of our metaphors. But of course some are happy to cry “Truth conditional semantics is dead! Long live to truth-conditional pragmatics!”Footnote 4

In this paper, we’ll argue that there are two false assumptions that generate the circle. One is that we need to identify what is said, in a canonical way, before Gricean considerations can be applied. The other is that simply because semantics underdetermines what is said, it does not provide a propositional content that can be the basis of reasoning. We’ll show that these assumptions are wrong and that abandoning them gives way to a natural account of the semantic content of an utterance, what is said, and the semantics/pragmatics distinction. So we claim, in terms of our metaphors, to square Grice’s circle, or untie the knot without slicing it, that is, without undermining truth-conditional semantics. We’ll start by considering Grice’s concept of what is said.

2 Grice on What is Said

Grice (1967/1989) famously distinguished between what a speaker says and what she implicates by uttering a sentence. Think about Anne and Bob talking about their common friend Carol, who both know that she recently started working in a bank. Anne asks: “How is Carol getting on in her job?” Bob replies: “Oh quite well… She hasn’t been to prison yet.” Bob is clearly suggesting something here; something related to Carol’s tendency to yield to the temptation provided by her occupation, as Grice would put it. However, that’s not something he said, but something he implicated in saying what he said.Footnote 5 But what did he say?

Grice’s remarks suggest that his concept of ‘what is said’ can be taken as equivalent to ‘the proposition expressed’ or ‘the content’ of the utterance.Footnote 6 He claims that to know what someone said by uttering a sentence one has to know

  1. (1)

    the conventional meaning of the sentence uttered;

  2. (2)

    the disambiguated meaning of the sentence in that particular occasion of use; and

  3. (3)

    the referents of referential expressions (Grice 1967/1989: 25).

This view of what is said fits well with (Kaplan's 1989) distinction between the character of a sentence, and the content of an utterance (or sentence-in-context). This also results in a seemingly perfect match between semantics and pragmatics: semantics deals with what is said; pragmatics deals with implicatures. The input to pragmatics is the output of semantics, in the form of disambiguated meaning plus context and fixing the reference of names.

But Grice’s concept of what is said turned out to be less simple and clear as initially thought. Perry (1986) argued that someone who utters “It is raining” normally expresses a proposition that includes the place of the raining event, even if the sentence does not include any expression articulating that element of the proposition expressed. A variety of phenomena arguably showed a similar point: quantifier domain restrictions, comparative adjectives, assertions about taste, and a long list of other phenomena seem to involve constituents of the proposition expressed that are not articulated by any element in the sentence uttered. A significant part of the debate between minimalism and contextualism, and all the isms around them, concerns the analysis of these elements: whether they are actually part of what is said or should relegated to some other category, like the category of implicitures (Bach 1994) or generalized conversational implicatures; whether they are actually unarticulated or they are values for some ‘hidden’ indexicals in the logical form of the sentence uttered.

One might think that this kind of pragmatic intrusion into determining what is said is all there is to the circle; and that just a minimalist approach as found in, for instance, (Cappelen and Lepore 2005), is enough to square it. The product of sentence meaning plus disambiguation and reference-fixing of names and indexicals might not yield what intuitively is said by the speaker, but it would have the merit of giving us a clear-cut notion of semantically expressed proposition. If you say, “It is raining,” the semantically expressed proposition is simply that it is raining; something true if it raining anywhere (roughly). That proposition is trivial, so pragmatics takes the hearer to a more promising one as what the speaker conveys: it is raining in X, where the identity of X is determined pragmatically.

Another view is that semantics alone doesn’t (always) get us to a proposition at all, even if we assume disambiguation and add context and reference fixing. We can only rely on obtaining a ‘proposition template’ (Carston) or a propositional radical (Bach). Moreover, whatever exactly the result is, it does not seem to constitute the right ‘input’ for implicatures; their derivation requires an ‘enriched’ proposition to serve as ‘what is said’: the content of the utterance with its implicitures.

On this issue, we side with Cappelen and Lepore, although with important differences. Like them, our view is that semantics provides us with propositions, and these propositions are not what is said. But our account is even more minimalist a conception than the one Cappelen and Lepore provide, and, we think, much more intuitive in a wide variety of cases.

To see the merits of our position, it is helpful to note that a Cappelen and Lepore style minimalism does not actually get us out of the Gricean circle. This is because the factors they fold into semantics: disambiguation and reference-fixing in particular, are often only resolved using pragmatic methods. As Levinson notes, reference fixing and indexical resolution and disambiguation, ‘which are prerequisites to determining the proposition expressed, may themselves depend crucially on processes that look indistinguishable from implicatures.’

Here is an example involving reference-fixing. Suppose that the Stanford Philosophy Department is meeting in the 1980s, with John Perry, John Etchemendy, John Dupre and Jon Barwise all in attendance. John Perry has been talking at length, while Jon Barwise has been waiting impatiently to say something. The chair, Nancy Cartwright says, “/djon/ needs some time to develop his views”. Is she referring to John Perry, using the name “John”, or one of the other Johns, or Jon Barwise, using his name? It seems that the listener will most likely try to figure out whether she is speaking literally, referring to John Perry and implicating that Barwise should calm down (unlikely), being non-literal, referring to John Perry, and implicating that he should shut up, (more likely, postulating a sarcasm within Cartwright’s repertoire), or speaking literally, referring to Jon Barwise (using the name ‘Jon’), and implicating that time is running out so he needs to be given a turn (most straightforward), or referring to Etchemendy or Dupre, and implicating that they should quit dozing and get involved (a distinct possibility).

What seems beyond question is the principle of underdetermination,

The linguistic meaning of a sentence underdetermines what is said by a speaker uttering that sentence.

If all there is to semantics is to give the linguistic meaning of sentences (types), then there is an obvious sense in which that claim is true. Sentences say nothing; utterances do or, better, speakers do by uttering sentences. The semantic meaning of a sentence type uttered on a particular occasion is not (often, according to ‘moderate’ contextualists; always, according to ‘radical’ ones) enough to determine what the speaker said by the utterance. She might not have said anything, but have been asking a question, giving an example, or rehearsing a line for a play. More relevantly for our purposes, the linguistic meaning of the uttered sentence does not seem to yield a fully truth-conditional or propositional content.

3 Minimal but Complete Semantic Contents

It’s helpful to start with indexicals. Suppose, out of the blue, in a crowded room, you hear the utterance

  1. (1)

    I am French.

You don’t even see who is doing the talking. Having no clue about who is talking, you would not really be able to say what the speaker said. Your semantic competence gets you only so far; it seems you wouldn’t grasp a proposition, but just a propositional template or propositional ‘radical’; a predicate like

  1. (2)

    x is French.

Since you don’t know who is speaking, you don’t know who this predicate has to be true of, for the utterance to be true.

On the other hand, semantics does provide you with enough to get started. You know that (1) is true iff,

  1. (3)

    ∃x (x is the utterer of (1) and x is French)

(3) Gives you a proposition, which may serve to get some reasoning started. Perhaps you reason that most likely (1) is true, and whoever said it is French; you can’t think off hand of any reason why someone would be claiming to be French if they weren’t; it didn’t sound sarcastic, but had more of an informative tone. This may be rather weak pragmatic reasoning, but it is pragmatic. You might look around for some reliable signs of a French person—someone who is smoking Gauloises cigarettes for example. If you spot such a person, and are right, you will have used pragmatic reasoning and semantics (what is required for the utterance to be true, plus what seems to be the intention behind the utterance) to help you figure out who said it, which allows you to figure out what is said.

This illustrates our basic strategy. Semantics provides slots, which provides truth conditions for utterances; by existentially quantifying we get a proposition; this proposition usually won’t be what is said, but it provides us with what is needed to start what we call “near side reasoning”; that is reasoning that gets us from perception of an utterance of a sentence and a grasp of the semantics, to what is said.

This picture is pretty much in tune with Borg’s (2004) minimalist view of semantic content, and with Grice’s comments on what one understands from an utterance in virtue of knowing the language. About an utterance of ‘He was in the grip of a vice’, he makes the following remark:

Given a knowledge of the English language, but no knowledge of the circumstances of the utterance, one would know something about what the speaker had said, on the assumption that he was speaking standard English, and speaking literally. One would know that he had said, about some particular male person or animal x, that at the time of utterance (whatever that was), either (1) x was unable to rid himself of a certain kind of bad character trait or (2) some part of x’s person was caught in a certain kind of tool or instrument (approximate account, of course) (Grice 1967/1989: 25).Footnote 7

As Grice’s remark suggests, we don’t even need disambiguation to get these sorts of truth-conditions. Given the meanings in (British) English of “He was in the grip of a vice”, there is a fully propositional if utterance-bound content of u, namely,

  1. (4)

    ∃x∃t (x is the person or animal the speaker of u is talking about and t is the time of utterance & at t either x was unable to rid himself of a certain kind of bad character trait or some part of x’s person was caught in a certain kind of tool or instrument)

Suppose you receive an unsigned postcard that reads

  1. (5)

    I am having a good time here.

Without identifying the writer, the time and the place of the writing, you would not know what she or he was saying to you. But this does not mean that you would fail to grasp a complete proposition from the postcard on the basis of your semantic competence. You would understand that the utterance would be true if and only if its author was having a good time at the time of the utterance at the place of the utterance. This is a perfectly truth-conditionally complete content.

These propositions are not what is said, and they aren’t even about the same things the speaker is talking about. (4) is about the utterance u, not about the person the speaker refers to with ‘he’, or even about the speaker himself or herself. We call such propositions utterance-bound contents. We want to emphasize three things about such contents before going further:

  1. (a)

    Utterance-bound contents are utterance-bound with respect to the utterance they are about. (4) is the utterance-bound content of u, but it is not the utterance-bound content of (4).

  2. (b)

    We do not claim, to repeat ourselves, that the utterance-bound content of an utterance is what the utterance expresses, or what the utterance says, or what the speaker of the utterance says.

  3. (c)

    We do claim that the utterance-bound content gives the truth-conditions of the utterance. That is, it gives the conditions that the utterance must meet, in order to be true; there must be various things, speakers, things the speaker refers to and the like, that stand in various relations to the utterance and fulfill further conditions. But saying this can be misleading. In philosophy, at least, one typically uses the term “truth-conditions” for the counterfactual truth-conditions, the conditions that a situation or world must satisfy, to be one in which the proposition expressed, or what is said, is true. The counterfactual truth-conditions usually line up with what we call the referential truth-conditions, which is what you get when you identify the witnesses of the various existential quantifiers and plug them in for the variables. For example, with respect to our last example, this might be the proposition

    That Hiram is/was having a good time on March 4, in Hawaii.

    This proposition isn’t about an utterance, and could be true in worlds in which the utterance u did not occur.

    Although potentially misleading, we think our use is a correct and literal use of the term ‘truth-conditions’. You get different truth-conditions for an utterance, depending on what you hold fixed and what you allow to vary. The truth-conditions are what else has to be the case, given what is held fixed, for the utterance to be true. Both utterance-bound and referential truth-conditions are truth-conditions; we aren’t replacing the ordinary philosophical concept, but noticing that it is part of a system of truth-conditions, or contents, that an utterance can have.

4 Utterance Contents and Implicatures

Even admitting that an utterance has a truth-conditionally complete content before disambiguation, reference fixing and any other near-side pragmatic considerations, it can be argued that, since this content does not amount to what the speaker said by her utterance, we need to perform disambiguation and reference assignment and even further pragmatic processes (like so-called ‘free enrichment’ processes) to get the content that can appropriately called what is said (or the ‘explicature’). This would be required by Grice’s picture of the inference of implicatures. Minimalists like Cappelen and Lepore share this view with contextualists like Carston (2002):

  • We agree with her that you need a contextually shaped content to generate implicatures in all of the cases she discusses. (…) What’s needed in order to derive the implicature in these cases is a contextually shaped content, i.e., a contextually shaped what-is-said.

  • (…)

  • More generally: We are happy to agree with Carston that an appropriate notion of what the speaker said must allow for contextual influences that go far beyond what the speaker said (Cappelen and Lepore 2005: 180–181).

So, if this morning someone invited you to a coffee and you uttered

  1. (6)

    I’ve had breakfast,

trying to implicate a negative answer, the implicature wouldn’t have go through, had he only got its utterance-bound content (7), or what Cappelen and Lepore take to be the proposition semantically expressed (8) (the result of semantic meaning plus disambiguation and reference assignment to referential expressions). Something like (9) which, arguably, includes elements that are not articulated in the uttered sentence, needs to be determined to infer the implicature:

  1. (7)

    There is a time previous to the utterance of (6), when the speaker of (6) has had breakfast.

  2. (8)

    X has had breakfast at some point or other.

  3. (9)

    X has had breakfast this morning.

In this and other examples, the concept of what is said required by Grice’s theory of implicatures seems to go beyond our fully truth-conditional but minimal level of content. Hence, we seem condemned to Grice’s circle, after all.

But we are not. First, and most importantly, we have a truth-conditionally complete content whose determination is independent of pragmatic reasoning, and thus keep us out of the circle. And, second, because, contrary to what Cappelen and Lepore suggest, it is not generally the case that the hearer needs arriving at a ‘contextually shaped’ what is said to understand the implicatures of an utterance. It is not even necessary that he gets what is said or what they call the proposition semantically expressed. It is sometimes enough to get the utterance-bound content of the utterance, or a content that with some undetermined (but existentially bound) element.Footnote 8 Suppose Kepa asked John for suggestions about whom to invite for an upcoming pragmatics conference. The conversation runs like this:

J: He is rather unreliable, doesn’t have much to say, and always takes a long time to say it.

K: Next.

In this case, Kepa need not resolve the referent for John’s use of the demonstrative ‘he’, and maybe John doesn’t intend him to resolve it. He trusts that grasping the utterance-bound truth-conditions of his utterance,Footnote 9 Kepa will infer that that guy would not be an appropriate candidate for lecturing at the conference. Or take our postcard example. You can guess, say, who the author is, but not the place he is talking about. You can, however, understand that he is implying he is postponing his trip back home. The moral is that even without determining what is said in the sense of disambiguating the expressions used and fixing the referents involved, Gricean inference of implicatures is often possible.

To sum up, Grice’s circle is avoided once a fully truth-conditional minimal semantic content is provided. This is our utterance-bound content, which does not require any pragmatic process of disambiguation, anaphora resolution or any other process that looks practically indistinguishable from implicature inference. The utterance-bound content itself can constitute a sufficient input for figuring out an implicature in some cases, while in other cases more facts about the utterance may need to be fixed. In some cases, the referential content of the utterance may suffice, in others one may need “the contextually-shaped what is said”. But even in these cases the pragmatic reasoning can begin with the utterance-bound content, and can be used to arrive at the more specific contents that are required. Suppose you take your watch to be repaired and the watchmaker tells you:

  1. (10)

    It will take some time to fix this watch.

The referential content of (10) is trivially true; any human act takes some time to perform. The enriched content of (10) (with the impliciture within brackets) would be something like

  1. (11)

    It will take some time [more than you might expect] to fix this watch,

which would allow you to infer the intended implicature, say, that you should take it easy. Now, some other shopper who overhears the conversation without being in position to fix the reference of the speaker’s use of this watch would easily understand the impliciture and implicature at issue. The inference of pragmatic contents starts from the utterance-bound content with or without reference fixing, disambiguation and other near-side pragmatic processes.

5 Truth-Conditional Semantics and Pragmatics

In our view semantics has to do with the conventional meanings of words and modes of combination, and its most central part is truth-conditional semantics. Truth-conditional semantics gives us the truth-conditions of utterances in terms of the constraints imposed by these meanings. They give us the utterance-bound or reflexive truth-conditions in terms of the utterance itself. They do not give us what is said or the proposition expressed by the speaker—speakers most often attempt to talk about things in the world, not about their own utterances. So our view of semantics is minimalist.

But being minimalist without allowing pragmatic ‘intrusion’ into semantic content does not necessarily involve sacrificing truth-conditional semantics, and being pushed into Grice’s circle. Our approach allows a clear-cut distinction between semantics and pragmatics that avoids it and offers the ideal toolkit to account for the relation between our knowledge of language and its use in communication.