The Semantics-Pragmatics Distinction: What It Is and Why It Matters
The distinction between semantics and pragmatics is easier to apply than to explain. Explaining it is complicated by the fact that many conflicting formulations have been proposed over the past sixty years. This might suggest that there is no one way of drawing the distinction and that how to draw it is merely a terminological question, a matter of arbitrary stipulation. In my view, though, these diverse formulations, despite their conflicts, all shed light on the distinction as it is commonly applied, in both linguistics and philosophy. Although it is generally clear what is at issue when people apply the distinction to specific linguistic phenomena, what is less clear, in some cases anyway, is whether a given phenomenon is semantic or pragmatic, or both. Fortunately, there are other phenomena that are uncontroversially semantic or, as the case may be, uncontroversially pragmatic. Their example will help us get clear on what the semantics-pragmatics distinction is.
Perhaps the main reason for introducing the semantics-pragmatics distinction is to provide a framework for explaining the variety of ways in which what a speaker conveys can fail to be fully determined by the (conventional) linguistic meaning of the sentence he utters:
• vagueness (and open-texture)
• semantic underdetermination
• non-truth-conditional content
• illocutionary force
The null hypothesis is that there is always some pragmatic explanation for how, in any given case, sentence meaning can underdetermine what the speaker means. For example, the null hypothesis about controversial claims of ambiguity (on tests for ambiguity see Atlas 1989, ch. 2) is that diverse uses of an expression are best explained not by different pieces of linguistic information (several conventional meanings) but by one piece of linguistic information combined with extralinguistic information. As Green has written,
The possibility of accounting for meaning properties and syntactic distributions of uses of linguistic expressions in terms of conversational inferences rather than semantic entailments or grammatical ill-formedness was welcomed by many linguists as a means of avoiding redundant analyses on the one hand and analyses which postulate rampant ambiguity on the other. (Green 1989, p. 106)
However, it is merely the null hypothesis that a given linguistic phenomenon has a pragmatic explanation. Particular phenomena and specific constructions obviously have to assessed on a case-by-case basis.
Another reason for invoking the semantics-pragmatics distinction is to shed light on a number of other distinctions:
• type vs. token
• sentence vs. utterance
• meaning vs. use
• context-invariant vs. context-sensitive meaning
• linguistic vs. speaker’s meaning
• literal vs. nonliteral use
• saying vs. implying
• content vs. force
Contrary to many of the formulations that have appeared since Morris’s initial formulation in 1938 (see the Appendix), the semantics-pragmatics distinction does not coincide with any of these other distinctions. Even so, it should respect them.
Properly formulated, it should take into account the wide range of items that have been described as semantic or pragmatic or both:
• phenomena: ambiguities, implications, presuppositions
• anomalies: paradoxes, contradictions, nonsense
• meanings: contents, interpretations
• knowledge: information, intuitions, processes
• rules and principles
It would require a detailed lexicographic analysis of the terms ‘semantic’ and ‘pragmatic’ to do full justice to their various applications. However, these should be kept in mind in the following discussion, which will emphasize the semantics-pragmatics distinction as it reflects the difference between linguistic and extralinguistic information available to language users.
Three disclaimers: (1) I will not use the term ‘pragmatics’ so broadly as to apply to the full range of phenomena falling under the heading of language use. That would take us too far afield, into such areas as social psychology, sociolinguistics, cultural anthropology, and rhetoric. I will restrict the discussion to those aspects of use that are directly related to acts of communication, and not even include perlocutionary acts and collateral speech acts (Bach and Harnish 1979, pp. 81-103). For it is in the context of communication that the question arises of where to draw the line between semantics and pragmatics. (2) Unless otherwise indicated, I will be treating sentences as the primary linguistic unit with respect to which the semantics-pragmatics distinction applies. This does not do justice to the fact that phrases can often be used as complete utterances nor to the alleged fact that there are certain intersentential semantic phenomena (for examples see Prince 1988). (3) I am not assuming any particular framework for semantics, formal or otherwise. I do assume that the meaning of a sentence depends entirely on the meanings of its constituents and its syntactic structure, but I am taking no position on whether sentence semantics should rely on the notion of truth conditions or propositions (however conceived). I will speak indifferently of a sentence’s truth condition, its truth-conditional content, and the proposition it expresses.
The semantics-pragmatics distinction has long been methodologically important in both linguistic and philosophy. It was implicit in philosophy a half century ago in discussions of pragmatic paradoxes and contextual implication (for a survey see Hungerland 1960), a forerunner of Grice’s notion of conversational implicature. It has often been invoked for corrective purposes. It was invoked by Strawson (1950), albeit implicitly, when he argued that Russell in his theory of descriptions had confused (linguistic) meaning and reference. Reference, Strawson contended, is something that speakers do, not words. Here Strawson anticipated the distinction between linguistic meaning and speaker’s meaning, which, along with the related distinction between what is said and what is implicated, became widely influential as the result of the work of Grice (collected in Grice 1989). Ironically, it was also Strawson (1952) who proposed a semantic account of presupposition. This was thought to make further trouble for Russell until presupposition came to be seen as a pragmatic phenomenon (Stalnaker 1974, Grice 1981/1989, ch. 17). Treating it as semantic led linguists down a blind alley for many years, searching for a solution to the “projection problem,” a problem that does not arise when presupposition is seen as pragmatic.
In the sixties, invoking the semantics-pragmatics distinction enabled philosophers to stem the excesses of ordinary language philosophy practiced by Austin and his followers. Their “linguistic botanizing” and rampant appeals to “what we would say” were overly ambitious in their attempt to get philosophical mileage out of subtle features of ordinary usage. Later Austin implicitly acknowledged the semantics-pragmatics distinction by contrasting locutionary and illocutionary acts (1960, pp. 93-101). Grice (1961/1989, ch. 15, and 1967/1989, ch. 2), by applying the notion of conversational implicature and wielding his “modified” Occam’s Razor, and Searle (1969, ch. 6), with his exposure of the “assertion” and the “speech act” fallacies, challenged proposed analyses of various epistemological, logical, and ethical terms, such as ‘looks,’ ‘knows,’ ‘or,’ and ‘good.’ Philosophers’ extravagant claims of semantic ambiguity were later decried by Kripke as “the lazy man’s approach in philosophy” (Kripke 1977, p. 268). Kripke illustrated how to avoid this by invoking the distinction between semantic reference and speaker’s reference to show that the difference between referential and attributive uses of definite descriptions, which had been thought to undermine Russell’s theory of descriptions, is merely pragmatic (see also Bach 1987a, chs. 5 & 6, and Neale
1990). Philosophers have since made similar moves on other important topics:
• contrastive explanations
• counterfactual conditionals
• domains of discourse
• illocutionary standardization
• indefinite descriptions
• logical form
• propositional attitude ascriptions
• relative terms
• similarity sentences
• speech act modifiers
In linguistics the category of pragmatics has served mainly as a bin for disposing of phenomena that would otherwise be the business of semantics (as part of grammar) to explain. Relegating such phenomena to pragmatics freed linguistic theory, already becoming more and more complex, of numerous additional complications. A notable exception to this strategy was the systematic attempt by generative semanticists, in their campaign to undermine the autonomy of syntax, to empty the “pragmatic wastebasket,” so-called by Bar-Hillel, who wisely advised linguists “to first bring some order into the contents of this wastebasket” (1971, p. 401). Many defied his advice and included everything but the kitchen sink in semantics. The performative hypothesis was the most prominent example (for a brief history see Sadock 1988). Historically, generative semantics is best remembered for generating the “linguistics wars” which have been chronicled in detail by Harris (1993).
In a more positive vein, the distinction between semantics and pragmatics has served to separate strictly linguistic facts about utterances from those that involve the actions, intentions, and inferences of language users (speaker-hearers). Howver, there are some linguistic phenomena that seem to straddle the semantics-pragmatics boundary:
• adjectival modification
• ambiguity vs. polysemy
• compounds and noun-noun pairs
• interpretation of quantificational phrases
And there are many linguistic phenomena which might seem at first glance to be pragmatic but, because of their syntactic basis, are arguably semantic. Space limitations prevent detailed discussion, but here are a few examples:
• constraints on anaphoric reference (c-command violations)
• empty categories
• implicit arguments
• implicit quantification over events
• thematic roles and complementation
• lexical alternations
• factive verbs
• negative polarity items
• it-clefts, wh-clefts, preposing, inversion, topicalization and other
devices of information packaging
• discourse modifiers and speech act adverbials
Most of these are syntactico-semantic phenomena that seem to explain certain co-occurrence and interpretational regularities. For example, implicit quantification over events helps explain the semantics of verbs, tense, and aspect, and the roles of adverbs (for a detailed account see Parsons 1990). The last three items on the list above involve semantic properties concerning use, not truth conditions. These properties are not pragmatic just because they pertain to use, for they are linguistically marked.
The semantics-pragmatics distinction has been formulated in various ways, generally without recognition that the different versions do not coincide. Historically, formulations have fallen into three main types, depending on which other distinction the semantics-pragmatics distinction was thought to coincide with:
• linguistic (conventional) meaning vs. use
• truth-conditional vs. non-truth-conditional meaning
• context independence vs. context dependence
The Appendix collects a variety of formulations that rely on one or another
of these distinctions. Here we will briefly review the three types and identify their shortcomings.
For purposes of clarifying the semantics-pragmatics distinction, the distinction between (linguistic) meaning and use is misleading at best. It neglects the case of expressions whose literal meaning is related to use. In addition to the obvious fact that features of illocutionary force can be linguistically encoded, notably by mood (Harnish 1994), there is the interesting case of expressions that are used to perform second-order or what Grice called “noncentral” speech acts (1989, p. 122). These are acts of commenting on the force, the point, or the role in the discourse of one’s utterance. Grice’s examples were limited to adverbs like ‘however’ and ‘moreover,’ but the list may be easily expanded to include such speech act adverbials as:
after all, anyway, at any rate, besides, be that as it may, by the way, first of all, finally, frankly, furthermore, if you want my opinion, in conclusion, indeed, in other words, now that you mention it, on the other hand, otherwise, speaking for myself, strictly speaking, to begin with, to oversimplify, to put it mildly
With these it seems that the only way to specify their semantic contibution (when they occur initially or are otherwise set off) is to specify how they are to be used (see Bach 1994a, pp. 148-149). Note that performatives do not fall in this category (Bach and Harnish 1979, ch. 10, and 1992).
Speech act adverbials also illustrate that an expression’s semantics can consist in non-truth-conditional meaning. Semantic presupposition would illustrate this too if there were such a thing, but Stalnaker (1974) and Grice (1981/1989, ch. 17) have made compelling cases that there is not. Even so, it may be granted that those linguistic devices, such as it-clefts and wh-clefts, which have been thought to encode semantic presupposition, do have some non-truth-conditional function. Like such devices as preposing, inversion, and topicalization, they serve to organize the presentation of information and to redirect focus.
Another example of non-truth-conditional meaning is provided by directly referential expressions, such as indexicals and demonstratives. As Kaplan (1989) has pointed out, if I say, “You are here,” it is not part of the truth condition of what I say that I am speaking to you at a certain place. The truth-conditional content of this sentence, relative to the context, is that the person being spoken to is where the speaker is, but this is a singular proposition involving that person and that place. It would be true even if the speaker were silent or not even there. What Kaplan calls the “character” of the terms ‘you’ and ‘here’ determines their contribution to the content (relative to the context) of the sentence being uttered, but character is not part of that content.
Now the notion of context is often invoked to explain how pragmatics complements semantics. It is a platitude that a sentence’s linguistic meaning generally does not determine what is said in its utterance and that the gap between linguistic meaning and what is said is filled by something called “context.” The intuitive idea behind this platitude is that there are different things that a speaker can mean, even when using his words in a thoroughly literal way (even that he is speaking literally is a matter of context—there is no such thing as Katz’s “null context” (1977, p. 14) but only informationally impoverished contexts). What one says in uttering the words can vary, so what fixes what one says cannot be facts about the words alone but must also include facts about the circumstances in which one is using them; those facts comprise the “context of utterance.”
It turns out, however, that context plays a role in semantics as well as pragmatics. As we saw above, with indexicals and demonstratives (and tense also), in these cases it is on the semantic side of the ledger that content varies with context. So the distinction between context invariance and context dependence does not provide the basis for drawing the semantics-pragmatics distinction. Confusion on this point, at least prior to Kaplan’s work, may have been caused by the use of the term ‘pragmatics,’ by such philosophers as Bar-Hillel (1954) and Montague (1974), to mean indexical semantics. Also, confusion has been caused by the fact that the limited notion of context relevant to the way in which the reference of terms like ‘you’ and ‘here’ is sensitive to context is rarely distinguished from the very broad notion of context that is relevant to pragmatics. Let me explain.
There are two sorts of contextual information, one much more restricted in scope and limited in role than the other. Information that plays the limited role of combining with linguistic information to determine content (in the sense of fixing it) is restricted to a short list of variables, such as the identity of the speaker and the hearer and the time and place of an utterance. Contextual information in the broad sense is anything that the hearer is to take into account to determine (in the sense of ascertain) the speaker’s communicative intention. It is often said that what a speaker means “depends on context,” is “determined by context” or is “a matter of context,” but this is not narrow context in the semantically relevant sense discussed above. When it is said that “Context makes it clear that ...,” what is meant is that there are items of information that the hearer can reasonably suppose the speaker to have intended him to take into account to determine what the speaker means. In this broad, pragmatic sense, which is also relevant to whether the speech act is being performed successfully and felicitiously, context does not literally determine content. So not just any sort of context variability is semantic. The variability must be provided for by lexical meaning and sentence grammar.
An important complication here is that there are many (indicative) sentences that do not express complete propositions even relative to a context. Though syntactically complete, they are semantically incomplete (Bach 1994a, 1994b). Here are some straightforward examples (given as the grammatical member of a minimal pair):
Fred finished/*completed yesterday.
Sam ate/*devoured earlier.
Jack tried/*attempted again later.
In each case, even though the verb lacks the complement that a similar verb requires, the sentence is syntactically complete. But the sentence is not semantically complete and the hearer must infer some completing material, e.g., ‘the job,’ ‘lunch,’ and ‘to call Jill,’ to understand the speaker. A pragmatic process of completion is required to arrive at a full proposition, at something with a determinate truth condition. These cases are also counterexamples to the truth-conditional conception of semantics. There is no theoretical basis for denying their semantic incompleteness by inventing hidden syntactic slots that must be filled in order for a complete proposition to be expressed. Rather, we must just acknowledge the fact that some sentences are semantically incomplete (and not just in need semantic values, as with indexicals) and that understanding utterances of them requires pragmatic supplementation.
There there is the case of sentences which, strictly and literally, express an unrestricted proposition but are typically used to convey something more specific:
I haven’t taken a bath [today].
Nobody [important] goes there any more because it is too crowded.
Abe didn’t have sex and [thereby] get infected; he got infected
and [then] had sex.
It is sometimes argued that because such sentences are standardly used without the bracketed material but such material is understood anyway, this material enters into what is said by the utterance, into its explicit content (Sperber and Wilson 1986, Recanati 1989). However, this material is not uttered and does not correspond to anything in the syntactic structure of the uttered sentence (even as an empty category in the sense of GB theory). So it is not explicit. It is not implied by what is said but that does not make it explicit either— it is implicit in what said. Such utterances are understood by way of a pragmatic process of expansion. Expansion, like completion, is a process required for the recognition of what I call “conversational implicitures,” as opposed to Gricean implic-a-tures (Bach 1994a, 1994b).
We have seen that the various traditional ways of formulating the semantics-pragmatics distinction either leave something out or draw the line at the wrong place. This is similar to what Levinson (1983, pp. 3-35) concluded in his survey of actual and possible formulations, although he ended up opting for the truth-conditional conception of semantics (he did so only provisionally and for historical rather than theoretical reasons). We need a better formulation. Otherwise, we will be left with what Horn (1988, p. 114) calls the “disjunctive attitude,” supposing, if only by default, that any phenomenon that is “too ill-behaved and variable to be treated coherently within the syntactic component, … [not] quite arbitrary enough for the lexicon or quite phonological enough for the phonology … must be pragmatic.”
A Better Formulation
What we need is a formulation of the semantics-pragmatics distinction that takes the above distinctions into account but does not rely on them too heavily. It needs to accommodate the following facts, that:
• only literal contents are semantically relevant
• some expressions, as a matter of meaning, are context-sensitive
• narrow context is relevant to semantics, broad context to pragmatics
• non-truth-conditional, use-related information can be linguistically
• rules for using expressions do not determine their actual use
These facts can all be accommodated on the supposition that semantic information pertains to linguistic expressions (sentences and their constituents), whereas pragmatic information pertains to utterances and facts surrounding them. Semantic information about sentences is part of sentence grammar, and it includes information about expressions whose meanings are relevant to use rather than to truth conditions. Linguistically encoded information can pertain to how the present utterance relates to the previous, to the topic of the present utterance, or to what the speaker is doing. That there are these sorts of linguistically encoded information shows that the business of sentence semantics cannot be confined to giving the proposition it expresses. Sentences can do more than express propositions. Also, as we have seen, there are sentences which do less than express propositions, because they are semantically incomplete.
Pragmatic information concerns facts relevant to making sense of a speaker’s utterance of a sentence (or other expression). The hearer thereby seeks to identify the speaker’s intention in making the utterance. In effect the hearer seeks to explain the fact that the speaker said what he said, in the way he said it. Because the intention is communicative, the hearer’s task of identifying it is driven partly by the assumption that the speaker intends him to do this. The speaker succeeds in communicating if the hearer identifies his intention in this way, for communicative intentions are intentions whose “fulfillment consists in their recognition” (Bach and Harnish 1979, p. 15). Pragmatics is concerned with whatever information is relevant, over and above the linguistic properties of a sentence, to understanding its utterance.
Consider some examples involving pronouns. There is no semantic basis for interpreting the pronouns one way in
Ann told Betty that she wanted to borrow her car.
and the opposite way in
Ann told Betty that she could not borrow her car.
The hearer relies on extralinguistic information to interpret one utterance one way and the other in the opposite way. The so-called “E-type” pronoun in
Most philosophers who have written a book think it is brilliant.
is interpreted as going proxy for the description ‘the book he wrote,’ and the “pronoun of laziness” in
John carried his luggage but everyone else checked it in.
is also interpreted descriptively— ‘it’ is not taken as being used to refer to John’s luggage (see Bach 1987a, pp. 258-261, and Neale 1990, pp. 180-191). In none of these cases is there any semantic requirement that the pronoun be interpreted in a certain way. The explanation for the preferred interpretation is pragmatic.
As part of linguistics and philosophy of language, pragmatics does not provide detailed explanations of how interpretation works in actual practice. This is a problem for cognitive and social psychology. For this reason it seems futile for linguists to seek a formal pragmatics. The task of explaining how utterances change context, for example, or how they exploit context, is not a job for linguistic theory by itself. The task is impossible without introducing general considerations about human reasoning and rational communication. Similarly, it is unreasonable to complain that theories like Grice’s account of conversational implicature provide no algorithm for conversational inference, so that, when applied to particular cases they simply pull implicatures out of a hat (see Sperber and Wilson 1986, Kempson 1988, Davies 1996). This is not just a problem for Grice’s theory.
At any rate, whereas semantic information is grammatically associated with the linguistic material uttered, pragmatic information arises only in relation to the act of uttering that material. (In fact, a stony silence can impart pragmatic information and thereby communicate something.) Whereas semantic information is encoded in what is uttered, pragmatic information is generated by the act of uttering it. No sentence encodes the fact that it is being uttered. Even the sentence ‘I am speaking’ is not analytic. The act of producing the utterance exploits the information encoded but by its very performance creates new information. That information, combined with the information encoded, provides the basis for the hearer’s identification of the speaker’s communicative intention. Contextual information is relevant to the hearer’s inference only insofar as it can reasonably be taken as intended to be taken into account, and that requires the supposition that the speaker is producing the utterance with the intention that it be taken into account. In contrast, the encoded information provides the input to the hearer’s inference in any context.
I foresee three main challenges to the semantics-pragmatics distinction, at least as it has been drawn here. They would contend that our formulation rests on one or another false assumption, (1) that semantics is autonomous from pragmatics, (2) that literal meaning is a viable notion, and (3) that communication involves Gricean reflexive intentions. In reply, I will suggest that each challenge identifies certain empirical complications for the application of the semantics-pragmatics distinction but does not undermine the distinction itself. For this reason, defending those assumptions against these challenges will help clarify the distinction.
1. Semantic autonomy
Occasionally it is claimed that pragmatics somehow impinges on semantics. Consider, for example, that words are often used in creative ways that depart from any of their conventional meanings, e.g., using nouns as verbs (Clark 1992, chs. 10 & 11) or cases of metonymy or deferred reference. Utterances of sentences like
Chicago always votes Democrat.
Philosophy has a tenure-track opening.
John was so thirsty he drank three mugs.
depart from their literal meanings, although people generally don’t think of such uses as not quite literal. In such cases the sentence possesses no meaning other than its usual conventional meaning(s)—it just is not being used in accordance with its meaning(s).
Whereas the difference between
Josh played his favorite violin yesterday.
Josh played his favorite concerto yesterday.
seems to have a clearly semantic basis (in terms of the different thematic roles of ‘concerto’ and ‘violin’), the autonomy of semantics relative to pramatics might be challenged on account of examples like the following:
John finished the newspaper/the letter/the meal.
Jack enjoyed the food/the movie/the day.
Jill wants a soda/a salad/a fork/a car.
What ordinarily counts as finishing a newspaper, a letter, or a meal varies from one case to another. Typically, you finish reading a newspaper, finish reading or writing a letter, and finish cooking or eating a meal. It seems to be a matter of semantics that verbs like ‘start’ and ‘finish’ are understood as having a verb (in gerundive form) in its complement, but it is a pragmatic matter which verb that is. The situation is similar with ‘enjoy’ and ‘want’ in the other examples above. Taken by themelves, these sentences are semantically incomplete in the sense described earlier. This does not mean, however, that the pragmatic processes required for understanding utterances of them somehow impinges upon their semantics. Nor is this shown, as Recanati (1989, 1995) has argued, by the fact that the completion is accomplished before the entire sentence is processed. The semantics-pragmatics distinction is concerned with the information available to the hearer, not with its real-time, online processing, which, it may be granted, is far from sequential.
Gazdar (1979, pp. 164-8) argues against the autonomy of semantics by means of examples of other sorts. One of his examples is:
To have a child and get married is worse than getting married
and having a child.
Since the alternatives here are semantically equivalent, given the logical conjunction reading of ‘and,’ how can we explain the force of an utterance of this sentence? Gazdar thinks that the correct pragmatic explanation has semantic import. However, as we saw earlier with a similar example, the proper pragmatic explanation appeals to the process of expansion, which has no semantic repercussions. It requires merely the supposition that the sentence is not being used with its strict, conventional meaning. On the expansion story, this follows from the fact that its utterance would normally be understood as including two implicit occurrences of the word ‘then.’
Gazdar also argues the the meaning of this permission sentence,
Inmates may smoke or drink.
is stronger than the combined meaning of the disjunction of permissions,
Inmates may smoke or inmates may drink.
and offers an account of its semantics that involves pragmatic considerations. However, this example may be disposed of in Gricean fashion. For if the utterance of ‘Inmates may smoke or drink’ is a permission, presumably it is a permission that can be complied with. The inmates can only be expected to interpret it in such a way that they can determine what they are permitted to do. If its import were either to permit smoking or to permit drinking without specifying which, there would be no way for an inmate to know how to comply with it.
2. Literal meaning
In formulating the semantics-pragmatics distinction, I have made no attempt to characterize the job of semantics. But as we have seen, there is more for a semantic theory of a language to do than to give a compositional account of the truth conditions of or the proposition expressed by each sentence, as a function of its syntactic structure and the semantic values of its constituents. But it would seem that the semantics-pragmatics distinction as formulated presupposes a well-defined level of lexical semantics and a viable distinction between literal and nonliteral meaning. There are several possible reasons for doubting that there is such a level.
I am not referring here to general skepticism about linguistic meaning, based on behaviorism about language use. Nor am I referring to doubts about linguistic meaning based on the familiar observation that most words are impossible to define, at least in terms of singly necessary and jointly sufficient conditions of application, and are vague or open-textured. These platitudes show not that Wittgenstein and Quine were right about linguistic meaning but only that it is not what philosophers used to think it to be. The two arguments I want to consider claim that the notion of literal meaning required by the semantics-pragmatics distinction cannot do justice to the general context-dependence of language.
One such argument is based on polysemy, as exemplified by the adjectives ‘sad,’ ‘long,’ and ‘dangerous’ as they occur in the following phrases:
sad person/sad face/sad day/sad music
long stick/long movie/long book
dangerous drug/dangerous game/dangerous road
The import of these adjectives varies with the noun they modify, but they do not seem to be cases of ambiguity, of linguistic coincidence (or else they would not have similarly-behaving counterparts in other languages). The argument is that since this variation in import is not due to ambiguity, it must have a pragmatic explanation. However, there is an alternative possibility, namely that polysemy involves what Pustejovsky (1995) calls “co-compositionality”: what varies from case to case is not a term’s semantic properties but how those properties interact with those of the term it is construction with. I do not endorse Pustejovsky’s ambitious theory of how this works, but certainly it is an improvement over what he calls “sense enumeration lexicons” (1995, p. 29). The relevant point here is that the phenomenon seems too systematic to be relegated to pragmatics. It does not justify the claim that pragmatics impinges on semantics.
The other argument relies on the observation that natural language is context-sensitive through and through. Contrary to the Gricean picture, it is argued, understanding an utterance is not just a matter of knowing the conventional meaning of what is uttered and, as necessary, resolving ambiguities, determining references, and distinguishing what is implicated from what is said. From this it is inferred that, even leaving aside disambiguation and reference fixing, there is often a pragmatic element in what is said, which, therefore, is not determined by the semantics of what is uttered. The general context-dependence of “interpretations” of utterances is supposed to show that what is said is not a purely semantic matter (Kempson 1988, Recanati 1989, 1996).
The trouble with these arguments is that they run roughshod over a number of straightforward distinctions. Ignoring Austin’s distinction between locutionary and illocutionary acts (1960, pp. 92-101), they fail to distinguish what is said from what is stated. They fail to distinguish what is said, in the strict and literal sense tied to the syntactic form of the sentence, with what is directly communicated in uttering the sentence, which may include elements that are not associated with anything in the sentence. They fail to distinguish context in the narrow sense described earlier, which is relevant to the interpretation of the sentence uttered, from context in the broad sense, which is relevant to the interpretation of its utterance, i.e., to identifying the speaker’s overall communicative intention.
It is trivially true that in the broad sense of context every utterance is context-sensitive. After all, it is never part of the meaning of a sentence that on a particular occasion of use it is being used to communicate. That is something the hearer presumes from the fact that the speaker is uttering the sentence. This “communicative presumption,” as Bach and Harnish (1979, p. 7) call it, comes into play even if what the speaker means does not extend beyond or depart in any way from the meaning of the sentence he utters. For it is never part of what a sentence encodes that it has to be used literally—the hearer must infer (even if only by default) that it is being used literally. The utterance does not carry its literalness on its sleeve. It might contain the word ‘literally,’ but even that word can be used nonliterally.
3. Gricean intentions
Our formulation of the semantics-pragmatics distinction relies heavily on a Gricean conception of communicative intentions, for it takes as key to the pragmatic side the idea that in any communication situation extralinguistic information comes into play because, and only because, such information is intended, or taken by the hearer as intended, to be taken into account. So another way of challenging our formulation of the semantics-pragmatics distinction would be to challenge the Gricean view of communication.
Such a challenge has been mounted by Sperber and Wilson (1986) with their so-called relevance theory (in relevance theory ‘relevance’ does not mean relevance). The “principle of relevance” states that, as a matter of general cognitive fact, people seek to maximize contextual effects at a minimum of processing cost. Apart from not explaining how to measure contextual effects and processing costs, how to make them commensurate with each other, or why there is always a unique way satisfying the principle (Bach and Harnish 1987), relevance theory ignores the fundamental fact that the hearer is to recognize the speaker’s intention partly on the basis that he is so intended. Instead, relevance theory seems to assume that in the context of communication everyone is an applied relevance theorist. That is, people are supposed to gear their utterances to their listeners’ inherent propensity to discover maximize contextual effects at a minimum of processing cost.
Contrary to Sperber and Wilson’s complaint that Grice’s account requires the hearer to know what the speaker’s intention is in order to identify it (1986, pp. 28-31 and 256-7), there is nothing paradoxical about the reflexivity of communicative intentions (Bach 1987b). For all that this reflexivity involves is that the hearer is to take into account the fact that he is intended to identify the speaker’s intention, whatever that intention is. That is, the hearer may presume that the speaker’s intention is identifiable under the circumastances. This leaves open, of course, the question of how the hearer, even when armed with that presumption, manages to figure out the speaker’s intention. The basic shortcoming of relevance theory is that it provides no place for this presumption. It replaces the distinctive feature of rational communication with an a priori generalization about human cognitive processes.
Relevance theory does not do justice to the fact that whereas semantic information is associated with the sentence uttered, pragmatic information is tied to the fact that the speaker is uttering it. Any contextual information, whether about the immediate situation (including what has been said previously), the conversants’ relationship, or their background knowledge, is relevant (in the ordinary sense of ‘relevant’) to the interpretation of the utterance only because it is intended, or can reasonably be taken as intended, to be taken into account. That is why, for example, the pragmatic paradoxes philosophers discussed a half century ago arise only because the speaker actually utters the seemingly paradox sentence, e.g.,
It is raining but I don’t believe it.
I am not speaking.
I am lying.
This fact was also essential to the notion of contextual implication that predated Grice’s notion of conversational implicature. It is essential to understanding why presupposition is a pragmatic phenomenon, something done by speakers not by their words, and why implicatures “are carried not by what is said but only by the saying of what is said, or by ‘putting it that way’” (Grice 1967/1989, p. 39).
There is nothing new in our formulation of the semantics-pragmatics distinction. It is relies on the familiar distinctions between sentences and utterances and between linguistic (grammatical) and extralinguistic information. What is new, if anything, is the way in which it accommodates various other distinctions without attempting to reduce the semantics-pragmatics distinction to any of these. The present formulation has aimed to:
1. simplify the task of semantic theory by identifying a principled reason which, when applicable, justifies not addressing certain phenomena that might otherwise seem the business of semantics to explain,
2. keep open the option that certain seemingly pragmatic phenomena might be correlated with or constrained by syntactic features in such a way as merit classification as semantic,
3. avoid burdening semantics with the false assumption that every (indicative) sentence expresses a proposition (even relative to a context) and does nothing else,
4. accommodate the fact that contextual parameters and speech act information can be linguistically encoded, but without equating context in the broad sense relevant to communication with context in the narrow sense relevant to providing values for the contextual parameters that determine or at least constrain the force of literal utterances,
5. respect intuitions about what is and what is not semantic without always accepting them at face value (sometimes intuitions are better accounted for not by explaining them but by explaining them away), and
6. justify and preserve the distinction between interpretation of a sentence and interpretation of an utterance and thereby the distinction between narrow linguistic competence and general communicative rationality.
These broad features of our account do not determine on which side of the the semantics-pragmatics boundary particular linguistic phenomena fall. Whether a given phenomenon has a semantic or a pragmatic explanation or, as is often the case, some combination of both, must be settled on a case-by-case basis. Obviously it is one thing to formulate the semantics-pragmatics distinction and another thing to apply it.
Appendix: A Chronology of Formulations
Semantics deals with the relation of signs to … objects which they may or do denote. Pragmatics concerns the relation of signs to their interpreters. (1938/1971, pp. 35, 43)
Syntax studies sentences, semantics studies propositions. Pragmatics is the study of linguistic acts and the contexts in which they are performed. There are two major types of problems to be solved within pragmatics: first, to define interesting types of speech acts and speech products; second, to characterize the features of the speech context which help determine which proposition is expressed by a given sentence. … It is a semantic problem to specify the rules for matching up sentences of a natural language with the propositions that they express. In most cases, however, the rules will not match sentences directly with propositions, but will match sentences with propositions relative to features of the context in which the sentence is used. These contextual features are part of the subject matter of pragmatics. (p. 383)
[I] draw the theoretical line between semantic interpretation and pragmatic interpretation by taking the semantic component to properly represent only those aspects of the meaning of the sentence that an ideal speaker-hearer of the language would know in an anonymous letter situation, … [where there is] no clue whatever about the motive, circumstances of transmission, or any other factor relevant to understanding the sentence on the basis of its context of utterance. (p. 14)
pragmatics = meaning - truth conditions (p. 2)
What we need in addition is some function that tells us about the meaning of utterances. … The domain of this pragmatic function is the set of utterances, which are pairs of sentences and contexts, so that “for each utterance, our function will return as a value a new context—the context as changed by the sentence uttered … . And we can treat the meaning of the utterance as the difference between the original context and the context arrived at by utterance the sentence. [This applies to only] a restricted subset of pragmatic aspects of meaning. (pp. 4-5)
Semantics provides a complete account of sentence meaning for the language, [by] recursively specifying the truth conditions of the sentences of the language. … Pragmatics provides an account of how sentences are used in utterances to convey information in context. (p. 139)
The Oxford Companion to Philosophy (Fotion 1995):
Pragmatics is the study of language which focuses attention on the users and the context of language use rather than on reference, truth, or grammar. (p. 709)
The Cambridge Dictionary of Philosophy (Lycan 1995):
Pragmatics studies the use of language in context, and the context-dependence of various aspects of linguistic interpretation. … [Its branches include the theory of how] one and the same sentence can express different meanings or propositions from context to context, owing to ambiguity or indexicality or both, … speech act theory, and the theory of conversational implicature. (p. 588)
The Blackwell Companion to Philosophy (Davies 1996):
The distinction between semantics and pragmatics is, roughly, the distinction between the significance conventionally or literally attached to words, and thence to whole sentences, and the further significance that can be worked out, by more general principles, using contextual information. (p. 124)
Atlas, J. (1989): Philosophy Without Ambiguity. Oxford: Oxford University Press.
Austin, J. (1960): How To Do Things With Words. Oxford: Oxford University Press.
Bach, K. (1987a): Thought and Reference. Oxford: Oxford University Press.
Bach, K. (1987b): On communicative intentions. Mind & Language, 2, 141-154
Bach, K. (1994a): Conversational impliciture. Mind & Language, 9, 124-162.
Bach, K. (1994b): Semantic slack. In S. Tsohadzidis (ed.): Foundations of Speech Act
Theory. London: Routledge, pp. 267-291.
Bach, K. and R. Harnish (1979): Linguistic Communication and Speech Acts.
Cambridge, Mass: MIT Press
Bach, K. and R. Harnish (1987): Relevant questions. Brain and Behavioral Sciences, 10, 711-712.
Bach, K. and R. Harnish (1992): How performatives really work: a reply to Searle. Linguistics
and Philosophy, 15, 93-110.
Bar-Hillel, Y. (1954): Indexical expressions. Mind, 63, 359-379.
Bar-Hillel, Y. (1971): Out of the pragmatic wastebasket. Linguistic Inquiry, 2, pp. 401-407.
Clark, H. (1992): Arenas of Language Use. Chicago: University of Chicago Press.
Davies, M. (1995): Philosophy of language. In N. Bunnin and E. Tsui-James (eds.):
The Blackwell Companion to Philosophy. Oxford: Blackwell, pp. 90-139.
Fotion, N. (1995): Pragmatics. In T. Honderich (ed.): The Oxford Companion to
Philosophy. Oxford: Oxford University Press, p. 709.
Gazdar, G. (1979): Pragmatics: Implicature, Presupposition, and Logical Form.
London: Academic Press.
Green, G. (1989): Pragmatics and Natural Language Understanding. Hillsdale, NJ:
Lawrence Erlbaum Associates
Grice, P. (1989): Studies in the Way of Words. Cambridge, Mass: Harvard University Press.
Harnish, R. (1994): Mood, meaning, and speech Acts. In S. Tsohadzidis (ed.):
Foundations of Speech Act Theory. London: Routledge, pp. 407-459.
Harris, R. (1993): The Linguistics Wars. Oxford: Oxford University Press.
Horn, L. (1988): Pragmatic theory. In Newmeyer, F. (ed.): Linguistics: The Cambridge
Survey, Vol. I. Cambridge, Eng.: Cambridge University Press, pp. 113-145.
Hungerland, I. (1960): Contextual implication. Inquiry, 3, 211-258.
Kaplan, D. (1989): Demonstratives. In J. Almog, J. Perry, and H. Wettstein (eds.):
Themes from Kaplan. Oxford: Oxford University Press, pp. 481-563.
Katz, J. (1977): Propositional Structure and Illocutionary Force. New York: Crowell.
Kempson, R. (1988): Grammar and conversational principles. In Newmeyer, F. (ed.):
Linguistics: The Cambridge Survey, Vol. II. Cambridge, Eng.: Cambridge University Press, pp. 139-163.
Kripke, S. (1977): Speaker’s reference and semantic reference. Midwest Studies in Philosophy, 2, 255-296.
Levinson, S. (1983): Pragmatics. Cambridge, Eng.: Cambridge University Press.
Lycan, W. (1995): Philosophy of language. In R. Audi (ed.): The Cambridge Dictionary
of Philosophy. Cambridge, Eng.: Cambridge University Press, pp. 586-589.
Montague, R. (1974): Pragmatics. In R. Thomason (ed.): Formal Philosophy. New
Haven: Yale University Press.
Morris, C. (1938/1971): Foundations of the theory of signs. In Writings on the Theory
of Signs. The Hague: Mouton, pp. 17-74.
Neale, S. (1990): Descriptions. Cambridge, Mass: MIT Press.
Parsons, T. (1990): Events in the Semantics of English. Cambridge, Mass: MIT Press.
Prince, E. (1988): Discourse analysis. In Newmeyer, F. (ed.): Linguistics: The Cambridge
Survey, Vol. II:. Cambridge, Eng.: Cambridge University Press, pp. 164-182.
Pustejovsky, J. (1995): The Generative Lexicon. Cambridge, Mass: MIT Press.
Recanati, F. (1989): The pragmatics of what is said. Mind & Language, 4, 295-329.
Recanati, F. (1995): The alleged priority of literal interpretation. Cognitive Science, 19, 207-232.
Recanati, F. (1996): Domains of discourse. Linguistics and Philosophy, 19, 445-475.
Sadock, J. (1988): Speech act distinctions in grammar. In Newmeyer, F. (ed.): Linguistics: The
Cambridge Survey, Vol II. Cambridge, Eng.: Cambridge University Press, pp. 183-197.
Searle, J. (1969): Speech Acts. Cambridge, Eng.: Cambridge University Press.
Sperber, D. and D. Wilson (1986): Relevance. Cambridge, Mass: Harvard University Press.
Stalnaker, R. (1972): Pragmatics. In G. Harman and D. Davidson (eds.): Semantics of
Natural Language. Dordrecht: Reidel, pp. 380-397.
Stalnaker, R. (1974): Pragmatic Presuppositions. In M. Munitz and P. Unger (eds.):
Semantics and Philosophy. New York: New York University Press, pp. 197-213.
Strawson, P. (1950): On referring. Mind, 59, 320-344.
Strawson, P. (1952): Introduction to Logical Theory. London: Methuen.