[in J. Keim Campbell, M. O'Rourke, and D. Shier, eds., Meaning and Truth, New York: Seven Bridges Press (2002), pp. 21-33]


Kent Bach


From ethics to epistemology to metaphysics, it is common for philosophers to appeal to “intuitions” about cases to identify counterexamples to one view and to find support for another. It would be interesting to examine the evidential status of such intuitions, snap judgments, gut reactions, or whatever you want to call them, but in this paper I will not be talking about moral, epistemological, or metaphysical intuitions. I’ll be focusing on semantic ones. In fact, I’ll be focusing on semantic intuitions about sentences, not individual words (although the contributions of individual words may ultimately be at issue in some of these cases), and on closely related intuitions about what is said in utterances of those sentences. Such intuitions play an important role in philosophy of language. For example, intuitions about the informativeness of identity statements give rise to Frege’s problem. Intuitions about the failure of substitution in attitude contexts are used to impose a constraint on an adequate theory of attitude ascriptions. And intuitions about sentences containing definite descriptions used referentially or incomplete definite descriptions have been relied upon to cast doubt on Russell’s theory of descriptions.

        Although I have my doubts about such appeals to intuition in regard to these long-standing problems in the philosophy of language, I will not state them here. The cases I will consider are philosophically less interesting, but they are the sorts of examples that provide data for answers to the controversial question of how to draw the line between semantics and pragmatics (for my answer see Bach 1999). Such examples are often cited, e.g. by François Recanati in his paper, “The Pragmatics of What is Said” (1989), and by the psycholinguists Ray Gibbs and Jessica Moise, in their paper “Pragmatics in Understanding What is Said” (1997), to undermine the orthodox view that, to put it crudely, semantics provides input to pragmatics without any feedback from semantics. Here are a few examples thought to provide evidence for so-called “pragmatic intrusion,” in which pragmatic factors allegedly contribute to semantic interpretation:

        (1)       Billy will get promoted if he works hard.

which is intuitively understood to mean that Billy will get promoted if and only if he works hard,

        (2)       Mary has three cars.

which is intuitively understood to mean that Mary has exactly three cars,

        (3)       Bobby hasn’t taken a bath.

which is intuitively understood to mean that Bobby hasn’t taken a bath lately, and

        (4)       Molly got infected and went to the hospital.

which is intuitively understood to mean that Molly got infected and then, because of her infection, went to the hospital.

        As I see it, these are all examples of what I call (in Bach 1994a) conversational “impliciture” (as opposed to Grice’s implic-a-ture). In each case, I claim, what the speaker means is distinct from he is saying, because what he means includes an implicit qualification on what he is saying, something that is not really part of what is said. However, people’s intuitions tend not to be sensitive to the difference, at least not until they’re sensitized. That’s because they tend to ignore what I’ll call the “Syntactic Correlation Constraint,” as expressed by Paul Grice’s stipulation that what is said must correspond to “the elements of [the sentence], their order, and their syntactic character” (1989, p. 87).

        Since this is pretty much all that Grice says about what is said, I should add a few things about how I understand it. What is said is determined compositionally by the semantic contents of the constituents (“elements”) of the sentence as a function of their syntactic relationship. To allow for the presence of tense and sometimes indexicals, we should add that what is said in a context is the semantic content of the sentence relative to that context. Notice that I do not speak of the semantic contents of utterances. If “utterance” means what is uttered, then an utterance is just a sentence. And if “utterance” means an act of uttering, then the content of an utterance is really the content of the speaker’s communicative intention, which can depart in various ways from semantic content. Also, although what is said is sometimes described as the “proposition expressed” by the sentence (relative to the context), this mistakenly assumes that every sentence expresses a complete proposition. In fact, the syntactic requirements on well-formed sentences do not exclude the case of sentences whose semantic interpretation is not a complete proposition. There are many sorts of sentence that do not express a complete proposition, not even relative to a context (see (15) and (16) below and Bach 1994a, sec. 2).

        At any rate, the Syntactic Correlation Constraint entails that if any element of the content of an utterance, i.e., of what the speaker intends to convey, does not correspond to any element of the sentence being uttered, it is not part of what is said. So when people intuitively think that what is said includes such elements, their intuitions are illicitly including something that just isn’t there. Now Grice did not think that people’s intuitions deserve cavalier dismissal. He could have just explained them away by appealing to the distinction between what is said and what is implicated in uttering a sentence or to the distinction between what a sentence means and what a speaker means in uttering it, but he was worried by what struck him as a kind of paradox:

We must of course give due … weight to intuitions. … For in order that a nonconventional implicature should be present in a given case, my account requires that a speaker shall be able to utilize the conventional meaning of a sentence. … This … seems to lead to a sort of paradox: if we, as speakers, have the requisite knowledge of the conventional meaning of sentences we employ to implicate, when uttering them, something the implication of which depends on the conventional meaning in question, how can we, as theorists, have difficulty with respect to just those cases in deciding where conventional meaning ends and implicature begins? If it is true, for example, that one who says that A or B implicates the existence of non-truth-functional grounds for A or B, how can there be any doubt whether the word ‘or’ has a strong or weak sense? I hope that I can provide the answer to this question, but I am not certain that I can. (1989, p. 49)


In my opinion, Grice accorded intuitions too much respect. In fact, there’s nothing at all paradoxical about how theorists can disagree about matters on which people’s intuitions tend to agree.

        However, Grice’s paradox will seem troubling to anyone who supposes that the central aim of semantics is to account for such intuitions, especially ones about the truth-conditions of sentences. But this anxiety, like many others, is unfounded. It is the central aim of semantics to account for semantic facts, not intuitions. People’s spontaneous judgments or “intuitions” provide data for semantics, but it is an open question to what extent they reveal semantic facts and should therefore be explained rather than explained away. Since, as I am suggesting, they are often responsive to non-semantic information, to what is implicit in what is said but not part of it, they should be treated cautiously. They should certainly not be given the respect accorded to them by Recanati’s so-called “Availability Principle,” which prescribes that intuitions about what is said be “preserved” in our theorizing. Nor should they be taken as seriously as they are by Gibbs and Moise, who “examined people’s intuitions” and claimed to find that data about what people say about what is said “lend support to theories of utterance interpretation [according to which] pragmatics strongly influences people’s understanding of what speakers say [as well as what they] communicate” (1997, p. 51).

        Aside from the question of the reliability of such intuitions and their relevance to semantics and its relation to pragmatics, there is the question of what role, if any, they play in the process of communication. It seems their role is marginal at best. In the course of speaking and listening to one another, we generally don’t consciously reflect on the semantic contents of the sentences we hear or on what is said in their utterance. We are focused on what we are communicating or on what is being communicated to us, not on what is said. Moreover, we don’t have to be able to make accurate judgments about what information is semantic and what is not in order to be sensitive to semantic information. To “preserve intuitions” in our theorizing about what is said would be like relying on the intuitions of unsophisticated moviegoers about the effects of editing on a film. Although people’s cinematic experience is dramatically affected by such factors as cuts and camera angles, there is no reason to suppose that their intuitions are reliable about what produces what effects. Intuitions about what is said may be similarly insensitive to the difference between the contribution that is made by the semantic content of a sentence and that made by extralinguistic factors to what an utterance communicates. So, I say, what worried Grice was not a real paradox but just an ordinary philosophical problem.

        Let’s get down to cases. In discussing them, I should stress that I am not claiming that semantic intuitions are totally unreliable and shouldn’t be trusted at all. Rather, they should be relied upon judiciously, and only after being fed an ample diet of examples, including contrasting ones. For instance, although it might intuitively seem that (1),

        (1)       John will get promoted if he works hard.

says or at least entails that John won’t get promoted if John doesn’t work hard, this apparent entailment can be explicitly canceled without contradiction:

        (1x)  John will get promoted if he works hard, though he might get promoted

                      even if he doesn’t work hard.

And intuitively it does not seem redundant to utter the strengthened version of (1),

        (1+) John will get promoted if and only if he works hard.

Here are a few more examples of faulty intuitions. In each case, rather than take the intuition at face value, we can describe what is going on in a way that explains both the occurrence of the intuition and its falsity.

        (5)       Jack and Jill went up the hill.

        (6)       Jack and Jill are engaged.

        (7)       Jill got married and became pregnant.

Although these sentences express complete propositions, in uttering them a speaker is likely to have meant something more specific, a qualified version of what he said:

        (5+) Jack and Jill went up the hill together.

        (6+) Jack and Jill are engaged to each other.

        (7+) Jill got married and then became pregnant.

where the italicized words, which are not part of the original sentence, indicate part of what the speaker meant in uttering (5), (6), or (7). He would have to utter those words (or roughly equivalent words--the exact words don’t matter) to make what he meant more fully explicit (let’s call the fuller version an “expansion” of the original and what is left out is its “implicit qualification”). Utterances of sentences like (5) - (7) illustrate what I call “sentence nonliterality” (Bach, 1994b, pp. 69-72), as opposed to constituent nonliterality, since no expression in the sentence is being used nonliterally. What the speaker means is not the exact proposition, as compositionally determined, that is expressed by the sentence, and the difference between the two propositions is not attributable to any particular constituent of the sentence. A speaker who utters (6), for example, is not saying that Jack and Jill are engaged to each other, any more than he would be saying this if he uttered “Jack and his sister Jill are engaged.” That he means they’re engaged to each other is implicit in what he is saying or, more precisely, in his saying of it. It is not part of what is said, since it passes Grice’s test of cancellability. That is, it may be taken back without contradiction. There is no contradiction in uttering (6x),

        (6x)  Jack and Jill are engaged but not to each other.

 or, for that matter, (5x) or (7x),

        (5x)  Jack and Jill went up the hill but not together.

        (7x)  Jill got married and became pregnant but not in that order.

Utterances of (5), (6), and (7) are all cancelable.

        Even so, for many people the apparent content of each of sentences (5) - (7) includes something that is not predictable from the compositional semantics of the uttered sentence. Taking semantic intuitions seriously would make life miserable for semanticists, even more miserable than it already is. It would require doing semantics from the top down: starting with the supposed meaning of a sentence and working down to the meanings of its constituents. Covert constituents would have to be posited to provide the “residue” of meaning not accounted for by the overt ones. Or, alternatively, special meanings, hence ambiguity, would have to be attributed to certain of the overt constituents, insofar as their ordinary meanings seem not to make the right contribution to what is said. All this can be avoided if we don’t take people’s seemingly semantic intuitions too seriously.

        Why   are such intuitions unreliable about the semantic contents of sentences like those we’ve considered? Part of the reason is that typical utterances of them involve sentence nonliterality. Unlike cases of metaphor or metonymy, there are no constituents which intuitively are being used nonliterally. Moreover, there is a recurrent pattern of nonliterality associated with such sentences. Phenomenologically, their nonliteral use seems literal, at least insofar as our intuitions are insensitive to the difference between conventionalization and mere standardization (Bach 1998). As with what Grice called “generalized” conversational implicature, where there must be specific contextual reasons for supposing that an implicature is not present (for a monumental study of generalized conversational implicature, see Levinson 2000), in the above cases the sentence is typically used to communicate something that is not predictable from its meaning alone. So it’s no wonder that when people are asked for their intuitions about such a sentence, they will tend to imagine it uttered in a normal context and count its typical implicit qualification as part of its content. They tend to attribute something to the conventional meaning of the sentence which in fact is attributable only to typical utterances of it.

        Recent experiments by Gibbs and Moise (1997) have sought to establish the reliability of semantic intuitions, as with examples like (8):

        (8)       Martha gave John her key and he opened the door.

 People judge that part of what is said is that John opened the door with the key Martha gave him. But Gibbs and Moise’s experimental design was clearly flawed. For one thing, it imposed a false dichotomy on their subjects by forcing them to choose between what is said and what is implicated. Subjects weren’t offered the in-between category of implicit qualification (that which is implicit in the saying of what is said). Also, they were not given the opportunity to make cancellability judgments or comparative judgments about what is said by explicitly qualified utterances as opposed to unqualified ones. Gibbs and Moise didn’t ask subjects if there is a contradiction in a sentence like (8x),

        (8x)  Martha gave John her key and he opened the door, but not with the key she

                      gave him.

Gibbs and Moise predict that subjects would find a contradiction here; I predict that they wouldn’t. Similarly, they didn’t ask subjects to compare (8) with (8+),

        (8+)       Martha gave John her key and he opened the door with the key she gave him.

and to judge whether they say the same thing. Gibbs and Moise predict that subjects would judge that (8) and (8+) do say the same thing; I predict that they wouldn’t.

        Even if Gibbs and Moise are right about people’s untutored intuitions about the original examples, it would be easy to sensitize their intuitions about what is said to Grice’s cancellability test for what is not said. Just present them with sentences like (5) - (7) followed by cancellations of what is not explicit in the utterance, as in (5x) - (7x) above. Ask them if they sense a contradiction or just a clarification. Or ask them, with a stress on “say,” whether what a speaker says in uttering explicitly qualified versions of (5), (6), or (7), i.e., (5+) - (7+) above, is the same as what a speaker says with (5), (6), and (7) themselves, and they are likely to discern the difference. If so, this contradicts the intuition that the implicit qualifications are part of what is said in the original utterances. So the verdict of intuition is reversed when we appeal to people’s cancellability judgments and their comparative judgments about what is said by explicitly qualified vs. unqualified utterances.

        I could discuss Gibbs and Moise’s experiments in detail (in fact, Nicolle and Clark (1999) have done so, and report that their own experiments often delivered different results, sometimes with people deeming clear cases of implicature to count as what is said), but the main difficulty with their research, which shows how misguided it was, is that it tested for the wrong thing. They thought they could get data about what is said, and thereby test the empirical validity of Recanati’s Availability Principle, by asking people what is said by a given utterance, or by asking them whether something that is conveyed by a given utterance is implicated or merely said. Evidently they assumed that what people say about what is said is strongly indicative of what is said. In fact, what it is indicative of is how people apply the phrase “what is said” and perhaps of what they mean by the word “say.” It tells us little about what is said, much less about the cognitive processes whereby people understand utterances.

        To appreciate how small a role semantic intuitions play in utterance comprehension, consider the case of ambiguity. There are many ambiguous sentences one of whose meanings is far more likely to be operative than the other. The following headlines illustrate what I mean:












We find these headlines funny because we notice their unintended meanings, but evidently these weren’t noticed by the editors who allowed the headlines to run. Similarly, we are not likely to notice the ambiguity of (9),

        (9)       Bill scratched the car with an umbrella.

considered by itself, or the ambiguity of (10),

        (10)       Bill scratched the car with a broken tail light.

considered by itself, but their ambiguity is obvious as soon as we compare them with each other or with the obviously ambiguous sentence,

        (11)       Bill scratched the car with a broken antenna.

Similarly, by itself neither of the following sentences is likely to seem ambiguous,

        (12)       The soldiers exchanged their arms for food.

        (13)       The soldiers used their arms to protect their faces.

but their ambiguity is evident once we compare them with each other or with the obviously ambiguous (14),

        (14)       The soldiers celebrated by waving their arms in the air.

Interestingly, it has long been known (Lackner and Garrett 1973) that people unconsciously access irrelevant meanings of ambiguous words, but only very briefly, and only immediately (about 150 msec.) after hearing them. People’s failure consciously to “intuit” the irrelevant meaning has no bearing on the process of utterance comprehension. And no one would seriously claim that their failure to do so is strong evidence against the existence of the ambiguity.

        I have stressed that people’s semantic intuitions tend not to respect the Syntactic Correlation Constraint. Now I’d like to consider some objections to that constraint and to the so-called “minimalist” conception of semantic content that goes with it.

        There are two objections that I’ll mention just briefly. The first notes the fact mentioned earlier, that some sentences, like (15) and (16), do not express complete propositions, not even relative to a context.

        (15)         Bonnie is ready.  (for what?)

        (16)         Clyde is finished.  (doing what?)

In such cases, there is something not semantically specified that is needed to yield a complete proposition. However, so the objection goes, what is said must be a complete proposition. Therefore, what is said in such cases is not a projection of the syntax of the sentence; it includes some element that does not correspond to any constituent, or feature of a constituent, of the sentence. However, why must what is said be a complete proposition? What’s wrong with using (15IQ) and (16IQ) to report, and report fully, what a speaker says in uttering (15) or (16)?

        (15IQ)  S said that Bonnie is ready.

        (16IQ)  S said that Clyde is finished.

It may be true that a speaker, in using a sentence to communicate something, must communicate a complete proposition, but it hardly follows that any sentence used to communicate a complete proposition must itself express one. Sentences that are syntactically complete but semantically incomplete do not. To understand such utterances the hearer must figure out how the speaker intends what is said to be turned into a complete proposition. I call this process “completion.” Utterances requiring completion, like those requiring expansion (utterances with an implicit qualification on what is said), carry implicitures along with what is said.

        A second objection is based on resisting what Robyn Carston has called “the compulsion to treat all pragmatically derived meaning as implicature” (1988, p. 176). Once it is recognized that the contribution of pragmatic processes is not limited to the determination of implicatures, “there is no reason,” according to Carston, “why pragmatics cannot contribute to the explicature, the truth-conditional content of the utterance,” which she equates with what is said. However, Carston is implicitly assuming that if something is not implicated, it is part of what is said. Since what she calls the explicature, which needn’t be fully explicit, can be the truth-conditional content of the result of an expansion or completion of the utterance, it cannot be identified with what is said.

        A third objection is based on the fact that on minimalism what is said is often false even in cases when the utterance of the sentence in question is true. Sentence (3), for example,

        (3)       Bobby hasn’t taken a bath.

though likely to be used to convey that Bobby hasn’t taken a bath lately, can itself be true only in the unlikely event that Bobby has never taken a bath. Similarly, (17) and (18), though used to convey truths, are likely to be literally false by minimalist standards:

        (17)     That car doesn’t look expensive--it is expensive.

        (18)     Nobody goes there any more--it’s too crowded.  (once uttered by Yogi Berra)

Why should the prediction that these sentences are literally false lead to an objection to minimalism? Because intuitively they are true. Well, that’s one consideration, but it has little weight, once we invoke the distinction between what is said and what is meant, and remember that intuitions tend to be insensitive to that distinction and to be responsive to as implicit qualifications, as explicitly included in (17+) and (18+),

        (17+)         That car doesn’t merely look expensive--it is expensive.

        (18+)         Nobody important goes there any more--it’s too crowded.

Another basis for the objection is that, given this distinction, it can only be the obvious falsity of what is said that explains the hearer’s inference to what the speaker means. Evidently, the objection assumes that minimalism must treat these cases as Gricean quality implicatures. In those cases, the hearer’s inference is triggered by his recognition that the utterance, if taken at face value, violates Grice’s first maxim of quality, “Do not say what you believe to be false” (1989, p. 27). However, the obvious falsity of these sentences has nothing to do with how the hearer figures out what the speaker is conveying (or with how the speaker intends him to do so). These cases we’re concerned with are quite unlike an utterance of, say, “I could eat a million of those potato chips,” which conveys how irresistible they are. In that case it is the obvious falsity of what is said that triggers the hearer’s search for something other than what is said. But in the examples in question it is not obvious falsity that does that. Consider one more example. Suppose a child is crying because of a tiny cut and his mother tries to calm him by uttering (19),

        (19) You’re not going to die.

 Obviously the mother is not assuring the kid of his ultimate immortality. But the operative pragmatic anomaly here is not obvious falsity but lack of relevant specificity. This is clear if we consider a positive version of the same utterance. An oncologist could say to a cancer patient who demands a frank prognosis, “I’m sorry to tell you, but there is nothing I can do. You’re going to die.” Presumably the patient won’t take the doctor to mean that he, like anyone else, is mortal. But it’s not the obvious truth of what is said that enables him to understand the doctor, it’s the presumption that the doctor is telling him something relevant to his medical condition. Similarly, utterances of the negations of the previous examples would typically have the same implicit qualifications as utterances of those sentences themselves. Obvious truth, like obvious falsity, has no bearing on the hearer’s inference in these cases.

        Another objection to minimalism claims that even if we accept the strict, minimalist conception of what is said, what is said in that sense can have no psychological reality unless it is something that a hearer must identify before inferring the speaker’s communicative intention. In other words, for what is said to matter psychologically, the hearer must identify what is said before identifying what is meant. But, so the objection goes, introspectively at least it seems that in many cases the first proposition one arrives at is not the “minimal” proposition (as Recanati 1989 calls it), the proposition which, according to the Syntactic Correlation Constraint, comprises what is said. Even if this is so, that is no objection to semantic minimalism. The process of utterance comprehension is obviously a very interesting topic for psychology, but it’s hard to see why facts about hearers’ cognitive processes should be relevant to what a speaker says. How could the fact (if it is a fact) that what is said sometimes has no psychological reality for the hearer show that it is a mere abstraction? All this shows is that hearers can infer what a speaker is conveying without first identifying what the speaker is saying. The semantic notion of what is said pertains to the character of the information available to the hearer in the process of identifying what the speaker is conveying, not to what goes on in this process (Bach and Harnish, 1979, pp. 91-93).

        Moreover, suppose that it is true that what is said, in the minimalist sense, is sometimes not consciously accessed. It is still consciously accessible. This is evident from the fact that people recognize, as we saw with examples (1) - (4) above, that implicit qualifications on these utterances are cancelable. Furthermore, even if in some cases the minimal proposition is not actually computed and plays no role in the interpretation process as it actually occurs, because of “local processing” on constituents of the sentence, it can still play some role. Even if a hearer doesn’t explicitly represent what is said by the utterance of a sentence, hence does not explicitly reject it, still he makes the implicit assumption that it is not what is meant. Implicit assumptions are an essential ingredient in default reasoning in general (Bach 1984) and in the process of understanding utterances in particular. Communicative reasoning, like default reasoning in general, is a case of jumping to conclusions without explicitly taking into account all alternatives or all relevant considerations. Even so, to be warranted such reasoning must be sensitive to such considerations. This means that such considerations can play a dispositional role even when they do not play an explicit role. They lurk in the background, so to speak, waiting to be taken into account when there is special reason to do so.

        I conclude that intuitive and related cognitive considerations do not undermine a minimalist conception of what is said. As Jerry Fodor says, “No doubt, intuitions deserve respect, … [but] informants, oneself included, can be quite awful at saying what it is that drives their intuitions. … It is always up for grabs what an intuition is an intuition of” (1998, p. 86). In the case of seemingly semantic intuitions, they are largely irrelevant to determining what is said. They are influenced by semantically irrelevant information, they tend to be insensitive to relevant distinctions, and they are likely to be biased in favor of understandings corresponding to things that people are relatively likely to communicate. Or so it seems to me, at least intuitively.





Bach, Kent: 1984, “Default Reasoning: Jumping to Conclusions and Knowing When To Think Twice,” Pacific Philosophical Quarterly, 65, pp. 37-58.

Bach, Kent: 1994a, “Conversational Impliciture,” Mind & Language 9, pp. 124-62.

Bach, Kent: 1994b, Thought and Reference, paperback edition, revised with postscript, Oxford University Press, Oxford.

Bach, Kent: 1998, “Standardization Revisited,” on line at http://userwww.sfsu.edu /~kbach/ standard.html; in A. Kasher (ed.), Pragmatics: Critical Concepts, vol. IV, Routledge, London, pp. 712-22.

Bach, Kent: 1999, “The Semantics-Pragmatics Distinction: What It Is and Why It Matters,” on line at http://userwww.sfsu.edu/~kbach/semprag.html; in Ken Turner (ed.), The Semantics-Pragmatics Interface from Different Points of View, Elsevier, Oxford, pp. 65-84.

Bach, Kent and Robert Harnish: 1979, Linguistic Communication and Speech Acts, MIT Press, Cambridge, MA.

Carston, Robyn: 1988, “Implicature, Explicature, and Truth-theoretic Semantics,” in R. M. Kempson (ed.), Mental Representations: The interface Between Language and Reality, Cambridge University Press, Cambridge, Eng., pp. 155-81.

Gibbs, Raymond and Jessica Moise: 1997, “Pragmatics in Understanding What is Said,” Cognition 62, pp. 51-74.

Grice, Paul: 1989, Studies in the Way of Words, Harvard University Press, Cambridge, MA.

Lackner, J.R. and Merrill Garrett: 1973, “Resolving Ambiguity: Effects of Biasing Context in the Unattending Ear,” Cognition 1, pp. 359-372.

Levinson, Stephen: 2000, Presumptive Meanings: The Theory of Generalized Conversational Implicatures, MIT Press, Cambridge, MA.

Nicolle, Steve and Billy Clark: 1999: “Experimental Pragmatics and What Is Said: A Response to Gibbs and Moise,” Cognition 69, pp. 337-354.

Recanati, François: 1989, “The Pragmatics of What is Said,” Mind and Language 4, pp. 294-328.