Midwest Studies in Philosophy, vol. 25: Figurative Language, Peter French and Howard Wettstein (eds). Blackwell Publishers (2001), pp. 249-263


Speaking Loosely: Sentence Nonliterality



This paper concerns a linguistic phenomenon so pervasive that it hardly ever gets mentioned, much less labeled, at least by theorists. Perhaps that's the most interesting thing about it. It's a perfectly familiar phenomenon, describable as "speaking loosely." But it's a particular way of speaking loosely, not to be confused with other kinds, like exaggeration and understatement. It's a special kind of nonliteral use.

        Words don't have nonliteral meanings (it is redundant to describe the meanings they do have as their literal meanings), but they can be used in nonliteral ways. You speak nonliterally when you say one thing and mean something else instead.[1] In familiar cases, such as metaphor and metonymy, particular expressions are used nonliterally. Such uses are commonly described as "figurative." But there's a different phenomenon, which I call "sentence nonliterality," as opposed to constituent nonliterality. Here a whole sentence is used nonliterally, without any of its constituent expressions being so used.

        Figurative speech is easily recognized as such. If I call someone an ape, a dog, or a pig, presumably I'm not describing him literally as an animal of any of those sorts. Similarly, if I comment on the behavior of a know-it-all by saying, "The brain is pontificating again," presumably I am using "the brain" to refer to a certain person. Brains, those chunks of gray matter in people's heads, don't pontificate; people do. I could add to the metaphorical mix by saying that, instead of "pontificating," the brain is mouthing off (or spouting off) again. When I say such a thing, it is obvious to my audience that I don't mean it literally, and they can easily figure out what I do mean. It is equally obvious to us theorists. Matters are different with sentence nonliterality. That is one reason why it's not a widely recognized phenomenon, at least not by theorists. But ordinary communicators have no problem with it. You can verify its pervasiveness by looking at any newspaper or magazine article or eavesdropping on any conversation, but in the discussion to follow I will have to use isolated sentences to illustrate this phenomenon.


Sentence Nonliterality

We hardly ever mean exactly what we say. It's not that we generally speak figuratively or that we're generally insincere (these are different ways of not meaning what one says). Rather, we commonly speak loosely, by omitting words that could have made what we meant more explicit, and we let our audience fill in the gaps. Language works far more efficiently when we do that. Literalism can have its virtues, like when we're drawing up a contract or programming a computer, but we generally opt for efficiency over explicitness. In most conversation, though, spelling things out is not only unnecessary, it just slows things down.[2] It is often misleading too, insofar as it guards against something that doesn't need to be guarded against.

        Let's start with some examples. Suppose I graciously tell my guest, "There's beer in the fridge." He opens the fridge and, contrary to what we both expected, finds not a single can or bottle but merely a small puddle of beer at the bottom of the fridge. I may have inadvertently misled him, but did I say something false? It seems not: there was some beer in the fridge.

        A baseball fan is glued to the tube during the course of a historic World Series between the Red Sox and the Cubs. The teams are tied at three games apiece. His wife is losing patience with him and, not knowing that the Series is best of seven, wants to know how much more baseball she'll have to put up with. He assures her, "They'll play only one more game." Since they will meet again in interleague play the following season, what he said was not really true, but that's not what he meant.

        An assistant professor at Harvard, worried that she won't get tenure, seeks the advice of a senior colleague. He assures her, without the slightest intention to mislead, "You'll be out of here if you don't publish at least five more articles in the next two years." She publishes seven more articles over the next two years, all in top journals, but is denied tenure anyway. Even so, what her senior colleague said to her was true. After all, publish or perish does not preclude publish and perish.

        It's not hard to see what's going in these examples. The speaker says something but means some qualified version of that. In each case, the speaker failed to make explicit part of what he meant. I could have said to my friend, "There's beer to drink in the fridge." The baseball fan could have said to his wife, "The Red Sox and the Cubs will play only one more game in the current World Series." And the professor could have said, "You won't get promoted if you don't publish at least five more articles in the next two years, but you will if you do." In each of the above examples, the speaker could have made fully explicit what he was trying to convey by including the italicized material (or its equivalent--the exact words don't matter) in his utterance. Usually we don't have to, of course, since ordinarily we can rely on our audience's ability to figure out the intended qualification.

        Using sentences nonliterally in this way is so common that we tend neither to be aware of doing it nor to think of it as not literal when others do it. But we do it all the time (as I did just then--I left out when we speak). Rather than insert extra words into our utterances in order to make fully explicit what we mean, we allow our listeners to read things into what we say. Even though we may not intuitively think of this phenomenon as nonliterality, because no specific words are being used figuratively, it is a way of not being literal. For the speaker says one thing but intends to convey something distinct from that.

        Using sentences as if we included words or phrases that are not there clearly is a kind of nonliterality, for what the speaker is trying to convey in these examples is not the proposition, as compositionally determined, that is expressed by the sentence. Consider the following example.

                (1)        Rick and Ann are engaged.

Someone who utters (1) is not saying that they are engaged to each other, any more than he would be saying this with "Rick and his sister Ann are engaged." Rather, this is implicit in what he is saying or, more precisely, in his saying of it.[3] That it is not part of what is said is clear from the fact that it passes Grice's test of cancellability: it may be taken back without contradiction. For example, there is no contradiction in saying, "Rick and Ann are engaged, but not to each other."

        If leaving words out is a kind of nonliterality and, more specifically, a kind of loose talk, how does it differ from other kinds? Sometimes we describe using vague expressions as speaking loosely, though this might better be described as speaking roughly. If you say, "It's a long way to Tucumcari," you're not being precise but still mean what you're saying (you're in Albuquerque and don't know just how far Tucumcari is from there). Then there are exaggeration and understatement. In both cases, you're distorting the truth for effect, either by stretching it or compressing it. You describe Parsifal as an "interminable opera" or Tiger Woods as a "good golfer." You're using 'interminable' to mean very long and 'good' to mean great. But the kind of loose talk we're concerned with is not a matter of using vague terms or of using particular words or phrases nonliterally, but simply a matter of leaving words out. So, for example, if you say "I haven't eaten" you don't include dinner today or anything of the sort. If you say "Everyone must wear a costume" you don't include who comes to my upcoming party. And if you say "Tigers have stripes" you don't include normal (or non-albino).

        These examples illustrate what I call conversational "impliciture," as opposed to Grice's "implicature." Implic-a-ture is an indirect constative speech act, whereby one says and means one thing and thereby asserts something else in addition.[4] In implic-i-ture, one says something but does not mean that; rather, what one means includes an implicit qualification on what one says, something that one could have made explicit but didn't.[5]

        It might seem that the qualification implicit in an utterance involves a nonliteral use of some particular constituent. Consider, for example, the case of using an obviously incomplete definite description. If a drama buff says to her husband, "The play will start in an hour," to refer to some particular play, say the one that they plan to attend that evening, it might seem that she is using 'the play' nonliterally, say for 'the play we plan to attend tonight.' If that is true, this is despite the fact that she is using its constituent words literally--she means the by 'the' and play by 'play.' It is true that if she had said, "The play we plan to attend tonight will start in an hour," she would have, by attaching a relative clause to 'the play,' made fully explicit what she meant. But this doesn't show that she meant 'the play we plan to attend tonight' by the phrase 'the play.' At most it shows that she was using 'the play' to refer to the play she and her husband planned to attend that night. The nonliterality is not attributable to any particular constituent of the sentence being used.

        This phenomenon I am calling sentence nonliterality seems pretty straightforward, but there are objections to how I characterize it. We will take up four such objections, but first we should take a closer look at what is going on in examples like the ones above. We can break this question down into two parts: what is the relation between what the speaker says and what the speaker means, and how does the audience figure it out?


The Relation between the Explicit and the Implicit

The implicitures involved in sentence nonliterality goes beyond what is said, but unlike implicatures, which are additional propositions separate from what is said, implicitures involve an unexpressed qualification on what is said. They are not a case of conveying one proposition and indirectly conveying another. Grice himself occasionally alluded to what I am calling implic-i-ture, as when he remarked that it is often "unnecessary to put in ... qualificatory words" (1989, p. 44). Although he did describe such cases as implic-a-tures, he appeared to have something distinctive in mind: "strengthening one's meaning by achieving a superimposed implicature" (p. 48; my emphasis). But what did he mean by "strengthening"? In our examples, what is the relation between what is said and what is conveyed?[6] The relation might seem to be logical, such that what is said is entailed by what is conveyed, as in the following examples:

                (2)     Jack and Jill went up the hill.

                (3)     Jack went to the cliff and jumped.

                (4)     Jill got married and became pregnant.

In each case, let's assume, the speaker means something more specific than the proposition expressed by the sentence being uttered, to wit:

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

                (3+)  Jack went to the cliff and jumped off the cliff.

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

The italicized material, not actually uttered, indicates the implicit qualification on what the speaker said. Now in what way does this implicit qualification strengthen what is said? Since in these examples its result, the impliciture, entails what is said, the strengthening might seem to be logical in character. But that can't be right, because precisely the same phenomenon occurs when the sentences in question are negated. For example, if the speaker uttered (~2) in order to convey (~2+),

                (~2)      Jack and Jill did not go up the hill.

                (~2+)    Jack and Jill did not go up the hill together.

the relation between what was said and what was conveyed would be the same, even though in this case the entailment is in the other direction.

        The relation between (2) and (2+), as well as that between their negations, involves what may be called conceptual strengthening, in that what is implicit is additional conceptual material, whether or not it logically strengthens the proposition in question that is expressed by the sentence actually uttered. That it is not logical strengthening is clear from the fact that an implicit qualification can be logically redundant. Take the following example:

                (5)     Jay has three cars.

Since (5) is true just in case Jay has at least three cars, adding 'at least' would not make any logical difference. But it could make a pragmatic difference. For normally an utterance of (5) conveys that Jay has exactly three cars, even though the sentence does not literally mean that (it is true even if Jay has more than three cars). In some circumstances, though, you could use (5) to mean merely that Jay has at least three cars, say if you were explaining why Jay is subject to a special parking fee (for owners of three or more cars). So it would make a pragmatic difference to include the words 'at least' in your utterance. Including them leaves open the possibility that Jay has more than three cars. Though logically redundant, it is not pragmatically redundant.

        The pragmatic phenomenon just noted illustrates an important point about the difference between implicit and explicit qualifications. Although expressly making the qualification, in this case with at least, would enable the speaker to make fully explicit what he was trying to convey, expressly making it could result in conveying something else as well. In the above case, expressly including the words at least forestalls the normal way of taking the utterance without them. You then convey that Jay could just as well have more than three cars as exactly three. And, in circumstances where an utterance of (5) would normally, though not literally, convey that Jay has exactly three cars, say in answer to the question "How many cars does John have?", including exactly expressly would foreclose the possibility that he has more than three.

        A more far-reaching point is that some sentences, given what they mean, cannot readily be used literally. This should not seem problematic, for there are many propositions that people are unlikely ever to intend to convey. That shouldn't suggest that there aren't sentences that literally express those propositions. To see how some sentences are hard to use in a strictly literal way, consider and compare the following pair:

                (6)        Dennis got herpes and had sex.

                (6')        Dennis had sex and got herpes.

These two sentences are logically equivalent, but it is hard to use them in the same way. An utterance of (6) is likely to be taken to convey that during a certain period Dennis got herpes and that during that period had sex. Reverse the order and matters are different: an utterance of (6') is likely to be taken to convey that Dennis had sex (on a certain occasion) and then, as a result of that, got herpes. It is hard for an utterance of either sentence to convey the bare conjunction that they both literally express. Even if you uttered (6'), say, and followed it with, "not that his getting herpes had anything had to do with his having sex," thereby cancelling the impliciture that he got herpes as the result of having sex, you would be taken to be precluding the possibility that he did. The bare conjunction does not preclude this.


Recognizing Implicitures

In examples such as we have considered, how is what is meant recognized, and reasonably expected to be recognized? Implicitures are like implicatures, insofar as they too exploit what Grice called "maxims" of conversation. He proposed a general Cooperative Principle and specific maxims--of "quality, quantity, relation [relevance], and manner" (1989, pp. 26ff)--to account for the rationale and success of conversational implicatures. The speaker intends his audience to figure out what he is conveying, and the audience figures that out, by way of recognizing what would be violation of a maxim if the utterance were taken at face value. On the supposition that the speaker is being cooperative, in particular that he is trying to convey something, which requires that his communicative intention be evident from what he is saying under the circumstances, the audience figures out what he is trying to convey.

        Grice seems to be basically right about this, but it is arguable that being cooperative (at least in making constative utterances) consists simply in being truthful and relevant, hence that there is no need for four separate maxims. For example, observing the maxim of quantity is just being relevantly informative. Also, the maxims might better be thought of as presumptions, for it is on the presumption that the speaker is being truthful and relevant (and intends to be taken to be) that the hearer figures out what the speaker means in saying what he says.[7]

        At any rate, it might seem that implicitures are like Gricean quality implicatures, whereby the audience recognizes what the speaker is trying to convey by drawing an inference from the obvious falsity of what the speaker is saying. [8] This is initially plausible with these examples,

                (7)     I haven't taken a bath.

                (8)     I have nothing to do.

The sentences are likely to be literally false (evaluated relative to the identity of the speaker and the time of their utterance), and are likely to be used in a more restrictive way. You'd probably use (7) to mean you hadn't taken a bath that day, and (8) to convey to your boss that you had no assigned work to do. However, it's not the obvious falsity of the sentences that matters, for in the same circumstances, utterances of the positive sentences,

                (7')     I have taken a bath.

                (8')     I have something to do.

would likely be made and be taken with the same implicit qualifications. But these sentences are very likely to be literally true. So the implicitures in question should not be understood on the model of quality implicatures.

        Besides, the sentences in most of our earlier examples were obviously true, not obviously false. In each case, what triggers the hearer's search for something other than what is said is not its obvious falsity but its lack of relevant specificity. As with any nonliteral utterance, with sentence nonliterality the audience is to recognize that the speaker couldn't plausibly be taken to mean exactly what he said. Assuming that the speaker is trying to communicate something and is therefore trying to make what he means evident, the audience has to find some salient connection between what is said and what is meant. Generally speaking, that involves finding some way of taking the utterance that is pertinent to the current purposes of the conversation.[9] Of course, this is not to say precisely how, in any given case, the hearer figures out just what that is. That is a difficult and unanswered question for the psychology of communication.


Possible Objections

Let's call the view that utterances like the ones considered above are not literal the sentence nonliterality account (SLA). SLA entails that what is said in such an utterance does not include what I am calling the implicit qualification. This is because the implicit qualification does not correspond to any constituent of the sentence. Here I'm assuming that what is said must satisfy what I'll call the "Syntactic Correlation Constraint," as expressed by Grice's stipulation that what is said must correspond to "the elements of [the sentence], their order, and their syntactic character" (1989, p. 87). What is said is determined compositionally by the semantic contents of the constituents ("elements") of the sentence as a function of their syntactic relationship.[10] Where there is lexical ambiguity, this compositional determination is relative to a given set of readings of the ambiguous terms, and because of the presence of tense and sometimes indexicals, what is said in the utterance of a given sentence is not independent of the context in which it is used. In short, what is said is the semantic content of the sentence relative to that context.[11] But it is independent of the speaker's communicative intention. In effect, the Syntactic Correlation Constraint says that every constituent of a proposition expressed by a sentence corresponds to a constituent of the sentence.

        The Syntactic Correlation Constraint entails that if any element of what the speaker intends to convey does not correspond to any element of the sentence he is uttering, it is not part of what he is saying. Of course it may correspond to what he is asserting, but I am not using 'say' to mean assert. In the jargon of speech act theory, saying is locutionary, not illocutionary. All of our examples and, it seems, much of everyday speech, involves conveying more than what one says, by using sentences nonliterally.

        SLA assumes that the implicit qualification on an utterance does not correspond to any constituent of the sentence being uttered. Also, it seems to presuppose that every such utterance has a determinate qualification, which the hearer must identify if he is to figure out what the speaker is trying to convey. Moreover, SLA may seem to require that the hearer can figure out what the speaker means only by consciously reasoning that the speaker could not mean merely what he is saying. Finally, it may seem that the Syntactic Correlation Constraint, which the SLA presupposes, does violence to our intuitions about what is said. These points suggest certain possible objections:

        1. What is allegedly implicit is really the value of a covert variable.

        2. What is implicit is not always determinate.

        3. Hearers don't make inferences of the sort SLA requires.

        4. SLA is counterintuitive--the utterances under discussion seem perfectly literal.


Objection 1. What is allegedly implicit is really the value of a covert variable.

SLA presupposes that the implicit qualification on an utterance does not correspond to any constituent of the sentence being uttered. This includes phonologically null constituents, such as variables and other empty categories (in the sense of GB theory). So it might be objected that in each of our examples, the place where I have suggested the speaker could have included an additional word or phrase is really a place where there is a hidden variable, whose value is contextually provided. The idea, in effect, is that what I am characterizing as a kind of nonliteral use is better understood on the model of using indexicals.

        The trouble with this objection is that there is no reason to believe in the presence of such covert variables. Our examples are not like sentences such as "George is ready," "Martha noticed," or "I've had enough," which are syntactically well-formed but, because they are missing an (overt) argument, appear not to express complete propositions. In those cases it plausible to suppose (but see notes 5 and 11) that the sentence's argument structure requires the filling of a slot, even if syntax does not require an overt expression to occupy that slot. But sentences like "I haven't taken a bath" and "There's nothing in the fridge" do express complete propositions, even if these are not the ones speakers are likely to use them to convey. The mere fact that a sentence is not readily used to convey the bare proposition does not show that it does not really express that proposition, but only that, given the Syntactic Correlation Constraint, the sentence is likely to be used nonliterally. If the covert variable hypothesis is to be a viable alternative to SLA, positive syntactic evidence must be adduced for the existence of such variables.

        In some cases positing a covert variable seems relatively plausible. With an utterance of (9), for example,

                (9)        Most people voted for Gore.

where the speaker is likely to be conveying that most people who voted in the last presidential U.S. election voted for Gore, we might suppose that there is a hidden variable of domain restriction on the quantified noun phrase 'most people'. Similarly, when Yogi Berra said, "Nobody goes there [to a certain restaurant] any more--it's too crowded," he meant that nobody important goes there any more. But why posit a hidden variable of domain restriction? Besides, if one were needed for (9), one would be needed for (10) too,

                (10)     All men are mortal.

even though there is no restriction on the class of men being said to be mortal. So it seems gratuitous to suppose that there are hidden domain markers that restrict the "universe of discourse" or "domain of quantification." Why transpose these technical notions from logic to natural language rather than suppose simply that the speaker said one thing but intended an implicit qualification on what he was saying?[12]

        The same can be said for the view that there is a covert time frame variable in sentences like (7) or (11),

                  (7)        I haven't taken a bath.

                (11)        Johnny will not do his homework.

Presumably a speaker of (7) would mean that he hasn't taken a bath that day, and a speaker of (11) would mean that Johnny will not do his homework for that night. But there is no reason to suppose that the day or night in question is the value of a covert variable rather than the content of an implicit qualification such as today. The mere fact that on SLA (7) and (11) come out as literally false (assuming that the speaker of (7) has taken a bath some time in the past and that at some time in the future Johnny will do his homework) does not suggest that they contain covert variables. To suppose that they do requires denying that, as they stand, they express propositions at all. This would be to treat them, contrary to appearances, as merely open sentences.


Objection 2. What is implicit is not determinate.

It might be objected that by using specific words or phrases to represent the implicit

qualifications in our examples, SLA implausibly requires that the speaker must have meant something determinate, and that in order to understand him his audience had to figure out just what this was.

        If the objection is that the audience must figure out the exact words the speaker had in mind but didn't use, surely that would be too strong a requirement. But SLA doesn't require that. Indeed, I was careful to describe the implicit qualification involved in sentence nonliterality as what the speaker could have made explicit if he had included certain words or their equivalents. The exact words don't matter. But what counts as equivalent here?

        So perhaps the objection is that it is not determinate what the implicit qualification is, not how it is expressed. I'll grant the point but deny it's an objection. Consider (7) again. In saying "I haven't taken a bath," the speaker might mean that he hasn't taken a bath that day or that he hasn't taken one lately or that he hasn't taken one since he last worked out or whatever, but for normal purposes of communication it doesn't matter precisely what he means, or whether the hearer identifies precisely that. Indeed, he might not mean any one thing precisely, and be quite prepared to concede that he didn't mean some one of those things as opposed to any of the others.


Objection 3: Hearers don't make inferences of the sort SLA requires.

The objection here is that when people hear utterances of sentences like the ones we've discussed, they don't first compute the bare, unqualified proposition expressed by the sentence and then, after deciding that the speaker couldn't plausibly be taken to mean that, infer that he means some implicitly qualified version of that.

        Let's grant the psychological claim underlying the objection. The question is whether SLA requires what the objection says it requires. In fact, SLA does not require that the hearers must compute the bare propositions. It concerns the semantics of sentences, not the psychology of understanding utterances of them. Moreover, all the objection suggests is that hearers can infer what a speaker is communicating without first identifying what the speaker is saying. But SLA is not committed to an account of the temporal order or other details of the process of understanding. For all it says, hearers can identify implicit qualifications on the fly, without having first to figure out the semantic content of the entire sentence.[13]

        Besides, even if a hearer doesn't explicitly represent the proposition literally expressed by a sentence, hence does not explicitly reject it as the one that is meant, 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 consciously 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 worth considering.


Objection 4: SLA is counterintuitive.

The objection here is that the utterances under discussion seem perfectly literal. Accordingly, SLA disregards ordinary intuitions about the truth or falsity, hence about the content, of what is said. For example, it seems that sentence (7), "I haven't taken a bath," as uttered on a certain day, is true if the speaker hasn't taken a bath that day, although on SLA it is false if he has ever taken a bath. This is counterintuitive.[14]

        In this and the other cases, I claim, what the speaker means is distinct from he says, since what he means includes an implicit qualification on what he says. Of course what the speaker means is true, but that includes the implicit qualification provided by today, which is not present in the sentence.[16] 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 the Syntactic Correlation Constraint on what is said.

        It is easy to sensitize people's intuitions to this constraint. One way is to get them to appreciate Grice's cancellability test for what is not said. Just present them with sentences like (2) - (4) followed by cancellations of the implicit qualification, as in (2xx) - (4xx),

                (2xx)        Jack and Jill went up the hill but not together.

                (3xx)        Jack went to the cliff and jumped but not off the cliff.

                (4xx)        Jill got married and became pregnant but not in that order.

and see if they sense a contradiction or just a clarification. Or ask them to compare what is said with explicitly qualified versions of (2), (3), or (4) to what is said with sentences (2), (3), and (4) themselves:

                (2eq)        Jack and Jill went up the hill together.

                 (2)          Jack and Jill went up the hill.

                (3eq)        Jack went to the cliff and jumped off the cliff.

                   (3)        Jack and Jill are engaged.

                (4eq)        Jill got married and then became pregnant.

                (4)        Jill got married and became pregnant.

They are likely to say that the explicitly qualified utterances say something not said with the original ones.

        These considerations aside, why should we be confident in the accuracy of people's untutored intuitions about what is or is not literally expressed by a sentence? To be sure, we have a perfectly good handle on what our words mean, on how they are put together into sentences, hence on what our sentences mean. Who could deny that competent speakers have such knowledge? But this is not theoretical knowledge. Even so, Grice regarded it "as a sort of paradox [that] 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?" (1989, p. 49). Grice's paradox may seem especially troubling if it is supposed, as it often is, that accounting for our ordinary judgments, our "intuitions," about the truth-conditions of sentences is the central aim of semantics.

        This worry 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 they are often responsive to non-semantic information, to what is implicit in what is said, they should not be given too much weight.

        Besides, these intuitions don't seem to play a role in ordinary communication. In the course of speaking and listening to one another, we generally don't consciously reflect on the semantic content of the sentences we hear or on what is said in their utterance. We are focused on what we are communicating and on what is being communicated to us, not on what is being 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 take intuitions overly seriously 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 cuts and camera angles, there is no reason to suppose that their intuitions are reliable about these effects or about how they are produced. 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 conclude, what worried Grice was not a real paradox.

        People's untutored intuitions tend to neglect the Syntactic Correlation Constraint. A related reason they are unreliable about the semantic contents of sentences like those we've considered, in contrast to cases of metaphor or metonymy, is that there are no constituents which intuitively are being used nonliterally. Moreover, there is a recurrent pattern of nonliterality associated with sentences like these. As with what Grice called "generalized" conversational implicature, where there must be specific contextual reasons for supposing that an implicature is not present, in our examples 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. [xvi]


Summing Up

I have tried to make a case for the existence of the phenomenon of sentence nonliterality. I have sketched how it works but really did nothing more than invoke Grice's well-known theory of conversation. Neither it nor I can explain in detail how people manage to convey more than what their sentences mean and how their listeners manage to understand them. That, I conceded, is a matter for the psychology of communication. Short of that, the most interesting thing about the phenomenon of sentence nonliterality is, as I suggested at the outset, that it is so pervasive that it gets no theoretical attention. Here are a few more examples (with likely implicit qualifications in brackets) to illustrate how prosaic it is:

                (12)        Helen poured some wine [intentionally].

                (13)        Helen spilled some wine [unintentionally].

                (14)        Roy has always been an honest judge [since he's been a judge].

                (15)        The stroke victim is not going to die [from that stroke].

                (16)        Dr. Atkins is not [what I would describe as] a physician but a quack.

As I also suggested, you can verify the pervasiveness of sentence nonliterality by looking at any newspaper or magazine article or eavesdropping on any conversation. You may find that to be much more interesting than theorizing about it.





Bach, Kent (1984), "Default Reasoning: Jumping to Conclusions and Knowing When to Think Twice," Pacific Philosophical Quarterly 65: 37-58.

_____ (1994a), "Conversational Impliciture," Mind & Language 9: 124-62.

_____ (1994b), Thought and Reference, paperback ed., revised with postscript. Oxford: Oxford University Press.

_____ (1995), "Standardization vs. Conventionalization," Linguistics and Philosophy 18: 677-86.

_____ (1999), "The Semantics-Pragmatics Distinction: What It Is and Why It Matters." In Ken Turner (ed.), The Semantics-Pragmatics Interface from Different Points of View. Oxford: Elsevier, 65-84.

_____ (2000), "Quantification, Qualification, and Context," Mind & Language 15: 262-283.

_____ (2001), "You Don't Say?," Synthese 127: 11-31.

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

Gibbs, Raymond and Jessica Moise (1997), "Pragmatics in Understanding What is Said," Cognition 62: 51-74.

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

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

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

Recanati, Fran┴ois (1989), "The Pragmatics of What is Said," Mind & Language 4: 295-329.

Stanley, Jason (2000), "Context and Logical Form," Linguistics and Philosophy 23: 391-434.

Stanley, Jason and Zoltan Szabo (2000), "On Quantifier Domain Restriction," Mind & Language 15: 219-261.




[1] You speak indirectly when you mean one thing and mean something else as well. This can occur when you say one thing and mean not only that but something else as well. It is possible to speak both nonliterally and indirectly in the same breath (Bach and Harnish 1979, pp. 71ff). You might use "I'm sure the cat likes having its tail pulled" to assert nonliterally that it doesn't and indirectly to ask your listener to stop pulling its tail.

[2] This is partly due to what Stephen Levinson calls "the articulatory bottleneck in linguistic communication," which he regards as a "design flaw ... in an otherwise optimal system. The bottleneck is constituted by the remarkably slow transmission rate of human speech (conceived of as the rate at which phonetic representations can be encoded as discriminable speech signals), with a limit in the range of 7 syllables or 18 segments per second" (2000, pp. 28). In contrast, Levinson adds, "the psycholinguistic evidence suggests that all other aspects of speech production and comprehension can run at a much higher rate." Levinson interprets the Gricean maxims as heuristics for getting through this bottleneck.

[3] Here I am alluding to Grice's point that what a speaker implicates in saying what he says is carried not by what he says but by his saying it (1989, p. 39). His point suggests a way of drawing the semantic-pragmatic distinction (see Bach 1999 for details). Semantic information is information encoded in what is uttered--these are stable linguistic features of the sentence--together with any extralinguistic information that provides (semantic) values to context-sensitive expressions in what is uttered. Pragmatic information is (extralinguistic) information that arises from an actual act of utterance. Whereas semantic information is encoded in what is uttered, pragmatic information is generated by, or at least made relevant by, the act of uttering it. The act of producing the utterance exploits the information encoded but by its very performance creates new or otherwise invokes extralinguistic information. That information, combined with the information encoded, provides the basis for the hearer's identification of the speaker's communicative intention. This pragmatic information is relevant to the hearer's inference only on the supposition that the speaker is producing the utterance with the intention that the information in question be taken into account.

[4] Grice also regarded figurative speech, such as metaphor and metonymy, as implicature, even though in these cases the speaker says one thing and means something else instead. Actually, Grice held that in speaking nonliterally one "makes as if to say" something (1989, p. 30) but does not really say it. Here he seems to have conflated saying with stating or asserting. In my opinion, he should not have classified figurative speech as implicature (Bach 1994a, pp. 143-4).

[5] In all the cases we will consider, the sentence that is uttered expresses a complete proposition (my claim is that the use of such a sentence to convey a distinct, implicitly qualified version of that proposition is not literal). A more controversial species of impliciture involves sentences which, though syntactically complete, are semantically incomplete--they do not express a complete proposition. This sort of impliciture includes a "completion" of that proposition. The existence of this phenomenon is claimed in Bach 1994a (sec. 2), but Stanley (2000) contends that in all of these cases there are covert variables. In effect, he assimilates this phenomenon to indexicality.

[6] What is conveyed is what the speaker means, i.e., attempts to communicate, in saying what he says. I use 'what is conveyed' rather than 'what is meant' because the latter can include what the sentence means as well as what the speaker means in uttering it.

[7] See Bach and Harnish 1979, pp. 62-5. We propose replacing Grice's Cooperative Principle with a "Communicative Presumption," to the effect that whenever someone says something to someone, he intends to be communicating something (p. 12). Grice's Principle is too strong, since communicators need be cooperative only to the extent that they make, and may be presumed to be making, their communicative intentions evident. Their intentions in other respects may be obscure, deceptive, or downright malicious.

[8] The obvious falsity of a metaphor is sometimes thought to be what triggers the hearer's inference to what the speaker means. But that can't be right, as is clear from "No man is an island," which is obviously true.

[9] For further discussion see Bach and Harnish 1979, pp. 65-70.

[10] I defend this strict conception of what is said in Bach 2001.

[11] 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 various sorts of sentence that do not express a complete proposition, not even relative to a context (see Bach 1994a, sec. 2).

[12] I have debated this special case previously (Stanley and Szabo 2000 and Bach 2000b).

[13] The sentence has its semantic content independently of whether the hearer has to ascertain that content before figuring out what the speaker means. Its semantic content, the proposition it literally expresses, is the information available to the hearer simply in virtue of hearing the sentence and quite apart from any question of what the speaker intends to be communicating. How the hearer exploits that information is another matter. For further discussion of this and related issues see Bach and Harnish 1979, especially pp. 91-93.

[14] Gibbs and Moise (1997) devised some experiments that purport to show that people's intuitions about what is said tend to include implicit qualifications. I could discuss Gibbs and Moise's experiments in detail, but suffice it to say that their research tested for the wrong thing. It didn't show that people's intuitions really are about what is said, even if people say they are. Gibbs and Moise thought they could get data about what is said by asking people what is said by a given utterance, by forcing them to judge that what was conveyed by a given utterance, even when it included an implicit qualification, was either implicated or merely said. Gibbs and Moise evidently assumed that what people say about what is said is strongly indicative of what is said. They really tested for how people apply the word "say" and the phrase "what is said." Little was established about what is said, much less about the cognitive processes whereby people understand utterances. Moreover, Nicolle and Clark (1999) have experimentally challenged Gibbs and Moise's findings, and report that their own experiments often delivered different results, sometimes with people deeming clear cases of implicature to count as what is said.

[15] Recanati (1989) even formulates a principle, the Availability Principle, which prescribes that intuitions about what is said be "preserved" in our theorizing. But he needs to explain why the intuitions he takes to concern what is said really do concern that, as well as why we should be confident in their accuracy.

[16] Our intuitions about literal content are, as suggested, especially unreliable when there is a recurrent pattern of nonliterality associated with particular locutions or forms of sentence. These intuitions are insensitive to the difference between conventionalization and cases of standardization (see Bach 1995), such as what I call standardized nonliterality (Bach 1994b, pp. 77-85) and the more commonly recognized phenomenon of standardized indirection (Bach and Harnish 1979, pp. 192-219), including what Grice called generalized conversational implicature (1989, pp. 37-39). Where there is standardization, the hearer's inference to what the speaker means is short-circuited, compressed by precedent (though capable of being worked out if necessary), so that the literal content of the utterance can be bypassed. Hence the imporession of conventionalization. For a monumental study of generalized conversational implicature, with a huge range of data, many of which illustrate what I regard as implicitures, see Levinson 2000.