The Myth of Conventional Implicature*

Kent Bach


Grice’s distinction between what is said and what is implicated has greatly clarified our understanding of the boundary between semantics and pragmatics. Although border disputes still arise and there are certain difficulties with the distinction itself (see the end of §1), it is generally understood that what is said falls on the semantic side and what is implicated on the pragmatic side. But this applies only to what is conversationally implicated. Grice’s category of conventional implicature throws a monkey wrench into his distinction, inasmuch as conventional implicatures derive from the meanings of particular expressions rather than from conversational circumstances. This monkey wrench needs to be removed. I will argue that there is no such thing as conventional implicature and that the phenomena that have been described as such are really instances of something else.

        In linguistics and philosophy it is common to suppose that certain words, such as ‘but’, ‘still’, and ‘even’, do something besides contribute to what is said in utterances of sentences containing them. So, for example, the difference between (1) and (2) supposedly consists not in what they say but merely in what is indicated by (the presence of) the word ‘but’:

            (1)           Shaq is huge but he is agile.

            (2)           Shaq is huge and he is agile.

According to common wisdom, the truth of (1) requires nothing more than the truth of (2), although in uttering (1) rather than (2) one is indicating that there is some sort of contrast between being huge and being agile. But one is not saying that. Nor is it even entailed by what is one saying. On the other hand, this proposition is not a conversational implicature, because its being indicated depends essentially on the conventional meaning of the word ‘but’. The common view is that it is a conventional implicature (§1).

        My aim is to debunk this view and its intuitive basis. There are two sorts of locution that have been thought to generate conventional implicatures. I will argue that expressions of the first kind, typified by ‘but’, ‘still’, and ‘even’, in fact contribute to what is said. The best evidence that they do is that they can occur straightforwardly in indirect quotation (§2). They seem not to contribute to what is said, I will suggest (§3), because intuitions about the truth or falsity of utterances containing them are insensitive to their contribution, which, though truth-conditional, is secondary to the main point of the utterance. Indeed, contrary to the common assumption of one sentence, one proposition, such utterances express more than one proposition (§4).

        There are locutions of another kind which, although they do not contribute to what is said, do not generate conventional implicatures either. They do something else. They are vehicles for the performance of second-order speech acts.[ ] A locution like ‘confidentially’, ‘in other words’, or ‘to get back to the point’ can be used to comment on some aspect of the speech act being performed in the utterance of the matrix sentence.[1] I call these locutions utterance modifiers, as opposed to sentence modifiers, because they do not modify the content of the sentence but instead characterize the act of uttering it.[2] In other words, although they are syntactically coordinate with the rest of the sentence, they are not semantically coordinate with it. Utterance modifiers will be taxonomized in §5 and alternative accounts of them will be discussed in §6. For purposes of illustration (as well as exposition) they will be used liberally throughout this paper.

        Grice himself warned that “the nature of conventional implicature needs to be examined before any free use of it, for explanatory purposes, can be indulged in” (1989, p. 46). In heeding his warning I aim to show that conventional implicature is, in the words Boër and Lycan used to denounce semantic presupposition, “a theoretical artifact of linguistic and philosophical semanticists” (1976, p. 81).[3] Both are a myth.


1. The Intuitive case for Conventional Implicature


If there are conventional implicatures, they must be conventional and they must be implicatures. An implicature is different from an entailment or a semantic presupposition, in that it is not necessary for the truth of the sentence. And for an implicature to qualify as conventional, it must depend on the conventional meaning of a particular locution in the sentence. A conventional implicature is, as Grice says, not “calculable.” In this way it is different from a conversational implicature, which depends on the fact that what is said is, in the context, not sufficiently plausible, informative, relevant, or otherwise appropriate and whose conveyance requires an inference based on the supposition that the speaker wouldn’t have said what he said if he hadn’t meant something more than that. The conventional implicature (CI-) thesis is that there are certain locutions which do not contribute to what is said and do not affect the truth or falsity of what is said and yet, by virtue of their conventional meanings, generate implicatures.[4]

        What reasons have been given for thinking there is such a thing as conventional implicature? I know of only one, rather weak argument for it (to be discussed below). The case for the CI-thesis seems to rest almost entirely on intuition. All those who share this intuition and see no need to back it up at least enjoy the good company of Frege and Grice.

        Grice is usually credited with the discovery of conventional implicature, but it was actually Frege’s idea—Grice merely labeled it. In “On Sense and Reference”, Frege wrote,

Subsidiary clauses beginning with ‘although’ … express complete thoughts. This conjunction actually has no sense and does not change the sense of the clause but only illuminates it in a peculiar fashion. [footnote: Similarly in the case of ‘but’ and ‘yet’.] We could indeed replace the concessive clause without harm to the truth of the whole by another of the same truth value; but the light in which the clause is placed by the conjunction might then easily appear unsuitable, as if a song with a sad subject were to be sung in a lively fashion. (Frege 1892/1994, p. 155)

Much later, in “The Thought”, Frege puts his idea this way:

With the sentence ‘Alfred has still not come’ one really says ‘Alfred has not come’ and, at the same time, hints that his arrival is expected, but it is only hinted. It cannot be said that, since Alfred’s arrival is not expected, the sense of the sentence is therefore false. … The word ‘but’ differs from ‘and’ in that with it one intimates that what follows is in contrast with what would be expected from what preceded it. Such suggestions in speech make no difference to the thought. (Frege 1918/1994, p. 522; my italics)

In other words, ‘still’ and ‘but’ (beyond its conjunctive import) have no bearing on the truth or falsity of what is said.

        Grice makes a similar point about ‘therefore’:

If I say (smugly), He is an Englishman; he is, therefore, brave, I have certainly committed myself, by virtue of the meaning of my words, to its being the case that his being brave is a consequence of (follows from) his being an Englishman. But while I have said that he is an Englishman, and said that he is brave, I do not want to say that I have said (in the favored sense) that it follows from his being an Englishman that he is brave, though I have certainly indicated, and so implicated, that this is so. I do not want to say that my utterance of this sentence would be, strictly speaking, false should the consequence in question fail to hold. (Grice 1989, p. 25)

‘Therefore’ is not the most convincing example, for it seems that the truth of the utterance does require that the second proposition be a consequence of the first. More plausible is Grice’s earlier example involving ‘but’,

            (3)           She is poor but she is honest.

where the putative contrast between being poor and being honest is, he claims, “implied as distinct from being stated” (Grice 1961, p. 127).[5]

        Frege and Grice, beyond appealing to intuition, do not try to justify what they say about ‘but’, ‘still’, and ‘therefore’. Grice does remark that conventional implicatures are detachable but not cancelable, but this cannot serve as a test for their presence. It does distinguish them from conversational implicatures, which are cancelable but not detachable (except for those induced by exploiting the maxim of manner, which depend on how one puts what one says), and from entailments, which are neither cancelable nor detachable. However, detachability is not an independent test. If a supposed implicature really were part of what is said, one could not leave it out and still say the same thing. To use ‘and’ rather than ‘but’, for example, would be to say less.

        Before attempting to evaluate the CI-thesis, we need to get clear on what it says and what it doesn’t say. This is important, inasmuch there are several misunderstandings about conventional implicature that have crept into the literature and need to be dispelled. The CI-thesis says that there are certain locutions which give rise to implicatures by virtue of their meanings. The propositions are said to be implicatures because their truth value does not affect the truth value of the entire utterance, so that the falsity of such a proposition is compatible with the truth of the entire utterance. So, according to the CI-thesis, the truth of (1) would not be affected if there were no contrast, real or presumed, between being huge and being agile. Accordingly, we may define conventional implicature as follows:


(CI) A proposition is a conventional implicature of an utterance just in case (a) the speaker (speaking seriously) is committed to the truth of the proposition, (b) which proposition that is depends upon the (or a) conventional meaning of some particular linguistic device in the utterance, but (c) the falsity of that proposition is compatible with the truth of the utterance.


As is clear from the passages quoted above, this is essentially what Frege and Grice had in mind. However, in the literature the conception encapsulated by (CI) seems to have been confused with several other ideas, none of which is essential to conventional implicature.

        One of Frege’s own remarks might confuse the issue. He characterizes the import of ‘but’ and ‘still’ as merely “hinted” or “intimated”, but this could be taken to mean that the relevant dimension is degree of speaker commitment. That can’t be right (or what Frege meant), for the speaker could be as much committed to what he is implicating as to what he is saying. Or it might seem that Frege is saying that the import of words like ‘but’ and ‘still’ is not fully explicit. There is a trivial sense in which (alleged) conventional implicatures are not explicit, as in Frege’s example, ‘Alfred has still not come’, where the import of ‘still’ is a complete proposition that is obviously not spelled out. Presumably the claim that such words generate conventional implicatures comes to more than this triviality. ‘OK’ or ‘No’, uttered by themselves, do not make explicit the propositions they express (in context), but this does not mean that they merely give rise to conventional implicatures.

        A third misleading, though common, way of describing what is involved in conventional implicature is to say that the relevant terms have “non-truth-conditional” meaning.[6] This misleadingly suggests that the conventional implicatures they generate are neither true nor false. But of course implicatures are true or false, and presumably different words generate ones with different truth conditions, as with the minimal pair in (4), for example:

            (4)   a.     John is a philosopher but he is rich.

                    b.     John is a philosopher so he is rich.

Obviously, the conditions under which the contrast indicated by ‘but’ obtains are different from those under which the consequence indicated by ‘so’ obtains. So the presence of ‘but’ or ‘so’ affects the truth conditions of something. The question is whether they affect the truth-conditions of what is said. Accordingly, the CI-thesis should not be read to mean that such terms have non-truth-conditional meaning.[7]

        As explained by (CI), a conventional implicature is a proposition which is conveyed due to the presence of a certain term with a certain meaning but whose falsity is compatible with the truth of the utterance. This was Grice’s conception and the one that gained currency among linguists, thanks to Karttunen and Peters (1979), who proposed it as a replacement for the notion of semantic presupposition, which by the late seventies had been discredited (see note 4). In their view,

a large set of cases that have been called presupposition are really instances of conventional implicature. The most obvious examples are those associated with particles like too, either, also, even, only, and so on. This class also includes the presuppositions of certain factive verbs, such as forget, realize, take into account, and so on, and those that accompany implicative verbs like manage and fail. … These are just a few examples; the list could be made much longer.[8] (1979, p. 11)

Their examples illustrate the sorts of terms that have been thought to generate conventional implicatures. I will call these terms ACIDs (alleged conventional implicature devices). Here is a representative list of those that I have seen:[9]




                  adverbs: already, also, barely, either, only, scarcely, still, too, yet

                  connectives: but, nevertheless, so, therefore, yet

                  implicative verbs: bother, condescend, continue, deign, fail, manage, stop

                  subordinating conjunctions: although, despite (the fact that), even though


        Karttunen and Peters offer only one argument for the CI-thesis. Applied to the occurrence of ‘even’ in (5),

            (5)           Even Bill likes Mary.

it is meant to show that “the truth of what [5] says depends solely on whether Bill likes Mary” (1979, p. 12). Suppose (5) is embedded in (6):

            (6)           John just noticed that even Bill likes Mary.

The crux of the argument is that (6) “does not mean that he has just noticed that other people like Mary or just noticed that Bill is the least likely person to do so” (1979, p. 13). But does it follow that either of these propositions is conventionally implicated? Karttunen and Peters assume that noticing a complex fact requires noticing its constituent facts. Their reasoning is that since (6) does not entail that John just noticed that other people like Mary or that Bill is the least likely person to do so, (6) says merely that John just noticed that Bill likes Mary, i.e., that ‘even’ does not contribute to what John is being said to have noticed. However, this line of argument is invalid, as (7) illustrates:

            (7)           John just noticed that Bill has three cars.

(7) could be true even if John has long known of two of Bill’s cars and just became aware of a third one.

          The same point applies to implicative verbs, such as ‘manage’. While functioning syntactically as a main verb, ‘manage (to)’ seems to modify the verb in its infinitival complement, entailing that the action in question requires effort or involves difficulty. However, this proposition is not a mere conventional implicature, for it is part of the content of sentences in which ‘manage’ occurs. The content of (8), for example,

            (8)           Bill managed to finish his homework.

includes both the finishing and the entailed difficulty. The same point made above about noticing applies here: one could notice that Bill managed to finish his homework even if one already knew that finishing it would be difficult.

        Despite the lack of compelling argument for the CI-thesis, the notion of conventional implicature continues to be used uncritically. For instance, in a recent paper in which he endorses a conventional implicature approach to ‘even’, Francescotti (1995) forcefully argues against various truth-conditional accounts and, with some ingenious examples, shows that the import of ‘even’ is contextually variable. However, this hardly supports his claim that it generates conventional implicatures. Francescotti does not argue but evidently just assumes that because the truth-conditional approach cannot accommodate contextual variability, only the conventional implicature approach can. Even more recently Rieber, who proposes an ingenious new theory of conventional implicature, does not find it necessary to show that there is such a thing (1997, p. 51n). He just gives a few stock examples of ACIDs, asserts that they “exhibit the peculiar feature of having a non-truth-conditional meaning” (1997, p. 51), and launches into his new theory (it will be discussed in §6).

        Look where you may, the case for conventional implicature seems to rest almost entirely on intuitions about cases. One is just supposed to see that (9), (10), (11), for example,

            (9)           Arthur was a lawyer but he was honest.

            (10)         Even though Arthur was a lawyer, he was honest.

            (11)         Arthur was a lawyer; nevertheless he was honest.

say nothing more than

            (12)         Arthur was a lawyer and he was honest.

Any suggestion that being a lawyer does not lend itself to being honest is, according to this intuition, merely a matter of implicature. The intuition is that even though a serious, literal user of (9), (10), or (11) commits himself to more than he would if he merely uttered (12), he is saying no more—if (12) is true, so are these other utterances. No support for this intuition is provided by the claim that conventional implicatures are detachable—that is just part of the intuition. For, as observed earlier, it just begs the question to use detachability as a test for the presence of a conventional implicature—if something really is part of what is said, you can’t say the same thing if you leave it out.

        In the next three sections, I will attempt to undermine the intuitive support for the CI-thesis. In §2 I will show that ACIDs pass a test that they would fail if they generated conventional implicatures instead of contributing to what is said. In §3 I will identify several factors that conspire to produce the spurious intuitions that make it seem that certain terms generate conventional implicatures. And in §4 I will propose a new way of looking at the semantic content of sentences containing ACIDs. From this perspective it becomes clear how CI-intuitions can arise and why they are spurious.

        Before proceeding, we should take account of certain issues raised by the notion of what is said, the other side of the contrast with what is implicated. This notion enters into a number of distinctions. In addition to being contrasted with what is conventionally implicated, what is said has been contrasted with what is conversationally implicated, with what is semantically presupposed, with what is meant, and with what is asserted. Each of these contrasts has its own theoretical significance, and this is not the place to compare and contrast them all (see Neale 1992). However, there is the underlying question of how strictly the notion of what is said should be construed. Presumably what is said corresponds to the constituents of the utterance (and to how they are combined syntactically). However, this does not mean that what is said must be made fully explicit. In the case of ellipsis, for example, what is not spelled out explicitly is still part of what is said. With (13), involving VP-ellipsis,

            (13)         Frege was a great philosopher of language, and so was Grice.

the speaker is saying, not merely implicating, that Grice was a great philosopher of language. On the other hand, what about the case of phrasal utterances? Suppose that (14), for example,

            (14)         Château Margaux

is given in answer to the question “what is your favorite wine?” Is the speaker saying that his favorite wine is Château Margaux? Well, he might have uttered ‘Château Margaux’ in answer to the question, “what is the most expensive wine you own?”, to “what is the most famous château of the southern Médoc?”, or to “what was Margaux Hemingway named after?” Even so, surely he has not merely implicated, conventionally or otherwise, that his favorite wine is Château Margaux. But if we grant that he has said this, we must allow that he has not said this fully explicitly.[10] There is a relevant difference between (14) and (13) however, because in (13) what is not spelled out is nevertheless recoverable without consideration of the extralinguistic or prior linguistic context.[11] In contrast, phrasal utterances like (14) do not, strictly speaking, say what they are used to convey in context. This would be Grice’s view, for on his strict conception, what is said must correspond to “the elements of [the sentence], their order, and their syntactic character” (1989, p. 87).[12] At any rate, for present purposes we should focus on utterances of complete sentences. For with an utterance of a complete sentence, there will be no element in what is said that does not correspond to some constituent of the sentence.[13] This leaves open the question of whether there can be constituents that do not correspond to anything in what is said but instead give rise to conventional implicatures.

        Although Grice thinks that his conception comports with “intuitive understanding of the meaning of say” (1989, pp. 24-25), it does depart somewhat from common usage (see Bach 1994, pp. 141-144). In particular, he stipulates that what is said falls under the category of what is meant.[14] On his stipulation, if one is not speaking literally and seriously, one is not saying anything but, as Grice puts it, merely “making as if to say” something. Not only is this unduly restrictive, in effect it conflates the locutionary and illocutionary levels of speech-act analysis. Indeed, Grice (and many others) tend to equate saying with stating. On the other hand, it seems reasonable for Grice to allow, while requiring that what is said correspond to the constituents and structure of the sentence, that disambiguation and reference fixation also contribute to the determination of what is said.

        Whether Grice’s conception of what is said is too restrictive or too relaxed, Sperber and Wilson (1986) and Recanati (1988) have gone so far as to deny that there is a level of what is said that even satisfies Grice’s criterion. They argue that what is said includes certain pragmatically determined elements beyond the uncontroversial ones that Grice allowed (fixing reference and resolving ambiguity), hence that what is said is not entirely explicit.[15] They argue further that in many cases no proposition is assigned to the sentence prior to the application of these processes. Their position is analogous to the view that understanding an utterance containing a metaphor does not require first giving it a literal reading, then judging that reading to be false, and only then, by applying Grice’s first maxim of quality (truthfulness), inferring the metaphorical import of the utterance. This is not the place to go into details (see Bach 1994, pp. 154-160), but suffice it to say that their view concerns the processing of semantic information, not the nature of it. Their argument assumes that any sentence whose utterance can be interpreted without computation of a literally expressed proposition does not express a proposition literally. This argument overlooks the important distinction between information available to the hearer and the specific pattern of inference followed by the hearer that exploits this information (Bach and Harnish 1979, pp. 89-91). The concept of what is said does not concern the nature of hearer’s inference to the intended content of the utterance but merely the semantically determined portion of the information available to that inference.

        Finally, it is important to recognize that the distinction between what is said and what is implicated is not exhaustive. Intermediate between the explicit and the implied is the implicit, or what I call conversational impliciture (Bach 1994; see the examples in note 14). Even if the category of conversational implic-a-ture needs to be complemented by conversational implic-i-ture, it is still a relatively uncontroversial category. Much more problematic is the category of conventional implicature. As we will see, to the extent that putative conventional implicatures really are implicatures, they are not conventional, and to the extent that they are conventional they are not implicatures.


2. The IQ Test


There is a very simple problem with the CI-thesis. It claims that using certain expressions (ACIDs), as a matter of their meaning, commits a speaker (using them seriously and literally) to a proposition that is not part of what he is saying. Its main support is the intuition that the falsity of this proposition is compatible with the truth of what is said, hence that this proposition is not part of what is said. However, proponents of the CI-thesis seem to have overlooked the fact that ACIDs can occur perfectly well in indirect quotations of utterances containing them. The reason this fact poses a problem for the CI-thesis is, quite simply, that the ‘that’-clause in an indirect quotation specifies what is said in the utterance being reported, and ACIDs can occur in specifications of what is said. Here are some examples:

            (1)           Shaq is huge but he is agile.

            (1IQ)       Marv said that Shaq is huge but that he is agile.

            (15)         Shaq can dunk and block shots too.

            (15IQ)    Marv said that Shaq can dunk and block shots too.

            (16)         Even Shaq can make some free throws.

            (16IQ)    Marv said that even Shaq can make some free throws.

            (17)         Shaq managed to make four out of nine free throws.

            (17IQ)    Marv said that Shaq managed to make four out of nine free throws.

So far as I know, advocates of conventional implicature have not confronted, much less explained away, the fact that such expressions as ‘but’, ‘too’, ‘even’, and ‘manage’ occur straightforwardly in indirect quotation. It is true that an ACID can be used by the reporter to make an editorial comment on what he is reporting as being said, but an ACID can also contribute to what is being reported.[16] When that is the case, leaving the term out would render the specification of what was said less than fully accurate, as in the following indirect quotations:

            (1-IQ)         Marv said that Shaq is huge and that he is agile.

            (15-IQ)   Marv said that Shaq can dunk and block shots.

            (16-IQ)   Marv said that Shaq can make some free throws.

            (17-IQ)   Marv said that Shaq made four out of nine free throws.

The fact that these indirect quotations are incomplete and to that extent inaccurate shows that the propositions alleged to be merely conventionally implicated by (1), (15), (16), and (17) are not detachable (as conventional implicatures are supposed to be) but are in fact part of what is said.[17] The ACIDs in these examples, as well as the others on the list of ACIDs given earlier, contribute to what is said because they pass the IQ test:


(IQ test): An element of a sentence contributes to what is said in an utterance of that sentence if and only if there can be an accurate and complete indirect quotation of the utterance (in the same language) which includes that element, or a corresponding element, in the ‘that’-clause that specifies what is said.


The IQ test is formulated in accordance with Grice’s strict construction of ‘what is said’, so as to apply only to indirect quotations that respect the constituent structure of the utterance being reported (a looser standard would make passing the IQ test too easy and bias it against the CI-thesis). Even so, the qualification ‘or a corresponding element’ must be included in this formulation to allow for indirect quotations that require adjustment for tenses or indexicals, e.g., ‘was’ for ‘is’, ‘then’ for ‘now’, and ‘he’ or ‘she’ for ‘I’. Similarly, the parenthetical ‘in the same language’ is included because if an utterance is reported in a different language, there may be no available oblique clause of the same syntactic form. Also, it should be understood that the IQ test applies only to fully indirect and not to “mixed” quotation, where part of the material in an otherwise indirect quotation is intended to be taken as directly quoted.[18] The IQ test can only exclude elements that do not contribute to what is said in the sense of propositional content—any element, even an interjection, contributes to what is said in the sense of what is uttered.

        Not all expressions commonly cited as sources of conventional implicatures pass the IQ test. In fact, there are many locutions that flunk it: their occurrence in an utterance defies inclusion in specifications of what is said. As the following examples illustrate, some locutions do not fit comfortably into indirect quotations of utterances containing them, at least not if used to report part of what was said:

            (18)         Moreover, Bill is honest.

            (18IQ)    #John said that moreover, Bill is honest.

            (19)         Now that you mention it, New York is a great place to visit.

            (19IQ)    #John said that now that he mentioned it, New York is a great place to visit.

            (20)         In other words, Bill is a liar.

            (20IQ)    #John said that in other words, Bill is a liar.

These utterance modifiers, as I call them, do not contribute to what is said, but that does not mean, as Grice suggested, that they give rise to conventional implicatures. Rather, they are vehicles for the performance of second-order speech acts, as Grice also suggested.[19] For example, “the meaning of ‘moreover’ is linked to the speech-act of adding, the performance of which would require the performance of one or another of the central speech-acts” (Grice 1989, p. 125). In using it at the beginning of a sentence whose utterance adds to what was previously said, one is not implicating but explicitly indicating that one is adding something. In using ‘in other words’, one is not implicating but explicitly indicating that the balance of the utterance will reformulate something just said.

        Utterance modifiers, such as ‘moreover’, ‘now that you mention it’, and ‘in other words,’ are used for the benefit of an audience. As they occur in sentences like (18) - (20), they do not contribute to the expression of a thought but characterize the expression of it (if one thinks in words, one does not think in those words). For this reason, not only do they flunk the IQ test, they do not belong in propositional attitude ascriptions, such as the following belief reports,

            (18BR)    #John believes that moreover, Bill is honest.

            (19BR)    #John believes that now that I mention it, New York is a great place to visit.

            (20BR)    #John believes that in other words, Bill is a liar.

They cannot contribute to ascribed belief contents.

        Because of the second-order function of an utterance modifier, its semantic content is not coordinate with that of the rest of the sentence. Rather, it characterizes what one is doing in the act of uttering the rest of the sentence (hence the label ‘utterance modifier’). If it is a connective, it is a discourse as opposed to a content connective. To appreciate the difference, compare the uses of ‘although’ in the following two utterances:

            (21)         Although the judge issued a gag order, my client will appear on Hard Copy.

            (22)         Although the judge issued a gag order, my client has an airtight alibi.

In (21), the content of the main clause contrasts with the content of the subordinate clause. The use of ‘although’ indicates that there is some sort of clash between the two. In (22), on the other hand, there is no suggestion of any contrast between the client’s having an alibi and the gag order. Rather, the speaker is using the ‘although’ clause to perform the second-order speech act of indicating that his first-order speech act, of making a statement about the case, is a violation of the gag order. The same contrast in uses is exhibited by ‘nevertheless’:

            (23)         The judge issued a gag order; nevertheless my client will appear on Hard Copy.

            (24)         The judge issued a gag order; nevertheless my client has an airtight alibi.[20]

Notice that the occurrences of ‘although’ in (21) and of ‘nevertheless’ in (23) pass the IQ test, but not those in (22) and (24). Compare (21IQ) and (22IQ), for example:

            (21IQ)    The lawyer said that although the judge issued a gag order, his client

                            would appear on Hard Copy.

            (22IQ)    #The lawyer said that although the judge issued a gag order, his client

                            had an airtight alibi.

The trouble with (22IQ) is that the content of the ‘although’-clause is not part of what the lawyer said. When used as part of an utterance modifier, ‘although’ (or ‘nevertheless’) flunks the IQ test.

        Care must be taken in applying the IQ test. For example, if we quoted just the second half of (23), as in (23-IQ),

            (23-IQ)   The lawyer said that nevertheless his client would appear on Hard Copy.

it would seem that ‘nevertheless’ as it occurs in (23) flunks the IQ test. However, a complete quotation of (23) would show that it passes the test:

            (23IQ)    The lawyer said that the judge had issued a gag order [but] nevertheless

                                his client would appear on Hard Copy.

False negatives can result from applying the IQ test to indirect quotations taken out of context.

        To sum up, occurrences of locutions that pass the IQ test contribute to what is said and locutions that function as utterance modifiers flunk it. Utterance modifiers, which will be catalogued in §5, do not contribute to what is said but indicate something about the act of saying it. They do not encode an element of thought but are essentially communicative devices. But most of the expressions which have been put forward, from Frege and Grice on, as sources of conventional implicatures pass the IQ test and do contribute to what is said.


3. Undermining the CI-Intuition


The fact that ACIDs pass the IQ test presents a major difficulty for the CI-thesis, but what about its intuitive support? To undermine that I will begin by using ‘but’ as a case study. Our observations about ‘but’ will then be extended to other ACIDs (but not to utterance modifiers). There are several factors, in my view, which conspire to make it seem that the contrastive import of ‘but’ does not contribute to what is said and gives rise merely to a conventional implicature.[21]

        The first factor is that ‘but’ does not encode a unique contrastive relation. As a result, its import can vary with the context. The most natural way of taking ‘but’ in (1), especially out of context,

            (1)           Shaq is huge but he is agile.

is as indicating that being huge tends to preclude being agile, but that is not the only contrastive relation between these properties. For an alternative, suppose we’re talking about NBA centers, most of whom are both huge and agile. I make a list of exceptions (centers I think are huge and clumsy), but mistakenly put Shaquille O’Neal on the list. You point out my mistake and say, “Shaq is huge but he is agile”. The contrast here between being huge and being agile is that having the second property distinguishes Shaq from the others on the list. And if I insist that Shaq is huge and clumsy and you again say, “Shaq is huge but he is agile”, you are indicating a third sort of contrast between being huge and being agile, namely that being agile, unlike being huge, is not a property I ascribed to Shaq. So there is no unique sort of contrast that is determined by the meaning of the word ‘but’. This shows that there is no unique contribution that ‘but’ makes to truth-conditional content, so that any specific claim about its truth-conditional content is vulnerable to counterexample, but it does not suggest, much less show, that ‘but’ generates a conventional implicature. This variability suggests that the different contextual effects are conversational, not conventional (I would call them implicitures rather than implicatures).

        A second factor contributing to the CI-intuition is that the contrast indicated by ‘but’ is often common ground rather than part of what the speaker is asserting. This does not mean that the relevant contrast is not part of what is said, for not all aspects of what is said need be of equal prominence. Being common ground, the contrast is pragmatically presupposed (see note 4), but this suggests not that it is conventionally implicated but only that it is informationally and conversationally subordinate to the main content of the utterance. The point here is analogous to part of Boër and Lycan’s diagnosis of the myth of semantic presupposition. They point out that, as with clefts and nonrestrictive relative clauses, “information that is plainly part of the semantic content of a sentence may have been placed … in so unemphatic a position … that we are disinclined to admit that that information is part of what that sentence says” (1976, p. 74).[22] The same point could be made about the alleged conventional implicature induced by the standard contrastive use of ‘but’.[23]

        A third factor, related to the second, is that the CI-intuition is produced as the result of a forced choice. One is forced to judge an utterance containing ‘but’ as simply true or simply false. Matters may not be so simple, however, for there may be more choices. What would we say about (25),

            (25)         Shaq is huge but he is rich.

given that there is no contrast (not even a presumptive one) between being huge and being rich? Advocates of the CI-thesis would acknowledge that there is something wrong with (25), even though Shaq is both huge and rich, but the lack of a relevant contrast does not incline them to regard (25) as false. That is, given the choice between judging (25) true and judging it false, they judge it true. They see it has having the same truth condition as (26),

            (26)         Shaq is huge and he is rich.

so that the contrast indicated by ‘but’ does not affect truth value and thus does not enter into what is said. But what if we do not have to choose between saying that (25) is categorically true or categorically false?

        No such choice is required in the case of sentences containing nonrestrictive relative clauses or appositives, as in

            (27)         Ann’s computer, which she bought in 1992, crashes frequently.

            (28)         Beth’s husband, a plumber, never washes the dishes.

If Ann bought her computer in 1993 and Beth’s husband is an electrician, are (27) and (28) false? No doubt many people, if forced to make a choice, would say the sentences are true anyway. But would they want to say that what is expressed by the material between the commas is not part of what is said? Surely not. It is either true or false that Ann bought her computer in 1992 and that Beth’s husband is a plumber. Are these propositions part of what is said? Well, whatever they are, clearly they are not merely implicated.

        I take it as uncontroversial that nonrestrictive relative clauses and appositives (and parentheticals) can be assessed as true or false. The relevant point is that intuitions about the truth or falsity of utterances containing them tend to ignore them. Why should this be? They seem to get ignored because they are not prominent enough to count (see Boër and Lycan’s observation quoted above). With (27) and (28), the proposition expressed by the main clause is the one whose truth value is intuitively judged to bear on that of the whole utterance.[24] For the same reason, in judging the truth or falsity of utterances containing ‘but’, we tend to ignore its contrastive force. To appreciate how this can happen, consider the effect of replacing ‘but’ with phrases that spell out the different contrastive forces (mentioned above) that might be intended in an utterance of (1) :

            (1)  a.      Shaq is huge and, unlike most huge people, he is agile.

            (1)  b.      Shaq is huge and, unlike others on the list, he is agile.

            (1)  c.      Shaq is huge and, contrary to what you said, he is agile.

Taken in context, the italicized phrases clearly can be assessed as true or false (to the extent that anything less than a complete clause can be true or false). Even so, the falsity of the proposition associated with the phrase does not seem sufficient for the falsity of the entire utterance. This is so even though (1a), (1b), and (1c) do not merely implicate the relevant contrast but actually say what the contrast is. So the mere fact, if it is a fact, that the falsity of a proposition expressed by part of a sentence does not affect the truth of the whole does not show that this proposition is a conventional implicature.

        The same is true of (1) itself, where the relevant contrast is not made explicit. In general, intuitions about the truth or falsity of utterances containing ACIDs tend to ignore the secondary proposition being expressed. This is clearly what happens with utterances containing nonrestrictive relative clauses or appositives (or parethenticals), where the truth value of the material set off between the commas (or parentheses) tends to be downplayed. In the next section I will suggest that this is possible because such utterances do not express one composite proposition but two separate and separately evaluable ones, one of which is peripheral to the main point of the utterance. When the secondary proposition is false but the primary one is true, intuitions about the truth or falsity of the whole utterance are forced. If we are forced to choose between true and false, we say “true”, we do so reluctantly, because we recognize that something isn’t right. In my view, calling the secondary proposition a conventional implicature isn’t right either.

        There is a fourth and final factor underlying the CI-intuition about a binary connective like ‘but’: because specifying the two conjuncts exhausts the entire content of a sentence like (1) except for the contribution of ‘but’ itself, spelling out its contribution requires adding an extra clause, as in (1+):

            (1+)        (i) Shaq is huge, (ii) Shaq is agile, and (iii) there is a certain contrast

                             between being huge and being agile.[25]

If we had to decide whether (1+) specifies just what (1) says or more than what it says, we would probably be disinclined to count (iii) as part of what is said in (1). (1+) has one more clause than (1), and ‘but’ by itself does not seem to have the force of an entire clause. Indeed, (iii) in (1+) mentions the properties of being huge and being agile, which already figured in (i) and (ii), for a second time. So spelling out the contrast indicated by ‘but’ requires one more clause than is contained in the original utterance. Consequently, if the import of ‘but’ is part of what is said and we want to spell out its import in a specification what is said, we would have to include all three clauses, as in the following extended indirect quotation:

            (1EIQ)    Marv says that Shaq is huge, that he is agile, and that there is a

                                certain contrast between being huge and being agile.

I suggest that one reason we are disinclined to count the content of the last clause of (1EIQ) as part of what is said is that (1EIQ) renders what is said in (1) as having more conjuncts than it has clauses.[26]

        However, there is a natural, more straightforward way of reporting what someone says in uttering (1):

            (1IQ)       Marv says that Shaq is huge but that he is agile

Here the word ‘but’ is included in the specification of what is said. It seems to me that the tendency of Grice to suggest, and others to agree, that the contrastive import of ‘but’ is merely a conventional implicature is the result of insisting on separating the contrastive import of ‘but’ from its conjunctive import, as in (1EIQ). One is led to posit a conventional implicature here only by implicitly forcing the specification into a set of independent conjuncts, whereupon it must either include one clause too many or omit the contrastive import of ‘but’. However, no extra clause is needed if the word ‘but’ is included in the specification of what is said, as in (1IQ). In the next section I will offer an explanation of why ‘but’ and similar terms can make a propositional contribution— they function as a kind of propositional operator—without adding an extra conjunct to the proposition expressed by sentences in which they occur.

        I have identified four factors that contribute to the CI-intuition about ‘but’. They seem to operate also in respect to other terms that are often said to generate conventional implicatures rather than contribute to semantic content. First, as the following examples illustrate, uses of the ACIDs ‘so’, ‘even’, and ‘still’ require contextual filling in:

            (29)         So Ann gave up smoking.

            (30)         Bev even likes Jesse Helms.

            (31)         Cal is still on the phone.

 In (29) ‘so’ indicates that there is a certain reason Ann did what she did but it doesn’t specify that reason, and in (30) ‘even’ indicates that there are certain more likable (or less unlikable) people that Bev likes, but it does not specify who. Less contextual filling in is required in the case of ‘still’, no doubt because there is only one possible dimension (time) to which ‘still’ is relevant.[27] However, no matter how much or how little contextual filling in is required, the need for it does not show that these terms generate conventional implicatures. Moreover, if the contextual details were supplied, say as in (29+) - (31+),

            (29+)      Cigarettes went up to $5 a pack, so Ann gave up smoking.

            (30+)      Fond of Southern politicians in general, Bev even likes Jesse Helms.

            (31+)      Although his mother called an hour ago, Cal is still on the phone.

the terms themselves would play the same roles as before. So context sensitivity is irrelevant to conventional implicature, as Grice pointed out when contrasting it with conversational implicature.

        The other three factors explain how our intuitions can downplay the truth-conditional import of ACIDs. First, our intuitive judgments of truth and falsity tend to slight the difference between what is stated in an utterance and its full propositional content, which may also include common ground. In order for (29), (30), and (31), to be understood (in the assumed context) without further explanation, it must be understood that cigarettes went up to $5 a pack, that Bev is fond of Southern politicians, and that Cal has been on the phone. Spelling them out would only repeat what was already understood. As we saw in the case of ‘but’, this leads us to exclude their import from what was said.

        Next, notice that if we do spell out these propositions, by replacing ‘so’, ‘even’, and ‘still’ with clausal paraphrases that make explicit their import in the context,

            (29p)      Because cigarettes went up to $5 a pack, Ann gave up smoking.

            (30p)      Not only is she fond of more likable Southern politicians, Bev likes Jesse Helms.

            (31p)      Cal has been on the phone (with his mother for an hour); he is now on the phone.

the falsity of the secondary proposition (the one which in (29) - (31) is alleged to be merely a conventional implicature) becomes relevant to the truth or falsity of the whole utterance. This is evident if we consider the following responses:

            (29r)       That’s not true. She gave up smoking because it is hazardous to your health.

            (30r)       That’s not true. Helms is the only southern politician Bev likes.

            (31r)       That’s not true. Cal hasn’t been on the phone at all—it’s off the hook.

As rejoinders to (29p) - (31p), they are perfectly natural. On the other hand, they are inappropriate as responses to (29) - (31) because the secondary proposition is in the background, so that the possibility of its falsity tends to be disregarded. And if one were forced to judge (29) - (31) as true or false when the primary proposition is true and the secondary proposition (the alleged conventional implicature) is false, one would judge that they are true.

        The last factor is that making the force of ACIDs explicit requires one more clause than is contained in the original utterance. When this force is spelled out, as in (29p) - (31p), the result seems semantically inequivalent to the original utterance just because it has an extra clause. Notice that this extra clause repeats material from the rest of the sentence (or the previous utterance). This suggests that words like ‘so’, ‘even’, or ‘still’ function as operators on that material to yield the additional propositional contents that they do.


4. Expressing Multiple Propositions


I have identified four factors that contribute to the CI-intuition about the conventional import of ACIDs. There is, I suggest, an underlying explanation for the effect of at least the last three. They are aided and abetted by the clandestine operation of a certain insidious assumption:

            (OSOP)  Every indicative sentence expresses exactly one proposition.

This widespread but less than obvious assumption of one sentence, one proposition is just a version of the grammar-school dictum that a complete sentence expresses a complete thought, i.e., exactly one thought. I am not suggesting that philosophers or linguists have explicitly defended OSOP (although it does seem to operate in much semantic theorizing) and, in fact, it has been explicitly rejected.[28] All I am suggesting here is that it functions as an implicit assumption underlying people’s intuitions about what sentences do and do not say.

        What could have led people to make this assumption, however implicitly? Perhaps the culprit was not grammar school but logic class. We got used to the standard truth-functional connectives ‘and’ and ‘or’ and to how they are represented in the propositional calculus: ‘p’ conjoined with ‘q’ yields the conjunctive proposition ‘p & q’; ‘p’ disjoined with ‘q’ yields the disjunctive proposition ‘p v q’. However, this does not show that in general a sentence whose content includes more than one proposition expresses a composite proposition (with those individual propositions as components). As we saw earlier, there are two propositions expressed by (27), for example,

            (27)         Ann’s computer, which she bought in 1992, crashes frequently.

that Ann’s computer crashes frequently and that she bought it in 1992. (27) expresses these two propositions, but it does not express their conjunction. OSOP arbitrarily requires that such sentences express no more than one proposition (it also disregards the fact noted earlier (note 14) that many syntactically complete sentences fail to express even one proposition). With sentences containing ACIDs like ‘but’, ‘so’, ‘even’, and ‘still’, there is no such thing as the proposition expressed—in these cases what is said comprises more than one proposition. And when the sentence does so without expressing the conjunction of these propositions, and these propositions differ in truth value, the sentence as a whole is not assessable as simply true or simply false.

        Once we reject the assumption of one sentence, one proposition, we are no longer forced to choose to between treating the import of an ACID as either an entailment or a conventional implicature. If what is said can comprise more than one proposition, the presence of an ACID can be responsible for one of them. This proposition does not have to be regarded as either a conjunct of the one proposition expressed by the entire sentence or as not part of what is said at all but merely a conventional implicature. To see how the presence of an ACID could result in the expression of an additional proposition, look again at (29) - (31) and the result of paraphrasing its ACID with a clause:

            (29)         So Ann gave up smoking.

            (29p)      Because cigarettes went up to $5 a pack, Ann gave up smoking.

            (30)         Bev even likes Jesse Helms.

            (30p)      Not only is she fond of more likable Southern politicians, Bev likes Jesse Helms.

            (31)         Cal is still on the phone.

            (31p)      Cal has been on the phone (with his mother for an hour); he is on the phone.

Clearly the falsity of the additional proposition is sufficient for the falsity of the paraphrases. (29p) is false if Ann did not give up smoking for the reason in question, (30p) is false if Bev is not fond of other Southern politicians, and (31p) is false if Cal hadn’t been on the phone. In these cases the additional proposition is not even a candidate for being a conventional implicature—it is obviously part of what is said. Now I am not suggesting that this is just as obvious with (29), (30), and (31). In these cases, the additional proposition in question is not expressed as a separate clause (as a conjunct in a conjunctive proposition), for there is a different way in which this proposition comes into play.

        Terms like ‘but’, ‘so’, ‘still’, and ‘even’ function as operators of a special sort. I call them preservative operators because in operating on a sentence (or phrase) to yield a new proposition, they preserve the original proposition. If, for example, ‘O’ is a unary preservative operator on sentences and expresses the property of being F, and ‘S’ expresses the proposition that p, then ‘O(S)’ expresses both the proposition that p and the proposition that F(p). To illustrate, the occurrence of ‘still’ in (31) indicates that the same state of affairs expressed by the sentence that ‘still’ operates on obtained during some interval up to the reference time. In the case of ‘so’, which is a binary operator, its occurrence in (29) indicates that what is expressed by the sentence that follows it is a consequence of what was expressed by the preceding one (which may be merely understood). Obviously, the

suggestion that ACIDs are preservative operators needs to be developed in detail. Such an account would indicate, for each ACID, whether it is a unary or a binary operator, how, if the term is binary, it picks up the first of its operands from prior discourse or context, what property or relation it ascribes to its operand(s), and how the precise character of this property or relation is determined by context.

        As observed earlier, our intuitions about sentences like (29), (30), and (31) tend to be insensitive to the falsity of the secondary proposition being expressed. The force of these intuitions is neutralized once we allow that sentences can express two (or more) propositions of different degrees of prominence, and their ranking in prominence depends not only on linguistic form but also on contextual factors. Whatever the exact story of how this ranking is determined, it pertains not to the semantic question of which proposition(s) an utterance expresses but only to the psychological explanation of which proposition(s) our intuitions of truth and falsity take into account and which ones they ignore.

        The OSOP has provided implicit support for the CI-intuition (as well as earlier intuitions about semantic presupposition) by forcing the terms of the debate. It requires us to decide whether a certain proposition, e.g. with (31) that Cal was on the phone, is entailed or merely implicated (or presupposed, when presupposition was the issue). We are thus forced to judge whether (31) expresses merely the simple (31a) and conventionally implicates (31b) or expresses their conjunction, thereby entailing both.

            (31)  a.   Cal is on the phone.

            (31)  b.   Cal has been on the phone.

But there is another option. On the multiple-proposition view (31) expresses two propositions, (31a) and (31b)—it does not express their conjunction and merely entail them. Judgments about (31) are forced if, when (31a) is true and (31b) is false, we have to choose between a simple “true” and a simple “false”. The multiple-proposition view presents another option.

        Not only does OSOP underlie the CI-intuition, it lurks behind the only argument I am aware of for the CI-thesis. This is the argument mentioned earlier, due to Karttunen and Peters, who point out that (6), repeated here,

            (6)           John just noticed that even Bill likes Mary.

“does not mean that he has just noticed that other people like Mary or just noticed that Bill is the least likely person to do so” (1979, p. 13). I claimed that it does not follow that this proposition is conventionally implicated but only that noticing something, in this case (5),

            (5)           Even Bill likes Mary.

does not require noticing everything necessary for it.

        Now it is interesting to note that Karttunen and Peters observe that there are two grades of criticism that can be leveled at someone who asserts (5) when (5a), (5b), and (5c) are not all true:

            (5)       a.  Bill likes Mary.

                        b.     Other people besides Bill like Mary.

                        c.     Of the people under consideration, Bill is the least likely to like Mary.[29]

An utterance of (5) is deemed false if (5a) is rejected but “partial credit” is still in order if only (5b) or (5c) is false. As Karttunen and Peters describe the situation, the “stronger criticism” is called for if the speaker’s “principal commitment, which is to the truth of [(5a)], has run afoul of the facts. The milder criticism is warranted only when the principal commitment accords with the facts and the speaker’s error concerns one of his subsidiary commitments” (1979, p. 12; my italics). Karttunen and Peters’s argument for conventional implicature relies on the assumption that what is said is exhausted by the “principal commitment”. Only thus is there any reason to consign the subsidiary commitment to the status of conventional implicature.[30]

        Although it is arbitrary to assume one sentence, one proposition, the suggestion that a sentence can express more than one proposition might still seem problematic. In fact, it is problematic within the framework of a truth-conditional semantics, because a sentence that expresses more than one proposition, as opposed to a conjunction of propositions, does not have a unitary truth condition. A semantic framework in which meanings of sentences are given not in terms of truth conditions but in terms of things that have truth conditions, namely propositions, is better suited for handling this problem.[31] Then we can speak of the truth or falsity of the different propositions that are expressed by the utterance of one such sentence without having to judge the utterance as a whole as true or false. This would eliminate any temptation to speak of ACIDs as having “non-truth-conditional” meaning.

        The multiple-proposition view explains a phenomenon observed earlier (§3). I pointed out that ACIDs pass the IQ test even though the propositions that they give rise to cannot be made explicit by means of an extra clause in the indirect quotation. The multiple-proposition view explains why including an additional conjunct renders the indirect quotation unfaithful to the utterance being reported, since the utterance does not express a single proposition containing the one introduced by the ACID as a conjunct. In reporting an utterance containing a unary ACID, such as ‘still’, we can use the ACID in the report. What we can’t do is report the two propositions that are expressed, such as (31a) and (31b), by conjoining them in a single ‘that’-clause in an indirect quotation like (31EIQ):

            (31EIQ) Don said that Cal is on the phone and that Cal has been on the phone.

For what was said was not a conjunctive proposition. On the other hand, if we separate the propositions, as in (31'EIQ),

            (31'EIQ) Don said that Cal is on the phone and said that Cal has been on the phone.

this misleadingly suggests that he said the two things separately, one after the other, when what he said was simply (31),

            (31)         Cal is still on the phone.

The best way to report this utterance is with (31IQ),

            (31IQ)    Don said that Cal is still on the phone.

Thus, the proposition generated by an ACID like ‘still’ cannot be included explicitly in a faithful indirect quotation of an utterance containing the ACID. But including the additional proposition is not necessary, because ACIDs pass the IQ test.


5. Utterance Modifiers and Second-order Speech Acts


Utterance modifiers are locutions used to comment on the main part of the utterance in which they occur. Typically they occur at the beginning of a sentence, but they can occur elsewhere, set off by commas or pauses. Because they are not semantically coordinate with the rest of the sentence, they flunk the IQ test and, as a result, do not contribute to what is said in utterances containing them.

        As vehicles for performing second-order speech acts, utterance modifiers characterize some aspect of the first-order speech act performed in uttering the rest of the sentence. I propose to use some Austinesque neologisms to categorize them in terms of the type of comment they are used to make. So, for example, ‘moreover’ and ‘furthermore’ are additives, indicating that the main utterance (the first-order speech act being commented on) adds to what was already said. ‘To change the subject’ and ‘to get back to the point’ are topicals, indicating something about the subject matter of the main utterance. ‘In other words’ and ‘so to speak’ are formulationals, indicating something about how the main utterance is being put. ‘Speaking as a friend’ and ‘man to man’ (as well as the more specialized ‘Marine to Marine’, as used by F. Lee Bailey to Mark Fuhrman) are relationals, indicating the personal relation in the context of which the main utterance is being made. And ‘in case you’re interested’ or ‘since you’ll find out anyway’ are explanatories, giving a reason for making the main utterance.[32]

        As the above examples suggest, utterance modifiers may be categorized by the kind of comment they make on the main utterance. The taxonomy below distinguishes fourteen such categories, and representative locutions are included under each.[33] The labels for the following categories should be self-explanatory.


Types of Utterance Modifiers


1. topicals: anyway, as for …, by the way, incidentally, in that regard, speaking of …, to change the subject, to digress for a moment, to get back to the subject, while we’re on the subject

2. positionals: first of all (secondly, etc.), next, finally, in conclusion, lastly, to sum up

3. additives: along the same lines, also, besides, by the same token, equally, furthermore, in addition, in fact, likewise, moreover, more specifically, not only that, on top of that, similarly, what is more

4. illustratives: by way of example, for example, for instance, for one thing, if I may illustrate, in particular, to illustrate, to show you what I mean

5. conclusives: accordingly, after all, all in all, all things considered, to conclude

6. concessives: although …, anyway, at any rate, be that as it may, even so, in any case/event, just the same, nevertheless, regardless, still, whatever

7. contrastives: but, conversely, however, in contrast, instead, on the other hand, rather, yet

8. formulationals: alternatively, as it were, briefly, crudely, in a nutshell, in a word, in effect, in other words, in plain words, in technical terms, metaphorically speaking, more precisely, so to speak, speaking loosely, strictly speaking, that is, to be specific, to change the analogy/ image/metaphor, to elaborate, to make a long story short, to oversimplify, to put it mildly, to put it more accurately, to use a cliché, without exaggeration, without going into detail

9. emphatics: above all, indeed, it cannot be overemphasized/stressed too much that, mark my words, most importantly, to say the least

10. veracitives: believe it or not, in all candor, frankly, seriously, to set the record straight, to tell (you) the truth, truthfully

11. secretives: between you and me, confidentially, if you don’t tell anybody, off the record

12. relationals: as your friend/doctor/mother (etc.), man to man, personally, since I’m your friend/doctor/mother (etc.), speaking as a friend

13. mitigatives: I don’t mean to interrupt/intrude but, I don’t mean to pressure you but, if I may interrupt, if it isn’t too much trouble, if you ask me, if you don’t mind, I’m no expert but, I’m sorry to have to ask you this but, it’s none of my business but, needless to say, since it would be easier for you, since you’re so forgetful, you’re entitled to your opinion but

14. explanatories: although I shouldn’t say anything, although it’s none of your business, before I forget, by way of explanation, for your information, if I remember correctly, if you really want to know, if you want my opinion, in case you don’t know, in case you haven’t heard, in case you’re interested, in case you’ve forgotten, just so you know, just to remind you, now that you mention it, since you’ll find out anyway, since you’re in a position to know, since you’re so smart


This taxonomy could surely be expanded or otherwise refined, and no doubt some of these locutions could be cross-listed. Also, I am not claiming (except for those that are speech-act idioms, such as ‘mark my words’, ‘just so you know’, and ‘now that you mention it’) that these locutions function exclusively as utterance modifiers—they function as such only when they occur at the beginning of a sentence or are otherwise set off. In this regard, notice that a couple of the ACIDs taken up earlier, ‘but’ and ‘still’, appear above. They belong there because they also have uses as utterance modifiers. Sentence-initial ‘but’ is generally the ‘but’ of rebuttal, used to introduce a reason or evidence against something previously asserted. If someone says that Shaquille O’Neal is well-coordinated and you reply, “But Shaq can’t make free throws,” you are not expressing a contrast of the sort expressed in (1) but rather are objecting to the claim just made. As for ‘still’, it is included in the category of concessives not as a marker of continued activity but as an utterance modifier (typically when it occurs correlatively with ‘although …’). Finally, notice that although ‘also’ appears in the category of additives, ‘too’ does not. Unlike ‘also’, ‘too’ does not occur in sentence-initial position (at least not in my idiolect), and in other positions it makes the sort of contribution to content identified in §2.

        As we saw in §2, utterance modifiers flunk the IQ test. The reason for this is that their role is to comment on the main utterance. If the entire utterance could be reported in one indirect quotation, the utterance modifier (or some paraphrase of it) would have to be included in the indirect quotation. But including it would not report the quoted speaker as having commented on his main utterance. This is clear when we consider the following attempts to include an utterance modifier in an indirect quotation.

            (32)         Confidentially, Al’s wife is having an affair.

            (32IQ)    #Bill said that confidentially, Al’s wife was having an affair.

            (33)         Man to man, your wife is having an affair.

            (33IQ)    #Bill said to Al that man to man, his wife was having an affair.

            (34)         In case you’re interested, your wife is having an affair.

            (34IQ)    *Bill said to Al that in case he was interested, Al’s wife was having an affair.

So, for example, (32IQ) does not report Bill as having characterized his statement as confidential. If anything, it reports the affair as being confidential. On the other hand, if the utterance is being reported as confidential, ‘confidentially’ would precede the ‘that’-clause, as in (32'IQ),

            (32'IQ)   Bill said confidentially that Al’s wife was having an affair.

But there it would require a full clause to report Bill as characterizing his own utterance is confidential:

            (32''IQ)  Characterizing his utterance as confidential, Bill said that Al’s wife

                                was having an affair.

 Similarly, (33IQ) does not report Bill as having characterized his statement as man to man but, if anything, reports the affair as being man to man. If ‘man to man’ precedes the ‘that’-clause,

            (33'IQ)   Bill said man to man to Al that his wife was having an affair.

then Bill’s utterance to Al is being reported as being man to man, but it does not indicate that Bill was so characterizing it.

        Finally, I should mention two other categories of expression which, when they occur in sentence-initial position, might seem to be utterance modifiers but are in fact content modifiers:

      implicatives: as a result, consequently, hence, in that case, so, therefore, thus

      assessives: amazingly, coincidentally, disappointingly, (un)expectedly, (un)fortunately,      incredibly, inevitably, ironically, luckily, naturally, obviously, oddly, of course,                predictably, regrettably, sadly, surprisingly, undoubtedly

These locutions do not comment on the main act of utterance but rather contribute to the content of the utterance. Implicatives should not be confused with conclusives, a category of utterance modifier. As Bellert has suggested (1977, p. 342), sentences containing assessives (she calls them evaluatives), which are factive, are used to express two propositions, the proposition expressed by the matrix sentence and the proposition that the fact stated by that proposition has the property expressed by the assessive. That is, assessive adverbs are preservative operators. (35), for example,

            (35)         Surprisingly, Walkabout won the Kentucky Derby.

expresses the proposition that Walkabout won the Kentucky Derby and the proposition that the fact that Walkabout won the Kentucky Derby was surprising. In this respect, ‘surprisingly’ and the other adverbs on the list of assessives are different from modal adverbs like ‘necessarily’ and ‘certainly’. (35’), for example,

            (35)       Certainly Walkabout won the Kentucky Derby.

expresses just one proposition, that it is certain that Walkabout won the Kentucky Derby. In other words, a speaker of (35’) says just one thing, whereas a speaker of (35) says two. This difference between (35) and (35’) provides additional support for the multiple-proposition hypothesis.

        As we saw with ‘but’ in §2, content modifiers, unlike utterance modifiers, can occur within the ‘that’-clause of an indirect quotation and contribute to the specification of what was said. Consider the following examples, involving the implicative ‘therefore’ and the assessive ‘predictably’:

            (36)         Therefore, I exist.

            (36IQ)    René said that therefore, he existed.

            (37)         Predictably, Dave is late again.

            (37IQ)    Ed said that predictably, Dave is late again.

In these cases, the term can occur unproblematically in indirect quotation, although its inclusion will make sense only if the context of the reported utterance is understood.

        Utterance modifiers flunk the IQ test. The existence of such locutions, which encode second-order illocutionary forces,[34] poses an interesting challenge for semantics. The challenge is to explain how a word or phrase in a sentence can be syntactically in construction with the matrix clause but not be semantically coordinate with it. And I have certainly not explained how this is possible. But at least the examples show that this is a genuine phenomenon, and present approaches to semantics do not seem to account for it. It would be desirable to develop a conception of semantics that reckons with the fact that sentences can contain locutions that encode second-order illocutionary forces, as well as with the fact that certain sentences (containing ACIDs) can express more than one proposition. The syntax of both sorts of sentences has been investigated (see Espinal 1991 and references there), and perhaps progress on their semantics can build on that.


6. Alternative Accounts


Although I do not accept Grice’s doctrine of conventional implicature, I have endorsed his view that certain locutions are vehicles for the performance of second-order speech acts. Two apparent alternatives to this view have been suggested recently but, insofar as they are plausible, they do not seem to be genuine alternatives.

        One suggestion is that the locutions in question encode “procedural” rather than “conceptual” meaning. As Wilson and Sperber explain this idea, such locutions “should not be seen as encoding concepts. They do not contribute to the truth conditions of utterances, but constrain the inferential phase of comprehension by indicating the type of inference process that the hearer is expected to go through” (1993, p. 11). The force of calling their meanings “procedural” is, as Blakemore puts it, that they “constrain the hearer’s interpretation of utterances” by “provid[ing] an instruction for performing computations over th[e] proposition” expressed by the utterance (1990, pp. 205 and 211). But what does this amount to? Surely ordinary speakers do not have sophisticated intentions, expressible only in the jargon of cognitive psychology, concerning how the use of certain locutions should affect the cognitive processes going on in hearers’ heads. But if the proceduralist claim does not concern the intentions involved in the use of utterance modifiers, it would seem to apply to locutions of any sort.[35] After all, in some way or another anything one utters “constrains the inferential phase of comprehension” and “provides an instruction for performing computations”. Understanding the use of any bit of language requires engaging in a cognitive process, and particular expressions provide the input to that process. What distinguishes the locutions under discussion is not they encode processing instructions but that they encode speech act information. Morever, as Rieber points out in his criticism of the proceduralist view, “it is not clear why procedural meaning must be non-conceptual” (1997, p. 66). It seems, then, that to deny that the meanings of the locutions in question are “conceptual” comes to nothing more than the denial that they “contribute to the truth conditions of utterances”.[36] They do not do so because they are utterance modifiers.

        Rieber himself makes a different proposal. His idea is that locutions that generate conventional implicatures are “tacit performatives”. What exactly does that mean? If it means nothing more than that these locutions encode speech act information, then it is just a restatement of Grice’s idea that they are devices for performing (second-order) speech acts—utterance modifiers are tacit performatives in that sense. Indeed, in that minimal sense, even the imperative and interrogative moods would qualify as tacit performatives: they too encode speech act information (see note 33). They are not explicit performatives because, although they indicate that the speaker is performing an act of a certain sort (making a request, asking a question), they do not say what sort of act that is. The same is true of utterance modifiers.

        Presumably Rieber intends something stronger. Note here that he is concerned not with utterance modifiers but with certain ACIDs, such as ‘but’, ‘so’, ‘still’, and ‘although’, which he calls “discourse connectives”. These are terms which in his view contribute to content (although, as illustrated at the end of §2, ‘although’ can also be used as an utterance modifier). In calling discourse connectives “tacit performatives”, Rieber means that they are synonymous with, or at least are paraphrasable literally by, performative sentences (first-person present-tensed sentences containing performative verbs). His candidate verb is ‘suggest’. So on his analysis (1) says something like (1R),

            (1R)    Shaq is huge and (I suggest that this contrasts) he is agile.

The parentheses indicate that a tacit performative does not contribute to the truth or falsity of the whole utterance.

        There are several difficulties with Rieber’s account. For one thing, his choice of ‘suggest’, meaning “tentatively put forward a proposition” (1997, p. 54n), falsely assumes that there must be something tentative in the use of ‘but’. Yet as we saw earlier, an alleged conventional implicature need not be put forward more tentatively than what is said. Consider how Rieber would render (31) - (33), where the parenthetical material is supposed to be synonymous with ‘so’, ‘still’, and ‘even’, respectively:

            (31R)      [Cigarettes went up to $5 a pack.] (I suggest that this is a consequence)

                            Ann gave up smoking.

            (32R)      [Fond of Southern politicians in general,] Bev (I suggest that this is

                            more surprising) likes Jesse Helms.

            (33R)      (I suggest that this has been the case for some time) Cal is on the phone.

In each case, a speaker could be committing himself strongly to what Rieber’s analysis represents as merely suggested. So there is no unique performative with which any of the discourse connectives in question is synonymous. Indeed, as it turns out, he is not committed to his candidate ‘suggests’—he uses it only “for specificity”—and thinks that “which performative works best is a question that can be left open” (1997, p. 53). However, since different candidates will not be synonymous with each other, it needs to be shown that a given discourse connective is synonymous with a particular performative. Otherwise, there is no coherent way to take literally the claim that discourse connectives are synonymous with performatives.[37]

            Secondly, Rieber’s view gives the wrong explanation of why so-called conventional implicatures are not cancelable. Consider what happens if a speaker of (1) denies that he is suggesting that the contrast indicated by ‘but’ obtains:

            (38)         Shaq is huge but he is agile [(1)], although I am not suggesting that

                            there is any contrast between being huge and being agile.

Given that he is suggesting a contrast with his use of ‘but’, the speaker is guilty not of a logical contradiction but of a pragmatic one—he is adding something that undercuts the sincerity of the earlier part of his utterance.[38] With (39), on the other hand,

            (39)         Shaq is huge and (I suggest that this contrasts) he is agile,

                            although I am not suggesting that there is any contrast between

                            being huge and being agile.

the speaker is contradicting himself logically. But if cancelling a conventional implicature is a logical contradiction, as Rieber’s account seems to imply with (39), then so-called conventional implicatures are not the genuine article—they are part of what is said.

        Finally, locutions like ‘but’, ‘so’, ‘still’, and ‘even’, pass the IQ test, and are therefore capable of contributing their content to what is said. So consider what happens if we embed (1R), for example, in indirect quotation:

            (1RIQ)        Marv said that Shaq is huge and (I suggest that this contrasts) that he is                                                           agile.

Rieber’s analysis requires that the speech reporter, not the reported speech, be making the contrast, and yet (1IQ) can just as well be used to ascribe the making of the contrast.

            (1IQ)       Marv said that Shaq is huge but that he is agile.

Rieber acknowledges this difficulty (1997, p. 57) but does not explain how his account can handle it.

        It seems, then, that the two alternative views are plausible only insofar as they can be interpreted as restatements of the view that certain locutions are used to perform second-order speech acts. Calling the meanings of these locutions “procedural”, if this is to distinguish their meanings from those of other expressions, and calling them “tacit performatives”, if this is not to attribute to them properties that they do not have, would then just be ways of saying that they are devices for performing (second-order) speech acts.


7. Summing Up


As Grice himself warned, “the nature of conventional implicature needs to be examined before any free use of it, for explanatory purposes, can be indulged in” (1989, p. 46). When we do examine it, we find that there are no clear examples of it. The phenomena that have been thought to be conventional implicatures turn out to be examples of something else. In some cases, the propositions that are alleged to be conventional implicatures are actually aspects of what is said. With ACIDs, which pass the IQ test, the proposition that is allegedly a conventional implicature is really part of what is said. I suggested how the falsity of that proposition could seem irrelevant to the falsity of the entire utterance and thereby lead to the CI-intuition. However, this intuition is illusory, essentially depending on the false assumption that a sentence can express only one proposition. In other cases, the expressions in question are utterance modifiers. They do not contribute to what is said, but they do not generate conventional implicatures either—they are are vehicles for the performance of second-order speech acts. Utterance modifiers pose an interesting challenge for any viable semantic theory, for their semantic contents are not coordinate with the semantic content of the rest of the sentence but operate, so to speak, at one level up.





Bach, K.: 1987, Thought and Reference, Oxford University Press, Oxford.

Bach, K.: 1994, ‘Conversational Impliciture’, Mind and Language 9, 124-162.

Bach, K. and and R. Harnish: 1979, Linguistic Communication and Speech Acts, MIT Press, Cambridge, Mass.

Bach, K. and and R. Harnish: 1992, ‘How Performatives Really Work’, Linguistics and Philosophy 15, 93-110.

Barker, S.: 1993, ‘Even, Still, and Counterfactuals’, Linguistics and Philosophy 14, 1-38.

Bellert, I.: 1977, ‘On Semantic and Distributional Properties of Sentential Adverbs’, Linguistic Inquiry 8, 337-351.

Blakemore, D.: 1990, ‘Performatives and Parentheticals’, Proceedings of the Aristotelian Society 91: 197-213.

Boër, S. and W. Lycan: 1976, The Myth of Semantic Presupposition, Indiana Linguistics Club, Bloomington, Ind.

Cappelen, H. and E. Lepore: 1997, ‘Varieties of Quotation’, Mind 106, 429-450.

Espinal, T.: 1991, ‘The Representation of Disjunct Constituents’, Language 67, 726-762.

Fabb, N., 1990: ‘The Difference between English Restrictive and Nonrestrictive Relative Clauses’, Journal of Linguistics 26, 57-78.

Francescotti, R.: 1995, ‘Even: The Conventional Implicature Approach Reconsidered’, Linguistics and Philosophy 18, 153-173.

Fraser, B.: 1996, ‘Pragmatic Markers’, Pragmatics 6, 167-190.

Frege, G.: 1892/1994, ‘On Sense and Reference’, in R. Harnish (ed.), Basic Topics in the Philosophy of Language, Prentice-Hall, Englewood Cliffs, N.J., pp. 142-160.

Frege, G.: 1918/1994, ‘The Thought: A Logical Inquiry’, in R. Harnish (ed.), Basic Topics in the Philosophy of Language, Prentice-Hall, Englewood Cliffs, N.J., pp. 517-535.

Grice, P.: 1961, ‘The Causal Theory of Perception’, Proceedings of the Aristotelian Society, Supp. Vol. 35, pp. 121-52.

Grice, P.: 1989, Studies in the Way of Words, Harvard University Press, Cambridge, Mass.

Harnish, R.: 1994, ‘Mood, Meaning, and Speech Acts’, in S. Tsohatzidis (ed.), Foundations of Speech Act Theory, Routledge, London, pp. 407-459.

Kaplan, D.: 1977/1989, ‘Demonstratives’, in J. Almog, H. Wettstein, and J. Perry (eds.), Themes from Kaplan, Oxford University Press, Oxford, pp. 481-563.

Karttunen, L. and S. Peters: 1979, ‘Conventional Implicature’, in Syntax and Semantics, vol. 11: Presupposition, C-K Oh and D. Dineen (eds.), Academic Press, New York, pp. 1-56.

King, J.: 1995, ‘Structured Propositions and Complex Predicates’, Noûs 29, 516-535.

Lambrecht, K.: 1994, Information Structure and Sentence Form, Cambridge University Press, Cambridge, Eng.

Levinson, S.: 1983, Pragmatics, Cambridge University Press, Cambridge, Eng.

Neale, S.: 1992, ‘Paul Grice and the Philosophy of Language’, Linguistics and Philosophy 15, 509-559.

Recanati, F.: 1988, ‘The Pragmatics of What is Said’, Mind and Language 4, 295-329.

Rieber, S.: 1997, ‘Conventional Implicatures as Tacit Performatives’, Linguistics and Philosophy 20, 51-72.

Sperber, D. and D. Wilson: 1986, Relevance, Harvard University Press, Cambridge, Mass.

Stainton, R.: 1995, ‘Non-sentential Assertions and Semantic Ellipsis’, Linguistics and Philosophy 18, 291-296.

Wilson, D. and D. Sperber: 1993, ‘Linguistic Form and Relevance’, Lingua 90, 1-25.

Ziff, P.: 1972, ‘What is Said’, in G. Harman and D. Davidson (eds.), Semantics of Natural Language, Reidel: Dordrecht, Holland, pp. 709-721.

Zwicky, A.: 1971, ‘On Reported Speech’, in C. Fillmore and T. Langendoen (eds.), Studies in Linguistic Semantics, Holt, Reinhart, and Winston, New York, pp. 72-77.




**Thanks to Bruce Fraser, Mike Harnish, Jeff King, David Sosa, and two Linguistics and Philosophy reviewers for their very helpful comments and criticisms, most of which I heeded.


[1] Here and throughout I use the phrases ‘speech act’ and ‘utterance’ to include language acts by whatever means, not just oral ones.

[2] This does not make them metalinguistic, unless, like ‘in other words’ or ‘in short,’ they pertain to the wording of the utterance, not just to the act of making it.

[3] It is ironic that this is how Karttunen and Peters, who popularized the notion of conventional implicature among linguists, felt about semantic presupposition. They thought that this was a spurious category and that alleged instances of it comprise “a wide range of different things have been lumped together under this single label. ... we propose to do the sensible thing and, namely to divide up this heterogeneous collection and to put the particular cases into other categories of phenomena” (1979, p. 2). The irony is that conventional implicature was one of these categories.

        Semantic presupposition had already been debunked by Grice (1989, ch. 17, which had circulated in the seventies) and others, most thoroughly by Boër and Lycan (1976) in “The Myth of Semantic Presupposition”, to which more than the title of the present paper is indebted. They argue, after examining a wide range of expressions and constructions, that the intuitions thought to support claims of semantic presupposition are really intuitions about pragmatic presupposition. Whereas a semantic presupposition of a sentence is a necessary condition for its having a truth value, a pragmatic presupposition is not a property of the sentence at all but of its utterance—it is a condition on the successful and felicitious use of the sentence. Pragmatic presuppositions are defeasible, contextually variable, unprojectible, and in some cases cancelable.

[4] Strictly speaking, this should be qualified to allow for locutions that do both. For example, ‘but’ not only produces a conventional implicature (according to the CI-thesis) but also makes a conjunctive contribution to what is said.

[5] As Grice explains, he does “not feel compelled to accept the hypothetical ‘If she was poor but honest, then there is some contrast between poverty and honesty, or between her poverty and her honesty’.”

[6] They have been described in this way by Karttunen and Peters (1979, p. 12), Levinson (1983, p. 127), and Rieber (1997, p. 51), among many others. This description is more plausibly applied to certain constructions, such as clefting, passivization, and topicalization, which play “information-packaging” roles (see Lambrecht 1994 for a book-length treatment of these and other such constructions). Such constructions used to be thought to generate semantic presuppositions, but Boër and Lycan forcefully argued that if they give rise to anything semantic, it is entailments, not presuppositions (1976, pp. 25-28).

[7] Nor should it be supposed that any term with non-truth-conditional meaning, such as ‘oh’ and ‘well,’ ipso facto generates conventional implicatures. Such terms do pose the problem of explaining what it is for terms to have non-truth-conditional meaning. The case of utterance modifiers will be taken up in §5, but there are other interesting phenomena that will not be taken up here, involving what Frege called “coloring”, or what is popularly known as “connotation”, and contrasts like ‘policeman’ vs. ‘cop’, ‘essen’ vs, ‘fressen’ (in German), and ‘vous’ vs. ‘tu’ (in French). Now it might be argued, as Levinson (1983, p. 128-129) does with the familiar vs. the formal second-person singular pronoun (in languages like French or German, where the difference is marked), that using the second member of each pair rather than the first must, since it does not affect what is said, produce a conventional implicature. In my view, however, it does not do that either. Insofar as which term one uses in each pair is a matter of appropriateness (for whatever reason—it is different in different cases), using one term rather than the other indicates that the condition for its appropriate use has been met. However, this is not something one specifically communicates, much less conventionally implicates. In general, after all, utterances do not communicate that the conditions for their appropriate performance have been met. Implicature does occur when one uses one term when the other is appropriate, e.g. uses ‘fressen’ rather than ‘essen’ to describe a person’s eating, but the implicature here is conversational. Similarly, although using ‘tu’ rather than ‘vous’ normally does not communicate that one is speaking to an intimate (or an inferior), if one switches from using ‘vous’ to ‘tu’ in addressing someone, the switch conversationally implicates a change in the status of the relationship.

[8] Karttunen and Peters say here that “presuppositions of cleft and pseudocleft constructions also seem to be genuine examples of conventional implicature”, but, as noted above (note 7), Lambrecht (1994) has given a plausible account of them as information-packaging devices.

[9] There are several common examples that I have left off this list, including ‘furthermore’, ‘in addition’, and ‘moreover’. For reasons that will become evident later, they will appear under the heading of “additives” in the taxonomy of utterance modifiers in §5.

[10] This is not to suggest that the utterance is of an elliptical sentence. Such a suggestion, whether taken syntactically or semantically, is soundly refuted by Stainton (1995).

[11] Of course if there is an ambiguity in the elliptical utterance, as in (i),

            (i)      Bill Gates knows a richer man than Ross Perot. [is/knows]

the determination of what is said does require contextual information, but only in the way that any sort of disambiguation requires contextual information.

[12] Whether or not one adopts Grice’s strict conception of what is said, one can distinguish different degrees of explicitness in an utterance (Bach 1987, pp. 69-74). Correspondingly, reports of what someone says can be less than faithful and complete and still be regarded as correct. As Zwicky (1971) and Ziff (1972) have shown with the help of a rich variety of examples, what counts, for contextual purposes, as an accurate report of what someone says is subject to variable standards of fidelity and completeness. For present purposes it is necessary to maintain a strict standard on what is said. This standard can be maintained on the supposition that what is said, the proposition expressed by a sentence, is a structured proposition whose constituents and form mirror that of the sentence itself. See King 1995 for a lucid exposition of this view.

[13] Worth noting here are two types of complete sentence whose utterance does not make completely explicit the proposition being conveyed by the speaker. In the first case, illustrated by (i) and (ii),

            (i)      Jack has finished.

            (ii)    Jill is late.

the sentence is syntactically complete (well-formed) but semantically incomplete—it does not express a complete proposition (even after references are fixed). Utterances of (i) and (ii) do not specify what Jack has finished or what Jill is late for. In order for what is being conveyed to be understood, understanding such utterances require what I call the pragmatic process of completion (Bach 1994). In other cases, like (iii) and (iv),

            (iii)   I haven’t had breakfast [today].

            (iv)   I hope everyone [in my audience] has a handout.

the sentence does express a complete proposition, but part (in brackets) of the proposition the speaker is conveying is not made explicit. Understanding utterances of this sort requires a process of what I call expansion.

[14] Grice acknowledges that his “favored sense of say” is somewhat stipulative, as is his contrast between what is said and what is conventionally implicated. He writes,

I would wish to maintain that the semantic function of the word ‘therefore’ is to enable a speaker to indicate, though not to say, that a certain consequence holds. Mutatis mutandis, I would adopt the same position with regard to “but” and “moreover”. My primary reason for opting for this particular sense of “say” is that I expect it to be of greater theoretical utility than some other sense of “say” would be. (1989, 121)

Grice does not explain how we are supposed to generalize from his very small set of examples (‘therefore’, ‘but’, and ‘moreover’), and he does he not say what the theoretical utility is of his sense of ‘say’. If anything the category of conventional implicature just complicates Grice’s account of the relation between saying and meaning. Because of the alleged existence of conventional implicature, he cannot define saying that p in terms of uttering something that means ‘p’ (Grice 1989, p. 88). So if there were no such thing as conventional implicature, he could draw his distinction between what is said and what is conversationally implicated much more neatly than in fact he does.

[15] This has not kept Sperber and Wilson and their followers from calling these inexplicit elements ‘explicit’ and from using the neologism ‘explicature’ to describe the pragmatic process required for their identification. I prefer the more accurate neologism, ‘impliciture’ (Bach 1994).

[16] Rieber very briefly discusses the “difficult case” of indirect quotation. He tentatively suggests that ‘even’ (the example he uses) “is here being used metalinguistically” (1997, p. 57), but offers no rationale for this suggestion. Also, he claims that the use of ‘but’ in indirect quotation, as in (1IQ), more naturally indicates a contrast being made by the reporter than by the person being quoted, but it can be just as natural to take the contrast as part of what is being reported. Which is more natural in a specific case depends on contextual factors.

[17] Note that it would not support the CI-thesis to concede that ACIDs do contribute to what is said (since they can occur in indirect quotation) but argue that their special contribution does not concern the truth conditions of what is said. For the CI-thesis claims that these terms generate conventional implicatures, and conventional implicatures are true or false. Besides, it would just beg the question, just as relying on detachability as a test would beg the question, to insist that these propositions are not part of what is said.

[18] In written language, quotation marks would be used for the directly quoted material, as in (i),

            (i)            Monica said that Linda was “a traitor,” not a friend.

There are no oral quotation marks, but one can say “quote ... end quote” or use “finger quotes”. Mixed quotation is discussed at length in Cappelen and Lepore 1997.

[19] Grice called them “noncentral” (1989, p. 122) or “higher-order” (1989, p. 362) speech acts (Levinson calls them “discourse deictics” (1983, p. 128) and evidently endorses Grice’s view that they generate conventional implicatures). I prefer ‘second-order’ to ‘higher-order’ because I see no need to allow for the merely theoretical possibility of orders of speech acts higher than second.

[20] Rieber’s criticism of an earlier version of my account, as applied to ‘although’ and ‘nevertheless’, overlooks the distinction between discourse and content connectives. His examples (14 and 15 on p. 71 of Rieber 1997) are not on point,

            (i)      Although it’s sunny, it’s quite cool.

            (ii)    Buying lottery tickets may be irrational; nevertheless millions of people do so.

because ‘although’ and ‘nevertheless’ are used here not as discourse but as content connectives.

[21] One consideration that I will not take into account (it would only help my case) is the fact, noted earlier (note 13), that what counts as an accurate and faithful report of what someone says is subject to standards that can vary widely from case to case. Judgments about what is or is not part of what is said are bound to be slippery on this basis alone, quite apart from questions about conventional implicature.

[22] It is “natural and harmless”, they add, to regard such information as presupposed rather than asserted, but this is only in the innocuous, pragmatic sense of ‘presupposition’.

[23] Part of the confusion here is that intuitions about what is said tend to conflate what is said with what is asserted, which may include less than the full propositional content of the utterance. What is asserted is the content of an illocutionary act, not the locutionary act of saying. What is said comprises the full content of the locutionary act (the semantic content of the utterance), but that may include more than what intuitively is taken to be asserted. This is evident if we consider how denials of assertions are understood. If someone disputes (1), for example, by saying, “No, that’s not true” or “I disagree”, he would be taken to be disputing one of the conjuncts. But he can dispute the alleged contrast as well. It’s just that if he does not make this explicit, he will be misunderstood. He could make it explicit by saying, “Shaq is huge and agile all right, but as a seven-footer I can assure you that he is not huge BUT agile”. This fact about denials is symptomatic of the fact, as we will see next, that intuitions about truth or falsity are not sensitive to the entire semantic content of utterances containing ACIDs.

[24] This may be connected to the claim made by several linguists (cited in Fabb 1990) that nonrestrictive relative clauses are not constituents of their host sentences. Note, however, that the longer or more important the material between the commas, the more one is inclined to regard it as relevant to the truth or falsity of the entire utterance. Compare (20) and (21) with (20') and (21'):

            (20') Ann's computer, which is essential to her work, crashes frequently.

            (21') Beth’s husband, a doer of many household tasks, never washes the dishes.

Here the material between the commas is important enough to count toward intuitive assessments of truth or falsity. When it is that important, its failure to hold true seems sufficient for the falsity of the entire utterance. If Ann’s computer is not essential to her work, (20') seems false. If Beth’s husband does not do many household tasks, (21') seems false. Also, compare (20) and (21) with the result of interchanging the main and the secondary contents:

            (20)  Ann’s computer, which crashes frequently, she bought in 1992.

            (21)  Beth’s husband, who never washes the dishes, is a plumber.

For detailed discussion of such phenomena see Boër and Lycan 1976, pp. 18ff.

[25] More specifically, the contrast is that being huge tends to preclude being agile, but, as we saw earlier, the specific contrast being conveyed is not determined by the conventional meaning of ‘but’, and so cannot be implicated conventionally.

[26] The situation is rather like reporting (i) with (iEIQ) rather than (iIQ),

            (i)            I know that grass is green.

            (iEIQ)     Max said that he has a justified true belief that grass is green.

            (iIQ)        Max said that he knows that grass is green.

even if knowledge is justified true belief. By introducing extra structure, (iEIQ) is not faithful to the syntax of what is being reported.

[27] This does not apply to the nontemporal use of ‘still’ in ‘even if’ conditionals. These are discussed in Barker 1993.

[28] For example, Bellert has suggested that with subject-oriented adverbs like ‘cleverly’ and ‘wise’, there are “two propositions asserted in one sentence” (1977, p. 340). The rejection of OSOP in connection with nonrestrictive relative clauses is implicit in Fabb’s claim, shared by several authors he cites, that an “NRR is not syntactically related as a constituent to the sentence which contains it” (1990, p. 75). And rejection of OSOP is evident in Espinal’s (1991) discussion of various other kinds of “parenthetical” or “disjunct” constituents. OSOP is also challenged by Stephen Neale, who, in a work in progress entitled “Coloring and Composition”, applies the multiple-proposition view to various problematic constructions of interest to philosophers.

[29] As Francescotti (1995, p. 156) and others have noticed about such cases, (10c) is too strong: (10) could be true even there are people under consideration who are less likely than Bill to like Mary. See Francescotti for a review of the literature on ‘even’.

[30] Karttunen and Peters also point out that “challenging [conventional implicatures] necessitates a digression away from what was actually said. It brings about a disruption in the flow of the discourse” (1979, p. 14). They illustrate this point by contrasting different ways of disputing an utterance of (5), or of answering its interrogative counterpart:

        (5?)       Does even Bill like Mary?

Responding “No” indicates a rejection of (5a), whereas something more complicated is required if one is disputing (5b) or (5c). But again, it is only by assuming one sentence, one proposition that it can be inferred that (5b) and (5c) are mere conventional implicatures (or mere entailments, for that matter). Karttunen and Peter’s observations are accurate, but their argument that the propositions generated by ACIDs are conventional implicatures is invalid. They draw this conclusion on the grounds that since the propositions in question are invariant under various operations, such as embedding, denying, and questioning, they are not part of what is said. However, this argument requires the assumption of one sentence, one proposition. Abandoning this assumption leaves open, of course, the task of explaining why these propositions are invariant under the various operations.

[31] While expressing my preference for a propositional over a truth-conditional semantics, I acknowledge that there is disagreement as to what propositions are (I prefer a structured proposition conception over a possible-worlds approach) and that the very idea of propositions is in some circles still controversial.

[32] Notice that these locutions are not explicit performatives. Even those that describe a type of speech act are not vehicles for performing a speech act of that type. For example, ‘to change the subject’ does not change the subject, ‘in addition’ does not add to what was just said, and ‘in other words’ does not reformulate what was previously said. In each case, it is the main utterance that does this. The utterance modifier is used to perform a second-order act of commenting on the main utterance.

[33] For the category of mitigatives, as well as for a number of sample locutions, I am indebted to Fraser’s (1996) taxonomy of what he calls “pragmatic markers”.

[34] It is clear just from the case of grammatical mood that semantics must recognize indicators of first-order illocutionary force. For discussion see Harnish 1994. However, in my view performative verbs do not encode illocutionary forces (Bach and Harnish 1979, pp. 203-209; Bach and Harnish 1992).

[35]This point applies also to a reviewer’s suggestion that the semantics of ACIDs is use-conditional rather than truth-conditional: “Suppose there is in our language a convention like the following: ‘Use but only if there is a contrast of a certain type between the two conjuncts’.” However, analogous rules could be formulated for any clearly descriptive term, e.g., for using the word ‘rose’ as: “Call something ‘a rose’ only if (you think) it is a rose”.

[36] Rieber also questions their analogy of utterance modifiers to indexicals, pointing out that

indexicals do have conceptual content. For example, understanding the word ‘you’ requires knowing that it is to be used to refer to the addressee(s). The point of Wilson and Sperber’s analogy is that this conceptual content does not enter into the content of an utterance containing the expression. If I say, “You are wonderful”, I am using ‘you’ to refer to whoever I am addressing, but I am not saying that the person I am addressing is wonderful. What I am saying could be true even if I were not saying anything, in which case there would be no one I am addressing, provided that the person I am in fact addressing is wonderful. The point of saying that the semantic import of an indexical is nonconceptual is not that understanding an indexical does not involve using concepts, but that such concepts do not enter into the content of the utterance containing the indexical. The referent of the indexical is what enters in, not the means by which the referent is determined. See Kaplan 1977.

[37] Rieber’s synonymy claim is puzzling in any case. Although the exclamations ‘Ouch!’ expresses pain and ‘Yuck!’ expresses disgust, no one would claim that ‘Ouch!’ is synonymous with ‘That hurts!’ or that ‘Yuck!’ is synonymous with ‘That’s disgusting!’. Rieber thinks his “analysis preverves the logical form of the original sentences” and “gets the truth conditions right” (1997, 54), but it doesn’t even get the grammatical form right, much less the truth conditions. If words like ‘but’, ‘so’, ‘still’, and ‘even’ are tacit performatives, then, like an explicit performative such as ‘I suggest …’, they ought to be modifiable by adverbs and be capable of being hedged. But whereas forms like those in (i) are fine,

            (i)      I cautiously suggest, I strongly suggest, I might suggest, I would like to suggest

words like ‘but’ and ‘still’ are not similarly modifiable. It has hard to see how conjunctions or adverbs could be semantically equivalent to such syntactically different expressions as performative sentences.

[38] The situation is like that produced by (i), the standard illustration of Moore’s paradox,

            (i)      It is raining but I don’t believe it.

where the speaker asserts that it is raining and then denies the condition for the sincerity of that assertion.