Pragmatics

The following is a tutorial essay for Philosophy of Logic and Language with James Kirkpatrick. In effect, it is a quick agenda for an hour-long discussion on a guiding question, which for this week was ‘Do conventional implicatures exist? If so, what is the difference between conversational implicature and conventional implicature? If not, why not?’ A free-standing and much more readable version of some of the material may eventually go up on my Substack.

I. Conventional implicatures

An implicature of an utterance is any proposition to which the utterance commits its speaker, but which may not be true even when the utterance is true. Thus, the content of an utterance is not an implicature: the utterance is true if only if the content is true. Similarly, a presupposition of an utterance is not an implicature: the utterance is true only if its presuppositions are all true.

An implicature of an utterance is cancellable just in case the utterance could be made without committing to that implicature. An implicature is conventional just in case it is not cancellable—that is, a conventional implicature of an utterance is one that the utterance has necessarily. An implicature is conversational just in case it must be calculated pragmatically.

For instance, consider the sentence (1).

(1) Even Ann has stopped smoking, for now.

An utterance of (1) is true just in case Ann has stopped smoking for now, and so has that proposition as its content. It presupposes that Ann smoked at some point in the past; and indeed, this is not an implicature, because it cannot be true that Ann has stopped smoking for now unless it is true that Ann smoked at some point in the past.

A conversational implicature of (1) is that the speaker does not know that Ann will not resume smoking: if the speaker did know that Ann would not resume smoking, they would have left off the ‘for now’, so the speaker is committed to this; however, this implicature is cancellable: the speaker may continue with ‘Of course, she won’t resume—it’s just that we’re interested in current smokers’ and so remove such an implicature. (And, since the result could be true, the original implicature indeed could be false while (1) is true.)

Finally, a conventional implicature of (1) is that Ann was relatively unlikely to stop smoking. Although (1) could be true without this being true, the speaker is committed to its being true. But the implicature is not easily cancellable: continuing with ‘Of course, she was the most likely to stop—we’re just interested in whether someone in fact stopped smoking’ would be perplexing.

II. Against error theories against conventional implicatures

If verdicts like those above about cases like (1) are correct, then there are conventional implicatures. Of course, one could always deny such data, perhaps by trying to explain why we reach such verdicts despite their inaccuracy. But the default assumption should be that the verdicts are more or less accurate, especially since the category of conventional implicatures is built from categories that we should robustly accept. In the absence of a special reason for thinking that this category is empty, or that our intuitions are especially unreliable here, the default presumption should be in favor of our intuitions.

For instance, the error theories of Bach (1999) seem to overgenerate. He lists four reasons for distrusting our intuitions about there being conventional implicatures—in particular, for thinking that uncancellable commitments need not be entailed by the content of the utterance: first, that conventional implicature can vary with context; second, that conventional implicature is often unemphasised and part of the common ground; third, that the judgements are artefacts of a forced choice; and, fourth, making explicit the commitment requires adding an additional clause.

Briefly, the issues are as follows. Firstly, that conventional implicatures can vary with context does not suggest that they are not necessary: it may be that, roughly, what is necessary is an implicature with the same character, but perhaps a different content. We have no trouble noticing that ‘She has stopped smoking’ entails that the subject has stopped smoking, not just that someone or other has stopped smoking, despite the fact that the contribution of ‘she’ varies with context.

Secondly, we have no trouble recognising entailments of content that are unemphasised or already in the common ground. For instance, consider (2), from Bach’s (1999) own discussion.

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

That Ann bought her computer in 1992 is unemphasised here, and may even (seem to be) part of the conversational common ground in some contexts. Nevertheless, contra Bach (1999)—and perhaps Potts (2005)—it is clear that (2) is not true if Ann’s computer was bought in 1993 rather than 1992. And indeed, one can include the aside within indirect speech reports about utterance of (2), without endorsing them.

(3) He said that Ann’s computer often crashes, and that it was from 1992. But he’s wrong: although it crashes frequently, it was from 1993.

This similarly addresses Bach’s (1999) worries about a forced choice: we often make the opposite choice, even when forced, as in judging that (2) is false if the aside is false.

Finally, his fourth worry seems to overgenerate against the presupposition in (1): although making explicit the assumption that Ann smoked at some point in the past would require an extra clause, we have no trouble recognising it as entailed by the content.

III. The theoretical role of conventional implicature

The category of conventional implicature has a productive theoretical role; beyond merely supplying verdicts about individual cases, this suggests that there indeed are conventional implicatures (since our best linguistic theories invoke the category). Two interesting cases are in the semantics of pejoratives (following Williamson 2009) and in the semantics of imperatives.

An interesting account of pejoratives—with a major problem that we’ll address shortly—is as follows. A pejorative term, like ‘clanker’ for robots, has its meaning given by the inferences that it licences. In particular, there are introduction and elimination rules given for the term.

(‘Clanker’ Introduction) From ‘X is a robot’, infer ‘X is a clanker’. (‘Clanker’ Elimination) From ‘X is a clanker’, infer ‘X is incompetent’.

Put together, the rules for introducing and eliminating ‘clanker’ licence the inference from ‘X is a robot’ to ‘X is incompetent’. Thus, any language with a term obeying such rules licences this inference. This offers an intriguing diagnosis of whatever is going on with pejorative expressions, and in particular, of their harmfulness. Call this the ‘tonk-like’ view of pejoratives, after the trivialising operator ‘tonk’.

(‘Tonk’ Introduction) From ‘A’, infer ‘A tonk B’. (‘Tonk’ Elimination) From ‘A tonk B’, infer ‘B’.

However, there are two issues with this. Firstly, it vindicates the inferences made in a language containing the problematic term; although it can criticise the languages themselves, they hold that such inferences are indeed valid in such a language. This is especially concerning given that English does have pejorative terms; if they indeed behave in the tonk-like way, then the inference from ‘X is a robot’ to ‘X is incompetent’ is valid in English. Secondly, it doesn’t seem to fully tell us what ‘clanker’ really means—that is, what the content of the term ‘clanker’ is. This feels unsatisfying, given a background in which we know the contents for most other English terms.

What we would really like is an account that explains (but does not vindicate) the problematic ‘clanker’ inferences, and gives us the (possibly empty) content of the term itself. One way of doing this is holding that the content of ‘clanker’ is just the same as the content of ‘robot’, which explains (indeed, vindicates) the ‘clanker’ introduction rule. However, we may want to add that, by using the term ‘clanker’, instead of ‘robot’, one commits oneself to thinking that robots are incompetent, and cannot disavow such a commitment; but, nevertheless, ‘X is a clanker’ may be true while ‘X is incompetent’ is not true, and so the inference is blocked. The category of conventional implicatures is exactly what is needed: ‘clanker’ carries the conventional implicature that all robots are incompetent.

Next, consider the semantics of imperatives. Both indicatives and interrogatives can function like imperatives; consider the following.

(4) You will pass the salt.

(5) Could you pass the salt?

(6) Pass the salt!

By uttering (4), (5), or (6), the speaker commits themselves to (something like) wanting the audience to pass the salt. In no case does this seem to be the content of the utterance. It certainly is not for (4), and (5) and (6) either have no content or (as argued last week) the same content as (4). In the latter case, the commitment is in each case an implicature. But in (4) and (5), it is cancellable; in (6) it is not.

(7) You will pass the salt—not that I’d like you to, I’m just predicting you will.

(8) Could you pass the salt? I wouldn’t like you to, I’m just curious.

(9) # Pass the salt! I wouldn’t like you to, but I’m imploring you to do it.

To the extent that (9) sounds acceptable, it seems to just be an artefact of our paraphrase being somewhat inadequate; in any case, it is clearly much worse than (7) or (8). This enables a diagnosis of how (6) can have the same content as (4). In particular, it is possible that the speaker’s wanting the audience to pass the salt is a conversational implicature of (4) or (5), since it is unlikely that the speaker otherwise knows that the audience will pass the salt, or that the speaker needs to ask whether the audience is capable of passing the salt. But such calculations can be rendered unnecessary if the implicature is conventionalised by the grammar: this predicts the introduction of the imperative mood, to avoid things like mistakes in calculation.

So, the existence of conventional implicature is not only a result of robust intuitions; it is also theoretically promising for a referentialist semantics, in enabling a proper account of pejoratives; and for an account of meaning which privileges propositions, in enabling an account of the imperative mood.