Types of implicatures and their diagnostics

↵ Back to class homepage

Grice distinguished a few types of implicatures. Most of these categories are problematic, and the distinctions between them remain controversial, but it's still useful to review them to become aware of relevant pragmatic issues (are you noticing a theme yet in this class?). Also, these terms and concepts are still commonly used, so you need to know what they are if you want to understand people talking about pragmatics.

Let's look at a few different examples:

  1. C'est anglais, mais c'est bon ! ("They're English, but they're good!")
  2. Sammy and Chris fell in love and got married.
  3. Did you win the lottery and not tell me? (uttered in the Breaking Bad scene we discussed in the previous module)

When (1) suggests that English stuff is usually not good, that's an example of what Grice calls conventional implicature. On the other hand, the things suggested by (2) and (3) are both examples of what Grice calls conversational implicature, but they're different types. The suggestion in (2), that Sammy and Chris fell in love first and got married second, is what Grice calls a generalized conversational implicature. On the other hand, the suggestion in (3), that Walt and Skyler don't have enough money for a fancy house, is what Grice calls a particularized conversational implicature. Let's break down what these distinctions mean.

Conventional implicatures

These are implicatures which, according to Grice, don't come from context or even from reasoning about speaker's beliefs and intentions, but come from the way language is conventionally used. e.g., as speakers of English we know that "but" is used in a way that implies some sort of contrast, even though it doesn't literally mean that.

Many conventional implicatures feel, at first blush, like regular old literal meaning. For example, one of Grice's examples is "The queen is English and therefore brave"; he argues that this utterance implies, but doesn't literally say, that the quality of brave-ness follows from the quality of English-ness, i.e., all English people are brave. Intuitively, it seems like the second part (i.e. the idea that all English people are brave) just comes from the literal "meaning" of therefore. Nevertheless, many pragmaticists argue that, even though these implicatures come from the conventional usages of certain words (like but and therefore), the implicatures themselves are things that are only suggested by the utterance, rather than explicitly said; at the end of this module we will review some evidence for this. Anyway, the conventional implicature associated with "but" seems to be the one that holds up the best, and thus whenever people talk about conventional implicatures they almost always use this example. Another pretty good example is "even"; if I say "Even John likes this pizza", it seems to imply (without literally saying) that John is someone who is picky or who normally doesn't like pizza.

Of all the types of implicatures, this is the trickiest category, because it doesn't differ obviously from semantic meaning in the way the other sorts of implicature do. Some people who study pragmatics and semantics doubt the distinction between "conventional implicature" and truth-conditional meaning at all. Some also say Grice shouldn't have called them "conventional implicatures" at all, because even if they are different from semantic, truth-conditional meaning, they don't really seem like "implicatures", in that they don't require reasoning about speakers' intentions and the Cooperative Principle. Grice calls them implicatures anyway, though, because they are not part of the semantic (truth-conditional) meaning; for Grice, meaning is divided into "what is said" (which comes from semantics and is truth-conditional) and everything else—everything that doesn't come from semantics and is not truth-conditional is an "implicature".

No matter what conventional implicatures are, they do seem to raise a challenge to the idea of a clear-cut distinction between semantics and pragmatics, as they sort of seem to have some properties of each (although see Zufferey et al., chapter 5, for an explanation of how this conundrum can be solved while maintaining a semantics-pragmatics distinction). But, as mentioned above (and as we will see more concretely below), conventional implicatures are different from semantic meaning—and similar to pragmatic meaning and implicatures—in a very important way: they aren't truth-conditional. This is the main reason that traditional pragmaticists group conventional implicatures together with conversational implicatures (i.e., as part of pragmatics) rather than together with semantic meaning (see e.g. Feng, 2010, chapter 2, for more discussion of this point).

Conversational implicatures

Conversational implicatures are the "real" implicatures, which come from reasoning about what a speaker must have intended. Most of the examples we've seen, and most of the rest of this class, will be focused on conversational implicatures.

As we saw above, though, these can be divided into "generalized" and "particularized" conversational implicatures. Grice himself didn't make a big deal of this distinction (he pretty much just mentioned it in passing and then didn't talk about it again), but many pragmatics scholars since then have been crazy about it.

"Generalized" conversational implicatures are ones that don't depend on context; uttering certain things will (supposedly) always trigger these implicatures. Looking at the example (2) above, it seems reasonable to assume that "X happened and Y happened" always implies "X happened before Y", regardless of the context in which it is uttered.

You might notice now that generalized conversational implicatures sound similar to conventional implicatures—both seem intimately connected to certain words or phrases, and neither depends on the context. I've mixed these up myself before. Towards the end of this module we'll look at how they are different and why this difference is important.

"Particularized" conversational implicatures are much more straightforward. They depend on the context and on reasoning about the speaker, so they only arise in certain contexts. Consider example (3) above. "Did you win the lottery and not tell me?" implies "We don't have enough money for the house you want" in the context of this particular conversation, but in another context it would not imply that (hence why I had to specify the context when I shared the example).

Many pragmaticists, from a variety of theoretical perspectives, reject the division between "generalized" and "particularized" conversational implicatures, and instead argue that all conversational implicatures are "particularized" and are derived via the same mechanisms (see, e.g., Katsos & Cummins, 2010, for review). For our purposes in this class, however, there is one practical way this distinction will be useful: for clarifying the difference between conventional and conversational implicatures. If we compare a conventional implicature like (1) to a particularized conversational implicature like (3), the difference is so obvious that we don't really need any special ways to notice the difference. But comparing (1) and (2) is much trickier; how can we figure out that (1) is a "conventional implicature" and (2) is a "conversational implicature"? The difference here is not so intuitive, so we will need some special techniques to recognize the difference. Let's turn to those now.

Diagnostics for implicatures

A "diagnostic" is a straightforward test for something (e.g., if you think you have the flu, you go to a doctor and they administer some kind of test which will tell you whether or not you have the flu). Pragmaticists have thought of some good diagnostics for recognizing different kinds of implicatures.

Distinguishing conversational implicatures from conventional implicatures and truth-conditional meaning: cancellability

The most important and most useful diagnostic is cancellability: testing whether an implicature could be cancelled (or prevented from arising in the first place). A conversational implicature can be cancelled without contradicting oneself. For example, we can explicitly say that the implicature in (2) is false, but our utterance won't sound weird or self-contradictory: "Sammy and Chris fell in love and got married. But not in that order!" Likewise for the one in (3): "Did you win the lottery and not tell me? I'm not suggesting that we don't have enough money; I'm just literally asking because I found this lottery card." Maybe these might sound a bit funny, but they certainly sound more acceptable than what happens if we try to cancel an utterance's literal, truth-conditional meaning; for example, "Sammy and Chris fell in love and got married, but they didn't get married" is self-contradictory and paradoxical.

Importantly, only conversational implicatures are cancellable. Conventional implicatures are not. Let's look again at example (1), where a box of British cookies (McVitie's) being sold in France had a label on them that says, "They're English, but they're good!" This conventionally implies that English food is not usually good (and this is why people find this box of cookies hilarious). Trying to cancel this conventional implicature sounds pretty weird: "They're English, but they're good—I'm not trying to suggest that English stuff is usually bad, though." To me, this sounds much worse than the cases where we cancelled conversational implicatures; it sounds more like the case where we cancelled literal meaning.

Something very related to the cancellability test is to try uttering something in a situation where its implicature is not true. For example, if the reality is that Sammy and Chris got married first and then later fell in love (maybe they had an arranged marriage, or a green-card marriage, or something like that?), and you say "Sammy and Chris fell in love and got married", I feel like this sounds misleading, but not necessarily false. This is because, again, the conversational implicature of this utterance (the suggestion that they fell in love first and got married second) is not part of its literal, truth-conditional meaning. On the other hand, if Sammy and Chris haven't gotten married at all, then uttering "Sammy and Chris fell in love and got married" would be downright false, not just misleading.

Here again we can contrast conventional implicatures with conversational implicatures. How does it sound if I say "They're English, but they're good" in a situation where English food really is good? Perhaps this is hard to imagine, since English food is famously terrible and even English people think so. Maybe we can use a more realistic example: for instance, Chongqing food is famous for being spicy, so how would it sound if I said "This hotpot is Chongqing style, but it's spicy!" Here I have just uttered a sentence in a situation where its conventional implicature (the suggestion that Chongqing food is not normally spicy) is obviously false. To me, this sentence sounds very wrong... not just misleading, but infelicitous and somehow ill-formed. Perhaps it's not as obviously bad as "Sammy and Chris fell in love and got married" when they haven't gotten married, but it's certainly far worse—and worse in a different way—than "Sammy and Chris fell in love and got married" when they actually got married first and fell in love second. So here we see another example of how we can distinguish between conventional and conversational implicatures. Conversational implicatures are cancellable, and when an utterance's conversational implicature is false the utterance itself is still true and well-formed; conventional implicatures have neither of those features.

(The cancellability test is not without problems. Levinson (chapter 3.2.1) goes into some detail on problems with cancellability and other diagnostics. But for our purposes in this class, the cancellability test is a good start.)

Distinguishing conventional implicatures from truth-conditional meaning

Cancellability can help us tell the difference betwen conversational and conventional implicatures, but how can we tell the difference between conventional implicatures and literal, truth-conditional meaning?

One difference between them is intuitive: in the examples we saw related to cancellability above, there do seem to be subtle differences between conventional implicatures and literal meaning. Saying something when its literal meaning is false makes a sentence completely false, whereas saying something when its conventional implicature is false makes a sentence weird, infelicitous, or somehow ill-formed. If you say something whose literal meaning is false, then people might correct you or say that you're wrong (e.g., if you point at a Korean dish and say "This is Chongqing food but it's spicy!", people might say, "It's not Chongqing food, it's Korean food!"). On the other hand, if you say something whose conventional implicature is false, then people are more likely to tell you that what you said doesn't make sense (if you point at some Chongqing hotpot and say "This is Chongqing but it's spicy!", people will be more likely to say, "Why did you say but? Chongqing food is always spicy!"). In other words, if you say something whose literal meaning is false, then people will object to what you said, but if you say something whose conventional implicature is false, then people will object to the way you said it. This is one of the key differences between conventional implicatures and literal meaning (although it seems pretty subjective).

This distinction also works for other conventional implicature examples we saw above. For example, if the queen really is British (this is certainly true) and really is brave (I don't know if this is true and in fact I suspect it is not [since the only appropriate and brave thing for a monarch to do in the 21st century is to abolish the monarchy, and she has not yet done that], but let's pretend for now that it is true), but it's not true that British people are necessarily brave, then "The queen is British, and therefore she is brave" sounds weird but not necessarily false. Likewise, saying "Even Jeff Bezos can afford that house" sounds wrong (because "even X can Y" implies that X is someone/something who would normally be unlikely to Y, whereas in reality we know that Jeff Bezos is an evil rich supervillian so it's not at all unlikely that he could afford some house), but doesn't sound false.

Utterances vs. meanings

One final thing I must emphasize is that sentences or utterances themselves are not implicatures. An utterance has multiple meanings, and some of those meanings might be implicatures. For example, it makes no sense to say that example #1 from the beginning of this module is a conventional implicature or that example #2 is a generalized conventional implicature. Instead, example #1 is an utterance which means several things at once: it means that the cookies are English and they're good (its semantic meaning) and, at the same time, it also means that English cookies are usually bad (a conventional implicature). Likewise, example #2 is an utterance which means that Sammy and Chris fell in love (part of its literal meaning), Sammy and Chris got married (also part of its literal meaning), and might also mean that their falling in love happened first and their getting married happened second (a generalized conversational implicature). These meanings are all there at the same time. It doesn't make sense to try something like the cancellability test, or any other diagnostic, on a sentence or utterance; you try this test on one of its meanings, which means you need to clearly specify which meaning you are testing.

To illustrate the importance of talking about meanings instead of sentences or utterances, and the fact that one utterance can have a bunch of meanings at the same time, let's wrap up by considering the two graphs below. First, here's a graph of the traditional Gricean way of dividing up the meaning of an utterance into its parts. Meaning is divided into semantic meaning ("what is said") and implicatures (everything else). Implicatures can be further broken down into conventional and conversational implicatures, and conversational implicatures can be broken down into generalized and particularized conversational implicatures.

Now imagine that I went to see a movie with a few other people who both you and I are friends with, but you weren't able to come to the movie with us. Later you chat with me and ask me how the movie was. I say, "Even John wanted to get up and leave!" This utterance means several things at once; in fact, in this context, it has every kind of meaning from the chart above. As illustrated below, it literally just means that John wanted to get up and John wanted to leave. But also, given the context, it seems to suggest that the movie was bad, and this suggestion is a particularized conversational implicature—we can infer that the reason someone would want to get up and leave is because the movie was bad. Furthermore, because of the word even, it suggests that John is somehow the least likely person to want to get up and leave from a movie (maybe he's a big fan of this type of movie, or maybe he's just very polite and thinks it's rude to leave during a movie); this is a conventional implicature. And "get up and leave" might implicate, through generalized conversational implicature, that he wants to get up first and leave second (although this part of the meaning doesn't seem very important for the overall interpretation of this particular utterance). The important thing to see here is that one utterance has all four types of meaning we've seen here, so you can't just say that this utterance "is" a conventional implicature or "is cancellable" or anything like that; you need to figure out which particular meaning within the utterance you want to analyze and then use the tests on that particular meaning to figure out which kind of meaning it is.

Summary

The chart below summarizes the differences between the four kinds of meaning we have encountered so far: semantic meaning, plus the three kinds of implicatures discussed in this module.

As we can see here, what separates semantic meaning from all kinds of implicatures is that semantic meaning is truth-conditional and implicatures are not—in other words, saying something when its semantic meaning is false makes the utterance false, whereas saying something when its implicature is false makes the utterance weird or misleading but not false.

What separates semantic meaning and conventional implicatures from conversational implicatures is that the first two are not cancellable, whereas conversational implicatures are. If I say something that has a conversational implicature and then we immediately deny that implicature (e.g., "Sammy and Chris got fell in love and got married... but not in that order"), it doesn't sound like I have contradicted myself, it just sounds like I have offered a clarification; on the other hand, if I deny the semantic meaning or conventional implicature of what I said (e.g., "That's a red apple... but it's not an apple" or "Hanako is Japanese and therefore he loves anime... but liking anime doesn't necessarily follow from being Japanese") then it sounds like I have contradicted myself.

And what separates particularized conversational implicatures from the other three types of meaning is that particularized conversational implicatures depend on the context, whereas the other types of meaning do not. "Apple" means apple no matter what context, and "but" has a conventional implicature no matter what context, but "Did you win the lottery and not tell me?" only means "we don't have enough money for the kind of house you want" in a particular kind of context.

Video summary

In-class activities

Here are a few sentences:

All of these sentences say (literally, truth-conditionally) that Hobbes loves tuna fish sandwiches. They also convey that Hobbes is a tiger.

What kind of information is this second bit (the fact that Hobbes is a tiger)? A truth-conditional entailment, a conventional implicature, or a conversational implicature?

Have students discuss this and justify their conclusions.

(Hints: in the first three examples, the claim that Hobbes is a tiger is within an appositive, and traditionally these have been thought to contribute to conventional implicatures rather than truth conditions [see, e.g., Potts, 2005]. But there has been a lot of debate over this, and some authors [e.g. Syrett & Koev, 2014] have claimed that they are truth-conditional. See Kroll & Rysling, 2019, for a review. You don't need to present all this information to the class, and they don't need to know this information in order to do the discussion; this is just background information for you. Anyway, if appositives contribute to conventional implicatures rather than to truth-conditions, then the first three constructions here should feel "less false" than the other two constructions if Hobbes really isn't a tiger [you can test this by replacing this with a sentence where the information in the appositive really is false, such as "Guangzhou, a city in the north of China, is my hometown".)

If they figure out those easily, here's some more they can try. The following three sentences each convey that not all of my cats have stripes. Again, have students figure out how that information is conveyed in each utterance (i.e., is it part of the literal meaning, a conventional implicature, a generalized conversational implicature, a particularized conversational implicature, or none of the above?).

Levinson (chapter 3.2) discusses another diagnostic for conversational implicatures, called reinforceability. The idea is that when we say something that makes a conversational implicature, we can further add something which emphasizes (reinforces) the implicature and it won't sound weird or redundant. On the other hand, other kinds of meaning supposedly don't allow this. Compare example (1) below, where the underlined portion reinforces the implicature (the interpretation of some as meaning "not all"), versus example (2), where the underlined portion reinforces the literal meaning (the interpretation of some as meaning "more than zero"); (1) sounds acceptable and (2) does not:

  1. Some of the boys went to the soccer match, but not all.
  2. Some of the boys went to the soccer match, but not none.

Have students discuss the reinforcability test. Is it useful for distinguishing between conversational implicatures and other types of meaning? Are there other, better ways to explain the contrast between (1) and (2) above? Are conversational implicatures always reinforceable, and are other meanings always not reinforceable?

A few points of my own I've noticed, which may or may not come up during the discussion:

Given examples like these, it seems like we need an explanation for just what sorts of things count for the "reinforceability" test. For example maybe we can explain away the presupposition examples by saying that the reinforcement here is unusual and marked, and the speaker is intentionally saying something unusual to emphasize a point (that is, indeed, one of the main uses of pragmatics), and the very fact that it is unusual and attention-grabbing is evidence that this sort of reinforcement is not normal. For the entailment example, we could explain it using mutual knowledge; implicatures arise from mutual knowledge (i.e. when both the speaker and hearer know something, and know that each other know it) and the hearer is able to use the Cooperative Principle and the mutual knowledge to reason about what the speaker must have meant; clearly for the examples I gave, something is not mutual knowledge (since it needed to be explained to the hearer), and that explains why it is not an implicature, although this still doesn't explain how the "reinforceability" test can be useful for distinguishing implicatures and non-implicatures.

In the module we saw how conventional implicature is different from semantic (truth-conditional) meaning because conventional implicatures can't make an utterance false, just weird. We noted that this distinction is a bit subjective.

Another diagnostic for conventional implicatures, which is a bit more complicated but might be a bit less subjective, is that a conventional implicature can be separated from truth-conditional meaning in a way that truth-conditional content itself cannot. As Geurts (chapter 1) points out, a sentence like "That nurse is new but she's very skilled" conventionally implies there is some contrast between being new and being skilled (i.e., people who are new to some profession are usually not skilled at that profession yet), but in a sentence like "Half of the nurses here are new but skilled", the implicature does not seem to be part of what "half of" is describing. In other words, this sentence seems to be saying that, out of all the nurses, half (but not all) of them are new-and-skilled; it does not seem to be saying that the contrast between being new and being skilled only exists for half of the nurses but not for all of them. We can see the same sort of separation with therefore. "Half of the nurses are experienced and therefore skilled" does not seem to suggest that the connection between being experienced and being skilled only exists for half of the nurses; rather, the idea seems to be that there is always a connection between being experienced and being skilled. (i.e., the sentence does not mean that half of the nurses are experienced-therefore-skilled while the other half of the nurses are skilled-in-spite-of-being-experienced!)

The examples above are related to a kind of behaviour called "projection", where some part of meaning within one clause can apply outside of that clause. For example, consider a sentence like "John thinks that's a black dog, but I don't". This sentence has an embedded clause, "that's a black dog", and the literal semantic meaning of this clause says two things: (1) that's a dog, and (2) that's black. When I say "I don't think that's a black dog", the negative part ("I don't think") could be negating either part of that meaning in the embedded clause: maybe I don't think that's a black dog because I don't think it's black, or maybe I don't think it's a black dog because I don't think it's a dog. This is a property of how semantic meaning works. Conventional implicatures, on the other hand, don't work in this way. Think of a sentence with a conventional implicature in an embedded clause, like "John thinks that nurse is new but skilled, but I don't". The embedded clause ("that nurse is new but skilled") has three parts to its meaning: (1) that nurse is new, (2) that nurse is skilled, and (3) new nurses usually aren't skilled. (1) and (2) are part of the semantic meaning (entailments), and (3) is a conventional implicature. Importantly, only (1) or (2) are negated in this example. If I say "I don't think that nurse is new but skilled", I might have said that because I don't think the nurse is new (negating meaning #1) or because I don't think the nurse is skilled (negating meaning #2); but it doesn't seem natural for me to say "I don't think that nurse is new but skilled" to mean that I think the nurse is new and I think the nurse is skilled but I don't think there's a contrast between being new and being skilled. In other words, meaning #3, the conventional implicature, can't be cancelled by the negation. That is to say, conventional implicatures can be separated from semantic meaning, just as we saw in the examples in the previous paragraph.

Here we've looked at this separation with but and therefore. Can you illustrate this distinction with other conventional implicature triggers, like even?


⟵ Violating maxims
One utterance, multiple implicatures ⟶

by Stephen Politzer-Ahles. Last modified on 2022-03-05. CC-BY-4.0.