15.0 Questions and Answers

We owe the term ‘implicature’ to Paul Grice. He introduced a two-way division within what he called the ‘total signification of an utterance’ between ‘what is said by the sentence’ and ‘what is implied by the speaker’. So, a speaker who says ‘It is a lovely day” produces a sentence which says simply that it is a lovely day but they may thereby imply a further content, such as it is a horrible day (if the speaker is heard as being ironic) due to features of the context in which the sentence is produced. This distinction between the contents of linguistic items per se versus pragmatic, communicated contents is intuitively compelling, and it provides a way in which the project of semantics might seem feasible (as semantics no longer need account for everything a speaker might convey by her utterance), thus it is no surprise that it has been foundational in a vast sweep of theorizing in philosophy and linguistics following Grice.

Recent work in philosophy of language and linguistics has seriously muddied the waters around implicatures. Whereas once (in Grice’s heyday) it seemed reasonable to think that maybe we could limit the input of contextual features to the content of ‘what is said by a sentence’ to a handful of (fairly well-behaved) processes, like disambiguation and reference assignment, now the situation looks much more murky. I will survey the problems (such as the relativization apparently necessary for taste predicates, the hidden standards for comparative adjectives, and the completion of apparently incomplete expressions like ‘enough’ and ‘ready’) which have so exercised recent theorists.

Moves to introduce kinds of richly pragmatically enhanced but nevertheless semantically relevant contents (Sperber and Wilson’s ‘explicatures’, Bach’s ‘implicitures’, Recanati’s ‘what is said’) provide an exciting recent lens through which to ask what force Grice’s original machinery can still wield in the contemporary debate. I will also consider (and argue against) Lepore and Stone’s recent rejection of the category of ‘implicatures’ and their claim that all pragmatically enhanced content is a question of unruly imagination, suggesting instead that there remains a well-defined place for implicatures (contrasting with explicatures) in our account of communicated content.

Grice gave us a range of now classic examples of the phenomenon he thought of as implicatures – the reference-writer who states just ‘Jones has nice handwriting’ but thereby conveys that Jones lacks philosophical talent, the speaker who tells a stuck driver ‘There is a garage around the corner’ and succeeds in conveying that the garage is open and sells petrol, and so on and so forth. These examples are intuitively compelling and are crucial for understanding what implicatures are supposed to be. However, the intuitive pull of the examples has, I will suggest, served to obscure a problem at the heart of the Gricean account. For it seems that within the account are two features which are able to pull in different directions: Grice’s notion of ‘what is said by a sentence’ (the content which lies in opposition to implicatures) is defined both in terms of the kinds of contextual inputs it allows (limited to disambiguation and reference-fixing) and in terms of its role in the communicative forum (as the content which the speaker directly asserts or conveys). Implicatures, on the other hand then, can be understood either in terms of the amount/kind of pragmatic processing that goes in to their recovery or in terms of the role that content plays (i.e. as indirectly conveyed or communicated). Yet again these two conditions seem able to come apart. It is thus key to correctly analyzing the phenomenon of implicatures that we get clearer on the criterion that we are using to identify or individuate them.

The most serious question pertaining to the phenomenon of implicatures is whether they should be understood in terms of the sorts of processes involved in generating or recovering a given content or in terms of the role that content plays in a communicative setting. I’ll suggest that improving our grip on these two (potentially conflicting) elements allows us to deliver a much clearer understanding of what implicatures are and what role they should play in theorizing. A further outstanding question (gestured at but left open in the chapter) concerns the role of speaker intentions in determining utterance meaning (see Borg et al 2022).

15.1 Introduction

The term ‘implicature’ is owed to Paul Grice. In a seminal contribution to the philosophy of language and linguistics, Grice took what he called the ‘total signification of an utterance’ (i.e. the complete content someone communicates by producing a linguistic signal) and divided it in two, distinguishing between ‘what the speaker says’, on one hand, and ‘what the speaker implies’, on the other. For instance, imagine a speaker, Hashim, who says “It is a lovely day”: Hashim says simply that it is a lovely day but he may thereby imply a further content, such as it is a horrible day (if he’s being ironic). Or again, take the person who writes just “Smith has lovely handwriting” as a letter of reference for Smith’s application for a philosophy job (Grice Reference Grice1989: 33). The writer produces a sentence which says that Smith has lovely handwriting but they may thereby convey or implicate some further proposition such as Smith is not a very good philosopher.Footnote ¹ I will look more closely at how Grice defines ‘implicature’ and the opposing notion of ‘what is said’ below (Section 15.2), but intuitively the distinction between the contents expressed by linguistic items per se (what the speaker says, in Grice’s “favoured sense”) versus pragmatic, communicated contents (what the speaker implies) seems intuitively both clear and compelling. Furthermore, as I’ll argue in Section 15.1, drawing the distinction seems to really matter, as it holds out the promise of rescuing a project we might term ‘formal semantics’ from a plethora of utterly obvious counterexamples. Thus it is no surprise that Grice’s notion of implicature has become foundational in a vast sweep of theorizing in philosophy of language and linguistics.

However, recent developments have served to throw doubt on Grice’s taxonomy for the total signification of an utterance, with both sides of his divide coming under fire. So some have questioned the existence of implicatures (Lepore & Stone Reference Lepore and Stone2014), while others have questioned how he individuated his category of ‘what is said’ (e.g. Sperber & Wilson Reference Sperber and Wilson1986, Recanati Reference Recanati2004, amongst many others). These challenges to Grice’s framework will be examined in Section 15.3. However, I will argue that, on closer inspection, the recent developments canvassed do not serve to show that Grice’s notion of implicature is ill-founded, nor that his “favoured sense” of what is said is unnecessary. What they do serve to highlight is a peculiar tension in Grice’s original account. For it seems that Grice merged two distinct features when defining what the speaker says versus what the speaker implicates: the idea of a content dictated by word meaning and structure alone, on the one hand, versus the idea of an asserted or directly expressed proposition on the other. What the challenges to Grice’s framework reveal is that these features can and often do come apart. Yet once we resolve this tension, I’ll suggest, it is possible to deliver an account of the total signification of an utterance which is both (fairly) faithful to Grice’s original account and which is able to do a great deal of explanatory work.

15.2 The Role of Grice’s Theory of Implicatures

One of the reasons Grice’s notion of implicatures has seemed so important to philosophers and linguists, I suggest, is because it holds out the promise of rescuing an attractive but prima facie doomed project in theorizing about linguistic meaning. According to this project, which we might label ‘formal semantics’, a core explanatory datum for a theory of meaning is the fact that ordinary speakers can produce an apparently indefinite number of novel sentences. Furthermore, hearers are commonly able to grasp the meaning of novel sentences – i.e. those they have never heard before – with ease (provided they know the meaning of the words and are familiar with the grammatical structure involved). And speakers and hearers can do these things even though the cognitive resources they are able to dedicate to linguistic understanding are constrained. From this perspective, then, a key explanatory task for a theory of meaning is to show how an indefinitely productive capacity can rest on a finite knowledge base.Footnote ² The answer formal semantics gives is that sentence meaning is strictly compositional, i.e. that the meaning of a complex linguistic whole is a product just of the meanings of the parts of the whole (e.g. the words it contains) and the way those parts are put together. If language is compositional then the indefinite nature of linguistic understanding is explained because our productive capacity is based on knowing only a limited set of lexical items and a limited set of rules for putting those items together. For instance, someone who knows the meanings of “horse,” “cow,” “brown,” “white,” and the relational term “_is bigger than_” will (normally) be able to generate all the following sentences with ease:

1. (1) The horse is bigger than the cow.

2. (2) The cow is bigger than the horse.

3. (3) The white horse is bigger than the cow.

4. (4) The brown horse is bigger than the cow.

5. (5) The horse is bigger than the white cow.

6. (6) The horse is bigger than the brown cow.

7. (7) The white horse is bigger than the brown cow.

8. (8) The brown horse is bigger than the white cow.

9. (9) The white cow is bigger than the brown horse.

10. (10) The brown cow is bigger than the white horse.

From a limited knowledge base of just five lexical items, then, the subject’s linguistic understanding projects out to a much larger number of sentences. Compositionality thus explains what is sometimes called the ‘productivity’ and ‘systematicity’ of natural language understanding.

However, the obvious challenge to any such formal semantics programme is that we cannot predict linguistically communicated content just on the basis of words and structure alone – linguistically communicated content is not compositional in the required way. To see this, take the classic case from Grice (Reference Grice1989: 32), where a driver (A) says to a passerby (B) “I am out of petrol” and B replies “There is a garage around the corner.” As Grice suggests, what B communicates in this context is that the garage is open, sells petrol, etc. Yet this isn’t content that we can get just by paying attention to the words B says and the way those words are put together. Or again, consider an utterance of “Someone ate the last cookie” (which in the right context can communicate that Hardeep ate the last oatmeal cookie from the jar) or “Students can take logic or maths” (which in some contexts might convey that they can take one or the other or both, but in other contexts that they can take one or the other but not both). In each case communicated content seems to outstrip compositional, lexico-syntactic content. This recognition – that what we communicate by what we say is often much more than our words strictly and literally mean – seems to throw the entire project of formal semantics into chaos, undermining the very idea of predicting the meaning of complex wholes simply in terms of the meaning and structure of parts.

It is at this point, then, that Grice’s model of communication comes to the rescue by showing that a semantic theory need not provide a treatment of all the content communicated by a speaker. Instead we could maintain that there is a difference between what a sentence itself says (the sort of content that a semantic theory might be able to capture) and what a speaker implies by uttering the sentence, where the latter is a pragmatically conveyed, indirectly communicated bit of content. This noncompositional implied content is nevertheless, according to Grice, something we can theorize about, recognizing the underlying principles which give rise to it.

15.3 Grice’s Theory of Implicatures

Grice divides up the ‘total signification of an utterance’ (what I’ll label ‘utterance meaning’) as shown in Figure 15.1.Footnote ³

Figure 15.1 Grice’s account of the elements of utterance meaning

Focusing first on the notion of ‘what is said by the speaker’, Grice states that he is using the term in a “special, favoured sense” (Reference Grice1989: 25), whereby this kind of content is fixed by the conventional meaning of the words a sentence contains together with certain, limited pragmatic processes (namely disambiguation and reference assignment). So, for instance, to get at what is said by the speaker who utters “He was in the grip of a vice” we need to know both the meaning of the words and sentence structure, and certain facts from the context of utterance, namely the referent of “he” and the disambiguation of “vice” (meaning moral failing or piece of machinery). Grice gives us, then, what I’ll sometimes term below a ‘structural’ or ‘process-based’ definition of ‘what is said’, in terms of the degree of isomorphism between the formal elements of the sentence (e.g. the words) and the elements of the proposition delivered, or in terms of the kinds of processes required for getting at the content (lexico-syntactic interpretation plus disambiguation and reference assignment). Although Grice himself didn’t use the terms ‘semantics’ and ‘pragmatics’ it has become extremely common to identify ‘what is said by the speaker’ with semantic content, and ‘what is implied’ with pragmatic content.

The second element of utterance meaning – implicatures – give us the content that a speaker contributes to the conversation in a more indirect fashion, implying rather than stating. As we can see from the above diagram, Gricean implicatures come in a variety of different kinds, individuated by the degree to which they depend on particular features of the context of utterance. ‘Conventional implicatures’ are carried by conventional features relating to the meaning of the expressions the speaker chooses to use. For instance, an utterance of “She was poor but honest” implies that the speaker thinks there is some kind of contrast or tension between being poor and being honest, but it does so simply via the choice of the expression – ‘but’ rather than ‘and’ – instead of via more fine-grained contextual features.Footnote ⁴ ‘Nonconventional’ or ‘conversational implicatures’, on the other hand, depend more closely on the context in which an utterance is made. Generalized conversational implicatures are ones that would typically or normally arise given the utterance of a certain form of words in the given context. For instance, consider the so-called ‘scalar implicature’ from “Some of Year 8 ate cookies” to Not all of Year 8 ate cookies. It seems there is a scale of strength of expression which runs [one – some – all] and, in this context, selection by the speaker of a weaker item on the scale allows the hearer to infer the negation of the stronger item – not (all) – as implied by the speaker (see Horn Reference Horn1972). Finally, particularized conversational implicatures, like the move from “Someone ate the cookies” to Hardeep ate the cookies, depend intrinsically on quite specific features of the one-off utterance situation.

In addition to distinguishing these different categories of implicature, Grice also provided a mechanism for the recovery of conversational implicatures, in terms of what he called ‘calculability’ (Grice Reference Grice1989: 30–31). A conversational implicature should be capable of being worked out by a hearer using the following kind of schema:Footnote ⁵

A speaker S conversationally implicates that q by saying that p only if:

1. (i) S is presumed to be following the general principles of good communication (what Grice termed the ‘conversational maxims’, the most general of which is the assumption that speakers are being cooperative);

2. (ii) the supposition that S is aware that (or thinks that) q is required to make S’s saying or making as if to say p consistent with this presumption;

3. (iii) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (ii) is required.

So, returning to the Gricean garage example above, if we just take B’s utterance ‘on face value’ (i.e. as communicating only the content that there is a garage around the corner) then their utterance seems to be irrelevant to A’s prior statement. That is to say, taking B as communicating only this piece of content would entail viewing B as flouting one of Grice’s general conversational maxims (in this case: Be Relevant). However, there is a further bit of content naturally available – that the petrol station is open (call this content ‘q’) – which would make B’s contribution conversationally cooperative. The content given in q thus counts as an implicature because: the supposition that B thinks that q is required if B is to be seen as abiding by the conversational maxims; furthermore B hasn’t stopped A thinking that q; so B has implicated that q.Footnote ⁶ This chain of reasoning shows that the implicature is calculable.Footnote ⁷

As well as the requirement for calculability, Grice provided two further tests for deciding whether a given bit of communicated content counts as a conversational implicature, as follows:

1. (i) Nondetachability: if we keep the context and the content fixed, we must get the implicature, as Grice says (Reference Grice1989: 58), “it is not possible to find another way of saying the same thing, which simply lacks the implicature in question.” So, in our above example, if, in response to A’s utterance, B had said “There’s a gas station around the corner” or “There’s a petrol station just past the bend,” the implicature that the station is open would remain.

2. (ii) Cancellability: conversational implicatures can be retracted without contradiction. Grice (Reference Grice1989: 58) notes two different types of cancellation: explicit vs. contextual. Explicit cancellation involves “the addition of a clause that states or implies that the speaker has opted out,” whereas contextual cancellation occurs where the “context…makes it clear that the speaker is opting out”. So, for instance, a speaker can add “but not p” or “I don’t mean to imply p” (explicitly retracting a content that might otherwise be implied). Or we can find contexts in which the utterance of this form of words does not carry the implicature; for instance if B says “There is a garage around the corner” in response to the question “What facilities does this village have?”, B’s utterance may fail to imply that the garage is open.

While these three tests (calculability, nondetachability, cancellability) certainly serve to flesh out our intuitive grip on what counts as a conversational implicature, it seems they do not provide necessary and sufficient conditions for identifying where a given bit of communicated content sits within Grice’s overall taxonomy (i.e. whether some bit of communicated content counts as what is said, versus what is conventionally implied, versus what is conversationally implied).Footnote ⁸ I will look briefly at failures of necessity in this section, before focusing on failures of sufficiency in the next section (Section 15.3).

First, we should be clear that Grice did not expect his tests to draw the divide between ‘what is said’ and ‘what is implicated’ in general because he explicitly notes that the tests are not passed by conventional implicatures. For instance, the move from “She was poor but honest” to the implicature that the speaker thinks there is some contrast between being poor and being honest is neither calculable nor nondetachable. I don’t need to see the speaker as flouting some principle of good communication to know that I’m entitled to derive the implicature that there is a contrast between being poor and being honest (that is to say, the implicature is not calculable) and if the speaker uses a different way of saying the same thing (such as uttering “She was poor and honest”) the implicature disappears (conventional implicatures are thus detachable). At most, then, Grice’s tests might serve to hold apart ‘what is said by the speaker’ from ‘what is conversationally implicated’. However, it might turn out that this worry is in fact somewhat moot since some theorists have argued that the category of conventional implicature itself is problematic. So, Bach Reference Bach1999 argues that conventional implicatures simply don’t exist, while Potts Reference Potts2005 argues that they exist but require a very different treatment to the one Grice envisaged. Thus it may turn out that concentrating our attention just on the divide between ‘what is said’ and ‘what is conversationally implied’ is warranted.

Yet, even if we do narrow our focus in this way, questions about the necessity of the tests remain. For a start, as Grice himself acknowledged, manner implicatures (which depend on the way in which a given content is expressed) clearly fail the test of nondetachability. The speaker who says “Miss X produced a series of sounds that corresponded closely with the score of ‘Home Sweet Home’” implicates something they would not have done if they had instead uttered “Miss X sang ‘Home Sweet Home,’” despite the fact that the latter sentence appears to be just ‘another way of saying the same thing’ (Grice Reference Grice1989: 37). Thus Grice himself explicitly excludes manner cases from the detachability test (Reference Grice1989: 58, emphasis added): “[I]nsofar as the manner of expression plays no role in the calculation, it will not be possible to find another way of saying the same thing, which simply lacks the implicature in question.”

So, something can be a conversational implicature but fail the nondetachability test.Footnote ⁹ Questions have also been raised about Grice’s cancellability test. For instance, it has been argued that some conversational implicatures based on Grice’s maxim of Quantity (which commands speakers to be as informative as required) cannot be easily cancelled. As noted above, many Quantity implicatures rely on the existence of a background scale (like the implicature from “some children ate cookies” to not all children ate cookies) and these cases standardly seem easy to cancel – saying “Some of the children ate cookies, in fact all of them did” is potentially conversationally odd, but the speaker doesn’t directly contradict themselves. However, things seem less clear with nonscalar Quantity implicatures (see Rett Reference Rett2020 for discussion). For instance, consider:

Example (11) implicates, via a flouting of the Quantity maxim, that the woman John met was not one he had a pre-existing relationship with. The tautology in (12) implicates that war has a number of canonical properties that cannot be avoided. In these cases (as Hirschberg Reference Hirschberg1991 and Rett Reference Rett2020 point out) cancellation seems less acceptable than in the scalar case:

Rett Reference Rett2020 goes on to argue that whether a conversational implicature is cancellable or not depends on further features of the conversational context, with implicature content which is crucial to the conversation (roughly, what is known as ‘at-issue content’) failing cancellation tests, while not-at-issue implicature content is cancelleable.Footnote ¹⁰ Yet this is to recognize that, at least as it stands, cancellability alone is not a good candidate for a necessary feature of all conversational implicatures.

Finally, a further challenge to the claim that all conversational implicatures are cancellable comes from content which is also entailed by the semantic content of the sentence produced. For instance, take the following exchange:

1. A: Does John drink slivovitz?

2. B: He doesn’t drink any alcohol.

B’s utterance seems to both entail and imply that John does not drink slivovitz (Sperber & Wilson Reference Sperber and Wilson1986: 61; Carston Reference Carston2002: 139), yet as an entailment this is content which is not cancellable (see Haugh Reference Haugh, Capone, Lo Piparo and Carapezza2013 for further discussion). Although these kinds of cases are not universally accepted, they do once again raise a question about whether cancellation without contradiction marks a universal feature of all conversational implicatures. Of course, the brief discussion of this section doesn’t suffice to show that there are no necessary features had by all conversational implicatures, for it may be that Grice’s original three tests could be refined or reinforced in some way to yield necessary tests.Footnote ¹¹ However, I want to leave this question to one side now and turn instead to the question of whether Grice’s tests are sufficient – that is to say, if some piece of communicated content passes the three tests is that sufficient for us to conclude that the content in question is a conversational implicature? For recent work appears to show that (whether they are necessary or not) Grice’s tests are far from sufficient for identifying implicatures. To see this, let’s turn to examine recent developments at the semantics/pragmatics border and see what these can tell us about how we should characterize implicatures.

15.4 Explicatures vs. Implicatures

Recent work in philosophy of language and linguistics has seriously muddied the waters around implicatures. Whereas once (in Grice’s heyday) it might have seemed reasonable to think that we could limit the input of contextual features to the content of ‘what is said by a speaker’ to a small handful of (fairly well-behaved) processes (i.e. disambiguation and reference assignment), now the situation looks decidedly more murky. To see this, consider the following kinds of cases (where the material after the arrow indicates what a speaker might communicate in a given context, while the underlined part indicates content that does not appear to be contributed by any element contained in the uttered sentence):

These kinds of cases (respectively: (15) comparative adjectives, (16) quantifier restriction and incomplete definite descriptions, (17) incomplete expressions, (18) bridging inferences, and (19), (20) loose talk), amongst others, have recently been at the center of discussions in philosophy about how and where to draw the boundary between semantics and pragmatics. What makes all these kinds of cases so interesting is that they appear to be examples of a speaker directly communicating something by their utterance (so content which ought to line up with ‘what is said by the speaker’ in Grice’s framework) which outstrips the content that can be recovered just by appeal to the constituents of the sentence, together with disambiguation and reference assignment (so which ought to count as implicature content given Grice’s structural definition of what is said by the speaker).Footnote ¹³ In the right context of utterance, the first content hearers’ will recover from an utterance of “Jill is tall” or “Bill got married and had children” seems to be Jill is tall for a jockey or Bill got married and then had children. The explicitness of these further contents – the fact that they seem to capture what the speaker is directly committed to – seems to be in marked contrast to the indirectness of paradigm implicatures, such as the handwriting case with which this chapter opened. What we have in cases like (15)–(20), it has been argued, are thus examples of semantic underdetermination: the semantic content of the sentence (i.e. what is said by the speaker) cannot be compositionally determined on the basis of the meanings of the parts of the sentence and their mode of composition, but this runs counter to Grice’s original way of dividing up the total signification of an utterance.

Roughly speaking, there are three main arguments given for the existence of semantic underdetermination (see Borg Reference Borg, Preyer and Peter2007):Footnote ¹⁴

• ‘Context-shifting cases’ (stressed in Travis Reference Travis1996, Reference Travis, Hale and Wright1997) where utterances of the very same (apparently nonindexical) sentence type intuitively shift in truth value across contexts, even though the relevant feature of the context remains unchanged. So, for instance, in Travis’s famous example (Travis Reference Travis, Hale and Wright1997: 89), we are asked to consider an utterance by Pia of “These leaves are green,” said of some naturally red leaves which have been painted green. If Pia’s utterance is made in response to an artist who is looking for something green to add to her composition, it seems Pia intuitively says something true, but if her utterance is made to a scientist who is looking for leaves for an experiment on photosynthesis, what Pia says seems intuitively false. Yet this apparent change in truth value occurs even though nothing alters with the leaves themselves. This is taken to show that, contrary to initial appearances, something about the meaning of “these leaves are green” must be context-sensitive (and as for this sentence, so for many/most/all other sentences).

• ‘Inappropriateness cases’: in very many contexts, the compositionally derived content will not be appropriate and the speaker should thus be taken to be asserting some contextually enriched content; e.g. the speaker who says “I’ve eaten” standardly conveys something like I’ve eaten recently, instead of the much weaker compositionally derived content which simply requires that the speaker has eaten at some point in the past.

• ‘Incompleteness cases’ seem to demonstrate that some well-formed sentences fail to express complete propositions without contextual enrichment. So “Jill is ready” or “Steel is strong enough,” although grammatically well-formed, seem to require contextual completion before they can be evaluated for truth or falsity, but (at least arguably) nothing in the lexico-syntactic elements of these sentences marks this context-sensitivity.

The pragmatically enriched contents (i.e. the contents on the right-hand side of (15)–(20)) are often labeled explicature contents (following the terminology of Sperber & Wilson Reference Sperber and Wilson1986). Explicature content is directly conveyed but pragmatically enriched content; that is to say, content which captures what is said or asserted but which is enriched by features contributed by robust contextual reasoning, rather than being dictated by compositional linguistic content alone. Importantly, explicature contents seem to pass Grice’s three tests above. The content seems generally nondetachable (saying “Steel is sufficiently strong” or “Bill got married and had kids” would convey the same richer explicature content as conveyed by (17) and (18)), and the content is also cancellable – the speaker can retract apparently without contradiction (e.g. continuing (18) with “although not in that order”). Furthermore, at least in Sperber and Wilson’s Reference Sperber and Wilson1986 ‘Relevance Theory’ model, explicatures are clearly calculable. According to this model, explicature content is recovered through the operation of the Maxim of Relevance, with the content on the right-hand side of (15)–(20) being (in the right context) the most relevant content the hearer can recover (for reasonable processing costs). It is this which gives the contents their status (in the right context) as ‘what the speaker says’.Footnote ¹⁵

Yet despite passing these tests, explicatures seem to be directly expressed content, rather than being indirectly conveyed implicatures. The speaker who utters “Bill got married and had children” will, in most contexts, be heard as asserting, not merely implying, that Bill got married and then had children. If this is right, it shows that Grice’s tests for conversational implicatures are not sufficient – there are contents which pass the tests yet which we do not want to count as indirectly conveyed implicatures. The framework of utterance meaning inherited from Grice then – of a binary divide between, on the one hand, ‘what is said’ content which is recoverable on the basis of words and structure alone and, on the other, implicated content which is the result of rich contextual processing – seems fatally flawed in light of cases like (15)–(20), for many contents which intuitively constitute ‘what is said by the speaker’ turn out to require rich contextual reasoning to recover.

Once we recognize this problem, there seem to be two possible options. On the one hand, we might take cases like (15)–(20) as simply fatal for Grice’s framework, showing that his way of carving up the total significance of an utterance cannot be maintained. As part of this kind of move we might decide to simply reject his category of implicature as itself ill-formed and explanatorily/theoretically empty. This move – what we might term ‘implicature eliminativism’ – has been suggested by Lepore and Stone Reference Lepore and Stone2014. As they write (Reference Lepore and Stone2014: 6) “Put most starkly: we have no use for a category of conversational implicatures, as traditionally and currently understood.” They deny the theoretical value of positing a distinct category of ‘implicatures’ since (they argue) any such label fails to pick out a unified, homogenous category. For Lepore and Stone the fundamental divide in a linguistic communicative act comes between ‘convention-based contributions to the conversational record’ and ‘imaginative, open-ended invitations to explore content’, which do not contribute to the conversational record (Lepore & Stone Reference Lepore and Stone2014: 242). The former kinds of contribution qualify as genuinely semantic content, while the latter kind of personal explorations qualify as individualistic imaginative endeavors, yielding at best indirectly conveyed contents. Lepore and Stone’s objection to implicatures then is that the kinds of cases Grice used to motivate his category cross-cut this divide. So, some ‘implicatures’ (such as scalar cases, indirect requests, and bridging inferences) qualify as convention-based contributions to the conversational record and hence qualify as genuinely semantic content. While others (e.g. metaphor, irony, sarcasm) qualify as open-ended invitations to engage with what the speaker says (with no determinate proposition to be added to the conversational record). Thus, for Lepore and Stone, ‘implicatures’ as Grice conceived them span across the semantic/pragmatic divide. Since no unified treatment could accommodate such a disparate set of cases, they conclude that Grice’s category of ‘conversational implicatures’ should be eliminated as theoretically unhelpful.

While Lepore and Stone’s approach is clearly radical and worth fuller consideration than I can give here, we might note that there are some prima facie worries with the model they provide. First (as other theorists have noted, e.g. Szabó Reference Szabó2016; Stainton & Viger Reference Stainton, Viger and Preyer2018), Lepore and Stone don’t provide a particularly robust account of ‘convention’ or what has to be in place for content to be ‘fixed by convention’. Yet this lacuna apparently leaves them open to the challenge that there are plenty of linguistic expressions that have a conventional usage but which fail to express a semantic content that hearers should add to the conversational record. For instance, as Stainton and Viger Reference Stainton, Viger and Preyer2018 note, the rule attaching to the term “gesundheit” is something like “used as a conventional response when someone sneezes,” but the fact that the term has a clear conventional use does not entail that it contributes a determinate semantic content. Being a linguistic expression associated with a clear linguistic convention is not sufficient, it seems, to guarantee a semantic contribution.

Second, and more problematic, is the fact that there are also nonconventional, imagination-based expressions which it seems must be allowed to make a contribution to the conversational record. The conversational record is supposed to keep track of linguistically relevant moves, in order to help a speaker to decide on an appropriate conversational contribution at any given point. Yet it is clear that, in order to decide what to say at any point in a conversation, one needs to know about pragmatic content just as much as (perhaps even more than) semantic content. I won’t be able to contribute appropriately to a conversation where I’ve heard you say “Moeen is a fine friend” if I fail to realize you’ve spoken ironically, or if I don’t realize when you say “Jill is ready” that you’ve said she is ready for her exam. Planning one’s next conversational move depends fundamentally, it seems, on a grasp of metaphorical content, irony, loose talk, and so on (Green Reference Green and Preyer2018). If all of these kinds of contents are placed ‘off record’, as Lepore and Stone seem to envisage, then hearers will need to run a second, more liberal kind of conversational record, one which tracks both kinds of linguistically conveyed content, alongside the one that tracks only conventional, semantic contributions. Yet this seems to multiply conversational records (and cognitive load) unnecessarily.Footnote ¹⁶

So, instead of eliminating the category of implicatures and entirely scrapping Grice’s framework for dividing up utterance meaning, let’s turn to a second, very widely endorsed response to cases like (15)–(20). On this approach the lesson to be drawn from the problematic cases is that we should refine Grice’s bipartite model, replacing it with a tripartite divide which recognizes:

1. (i) standing sentence meaning (exhausted by lexico-syntactic features but often/always subpropositional content).

2. (ii) directly asserted, pragmatically enhanced propositional content – what the speaker says (what I’ll term, following Sperber & Wilson Reference Sperber and Wilson1986, ‘explicature content’).

3. (iii) indirectly conveyed content – what the speaker implies.

This three-way divide gives us the general shape of what have come to be known as ‘Contextualist’ models.Footnote ¹⁷ According to this school of thought, what a speaker says (i.e. the content which captures what the speaker directly expresses or asserts) is shot through with rich pragmatic effects. Grice’s notion of what is said by a speaker, if this is to be propositional/truth-evaluable, asserted content, cannot, Contextualists argue, be content that is recovered on the basis of just a small subset of well-behaved contextual processes (like disambiguation and reference assignment), rather the full panoply of pragmatics (as partially illustrated by (15)–(20)) must be allowed to enter.

Cases like (15)–(20) provide a valuable lens through which to ask what (if any) force Grice’s original machinery can still wield in the contemporary debate. However, I want to suggest that, in answering this question, we need to be careful about exactly what lessons we draw from them. For what these kinds of cases show, I contend, is not that the Gricean assumption about what is said – namely, that there is a useful and important notion of linguistic content which cleaves very tightly to lexico-syntactic content – is wrong. Rather what they show is that the move (somewhat tacitly made in Grice and more explicitly made post-Grice) to identify semantic content with what is said by a speaker is a mistake. To see this, however, we need to take something of a detour through the Contextualist picture of a semantic theory infiltrated throughout by rich pragmatic processing, asking how this approach distinguishes explicatures from implicatures. The question we need to ask, for any contextually given piece of content, is whether it contributes to what is said or asserted (explicature content) or to what is indirectly conveyed (implicature content)? This has proved a notoriously difficult question for Contextualists to answer and I will argue that the only plausible answer to be given serves to reveal the need for propositions determined on the basis of lexico-syntactic properties (together with disambiguation and reference assignment) alone (that is to say, it shows that we do need contents which answer to Grice’s structural definition of ‘what is said by a speaker’).

Testing for Implicatures vs. Explicatures

The question we face is how to individuate explicature content from implicature content – if someone says “Naoki took out his key and opened the door” what makes the content Naoki took out his key and opened the door with his key the proposition expressed or asserted, as opposed to being a content merely implied? Or again, in our above example, should we treat the content that there is an open garage around the corner as part of the explicature – what the speaker directly conveys – or as part of an implicature (as Grice himself thought)? A range of tests have been proposed to answer this crucial question, including formal constraints and what I would term ‘processing accounts’, however as I’ve argued elsewhere (Borg Reference Borg2016), none of these tests seem able to draw the explicature/implicature divide at the place where we intuitively want it. To see this, let’s turn now to look very briefly at some of these proposals.

According to Sperber and Wilson’s canonical account (Reference Sperber and Wilson1986: 182), an explicature is a “pragmatic development of the logical form” of the sentence produced. Thus, we might think to define implicatures negatively, simply as contextually derived contents that are not mere developments of extant logical form. So, for instance, we might think that (21a) counts as an explicature of (21), while (21b) could only be an implicature:

While this proposal has intuitive force and prima facie seems to deliver the results we want (e.g. making the right ruling for (21)), it does face difficulties. For instance, consider (21c):

Here it seems we have ‘developed’ the original logical form, via conjunction introduction; so should we count (21c) as potential explicature content or only a potential implicature? To decide how to answer this question it seems that we either need a fuller account of ‘development’ or we need some additional test we can apply to developed logical forms to decide if they constitute explicatures or implicatures. Proposals have been made along both these lines. For instance, Carston and Hall (Reference Carston, Hall and Schmid2012) have suggested that any ‘nonlocal’ development, where this means allowing pragmatic processes to modulate a whole proposition rather than a part, must count as implicature content. This kind of ‘locality constraint’ seems to do the work required in ruling (21c) as an implicature rather than an explicature, but it too runs into difficult cases. For instance, consider (21d):

This is arguably a local development, yet it captures content that above we thought intuitively would only constitute an implicature not an explicature of (21).

Alternatively, then, we might turn to other features to mark the difference here. For instance, both Carston Reference Carston2002 and Recanati Reference Recanati2004 have appealed to the ‘scope test’, requiring explicature content to fall within the scope of operators such as conditionals and negation, while implicature content can fall outside. However, as Carston herself notes, such a test seems problematic. First, because it seems to draw the line in intuitively the wrong place, for instance counting the content generated by sarcastic utterances as explicature content. Thus, as Camp Reference Camp2006 has stressed, if I say “If you have one more great idea like that you’ll be sacked” the sarcastic content – that your ideas are not great – seems to be part of the overall truth conditions of what I say (falling within the scope of operators).Footnote ¹⁸ Yet intuitively such sarcastic content is not something I assert directly but instead has a much more indirect, implicature-like feel. Furthermore, scope tests (and others like it) seem to operate at the wrong level for individuating explicatures from implicatures. For scope tests operate at the sentence level, targeting type sentences, while at least particularized conversational implicatures are most certainly properties of token utterances.

Finally, then, we might think to look to the realm of psychological processing to do the work required instead. So Sperber and Wilson Reference Sperber and Wilson1986 suggested that the only way to individuate explicatures from implicatures is to use their general relevance framework, whereby propositions that are highly relevant for a hearer (which they define as having high cognitive effects for low processing costs) will constitute explicature content, while less relevant content will count as implicatures. Recanati Reference Recanati2004 also defined explicature content (what he calls ‘what is said’) in psychological terms, as the first full proposition available to a hearer on processing a given utterance, with all implicatures requiring further inferential work from this first available content (see Footnote n. 14 above). However, it is not at all clear that turning to psychological criteria will draw the division in the place we want either. For, on the one hand, content classed as classic Gricean implicatures and which definitely seems to be in some way indirectly conveyed (e.g. communicating Jones is not a good philosopher by saying “Jones has nice handwriting”) might nevertheless constitute the most relevant content for the hearer. While, on the other hand, as Grice himself stressed it seems that sometimes an implicature can be ‘intuitively grasped’, that is grasped without entering into a complex psychological derivational procedure. As Grice writes (Reference Grice1989: 31, my emphasis):

This seems to suggest that Grice himself allowed that sometimes an implicature may be the first recovered content a hearer entertains (which would, contra intuition, make it count as an explicature according to the current suggested definition). For reasons of space, I won’t explore the apparent problems with each of the extant criteria for holding apart explicatures and implicatures any further (see Borg Reference Borg2016 for further discussion), for instead I want to turn at this juncture to what I take to be the only plausible candidate for individuating explicatures from implicatures, namely socio-linguistic features.

15.5 Using Socio-Linguistic Notions to Distinguish Elements of Utterance Content

As I’ve argued elsewhere (Borg Reference Borg2017), I think we should move towards an understanding of the explicature/implicature divide which locates the difference in the socio-linguistic realm, arising from the relation of language to social norms and cultural expectations. On this approach, a socio-linguistic implicature is an element of communicated content that a speaker is not held strongly conversationally liable for, whereas explicature content is communicated content that a speaker is judged strongly conversationally liable for. Conversational liability I take to be a matter of degree, where a speaker can be held more or less responsible for communicating a given piece of content and where it is commensurably more difficult or more easy for the speaker to retract or reject the content in question as forming part of what she is committed to by her utterance. As with the Contextualist model above then, I want to agree that some richly pragmatically enhanced contents count as directly asserted (so constitute part of Grice’s notion of ‘what is said by the speaker’ when this notion is defined in terms of directly asserted, rather than indirectly conveyed, content), while other such contents count only as indirectly conveyed implicatures. A speaker who says (in the right context) “Bill got married and had children” may well succeed in asserting that Bill got married and then had children, or someone who says “Jill is ready” may assert that Jill is ready to bowl, even though these asserted contents outstrip what we can recover on the basis of lexico-syntactic constituents alone. These contents will count as asserted, I suggest, just in case the speaker is held strongly conversationally liable for them by ordinary interlocutors, i.e. if speakers cannot, or find it extremely hard, to retract or reject them as forming part of what they said.Footnote ¹⁹ It is notions of responsibility and liability then that give us the divide Contextualists want between explicature and implicature content.

However, I suggest that once we turn to the socio-linguistic realm to distinguish these kinds of contents, what we also find is that there is an ineliminable role for propositional contents which are fixed just by word meaning, sentence structure, disambiguation and reference determination (i.e. contents which map to Grice’s notion of ‘what is said by the speaker’ when this is defined in structural terms). That is to say, if we allow that socio-linguistic factors (to do with the degree of responsibility a speaker is held to have for a given communicated content) provide the most plausible ground for distinguishing between explicature and implicature content, then we cannot avoid also positing a notion of minimal content that tracks Grice’s structural version of ‘what is said by the speaker’. If this is right, it yields a slight tweak on the three-way divide within the total signification of an utterance provided by the Contextualist: if a speaker U utters a sentence s and thereby conveys p, p will fall into one of the three categories shown in Table 15.1.

Here (a) captures Grice’s structural definition of ‘what is said by the speaker’. Elsewhere, I’ve labeled this kind of content ‘minimal content’ (see Borg Reference Borg2004, Reference Borg2012 for further discussion of this terminology). Minimal content differs from ‘standing sentence meaning’ in the Contextualist model above as it is held always to be genuine propositional or truth-evaluable content, rather than some subpropositional fragment of meaning. So why might we think that this level of content must be propositional in nature? The answer to this question is: because there are socio-linguistic practices which are clearly grounded in this kind of content. So, for instance, consider the important socio-linguistic category of lying. It seems that paradigm cases of lying involve a speaker producing a sentence with a literal meaning p where they know or believe p to be false (perhaps with the additional clause that they intend to get their audience to believe p). As many authors (e.g. Camp Reference Camp2006; Goldberg Reference Goldberg2015; Saul Reference Saul2013; Michaelson Reference Michaelson2016) have suggested, paradigm judgments of lying seem to track something very like minimal content. So, take an utterance of:

Imagine that the speaker knows that Maya took out her key but did not use it to open the door (perhaps the door was already unlocked and she had taken out her key to unlock something in the room). Here the utterance is clearly misleading, and it may well be intended to be misleading, but intuitively it seems it does not count as a lie. Although the speaker might well be conversationally committed to the bridging inference that Maya opened the door with her key, since this is something most ordinary interlocutors would assume (and after all it is this which grounds our judgment that the speaker in this case is being misleading), intuitively the speaker is not judged to have the kind of responsibility for the bridging content that would underpin an accusation of lying.Footnote ²⁰ To lie using an utterance of (22) it seems that the speaker must know or believe either that Maya didn’t take out her key or didn’t open the door (i.e. she must know or believe that the minimal content of (22) is false).

Or again, take the well-entrenched distinction between following the rules of a game and abiding by the spirit of the game. The rules of a game are written down or orally preserved and they state the conditions players must observe if they are to play the game. Yet the rules will often allow for a range of actions which, if a player were to undertake them, would ruin the game for others. These activities – not strictly prohibited by the rules but nevertheless not acceptable – constitute breaches of the spirit of the game and players can be criticized when they perform them, even though there are no explicit penalties or sanctions that can be imposed (as there would be with breaches of a rule). Playing in line with the spirit of the game constitutes what we recognize as fair play. So, for instance, nothing in the rules of soccer say that (if the game hasn’t been stopped by the referee) a striker can’t shoot when the opposing goalkeeper is incapacitated. Yet most people recognize that shooting at an open goal when the keeper is lying on the floor injured is unsporting – it is failing to play within the spirit of the game. When a player abides by this sense of what the spirit of the game demands they are commended as examples of good sportsmanship (as in a match in 2000 when West Ham striker Paolo di Canio caught the ball, rather than taking a shot, because goalkeeper Paul Gerrard was lying injured).Footnote ²¹ This kind of commendation is not generally forthcoming when someone merely abides by the rules of the game. This well-established, socio-linguistic distinction again depends, I want to argue, on recognizing a minimal propositional content that is strictly bound by the lexico-syntactic content of the sentence (i.e. a kind of content which is defined by Grice’s structural sense of ‘what is said by the speaker’) and which gives us the rules of the game. This differs from the explicature content that all reasonable interlocutors may take the rule-setter to have committed herself to by her statement of the rules (the pragmatically delivered spirit of the game). Furthermore, this kind of distinction matters in other areas as well, such as in legal discourse and similar contexts where interpretation of a text, perhaps as opposed to a speaker, matters (e.g. academic dissertations and textualist interpretations of religious scripture); see Borg Reference Borg2017; Borg and Connolly Reference Borg, Connolly, Lepore and Sosa2022 for further discussion.

What cases like (15)–(20) demonstrate then, I want to suggest, is not that Grice’s category of implicatures cross-cuts the semantics/pragmatics divide and so must be eliminated (Lepore and Stone’s proposal), nor do they show that the only important distinction within the total signification of the utterance comes between content the speaker directly asserts and content she merely implies. Rather what they show is that Grice’s original category of ‘what is said by the speaker’ merged two different individuating features, features which can and often do come apart.

On the one hand, Grice gave us a formal, structural definition of ‘what is said’, in terms of content delivered by the words and structure of the sentence together with the limited pragmatic processes of disambiguation and reference assignment. On the other hand, however, by labeling this notion ‘what is said by the speaker’ he seems to have intended to capture our intuitions about the content a speaker asserts or directly communicates. But as cases like (15)–(20) show (given our working assumption that Indexicalist approaches won’t work, see Footnote note 13), what is asserted/directly expressed cannot be captured by lexico-syntactic elements, plus disambiguation and reference determination, alone. Instead rich pragmatically derived content must be allowed to figure. Yet none of this is tantamount to showing that the content which lines up with Grice’s structural definition is otiose.

Contextualist accounts claim that:

1. (i) Sentences [often/always/only] express propositions given rich pragmatic inference.

2. (ii) It is explicatures alone (not sentence-level contents) that have contextual effects (in terms of affecting agent’s belief sets, etc.) and are therefore worthy of conscious attention (Sperber & Wilson Reference Sperber and Wilson1986: 193).

Yet as I’ve tried to suggest, once we turn to look at socio-linguistic practices we see that both these claims are problematic, as we cannot make sense of the practices surrounding things like lying, perjury, misleading, deceiving, defamation, rules vs. spirit of games, and textualist interpretations in the law and religion, without positing a propositional content delivered by words and structure (together with disambiguation and reference determination) alone.

When subsequent writers took Grice’s notion of what is said by the speaker as capturing semantic content, and combined this with the idea that ‘what is said’ must capture what the speaker intuitively asserts, this led to the view that semantics must be treated as ineliminably shot through with rich pragmatic processes. As Soames (Reference Soames2009: 320) notes (when discussing cases like (15)–(20)):

Yet once we recognize the tension in Grice’s original account and see that the two elements of his definition pull apart, we can also recognize, I contend, that the most explanatorily robust move is to retain both elements of his account of ‘what is said by the speaker’, positing both a structurally constrained minimal content and a socio-linguistic, pragmatically enhanced notion of explicature content.

Thus I want to recommend the kind of neo-Gricean framework for the total signification of an utterance displayed in Table 15.2.

The divide between (i) and (ii) – semantic versus pragmatic content – should be construed (as it was by Grice’s structural notion of ‘what is said by the speaker’) as grounded in the kinds of processes licensed in production/recovery of the given content-type: limited, lexico-syntactically based processes for semantic content; rich, abductive, potentially all-things-considered processes for pragmatic content.Footnote ²² This process-based account yields a binary distinction between propositions which capture the literal, standing meaning of linguistic expressions on the one side (allowing that semantic content is suitable for a potentially modular account of the cognitive capacities which underpin it, see Borg Reference Borg2004, chapter 2), and propositions which capture pragmatically conveyed utterance content, a content which is not strictly compositional and which involves processes capable of considering anything a speaker or hearer knows. The divide between (a) and (b) on the other hand – asserted versus implied content – is not a division in kind but is rather a vague, hazy boundary drawn in terms of socio-linguistic notions like praise/censure of the speaker, judgments of conversational responsibility and conditions of retraction. Sometimes (but very far from always) speakers assert the literal, minimal semantic content of the sentences they produce, while at other times they assert pragmatically enhanced contents (explicatures).Footnote ²³

This model differs from the Gricean picture with which we began, in that it allows that contextually determined content (where the processes involved go beyond the ones Grice countenanced as part of what is said by the sentence, i.e. beyond mere disambiguation and reference determination, to include such things as pragmatic broadening and narrowing) can count as asserted content. However it follows the original Gricean framework in maintaining that there is an important kind of propositional content (what I would term ‘minimal semantic content’) which depends only on the elements of the sentence produced (broadly speaking, the words it contains) and the way they are put together. And it contrasts this kind of content with all and any content deriving from rich, all-things-considered contextual processing. It then recognizes Grice’s crucial division between asserted and merely implied content but holds that this is a socio-linguistic distinction, to do with degrees of liability and conditions for retraction, which are orthogonal to the question of literal, semantic content versus nonliteral, pragmatically determined content. By clarifying the core relationship between the semantics/pragmatics divide and socio-linguistic notions like assertion and implicature, then, I suggest that we can arrive at a truly satisfying account of the intuitive category of implicature content that Grice did so much to bring to our attention.

15.6 Conclusion

I have suggested in this chapter that Grice’s technical notion of ‘implicature’ is both intuitively appealing and does important explanatory work, capturing content that we take speakers to have indirectly communicated. However, the framework within which implicatures were originally introduced – which holds apart content delivered via lexico-syntactic processing and content delivered via rich pragmatic processing, and claims that the latter is always indirectly conveyed implicature content – is problematic. For, as Contextualists have stressed and contra to Grice’s original model, not all rich pragmatic content is implied, some is directly asserted explicature content. Grice’s three tests, then, turn out to be good tests for richly inferential pragmatic content, but they are not good tests of indirectly conveyed implicature content. In light of this recognition, Contextualists have proposed a refinement of Grice’s original framework, which recognizes three kinds of content: incomplete sentence meanings, explicatures and implicatures. However, this revision leaves Contextualists with a fundamental question to answer concerning how to individuate explicature content from implicatures.

I suggested that the only plausible answer to this individuation question is to look to the socio-linguistic realm, asking what content a speaker is held most strongly conversationally liable for. However, once we turn our attention to the socio-linguistic realm, I argued that we find that the content individuated in terms of Grice’s structural definition of ‘what is said by the speaker’ (content individuated by word meaning, compositional structure, disambiguation and reference assignment alone, which I above termed ‘minimal semantic content’) must, contra the Contextualist suggestion, be treated as genuinely propositional content. We need propositional contents which answer Grice’s structural definition of ‘what is said by the speaker’ in order to capture things like the lying/misleading and the rules/spirit of the game distinctions. It seems then that we need both (i) the distinction between content Grice identified via his structural definition of ‘what is said by the speaker’ and content which is arrived at via richer pragmatic processing, and (ii) the distinction between directly conveyed and indirectly conveyed content (the explicature/implicature divide). Although terminology is somewhat vexed in this area, I suggested that we might think of the first distinction as the distinction between semantics and pragmatics, and the second as the distinction between assertion and implication. The moral of this re-visiting of Grice’s notion of implicature, and his framework in general, then, is that if we want to understand the terrain here properly, we need to appreciate that the semantic/pragmatic and asserted/implied distinctions cross-cut one another. The divide between semantics and pragmatics should, I suggest, be understood as a type-level processing distinction, where semantic content is delivered by processes that look no further than to what is given at the lexico-syntactic level, while pragmatic content is delivered by rich, abductive, all-things-considered reasoning. On the other hand, the asserted/implied distinction is a token-level socio-linguistic distinction, which cleaves content apart in terms of the role that that content plays in our communicative exchanges. Contra Lepore and Stone, I would argue that Grice’s category of implicature remains an extremely useful category in contemporary theorizing about linguistic meaning, but as a speech act category, rather than as the opposite of semantics.

16.0 Questions and Answers

We do not live in a void. External information from the world is constantly perceived and processed by the senses. In order to facilitate this process, our cognition relies on a number of background assumptions and predictions based on previous experience. One of the most interesting linguistic aspects of this fact is the phenomenon of presupposition. Almost anything we say presupposes a number of things; this is why presuppositions have been of interest to linguists for the last 50 years. However, the issue of presuppositions also goes to the heart of analytic philosophy: key topics such as the analysis of assertion (Schlöder, this volume), attitude ascriptions (Kratzer, this volume), argumentation (Pavese, this volume), definite descriptions (Coppock’s and Kamp’s contributions to this volume), indexicality (Bary, this volume), implicatures (Borg, this volume), modals (Mandelkern, this volume), perspective sensitivity (Anand and Toosarvandani, this volume) and vagueness (Carter, this volume), among others, rely crucially on an understanding of presuppositions.

Traditional linguistic approaches to presupposition were mostly concerned with the interaction of presuppositions with various logical connectives and other embedding contexts, the so-called projection problem. Recent developments in linguistics have started to link the analysis of presuppositions to general processes of cognition and reasoning, such as attention, probabilistic reasoning and information structure. The bulk of the paper reviews these recent developments.

One key ingredient for adequately analyzing presuppositions is to understand its interaction with the discourse context. In order to be able to do this, we need more good quality data. This calls for more large-scale corpus studies as well as verification of data in an experimental setting.

Some outstanding questions concern the acquisition of presuppositions, perspective-taking (see Anand & Toosarvandani, this volume), the interaction of presuppositions and discourse structure (see Hunter & Thompson, this volume). Another question is whether presuppositions form one coherent group or they should be thought of as different types of presuppositional phenomena, and if yes, along what dividing lines. Finally, the answer to the ultimate questions, why do we have presuppositions at all, and why exactly we see the presuppositions that we see, looms still in the distance.

16.1 Introduction

What would it be like to speak a language without presuppositions? Is it even possible to have such a language, and if yes, would we want to speak it? Here is a game to try: let’s transcribe the fist sentence of the novel Wuthering Heights by Emily Brontë (1) into a text that does not contain expressions that are typically assumed to trigger presuppositions, cf. (2):

A first attempt at a presuppositionless transcription:

An immediately visible result of this little exercise is that spelling out the presuppositions of various expressions creates a rather tedious text. I have assumed that the presuppositional expressions above are the verb return from x (which presupposes that its subject went to x at a previous time), the definite descriptions my landlord, the neighbour (which presuppose that there is a unique individual that the description refers to) and solitary (=sole) (which presupposes the clause that it modifies, and asserts an exhaustification of it). Just trying to spell out all these presuppositions creates a text that almost certainly no one would want to read any further. One reason for this tediousness is repetitions. There is something about presuppositions that helps package information in an efficient way.

In fact, the above paraphrase was based on a limited understanding of presuppositions that assigns them to a relatively small subset of the lexicon, also called conventional presupposition triggers (see an example list in Section 3.1). If we think of presuppositions as a broader category that encompasses sortal and compositional restrictions of lexical items (cf. Magidor Reference Magidor2013; Asher Reference Asher2011) as well as bits of world knowledge and assumptions of conversational partners (a.k.a. conversational presuppositions), a presupposition-free paraphrase becomes impossible. This is because the paraphrase has a problem of regression: any word that we use to explicate the sortal restriction of another word has sortal presuppositions itself and understanding our paraphrases require a significant amount of world knowledge and assumptions about our conversational partners as well. Here is thus a second (failed) attempt at paraphrasing (1):

Another irreducible and arguably presuppositional aspect of the above example is the interpretation of anaphors. One reason is the gender restriction on pronouns is commonly thought of as a presupposition on their use. But more importantly, the requirement that an anaphoric pronoun needs to be resolved to a suitable referent in the discourse is usually thought about as a presupposition on using the pronoun. However, spelling out the anaphoric component of pronouns results in metalinguistic statements (“a unique discourse referent of he can be found in some preceding discourse”):

Thus it seems all but impossible to transform English into a language that does not use conversational and sortal presuppositions, or anaphoric pronouns. Eliminating certain conventional triggers such as change of state verbs, definite descriptions or focus particles from our language is perhaps manageable (though one might wonder about the possibility of eliminating some other conventional triggers, for example the factive verb know or discourse connectives; see Pavese Reference Pavese2020 for discussion of the latter) but at the cost of creating a highly tedious and cumbersome text. If we add the content of presuppositions to our sentences explicitly – to the extent that it is possible – we end up endlessly repeating bits of information.

Another aspect of the original example (1) that is lost in the paraphrases is the illusion of familiarity that Brontë creates between the reader and the protagonist. This illusion might be due to the requirement according to which the content of presuppositions should be already present in the common ground of the interlocutors at the time of uttering a presuppositional sentence (see discussion in the next section). When this is not the case, and there is no information that contradicts them in the common ground, presuppositions are thought to be accommodated, i.e. silently added to the common ground. Since (1) is the first sentence of a novel, there is no common ground between the reader and the narrator other than general world knowledge. But the use of presuppositions and the resulting accommodation process forces the reader to create one, which in turn produces an illusion of familiarity between the reader and the protagonist.Footnote ²

Presuppositions are an irreducible property of natural language use. They have a crucial role for creating coherent discourse, managing new and old information. They have a crucial role in keeping track of discourse referents, whether concrete or abstract. They have a crucial role in efficient information packaging at the lexical level: they way we fold concepts into words interacts fundamentally with presuppositions. They even have a crucial role in managing social interaction and bonding between conversational partners in dialogue.

This is why it is hard to find a topic in semantics and analytical philosophy that does not interact with presuppositions. The study of assertion (Schlöder, this volume), attitude ascriptions (Kratzer, this volume), argumentation (Pavese, this volume), definite descriptions (Coppock, Kamp, this volume), discourse structure (Hunter, this volume), indexicality, implicatures (Borg, this volume), modals (Mandelkern, this volume), perspective sensitivity (Anand and Toosarvandani, this volume), vagueness (Carter, this volume) and more … all interact crucially with presuppositions.

And yet, there is little agreement about what presuppositions are as a theoretical phenomenon, what are examples of it, whether it is one phenomenon or many different phenomena, how presuppositions arise, what the content of particular presuppositions is, and many other issues. Perhaps the only thing that everyone agrees about is that presuppositions project: the presuppositional inference can survive in contexts that are normally entailment-cancelling: the so-called family of sentences test:

1. (5)

   a.	I doubt that Mr. Lockwood has returned from his visit.
   b.	Has Mr. Lockwood returned from his visit?
   c.	If Mr. Lockwood has returned from his visit, the lights should be on.
   	all imply: Mr. Lockwood went for a visit.

The agreement ends there: why projection happens, why it shows the particular properties it does and what the fine-grained structure of the data is like is a matter of heated debates in both linguistics and philosophy. Indeed, the problem of explaining presupposition projection dominated the presupposition literature in the last 50 years leading to an abundance of projection theories that propose to explain the purported basic facts of projection.

Recently though, scholars have started addressing various questions in presupposition theory that are complementary to those asked by classical theories of projection: whether the presuppositions of various triggers show the same properties, why presuppositions arise to begin with (a.k.a. the triggering problem) and whether contextual and linguistic factors can influence presupposition projection and interpretation. After a discussion of major theories of what presuppositions are, this chapter overviews some of these recent developments. Unconventionally, I will not present in detail how the classic theories of presupposition predict the (alleged) projection facts (see Soames Reference Soames, Gabbay and Guenther1989; Heim Reference Heim1990; Geurts Reference Geurts1999; Beaver Reference Beaver2001; Kadmon Reference Kadmon2001; Potts Reference Potts, Lappin and Fox2015; Beaver & Geurts Reference Beaver, Geurts and Zalta2014, among others, for excellent overviews). Still, the conclusion I reach about presuppositions is valid for projection as well: it is complicated. Facts of presuppositions are the result of a dauntingly complex interplay of a number of lexical, contextual and extralinguistic factors and cannot be described by a simple beautiful formula one can print on a T-shirt.

16.2 What Is Presupposition?

16.3 Some Recent Developments and Outstanding Questions

There are many thorny issues in presupposition theory; this section presents a personal selection. I discuss whether all presuppositions are the same and if not, whether we can establish different classes of them. Second, I present some recent attempts at explaining why we have presuppositions in the first place, aka the triggering problem. Third, I give an overview of various linguistic and pragmatic factors that influence projection and the interpretation of presuppositions, and discuss the challenges these facts pose for projection theories. Finally, I briefly comment on the problem of presupposition projection from the scope of attitude verbs.

16.4 Conclusion

By way of conclusion, let me quote a paragraph from Kamp and Rossdeutscher (Reference Kamp and Rossdeutscher1994) which suits presuppositions perfectly:

17.0 Questions and Answers

The difficulty of simultaneously accounting for logical and semantic properties which seem to jointly obtain is one thing that makes modals and conditionals endlessly interesting to linguists and philosophers.Footnote * This, in turn, has important ramifications for the foundations of semantics and pragmatics. And modals and conditionals play a central role in rational decision-making and practical reasoning, making their proper treatment of broad interest across cognitive science.

Especially of interest to me is work that has explored surprising connections and interactions, for instance, between modality, presupposition, and anaphora; between modality and attitude predicates; and between modality and moral psychology.

A successful account of modals and conditionals will compositionally account for the interaction of ‘if’-clauses with overt modals, while also accounting for the meaning and logic of conditionals that do not contain overt modals. It will give a systematic account of the interaction of modals and conditionals with other logical connectives and attitude predicates. It will situate the theory of modals and conditionals within a psychological theory of our modal representations, and in the context of other semantic/pragmatic systems, like anaphora and presupposition.

What is the logic of the conditional? And, relatedly, what is the underlying structure of the conditional and the nature of the interaction of ‘if’-clauses with overt modals? What is the precise connection of modality to anaphora and presupposition? How unified is the semantics and psychology of different kinds of modality (epistemic, deontic, agentive, circumstantial) with each other, and with conditionals? How do modals and conditionals interact with attitude predicates?

17.1 Introduction

Modals (‘It might rain’; ‘You must eat a cookie’; ‘I can fly’) and conditionals (‘If it rains, the picnic will be cancelled’; ‘If you want a cookie, let me know’; ‘If I had wings, I would have been able to fly’) play a starring role in philosophical and linguistic research. The ability to think modally distal thoughts is central to the human capacity to plan and choose; and the ability to express such thoughts is central to the human capacity for collective action.

Modals and conditionals have yielded a rich bounty of puzzles about logic, semantics, and pragmatics. In light of the obvious futility of giving an overview of these puzzles, I will organize this chapter around just three topics, focusing further in each case on just a few questions that particularly fascinate me. The result is partly autobiographical; but that seems inevitable for the task set out for this volume, and I hope to thus give a taste of some of the many interesting issues raised, together with pointers to further resources.

The first topic I explore is epistemic modals: words like ‘might’ and ‘must’, on a broadly epistemic reading. I survey a handful of puzzles about epistemic modals, puzzles which also touch on questions about attitude predicates, propositions, anaphora, and presupposition (for more on which, see Angelika Kratzer’s, Yael Sharvit and Matt Moss’s, Elizabeth Coppock’s, and Márta Abrusán’s contributions to this volume).

The second topic is conditionals. One reason the conditional has played a central role in logic, semantics, and pragmatics is because it does not seem amenable to a bivalent truth-functional analysis. For this reason, the conditional constitutes a major challenge for the Gricean research program which aims to treat the logical connectives and operators of natural language – ‘and’, ‘or’, ‘not’, ‘if … then …’ – as the truth functions of classical logic, and explain their communicative complexity via the interaction of simple truth conditions with complex pragmatic reasoning (see Emma Borg’s chapter). If conditionals are not truth functions, what are they? A prominent view in philosophy says that $⌜$If p, then q$⌝$ is true just in case the closest p-world(s) are q-world(s).Footnote ¹ A prominent view in linguistics says that $⌜$If p, then q$⌝$ is a restricted epistemic modal claim, which says that q must be true, on the assumption that p is true. In my discussion I will focus on points of convergence and divergence between these approaches.

The final topic I will explore is practical modality: modal claims about what one should do (deontic modals: ‘You should give me a cookie’), and about what one can or cannot do (agentive modals: ‘You can pass this test’). This topic has obvious ramifications for philosophical questions about morality, ability, and their connections. I will focus on recent work which suggests that these modals build on a representation of a set of available actions, and explore a puzzle about how we represent that set of actions when we give orders.

17.2 Epistemic Modals

Epistemic modals are words like ‘might’ and ‘must’ on a broadly epistemic interpretation, as in (1):

1. (1)

   a.	Susie might bring her new girlfriend to the party.
   b.	Latif must be furious.

The standard analysis of modals in natural language, growing out of work in modal logic (Kripke Reference Kripke1963; Kratzer Reference Kratzer1977, Reference Gibbard, Harper, Stalnaker and Pearce1981; Lewis Reference Lewis1979), treats them as quantifiers over accessible worlds.Footnote ² The idea is that modals are evaluated relative to a binary accessibility relation between worlds. An existential modal sentence ($⌜$Might p$⌝$, $⌜$May p$⌝$, etc.) is true at a world w just in case some world accessible from w makes p true; a universal modal sentence ($⌜$Must p$⌝$, $⌜$Have to p$⌝$, etc.) is true at w just in case every world accessible from w makes p true. Different kinds of modality are associated with different kinds of accessibility relations. Epistemic modals are associated with a broadly epistemic one, on which a world w can access a world $w^{\prime }$ just in case $w^{\prime }$ is compatible with the relevant evidence or knowledge in w.

17.3 Conditionals

I will turn now from modals to conditionals, which have been a topic of lively philosophical debate since antiquity. The literature on conditionals is thus extraordinarily large. For some helpful overviews, see e.g. Edgington Reference Edgington1995; Bennett Reference Bennett2003; von Fintel Reference von Fintel, von Heusinger, Maienborn and Portner2011; Kaufmann and Kaufmann Reference Kaufmann, Kaufmann, Lappin and Fox2015; Gillies Reference Gillies, Hale, Miller and Wright2017; for some of the earlier history of the debate, see Mates Reference Mates1953 and Algra et al. Reference Algra, Barnes, Mansfeld and Schofield1999: part II. Let me emphasize again that I make no pretense of giving an overview here. (Conditionals have also played a central role in philosophical work well beyond philosophy of language, for instance in the theory of rational decision [Stalnaker Reference Stalnaker, Harper, Stalnaker and Pearce1980b; Gibbard & Harper Reference Gibbard, Harper, Stalnaker and Pearce1981] and (relatedly) causation [Lewis Reference Lewis1973a].) In my brief space here, I will start by explaining why the traditional identity of the conditional with the material conditional is not viable. Then I will introduce two influential theories of the conditional: one, from the philosophical literature, which regards ‘if’ as a two-place operator; and one, from the linguistics literature, which regards ‘if’ as simply providing a restriction on modals in the conditional’s consequent. I will argue that there is more disagreement between these approaches than first appearances suggest.

17.4 Practical Modality

In this final section, I will turn to a class of modals which I’ll call practical modals. This class comprises, first, deontic modals – modals that communicate permissions, obligations, and requirements, as in (16):

1. (16)

   a.	You may have a cookie.
   b.	You should visit your grandmother.
   c.	John must stop cheating on his husband.

And, second, agentive modals: modals which ascribe abilities and compulsions, as in (17):

1. (17)

   a.	Dumbo can fly.
   b.	I have to sneeze.

Anankastic modals, which represent something like conditional practical necessity, as in (18), are plausibly also in this class:

Categorization here is controversial: one might think, for instance, that anankastic modals are just restricted compulsion modals. One might also think that agentive modals are just circumstantial modals: modals which say how things could or must go given local circumstances. There seems to be a difference, however, between the agentive (19a) and the circumstantial (19b):

1. (19)

   a.	Susie can hit the bullseye.
   b.	It could be that Susie hit the bullseye.

If Susie is an untalented dart player, we may be disinclined to accept (19a), whereas (19b) still seems true, since of course she could hit a bullseye. In other words, (19b) seems to say something about mere compatibility with local circumstances, while (19a) says something stronger – something, intuitively, about Susie’s abilities.

There is a related point in the neighborhood concerning deontic modals. While there are perfectly well-formed “impersonal” deontic modal claims, like ‘There have to be 50 chairs in the living room by 5 p.m.’ (Bhatt Reference Bhatt and Harley1998), it doesn’t look like we can generally use deontic modals to simply describe preferable states of affairs which don’t involve agents. For instance, suppose that your doctor tells you that you need to go running in the sun more. The best states of affairs, then, are ones where you run in the sun. However, while (20a) seems true in this situation, (20b) seems weird (on the intended deontic reading):

1. (20)

   a.	You should run more.
   b.	It should be sunny more.

This raises an interesting possibility: perhaps practical modals always concern actions. This hypothesis would have both linguistic and philosophical significance. It would suggest there is a distinction between modals which take propositions and those which take actions as complements, a distinction which maps onto a contrast between theoretical modality – claims about possibility, necessity, and conditionality – versus practical modality – claims involving ability, permission, and so on.

In Mandelkern et al. (Reference Mandelkern, Schultheis and Boylan2017), we develop this idea by arguing that ability ascriptions depend on an underlying representation of a set of actions which we treat as practically available to the relevant agent. Our aim there is to rehabilitate (in improved form) the classical conditional analysis of ability, which says that $⌜$S can q$⌝$ is true just in case S would do q if she tried to. This kind of theory has obvious appeal: for instance, our judgments about whether Susie can hit the bullseye seem to map neatly onto our judgments about whether she would, if she tried. In fact, it’s natural to think the latter is rather unlikely, though not certainly false; and this seems true of the former, too. But, as is well known, this account faces some rather dramatic counterexamples; for instance, if you’re going to a movie, you may be inclined to say you can’t go to dinner with your friend – but of course, if you tried to go to dinner, you would have no trouble doing so (see Thomason Reference Thomason2005). These problems can be circumvented by relativizing the conditional analysis to a set of actions: we say that $⌜$S can q$⌝$ is true just in case there is an action A among the actions practically available to S, such that if S tried to do A, S would do q (see Boylan Reference Boylan2021 for more recent developments).

Whether this strategy is successful depends on whether a principled account of practical availability can be given. We make some preliminary remarks about the notion in those papers, but there is much more work to be done here. Here I would like to draw out a connection to the high-level hypothesis that there is a distinctly practical kind of modality. The picture that results here is structurally reminiscent of theories of deontic modals put forward by Cariani et al. (Reference Cariani2013); Cariani (Reference Cariani2013), which make the evaluation of deontic modals depend on a partition of logical space – a partition that we can think of as a set of actions. Likewise, in Mandelkern and Phillips Reference Mandelkern, Phillips, Maspong, Stefánsdóttir, Blake and Davis2018, we use experimental results concerning order effects to argue that ascriptions of freedom and force similarly are built on domains of actions, not just possibilities. And all this, in turn, goes naturally with standard approaches in decision theory, which take for granted a background set of actions available to the agents. It seems plausible to me that there is a potential for unification across these domains: namely, all these models of practical reasoning draw on the same core representation of practically available actions.

While I have emphasized here the distinctness of the class of practical modals, it is worth also noting here that, if our theory of agentive modality is correct, then at least that particular species of practical modality is intimately connected to our judgments about conditional facts. Along the same lines, Mandelkern and Phillips (Reference Mandelkern, Phillips, Maspong, Stefánsdóttir, Blake and Davis2018) argue that the set of practically available actions is constrained by a theoretical representation of the causal structure of a given scenario (cf. Phillips & Cushman Reference Phillips and Cushman2017; Phillips & Knobe Reference Phillips and Knobe2018 for related work on the psychological representation of modality). Causal decision theory (Stalnaker Reference Stalnaker, Harper, Stalnaker and Pearce1980b; Gibbard & Harper Reference Gibbard, Harper, Stalnaker and Pearce1981) likewise ties together practical modality – what one ought to do – with theoretical modality – what would happen if one did such-and-such.

Let me conclude this discussion by highlighting a puzzle about practical modality from Silk Reference Silk and Timmons2015, Reference Beddor and Egan2018; Mandelkern Reference Mandelkern2021b concerning orders.Footnote ²³ We can use deontic modals to give orders; we can also use other constructions, like imperatives or performatives, which are presumably closely related:

1. (21)

   a.	You have to give me your cookie.
   b.	Give me your cookie!
   c.	I order you to give me your cookie!

An important fact about giving orders is that there is nothing wrong with giving an order when you aren’t sure you will be obeyed. Susie could say any of (21a)–(21c) to John, knowing that John is very unlikely to part with his cookie. She might communicate this to an onlooker with a construction like one of the following:

1. (22)

   a.	He might not do it.
   b.	I’m not sure if he will give me his cookie.

The puzzle is that, in spite of this, there is something very weird about Susie telling John that he might not obey her at the same time that she is ordering him to give her the cookie:

1. (23)

   a.	# You have to give me your last cookie, but you might not.
   b.	# Give me your last cookie! I’m not sure if you will.
   c.	# I order you to give me your last cookie, but you might not.

The weirdness can be brought out if we contrast variants which are not used to give orders, like weak deontic modals or verbs of desire:

1. (24)

   a.	You should give me your last cookie, but you might not.
   b.	I want you to give me your last cookie, but I don’t know if you will.

Sentences which both give an order and express uncertainty about whether it will be carried out (which I call practical Moore sentences) are generally infelicitous. But this is puzzling: if there’s nothing wrong with giving an order while being unsure whether it will be carried out, what is wrong with giving an order and simultaneously saying that it might not be carried out? One thing that I want to point out here is that it’s not clear what kind of puzzle this is. Is it a puzzle about the semantics of deontic modals, imperatives, and performative sentences? About the speech act of ordering? Or about the moral psychology of ordering? Or all of these, or something else? In Mandelkern Reference Mandelkern2021b, I argue that these sentences reveal a surprising norm on ordering: namely, in giving an order, you must act towards your orderee as though they will obey you. If this is correct, it might be revealing about the structure of conversational norms more generally; whether or not this is correct, this is an area where further exploration is clearly needed.

17.5 Conclusion

I have focused on a handful of puzzles concerning epistemic modals, conditionals, and practical modals, respectively. I have just brushed the surface of a rich and enormous literature, and I have done so in a necessarily idiosyncratic and autobiographical way. By delving into the details of these few topics, I hope to have said enough to show how much interesting work has been and remains to be done here by both philosophers and linguists.