Relevance Theory, Semantic Content and Pragmatic Enrichment

Abstract

Work in the last two decades on semantics and pragmatics has given rise to a multitude of different positions on where to draw the distinction between them, whether such a distinction can be drawn at all, and to what extent, if any, pragmatics contributes to semantic content. I outline the relevance-theoretic view that the semantics-pragmatics distinction corresponds to the distinction between linguistically encoded and pragmatically provided meaning, and the reasons for the rejection of any intermediate level of semantic content such as a minimal proposition. In the second part of the paper, I survey a range of recent frameworks (indexicalism; certain versions of minimalism) that potentially avoid those objections, and consider whether these approaches motivate the need for some variety of semantic content as a psychologically real or theoretically valid level of representation distinct from encoded meaning and explicit utterance content.

Thanks to Robyn Carston, Rob Stainton and Catherine Wearing for their comments on earlier forms of much of this material, which is drawn from my University of London PhD dissertation. This work is funded by a British Academy Postdoctoral Fellowship.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction: Semantics, Truth Conditions, and Explicature

The distinction between pragmatics and semantics is widely accepted to be a distinction between, respectively, meaning that is recovered by principles or maxims of pragmatic inference, and meaning that is determined largely by linguistic mechanisms. However, the interplay of linguistic and inferential factors, particularly at the level of explicitly communicated content, has given rise to a variety of different positions on how and where to draw the semantics-pragmatics distinction.

Traditionally, the guiding assumption of many semantic theories (from Frege 1892, through Russell 1905; Davidson 1967; Kaplan 1977/1989, to Larson and Segal 1995) has been that semantics concerns truth conditions (see Carston 1999: Sect. 2 for discussion of the aims of some of these semantic programmes). Grice’s (1967) influential work showed that, in addition to what a speaker says, which is both largely conventional, and that content on the basis of which her utterance would be judged true or false, a speaker may also convey implicatures that do not affect the truth-value of what she says; these conversational implicatures are calculated by assuming that the speaker is being cooperative and adhering to certain expected standards of truthfulness, relevance, informativeness, and manner of expression (Grice’s Cooperative Principle and Maxims; see Grice 1975/1989: 26–27).¹ This suggests a natural way of drawing the semantics-pragmatics distinction: semantics would correspond to the truth-conditional content of the utterance (Grice’s what is said), and pragmatics to any conveyed meaning that falls outside the truth-conditional content (implicatures).

On this way of drawing the distinction, semantic content can be equated with Grice’s what is said, and has the following two features: it is the truth-conditional content of the utterance, and it is determined almost entirely by the conventional, encoded meaning of the linguistic expressions used. As Grice acknowledged, the truth-conditional content is not completely free of contextual input: In some brief comments on an utterance of “He is in the grip of a vice,” he says, ‘for a full identification of what the speaker has said, one would need to know (a) the identity of x, (b) the time of utterance, and (c) the meaning, on the particular occasion of utterance, of the phrase in the grip of a vice’ (1975/1989: 25). However, he did not seem to see these processes of reference assignment and disambiguation as requiring appeal to his conversational maxims; instead, the idea seems to be that they are resolved more automatically, the requisite values being something like objective features of the context of utterance (see Carston 2002: 105–106 on Grice’s appeal to a criterion of best contextual fit).

This equation of semantics with the truth-conditional content of the utterance quickly runs into difficulties when one tries to apply the semantics-pragmatics distinction to the following kinds of examples:

• 1. Anna: How’s Max doing now? Is he any better than last time I saw him?

Ben: Well, his wife’s left him and he’s started drinking again.

• 2. After he bulldozed his way past Berdych in straight sets, Nadal is everyone’s favourite to make it three grand slam titles in a row.

• 3. A: The White House doesn’t visit Tip O’Neill in his Congressional office.

B: Old grudge.

Consider, first, Ben’s reply in (1). It seems clear that he is communicating that Max is not doing well, and this communicated meaning is an implicature. What about the truth-conditional content? Grice allowed for reference assignment (and disambiguation), so context can provide a value for the pronouns “he/his/him.” But does this result in the truth-conditional content that the speaker intends to express? According to the intuitions of ordinary speaker-hearers—and, as most theorists now agree—the answer is no. The linguistically-encoded denotation of “drinking” covers events of drinking water, tea, cough syrup, and so on, but what the use of “drinking” in Ben’s reply is intended to denote is likely to be a subset of the events denoted by the linguistic meaning: the denotation is narrowed to exclude many of them and cover only those events—that is, drinking of alcohol—relevant in the context. There is also a cause-consequence relation between the conjuncts, which Grice would have treated as a conversational implicature (a generalized one: see Grice 1975/1989: 37–38), but Carston (1988) argued that such non-truth-functional meanings conveyed by “and”-conjunctions are often best treated as contributing to the truth-conditional (i.e. explicit) content. Here, this causal meaning does seem to be part of that content on the basis of which one would agree or disagree (a third speaker in the exchange in (1) could directly contradict Ben by saying “No; what happened was he started drinking and his wife left him”),² so the truth-conditional content is something like this, with pragmatically supplied elements underlined³:

• 4. Max’s _i wife has left him _i & as a result he _i has started drinking excessive quantities of alcohol.

Turning to (2), apart from reference assignment, there is a metaphorical use of “bulldozed”; as I discuss later in the chapter, many such loose uses are best seen as contributing to truth-conditional content. Other elements of this content that appear to arise here independently of semantic (linguistic) motivation or constraints include, arguably, a cause-consequence relation between the two clauses, much as in (1), and the domain of “everyone” would be restricted to something like everyone who is interested in this match. In (3B), taken from Barton (1990), the underdetermination of truth-conditional content by conventional meaning is even more extreme: linguistic decoding arguably produces just a determiner phrase, old grudge, yet the content expressed is, roughly, that is because of an old grudge. ⁴

In all three cases, the linguistic meaning, even when saturated with contextual values for the indexical elements, falls far short of determining truth-conditional content. So the problem is that the two criteria—that semantics be conventional and that it be truth-conditional—pull in opposite directions. What corresponds to the intuitive truth-conditional content of an utterance—the proposition that is explicitly expressed by the speaker—is, as illustrated in (1)–(4), often the result of extensive pragmatic development, and departs radically from any content that could be considered semantic (these apparently pragmatically motivated contributions to explicit content, known as ‘free pragmatic enrichment’, will be discussed further in Sect. 3). Thus many theorists reject the idea that the semantics-pragmatics distinction corresponds to the distinction between what is explicitly expressed (the intuitive truth-conditional content), and what is implicated.⁵ Still, there continues to be little consensus in the literature as to where the distinction should be drawn.

This chapter will take as its starting-point the relevance-theoretic view that semantics is standing (encoded) linguistic meaning (hence context-invariant), while pragmatic meaning is any meaning that is recovered by appeal to contextual (pragmatic) factors, including all those pragmatic processes that contribute to determining explicit utterance content, that is, including reference assignment and disambiguation (see in particular Carston 2002 and 2008a). In the first part of the chapter, I recap Carston’s arguments for drawing the distinction this way and against the positing of some level of semantic content intermediate between linguistically encoded meaning and explicit content (henceforth ‘explicature’), focusing on her discussion of Bach (2001) and Cappelen and Lepore (2005). In the second part (Sects. 3,4, and 5), I consider various recent approaches to semantics that potentially avoid Carston’s objections and make sense of the idea of semantic content, motivating the idea that semantics does deal in truth conditions or ‘content’ after all. The first two, which I argue do not motivate a revision of the relevance-theoretic distinction, are the ‘hidden indexical’ approach defended by Stanley (2007) and the very minimalist versions of minimalism proposed by Borg (2004) and Korta and Perry (2006, 2008). Finally (Sect. 5), I consider a version of the ‘semantic relativism’ that has gained popularity in the last few years, and provide a brief preliminary assessment of its implications for the issues addressed in this chapter.

2 Relevance Theory’s Semantics-Pragmatics Distinction

Relevance theory (Sperber and Wilson 1986/1995; Carston 2002), along with many contextualist philosophers of language such as Bezuidenhout (2002), Neale (2007), Recanati (2004) and Stainton (2006), takes the view that a theory of utterance comprehension requires only three psychologically real levels of representation. There is the linguistically encoded meaning, or ‘logical form’ of the utterance, plus the two kinds of communicated assumption introduced in the last section: explicatures, which are arrived at by pragmatically developing the logical form, and implicatures, inferred from explicatures plus contextual assumptions. Explicatures answer to our intuitions about the truth-conditional content of an utterance but, as illustrated above, may have extensive pragmatic input that goes beyond anything mandated by the linguistic meaning, a fact that argues against equating the semantics-pragmatics distinction with the explicature-implicature one. Having dropped Grice’s truth-conditionality requirement on semantics, we are left with the criterion of conventionality. The rest of the argument goes as follows. There is no level of representation that is both distinct from conventional—that is, linguistically encoded—meaning, and more closely tied to it than explicature is, nor is there any theoretical use for isolating any such entity. That leaves the distinction between encoded and inferred meaning as the only useful place to draw the semantics-pragmatics distinction (Carston 1999, 2002, 2008a).

Underlying this approach is a view of linguistic meaning on which its role is not to determine truth conditions, but to act as one clue, among others, to the speaker’s meaning. We can communicate propositional—truth-conditional—contents without using language at all, drawing on various cues (gestures such as pointing, facial expressions, other paralinguistic behaviour, the contextual salience of objects), and linguistic meaning is another of these potential cues, albeit one that allows the communication of far more complex contents than would be communicable non-verbally. There is, then, no reason why linguistic (semantic) decoding should produce something that can be given a truth value, or why it should even articulate at logical form all the elements of the truth-conditional content (leaving variables or slots to be saturated with the kinds of values specified by the linguistic meaning). As Carston (2002: 29–30) puts it, ‘underdeterminacy is an essential feature of the relation between linguistic expressions and the propositions (thoughts) they are used to express. […] public-language systems are intrinsically underdetermining of complete (semantically evaluable) thoughts because they evolved on the back, as it were, of an already well-developed cognitive capacity for forming hypotheses about the thoughts and intentions of others on the basis of their behaviour’. Although Carston is defending an ‘essentialist’ underdeterminacy position here (one shared by Recanati 1987, 1994, 1996, Searle 1978, 1980, 1992, and Sperber and Wilson 1986/1995), this evolutionary justification is also compatible with a weaker version of underdeterminacy, on which the linguistic expressions used do, generally, underdetermine the thought expressed, but ‘this is merely a matter of effort-saving convenience for speakers and another sentence which fully encodes the proposition/thought could always be supplied’ (Carston 2002: 29).⁶ The acceptance of pervasive underdeterminacy, whether one thinks of it as essential or merely convenient, is based on the recognition that we have a highly developed theory-of-mind capacity to bridge the gap between linguistic meaning and communicated thoughts, so that encoding of fully propositional forms is often unnecessary and uneconomical. Rather, the role of linguistic meaning is just to encode what is necessary, which, as illustrated particularly well by example (3) above (B’s utterance of “Old grudge”), is often only a fragment of a propositional form. As Sperber and Wilson say, ‘all that is required is that the properties of the ostensive stimulus (e.g. the utterance) should set the inferential process on the right track; to do this they need not represent or encode the communicator’s informative intention in any great detail’ (1986/1995: 254).

The idea that semantics should determine truth conditions/propositions or at least some genuine level of content, though, continues to exert a strong pull, even while many authors agree that semantics radically underdetermines the communicated content. Given the variety of pragmatic processes that are held to contribute to explicature (disambiguation, reference assignment, various kinds of entirely pragmatically-motivated enrichments and meaning modulations illustrated in (1)–(3) above), there are, clearly, several places where it would be possible to delineate a content intermediate between linguistic meaning and explicature. Two such accounts are the minimal propositions of Cappelen and Lepore (2005) and the (sometimes sub-propositional) ‘what is said’ of Bach (2001); in the rest of this section, I present these accounts and their motivations, and argue against them, drawing mainly on Carston’s (2008a) previous discussion. In the following sections, I look at some more recent approaches and consider whether they require any revision or supplementation of the relevance-theoretic picture.

According to Cappelen and Lepore (2005), semantic context-sensitivity is displayed by only a limited set of expressions consisting mostly of overt indexicals and demonstratives (Kaplan’s (1977/1989: 489) Basic Set of Context Sensitive Expressions).⁷ The result of decoding, disambiguation, and saturating these overt context-sensitive elements is the semantic content. They agree that the semantic content would not generally be what the speaker intended to communicate, and that recovering what the speaker ‘said’, or ‘asserted’, or ‘claimed’ (that is, explicature) usually requires far more pragmatic processing. However, Cappelen and Lepore appear to take the view that there is nothing enlightening to say about how explicature is arrived at, since it ‘depends on a potentially indefinite number of features of the context of utterance and the context of those who report on (or think about) what was said by the utterance’ (2005: 4). An interesting systematic account of how these myriad pragmatic features interact with linguistic meaning is seen as a hopeless prospect. What is tractable, and what semantic minimalists should limit themselves to, is to concern oneself with pragmatics only when necessary—that is, where its interaction with linguistic meaning is mandated by the linguistic meaning itself.

So Cappelen and Lepore are not in the business of explaining how we grasp speaker’s meaning—explicature—but do think that their minimal semantic content has a role to play in an account of communication, because this content—a minimal proposition—is required to explain how it is possible to ‘share content’ across contexts.

The two aspects of Cappelen and Lepore’s proposals that have been most disputed, which I’ll address here, are whether this semantic content is a proposition, and whether it is shared. First, Cappelen and Lepore’s (2005) claim that the minimal semantic content is a proposition initially seems clearly wrong: The result of decoding, disambiguation, and saturation of the overt indexicals often does not produce something that we are able to evaluate the truth or falsity of. The minimal semantic content of an utterance of “He is ready,” for example, would just be, say, John is ready. Intuitively, we need to know more than that to grasp a truth-evaluable content—that is, for a proposition to be expressed. But Cappelen and Lepore (2005: 87–112) appeal to various tests in support of their claim, one of which, the ‘Inter-Contextual Disquotational Indirect Report’ test, is illustrated here:

Context C1: In a conversation about exam preparation, someone raises the question of whether John is well prepared. Nina utters “John is ready.”

Context C2: Three people are about to leave an apartment; they are getting dressed for heavy rain. Nina utters “John is ready.”

Cappelen and Lepore claim that they can truly report that “In both C1 and C2, Nina said that John is ready.” The content shared across the two contexts of utterance (and the report) is that John is ready (punkt), and that this minimal content is a proposition is established by the fact that the report is supposedly truth-evaluable as it stands, without any felt need to ‘complete’ it by specifying what John was ready for in each context.

This test, and the conclusions Cappelen and Lepore draw from it, have already been extensively criticized (Bezuidenhout 2006; Carston 2008a; Szabó 2006, among others). Even if we agree with Cappelen and Lepore that the report is truth-evaluable, this does not show that the reported utterance—what is embedded under “said that”—itself is a proposition. Rather, the correct conclusion to draw is that the ‘content’ of “John is ready” that is shared across the various utterances is the linguistically encoded meaning, free from any pragmatic contribution. Wedgwood (2007) illustrates this by applying the test to a pair of utterances containing an ambiguous expression. Consider two utterances by Nina, in different contexts, of “John went to the bank.” Imagine we know that, in the first context, she was referring to the financial bank, and in the second, to the river bank. We can truthfully report “In both C1 and C2, Nina said that John went to the bank.” Given the use they made of this test with the utterances of “John is ready,” Cappelen and Lepore should maintain that Nina’s two utterances of “John went to the bank” had a shared content (abstracting away from tense, and treating proper names as constants). But the only possible candidate for shared content would include a disjunction of the two meanings of “bank.” The minimal proposition expressed by Nina’s first utterance was John went to bank1; that expressed by her second utterance was John went to bank2. The only level of content that is shared between the two utterances is a level prior to the recovery of the minimal proposition—that is, the linguistically encoded meaning. And as for why we judge the report (“In both C1 and C2, Nina said that John is ready/that John went to the bank”) true, I suggest that that must be because we are not entirely disquoting: we are simply judging it as a true report of Nina’s words, rather than of the contents of the two utterances.

This leaves Cappelen and Lepore with no argument for the propositionality of their minimal semantic contents, and, more recently, at least one of them appears to have dropped the insistence that this semantic content is propositional (see, for example, Cappelen 2007). A more important issue, though, is the role that they see their semantic content—propositional or not—playing in an explanation of communication. It is essential for philosophy of language to explain how content can be shared across contexts, and, according to Cappelen and Lepore, only semantic minimalism ‘can account for how the same content can be expressed, claimed, asserted, questioned, investigated, etc. in radically different contexts. It is the semantic content that enables audiences who find themselves in radically different contexts to understand each other, to agree or disagree, to question and debate with each other. It can serve this function simply because it is the sort of content that is largely immune to contextual variations’ (2005: 152).

The minimal semantic content is, then, supposed to be what we can expect people to grasp, even if they do not have enough information about the context to allow them to recover the speaker’s full message. The intuitive truth-conditional content cannot play this role, as this level of communicated content incorporates the results of too much pragmatic inference. So semantic content has an important role as a fall-back content that the speaker can rely on the hearer grasping. However, as has been pointed out (Wedgwood 2007; Carston 2008a), this is a role that the linguistically encoded meaning is far better able to play than Cappelen and Lepore’s semantic content. This is because linguistic meaning is recovered by decoding, an automatic, algorithmic process, in contrast with pragmatic inference which, by virtue of being a process of hypothesis formation and confirmation, always takes place at some risk. And, since Cappelen and Lepore’s semantic content is partly, and often largely, pragmatic—being the result of decoding plus the pragmatic processes of disambiguation and saturation⁸—speaker and hearer may diverge substantially. The only content that can be guaranteed to be shared is the linguistically encoded meaning; any content that incorporates the results of pragmatic inference cannot serve as a fall-back content that the speaker can rely on the hearer grasping if the hearer lacks knowledge of the context. So this version of semantic minimalism makes no progress on how we manage to faithfully share content well enough to enable us to debate, investigate, hold people responsible for what they say, and so on.

Cappelen and Lepore’s semantic minimalism is one of the recent varieties of what Recanati (2004: 51) calls Syncretism, a view that is a compromise between ‘Literalism’—by which Recanati seems to mean Grice’s view, on which what is said departs as little as possible from ‘literal’ (i.e. encoded) meaning—and the contextualist and relevance theoretic approach on which what is said is much more pragmatically developed. Syncretism considers both these notions of what is said to be legitimate, giving a four-level picture, with two levels of ‘literal meaning’—sentence meaning and the minimal what is said—and two levels of speaker’s meaning—the pragmatic what is said (explicature), and conversational implicature.

Several philosophers have defended this four-level picture: Soames (2002) contrasts what is said in the semantic sense with what is said in the pragmatic sense, which is what the speaker states or asserts, while Salmon (1991: 88) distinguishes ‘saying or asserting in the strict and philosophical sense’ from ‘saying in the loose and popular sense’; the latter ‘what is said’ is the content of the speaker’s speech act, but does not cover what is merely implicated in Grice’s sense. Both these authors’ conceptions of the semantically expressed proposition include the results of saturation, like Cappelen and Lepore’s, so are susceptible to the same objection: as Stanley (2007: 233) says in his review of Recanati (2004), ‘The central problem for the Syncretic View is that the notion of semantic content appealed to in the theory threatens to be an idle wheel in an explanation of linguistic practice’.

A different version of Syncretism, however, is defended by Bach (1994, 2001), who, while sharing with Cappelen and Lepore and Soames the view that a semantic content is needed as a shared fall-back content, delineates semantic content differently than other syncretists. He distinguishes a level of what is said, separate from explicature (which he calls impliciture⁹) and implicature, on the basis of the kind of context needed to derive semantic content versus other contents. I consider here just his (2001), which represents his latest view. For discussion of his (1994)—similar to Cappelen and Lepore (2005) except that he does not insist that semantic contents are propositions—see Carston (2002; Sect. 2.5).

According to Bach (2001), ‘what is said’ results from saturating ‘pure’ indexicals, which are expressions such as “I,” “today,” “here,” and “now.” The result of saturating other indexicals (“he,” “that,” “we,” etc.), and of ‘completion’ and ‘expansion’¹⁰ is impliciture. What is said may be subpropositional (a ‘propositional radical’), and so need not be a communicated assumption. Within communicated assumptions, Bach recognizes the same binary distinction as relevance theorists and other contextualists, between developments of logical form (Bach’s impliciture, RT’s explicature), and conversational implicatures.

Bach delineates his ‘what is said’ on the basis of the kind of context necessary for recovering it: It consists of linguistically encoded meaning plus saturation of pure indexicals, whose referents, supposedly, can simply be read off of the context—e.g. “I” is automatically assigned the speaker as referent; “now” is assigned the time of utterance, and no pragmatic inference comes in. Only ‘narrow’ context is required—that is, the objective features of the context, such as speaker, addressee, location and time of utterance. Pure indexicals contrast with other indexicals and demonstratives, such as “he” and “that,” which require consideration of speaker intentions and other contextual features (‘broad context’).

Bach’s reason for drawing this distinction is to single out ‘the linguistically determined input to the hearer’s inference to what, if anything, the speaker intends to be conveyed in uttering the sentence’ (2001: 15). The thought here is that, since the value of a pure indexical is information that simply ‘arises from the act of uttering’ without consideration of speaker intentions, it can be considered the linguistically determined input, and can be guaranteed to be shared among interlocutors.

This appears, initially, to be a shared content that can serve as the input to pragmatic reasoning and that avoids the problems of Cappelen and Lepore’s approach by distinguishing between two types of indexicals (and the two types of context that their saturation requires). But the problem with this is that narrow context is not enough to fix the value of “here” (which the speaker could be intending to use to mean in this room, in this building, in this city, or any number of other values) or “now” (in the 21st century, at 2 pm on August 30 2008?). The idea that narrow context is sufficient is only remotely plausible for “I,” but even with “I,” there are several kinds of cases for which pragmatic inference, and appeal to wide context, is necessary for working out the referent. Cases familiar from the literature include these:

• 5. Professor X is out of his office. A colleague, seeing students calling at Prof X’s office to find him, writes a note saying “I am not here now” and sticks it on Prof X’s door (Predelli 1998).

• 6. George Bush utters “The founding fathers invested me with the power to appoint members of the Supreme Court” (Nunberg 1993).

So, given that even assigning reference to “I” can involve appeal to broad context, the most natural move would be to accept that the linguistically determined input is just the standing—context-invariant—linguistic meaning, and class “I” together with the rest of the indexicals.

After introducing the relevance-theoretic distinction between semantics and pragmatics, in the rest of this section, I’ve considered the following two alternative ways of drawing the distinction:

• (a) Semantics as linguistically encoded meaning plus contextual values for all indexicals (i.e. semantics as ‘minimal proposition’); pragmatics as the rest of speaker meaning.

• (b) Semantics as linguistically encoded meaning plus contextual values for pure indexicals; pragmatics as the rest of speaker meaning.

The discussion so far clearly supports Carston’s conclusion that ‘not only are these not good ways to draw the distinction between semantics and pragmatics, they are not worthwhile distinctions of any other sort either; that is, they do no useful work’ (2008a: 322). However, several other recent theories avoid the objections discussed above. In the next sections, I look at these in turn, and consider whether they motivate a revision to the relevance-theoretic distinctions, by justifying the positing of a level of genuine content (as opposed to linguistically encoded meaning) that is semantic.

3 Covert Indexicals versus Free Pragmatic Enrichment

The first of these views is in much the same spirit as Cappelen and Lepore’s minimalism, in that it sees semantic content as being propositional, and resulting only from decoding and disambiguation plus saturation of all indexicals. Where it differs from minimalism is in its claim that what the semanticist should be trying to account for is the intuitive truth-conditional content. On this, it agrees with relevance theorists and contextualists that our intuitions are evidence that the content on which these truth-value judgments are based is a psychologically real level of representation, while there is no evidence for the existence of the more minimal content posited by Cappelen and Lepore and other syncretists.

Recall that the two Gricean criteria on semantic content (Grice’s what is said) were that it be conventional (departing little from linguistic meaning, allowing only for saturation and disambiguation), and that it be the speaker-meant truth-conditional content. These two criteria proved contradictory, and one had to be dropped, as a host of examples—such as (1)–(3) in Sect. 1—showed that (intuitive) truth-conditional content incorporated the results of more pragmatic processes than just saturation and disambiguation. Another example that is central to the debate addressed in this section is (7):

• 7. It’s raining.

The truth value of weather reports is agreed to depend on the location that the weather in question is supposed to occur at, so (7), when appropriately contextualised, would have the explicature It is raining in London. This location constituent appears to be provided entirely pragmatically, on grounds of relevance; if so, then this example would reinforce the message of the above examples, that semantic content cannot be both conventional and truth-conditional. But the approach that I discuss here manages to reconcile these apparently incompatible requirements.

This approach claims that explicature is fully linguistically articulated, and the only pragmatic processes affecting it are linguistically mandated. Stanley (2000: 391) states the view thus: “all truth-conditional effects of extra-linguistic context can be traced to logical form.” Given the agreement on what the object of explanation is—the non-minimal, intuitive conception of truth-conditional content—the challenge for the defender of this view is to account for the elements of explicature that do not appear to be the values of anything in the overt linguistic form. His explanation is that there are covert indexicals attached to certain lexical items, which are, therefore, present in logical form whenever the item is tokened. For instance, weather verbs might encode a location variable, and, on Stanley and Szabó’s (2000) proposal, every nominal encodes a pair of domain indices, which accounts for domain restriction and the completion of definite descriptions, and so on. (8)–(10) show, in rough form, the kinds of structures envisaged:

• 8. It is raining <at Loc L>

• 9. The <candidate, f(i)> was late

• 10. Every <bottle, f(i)> is empty

If this approach could be extended to account for all effects of context on explicature, then, apart from disambiguation, the only pragmatic process involved in getting from logical form to explicature would be saturation of (overt and covert) indexicals: In accordance with what Stanley suggests in the above quote, there would be no contribution from pragmatic processes that are not linguistically mandated.

The motivation for this approach, which I label here ‘indexicalism’, is as follows. The traditional, straightforward account of how an utterance’s content is grasped is that we work out the semantic values of the expressions used (decode and disambiguate; assign referents to indexicals, demonstratives, tense indicators, etc.) and combine these values according to the rules of semantic composition that are part of our linguistic knowledge. The attractiveness of such a systematic account of semantic content is at least partly what motivates many of the minimalist approaches discussed in this paper. If it could be extended to the intuitive, rather than minimal, content, then it would enable a systematic and constrained account of the relation between utterances and the propositions they explicitly express. But if free—that is, non-linguistically mandated—pragmatic processes ‘intrude’ on explicature, then the prospect of a systematic account of how we get from linguistic meaning to explicature diminishes. The indexicalist is sceptical that these alleged free pragmatic processes can be shown to be adequately constrained by purely pragmatic factors, and claims that they would massively overgenerate interpretations of utterances at the level of explicature. For example, Stanley (2005: 225–226) asks, if free pragmatic enrichment can supply the quantifier domain so that an utterance of “Every Frenchman is seated” can have the explicature Every Frenchman in my phonology class is seated, what prevents it from enriching the utterance in a different way, e.g. to Every Frenchman or Dutchman is seated? This is an interpretation which, according to Stanley, is impossible but is predicted to be able to occur by free pragmatic enrichment.

I do not address this systematicity objection in this paper; for more detail, see Stanley (2002, 2005), and for responses to it, see Carston and Hall (in preparation), Hall 2008, 2009, 2014, Recanati (2010: introduction and Chap. 1). My concern with it here is its role in justifying the idea that explicature—intuitive truth-conditional content—should correspond to semantic content and so requires the positing of extensive covert indexicality in the linguistic logical forms of sentences. This approach, in principle, promises to better substantiate the idea of semantic content than do the minimalisms discussed in Sect. 2, as it suggests that what is agreed to be a psychologically real level of representation—that level to which our truth-value judgments about the utterance are responsive—is the result of only linguistic decoding plus processes that are mandated and constrained semantically.

In practice, though, for this approach to work requires all pragmatic contributions to explicature to plausibly be traceable to linguistic form. Otherwise, what can be considered the semantic content (the product of decoding, disambiguation, and saturation) will fall short of explicature—a result that undermines both of the justifications for positing it (that it is a psychologically real level of content, and that free pragmatic effects on explicature must be excluded thanks to their unconstrained nature). Below, I present a range of types of pragmatic contributions to explicature for which, I argue, it is not plausible that they are underpinned by covert structure.

Before that, it should be noted that there does seem to be linguistic evidence that a small number of pragmatic effects on explicature are linguistically mandated by a variable in the linguistic meaning. This evidence concerns relational terms such as ‘local’, ‘home’, ‘enemy’, ‘foreign’:

• 11.a. [Every reporter]_i was sponsored by her_i local bar

• b. *Her_i local bar sponsored [every reporter]_i

• 12.a. Every reporter was sponsored by a local bar

• b. *A local bar sponsored every reporter (Carston 2002: 200)

The expression “local” in (12) seems to behave syntactically very like the overt pronoun “her” in (11), giving rise to so-called ‘weak crossover effects’: neither expression, in the (b) sentences here, can be bound by the quantified phrase “every reporter,” and Carston (2002) and Recanati (2004) agree that these are plausible cases of covert variables (though for some doubts about the weak crossover evidence, see Pupa and Troseth 2011). Other likely cases are “ready” and “enough,” which seem inherently underdetermined, and require a specific value (ready for x, enough of y) to be provided on every occasion of utterance. A linguistic variable mandating saturation, then, may well be the right analysis of these expressions.

But the expressions for which there is such evidence are very limited in number, leaving the indexicalist with no linguistic evidence for many of the hidden indexicals he is positing.¹¹ The strongest remaining argument for his overall approach is his pessimism about the prospects of a sufficiently constrained account if one posits free pragmatic enrichment. Providing a positive account of how free enrichment is constrained is beyond the scope of this paper; instead, what I do in the rest of this section is show that, if that concern were valid, then the indexicalist would be just as susceptible to it as the pragmatic enrichment approach, because there are several kinds of contextual contribution to truth conditions that are not plausibly traceable to linguistic logical form. In the rest of this section, I present a selection of the examples used by Hall (2008: Sect. 3) to make this case.

The first phenomenon I’ll consider is what Stanley himself acknowledges is the central worry for the indexicalist, which is deferred reference, or metonymy, illustrated by (13a) [with the explicature given in (13b)] and (14):

• 13.a. The ham sandwich wants his bill.

• b. The person x who ordered the ham sandwich wants his _x bill.

• 14. I’m parked out back.

Following Nunberg (1995), most authors accept that the deferred content [e.g. person who ordered the ham sandwich in (13)], rather than the encoded, or ‘literal’, meaning, contributes to the explicature of the utterance. Supporters of this approach include those on both sides of the present debate: Recanati (1993, 2004), Carston (2002), Sag (1981), Stern (2000, 2006), and Stanley (2005). Stanley also agrees with relevance theorists and contextualists in dismissing the idea that there is a covert metonymy operator or indexical in logical form, on the model of Stern (2006)’s metaphor operator. One reason that a metonymy operator/indexical is unappealing is that it would place no constraints on the sort of deferred interpretation the hearer arrives at, because the relation between literal and deferred meaning is different from case to case. Consider just the two examples above: (13) requires a move from reference to a culinary item to reference to its orderer; (14), from a property of a car to a property of a person. As Stanley (2005: 229–230) says, beyond the fact that the property encoded by the metonymically used expression must provide a guide of some sort, the content is determined entirely by pragmatics, and the syntax/semantics places no constraints on what the pragmatics can do. A metonymy operator would be, therefore, redundant. A second reason for rejecting the metonymy operator approach is that to account for such figurative effects semantically would require a massive multiplication of linguistic operators for which there is no evidence: Almost every simple or complex expression can be used in a variety of figurative ways, so practically every word and phrase in the language would require not just an operator for metonymy, but also one for metaphor, and perhaps for other figures such as hyperbole and meiosis. In summary, Stanley agrees with the defenders of pragmatic enrichment that metonymy contributes to truth conditions, and that it is not underpinned by a linguistic operator, and this amounts, I think, to a concession that this is a genuine case of a free pragmatic effect on explicature.

One might feel uncomfortable relying on figurative cases to support pragmatic enrichment, feeling that, in ‘normal’, non-figurative speech, the indexicalist thesis should hold. But, as I discuss below, there are many kinds of non-figurative utterances to make the case, which metonymy, therefore, simply reinforces.

One set of examples where truth conditions and semantics diverge are (15a)–(17a), where the overt meaning alone can determine a proposition (modulo saturation), but this is clearly not a proposition that the speaker wants to communicate, or that the hearer recovers [likely explicatures are given in (b)]:

• 15.a. It will take time for that cut to heal.

• b. It will take a long time for that cut to heal

• 16.a. Mary has a brain.

• b. Mary has a very good brain

• 17.a. You’re not going to die.

• b. You are not going to die from that scratch.

The propositions determined by the linguistic meaning of (15a) and (16a) are trivially true, while (17a), taken literally, is patently false. Utterances of such sentences, though, are not perceived as obviously non-literal and are accepted as non-trivially true in appropriate contexts, so the extra elements of meaning, such as from that scratch in (17), are unlikely to be mere implicatures. Rather, what forms the intuitive truth-conditional content in each case is (b), on the basis of which the utterance would be evaluated, agreed/disagreed with, etc. (e.g. in reply to (15a), one might say “No it won’t, I’m having the stitches out tomorrow”). However, hidden indexicals or other covert elements are highly improbable in these cases. The interpretations of “time” and “brain” in these utterances are not instances of domain restriction, so the domain variables allegedly attached to the nominals would not account for these effects, and the meaning of “die” seems very unlikely to encode a variable for ‘cause of death’. That these are highly occasion-specific effects of context makes it very difficult to accept that they should be provided for in the linguistic meaning by an indexical triggering mandatory saturation (which would presumably have to occur even in the many cases where the provision of a value would be redundant). These examples look like further cases of genuine free pragmatic effects on truth-conditional content, which cannot be traced to an element of the syntax or semantics.¹²

Finally in this section, I look at a construction that both camps in the debate have discussed in detail, and argue that the indexicalist solution proposed so far is inadequate. An utterance of (18) would often be understood as communicating a causal relation between the events referred to by the conjuncts:

• 18. If Hannah insulted Joe and Joe resigned, then Hannah is in trouble.

Since the causal connection falls within the scope of the operator ‘if…then’, practically everyone agrees that it contributes to truth conditions. King and Stanley are no exception here: they write: ‘[(18)] seems to express the proposition that if Hannah insulted Joe and Joe resigned as a result of Hannah’s insult, then Hannah is in trouble’ (2005: 158). This requires, for them, that the linguistic meaning of the sentence contain some element that can take as a result as its context-specific value. Here, they appeal to Stalnaker (1999)’s work on the semantics of indicative conditionals. According to Stalnaker, an indicative conditional is true if and only if the consequent is true in every one of the most relevantly similar worlds in which the antecedent is true, and this similarity relation counts only those non-actual worlds compatible with the mutually accepted background assumptions as similar worlds for purposes of semantic evaluation (King and Stanley 2005: 154). If this is going to help King and Stanley account for examples like (18), the idea has to be that the conditional structure would come with a linguistic parameter, requiring saturation, which specifies the similar-worlds constraint, thus requiring the selection of the most relevantly similar worlds in the context set. In a context in which ‘the speaker has in mind a causal relationship’ between the events described in the conjuncts, the most relevantly similar worlds will be just those worlds in which that causal relationship holds. This predicts the reading of (18) on which a causal relation is part of the truth-conditional content (ibid: 160).

One important problem with this solution is this (see Hall 2008: Sect. 3 for others): Since it is the semantics of the conditional that is supposed to account for the incorporation of the causal relation into truth conditions, it follows that, for King and Stanley, the explicature of the unembedded conjunction (19a) does not include the causal relation. Instead, this would be conversationally implicated, and is, according to them, calculable from the explicature using Gricean maxims:

• 19.a. Hannah insulted Joe and Joe resigned.

• b. Explicature: Hannah insulted Joe & Joe resigned.

c. Implicature: Joe resigned because Hannah insulted him.

Is this supported by our intuitions about truth conditions? I don’t think so, but intuitions are perhaps not entirely clear with isolated utterances such as (19a). They seem to be sharper when (19a) is presented as one premise of an argument, as in (20), which looks like an obvious case of modus ponens, and which most people would judge to be a valid argument:

• 20.a. If Hannah insulted Joe and Joe resigned, then Hannah is in trouble.

• b. Hannah insulted Joe and Joe resigned.

• c. Hannah is in trouble.

For King and Stanley, the proposition expressed by (20b) must have a different propositional form than that of the antecedent of the proposition expressed by (20a). That is, the argument is not in the modus ponens form ‘If P then Q. P. Therefore Q’, but is rather ‘If P then Q. R. Therefore Q’, which is not a valid inference. So, the indexicalist account makes a wrong prediction.¹³

This outcome does not sit well with King and Stanley’s concern to respect intuitions about truth conditions, which are what they take to be the primary object of semantic theorizing (see, especially, Stanley and Szabo 2000: 240 and King and Stanley 2005: 141). They offer no explanation for why our intuitions about the validity of the argument in (20), which depend on the intuitions about the truth conditions of (18) and (19), should not be respected, but are forced into this position by the fact that none of the overt expressions in the unembedded conjunction are plausible candidates for carrying an appropriate hidden indexical or a semantic parameter that could pick up a similarity relation. So (19) is another case where an aspect of intuitive truth-conditional content cannot be accounted for linguistically but seems to rely on a process of free pragmatic enrichment.

In conclusion, then, there are several types of case where context clearly does affect truth conditions, but where this effect cannot be accounted for by any linguistic trigger requiring contextual contribution. In other words, there are elements of meaning that, according to the truth-conditions criterion shared by both camps, are part of explicature, but that cannot be considered semantic.

The indexicalist approach denies the existence of free pragmatic effects on explicature—a position motivated by the suspicion that the only way that a systematic, explanatory account of our grasp of explicit utterance content will be tractable is to assume that it is semantic content: that is, that the role of the linguistic meaning is to provide the truth-conditional content of the utterance, modulo contributions that are triggered and constrained semantically. The examples surveyed in this section, though, have shown that covert indexicality, while plausible for certain phenomena such as relational expressions and perhaps even domain restriction, cannot be as extensive as is required to support the indexicalist thesis. Ultimately, then, this approach does not rescue the idea of semantic content.

4 Genuinely Minimal Minimalisms and Multipropositionalism

Having discussed in Sect. 2 some views that posit minimal semantic contents, which are not the ‘enriched’ contents (explicatures) that speakers communicate, but which also differ from standing linguistic meaning in that some or all of the overt indexical and other contextual elements are saturated, the conclusion was that no convincing case has been made for such a level of content: It is an ‘idle wheel’ in an account of communication, and any semantic (or semantically-mandated) content falls well short of the communicated content. Nonetheless, there is an enduring feeling that semantic theory should deal in truth conditions and propositions—that the function of linguistic meaning is to determine these, even if they are not what are judged the intuitive truth conditions of the utterance. Despite the arguments presented in the last two sections, this view can be reconciled with the relevance-theoretic position on the semantics-pragmatics distinction. What are called for here are genuinely minimal propositions, determined solely by standing linguistic meaning. I discuss two recent accounts that posit such entities—Borg (2004), and Korta and Perry (2006, 2008).

Borg (2004) presents a semantic theory that she sees as meshing with Chomskyan/Fodorian views on mental architecture. These views hold that the perception and language encoding/decoding systems are modules, translating visual, linguistic, and other perceptual signals into representations in a common format—a language of thought or ‘Mentalese’—which these modules deliver to the central system (or systems¹⁴), which is where reasoning processes, including pragmatic inference, take place. (Language, of course, is also an output system, encoding central-systems representations). These peripheral input systems are ‘encapsulated’, that is, they do not have access to information in the central system (or in the other input systems), so a given input into one of these modules will always deliver the same output to the central system.

Semantics deals with linguistic decoding, so must, for Borg, be part of the language faculty, encapsulated from central-systems processes. This means that no pragmatic inference can affect semantics: Semantics can only be that portion of communicated meaning that is formally tractable—that can be described without any influence from pragmatics. As discussed above (Sect. 2 on Bach 2001), and as Borg agrees, assignment of reference to indexicals is not formally tractable, because of its inevitable dependence on recognition of speaker intentions (she considers and rejects various ways of making it so in Chap. 3 of her 2004 book). So Borg concludes that semantics is purely the result of linguistic decoding, and does not incorporate anything else often included under the label ‘semantics’, such as assigning referents (not even to “I”).

This much is something that relevance theorists and many contextualist philosophers (in particular Carston, Recanati, Sperber and Wilson) would fully endorse. The difference between Borg and these authors is that she assumes the output of semantics must be propositional. Indexicality is the obvious problem for this view, since, as she argues at length, the output of a modular semantics cannot include the results of reference assignment: In the case of indexicals, semantics would just give the Kaplanian ‘character’ (Kaplan 1977/1989), which is their context-independent standing meaning (for example, that “I” refers to the person who utters it). Kaplan distinguishes ‘character’ from ‘content’: the former is not a constituent of a proposition; instead, it is an indication of how to recover the content (the referent), which is what does form a constituent of a proposition. So it would appear that the result of semantics cannot be a complete proposition: it contains variables with attached instructions or constraints, specifying what kind of value should saturate it (for instance, the result of decoding “he” would be something we can represent as [X_{male, singular, animate}]—a placeholder for the concept that will be pragmatically supplied, where the ‘X’ is a variable or slot requiring saturation).

Borg’s solution is to suggest that the output of linguistic decoding in the case of indexicals and demonstratives is a syntactically-provided singular concept. So the result of decoding an utterance of “That is red” is α is red. α is a singular concept, the semantic content of “that,” and comes with a constraint on how it is to be integrated with the hearer’s ‘wider cognitive environment’—an instruction for what sort of concept is to be pragmatically supplied to grasp the content that the speaker intends to assert. In effect, the semantic content of an indexical or demonstrative on this proposal is the character of the expression under a Kaplanian dthat operator. For example, the semantic content of “That is red,” α is red, can also be characterized as [dthat (salient object) is red].¹⁵ Although, as a result only of semantic processing, the hearer will not be able to work out the intended referent (and may not be able to assign a referent at all if he lacks sufficient knowledge of the context), the syntactically-tokened concept is the kind of thing that can have content; it is object-dependent. As the utterance has object-dependent truth-conditions, the semantic content that the hearer recovers counts as being truth-evaluable (propositional) even if the hearer is not in a position to know what the truth conditions are.

It is a contentious part of Borg’s view that these very ‘thin’ singular concepts, which result purely from semantic decoding processes, can endow the semantic content with truth-conditions and propositionality: one could just as well say that the subpropositional logical form assumed by relevance theorists has truth conditions and genuine content in this sense.¹⁶ The reason for her insistence on the output of semantics being propositional is not clear; that aside, though, her approach seems compatible with that of relevance theory, drawing the semantics-pragmatics distinction in the same place and recognizing that there may be a wide gap between the output of semantics and the proposition explicitly expressed. And, in fact, her thin, singular concepts might turn out to be a useful way of cashing out the relevance-theory talk of indexicals triggering slots, or placeholders for concepts, as suggested by Carston (2008b).

How Borg’s minimalist semantics differs from the RT picture, then, is not that it posits some extra level of semantic content between encoded linguistic meaning and explicature, but that it gives a slightly different, but probably compatible, account of how the linguistic meaning is represented and integrates with the pragmatic inference system. On the RT view, linguistic decoding produces an incomplete propositional template or schema, with a number of slots corresponding to indexicals and demonstratives (plus similar elements such as tense operators). Each slot consists of a variable-like element with constraints on what sort of concepts are to be supplied as their value. Borg’s approach to indexicals and demonstratives suggests a different way of thinking about what their decoding results in. This variable-plus-constraint complex is what an indexical or demonstrative is likely to be when considered as part of our linguistic knowledge, independent of any utterance of it—the format in which its meaning would be stored in the lexicon. However, when an indexical is uttered the result is not that the standing meaning, which is not a concept, simply appears unaltered in the output of semantics, but rather that the act of utterance brings about the tokening of a singular concept, which carries with it the constraint (or ‘character’) of the indexical. This seems to provide a plausible alternative to the conception of logical form so far assumed by relevance theory.

A different kind of genuinely minimal propositional content, directly determined by the rules of the language, completely independently of the speaker’s meaning, is the ‘reflexive’ proposition (cf. Perry 2001). It is ‘reflexive’ because it makes reference to the utterance itself in describing the truth conditions: an utterance u of “I am here now” expresses the reflexive proposition that the utterer of u is at the place u is uttered at the time u is uttered. As Recanati (2004: 66–67) says, this way of preserving the notion of ‘what the sentence says’, in a purely semantic sense, does not support the syncretic view (e.g. Cappelen and Lepore, Soames) with its four levels, because the reflexive proposition ‘does not incorporate those contextual ingredients whose provision is linguistically mandated; […] it is directly and immediately determined by the linguistic meaning of the sentence.’ So we appear to retain the RT picture with its three levels of content: the standing linguistic meaning (and the reflexive proposition it directly determines), explicature, and implicature.

Korta and Perry (2006, 2008), however, criticize the RT and contextualist picture, because, as they illustrate, often what would seem to be the explicature is not what serves as input to the hearer’s inference to implicatures. What is required instead may be one of various levels of partly token-reflexive or descriptive content that they propose,¹⁷ from (and including) the reflexive proposition up to (but not including) the ‘proposition expressed’ by which they mean the pragmatic what is said, or explicature, incorporating the result of saturation, disambiguation, and enrichment. It is this proposition, as opposed to the partly token-reflexive and descriptive propositions, that they see as the ‘official content’. Korta and Perry work through several examples to demonstrate how these propositions of varying degrees of token-reflexivity enter into comprehension: In some cases, one of these propositions might have a role to play when there is some difficulty in recovering the proposition expressed, but in other cases, it seems that a partly token-reflexive proposition would be equally or even better suited as input to the hearer’s inference to implicatures than the fully saturated (enriched, etc.) proposition expressed would be. This initially looks to motivate the idea of more levels of content in between standing linguistic meaning (plus fully reflexive proposition) and explicature.

Here are some of Korta and Perry’s examples. They consider Grice’s (1975/1989: 32) example of the motorist (A) who, having run out of petrol, is standing by his car and is approached by B:

• 21. A: I’m out of petrol.

• B: There’s a garage round the corner.

They concentrate on how B would interpret A’s utterance, of which they write, ‘It is natural to take A’s opening remark as implicating that he would like some help in finding petrol for his car’. Then they ask us to suppose that A is Harold Wilson,¹⁸ and continue:

• According to the theories of names and indexicals that are now widely accepted, A would have then expressed the same proposition in this scenario [i.e. 21′ below], the singular proposition individuated by Harold Wilson and the property of being out of petrol.

• [21']. A: Harold Wilson is out of petrol.

• B: There is a garage round the corner. (Korta and Perry 2006: 169)

That is, the same proposition is expressed by A in both scenarios (21) and (21′), but, while B’s utterance in (21) is an appropriate thing to say in reply, Korta and Perry comment that ‘In scenario [21′] there is no motivation for B’s remark. What does the proximity of a garage to the participants in the conversation have to do with Harold Wilson’s being out of petrol?’ (ibid: 170)

Korta and Perry’s point is this: It is the implicature of A’s utterance in (21), that A would like help finding petrol, which prompts B’s reply. But this implicature is not inferred from the proposition expressed by A’s utterance. If it was, then A’s utterance in (21′), since it expressed the same proposition, should have led to the recovery of the same implicature, to which B’s reply would be an appropriate response. But B’s response in (21′) does not seem to be motivated. So the implicature of A’s utterance must have been inferred from something other than the proposition expressed. For this, they suggest, what the hearer (B) starts with is the reflexive proposition, the speaker of u is out of petrol at the time of u, from which, together with the contextual information that the speaker is the person B is now looking at, he infers a proposition the person I am now looking at is out of petrol now. This proposition is something less than the ‘official’ proposition expressed, in that it doesn’t have Harold Wilson assigned as referent of “I” (or a specific time assigned to the tense indicator): But it’s this description of the proposition expressed that is the basis for inferring the implicature, and thus for B’s helpful reply.

A second example has ‘JP’ saying to ‘KK’, while KK is driving,

• 22. He is going to drive his car into yours.

They say, ‘JP is, let us imagine, referring to FR, a famous philosopher, who is careening down the street in the opposite direction. There is a pretty clear implicature that KK would do well to engage in evasive maneuvers, to avoid getting hit. But how is KK supposed to figure this out?’ The mode of presentation under which KK needs to be thinking of the person JP is referring to, in order to figure out the implicature, is something like The person driving that car that is weaving across the street. As Korta and Perry say, ‘To take proper evasive action, it is unnecessary, and perhaps counterproductive, to recognize JP’s referent as the famous philosopher FR. KK might become awestruck [etc.]’ (ibid: 180–181).

The apparent lesson of these examples (and others given in Korta and Perry 2006, 2008) is that we need to recognize various levels of propositional content somewhere between the reflexive proposition (equivalent to standing linguistic meaning) and the official proposition expressed. These in-between levels are utterance-bound and descriptive to varying degrees: rather than having the referents assigned to the indexicals, demonstratives, and tense indicators, what hearers sometimes need in order to work out the implicatures, and respond appropriately, is a proposition with, e.g. “I” assigned the value the person standing in front of me, or “that” assigned the value the thing the utterer of u intended to refer to. Does this require a revision of RT’s three-level picture, which has linguistic meaning being developed into explicature (also known as the proposition expressed, what is said, etc.), with no levels of semantic or sentence content in between?

Despite initial appearances, it does not undermine the RT semantics-pragmatics distinction presented earlier, because RT takes the explicature to be, roughly, the representation that the hearer does in fact entertain as the development of the linguistic meaning, and this is different from what Korta and Perry describe as ‘the (official) proposition expressed’ or ‘official content’. Recall their discussion of Grice’s example, involving an utterance of “I am out of petrol.” They claimed that, if the speaker is Harold Wilson, then the proposition expressed is the singular proposition individuated by Harold Wilson and the property of being out of petrol. That may be the case, but it is not relevant to an account of how what is communicated here is recovered. In the sense in which Korta and Perry are using the notion, the ‘proposition expressed’ by A’s utterance is Harold Wilson is out of petrol, by virtue of Harold Wilson being the speaker, irrespective of whether the hearer even knows the speaker’s identity. Let’s assume that the hearer does not know who the speaker is (and that the speaker has no reason to assume that he does). Then the official proposition expressed might be as Korta and Perry claim, but this is not what the hearer recovers, or is expected to recover, as the explicature of the utterance. The explicature is not the (external) content, but the representation that hearers recover of that content (which fulfils the speaker’s communicative intention).

In example (21), then, the explicature of A’s utterance is just what Korta and Perry say is the proposition that forms the basis for the hearer’s inference to the implicature: it can be glossed as something like The person I am now looking at is out of petrol now, though may be better represented as α is out of petrol now, with ‘α’ being a singular concept with the relevant person as content, and with a mode of presentation that can be paraphrased as ‘the person I am now looking at’. It might be that, in some cases, two or more such propositions, of different degrees of utterance-bound-ness/descriptiveness, are required as the basis for inference to different implicatures of the same utterance, or that the hearer would probably represent the full referential content but also a less than fully referential content. In such cases, relevance theorists would simply say that the utterance has more than one (basic¹⁹) explicature, which are all speaker-meant, unlike the semantic contents posited by Cappelen and Lepore, Soames, and Bach as intermediate between linguistic meaning and speaker meaning.

To summarize this section, both of the views discussed here are compatible with relevance theorists’ and other contextualists’ views on the semantics-pragmatics distinction described in Sect. 2. Borg draws the distinction in the same place as relevance theory, but provides a slightly different take on the nature of the representation of the linguistically encoded meaning than that assumed so far by relevance theory and contextualists; it remains to be seen which alternative is correct. Korta and Perry highlight the need to recognize that explicatures need not be ‘fully’ enriched, saturated, etc., but that any of a number of different varieties of representation of different degrees of utterance-boundness and descriptiveness may be the explicature(s).

The approaches discussed in this paper so far do not provide any grounds for revising relevance theory’s semantics-pragmatics distinction or justify a level of semantic content distinct both from linguistic meaning (and propositions directly and immediately determined by it) and from explicature. However, a position that remains to be considered in relation to these questions is the relativism about utterance content that has sparked intense debate in the last few years. I leave for another time a detailed comparison of relativism with relevance theory and contextualism. In the following, final, section I just present some of the general features of relativism(s), drawing mainly on the ‘moderate’ version defended by Recanati (2007), and make some brief preliminary remarks to try and assess its implications for the issue discussed in this paper.

5 Moderate Relativism, Semantic Content, and Unarticulated Constituents

The central idea of semantic relativism is that certain elements that appear to contribute to the truth-conditions of utterances, so on a relevance-theoretic or contextualist account would probably be considered part of explicature, are better treated as lying outside of that content. Rather than being implicatures, though, they are parameters in the circumstances of evaluation or context of assessment, against which the content is evaluated for truth. Common examples are taste predicates (“Seaweed is delicious”), knowledge claims (“I know that my car is on the drive”), and epistemic modals (“Fred might be in Boston”). Assuming the values of indexicals are fixed, the content of each of these utterances does not vary across contexts; instead, what varies is a standard of taste or of knowledge. When Mary utters sincerely “Semantics is fun,” the content is true relative to Mary’s taste; when Bob replies “Semantics is not fun,” the content is true relative to Bob’s taste. This explains why Mary and Bob can give the appearance of disagreement, while one would not want to say that either of them is at fault.²⁰

Here, I look at a variety of relativism that is relevant to the issues discussed in this paper because it directly addresses the question of free pragmatic processes contributing to explicature, processes whose existence I have used as an argument against the positing of a level of semantic content distinct from linguistic meaning. Relevance theorists distinguish, broadly, two types of free pragmatic enrichment. In some cases, such as when “It’s raining” is given a location-specific interpretation, the explicature contains an extra constituent, an ‘unarticulated constituent’ (henceforth abbreviated to UC), that is not the value of an element of linguistic form. Similarly when “and” is interpreted as & as a result, as seen in several of the above examples. However, in other cases, what is going on seems better construed as not the addition of extra concepts, but the adjustment of an encoded concept so that the concept understood as communicated by the use of a word is different than the encoded concept—it may be narrower, looser, or some combination of the two. Examples of this pragmatic adjustment of word meaning or ‘lexical modulation’ seen above include “drink” being used in example (1) to denote only alcoholic drinks, and “raw” being used of a steak that has received some (inadequate) amount of cooking; a few more examples are given here²¹:

• 23. To buy a house in London you need money.

• 24. Put the empty bottles in the garbage.

• 25. This water’s boiling.

Grasping the explicature of each of these examples is very likely to involve an optional process of meaning modulation (of the concept encoded by the italicized word in each case). In most contexts, the proposition that buying something requires (some, or any amount of) money will be trivial and uninformative, so the lexically encoded concept money is likely to be narrowed to a concept, represented as money*, that denotes just those quantities that would count as sufficiently large amounts of money in the context of London house-buying. If we imagine (24) uttered in a context where people are clearing up after a party, then the bottles in question may well contain dregs of beer, cigarette ash, etc., so will not be strictly empty; similarly, the concept encoded by “boiling” in (25) may be adjusted if the expression is used as an approximation or hyperbole.

The addition of extra components—unarticulated constituents—and lexical pragmatic modulation are the two types of free pragmatic enrichment that relevance theorists distinguish.²² The latter is a relatively new idea, and it is not yet clear how many of the cases previously treated as unarticulated constituents should be reanalyzed as modulation, though some clearly do not look susceptible to this treatment. Consider again the case of weather predicates, discussed earlier, where “It’s raining” has the explicature it is raining in london. It is difficult to construe rain-in-london as a narrowed form of the concept rain. Similar remarks apply to the enrichment of certain “and”-conjunctions, such as the cause-consequence interpretation in (1) [and the similar causal interpretation in (2)]. So at least some cases are likely to remain best analysed as unarticulated constituents rather than lexical pragmatic modulation. Anyhow, the idea is that both types of enrichment contribute to explicature, either changing or adding a constituent of the representation. However, Recanati (2007) takes a different view: he suggests that, while modulation contributes to the explicit content (what he labels the ‘lekton’), unarticulated constituents do not; instead, they are part of the situation of evaluation. Lekton and situation of evaluation together form the entire truth-conditional content, which Recanati calls the ‘Austinian proposition’.

To illustrate, consider weather predicates, on which much of the discussion is focused. The truth of an utterance of “It’s raining” virtually always depends on the situation at a specific location. However, for Recanati, the location UC is not part of the lekton but a parameter of the circumstances of evaluation. There are two variants of this scenario: first, in much of our weather talk, as when we look out the window and utter “It’s raining” to someone else in the same room, we are not thinking about the location we are in as opposed to any other location.²³ So in this case, the location has no cognitive significance, and is not represented at all. Yet the truth of the utterance still depends on the location.

The second variant is where we do cognitively articulate the location, for example, if someone is on the phone to a friend in another city, who reports “It’s raining,” the hearer, in comprehending the utterance, needs to entertain a concept of the location being talked about, as opposed to his own and any other salient locations. Recanati rightly claims that, from the fact that the location is represented, it does not follow that it is part of the lekton (2007: 224–230)—which is to say, it does not follow that its representation is incorporated into explicature. Instead, the location remains part of the situation of evaluation. Other elements of the overall truth-conditional content that would submit to the same treatment are the judges of taste and available information that appear to affect the truth values of utterances of “This is delicious” and “He might be in Boston,” mentioned above.

In earlier sections I rejected two candidates for a semantic content distinct from merely the linguistically encoded meaning: Cappelen and Lepore’s minimal propositions resulting from linguistic decoding plus saturation of all indexicals, and Bach’s what is said resulting from linguistic decoding plus saturation of pure indexicals. But now we have a third candidate, the lekton, which is, according to Recanati, a genuine level of content—the content developed out of logical form that is explicitly represented in thought, or, the semantic interpretation of the sentence relative to context. In the case of utterances, all constituents of the lekton are the value of an element of the uttered sentence; the difference from the minimalist and indexicalist positions is that the lekton includes not only the values of indexicals, but also the results of lexical pragmatic modulation, illustrated in (23)–(25) above. What advantages does this have over the standard contextualist and RT view? Firstly, it could be more plausible for the cases in which we are not contrastively thinking of the location we are at, or of our own taste or knowledge as opposed to others’. Second, Recanati (2010) suggests, the ‘no UCs in the lekton’ position may also have the advantage of allowing the preservation of a compositional account of our grasp of explicit content, and thus should find favour with (or less opprobrium from) those indexicalists (Stanley, Szabó, Martí, King) and others who are sceptical about the prospects of a systematic account incorporating free pragmatic enrichment.

This is only one among several approaches that are discussed under the label ‘relativism’. I leave a detailed assessment to further work; here, I just raise a few questions about the idea that there are no UCs in the lekton/explicit content.

First, consider again the normal utterances of “It’s raining” where, Recanati suggests, the location has no cognitive significance: we are not thinking of our location as opposed to any other. I agree with the intuition here, that we do not have a contrastive representation of the location, so the explicature should probably not be characterized as, for example, It is raining in London. However, this would mean that there is no difference in the cognitive significance of the location in these two scenarios: (a) when someone looks out the window and utters “It’s raining,” and (b) when someone utters “Mary’s singing” (in a context where the location is irrelevant to the truth of the utterance).²⁴ This does not seem quite right, as the former case requires the thought to be anchored to a location in a way that the latter doesn’t. An alternative to the explicature given above, then, would be one that does incorporate a representation of the location in the weather case, but, rather than a full-fledged concept, it would be more like a mental indexical or demonstrative, much like singular concepts of individuals that would be expressed with, e.g. “that guy,” “him,” or “this.” As Korta and Perry (2006)’s examples, discussed in the previous section, showed, often what is needed in the representation that the hearer entertains is not a fully identifying conception of the object or person referred to by a referring term; the same should apply to locations, times, etc. If this is correct, then the explicature would be better characterized as it is raining-here (where here is not a natural language indexical but a mental one whose content is fixed).

The more relevant issue here, though, is whether the location constituent, even if represented (whether by the full-fledged, contrastive concept of that location, or by a demonstrative-like element) is incorporated into explicature. One reason to be sceptical is that a lot of the motivation for relativism appears to come from the phenomenon of faultless disagreement, for example where Mary says “Snails are delicious” and John says “Snails are not delicious.” Something similar happens with the other paradigm cases for relativism: for example, Mary says “Fred might be in Boston” and John says “He can’t be: I saw him just twenty minutes ago.” Despite John’s contradiction (which we assume to be correct) of Mary, there is a sense in which what she said seems correct or even true in that it was compatible with her knowledge at the time of utterance. Yet in the case of weather and locations, there is no similar phenomenon: If Mary, reporting on the situation where she is in Paris, utters “It’s raining,” and Fred utters “It’s not raining,” then there is no appearance of faultless disagreement: either one of them is wrong, or else they are not talking about the same location, thus not disagreeing. So in the former cases, it is easy to see how the standard of taste/speaker’s grounds can be considered a parameter relative to which the content is evaluated; in the “rain” case, the location does not behave the same way, but as a constituent of the content to be evaluated. The same applies to other likely UCs, such as the causal and temporal interpretations of various kinds of conjunctions, and completion of incomplete descriptions.

This leaves the question of how pragmaticists should treat taste predicates, epistemic modals, and so on: do we need to recognize a content (lekton) versus circumstances of evaluation (or context of assessment) distinction for these? The obvious way for RT to go would seem to be to say that the reference to the judge, rather than being a parameter of evaluation, is incorporated into explicature. Several authors, under the label ‘contextualism’, have given indexicalist accounts of taste predicates (e.g. Glanzberg 2007) and epistemic modals (e.g. von Fintel and Gillies (2011). The way that RT would implement a contextualist account would likely be to analyse the reference to a judge as an unarticulated constituent. I leave this open, but my tentative suggestion is that, in most of the cases at issue, neither solution applies, because the relevant parameter (the judge according to whose taste or knowledge the claim is made) is not communicated. Consider this dialogue, already mentioned above:

• 26. Mary: Semantics is fun.

Tom: Semantics is not fun.

Both utterances in this dialogue seem to be presented as objective statements; their explicatures could be represented as semantics is fun/not fun punkt (or tout court). The fact that we are unwilling to judge either speaker wrong, despite their disagreement, gives the appearance that the truth-conditional content being evaluated—hence the explicature—includes a standard of taste, but if the disagreement is genuine, then this is likely to be an illusion. There are two possibilities for accounting for it. The first is that what we are evaluating here is not the truth of the utterances, as we lack the means to determine which is true (if either: in some cases it may be that there is no fact of the matter). Instead, when asked to judge truth or falsity, all that we can evaluate in such cases is something closely related, which is the appropriateness of the utterance; the appropriateness depends on the sincerity of the speaker, which depends, in turn, on her personal taste. On this count, both utterances in (26) are faultless, though not both can be true. The second alternative is that my claim that no reference to the judge is communicated is quite compatible with some versions of relativism—in which case both utterances may be true from the relevant point of assessment. After all, the issue of how what is communicated is assessed for truth or falsity is not the concern of pragmatic or linguistic theory. So if it is correct that no reference to the judge is communicated, then this would be a reason to reject the proposals by Lasersohn (2005) and Stephenson (2007) that are contextualist-relativist hybrids, employing a linguistic parameter that can be bound to a judge in the context of assessment.

In conclusion, then, I have given some reasons to be sceptical of the import of relativism for the issues discussed in this paper, and possibly for pragmatic theory more generally. These brief considerations are, of course, far from decisive, and a much more detailed comparison of relevance theory and contextualism with a wider variety of relativist views will be necessary to establish whether the RT position on the semantics-pragmatics distinction and the non-existence of semantic content can ultimately be maintained.