SAYING (NOTHING) AND CONVERSATIONALIMPLICATURES

: I defend an alternative theory of conversational implicatures that does without Grice ’ s notion of making-as-if-to-say. This theory characterises conver-sationally implicatingthat p as a waytomean that p by sayingthat q or by saying nothing. Cases that Grice ’ s theory cannot capture are captured, and cases that Grice ’ stheorymisdescribesarecorrectlydescribed.Adistinctionbetweenconver-sational implicatures and pragmatic inferences from what speakers express is required, as well as a non-implicature treatment of figurative speech.


Introduction
Grice characterises conversationally implicating that p as a way to mean that p by saying that q or by making-as-if-to-say that q. In this article I defend an alternative theory according to which conversationally implicating that p is a way to mean that p by saying that q or by saying nothing. On the one hand, this alternative theory captures cases that Grice's theory does not capture: It notably counts as implicatures certain cases in which a speaker means that p by remaining silent. On the other hand, the alternative theory does not capture cases that Grice's theory captures. Notable here is the exclusion of figurative speech: Figuratively meant propositions do not count as conversational implicata. I shall defend these inclusions and exclusions in due course. I shall also argue that some cases that are captured by both theories are correctly characterised by the alternative theory only. These are cases in which a speaker non-figuratively means something by uttering a sentence although she says nothing in Grice's sense of 'say'.
Section 2 of this article is dedicated to preliminaries about Gricean pragmatics. In Section 3, I present two cases of intentional silence and argue that they involve conversational implicatures in which an agent means that p by saying nothing. This meaning-that-p-by-saying-nothing analysis is extended in Section 4 to cases of non-figurative speech in which the speaker utters a sentence and yet says nothing in Grice's sense. In Section 5, I compare my preferred theory with another alternative theory that does without the notion of making-as-if-to-say. In Section 6, I discuss the exclusion of figurative speech from my preferred theory.

GRICEAN PRAGMATICS
Participants in a talk exchange sometimes 'share a purpose', as Grice puts it (Grice, 1989, p. 26). They attempt to make contributions which further this common purpose, and so are guided by Grice's Cooperative Principle: 'Make your contribution such as is required […] by the accepted purpose or direction of the talk exchange in which you are engaged' (Grice, 1989, p. 26). One sort of common purpose is sharing information about a certain topicfor example, what happened at the party last night. It is this sort of purpose on which Grice focuses. Next comes the question of the features that make speech contributions 'such as is required' by this purpose. This question is answered by Grice with four 'maxims': Quantity, Quality, Relation, and Manner. The maxims spell out the features that speech contributions must have in order to further the purpose of sharing information.
Conversation as a collective activity governed by the Cooperative Principle constitutes the first tenet of Gricean pragmatics. The second tenet is that speakers intend some of the things they mean to be retrieved by their audience through inferences. I have just helped myself to the notion of what a speaker means. Grice and others following him have put forward several analyses of this notion, but I will not present them here. I will just note that in this tradition meaning that p is acting with an intention of a certain kind: a communicative intention. A communicative intention is directed at an audience, and more specifically it is an intention to produce a certain cognitive effect on an audiencefor example, intending that the audience take the speaker to have a certain belief. Secondly, communicative intentions are characterised by the privileged role that their recognition plays in their fulfilment. On the strongest conception of this privileged role, the recognition of a communicative intention constitutes its fulfilment: to recognise it is to fulfil it (Bach & Harnish, 1979). Thirdly, the content of a communicative intention associated with meaning that p embeds the proposition pas in for example, the intention that the audience take the speaker to believe that p. 1 Back to the inferential tenet of Gricean pragmatics, the first crucial feature of the intended inferences to what speakers mean postulated by Grice -henceforth pragmatic inferencesis that they are guided by an assumption of cooperativity. Speakers intend their audience to assume that they are observing the Cooperative Principle. Of course, more than an assumption of cooperativity is needed to infer what speakers mean. The audience is expected to use various additional pieces of knowledge (Grice, 1989, p. 31). The second crucial feature of pragmatic inferences is that they can be represented as explicit arguments. These arguments start from a premise about some observable aspect of the speaker's utterance (e.g., the speaker used the word 'X'), or some already registered aspect of what the speaker did with their utterance (e.g., the speaker said that p). They end with a conclusion about what the speaker means. The relation between the inferential processes actually involved in interpretation and argument-representations is a fraught issue (Dänzer, 2021;Geurts & Rubio-Fernandez, 2015;Saul, 2002b). Here is a plausible claim about this relation: Argument-representations display explicit conscious reasoning of which normal interpreters are capable. This claim is consistent with the view that the inferences performed by interpreters do not generally involve explicit conscious reasoning, and also consistent with the view that speakers do not generally intend their audience to infer what they mean through explicit conscious reasoning.
Grice presents the argument-representation aspect of his theory as a necessary condition for conversational implicatures: A speaker conversationally implicates that p only if the intended inference to p is amenable to representation as an explicit argument (Grice, 1989, p. 31). This demand may however be extended to all pragmatic inferences. Here is a non-implicature example adapted from Geurts and Rubio-Fernandez (2015). A speaker utters 'The chestnuts are shedding their leaves', intending her audience to infer which of the meanings of 'chestnuts' she is using. Here is the argument-representation suggested by the authors (the starting premise is in bold, and the conclusion is underlined): -she has used the word 'chestnut'; -there is no reason to suppose that she is not observing the Cooperative Principle; 1 Little in this article hinges on the detail of the analysis of speaker-meaning. When the detail does matter I will mention it. For Grice's seminal ideas on speaker-meaning and some complications, see chapters 14, 5, and 6 of (Grice, 1989). For a critical overview of Grice and his successors on the analysis of speaker-meaning, see section 5 of (Neale, 1992 -she could not be doing this unless she intended to refer to trees of the genus Castanea, for this is one of the standard meanings of the word, and it fits our discourse purposes better than any of the others; -she intends me to think, or is at least willing to allow me to think, that she intended to refer to this type of tree; -and so this is what she meant. (adapted from Geurts & Rubio-Fernandez, 2015, p. 448) 2.2. CONVERSATIONAL IMPLICATURES Conversational implicatures are pragmatic inferences whose argumentrepresentation starts with the premise that the speaker said that q and ends with the conclusion that the speaker meant that p. 2,3 Conversational implicatures are thus associated with the following argument-schema: • The speaker said that q.
• The speaker meant that p.
The central notion in Grice's theory of implicatures is that of a speaker implicating that p. There is at least some overlap between this notion and pre-theoretical notions such as suggesting that p, implying that p and so on (Grice, 1989, p. 24). The impetus for the introduction of the notion of implicating that p is thus the truism that people often mean more than they say. At first blush, to implicate that p is to mean that p by saying that q. 4 Grice distinguishes three kinds of conversational implicatures, according to the degree of compliance with the maxims exhibited by the speaker (Grice, 1989, pp. 31-37). To mention the kind of implicatures that will be most relevant in this article, 'group-C implicatures' are ones associated with the overt violation of a maxim. Grice calls such overt violation the 'exploitation' of maxims (Grice, 1989, p. 30). With a group-C implicature, a speaker intends her audience to infer that she means that p by reconciling her overt violation of a maxim with her observance of the Cooperative Principle. 5 2 'Implicature' is sometimes used in the literature to refer to the speaker-meant proposition in the conclusionfor example, p in the main text. I prefer to use Grice's own 'implicatum', or 'implicated proposition' for this purpose. I reserve 'implicature' for the intended inference. 3 This article is not concerned with the conventional implicatures that Grice distinguishes from conversational ones (Grice, 1989, pp. 25-26). I sometimes use 'implicature', 'implicate', and 'implicatum' for the sake of brevity, but it should be clear that I am referring to the conversational variety. 4 I choose not to present Grice's three-clause characterisation of S implicates that p (Grice, 1989, pp. 30-31). Difficult interpretative questions are associated with the three-clause characterisation, some of which lead Saul (2002a) to argue that what speakers conversationally implicate is not settled by what they mean beyond what they say.
The foregoing characterisation of conversational implicatures overlooks a caveat made by Grice, one that is crucial in the context of this article. The starting premise of the argument-representation of a conversational implicature can either be that speaker S said that p, or that she made-as-ifto-say that p. This raises the question of the nature of saying that p as well as that of making-as-if-to-say that p.
Starting with saying that p, Grice's notion notoriously fails to align with our pre-theoretical notion of saying. His has two dimensions. The first dimension is the entailment from S says that p to S means that p (Grice, 1989, pp. 86-88). This is where Grice's notion clashes with the ordinary notion: An ordinary report that S said that p may be acceptable even though S did not mean that p. The second dimension of Grice's notion is that S says that p only if p is closely related to the conventional meaning of the sentence uttered by S (Grice, 1989, p. 25). Grice further specifies what counts as closely related to conventional meaning, but we need not go into details right now. Bringing the two dimensions together, S says that p if and only if (i) p is closely related to the conventional meaning of the sentence uttered by S and (ii) S means that p.
I will use 'say' as Grice does throughout this article. Let me clear the air: I agree that Grice's choice of 'say' as a label for the relation he targets is unfortunate given its departure from ordinary usage. If this label is too much of an irritant for the reader, they can replace it with the label of their choice ('state', 'assert' …). I choose to use Grice's label because it makes the comparison between his theory and my alternative theory easy. Now, it is possible to object not to the label 'say' but to the theoretical role given to the relation picked out by the label. As will become clear by the end of this article, I view the relation picked out by Grice's 'say' (modulo his conception of a close relation to conventional meaning) as essential to a theory of conversational implicatures. This view can be criticised independently of terminological choices.
Let me turn to making-as-if-to-say that p. Grice wants to treat figuratively meant propositions as conversational implicata. But when it comes to figurative speech, the speaker does not mean the literal content of her sentence, and therefore does not say it in Grice's sense. Therefore, implicating a figurative content p cannot be a matter of meaning that p by saying that qwhere q is the literal content. Grice's rescue-move is to introduce the further notion of making-as-if-to-say. A speaker implicates a figurative content by making-as-if-to-say a literal content. The notion of making-as-if-to-say can be characterised as follows: S makes-as-if-to-say that p if and only if (i) p is closely related to the conventional meaning of the sentence uttered by S, and (ii) S does not mean that p. 6 The resultant theory holds that implicating that p is a way of meaning that p by saying that q or by making-as-if to say that q.

Implicating something by remaining silent
According to Grice, one can implicate that p only by saying or making-as-ifto-say that q. One must then at the very least utter a sentence in order to implicate something. In this section, I put aside the letter of Grice's theory to assess the idea that one can implicate something by remaining silent.
First, one can mean something by remaining silent. Here is an example. S's friend asks S what she thinks of Mr. X. S looks at her friend intently and remains silent. What is required of S to have meant, for example, that she does not think highly of Mr. X (that p)? S must first have an audience-directed, p-embedding intention: S must intend her friend to, for example, take her to believe that p. Whatever the exact content of this intention, it seems possible for S to have it. For S to mean that p, S's intention must further be fulfillable by being recognised. S must then reasonably expect her friend to recognise her intention. For this to be the case, S and her friend must know enough about each other. Say that S is reluctant to bad-mouth people, and that this is common knowledge between the two friends. In such circumstances, it seems that S may reasonably expect her friend to recognise that she intends her to think that she does not think highly of Mr X. And so it is plausible that S means that p. Here is another example. S meets a friend of hers who is looking gloomy. It is common knowledge that this friend is sometimes reluctant to talk about her mood. S ventures to ask: 'So what's troubling you?'. S's friend stares at S intently and remains silent. For the same reasons as in the first case, it seems that S's friend means that she does not want to discuss her mood.
So one can mean something by remaining silent. But can one implicate something by remaining silent? Here is a reason to think not. 7 Pragmatic inferences from either saying that q or making-as-if-to-say that q involve the speaker's reliance on the audience's knowledge of the linguistic meaning of an uttered sentence. One may think that this feature is a defining feature of the pragmatic kind of conversational implicature. If so, pragmatic inferences from silence cannot be a species of the kind of conversational implicatures. 6 Some authors seem to think that condition (ii) is not necessary (Braun, 2011;Simons, 2017b). If they are right, then saying that p entails making-as-if-to-say that p. Given Grice's exclusive deployment of the notion of making-as-if-to-say that p for cases in which the speaker does not mean that p, and given the chosen terminology ('make-as-if-to-X' resembles 'pretend to X', and typically one cannot both pretend to X and X), I take making-as-if-to-say that p to be inconsistent with saying that p. In this I follow Bertolet (1983). I agree that the speaker's reliance on the audience's linguistic knowledge seems a significant feature. However, one cannot tell if a common feature between pragmatic inferences is a kind-constituting one just by looking at it. Similarly, one cannot tell if a dissimilarity between pragmatic inferences justifies a species-differentiation or a kind-differentiation just by looking at it. Take Grice's theory: According to it, there are implicatures from saying that q and implicatures from making-as-if-to-say that q. Only implicatures of the former sort are inferences from the fact that the speaker means a proposition. One might think that this feature is a merely species-constituting one, as Grice does. But one might alternatively think that it is a kind-constituting feature. Indeed, several authors take it to be constitutive of the kind of conversational implicatures (Bach, 1994;Carston, 2017;Recanati, 2017). As I see it then, assessing the boundaries of a pragmatic kind should happen in the context of assessing a broader pragmatic taxonomy that includes this kind. Later in this article I will sketch the broader pragmatic taxonomy resulting from my chosen inclusions in the pragmatic kind of conversational implicatures. What I can do in the present section is argue that there are similarities between pragmatic inferences from saying that p and pragmatic inferences in my silence cases which make it plausible that both belong to the same kind.
One crucial similarity is the violation of conversational norms. In both of my silence cases, someone deliberately remains silent after being asked a question. Conversational norms demand that this person say something, she knows it, and yet she says nothing. 8 Here I assume that the purpose of sharing information does not merely demand that one utter a sentence after being asked a question. It demands that one answer the question (if possible), by making a linguistic contribution in compliance with the maxims. Hence, one is expected to say something in Grice's sense. Someone who says nothing when asked a question thus violates conversational norms. There are several, compatible ways of looking at the violation in question. It may first be regarded as the upper limit of the violation of the first sub-maxim of Quantity ('Make your contribution as informative as is required'). Trivially, by remaining silent when asked a question one is overtly making an insufficiently informative contribution. However, there might be more going on than an exploitation of Quantity. By remaining silent, one might be 'opting out' of the Cooperative Principle. That is, one might be signalling that one is 'unwilling to cooperate' (Grice, 1989, p. 30). Consider my second case, in which S's gloomy friend chooses not to answer a question about her mood. It looks like S's gloomy friend is opting out. 8 Swanson contends that there are omissive implicatures, that is, implicatures driven by the recognition that the speaker did not do something they were expected to do (Swanson, 2017). Both the silent cases in this section and the linguistic cases in the next section are arguably cases of omissive implicatures in Swanson's sense.
This 'opting out' construal of my second case seems to raise a new problem. In Grice's theory, opting out is inconsistent with implicating something. Conversational implicatures rely on an assumption of cooperativity: '… to assume the presence of a conversational implicature, we have to assume that at least the Cooperative Principle is being observed' (Grice, 1989, p. 39). Discussing similar cases of intentional silence, Kasher claims that they involve both implicatures and uncooperativity (Kasher, 1976). Recognising that this generates an internal inconsistency in Grice's theory, Kasher concludes that the driving force behind conversational implicatures is something other than the Cooperative Principle. I wish to make a different move at this juncture. I want to claim that the case of S's gloomy friend involves both a conversational implicature and a low yet sufficient degree of cooperativity. This move demands that we distinguish between two degrees of cooperativity. S's gloomy friend refuses to share information about what's troubling her, and so she opts out of the Cooperative Principle to the extent that she refuses to share information about the question under discussion. We might call the corresponding form of cooperativity local cooperativity. S's gloomy friend is then not locally cooperative. However, she is still cooperative to the extent that she is sharing information about something or other. We might call this form of cooperativity global cooperativity. 9 When a speaker is not locally cooperative but still means something, she expects the interpreter to retreat to an assumption of global cooperativity. I propose that we regard implicating that p as consistent with this kind of mixed cooperativity.
This amendment may seem to stretch Grice's theory, but in fact, some cases described by Grice as conversational implicatures arguably involve mixed cooperativity. Grice presents as a group-C implicature a case in which a speaker utters a sentence whose literal content p is wildly irrelevant to the previous utterance made by someone else. The irrelevant speaker does this to mean that the previous utterance constituted a social gaffe, and Grice claims that the speaker conversationally implicates this (Grice, 1989, p. 35). This case arguably involves mixed cooperativity: The obvious irrelevance of the speaker's utterance constitutes an overt violation of local cooperativity; and the intended inference to what the speaker means relies on an expected retreat to the weaker assumption that the speaker is globally cooperativethat is, that she means something or other. Hence, Grice himself seems to regard implicating that p as consistent with mixed cooperativity.
An agent who means that p by remaining silent when conversational norms demand that she say something therefore intends her audience to recognise that she means that p through an inference involving (i) the recognition that she overtly violates conversational norms and (ii) the preservation 9 Kasher introduces a distinction between local and global aims in talk exchanges (Kasher, 1976, pp. 201-202), but his distinction is orthogonal to mine. of an assumption of cooperativity (sometimes only global cooperativity). These features make a group-C implicature treatment of such cases plausible.
There is a further feature shared by my silent cases and cases of meaning that p by saying that q. In both, the pragmatic inference starts with the intended recognition of a fact about an act of saying (in Grice's sense). In paradigmatic cases of conversational implicatures, the relevant fact is that the speaker said that p. In my cases of intentional silence, it is rather that the silent 'speaker' said nothingthat is, that no act of saying has been performed. To see this, let me turn to the argument-representation of the pragmatic inference in my second silence case. S's gloomy friend means that she does not want to discuss her moodcall this proposition p. What does the argument-representation of the intended inference look like? First, S is expected to recognise that her friend said nothing: -She said nothing.
This recognition is not expected on the basis of the agent's silence alone, but rather on the basis of her silence plus the conversational demand that she say something. Let me explain this. First, not every intentional silence is intended to be interpreted. When at a certain point in a talk exchange an agent is not expected to say anythingfor example, when it is not her turn to speakher silence may be intentional, but she does not intend this silence to be interpreted. Secondly, there are cases in which an agent means that p by doing something other than producing sounds, but the fact that the agent says nothing is not part of the intended inference to what she means. This usually happens when conversational norms do not demand that the silent agent say something. Here is a simple example. A parent enters the untidy bedroom of their teenage son. They catch the eye of their son, point to the pile of dirty clothes on the floor and roll their eyes. They mean something (quite vague), but they do not intend their teenage son to infer what they mean by recognising that they said nothing. Candidate cases of implicating that p by remaining silent are ones in which the intended inference to what the silent agent means includes the recognition that she said nothing. And one necessary condition for this inclusion is that conversational norms demand that something be saidfor example, when a question is asked. 10,11 10 Thanks to an anonymous referee for making me address the distinction between implicating that p by being silent and meaning (but not implicating) that p by being silent. 11 Goldberg (2020, chapter 8) uses Gricean tools to illuminate yet another aspect of silence in conversation. When a conversational participant stays silent following a contribution by another participant, the inference that the former participant endorses the contribution is licensed (except in 'oppressive' and other specific circumstances). Back to the argument-representation of my second silence case, after recognising that her gloomy friend said nothing S is expected to hold firm and assume that her friend was nonetheless being (globally) cooperative: -She is being (globally) cooperative (i.e., she means something).

SAYING (NOTHING) AND CONVERSATIONAL IMPLICATURES
From here onwards, S is expected to use her knowledge about her friend (notably that her friend tends to be reluctant to talk about her problems) to reach the following conclusion: -She meant that she does not want to discuss her moodthat p. This is not the full argument-representation of the intended inference, but rather a bare-bones reconstruction focusing on its essential moving parts. These essential moving parts are: (i) the recognition that the silent 'speaker' said nothing; and (ii) an assumption of (global) cooperativity. The audience's knowledge about the speaker is expected to allow them to go the extra-mile and conclude that the speaker meant that p.
I wish to use these cases of intentional silence as a springboard to an alternative theory of conversational implicatures. This alternative theory holds that implicating that p is a matter of meaning that p by saying that p or by saying nothing. One advantage of this theory over Grice's theory is already apparent: It is able to capture cases that Grice's theory cannot capture. The attentive reader will notice that this advantage extends beyond cases of silence, since remaining silent is just one way of saying nothing. Coughing, uttering a random string of syllables, or humming 'Dancing Queen' are all ways of saying nothing which may be employed by a speaker to implicate something. In the next section, I argue that a class of cases captured by both Grice's theory and the alternative theory is more accurately characterised by the latter.

Implicating something by saying nothing by uttering a sentence
Drawing on (Neale, 1992), Simons suggests that paradigmatic cases of conversational implicatures in which the speaker says that p systematically have a variant in which the speaker utters the same sentence but does not say that p, because she does not mean that p (Simons, 2017b, pp. 541-544). These variants are not supposed to be cases of figurative speech. If Simons is correct, Grice needs to invoke his notion of making-as-if-to-say beyond figurative speech. I will not take a stance on the pervasiveness of these variants, but I will use a variant of one of my cases of intentional silence to illustrate their existence. S asks her gloomy friend: 'So what's troubling you?'. S's gloomy friend stares at her intently and replies in a flat tone: 'I'm doing great'. She is not engaging in sarcasm but rather conveying that she does not want to discuss her mood. This is a stipulation, albeit an acceptable one: This case is possible. The utterance of S's gloomy friend is overtly false: It is common knowledge between the two friends that she is not doing great. As Bach and Harnish put it, this is a case of 'obvious insincerity' (Bach & Harnish, 1979, pp. 57-59). 12 I take this feature of the case to imply that S's gloomy friend does not mean that she is doing great. This requires a commitment to the following inconsistency: A rational speaker cannot both utter a sentence whose content p is false as a matter of common knowledge between participants and mean that p. 13 Assuming some analysis of meaning that p that implies the proposed inconsistency, S's gloomy friend did not say that she is doing great, in Grice's sense of 'say'.
What did she say then? The question here is whether there is a further proposition closely related to the conventional meaning of the uttered sentence and which the speaker means. Grice has his own preferred interpretation of 'closely related to conventional meaning' (Grice, 1989, pp. 87-88). Only propositions obtained by (i) the syntactic and semantic decoding of the uttered sentence, (ii) the resolution of ambiguity, and (iii) the determination of the content of context-sensitive expressions count as closely related to the conventional meaning of the uttered sentence. In the case at hand, there seems to be no proposition falling within this range such that the speaker means it. And so the speaker said nothing. It is important to note that this verdict is not conditional on endorsing Grice's conception of a close relation between proposition and conventional meaning. On a view inspired by Recanati (2001) and Carston (2002), a proposition counts as closely related to the conventional meaning of an uttered sentence either if it is identical with the content of the initial representation arrived at by syntactic and semantic decoding of the uttered sentence, or if it results from 'local' pragmatic developments or substitutions of parts of this initial representation. 14 Even according to this more relaxed conception, it seems that S's gloomy friend said nothing at all. 12 As in cases of figurative speech, Grice would contend that the speaker exploits the first sub-maxim of Quality -'Do not say what you believe to be false' (Grice, 1989, p. 27). Unfortunately, Grice's formulation of the first sub-maxim of Quality is confusing. In the case at hand, as in cases of figurative speech, the speaker does not say the false content of her uttered sentence by Grice's own lights. 13 The proposed inconsistency has an analogue in Stalnaker's theory of assertion (Stalnaker, 1999). One of the 'rules of assertion' states that asserting that p is inconsistent with p being false in every possible world in the context set (i.e., the set of possible worlds representing the common ground between participants). Interestingly, Bach and Harnish argue that obvious insincerity is consistent with meaning that pin their preferred terminology, with expressing the belief that p (Bach & Harnish, 1979, pp. 57-59). 14 In relevance-theoretic parlance such propositions, when they are meant by the speaker, are 'explicated' rather than 'said' (Carston, 2002;Sperber & Wilson, 1995 Granting that the speaker said nothing, it must still be established that she implicated that she does not want to discuss her moodthat p. First, the speaker means this proposition, for the reasons I gave in the previous section in the corresponding silent case. Secondly, by saying nothing the speaker 'exploits' the conversational demand that she say something. She intends her audience to infer that she means that p (i) by recognising that she exploits conversational norms and (ii) by maintaining an assumption of cooperativity. On the face of it, this case involves a group-C implicature.
What is the argument-representation of this conversational implicature? I claim that it is the same as in the corresponding silent case. The premise that the speaker said nothing and the premise that she is (globally) cooperative lead to the conclusion that the speaker means that p.
-She said nothing. -She is being (globally) cooperative.
-She meant that she does not want to discuss her moodthat p.
Of course, the overall intended interpretation is not the same as in a case of saying nothing by remaining silent. In a case of saying nothing by uttering a sentence, prior inferential work is expected of the interpreter to arrive at the recognition that the speaker said nothing. The speaker utters her sentence with its conventional meaning, which determines in the context of utterance the proposition that she is doing great (or so we can assume here). We might say that the speaker expresses that she is doing great, borrowing the terminology from Simons (2017aSimons ( , 2017b. The speaker intends her audience to recognise this. The speaker further expects her audience to assume that she said something. The idea here is that there is a general default expectation that the speaker means some proposition closely related to the conventional meaning of her sentence. 15 This expectation may be construed as one that the speaker speaks literally, or one that she speaks directlythat is, that she performs a direct illocutionary act in the sense of (Bach & Harnish, 1979). 16 The proposition that the speaker is doing great is eligible for being said, since it is identical to the meaning of the uttered sentence in its context of utterance. However, the audience is expected to recognise that the speaker did not mean this proposition, and therefore did not say it in Grice's sense. The audience is finally expected to realise that the speaker said nothing, thereby cancelling the assumption that she said something.
To summarise, this part of intended interpretation can be represented by the following argument: 15 If 'closely related to conventional meaning' is interpreted in Grice's way, then this expectation is just the expectation that the speaker says what she expresses. I return to the relation between what is expressed and what is said in the next section. 16 See Bach and Harnish (1979, chapter 4) for a discussion of how assumptions of this sort come into play in the interpretation of utterances. Endorsing a theory according to which implicating something is a matter of meaning it by (i) saying something else or (ii) saying nothing commits one to the view that the part of intended interpretation leading to what (if anything) is said does not belong to the implicature. This is true not just of cases in which the speaker says nothing, but also of cases in which the speaker says something. Intended inferences to what the speaker says are pragmatic inferences alright, but they are not conversational implicatures. I will return to this commitment of the theory in more detail in the next section.
For now, let me consider a reply on behalf of an advocate of Grice's theory. It is the recognition that the speaker made-as-if-to-say that q that drives the inference to what the speaker means, not the recognition that she said nothing. The first part of intended interpretation in my 'I'm doing great' case should be represented as follows: -She expressed that she is doing greatthat q.
-She said something.
-She did not say that q.
-She only made-as-if-to-say that q.
And the part corresponding to the conversational implicature should be represented as follows: -She made-as-if-to-say that q.
-She is being (globally) cooperative. -She meant that she does not want to discuss her moodthat p.
Here is my counter-reply. The first part of the argument-representation just presented cannot be correct. The question of what the speaker said has been raised by the assumption that the speaker said something. Recognising that the speaker made-as-if-to-say that q does not settle this question. Assuming that the speaker is rational, she cannot be intending her audience to leave an interpretive question hanging in this way. Of course, there may be overriding reasons for an interpreter not to complete the interpretive task of establishing what the speaker said (their house is on fire!). But it is another thing to leave the question of what the speaker said hanging, and then plough on to determine what the speaker meant but did 17 Thanks to an anonymous referee for convincing me that this is the order in which the premises should be presented. not say. A rational speaker would not intend her audience to do this. In the case at hand, the question of what the speaker said is settled by recognising that the speaker said nothing. Hence, there is no way around the intended recognition that the speaker said nothing.
If there is no way around the recognition that the speaker said nothing, and if it is possible to infer what the speaker means from this recognition and an assumption of (global) cooperativity, then there is no additional role to play for the premise that the speaker made-as-if-to-say that q. This premise would be superfluous in the overall argument-representation. To the extent that a speaker is rational, she does not intend her audience to use superfluous information, and this should be reflected in the argument-representation of the intended inference. Hence, an alternative theory of conversational implicatures according to which implicating that p is a matter of meaning that p by saying that q or by saying nothing characterises cases of this sort more accurately than Grice's theory does.
There is another alternative theory of conversational implicatures that dispenses with the notion of making-as-if-to-say. In the next section, I present this theory and compare it with my preferred alternative theory. This gives me the opportunity to clarify the broader pragmatic picture in which my preferred theory fits.

Implicating something by expressing that p
The speaker-meaning dimension of Grice's notion of saying that p leads him to add the notion of making-as-if-to-say that p to his theory of implicatures, in order to catch figurative contents in its net. One alternative is to remove both of Grice's notions and replace them with a single notion that is neutral with regard to speaker-meaning. On this view, all conversational implicatures instantiate the following argument-schema: -The speaker X'd that q.
-The speaker meant that p.
To capture cases of making-as-if-to-say as well as cases of saying, the X relation must be like saying but with no speaker-meaning strings attached. Simons' expressing relation introduced in the previous section seems to fit the bill. Indeed, expressing that p consists in uttering a sentence that means that p in the context of utterance with the intention to use it with this meaning (Simons, 2017b, p. 543). 18 Other authors put forward speaker-meaning- 18 Simons does not regard her project as that of putting forward an expressing theory of conversational implicatures. As she sees it, she only emphasises the 'continuity' between conversational implicatures and pragmatic inferences starting from what a speaker expresses (Simons, 2017b,  neutral relations similar to expressing. Bertolet (1983) suggests representing oneself as saying for X. Braun (2011) suggests locuting. 19 The expressing theory of conversational implicatures differs from my preferred theory in the following way. Grice's notion of saying picks out a relation that is partly a matter of communicative intentions. And so according to my preferred theory, conversational implicatures are pragmatic inferences starting from the recognition of facts about a relation of this sort. This point could be, and has been formulated in speech-act-theoretic terms: 'conversational implicatures […] are implications of an act of "saying" ' (Recanati, 2010, p. 143). Recanati is here thinking of the act of Grice-saying as an illocutionary act. 20 By contrast, according to the expressing theory, implicatures start from the recognition that a mere locutionary act has been performed (Braun, 2011). My preferred alternative theory of conversational implicatures thus commits me to the idea that Grice-saying or more broadly what Bach and Harnish (1979) call direct illocutionary actsconstitutes a joint in pragmatic nature. This joint separates two kinds of pragmatic inferences: those from what is expressed to what is said, and those from what is said. Only the latter are conversational implicatures. By contrast, the expressing theory does not regard Grice-saying or related notions involving speaker-meaning as a joint in pragmatic nature. All inferences from what a speaker expresses to what she means are of the same kind, a kind we can label 'conversational implicature'.
Which map of the pragmatic realm is more accurate? To answer this question it will help to understand what the difference between the two maps amounts to concretely. Let me start with the argument-representation in my 'I'm doing great' case, from the perspective of the expressing theory. The expressing theory captures this case differently from my preferred theory. The expressing theory does not divide the overall intended interpretation in two as I did in the previous section, but rather treats it as one single pragmatic inference from what the speaker expresses to what she means: -She expressed that she is doing greatthat q.
-She said something.
-She did not say that q.
-She said nothing. -She is being (globally) cooperative.
-She meant that she does not want to discuss her moodthat p. 19 Braun's locuting is however broader than expressing, since one can locute the content of an interrogative sentence. I should further note that the theory in which Braun deploys the notion of locuting is different from the one discussed in this section. 20 The project of characterising illocutionary acts in terms of communicative intentions is pursued in great detail in (Bach & Harnish, 1979).
Contrast this with the two-argument representation I gave in the previous section. There was first a pragmatic inference from what is expressed to what is said: -She expressed that she is doing greatthat q.
-She said something.
-She did not say that she is doing greatthat q.
-She said nothing.
Then there was a second pragmatic inferencewhat I regard as the conversational implicaturefrom what is said to what is further meant: -She said nothing.
-She meant that she does not want to discuss her moodthat p.
According to this two-step picture, the role of the recognition of what the speaker expressed is only that of guiding the interpreter to the recognition of what (if anything) the speaker said. Once this task has been completed, the recognition of what the speaker expressed has exhausted its role in interpretation. To further derive what the speaker meant beyond what she said, one need not consider again what she expressed.
We are now in a position to reflect on the consequences of endorsing the one-step picture associated with the expressing theory or the two-step picture associated with my preferred theory. One odd prima facie feature of the one-step picture is that some pragmatic inferences seem to 'disappear' depending on whether the speaker only means what she says, or means what she says and some more. When a speaker means only what she says in Grice's sense, there is a pragmatic inference from what she expresses to what she means and says. But when she means what she says plus further propositions, then there is just one big pragmatic inference from what she expresses to the further propositions she means. The inference from what she expresses to what she says does not stand on its own anymore.
Advocates of the expressing theory could avoid this consequence by denying that there are pragmatic inferences from what is expressed to what is said in the first place. They could contend that there is no gap between expressed content and said content and, hence, no need for inferences from the former to the latter. However, this view is far from uncontroversial given the nature of the expressing relation. A content is expressed by a speaker only if it is the meaning of the sentence uttered by the speaker in the context of utterance. The view that at least in some cases the meaning of a sentence in context is less than a proposition has been aptly defended in the last thirty years (Bach, 1994;Carston, 2002;Recanati, 2010 in some cases the meaning of a sentence in context is a proposition but not one the speaker means is even more common in philosophy -I would cite the same authors plus (Borg, 2004) and (Cappelen & Lepore, 2005). If at least one of these two views is right, then it is not true in general that no gap exists between what is expressed and what is said in Grice's sense. 21 The possibility of a gap between what speakers express and what they say should make us think twice before adopting Grice's restrictive conception of a close relation to conventional meaning in his definition of saying that p. If this close relation does not extend beyond assigning contents to context-sensitive expressions, then it is not clear that there usually is a proposition closely related to the conventional meaning of the uttered sentence which the speaker meansthat is, not clear that there usually is something the speaker says in Grice's sense. This tension in Grice's notion of saying that p is well documented (Carston, 2002). Fortunately, there are more relaxed conceptions of a close relation to conventional meaning, as mentioned in the previous section. It is then safer to adopt such a relaxed conception, which would bring Grice's notion of saying in line with Recanati's (Recanati, 2001) or with the notion of 'explicating' a proposition in relevance theory (Carston, 2002;Sperber & Wilson, 1995). Now putting aside general views about the relation between what is expressed and what is said, linguistic cases of saying nothing such as my 'I'm doing great' case raise a challenge for the one-step picture associated with the expressing theory. In such cases, the speaker expresses a proposition but says nothing. Some inferential work is needed to go from what the speaker expresses to what she says (nothing). Hence, there really seem to be two interpretive steps in these cases. One option is left to the advocate of the expressing theory. She should deny that in these cases intended interpretation goes through the recognition that the speaker said nothing: If this interpretive stop is not mandatory, intended interpretation directly takes the interpreter from what is expressed to what is meant. Now, in the previous section I argued that there is no way around the recognition that the speaker said nothing if the assumption that the speaker said something is part of intended interpretation. This conditional seems safe to me. But what about its antecedent? Could one deny that the assumption that the speaker said something is part of intended interpretation? As I mentioned in the previous section, this assumption could be construed either as one of literalness or as one of directness. The advocate of the expressing theory should then argue that neither sort of assumption comes into play when a speaker utters a declarative sentence in the course of a talk exchange aiming at sharing information. Unless such an argument is provided, it must be conceded that linguistic cases of 21 (Stanley, 2007) is a sustained attack on both views. Stanley argues that for every literal utterance of a declarative sentence, assigning denotations to constituents of the logical form of the sentence yields through composition a proposition that is meant by the speaker (Stanley would use 'asserted' rather than 'meant'). implicating something by saying nothing involve an intended inference from what the speaker expresses to what she says. Hence, the advocate of the expressing theory must concede that at least in these cases a two-step picture is necessary: Some inferential work is needed to go from what is expressed to what is said, and some extra inferential work is needed to go from what is said to what is meant but not said.

SAYING (NOTHING) AND CONVERSATIONAL IMPLICATURES
I have given reasons to favour the broader pragmatic picture implied by my preferred alternative theory of conversational implicatures. In the next section, I discuss the most notable exclusion from this theory: that of figurative speech.

Figurative speech and conversational implicatures
Grice treats figuratively meant contents as conversational implicata (more specifically as group-C implicata associated with the exploitation of the first sub-maxim of Quality). Speakers implicate a figurative content by makingas-if-to-say a literal content. As for the expressing theory, it is able to count figurative contents as implicata, since speakers mean a figurative content by expressing a literal content.
In contrast with these theories, my preferred theory is unable to count figurative contents as conversational implicata. On the one hand, if a loose construal of 'close relation to conventional meaning' is endorsed, then perhaps a non-literal speaker says the figurative content when metaphorically uttering 'You are my sunshine' or sarcastically uttering 'He's a fine friend'. Hence, the pragmatic inference to the figuratively meant content does not count as an implicature. On the other hand, if Grice's restrictive construal of 'close relation to conventional meaning' is endorsed, the non-literal speaker says nothing when uttering 'You are my sunshine' or 'He's a fine friend'. According to my preferred theory, neither expressing nor makingas-if-to-say it plays a role in conversational implicatures. The problem with this is that a figurative content cannot be retrieved without the help of the corresponding literal content. To infer that a speaker meant a figurative content p, the interpreter must use the fact that the speaker Y'd that q, where q is the literal content of the uttered sentence and Y is some relation that does not involve meaning that q. Recognising that the speaker said nothing and deploying any amount of additional knowledge will not suffice.
Is this incapacity to capture pragmatic inferences from literal contents to figurative contents a problem for my preferred theory? One reason to treat figuratively meant contents as conversational implicata is to have a unified theoretical treatment of various phenomena falling into a broad pre-theoretical domain. This broad pre-theoretical domain is the transmission of non-linguistically-encoded information in conversation. However, as we have seen throughout this article, there are differences among cases in this broad pre-theoretical domain that motivate the drawing of theoretical boundaries within it. There are reasons of this sortwhich I cannot review herenot to treat figurative contents as conversational implicata (Camp, 2006(Camp, , 2012. It is even possible to treat different sorts of figurative contents as distinct species of non-linguistically-encoded, non-implicated contents. For instance, it is possible to hold an 'explicature' view of metaphor while holding an 'echoic' view of sarcasm, as Carston (2002) does.
Focusing only on metaphor, let me just mention two alternatives to an implicature treatment. According to Stern (2000), an unpronounced operator inserts itself in the logical form of sentences and takes sub-sentential expressions to sets of properties metaphorically associated ('m-associated') with these expressions in the context of utterance. On this view, the mapping of an expression onto a metaphorical content takes place in the compositional machinery of semantics, without the mediation of a literally expressed proposition. Perhaps contrary to intuition, metaphorical content is in some sense linguistically encoded. This is one way of implementing the idea that metaphorical contents are said rather than implicated. Another way to implement this idea turns on the loosening of Grice's conception of a close relation to conventional meaning. Given a suitable loosening, it is possible to construe metaphorical propositions as the result of 'local' modifications of linguistically determined representations, and thus as said rather than implicated. As I see it, the existence of well-motivated non-implicature theories of various species of figurative speech provides enough reason to grant that a theory of conversational implicatures does not stand or fall on its ability to capture figurative speech.
Let me conclude this section by noting an awkward feature of both Grice's theory and the expressing theory when it comes to figurative speech. A teacher is asked: 'Do you need a break?'. The teacher might answer either 'Grading these exams is exhausting' or 'Grading these exams is killing me' to convey that she needs a break from grading. 22 On a pre-theoretical level, there seems to be no substantial difference between conveying that one needs a break from grading by speaking literally and conveying the same thing by speaking figuratively. Grice's theory and the expressing theory struggle to reflect this: While they can capture the former case, they cannot capture the latter case. This is because the inference from the literal expressed/made-as-if-to-say content that grading these exams is killing the speaker to the figurative content that grading these exams is exhausting already counts as a conversational implicature according to both theories. And neither theory counts inferences from implicata as conversational implicatures. Hence, neither theory can capture the inference from the figuratively meant content that grading these exams is exhausting to the further content that the teacher 22 Thanks to an anonymous referee for offering me this example.