1 Introduction

In the debate over the definition of lying there is a trend (Carson, 2006; Horn, 2016; Saul, 2012; Stokke, 2018) according to which a person lies only if they literally say something they believe to be false. In recent years, Saul (2012) and Stoke (2018) have raised the concern in philosophy of language about what notion of what is said is best suited for says-based definitions of lying. I shall not take a stand in the debate on the definition of lying.Footnote 1 My paper is in the field of philosophy of language. My concern is the following: if says-based definitions of lying are correct (or assuming that they are), what notion of what is said do they require? And what is the impact on the semantics/pragmatics divide? (See Michaelson, 2016; Vignolo, 2021, 2022 for similar concerns).

I argue that a conventions-based notion of utterance content inspired by Korta and Perry’s (2007a) locutionary content and Devitt’s (2021) what is said fits the desiderata for the role of what is said in says-based definitions lying and fares better than the notions of what is said that have been so far presented by Saul and Stokke.

Saul’s and Stokke’s says-based definitions of lying and their notions of what is said are the following:

Saul’s definition of lying:

If the speaker is not the victim of linguistic error/malapropism or using metaphor, hyperbole, or irony, then they lie iff (1) they say that p, (2) they believe p to be false, (3) they take themself to be in a warranting context. (Saul, 2012: 3).

Saul’s notion of what is said (S stands for a sentence and C for a context):

A putative contextual contribution to what is said is a part of what is said only if without this contextually supplied material, S would not have a truth-evaluable semantic content in C. (Saul, 2012: 57).

Stokke’s definition of lying:

A lies to B iff there is a proposition p such that (1) A says that p to B, (2) A proposes to make it common ground that p, (3) A believes that p is false. (Stokke, 2018: 31).

Stokke’s notion of what is said (μc(S) stands for the minimal content of the sentence S in context c):

What is said by S in c relative to a QUD q is the weakest proposition p such that (i) p is an answer (partial or complete) to q and (ii) either p ⊆ μc(S) or μc(S) ⊆ p.Footnote 2 (Stokke, 2018: 99).

Saul introduces condition (3) in her definition of lying in order to cope with bald-faced lies, when speakers do not have the intention to deceive their audience about what they say. Banning the intention to deceive from the definition of lying opens a problem of overgeneration in all those cases in which speakers use language non-literally or ironically knowing that what they literally say is false. Saul adopts Carson’s (2006) suggestion to add condition (3). In a warranting context the speaker guarantees the truth of what is said. Stokke’s proposal is more perspicuous because it appeals to the received Stalnakerian framework of the common ground. I shall ignore this difference between Saul’s and Stokke’s definitions and focus on their notions of what is said.

2 Desiderata for What is Said

Saul’s and Stokke’s analyses together put forward four main desiderata for contents to cover the role of what is said in says-based definitions of lying: they must be (i) fallback contents (in the sense to be explained below), (ii) truth-evaluable, (iii) non-minimal, and (iv) sensitive to questions under discussion (QUDs). A word of caution is in order. These desiderata arise from the point of view of says-based definitions of lying. I take for granted the intuitions and the arguments that Saul and Stokke employ to justify them. Of course, scholars who are critics of says-based definitions of lying do not share many of these intuitions and reject many of Saul’s and Stokke’s arguments. As I said, taking a stand in the debate on the definition of lying is not the issue of my paper.

  • (i) Fallback content.

The rich Catholic (Saul, 2012: 37).

A rich man decides to leave his fortune to Jack, as long as Jack has lived his life in full compliance with the precepts of Catholicism. The rich man asks John for information about Jack. John intends to favour his friend and, knowing that Jack had two children before getting married, he says:

  1. (1)

    Jack got married and had two children.

Among the philosophers who support says-based definitions of lying there is wide consensus that this is a case of misleading, not of lying. John expects that the rich man understands that Jack got married before having two children, but this is not what John says. John says that Jack got married and had two children in one order or the other.

The notion of saying according to which John is not lying is independent from what the hearer does or is expected to do. John expects that the rich man understands the content that Jack got married before having two children, but his assertion is not semantically committed to that content. John retains deniability for the disbelieved information that Jack got married before having two children. If charged with lying, John can protest that the content of his assertion is that Jack got married and had two children in one order or the other.Footnote 3

The contents that cover the role of what is said in says-based definitions of lying must be of a kind that allows speakers to fallback to them when they try to mislead their audience without lying. They are contents which speakers can claim responsibility for, and commit their assertions to, when they discharge the accusation of lying. That speakers can claim responsibility for those contents and commit their assertions to them implies that their assertions can be taken as literally expressing those contents, though the assertions so construed are not necessarily cooperative. According to says-based definitions of lying, one mark that distinguishes misleading from lying is that speakers retain deniability for the disbelieved information. Liars cannot deny commitment to the disbelieved information because they convey it by literally saying it. Misleaders, instead, can retreat from the disbelieved information they convey because they do not say it. As Stokke (2018: 84–87, 112) points out, when misleading speakers deny commitment to the disbelieved information they disclose that they were uncooperative with their assertions. Still, in order for discharging the accusation of lying, there must be a propositional content for their assertion, i.e. what they say, that makes their assertion literal (direct) and preserves their role of asserters, though uncooperative ones. The point is typically illustrated with the following example:

Paul’s party (Stokke, 2018: 76).

Dennis is going to Paul’s party tonight. He has a long day of work ahead of him before that, but he is very excited and can’t wait to get there. Dennis’s annoying friend Rebecca comes up to him and starts talking about the party. Dennis is fairly sure that Rebecca won’t go unless she thinks he’s going, too.

(2) Rebecca::

Are you going to Paul’s party?

Dennis::

I have to work.

In (2) Dennis conveys the disbelieved information that he will not go to Paul’s party, but he is not lying (according to says-based definitions of lying), because he is not literally saying it. Dennis is misleading. Utterance (2) leaves Dennis free to sincerely deny that he said that he would not go to Paul’s party. Suppose that, after finding out that Dennis went to Paul’s party, Rebecca charges him with lying. After utterance (2) Dennis might reply:

  • (3) I did not lie. I did not say that I would not go to Paul’s party. I just said that I had to work.

Of course, with this reply Dennis discloses that he was uncooperative. His discharge of the accusation of lying, requires that he expressed a propositional content that makes his assertion literal (direct). Dennis’s assertion must be taken as literally expressing that content in order for him to claim responsibility only for asserting that content.

  1. (ii)

    Truth-evaluable.

The murderous nurse (Saul, 2012: 48).

Dave is lying in bed, and two nurses are discussing the treatment he needs. Ed holds up a bottle of heart medicine, points at it just in front of Fred, and asks (4):

  • (4) Has Dave had enough?

Fred replies with (5):

  • (5) Dave has had enough.

Fred plans Dave’s death by denying him the heart medicine for his daily therapy. When he uttered (5), Fred proposed that something like (5*) became common ground, which he knew to be false.

  • (5*) Dave’s had enough heart medicine for his daily therapy.

Saul points out that most of us have the intuition that Fred lied. The intuition that Fred lied is captured if what Fred said is that Dave’s had enough heart medicine for his daily therapy. This consideration rules out all non-truth evaluable contents such as Bach’s (2006) propositional radicals. On Bach’s view, the semantic content of (5) is not a full proposition, but a propositional radical that lacks determinate truth conditions. The contents that figure in says-based definitions of lying must be truth-evaluable since those definitions require that liars believe that what they say is false.

  1. (iii)

    Non-minimal.

The sabotage of an appointment (Saul, 2012: 63).

Mary wants to sabotage Helga’s friendship with Iggy, a known despiser of the unpunctual. Iggy shows up at the house and Mary says:

  • (6) Helga is not ready.

Mary has the intention to deceive Iggy that Helga is not ready to go out with him, which Mary knows to be false. Saul points out that most of us have the intuition that Mary lied. This intuition is captured if Mary said that Helga is not ready to go out with Iggy. This consideration rules out minimal contents, both in Cappelen and Lepore’s (2005) account and in Borg’s (2012) account. In Cappelen and Lepore’s account, the minimal content of (6) is that Helga is not ready simpliciter. They do not say what it is to be ready simpliciter and leave it as a matter for metaphysicians. However, they agree that speakers might not be able to tell whether it is true or false that one is ready simpliciter. This is incompatible with the above definition of lying. In order for Mary to be lying, she must believe that what she says is false. Yet, Mary, like most of us, might have no idea whether it is true or false that Helga is not ready simpliciter.

In Borg’s account, the minimal content of (6) is that there is something for which Helga is not ready. This content is trivially true. It is, then, very unlikely that Mary believes that what she said is false, if what she said is Borg’s minimal content. This makes Borg’s minimal content that Helga is not ready for something as unsuited as Cappelen and Lepore’s content that Helga is not ready simpliciter.

The point can be generalized to many minimal contents: speakers do not have a grip on their truth value or they know that they are trivially true, making it impossible to lie with them in light of says-based definitions of lying.

  • (iv) Sensitive to QUDs.

Christmas Party (Stokke, 2018: 82).

At an office Christmas party, William’s ex-wife, Doris, got very drunk and ended up insulting her boss, Sean. Sean took the incident lightly, and their friendly relationship continued unblemished. More recently, the company was sold, and Doris lost her job in a round of general cutbacks. But, despite this, Doris and Sean have remained friends. Sometime later, William is talking to Elizabeth, who is interested in hiring Doris. However, William is still resentful of Doris and does not want Elizabeth to give her a job. Consider the following dialogues:

(7) Elizabeth::

Why did Doris lose her job?

William::

Doris insulted Sean at a party.

(8 ) Elizabeth::

How is Doris’s relationship with Sean?

William::

Doris insulted him at a party.

Stokke reports the intuition that William lied in (7) and did not lie in (8). With respect to says-based definitions of lying, this difference can only be explained by making what William says vary from (7) to (8). In turn, the difference in what William says can be explained by making what he says sensitive to the QUD he is addressing, because William utters exactly the same sentence.

I argue that a conventions-based notion of utterance content inspired by Korta and Perry’s (2007a) locutionary content and Devitt’s (2021) what is said meets the desiderata (i) to (iv). In the next section, I discuss the points that Korta and Perry’s locutionary content and Devitt’s what is said have in common with respect to linguistic conventions. In Sect. 4, I argue that such a conventions-based notion of utterance content is what says-based definitions of lying require for the role of what is said.

3 Conventions and Utterance Content

Korta and Perry (2007a: 172) present the notion of locutionary content as a kind of utterance content that is determined by the following intentions of the speaker (assuming English is the language):

  1. (i)

    The intention to produce grammatical phrases of English, by speaking, writing, typing, signing or other means;

  2. (ii)

    Doing so with appropriate intentions that resolve:

    1. a.

      which words, of those consistent with the sounds uttered (or letters typed…), are being used;

    2. b.

      which meanings of those permitted by the conventions of English for the words and phrases being used, are being employed;

    3. c.

      which of the syntactic forms consistent with the order of words, intonations, etc. are being employed;

    4. d.

      nambiguities; that is, issues about the reference of names which various persons, things, or places share;

    5. e.

      the primary reference of demonstratives and other deictic words and issues relevant to the reference of indexicals;

    6. f.

      anaphoric relations;

    7. g.

      the values of various other parameters that are determined by the speaker’s intentions.

Devitt’s (2021: 38) notion of what is said has many points in common with Korta and Perry’s notion of locutionary content. Devitt develops his notion of what is said as an utterance content that is determined by speakers’ exploitation of rules of language, largely established by conventions. Devitt argues that three sorts of properties are constitutive of what is said:

  1. (i)

    Properties arising from linguistic conventions;

  2. (ii)

    Disambiguation;

  3. (iii)

    Saturation.

Apart from terminological differences,Footnote 4 Korta and Perry’s locutionary content and Devitt’s what is said overlap in some important respects. First, both contents are non-Gricean. The speakers’ intentions (intentional statesFootnote 5) that are determinative of both kinds of contents are expressive and not communicative. Such intentions are independent of what the hearer does or is expected to do. This is not to say that Korta and Perry’s locutionary content and Devitt’s what is said are never meant in the Gricean sense. On the contrary, typically they are (part of) what speakers intend to communicate to the audience. Yet, the fact that a content p is said in the expressive sense does not entail that p is meant in the Gricean sense.

Second, there are important connections between Devitt’s properties (i) to (iii) and Korta and Perry’s (a) to (g). Devitt’s properties (i) arising from linguistic conventions correspond to Korta and Perry’s (a) and (b). Devitt’s disambiguation (ii) covers to a large extent Korta and Perry’s (b), (c), and (d). Devitt’s saturation (iii) covers to a large extent Korta and Perry’s (e), (f), and (g). Let us see why.

Devitt holds that the meaning of, for example, quantifiers, genitives, ‘it is raining’, and ‘ready’ demands saturation in context by an implicit reference the speaker has in mind. The linguistic conventions that govern the use of those expressions demand that the speaker has in mind a particular restriction of the domain of quantification for a quantifier, a particular relation for a genitive, a particular location where it is said to be raining, a particular course of action for which someone is said to be ready. Devitt says that the conventional meaning of those expressions has an implicit slot to be filled. Although Perry dislikes the term ‘saturation’, his view (subscribed by Korta) on unarticulated constituents is quite close (and precedent) to Devitt’s view on slots to be filled. Perry (2007) is a helpful text where to look for evidence of the conceptual connections between Perry’s unarticulated constituents and Devitt’s slots to be filled. For example, Perry (2007: 553) says that the presence of an unarticulated location in the content of ‘it is raining’ is demanded by the conventions that govern the use of that sentence. Devitt has the same view about weather reports. In the same vein, Perry (2007: 558) says that the rule that governs the meaning of ‘late’ demands that the speaker has in mind a particular event for which someone or something is said to be late. This is the same view that Devitt holds with respect to ‘ready’ and ‘enough’.

One important idea that Korta/Perry and Devitt share is the rejection of what Devitt calls ‘the tyranny of syntax’. Neither Perry’s unarticulated constituents nor Devitt’s slots to be filled are explained by positing elements in syntactic structures or logical forms. Korta/Perry and Devitt reject commitment to the Isomorphic Principle according to which there must be a correspondence between the structure of a sentence and the structure of the content it expresses.

Perry’s unarticulated constituents and Devitt’s slots to be filled are to be found in linguistic rules, which are posited to offer the best explanation of regularities of use. Perry (2007: 550) says that linguistic rules are registered in the ‘lexicon’. The lexicon Perry has in mind is a dictionary that supplies the information an artificial system would need in order to parse and use natural language sentences. That is the kind of information that is (implicitly) used by humans and needs to be made explicit for artificial systems. The lexicon registers thematic rules, like the one that demands a location for ‘it is raining’, and permissive rules, like the one that says that ‘it is raining’ can be used to say of a location that it has many of a number of properties: from sprinkling to monsoon-like rain. Perry (2007: 555) says that most of the conventions that govern words are permissive and permit to use them to stand for a number of things, often related to each other. Perry’s thematic rules correspond to Devitt’s rules that demand slots filling, and Perry’s permissive rules correspond to Devitt’s conventions that specify a plurality of related meanings,Footnote 6 which Devitt pushes to the extent of putting in doubt the existence of most generalized implicatures as a pragmatic phenomenon (e.g. the temporal and the causal/explanatory meaning of ‘and’).

There are important philosophical differences between Korta and Perry’s view and program and Devitt’s ones, and I will not oversimplify them. The central point is that they acknowledge a level of utterance content that is determined by the fact that the speaker exploits linguistic conventions and takes on the burden of filling the slots that those conventions demand to be filled. Linguistic conventions and what they demand generate a form of linguistic liability of speakers for the contents of their utterances that is (constitutively) independent from communicative intentions. This form of linguistic liability for utterance contents is independent of what hearers do or are expected to do.

In the next section I argue that a notion of utterance content inspired by Korta and Perry’s locutionary content and Devitt’ what is said is well suited to says-based definitions of lying.

4 Applications

4.1 The Rich Catholic

The intuition that John did not lie is predicted if John says that Jack got married and had two children in one order or the other. The conventions-based notion of utterance content explains that in a straightforward way. On that view, the conjunction ‘and’ is governed by a plurality of rules,Footnote 7 one for the truth functional meaning, one for the temporal meaning, and one for the causal/explanatory meaning. John exploits the truth functional meaning for expressing the content that Jack got married and had two children in one order or the other. John’s exploitation of the truth functional meaning is independent of his expectation about what his audience will do. If charged with lying, John might reply as followsFootnote 8:

  • (9) I did not lie. I did not say that Jack got married before having two children. I said that Jack got married and had two children in one order or the other.

In this reply John declares that it is the truth functional content that was the content of his assertion. John’s assertion is a fine response to the rich Catholic’s request, given the context and the lack of specificity of the request for information about Jack. John’s communicative strategy is tricky, but he is not lying. John’s communicative strategy is tricky because he plays with the plurality of conventions governing ‘and’. John intentionallyFootnote 9 takes responsibility for the truth functional meaning and at the same time has the expectation that the rich man will misunderstand his utterance with the temporally enriched meaning. John plans his misleading strategy relying precisely on the plurality of the meanings of ‘and’.

The conventions-based notion of utterance content has some important advantages over Saul’s notion of what is said. Saul’s notion of what is said emerges from composition of literal contents plus enrichments that are needed to guarantee truth-evaluability. Saul’s account predicts that John says that Jack got married and had two children in one order or the other, since the truth functional content does not require temporal enrichment for truth-evaluability. However, matters get more complicated with a slight variation of the scenario. Suppose the rich man asks John the following direct question and John gives the following answer:

(10) Rich man::

Did Jack get married and have two children or did he have two children and get married?

John::

Jack got married and had two children.

In this scenario the intuition that John lied is strong.Footnote 10 John cannot defend himself from the charge with lying as follows:

  • (9) I did not lie. I did not say that Jack got married before having two children. I said that Jack got married and had two children in one order or the other.

Assuming that the rich Catholic speaks literally, non figuratively, and is not ironic, and assuming that John is aware of that, the only interpretation of the question is that the rich Catholic wants to know the temporal order between Jack’s marital status and his parental status. (There are different explanations of how it comes that the rich Catholic’s question has that content. The conventions-based view maintains that the rich Catholic exploits the convention of using the conjunction ‘and’ with the temporal meaning). John cannot present his assertion as an answer to that question under that interpretation and claim that it has the truth functional content. His assertion would not be an answer to that question. Of course, John might pretend that he did not understand the question as a question about the temporal order between Jack’s marital status and his parental status and claim that he said that Jack got married and had two children in one order or the other. But in that case John must retreat his assertion as an answer to the rich Catholic’s question. Thus, either John presents his assertion as an answer to the question with the temporal content or he claims the truth-functional content for his assertion but retreats it as an answer to the question.

Saul’s account has a problem of undergeneration. The temporal enrichment is not required to get truth-evaluability. In Saul’s account, John says that Jack got married and had two children in one order or the other. Moreover, John believes that it is true that Jack got married and had two children in one order or the other, and (he believes that) he is in a warranting context. John is not lying according to Saul’s definition and notion of what is said. But this goes against our intuitions.

As said, the conventions-based notion of utterance content offers a neat explanation of this case. The rich man uses ‘and’ with the temporal meaning in his question, because he wants to know the temporal order between Jack’s marital status and his parental status. If John understands the question as about the temporal order between Jack’s marital status and his parental status, he is forced to use ‘and’ with the same temporal meaning, if he presents his assertion as an answer to that question under that interpretation. It is not simply a matter of cooperativeness.Footnote 11 John’s use of ‘and’ with the temporal meaning is forced by the convention of language that the rich Catholic exploits in his assertion, by his understanding of that question with the temporal content, and by his willingness to answer that question.

Stokke’s account of what is said correctly predicts that John is lying in the above scenario. The rich man raises a local QUDFootnote 12 about the temporal order of Jack’s marital status and parenthood. The content that Jack got married before having two children is the weakest proposition that answers that QUD and entails the minimal content. Yet, Stokke’s notion of what is said is not without difficulties. Consider the following dialogue.

A::

How many children does Jack have?

B::

Jack got married three years ago and had one child.

There seems to be nothing inappropriate in B’s answer. Of course, B might have responded with ‘Jack has one child’. But suppose that B intends to convey the richer information that Jack got married three years ago and then he had just one child. A regular way for conveying that content is to assert the sentence ‘Jack got married three years ago and had one child’. There is large consensus among philosophers of language that the information that John got married three years ago and then he had just one child is the content that is said by B’s utterance. The temporal relation between Jack’s getting married and his having a child and the exact number of his children are constitutive parts of what is said. Although there is disagreement among philosophers on whether the temporal order and the exact number of children are the result of free pragmatic processes or processes that are governed by conventions of language, there is wide agreement that they are not implicatures (generalized).Footnote 13

This case raises a difficulty for Stokke’s notion of what is said. There is an explicit QUD, which does not ask anything about Jack’s marital status, let alone about the temporal order between Jack’s getting married and his having a child. The weakest proposition that entails the minimal content of the uttered sentence and is a (partial or complete) answer to the QUD is that Jack got married three years ago and had one child in one order or the other. There might also be a difficulty as to the exact number of Jack’s children. For, if one holds that the minimal content of ‘one’ is at least one, then the weakest proposition that entails the minimal content of the uttered sentence and is a (partial or complete) answer to the QUD is that Jack got married three years ago and had at least one child in one order or the other, which selects the set of all possible world in which Jack is married and has one or more children independently of the temporal order between his marital and parental status.

As far as I can see, there are two options for Stokke. The first is to maintain that the temporal order of Jack’s marital and parental status and the exact number of his children are implicatures. As said, this is not what most philosophers of language agree on nowadays. The second is to introduce an intermediate level of content between his what is said and implicatures. This is not a parsimonious manoeuvre, and parsimony is an epistemic virtue of theories. I acknowledge that to say this much is far from providing a knockdown argument against Stokke’s account, but it is enough for pointing out some hard difficulties that affect his notion of what is said.

4.2 The Murderous Nurse and the Sabotage of an Appointment

Saul’s intuitions are that Fred says that Dave has had enough heart medicine for his daily therapy and Mary says that Helga is not ready to go out with Iggy. Fred and Mary lie. The conventions-based notion of utterance content captures these intuitions. The meanings of ‘enough’ and ‘ready’ call for saturation. Fred takes responsibility for saturating ‘enough’ with heart medicine for the daily therapy and Mary for saturating ‘ready’ with going out with Iggy (more on this point below).

Cases involving incomplete expressions like ‘enough’ and ‘ready’ raise difficulties for both Saul’s and Stokke’s account of what is said. Saul agrees that Fred says that Dave has had enough heart medicine for the daily therapy and Mary says that Helga is not ready to go out with Iggy. Saul’s account of what is said, however, does not rule out Borg’s minimal contents that Dave has had enough of something and that Helga is not ready for something, since they are truth-evaluable. Saul does not provide a principled explanation for ruling out these minimal contents except that taking them to be what is said is in contrast with the intuition that Fred and Mary lie. Saul (2012: 58) acknowledges this point and rejects Borg’s minimal contents on the basis of what she takes to be wide consensus among scholars. Yet, Saul’s manoeuvre is ad hoc. Without specific saturations we could not get contents that are in line with the intuitions that Fred and Mary lie. But there is no explanation of why Fred and Mary are committed to those saturated contents. The conventions-based notion of content provides instead an explanation in terms of linguistic conventions and linguistic liability. The linguistic conventions for ‘enough’ and ‘ready’ demand saturation, and the context of the conversation constrains the way speakers can intelligibly and rationally perform it.Footnote 14 Here lies an important theoretical advantage of the conventions-based notion of utterance content over Saul’s notion of what is said. Let me elaborate on this point.

Consider the sabotage of an appointment case. The prevalent intuition is that Mary lied. In order to get this prediction, there must be a proposition that is said by Mary. The difficulty is to tell what in fact Mary said. The dialogue between Mary and Iggy can be described as if Mary said and Iggy understood any completion among the following ones (the list is not exhaustive):

  • (A) Helga is not ready to go out with you.

  • (B) Helga is not ready to meet you.

  • (C) Helga is not ready to go to the theatre.

  • (D) Helga is not ready to go to the city centre.

  • (E) Helga is not ready to leave the house.

  • (F) Helga is not ready to anything but to stay at home.

All completions (A)-(F) fit well Mary’s deceptive plan, provided that Iggy understands one of them (assuming a certain common ground, for example that Iggy wants to take Helga to the theatre, that the theatre is in the city centre, etc.). Saul (2012: 64–65) maintains that what Mary says is indeterminate across a range of admissible completions and revises her definition of lying as follows:

If the speaker is not the victim of linguistic error/malapropism or using metaphor, hyperbole, or irony, then they lie iff A or B holds:

A::

(1) they say that p, (2) they believe p to be false, (3) they take themself to be in a warranting context.

B::

(1) they say something indeterminate across a range of acceptable complete propositions P1… Pn, (2) for each complete proposition in the range P1… Pn, they believe that proposition to be false, (3) they take themselves to be in a warranting context.

Saul’s view is poorly specified, as she (2012: 64) acknowledges. It moves from the uncontroversial insight that typically in an utterance of a semantically underdetermined sentence the speaker is willing to answer the request for more explicitness about what is said with one of the completions that are contextually relevant, and is happy if the hearer understands one of them. For example, if asked to be more explicit, Mary might be willing to respond that she said that Helga is not ready to go out with Iggy, or that Helga is not ready to go to the theatre, or that Helga is not ready to leave the house, and so on. And Mary is happy if Iggy understands one of those propositions. The insight is uncontroversial, but one important aspect is in need of explanation. That is the selection of the range of the contextually relevant completions. Not all completions are admissible. The range of admissible completions must be restricted to those that are contextually relevant. But what counts as a contextually relevant completion?

The conventions-based notion of utterance content provides an explanation for this central point. The fact that typically in an utterance of a semantically underdetermined sentence if the hearer asks the speaker to be more explicit about what they said, the speaker is willing to make a completion explicit is underestimated. The point is not simply that the speaker is willing to provide a completion. The point is that the speaker is compelled to do so by a semantic rule. Consider the following dialogue:

(11) Mary::

Helga is not ready.

Iggy::

What for?

Mary::

For something, but I have no idea what.Footnote 15

Mary’s response is odd. The oddity is not simply a matter of lack of cooperativeness due to violation of the maxim of quantity. Compare the above dialogue with the following one:

(12) Mary::

Helga is not ready.

Iggy::

What is Helga not ready for?

Mary::

Helga is not ready to take her logic exam.

Mary’s response ‘Helga is not ready to take her logic exam’ is not cooperative as it violates the maxim of relation. It might also be irrational, given the context. Nonetheless, Mary looks competent in her use of ‘ready’. In fact, Iggy understands that Mary said that Helga is not ready to take her logic exam, although he likely does not understand why Mary said that. Mary is uncooperative and might seem irrational, but Iggy takes her utterance to have a meaningful content. On the contrary, Mary’s response ‘For something, but I have no idea what’ reveals incompetence on the use of ‘ready’ in the previous assertion.

This is not to deny that there are contexts in which the speaker might assert, say, that Helga is not ready for something. Think of a philosophical talk on the metaphysics of readiness. Notice, however, as Bach (1994) pointed out long ago, that when speakers intend to assert that Helga is not ready for something they do not utter the sentence ‘Helga is not ready’, but the sentence ‘Helga is not ready for something’.

A straightforward explanation of the oddity of the ‘no idea’ response is that Mary is compelled to provide a completion by a convention of language. One cannot competently use ‘ready’ without taking on the semantic burden of making explicit what someone or something is said to be ready for. The burden is semantic because without a completion the utterance is meaningless and a misuse of language. Without a completion the speaker is not exploiting the conventional meaning of ‘ready’, but abusing that meaning. The commitment to provide a completion is prior and independent of communicative intentions. It comes from the speaker’s use of an expression in accord with its conventional meaning, and it belongs to the expressive level (one might call this the locutionary level), not to the communicative level. The intentions that are constitutive of this conventions-based utterance content are expressive intentions, not communicative, audience-directed intentions. The proposition expressed by the utterance of a semantically underdetermined sentence need not be the content that the speaker expects the hearer to recognize as the proposition said. Rather, it is the content to which the speaker is committed by the conventional meanings of the expressions that occur in the uttered sentence. Conventions of language generate the commitment of the speaker to the content of the utterance.

This picture yields a reassessment of semantic underdetermination in line with Korta and Perry’s and Devitt’s orientations. Semantic underdetermination is explained as a linguistic phenomenon similar to the context dependence of overt indexicals and demonstratives (e.g. the Basic Set of Cappelen and Lepore 2005). The idea is that the contextual supplementations that enter the contents of utterances of sentences containing semantically underdetermined expressions are linguistically governed by conventions of language that are codified in conventional meanings like Kaplanesque characters of indexicals and demonstratives. Expressions like ‘ready’ are semantically underdetermined in a way that is analogous to the way in which the indexical ‘here’ and the demonstrative ‘that’ need a reference assignment in context. I am not claiming that indexicality and semantic underdetermination are one and the same phenomenon. I am claiming that they are analogous in the sense that both are linguistically governed by conventions of language. One cannot competently use ‘ready’ without taking on the semantic burden of making a contextual completion explicit, just as one cannot use ‘here’ and ‘that’ without having a particular referent in mind and taking on the semantic burden of making it explicit.Footnote 16

Since there must be a completion to which the speaker is committed in virtue of conventions of language, the proposition said by the utterance of a semantically underdetermined sentence can be identified with the proposition that the speaker is willing to explicitly express if asked to do so. That proposition must exists and can be identified with what is said by the utterance of a semantically underdetermined sentence, though it is said not in the Gricean sense that requires audience-directed intentions. Thus, Wettstein’s (1981: 247) objection that often not even the speaker is able to tell what is the completion that they intended to communicate at the very moment of the utterance misses the target.Footnote 17 The completion need not be consciously represented in communicative intentions at the time of the utterance. In ordinary communication, typically the speaker is not asked to be fully explicit and does not need to think of a particular contextual completion. But the speaker takes on the semantic burden of providing a completion and must have the disposition to do so, on pain of being linguistically incompetent.

Once the existence of the proposition said is established by the fact that the speaker is exploiting conventions of language and is linguistically compelled to provide a completion, the range of the contextually relevant completions can be delimited. As said, scholars rightly point out that the speaker is happy if the hearer understands one of the contextually relevant completions. The point is to say what counts as a contextually relevant completion. I am not in the position to provide a detailed account, but the idea seems intuitive enough. Suppose Mary has the disposition to provide the completion that Helga is not ready to go to the theatre. The proposition that Helga is not ready to go to the theatre is what is said by her utterance. That proposition fits well Mary’s communicative plan, of course. The range of the contextually relevant propositions can be delimited. A proposition is contextually relevant if and only if it stands in the relation of communicative equivalence to the proposition that Helga is not ready to go to the theatre. I am not able to give the analysis of the relation of communicative equivalence. Yet, pre-theoretically, two propositions are communicatively equivalent if and only if they can be replaced one by another to fulfil the speaker’s communicative plan. A proposition counts as contextually relevant if and only if it is communicatively equivalent to the proposition said. In Mary’s example, a proposition is contextually relevant if and only if it is communicatively equivalent to the proposition that Helga is not ready to go to the theatre (assuming that Mary has the disposition to make explicit this completion). Mary is happy if Iggy understands one of the propositions (A)-(F) because each of them fits well her communicative plan.

To sum up, the advantages of the conventions-based notion of utterance content over Saul’s notion of what is said are that it provides a principled way for ruling out Borg’s minimal contents and an explanation of the commitment of speakers to contents when they utter underdetermined expressions. Since this kind of commitment is linguistic, as it originates in conventions of language, it explains the responsibility of liars for disbelieved information that says-based definitions of lying try to capture with the notion of what is said and need in order to distinguish lying from misleading.

Saul (2012: 64) says that with regard to incompleteness the range of possible completions is restricted by speaker intentions, audience interpretation, and salience. When these agree things are fairly simple, and when they do not agree intuitions are murky. Saul underestimates the matter at stake. First, audience interpretation and salience have no constitutive role for the contents that serve as what is said in says-based definition of lying, as the discussion of the rich Catholic case showed. The case can be strengthened with the following example. Imagine a scenario identical to the murderous nurse case except that Fred is myopic, misidentifies the bottle of heart medicine for a bottle of water, and says ‘Dave has had enough’ with the intention of expressing the content that Dave has had enough water, and knowing that it is true that Dave has had enough water. Even if it happens that Ed misunderstands Fred saying that Dave has had enough heart medicine, Fred has not lied.

Second, it is fundamental that the speaker intentions are expressive and not communicative and audience-oriented, otherwise the distinction between lying and misleading that says-based definitions aim at preserving is lost. Moreover, since Wettstein (1981), it is well known that audience-oriented intentions are not suited for constituting contents with completions. But, and this is the crucial point, the expressive intentions must also generate the commitment of the speaker to the expressed content, because commitment is constitutive of lying. It is not enough to give an account of how the speaker expresses a propositional content because, say, the speaker has the intention to express that content. The account must also explain the commitment of the speaker to the expressed content. On this crucial point Saul’s account is silent and ex post since it simply registers that fact that the speaker is willing to provide a completion if asked to do so. The conventions-based view, instead, provides an explanation of the commitment. Liars are committed to the contents that they cannot avoid expressing when they (literally) use words with their conventional meaning. With regard to incompleteness, conventions of language guarantee that there is a propositional content expressed. It is the content that the speaker is willing to explicitly express if asked to do so. Convention of language compels the speaker to provide a completion, if asked to do so, on pain of being linguistically incompetent.

Incomplete expressions like ‘enough’ and ‘ready’ raise difficulties for Stokke’s account too. Consider a slight variation of the sabotage of an appointment case and imagine Iggy asking:

  • (13) Is Helga at home?

Iggy raises the local QUD as to whether Helga is at home. Mary answers:

  • (6) Helga is not ready.

Mary does not address the local QUD raised by Iggy. Therefore, among the possible saturations for ‘ready’, none is an answer to the local QUD. Stokke’s (2018: 109) says that when an utterance is indeterminate across a range of minimal contents and cannot be construed as saying anything relative to the local QUD, the interpretation takes as what is said the union of all possible completions, which can be paraphrased with a generalized proposition. In Mary’s case, the generalized proposition is identical to Borg’s minimal content that Helga is not ready for something (see Stokke, 2018: 113 for a similar example).

The price Stokke has to pay in order to avoid the absurd conclusion that nothing is said in (6) is very high. In fact, the generalized proposition that Helga is not ready for something is trivially true, and this makes it impossible for Mary to lie. This goes against the intuition that Mary lies in (6).

One might object that this argument against Stokke’s view starts from a false premise. I assumed that since Mary is not addressing the local QUD raised by Iggy, she is not addressing any other local QUD and, therefore, is defaulting to the Big Question, as Stokke (2018: 104) says. The Big Question does not help select any particular completion and, on Stokke’s view, when there is no selected completion the interpretation takes as what is said the union of all possible completions, which can be paraphrased with a generalized proposition. One might object that my reconstruction of Mary’s case is mistaken. One might argue that Mary is addressing an implicit QUD about whether Helga is ready to go out with Iggy (or any other relevant completion). Therefore, according to Stokke, Mary is saying that Helga is not ready to go out with Iggy, and she is lying.

My counter-objection is that my argument rests on a case that is structurally identical to a case that Stokke (2018: 86) discusses. Furthermore, even if my case were not structurally identical to Stokke’s one, Stokke’s case is already enough to mount an argument to the conclusion that Stokke’s notion of what is said is not the best theoretical choice for says-based definitions of lying. Here is Stokke’s case.

Larry’s book (Stokke, 2018: 86).

Larry is keen on making himself seem attractive to Norma. He knows she’s interested in logic, a subject he knows nothing about. From talking to her, Larry has become aware that Norma knows that he has just finished writing a book, although she doesn’t know what it’s about. In fact, the book Larry wrote is about cats. Recently, Larry also joined an academic book club where the members are each assigned a particular book to read and explain to the others. Larry has been assigned a book about logic.

(14) Norma::

Do you know a lot about logic?

Larry::

My book is about logic.

Stokke says that Larry does not lie. I agree on Stokke’s verdict, but reject Stokke’s explanation of why Larry does not lie. Stokke holds that Larry’s utterance is indeterminate across a range of contents. None of them entails an answer to the local QUD raised by Norma. The difficulty for Stokke is to specify a content that is said by Larry’s utterance. As noted above, Stokke’s says that when an utterance is indeterminate across a range of contents and cannot be construed as saying anything relative to the QUD, the interpretation takes as what is said the union of all possible completions, which can be paraphrased with a generalized proposition. In Larry’s case, the generalized proposition is the following (Stokke, 2018: 109):

  • (15) Larry bears some relation to a book about logic.

The generalized proposition is trivially true and helps capture the intuition that Larry’s is not lying. Yet, Stokke’s explanation is wrong. Stokke says that Larry conveys the disbelieved information that he knows a lot about logic through a conversational implicature. Stokke does not give a full account of the rational reconstruction of the implicature, but he (2018: 110) claims that the generalized proposition, as what is said, is the input for the derivation of the implicature. I think that his sketch of the rational reconstruction of the implicature is wrong, because generalized propositions are an idle wheel. Stokke (2018: 110) suggests that Larry exploits the fact that Norma believes that he wrote a book about logic and the Maxim of Relation. I agree on this. Yet, the explanation of how Norma comes to believe that Larry wrote a book about logic is that she understands Larry’s use of the possessive ‘my’ as if Larry intended to saturate it with the relation of being the author of the book. She is led to that understanding by the information that Larry has just finished writing a book and by the linguistic convention that the possessive ‘my’ is in need of saturation. Larry expects that Norma understands his utterance with the content that the book he wrote is about logic. And this content is the input for the derivation of the implicature that Larry knows a lot about logic. Larry avoids lying to the extent that he can retreat to the content that the book he has been assigned to read is about logic. Stokke’s generalized proposition that Larry bears some relation to a book about logic plays no theoretical role.

Notice that it would be risky for Larry to convey that he knows a lot about logic by means of an implicature that takes as an input the generalized proposition. The content that he bears some relation to a book about logic is not informative, since it is trivially true. This might lead astray Norma to wonder why Larry said something that is not informative and prevent her from grasping the implicature.

There is another important difficulty for the view that generalized propositions play the role of what is said. Generalized propositions are not contents to which speakers can retreat in order to discharge the accusation of lying. They cannot play the role of fallback contents, which is one of the desiderata for the role of what is said in says-based definitions of lying. Suppose Norma finds out that Larry did not write a book about logic and knows nothing about logic, and charges Larry with lying:

(16) Norma::

You lied, you did not write a book about logic.

The following answer by Larry is implausible:

(17) Larry::

I did not lie. I meant that I bear some relation to a book about logic.

There are at least two difficulties here. First, given the context and the common ground, the information that Larry bears some relation to a book about logic is trivially true and not informative. Thus, it is not convenient to Larry to claim responsibility for a content that is trivially true. What is the point of claiming responsibility for a trivial truth? Generalized propositions are trivially true or trivially false and, therefore, in most conversational contexts it is not rational to claim responsibility for them as contents of assertions or constative speech acts in general (more on this point below). Second, it is doubtful that ordinary speakers are aware of generalized propositions and manage to plan their misleading strategies relying on them. On the contrary, it is certain that possessives are regularly used for expressing specific completions. Arguably, ordinary speakers do not understand those attempts to discharge the accusation of lying that do not appeal to specific completions. It is worth noting that Stokke himself (2018: 86) envisages the following continuation of the dialogue, in which Larry tries to avoid commitment to lying:

(18) Norma::

Do you know a lot about logic?

Larry::

My book is about logic.

Norma::

Oh, you wrote a book about logic?

Larry::

No, I just meant that the book I am assigned to read is about logic.

In this continuation of the dialogue, Larry does not retreat to the generalized proposition in order to avoid commitment to lying, but to a particular completion.Footnote 18 Of course, in doing so, Larry discloses his behaviour as uncooperative, but he can defend the literalness of his utterance, though clearly misleading, and above all his role of asserter.

This last point about the role of asserter is central and goes deeper than mere lack of cooperativeness due to the fact that the assertions of most generalized propositions are not informative (being those propositions trivially true or trivially false in most conversational contexts). Stokke (2018: 110) says that Larry is naturally seen as asserting that he bears some relation to a book about logic. On Stokke’s account this means that Larry is naturally seen as proposing to add to the common ground the information that he bears some relation to a book about logic. The problem with this view is that it is not rational for Larry to propose to add that information to the common ground, because that information is already part of the common ground. Given the context of conversation, it is mutually accepted that Larry exists, that books about logic exist, that for any of them, Larry bears some relation to it (for example, given any book about logic, Larry is or is not the author of that book). Given the context of conversation, then, it is already mutually accepted that Larry bears some relation to a book about logic. Therefore, it is not possible for Larry to rationally intend to propose to add that information to the common ground. Contrary to Stokke’s view, Larry cannot be seen as rationally asserting that he bears some relation to a book about logic.

One might try to defend Stokke’s view by claiming that although he actually holds that Larry is asserting that he bears some relation to a book about logic, that claim is not a necessary feature of his view. One might grant that Larry does not address the QUD raised by Norma and defaults to the Big Question. The Big Question does not select any particular completion and the interpretation takes as what is said the union of all possible completions, which can be paraphrased with the generalized proposition. That generalized proposition is the content of Larry’s utterance. Still, one might claim that Larry’s utterance is deprived of direct illocutionary force. Larry is not making a direct assertion. This view is in line with the Stalnakerian framework, according to which one cannot directly assert something that is accepted as trivially true or trivially false in the conversational context, because it is already part of the common ground and, thereby, it is not rational to propose to add that content to the common ground.Footnote 19 This view returns the verdict that Larry is not lying when combined with Stokke’s definition and avoids the problematic conclusion that Larry asserts that he bears some relation to a book about logic.

This attempt to defend Stokke’s account fails. According to it, Larry’s utterance is deprived of literalness in the sense that its literal content is deprived of direct illocutionary force. More or less this can be compared with what happens with typical cases of irony and non-literal or figurative uses of language. In those cases, what is literally said is not the content of a direct illocutionary act. The problem with this view is that it is implausible to attribute to Larry the intention of speaking non-literally. It is implausible both on the side of the psychological processes in the production and understanding of utterances and on the side of the rational reconstruction of the logical relations among the information that is available in those processes. Generalized propositions are idle wheels on both sides.

First, there is no evidence that ordinary speakers are aware of generalized propositions and that they plan their communicative strategies relying on them. On the contrary, as noted above, when Stokke envisages how Larry might try to discharge the accusation of lying, he envisages Larry replying that he meant that the book he was assigned to read is about logic. Stokke does not envisage Larry replying that he meant that he bears some relation to a book about logic or that he was not speaking literally or not making a direct assertion. After all, how could Larry rationally intend not to make a direct assertion by uttering the sentence ‘My book is about logic’ in that context? How could Larry expect that Norma does not interpret his utterance literally as a full-fledged direct assertion?

Second, there is no need for generalized propositions in the rational reconstruction of the logical relations among the information that is available during the production and understanding of utterances. From the point of view of the rational reconstruction of such processes, Korta and Perry (2007b) have forcefully argued that the information about conventional meanings and what they demand to be filled in is the input for a semantic module or semantic competence that returns token-reflexive contents.Footnote 20 Token-reflexive contents together with contextual information are input to derive the propositions expressed by utterances.Footnote 21 The propositions expressed are input to derive implicatures and indirect speech acts. In this rational reconstruction of the logical relations among the information available in the production and understanding of utterances, generalized propositions play no theoretical role. No one doubts that Norma understands that Larry knows a lot about logic through a conversational implicature. However, the input for the implicature is the proposition that the book that Larry wrote is about logic. That proposition is the result of combining the conventional meaning of the possessive ‘my’ with the contextual information that Larry wrote a book. No theoretical role is left for the generalized proposition that Larry bears some relation to a book about logic.

4.3 Christmas Party

Stokke employs the Christmas Party case to argue in favour of the view that what is said is a function of the question under discussion and the minimal content of the uttered sentence. William is lying in (7) and is not in (8), although he utters the same sentence with the same minimal content in both occasions. Stokke explains the difference between (7) and (8) in terms of a difference in what is said. Williams says that Doris lost her job because she insulted Sean at the party in (7), and that Doris insulted Sean at the party in (8). Stokke argues that his notion of what is said is needed to explain this difference. The aim of this subsection is to show that Stokke’s argument is not strong enough to establish the conclusion that his notion of what is said is needed in order to cope with sensitivity to QUDs. I contend that the conventions-based notion of what is said provides an alternative explanation with some theoretical advantages over Stokke’s explanation.

Van Elswyk (2020) raised an objection to Stokke’s account. I argue that the advantage of the conventions-based notion of utterance content is that it provides an explanation of both Stokke’s case and Van Elswyk’s counterexample. Van Elswyk rephrases William’s answer with two changes that have no truth-conditional import.

(19) Elizabeth::

Why did Doris lose her job?

William::

Well, Doris did insult Sean at a party.

Since neither the discourse marker ‘well’ nor the difference, for a verb V (insult), between Ved (insulted) and did V (did insult) contributes to truth conditions, the sentence ‘Doris insulted Sean at the party’ and the sentence ‘Well, Doris did insult Sean at the party’ express the same minimal content. The local QUD does not change from (7) to (19) and, thereby, Stokke’s account predicts that William says that Doris lost her job because Doris insulted Sean at the party in (19) too. William, then, ought to be lying in (19). Van Elswyk argues that this prediction is false. The presence of ‘well’ conveys that William’s reply is not an answer to the local QUD, although the presence of ‘did’ signals that his reply might provide evidence germane to the local QUD. As Van Elswyk points out, William can deny the information that Doris lost her job because she insulted Sean at the party as follows:

(20) William::

Well, Doris did insult Sean at a party. But Doris did not lose her job because she insulted Sean.

Likewise, if charged with lying, William can defend himself as follows:

  • (21) I did not lie. I only said that Doris insulted Sean at the party.

Van Elswyk draws the conclusion that Stokke’s account suffers from overgeneration and the proper notion of what is said that is required by says-based definitions of lying remains elusive. I agree with Van Elswik on the first part of his conclusion but disagree on the second part.

In the second part of his conclusion Van Elswyk overlooks the role of conventions. Speakers regularly respond p to question why q? in order to say q because p. Extending Devitt’s view on sub-sententials (2021: ch. 12), a viable explanation for this regularity might call for a linguistic convention. Speaker conventionally use p as a shorthand for q because p in response to the question why q?Footnote 22 This is not to say that p is ambiguous. P retains its conventional meaning, but there is a linguistic convention of using p with its conventional meaning in response to the question why q? as a shorthand for q because p.Footnote 23 William exploits this convention in (7) and says that Doris lost her job because she insulted Sean at the party. William lies. In (8) there is no why question and William does not exploit the same convention. William says that Doris insulted Sean at the party. William does not lie. In (19) there is a why question but William uses the marker ‘well’ just to signal that he is not exploiting the convention of uttering p as a shorthand for q because p. William says that Doris insulted Sean at the party. William does not lie.

It is worth noting that by appealing to conventions we obtain an explanation of why in (7) William is committed to the content that Doris lost her job because she insulted Sean at the party. It is not sufficient to give an account of the fact that William expresses that content. It is also necessary to give an account of William’s commitment to that content. In (7) William cannot utter ‘Doris insulted Sean at the party’ without saying that Doris lost her job because she insulted Sean at the party. If charged with lying, William cannot defend himself as in (21). William’s commitment is explained because he is responsible for that content in consequence of his exploitation of a linguistic convention. One way to avoid that commitment is to make explicit the suspension of the convention by using a marker like ‘well’ in (19). This account, then, explains well also Van Elswyk’s counterexample.

I close this section by addressing the objectionFootnote 24 that the view that there is a convention of using p as a shorthand for the answer q because p in reply to the question why q? is in tension with cases like Paul’s party. One might think that there is a convention that if a speaker asks a yes/no question about whether the addressee will attend an event E and the addressee does not reply with ‘no’ but with p, and p is a reason for not attending E, p is a shorthand for ‘I will not attend E, because p’. If there is such a convention, Dennis would be lying when he responds ‘I have to work’ to the question whether he will attend Paul’s party, contrary to what is assumed in this paper.

The objection raises an interesting point but it is not an objection to the view, defended in this paper, that a conventions-based notion of what is said is the most promising for says-based definitions of lying. Let me elaborate on this reply. Suppose there is a convention of using p for ‘I will not attend E, because p’ in response to an invitation to attend an event E. Then, the first conclusion to draw would be that Dennis is lying when he answers ‘I have to work’, since what he says is that he will not got to Paul’s party because he has a lot of work to do. But the second conclusion to draw would not be that the conventions-based notion of what is said does not explain this case. The second conclusion to draw would be that the intuition of most philosophers that Dennis does not lie is mistaken. And if it is a fact that Dennis lies, the conventions-based notion of what is said explains it just by acknowledging the role of the above convention in the constitution of what is said. Whereas, neither Saul’s nor Stokke’s notion of what is said explain it. Thus, the existence of a convention like the above one would not pose a difficulty to the conventions-based notion of what is said. Rather, it would require a revision of most philosophers’ intuition that Dennis does not lie.

However, there are reasons, both empirical and theoretical, to believe that there is no such a convention. Domaneschi and Vignolo (2022) have conducted an experiment testing whether laypeople are disposed to use particularized implicatures for the purpose of conveying disbelieved information without lying. Participants are invited to imagine a fictional scenario in which (i) they have bet with a friend that they will not lie for the next three days and (ii) they want to convey disbelieved information to a fictional character without lying in order to win the bet. One of the vignettes of the experiment is modeled upon Dennis’s case. Participants are invited to imagine a scenario in which they want to convey the disbelieved information that they will not go to Paul’s party but, in order to win the bet, they do not want to lie. Participants are presented with a target sentence and are asked to rate it on a 1–5 points Likert scale, with 1 indicating strong disagreement that the target sentence is helpful to convey the disbelieved information without lying and 5 indicating strong agreement. The target sentence conveys the disbelieved information explicitly as what it says (i.e. ‘I will not go to Paul’s party’), or as a presupposition or a generalized implicature (i.e. ‘I regret that I will not go to Paul’s party’) or as a particularized implicature (i.e. ‘I have a lot of work to do by tomorrow’). The collected data show that the participants’ agreement that the sentence ‘I have a lot of work to do by tomorrow’ is helpful for the purpose of communicating the disbelieved information (that the speaker will not go to Paul’s party) without lying is above the midpoint of the scale (3). This is evidence that there is no convention of using p to say ‘I will not attend the event E because p’.Footnote 25 Otherwise, participants’ agreement that the sentence ‘I have a lot of work to do by tomorrow’ is helpful for conveying the disbelieved information without lying would not be so high. In fact, the utterance of that sentence would be a plain lie.

There are also theoretical reasons for being skeptic about the existence of a convention like the one envisaged in the above objection. The first thing to note is that the convention lacks in generality. Compare it with the convention of using p as a shorthand for q because p in reply to the question why q?. P is introduced as the cause of q due to the mere fact that it is an answer to a why-question and independently of the content of p and q. No matter how surprising, unexpected, unusual, etc. p is as a cause of q, the speaker says that q because p. On the contrary, the usage of p as a shorthand for ‘I will not attend E because p’ works only against the background assumption that p is a reason for not attending E. This point suggests that the communication that the speaker will not attend the event E is a pragmatic phenomenon, in fact a particularized implicature.

There are also theoretical resources for explaining the regularity of such a pragmatic phenomenon. As Jary (2013) and Kissine (2016) point out,Footnote 26 there are particularized implicatures that are not necessarily recovered through inferences about the speaker’s communicative intentions but are recovered because the content expressed by the utterance makes accessible some background assumptions that allow the conclusion of the implicature. For example, the fact that the speaker says that she has a lot of work to do makes accessible the background assumption that having a lot of work to do by the following day is a good reason for not attending an event in the evening, which warrants the implicature that the speaker will not attend the event in the evening.

Moreover, the strategy of introducing conventions like the one of using p in response to the invitation to attend an event E to say that the speaker will not attend E because p, when p is a reason for not attending E, suffers from overgeneration. One might think that there is a convention of using p in order to say that the speaker will not do X in response to the invitation to do X, when p is a good reason for not doing X. For example, one might think that there is a convention of using ‘I have already had breakfast’, in response to the invitation to have something to eat, in order to say that the speaker will not have anything to eat because they have already had breakfast, given that having already had breakfast is a good reason for refusing to have something to eat. Yet, there is a very large consensus among scholars that the information that the speaker will not have anything to eat is a particularized implicature.

To sum up, there are empirical and theoretical reasons for believing that there is no convention like the one envisaged in the above objection and, thereby, for believing that the type of solution advocated here to deal with cases like the Christmas Party is not in tension with cases like Paul’s Party.

4.4 Summary on Theoretical Desiderata

Propositional contents must meet the following desiderata in order to cover the role of what is said in says-based definitions of lying. They must be (i) fallback contents, (ii) truth-evaluable, (iii) non-minimal, (iv) sensitive to QUDs. In Sects. 4.1, 4.2, 4.3 I argued that conventions-based utterance contents meet these desiderata. Conventions-based utterance contents are fully propositional and truth-evaluable and meet desideratum (ii), which is demanded by the condition in the definitions of lying that the speaker disbelieves the conveyed information. With regard to contents expressed by incomplete expressions, this condition rules out Bach’s propositional radicals, since they are not truth-evaluable, and Cappelen and Lepore’s minimal contents, since in most cases speakers have no idea on their being true or false. Borg’s minimal contents are ruled out too. The point is that many of Borg’s minimal contents expressed by incomplete expressions are generalized propositions that are trivially true. Therefore, according to say-based definitions of lying, it is impossible to lie with them, which goes against our intuitions and data about everyday use of language. Conventions-based utterance contents are not like minimal contents in Cappelen and Lepore’s account, nor in Borg’s account, and meet desideratum (iii). With this respect, conventions-based utterance contents have important theoretical advantages over Saul’s and Stokke’s what is said. First, Saul does not provide a principled way in order to rule out Borg’s minimal contents in those cases in which the speaker lies with incomplete sentences. Stokke appeals to Borg’s minimal contents in order to explain in some other cases why the speaker is not lying but misleading with implicatures. Conventions based utterance contents provide a principled way to rule out Borg’s minimal contents for explaining cases of lying and a better explanation of cases of not lying than Stokke’s explanation, since speakers are very often not aware of generalized propositions and generalized propositions are idle wheels in the calculation of implicatures.

Second and more important, the intentions that are constitutive of conventions-based utterance contents are not communicative but expressive. They are not audience-oriented intentions about what the audience will rationally understand. They are intentions of using words with their conventional meaning. This point is central for explaining the kind of commitment that, according to says-based definitions of lying, ties liars to disbelieved information. The commitment is not explained in terms of conversational responsibility for what the speaker is rationally understood to communicate or what the speaker expects the audience will rationally understand. Indeed, scholars who defend says-based definitions of lying employ deniability of disbelieved information as a diagnostic test for the commitment that is constitutive of lying. When speakers defend themselves from the accusation of lying by denying responsibility for the disbelieved information, they make manifest that they were intentionally uncooperative and, thereby, deceitful. Their defence, then, is not from the accusation of being deceitful, but from the accusation of being deceitful by lying. From this point of view, the commitment that ties liars to disbelieved information is stricter than the conversational responsibility that is based on Gricean audience-oriented intentions and cooperativeness. Conventions of language explain this kind of commitment. Liars are committed to those contents that they cannot avoid expressing when they use words in accordance with their conventional meaning.

This characteristic makes conventions-based utterance contents particularly apt to meet desideratum (i). Misleaders disclaim responsibility for the disbelieved information and claim responsibility for fallback contents. The view that the commitment in question resides in the exploitation of conventions of language provides an explanation that is better than Saul’s and Stokke’s explanation. It explains cases, like the modified version of The rich Catholic case, which Saul does not explain. It is more parsimonious than Stokke’s account and rules out Borg’s generalized propositions, which do not work as fallback contents, especially from the Stalnakerian perspective that Stokke endorses, but are nonetheless invoked by Stokke for explaining some cases of not lying.

Finally, conventions-based utterance contents meet desideratum (iv), at least with regard to the cases that Stokke employs for arguing in favour of this desideratum. With respect to this desideratum, too, conventions-based utterance contents have some theoretical advantages over Saul’s and Stokke’s notion of what is said. The advantage over Saul’s account is straightforward because Saul does not explain sensitivity to QUDs. Stokke’s account suffers from overgeneration as Van Elswyk pointed out. The advantage of the conventions-based view is that it provides an explanation of Stokke’s cases and a solution to Van Elswyk’s problem.

5 Conclusions

The ongoing debate over the distinction between lying and misleading has important reflections in philosophy of language. Philosophers capture the distinction between lying and misleading in terms of commitment to disbelieved information. Liars are committed to the disbelieved information they communicate, whereas misleaders are not. Says-based definitions of lying explain commitment in terms of the content that is literally said. This approach to lying raises the question about what notion of what is said is best suited to that theoretical role. I argued that a conventions-based notion of utterance content whose constitutive intentions on the part of the speaker are expressive and not communicative and audience-oriented fits well the theoretical desiderata and has important advantages over other notions of what is said that have been so far presented.

One main idea underpins my arguments. Philosophers who defend says-based definitions agree that when misleading speakers defend themselves from the charge of lying, they disclose that they were not cooperative as they were violating one or more maxims of the principle of cooperation by, for example, communicating only part of the information at their disposal or by not addressing directly the question under discussion. In doing so, misleading speakers make manifest their deceitful intentions. Thus, their defence is not against the accusation of being deceitful, but against the accusation of being deceitful by lying. The commitment that ties liars to disbelieved information is not accounted for in terms of what the speaker is rationally understood to communicate or what the speaker expects the audience will rationally understand. This is to say that the commitment is not accounted for in terms that involve Gricean communicative intentions. Indeed, clever misleaders plan their deceitful strategy relying precisely on what the audience will rationally take them to communicate, although they deny responsibility for the disbelieved information. The responsibility in question, then, is stricter than the conversational responsibility for the communicated information, i.e. the Gricean total signification of the utterance that includes what is implicated. The liar is responsible only for the content that is conveyed in a strictly linguistic sense.

Conventions of language come into play for explaining this form of strictly linguistic commitment. The liar is committed to that part of the total signification of the utterance that is constituted by the expressive intention of using words with their conventional meaning. It is the part of the total signification of the utterance that speakers cannot avoid expressing when they use words in accordance with their conventional meaning. The role of conventions of language is eminent in cases of underdetermination. No one doubts that speakers can lie by using sentences that contain expressions like ‘ready’, ‘enough’, ‘late’, etc. Pragmatic accounts of underdetermination explain how speakers express fully propositional contents (explicatures, imlicitures, pragmatic modulations). Such accounts, however, do not explain the commitment of speakers to those propositional contents. As said, the commitment is not to be explained in terms of communicative intentions and conversational constraints, but it can be explained in terms of conventions of language. Conventions of language do not determine worldly meanings for incomplete expressions, but rules of use on the model of indexicals and demonstratives.Footnote 27 This is the view that Korta, Perry, and Devitt maintain in overt polemic with the way the debate over the semantics/pragmatics divide has been so far conducted between semantic minimalism on the one side and linguistic pragmatism on the other. In this paper I argued that the debate over the distinction between lying and misleading highlights some important theoretical advantages of the view that Korta, Perry, and Devitt maintain in philosophy of language.