Deceiving without answering

Abstract

Lying is standardly distinguished from misleading according to how a disbelieved proposition is conveyed. To lie, a speaker uses a sentence to say a proposition she does not believe. A speaker merely misleads by using a sentence to somehow convey but not say a disbelieved proposition. Front-and-center to the lying/misleading distinction is a conception of what-is-said by a sentence in a context. Stokke (Philos Rev 125(1):83–134, 2016, Lying and insincerity, Oxford University Press, Oxford, 2018) has recently argued that the standard account of lying/misleading is explanatorily inadequate unless paired with a theory where what-is-said by a sentence is determined by the question under discussion or qud. I present two objections to his theory, and conclude that no extant theory of what-is-said enables the standard account of the lying/misleading distinction to be explanatorily adequate.

1 The lying/misleading distinction and what-is-said

Language equips us to deceive others in at least two ways. Through the use of a sentence in a context, we can lie to others or merely mislead them. What lying and misleading share in common is that they involve the use of a sentence in a context to communicate a proposition that the speaker disbelieves. How these varieties of deception differ is our present topic.

The standard explanation of the distinction between lying and misleading bases their differences in how a disbelieved proposition is communicated by a speaker (Adler 1997; Saul 2012; Stokke 2016). To lie, a speaker uses a sentence in a context to say a proposition she does not believe. A speaker misleads by using a sentence to somehow convey but not say a disbelieved proposition. A proposition may be conveyed but not said because it is a conversational implicature, a presupposition, or another variety of not-at-issue content. Let’s abbreviate the standard account of the lying/misleading distinction as ${\textsc {sa}}_{\textsc {l/m}}$.

Front-and-center to ${\textsc {sa}}_{\textsc {l/m}}$ is an account of what-is-said by a speaker’s use of sentence in a context. What is said is the primary content the sentence’s use contributes to a conversation. An account of what-is-said is tightly connected with a conception of semantic content. The semantic content of a sentence is what is determined in a context by combining the meanings of the sentence’s constituent expressions according to grammatically associated composition rules. Many identify what-is-said by a sentence in a context with its semantic content in that context. Others deny the identification.

A major point of controversy about what-is-said concerns the contribution made by context. Features of context such as speaker intentions, hearer beliefs, discourse structure, prominent objects or properties, practical stakes, and more have all been argued to exert influence on what-is-said. Saul (2012, 57) argues that ${\textsc {sa}}_{\textsc {l/m}}$ requires a theory of what-is-said that abides by the following constraint on the role of context:

needed for truth evaluability (nte)

A putative contextual contribution to what is said is a part of what is said [by a sentence S] only if without this contextually supplied material, S would not have a truth-evaluable semantic content in a context.

nte limits the role of context to determining the truth-evaluable content of a sentence. Let’s use ${\textsc {wis}}_{\textsc {nte}}$ as shorthand for a theory of of what-is-said that is compatible with nte. Since a proposition is the truth-evaluable content had by a declarative sentence, ${\textsc {wis}}_{\textsc {nte}}$ theories agree that what-is-said is the initial proposition produced after context interacts with semantic content.¹ Let’s call that proposition the minimal proposition. With regards to Fig. 1, ${\textsc {wis}}_{\textsc {nte}}$ theories affirm only the role of context represented by the first solid arrow. They differ on how context ensures that a minimal proposition is produced from a declarative’s semantic content.

Fig. 1

nte (in)compatibility

In contrast, non-${\textsc {wis}}_{\textsc {nte}}$ maintain that what-is-said can ocassionally be non-identical with the minimal proposition produced after context contributes to semantic content.² So they affirm that context can play the roles represented by the solid and dotted arrows in Fig. 1.

Stokke (2016, 2018) disputes that nte is enough for ${\textsc {sa}}_{\textsc {l/m}}$. He presents a question–reply exchange in which the reply is a lie but ${\textsc {sa}}_{\textsc {l/m}}$ cannot identify it as a lie if paired with a ${\textsc {wis}}_{\textsc {nte}}$ theory. His take-away is that ${\textsc {sa}}_{\textsc {l/m}}$ is sensitive to the structure of the discourse. To accommodate such discourse-sensitivity, Stokke pairs ${\textsc {sa}}_{\textsc {l/m}}$ with a new theory of what-is-said that makes it sensitive to which questions are asked in a discourse. Accordingly, Stokke uses ${\textsc {sa}}_{\textsc {l/m}}$ to motivate a non-${\textsc {wis}}_{\textsc {nte}}$ theory where what-is-said need not be the minimal proposition produced after context contributes to a declarative’s semantic content.

This paper raises more trouble. I begin in Sect. 2 by detailing Stokke’s take on linguistic deception. In Sect. 3, I consider the question–reply exchange he presents to motivate that ${\textsc {sa}}_{\textsc {l/m}}$ requires a discourse-sensitive theory of what-is-said. Then I offer two question–reply exchanges that are mishandled by ${\textsc {sa}}_{\textsc {l/m}}$ when paired with Stokke’s own theory of what-is-said. In the first, a misleading statement is mistakenly predicted to be a lie (Sect. 4). In the second, a lie is mistakenly predicted to be sincere (Sect. 5). Strikingly, neither exchange is mishandled by a ${\textsc {wis}}_{\textsc {nte}}$ theory. In Sect. 6, I conclude with a discussion of what the failings of ${\textsc {wis}}_{\textsc {nte}}$ theories and Stokke’s non-${\textsc {wis}}_{\textsc {nte}}$ theory tell us about context’s role in determining what-is-said as required by ${\textsc {sa}}_{\textsc {l/m}}$.

2 Stokke on linguistic deception

Stokke’s perspective on the lying/misleading distinction has three parts: an account of lying, an account of discourse structure, and an account of what-is-said that anchors his account of lying to discourse structure. I consider each part in turn and start off with his account of lying.

Stokke (2013, 2016, 2018) views lies as disbelieved assertions. The way in which he understands the speech act of assertion is in the tradition of Stalnaker (1978, 2002). In this tradition, conversations are joint inquiries that take place against a backdrop of mutually believed or accepted information known as the common ground. As participants learn from each other, the common ground grows. So an assertion is a proposal to add a proposition to the common ground. That yields the following account of lying:

stalnakerian lying (sl)

A lies in uttering a sentence S if and only if (i) S says that p, (ii) By uttering S, A proposes to make it common ground that p, and (iii) A believes that not-p.

What sl makes transparent is that lying is an attempt to disrupt inquiry. When a speaker lies, she proposes to add a proposition to the common ground that she regards as unworthy of acceptance.

But what does it take for a sentence to say that p as required by condition (i)? Stokke notes that the semantic content of a declarative often determines a minimal proposition. Consider (1). Composing the expressions according to grammatically associated rules is sufficient to determine the proposition that Jerome cooked dinner.

• (1) Jerome cooked dinner.

But Stokke observes there are still cases where a semantic content can fail to determine a minimal proposition in a context. For example, a speaker may utter (2) while vaguely gesturing towards a nearby table brimming with freshly prepared food.

• (2) That looks tasty.

It is not clear what value to give to the demonstrative that in that context. Its semantic value is underspecified. The speaker could be talking about all of the food or a particular dish. As a result, a minimal proposition is not determined by the semantic content in a context.

When a unique proposition fails to be determined, Stokke maintains that the semantic content determines a range of candidate propositions that differ only in the value assigned to the context–sensitive expression.³ For example, the semantic content of (2) determines a cloud of propositions that corresponds to the sentences below.

(2a):

That tilapia looks tasty.

(2b):

That salad looks tasty.

(2c):

That loaf of bread looks tasty.

$\vdots$

The task of a theory of what-is-said, as Stokke sees it, is to uniformly explain what-is-said when a unique proposition is determined by a declarative’s semantic content and when a range is determined.

To offer such a theory, Stokke follows Roberts (1996/2012) in maintaining that discourses are organized as a series of question–reply exchanges. Such a view of discourse structure is a natural development of the Stalnakerian view that conversation is a joint inquiry into what the world is like. After all, what is the world like? is itself a question. Structuring discourse as a series of question–reply exchanges captures how conversational participants aim to answer the Big Question by incrementally answering smaller questions. Every state of a conversation contains a question under discussion or qud. The qud may be overt by being asked. Or, it may be covert.

The qud-based approach to discourse structure is made precise with a semantics for questions. On the standard semantics, the meaning of a question is a set of propositions (Hamblin 1973). Adopting the familiar assumption that propositions are sets of possible worlds, polar questions like (3) contain just two propositions in their denotation.

• (3) $\llbracket$Did Lisa bring wine?$\rrbracket$ = $\begin{Bmatrix} \{w \mid {\text {Lisa brought wine in } }{w}\},\\ \{w \mid{ \text {Lisa did not bring wine in }}{ w}\} \end{Bmatrix}$

• (4) $\llbracket$Who brought wine?$\rrbracket$ = $\begin{Bmatrix} \{w \mid {\text {Lisa brought wine in } }{w}\},\\ \{w \mid {\text {Megan brought wine in }} {w}\},\\ \{w \mid {\text {Will brought wine in }} {w}\},\\ \dots \end{Bmatrix}$

In contrast, constituent or wh-questions like (4) contain more than two propositions. Their propositions are the result of composing relevant individuals from the conversational context with the property $\lambda x$.{w$\mid$x brought wine in w}. Within such a semantics for questions, we can distinguish partial and complete answers. A partial answer to a question entails the truth or falsity of at least one proposition in its denotation. A complete answer to a question entails the truth or falsity of every proposition.

Stokke (2016, 2018) proposes that the qud uniformly determines what-is-said by a sentence in a context. The basic idea is that we look to the broader structure of the discourse in which a declarative occurs to figure out what it says. Since the discourse is organized as a question–reply exchange, the declarative always sits under a covert or overt qud. We then consider partial or complete answers to a question as our candidates for what-is-said. Whichever answer is the weakest that entails a minimal proposition determined by the semantic content of the declarative is what-is-said in the discourse. Where $\mu _{c }$(S) denotes a minimal proposition determined by a sentence S’s semantic content in a context c, Stokke (2018, 99) offers the following definition:

what-is-said (${\textsc {wis}}_{\textsc {qud}}$)

What is said by S in c relative to a qudq is the weakest proposition p such that (i) p is an answer (partial or complete) to q? and (ii) either p$\subseteq$$\mu _{c }$(S) or $\mu _{c }$(S) $\subseteq$p.⁴

To see ${\textsc {wis}}_{\textsc {qud}}$ in action, reconsider examples (1) and (2). For sentences like (1) where the semantic content in a context determines a minimal proposition, ${\textsc {wis}}_{\textsc {qud}}$ identifies that proposition with what-is-said. For example, assume that Who cooked dinner? is the covert qud for (1). Then what-is-said is just the proposition that Jerome cooked dinner because it is a partial answer to the qud and it is identical to $\mu _{c }$(1). When it comes to sentences like (2) where the semantic content in a context determines a range of propositions, the qud acts as a sieve to separate out the candidate propositions that are entailed by answers to the qud. Suppose that the salad contains glutenous croutons and that the loaf of bread is not gluten-free. Then if Does anything gluten-free look tasty? is the qud for (2), both (2b) and (2c) are filtered out because they are not entailed by answers. That makes the answer that the tilapia looks tasty what (2) says because it is what entails a proposition contributed by (2)’s content in the context.

We now have each component of Stokke’s account of lying/misleading. The standard story espoused by ${\textsc {sa}}_{\textsc {l/m}}$ is that a speaker lies by saying a proposition she disbelieves. She misleads when she conveys but does not say a disbelieved proposition. By understanding what-is-said as being sensitive to the qud with ${\textsc {wis}}_{\textsc {qud}}$, what qualifies as a lie is qud-sensitive as per condition (i) of sl above. A new twist on ${\textsc {sa}}_{\textsc {l/m}}$ is therefore produced. To lie is to commit to a disbelieved answer to the immediate qud in the discourse. A speaker can mislead without lying by not committing to a disbelieved answer while still conveying a disbelieved proposition in the discourse.

3 Lying about why

But why pair ${\textsc {sa}}_{\textsc {l/m}}$ with ${\textsc {wis}}_{\textsc {qud}}$ as opposed to a ${\textsc {wis}}_{\textsc {nte}}$ theory? Stokke (2016, 88) submits the following vignette followed by two question–reply exchanges as evidence.

deceptive william

At an office Christmas party, William’s ex-wife, Doris, got very drunk and ended up insulting her boss, Sean. Nevertheless, 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 cut-backs. But, despite this, Doris and Sean have remained friends. Sometime later, William is talking with Elizabeth, who is interested in hiring Doris. However, William is resentful of Doris and does not want Elizabeth to give her the job.

• (5) (a) elizabeth: Why did Doris lose her job?

  (b) william: Doris insulted Sean at a party.

• (6) (a) elizabeth: How is Doris’s relationship with Sean?

  (b) william: Doris insulted Sean at a party.

William utters the same declarative sentence in each reply. And yet, what-is-said with that sentence is not the same between (5b) and (6b). In (5b), William says that Doris lost her job because she insulted Sean at a party. He says as much even though the sentence he used is only Doris insulted Sean at a party with its associated semantic content. William says something different in (6b). The reply in (6b) says Doris insulted Sean at a party.

Whether William lies or merely misleads in the replies above depends on what each reply says. William disbelieves that Doris lost her job because she insulted Sean at a party, but he believes that Doris insulted Sean at a party. William therefore lies in (5b) because he says Doris lost her job because she insulted Sean at a party. However, he merely misleads in (6b). He believes that Doris insulted Sean at a party and that is all that (6b) says. Though (6b) conveys that Doris’s relationship with Sean is poor, that is not said.

From the difference between (5b) and (6b), Stokke (2016, 94) concludes that pairing ${\textsc {sa}}_{\textsc {l/m}}$ with an ${\textsc {wis}}_{\textsc {nte}}$ theory “cannot count the utterance [by William] as saying that Doris lost her job because she insulted Sean at a party, and hence Saul’s account cannot agree with the judgment that the utterance is a lie.” The reason why it cannot is that what William says with (5b) noticeably differs from the minimal proposition determined by its semantic content. The proposition determined is that Doris insulted Sean at a party. But what-is-said is that Doris lost her job because she insulted Sean at a party. In contrast, ${\textsc {wis}}_{\textsc {qud}}$ equips ${\textsc {sa}}_{\textsc {l/m}}$ to identify William’s reply as a lie. The weakest answer to (5a) entailing that Doris insulted Sean at a party is the answer Doris lost her job because she insulted Sean at a party. ${\textsc {wis}}_{\textsc {qud}}$ therefore identifies that stronger answer as what-is-said by (5b). Since (6b) is not entailed by the weakest answer to (6a), ${\textsc {wis}}_{\textsc {qud}}$ does change what it says from its minimal proposition.⁵

4 Not lying about why

In Sect. 3 we witnessed that question–reply exchanges featuring why-questions compel pairing ${\textsc {sa}}_{\textsc {l/m}}$ with ${\textsc {wis}}_{\textsc {qud}}$ as opposed to a ${\textsc {wis}}_{\textsc {nte}}$ theory. This section flips the script. Instead of motivating ${\textsc {wis}}_{\textsc {qud}}$, exchanges featuring why-question provide evidence that ${\textsc {sa}}_{\textsc {l/m}}$ requires a ${\textsc {wis}}_{\textsc {nte}}$ theory.

I offer the exchange below. It is otherwise identical to the exchange in (5), except that (7b) differs from (5b) in two ways. First, the sentence is fronted by the discourse marker well. Second, the verb ending -ed is absent and temporal adverb did is present.

• (7) (a) elizabeth: Why did Doris lose her job?

  (b) william: Well, Doris did insult Sean at a party.

Importantly, neither difference renders (7b)’s truth-conditions different from (5b)’s. Discourse markers like well do not contribute to truth-conditions: Well, Doris insulted Sean at a party and Doris insulted Sean at a party are true in all the same worlds. Similarly, the difference, for a verb ${\mathcal {V}}$, between ${\mathcal {V}}$ed and did${\mathcal {V}}$ is not truth-conditional either. They both indicate that event denoted by the verb insult happened in the past such that Doris insulted Sean at a party and Doris did insult Sean at a party are true in all the same worlds. So the minimal proposition had by (7b) in a context is the same as (5b).

The combined effect of these differences is that (7b) is not presented as an answer. The discourse marker conveys that the speaker’s utterance does not fully cohere with the discourse. As Schiffrin (1987, 102–103) puts it, “well$\ldots$anchors its user in an interaction when an upcoming contribution is not fully consonant with prior coherence options”.⁶ Since a reply coheres only when it is a partial or complete answer to the prior question in the strict sense from Sect. 2, well in (7b) signals that the reply does not settle the question. Instead, his reply merely offers evidence relevant to resolving why Doris lost her job. The presence of did in particular allows William to highlight that Doris’s insulting of Sean at the party is germane to Elizabeth’s prior question.

To give replies in exchanges like the above a name, let’s call them helpful non-answers.⁷ Confirmation that helpful non-answer are—as their name suggests—non-answers is that they are compatible with a denial of the answer they provide supporting evidence for.

• (8) (a) elizabeth: Why did Doris lose her job?

  (b) william: Well, Doris did insult Sean at a party. But Doris didn’t lose her job because she insulted Sean.

A discourse like (8b) illustrates. In response to Elizabeth’s question, William reports that Doris insulted Sean at a party, which provides evidence for the answer that Doris was fired because she insulted Sean, but denies the truth of that answer without either contradiction or absurdity.⁸

Helpful non-answers matter to ${\textsc {sa}}_{\textsc {l/m}}$. What-is-said by William with (7b) is that Doris insulted Sean at a party. Unlike (5b), he does not say that Doris lost her job because Doris insulted Sean at a party. Interpreted against the backdrop of deceptive william, William does not thereby lie in (7b). He merely misleads. He believes what-is-said. Support that William misleads as opposed to lies in (7b) is found in how William can defend himself when Elizabeth finds out the truth. For contrast, consider a continuation of the question–reply exchange in (5) where William does lie.

• (5) (a) elizabeth: Why did Doris lose her job?

  (b) william: Doris insulted Sean at a party.

  $\vdots$

  (c) elizabeth: You lied to me! You answered that Doris lost her job because Doris insulted Sean at a party.

  (d) william: #No I didn’t! I only told you that Doris insulted Sean at a party. And that’s true.

William’s response in (5d) to Elizabeth’s accusation is not credible. His reply in (5b) did commit him to the answer that Doris lost her job because Doris insulted Sean at a party. Though he may try to deny as much, his denial will not be acceptable in the discourse.

• (7) (a) elizabeth: Why did Doris lose her job?

  (b) william: Well, Doris did insult Sean at a party.

  $\vdots$

  (c) elizabeth: You lied to me! You answered that Doris lost her job because Doris insulted Sean at a party.

  (d) william: No I didn’t! I only told you that Doris insulted Sean at a party. And that’s true.

In contrast, William’s defense in (7d) is a credible response to Elizabeth’s accusation. Given the hesitancy conveyed by (7b) due to well and fall-rise intonation, William did not commit to the answer that Doris lost her job because Doris insulted Sean at a party. As a result, he can assure Elizabeth that he did not lie to her. He can retreat to the claim that he merely provided the evidence that Doris insulted Sean to help her arrive at an answer.

${\textsc {wis}}_{\textsc {qud}}$ runs aground on helpful non-answers like (7b).⁹ It has the same minimal proposition as (5b) for reasons discussed. ${\textsc {wis}}_{\textsc {qud}}$ therefore predicts for (7b) as it did for (5b) that William says that Doris lost her job because Doris insulted Sean at a party. As stipulated by deceptive william, William disbelieves that particular proposition. Consequently, ${\textsc {sa}}_{\textsc {l/m}}$, if paired with ${\textsc {wis}}_{\textsc {qud}}$, erroneously identifies William’s reply in (7b) as a lie.

In contrast, ${\textsc {sa}}_{\textsc {l/m}}$ does not mispredict that William lies as opposed to misleads if paired with a ${\textsc {wis}}_{\textsc {nte}}$ theory. On such a theory, what-is-said cannot outstrip the minimal proposition of (7b). Since the minimal proposition is that Doris insulted Sean at a party, William is not identified as lying. He merely misleads by encouraging Elizabeth to draw the conclusion for herself that Doris lost her job because she insulted Sean at a party. Where ${\textsc {wis}}_{\textsc {qud}}$ runs aground, ${\textsc {wis}}_{\textsc {nte}}$ theories sail.

5 Super liars

So far, we have considered only exchanges where an answer to the qud entails a declarative’s minimal proposition. But what happens if there are exchanges where no answer to a question entails a proposition determined by a reply’s semantic content? In such scenarios, Stokke (2016, 2018) maintains that the qud defaults to What is the world like?, the Big Question to which all questions are subquestions. As Stokke (2018, 104) puts it, “when there is no answer to the local qud that entails any candidate minimal proposition, the utterance is interpreted against the Big Question.” That default has no effect on what-is-said by a sentence whose semantic content determines a unique proposition because that proposition will be entailed by a partial answer to the Big Question.

In contrast, defaulting has a sizable effect on what-is-said by a declarative sentence whose semantic content determines a range of propositions. The Big Question cannot filter any out. Each is entailed by one or more partial answers to the Big Question. Consider a question–reply exchange like (9) with the background provided by Stokke (2016, 92). Larry’s reply to Norma is not entailed by an answer to Norma’s question.

unlogical larry

Larry is keen on making himself seem attractive to Norma. He knows she’s interested in logic—a subject he himself 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. But he hasn’t even opened it.

• (9) (a) norma: Do you know a lot about logic?

  (b) larry: My book is about logic.

The semantic of Larry’s reply also does not determine a unique proposition because there are multiple relations that the possessive my could be contributing as a context–sensitive expression.

• (10) (a) The book Larry checked out from the library is about logic

  (b) The book Larry bought for book club is about logic

  (c) The book Larry is writing is about logic

  $\vdots$

Each of these candidate propositions will be entailed by some partial answer to the Big Question. To get around this problem, Stokke (2018, 109) proposes that “the interpretation pulls back even further and takes as what is said a generalized proposition.” That generalized proposition is the proposition created by existentially binding the unsaturated component of the context–sensitive expression’s semantic value. For Larry’s reply to Norma, Stokke says that (11) is what-is-said by (9b).

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

Without knowing exactly which relation is contributed by the possessive in (9b), we are left with the generalized proposition.

It may be that underspecified declaratives can sometimes be used to say a generalized proposition. But maintaining that they can only be used to do so is untenable. A generalized proposition is unrestricted in what can make it true. Larry standing in any relation whatsoever to a book about logic will suffice for (11)’s truth. However, the range of candidate propositions determined by a semantic content is restricted in a context. One way it is restricted is by what propositions are already common ground. Conversational participants minimally assume that an assertion like Larry’s is informative. As a result, a generalized proposition would not be a candidate proposition because it is already mutually believed that Larry stands in some relation to a book about logic. For example, they may believe that he is heavier than, wider than, older than, more interesting than, more beautiful than, more useful than, and more valuable than such a book. Instead, participants would settle on a finite number of contextually relevant relations that would contribute to a proposition capable of growing the common ground by being informative. Consequently, Stokke’s pull-back proposal identifies what-is-said with a proposition that no competent conversational participant would take Larry as saying in (9b).

There is another problem. A common response to an underspecified sentence is a clarification request. That Larry does not say (11) with (9b) is further evidenced by how Larry can and cannot felicitously clarify what was said with his utterance. To see as much, consider the continuation of the initial exchange between Larry and Norma.

• (12) (a) norma: Do you know a lot about logic?

  (b) larry: My book is about logic.

  (c) norma: Your book? / How is the book yours?

  (d) larry: #I bear some relation to a book about logic.

Were Norma to seek clarification from Larry about what he meant by (12b), Larry cannot cite the generalized proposition. He is not competently using my if he does. In contrast, Larry can clarify that he is related to a logic book in one of the ways specified in (10). For example, My library book or My book club book are felicitous replies in (12d) to Norma’s query.

I conclude that the pull-back proposal is implausible. Implausibility compounds when the proposal is paired with ${\textsc {sa}}_{\textsc {l/m}}$. As discussed earlier, the range of propositions determined by a sentence’s semantic content is not without limit. It is restricted in a context by what is common ground to participants. To set up a problem for pairing the pull-back proposal with ${\textsc {sa}}_{\textsc {l/m}}$, suppose that (10a) through (10c) are the only available interpretations of my for (9b). Suppose further that Larry disbelieves each of these propositions: he does not believe that the book he checked out from the library is about logic, that the book he bought for book club is about logic, or that the book he is writing is about logic. Larry is what we should therefore call a super liar. There exists no situation in which Larry is not a liar that is identical to the one described except that the context–sensitivity of my is resolved. Larry, in other words, lies relative to every proposition he could say with (9b) as constrained by the context.

Do super liars lie? I submit that they do. An alternative view is that super liars are neither liars nor sincere speakers. It is indeterminate which they are until the context–sensitivity of the relevant expression is resolved. I do not find this alternative compelling for (9b) because there is no resolution of my where Larry is not a liar. It is determinate in the world of the context that he lied; it is only unsettled what proposition he lied with.¹⁰

Two additional reasons why a super liar lies can be given. Lying differs from misleading in that speakers can reasonably deny that they communicated the misleading proposition whereas they have no plausible deniability when lying. The continuation of Larry and Norma’s exchange below shows that Larry lacks plausible deniability.

• (13) (a) norma: Do you know a lot about logic?

  (b) larry: My book is about logic.

  (c) norma: Oh, you’ve checked out a book from the library about logic, bought one about logic for book club, or been writing a logic book?

  (d) larry: #No, those are not the ways the book is mine. / #No, that’s not what I meant.

When Norma seeks clarification by listing the various things Larry could have said in (13b), Larry cannot get off the hook. In denying that the logic book is his either because he wrote it, checked it out from the library, or bought it for book club, Larry denies all of the available candidates for what he said as constrained by the common ground. He has no ability to retreat to a candidate proposition that he believes to give him plausible deniability.¹¹

Lying also differs from misleading in that a lie can be met with explicit denial from a third-party. A merely misleading utterance cannot. If the third-party wants to challenge the speaker, they have to challenge the speaker’s motives instead of what the speaker said. The continuation in (14) shows that Larry’s utterance can be met with denial.

• (14) (a) norma: Do you know a lot about logic?

  (b) larry: My books is about logic.

  (c) garry: No, it isn’t! You have no book about logic. Your library book is about cats, the book you bought for book club is about cats, and the book you’re writing is about cats.

Garry’s denial of Larry’s utterance is in no way defective. He rejects the proposition that Larry’s book is about logic in any relevant way and elaborates that Larry is exclusively interested in cat books.

Here then is the problem that super liars pose for pairing the pull-back proposal with ${\textsc {sa}}_{\textsc {l/m}}$. In exchanges like (9) where the qud defaults to the Big Question such that a generalized proposition is said, Larry ceases to be a liar. The reason why is that Larry believes the generalized proposition. Without a doubt, Larry believes that he bears many relations to a book about logic (e.g. heavier than, wider than, older than, more interesting than). While Larry is a super liar, the pull-back proposal makes him an honest man.

What happens if the pull-back proposal is ditched? Then ${\textsc {wis}}_{\textsc {qud}}$ predicts that nothing whatsoever is said by (9b) and sentences like it. That prediction is made because ${\textsc {wis}}_{\textsc {qud}}$ requires what-is-said to be the weakest answer to the qud.¹² And there is not a weakest answer entailing a proposition determined by (9b) when there are multiple answers to the Big Question that are unordered by logical strength because they entail different propositions determined by (9b). Predicting that nothing is ever said is incorrect. It also mucks up the explanatoriness of ${\textsc {sa}}_{\textsc {l/m}}$. As before, suppose Larry is a super liar. Then sl is incapable of identifying him as a liar. As per condition (i), lying requires the speaker to say something and ${\textsc {wis}}_{\textsc {qud}}$ predicts that nothing is said by (9b).

So the pull-back proposal is needed to prevent ${\textsc {wis}}_{\textsc {qud}}$ from erroneously predicting that nothing is said by sentences whose content determines a range of propositions that are each entailed by an answer to the Big Question. But the pull-back proposal is implausible by limiting speakers to only saying a generalized proposition. Either way, Stokke’s account runs into trouble by having the qud uniformly determine what-is-said.

Once again, ${\textsc {wis}}_{\textsc {nte}}$ theories do not share in the problems. By not being limited to the qud to determine what-is-said, there is no need for a ${\textsc {wis}}_{\textsc {nte}}$ theory to maintain either that nothing is said or that only a generalized proposition is said. Instead, what-is-said can be settled on a case-by-case basis. When it comes to declaratives like (9b) where the speaker puts a cloud of propositions into play, the utterance can be correctly identified as a lie by adopting a proposal already argued for by Saul (2012, 64–65). In cloudy cases, speakers lie when they disbelieve every candidate proposition. Otherwise, they merely mislead because they can retreat to a believed proposition that ensures they have plausible deniability. That explanation, which takes a cue from supervaluationist approaches to vagueness, guarantees that super liars cannot be rehabilitated merely by a theory of what-is-said.

6 What-is-said

According to ${\textsc {sa}}_{\textsc {l/m}}$, lying differs is from misleading on the basis of how a disbelieved proposition is conveyed. To lie, a speaker uses a sentence to say a proposition she believes to be false. A speaker merely misleads by using a sentence to somehow convey but not say a disbelieved proposition. To yield any predictions, ${\textsc {sa}}_{\textsc {l/m}}$ requires a theory of what-is-said.

Saul (2012) casts a wide net. Any theory of what-is-said compatible with nte enables explanatory success for ${\textsc {sa}}_{\textsc {l/m}}$. However, Stokke (2016, 2018) argued that such theories are too restrictive in the role given to context. By limiting context to completing a declarative’s semantic content into a proposition, ${\textsc {sa}}_{\textsc {l/m}}$—paired with an a ${\textsc {wis}}_{\textsc {nte}}$ theory—underpredicts. It fails to identify lies in question–reply exchanges involving why-questions. Stokke’s alternative is ${\textsc {wis}}_{\textsc {qud}}$, a theory of what-is-said that relies on the qud. But the discourses in Sects. 4 and 5 show that the role he gives to context is problematic too. When paired with ${\textsc {wis}}_{\textsc {qud}}$, ${\textsc {sa}}_{\textsc {l/m}}$ does worse. It overpredicts by identifying misleading utterances as lies in question–reply exchanges where the reply is a helpful non-answer. It also fails to identify lies in question–reply exchanges where the reply contributes a range of propositions and none are entailed by an answer to the qud.

What do we learn from the preceding discussion of lying/misleading? The broader conclusion that Stokke initially drew from his counterexample to nte was that ${\textsc {sa}}_{\textsc {l/m}}$ requires a discourse-sensitive theory of what-is-said. I am sympathetic to this conclusion. But ${\textsc {wis}}_{\textsc {qud}}$ will not do. In inquiring into what theory can navigate the data without underpredicting or overpredicting, we therefore face a choice. Either pursue a theory that permits discourse structure to sometimes but not always determine what-is-said in a context or a theory that requires discourse structure to always determine what-is-said.

Neither option is without challenge. I close by sketching why. A potential virtue of a theory in which discourse structure only sometimes plays a role is that a qud-based account of discourse structure can still be used.¹³ Perhaps what-is-said is sometimes the weakest answer to the qud that entails a declarative’s minimal content [e.g. discourse (6)] and perhaps it is sometimes not [e.g. discourses (7) and (9)]. The corresponding vice of such a theory is that it loses predictive power. Without detailing the conditions under which discourse structure determines what-is-said, we would not know what-is-said for a wide range of cases. By depending upon a theory of what-is-said, ${\textsc {sa}}_{\textsc {l/m}}$ loses its predictive power too. So this option takes a step back. Stokke’s ${\textsc {wis}}_{\textsc {qud}}$ was admirable as a non-${\textsc {wis}}_{\textsc {nte}}$ theory because it was noticeably predictive.¹⁴ Weakening the role of discourse structure creates a significantly less predictive theory.

With the problems we have seen, taking the second option presumably requires a different conception of discourse structure. A prominent alternative to a qud-based approach is one that posits coherence relations along with rules specifying how segments of discourse can be linked. What a relations-based approach offers is a more fine-grained way of structuring discourse than a qud-based approach. To illustrate, consider the familiar pair of question–reply discourses a final time.

• (5) (a) elizabeth: Why did Doris lose her job?

  (b) william: Doris insulted Sean at a party.

• (7) (a) elizabeth: Why did Doris lose her job?

  (b) william: Well, Doris did insult Sean at a party.

On a qud-based approach, the only way a reply is related to a question is by standing in an entailment relation to an answer to that question. In particular, either a reply entails an answer or an answer entails a reply in conjunction with what is common ground. Since both (5b) and (7b) are entailed by the same answer to the preceding why-question, the qud-based approach has to give the same structure to (5) and (7). But a relations-based approach organizes question–reply exchanges according to the coherence relations that link them. So it could explain how (5) and (7) differ by maintaining that the question and reply in each are linked by a different coherence relation.

Nevertheless, fine-grainedness is a blessing and a curse. Though a relations-based approach has the resources to explain the difference between (5) and (7), it has to first specify what relations there are in a principled manner. That task is non-obvious and highly disputed. For example, Kehler (2002) offers four while Martin (1992) offers over 100 hundred different relations. Extant theories of what-is-said anchored to a relations-based approach to discourse structure cannot be used off the shelf either. They either are a ${\textsc {wis}}_{\textsc {nte}}$ theory, do not include coherence relations that structure question–reply exchanges, and/or specify relations that do not readily apply to why-questions (Asher and Lascarides 2003; Pagin 2014; Stojnic et al. 2017). So considerable work would need to be done to develop a discourse-sensitive theory of what-is-said that can jointly explain the data found in Stokke (2016) and the pages above.

I conclude that the proper theory of what-is-said, as required by the standard account of the difference between lying and misleading, remains elusive. The theory of Stokke (2016, 2018) and any theory compatible with nte explain some but not all of the data.