1 Introduction

David Plunkett and Timothy Sundell in a series of influential papers argue that many of our disagreements are—contrary to what we think—metalinguistic. They divide disagreements into canonical and non-canonical instances, where canonical disagreements are those that concern the literal content expressed by the speakers. They claim that a disagreement doesn’t have to be canonical to be genuine and worth having. Many non-canonical disputes are in fact genuine disagreements, which reflect a conflict in the attitudes of the speakers (e.g., in their beliefs, desires, intentions, etc.),Footnote 1 even though that conflict doesn’t concern the literal content of the speakers’ utterances.Footnote 2 The observation that in genuine disagreements the literal contents of the utterances might be compatible allows Plunkett and Sundell to argue—contrary to what is usually assumed—that speakers who say “a is P” and “No, a is not P” can genuinely disagree even if they don’t mean the same by “P.”Footnote 3 In such a case, the literal contents of their utterances are not inconsistent, but there might be other related attitudes that are also expressed and which are incompatible. One type of such non-canonical disagreement is metalinguistic negotiation. Plunkett and Sundell argue that in exchanges like:

  1. (1)

    Secretariat is an athlete. No, Secretariat is not an athlete;

  2. (2)

    Waterboarding is torture. No, waterboarding is not torture;

  3. (3)

    This chili is spicy. No, this chili is not spicy;

  4. (4)

    Philip is tall. No, Philip is not tall,

the speakers have different meanings in mind when they use the words “athlete,” “torture,” “spicy,” and “tall,” and therefore their disagreements are not canonical.Footnote 4 They are, however, involved in genuine disagreements, namely in metalinguistic disagreements over how the relevant words should be used. Thus, when you hear that two speakers, Oscar and Callie, utter in turn “That chili is spicy” and “No, that chili is not spicy,” it might seem to you that they are disputing whether the chili in question is spicy or not, but in fact they are negotiating the standard for spiciness adequate to their situation. Plunkett and Sundell claim that since Oscar and Callie have already tasted the chili, “they already agree on what the chili actually tastes like” (2013, p. 15) and are negotiating the appropriate use of the predicate “spicy.” They disagree because Oscar accepts the content “we should use ‘spicy’ in such a way that it applies to the chili,” whereas Callie accepts the incompatible content “we should not use ‘spicy’ in such a way that it applies to the chili” (2013, p. 15). Plunkett and Sundell want to remain neutral on how precisely those contents are expressed, but say that via their assertions the speakers “also pragmatically advocate for the parameter settings by virtue of which those propositions are asserted” (2013, p. 15).Footnote 5

Thus, metalinguistic negotiation is negotiation in which the speakers using an expression metalinguistically to “communicate information about the appropriate usage of that very expression in context” (2013, p. 3) advocate for competing views of how the term in question should be used. Such negotiations are part of conceptual engineering or conceptual ethics. They are not merely verbal disputes over appropriate use (such disputes are usually not worth having) but reflect hidden disagreement about a normative issue, such as whether animals deserve the same kind of praise as humans (as might be the case in (1)) or whether waterboarding is morally wrong (as might be the case in (2)).

Plunkett and Sundell claim that metalinguistic negotiations are often implicit and admit that “they may not at first appear—either to the speakers themselves or to the theorist—to reflect disagreements about conceptual choice” (2013, p. 3). Nevertheless, they argue that such negotiations are “common” (2013 p. 15), “pervasive” (2013, p. 7), or even “ubiquitous” (2013, p. 4).

In this paper, I’ll argue that metalinguistic negotiations are much less frequent than Plunkett and Sundell claim and in fact happen very rarely in everyday life. Such negotiations require a particular context, which is anything but common in ordinary conversations. In particular—contrary to what Plunkett and Sundell suggest—typical disagreements involving relative gradable adjectives like “tall,” “rich,” and “spicy” should not be interpreted as metalinguistic but rather as object-level disputes concerning the issue of whether a given man is tall or rich and whether a given dish is spicy. Moreover, as Plunkett and Sundell acknowledge, metalinguistic interpretation involves i.a. two controversial claims: the claim that speakers don’t know what they say and the claim that they directly communicate metalinguistic contents regarding how the relevant expressions should be used. These two claims generate two challenges that the metalinguistic-negotiation view should meet. Firstly, it should clarify why the speakers are oblivious to what they are saying and communicating, and secondly, it should explain the mechanism that transforms what seems like a typical object-language disagreement into a metalinguistic dispute. Regarding Plunkett and Sundell’s answer to the first challenge, I’ll argue that the theoretical cost of postulating massive semantic and pragmatic blindness in otherwise competent speakers is too high.Footnote 6 Regarding what they say in relation to the second challenge, I’ll claim that metalinguistic contents can only be conveyed when speakers uttering apparently contradicting claims know that they are using terms with different meanings. If this is so, then metalinguistic negotiations are certainly not “ubiquitous”.

In Sect. 2, I present some further examples of apparent metalinguistic negotiations involving relative gradable adjectives. In Sect. 3, I describe the error that speakers are committing according to Plunkett and Sundell’s view and their explanation of why that error is not serious (3.1). In Sect. 3.2, I argue that this explanation is not convincing and that the error postulated is much graver than they admit. In Sect. 4, I claim that in typical disagreements involving relative gradable adjectives it is dubious whether any metalinguistic contents are conveyed and note that no explanation is provided as per how they could be communicated (4.1). In particular, I demonstrate that they cannot be conversational implicatures (4.2). I focus on disagreements involving relative gradable adjectives, but my conclusions apply to many other disagreements (e.g., some of those involving “torture” or “athlete”), since I argue that metalinguistic negotiation can only occur in a specific context. In Sect. 5, I revise the three criteria for recognizing metalinguistic negotiations recently offered by Andrés Soria-Ruiz (2021). I argue that his second criterion is not helpful for identifying metalinguistic negotiations, but the other two are appropriate and in fact can be used to demonstrate that typical disagreements involving relative gradable adjectives are not metalinguistic negotiations.

2 Metalinguistic negotiation

Traditionally it was assumed that if one speaker says “a is P” and another says “No, a is not P,” and they mean different things by “P,” they are not really disagreeing but rather talking past each other. Against this, Plunkett and Sundell argue that in such situations the disagreement might be genuine, and moreover they claim that disagreements of this sort are omni-present.

Probably the best-known example of metalinguistic use of an apparently object-language sentence “a is P” comes from Chris Barker. In an often-cited passage, Barker discusses a metalinguistic use of

  1. (5)

    Feynman is tall.

If (5) is used as an answer to the question “What counts as tall in this country?” in a situation in which Feynman’s height is well-known, what the speaker does is not “provide any new information about the world” but rather give “guidance concerning what the prevailing relevant standard for tallness happens to be in our community” (2002, p. 40).Footnote 7 For Barker, “negotiating standards is a normal, typically automatic, part of ordinary discourse” (2013, p. 241).

If someone challenged (5) and said

  1. (6)

    Feynman is not tall,

the resulting disagreement would be metalinguistic, but it wouldn’t be an example of metalinguistic negotiations in Plunkett and Sundell’s sense, for it would be descriptive rather than normative (2013, p. 14). The speakers are debating what the appropriate standard is and not what it should be. However, it is easy to imagine examples of more relevant metalinguistic negotiations. Plunkett describes a situation in which Ann and Dan are opening up a mini-golf course together. Ann wants the course to cater to expert golfers, whereas Dan wants it to cater to beginner golfers. They have to determine which holes are hard. Dan says “That hole that we just played together was hard,” and Ann contradicts him by saying “No way. That wasn’t hard” (Plunkett 2015, p. 839). If we assume that they are both equally good at golf and played the hole equally well, it seems that a plausible interpretation of their exchange is that they are trying to determine which meaning of “hard” should be accepted for their purposes. Sundell describes a similar case in which Alphie and Betty are opening a new restaurant together and have to determine which knives should be considered sharp. They both try a particular knife. Alphie, who had previously worked at a delicatessen, says “This knife is sharp,” while Betty, who worked at a sushi-restaurant, says “No, it’s not sharp.” Sundell argues that in this situation they are not communicating information about the knife. Nor are they engaging in a descriptive dispute concerning the pre-established standard of “sharpness” (Sundell 2016, pp. 11–12). Instead, they are negotiating what the threshold for sharpness should be in their joint venture.

I admit that these two scenarios may be regarded as examples of metalinguistic negotiation. They are unique in that the implicit issue that the participants are trying to settle might indeed be interpreted as metalinguistic. However, Plunkett and Sundell argue that such metalinguistic negotiations happen very frequently in our lives and in fact “disagreements about what should count as ‘tall’ during our basketball draft, or ‘cold’ in our shared office, or ‘rich’ for our tax base” (2013, p. 15) are all metalinguistic negotiations. As we have seen, Oscar and Callie’s disagreement over chili is also supposed to be metalinguistic.

Already in their 2013 paper, Plunkett and Sundell admit that speakers might resist seeing their disputes as metalinguistic, and in a recent paper they add that the speakers “seem not to understand their own debates as being about language in any sense” (2021a, p. 143) and “might find [such analyses] highly counterintuitive” (2021a, p. 144).

Thus, we arrive at the two main challenges that Plunkett and Sundell have to meet. Firstly, they have to explain why speakers are usually not aware that they are involved in such negotiations and would stubbornly protest if asked whether they are, and secondly, they should provide a convincing reason for us to think that those debates really are metalinguistic, i.e., they should explain the mechanism behind such exchanges.

3 First challenge to metalinguistic negotiation: explaining speaker error

3.1 Speaker error and its justification

As far as the first challenge is concerned, Plunkett and Sundell postulate a global speaker error and argue that postulating such an error is less theoretically costly than it might seem. According to them, one general reason why speakers are reluctant to see their disagreements as metalinguistic is that they tend to perceive debates over language as insignificant, merely verbal. Since they don’t think that their disputes are insubstantial, they balk at attempts to describe their exchanges as metalinguistic. Plunkett and Sundell repeatedly stress, however, that metalinguistic negotiations are not insignificant, because they are closely connected to substantive disagreements. As they put it, “metalinguistic negotiations can be a way of arguing about object-level issues” (2021a, p. 148). When people debate whether Secretariat should be called “an athlete” or not, they might also disagree about “who gets certain kinds of fame, praise, or recognition” (2021a, p. 148). The former issue is not in itself substantial, but the latter is.

Nevertheless, the “immediate disagreement” is about words and concepts:

Metalinguistic negotiations express disagreements that, in the first instance, are about language and thought—about the resolution of context sensitivity, or the precisification of vague terms, or which concept to pair with some term with a range of candidate meanings, and so on. That means that to the extent that these disputes reflect disagreements about non-linguistic object-level matters they do so at best indirectly. (2021a, p. 150)

Thus, Plunkett and Sundell argue that speakers are generally wrong about the insignificance of disputes about language and thought. That’s probably the reason why the speakers also make two more specific mistakes: they are mistaken in believing that they disagree over the literal semantic content of their utterances and “they are mistaken in believing that their disagreement is not, in the first instance, about language and thought” (2021a, p. 151). In other words, the speakers think that their disagreement is canonical and directly concerns object-level issues, whereas in fact it is non-canonical and concerns those issues only indirectly. In particular, the speakers think that their utterances are contradictory, whereas in fact what they are saying is consistent. Plunkett and Sundell acknowledge that they should provide an explanation of why speakers commit these mistakes and proceed to deliver one.

First of all, they claim that the first mistake comes down to not distinguishing semantic contents from pragmatic ones.Footnote 8 And they argue that that distinction is not uncontroversial and can be confusing even for specialists. Their observation is reminiscent of that made by Kent Bach in “You don’t say,” who suggests that relying on the intuitions of ordinary speakers regarding the semantic/pragmatic distinction is like

relying on the intuitions of unsophisticated moviegoers about the editing on a film. Although people’s cinematic experience is dramatically affected by cuts and camera angles, there is no reason to suppose that their intuitions are reliable about these effects or about how they are produced. (2001, pp. 26–27)

Roughly put, ordinary speakers are not properly qualified and do not have intuitions that would allow them to make subtle distinctions concerning pragmatics and semantics. Moreover, the distinction itself is notoriously hard to grasp and even linguists and philosophers of language disagree about whether certain phenomena (such as scalar implicatures) are semantic or pragmatic in nature.

As far as the second mistake regarding the subject matter of the disagreement is concerned, Plunkett and Sundell claim that since metalinguistic negotiation can be a way of (indirectly) communicating disagreements about object-level issues, their account of metalinguistic negotiation “is a theory that vindicates [the] ordinary speaker intuitions that they’re really disagreeing, and specifically that they’re really disagreeing about [such things as] public policy and waterboarding.” (2021a, p. 158).

Thus, Plunkett and Sundell claim that the error they postulate concerns speakers having false folk-linguistic beliefs about communication and the way it proceeds rather than speakers misapplying their own words, which would be more worrisome (2021a, p. 165). False beliefs about communication can be explained by the lack of relevant linguistic training, and since they don’t result in speakers using language incorrectly, the fact that speakers have them is not that worrying.Footnote 9

3.2 Semantic and pragmatic blindness

Although clever, I don’t think Plunkett and Sundell’s attempt at downplaying the gravity of the speaker error is ultimately successful. Were metalinguistic negotiations as frequent as Plunkett and Sundell think, that error would be massive and it would put into question our competence as speakers.

Ordinary speakers might not be experts as far as the semantic/pragmatic distinction is concerned, but it seems to me that in general they are not as bad as Bach as well as Plunkett and Sundell seem to postulate. In many cases, speakers are aware of the distinction between what they say and what they additionally communicate.Footnote 10 For instance, the distinction between lying and misleading, although not uncontroversial, does exist in natural language and arguably depends on the distinction between what is said and what is merely communicated (see, e.g., Saul 2013). Moreover, normally speakers—at least on reflection—are aware of both contents.Footnote 11 In metalinguistic negotiation the situation is different. Most speakers are mistaken about what they say (they think that their utterances are directly contradictory) and do not know that they are communicating metalinguistic content. Moreover, as Plunkett and Sundell admit, it is not only ordinary speakers who are so ignorant. They acknowledge that also lawyers, judges, policy makers, and philosophers don’t think of their disagreements as being about words and thoughts. In defence they express some sympathy with the view that “practitioners are bad theorists” (2021a, p. 151) and while one might be tempted to agree that the view is true of lawyers, judges, and policy makers, applying it to philosophers, philosophers of language in particular, would have disastrous consequences: it would undermine the whole raison d’etre of the philosophy of language. And it has to be noted that philosophers of language, who are both practitioners and theorists and arguably have the required knowledge, commit the same alleged errors as ordinary speakers.

Thus, I think that the error, postulated by Plunkett and Sundell, is serious. It consists in the fact that speakers do not know what they are talking about (they think they are talking about the world, whereas in fact they are talking about the language), and they think that their utterances are contradictory, whereas in fact they are compatible. The fact that even trained philosophers make that error makes it even more problematic. The lack of clarity surrounding the semantic/pragmatic distinction is not sufficient as an explanation, because it’s not just a matter of confusing contents of different types. Ordinary and non-ordinary speakers alike are not even aware that metalinguistic content is being communicated in their exchanges. Plunkett and Sundell remark that “as long as we succeed in communicating what we intended, the mode of communication is usually not important” (2021a, p. 157), but the problem here is that in many cases the speakers do not intend to communicate the metalinguistic contents that the authors see them as conveying.

4 Second challenge to metalinguistic negotiation: Explaining the mechanism

4.1 Dubious metalinguistic contents

The second challenge consists in explaining why disagreements that look like typical object-language disputes are in fact metalinguistic negotiations.Footnote 12 I’ve already described how Plunkett and Sundell handle this challenge. Here I’ll endeavor to argue that their explanation is not sufficient. To be clear, I don’t want to deny that speakers do sometimes engage in metalinguistic negotiation. That would be foolish (and would also involve postulating speaker error). What I want to challenge, however, is the ubiquity of such negotiations. Metalinguistic negotiation requires a particular context, which doesn’t occur that often in everyday conversations. In particular, I want to argue that typical, everyday discussions involving relative gradable adjectives are not metalinguistic.

As Cappelen points out, Barker’s paradigmatic example of a metalinguistic use of “Feynman is tall” is unusual. The question “What counts as tall in your country?” is not typical and there is no reason to model other—more typical—cases on this one (Cappelen 2018, p. 178). In this example it is indeed likely that the answer should be interpreted metalinguistically, since the question is clearly metalinguistic.Footnote 13 Similarly, the examples involving “hard” and “sharp” described above are set up in such a way that the metalinguistic interpretation is the most likely one. In these examples the speakers try to establish what should count as sharp and hard for their respective purposes.Footnote 14 Example (1) above is also arguably most naturally interpreted metalinguistically as a disagreement over the meaning of “athlete” and example (2) might be seen this way too, but examples (3)–(4) are not like this. The relevant disagreements seem to be object-language disagreements concerning spiciness and tallness rather than the meaning of the words “spicy” and “tall.”Footnote 15 Moreover, it seems to me that it is examples (3) and (4) that are typical and ubiquitous. From my perspective, the intuition that in such situations the speakers are directly expressing their object-level beliefs is hard to deny. As we have seen, in their explanation Plunkett and Sundell appeal to the subtlety of the distinction between semantics and pragmatics. What their explanation is lacking, however, is a clear explication of how it happens that the pragmatic content expressed is what it is. Plunkett and Sundell don’t want to commit to a particular pragmatic mechanism as responsible, but nothing they say makes it any more plausible.

The common, pre-Plunkett-and-Sundell, picture of a disagreement like (3) is this: The speakers think that they disagree over the literal content they express. One says “That chili is spicy,” the other says “No, that chili is not spicy.” Both think that they disagree over whether the chili in question is spicy, and both think that the second speaker contradicts the first. Do the speakers communicate anything else? Well, that depends on the context. If they are in a context in which a decision must be made whether to add more spice to the chili, then presumably the first speaker will communicate that no more spice should be added, while the second will communicate that more spice is needed. If the context is one in which a decision has to be made whether to give the chili to a toddler, the first speaker will imply that it’s not a good idea, and so on. If they are just eating the chili, then they might not be communicating anything more than how it tastes to them. It might additionally be the case that those disagreements are motivated by deeper disagreements lurking in the background. Suppose we assume that the first speaker—but not the second—thinks that spices are really unhealthy and nobody should consume them. In such a case, the first speaker’s utterance might be motivated by his conviction that people shouldn’t eat food to which spices have been added. Although those speakers do disagree on that, we would not say that that is what the disagreement is about. It’s just not true that when we say something we express all our background beliefs that led us to say what we said. Plunkett notices the difference between the reasons why speakers enter into a dispute and the immediate topic of disagreement (2015, 844). He remarks that in the Secretariat case the immediate topic is which concept of “athlete” should be used. This is a plausible description of (1), but that does not seem to be a correct account of what is going on in (3). In (3)—as well as in countless cases like it—the immediate topic of disagreement is whether the object in question has a given property or not. In such discussions the immediate topic is not the threshold of the relevant predicate.

Plunkett and Sundell see it differently. As we’ve seen, they claim that although the speakers think that they disagree over the literal content they express, they in fact do no such thing. The literal content of their respective utterances is “The chili is spicy1” and “No, the chili is not spicy2” and they are not inconsistent. Their primary disagreement is over pragmatically conveyed content having to do with how predicates like “spicy” should be used. The speakers may or may not also be in other (background) disagreements with their interlocutors. In some cases, those background disagreements are also pragmatically expressed, and in such cases the original disagreements are substantial.

Thus, on the traditional picture, the situation looks like this:

Literal content: The chili is spicy // No, the chili is not spicy.

Conversationally implicated contentFootnote 16: We shouldn't/should add more cayenne pepper to the chili.

Whereas on Plunkett and Sundell’s picture:

Literal content: The chili is spicy1 // No, the chili is not spicy2.

Pragmatically communicated content: We should/shouldn’t use “spicy” in such a way that it applies to that chili plus additional contents over which the speakers disagree and which motivate their explicit disagreement.

4.2 Metalinguistic negotiation as conversational implicature

Plunkett and Sundell don’t commit to a particular pragmatic mechanism, but they don’t rule out the possibility that metalinguistic contents are conversational implicatures (see, e.g., 2013; 2021a, p. 153). It seems clear, however, that on their view the pragmatically communicated content cannot be due to conversational implicature. As we know, Grice characterizes conversational implicature in the following way:

A man who, by (in, when) saying (or making as if to say) that p has implicated that q, may be said to have conversationally implicated that q, provided that (1) he is to be presumed to be observing the conversational maxims, or at least the Cooperative Principle; (2) the supposition that he is aware that, or thinks that, q is required in order to make his saying or making as if to say p (or doing so in those terms) consistent with this presumption; and (3) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (2) is required. (1989, pp. 30–31)

For Plunkett and Sundell’s pragmatic contents, neither (2) nor (3) are satisfied. Since the speakers are not even aware that they communicate contents like “we should use ‘spicy’ in such a way that it applies to the chili", they cannot have the thoughts that Grice would want them to have if they were to conversationally implicate those contents.Footnote 17 Also, those pragmatically communicated contents are not cancellable.Footnote 18 It doesn’t make sense to say either of the following:

#The chili is spicy, but we should not use “spicy” in such a way that it applies to the chili.

#The chili is not spicy, but we should use “spicy” in such a way that it applies to the chili.

A detailed account on which metalinguistic negotiations crucially involve conversational implicature has recently been suggested by Poppy Mankowitz.Footnote 19 Mankowitz argues that such negotiations involve conversationally implicating metalinguistic propositions about the aptness of using certain expressions (2021, p. 5611). Her view, however, also cannot account for the chili example. She describes pragmatic reasoning which involves i.a. the following steps:

  1. (i)

    “an assessor doubts that the (real or imaginary) speaker is proposing to add the non-metalinguistic proposition expressed by her utterance to the common ground” (2021, p. 5614);

  2. (ii)

    “an assessor becomes aware that the violation of Gricean maxims need not be attributed to the speaker if the speaker is understood to be using her utterance to convey a metalinguistic proposition” (2021, p. 5615); in this step “the assessor entertains the hypothesis that the speaker is uttering the sentence in order to convey a metalinguistic proposition concerning the aptness of certain expressions in the sentence”Footnote 20 (2021, p. 5616).

Mankowitz goes on to illustrate how her model can be used to account, e.g., for the Secretariat case.Footnote 21 Nevertheless, it seems clear that this mechanism cannot be used to explain the chili case, where the assessor has no reason to doubt that the speaker is proposing to add the object-language proposition to the common ground, since there is no apparent violation of Gricean maxims. The assessor takes the speaker’s utterance at face value and does not attempt to reinterpret it in any way. She thinks that the utterance “This chili is spicy” is false, so will object to adding it to the common ground, but she has no reason to think that the speaker violates any of the maxims.Footnote 22 In particular, in a typical chili case the assessor does not “entertain the hypothesis that the speaker is uttering the sentence in order to convey a metalinguistic proposition” (2021, p. 5616).

Mankowitz notices that already at step (i) the assessor may be aware (and assume that the speaker is also aware) that the meanings they attach to the expression uttered are different (2021, p. 5615, fn. 15). Similarly, Belleri (2017) argues that the speakers immediately infer that they are purposefully using different senses of the relevant expressions. And Sundell in his 2015 paper suggests that it is characteristic of metalinguistic negotiations that the speakers mean different things by the terms used in the disagreement (2015, p. 847). He claims that good evidence for this is that speakers “have dispositions to systematically apply the same term in divergent ways in the same (non-defective) conditions” (2015, p. 847).

While this assumption is plausible in the Secretariat case and possible in cases of certain ontological disputes,Footnote 23 it is not likely in the chili case, unless the chili under discussion is a clear case of a spicy dish. If the chili in question were definitely spicy, then indeed Oscar might have wondered whether Callie is using “spicy” in the same way as he is. However, if the chili is borderline spicy, then Oscar will take Callie’s utterance at face value. Usually, we do not take disagreements over borderline cases of vague expressions as evidence of divergent meanings. The meaning of such expressions is such that it permits differing opinions. Speakers who disagree over the chili might not have dispositions to systematically apply “spicy” in divergent ways. If the chili is borderline spicy, the speakers might agree on most other cases and yet disagree about whether that particular chili is spicy. Unlike in the Secretariat case and unlike what Plunkett suggests, the speakers need not have different preferred concepts for use in the given context (2015, p. 841). As Plunkett and Sundell notice, in disagreements over relative gradable adjectives like “spicy,” the speakers may agree on the character and disagree only over the correct extension of the given adjective. And their divergent assessment of the borderline cases of “spicy” is not enough to send the speakers in search of metalinguistic reinterpretation.

Thus, although Plunkett and Sundell appeal to Grice, it seems that the Gricean picture cannot be of help here,Footnote 24 and we are left in the dark as to how those additional contents are communicated.

I’m focusing here on relative gradable adjectives, but it is worth noticing that the speakers’ awareness that they are systematically using a given term differently is crucial for metalinguistic negotiation no matter what the subject. It is thus clear that such negotiations are not as frequent as Plunkett and Sundell try to suggest.Footnote 25 In order to conclude that disagreement about torture or athletes is metalinguistic, the context of that disagreement must be of a specific type. If Amir and Betty are so far in agreement that various interrogation techniques are in fact torture, and disagree now about waterboarding, one cannot straightaway conclude that they mean different things by “torture.” Only if they know that they systematically differ in how they apply that term (and they know that they both know it), can they assume that their meanings are divergent. In such cases, it is indeed likely that the (implicit) question that they are trying to settle is metalinguistic.

As I mentioned earlier, Plunkett and Sundell themselves express some doubts about whether Grice’s framework is the best to capture how metalinguistic negotiations happen and stress several times that they want to remain neutral on the nature of the pragmatic mechanism supporting their claim. Thus, one might object that pointing out that that mechanism cannot be that of conversational implicature does not affect their view.Footnote 26 However, the dialectic of the situation is this: Plunkett and Sundell propose a metalinguistic interpretation of disagreements that until now have been regarded as object-level disputes and that goes against competent speakers’ convictions. Unless they justify that interpretation and explain how those disagreements happen, their proposal is not complete and is hardly a real contender for being the correct account of what goes on in such disputes.Footnote 27

5 Criteria for recognizing metalinguistic negotiation

In a recent paper, Soria-Ruiz proposes three criteria that together can serve as an indicator that the debate is metalinguistic. He doesn’t investigate the mechanism by which metalinguistic negotiations are supposed to happen, but concentrates on the features which can be of help in identifying metalinguistic negotiation.Footnote 28 For him a debate is metalinguistic negotiation iff it satisfies M, where M is a set of 3 linguistic markers:

  1. (1)

    Felicitous consider-embeddings: when a given term is used metalinguistically, we may felicitously embed the utterance under “consider.” For instance, the fact that “I consider Leo to be vegetarian” is felicitous suggests that “vegetarian” is used here metalinguistically. “Consider” triggers the supposition that the use of the term is controversial (2021: 8).

  2. (2)

    “The possibility of replying to a statement using a metalinguistic comparative in a literal, non-ironic/humorous sense” (2021: 11). The idea here is that if the speaker replies to an utterance of “Waterboarding is torture” with the metalinguistic comparative “Waterboarding is more interrogation technique than torture” and his reply can be interpreted literally, that is an indication that the debate is metalinguistic (unlike “The sushi is more disgusting than delicious” uttered in reply to “The sushi is delicious,” which can only be interpreted as mocking or ironic (2021: 12)).

  3. (3)

    The most relevant QUD is metalinguistic (“What should count as F?” rather than “What is x like?”). If the most relevant QUD is not explicit, it can be revealed by the way in which the conversation develops. Take the following disagreement:

Amir: Feynman is tall.

Berta: No, Feynman is not tall.

If the QUD of the context in which it occurs is “What counts as tall around here?” then Amir’s reply:

Amir: Well, whatever. Friedman is tall

is felicitous, whereas the reply

Amir: ? Well, whatever. He wears glasses 

is not felicitous (Soria-Ruiz 2021, p. 14). The situation would be reversed if the QUD were “What does Feynman look like?”

If we apply the criteria proposed by Soria-Ruiz to debates involving relative gradable adjectives, we’ll see that such debates are not usually metalinguistic.

First of all, notice that the second condition is not helpful in deciding whether the chili example is metalinguistic or not, since it only reflects how the second speaker, the one uttering the metalinguistic comparative, interprets the exchange. It doesn’t tell us anything about how the first speaker meant her utterance. She might have uttered an object-language claim, that was wrongly interpreted metalinguistically by the second speaker. This is evidenced by the two dialogs cited by Ruiz:

Amir: The sushi is delicious! Don’t you think?

Berta: The sushi is more beautiful than delicious.

Amir: The sushi is delicious! Don’t you think?

Berta: The sushi is more disgusting than delicious (2021, p. 12).

According to Soria-Ruiz, in the second dialogue Berta’s remark can only be interpreted ironically, which suggests that it’s not metalinguistic, while in the first dialogue her remark is literal and therefore metalinguistic. It has to be noted, however, that Amir’s claim is the same in both dialogues concerning the sushi and might easily be interpreted non-metalinguistically. Thus, while I agree that Berta’s second reply indicates that she wants to turn the debate into a metalinguistic one, it might well be that Berta is in fact changing the topic and replies metalinguistically to Amir’s non-metalinguistic remark. In such a case their original debate is not an example of metalinguistic negotiation. In the chili case, Callie could reply with “The chili is more hot than spicy” as well as with “The chili is more tasteless than spicy,” and while her first reply would be indicative of her metalinguistic intention, it wouldn’t tell us anything about how Oscar meant his utterance.

So, let’s focus on the other two conditions. Take again our chili example. “That chili is spicy” can felicitously embed under “consider,” but this criterion is too weak to be used on its ownFootnote 29 and has to be paired with the third one. The third criterion is crucial and expresses nicely what I’d been hinting at earlier. In metalinguistic negotiation the most relevant QUD is “What should count as F?”. However, debates revolving about such a metalinguistic QUD – no matter whether explicit or implicit – are hardly typical and certainly not ubiquitous. In particular, if the most relevant QUD in our scenario concerning the chili were metalinguistic, then the following continuation would be felicitous:

Oscar: The chili is spicy.

Callie: The chili is not spicy.

Oscar: ? Well, whatever. The tikka masala is spicy.

It is clear however that it is not felicitous. The reason for this is that Oscar and Callie are discussing the taste of the chili and the question they are answering is “What is the chili like?” rather than “What should count as spicy?”.

Thus, it appears that two of the three linguistic markers suggested by Soria-Ruiz can indeed be used to distinguish metalinguistic negotiation from other kinds of debate. Moreover, they strengthen our earlier contention that to qualify as an instance of metalinguistic negotiation, the debate must center around a metalinguistic question and provide diagnostics that allows one to determine whether this is the case. However, the result is that metalinguistic negotiations are relatively rare and definitely not ubiquitous.

6 Conclusion

I’ve tried to argue that metalinguistic negotiation is not as pervasive as Plunkett and Sundell seem to think. If it were so ubiquitous, speakers would be constantly mistaken about what they are directly talking about. Postulating such an error is too high a theoretical cost to pay. I’ve focused on the disagreement concerning “spicy” because in my opinion relative gradable adjectives are particularly ill suited to the analysis that Plunkett and Sundell propose. If we bracket the problems surrounding speaker error and the mechanism involved, the appeal to normative or evaluative issues that are the real subject of the disagreement seems quite plausible when words like “athlete” or “torture” are concerned. However, it is much less convincing that speakers who quarrel over whether the chili is spicy (or whether Philip is tall) have such substantial matters in mind. In typical situations, the speaker who says “The chili is/is not spicy” or “Philip is/is not tall” doesn’t do so because of some background “moral/social/political/legal views” (2021a, p. 149) that she has. Thus, it seems to me that it is a mistake to lump together examples (1)–(2) and (3)–(4) and analyze them in the same way.

Moreover, I’ve argued that the speaker error that Plunkett and Sundell have to postulate is grave, since they in effect accuse the speakers of being incredibly incompetent both semantically and pragmatically. Additionally, a pragmatic Gricean explanation of how the metalinguistic contents are conveyed cannot get off the ground in contexts in which speakers are not aware that they are using words with different meanings or answering a metalinguistic question. And we are not told how otherwise such contents could be conveyed.

In general, it appears that people engage in metalinguistic negotiations when they try to answer the explicit or implicit metalinguistic question of what should count as F in a given context. If their most relevant QUD is normative and metalinguistic, then indeed it’s likely that the debate constitutes metalinguistic negotiation. Oftentimes in such cases the speakers are aware that they are using terms with different meanings, hence the real content of their utterances must be different form the literal one (otherwise they wouldn’t be negating each other’s utterances). However, in typical cases the most relevant QUD is not metalinguistic, as witnessed by the possible ways in which the debate may felicitously continue (see Sect. 5). It thus appears that typical exchanges involving relative gradable adjectives, like “The chili is spicy”/ “The chili is not spicy” or “Philip is tall”/ “Philip is not tall” are directly about object-level issues.Footnote 30 If Philip is borderline tall, then the speakers are free to either classify him as tall or as not tall, provided other interlocutors agree and it is consistent with previous decisions made in the conversation (see Shapiro 2006). Making a pronouncement on someone’s tallness can lead to a precisification of the threshold of “tall.” A decision to regard a given borderline case as having or not having the property in question is ipso facto a decision about the boundary of the extension of the predicate, but – contrary to Plunkett and Sundell – this is not an “immediate topic” of the conversation.