1 Introduction

Everyone has said things that shouldn’t have been said. One may have spoken falsely or, if not falsely, perhaps misleadingly. Other times, one may have been unkind, offensive, or just callous. For these or other reasons, retracting might be the thing to do. The reasons why one might retract are different in the two sorts of cases. It is not the same to speak falsely or misleadingly as it is to be offensive or unkind. One may retract in the former case because it is wrong to induce others in error. One may retract in the latter case because it is wrong to disrespect others. Sometimes, one may retract what was said even if one did nothing wrong. Either way, what one does is a speech act, albeit a fairly unspecific and second-order one. One may retract not just an assertion, but also a bet, a question, an order, a conjecture, a proposal, etc. Nonetheless, it is not the goal of this paper to offer a unified account of retraction.

The focus of this paper is on a specific way retraction has been regarded in the recent literature on truth relativism, in particular on certain cases where speakers are presumed to be under the obligation to take back what they said. This is a view that has been argued for by MacFarlane (2005b, 2014). Intuitions about retractions are supposed to support relativist semantic analyses of sentences of particular disputed kinds. The connection between the requirement to retract, on the one hand, and falsity, on the other, is such that the presumption that one ought to retract should show that what one said is false. Contextualists arguably make the wrong predictions about the judgments of truth and falsity, given the presumed obligation to take back what was said (or even the apparent naturalness of retracting), since contextualists predict that what was said is true. Epistemic modal sentences, sentences attributing knowledge, or personal taste sentences, among others, have been argued to be better accommodated within an assessment-relativist semantic theory. The truth of those sentences as used would depend not just on the context of use, but also on the context where their use is assessed.

Do we need the new assessment-relative truth-predicate? In other words, does assessment-relativism allow us to explain something that more standard contextualist theories cannot explain? This is a theoretical decision, and general considerations of simplicity and economy are relevant. Unless retracting assertions of sentences of the disputed kind is something that assessment-relativism can account for, but contextualism cannot, or that it is something that assessment-relativism can account for better than contextualism, then assessment-relativism is an unmotivated semantic complication. In this paper, I focus on retractions because of their centrality to the argument for relativism. I will show that there is no obligation to retract an assertion that was not only justified but true when it was made.

The objections against the obligatoriness of retractions rule out a possible line of defense for the relativist. The relativist could try to secure strategies to explain the permissibility of not retracting while continuing to hold that there is a requirement to retract. MacFarlane tries to explain away such presumed violations of the normative requirement to retract by appeals to the grounds or justification the speaker may have had in asserting. However, as will be shown, that line of defense fails to address the crucial issue in contention: that speakers cannot be required to retract an assertion that complied with a constitutive norm of assertion. This claim—that there is no such requirement—is supported by recent experimental work by Knobe and Yalcin (2014), which I will also briefly expound. Without the obligation to retract, assessment-relativism is unmotivated.

Although retractions are not obligatory in the conditions described by relativists, we may still be interested in understanding the speech act of retraction. Offering an account of such a speech-act is beyond the scope of this paper, but some considerations on what retraction is will be offered.

Section 1 distinguishes between indexical and nonindexical contextualism on the one hand, and assessment-relativism on the other, and concludes by explaining the connection between the requirement of retractions and the point of assessment-relative truth in the practice of assertion. Section 3 offers a review of three important objections to the obligatoriness of retractions. The objections show that there is no such obligation, that speakers who hold their ground are neither insincere nor irrational, and moreover that the imposition of that obligation would commit speakers to either irrationality or insincerity. This section further shows that MacFarlane fails to explain away the primary permissibility of not-retracting. Taken together, these objections make a strong case against the alleged support that retractions give to relativism. The objections are reinforced by recent experimental results that are also discussed in Sect. 4. Finally, Sect. 5 offers a some considerations on what retraction is.

2 Retraction and relativism

2.1 Introducing indexical contextualism, nonindexical contextualism and assessment-relativism

MacFarlane (2009) makes a distinction between what he considers to be two kinds of contextualist positions: indexical and nonindexical contextualism, which he characterizes on the basis of the different ways the two positions account for the context-dependence of semantic expressions in a language:

  1. 1.

    An expression is indexical iff its content at a context depends on features of the context.

  2. 2.

    An expression is context-sensitive iff its extension at a context depends on features of the context.

(MacFarlane 2009, p. 232)

That the difference between the forms of context-dependence, indexicality and context-sensitivity, exists is manifest in the analysis of sentences like

  • Tomorrow comes after today.

The content of the sentence as uttered in a given context depends on the time (day) of the context, but the sentence has the same extension (true) whenever it is used. So, a sentence can be indexical without its truth being context-sensitive. MacFarlane (2009) claims that sentential truth may also be context-sensitive without the sentence satisfying indexicality. In other words, the truth-value assigned to a given sentence may vary with features of context without it being the case that the content expressed in context so varies.

The framework for the distinction is provided by possible world semantics. Temporalists generalize the framework to relativize sentential truth not just to possible worlds but also to times. It is standard to consider that if a given uttered sentence expresses a contingent proposition, for instance, that Mario Draghi is the president of the European Central Bank in 2015, then this proposition is true in the actual world, but there are possible-worlds where it is false.

To generalize this idea to the temporal case is to say that certain sentences express a single proposition, which can be true or false at different times. For instance, the proposition that it is raining in Lisbon is true on 15 April 2015, but was false on 10 April 2015. Eternalists, in contrast, would agree with the variability in truth-value of the sentence ‘it is raining in Lisbon’ as it is uttered on different days, and agree with the temporalists about the actual truth (or falsity) of each utterance of that sentence, but do not agree that the sentence expresses the same proposition on those different days of use. Instead, they claim, the reason why the sentence has different truth-values on different occasions of use is due to the fact that it expresses different contents, different propositions, on those different uses, something like that it is raining in Lisbon on 15 April 2015 and that it is raining in Lisbon on 10 April 2015.

Kaplan (1989) says,

If c is a context, then an occurrence of \(\phi \) in c is true iff the content expressed by \(\phi \) in this context is true when evaluated with respect to the circumstance of the context. (Kaplan 1989, p. 522)

The circumstance of the context are those features of context like a possible world, or a time, that are necessary to evaluate the occurrence of \(\phi \) in c, i.e., to evaluate the content or proposition expressed as true or false at the world and time that the context where the sentence is used determines as relevant.

In recent years, there has been an active debate concerning the correct semantic analysis of a variety of types of expressions, including knowledge attributions, deontic and epistemic modals, conditionals, and value and taste predicates.Footnote 1 There are authors that defend the ‘context-sensitivity of the extension’ of expressions of the kind in consideration and that self-identify as relativists (see for instance Kölbel (2004b) or Recanati (2007)). There are authors that defend what MacFarlane calls ‘indexicality’ but who also self-identify as relativists (see for instance López de Sa (2015)). Given the latitude of possible “relativisms” in the debate, for the sake of simplicity and comprehension I will here follow McFarlane’s terminology and call the proponents of indexicality ‘indexical contextualists’, and proponents of the context-sensitivity of truth ‘nonindexical contextualists’.

In an important sense, both indexical and nonindexical contextualists are contextualists: the truth value assigned to a sentence as uttered at a context depends, ultimately, on features of the context of utterance, either because the context contributes to the content expressed by the sentence in context, or because context contributes to the selection of relevant parameters in the circumstances of evaluation. In the latter case, those parameters are the ones the utterance, or the assertion, concerns (see Perry 1993 on ‘concern’).Footnote 2

MacFarlane’s semantic proposal is radical in comparison with either kind of “contextualism”. By adding contexts of assessment, a different relativization of truth is made possible. Whereas on indexical and nonindexical contextualism, the very same utterance of a sentence is either true or false absolutely, on assessment-relativism the very same utterance of a sentence may be true with respect to one context of assessment and false with respect to another context of assessment. Assessment-relative truth can be deployed in offering the semantics of a variety of expressions. For instance, MacFarlane (2011) discusses how assessment-relativist semantics yields different results from contextualism about epistemic modals might and must.

According to the canon, Kratzer’s context-dependent analysis, the modals ‘might’ and ‘must’ are treated semantically as quantifiers over possibilities. The context of utterance contributes to determine the specific kind of modality at stake in each use (See Kratzer (1991), and for a recent defense Dowell (2011)). The logical form of a sentence like (1) is given in (2):

  1. 1.

    There might be ice cream in the freezer.

  2. 2.

    Might \((B)\phi \)

In (2), B is the modal base or domain of quantification that is determined in the context of use, and \(\phi \) is the prejacent—the sentence that is evaluated as true or otherwise with respect to each possible world in the modal base. For the cases of epistemic modals, the context of use determines a set of possible worlds compatible with the relevant epistemic state of the context of use (the information available at the context). (2) is true as used in the context if \(\phi \) is true in at least one world w in the modal base B of the context.

MacFarlane distinguishes the semantic analysis given by contextualism and that given by assessment-relativism. He represents an information state as a set of worlds—the worlds that are open possibilities given the information available. The compositional semantics offered recursively defines truth relative to a context and an index consisting of a world, time, information state, and assignment.Footnote 3

According to a Contextualist Postsemantics, \(\ulcorner \lozenge _{e}\phi \urcorner \) is true at c just in case \(\llbracket \lozenge _e \phi \rrbracket _{\langle w_c, t_c, i_c, a\rangle }^c = \) True for some \(w' \in i_c\), that is, just in case \(\phi \) is true at some world in the information state of the context of use. A similar derivation shows that, on a Relativist Postsemantics, \(\ulcorner \lozenge _{e}\phi \urcorner \) is true as used at c and assessed from \(c'\) just in case \(\llbracket \lozenge _e \phi \rrbracket _{\langle w_c, t_c, i_{c'}, a\rangle }^c = \) True for some \(w' \in i_{c'}\). (MacFarlane 2014, p. 264).

The difference between the contextualist postsemantics and the relativist postsemantics concerns which context determines the information state relevant for the truth of \(\ulcorner \lozenge _{e}\phi \urcorner \). For the contextualist, the truth of the epistemic modal depends only on information available at the context of use c, but for the relativist truth depends on information available at the context of assessment, that is, \(\ulcorner \lozenge _{e}\phi \urcorner \) is true just in case it is true at some world in the information state of the context of assessment. In other words, it is the information available at a context of assessment, not of utterance, that determines whether an utterance (a use) of an epistemic modal sentence like (1) ‘there might be ice-cream in the freezer’ is true.

MacFarlane offers assessment-sensitive compositional (post) semantics for a variety of expressions—crucially, those that have been at the focus of the debate between indexical and nonindexical contextualists. Knowledge attributions, conditionals, and value and taste predicates can be accommodated in an assessment-sensitive theory, not just epistemic (or deontic) modals. For instance, a use of a sentence like

  1. 3.

    Pocoyo is funny.

is only assigned a truth-value relative to the contexts of utterance and a standard of taste provided by a context of assessment. The perspective or standard of taste relevant is to be that of the assessor at the context of assessment, just like the information state required to assign a truth-value to the epistemic modal (1) was to be provided by an assessor. By contrast, a contextualist takes the relevant perspective to be the standard of taste determined at the context of utterance.

2.2 Constitutive rules and the point of true\(_{c_U, c_A}\)

MacFarlane explicitly acknowledges that making sense of assessment-relative truth, i.e. making sense of the point of introducing assessment-sensitive truth in a semantic theory, must be revealed by the pragmatic implications of the theory for assertion (see MacFarlane (2005b) and MacFarlane (2014), Chap. 5). MacFarlane refers to Dummett’s similar concerns about the point of truth and the practice of assertion:

[S]omeone who had an extensionally correct Tarskian truth definition for a language but did not understand the significance of characterizing sentences as true would not grasp the concept of truth. (MacFarlane 2014, p. 99)

Dummett himself had said,

[W]hat has to be added to a truth-definition for the sentences of a language, if the notion of truth is to be explained, is a description of the linguistic activity of making assertions. (Dummett 1978, p. 20)

Contexts of utterance can be understood in at least two ways—as the concrete situations in which sentences are uttered, and as formal sequences of parameters (fixed at concrete situations of use). The latter are formal theoretical models of those features that are considered necessary for a semantic representation of the content and truth-conditions of a given sentence as used. Contexts of assessment can also be understood as concrete situations in which a previous use of a sentence is assessed and as formal sequences of parameters (fixed at the respective concrete situations of assessment). It is to contexts qua formal entities that sentential truth is relative in the theory. But it is not the relativization of sentential truth to formal sequences of parameters that makes assessment-relativism a radical departure from other theories.

Although, technically, true at context of use \(c_U\) and true at context of use \(c_U\) and context of assessment \(c_A\) are different properties pertaining to different semantic theories, there are cases where the latter is reducible to the former. This occurs when a context of assessment and a context of use are identical, or whenever the relevant parameter of the context of assessment is determined at the context of use. Obviously, a semantic theory that introduced contexts of assessment, but declared that it is the context of use that fixes, say, the relevant information state, or the relevant standard of taste, in the context of assessment would make assessment-sensitive semantics entirely redundant.

Assessment-relativism is distinctive because of the practical normative difference that sets it apart from either indexical or non-indexical contextualism. It is this difference that gives significance to truth at a context of use and a context of assessment, and makes a normative difference to what a speaker ought to do in specific conditions. The normative difference concerns the putative obligations a speaker acquires in virtue of asserting. The relativist must offer a pragmatic account of assertion to elucidate the point of assessment-sensitive truth. Like Dummett (1978), MacFarlane seeks to characterize the significance of assessment-relative truth by characterizing the activity of making assertions. MacFarlane connects truth with the practice of assertion by adopting a constitutive account, where a truth rule is a constitutive norm.

Before moving forward, let me first signal what is at stake in adopting a constitutive theory of assertion. Pagin (2015) elucidates the features of a constitutive norm of assertion, what characterizes a constitutive rule as such. A constitutive rule N is one that applies specifically to assertion (N1), it uniquely identifies assertion (N2), and it is essential to assertion that it is subject to N (N3). The rule constitutes assertion N4. Nonetheless a particular assertion can be subject also to other rules (e.g., etiquette, grammaticality, relevance, etc.) (N5), but an assertion that violates N is incorrect (N6). As Pagin (2015) puts it, “[a]n assertion that violates the norm is still an assertion. Nothing but an assertion could violate the norm.”Footnote 4

A general truth rule would be

  • Truth Rule At a context c, assert that p only if p is true at c. (MacFarlane 2014, p. 101)

This makes a connection between what constitutes assertion, and a predicate true at c. MacFarlane needs a constitutive rule suitable for assessment-sensitive (post)semantics, one that relativizes truth to a context of use and a context of assessment. After discussing alternatives, MacFarlane offers a reflexive rule:

  • Reflexive Truth Rule An agent is permitted to assert that p at a context \(c_1\), only if p is true as used at \(c_1\) and assessed from \(c_1\). (MacFarlane 2014, p. 103)

It is not irrelevant how one arrives at a Reflexive Truth Rule, where the context of use and context of assessment are identical. It can’t be that speakers are permitted to assert p only when p is true as used at their context of use and a (or some) context of assessment distinct from the context of use, since there are indefinitely many such contexts. It is also unreasonable to require truth at all contexts of assessment (and MacFarlane admits that truth at most contexts of assessment would be a mysterious requirement). However, with the reflexive rule, contextualist and relativist postsemantics are pragmatically indistinguishable.

Imagine that a subject A asserts that there might be ice-cream in the freezer at \(c_1\). A’s assertion is subject to the Reflexive Truth Rule, and, as such, she makes a correct assertion only if it is true that there might be ice-cream in the freezer, given the information available at \(c_1\)—or, to be more precise, only if her use of the sentence “there might be ice-cream in the freezer” at \(c_1\) is true as used at \(c_1\) and assessed from \(c_1\). This means that the Reflexive Truth Rule fails to distinguish between contextualism and relativism, and hence that it does not serve on its own to give significance to assessment-relative truth. Since the Reflexive Truth Rule yields the same predictions about the correctness conditions for the making of assertions in a contextualist and in an assessment-relativist postsemantics framework, we could drop the reference to contexts of assessment entirely. Contexts of assessment, qua formal sets of parameters, would be idle wheels.

It is the obligatoriness of retractions, according to MacFarlane, that reveals the significance of assessment-relativism. Making sense of assessment-relative truth, as he stresses, is a matter of making sense of the normative commitments undertaken by speakers. If we cannot make sense of those commitments, then we cannot make sense of assessment-relativism. As he says,

The basic thought is that the pragmatic difference between R[elativism] and C[ontextualism] manifests itself in norms for the retraction of assertions rather than norms for the making of assertions. R[elativism] predicts that an assertion of p at \(c_1\) ought to be retracted by the asserter in \(c_3\), while C[ontextualism] predicts that it need not be retracted. Thus, the Reflexive Truth Rule is not so much wrong as incomplete. It needs to be supplemented by a constitutive norm for retraction:

  • Retraction Rule An agent in context \(c_2\) is required to retract an (unretracted) assertion of p made at \(c_1\) if p is not true as used at \(c_1\) and assessed from \(c_2\).

By “retraction,” I mean the speech act one performs in saying “I take that back” or “I retract that.” The target of a retraction is another speech act, which may be an assertion, a question, a command, an offer, or a speech act of another kind. The effect of retracting a speech act is to undo the normative changes effected by the original speech act (MacFarlane 2014, p. 108) (My emphasis).

There are two ways to interpret what the Retraction Rule is. The first interpretation, which I think is the correct one, sees the Retraction Rule as aiming to be a second constitutive rule of assertion, besides the Reflexive Truth Rule. This makes sense in the overall picture of connecting a compositional assessment-sensitive semantic theory with the practice of making assertions. It is also a plausible interpretation given the fact that retractions, in general, are not specific to assertions. One can retract a promise, a proposal, a bet, an order, a supposition, a question, etc.Footnote 5 MacFarlane further says,

Note that the Retraction Rule obliges retraction under certain conditions, while the Reflexive Truth Rule forbids assertion under certain conditions. This is as it should be. In asserting, the fault lies in commission (asserting something untrue), while in retracting, the fault lies in omission (failing to retract something untrue). There is nothing inherently wrong with retracting an assertion one still thinks is true—one may not want others to rely on one’s word in this matter, or one may not want to take on the obligation of defending the assertion—and doing so is not ‘insincere’ in the way that asserting something one does not believe to be true is. (MacFarlane 2014, p. 109)

The fact that what one retracts is an assertion (“there is nothing wrong with retracting an assertion one still thinks is true”) reinforces the idea that the Retraction Rule is to be taken on a par with the Reflexive Truth Rule on a normative constitutive theory of assertion.

The second interpretation of the Retraction Rule takes it as a constitutive rule of retraction. This seems to be MacFarlane’s interpretation—he does say “the Reflexive Truth Rule needs to be supplemented with a constitutive rule for retraction”, but, given the previous considerations, we have good reasons to read the Retraction Rule as a second constitutive norm of assertion.Footnote 6 Given that different speech acts can be retracted, presumably there would be different kinds of retraction for each distinct kind of speech act. The argument in this paper is not, in any case, affected by interpreting the Retraction Rule as constitutive of assertions or as constitutive of retractions.

Setting aside exegetical issues, how does the Retraction Rule reveal the normative difference in practice between a Contextualist postsemantics and a Relativist postsemantics? Imagine that A asserts at \(c_1\) that there might be ice-cream in the freezer. A’s assertion is subject to the Reflexive Truth Rule, and, as such, her assertion is permissible only if it is true that there might be ice-cream in the freezer, given the information available at \(c_1\). Let us further assume that at the context of assessment \(c_1\) it is indeed true that there might be ice-cream in the freezer. But now, at \(c_2\), A learns that the kids finished the ice-cream last night and there is nothing left. Although it was permissible for A to assert as she did at \(c_1\), what should she do when she learns that there is no ice-cream left?

We know A may want to retract (“there is nothing wrong with retracting an assertion one still thinks is true”), but does she do something incorrect if she does not? If she does something wrong by not retracting her assertion, that would support the Retraction Rule and would give assessment-sensitive truth the practical significance it requires. But if she does nothing wrong if she does not retract, then the Retraction Rule is not constitutive of the practice of assertion, and fails to give any practical significance to assessment-sensitive truth.

There are two ways that intuitions about retractions may be used by the relativist against contextualism:

  1. (I)

    Intuitions about retraction may be used against indexical contextualism. Upon learning that there is no ice-cream left, it is natural for A to retract her assertion that there might be ice-cream in the fridge. It follows that the content of her assertion must not be that her information was compatible with there being ice-cream in the fridge, as the indexical contextualists claim, since that content is true (MacFarlane 2014, p. 260).

  2. (II)

    Intuitions about retraction may be used against non-indexical contextualism. Upon learning that the kids finished the ice-cream, A ought to retract her assertion that there might be ice-cream in the freezer. Non-indexical contextualists lack the resources to explain this fact, since their account of the norm of assertion appeals only to the notion of truth relative to the context of assertion (a feature A’s assertion has) (MacFarlane 2014, p. 256).Footnote 7

The next section reviews three recent objections to the use of intuitions about retractions in favor of relativism. These objections clearly put into question (II) above, by showing that it is permissible not to retract. Moreover, it is shown that MacFarlane’s attempt to dismiss the seriousness of the permissibility of not retracting does not address the crucial point at issue. Section 4 will review some recent experimental results on speaker’s intuitions about epistemic modal claims. These reveal that the alleged connection between retraction and falsity is weaker than assumed, putting (I) into question. The objections to relativism, and the experimental results, both tell us that retraction data do not support assessment-relativist semantics.

3 Objections—not obligatory or not rational

Relativists emphasize the importance of retracting a past assertion when the asserted sentence is false at the context of the putative retraction. The examples typically offered by relativists are meant to confirm the claim that retractions of past perspective-dependent claims are mandatory when the relevant parameter has changed. Consider for instance, that Mimi says (4) when she is 3, but says (5) when she is 9:

  1. 4.

    Pocoyo is funny!

  2. 5.

    Pocoyo is not funny; I was so wrong when I was 3.

If Mimi has done what she ought to do in retracting, assessment-relativism makes the right normative prediction, but, as the previous section highlights, contextualism does not.

Several authors have recently developed objections against the normative prediction made by assessment-relativism. I will review what I take to be the main objections available in the current literature, in particular, the reasons given by von Fintel and Gillies (2008), Marques (2014), and Ross and Schroeder (2013) against the obligatoriness of retraction. Taken together, the objections show that there is no obligation to retract assertions of sentences of the kinds considered, that speakers who hold their ground are neither insincere nor irrational, and moreover that the assessment-relative retraction norm would commit speakers to either insincerity or irrationality. Given this, the argument against (nonindexical) contextualism is unsound ((II) above in previous section). It’s not true that speakers have an obligation to retract, and, hence, it’s not true that retraction is the speech act that gives significance to assessment-relative truth. One may still be intrigued about the naturalness of a retraction in the various cases considered, and inquire whether this suggests that the naturalness of retracting is better explained by the falsity of the content asserted. This would weigh against (indexical) contextualism, according to which the asserted content is true ((I) above in the previous section). The putative connection between retraction and falsity is put into question by recent experimental results, which are discussed in the next section.

Von Fintel and Gillies discuss several might claims where the prediction that one must retract fails. Consider for instance, that A asserts (1) and later finds that the kids finished the ice cream.

  1. 1.

    There might be ice cream in the freezer.

According to the relativist, when it is revealed that there is no ice cream left, the asserter of (1) ought to retract with something like “Oh, I guess I was wrong” or “I take that back, there’s no ice cream left then”.

The point von Fintel and Gillies make (p. 81 ff), however, is that not all mights are retracted in the face of new evidence. Often, speakers resist an invitation to retract. A speaker could resist thus:

  1. 6.

    Look, I didn’t say there is ice cream in the freezer; I said there might be. Maybe the kids finished it last night. Sheesh.

They further point out that constructions under if are particularly resistant:

  1. 7.

    If the kids haven’t found it, there might be ice cream in the freezer.

There is no impression that the speaker should retract her assertion of ‘there might be ice cream in the freezer’ when it is embedded in a conditional like (7). Edgington (1995) would consider that examples like (7) actually count as conditional assertions. For Edgington, a conditional assertion of ‘If the kids haven’t found it, there is ice cream in the freezer’ is an unconditional assertion that there might be ice cream in the freezer, when it is true that the kids haven’t found the ice cream. If the kids had left a note on the freezer saying ‘we ate all the ice cream!’ there would still be no expectation that the conditional assertion should be retracted. The construction: “I didn’t say there is... I said there might be” seems to be a perfectly adequate and appropriate way to reply to criticism.

In previous work, Marques (2014) focuses on aesthetic and personal taste predicates instead of deontic modals, and argues that a speaker who refuses to retract a past assertion after a change of perspective (standard of taste) is neither irrational nor insincere. Recall Mimi, aged 3 years old, saying “Pocoyo is funny”. Suppose that, at 63 years of age, she recalls what she thought of Pocoyo when she was 3. According to the relativist, she should retract, because she is not now disposed to find Pocoyo funny. Presumably, if the truth of “Pocoyo is funny” varies across time, it’s because Mimi’s dispositions towards children’s characters change over time. Her dispositions to find Pocoyo funny earlier in life are no more veridical than her dispositions not to find it funny later in life. Her utterance of “Pocoyo is funny” is true when she is 3, and her utterance of “Pocoyo is not funny” is true when she is 63. But she should, according to the relativist, retract her earlier assertion when she is 63. She should, moreover, be already committed to retracting when she initially asserts “Pocoyo is funny”.

This description of the data is generalized beyond Mimi when she is 3 and applies to anyone who makes an assertion about taste, humour or aesthetics: the utterance of a sentence S was true in the past at \(c_1\). The utterance of \(\lnot S\) is true now at \(c_2\). The relativist holds that the speaker should retract at \(c_2\) the past assertion made at \(c_1\). In asserting S at \(c_1\), the speaker should already be committed to retracting at any future context c when their perspective (standard of taste, humour, etc) has changed suitably. Everyone is in a position to know that S is a sentence that they will not use in the future when their standards (or perspectives) change.

The Retraction Rule imposes an obligation to take back the assertion of S when S is not true as used at \(c_1\) and assessed at \(c_2\). What motivates the Retraction Rule? Surely, the presumed obligation derives from the assumption of the falsity of the sentence as used at the context of assessment. A retraction does require admitting fault. By retracting, Mimi would not be simply signaling that she no longer has the inclination to find Pocoyo funny. If the Retraction Rule is in place, Mimi is also admitting fault. But instead of retracting, she could rather reply:

  1. 8.

    I used to find Pocoyo funny, and I do not anymore. Still, I was not wrong when I was 3 years old and found it funny. It was the funniest thing back then!

In uttering (6), Mimi informs the audience that indeed she no longer has an inclination she had when she was 3 and refuses to retract her past assertion of “Pocoyo is funny”. Why should she be expected to retract?

As MacFarlane admits, people with different dispositions are not better positioned to assess a given matter than the subject who is being challenged. The context of assessment that matters for the constitutive rule of assertion, the Truth Rule, is the asserter’s own context, and hence one is not under any obligation to retract when confronted with others that do not share the same context of assessment.Footnote 8 Yet, mutatis mutandis, the same is true in the individual case. If one’s dispositions towards cartoons, or food, change across time, one’s later responses and reactions are not more veridical than one’s earlier responses and reactions. A fortiori, one has no obligation to retract one’s earlier assertion, which was correct when made and complied with the constitutive Truth Rule of assertion, at a later time.

For there to be any obligation to retract claims of the kind discussed, we would have to perceive speakers who resist retraction as somehow irrational, or insincere. Subjects would have to be perceived as violating a normative requirement, in a way that lying and misleading are perceived to be a violation of norms of communication.Footnote 9 However we have no perception of insincerity or irrationality, either in the might cases discussed by von Fintel and Gillies, or in the funny case just considered.

Finally, Ross and Schroeder (2013) argue that it would in fact be irrational for an asserter who is rational and sincere to commit to retracting. They begin by noting that speakers who are rational and sincere in their assertions, and who know that in the future will rationally and sincerely assert the negation of what they had previously asserted, can still make claims like the ones we have discussed thus far (like (1) or (3)). They call this the reversibility thesis. They argue that the only invariantist position that can accommodate the reversibility thesis is relativism. This raises a dilemma for the relativist, because reversibility and disagreement (and retraction) are incompatible.

Imagine a subject, Ankita, who is writing a story for a newspaper on a local murder investigation. In the morning, it is unknown whether Axeworthy is the murderer. Ankita knows that in the afternoon a DNA test will establish whether or not he is. In the morning, Ankita asserts m, and in the evening asserts \(\lnot m\), where m is:

  1. 9.

    m: Axeworthy might be, and might not be, the murderer.

Now, when Ankita asserts m, she is sincere and rational. Presumably she would undertake the commitment to vindicate her assertion of m at a later context, too. But in the evening she will no longer be in a position to assert m sincerely. She will assert \(\lnot m\), and hence be committed to vindicating this latter assertion. And so, Ankita will not be in a position to vindicate the earlier assertion of m as well, since it will be common knowledge by then whether Axeworthy is the murderer.

Ross and Schroeder’s argument is essentially this. If Ankita is sincere when she asserts in the morning, then she must intend to vindicate or retract her assertion in the face of a challenge in the evening. And so, if she intends the disjunction, she must intend to retract. But since she knows that in the evening what she asserts will not be true, she will be unable to vindicate its truth in the evening. So she knows that she will have to retract. But if she knows her assertion will be indefensible in the face of being challenged in the evening, and if she already intends to retract her assertion in the face of any such challenge, then her present assertion would hardly seem to qualify as sincere (Ross and Schroeder 2013, pp. 69–70). Knowing all this, Ankita could not rationally intend to vindicate the truth of her assertion as made in the morning later in the evening. But, on MacFarlane’s theory, this is the intention that Ankita must have. Hence, Ankita could not rationally and sincerely make her assertion in the morning.

The objections to the Retraction Rule are, hence, of two kinds. First, there are counterexamples to the claim that there is an obligation to retract an assertion of a sentence that is not true (at the context of use and context of assessment). The counterexamples concern both epistemic modals and taste predicates. Second, there are considerations concerning how a given rule impacts on the rationality and sincerity of speakers of a language with such a rule in place: speakers are neither irrational nor insincere if they don’t follow that rule, and would be either irrational or insincere if they had to (whether the Retraction Rule is interpreted as constitutive of assertion or of retraction.) MacFarlane offers a direct response to the first kind of objection above, which is also an indirect response to second kind of objection. But his response does not touch on the core of either, as I’ll now argue.

The Retraction Rule is offered as a constitutive rule, and I have claimed that it is best to interpret it as aiming to be constitutive of assertion. Interestingly, the adoption of a Retraction Rule has another Dummettian inspiration. Dummett also refers to retraction in connection with assertion and truth, and says:

Of course, we can talk about what is required to be the case by an assertion; but this notion relates, once again, to how we recognize the assertion as incorrect...There is a well defined consequence of an assertion’s proving incorrect, namely that the speaker must withdraw it, just as there is a well-defined consequence of disobedience. (Dummett 1978, p. 22)

Now, MacFarlane’s resistance to the counterexamples has two components. First, he justifies the rationality and legitimacy of speakers’ refusal to retract. Second, he claims that retracting is not an admission of error. He says,

Suppose one’s evidence strongly suggests that Uncle Jack is coming to lunch, and on the strength of that evidence you assert that Uncle Jack is coming. A bit later, Aunt Sally calls to say that Uncle Jack has broken his leg. This makes it quite unlikely that he is coming, so you retract your assertion. Nonetheless, you were perfectly reasonable in making it, and cannot be criticized for having done so. Retracting it is not admitting fault. (MacFarlane 2014, p. 110)

It is curious to see the contrast between the diagnosis that, according to MacFarlane, to retract is not to admit fault, and Dummett’s straightforward connection between the constitutive norm of assertion (Truth), the fact that violating that norm has a well-defined consequence—that the assertion is incorrect, and that that requires one to withdraw the assertion. One must withdraw, if Dummett is right, because the assertion is incorrect.

Once truth is relativized to a context of use and context of assessment, what is the connection between a retraction and an admission of error? The Retraction Rule mandates that one retracts when the asserted sentence is false at the context of use \(c_U\) and the context of assessment \(c_A\). If it is false because it is not true as used, then one violated the Truth Rule. In this case, one should retract because one violated a constitutive norm of assertion, and that means that the assertion is incorrect—to retract here is to admit error. Let us now suppose that a sentence is not true at the context of use and a context of assessment, because it is not true at the context of assessment \(c_A\). Now the Reflexive Truth Rule was not violated. Yet, to say that to retract in this condition is not to admit fault is to require speakers to draw a distinction between retractions that are admissions of error under some conditions of falsity, and retractions under other conditions of falsity that are not admissions of error. Moreover, if one does not retract, then one is incurring in error, given the presumption that the Retraction Rule is a constitutive rule.

There is an essential connection between a constitutive rule of a speech act like assertion and the primary propriety and impropriety of that act. These are distinguished from other nonessential senses of propriety or correctness, as is stated by DeRose (2002).

As happens with other rules, a kind of secondary propriety/impropriety will arise with respect to [the Knowledge Norm of Assertion]. While those who assert appropriately (with respect to this rule) in a primary sense will be those who actually obey it, a speaker who broke this rule in a blameless fashion (one who asserted something she didn’t know, but reasonably thought she did know) would in some secondary sense be asserting properly...(DeRose 2002, p. 180)

DeRose’s idea develops a point by Williamson (1996, 2000), who argues for a constitutive knowledge norm of assertion, and claims that it may be reasonable to assert that p if one reasonably believes that one knows that p (Williamson 2000, pp. 256–257).

The central point of MacFarlane’s resistance to the first objection concerns the reasonableness and the grounds of the resistance to retract. In this, his strategy is similar to that of DeRose and Williamson. However, in a direct reply to the argument by von Fintel and Gillies (2008), MacFarlane argues that although a speaker may have made an assertion “responsibly”, and on “excellent grounds”, the speaker anyway has an obligation to retract because the sentence asserted is false (at the context of use and the context of assessment). This makes his reply puzzling (the case discusses whether keys might have been in the drawer, rather than ice-cream in the freezer).

On this interpretation, Alex’s claim “they might have been there” is intended to explain the reasonableness of her assertion. It shows that, in making the assertion, Alex was confirming to the Reflexive Truth Rule. But it does not support a refusal to retract. For that, Alex would need to be in a position to assert not just that for all she knew, the keys might have been in the drawer, but that the keys might have been in the drawer. And if the relativist is right, she can’t assert that, because she knows now that the keys weren’t in the drawer when she made her claim. (MacFarlane 2014, p. 258)

MacFarlane attributes reasonableness to the assertion, and had conceded on these grounds that Alex could resist to retract. This is to ascribe secondary propriety or correctness to the assertion, and to the resistance to retract. But the argument is puzzling. The reason given for the warrant was the compliance with the Reflexive Truth Rule —constitutive norm of assertion. So the assertion was not only secondarily correct, it was also primarily correct. The resistance to retract is grounded not merely on being warranted or being responsible, but on complying with a norm. It is in this sense that the counterexamples to the obligatoriness of retractions are refusals to admit fault in the primary sense, and show that there is no obligation to retract.

MacFarlane’s argument comes down to the following two claims:

  1. (A)

    A speaker must retract at a context of assessment \(c_2\) an assertion made at a context \(c_1\) of a sentence S, if the speaker cannot use S at \(c_2\) to make an assertion. (Notice that MacFarlane requires that the speaker be in a position to assert using the epistemic modal sentence).

  2. (B)

    A speaker cannot make an assertion using a sentence S at a context \(c_2\) if the use of S at \(c_2\) is not true as used at \(c_2\) (and assessed from \(c_2\)).

Everyone in the debate can accept (B) (or the equivalent reformulation of (B) with truth relative only to contexts of use), i.e., everyone has a way of stating the (Reflexive) Truth Rule. Upon learning that there is no ice-cream left at \(c_2\), the speaker can’t truthfully use the sentence ‘there might be ice-cream in the freezer’, because the relevant epistemic states that contribute to determine what possibilities are compatible with the truth of the prejacent are different at \(c_1\) and at \(c_2\). But (A) doesn’t follow. Even though a speaker can’t use a sentence at a context to make a truthful assertion, it does not follow that that speaker must retract a truthful past assertion made with that very same sentence. It does not follow in general for indexical or context-sensitive sentences (‘I’m hungry’, ‘it’s raining’, etc), and it doesn’t follow for epistemic modals either.

(A) is a restatement of the claim that there is an obligation to retract. But MacFarlane did not address the main contention of the objections: it is because speakers make assertions that are primarily correct, in that they comply with the presumed constitutive truth norm, that they cannot be under any obligation to retract those assertions. And it is because of this primary correctness that speakers are neither irrational nor insincere if they don’t retract, not just because they are “responsible” or have “good grounds”.

This argument severely weakens assessment-relativism. If retraction is not mandatory when a past asserted sentence is false at a current context of assessment, and if it being mandatory would commit speakers to insincerity or irrationality, then assessment-relativism is not a genuine alternative semantic theory.

The next section reviews recent experimental work that confirms the permissible but not obligatory status of retraction, and confirms, moreover, the divorce between judgments of falsity and judgments about retractions.

4 Experimental results on retraction and falsity

The previous section argued that there is no obligation to retract a past assertion made using a sentence S when one can’t truthfully use S in the new context. This means that there is no normative difference between assessment-relativism and contextualism, and that we are still lacking an account of the point of assessment-relative truth in communication. Yet, one may wonder whether the naturalness of retracting can at least weigh against indexical contextualist semantics.

In recent experimental work, Knobe and Yalcin (2014) tested relativist semantics for bare epistemic modals as stated in the thesis (J) below. Their results reveal that native speakers’ judgments about the falsity of an epistemic modal sentence and judgments about the appropriateness of a retraction of an epistemic modal sentence diverge. This shows that even if speakers are inclined to think that a retraction of an assertion of a sentence with an epistemic modal is natural, that cannot be taken as evidence that speakers think that the assertion is false. And if they don’t agree that the assertion is false, then (J) below is false:

  1. (J)

    Competent speaker/hearers tend to judge a present-tense bare epistemic possibility claim (BEP) true only if the prejacent is compatible with their information (whether or not they are the producer of that utterance); otherwise the BEP is judged false. (Knobe and Yalcin 2014, p. 3)

Knobe and Yalcin (2014) carried out four experiments to test speakers’ intuitions about intercontextual “eavesdropper” and retraction cases. They focused only on cases that involve bare epistemic modals, and did not cover deontic modals, or the various value and taste predicates that have been discussed in the recent literature on relativism. They tested two crucial cases that are brought up against the contextualist account of bare epistemic modals. The first is this:

FAT TONY

Fat Tony is a mobster who has faked his own death in order to evade the police. He secretly plants highly compelling evidence of his murder at the docks. The evidence is discovered by the authorities, and word gets out about his apparent death. The next evening, from his safehouse, Fat Tony watches a panel of experts on the news discussing the question of whether he is dead. Expert A has had a good look at the evidence found at the scene. “Fat Tony is dead,” he says. Expert B has also had a good look at the evidence, but his assessment is more cautious. “Fat Tony might be dead,” B says. (Knobe and Yalcin 2014, p. 4)

In their first experiment, they randomly distributed among participants one of the following sentences

  • (NONMODAL-TRUE) What Expert A said is true.

  • (NONMODAL-FALSE) What Expert A said is false.

  • (MODAL-TRUE) What Expert B said is true.

  • (MODAL-FALSE) What Expert B said is false.

Each participant was asked to rate the relevant statement on a Likert scale from 1 (‘completely disagree’) to 7 (‘completely agree’). The results of the first experiment are shown in Fig. 1 below, and show that people’s judgments reveal that (J) is false.

Fig. 1
figure 1

(Knobe and Yalcin 2014, p. 9, Fig. 1)

The mean rating for judgement of falsity of the nonmodal was approximately 6 on the scale, whereas the mean rating for the judgment of falsity of the modal was below 2 on the scale. In contrast, speakers were much more inclined to agree that the modal was true (mean rating of almost 5) than to agree that nonmodal was true (agreement below 2).

Whatever other results are obtained with respect to the retraction of modal and nonmodal sentences, it will have to be considered against the fact that speakers judge (J) as false—speakers don’t judge a present tense bare epistemic modal only with respect to their current state of information. This entails that assessment-relative semantics makes the wrong predictions about native speakers’ judgments about truth at contexts of use and contexts of assessment.

Knobe and Yalcin’s third and fourth experiments reveal interesting results on speaker’s intuitions concerning retractions. In the third experiment, participants received the FAT TONY vignette complemented with the following information:

Shortly thereafter, new evidence comes to light, and everyone now agrees that Fat Tony is actually alive.

  • Expert A then says, “I was wrong—Fat Tony was actually alive.”

  • Expert B also says, “I was wrong—Fat Tony was actually alive.”

(Knobe and Yalcin 2014, p. 6)

Participants, who were randomly assigned to either the modal or the nonmodal condition, were then asked to assess the following:

  • Expert A [B] was right to say “I was wrong.” (idem)

In this experiment, participants agreed that it would be right to retract in the nonmodal condition (mean rating above 6 in the Likert scale), i.e., that it would be right for expert A, who said “Fat Tony is dead”, to retract. But the interesting result was that participants in the modal condition gave the statement an intermediate rating (approx. 4 on the scale), which was significantly lower than the one obtained for the nonmodal condition.

The results of this experiment were not considered conclusive for the relativist’s prediction about retraction, in particular it did not conclusively favour MacFarlane’s Retraction Rule, which requires retraction. The Retraction Rule predicts that the mean rating of the modal condition should converge with the mean rating of the nonmodal condition, which was not confirmed. Given that this experiment did not give conclusive evidence, Knobe and Yalcin carried out one further experiment that focused on a distinct kind of case. Moreover, they wanted to disentangle people’s intuitions concerning the truth or falsity of the past claim from the “appropriateness” of the retraction of that claim.

The last case tested was closely modeled on one given by MacFarlane (2011):

Sally and George are talking about whether Joe is in Boston. Sally carefully considers all the information she has available and concludes that there is no way to know for sure. Sally says: “Joe might be in Boston.” Just then, George gets an email from Joe. The email says that Joe is in Berkeley. So George says: “No, he isn’t in Boston. He is in Berkeley.” Participants in the nonmodal condition received a vignette that was exactly the same, except that Sally says “Joe is in Boston.” (Knobe and Yalcin 2014, p. 14)

Participants in this experiment were randomly assigned to answer one of two questions. Some were asked whether (a) it would be appropriate for the speaker to retract, by being asked if they agreed with the statement “it would be appropriate for Sally to take back what she said”. The remaining participants were asked (b) if they agreed with the statement “what Sally said is false”. The results are represented in Fig. 2 below.

Fig. 2
figure 2

(Knobe and Yalcin 2014, p. 15, Fig. 4)

In the nonmodal case (a), Knobe and Yalcin’s results did not show any significant difference between the appropriateness of retraction and the falsity of the claim (mean rating for appropriateness to retract above 6 and mean rating for falsity at 6 on the Likert scale). But, in the modal condition (b), the results show two things: there was again a disparity in the mean ratings for judgments of falsity (around 3 on the scale) and those of appropriateness to retract (around 5,5 on the scale). Basically, although people don’t think that what Sally said, in the modal condition, is false, they tend to agree that it would be appropriate for her to retract.

As Knobe and Yalcin say, “[t]aken together, these four experiments tell a consistent story. What they suggest is that (J) is mistaken.” (Knobe and Yalcin 2014, p. 17). (J) is mistaken because judgments about the appropriateness to retract and judgments of falsity converge in nonmodal cases, but diverge in the modal case. This divergence is such that, whatever the reason for people to tend to agree that a retraction is appropriate, that cannot be due to the falsity of the original assertion of the epistemic modal. This undermines the support that intuitions about retraction are meant to give to relativist semantic analysis of epistemic modals and against (indexical) contextualist analysis.Footnote 10

Now, MacFarlane conceded that “there is nothing inherently wrong with retracting an assertion one still thinks is true” (MacFarlane 2014, p. 109). Given this, it is perhaps interesting that people judge it appropriate to retract a true assertion, or at least an assertion that one does not think is false. However intuitions on the naturalness or appropriateness of retraction do not motivate a radical revision of semantic theories.Footnote 11

The final section offers some tentative considerations about why speakers may think that it is adequate to retract true claims.

5 Permissible retractions as changes in the common ground?

In previous sections, three objections to the appeal to retractions in contemporary semantic relativist theories were reviewed, as well as recent experimental results involving speakers’ judgments on retraction and falsity. The objections and the experimental results tell us a consistent story: that retraction does not motivate assessment-relativist semantics.

In Sect. 2.2, I distinguished the two ways retractions can be used against contextualism:

  1. (I)

    Intuitions about retraction may be used against indexical contextualism. Upon learning that there is no ice-cream left, it is natural for A to retract her assertion that there might be ice-cream in the fridge. It follows that the content of her assertion must not be that her information was compatible with there being ice-cream in the fridge, as the indexical contextualists claim, since that content is true (MacFarlane 2014, p. 260).

  2. (II)

    Intuitions about retraction may be used against nonindexical contextualism. Upon learning that the kids finished the ice-cream, A ought to retract her assertion that there might be ice-cream in the freezer. Non-indexical contextualists lack the resources to explain this fact, since their account of the norm of assertion appeals only to the notion of truth relative to the context of assertion (a feature A’s assertion has) (MacFarlane 2014, p. 256).

In the first place, the objections in Sect. 3 show that the Retraction Rule can’t be a constitutive norm of assertion (or of retraction). Speakers are not under the obligation to retract an assertion of a sentence that is true as used but false at the context of assessment (if anyone has difficulty with understanding what “true as used but false as assessed” means, replace that with “true at the context of use \(c_1\) and context of assessment \(c_1\), but false at context of use \(c_1\) and context of assessment \(c_2\)). Hence, retractions don’t reveal the significance, or the point, of assessment-relative truth. Intuitions about retraction cannot be used against nonindexical contextualism ((I) above).

In the second place, the experimental results with epistemic modals further show that people’s judgments on the appropriateness of retracting an assertion of an epistemic modal do not converge with their judgments about the falsity of those assertions. That means that even if it is natural to retract, the motive for retracting is not the falsity of the assertion. Intuitions about retraction cannot be used against indexical contextualism, either ((II) above).

The negation of the claim that a retraction is obligatory is the claim that retractions are permissible. Being permissible means, only, that retraction are neither forbidden nor obligatory. That is compatible with a retraction being appropriate for reasons that do not concern directly the truth or falsity of the semantic content of a sentence as used in a context (I will no longer take into account contexts of assessment, given the strength of the arguments against its pragmatic significance).

Whatever explanation there is of the appropriateness to retract, it must respect two features of retractions:

  1. (i)

    It is permissible to retract a past assertion that is true, and

  2. (ii)

    It may be appropriate to retract a true assertion for reasons that do not directly concern truth.

Finding an explanation for these aspects of retractions is not to replace the Retraction Rule with a new rule. It is also not an attempt to find a distinct constitutive rule of assertion, or even a constitutive rule of retraction itself. It is merely an attempt to understand the conversational mechanisms of retraction.

This paper started with a broad description of features of retraction: people may retract because they don’t want to induce others in error, but people may also retract when they were offensive or disrespectful towards others, and may retract various kinds of speech acts. Although this is speculative, I suspect that people are more inclined to think that a speaker is required to retract something offensive than something false. Nonetheless, the two features (i) and (ii) above, and the possible retractions of offenses, suggests a possible avenue of inquiry. The remaining of this article will offer some preliminary considerations on how an account of retraction could proceed.

Derogatory discourse, and in particular uses of pejoratives, reveal several features that, as many authors argue, go beyond semantic content. Pejoratives express negative psychological attitudes of variable derogatory force, and are associated with various taboos (for an overview, see Hom (2010)). Authors like Richard (2008) claim that pejorative expressions have no truth-conditional content, and slurring assertions by compositionality lack truth-conditional content. Macià (2002) and Schlenker (2007) claim that pejoratives carry presuppositions that the speaker thinks that certain people can (or should) be treated in a contemptuous way. To accept an assertion with a pejorative is to accommodate these presuppositions in the common ground. Potts (2005) defends a hybrid account of pejoratives. Pejoratives conventionally implicate negatively valanced content, and this conventionally implicated expressive content is not truth-functional.

Uses of pejoratives are offensive, and speakers who use pejoratives are often chastised in a way that speakers who make false assertions are not. On some of the available accounts of pejoratives, it may be appropriate for a speaker to retract an offense that results from a slurring assertion, even if that assertion is not false (different accounts of pejoratives will offer different explanations of why the assertion is not false).Footnote 12

The combination of true assertions with conflict has been explored also in the literature on disagreement. Recently, authors like López de Sa (2008), García-Carpintero (2008), Sundell (2011), Huvenes (2012), Marques and García-Carpintero (2014), and Marques (2015) offered alternative explanations of how contextualists can appeal to a combination of pragmatic mechanisms to account for disagreements expressed with epistemic modal or personal taste claims. The mechanisms include presuppositions of commonality (López de Sa 2008), and metalinguistic disputes over the choice of salient standards (García-Carpintero 2008; Sundell 2011), for instance). Contextualists have also appealed to conflicts of non-doxastic attitudes (Sundell 2011; Huvenes 2012; Marques 2015). A speaker may deny or accept a variety of related information beyond the semantic content asserted. One may deny via metalinguistic negation content that was not literally asserted, but presupposed or implicated.Footnote 13

Interestingly, Knobe and Yalcin (2014) offer an hypothesis that could accommodate retractions of derogatory and slurring assertions, and could also fit in a broader account of the pragmatics of conflicts in communication. They say,

This surprising fact [that retraction judgments and falsity judgments can come apart] suggests that retraction is not—or not generally—a way of manifesting a view about the truth value of a claim. We might therefore seek some other kind of theoretical understanding of retraction. One possible approach would be to view retraction as a phenomenon whereby speakers are primarily indicating that they no longer want a conversational common ground incorporating the update associated with a sentence that they previously uttered. On this approach, what is retracted is a certain conversational update; retraction is in part a means of undoing or disowning the context change or update performed by a speech act. (Knobe and Yalcin 2014, p. 16)

Knobe and Yalcin suggest that the experimental data on bare epistemic modals could be explained by their conversational dynamics and not by their truth-conditions. They further suggest that retraction would be required when speakers should signal that they no longer want to be in a given context. For instance, by retracting an assertion of an epistemic modal sentence, the speaker could be indicating that she no longer wishes to be in a context that could be updated with the prejacent of the modal sentence (“there is ice-cream in the freezer”, or “Joe is in Boston”). However, if it is already common knowledge that the context cannot be updated with the prejacent of the modal sentence, it would be pointless to retract. I assume that Knobe and Yalcin would agree that the speaker not have to signal that she does no longer wants to be in a context \(c_n\) when she is not in \(c_n\).

A slightly distinct explanation from the case of epistemic modals would be required for retractions of personal taste assertions. A possibility would be to extend to retractions the idea of metalinguistic negotiations as theorized by Sundell (2011), and Plunkett and Sundell (2013). Metalinguistic negotiations are disputes over pragmatically conveyed information, and not over literal semantic content. A (metalinguistic) dispute over taste may be a dispute over what standard of taste should be commonly adopted by two disputants. In this spirit, a retraction of a past (true) assertion of personal taste might be a disavowal of the previously endorsed standard of taste, rather than a concession of the falsity of that assertion. Mimi may retract an assertion of “Pocoyo is funny!” when she is 9, because she wants it to be clear to everybody that she is not a little girl anymore. But she may refuse to retract when she is 63, because of her endearing memories from childhood, and because it is common knowledge that she does not have the same childish standards of taste.

Derogative discourse poses different challenges to a possible account of retraction. It may be impossible to undo the effects of some seriously offensive remarks, and the best the offender can do might be to atone for the offense. If this is right, retraction may do more than to effect changes in the conversational common ground. A retraction may be, in a very fundamental sense, a recognition of fault and an attempt to make amends, even if the fault itself cannot be undone.

This paper argued that there is no obligation to retract true past epistemic modal or personal taste assertions. There are enough resources available to offer an explanation of the permissibility of retractions in these central cases without the need to revise semantics. We know, moreover, that a constitutive rule for retractions that imposed an obligation to retract on speakers would commit them to either insincerity or irrationality. This undermines the point of assessment-relativism. We are left with contexts of assessment qua mere formal sets of parameters, idle wheels in the semantics.