Abstract
Differences in informativeness regarding truth-conditional and presuppositional content elicit scalar inferences. Many sentences carry not-at-issue, non-presupposed content, e.g. conventional implicatures. This raises the question whether differences in informativeness due to differences in CI content license scalar inferences as well (let us call them anti-conventional implicatures). In previous work, I have shown that certain expressions do license anti-conventional implicatures, and I accounted for the data by postulating a pragmatic principle called Maximize conventional implicatures!, which directs the speaker to use the alternative with the stronger conventional implicatures among those that are formal alternatives and are contextually relevant. Slurs pose a challenge for this view, since they do not seem to license scalar inferences. In this article, I advance an explanation of the data about slurs that is compatible with Maximize conventional implicatures! The gist of the argument hinges upon the notion of linguistic markedness. It has been argued that scalar inferences are blocked if linguistic markedness is not held fixed. I contend that slurs are sociolinguistically marked, and that this is what accounts for the fact that the use of a slur’s neutral counterpart does not license the expected scalar expressive inference.
Keywords:
Scalar Inferences; Conventional Implicatures; Slurs; Epithets; Honorifics
1. Introduction
In conversation, language users frequently take into account not only what has been said, but also alternative sentences that could have been said instead. It has been widely acknowledged that the comparison with unsaid alternative sentences may have a significant impact on the interpretation, the grammaticality or the felicity of an utterance. Hence, it is not surprising that theorizing about alternatives plays a central role in several theoretical frameworks (Gotzner and Romoli, 2022; Nicolae and Sauerland, 2020; Gotzner, 2017, a.o.). The article focuses on a particular alternative-based phenomenon, namely scalar inferences, broadly understood. Very roughly put, scalar inferences arise as a result of the comparison between the sentence uttered and a number of relevant more informative alternatives.53 The most discussed instance of this phenomenon is that of scalar implicatures, an example of which is given in (1):
(1) Some students passed the exam.
⇒¬(all the students passed the exam)
Under the right circumstances, an utterance of (1) elicits the inference that not all the students passed the exam. Roughly put, according to a Gricean approach this inference is based upon the assumption that the speaker is cooperative, in particular that she would utter the most informative sentence she is in position to utter, among those that are relevant in the context. Arguably, in this particular case the hearer compares (1) with (2):
(2) All students passed the exam.
Since (2) is relevant and more informative than (1), the hearer reasons that if the speaker had been in position to use (2) she would have used it. The conclusion is that the speaker was not in position to use (2). This may be merely because (2) is not in the speaker’s belief set, in which case a so-called ignorance inference is derived, i.e. that the speaker neither believes nor disbelieves (2). But sometimes the inference elicited is stronger, namely that the negation of (2) is in the speaker’s belief set. In a Gricean framework, the latter inference, a scalar implicature, is usually derived by introducing the assumption that the speaker is opinionated about the issue (that is, that either (2) or its negation is in the speaker’s belief set).54
In the case of scalar implicatures, the hearer compares the original sentence with its relevant alternatives as regards their at-issue content. But many sentences carry not-at-issue, presuppositional content as well. Most importantly, those sentences can be ordered in terms of informativeness as regards their presuppositional content. Scalar inferences may also be licensed in these cases, as shown by Heim (1991) and many others, by virtue of the comparison between alternatives that differ in informativeness with respect to their presuppositional content. Following Percus (2006), I will call such inferences antipresuppositions.55 By way of illustration, consider (3):
(3) Juan believes that vaccines are harmful.
⇒ ¬(vaccines are harmful)
Under the right circumstances, sentence (3) may license the inference that the speaker believes that vaccines are not harmful. Arguably, the reasoning underpinning this inference is parallel to the one briefly discussed for scalar implicatures, although it works at the level of presuppositional content: the hearer assumes that the speaker would utter the sentence with the most informative presupposition among those relevant alternatives whose presuppositions are common ground. This conversational principle has been labeled Maximize Presupposition! (see fn. 9 below) Arguably, in the case at hand the hearer compares (3) with (4):
(4) Juan knows that vaccines are harmful.
Assume that (4) is contextually relevant, and that ‘know’ encodes the same truth-conditional content as ‘believe’, although it carries a stronger presupposition, to wit, that its complement is true (a factive presupposition). The hearer then reasons that if the speaker had been in position to use (4), she would have used it. If she did not, it must be because (4)’s presupposition is not common ground. This may be merely because such presupposition (vaccines are harmful) is not in the speaker’s belief set. But, assuming that the speaker is opinionated about the issue, a stronger inference may be licensed, to wit, that the negation of the factive presupposition is in the speaker’s belief set.56
Many expressions have been argued to carry non-presupposed, not-at-issue content, either as a conventional implicature, e.g. nominal appositives, supplementary adverbs (‘luckily’, ‘amazingly’), and expressives (e.g. epithets, honorifics and expressive adjectives, see Kaplan, 1999, Potts, 2005, Gutzmann, 2015, 2019 and McCready, 2010), or in the form of illocutionary content, e.g. ‘wow!’ or ‘alas’ (Rett, 2021). So, sentences may also differ in terms of informativeness as regards these kinds of contents. Thus, for example, the sentence ‘That bastard Kaplan got a promotion’ is more informative than the sentence ‘Kaplan got a promotion’ at the level of CI content, albeit not at the level of truth-conditional content. The question then arises whether CI content57 prompts alternative-based scalar inferences parallel to scalar implicatures and antipresuppositions. In other words, one may ask whether sentences are sometimes compared with more informative CI alternatives, namely sentences that encode a stronger CI content, thereby licensing the inference that using the alternative would not have been appropriate in the context. Let us call potential scalar inferences at the level of CI content anti-conventional implicatures (ACIs), by analogy with antipresuppositions.
Following previous work from McCready (2019, pp. 53-54) and Oshima (2023), I have argued, based upon the analysis of a variety of expressions (nominal appositives, supplementary adverbs, epithets and certain honorifics, expressive adjectives and certain emotive markers), that ACIs constitute a bona fide phenomenon (Lo Guercio, ms). Moreover, I advanced the view that conversations are governed by a pragmatic principle called Maximize conventional implicatures! (MCIs!) which, roughly put, directs the speaker to use the strongest CI alternative among those that are formal alternatives (in a sense to be specified below) and contextually relevant. However, such a view is challenged by an apparent counterexample: as we shall see, slurs are prima facie predicted to elicit ACIs, a prediction that is not borne out (see also Lo Guercio, 2021).58 The main goal of this article is to take up this challenge.
In the remainder of the paper, I offer an account for the contrast between slurs and other expressions carrying conventional implicatures as regards ACIs. To put it bluntly, I argue that generally slurs do not license ACIs because they are sociolinguistically marked. As noted by Horn (1984) and Rett (2015, 2020) a.o., markedness considerations may prevent scalar inferences to arise. I contend that this is what is going on in the case of slurs, where the use of a sentence containing the slurs’ neutral counterpart typically does not license the inference that uttering the stronger CI alternative formed by substituting the slur for its neutral counterpart would not have been appropriate in the context. If this is on the right track, it is possible to account for the foregoing fact about slurs in a way that is congenial with the proposed account of ACIs for other expressions, thereby retaining a uniform theory.
Incidentally, note that the foregoing observation about slurs poses a significant challenge for content-based views of slurs in general. Content-based views place the pejorative load of slurs at some level of content, be that truth-conditional content, presuppositional content or conventional implicature, so according to all of these views slurs are more informative than their neutral counterparts at some level. In light of this, one expects the use of a slurs’ neutral counterpart to license a scalar inference of some sort, either a scalar implicature, an antipresupposition or an ACI (depending on the kind of content-based view one prefers), contrary to facts. Thus, if correct, the argument below serves, on the one hand, to defuse a potential argument against the existence of ACIs, and on the other, to address a broad challenge for content-based theories of slurs in general.
The article is structured as follows. In the next section, I briefly review the evidence for the existence of ACIs. For reasons of space, I focus only on epithets and honorifics. In section 3, I discuss the challenge from slurs. In section 4, I show how it is possible to account for the behavior of slurs in a way that is compatible with the view of ACIs presented in section 2. Finally, I draw some conclusions.
2. Anti-conventional implicatures and the problem of slurs
Elsewhere I have argued that ACIs constitute a bona fide phenomenon (see Lo Guercio, ms for details). Here I present only the gist of the argument. I grounded the case on a number of observations concerning nominal appositives, supplementary adverbs, epithets, the Spanish honorifics ‘Don/Doña’ and ‘señor(a)’, expressive adjectives and certain emotive markers. However, for reasons of space, in this section I will focus only on epithets and Spanish honorific ‘Don/Doña’, for which I assume a standard semantics (Potts 2005, McCready 2019):
5) ⟦bastard⟧ = λx. x is a bastard: ⟨ea,tc⟩59
6) ⟦Don/Doña⟧ = λx. sc respects x: ⟨ea,tc⟩60
(where s c is the speaker of the context) Thus, I assume that both epithets and honorifics are pure expressives in Potts’ sense (2005): they take an at-issue argument and output a not-at-issue proposition at the CI dimension.
To see how expressives may trigger ACIs, start by considering an out-of-the-blue utterance of (7) and (8) (where the (a) sentence is the sentence uttered, and the (b) sentence is the CI alternative against which it is compared):
7) a. John arrived late.
⤃¬(John is a bastard)
b. That bastard John arrived late.
8) a. Diego entró.
Diego entered
⤃¬(the speaker respects Diego)
b. Don Diego entró.
HON Diego entered.
The examples above suggest that epithets and Spanish honorific ‘Don/Doña’ do not license ACIs. Thus, for example, the utterance (7a) does not license the inference that the CI content encoded in its alternative (7b) is false, to wit, that John is not a bastard. Mutatis mutandis for (8). Examples (7) and (8), however, contrast with (9) and (10) below, where the relevant inferences are attested:
9) a. John arrived first, then that bastard Pedro arrived.
⇒¬(John is a bastard)
b. That bastard John arrived first.
10) a. Primero entró Donato y después entró Don Pedro.
First entered Donato, and then entered HON Pedro
⇒¬(The speaker respects Donato)
b. Primero entró Don Donato.
First entered HON Donato
The question then arises, how are we to explain this contrast?
I argued that this kind of data can be accounted for by the hypothesis that conversations are governed by the conversational principle Maximize conventional implicatures! (hereinafter, MCIs!), a principle inspired in analogous conversational principles that have been put forward in order to account for scalar implicatures and antipresuppositions:61
11) Maximize conventional implicatures! Do not use sentence φ if there is another sentence φ’ ∈ F(φ) such that:
a. φ’ carries a stronger conventional implicature than φ
b. φ’ is contextually relevant
c. the negation of φ’ does not contradict the negation of the CI content of any other contextually relevant alternative given φ
Before showing how MCIs! accounts for the data, a few clarifications are in order. First, the account is framed within Fox and Katzir’s Structural theory of alternatives (Fox and Katzir 2011, see also Katzir 2007). Very roughly put, the main insight of this view is that the alternatives to be taken into account in order to draw scalar inferences must be at most as structurally complex as the original utterance. The idea is cashed out in Fox and Katzir’s (2011) implementation of the view via the notion of a formal alternative, which they define as below follows: sentence φ’ is a formal alternative to sentence φ if it is possible to transform φ into φ’ by a finite series of deletions, contractions, and replacements of constituents in φ with constituents of the same syntactic category taken from a given substitution source L(φ), which is the union of the lexicon of the language, the set of all subtrees of φ, and the set of contextually relevant (e.g. by virtue of having been mentioned) alternatives in the context (see Fox and Katzir 2011 and Lo Guercio, ms for more details):
Formal alternatives
The set of formal alternatives of sentence S in context c, F(S), is the set of sentences derivable by successive replacement of constituents of S with items in the substitution source of S in c
Substitution source
An item α is in the substitution source of S in c if:
-
is a constituent that is contextually relevant in c (e.g. by virtue of having been mentioned); or
-
α is a sub-constituent of S; or
-
α is in the lexicon.
Crucially, within this view the alternatives constructed through the above mentioned mechanism are considered to be structurally no more complex than the original sentence. Second, we say that sentence φ has a stronger CI content than sentence φ’ iff {c: φ is felicitous in cw} ⊂ {c: φ’ is felicitous in cw}.62 Third, contextually relevant alternatives must obey the following restrictions (where C is the set of relevant alternatives, S is the sentence uttered and F(φ) is the set of formal alternatives to φ):
Closure condition
-
C ⊆ F(S)
-
S ∈ C
-
There is no S0 ∈ F(S)/C s.t. the meaning of S0 is in the Boolean closure of C
Finally, point (c) of MCIs! requires that the alternatives be innocently excludable (see Fox and Katzir 2011) at the level of CI content, that is, that it is consistent to maintain the CI content of the original sentence while negating the CI content of its alternatives.
MCIs! supports the following reasoning: if a cooperative speaker has used φ, then for all contextually relevant formal alternatives φ’ that have a stronger CI content than φ, she could not have used φ’. This may be merely because neither φ’ CI content nor its negation is in the speaker’s belief set, or because the negation of φ’ CI content is in the speaker’s belief set, in which case, assuming that the speaker is opinionated about the issue, one may infer that the negation of φ’ CI content is in the speaker’s belief set. The former inference is akin to the ignorance inferences mentioned at the introduction; the latter is a scalar inference. I will be concerned only with scalar inferences below.
Now we are in position to account for the observations above. Examples (7) and (8) show that out-of-the-blue uses of epithets and the Spanish honorific ‘Don/Doña’ do not license ACIs. Crucially, MCIs! predicts ACIs to be licensed only insofar as the alternative sentences under consideration are formal alternatives to the original one. Since in this case the b-sentences are not formal alternatives in the required sense, ACIs are predicted not to arise, in conformity with our intuitions. To see the point, consider (7) for example. The structure ‘That bastard John’ is not in the substitution source, since it is not a lexical item, is not contextually relevant (it has not been mentioned, and let us stipulate, is not salient in the context), and is not a subtree of the original sentence. In consequence, (7b), the sentence that results from substituting ‘that bastard John’ for ‘John’ in (7a), is not a formal alternative to (7a) (similar considerations apply to example (8)), hence the ACI is predicted to be absent.
By contrast, in examples (9) and (10), despite appearances, the b-sentences are formal alternatives to the original sentence, so MCIs! correctly predicts the corresponding ACIs to arise. To see the point, take (9a). In this case, the epithet occurs in the second conjunct, so it becomes part of the substitution source (it is contextually relevant by virtue of having been mentioned), and is thereby available for constructing the alternative in (9b) by a finite series of deletions, contractions and replacements.63 Then we can derive the expected inferences by appealing to MCIs!: if the speaker is cooperative and has used (9a) in spite of having alternative (9b) available, which is contextually relevant, has a stronger CI content and is a formal alternative to (9a), then assuming that the speaker is opinionated about the issue, the hearer is licensed to infer that John is not a bastard is not in the speaker’s belief set. Mutatis mutandis for ‘Don/Doña’ in (10).
To sum up, there is a contrast between (7)-(8), on the one hand, and (9)-(10) on the other, and MCIs! correctly predicts this contrast. This supports the thesis that the inferences licensed by the latter group of examples constitute ACIs. In other words, the previous discussion indicates that differences in informativeness contributed by conventionally implied content, like epithets and honorifics, can license scalar alternative-based inferences.64
Before moving on, however, it is worth pausing in order to put the claims made above into perspective. The foregoing discussion inquires into the inferences that interpreters may draw based upon certain Gricean considerations typically thought to govern cooperative conversations, such as informativeness, relevance and complexity. Admittedly, however, conversations may be affected by other kinds of considerations, such as social protocols, rules of etiquette, and so on. Moreover, these considerations may affect the kind of inferences language users are compelled or licensed to draw in specific contexts. For example, there are highly regulated contexts in which not using an honorific may license the inference that the speaker does not respect the addressee, e.g. if someone calls a judge by her name instead of employing the expected honorific formula ‘your honor’. To be sure, some of these cases could in principle be accommodated by the proposal: in such cases the honorific formula is highly salient in the context (it is the formula prescribed by a social/legal norm) hence it may be argued to be part of the substitution source, thus to be available in order to construct the required alternative, so that it counts as a formal alternative. To be sure, a full account of ACIs would require taking into account these issues, as well as inquiring into the interplay between principles of cooperative conversation and other kinds of social norms. One may think about the issue in terms of conflicting conversational rules (more on this below), or alternatively, by taking rules of etiquette and other sociopragmatic features as partially constraining which alternatives are contextually relevant (that is, which alternatives are a ‘live option’ for the speaker).65 For reasons of space, and because addressing this problem would take me far from the main topic of the article, namely slurs, I leave it for future research.
3. The challenge from slurs
As shown in the previous section, MCIs! correctly predicts the behavior of epithets and Spanish honorific ‘Don/Doña’ in different contexts. However, not all expressions carrying CI content can be easily handled by this approach. As we shall see, slurs pose a prima facie challenge.
Before jumping into the discussion of slurs, however, it is worth underscoring certain complications upfront. Firstly, there is no consensus concerning the correct analysis of slurs. The literature can be divided between content-based theories, which contend that slurs encode pejoration at some level of conventional meaning, and non-content-based theories, which maintain that pejoration is achieved by pragmatic means.66 As we shall see, slurs challenge MCIs! only under the assumption that they encode pejoration at the level of their conventionally implied content. Secondly, slurs are taboo or socially sanctioned terms in a broad set of contexts. In those contexts, there is a strong preference against their use in general, even in mentions, echoic and ironic uses. This may function as a confounding factor, since speakers often have an excuse not to use a slur, beside considerations of informativeness, relevance or complexity. Thirdly, even among those who defend a content-based theory of slurs there is no consensus concerning whether slurs require the existence of a so-called neutral counterpart, to wit, an expression that possesses the same truth-conditional content but lacks pejorative content, an issue that will prove important for the upcoming discussion.67
In what follows, I focus on a relatively weak slur, ‘sudaca’, a pejorative for South Americans commonly employed both in Spain and South America.68 As regards slurs’ semantics, I assume a content-based view. As noted earlier, the choice is not innocuous; if one assumes a pragmatic theory, slurs pose no challenge for MCIs! More specifically, I assume that slurs’ pejorative content is conventionally implied, as is strongly suggested among other things by their hyper-projective behavior. Concretely, I adopt the following lexical entry for ‘sudaca’ in line with McCready (2010):
12) ⟦sudaca⟧w = λx. x is South American in w ⧫
BAD(∩South Americans): ⟨e a ,t a ⟩ × t s69
Finally, I assume that many slurs, and in particular ‘sudaca’, have neutral counterparts. Albeit not completely undisputed, these assumptions are not especially controversial compared with other views about slurs. Be that as it may, it is worth making them explicit, so as to put the following discussion into perspective.
Slurs constitute a challenge for MCIs! To see the point, consider ‘sudaca’. Firstly, both ‘sudaca’ and its neutral alternative ‘sudamericano’ (South American) are lexical items, so both belong to the substitution source from which alternatives are to be constructed. In consequence, substituting one for the other in a sentence gives a formal alternative as a result (that is, the replacement does not add structural complexity). Secondly, ‘sudaca’ is expressively more informative than ‘sudamericano’:
13) {c: sudaca is felicitous in c w } ⊂ {c: sudamericano is felicitous in c w }
This is because, all other things being equal, the former restricts contexts of felicitous use to those where the speaker holds a negative attitude towards South Americans, whereas the latter encodes no restriction at all in this regard. In light of this, MCIs! predicts that an utterance of the sentence formed by substituting ‘sudamericano’ for ‘sudaca’ should license (assuming that the speaker is opinionated about the issue) the inference that the speaker believes that South Americans are not despicable. The reasoning goes as follows. If the speaker is cooperative and has used ‘sudamericano’ in a context where using ‘sudaca’, which is a contextually relevant formal alternative with a stronger CI content, is an open possibility, this must be because she could not have used the latter. Assuming that the speaker is opinionated about the issue, the hearer is licensed by MCIs! to infer that the content South Americans are not despicable is in the speaker’s belief set. However, this prediction is not borne out:
14) a. Juan es sudamericano.
Juan is South.American
⤃¬(South Americans are despicable)
b. Juan es sudaca.
Juan is South.American(pejorative)
This represents a challenge for the account of ACIs discussed in section 2, and raises the question why slurs contrast with epithets and honorifics in this regard.
I presented the problem against the backdrop of both MCIs! and a conventional implicature view of slurs, but as mentioned earlier the problem arises under any content-based view. On the one hand, I have argued elsewhere (Lo Guercio, 2021) that the problem arises for presuppositional views of slurs. Under that approach, ‘sudaca’ has a stronger presupposition than its neutral counterpart ‘South American’, so using the latter is predicted to trigger a similar kind of inference by the conversational principle Maximize Presuppositions!, which directs the speaker to use the sentence with the strongest presuppositional content among those relevant alternatives whose presuppositions are satisfied in the context (see fn. 9). Likewise, an analogous kind of problem could be formulated by assuming a truth-conditional account of pejoration (Hom 2008). Such a view is also committed to the idea that slurs are more informative than their neutral counterparts, although at the level of truth-conditional content. As a consequence, the use of the neutral counterpart is predicted to license a scalar implicature by a suitable version of the Conversational principle (see fn. 9), which directs the speaker to use the most informative alternative among those that are weakly assertable in the context.
4. Meeting the challenge
In this section, I contend that slurs’ behavior regarding ACIs can be accounted for in a way that is compatible with MCIs! As we shall see, the gist of the argument lies in the concept of linguistic markedness. Before advancing the argument, however, it is worth briefly considering two closely connected alternative lines of reasoning that do not work (see Lo Guercio, 2021 for details).
4.1 Two failed attempts
First, one may attempt to account for the absence of ACIs by noting that slurs are taboo words (Anderson and Lepore 2013). According to this view, there is a very strong social norm that prohibits any occurrence of a slur.70 The idea is that this social norm gets in the way of pragmatic inferences because it creates a conflict between two different kinds of norms: on the one hand, the default conversational principles governing cooperative conversation, and on the other hand, the social rule that prohibits the use of slurs. One may argue that ACIs are absent because the latter takes precedence over the former, so that even if using the slur is an open possibility in the context, agents would rather flout MCIs! by uttering the slur's neutral counterpart than breake the taboo. Since the alleged taboo dimension of slurs is typically common knowledge, there is no shared presumption that speakers will use the expressively strongest alternative in this case.
If this is on the right track, there is a principled explanation for the absence of ACIs which is not in tension with the proposed view based on MCIs! However, such a strategy is controversial. On the one hand, there are reasons to reject Anderson and Lepore’s approach to slurs. As Nunberg (2018) points out, arguably not all slurs are taboo in this way, and even those that are taboo have not always been so. Moreover, their view has been criticized for reversing the order of the explanation: it is the pejorative content of slurs that makes them taboo, not the other way around. On the other hand, even granting Anderson and Lepore’s view, the strategy is undermined by the consideration of mixed sentences such as (15) below:
15) This country has been taken over by Chinks and South Americans. They steal our jobs and commit crimes, and the government does nothing about it.
In (15) the use of ‘chinks’ makes it clear that the speaker is willing to break the taboo about slurs. Hence, the prevalence of a social norm prohibiting slurs over MCIs! cannot be what explains the speaker’s avoidance of ‘sudaca’ immediately after. Still, (15) does not seem to carry an ACI to the effect that the South Americans are not despicable.
The second line of reasoning is closely connected to the first one. Let us grant that slurs are not prohibited words in the strong sense favored by Anderson and Lepore. Even so, by using a slur a speaker surely risks damaging her own reputation, if the slur is uttered in the presence of someone who does not share her derogatory attitudes. That is, using a slur may come at a social cost. In light of this, one may argue that avoiding a possible social sanction typically works as an incentive for bigoted speakers to avoid such expressions, even if this means flouting the default principles that govern cooperative conversation, such as MCIs! Since this fact is arguably common knowledge across a broad range of contexts, plausibly interpreters do not assume that in using the neutral counterpart the speaker’s utterance was governed by MCIs!, and so they do not draw the predicted ACI. Put differently, the idea is that there is no shared presumption that speakers are fully cooperative when it comes to slurs. Again, if this is correct, there is a principled explanation for the systematic absence of ACIs following the use of the slur’s neutral counterpart.
This strategy can also be criticized, however, since arguably not-fully-cooperative contexts do not necessarily block pragmatic inferences, they just make them unsafe. As argued by Asher and Lascarides (2013), in not-fully cooperative contexts participants are able to draw the same kind of inferences they would draw in fully cooperative ones, insofar as they are rhetorically cooperative, that is, insofar as they act as if they were Gricean-cooperative, even if they know they are not. Now, a bigoted speaker who avoids using a slur in order to circumvent the social costs that could ensue is not Gricean-cooperative concerning her own prejudices, but arguably she must appear as Gricean-cooperative to the audience, for if it was clear in the context that the speaker avoided the slur with the intention of hiding her true attitudes towards the relevant group, she would probably be censured anyway. In other words, she is being rhetorically cooperative. But rhetorical cooperativity is all it takes to generate the relevant inferences. So, even if it is common knowledge that the speakers' utterance may not be governed by MCIs!, insofar as she can be taken to be rhetorically cooperative the interpreter should be able to compute the corresponding ACI (that the speaker does not hold a negative attitude towards the target group), even if she has reasons to be skeptical about the speaker’s true attitudes.
Since these strategies to meet the challenge have problems, it is worth exploring an alternative. I develop such an alternative in the following subsection.
4.2 Slurs are marked expressions
The key to explain away the apparent tension between the facts discussed in section 2 and those presented in section 3, I shall argue, is the notion of linguistic markedness. According to Rett (2020), the notion of markedness in linguistics is robust and captures the fact that “natural language seems to organize members of pairs or n-tuples of linguistic elements in terms of primacy.” Although there is no unified theory of markedness, we can say that broadly speaking markedness is a measure of relative linguistic effort or cost. More specifically, it has been noted (see Bybee 2011 and references therein) that being the marked member of an n-tuple of linguistic expressions correlates with a number of properties like higher structural or morphological complexity, less familiarity, lower frequency, narrower internal distribution, or late acquisition, among others.
The role of linguistic markedness in pragmatic reasoning has been broadly discussed by Horn (1984, 1989) and Rett (2015, 2020), a.o., mostly in relation to scalar and manner implicatures. Horn reduces Grice's four Maxims to two principles, the Q-principle and the R-principle:
The Q Principle (Hearer-based) (Horn 1984: 13):
a. Make your contribution sufficient (cf. Quantity1)71
b. Say as much as you can (given R)
(Lower-bounding principle, inducing upper-bounded implicata)
The R Principle (Speaker-based) (Horn 1984: 13):
a. Make your contribution necessary (cf. Relation, Quantity2,72 Manner)
b. Say no more than you must (given Q)
(Upper-bounding principle, inducing lower-bounded implicata)
In this view, scalar inferences are generated via the Q-principle, and work through informativeness-based competition. Roughly put, the Q-principle directs the speaker to use the most informative sentence she is in position to use, modulo relevance. Thus, the principle leads to a reasoning as the one we have been discussing in the previous sections. On the other hand, at least certain manner implicatures are generated via the R-principle. More specifically, Horn focuses on manner implicatures driven by markedness-based competition, that is, cases where the use of a marked expression generates certain contextual effects due to its contrast with a contextually salient unmarked expression. Roughly put, the markedness-based reasoning licensed by the R-principle goes as follows: assuming that the speaker is cooperative, if she has used marked sentence φ, despite having at her disposal an alternative sentence φ' which is relevant, equally informative and less marked than φ, it must be because using φ was necessary in order to convey what she wanted to convey, and this in turn must be because she meant something different from what she literally said. As a result, according to Horn the use of unmarked expressions tends to be associated with unmarked (`normal' or stereotypical) situations, whereas the use of marked expressions tends to be associated with marked (atypical) situations (see Atlas 1981 and Levinson 2000 for other proposals in the same spirit).
Most importantly, it has been noted by Rett (2015, p. 99) that informativeness-based quantity inferences and markedness-based manner inferences are closely connected. As their name indicates, the former work by comparing alternatives ordered in terms of informativeness, but crucially they are licensed only insofar as markedness is held fixed (note that this is in line with the main insight behind the structural theory of alternatives: greater structural complexity makes a sentence comparatively more marked, and as a consequence scalar inferences are blocked). Conversely, markedness-based inferences work by comparing alternatives in terms of markedness, but crucially they are licensed only insofar as informativeness is held fixed. Thus, we have the following picture:
Quantity implicatures
-
involve Q-scales, which hold fixed markedness (i.e.morphological or syntactic complexity) and order elements with respect to informativeness
-
are calculated relative to the Q-Principle
Manner implicatures
-
involve M scales, which hold fixed informativeness and order elements with respect to markedness (i.e.~morphological or syntactic complexity)
-
are calculated relative to the R-Principle (when a marked form is used) or the Q-Principle (when an unmarked form is used)
(Rett 2015, p. 99)
Like scalar implicatures and antipresuppositions, ACIs constitute a special kind of informativeness-based quantity inference. Hence, it is expected that they emerge only insofar as markedness is held fixed. We have seen that ACIs emerge only if certain formal alternatives are contextually relevant, and we saw that formal alternatives must be at most as structurally complex as the original sentence. Put differently, this means that ACIs arise only if structural complexity is held fixed. However, structural complexity is only one way in which an expression may be marked; arguably an expression can be marked in virtue of possessing other properties as well, like greater morphological complexity or a narrower internal distribution, among others.73 With this in mind, we may account for the absence of ACIs by arguing that slurs, although as structurally complex as their corresponding neutral counterparts, are nevertheless more marked. As a result, they are not considered as legitimate alternatives for computing ACIs. In other words, we may say that slurring alternatives are not formal alternatives to the original sentence, and so must be ignored for the derivation of ACIs. This would require advancing a suitably modified version of MCIs! where the broader notion of markedness (understood as encompassing structural complexity as well as other properties) substitutes for structural complexity. But are slurs marked?
As argued by Nunberg (2018), theories of slurs that include the existence of a neutral counterpart as a defining, or at least a characteristic property of slurs (see Bolinger 2017, Croom 2013, Camp 2013, Bach 2012, Bianchi 2014, Jeshion 2013, Whiting 2013, Hornsby 2001 and Williamson 2009, among others) implicitly assume that slurs are marked in some sense. For these views, the existence of a neutral counterpart plays a fundamental role for the explanation of slurs' pejorative effects: it is at least in part the contrast between the slur and its expressively unloaded alternative that makes the slur pejorative. In Nunberg's view markedness does all the explanatory heavy lifting: he maintains that the pejorative import of slurs emerges as a kind of routinized R-inference based on the comparison between the marked expression, namely the slur, and its unmarked neutral counterpart.74 It is worth noting, however, that one may grant that slurs are marked, but assign markedness a different theoretical role. In effect, broadly speaking one could understand content-based views as maintaining that what Nunberg characterizes as a routinized conversational manner implicature is in fact fully conventionalized, that is, part of the meaning of the expression. Still, the kind of markedness Nunberg identifies (more on this in a moment) is present and may even play a role in accounting for certain phenomena connected to slurs, e.g. reclamation (see Lo Guercio y Carranza 2024, Burnett 2020).75 Thus, the idea that slurs are marked expressions compared to their neutral counterparts is compatible with, and in fact seems to play an important explanatory role in, many theories of slurs, even content-based theories.
Now, in virtue of what are slurs marked? According to Nunberg, slurs are sociolinguistically marked; more specifically, a slur is not marked because it is structurally or morphologically more complex than its neutral counterpart, but because it is associated with members of certain social collective who share a common interest in having that specific word to refer to the target group (Nunberg 2018, 264-276). In terms of Agha (1998), we may say that slurs are personified, that is, inherently associated with certain types of speakers and speech communities through a process of iconic motivation. Neutral counterparts, by contrast, in addition to being semantically neutral, in the sense of not encoding conventional pejorative content, are sociolinguistically neutral, in the sense of not serving as pointers of any special social identity. Thus, while ‘sudamericano’ is the default Spanish term to refer to South Americans in any context, and it is not iconically associated to any particular group of language users, social identity, or specific speech community, ‘sudaca’ is sociolinguistically marked as the form typically employed by xenophobes, hence iconically associated with them.76 If this is on the right track, we can account for the absence of an ACI in an example like (14) by saying that although ‘sudaca’ is truth-conditionally equivalent and as structurally equally complex as its neutral counterpart ‘sudamericano’, it is also more marked in the above sense. For this reason, a suitably modified version of MCIs! where a broader notion of markedness substitutes for structural complexity when it comes to constraining formal alternatives stipulates that the alternative slurring sentence must be ignored (by virtue of not being a formal alternative), and thus predicts the ACI not to arise, in accordance with our judgments.
It is worth noting that slurs are not alone in this: Nunberg’s idea, as well as the present proposal, resorts to independently motivated ideas broadly discussed in sociolinguistics. Sociolinguists have long noted that speakers both make and notice stylistic distinctions in speech (e.g. differences in phonological features, grammatical patterns and equally informative lexical items) and exploit them both to present themselves with a certain social persona and to locate other speakers in the social landscape (Eckert 2000, 2006, 2008).77 Thus, for example, Campbell-Kibler’s (2007) work on the pronunciation of -ING in English shows that the velar variant is associated with education, intelligence and articulateness (or, alternatively, with pretentiousness and uptightness), while the apical variant is associated with lack of education and inarticulateness (or, alternatively, with relaxed, easygoing and unpretentious speakers, see also Eckert, 2000). Likewise, both grammatical patterns and lexical choices may be associated with certain types of language users, and thereby be exploited by speakers in order to place themselves and others in the social landscape. As an example of the former, consider the case of so-called capicúa sentences in Río de la Plata Spanish (constructed by iterating the phonological word containing the verb of the sentence, e.g. ‘¡Vino Ramón, vino!’ -lit. ‘came Ramón, came’), which are typically associated with uneducated speakers from poor neighborhoods. Finally, the word ‘compañero’ (lit. ‘partner’ or ‘buddy’) as it is used in Río de la Plata Spanish offers an example of the way in which certain lexical items may become associated with a specific social persona, and subsequently employed by language users as a pointer of their social location. In Argentina, the word ‘compañero’ has undergone a process in virtue of the place it has occupied (and still occupies) within Peronist rhetoric78 whereby it has become stereotypically associated with a kind of speaker (the Peronist speaker) and thereby with a particular social identity (at least when the word occurs in political contexts). As such, speakers may use the term in order to signal their own political affiliation, as well as to mark an opposition with an alternative, anti-peronist identity.79
Let us take stock of the argument. ACIs are informativeness-based inferences. As such, they may be blocked if markedness across relevant alternatives is not held fixed. Arguably, slurs are sociolinguistically marked. Hence, ACIs are blocked in the case of slurs because speakers ignore slurring alternatives when they compute ACIs, due to the former being marked. Put differently, the idea is that in contexts where using a slur is an open possibility one cannot assume that the speaker’s conversational moves are governed by MCIs!, since she may avoid the slurring alternative not because she rejects the pejorative content of the slur but because she wants to eschew identifying herself with the slur’s associated social identity. If this is on the right track, the proposed explanation of ACIs for the case of epithets and honorific ‘Don/Doña’, based on MCIs!, can be maintained.
Before concluding, it is worth discussing the following case discussed by Cepollaro (2017, p. 134), which appears to be a counterexample to what I have defended so far:80
…if a KKK member talks about ‘African-American people’ in a context where [the N-word] is the standard term for black people, she is choosing to flout the contextual expectations, thus possibly signaling her endorsement of non-racist contents.
In effect, Cepollaro’s example suggests that sometimes slurs do license ACIs. One can resist the example, however, on the grounds that it includes a confounding factor, to wit, arguably ‘African-American people’ is not neutral in the required sense, but it is sociolinguistically marked as the socially and politically preferred/appropriate option, specially for white speakers. Thus, this is not a case where there is contrast between a marked and an unmarked term, but between two marked terms.81
An analogy with the pair ‘compañero/camarada’ may serve to highlight the point. If, in the context of a peronist political meeting, someone chooses to use the term ‘camarada’ instead of the expected expression ‘compañero’, she will certainly license the audience to infer that she does not share a peronist outlook. But to be sure, this is not a scalar inference: the term ‘camarada’ is marked as the default option for people with a communist political affiliation, hence in using such a word the speaker conveys that she shares this social identity. The inference that the speaker is not a peronist arises because communism and peronism are incompatible views about the world, plus the assumption that the speaker’s belief set is consistent, broadly speaking. An analogous explanation can be provided in order to account for Cepollaro’s example: ‘African-American’ typically signals the speaker’s affiliation with a social identity which is sensitive to the demands for equal rights by African-Americans. It is on account of this inference that the audience is able to further conclude that she does not share her partners’ racist attitudes.
One may press the point by arguing that similar examples could be constructed where, unlike ‘African-American’, the term employed is not marked in the relevant sense. Thus, one may imagine a context in which ‘sudamericano’, which is the default term for referring to South Americans (and so is not marked in the relevant sense), is used. Plausibly, in such a case by opting out of the default and employing ‘sudamericano’ instead of its slurring alternative ‘sudaca’ the speaker manages to signal that she does not hold negative attitudes against South Americans. Admittedly, these kinds of examples are possible.
However, an alternative analysis is available, according to which these kinds of inferences are not ACIs. More specifically, one may analyze such inferences as follows. By using a slur, the speaker makes at least two moves: first, it advances a derogatory content in virtue of the slurs’s conventional meaning, namely that she holds a negative attitude towards South Americans; second, she signals her affiliation or identification with a certain social identity, by virtue of the slur’s so-called social meaning. Now, the relevant ACI is blocked in this case because the slurring alternative is marked, as I argued before. However, one still is able to contrast the utterance with its slurring alternative in terms of their social meaning. More specifically, in a context in which using the sociolinguistically marked slur is the default option, using instead its sociolinguistically neutral counterpart conveys that the speaker does not identify herself with the relevant group of speakers. Then, based upon that initial inference, the audience may further infer that the speaker does not hold negative attitudes against South Americans. As said earlier, this inference exploits the comparison between the utterance and its slurring alternative, but it is nevertheless not an ACI, since the contrast that is being exploited for the generation of the inference is that between a sociolinguistically marked term and a sociolinguistically neutral term, and not that between an expressively weaker and an expressively stronger term. Thus, we can disregard such apparent counterexamples on principled grounds.
5. Conclusion
In this article, I discussed data showing that slurs do not license ACIs, unlike other kinds of expressives like epithets and honorifics. This represents a challenge to the view that conversations are governed by the conversational principle MCIs!, which directs the speaker to use the sentence with the stronger Ci content among those formal alternatives that are contextually relevant. In order to meet the challenge, I argued that i) scalar inferences are typically blocked if the required alternatives are marked, and ii) slurs are sociolinguistically marked. If this is on the right track, the potential problem from slurs can be explained away on principle grounds in a way that fits comfortably with MCIs! Finally, it is worth mentioning that the present view offers content-based views a way out of a potential problem. It was mentioned in the introduction that all content-based theories of slurs face a challenge regarding alternative-based scalar inferences: if slurs encode pejorative content as part of their conventional meaning at any level (truth-conditional content, presuppositional content or conventionally implicated content), then the use of a sentence containing an occurrence of the slur’s neutral counterpart should trigger a scalar inference of some kind, be that an antipresupposition, a scalar implicature, or an ACI. In addition to undermining a potential problem for MCIs!, the previous discussion offers a principled way of meeting the aforementioned challenge: the reason why the differences in informativeness between slurs and their neutral counterparts do not elicit scalar inferences of any kind is that the mechanism by which scalar inferences are computed is sensitive to considerations of markedness and slurs are sociolinguistically marked.
6. References
- Agha, A. (1998). Stereotypes and registers of honorific language. Language in society, 27(2), 151-193.
-
Anderson, Luvell, and Ernie Lepore. 2013. Slurring words. Nous 47(1): 25-48. https://doi.org/10.1111/j.1468-0068.2010.00820.x
» https://doi.org/10.1111/j.1468-0068.2010.00820.x - Ashwell, L. (2016). Gendered slurs. Social Theory and Practice, 42(2): 228-239.
- Asher, N. and Lascarides, A. (2013). Strategic conversation. Semantics and pragmatics, 6: 1-62.
- Atlas, J. D., and Levinson, S. C. (1981). It-clefts, informativeness and logical form: Radical pragmatics. In Peter Cole (ed.), Radical pragmatics (pp. 1-62). Academic Press.
- Bach, Kent. 2014. “Mean and Nasty Talk: On the Semantics and Pragmatics of Slurs.” Pacific APA, San Diego, April 19, 2014.
- Bianchi, C. (2014). Slurs and appropriation: An echoic account. Journal of Pragmatics, 66, 35-44.
-
Bolinger, Reneé Jorgensen. 2017. The pragmatics of slurs. Nous 51(3): 439-462. https://doi.org/10.1111/nous.12090
» https://doi.org/10.1111/nous.12090 - Breheny, R., Klinedinst, N., Romoli, J., and Sudo, Y. (2018). The symmetry problem: current theories and prospects. Natural Language Semantics, 26: 85-110.
- Burnett, H. (2020). A persona-based semantics for slurs. Grazer philosophische studien, 97(1), 31-62.
- Bybee, Joan. 2011. Markedness: Iconicity, economy, and frequency. In J. J. Song (ed.), Oxford Handbook of Linguistic Typology, 131-147. Oxford: Oxford University Press.
- Camp, E. (2018). A dual act analysis of slurs. David Sosa (ed.) Bad words: Philosophical perspectives on slurs, pp. 29-59. Oxford, Oxford University Press.
- Caso, R. y Lo Guercio, N. (2015) What bigots do say: a reply to DiFranco Thought 5, 265-274.
-
Cepollaro, Bianca, and Isidora Stojanovic. 2016. Hybrid evaluatives: In defense of a presuppositional account. Grazer Philosophische Studien 93(3): 458-488. https://doi.org/10.1163/18756735-09303007
» https://doi.org/10.1163/18756735-09303007 - Cepollaro, B. (2017). The Semantics and Pragmatics of slurs and thick terms. Ph.D. Dissertation.
- Chemla, E. (2008). An epistemic step for anti-presuppositions. Journal of Semantics, 25(2):141-173.
- Croom, A. M. (2013). How to do things with slurs: Studies in the way of derogatory words. Language & communication, 33(3), 177-204.
- Eckert, P. 2000. Linguistic Variation as Social Practice: The Linguistic Construction of Identity in Belten High. Oxford: Blackwell.
- Eckert, P. 2006. “Communities of practice”. In: J. Mey (ed). Concise encyclopedia of pragmatics, 2nd edition. Oxford: Elsevier, 109-112.
- Eckert, P. 2008. “Variation and the indexical field.” Journal of sociolinguistics, 12(4), 453-476.
- Fox, D. (2007). Free choice and the theory of scalar implicatures. In Sauerland, U. and Staveva, P., editors, Presupposition and implicature in compositional semantics, pages 71-120. Palgrave, New York.
- Fox, D. and Katzir, R. (2011). On the characterization of alternatives. Natural language semantics, 19:87-107.
- Gotzner, N. (2017). Alternative sets in language processing: How focus alternatives are represented in the mind. Springer.
- Gotzner, N. and Romoli, J. (2022). Meaning and alternatives. Annual Review of Linguistics, 8:213-234.
- Gutzmann, D. (2015). Use-conditional meaning: Studies in multidimensional semantics, volume 6. Oxford University Press, Oxford.
- Gutzmann, D. (2019). The Grammar of Expressivity. Oxford University Press.
- Heim, I. (1991). Artikel und definitheit. Semantik: ein internationales Handbuchder zeitgennössischen Forschung, pages 487-535.
-
Hom, Christopher. 2008. The semantics of racial epithets. The Journal of Philosophy 105(8): 416-440. https://www.jstor.org/stable/20620116
» https://www.jstor.org/stable/20620116 - Horn, L. (1984). Towards a new taxonomy for pragmatic inference: Q-and R-based implicature. Meaning, form and use in context: 11-42. Georgetown University Press.
- Horn, L. (1989). A Natural History of Negation. Chicago: Chicago University Press.
- Hornsby, Jennifer. 2001. “Meaning and Uselessness: How to Think about Derogatory Words.” Midwest Studies In Philosophy 25(1): 128-41.
- Irvine, J. T., & Gal, S. 2000. “Language ideology and linguistic differentiation.” In P. V. Kroskrity, (Ed.), Regimes of language: Ideologies, polities, and identities. Santa Fe: School of American Research Press, 35-84.
- Jeshion, R. (2013). Expressivism and the Offensiveness of Slurs. Philosophical Perspectives, 27(1), 231-259.
- Kaplan, D. (1999). The meaning of ouch and oops. In Cornell conference on context dependency.
- Campbell-Kibler, K. (2007). Accent,(ING), and the social logic of listener perceptions. American speech, 82(1), 32-64.
- Magri, G. (2009). A theory of individual-level predicates based on blind mandatory scalar implicatures. Natural Language Semantics , 17(3):245-297.
- Marques, T., and García-Carpintero, M. (2020). Really expressive presuppositions and how to block them. Grazer Philosophische Studien , 97(1), 138-158.
- McCready, E. S. (2010). Varieties of conventional implicature. Semantics and Pragmatics, 3:8-1.
- McCready, E. (2019). The semantics and pragmatics of honorification: Register and social meaning (Vol. 11). Oxford University Press, USA.
- Neufeld, E. (2019). An essentialist theory of the meaning of slurs. Philosophers’ Imprint, 19(35).
- Nicolae, A. C. and Sauerland, U. (2020). Quantity implicatures. In The Wiley Blackwell Companion to Semantics, pages 1-28. Wiley Online Library.
- Nunberg, Geoffrey. 2018. The social life of slurs. In New work on speech acts, eds. Fogal, Daniel, Daniel Harris, and Matt Moss, 237-293. New York: Oxford University Press.
- Levinson, Stephen. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge, MA: MIT Press.
-
Lo Guercio, N. (2021) Slurs and antipresuppositions. Natural Language Semantics 29, 377-400. https://doi.org/10.1007/s11050-021-09178-y
» https://doi.org/10.1007/s11050-021-09178-y - Lo Guercio, N. (Manuscript). MAximize Conventional impliocatures!
-
Lo Guercio, N., & Carranza, F. 2024. "A metapragmatic stereotype-based account of reclamation." Analytic Philosophy, 00, 1-22. https://doi.org/10.1111/phib.12345
» https://doi.org/10.1111/phib.12345 - Orlando, E. and Saab, A. (2020). Slurs, stereotypes and insults. Acta Analytica, 35(4):1-23.
- Oshima, D. Y. (2023). The semantics and sociopragmatics of the japanese honorific titles san, kun, and chan: some focal points of variation. Journal of East Asian Linguistics, 32(2):169-200.
- Percus, O. (2006). Antipresuppositions. In Theoretical andEmpirical Studies of Reference and Anaphora: Toward the establishment of generative grammar as an empirical science. [Report of the Grant-in-Aid for Scientific Research (B)], page 52-73. Japan Society for the Promotion of Science.
- Potts, C. (2005). The logic of conventional implicatures, volume 7. Oxford University Press.
- Rett, J. (2015). The semantics of evaluativity. OUP Oxford.
- Rett, J. (2020). Manner implicatures and how to spot them. International Review of Pragmatics, 12(1):44-79.
- Rett, J. (2021). The semantics of emotive markers and other illocutionary content. Journal of Semantics, 38(2): 305-340.
- Sauerland, U. (2008). Implicated presuppositions. In Steube, A., editor, Sentences and context: Language, context, and cognition, pages 581-600. Mouton de Gruyter, Berlin.
-
Singh, R. 2009. ‘maximize presupposition!’ and informationally encapsulated implicatures. In Proceedings of Sinn und Bedeutung, 13, 513-526. https://doi.org/10.18148/sub/2009.v13i2.569
» https://doi.org/10.18148/sub/2009.v13i2.569 - Schlenker, P. (2007). Expressive presuppositions. Theoretical Linguistics, 33(2):237-245.
- Schlenker, P. (2012). Maximize presupposition and gricean reasoning. Natural language semantics , 20(4):391-429.
- Swanson, E. (2010). Structurally defined alternatives and lexicalizations of xor. Linguistics and Philosophy, 33:31-36.
- Whiting, D. (2013). It's not what you said, it's the way you said it: Slurs and conventional implicatures. Analytic Philosophy, 54(3), 364-377.
- Williamson, T. (2009). Reference, inference, and the semantics of pejoratives. In J. Almog & P. Leonardi (Eds.), The philosophy of david kaplan (pp. 137-158). Oxford: Oxford University Press .
