Provocative Insinuations as Hate Speech: Argumentative Functions of Mentioning Ethnicity in Headlines

We explore a particular type of propagandistic message, which we call “provocative insinuation”. For example: ‘Iraqi refugee is convicted in Germany of raping and murdering teenage girl’. Although this sentence seems to merely report a fact, it also conveys a potentially hateful message about Iraqi refugees. We look at the argumentative roles that these utterances play in public discourse. Specifically, we argue that they implicitly address the question of the integration of refugees and migrants, and in fact aim to tilt the audience against these groups by strongly inviting hearers to make generalisations based on “strik‑ ing” properties. We examine different strategies to counteract the conveyed hateful message.


Introduction
German Chancellor Angela Merkel decided in 2015 to open the country's borders to migrants fleeing war and turmoil in Syria, Iraq, and Afghanistan. The peak of asylum applications in the country was in 2016 (722.370 applications). This inflow, and the backlash against it via the growing popularity of conservative and xenophobic discourse, sparked a major public debate on refugee integration throughout 2018-2019.
In 2019, the internationally well-renowned journal New York Times (NYT) reported on the conviction of Ali Bashar Ahmad Zebari, a 21-old asylum seeker from Iraq-Kurdistan, for the May 2018 rape and murder of German-born Susanna Maria Feldmann, who was 14 at the time. The crime was among various high-profile crimes by asylum seekers, which spurred public debate regarding Germany's migrant policy.
As reported by the New Yorker magazine (not to be confused with the New York Times daily), "Susanna Feldmann has, in her death, been swept up in a grand argument about the lingering effects of the refugee crisis and the future of the country." 1 In Sect. 3.2 below, we analyse in more detail the structure of the "grand argument" in question. Here, we focus on the New York Times' headline, which ran as follows: 2 (1) Iraqi refugee is convicted in Germany of raping and murdering teenage girl.
We start from the assumption that a sentence like (1) invites a negative inference against Iraqi refugees. 3 There are two important things to note about this inference. First, the inference does not arise with any predicate in subject position. This can be seen by replacing 'Iraqi refugee' with another predicate, e.g., 'Austrian postman'. The hateful inference disappears: (2) Austrian postman is convicted in Germany of raping and murdering teenage girl.
Intuitively, the reason why the inference goes away in (2) is that we do not hold particularly damaging stereotypes against Austrian postmen. But it is easy to see that, in a hypothetical context where such stereotypes functioned, a sentence like (2) would trigger the same kind of negative inference as (1).
Secondly, the inference does not arise merely because one is saying something negative about Iraqi refugees. Consider the following headline published in a Spanish newspaper: 4 (3) Cuatro vecinos gitanos evitan que una mujer sufra una violación en plena calle.
[Four Roma neighbours prevent a woman from being raped in the middle of the street.] In this case, even though the headline reports a positive outcome, one cannot shake the impression that mentioning the ethnicity/background of the four people in some way raises the issue of whether individuals of that ethnicity/background are often criminals. It is easy to read the headline as highlighting this event as exceptional.
In sum, the appearance of social, ethnic, or national group terms together with actions that pose a risk for society leads to the inference that such actions are typically produced by members of these groups. This paper addresses three research questions in relation to this linguistic phenomenon. First, what kind of inference are these? In previous work (Domínguez-Armas and Soria-Ruiz 2021), we described the inference triggered by sentences like (1) as a provocative type of insinuation. These inferences are insinuations because the speaker is not directly and explicitly aiming to convey a negative message. And they are provocative, because they place the hearer in a difficult position, since they (i) provoke the hearer's outraged response but (ii) almost no form of reply seems to counter its negative effects (see Cepollaro et al. 2023). Here, in Sect. 2, we now go a step further and characterise these inferences as inviting racist explanations. Second, what is the argumentative function of provocative insinuations in public debates? In Sect. 3 we propose that provocative insinuations invite a generalisation based on a 'striking' property of the group mentioned in the subject. In short: mentioning the criminal's background suggests that individuals belonging to that group commit such crimes, where this generalisation is based on the fact that the particular crime is striking, in the sense of being socially relevant and dangerous (Leslie 2017). This generalisation then serves as a premise in practical argumentation regarding policies towards the group in question. Third, what can we do about it? In Sect. 4 we analyse the effectiveness of different communicative strategies to counter the hateful effects of provocative insinuations like (1).

What Kind of Inferences Are These?
We have described elsewhere (Domínguez-Armas and Soria-Ruiz 2021) headlines like (1) as a specific type of insinuations (Camp 2018;Fricker 2012). 5 Insinuations are pragmatic inferences strategically designed to remain 'offrecord'. 6 Imagine a real estate agent talking to a couple of potential buyers from a different ethnic background from the local majority (Camp 2018, p. 43): (4) Perhaps you would feel more comfortable locating in a more… transitional neighbourhood, like Ashwood?
On top of the "on-record" suggestion to look for a house elsewhere, the "off-record" insinuation is that they would not be welcome in this neighbourhood, probably because of their ethnicity. Another standard example of insinuation is the following. Suppose a speeding driver says the following to the police officer after receiving a ticket (Lee and Pinker 2010;Pinker et al. 2008, cited in Camp 2018: (5) I'm in a bit of a hurry. Is there any way we can settle this right now?
Although the explicit message is an innocent question, the implicit insinuation is an invitation to the police to waive the official ticket and accept a bribe instead. Finally, here is another of Camp's examples: (6) You know that Obama's middle name is Hussein. I'm just saying.
At first sight, (6) simply informs the audience of Obama's middle name. Nevertheless, when uttered in contexts where Islam is conceptually linked with terrorism, and appealing to the audience's background knowledge that "Hussein" is a Muslim name, it insinuates that Obama has some sort of terrorist ties (Camp 2018, p. 43;. In cases such as (4)-(6), the 'on-record' content is seemingly innocuous, cannot be denied by the speaker, and serves a controversial conversational point. By contrast, the 'off-record' content is controversial, and while it can be identified and criticised, it is designed to be conveyed without the speaker's being held accountable for it (Camp 2018;Fricker 2012). Fricker (2012) argues that insinuations are deniable, as the speaker can cancel their implicit content, and disavowable, as the speaker can object to having had the intention to convey the implicit content, both without appearing uncooperative. For instance, if the real estate agent is accused of bigotry ('are you suggesting that we would not be welcome in this neighbourhood due to our ethnicity?'), they can well react with 'Oh dear me, I didn't mean to suggest anything like that. I only meant that with so many families with young children here, you might not find as many people to socialise with as in a more up-and-coming neighbourhood' (Mazzarella 2021, p. 6). Alternatively, the real estate agent can simply answer 'Of course, you would be welcome in Ashwood'. The former reply to the accusation disavows the speaker's intention to convey the implicit message, while the latter denies its content.
In previous work, we characterised examples like (1) as an insinuation (Domínguez-Armas and Soria-Ruiz 2021). These sentences seem to fit the overall structure described by Fricker, Camp, and others. On the one hand, sentences like (1) convey the 'on-record' content that an individual of a certain background committed a crime. On the other hand, they convey the 'off-record' content that such actions are typically produced by members of such groups. Thus, the 'off-record' content associates the background of the individual with the committed crime. Moreover, characterizing (1) as a type of insinuation bears out the prediction that the implicit content of (1) can be denied or disavowed by the speaker without appearing uncooperative (Fricker 2012;Oswald 2022). Imagine the following dialogue: (7) a. [An] Iraqi refugee was convicted in Germany of raping and murdering a teenage girl. b. Are you suggesting that Iraqi refugees are rapists/ murderers? a. Not at all, Iraqi refugees are great people! / I didn't mean that! Whatever one thinks about the sincerity of speaker a, it seems clear that they can get away with denying or disavowing the problematic inference, just like in the examples reviewed before.
This being said, insinuations form a very varied lot (Camp 2018). We proposed to categorise (1) as a specific type of insinuation, namely provocative insinuations (PIs). The motivation for this label was the idea that (i) these utterances provoke the hearer to react against the utterance's racist undertones, but (ii) for almost any form of counterspeech that a hearer wants to use against the insinuation, it seems that the insinuator can deflect any accusations of misbehaviour (see Cepollaro et al. 2023).
We argued that (1) strongly invites the inference that Iraqi refugees tend to cause these actions by linking the individual's nationality to dangerous actions. We can now go a step further and suggest that the racist inference triggered by (1) (in contexts where it does trigger this inference, which is not always the case) seems to have an explanatory component. Note that the inference predicted is not that Iraqi refugees are dangerous or evil; strictly speaking, it is that the choice of the nationality is somehow explanatory: the criminal's ethnic background explains why they committed the crime. This explanation has various ramifications; it may mean that being of such background is a statistically good predictor of such criminal acts, or it may involve a deeper, causal connection between being of such background and being a criminal, or it may even involve the quasi-essentialist view that individuals of such background are prone to crime in virtue of being of such background. 7 It may not be obvious why people find such explanations acceptable, and why they are racist. We tackle both issues in the next section.
Before moving on, it bears mentioning that these examples feature an interesting interplay of linguistic and world knowledge: on the one hand, these inferences arise thanks to the linguistic structure of sentences like (1). Suppose that, instead of mentioning the criminal's nationality/political status, the sentence started with 'Someone is convicted…'-clearly, the hateful inference would not arise. But that is not the whole story. As illustrated by the contrast between (1) and (2) above, world knowledge plays a crucial role as well-without the existence of deep, engrained prejudice against migrants and refugees coming from poor countries (which are generally not harboured against postmen from rich, neighbouring countries), these inferences could not arise either. 8

Why are Provocative Insinuations Hateful?
All this being said, one may insist that sentences like (1) are merely informative statements about actions committed by an individual that simply happens to be of a certain background. One may complain that (1) cannot be read as a message of hatred against Iraqi refugees. After all, hate speech is standardly characterised as public expressions that spread, incite, or justify discrimination, subordination, and hostility against its victims (Torres da Silva 2021; Waldron 2012). And (1) does not explicitly incite hate against Iraqi refugees; by mentioning the nationality of the perpetrator, the headline does not incite action against persons with the same background. However, we argue that (1) can still be read as contributing to hate speech in a surreptitious way. In this way, (1) can be considered a form of soft hate speech (Assimakopoulos et al. 2017;Serafis 2022). Soft hate speech is constituted by expressions that appear to be neutral but nonetheless unduly target and belittle certain identifiable groups without explicitly inciting discrimination and hostility (Assimakopoulos et al. 2017). To justify our view that this is a form of soft hate speech, in this section we explore three aspects of PIs: the communicative purpose of PIs (Sect. 3.1), the reconstruction of PIs as part of a broader, public argument (Sect. 3.2), and the type of argument that PIs contribute to (Sect. 3.3).

The Communicative Purpose of PIs
We are sympathetic to the idea that discourse is naturally organized as aiming to resolve a 'question under discussion' (QUD) (Roberts 2012). In a conversation, interlocutors establish QUDs that 'tell [s] you what the discourse is about' (Roberts 2012, p. 8, italics added). Consider a situation where Aeden and Anyah are talking about a party that Maria organised at her house. Aeden asserts 'Maria invited Jan'. This assertion could be an answer to various questions, depending on what kind of information speakers are interested in. It could be an answer to the simple polar question Did Maria invite Jan? But it could also be an answer to more general questions, such as Who did Maria invite? or Who invited Jan? or even What did Maria do to Jan? These are what Roberts calls questions under discussion, a query that guides discourse and involves particular predictions about how speakers will structure their contributions to the conversation. Importantly, the QUD of a conversation need not be asked explicitly; it may be inferred from various contextual and conversational cues. 9 We want to suggest that (1) is a partial answer to a larger communicative exchange structured by the following QUD (Q1): 'What should be our policy towards Iraqi refugees?'. In particular, we see sentences as (1) as offering reasons to answer that question with a recommendation to not allow Iraqi refugees in (Germany). 10 While this connection might be seen as spurious or far-fetched, it is routinely adduced in the discussion of the case. Above, we already cited an account that the public debate over the rape and murder case behind example (1) contributed to "a grand argument about the lingering effects of the refugee crisis and the future of" Germany. 11 One distinct possibility of grasping this "grand argument" from various stories of refugees allegedly commonly raping "native German" women is as follows: The truth behind these baseless rumors is important because such stories influence Germans' image of refugees. They play into age-old clichés about the threat of foreign rapists. Few other arguments were cited as frequently by people in Germany in recent years for wanting to keep refugee camps from being opened in their immediate proximity. Once "they" are here, the argument went, the streets would no longer be safe for women or children unaccompanied by men. ("Fact-Check: Is There Truth to Refugee Rape Reports?" Der Spiegel, 17 January 2018). 12 Anti-immigrant, right-wing activists went as far as coining the derogatory term "rapefugees" and producing a website where all alleged rapes and other "refugee crimes" are listed (Der Spiegel, ibid.). While fact-checking journalists brand them in most instances as "baseless rumors", these fear-mongering arguments have been part-and-parcel of the public debate in Germany, and elsewhere, especially after the 2015 influx of refugees.
It should be noted, however, that (Q1) is different from Robert's traditional examples of QUDs. In the previous examplee.g., 'Who did Mary invite?'-the various possible answers address explicitly that QUD. The sentences 'Mary invited 9 One such cue is prosody: if Aeden puts prosodic stress on 'Maria' ('MARIA invited Jan'), then it is intuitive to assume that the QUD that his utterance addresses is Who invited Jan?; if the stress is on 'Jan' ('Maria invited JAN'), the relevant QUD would be Who did Maria invite? See Dretske (1972) for an original discussion of such "contrastive statements" vis-à-vis the questions asked. 10 Note that statements akin to (1) may not be contributing to the same argument in all contexts in which they are uttered. The context of (1) contributes decisively to the argumentative interpretation described in this paper. 11 https:// www. newyo rker. com/ magaz ine/ 2019/ 01/ 28/ how-a-teensdeath-has-become-a-polit ical-weapon. 12 https:// web. archi ve. org/ web/ 20190 11110 0913/ http:// www. spieg el. de/ inter natio nal/ germa ny/ is-there-truth-to-refug ee-sex-offen se-repor ts-a-11867 34. html. 8 We thank an anonymous reviewer for highlighting this issue to us.
Alice' and 'Mary invited Grace' are explicit, formal answers to the QUD 'Who did Mary invite?'. This is not the case for (1). 13 (1) ('Iraqi refugee is convicted in Germany of raping and murdering teenage girl') does not explicitly answer (Q1) ('What should be our policy towards Iraqi refugees?'). To the contrary, (Q1) is generated by a complicated process involving contextual understanding of salient societal issues about the discussion on refugee integration in Germany. Therefore, it is an implicit QUD. 14 The most immediate, literal QUDs that (1) answers would be: 'Who is convicted of raping and murdering teenage girl?' or 'What is an Iraqi refugee convicted of in Germany?' But we argue that such explicit QUDs are in fact subquestions of more general questions: 'Are Iraqi refugees rapists/murderers?' or 'Who are Iraqi refugees?'. These, in turn, are subquestions of a superquestion under discussion, namely, 'What should be our policy towards Iraqi refugees?' (sub-and super-questions are Roberts' terms). While that explicit subquestion is textually derivable, in our case the superquestion is implicit, as it needs to be contextually derived, based (partially) on world knowledge.
Furthermore, (Q1) is concerned with practical reasoning because it asks about a course of action in a particular situation (Fairclough and Fairclough 2012;Lewiński 2021). To illustrate practical reasoning, consider the case of two students leaving a seminar. One of the students might say 'we could go to a nearby pub I know' thereby establishing the practical question 'where could we go for a drink?' as the QUD of the conversation. Importantly, this QUD is different from, e.g., 'Are drinks cheaper at the coffeeshop or the pub?'-the former asks about what to do; the latter asks about what to believe (Lewiński 2017).
(Q1) can be described in a similar vein. (Q1) asks what to do in Germany towards the integration of Iraqi refugees. (Q1) has a set of alternative responses that concern different possibilities of action: e.g., 'we should let Iraqi refugees in'; 'we should not let Iraqi refugees in'; 'we should let only qualified Iraqi refugees in', etc. Furthermore, note that this QUD is different from the question 'What should be our attitude towards Iraqi refugees?'. The former concerns how speakers should act towards refugees, while the latter is about how speakers should feel about refugees. 15 An important aspect of the relationship between interlocutors' utterances and the purported QUD that these utterances address is that, depending on the QUD, one can infer the type of speech act that speakers intend to perform (Roberts 2012, p. 61). In the example of speakers gossiping about who Maria invited to the party, the QUD in the dialogue indicates that the speaker is asserting, for instance, that Maria invited Jan or Veronica. By contrast, in the example of the speakers discussing where to go for a drink, the QUD 'Shall we go the coffeeshop or the pub?' characterises the speakers' contributions as recommendations, suggestions, or proposals for/against each option (even if these utterances look superficially like assertions; see Lewiński 2021; Corredor 2023). Thus, assuming that (Q1) is the QUD that sentences like (1) address, an utterance of (1) might be described as a warning, rather than as a mere assertion. That is, (1) could be read as an invitation to be cautious about Iraqi refugees, rather than as a statement of fact: '[An] Iraqi refugee is convicted in Germany of raping and murdering a teenage girl… so watch out!'.
In sum, we propose that the communicative purpose of PIs such as (1) is to answer the implicit and practical QUD (Q1) 'What should be our policy towards refugees?'. We will now study how a sentence like (1) addresses (Q1). 13 We thank Claudia Picazo Jaque and Solmu Anttila for drawing our attention to this aspect of our analysis. 14 As a reviewer suggested, one might object that (Q1) is only connected to (1) in virtue of having 'Iraqi refugee' as part of its propositional content. So in principle, any other question about Iraqi refugees would be equally justified as an implicit QUD. We acknowledge this, but let us highlight two issues: first, we can accept that more than one QUD might be addressed by these utterances. We settled on Q1 for heuristic purposes, but questions such as 'Are Iraqi refugees violent?' or 'Are Iraqi refugees employed?' might equally be thought to be addressed implicitly by (1). One may worry that this overgenerates candidate QUDs for (1), but our second point is that contextual factors can substantially shrink the space of possible QUDs that a sentence like (1) might address. Note that (1) is a headline published by the New York Times, at a time when there was an existing debate about the integration of refugees in Germany. These contextual cues can considerably restrict the available QUDs, for example, to questions of social relevance about Iraqi refugees. 15 Both questions might be contextually connected (Roberts 2012, p. 12). QUDs can be contextually entailed in contexts with the same common ground, so that answering one of the QUDs yields a partial or complete answer to the other QUD. This is always the case with practical reasoning, which inescapably involves "theoretical" (that is, epistemic, factual) questions about what to believe: what the problematic circumstances to be solved are, what the possible solutions are, etc. On any model of practical reasoning-with the classic Beliefs-Desires-Intentions model being the most explicit case-the questions of beliefs or attitudes are relevant sub-questions with respect to the "superquestion" of what to do. Of course, over and above this conceptual relation, in any particular context where (1) is uttered, the connection of the above-mentioned practical and theoretical QUDs could be analysed by conducting, e.g., an experimental study. This can be done by asking a representative sample of subjects whether apparently "factual" sentences like (1) bear on the practical superquestion of 'What should be our policy towards Iraqi refugees?' (Note that Der Spiegel's report mentions this as an established social fact in Germany, but without citing background sources, at least in the version we have access to.) However, such a task is out of the scope of this paper. We thank Frank Zenker for drawing our attention to a possible experimental study.

Provocative Insinuations as Part of a Complex Argument
As noted before, the relation between (1) and (Q1) is different from the relation between sentences and QUDs in the standard examples given by Roberts (2012). In her examples, QUDs are generated following the syntactic and semantic structure of the relevant sentences. Instead, we think that the relation between (1) and QUD (Q1) is argumentative. Argumentation is the communicative activity of producing and exchanging reasons in situations of doubt or disagreement (Dutilh Novaes 2022; Jackson 2015; Lewiński and Mohammed 2016). For example, the discussion of the students deciding where to go after the seminar is partly an activity of producing reasons for/against going to each place. It is, as such, an instance of practical argumentation, that is, practical reasoning performed in situations of doubt or disagreement. Such argumentation takes as premises our desired goals and values and various (necessary, satisfactory, or the best) actions-qua-means to realise them. It then concludes that a specific action should be taken as, all things considered, it is the best (e.g., most efficient or noblest) or at least a satisfactory means to reach the goals. Or, indeed, that the action should be avoided, as it thwarts our goals or values, all other things being equal (Lewiński 2017;. Public debate over the question of policies towards refugees (our Q1) is, overall, an instance of such practical argumentation. Various parties to the debate would feed different values and goals into the argument (e.g., compassion towards people in need, national security, the need to preserve the "ethnic purity" of a country, economic advantages and disadvantages of accepting refugees, etc.), rank those they deem relevant or acceptable and, based on them, reach contrasting conclusions, as we described above. However, practical argumentation is also part of a larger argument structure, a complex argument where various types (schemes) of argument are combined to jointly support the conclusion (see van Eemeren and Grootendorst 2004; Freeman 2011). Arguments from example, from analogy, from authority, and causal arguments are routinely used to support practical arguments. This, we argue, is the case in our example. 16 (1) ('Iraqi refugee is convicted in Germany because of raping and murdering teenage girl') is embedded in a complex (and implicit!) scheme of practical argumentation that defends the standpoint 'we should not let refugees in'.

1.
We should not let Iraqi refugees in [conclusion]. 1.1 We want to live in security [value premise]. 1.1' Letting Iraqi refugees in thwarts our security [meansgoal premise].
While 1.1 is not manifestly controversial (depending on the definition of "security"), (1.1') is. The anti-immigrant argument would defend it by claiming that Iraqis are rapists/ murderers who threaten our security and, as such, should not be let in. As presented in Fig. 1, this basic argument can be further contextually reconstructed. (For simplicity, we skip the inferential step regarding allegedly thwarted security and focus on the remainder of the structure that is directly relevant to the insinuation we analyse here.) The argument structure presented in Fig. 1 differs in complexity from the simple practical argument scheme presented above: in contrast to the simple scheme, here we find various premises that together license the passage to the conclusion of the argument. Moreover, the complex argument structure remains predominantly implicit: only the PI (1) (premise 1.1.1.) is explicitly uttered by the speaker. 17 By way of uttering the PI (1) 'Iraqi refugee is con-victed…' (premise 1.1.1), an intermediary conclusion, 'Iraqi refugees are rapists/murderers' (premise 1.1), is licensed with the support of 'raping and murdering are striking properties' (premise 1.1.1'). Then, premise 1.1 and 'we should not let rapists/murderers in' (premise 1.1') leads to the conclusion of the argument 'we should not let refugees 17 For the rationale and methods of an 'analytical reconstruction' of the argumentatively relevant elements of discourse, see van Eemeren & Grootendorst (2004, Ch. 5). Such reconstruction is an analytic overview of the elements in discourse that are pertinent to the disagreement and includes the relations between different arguments defending an arguer's standpoint that often remain unexpressed. This implicitness-based on the presumption that much of the argument structure belongs to the common ground and its explicit performance would be redundant-has been a tenet of argumentation theory since Aristotle's concept of an enthymeme, an argument that is not explicitly performed in its entirety, because hearers already have its key elements "in mind" (en thymos). 16 In doing this, we go against the position of Hitchcock (2007) who explicitly excludes inferences involved in insinuation from the class of arguments: "[I]nsinuation is not argument. Although someone who insinuates something invites the hearer or reader to draw a conclusion from their words, the words themselves do not draw that conclusion. One can of course identify and discuss the argument that the insinuator invites us to construct for ourselves. Typically, however, this argument is rather indeterminate, precisely because insinuation merely suggests." (Hitchcock 2007, p. 120). We are not convinced by Hitchcock's position primarily because, by his criterion, any non-literal argument whereby "the words themselves do not draw [a] conclusion" would not count as argument, including all incomplete, enthymematic, and otherwise inexplicit arguments. This is a heavy theoretical cost, as such arguments are, arguably, the most common in ordinary discourse. We thus choose the path of "identify[ing] and discuss[ing] the argument that the insinuator invites us to construct for ourselves".
Footnote 16 (continued) in' (standpoint 1) and thereby addressing (Q1) 'what should our policy towards Iraqi refugees be? '. 18 In this section, we have described PIs such as (1) as embedded in a practical argument scheme that, further, via complex argument structure connects those sentences to a practical and implicit superquestion under discussion. In the following section, we examine the inferential step from premises (1.1.1') and (1.1.1.) to the intermediary conclusion (1.1) as contributing to a distinctive type of argument.

What Type of Argument Do PIs Contribute to?
We find particularly interesting the argumentative step from premises (1.1.1') and (1.1.1) to the intermediate conclusion (1.1) as it relies on a generalisation. Generics make generalisations about patterns in the world (descriptive generics) or about how the language should be used (definitional generics) (Krifka 2012;Mari et al. 2012). We consider that (1.1) belongs to the first type. That is, we suggest that (1) leads to a generic conclusion about Iraqi refugees, i.e., 'Iraqi refugees are rapists/murderers' (premise 1.1). From the existence of an individual from a particular group (Iraqi refugee) who instantiates a property (being a rapist and a murderer), the conclusion that individuals of the same group generally have the same properties is licensed. The formulation of an argument leading to a generic conclusion could be as follows:

Conclusion
Fs are G

Because
There is an individual x such that Fx & Gx.
By stating the premise that an individual x belongs to a group F and that the individual has a certain property G, one infers the conclusion that individuals of the same group F have the property G. Filling in the gaps, the argument scheme behind (1) can be formulated as:

Conclusion
Iraqi refugees are rapists/murderers.
However, this cannot be the end of the story. Note that, in general, a single instance is not sufficient to draw a generic statement. A purple plant isn't enough to conclude that plants are purple. So how could the argument from 3.1 to 3 ever be acceptable? We think that there is a way in which this argument can be compelling, but it requires a small detour into the semantics of generic statements. Generics relate two different sets of individuals, F and G, but unlike explicitly quantified statements, generics do not carry information about how many elements of F are elements of G (Leslie 2007(Leslie , 2011(Leslie , 2014Leslie and Lerner 2016;Mari et al. 2012). For example, 'tigers are striped' says something about how the set of tigers is related to the Iraqi refugees are rapists/murderers 18 We thank Lilian Bermejo-Luque and Christopher Tindale for their comments useful in the argumentative reconstruction represented in Fig. 1. set of striped things, but it does not say how many tigers have stripes. 19 Different types of generics introduce different restrictions on the relationship between the relevant sets F and G. Leslie (2008) identifies three types of generics. Firstly, 'majority generics' are statements that require most of the members of F to fall under G. For example, a sentence like 'cars have radios' is true in virtue of the fact that most cars have radios. If the majority of cars were not equipped with radios, then the statement 'cars have radios' would be false.
Secondly, 'characteristic generics' require the property denoted by G to be a characteristic trait of F. Characteristic generics do not require a majority of Fs to fall under G. Rather, they require that property G is typically shared by the members of F. For example, 'ducks lay eggs' does not require the majority of ducks to lay eggs (in fact, only female ducks lay eggs), but only that this property is typically shared by ducks.
Thirdly, 'striking property generics' require only a few instances to be true. Striking property generics refer to properties that 'we have a strong interest in avoiding because [they are] socially dangerous or harmful' (Leslie 2017, p. 397). Consider the statement 'mosquitoes carry the West Nile virus'. Although only mosquitoes of the genus Culex carry the West Nile virus (which accounts for 1% of mosquitos) the sentence 'Mosquitoes carry West Nile virus' is perceived as true. 20 The argument scheme in 3 cannot be reconstructed as supporting a 'majority' or 'characteristic' generic because it presents a single case of an F (Iraqi refugee) that is also G (rapist/murderer). But we may consider the argument scheme in 3 as an instance of a striking property generic. After all, rape and murder are social threats, and therefore they can be described as 'striking properties' according to Leslie (2008). By presenting the case of a single Iraqi refugee who raped and murdered a teenage girl in (1), the conclusion that 'Iraqi refugees are rapists/murderers' is, apparently, licensed with the support of the premise that raping and murdering are striking properties. The argument scheme can therefore be formulated as in Fig. 2.
Importantly, the social dimension of generics has been the focus of interest in recent philosophy of language (Leslie 2017;Rhodes et al. 2018;Saul 2017). In particular, the function of striking property generics in implying pernicious views about groups has been highlighted (Haslanger 2011;Leslie 2017). Haslanger (2011) argues that striking property generics implicate racist and sexist beliefs because they tend to essentialise the property in question. For example, 'women are submissive' implicate that women are disposed, by nature, to be submissive. "In choosing a generic, it appears that one is saying of a kind of thing, specified in the statement, that its members are, or are disposed to be G (or to G) by virtue of being of that kind" (Haslanger 2011, p. 193, italics in the original). If unchallenged, this licenses the further conclusion that e.g. 'women are submissive by virtue of what it is to be a woman'. 21 Iraqi refugees are rapists/murderers 21 Saul (2017) objects against the view that generics carry racist and sexist beliefs per se. Generic statements, she argues, can serve to fight racist and sexist social biases (Saul 2017, p. 14, italics in original). For instance, generic statements such as 'women are expected to want children', 'black people face discrimination' or 'LGBTQ + people are subjected to violence' have an important role as campaigners for social justice. 19 Leslie (2008) maintains that there is no specific 'generic operator' that expresses the meaning of generic statements. Moreover, generics are not limited to plural sentences (e.g. 'tigers are striped'): they can be expressed with indefinite singulars (e.g. 'a tiger is striped') and definite singulars (e.g. 'the tiger is striped') (Leslie 2008). 20 Other examples of striking property generics are 'sharks attack bathers', 'tigers eat people', and 'Pit-bulls maul children' (Leslie 2017). Although these cases are very rare, these generics are perceived as true in virtue of the property of being 'striking'. Accordingly, if (1) is reconstructed as in Fig. 2, the implication is that Iraqi refugees are disposed, by nature, to be rapists and murderers. This 'essentialising' tendency of striking property generics links them, again, to soft hate speech: even though (1) doesn't explicitly incite discrimination and hostility and thus appears to be neutral (Assimakopoulos et al. 2017), it implies negative 'essentialising' views about Iraqi refugees.
This section described (1) as part of a complex practical argument that answers the (Q1) 'What should our policy towards Iraqi refugees be?'. (1) is presented as part of a complex, even if largely implicit, argument (see Fig. 1) that defends the standpoint that 'We should not let Iraqi refugees in' and includes a premise that essentialises rape and murder as properties to which Iraqi refugees are naturally disposed. Therefore, we conclude, (1) is a form of soft hate speech because mentioning the nationality of the perpetrator invites the generic conclusion that members of the group commit such crimes, and that, in turn, suggests that they do so in virtue of their nature. The next section explores communicative ways of dealing with PIs such as (1).

What Can Be Done?
In previous work, we have studied different ways of dealing with PIs and concluded that PIs provoke a variety of responses in the audience that are broadly ineffective in blocking their effects (Domínguez-Armas and Soria-Ruiz 2021). Therein lies, as we noted above, the provocative character of these utterances: they trick the hearer into complaining, but they are designed in such a way that, at first sight at least, there is little to complain about. Specifically, we examined strategies such as (a) ignoring these utterances and (b) confronting them directly by saying, for example, 'are you suggesting that Iraqi refugees are rapists and murderers?' With much of the literature (Langton 2018), we concluded that ignoring the PIs can allow the hateful inference to be accommodated into the common ground of the conversation (Lewis 1979;Stalnaker 2002). A direct confrontation, on the other hand, risks shifting the agenda of the conversation, turning it into an explicit discussion of the hateful inference-which may give it visibility and some initial plausibility.
We now want to examine different communicative strategies to counter the harms of forms of hate speech (so-called counterspeech: Gelber 2021; Howard 2021; Langton 2018; Lepoutre 2019) such as PIs, taking into account the argumentative reconstructions explored in the previous section. This section looks at three ways of criticising the argumentative inference behind (1) (see Krabbe

Blocking Premise 1.1.1
Given that (1) is true, the speaker of (1) can easily deflect the accusation that they are lying or reporting unverified information. And given that PIs are deniable and disavowable (Sect. 2), they can also deflect a direct confrontation regarding the insinuated content. For example, if the speaker is confronted with 'are you suggesting that Iraqi refugees are rapists and murderers?' they can respond with 'No, Iraqi refugees are great people!', 'I didn't say that!', 'I'm just stating facts!' or 'You said it, not me! But now that you mention it…' (see Camp 2018, p. 46).
However, one can challenge the implicit content of PIs by means of 'blocking' (Langton 2018). To block is to "hinder the passage, progress, or accomplishment of something by, or as if by, interposing an obstruction" (Langton 2018, p. 145). Blocking occurs when speakers interfere by hindering the accommodation of implicit information taken for granted by other participants in a conversation. For example, consider a situation in which a football spectator shouts to a sluggish player 'Get on with it, Laurie, you great girl' and an alerted bystander replies 'Hey, what's wrong with a girl?' (Langton 2018, p. 145, italics in the original). The initial speech act was uttered primarily to express frustration at a sluggish player. But by calling the player a 'girl', the football fan presupposes a host of gender stereotypes. 23 The alerted bystander is resisting them by making such presuppositions explicit. That is, their intervention attempts to block those negative presuppositions (Langton 2018, p. 147).
Someone might block the acceptability of (1) in the context in which it was uttered in various ways. 24 For example, by saying 'Hold on a second, why immediately latch on to "Iraqi refugee"? Don't you know the guy had raped before, to start with?'. This response complains about the biased and unfair selection of nationality/political status as offering an appropriate explanation for the crimes committed by the individual, as opposed to a range of alternative, truthful predicates about the same individual ('suspected of another rape', 'electrician', 'unemployed young man ', etc.). This is an important possibility pointed out by German police officers: [W]hen compared to the German population, immigrants are more frequently young and male and are more likely to live in a large city, lack education, be unemployed and have no income. 'These can all be factors that promote criminal behavior.' ("Fact-Check: Is There Truth to Refugee Rape Reports?" Der Spiegel, 17 January 2018).
This contrastivity in selecting the predicate that picks out the subject-which then "elicits" the inference that that property explains the fact that occurred-is an important argumentative resource. It can indeed be used to block the surreptitiously conveyed presupposition by proposing alternative, better explanations for the specific case at hand via other predicates. "Suspected of another rape" is as true as "Iraqi refugee" but opens entirely different complex argument structures, notably those that would support a conclusion like "We should tighten our policies towards rapists, regardless of their nationality".
Another way of blocking premise 1.1.1 is to reply 'And a German was convicted of the same crime a week ago' (Domínguez-Armas and Soria-Ruiz 2021). This response also relies on a conversational eliciture that triggers the inference 'Germans are rapists and murderers' due to the choice of predicate. But such an answer mimics the pragmatic mechanism of PI (1), and thereby criticises its acceptability as a premise: if (1) is acceptable, the same argumentative structure should arise from 'A German was convicted of the same crime a week ago'. 25 One might argue that these blocking strategies are not ultimately successful. The speaker of (1), confronted with the accusation that their choice of words is malevolent, may say, for example, that they are 'just giving [us] the bare facts', as if sentences like (1) were innocent descriptions of an event. Or they may respond by deflecting the same accusations as appeals to political correctness, insensitive to the facts. These confrontational responses are based on the idea that a sentence like (1) is a bare report of reality that "just tells it like it is". To this, however, the counter-speaker can reply that their proposed revisions of sentences like (1) (e.g., 'electrician is convicted…') are just as successful at describing the "bare facts" or "telling it like it is", and thus that the defender of (1) has no advantage to claim for their preferred portrayal of the events.

Undercutting the Inference to the Intermediate Conclusion 1.1
In sum: undercutting the inference to the intermediate conclusion is substantially more difficult if the argument that (1) contributes to is an argument from a striking property generic. And since we have independent reasons to think that (1) indeed contributes to such an argument, this strategy would seem to be ineffective as a way of countering the pernicious effects of (1).

Defeating the Intermediate Conclusion 1.1
A third possibility of countering (1) would be to defeat the intermediate conclusion 1.1 by presenting overriding evidence against it (Krabbe and van Laar 2011). For instance, imagine a speaker who argues that Mrs. Wilson, who recently passed away, disinherited her daughter because her will leaves the daughter $1.00. Another speaker could rebut such an argument by bringing up the question 'How do you know that Mrs. Wilson wasn't mentally incompetent when she made her will?' (Pollock 1995, p. 161, cited in Krabbe andvan Laar (2011)). The question is a defeater since it challenges the conclusion that Mrs. Wilson disinherited her daughter, arguing instead that she was incapable of acting rationally at the time.
Recall that the intermediate conclusion 1.1 was reconstructed as 'Iraqi refugees are rapists/murderers'. We suggest that such defeaters will be unsuccessful if 1.1 is reconstructed in this way, because arguments based on striking properties require only a few true instances to license the conclusion. The overwhelmingly positive contributions of refugees to society are not sufficient to override the strikingness of rape and murder, precisely because such contributions are positive. As with the undercutting confrontation, the speaker of (1) can reply with 'I don't care if this is an exception, one example is enough to see that they [Iraqi refugees] are rapists/murderers', or 'rape and murder are so dangerous that we cannot afford any risk'.
To sum up: insofar as we favour the reconstruction of (1) as part of an argument based on a striking property, undercutting the inference to the intermediate conclusion 1.1 and rebutting the intermediate conclusion 1.1 seem to be ineffective. Therefore, we conclude that challenging the choice of language (Sect. 4.1) seems to be the most promising way to deal with provocative insinuations.

Conclusion
This paper argued that headlines such as (1) ('Iraqi refugee is convicted in Germany of raping and murdering teenage girl') convey the inference that Iraqi refugees typically commit rape and murder. We raised three questions about (1). First, what kind of inference is involved? We characterise (1) as a specific type of insinuation, namely provocative insinuation (PI). On the one hand, (1) conveys the (true) on-record content that an individual from a certain background committed a crime. On the other hand, (1) conveys an (false) off-record content that associates the individual's background with the crime committed, as if the crime was explainable by the social, ethnic, or national group to which the individual belongs. Second, why are PIs hateful? We describe (1) as part of a complex argumentative structure that addresses an implicit and practical question under discussion, i.e., 'what should be our policy towards Iraqi refugees?'. We conclude that in uttering (1) the speaker is favouring a negative stance towards that question, by way of licensing a generic and essentialising conclusion about Iraqi refugees (i.e., 'Iraqi refugees are rapists and murderers'), and thus arguing against their integration. The generic conclusion inferred from (1) constitutes a form of soft hate speech that does not explicitly claim discrimination against refugees, but which (if unchallenged) becomes a normalised label for the subjects (e.g., 'Iraqi refugees are rapists/murderers'). Third, what can be done to counter PIs? The paper contended that challenging the choice of language seems the most promising strategy to confront (1) because it complains about the choice of nationality as providing the best explanation of the crime committed.
We trust that future work, both conceptual and experimental, can further advance our understanding of the argumentative mechanisms behind provocative insinuations that we have identified here. the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.