Stereotyping and generics

We use generic sentences like ‘ Blondes are stupid ’ to express stereotypes. But why is this? Does the fact that we use generic sentences to express stereotypes mean that stereotypes are themselves, in some sense, generic? I argue that they are. However, stereotypes are mental and generics linguistic, so how can stereotypes be generic? My answer is that stereotypes are generic in virtue of the beliefs they contain. Stereotypes about blondes being stupid contain a belief element, namely a belief that blondes are stupid. This belief is an attitude taken towards the same proposition expressed by the sentence ‘ Blondes are stupid ’ , hence why we use the latter to articulate the former. This generic account of stereotypes can help us better understand their inner workings. I focus on one feature of generics, variability in the types of facts that can make them true, and argue that it can explain how stereotypes shape inferential patterns and thereby guide how we treat members of stereotyped groups. This feature, in turn, illuminates the harms caused by stereotyping and suggests some courses of action.


Introduction
Stereotypes play an important role in shaping our social behaviour. We associate whole groups with characteristics to guide our interactions with their members. Believing that management consultants are heartless or bin men smelly causes us to modify our behaviour towards individual management consultants or bin men. For example, we might avoid asking management consultants for help or keep our distance from bin men. But what are these stereotypes and how do they guide our behaviour? In this paper, I give an answer that takes its cue from the form stereotypes typically take.
Here are three sentences that express stereotypes: (1) Blondes are stupid (2) British people have bad teeth (3) Bus drivers are grumpy (1) -(3) are generics that express stereotypes, mental states associating social groups with properties. Generic sentences articulate generalisations, but do not include quantifier expressions like 'most' or 'many', and so can be contrasted with sentences like 'All blondes are stupid'. We often use generics to verbalise stereotypes, but is this coincidental or symptomatic of a deeper connection between the two? I argue that there is indeed a deeper connection, namely that stereotypes are partly made up of generic beliefs. The basic idea is as follows. Generics are linguistic items that express propositions: generic propositions. Such propositions can be expressed linguistically, but we can also take a variety of attitudes towards them. We can hope for them, dread them, expect them, or, we can believe them. Such beliefs are generic beliefs. In cases in which we believe a generic proposition of a certain type concerning a social group, we hold a stereotype about that group. When this belief causes other beliefs to be formed, serves as a reason for action, or is expressed verbally, we are stereotyping.
In the second part of the paper, I put this account of stereotypes to work and argue that one feature of generic contentsthat they can be made true by several kinds of factsexplains their inferential power. Generics can be made true by facts about what most members of a group are like, what is characteristic of them, what they can be like, or even what they should be like. However, this feature can cause those who hold stereotypes to form the expectation that their belief is not merely made true by one but by several related types of facts about the group and property in question, thereby disposing them to infer that several related non-generic generalisations are true. This can explain the inferential power of stereotypes and, in doing so, some of the harms they cause. I end by considering how my proposal could inform our response to stereotyping.
2. What are generics? What is stereotyping?

Generics
Generics are linguistic items. Sentences like Birds fly, A frying pan has a handle, and The child learns to read before it learns to write are called generics. 1 Sentences of the same syntactic forms as generics can, however, receive non-generic, existential interpretations e.g. A bird is in the tree, Birds are in the tree, or The bird is in the tree. While those sentences share the same syntactic form as generics, what sets generics apart is that they are not about any particular birds, frying pans, or children but are rather about birds, frying pans, and children in general.

Stereotyping
Stereotypes are mental items. They are mental states that associate properties with social groups 2 and, derivatively, those who compose the groups. 3 Some argue in favour of a narrower understanding of stereotypes, such as that the properties in question need to be associated more strongly with one group than another. 4 I will not take a position on this matter here but use examples of stereotypes that involve groups of people and properties that are thought particularly prevalent or otherwise noteworthy.
One important question about stereotypes is whether the properties in question need to be negative, or otherwise stigmatised. 5 These questions are bound up with what we want the notion of a stereotype to do: to pick out a general cognitive tendency or its harmful offshoot. In this paper, I will use examples of stereotypes that ascribe a stigmatised property to a social group because they are the generalisations I care about most, but what I have to say is compatible with a less restrictive understanding.
Stereotyping, meanwhile, is not a mental state but a kind of activity that involves stereotypes and can take different forms. One important 1 Generics are individuated by a certain type of meaning, though they also share common syntactic forms. In English, generic sentences are typically bare plurals (like Birds fly), indefinite singulars (like A bird flies), or definite singulars (like The bird flies). Most of the research on the link between generics and stereotyping has focused on bare plural generics and I will follow suit. For work on non-bare plural generics, see Greenberg (2004Greenberg ( , 2007. 2 I am using 'social groups' in an intuitive sense here to include what Young (2011) would call 'aggregates'. 3 For example, Begby (2021: 27). 4 For example, Puddifoot (2017: 139). 5 Those in the psychology literature tend to not require that stereotypes be stigmatised, whereas philosophers often do. See Jussim (2012) for an example of theorising of the first kind, and Beeghly (2015), Puddifoot (2017), Blum (2004) for examples of the second.
thing to note is that stereotyping can, but need not, include verbal expression. Take the following example. I hold a stereotype about fans of snooker player Ronnie O'Sullivan, which is that they are prone to heckling. As a result, I might expect a given Ronnie O'Sullivan fan to be prone to heckling and treat them in a hostile manner. This seems to be an example of stereotyping, regardless of whether I express the stereotype I hold verbally. Beeghly (2015) distinguishes various ways in which stereotypes can feature in judgments. Stereotype activation occurs when a stereotype is triggered in some way. The stereotype can then bring about beliefs, act as a reason for actions, or be expressed in speech.
First, stereotypes can play a causal role in forming beliefs. For example, my belief that Ronnie O'Sullivan fans are prone to heckling, together with the belief that Briony is a Ronnie O'Sullivan fan can cause my belief that Briony is prone to heckling. Second, stereotypes can be reasons for courses of action. For example, if I was to attend a snooker tournament, I might try to find a seat away from Briony because I believe Ronnie O'Sullivan fans like her to be prone to heckling and want to avoid them. Third, the stereotype can be communicated linguistically. For example, I might explain to my friend why I am trying to find seating elsewhere by gesturing towards Briony and, using a generic, say 'Ronnie O'Sullivan fans are hecklers'.
Often, several of these types of stereotyping occur simultaneously or in succession, but they need not. My stereotype about Ronnie O'Sullivan fans led to my forming the belief that Briony was prone to heckling, which in turn gave me a reason to find another seat, something I justified by expressing my stereotype verbally. However, many cases of stereotyping involve some but not all types. For example, I may form the belief about Briony and avoid her, but not say anything. That's why it's important not to focus on one type of stereotyping to the exclusion of others.

What's the connection between generics and stereotyping?
Several considerations seem to point towards the idea that there is a particularly close connection between generics and stereotyping. 6 First, English-speakers often use generics rather than overtly quantified generalisations to express stereotypes they hold. Instead of saying 'All blondes are stupid', they will say 'Blondes are stupid'. Many strengthen this claim to say that stereotypes are usually or typically expressed using generics. For instance, Blum (2004, 259) calls generics 'the typical form' of a stereotypical generalisation while Beeghly (2015, 676) says that 'stereotypes expressed in speech tend to take the form of what linguists call generics'. Second, as Blum (2004, 251) points out, stereotypes are typically held in a manner that renders them largely, though not entirely immune to counter-evidence. 7 This is particularly interesting, given that one aspect of generics that is often thought to distinguish them from universally quantified generalisations is that they tolerate exceptions. The fact that penguins can't fly makes All birds fly false, but Birds fly remains true. 8 This behaviour mirrors the ease with which stereotypes can survive what seem like exceptions. 9 Not very many who hold stereotypes about blondes being stupid believe that all blondes are stupid. As a result, they don't take the odd non-stupid blonde as a reason to change their mind. 10 Here's a straightforward explanation for these seeming connections: stereotypes are generic. Though this seems to be assumed in some discussions of stereotyping, I am not aware of many that explicitly discuss it. 11 One reason for this timidness might be an issue Beeghly (2015, 677) identifies. She considers whether stereotypes might be a subset of 7 For some further examples, see Begby (2013), Anderson, Haslanger, and Langton (2012), and Wodak, Leslie, and Rhodes 2015), all of whom also link this property to the exception tolerance of generics. 8 This observation has led to a whole cottage industry of theorists trying to explain exactly which exceptions and how many of them a generic can tolerate and still be true. See Leslie (2008) and Sterken (2017) for an overview. Consequently, an explanation of how stereotypes are resistant to evidence will depend, at least in part, on an understanding of the exception tolerating feature of generics. 9 For empirical work on this general feature, see Leslie, Khemlani, and Glucksberg (2011) and discussion in the social domain Leslie (2015). 10 Both points need to be qualified. While it is indeed true that stereotypes, at least in English, are often expressed using generics, they can also be expressed using overtly quantified generalisations. One thing to point out is that Blum and Beeghly contrast the use of generics with that of universally quantified generalisations. However, universally quantified generalisations are not the only kind of overtly quantified generalisation. Likewise, while it is definitely not the case that all those who believe stereotypes about blondes being stupid believe that all or most of them are, it is nevertheless the case that many of them do. Though I will argue that there is an important, constitutive connection between stereotypes and generic contents, this is not incompatible with the possibility of some stereotypes having a more universal flavour as these are beliefs we hold alongside beliefs in generic generalisations. I return to how these relate to one another in §5. 11 Two exceptions: . Jeshion (2013: 319) includes a discussion of stereotypes as generic in the context of accounts of slurs but does not explore this view in detail. . Begby (2013: 2,14), (2021:chapter 5) argues that stereotypes have generic contents and uses this to explain how stereotypes can be resistant to counterevidence, or in his words 'epistemically robust'. I see my proposal as being compatible with Begby's, for reasons similar to those given in Lemeire (2020).
generics but points out that that can't be the case because while generics are linguistic, stereotypes are not. However, I think that there is a view in the vicinity that is worth exploring. It's true that generics are linguistic and stereotypes mental. However, the content of generics, the propositions they express, might nevertheless have a distinctive generic character that could be the object of propositional attitudes involved in stereotyping. We might then think of stereotypes as generic in that they involve the same generalisations as those expressed by generics. I will develop such a view in the rest of this paper and argue that it helps us understand both how stereotypes guide our social behaviour and in doing so, some of the harms stereotyping causes.

Stereotypes involve generic beliefs
The basic idea is this: Generics are linguistic items that express propositions, namely generic propositions. Such propositions can be expressed linguistically, but we can also take a variety of attitudes towards them. We can hope for them, dread them, expect them, or, we can believe them. Such beliefs are generic beliefs. They are generic because the belief is a belief in a generalisation that can be expressed by a generic sentence. In cases in which we believe a generic proposition ascribing a stigmatised property to a social group, we hold a stereotype about that group. When this belief causes other beliefs to be formed, serves as a reason for action, or is expressed verbally, we are stereotyping.
I will say more about the nature of these beliefs shortly, but I first want to consider an initial worry about thinking of stereotypes as belief-like at all. As beliefs are propositional attitudes, they are attitudes taken towards propositions. However, many believe at least some of the content of stereotypes to be non-propositional. If that were so, at least some of the content of a stereotype could not be the kind of thing that is or isn't believed. Hence, stereotypes couldn't just be kinds of beliefs.
For instance, Smith and DeCoster (2000) argue that our associations with social groups include components like emotions and images that are not propositional. Whiteley (2022) argues that stereotypes can involve patterns of attention that influence which characteristics of individuals are salient to us. These patterns of attention are non-propositional and therefore cannot be part of beliefs.
I am sympathetic to their arguments and hence do not want to try to accommodate these elements within a standard model of belief. Instead, I want to argue that although stereotypes don't solely consist of generic beliefs, they do have a belief component. Stereotypes contain beliefs that structure the non-propositional elements in question. In addition to, say, emotions of hostility, we have beliefs that link those emotions to a target group and particular properties that give rise to the hostility. 12 The idea then is that stereotypes are mental states that have a belief component alongside other, non-propositional elements like emotions, images, or patterns of attention. The belief component takes the form of a generic belief. As stereotyping is the featuring of stereotypes in cognitive processes, communication, or as reasons for actions, generic beliefs play these roles alongside non-propositional components. Generic beliefs can cause other beliefs, act as justification for actions, and be communicatedas generic sentences.
Take the stereotype of blondes being stupid. This stereotype has a belief component and other, non-propositional components. The belief component is a belief that blondes are stupid. This belief is an attitude taken towards a proposition expressed by the generic sentence Blondes are stupid. The stereotype might have other non-propositional elements that bring up specific images, like those of famously stupid blondes, bring up emotions, say of condescension, or lead us to pay attention to certain things about blondes to the exclusion of others. The stereotype can cause other beliefs, for instance, that Maria, qua blonde, is stupid, cause actions, to doubt what she says, or be expressed by saying 'Blondes are stupid'.
This proposal enables us to straightforwardly make sense of the seeming connections between generics and stereotyping I mentioned previously. First, speakers often use generic sentences to express stereotypes simply because those stereotypes themselves contain generic beliefs. Second, stereotypes are at least somewhat resistant to evidence because generic propositionsthe objects of generic beliefstolerate exceptions.
In the remainder of this paper, I want to put the proposal I have put forward to work by focussing on a different point of explanation. I will argue that one feature of generic contents, variability in what kinds of facts can make them true, is important to understanding how stereotyping guides our behaviour towards stereotyped groups by disposing us to characteristic inference patterns.

Truthmaker variability
One feature of generics several theorists have drawn to attention recently is the variability among truthmakers for generics. These arguments can come in at least two forms.
First, some argue that generics are context-sensitive 13 or semantically incomplete 14 and can be made true by facts of differing quantificational force and flavour depending on the context in which they are uttered. 15 According to this line of thought, an utterance of Dogs bark can be made true by the fact that most dogs bark in a context in which statistical statements are salient, whereas in a context in which statements about evolutionary biology are salient, an utterance of the same generic could only be made true by facts about how dogs, by their nature, bark.
Second, some argue that the generalisations generics express are themselves non-specific and hence, even with the context of utterance held fixed, one and the same generic sentence can be made true by different kinds of facts. For example, Greenberg (2004Greenberg ( , 2007 argues that generics are vague, Lemeire (2020) that they express disjunctive statements, and Bosse (2021) that they existentially quantify over non-generic generalisations.
Variability in the truthmakers for generics can, according to these views, either apply across or within contexts of utterance. These proposals are similar in one particular regard: they say that the facts that can make true an utterance of a generic can be of varying types and flavours. Where they differ is whether these varying facts can make true one and the same generic sentence or just individual utterances of it. I will argue that this variability in truthmakers has an important effect on the types of inferences stereotypes dispose us to, and with them, the harms they cause.

Believing stereotypes and inferring
Generic beliefs are inferentially powerful. Our belief that dogs bark seems to guide our behaviours around dogs, both as a group and individually. For instance, believing that dogs bark might put us on guard when we knock on a door with a sign warning of a dog or lead us to believe that barking sounds from the house next door are caused by dogs. 16 13 For example, see Sterken (2015) and Nickel (2016). 14 See Nguyen (2019). 15 According to Nguyen, it actually isn't the generic tokens themselves that are made true or false, but pragmatic contents (implicitures) they convey. I ignore this distinction for ease of exposition here. 16 For empirical work on the inferential power of generics, see Leslie, Khemlani, and Glucksberg (2011), who argue that generic beliefs give rise to beliefs in other generalisations, notably beliefs that the A recent paper by McKeever and Sterken (2021) describes several distinctive inference patterns that generic stereotypes give rise to, including Amplification and Shifting. Amplification occurs when individuals accept generic claims based on weak evidence, but then take such evidence to warrant strong claims. For example, individuals form the belief that blondes are stupid on the basis of relatively weak evidence but then take that belief to warrant stronger statistical, modal, and normative claims. Shifting occurs when individuals accept a generic based on evidence for a non-generic generalisation of a particular flavour, but later take that same generic to support a different non-generic generalisation. I want to explain why this type of inference pattern exists in a way that makes use of the idea that various kinds of facts can make true generic contents.
Though generic propositions and, derivatively, the propositional attitudes towards them can have different truth-makers, those truthmakers often co-occur. It is true that most dogs bark, but also that they are capable of barking, disposed to it and that barking is characteristic of them. Not only are possible truthmakers for generics often true and false at the same time, but they typically explain one another. Most dogs bark because dogs are capable of and disposed to barking. Even if the generalisations don't always explain one another, we often think that they do. For example, we often believe that things have essences and that these ground both normative and other properties independently of whether this actually is the case.
I propose that these explanatory relationships between possible truthmakers for generics lead us to expect that they co-occur. 17 Holding a generic belief disposes us, with qualifications I discuss shortly, to expect that the belief in question is made true by not just one, but several possible truthmakers. In the social domain, holding a stereotype about a group of people disposes us to expect that the connection between the social group and characteristic in question is grounded in several types of facts that explain one another. For example, believing that blondes are stupid disposes us to believe that blondes are by their nature stupid, therefore disposed to be stupid, which in turn makes it property in question is prevalent. Pelletier and Asher (1997) argue that generic beliefs license defeasible default inferences. My proposal here is compatible with and indeed intended to complement these accounts. 17 Lemeire (2020 and personal communication) argues for similar such co-occurrence expectations, albeit based on a slightly different semantic account of generics and hence different kinds of expectations.
the case that most of them are stupid. Many of the inferences these cooccurrence expectations dispose us to are unjustified, false, and lead us to harmful behaviours, such as treating blondes are as though they are likely stupid. This is not to say that a generic belief automatically causes us to believe that all other generalisations about the kind and property in question are true. We often hold generic beliefs despite knowing that some possible truthmaker about the kind and property isn't true, for example, we can believe that dogs bark without believing that all dogs bark. Further, we can have general background beliefs about the kind and property in question, for example, that the kind is relatively heterogeneous or that the property in question tends to only be had by members of proper subsets of the kind, which constrain co-occurrence expectations. In addition, we can have general background beliefs about the direction of these explanations, for example, that it's not in the nature of members of a kind to have a property because many of them have it, but the other way around. My proposal here is that in the absence of such background beliefs, generic beliefs give rise to inferences about the possible truthmakers for generics. 18 This proposal can explain the inference patterns Sterken and McKeever identify in stereotyping as particular types of inferences based on cooccurrence expectations. Both amplification and shifting involve the forming of a generic belief, where that generic belief then causes the expectation that other generalisations are also true, in the case of amplification, stronger ones than the evidence that caused the original generic belief warranted. In both cases, generic beliefs are formed (based on more or less specific evidence) which, barring incompatibilities with background beliefs, in turn, give rise to co-occurrence expectations and hence cause beliefs in more specific generalisations. Our prior beliefs about social kinds and the nature of the stigmatised properties that are typically ascribed in stereotypes interact with cognitive biases to give rise to the distinctive inference patterns McKeever and Sterken describe.
The proposal can also illuminate the link between the generic character of stereotypes and the phenomenon of essentialising. Essentialising occurs when we infer that a given association between members of a kind and a property is grounded in and explained by further underlying, stable properties shared by members of a group. Work by Leslie (2014), Gelman, Ware, and Kleinberg (2010), Rhodes, Leslie, and Tworek (2012), and Wodak, Leslie, and Rhodes (2015) suggests that generic sentences might play an instrumental role in essentialising in that hearing about an association between a kind and property via a generic sentence might lead us to essentialise the kind and property in question. 19 On my proposal, essentialising is one instance of this broader inference pattern and happens when the co-occurrence expectations lead us from a generic belief to the belief that the connection between a kind and property is grounded in underlying, stable properties shared by members of the kind, a specific kind of generalisation.
So far, I have argued that stereotypes dispose us to characteristic inference patterns based on co-occurrence expectations about what makes them true. These inference patterns can explain some of what is wrong with stereotyping and what we might be able to do about it. The idea is that stereotypes dispose us to inferences that are often false and, typically, harmful. For example, believing that the residents of X-burgh are lazy disposes us, in the absence of conflicting information, to believe that laziness is prevalent among X-burghers and even that it is the result of some shared X-burgh essence. These inferences, in turn, can lead us to treat X-burghers as a group with contempt and even be suspicious of or hostile to individual X-burghers. In the last section of this paper, I turn to the question of what if anything we can do about this.

What to do
Some have advocated a prohibition on the use of generic sentences because of their potential for harm. 20 However, apart from questions about practicality and the positive role generic sentences can play in challenging stereotypes, this measure only tackles stereotyping insofar as it involves the expression of those stereotypes using generic sentences and not the holding of the beliefs themselves or the behaviour that these beliefs give rise to. 21 I will approach the wider question by first considering what we should do when we witness verbal stereotyping that involves utterances of generics. Insofar as stereotypes dispose us to false beliefs about social groups by virtue of the inferential patterns they give rise to, we might be tempted to take those utterances to be misleading speech acts and criticise them on this basis. 22 One prominent strategy in the literature on oppressive speech, of which stereotyping generics are often cited as examples, takes the form of counterspeech. Several authors that take this strategy have offered instruction on how speech whose harm comes about as a result of pragmatically communicated content should be responded to. For example, Langton (2018) argues that we can stop content communicated implicitly through presupposition by challenging the presuppositions and thereby preventing it from entering the common ground of the conversation through accommodation. Sbisà (1999) has argued that presupposed or otherwise implicitly conveyed content can be made explicit through what she calls 'explicitation procedures' to make it possible to challenge them and hold speakers accountable.
Alternatively, Haslanger (2011) suggests using metalinguistic negation to challenge the pragmatically conveyed content of stereotyping generics. 23 Her idea is that we can simply deny assertions of generics, not because what they semantically express is false, but on the basis that their pragmatic content is. So, for example, was someone to utter 'Women are submissive', Haslanger recommends rejecting their utterance on the basis that it conveys, via implicature, the false claim that women are, by their natures, submissive.
An argument by McGowan (2009) should give us pause for thought. McGowan argues that the counterspeech strategy is not sufficiently attentive to real-life conversational dynamics. She argues that oppressive speech has a sticky quality to it that makes it hard to undo it by using more speech. Once a conversational norm is enacted through oppressive speech, trying to undo that norm is like trying to 'unring a bell ' (2009:403). Simpson (2013) has generalised McGowan's argument by tracing the 22 One might worry that to mislead requires the intention to deceive, something that doesn't always seem to be the present. I agree that those engaged in verbal stereotyping don't always intend to mislead, however, that doesn't stand in the way of characterising their utterances as misleading. Saul (2018) has coined the term 'negligent falsehoods' for exactly such utterances. Negligent falsehoods are falsehoods 'propagated without sufficient attention to ascertaining the truth of the matter. ' (2018, 249) According to Saul, this can occur even when speakers' utterances are true, but cause interlocutors to believe falsehoods. 23 Horn (1989) stickiness of oppressive speech to a general mechanism, namely that it is much easier to prevent something becoming salient than it is to make something already salient unsalient. But what's so bad about making the ascribed content salient? Isn't that necessary in order to reject it? Lepoutre (2019) connects this discussion to various studies, which appear to show that raising the salience of a piece of ignorant speech, even on the way to rejecting it, can itself cause people to be more likely to adopt the false belief. 24 For instance, Lewandowsky et al. (2012) report efforts by a health authority to correct myths about the dangers of vaccines by producing leaflets that challenged the myths, but which in fact led those who read them to be more likely to fall prey to them. Such research seems to suggest that making content salient may itself be a problem and hence, that strategies that rely on it are ill-fated.
Where does this leave us with respect to stereotyping generics? At the very least, it should make us nervous about the effects of well-meaning interventions like the ones recommended by Langton, Sbisà, and Haslanger. If the above effect is a general one, then blocking manoeuvres, explicitation attempts, or metalinguistic denial may be not just ineffective, but actively harmful in causing interlocutors to believe stereotypes. Instead, Lepoutre recommends what he calls 'positive counterspeech' (2019, 34). Positive counterspeech is speech that engages with ignorant speech, not by denying it but by making a claim that is incompatible with the ignorant speech and hence entails its falsity.
Applied to generics, the idea would be that instead of responding to utterances of stereotypes by asking for clarification or metalinguistically denying them, we ought instead to produce speech that itself implicitly conveys that the stereotype or the inferences it gives rise to are false. This sounds good but would be exceedingly difficult. First, the question of what presents a genuine counter-example to generics is very hard to answer. 25 Second, it would be difficult to convey the falsity of the harmful inferences without repeating various non-generic generalisations that could, in the case of stereotyping generics, cause one's interlocutors to believe them.
To end on a slightly less bleak note, here is an alternative idea. I've argued that at least part of what gives stereotypes their potential for 24 Ignorant and oppressive types of speech are distinct from one another, but the types of verbal expressions of stereotypes using generics that I discuss here are discussed as examples of both. 25 See discussion of this difficulty with regards to stereotyping in Anderson, Haslanger, and Langton (2012) and Lemeire (2020).
harm is that they are inferentially powerful. This mechanism relied on the idea that generic beliefs dispose those who hold them to expect that if at least one non-generic generalisation about a kind and property is true, then so are several. However, I also argued that background beliefs can disrupt these co-occurrence expectations. To illustrate the idea a bit more, imagine someone telling a child that kangaroos have pouches.
Not knowing much else about kangaroos or marsupials more generally, the child is likely to infer that most if not all kangaroos have pouches. However, finding out that male kangaroos don't produce milk and that the reason marsupials have pouches to carry their young in is for them to sit in and drink milk can constrain the co-occurrence expectation and hence block the prevalence inference. The background beliefs we have about kinds and properties shape the types of inferences that our generic beliefs about them dispose us to draw.
In the case of social groups, we often lack such background knowledge, and indeed have strong essentialist biases that affirm co-occurrence expectations in its absence. These might be responsible for the especially strong co-occurrence expectations arising from generic beliefs about social kinds that are reflected in the essentialisation data. If we could disrupt the mechanisms responsible for the strong co-occurrence expectations that generic beliefs about social kinds give rise to, then we might be able to dull the stereotypes' potential for harm. Addressing co-occurrence expectations would also be helpful for responding to stereotyping in a way that doesn't just consist in telling people to not verbally express the stereotypes they hold, but to lessen their cognitive influence.
Strategies for this might take the form of simply increasing the number of accurate beliefs about social kinds, and thus make us less likely to allow generic beliefs to fill in the gaps. Alternatively, we might address co-occurrence expectations directly by focussing efforts to help us distinguish co-occurrence expectations that are wellfounded from those that are not. For example, this would involve improving understanding of the idea that in the social domain, a property's being prevalent in a group does not always give us good reason to believe that members of the kind share a common nature, or indeed that there is a normative requirement on them to have that property. The hope is that making us more generalisation-literate in this way will tackle the formation of false and harmful beliefs about social groups closer to their root.

Conclusion
In this paper, I provided an account of the role generics play in stereotyping. I argued that stereotypes involve generic beliefs alongside other nonpropositional content. Such beliefs, I suggested, are inferentially powerful in that they give rise to co-occurrence expectations in a variety of truthmakers for the generic beliefs. Understanding what lies behind these characteristic inference patterns can explain some of the harms stereotyping causes and also offer us a way of adressing them.