1. Social Categories in Inquiry: Explanation and Generalization

Social categories are categories we belong to in virtue of our memberships in social groups. These include races, ethnicities, religious affiliations, genders, sports teams, community groups, subcultures, and interest groups, with examples such as Kurdish people, Scottish people, African American people, men, women, Muslims, Christians and members of the local pick-up basketball league. In this section I will focus on two related aspects of the role that social categories play in inquiry: explanation and generalization.

Social categories are apparently indispensable to explanations of certain phenomena. This is evident in a range of examples that run from the extremely profound to the more light-hearted. For instance, we cannot explain political events taking place in Catalonia without a thorough understanding of the relevant social categories, in particular the category ‘Catalan’, and the relationship between that category and other social categories in Spain. A person attempting to explain and understand the history of Northern Ireland in the twentieth century would not be able to do so without a thorough understanding of the distinction between ‘Catholic’ and ‘Protestant’ and the history of the interactions between these groups in that region. These are very serious cases, but there are cases with less political significance where the social category is still explanatorily indispensable. For instance, when I moved from the UK to the US to attend graduate school, I was puzzled by the presence of large houses occupied by apparently affluent young men who all seemed to dress in a somewhat similar style. It was only when I became familiar with the social category ‘fraternity members’ that I could understand what was going on and thus effectively negotiate the social landscape.

The claim that such categories are indispensable to explanations leads to the question of whether such indispensability is in principle or in practice. For the purposes of this discussion, all I mean to establish is a claim about in-practice indispensability: as human beings engaged in nonidealized, incomplete scientific and academic inquiry, we will miss something significant if we choose not to appeal to social categories when formulating explanations of certain phenomena. This is reflected in everyday inquiry, such as the cases discussed above, and also in the social sciences, such as economics, as well as in other areas of social inquiry, such as history, anthropology, and critical social thought. The claim that such categories are explanatorily indispensable is familiar from long-standing debates about explanation and reductionism (for classic treatments see Fodor Reference Fodor1974; Putnam Reference Putnam1975).

Social categories vary significantly in what I call their fragility with respect to context. The notion of ‘context’ is slippery, but the relevant idea here is that fragility is responsive to social structure, rather than, say, linguistically determined notions of context, such as the topic of a conversation. Some social categories are fragile, by which I mean that their application is highly dependent upon particular details of social context, which can include factors like geographical location and time, but also features such as the existence of particular kinds of hierarchies or of particular kinds of professions. If these highly specific contextual conditions change, then the social category will cease to apply. For instance, social categories associated with trends are often very fragile. Consider the cases of the beatnik and the football casual. Beatniks formed a trend group that existed mostly in the USA in the 1950s and 1960s and were associated with the Beat generation of poets and musicians. Football casuals were a social group that existed in Scotland in the 1980s; football casuals would travel to football matches wearing expensive sportswear, and they were known for committing acts of violence against the supporters of other football teams. There is no longer any such thing as a beatnik or a football casual because when we move away from a very specific social context, we find that these social categories simply cease to apply.

Some social categories are fragile in that they display this kind of highly specific dependence upon context, but many are more robust, and persist across a variety of different social contexts. For instance, religious and ethnic categories, such as ‘Jewish people’ or ‘Catholic people’, are more robust and less fragile than social categories associated with subcultures or particularly context-specific professions, such as vloggers or chimney sweeps. These less fragile categories persist across national boundaries, different social classes, and across more substantial periods of time. Overall, a social category is more fragile the more sensitive its applicability is to particular details of social context, and less fragile the less sensitive its applicability is to particular details of social context.

One might expect the fragility of social categories to be correlated with their explanatory indispensability, such that the more fragile a social category is, the less likely it is to be explanatorily indispensable. But this is not the case, as remarkably fragile categories can be indispensable to explanations of certain social phenomena. For example, consider the case of the category ‘fraternity member’. Without familiarity with this category, it is impossible to explain a variety of social phenomena, and yet this is a highly fragile category because if we move away from a highly specific social context, the category ceases to apply. Alternatively, consider the highly fragile case of the ‘six pocket woman’, a social category related to the role of pockets in the economy of Papua New Guinea, which is similarly indispensable to certain explanations (Taylor Reference Taylor2016: 4–5). Differences in the fragility of social categories may be correlated with other explanatory features—for example, less fragile categories might feature in explanations of more phenomena simply because they appear in a wider variety of social contexts—but highly fragile categories are often just as indispensable to explanations as less fragile categories. Thus, social categories are indispensable to explanations, and even highly fragile social categories display this feature.

Social categories play a related role in lawlike generalizations about the social world, a role that mirrors that of lawlike generalizations about the natural world. A lawlike generalization about the natural world is a generalization about patterns in events as they are constrained by the laws of nature (though a Humean would reject this talk of ‘constraint’). For example, statements of laws of nature, such as, ‘for a fixed amount of an ideal gas at a fixed temperature, pressure is inversely proportional to volume’, or ‘the total entropy of an isolated system can never decrease over time’, are examples of such lawlike generalizations. Generalizations that are not laws themselves but that follow from laws are also lawlike generalizations in the sense at hand, including ‘one cannot travel faster than the speed of light on I-95’. We use lawlike generalizations about the natural world for a range of different purposes. On the basis of these generalizations, I can form predictions about facts such as the volume of a gas with a particular pressure or the success of attempts to travel faster than the speed of light. I can explain why events happen by appealing to the lawlike generalizations that govern them. And I can use lawlike generalizations to support interventions in the natural world. For instance, if I want a body of gas of a particular volume, I can use lawlike generalizations to work out what conditions need to be in place for that goal to be achieved.

We use lawlike generalizations about the social world for similar purposes in a variety of different areas, including social science, social theory, and everyday life. For instance, much critical social and political theory involves attempts to formulate generalizations about the workings of unjust societies in order to permit understanding and explanation of injustice as well as to predict, intervene in, and prevent that injustice. Some good examples come from recent social and political philosophy. For instance, Nora Berenstain has written about the harm of ‘epistemic exploitation’, a phenomenon whereby people in marginalized situations are required to educate the nonmarginalized about their situations, often at significant psychological, social, and material cost (Berenstain Reference Berenstain2016). Talia Mae Bettcher has documented a specifically pernicious double bind faced by transgender people between the option of disclosing their trans status and thereby risking violence, murder, prejudice, and disrespect or, alternatively, not disclosing that status and also incurring the risk of violence, murder, prejudice, and disrespect (Bettcher Reference Bettcher2007). Kristie Dotson has written about a variety of different kinds of epistemic oppression and the social and epistemic change required to undermine that oppression (Dotson Reference Dotson2014). In each of these cases, the author identifies a pattern in social events that constitutes or contributes toward injustice. Similar patterns in social events were identified by historical writers, including Karl Marx and Friedrich Engels, whose work on the structure and function of ideology constituted an attempt to understand a pattern in social events and the role of cultural and educational institutions in that pattern (Marx & Engels Reference Marx and Engels1970). These cases come from emancipatory social thought, rather than from social science, and this is a deliberate choice because the role of social categories in these cases is different from the role of natural categories in scientific explanation and generalization, and yet both still undeniably feature in explanation and generalization.

Such generalizations can play a role in explanation, prediction, and intervention very similar to that played by lawlike generalizations about the natural world. For example, if I want to explain why a particular person's testimony in a trial was for no apparent reason discounted in the judge's summing-up, I might appeal to Dotson's generalizations about epistemic oppression. If I want to understand why rates of consumption of luxury goods, such as designer shoes, do not strictly correlate with income, I might appeal to neo-Marxist generalizations about commodity fetishism. Similarly, I can use these lawlike generalizations to form predictions. On the basis of Berenstain's work I may predict that, for instance, as a woman in philosophy in certain situations I may be called upon to describe the situation of women in philosophy and that my answer to that call may be met with skepticism and resistance even on the part of those who asked for the information.

The extent to which lawlike generalizations play an identical, rather than merely similar, role in generalizations about the natural world and about the social world depends on the similarities and differences between natural laws and other kinds of generalizations. For instance, some hold that laws of nature are exceptionless, while social generalizations are not, but others hold that laws of nature do admit of exceptions. Likewise, some hold that laws of nature are universal and global and hence very different from social generalizations, while others hold that there are local laws of nature, much as there are local social generalizations (for discussion see Cartwright Reference Cartwright1980; Lange Reference Lange1993). These issues have a particular relevance to questions about the modal status of social and natural generalizations and to questions about whether all social generalizations can be undermined, which is important for understanding how intervention works in the social cases. These issues cannot be settled here, but we can recognize that natural generalizations and social generalizations play similar roles in explanation, prediction, and intervention without establishing precisely how similar.

2. Interpreting the Role of Social Categories in Inquiry

Some philosophers have argued that the role of social categories in inquiry favors certain metaphysical views of social categories. On this line of thought, social categories could not play a central role in inquiry without some shared metaphysical feature that makes it possible for them to do so. These approaches reflect a broadly realist view of explanation and generalization as closely tied to metaphysical structure; according to this, if categories play a central, legitimate role in explanation and generalization, then they must do so in virtue of having some metaphysically good-making feature. I will call these the ‘metaphysical’ views of social categories, and in this section I will survey two such approaches: the view that social categories are natural kinds and the view that social categories are at least reasonably joint-carving.

The view that social categories are natural kinds has been defended by Ron Mallon. Mallon adopts what he calls an ‘explanation-driven’ approach to social metaphysics, oriented around accounting for the apparent explanatory indispensability of social categories and explaining the apparent causal capacity of socially constructed entities (Mallon Reference Mallon2016). On Mallon's view, natural kinds (and hence social categories that feature in inquiry) are homeostatic property clusters, clusters of properties that are typically (though not necessarily) instantiated together because of an underlying causal mechanism that is responsible for their nonaccidental co-instantiation (Mallon Reference Mallon2016: 90; see also Boyd Reference Boyd1991). The homeostatic property cluster (HPC) view of kinds is an explicitly metaphysical view, because on this position the boundaries of natural kinds are determined by causal mechanisms.

Mallon argues that it is a constraint on an adequate account of social construction that it make sense of the explanatory role of social categories and that in order to do this we must embrace an HPC view of social categories that feature in explanations. As he puts it,

Later on in his book Mallon returns to this point, arguing that an account of social construction given merely in terms of social roles will be too weak to be adequate because it fails to account for the role that social categories play in explanation and generalization (Mallon Reference Mallon2016: 66–67). Other authors have defended natural kind approaches to social categories, including Muhammad Ali Khalidi who argues that some, but not all, social categories are natural kinds (Khalidi Reference Khalidi2013a, Reference Khalidi2013b).

Another metaphysical view of social categories is the position that social categories are comparatively metaphysically joint-carving. One influential notion of joint-carvingness is David Lewis’ account of metaphysical naturalness (Lewis Reference Lewis1983). According to Lewis, properties lie on a spectrum, at one end of which is a privileged, special group of properties that appear in the laws of nature, account for resemblance and causal powers, and ground other properties, and at the other end of which are properties that are complex, gerrymandered, and not fit to play such roles. The former are the perfectly natural properties and the latter highly non-natural properties. For example, on this view, the property of ‘being negatively charged’ is more natural than the property of being ‘something that I thought of last Wednesday’. Theodore Sider has extended and elaborated metaphysical naturalness into a broader notion of structure that extends beyond properties to include domains such as logical structure (Sider Reference Sider2011). Following Lewis, Sider connects structure to a variety of different areas, including laws, explanation, and reference. In particular, he argues that explanations are generally more successful when the categories that appear in those explanations are reasonably structural (Sider Reference Sider2011: 22–23). In so far as theories and explanations of the special sciences are adequate, according to Sider, the categories they feature must be at least reasonably joint-carving (Sider Reference Sider2011: 22). If social categories feature in legitimate, adequate explanations and generalizations, then those categories must be reasonably join-carving. So once again, we see that the role that social categories play in inquiry is taken to favor a metaphysical view of their nature.

There are other metaphysical views of social categories, but these two cases give us the rough idea—that in so far as social categories can play a central role in inquiry, they must do so because of a shared metaphysical feature. The views defended here are connected to and motivated by broader realist considerations about explanation and the nature of inquiry. I will not address the broader issues about realism here, but overall I do take these social cases to be good evidence against the viability of at least some versions of explanatory realism (for further discussion see Taylor Reference Taylor2018).

Metaphysical approaches to social categories have implications for the ontological unity of social categories. If all social categories play a central role in inquiry and if that role demands a metaphysical interpretation, then social categories must be to some extent metaphysically unified. The extent to which they are metaphysically unified will depend on the precise details of the view, but on the views considered here, social categories are unified in all being HPCs or all being reasonably metaphysically joint-carving, and this is settled by their role in inquiry. If we abandon the idea that a metaphysical approach is needed to account for the role of social categories in inquiry, then the ontological unification of social categories is not settled by such considerations. Social categories may turn out to be ontologically unified or not, but that matter would not be decided by their role in explanation and generalization. Accordingly, this discussion will have implications for the unification or otherwise of social categories, which I will return to in section 4.

3. Problem Cases: Empty Categories

Some social categories play a central role in inquiry but resist a metaphysical interpretation. I call them empty social categories, because they are social categories that have no metaphysical basis but around which social structure has formed. I will argue that because these categories are empty, they present counterexamples to metaphysical views of social categories, but that because social structure has formed around them, they still play a central, ineradicable role in inquiry.

To get a rough idea of the notion of an empty social category, consider three cases:

‘Pisces’: Pisces is an astrological category. A person born under the sign of Pisces is someone who is born within a certain set of dates and because of this is subject to astrological forces that determine further shared characteristics and a shared likelihood of life events unfolding in particular patterns. This is a widely known social category, along with other astrological categories, but because there are no astrological forces, this category is empty. There are people born within a certain date range in the year, but they are united only by that range of dates, not by astrological forces that determine further shared characteristics or the likelihood of any patterns in future events involving them.

‘Welfare Queen’: The ‘welfare queen’ is what Patricia Hill Collins has described as a ‘controlling image’, a harmful, oppressive stereotype that applies specifically to black American women (Collins Reference Collins2000). This controlling image portrays a category of women that does not exist: black women who deliberately have many children, often with different fathers, in order to take advantage of the American welfare and public housing systems. There is no such group of women, and the myth of the ‘welfare queen’ has been widely documented (see Collins Reference Collins2000; Levin Reference Levin2019).

‘Incel’ categories: The term ‘incel’ is used for a group of (typically) young men who self-identify as being ‘involuntarily celibate’. Members of this group typically hold a complex set of false beliefs about gender and the evolution of human sexuality, which includes a social taxonomy, featuring categories such as ‘Chads’ and ‘Stacys’. ‘Chads’ are males whose dominance and sexual success is undergirded by a complex network of facts about evolution and female heterosexual preference. ‘Stacys’ are attractive females, again united by a particular set of features associated with evolution and male heterosexual preference. However, because incel views on evolution, sexuality, and gender are false, these categories are empty. There are people who are attractive and people who have many sexual partners, but there are no ‘Chads’ or ‘Stacys’.

In each of these cases, a category is delineated by false beliefs about the world. In the Pisces case, the false belief is that there are astrological forces that, among other things, mediate a causal association between time and place of birth and certain personal characteristics. In the welfare queen case, the false belief is that there is a significantly large group of women who deliberately exploit the US welfare and housing systems by having a large number of children. And in the incel case, the false belief is that there are robust patterns in human evolutionary history and sexual preference that account for the dominance and attractiveness of people with certain kinds of bone structure, hair, place in American society, and so on. Because each of these categories is delineated by false beliefs, there is no metaphysical structure for the category to correspond to. We can see this point more specifically if we examine both the HPC view and the joint-carvingness view. On the HPC view, the social category is a cluster of properties that tend to be instantiated together, with an underlying causal mechanism responsible for their regular co-instantiation. In each of these ‘empty’ cases, there is no underlying causal mechanism, and so there cannot be a natural kind. On the joint-carvingness view, because the category is empty, there are no metaphysical joints to carve. Accordingly, these metaphysical views cannot accommodate empty social categories.

However, despite their emptiness, empty social categories can play a significant role in inquiry. For example, if we want to examine patterns in housing policy, attitudes toward race, gender, and welfare policy in the 1980s in the USA, we need a grasp of the social category ‘welfare queen’ even though that category is empty. Indeed, some have argued that a grasp of this category is essential to understanding phenomena ranging from the oppression of black women in the USA to Ronald Reagan's presidential campaigns (Collins Reference Collins2000; Levin Reference Levin2019). Similar facts are true about empty categories such as ‘Pisces’ or ‘Chad’—we need a grasp of these if we are to explain certain phenomena. This makes them different from completely made-up or imaginary social categories, and the difference is that social structure has formed around these categories despite their emptiness, making them necessary to explain certain aspects of that social structure.

Consider two contrasting examples for illustration of the claim that empty categories can be indispensable to inquiry. Imagine a community that lives in fear of what it calls ‘child snatchers’, men who kidnap children. This fear is unfounded as there have been no incidents of child snatching in this community, and so there are no ‘child snatchers’. However, the belief that there are ‘child snatchers’ is central to a variety of different customs, norms, and practices in the community, such as a mistrust of men around children and an entrenched preference for female childminders over male. If that were the case, then we would not be able to explain or understand these practices without a grasp of the social category ‘child snatcher’. For instance, if I were to attempt to explain why there is an entrenched preference for female childminders over male ones in this society, my explanation would have to include some reference to ‘child snatchers’ even though there are no such things. Compare this with a different scenario, in which I claim that there is a special trend group called x in British society and that members of x adopt a certain haircut and listen to a certain kind of music. However, I am making this up, and there simply is no such group. That is a false claim about social structure because nothing in this society is based on x; no practices or institutions have formed around it. Accordingly, inquiry into this society does not require any information about x, whereas inquiry into the society in the previous case does require a grasp of the category ‘child snatcher’. Something similar is true of each of the real-life cases described above. Even though ‘Pisces’, ‘welfare queen’, and ‘Chad’ are all empty categories, social structure has formed around them in that norms, practices, expectations, and institutions have been shaped around them. As a result, these categories appropriately feature in inquiry about that social structure despite their emptiness.

One might argue that in such cases what is indispensable for the explanation is not the social category itself but instead the claim that people have false beliefs and that this drives their behavior. If this is the case, then the empty category itself would not be indispensable to inquiry, and this objection to metaphysical approaches to social categories would fail. Furthermore, beliefs can be cited in standard causal explanations, and so their role in inquiry would be appropriately understood in metaphysical terms; thus, both the challenge to the metaphysical realist approach to categories and the challenge to the broader realist picture of categories in inquiry would be undermined.

If social categories were eliminable from inquiry, then equally adequate explanations of the same phenomena could be given without any mention of the relevant category. However, this is not the case. Imagine trying to understand and explain negative attitudes toward welfare systems in the USA in the 1980s and 90s, especially compared to other countries. such as the UK, where they were recognized as an essential part of social infrastructure. The answer will be complicated, but one part of that answer will include the fact that the ‘welfare queen’ was a recognized social category, and conventions and expectations were formed and actions were taken around and about this social category. Compare this with the explanation that ‘people had false beliefs about how people on welfare behaved’ or that ‘some people believed that some women had children deliberately to take advantage of the welfare system’. These attempts go a certain distance, explanation-wise, but they fail to capture the rich, perniciously harmful interaction between gender, race, poverty, and female sexuality encapsulated in the mythical category ‘welfare queen’. Merely telling us that people had false beliefs, even that they had particular false beliefs, does not go far enough. We need to know about the relevant social category in order to explain these phenomena fully.

However, one could press the point that even if the social categories are ineliminable from explanations, they are only ineliminable because they appear in people's beliefs and hence in causal explanations of the relevant phenomena. The objection here is not that the categories are eliminable from inquiry; instead, it is that the categories are ineliminable only because they appear in people's beliefs, and so these cases still do not provide a counterexample to metaphysical interpretations of the role of social categories in inquiry. However, if this were true, then an explanation that actually gives information about the appearance of the relevant social category in individual people's beliefs would be preferable to one that abstracts away from such detail as the former would give us more information about what, on this line of thought, is driving the explanation itself—namely, detail about causal interactions. However, this is not the case. Consider the case of the ‘welfare queen’ for illustration. Let us return to attempting to explain the negative attitudes toward welfare systems in the United States in the 1980s and let us say we add to this explanation a very detailed causal account of the beliefs of every individual who has a belief about ‘welfare queens’. This would be an unwieldy, messy, and massively detailed explanation. Compare this with an explanation that focuses on the myth or controlling image itself and abstracts away from this causal detail. This would be an equally legitimate explanation, perhaps even a better one, than the causal explanation. This kind of approach is much more in keeping with our actual explanatory practices than the explanation that gives detailed information about individual beliefs. If the causal detail really is what is driving the explanation in such cases, then we could reasonably expect more causal detail to improve the explanation. But in such cases, more causal detail makes the explanation worse. This last point leads to deep debates about reduction and the nature of explanation, and so there is more to say here. However, note that if we reject these cases on the basis of a broad background view on explanation, such as the view that all explanation must give information about causes, then we are not using the explanatory role of social categories as a guide to social ontology. Instead, we are using a background view on the nature of explanation as a guide to social ontology, which is a very different inference. Overall, the empty categories are not eliminable from enquiry, and they do not always appear in causal explanations.

Some might argue that metaphysical views of social categories can make sense of these cases. For instance, Mallon argues that certain categories are ‘covert’ in that we are fundamentally confused about the nature of those categories (Mallon Reference Mallon2016: 207–208). At first pass this appears to cover cases such as ‘Pisces’ or ‘welfare queen’ because those cases are also mired in confusion about the nature of the categories themselves. However, the confusion at the heart of Mallon's covert categories is confusion about whether the base of the category is natural or social. On the HPC view of natural kinds, the confusion is about whether the mechanism that underlies the co-instantiation of the relevant properties is social or natural. However, in either case the category is still an HPC, and as such its boundaries are still determined by the causal structure of the world, unlike in the empty cases, which do not track the causal structure of the world. Alternatively, Carl Craver has argued that in many cases, HPC kinds include a conventional element, in that there are perspectival aspects that go into the selection of the precise mechanism that underlies the property cluster, making the definition of the kind somewhat conventional (Craver Reference Craver2009). This might seem promising as a way to make sense of empty social categories. However, on Craver's view there still is a causal mechanism underlying the kind, a mechanism that is responsible for the co-instantiation of the relevant properties, unlike in the empty cases, in which the mechanism is absent.

Overall, then, the metaphysical approaches to social categories considered here struggle to handle these cases. One response could be to simply deny that these are social categories. But this is dialectically problematic. If metaphysical approaches to social categories are motivated by their role in inquiry, and we then find categories that play the same role in inquiry but cannot fit with the metaphysical view, then we have found counterexamples to the metaphysical view. This is far from a complete survey of metaphysical approaches to social categories, but in the next section I will describe and defend an alternative approach to social categories that not only handles these problem cases but is also compatible with a wide range of different approaches to social ontology.

4. Context-Dependent Naturalness

As mentioned in section 2, the notion of metaphysical naturalness was articulated and defended by David Lewis (Lewis Reference Lewis1983). Metaphysical naturalness is a property of properties, whereby some properties belong to an ‘elite minority’ that ground other properties and play roles in laws, causation, reference, explanation, and objective similarity. Metaphysical naturalness is context-independent, such that if a property is perfectly natural it is so across all contexts, regardless of factors like the interests of particular individuals or the goals of a particular community. However, elsewhere I have argued that there is another phenomenon that is in some ways similar to metaphysical naturalness, but that it is context-dependent (Taylor Reference Taylor2016). In much the same way that metaphysically natural properties are metaphysically privileged, certain properties are privileged with respect to a particular activity but not necessarily in general. For example, the property of ‘having positive charge’ is highly significant to someone who is practicing chemistry and less so to someone who is practicing psychiatry. Alternatively, the property of ‘being bipolar’ is highly important for a psychiatrist but not for a chemist. The importance and significance of the property (though not the property itself) is relative to the activity. An account of context-dependent naturalness takes these familiar ideas as the basis of a context-dependent alternative to metaphysical naturalness.

In what follows I will describe the basic motivations for context-dependent naturalness, the account of context-dependent naturalness, and explore some similarities and differences between context-dependent naturalness and Lewisian metaphysical naturalness before offering an application of context-dependent naturalness to social categories. The idea of context-dependent naturalness (hereafter c-naturalness) begins with a certain kind of context: an activity. On a broad, everyday notion of activity, playing a particular sport or game, pursuing different kinds of inquiry, working at a certain job, and (importantly for this discussion) negotiating social landscapes all count as activities. Activities have structure. The structure of activities is not objective or mind-independent because it is typically determined by the rules, conventions, and norms governing the activity in the communities in which it is practiced. But given those rules, conventions, and norms, the structure is there, and in order to perform activities we need to be familiar with that structure. For example, we all know that football is a game with goals and penalty kicks, whereas cricket is a game with wickets and runs. If we start talking about goals when we are discussing cricket, we have made a clear mistake about the structure of that game. This is not mind-and-context-independent fundamental structure, but it is structure of a kind, and we need to be able to limn that structure in order to perform the activities we want to perform.

The account of c-naturalness makes use of the idea that the structure of activities is captured by a framework of properties associated with that activity. Whenever we learn about a new game, a new kind of cooking, or a new academic discipline, for instance, we learn about the properties or categories relevant to the activity. For example, when I learn to play chess, I learn about the knight and about the checkmate, and a significant part of learning chess is taken up with learning about these classifications and learning how to apply them. Certain properties within these frameworks are particularly salient with respect to the relevant activity, which means that they are essential to that activity. A property is salient with respect to an activity if you cannot perform that activity without familiarity with this property. For instance, being bipolar is salient when you are working as a psychiatrist, but not when you are working as a physicist. Home runs are salient with respect to baseball, but not with respect to cricket. Certain properties within these frameworks are also versatile, which means that they are widely applicable with respect to the relevant activity. For example, batting is highly versatile with respect to baseball. Salience and versatility can come apart. Home runs are highly salient to baseball, but not particularly versatile, whereas batting is more versatile with respect to baseball as it is more widely applicable.

With these notions of property framework, salience, and versatility on the table, we can now formulate the account of c-naturalness:

As mentioned before, ‘activity’ should be understood generously, such that practicing a branch of the sciences, playing a sport, playing a particular kind of music, cooking, doing philosophy, and negotiating certain social landscapes all count as activities. I do not intend to offer strict conditions for an activity other than to suggest that if it is something you can do that is governed by some rules and conventions, implicit or explicit, and it brings with it a framework of properties, then it is most likely an activity. Activities can be related to one another in various different ways, which can be reflected in their sharing c-natural properties. Properties are particular to activities insofar as they are among the set of properties you have to learn about when you learn to perform this activity. To put this in a slightly different way, the properties that are candidates for c-naturalness are the ones in the framework of properties specifically associated with the activity. However, some activities may share certain properties, such as when batting is shared across cricket and baseball and roux is shared across French and Cajun cooking.

C-naturalness and metaphysical naturalness are very different phenomena. However, one central role of c-naturalness is analogous to a role played by metaphysical naturalness: the c-natural properties in a given context will feature in explanations of and generalizations about phenomena particular to that context. With metaphysical naturalness this worked through the connection between metaphysical naturalness and laws of nature because according to Lewis the perfectly natural properties feature in the laws, which in turn feature in explanations (Lewis Reference Lewis1983; Sider Reference Sider2011: ch. 3). Analogously, there are lawlike truths about activities, and lawlike generalizations about society mirror the structure of lawlike generalizations about the natural world. The properties that feature in these lawlike truths will be the more c-natural properties with respect to those activities. We should expect the more c-natural properties to feature in lawlike generalizations about activities and in explanations of features of those activities because the salience and versatility of c-natural properties reflects the fact that they capture the structure of the activities.

I have been exploiting a comparison between c-naturalness and metaphysical naturalness, but these are importantly different phenomena. One way to understand the differences between them is to look at the extent to which c-naturalness and metaphysical naturalness may coincide. If a property is fairly metaphysically natural, then this is true regardless of context, but a property is only highly c-natural relative to a particular activity. For certain activities, such as metaphysics or fundamental physics, the c-natural properties may turn out to also be metaphysically natural, but for many activities this will not be the case. For example, if we are negotiating a social landscape, then certain highly non-metaphysically natural social categories will be c-natural. A framework of metaphysically natural properties is salient and versatile only with respect to very limited areas of inquiry, such as metaphysics and certain parts of physics; therefore, although c-naturalness and metaphysical naturalness can coincide, we should not expect all or even most of the c-natural properties with respect to most activities to be metaphysically natural.

C-naturalness is independently plausible and useful for a variety of purposes (Taylor Reference Taylor2016). I propose that one useful application of c-naturalness is to social categories, as follows:

At first pass, this application captures a number of features of social categories, including the fact that social categories are indispensable to explanations of and generalizations about phenomena in particular social contexts. C-naturalness comes on a spectrum, which captures the idea that some social categories may be more significant for explanation and generalization than others. Furthermore, which social categories are c-natural will vary across different social contexts, and so we should expect social categories to vary in their importance or cease to apply when we move across different social contexts.

A c-naturalness approach to social categories can happily accommodate the ‘empty’ social categories that posed problems for the metaphysical views discussed in section 2. For instance, consider the astrological case. ‘Pisces’ is highly c-natural with respect to the practice of astrology. The practice of astrology is based on false beliefs about the world, but relative to that practice, the category ‘Pisces’ is highly c-natural, and so it is appropriate that the account of c-naturalness captures this fact. That a property is c-natural with respect to an activity does not make that activity scientifically grounded or intellectually reputable, and we have other resources for deciding upon such issues. This is also the case for many other empty categories, such as those that feature in historical science. For instance, ‘choler’ is fairly c-natural with respect to certain historical branches of medicine, and yet ‘choler’ cannot be either a natural kind or a reasonably joint-carving property.

Understanding social categories in terms of c-naturalness permits us to avoid metaphysically interpreting the role of such categories in inquiry, and so this approach is compatible with nonunified social ontologies as well as with unified approaches. Empty social categories cannot be given a metaphysical interpretation, but other social categories may warrant such treatment. General treatments of the nature of social categories offer a unified approach to social categories, incorporating various strategies for accommodating differences between cases, such as whether group membership is voluntary (Ásta 2018; Epstein Reference Epstein2017; Epstein Reference Epstein2015; Gilbert Reference Gilbert1989; Ritchie Reference Ritchie2013, Reference Ritchie2018). However, there are also specific treatments of particular social categories, including most notably accounts of race and gender categories, and some of these recommend treating those categories as metaphysically different from other social categories (Haslanger Reference Haslanger2000; Jenkins Reference Jenkins2016; Mills Reference Mills2000; Witt Reference Witt2011). On the c-naturalness approach the metaphysics of social categories is not determined by their role in inquiry, but this does not preclude adopting a metaphysical approach to certain social categories in much the same way that a property's being c-natural does not preclude it from also being fairly joint-carving.

The c-naturalness approach also combines well with a variety of different general views on the nature of social groups so long as those views do not take the metaphysics of social categories to be settled by their role in inquiry. For example, the conferralist framework defended by Ásta can be combined with this approach (Ásta 2018). On this view, social categories are conferred on the basis of the belief that the conferral tracks a certain base property. However, all that is required for the conferralist account is that the base property is believed to be there, not that it actually be present. This covers empty social categories, in which social significance grows up around a mythical, absent base property.

5. Concluding Reflections on the Limits of Metaphysics

Throughout this discussion I have been using the term ‘metaphysics’ so as to exclude c-naturalness as a metaphysical phenomenon. However, perhaps this is misguided. Recently there have been debates about the nature and limits of metaphysics, focusing on the legitimacy of social and feminist metaphysics. For example, Elizabeth Barnes has argued that according to some popular conceptions of metaphysics, social metaphysics, including feminist metaphysics, is impossible (Barnes Reference Barnes2014, Reference Barnes2017). Barnes points out that, according to Sider, metaphysics is concerned with questions about the fundamental structure of reality, and the substantivity or otherwise of metaphysical debates depends at least in part on whether the options being debated are equally fundamentally joint-carving. Jonathan Schaffer (Reference Schaffer2017), on the other hand, adopts a conception of metaphysics as primarily concerned with questions about what is fundamental and how the nonfundamental connects to the fundamental. Barnes argues that Sider's and Schaffer's conceptions of metaphysics leave no room for social metaphysics. Mari Mikkola has also documented what she describes as the ‘antagonism’ between mainstream metaphysics and feminist metaphysics (Mikkola Reference Mikkola2017). Sider and Schaffer have both responded to Barnes and Mikkola, arguing that their conception of metaphysics does accommodate social metaphysics (Sider Reference Sider2017; Schaffer Reference Schaffer2017).

Adopting a c-naturalness approach to social categories will not settle questions about whether or not metaphysics must be concerned with joint-carving structure and fundamentality and about whether social metaphysics is sufficiently joint-carving or sufficiently fundamental to count as legitimate metaphysics on such a picture. However, this case does provide an example (among many others) of what it could be like to use metaphysically grounded tools to address questions about social structure. C-naturalness is grounded in and borrows from Lewis's ideas about metaphysical naturalness, and c-naturalness captures a certain kind of context-dependent structure that is analogous to the metaphysical structure captured by metaphysical naturalness.

One might worry that the use of the term ‘structure’ here is inappropriate or that the kind of structure in question is simply too contingent, too flimsy, and too mind-dependent for investigation of it to count as genuine metaphysics. These are substantive concerns, and an interesting feature of social metaphysics is that it raises deep metametaphysical issues, such as how to understand claims about structure and whether mind-independence is a criterion for existence. However, getting facts about structure right in order to formulate better explanations, predictions and interventions sounds very much like metaphysics. Furthermore, the parallels between the cases of metaphysical naturalness and c-naturalness in terms of lawlike generalization and explanation show that this comparison is useful and illuminating. That is good reason to think of this application of c-naturalness to social categories and of discourse about c-naturalness in general as metaphysical inquiry. Doing so gives us an example of what it might look like to work with a more open understanding of metaphysical practice.