2.1. Special Purpose Kinds

While ethnobiology is a comparably young field that became institutionalized in the second half of the twentieth century, the documentation of biological knowledge has a long history in anthropological and ethnographic research (Hunn Reference Hunn2007; Svanberg and Łuczaj Reference Svanberg and Łuczaj2014). Much of this earlier research presented “primitive” biological knowledge as dominated by immediate practical needs. As Malinowski put it in a famous passage of Magic, Science and Religion: “The road from the wilderness to the savage’s belly and consequently to his mind is very short. For him the world is an indiscriminate background against which there stands out the useful, primarily the edible, species of animals and plants” (Reference Malinowski1948, 44). In the context of biological classification, this contrast suggested ethnotaxa that are shaped by practical concerns (e.g., classifications along criteria such as being dangerous, edible, medically useful, poisonous, or sacred) and therefore diverge from modern biological taxa that are distinguished on the basis of more general criteria such as the ability to interbreed.

Following Berlin, Breedlove, and Raven’s (Reference Berlin, Breedlove and Raven1966) later terminology of “special purpose” and “general purpose” kinds, the early utilitarian research assumed that ethnobiological knowledge is largely characterized by its “special purpose” character, while modern biological systematics exhibits a “general purpose” structure. To make this notion more specific for present purposes, let us define special purpose kinds as kinds whose extension is directly dependent on properties of human use such as being edible or poisonous. In this sense, edible mushroom would be a special purpose kind while bolete mushroom would not. It may be the case that all bolete mushrooms are edible in a certain environment and that the category is of obvious practical significance. However, only the extension of the former is directly dependent on use. Imagine that a nonedible bolete mushroom is introduced in the environment. While this mushroom would obviously be excluded from edible mushroom, it would still be included in bolete mushroom.

Special purpose kinds in the sense of this definition are well documented in many areas of ethnobiology. In the case of mushrooms, for example, Lampman has shown that Tzeltal Maya in Chiapas use a very general category of useless mushrooms bol lu’ that are otherwise highly disunified. While special purpose kinds can appear as very broad categories such as “useless mushroom,” they can also have a more specific character. Lampman (Reference Lampman2004, 191) points out that special purpose categories could result from an “allergic reaction to certain types of plant pollen; or they could arise from cultural convention, such as categories of plants that are appropriate for the ceremonies associated with the transition from childhood to adulthood.” Given this diversity of potentially relevant properties, we should expect special purpose kinds also to come with varying degrees of nonarbitrariness. In some cases, such as “useless mushrooms,” there may not be much more to say about the unification of members of the kind. In other cases, such as allergies or medical use, usefulness may indicate the presence of other (e.g., chemical) properties that are shared by members of a special purpose kind.

2.2. General Purpose Kinds

The emergence of ethnobiology as a field with a distinct identity is closely related to both theoretical and empirical criticism of earlier utilitarian perspectives. Levi-Strauss (Reference Levi-Strauss1966, 6) challenged utilitarianism with an “intellectualist” perspective according to which classification “meets intellectual requirements rather than … satisfying needs.” Furthermore, ethnobiological fieldwork clearly showed that a strict utilitarian perspective failed to account for the reality of ethnobiological classification. Conklin’s (Reference Conklin1954) dissertation is widely regarded as a watershed moment in ethnobiology as it provided a detailed systematics of Hanunoó plant classifications and included countless taxa that bore no resemblance to Malinowski’s “short road to the savage’s belly.” As Conklin summarized his results: “Hanunoó classify their local plant world, at the lowest (terminal) level of contrast, into more than 1800 mutually exclusive folk taxa, while botanists divide the same flora—in terms of species—into less than 1300 taxa” (Reference Conklin and Fishman1962, 426).

Not only Western science but also ethnobiology seemed to require acknowledgment of a wide variety of general purpose taxa that are not directly shaped by practical use. In contrast to the last section, we can define general purpose kinds as kinds whose extension is not directly dependent on properties of human use (e.g., edible or firewood) but rather on general biological (e.g., morphological and ecological) properties. Following Conklin’s groundbreaking studies, ethnobiologists of the 1960s and 1970s largely inverted the utilitarian account of the relation between epistemic and nonepistemic significance of kinds. Utilitarian accounts start with considerations of nonepistemic factors while allowing that special purpose kinds may also be epistemically significant by sharing many other properties (e.g., medical use can track chemical differences between plants). In contrast, intellectualism starts with epistemic considerations such as the recognition of “morphologically coherent groups” (Berlin, Breedlove, and Raven Reference Berlin, Breedlove and Raven1968, 296) while allowing that the identification of such groups will be useful from a nonepistemic perspective.

2.3. Mind-Independent Convergent Kinds

My definitions of special and general purpose kinds have been minimal in the sense that they only depend on the properties that determine the extension of a kind. From a philosophical perspective, one may wonder whether this is a metaphysically deep distinction. For example, edible mushroom would count as a special purpose kind, while mushroom that grows on wood would count as a general-purpose kind. However, it is not clear that we should consider the latter metaphysically more substantial or more natural than the former. On the contrary, we can easily imagine scenarios in which use-dependent (e.g., medical) similarities reflect further substantial similarities, while use-independent (e.g., color) properties do not lead to the identification of a more substantially unified kind. And indeed, intellectualist ethnobiologists often aimed for something much more ambitious than general purpose kinds in the sense of the last section. Rather than pointing out that the extension of ethnotaxa is often determined on the basis of use-independent (e.g., morphological and ecological) properties, an important part of the early ethnobiological project was the identification of natural kinds in a traditional “mind-independent” sense.

Much of this more ambitious program was motivated by the discovery of convergences between ethnotaxa and modern biological taxa that seemed to directly contradict classificatory relativism that dominated the anthropological tradition. For example, consider Diamond’s classic article “Zoological Classification System of a Primitive People” that found convergence between vertebrate categories (ámana aké) of the Fore of the New Guinea Highlands and taxa in biological systematics. Diamond concluded that this convergence illustrates the objectivity of ethnotaxa: “The nearly one-to-one correspondence between Fore ámana aké and species as recognized by European taxonomists reflects the objective reality of the gaps separating sympatric species” (Reference Diamond1966, 1102).

While the relation between intellectualist ethnobiology and natural kind philosophy of the 1970s (e.g., Kripke Reference Kripke, Davidson and Harman1972; Putnam Reference Putnam1973) is yet to be discussed in the research literature, the similarities are obvious. Indeed, many ethnobiologists of this time embrace a traditional perspective on natural kinds, from Bulmer’s (Reference Bulmer1970, 1087) claim that the taxonomic “hard core … simply has to be ‘general’ or ‘natural’ and consist of multi-purpose, multidimensional units which bear a definite correspondence to those applied by the biological scientist” to Berlin’s (Reference Berlin1992, 9) assumption that “in any local flora and fauna a single pattern stands out from all the rest” and Hunn’s (Reference Hunn1987, 147) insistence that “individuals of phylogenetically real groupings share a genetic essence.” In developing a typology of ethnobiological kinds, we can therefore define a class of mind-independent convergent kinds that follows the traditional philosophical picture of the discovery of mind-independent essences or “joints in nature” that ultimately also lead to convergence across cultures from indigenous societies to modern biological systematics.

2.4. Cognition-Dependent Convergent Kinds

While essentialist models of natural kinds soon became criticized in philosophy (e.g., Mellor Reference Mellor1977; Dupré Reference Dupré1981), ethnobiological criticism of the assumption of mind-independent convergence developed largely independent of the philosophical literature. One major challenge came from the increasing influence of the cognitive sciences and more specifically from the assumption that cross-cultural convergence can reflect general characteristics of cognitive processing rather than mind-independent discontinuities in nature. As Atran (Reference Atran1981, 64) put it: “How comes it that cultures belonging to such widely separated times and places invariably produce similar basic groupings? The only conceivable answer is that they employ identical cognitive processing over similar empirical domains.”

Atran’s point becomes especially plausible in the context of taxa that are found in many causally unrelated cultures around the world but are not recognized by a biological systematics that relies on phylogenetic relations. For example, taxa such as “thistle,” “tree,” “vine,” “bug,” “hawk,” or “sparrow” are widely recognized as kinds with coherent morphological and ecological forms, even if they are not recognized in biological systematics. “People naturally tend to find trees phenomenally compelling because of their evident ecological role in determining local distributions of flora and fauna. … Just go into a forest and see” (Atran Reference Atran1987, 150). Given the importance of general cognitive and perceptual mechanisms (see also Atran and Medin Reference Atran and Medin2010), we can define another class of cognition-dependent convergent kinds that are found in many causally unrelated cultures because of the interaction between biological (e.g., morphological and ecological) properties and cognitive mechanisms such as perceptual salience. The point is not that cognition-dependent convergent kinds are merely an invention of our own mind but rather that general cognitive mechanisms contribute to the classificatory relevance of properties. Using Reydon’s (Reference Reydon and Kendig2015) recent notion of “cocreation,” they may be best described as cases in which empirically discovered properties of organisms and distinctly human cognitive mechanisms “cocreate” a category.

2.5. Practice-Dependent (Convergent or Divergent) Kinds

Cocreation in the sense of the last section relies on (e.g., morphological and ecological) properties that are recognized as constituting a relevant category in the context of species-specific human cognition. However, the relevance of properties can also depend on cultural rather than general cognitive mechanisms. For example, Hunn (Reference Hunn1982, 837) suggested that “tree” may be recognized cross-culturally because of the “universal practical value of ‘trees’ rather than … the perceptual salience of ‘tree.’”

While the importance of cultural and cognitive factors has been highly controversial in the context of cross-cultural recognitions of “tree” (see Ellen Reference Ellen2006), there are also more straightforward examples for practice-dependent kinds such as Estrada-Medina et al.’s (Reference Estrada-Medina2013) research on classifications of soil types in Yucatán, Mexico. Estrada-Medina et al. found that local Mayan farmers identified soil types that have no extensional equivalent in Western soil taxonomies such as the World Reference Base for Soil Resources (WRB): “Many soils identified by farmers could be more than one WRB unit of soil and vice versa; in these cases, no direct relationship between both classification systems was possible” (1). Mayan classification of soil types uses a variety of properties including color, amount of gravel and rocks, hardness, drainage conditions, and soil depth. Rather than simply classifying soils along perceptually salient forms (as in case 2.4), Mayan classification of soil types reflects distinctions that matter for local agricultural practices such as “making milpa” (a traditional crop rotation system that includes corn, bean, and pumpkin). For example, Chichlu’um and Ch’och’ol constitute two different Mayan taxa of soil but would both be classified as Hyperskeletic Leptosol according to the WRB. However, Ch’och’ol is used for making milpa, while Chichlu’um is primarily used to grow chilli and sweet potatoes. We can therefore define another class of practice-dependent kinds that reflect interaction between empirically discovered properties (e.g., soil color, amount of gravel and rocks, hardness, drainage conditions, soil depth) and their practice-dependent relevance for classification. Note that a characterization of Chichlu’um as “practice dependent” does not imply that it is a “special purpose kind” in the sense of 2.1. The extension of the soil type is determined on the basis of properties such as fine gravel and brown color rather than directly in terms of agricultural use (say “soil used for growing pumpkins”).

Practice-dependent kinds can converge or diverge across cultures depending on the continuity of relevant practices. In the case of soil types, for example, classification seems to depend on local practices such as making milpa and the relevance of specific crops such as chilli and sweet potatoes. However, practice-dependent ethnotaxa can also converge insofar as communities share similar practices in areas such as agriculture and hunting. Furthermore, there is a lot of middle ground between the strictly local and the strictly universal. For example, making milpa is an agricultural practice that is used throughout Mesoamerica and may therefore be described as regional rather than locally restricted to Yucatán or universal. In analogy to the cognition-dependent convergent kinds from the last section, the contingency of practice-dependent kinds should not be confused with arbitrariness (Ludwig Reference Ludwig2016a). Even if a taxonomy is shaped by the concerns of a local community, it will still rely on empirically discovered (relations between) properties such as soil color, amount of gravel and rocks, hardness, drainage conditions, and soil depth.

2.6. Environment-Dependent Divergent Kinds

The locality of Mayan soil taxa can be partly explained in terms of equally local agricultural practices that influence the taxonomic relevance of properties and patterns. However, there is a second mechanism that can lead to taxonomies that are specific to a local community. Local communities do not only engage in different practices but also with different environments that exhibit different regularities. Even if two communities had exactly the same epistemic and social goals, they may still employ different taxonomies simply because there are different properties and patterns to be found in their respective environments.

The example of soil types can also illustrate this case, as the Mayan soil taxonomy in Yucatán clearly reflects the distinct soil profile of the Yucatán peninsula that involves “repetition of four geomorphic systems all over the area: coastal, karstic, tectono-karstic, and fluvio-paludal, each one showing specific soil-relief patterns” (Bautista and Zinck Reference Bautista and Zinck2010, 9). Even in other areas of Mesoamerica where Mayan agricultural practices such as making milpa are common, soil taxonomies can differ because soil profiles differ and communities therefore engage with different clusters of properties. This variability of taxonomies leads to a further type of environment-dependent divergent kinds that can be defined through their reliance on properties and patterns that are stable only within the context of a specific environment. While this locality of properties and patterns can lead to equally local taxonomies, environment-dependent divergent kinds are obviously far from arbitrary. For example, Chichlu’um comes with a cluster of soil properties from color to drainage conditions that is empirically discovered and stable enough to support agricultural practices even if it is limited to a specific environment such as the Yucatán peninsula.

2.7. Biosocial Kinds

Biological kinds often have social (e.g., economic or spiritual) properties. In many cases, however, the social properties are not themselves sources of taxonomic distinctions (Atran Reference Atran1990, 69). For example, jaguars have a range of spiritual properties across indigenous cultures of the Americas. However, jaguars would arguably be also recognized as a distinct kind if they did not have any spiritual significance whatsoever. In contrast to biological kinds that also happen to have social properties, we can define a more restricted class of ethnotaxa that are recognized only because they also have social properties.

To illustrate biosocial kinds in this narrower sense, consider Bulmer’s (Reference Bulmer1967) classic discussion of the taxonomic status of cassowaries. According to Bulmer, the Kalam (in 1967 still spelled as “Karam”) in the New Guinea Highlands employ a taxon yakt that includes all birds except cassowaries—large flightless birds that resemble ostriches and emus. Why are cassowaries not considered birds? Part of the answer includes behavioral, ecological, and morphological properties that are shared among typical birds but not cassowaries. As Bulmer explains, cassowaries are characterized by a distinct terrestrial habitat that contrasts with the aerial and arboreal habitat of other birds. This not only implies behavioral and ecological differences between cassowaries and birds (yakt) but also affects anatomical and morphological properties such as “heavy, strong and very human-like leg bones” (Bulmer Reference Bulmer1967, 10). However, Bulmer insists that the exclusion of cassowaries from the bird taxon yakt cannot be purely understood in terms of biological (anatomical, behavioral, ecological, morphological) properties but requires consideration of the unique social relationship between cassowaries and the Kalam community: “The cassowary is not a bird because it enjoys a unique relationship in Karam thought to man” (5), and the exclusion of cassowaries from yakt “by reference to objective features of its appearance and behaviour alone, could be to miss the point” (19). As Bulmer describes in detail, Kalam understanding of cassowaries is embedded in a complex system of social rules that puts cassowaries closer to humans than birds. For example, killing a cassowary is considered in many ways similar to homicide. In analogy to humans, blood of living cassowaries must not be shed, and the hunt therefore requires blunt weapons and is conceptualized in analogy to the struggle between two humans. Hunters who have killed a cassowary enter a ritually dangerous state asn that involves further rules such as avoidance of important crops for 1 month. The same state of asn is entered by killing a human but not by killing yakt (birds).

Bulmer’s detailed description of these kinds of social relations illustrates another important class of ethnobiological kinds that have to be at least in part understood in terms of their social properties. In contrast to biological kinds that happen to have social properties, biosocial kinds can be defined more narrowly by requiring that not only biological but also social properties are constitutive for a kind. Following Bulmer, yakt turns out to be a biosocial kind in this sense, as the exclusion of cassowaries cannot be understood on the basis of biological properties alone. This does not mean that biosocial kinds are exclusive to ethnobiology, and one may find similar phenomena in areas of the life sciences with important social implications. For example, consider debates about classification and species concepts in microbiology. It has become widely recognized in the literature that traditional species concepts are of limited applicability in microbiology and that microbial species often come with vast genetic and phenotypic diversity (Franklin Reference Franklin2007; Ereshefsky Reference Ereshefsky2010; O’Malley Reference O’Malley2014). This internal heterogeneity suggests that at least some microbial species may qualify as biosocial kinds in the sense that their recognition as species depends in part on their social significance for medical practice. Indeed, this position is embraced by Suárez (Reference Suárez2016) who explicitly argues for the legitimacy of medical concerns in delimiting bacterial species.

3.1. Success and Restriction Clauses

Magnus (Reference Magnus2012) has proposed an account of natural kinds in terms of categories that are uniquely suited for inductive and explanatory success in a given domain. More specifically, Magnus distinguishes between (1) a success clause that requires inductive and explanatory success in a domain of inquiry and (2) a restriction clause that requires that alternative taxonomies would be less successful. While Magnus’s account adopts some traditional ideas from the natural kind literature such as the focus on induction, it also expands the realm of natural kinds through its emphasis of domain relativity. By insisting that natural kinds have to be understood relative to a domain of inquiry, the framework includes categories with specialized purposes that would be excluded in many other accounts of natural kinds.

This implication is nicely illustrated by Magnus’s discussion of Alton Brown’s taxonomy of baked goods: “For example, traditional muffins, soda bread, pancakes, and waffles are all prepared using what Brown calls the muffin method. Biscuits, scones, shortcakes, grunts, and pie crusts are prepared using the biscuit method. Importantly, these methods are not merely a matter of preparation. They also correspond to different ways of leavening. In all the muffin method baked goods, air bubbles are generated by baking soda reacting with acid. In biscuit method baked goods, bubbles are generated by small bits of solid fat which melt during baking” (Magnus Reference Magnus2012, 133). The idea that muffin or biscuit may qualify as natural kinds sounds radical but is tied to nontrivial conditions. The point is not only that muffins and biscuits have different (e.g., chemical) properties that are outlined by Brown but that the categories must meet Magnus’s success and restriction clauses. For example, the category muffin would not only have to support the practices of bakers but would also have to be more successful than alternative taxonomies of baked goods. The case of baked goods illustrates that natural kinds in the sense of Magnus may be found in unexpected places, but it does not reduce to triviality.

The domain relativity of Magnus’s success and restriction clauses makes them helpful tools for analyzing local ethnotaxa that are unique to specific communities in the sense of 2.5–2.7. Local ethnotaxa fail to meet many traditional criteria of natural kinds, as they are typically not mind or interest independent, lack essences, and diverge from the categories of Western science. While this locality of ethnotaxa often leads to their disqualification (Ludwig Reference Ludwig2016b), Magnus’s account will recognize many of them as natural kinds. For example, recall the case of Mayan soil types such as Chichlu’um and Ch’och’ol. While these kinds of soil do not correspond to kinds in Western soil taxonomies such as the WRB, they may still meet Magnus’s criteria. After all, the WRB aims for a classification that is useful for pedologists in a large variety of environments and does not pay attention to the properties that are particularly relevant for Mayan farmers in Yucatán. Again, the point is not only that soil kinds such as Chichlu’um are characterized by empirically determined properties such as fine gravels and good water retention. Instead, Magnus’s success and restriction clauses require that taxa such as Chichlu’um are uniquely suited for inductive and explanatory success in the context of Mayan agriculture such as predicting which crops will successfully grow. As such, Magnus provides nontrivial criteria for engaging with local ethnotaxa and for distinguishing them from arbitrary classifications.

3.2. Categorical Bottlenecks

Magnus’s success and restriction clauses provide helpful tools for engaging with local ethnotaxa in the sense of 2.5–2.7. If we turn from issues of locality to cross-cultural convergence, the restriction clause may also help to understand cases that I have described as “practice-dependent convergent kinds.” The restriction clause requires that alternative taxonomies would be less successful in a given domain of inquiry. Assuming that shared (e.g., agricultural or hunting) practices constitute shared domains of inquiry, the restriction clause provides a compelling reason to expect cross-cultural convergence. If there is a category that is uniquely suited for specific inductive and explanatory practices, we should expect its recognition across cultures with sufficiently similar domains of inquiry.

Unfortunately, not all cases of cross-cultural convergence can be explained in terms of shared domains of inquiry such as shared agricultural or hunting practices. For example, consider cases such as jaguar that have often been discussed as cases of mind-independent convergent kinds in the sense of 2.3. What distinguishes these kinds is that they are recognized in very different taxonomies that serve very different purposes. For example, consider a child in rural Costa Rica who learns about jaguars as a potential threat and a Western biologist who is interested in jaguars in the context of a research project on phylogenetic relations between felidae. Even if their domains of inquiry could hardly be more different (understanding personal danger or phylogenetic relations), they will both recognize jaguars as a distinct kind. In this context, is seems more helpful to look at other proposals such as Franklin-Hall’s (Reference Franklin-Hall2015) account of natural kinds as categorical bottlenecks. Franklin-Hall’s general idea is that natural kinds can be identified with categorical bottlenecks that constrain conceptual options and therefore lead to recognition of the same categories among very different actors. More specifically, she suggests to identify natural kinds with “categories that well serve not only us, but also a variety of agents possessing any of a range of aims and capacities, a range that—though ultimately relative to us—extends well beyond those we presently enjoy” (926).

Franklin-Hall’s account of categorical bottlenecks is clearly more restrictive than Magnus’s success and restriction clauses in the sense that it excludes many local ethnotaxa that meet the unique epistemic aims of specific communities. However, it captures an aspect of “stance-independence” (Franklin-Hall Reference Franklin-Hall2015, 940) that is important for engagement with ethnotaxa such as jaguar that have been traditionally conceptualized as mind-independent convergent kinds in the sense of 2.3 and cannot be completely explained in terms of shared cognitive biases in the sense of 2.4 or shared practices in the sense of 2.5. Given that most philosophers have become uncomfortable with the strong metaphysical ideal of “mind-independent convergence,” Franklin-Hall’s categorical bottlenecks provide an alternative approach for analyzing surprising cases of cross-cultural taxonomic convergence. On the one hand, categorical bottlenecks are stance independent to a certain degree because they reflect the “robust cluster structure of our universe” (938). On the other hand, categorical bottlenecks are still defined relative to (a range of) epistemic aims and “ultimately relative to us” (927). Franklin-Hall’s categorical bottlenecks therefore provide a helpful tool for engaging with ethnobiological evidence about cross-cultural convergence without getting stuck with overly ambitious ideals of complete mind independence or with attempts to reduce convergence to shared cognitive structures and practical interests in the sense of 2.5 and 2.6.

3.3. Stable Property Clusters

Magnus’s and Franklin-Hall’s accounts focus on the epistemic features of natural kind categories. Rather than engaging in “deep metaphysics” (Magnus Reference Magnus2015), these accounts assume that natural kinds correspond to categories with certain epistemic features. As Franklin-Hall (Reference Franklin-Hall2015, 940) puts it: “natural kinds are groupings that match those categories that well serve actual inquirers along with (what I will call) ‘neighboring agents.’” Starting with epistemic features of natural kind categories can be helpful in analyzing ethnobiological classification by identifying subsets of ethnotaxa with unique epistemic features. While success and restriction clauses identify a larger set that includes many local ethnotaxa in the sense of 2.5–2.7, categorical bottlenecks identify a more restricted set of taxa that converge across cultures in the sense of 2.3.

However, many accounts in the natural kind literature develop a different strategy by also aiming for characterizations of the metaphysical features of natural kinds. Given that few contemporary philosophers of science endorse traditional natural kind essentialism, much of this literature starts with issues of property clustering as popularized by Boyd’s theory of homeostatic property clusters (Reference Boyd1991, Reference Boyd2010). For example, Slater’s (Reference Slater2015) account of stable property clusters (SPCs) provides an inclusive metaphysical framework for “natural kindness” that maintains the common focus on “property clustering” while avoiding metaphysically more ambitious assumptions about homeostatic mechanisms. Slater’s SPC account provides a complementary tool for engaging with ethnobiological classification by identifying a large set of ethnobiological kinds that involve stable clustering of properties.

As Ludwig (Reference Ludwig2017) has argued, such an inclusive SPC account is helpful for making sense of convergence-divergence patterns in ethnobiology. Consider a kind that comes with a stable cluster of properties such as [P, Q, R, S, T, U] and therefore allows probabilistic inferences from one subcluster such as [P, Q, R] to another subcluster such as [S, T]. Following the examples from sections 2.5–2.7, we should assume that many local taxa such as the Mayan soil type Chichlu’um constitute SPCs in Slater’s sense and therefore also support probabilistic inferences. For example, knowledge about some easily observable properties of soil, such as brown color, location on top of a mound, a lot of fine gravel, may be sufficient to make probabilistic inferences to other properties, such as good water retention, suitable for growing chilli. This SPC analysis is entirely compatible with locality in the sense that these inferences may only matter in the context of certain agricultural practices (2.5) and may only be stable in the context of the Yucatán peninsula (2.6).

While SPCs can be helpful to shed light on local ethnotaxa, they can also be used as a tool for explaining convergence beyond a local context. Consider a cluster [P, Q, R, S, T, U] but two different communities that have knowledge about different subclusters such as [P, Q, R, S] and [R, S, T, U]. In the case of jaguar, for example, it may be that both indigenous and Western scientists have knowledge of certain morphological properties [R, S], only the indigenous scientist has knowledge of certain ecological properties [P, Q], and only the Western scientist has knowledge of certain genetic properties [T, U]. Given that they have knowledge about different subclusters of the same SPC, however, it is not surprising that their taxonomies converge (Ludwig Reference Ludwig2017).

Although Slater’s account does not include all ethnotaxa in the sense of 2.1–2.7 (e.g., some special purpose kinds and cognition-dependent kinds may not involve SPCs), the inclusive character of SPCs makes them helpful tools for understanding convergence-divergence patterns in ethnobiology. While some highly robust SPCs such as jaguar are recognized from very different cultural perspectives, other SPCs such as Chichlu’um turn out to be more restricted.

3.4. Nodes in Causal Networks

While Slater’s inclusive notion of SPCs provides a helpful tool for engaging with diverse convergence-divergence patterns, there are also often good reasons to look beyond mere clustering by focusing on the causes of stable relations between properties. Khalidi has recently proposed an alternative account of natural kinds as “nodes in causal networks” that are “not just concatenations of properties but are ordered hierarchies of properties” (Reference Khalidi2015, 17). The basic idea is that natural kinds require primary properties that give rise to further secondary properties. In the classic case of gold, for example, Khalidi suggests that “primary causal properties … include atomic number 79 as well as a disjunction of mass numbers, which give rise in turn to a cluster of other causal properties (e.g. ionization energies, atomic radius, etc.)” (18).

Khalidi’s causal account of natural kinds reflects the assumption that a simple clustering account is “found wanting mainly because it leaves out the relationships that exist between the properties associated with a kind” (Reference Khalidi2013, 204). To illustrate this in the context of ethnobiology, consider the observation that many cultures recognize a taxon that includes both bats and birds. While philosophers may be inclined to consider such a bat-bird taxon as only superficially unified, Ludwig (Reference Ludwig2017, 203) argues that bats and birds actually share a large range of properties: “they fly, have wings, a light bone structure, a keeled sternum, a similar size range, streamlined bodies, high metabolism, migratory behaviour, similar natural enemies, a fruit- or insect-based diet, they spread seeds in the environment, reduce local insect biomass, and so on.” While a bat-bird taxon may therefore be described as involving some “natural kindness” in the sense of Slater’s SPCs, such an analysis leaves the existence of a phylogenetically heterogeneous property cluster unexplained. Looking for nodes in causal networks in the sense of Khalidi leads to a more substantial understanding of this classificatory practice as it can causally explain shared anatomical, behavioral, ecological, and morphological properties of bats and birds through shared environmental pressures and adaptation to a flying lifestyle.

While Khalidi’s account of natural kinds as nodes in causal networks can often provide a more substantial understanding of ethnotaxa, it is also more demanding in the sense that it arguably excludes a number of kinds that qualify as SPCs in the sense of Slater. As Khalidi points out (Reference Khalidi2015, 2), “clusters of properties [are sometimes] conventionally rather than causally related” and therefore do not constitute nodes in causal networks. Biosocial kinds in the sense of 2.7 will be sometimes at least partially conventional. For example, recall the exclusion of cassowaries from the bird taxon yakt (sec. 2.7) because of both biological differences (e.g., bone structure) between cassowaries and other birds as well as the specific social and spiritual significance of cassowaries. While we could construct a cluster of biosocial properties [P, Q, R, S, T, U] in the sense of the SPC model, such a cluster would exist at least in part because of conventional rather than causal relation between biological and social properties.

Furthermore, locally stable ethnotaxa in the sense of 2.6 may turn out to be SPCs in the sense of Slater but not qualify as nodes in causal networks in the sense of Khalidi. This possibility is especially obvious in the context of higher-order categories that group more specific ethnotaxa together (e.g., Berlin Reference Berlin1992, 138ff.). For example, consider an ethnomycological taxon that includes several more specific mushroom kinds. In addition to morphological properties (e.g., color, size, and form of the stipe), such a mushroom taxon may be useful because it also allows probabilistic inferences about where and when to find them (e.g., they tend to grow under pine trees and only between September and December) as well as what to do with them (e.g., they are all edible, but none of them have medical use). While such a category would qualify as an SPC in the sense of Slater, it does not have to reflect causal hierarchies in the sense of Khalidi. Of course, it could be the case that the mentioned similarities all reflect some underlying primary property and that Western taxonomists recognize a coextensional taxon due to phylogenetic relations between the included mushroom species. However, it could also turn out that a phylogenetic classification would include mushroom species with different ecological and use-dependent properties (say mushrooms that are not edible and do not grow under pine trees). The fact that an ethnotaxon such as a mushroom category constitutes a SPC is not sufficient for its status as a node in a causal network.

To sum up, Khalidi’s nodes in causal networks provide a useful tool for engaging with ethnotaxa both because they focus on causal structures beyond the scope of SPCs and because they identify a somewhat more restricted set of causally structured kinds in ethnobiology.

3.5. Integrating Dimensions of Nonarbitrariness

Let us take stock. I have argued that current accounts of natural kinds provide useful tools for engaging with taxonomic diversity in ethnobiology. The epistemic productivity of local ethnotaxa can often be specified by showing that they meet success and restriction clauses in the sense of Magnus. Taxonomic convergence that has often been associated with a strong metaphysical ideal of mind-independent convergence can be usefully reinterpreted in terms of Franklin-Hall’s categorical bottlenecks. Slater’s model of SPCs provides a helpful tool for understanding cross-cultural convergence-divergence patterns. Finally, Khalidi’s account of nodes in causal networks identifies an important subset of SPCs that are hierarchically structured and causally unified through primary properties.

While each of the discussed proposals provides helpful resources for analyzing ethnobiological classification, none of them will be preferable in all relevant contexts. For example, a broader account in terms of Magnus’s success and restriction clauses will not be adequate for analyzing many forms of cross-cultural convergence, while a more narrow account such as Franklin-Hall’s categorical bottlenecks will be of limited use for analyzing local taxa that are adapted to needs of a specific community. None of this is a problem, however, if the different accounts are interpreted not as competitors but as complements in a multidimensional framework of nonarbitrary classification.

It is precisely this specification of multiple dimensions of nonarbitrary classification that makes the recent natural kind literature valuable for understanding complex classificatory landscapes in ethnobiology. If philosophical accounts of natural kinds preselect one dimension as fundamental, they will often be too static to account for the diverse practices and criteria in ethnobiological classification. By recognizing accounts in the natural kind literature as complementary, however, we gain a more nuanced framework for analyzing classificatory practices. In the context of ethnobiological classification, researchers have often struggled with a polemical contrast between “universalist” perspectives that focus cross-cultural convergence in the sense of 2.3 and “relativist” programs that emphasize locality in the sense of 2.5–2.7 (see Ellen Reference Ellen2006). A multidimensional framework provides resources for moving beyond this simplistic contrast by allowing that an ethnotaxon can be nonarbitrary along some dimensions (e.g., by involving SPCs) but not others (e.g., by lacking categorical bottlenecks).