In this paper I will be reviewing literature, mostly from cognitive science, related to the human brain’s processes of categorization

VI.Why should ethnic groups be, cognitively, living kinds?

No cognitive theory is really complete unless it has a plausible evolutionary story to back it up. I here attempt to provide this theory with such a story. Since my argument relies heavily on the inductive inferences for hidden properties that essentialist thinking promotes in natural living kinds, and particularly at the generic-species level, I first review this aspect of living-kind cognition and then present the evolutionary argument.

A.Inductive inferences in natural kinds

Assume it is true that shared appearances are typical of members in a natural kind, but nevertheless not determinative of membership. Assume also that what is determinative of membership is the ‘essence’, whatever it is. Finally, assume that ‘essences’ are cognized as ‘hidden’: “more than meets the eye,” and therefore causative of a great many hidden properties and characteristics that are common to all category members but which are not immediately apparent. It seems reasonable then, that:⁹

(1) Knowledge that an item is a member of natural kind ‘A’ will lead to the automatic assumption that it has an ‘A essence’.

(2) Learning that this item has hidden property P should lead to the assumption that P is either caused by, or a part of, the essence, and thus to the generalization that P is true of other A’s.

(3) A’s that don’t look like the target item will be thought of as having property P anyway, and non-A’s bearing a strong similarity to the target will nevertheless be thought of as lacking P (unless the property is also specified for the natural kind they do belong in).

In a study involving 3-year-olds Gelman & Markman (1987) provide evidence for the above propositions, and for G&M’s explicit hypothesis that natural kinds will favor categorical rather than appearance-based inductive inferences. Their findings are very similar to what they earlier found for 4-year-olds (Gelman & Markman 1986), and so the 1987 study pushes the developmental threshold back by a year. They tested both animate/biological and inanimate natural kinds. I illustrate the spirit of the experiment by describing the word-and-picture condition.

Children were presented with a member of a named category (e.g. ‘a cat’) which constituted the ‘target’, and were told something about ite.g. “See this cat? This cat can see in the dark” (the children were pretested for naïveté concerning the properties they were informed about to ensure that they were in fact learning something new). Then they were shown other pictures: (1) similar to, and in the same category as, the target (e.g. a black cat with a white stripe down its back, just like the target, but sitting in a different position); (2) dissimilar to, but in the same category as, the target (e.g. a white cat in a different position from target); (3) very similar to, but in a different category from, the target’s (e.g. a black skunk sitting in the same way as the cat); (4) dissimilar, and in a different category from, the target (e.g. dinosaur).

For each of these pictures, the experimenter placed the picture on the table directly below the target and said, “This one’s an X [where X was the label, e.g. cat, skunk, or dinosaur]. See this X? Do you think it can see in the dark, like this cat?” After the child answered, the experimenter picked up the card that she had just talked about and placed the next nontarget card on the table—Gelman & Markman (1987:1536)

Where category membership and appearance supported opposite inductive inferences, children preferred categorical as opposed to perception-based induction.

Children drew more inferences to pictures of the same category but different appearance (e.g. from a leaf-insect [“bug”] to a black beetle [“bug”] than to pictures of similar appearance but a different category (e.g. from a leaf insect to a leaf). (…) This was true for both animal and inanimate domains, so it appears to be a fairly general assumption that children have by age 3.—Gelman & Markman (1987:1537)

It is important to note that children were not forced to choose between category and appearances, and could have made inductive inferences for both. Yet, they still preferred to rely on the categories; this supports proposition (3) at the beginning of this section.

Why do natural kinds promote categorical inductive inferences of non-obvious properties? Gelman & Markman (1986:184-185) observe that natural kind categories have rich, correlated structures, where a great many properties that correlate strongly are not obvious on first inspection, and therefore

extend far beyond our original categorization. For example, giraffes share a particular diet, life expectancy, gestation period, DNA structure, and so forth—attributes that are impossible to know by casual inspection. (…) The highly correlated structure of natural kinds suggests that new features learned about one category member will often be projected onto other category members as well.

A Darwinian unpacking of that statement might go like this.¹⁰ Any animal that relies heavily on learning will benefit by reducing the costs of the learning process. If we can reliably learn about whole suites of objects merely by examining one of them, then––by golly––evolution would have failed us if it didn’t provide mechanisms for doing so. Of course this argument applies to artifact categories (which are also characterized by rich, correlated structures) as much as to natural kinds. What makes induction in natural kinds special is that inductions are easily made for non-obvious (i.e. ‘hidden’) properties, and this is because members of natural kinds in fact do share many non-obvious properties.

Contrast this with, say, artifacts. If I showed you a ceramic pitcher and I told you that it brakes easily (a hidden property), you would be wrong to infer this as true for all pitchers, since they can be made of wood, stone, coconut shells, metal, etc., and, in modern times, out of unbreakable plastic (cf. Gelman 1988). The things members of an artifact category typically share in common is perceptually obvious: their parts and their interconnections are constrained to be similar because they must fulfill the same function; the hidden properties (such as the properties of the materials they are made of) can vary widely so long as the artifact itself fulfills the same function. Thus, if I show you that a ceramic pitcher breaks easily, you will generalize that to ceramic anythings (because ceramic is a ‘substance’); but you will not think of this as a ‘pitcher property’. Evidence for this can be found in Gelman & O’Reilly (1988:881)

Recent evidence (Atran et. al. 1997) supports the hypothesis that, where biological natural kinds are concerned, there is a privileged living-kind category level (or ‘rank’) for the purposes of making inductive inferences. This is the ‘generic-species’ level, which corresponds to the folk-species level in folk taxonomies (although it sometimes corresponds to scientific genera due to the fact that in most locales genera are monospecific and, when they are not, the species are very difficult to tell apart; Atran et. al. 1997:20-21). The generic-species level does not correspond to the ‘basic-level’ category in other taxonomical trees. The basic-level is the most informative on perceptual grounds (e.g. ‘chair’ rather than ‘furniture’, and rather than ‘kitchen chair’; Rosch et. al. 1976), and therefore privileged psychologically in salience, ease of processing, inductive potential, frequency of utterance, ease of learning, and representability. Note that there is a huge gain in information in going from, say, ‘furniture’ to ‘chair’ (the functions, shapes, and uses of chairs are considerably more restricted, and morphologically there are relatively very small variations across chairs, as against the same for different kinds of furniture), and a very small information gain in going from ‘chair’ to ‘kitchen chair’. On perceptual input alone, this also happens at the ‘life-form’ level in biological taxonomies (e.g. bird, tree, fish; see Rosch et. al. 1976). However, Atran et. al. (1997) found that the rank which acts as ‘basic-level’ for inductive inferences is the generic species. This seems to be the case because Atran et. al. tested inductive inferences for ‘hidden’ or non-obvious properties (e.g. susceptibilities to disease, the possession of particular proteins). Thus, whereas a domain-general mechanism does indeed seem responsible for the rank-similarity of the ‘basic-level’ in all sorts of taxonomical domains, the biological domain is different because of the nature of biological information, where countless non-obvious properties cluster at the generic-species level but not the life-form level. This discrepancy between biological taxonomies and other taxonomical trees suggests that there is indeed a privileged biological domain of cognition.

The idea is that universal, possibly innate, principles lead people to believe that visible morpho-typical patterns of each readily identifiable generic species, as well as non-obvious aspects of biological functioning, are causally produced by an underlying essence. The nature of this essence is initially unknown but presumed.— Atran et. al. (1997:37)

Atran et. al. (1997), by comparing the performances of the inhabitants of two very widely divergent cultures, and finding in both an absolute privilege of the generic-species level for the purposes of inductive inferences, support the hypothesis that the living-kinds or ‘folkbiology’ module is a human universal.