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

II.Natural kinds in context: a brief history of categorization theory

Until the 20^th c. categories were thought to be “abstract containers, with things either inside or outside the category. Things were assumed to be in the same category if and only if they had certain properties in common, and the properties they had in common were taken as defining the category” Lakoff (1987:6). An element in the Universe was either an ‘A’ or a ‘not A’—and any ‘A’ was such by virtue of satisfying a set of necessary and sufficient conditions. And that was that, without need of empirical support or further elaboration. This has come to be known as the ‘classical’ view. Very recently, but with lightning speed, views of categorization have moved away from the classical view. In its stead has developed an understanding of categories as mental constructs with a structure far richer than “inside/not inside”, and also a recognition that not all categories have the same structure.

A.Challenges to the classical view

Ludwig Wittgenstein (1953,1:66-67) first cracked the classical edifice by introducing the concepts of ‘family resemblances’, ‘centrality’, and ‘gradience’. He observed, first of all, that the members of some categories did not all share some properties in common (his most famous example is the category game) and concluded that what made such elements members of the category was not the satisfaction of a set of necessary and sufficient conditions, but rather that each possessed ‘enough’ of a set of properties that were characteristic (but not ‘necessary and sufficient’) of members of the category.

Think of the English category religion. There is a chain of family resemblances linking its members, and so long as a token has enough of these family resemblances (it doesn’t matter which), it will be a member of the class. This category includes such things as Shintoism, Shamanism and even Confucianism, at one end, and Islam, Tibetan Buddhism, and Zoroastrianism at the other. The promotion of a moral code, the performance of magic (or the manipulation of unseen forces with no clear physical explanation), stereotyped rituals, belief in supernatural entities, belief in the persistence of individual identity after bodily death (i.e. the soul), an expert class of practitioners (priesthoods), and an account of Creation are all characteristic of religion, but examples of members of the class which lack any of these can be found. Satanism, promotes an immoral code and Shamanism does not promote one either way. Zen Buddhism and Taoism perform no magic. Animism typically has no priests. Etc…There is no set of properties that all members of the class share. Notice, for example, the misgivings of an expert on Shamanism:

For a time it seemed to me best to write a book somehow without the words ‘religion’ or ‘shamanism’. ‘Religion’ seemed wrong for ideas and beliefs which are never set out as general theory and make use of relatively few abstract concepts, for which there is no holy founder, no organized institution, no moral dogmas, and no authoritative corpus of books. Above all, there is no tortuous justification of earlier beliefs enshrined in ancient texts.—Humphrey (1996)

This brings us to another of Wittgenstein’s contributions: centrality. The above quote reveals Humphrey’s intuitions (and mine, and probably those of most competent English speakers) that practices which are institutional, theological, historicized, and moral (in addition to being ritualized, magical, and supernatural) are more comfortably described as ‘religion’. Nevertheless, ritualized practices that attempt to influence, and codify reverence for, supernatural beings and forces (without the above additions, as, for instance shamanism) still strike us as somehow ‘religion’, if less comfortably so. Islam, then, is a ‘central’ (i.e. more typical) member of the category, whereas shamanism is not. Classical categories, as traditionally understood, did not allow for this because if every member equally satisfied all the defining criteria, then there wasn’t thought to be any basis on which to expect that some members would be “more equal than others.”

Finally, Wittgenstein observed that at least some categories were characterized by ‘gradual’ membership. For example, how do you crisply separate the set of poor people from the set of rich people? You can’t, not crisply. The same goes for young vs. old, blue vs. green, tall vs. short, etc. Some people are clearly young (babies), and some old (nonagenarians), but the boundaries of these categories are not clear. To age is to gradually lose membership in the category young and gain it in the category old.² Zadeh (1965) formalized these intuitions by generalizing set theory to include such ‘fuzzy sets’.

Eleanor Rosch pioneered experiments that could systematically investigate category structure, and synthesized advances by Wittgenstein and others into a theoretical framework that revolutionized psychology. She introduced the term ‘prototype’ to refer to the ‘cognitive reference point’ or most central member (or subcategory of members) of a category.

Rosch showed that a variety of experimental techniques involving learning, matching, memory, and judgments of similarity converged on cognitive reference points. She developed…experimental paradigms for investigating categories of physical objects. In each case, asymmetries (called prototype effects) were found: subjects judged certain members of the categories as being more representative of the category than other members. For example, robins are judged to be more representative of the category bird than are chickens, penguins and ostriches, and desk chairs are judged to be more representative of the category chair than are rocking chairs, barber chairs, beanbag chairs, or electric chairs. The most representative members of a category are called “prototypical” members.—Lakoff (1987:41)

Eventually, Rosch (1978) came to recognize that demonstrating prototype effects for a category did not in fact reveal its structure, for different structures are consistent with prototype effects. In particular, showing that there are typicality effects in people’s appreciation of various members of a category, or differential response-times in their identification of category members, for example, does not necessarily argue for a probabilistic categorical structure as in young people. The most forceful presentation of this argument is in Armstrong, Gleitman, and Gleitman (1983) where ratings of typicality and reaction times were examined for the category odd number. It was found that indeed some numbers, such as 3 or 7, were felt by subjects to be more representative of the category odd number than others, such as 109 and 2003, but “if subjects are asked directly whether typical odd numbers are or can be more odd than atypical ones, they will flatly deny it” (Keil 1989:30, my emphasis). In spite of this, as Rosch cogently emphasized, prototypical effects do constrain the range of possible category structures to those which can account for such effects. She also maintained that necessary and sufficient conditions (i.e. the classical view) cannot.

A representation of categories in terms of conjoined necessary and sufficient attributes alone would probably be incapable of handling all of the presently known facts, but there are many representations other than necessary and sufficient attributes that are possible.—Rosch (1978:40-41)

This last point is less defensible. I will try to convince you below that, despite their demonstrable prototype effects, natural kinds are classical categories, with necessary and sufficient conditions of membership.