Part One
On the content of prototype categories: questions of word meaning
Chapter 1
A survey of category types in natural language
Cecil H. Brown
The treatment of two or more distinguishable entities as if they were the same creates a category (cf. Mervis and Rosch 1981: 89). People create categories by assigning the same name or label to different things. When speakers of a language are in general agreement with respect to the different entities to which a single term applies, the pertinent category is a component of natural language. This chapter surveys types of category lexically encoded in natural language. Specifically, it focuses on categories whose membership is restricted to concrete objects such as plants, animals, toys, weapons, and tools, as opposed to abstract things such as war, love, religion, poetry, knowledge, and lies.
The present work attempts to show that categories of natural language can be profitably analysed by relating them to a system of category types defined in terms of three factors. These are (1) artifactual versus non-artifactual reference (+ AR vs. − AR); (2) Gestalt versus non-Gestalt motivation (+ GM vs. − GM); and (3) prototype/extension versus non-prototype/extension (+ P/E vs. − P/E). For example, as explained in detail presently, categories such as screwdriver, cup, pen, chair, rope, button, and train all belong to a single type of natural language category since all are plus for artifactual reference (+ AR), plus for Gestalt motivation (+ GM), and plus for prototype/extension (+ P/E). On the other hand, categories such as raccoon, robin, black walnut tree, and dandelion belong to a different category type since all are minus for artifactual reference (− AR), plus for Gestalt motivation (+ GM), and minus for prototype/extension (− P/E). There are, then, eight category types defined by all logically possible combinations of variables of these factors:
Category Type 1: − AR + GM − P/E
Category Type 2: − AR + GM + P/E
Category Type 3: −AR − GM − P/E
Category Type 4: − AR − GM + P/E
Category Type 5: + AR + GM − P/E
Category Type 6: + AR + GM + P/E
Category Type 7: + AR − GM − P/E
Category Type 8: + AR − GM + P/E
Artifactual reference
Concrete objects grouped in categories of natural language are either manufactured by humans (artifacts) or are natural kinds (non-artifacts) such as plants, animals, body parts, clouds, mountains, and rocks. A possible absolute universal of language is that artifacts and natural kinds are never included in the same category (putting aside ‘categories’ generated by metaphorical equations such as dipstick = penis). Thus, there is a clear distinction between categories which involve reference to artifacts (+ AR) and those that entail reference to non-artifacts (− AR).
Gestalt motivation
Concrete objects rarely are morphologically continuous, i.e., typically there is no continuum of objects grading from one to another with respect to similarity. Rather, there is usually a great deal of distinctiveness, making for obvious breaks or gaps among things. Hunn (1977) calls such gaps, when they apply to biological entities (non-artifacts), ‘discontinuities in nature’. Clearly, discontinuities perceived by humans are not restricted to natural kinds. Cups, mugs, and glasses are no more or no less discontinuities than are maples, oaks, and walnuts.
Hunn (1977: 41–75) focuses on psychological processes through which discontinuities are translated into natural language categories. He notes that discontinuities in nature are underlain by feature or attribute clustering. Bruner, Goodnow, and Austin (1956: 47) illustrate this by citing the example of birds in general, creatures possessing feathers, wings, a bill or beak, and characteristic legs. Any one of the latter features is highly predictive of the others. For example, if a creature possesses feathers, it will invariably also have wings, a bill or beak, and characteristic legs. Thus attributes of the discontinuity ‘birds in general’ cluster together, or in other words are highly correlated with one another.
Hunn, following Bruner et al. (1956: 47), proposes that the mutual predictability of clustering features can lead to an expectancy in the minds of humans that attributes involved will be found together. For example, through exposure to different kinds of bird, people build up in their minds an expectation that feathers, wings, and so on, go together. Such an expectation underlies the conceptual development of the configurational or Gestalt property of ‘birdness’. When such a property develops, inclusion of any particular object in a labelled bird category is contingent upon whether or not the object demonstrates the single feature ‘birdness’. As a result, clustering features pertaining to the bird discontinuity become psychologically subordinated to the single Gestalt property.
A Gestalt property arises through the recoding of features or attributes (Hunn 1977: 46). The concept of recoding, borrowed from information theory, involves the notions of ‘chunks’ and ‘bits’ of information. Data organized by a restricted number of immediate or simultaneous judgements constitute chunks (Miller 1967). The amount of information which each chunk contains is described as a number of bits of information. Recoding essentially consists in taking a great number of chunks, each of which contain but a few bits, and reorganizing them into fewer chunks with more bits per chunk. Miller (1967: 24) gives the following example:
A man just beginning to learn radio-telegraphic code hears each dit and dah as a separate chunk. Soon he is able to organize these sounds into letters and then he can deal with the letters as chunks. Then the letters organize themselves as words, which are still larger chunks, and he begins to hear whole phrases … I am simply pointing to the obvious fact that the dits and dahs are organized by learning into patterns and that as these larger chunks emerge the amount of message that the operator can remember increases correspondingly. In the terms I am proposing to use, the operator learns to increase the bits per chunk.
Bruner, Goodnow, and Austin (1956: 46) illustrate the recoding of attributes into a single Gestalt property by using the following biological example (cf. Hunn 1977: 47):
The student being introduced for the first time to microscopic techniques in a course in histology is told to look for the corpus luteum in a cross-sectional slide of rabbit ovary. He is told with respect to its defining attributes that it is yellowish, roundish, of a certain size relative to the field of the microscope, etc. He finds it. Next time he looks, he is still ‘scanning the attributes’. But as he becomes accustomed to the procedure and to the kind of cellular structure involved, the corpus luteum begins to take on something classically referred to as a Gestalt or configurational quality. Phenomenologically, it seems that he no longer has to go through the slow business of checking size, shape, colour, texture, etc. Indeed, ‘corpus luteumness’ appears to become a property or attribute in its own right.
Hunn (1977) restricts his discussion to the development of Gestalt properties relating to biological categories. However, it is clear that attribute recoding and resulting Gestalten are not limited to natural kinds. For example, as discussed in the above quotation of Miller, letters of radio-telegraphic code are recoded into words. Each word, then, constitutes a single Gestalt property. Words expressed in radio-telegraphic code are, of course, human artifacts. Such words differ from artifacts such as cups, mugs, and glasses, since they are not concrete objects. If words expressed in code can possess Gestalt properties, it seems clear that so can concrete objects manufactured by humans, so long as these objects fall into discrete discontinuities (a point discussed at length presently).
An important assumption of the present discussion is that Gestalt properties typically motivate categories which relate to discontinuities. This is not to propose that such categories are always motivated by Gestalt properties. As noted in the above quotation, a student learning to identify the corpus luteum in a rabbit ovary may begin to do so by ‘scanning the attributes’, so at first the category is defined in terms of several features rather than in terms of a single configurational property. However, objects pertaining to most categories of natural language relating to discontinuities, especially folk categories known to all or nearly all speakers of a language (as opposed to specialist categories such as corpus luteum) ordinarily do not require close scrutiny (for pertinent attributes) for the purpose of class inclusion. In addition, I do not mean to imply that for any one category relating to a discontinuity, a Gestalt property alone motivates the category (another point to be discussed at length presently).
Not all categories in natural language relate to discontinuities. This is particularly clear when abstractions such as lies (falsehoods) are considered. There are, of course, no perceptual things that belong to the category called lie and, consequently, no perceptual discontinuity with which it is connected. Lies, then, do not have in common a certain Gestalt property, rather they relate to what Lakoff (1987: 113) calls a ‘propositional model’: ‘Propositional models specify elements, their properties, and the relations holding among them.’ Coleman and Kay (1981: 28), for example, have proposed a propositional model relating to the category lie involving a speaker (S) who asserts some proposition (P) to an addressee (A):
(a) P is false.
(b) S believes P to be false.
(c) In uttering P, S intends to deceive A.
Thus, a lie is characterized by the properties (a) falsehood, which is (b) intentional, and (c) meant to deceive. None of these properties, of course, is a perceptual property of a thing.
Some categories of natural language encompass concrete objects but, none the less, are not underlain by discontinuit...