Problems identifying the attributes of prototype categories
The first type of problem concerns the nature of the semantic attributes on which judgements of prototypicality are based. In our discussion of categories we have simply isolated the attributes in an intuitive fashion, an apparently unproblematic procedure. For instance, it doesn’t seem unreasonable to suggest that people use the attribute ‘has a seat’ as part of the decision about whether to classify a particular object as a CHAIR. But Rosch herself acknowledges that the ease of identification for many attributes is deceptive (1978: 42). There are essentially three problems, which we deal with in turn:
• attributes can often only be identified after the category has been identified
• attributes are highly context-dependent
• there are many different alternative descriptions of the attributes of a given category
Attribute identification depends on category identification In the ‘has a seat’ case, for example, the identification of this attribute seems to paradoxically depend on a prior identification of the CHAIR category itself: how do we know, for instance, that an armchair ‘has a seat’ unless we have already categorized it as a chair? Why do we not treat the seat of the arm chair simply as a physical zone of the armchair without any particular functional significance, in the same way we treat, for example, the separately stitched piece of material which covers the shoulder section in a shirt? The answer seems to be that we can isolate the seat as a distinctive attribute of an armchair only because we already know that the armchair is designed to be sat on – that it is a chair. This is a paradox for the theory: examples are supposed to be assigned to a category in virtue of their attributes, but at least some attributes seem to depend for their identification on a prior identification of the category in question.
Rosch also points out (1978: 42) that some attributes, like ‘large’ for the category PIANO, depend on considerable background knowledge: pianos are large for pieces of furniture, but small for buildings. It could therefore be objected that attributes like this are not more basic cognitively than the whole objects to which they belong, and that they cannot be considered the basis for the categorization. As Rosch puts it, ‘it appeared that the analysis of objects into attributes was a rather sophisticated activity that our subjects . . . might well be considered to be able to impose only after the development of the category system’ (1978: 42; italics original).
Attributes vary with context In a similar spirit, Khalidi (1995: 404) notes that the kinds of features that subjects associate with certain concepts vary widely and almost without limit when one varies the experimental context in which they are tested. Rather than accessing a fixed set of features in conjunction with each concept, there is apparently no limit to the features that even a single subject associates with a certain concept depending on the context in question.
For example, members of the category MEAL will have very different attributes if the context is a hospital, a wedding banquet, a camping trip or the family dinner table. What would be a good example in one of these con texts will not be a good example in another, and the attributes on which prototypicality depends will vary similarly. The same remarks apply to the category MUSICAL INSTRUMENT: a plastic recorder is a good example and a bassoon a bad one if the context is an infants’ school music class, whereas these values are reversed if the context is a symphony orchestra. Similarly, the concept PIANO will be credited with different features depending on whether the context is taken to be producing music or moving furniture (Barclay et al. 1974, cited by Khalidi 1995: 405). Any attempt to specify the prototypical features of a category or the attributes of one of its members will therefore have to deal with the possibility that these features may change significantly from one context to another.
Alternative descriptions of attributes Another question arises even if we grant that a relatively fixed list of attributes could be constructed for a category: how do we know which descriptions of the attributes are psychologically real? For example, what are the attributes of the category TREE? Langacker (1987: 374) suggests ‘tall plant’, ‘with leaves’, ‘with branches’ and ‘with bark’. It is true that these attributes are among those which distinguish trees as a matter of fact, but we may not be entitled to assume that they enter into the conceptual representation of the category. It may be, for example, that the relevant attributes of TREE are actually best described as ‘made of wood’, ‘growing in ground’, ‘with long trunk’ and ‘sometimes covered in small green objects’. This description of the attributes makes no difference to the rankings of exemplars of trees: an oak will still be a prototypical tree, and a cactus will be an atypical one. But the nature of the attributes on which the prototypicality judgements are claimed to rest will reflect an entirely different understanding of the underlying structure of the category. Indeed, the category TREE might not depend on any underlying abstract features like ‘with bark’. Instead, it could be based around a particular example of a tree as stored in long-term memory. This, indeed, is precisely the hypothesis made in exemplar theories of categorization, which are alternatives to the prototype model in psychology (see Storms et al. 2000).
