Prototype experiments and metalinguistic belief
A final objection concerns the contrast between the evidence for a prototype model of semantics versus more traditional ones. At least some of the experimental evidence that motivates the postulation of prototypes can be criticized on the grounds that it is not evidence about how speakers actually use words, but evidence about how they think words are used or should be used. For instance, one of Rosch’s standard sets of instructions to the subjects of her experiments makes it clear that subjects are being asked to assess how far an example ‘represents what people mean’ when they use particular category terms (1975: 198), and subjects in one of Rosch’s classic experiments were asked how far certain words represented their ‘idea or image of what the category is’ (Rosch and Mervis 1975: 588). The problem here is that the results of these experiments are about subjects’ beliefs about language and the categories referred to in it, not about their actual language use itself:
they are, in short, metalinguistic. As such, they may be the result of an unpredictable range of prescriptive and other considerations which may not be operative in ordinary language use. Just as subjects’ ideas about how to define words are notoriously unreliable and unrepresentative of words’ actual use, so too their goodness of exemplar ratings may not tell us anything about the underlying meanings of the words concerned. This criticism is avoided to a certain extent by other experiments, such as the reaction time experiments mentioned in the previous section, which show various ways in which goodness of exemplar rating is correlated with actual processing time. But even these apparently less metalinguistic experiments may not be representative of people’s real-time categorizing behaviour in ordinary unmonitored discourse. In one type of experiment, for instance, subjects ‘typically are required to respond true or false to statements of the form X item is member of Y category’ (Rosch 1978: 38), with their speed in doing so correlating to the prototypicality of the exemplar in question. This experiment may reveal various psychological facts about categorization, but it could not be taken to reveal anything about the meaning of the words involved without the additional assumption that people’s natural language use involves the same principles observed in experimental situations, where subjects are consciously attending to issues of the truth and falsity of cate gory terms. Thus, while prototype theory may be well-founded as a theory of categorization, we should not assume that its results can be transferred immediately to the explanation of language use, since the naming options which people exercise in actual discourse may be affected by many other factors than the prototypicality of the referent.
This criticism does not have to apply to the necessary and sufficient conditions view of categorization. When assembling a list of necessary and sufficient conditions, an investigator can proceed simply by observing how words are actually used, and hypothesizing necessary and sufficient conditions to explain these uses; no-one has to be consulted in order to discover their beliefs about what words mean, as in the prototype approach. The centrality of subjects’ judgements about their own language use in prototype theory is a potential problem if there is any chance that subjects may simply be mistaken about the ways in which they use words. I may well say, when asked or tested by a prototype researcher, that tennis is a better example of a game than patience, but what if it turns out that in spite of this judgement I typically refer to tennis as a sport in my actual language use? The frequency of this sort of mismatch between sub jects’ self-reports and their actual behaviour is unknown; however, it is clearly an important issue that needs to be settled.
In spite of these problems, prototype models of categorization have been the source of a major reorientation in the practice of much semantic description. In spite of Rosch’s unwillingness to elevate prototype theory into a full-blown theory of mental representation, many semantic investigators now take it for granted that the meaning of all or most lexical items consists in a prototype structure. As a result, the semanticist’s role is to characterize only the most prototypical aspects of that structure, and a range of meanings outside it is only to be expected.
