Figure 102

A schematic diagram of the sort of hierarchical, semantic networks proposed by Collins and Quillian (1969).

On the whole the evidence against the defining-attribute view far outweighs that in favour of it. All of the so-called prototype effects outlined earlier go against its basic predictions.

On category judgements, all members of a category are not equally important or representative. In terms of the defining-attribute view, people should list the same attributes for all the members of a category (i.e., the defining set). However, people do not do this but tend to mention non-necessary attributes (Conrad, 1972; Rosch & Mervis, 1975). So, category members are not all equally representative. As Rosch and others have shown, some category members were rated as being much more typical than others (see also Rips, Shoben, & Smith, 1973). For example, a robin was considered to be a better example of a bird than a canary. Even categories that appear to meet the theory, like bachelor; show typicality effects. Tarzan is not a good example of a bachelor because, alone with the apes of the jungle, he did not have the opportunity to marry (see Fillmore, 1982; Lakoff, 1982, 1987). Furthermore, as we have seen, categories do not have clear boundaries but can be fuzzy and changing.

On the issue of conceptual hierarchies, the theory does not specifically predict the three-level structure and the centrality of basic-level categories. Furthermore, contrary to Collins and Quillian's findings, Smith, Shoben, and Rips (1974) have shown that more distant superordinates can be verified faster than immediate superordinates. So, when asked "Is a chicken a bird?" and "Is a chicken an animal?", contrary to Collins and Quillian's node-distance prediction, subjects responded faster to the latter than to the former. Hampton (1982) has shown that the defining-attribute prediction that hierarchies of concepts are transitive is not confirmed (i.e., as "An X is a Y" and "A Y is a Z" are true, "An X is a Z" is also true). Conrad (1972) has also shown that certain attributes of concepts were mentioned more often by subjects than other attributes and, hence, are considered to be more important or salient. For example, the attribute of a salmon is-pink is mentioned more often than the attribute has-fins. Not only does this suggest that attributes are not given equal weight by subjects but Conrad showed that, in Collins and Quillian's experiments, the fast verifications of some sentences were due to the attribute's salience and not the number of links.

On the predictive use of categories and concept instability, it should be clear that this theory is quite inadequate. It has no notion that attributes may be held in a probabilistic fashion by concepts and takes the view that all the attributes of a concept are equally available and present all of the time.

Evidence against the defining-attribute view

More generally, the theory suffers from a fundamental problem in determining exactly what are "defining attributes". Several generations of linguists, philosophers, and psychologists have failed to find defining attributes or semantic features for concepts (for examples of the latter see Hampton, 1979, 1995; McNamara & Sternberg, 1983). Some, therefore, argue that the whole enterprise of trying to break concepts down into their necessary and sufficient attributes is fundamentally ill-conceived (Fodor, Garrett, Walker & Parkes, 1980; Wittgenstein, 1958). Some concepts simply do not seem to have defining attributes. Consider Wittgenstein's example of the concept of a game. There are clusters of attributes that characterise sets of games (that they involve pieces, involve balls, involve one or more players) but hardly any attribute holds in all the members of the concept. Members of the category game, like the faces of the members of a family, bear a family resemblance to one another but they do not share a distinct set of necessary and sufficient attributes.

Saving the defining-attribute view

It is rarely the case that a theory is completely defeated by the evidence. Usually theories make a comeback in some modified form. In this vein, several variants of the defining attribute view have been proposed. One variation, feature comparison theory, admits that there are defining attributes and characteristic attributes (Rips et al., 1973; Smith et al., 1974). Using this additional assumption, a variety of effects can be handled (see Eysenck & Keane, 1995, Chapter 10 for details). Another variant on the defining-attribute theme, identified by Smith and Medin (1981) is based on Miller and Johnson-Laird's (1976) distinction between the "core" of a concept and its "identification procedure". The core of the concept consists of defining attributes and is important in revealing the relations between a given concept and other concepts. That is, the conceptual core of bachelor (male, single, adult) is important to revealing why a bachelor and a spinster (female, single, adult) are similar or why the terms "bachelor" and a "single male" are considered synonymous. The identification procedure plays a role in identifying objects in the real world and is responsive to their characteristic attributes. Thus, the core retains the defining-attribute theory while the identification procedure can account for typicality effects.

Armstrong et al. (1983) carried out a study that suggests evidence for this view. They examined concepts that clearly have defining attributes (e.g., even number, odd number, plane geometry figure) and found that members of these categories were judged to be more or less typical of the category. For instance, 22 was rated as being more typical of the concept even number than 18 and was also categorised faster. Thus, these concepts seemed to have a conceptual core and yet in categorisation tasks people made use of characteristic attributes. Similarly, McNamara and Sternberg (1983) asked subjects to list the attributes of several different types of nouns (artifacts, natural kinds, and proper names) and to rate the necessity, sufficiency, and importance of each attribute to the word's definition. An inspection of subjects' ratings revealed that they considered some of the words to have defining attributes (i.e., necessary and sufficient attributes) and characteristic attributes. But only half could be defined by the defining attributes produced by subjects. McNamara and Sternberg also showed that these same distinctions were implicated in the real-time processing of the concepts, when read as words.

If we assume that concepts have a conceptual core and then other characteristic features, then one would expect there to be linguistic hedges in the language to take this distinction into account. Lakoff (1973, 1982) has argued that such hedges exist and are signalled by terms like "true" and "technically speaking" or "strictly speaking". These terms qualify assertions we might make about category members. For example, if one says a "a duck is a true bird" the core definition of the concept bird is being explicitly marked, whereas the sentence "technically speaking, a penguin is a bird", marks the fact that you know a penguin is a non-representative example of the category but wish to include it within the category.

Stop Anxiety Attacks

Stop Anxiety Attacks

Here's How You Could End Anxiety and Panic Attacks For Good Prevent Anxiety in Your Golden Years Without Harmful Prescription Drugs. If You Give Me 15 minutes, I Will Show You a Breakthrough That Will Change The Way You Think About Anxiety and Panic Attacks Forever! If you are still suffering because your doctor can't help you, here's some great news...!

Get My Free Ebook


Post a comment