Prototype Theory Of Concepts

• Concepts have a prototype structure; the prototype is either a collection of characteristic attributes or the best example (or examples) of the concept.

• There is no delimiting set of necessary and sufficient attributes for determining category membership; there may be necessary attributes, but they are not jointly sufficient; indeed membership often depends on the object possessing some set of characteristic, non-necessary attributes that are considered more typical or representative of the category than others.

• Category boundaries are fuzzy or unclear; what is and is not a member of the category is ill-defined; so some members of the category may slip into other categories (e.g., tomatoes as fruit or vegetables).

• Instances of a concept can be ranged in terms of their typicality; that is, there is a typicality gradient which characterises the differential typicality of examples of the concept.

• Category membership is determined by the similarity of an object's attributes to the category's prototype.

view. This view is named after its fundamental proposal that categories have a central description, a prototype, that in some sense stands for the whole category. However, different theories characterise prototypes in different ways.

In some theories, the prototype is a set of characteristic attributes; there are no defining attributes but rather only characteristic attributes of differential importance within the concept (see e.g., Hampton, 1979; Posner & Keele, 1968; Rosch, 1978). An object is a member of the concept if there is a good match between its attributes and those of the prototype. In other prototype theories, the prototype is captured by a specific instance of the category, the best example of the concept (e.g., Rosch, 1978). So, for example, if robin is the best example for the bird category, then it would be the prototype. Another object is a member of the bird category if it shares many attributes with the best example. For the purposes of this chapter, we combine theses two variants of prototype theory in a single treatment that is summarised in Panel 10.3.

Evidence for the prototype view

Apart from the evidence we have already reviewed, there is a large body of evidence supporting the prototype account of categorisation. One of the most notable and early pieces of evidence came from cross-cultural studies on colour categories.

Colour categories

There are many different colour terms used in the languages of the world. Some cultures have terms for a wide variety of colours (e.g., in western Europe we have a huge diversity from magenta to sky-blue to red and so on), while other cultures have very few terms (e.g., the Dani of Papua New Guinea have only two colour terms for dark and bright). Berlin and Kay (1969) suggested that this diversity was only apparent if one distinguished between focal colours and non-focal colours. In their studies they identified basic colour terms using four criteria: (i) the term must be expressed as one morpheme, so something like sky-blue would be ruled out; (ii) its meaning cannot include that of another term, ruling out scarlet because it cannot be explained without reference to red; (iii) it must not be restricted to a particular domain of objects, ruling out terms like blond which really only apply to hair and possibly furniture; and (iv) it must be a frequently used term, like green, rather than turquoise. Berlin and Kay discovered that all languages draw their basic colour terms from a set of 11 colours. English has words for all of this set and they are black, white, red, green, yellow, blue, brown, purple, pink, orange, and grey.

Using the basic colour terms derived from this analysis, Berlin and Kay set about examining some 20 languages in detail, by performing experiments using a set of over 300 colour chips. In these studies, native speakers of the languages in question were asked two questions about the colour chips. First, they were asked what chips they would be willing to label using a particular, basic colour term. Second, they were asked what chips are the best or most typical examples of a colour term. What Berlin and Kay found was that the speakers of different languages agreed in their identification of focal colours; people consistently agreed on the best example of, say, a red or a blue. This together with the finding that subjects were uncertain about category boundaries, suggested that category membership was judged on the basis of resemblance to focal colours. These results were also found for cultures with a very limited colour terminology like the Dani. Rosch (when her name was Heider, 1972; also Rosch, 1975a) showed that the Dani could remember focal colours better than non-focal colours and that, even though they only had two colour terms, they could learn names of focal colours more quickly than those for non-focal colours. It should be pointed out that Lucy and Schweder (1979) have shown that some of these memory results need to be questioned because the colour array previously used to demonstrate the influence of focality on memory was discriminatively biased in favour of focal chips.

Thus, there seems to be a universality in people's categorisation of certain colours and in the structure of colour categories; in particular, it seems that these categories have a prototype structure. However, it is noteworthy that these categories have a strong physiological basis in the colour vision system (see Gordon, 1989). As such, some of these colour categories may be special cases. So, it is necessary to demonstrate similar effects for other categories.

Natural and artificial categories

Research on both natural categories (i.e., categories of things in the world, like birds and furniture) and artificial categories (e.g., numbers and dot patterns) has also supported detailed aspects of the prototype view. As we saw earlier, some members of categories are considered to be highly representative or highly typical. Subjects rate the typicality of instances of a concept differentially (Rips et al., 1973; Rosch, 1973).

These typicality effects have considerable generality; for instance, they have also been found in psychiatric classifications (Cantor, Smith, French, & Mezzich, 1980), in linguistic categories (Lakoff, 1982, talks of degrees of noun-ness and verb-ness) and in various action concepts (like to lie, and to hope; see Coleman & Kay, 1981, Vaughan, 1985). Furthermore, the most typical members of a concept play a special role in human categorisation.

First, the typicality gradient of members of a concept is a good predictor of categorisation times. In verification tasks (e.g., "A canary is a bird") typical members, like robin, are verified faster than atypical members like ostrich. This has proven to be a very robust finding (for reviews see Danks & Glucksberg, 1980; Kintsch, 1980; Smith, 1978; Smith & Medin, 1981). Second, typical members are likely to be mentioned first when subjects are asked to list all the members of a category (Battig & Montague, 1969; Mervis, Catlin, & Rosch, 1976). Similarly, Rosch, Simpson, and Miller (1976b) found that when subjects were asked to sketch the exemplar of a particular category they were more likely to depict the most typical member. Third, the concept members that children learn first are the typical members, as measured by semantic categorisation tasks (Rosch, 1973). Fourth, Rosch (1975b) has found that typical members are more likely to serve as cognitive reference points than atypical members; for example, people are more likely to say "An ellipse is almost a circle" (where circle is the more typical form and occurs in the reference position of the sentence) than a "A circle is almost an ellipse" (where ellipse, the less typical form, occurs in the reference position).

A final important finding is the extent to which estimates of family resemblance correlate highly with typicality. Using Wittgenstein's term family resemblance, Rosch and Mervis (1975) have shown that one can derive a family resemblance score for each member of a category by noting all the attributes that that member has in common with all the other members of the category. Rosch and Mervis found that typical members have high family-resemblance scores and share few (if any) attributes in common with related, contrast categories. This is rather direct evidence for the idea that the typicality gradient of a concept's instances is a function of the similarity of those members to the prototype of the category.

Conceptual hierarchies

Much of the work we have seen on basic-level categories and the three levels of generality (superordinate, basic, and subordinate) was specifically developed in the context of prototype theory. That is, one can think of a basic-level category as being organised around a prototype. So, just as there is a centrality of the prototype in making classification decisions, there is a centrality of the basic level as a focus for the maximally relevant category to consider in making such decisions.

Evidence against the prototype view

Three main criticisms can be made of the prototype view. First, not all concepts have prototypic characteristics. Hampton (1981) has shown that only some abstract concepts (like "science", "crime", "a work of art", "rule", "belief") exhibit a prototype structure. This difference occurs because of the endless flexibility in membership of some abstract categories, in contrast to concrete categories. For instance, it seems impossible to specify the complete set of possible rules or beliefs. Thus, there are limits to the generality of prototype theory.

The prototype view is also incomplete as an account of the sort of knowledge people have about concepts. People seem to know about the relations between attributes, rather than just attributes alone, and this information can be used in categorisation (Malt & Smith, 1983; Walker, 1975). Consider the following case (see also Holland, Holyoak, Nisbett, & Thagard, 1986). Imagine going to a strange, Galapagos-like island for the first time, accompanied by a guide. On the journey, one sees a beautiful, blue bird fly out of a thicket and the guide indicates that it is called a "warrum". Later in the day, we meet a portly individual and are told that he is a member of the "klaatu" tribe. A day later, wandering without the guide one sees another blue bird, like the first, and considers it to be another warrum; however, on meeting another fat native one does not assume that he is a member of the klaatu tribe. The reason being that we know that colour is a particularly diagnostic and invariant attribute of the bird category but physical weight is not a particularly diagnostic attribute of tribal affiliations and is known to be a highly variable attribute. Hence, we know that some attributes are more likely to vary than others. The fact that people can make reasonable guesses about the meaning of new terms on the basis of a single exposure to an instance is an important ability that prototype theory is silent about. The research we reviewed earlier on the predictive use of categories has made central use of these ideas and as such stretch beyond the explanatory reach of prototype theory.

Finally, the prototype view does not provide a good account of what makes some categories natural and coherent; what makes us group certain objects together in one category rather than in another. The traditional answer given by the prototype and other views is that similarity is responsible for category cohesion. Stated simply, things form themselves into categories because they all have certain attributes in common. However, similarity cannot be the only mechanism because we often form categories that are only tenuously based on shared attributes but which are nevertheless coherent. In reviewing the evidence on concept instability we saw that people can create categories on the fly, so-called ad hoc categories; from the perspective of prototype theory it is hard to imagine how such categories can cohere, given the lack of overlap between the attributes of category members (e.g., things-to-sell-in-a-garage-sale). Murphy and Medin (1985) point to the biblical categories of clean and unclean animals; clean animals include most fish, grasshoppers, and some locusts while unclean animals include camels, ostriches, crocodiles, mice, sharks, and eels (see also Douglas, 1966, Lakoff, 1987).

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


Responses

Post a comment