Prototype Theory

Now that we know these facts about the psychology of concepts, the question arises of how concepts are mentally represented. There are several theories about this but two main potential answers have been explored extensively in psychology research. The first, somewhat confusingly called the prototype theory, suggests that people have one summary representation of the category. This summary mental description is meant to apply to the category as a whole. (The significance of summary will become apparent when the next theory is described.) 
 
This description can be represented as a set of weighted features (Smith & Medin, 1981). The features are weighted by their frequency in the category. For the category of birds, having wings and feathers would have a very high weight; eating worms would have a lower weight; living in Antarctica would have a lower weight still, but not zero, as some birds do live there.

The idea behind prototype theory is that when you learn a category, you learn one typical, summary representation. Often these prototypes are an image or a description that applies to the category as a whole. For example, birds have wings and usually fly; some eat worms; some swim underwater to catch fish. People can state these generalizations, and sometimes we learn about categories by reading or hearing such statements (“The kimodo dragon can grow to be 10 feet long”).

When you try to classify an item, you see how well it matches that weighted list of features. For example, if you saw something with wings and feathers fly onto your front lawn and eat a worm, you could (unconsciously) consult your concepts and see which ones contained the features you observed. This example possesses many of the highly weighted bird features, and so it should be easy to identify as a bird. Notably, when comparing your new example to your prototypical summary representation for that category, your summary representation (and the combination of the strongly weighted features making it up) does not have to constitute a real-life category member. Rather, it could be something more like the most “average” category member Dr. Frankenstein could create based on the most frequently observed features.

This theory readily explains the phenomena we discussed earlier. Typical category members have more, higher-weighted features. Therefore, it is easier to match them to your conceptual representation. Less typical items have fewer or lower-weighted features (and they may have features of other concepts). Therefore, they don’t match your representation as well. This makes people less certain in classifying such items. Borderline items may have features in common with multiple categories or not be very close to any of them. For example, edible seaweed does not have many of the common features of vegetables but also is not close to any other food concept (meat, fish, fruit, etc.), making it hard to know what kind of food it is.

A different account of concept representation is the exemplar theory (exemplar being a fancy name for an example; Medin & Schaffer, 1978). This theory denies that there is a summary representation. Instead, the theory claims that your concept of vegetables is built on remembered examples of real vegetables you have seen. This could of course be hundreds or thousands of exemplars over the course of your life, though we don’t know for sure how many exemplars you actually remember.

How does the exemplar theory explain classification? When you see an object, you (unconsciously) compare it to the examples in your memory, and you judge how similar it is to examples in different categories. For example, if you see some object on your plate and want to identify it, it will probably activate memories of vegetables, meats, fruit, and so on. To categorize this object, you calculate how similar it is to each exemplar in your memory. These similarity scores are added up for each category. Perhaps the object is very similar to a large number of vegetable exemplars, moderately similar to a few fruit exemplars, and only minimally similar to some exemplars of meat you remember. These similarity scores are compared, and the category with the highest score is chosen.

Why would someone propose such a theory of concepts? One answer is that in many experiments studying concepts, people learn concepts by seeing exemplars repeatedly until they learn to classify them correctly. Under such conditions, it seems likely that people eventually memorize the exemplars (Smith & Minda, 1998). There is also evidence that close similarity to well-remembered objects has a large effect on classification. Allen and Brooks (1991) taught people to classify items by following a rule. However, they also had their subjects study the items, which were richly detailed. In a later test, the experimenters gave people new items that were very similar to one of the old items but were in a different category. That is, they changed one property so that the item no longer followed the rule. They discovered that people were often fooled by such items. Rather than following the category rule they had been taught, they seemed to recognize the new item as being very similar to an old one and so put it, incorrectly, into the same category.

Many experiments have been done to compare the prototype and exemplar theories. However, the experiments are somewhat limited in that they usually involve a small number of exemplars that people view over and over again. It is not so clear that exemplar theory can explain real-world classification in which people do not spend much time learning individual items (how much time do you spend studying squirrels or chairs?). Also, given that some part of our knowledge of categories is learned through general statements we read or hear, it seems that there must be room for a summary description (as with prototype theory) that is separate from exemplar memory.

The Importance of Context and Multiple Cognitive Systems

In assessing whether data best supports a classical (well-defined features), prototype, or exemplar model of concept formation, we find that it may depend on context and how the categories are being used (e.g., Barsalou).

Consider the “ad hoc category” (Barsalou, 1983) from the beginning of the chapter. It is not at all clear how dust, papers, a computer monitor, two pens, a cup, and an orange are related until one if given the context that they are on a computer desk.

Many researchers would now acknowledge that concepts are represented through multiple cognitive systems. For example, your knowledge of dogs may be in part through general descriptions such as “dogs have four legs.” But you probably also have strong memories of some exemplars (your family dog, Lassie) that influence your categorization. Furthermore, some categories also involve rules (e.g., a strike in baseball). The way in which these systems work together may ultimately be quite complex and is the subject of current study.