|
Rosch, E. & Mervis, C.B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573-605. Armstrong, S.L., Gleitman, L.R., & Glietman, H. (1983). What some concepts might not be. Cognition, 13, 263-308. |
 |
Family Resemblance
Armstrong, Glietman, & Glietman (1983) stated that people have a natural ability to categorize. There are several theories as to how we do it. These articles focused on feature theories. One of the feature theories, prototype theory, states that we pay attention to family resemblance and privileged properties of categories, not necessary and sufficient features. Family resemblance is useful when categorizing, but does not lead to an all-or-none decision about category inclusion. Instead, categories have a graded structure. This was a point that was driven home by both sets of researchers. However, simply stating that a category has a graded structure does not explain how categories arise. Rosch & Mervis (1975) say that a category has a family resemblance structure when each category member shares several attributes with other members of their own category and few features with members of other categories. However, it is not necessary that any single element is common among every member of the category. Members that meet these criteria, according to Rosch & Mervis are considered prototypes of the category. Armstrong, Glietman, & Glietman agree with the above, and add that members with more privileged properties are more prototypical of a category whereas members with fewer privileged properties are considered more marginal category members. The work of these authors, when taken together, begins to converge on an answer to the question of how a category is structured. Both researchers agree that family resemblance is important for category formation and that prototype formation is a part of category formation.
Armstrong et al. wanted to show that membership to a category is not clear cut. The authors used some of Rosch’s previous work in several experiments of well-defined and prototypic categories. They found that both well-defined and prototypic categories had graded membership (although this is not logical for well-defined categories) and that participants judged examples as less perfect more often in prototypic than well-defined categories. Armstrong et al. also showed that participants were willing to say that a category was well-defined on one hand, and then to give graded responses on the other hand.
Rosch et al. wondered why prototypical items are labeled as prototypes. In her first four experiments, she found that superordinate and basic level category members that share many attributes with other members of the category are rated as prototypical and are not prototypical of other categories. In the fifth and sixth experiments, Rosch varied family resemblance and contrast categories. She found that high family resemblance items and items that overlapped less with the contrast category were learned faster, categorized faster, and rated as prototypical whereas this was not the case for items with equal family resemblance. However, when family resemblance and overlap were opposed, they cancelled each other out. She, therefore, concluded that prototypes share many common attributes with their category members and share few with out-group category members.
In class, the students discussed several important points about each of these articles. When discussing the Armstrong et al. article, the first point discussed was that prototypes are difficult to define. The question of how prototypes and defining features differ was discussed as well as what makes a prototype a prototype. A clear consensus was never reached, which is not surprising given the topic is unresolved in the literature as well. However, the most important question that was pondered, and the question that consumed the majority of the conversation time, was the role of compliance in these experiments. For well-defined categories, participants should have rated all of the examples as equally appropriate members of the category. For instance, 451 shouldn’t be rated as a better or worse example of an odd number than 13. During the discussion, students tried to think of ways that the experiment could have been redesigned so that the participants would not give graded membership to items in a well-defined category. The group also discussed why even and odd numbers, although obviously belonging to a well-defined category, do seem to have a graded nature. One possibility that was pondered was the issue of familiarity and frequency. The lower even and odd numbers are used more often, and are therefore more familiar and may seem to be better examples of the categories.
The Rosch & Mervis article also led to an interesting discussion. The discussion centered on two issues. The first issue was a possible problem with the methodology in Experiments 3 and 4. During these experiments, the participants were told to pretend that they were playing a guessing game in which they were attempting to guess a word based on a list of attributes. They were told to imagine that their first guess was wrong. Their task was to provide what their second guess would have been. In the discussion, students talked about the confusing nature of these directions and whether or not the participants would have clearly understood that they were giving another instance of the same category. For instance, if one were given the attributes, “has legs, has a seat, is used for sitting upon” and then followed the given instructions, one would need to ignore one’s first guess of “chair” and report their second guess. However, the class worried that perhaps participants felt that they were supposed to ignore their first guess category and, instead, think of an obscure category that matches those same attributes.
The second issue that was discussed was whether Rosch et al.’s results would generalize to more abstract issues such as justice. A good deal of the discussion focused on defining justice and attempting to create a list of features that could be considered as a prototypical situation of justice. Very similar issues had been suggested before with other research we read and once again, the class came to the realization that justice and other abstract ideas were rather difficult to define and that prototypical features of justice are extremely difficult to define. The words, “I know it when I see it” arose several times.
Both of these studies provided unique contributions to the literature available on family resemblance. The authors agree in many ways about family resemblance and prototypes. Although both research teams focus on feature theory, Armstrong et al. posited their own theory called dual-position theory. Rosch, commonly considered the founder of family resemblance, would also most likely agree with Armstrong et al.’s theory. According to this theory, concepts have core descriptions and identification procedures. People use the identification function to decide which examples belong to a particular category and then use the core description function to decide about membership within the category, labeling some members as more prototypical than others. It seems as though the identification function that Armstrong et al. refers to is similar to the idea of comparing attributes of an example to members within and outside of the category, a concept backed by Rosch & Mervis. It also seems that Armstrong et al.’s core description function helps to explain the graded nature of categories, which is an element that Armstrong et al. and Rosch et al. agree is inherent in the structure of categories.
However, there is a need for continued research in this area due to the abundance of unanswered questions. Although it is commonly agree upon that categories have a graded structure, why is this so? What makes a prototype a prototype? Why are some categories easier to define than others and why are some hard to define when they should be well-defined and have an all-or-none membership assignment? With diligent continued research, perhaps the answers to some of these questions will become illuminated.