|
Armstrong, S.L., Gleitman, L.R., & Gleitman, H. (1983). What some concepts might not be. Cognition, 13, 263-308. Purpose This article focuses on the feature theories of mental concepts, specifically on prototype theories. The authors discuss human’s natural ability to easily divide their world into categories. Holistic theories propose that this ability is a result of a one-to-one relationship between words and mental categories such that the word itself represents a category and cannot be further analyzed. Feature theories posit that words are labels for complex mental categories containing smaller categories, which are grouped together by shared features and elements. There are two subtypes of feature theories: the definitional view and the prototype view. In the definitional view, membership in a category is based on a small set of necessary and sufficient features. Those items that have these features belong to the category and those that do not are not members of the category. Although this view appears simplistic, it is not easy to use. Finding the necessary and sufficient features for category membership is very difficult and has not been successfully accomplished. Also, it is possible to remove some of the necessary features without affecting judgments of category membership, which calls into question the necessity of these features. The prototype view is based on the idea of family resemblance. In this view, category membership is not based on necessary features. Rather, there are privileged properties, which are shared by members of the category, but they are not the only features, which determine category membership. The members with more privileged properties are more prototypical of the category, whereas those with fewer privileged properties are more marginal category members. Rosch (1973) has supported this view empirically. Participants were asked to rate objects as an exemplar of a category on a scale of best to worst. Rosch found that participants readily rated these objects demonstrating that membership to a category is not a clear-cut decision; there are some members which are more readily accepted as belonging to the category than others. Other empirical studies have found that when participants are given examples of a category and asked to verify their membership in the particular category, response times are faster for those examples that are considered the best exemplars of that particular category. According to Armstrong, Gleitman, and Gleitman, the prototype view, like the definitional view, is not without problems. It is just as difficult to identify prototypical features of a category, as it is to identify necessary and sufficient features. Also, prototype theories make reasoning with words about categories very difficult. The use and combination of different words and phrases when explaining categories causes the various features of items to subsequently become combined in a variety of ways complicating one’s understanding of an item’s category membership. In a set of their own experiments, the authors sought to replicate previous empirical findings. In three experiments, which relied on Rosch’s methodology with a few additions, the authors demonstrated that graded membership for categories is ubiquitous. Experimental Work Experiment 1: Participants were asked to rate items on how good of an example each was of their particular category. Two kinds of categories were used: well-defined, such as “odd number”, and prototypic, such as “support”. Using Rosch’s procedures, the authors developed sets of prototypic exemplars and their own sets of well-defined exemplars for the various categories. The participants were presented with six exemplars for each category and asked to rate each on a seven-point scale, one representing a good example of the category and seven representing a bad example of the category. There was a high degree of agreement among the subjects’ ratings of good and bad exemplars for both the well-defined and prototypic categories. For the well-defined categories, participants were able to provide graded responses for the presented items even though by definition these items could not be rated. For example, for the category of odd number, participants rated some numbers as better examples of an odd number than others. However, by definition they are all odd numbers. The pattern of graded responses was similar for both well-defined and prototypic categories but the mean scores were higher for the prototypic categories. Thus, participants seemed be more likely to judge an example as less than perfect in the prototypic category than in a well-defined category. Experiment 2: This experiment tested whether the findings of experiment one were simply a result of the participants’ willingness to follow directions. Since the directions were to rate exemplars, the participants might have simply complied even if they found the task to be ridiculous or impossible. In a replication of a category verification task conducted by Rosch, the authors presented participants with 64 sentences (e.g. “an ostrich is a bird”) on a computer screen one at a time and asked them to verify the truth of the sentences. The categories were either well-defined or prototypic and the participant’s task was to verify if the exemplar belonged to the category. Verification of typical items was tested for both the well-defined and prototypic categories. For both the prototypic and the well-defined categories, exemplariness was found to affect verification time with faster times being associated with the better exemplars of the categories. Experiment 3: Experiments one and two left the authors slightly puzzled. If participants truly rated items based on category membership, they should not have been able to make ratings for well-defined categories, for which there is equal membership. As a result, Armstrong et al. began to wonder if category membership was actually what the participants were using as the basis for their ratings. Thus, in the third experiment they investigated whether category membership was the same as category exemplariness. Participants were given categories, followed by six exemplars from each and asked to say if membership to that particular category was graded or all-or-none. They were then shown the same categories and exemplars again and asked to rate the exemplariness of each item in the categories. Even if the participants had stated that membership to a particular category was all-or-none, they were required to provide ratings for the exemplars. Participants rated some items as being better exemplars of a category than others, even when they had stated moments earlier that the category could not have graded responses by its very nature of being categorical. Based on these findings, it seems that the ratings might not be based on category membership but rather on exemplariness. Conclusions: The authors investigated the prototype theory and the family resemblance structure of categories, which have been commonly reported in past literature. They found that for both types of categories, participants provided graded responses, saying that some items in a category were better exemplars than others. Well-defined categories by definition should have resulted in all-or-none responses and not graded responses, but this pattern was not supported empirically. The authors relied on commonly used paradigms to measure category structure and exemplariness in their studies and based on their findings, they concluded that the paradigms were flawed. The methodology cannot be accurately measuring the structure of categories if graded responses are being provided for concepts, which by definition do not have graded membership. Lastly, the authors posited a dual-position theory, which stated that concepts have core descriptions and identification procedures. The authors used this theory to explain their findings of graded responses for members of well-defined categories. People may use the identification function to decide which examples belong to which categories and then use the core description to decide about membership within a particular category. The core descriptions of concepts will be different, thus allowing some items to be better examples than others of a category. Points for Discussion: The authors state that the paradigms used to study the structure of concept are flawed. However, they do not offer an alternative methodology. In their third experiment, they do use both Rosch’s methodology in addition to their own task, but they still find the same pattern of results. What are some possible methodologies, which could be used in these studies to better measure category structure? Maybe category structure cannot be truly studied as a theory; rather categories and concepts must be viewed as independent and unique. The role of compliance must be further considered. A compliant participant will follow the instructions given by the experimenter even if the task seems silly or impossible. As many studies have demonstrated, people find it very easy to divide the world into categories. Categorization seems to be a natural ability, thus rating items within a category will be an easy task even for items that by definition cannot be rated. Maybe it is simply our natural response to rate and understand items within a category as being on a continuum, which explains the pattern of findings. Is it possible to control for the effect of participant compliance? Or to control for this natural inclination to rate items? The authors attempt to distinguish between category membership and exemplariness. Isn’t this distinction related to the difference between the definitional view and the prototype view? In the definitional view, category membership is based on necessary and essential features, whereas in the prototype view, membership is based more on prototypical features. Is this distinction just a closer examination of the differences between the two views? Maybe it is necessary to think about concepts in different ways depending on the specific concept. |
|
This page was last updated: |