|
Lesley G. Hathorn
Barsalou, L. W. (1983). Ad hoc categories. Memory and Cognition, 11 (3), 211 – 217
Purpose
The purpose of this article was to examine the graded structure of ad hoc categories. Ad hoc categories are categories that exist for specialized functions. They include all categories created for a specific purpose, such as items to pack on a trip. They cut across correlational structure and are goal-oriented. The members of an ad hoc category usually are not categorized together unless they support a common goal, such as items to take on a camping trip. Common categories (such as fruit) have graded structure and representation in memory. Graded structure refers to the fact that some items are more typical examples of a category than others, that there are some items that do not fit clearly into a category, and that items that do not belong in the category vary in how similar they are to items that do belong in the category. Well-established category representation refers to the way information is organized (such as taxonomic organization) that enables information to be retrieved. Common categories have high free association with common superordinates and common contrast words. For example, apples have a high association with its superordinate “fruit” as well as other members in the same category such as “oranges” or “bananas” which form contrast words because these objects are all established in memory structure and become interrelated.
The Comparison-Network Model This model accounts for graded structures and well established category representations in common categories. The two parts to this model include the similarity comparison process and the spreading activation network. The similarity comparison process means that two concepts become more similar as the number of shared attributes increases and non-shared attributes decreases. In other words, typicality is highly correlated with family resemblance. The more similar a concept is to the other concepts in the category, the more typical it is. Spreading activation network describes, in three ways, why there are well-established category representations. The first is concept-to-instance associations in which there is a strong association between categories and items in the categories. This is top down processing. The second feature is that there are instance-to-concept associations in which items are easily grouped into their common categories. This is bottom up processing. The third aspect is that category concepts are highly associated with their properties (e.g. birds have wings – wings co-occur with flying). Barsalou sought to determine whether the comparison-network model could make predictions for ad hoc categories in the same way that it does for common categories. He did this by testing whether ad hoc categories posses graded structure, whether the graded structures differ from common categories and how the lack of structure (category representations in memory) of ad hoc categories results in different processing than that of common categories. In other words, processing is different because of the difference in concept-to-instance associations, instance-to-concept associations, and the lack of association of concepts with properties in ad hoc categories.
Experimental Work
Experiment 1
The purpose of this experiment was to determine whether ad hoc categories possess graded typicality. Subjects were given a vignette that described a specific context, an ad hoc category label relevant to the protagonist’s goals, and six items. They were asked to circle the items that belonged to the category and to rank the items for appropriateness from best to worst fit. They were given the six items in this experiment. Results showed that ad hoc categories possess graded structure. This was demonstrated by subjects’ high agreement on the relationship between items and categories. For unclear items, there was less agreement for category membership but not for typicality for all categories. There was agreement on which items were typical, but not on where to draw the line for which items belong in a category.
Experiments 2a and 2b These experiments were to test how the graded structure of ad hoc categories compared to the graded structure of common categories. The gradient structure of common categories occurs in two ways. There are items that are viewed as typical and when subjects generate examples, there is a high frequency of some examples. Experiment 2 compared these aspects of graded structure in common categories to ad hoc categories.
Experiment 2a Subjects generated four exemplars each for nine common and nine ad hoc categories. The common categories were those used by Rosh (1970a), for example, birds, fruit, and sports. The ad hoc categories were unusual ones that most people were unlikely to have exemplars for, such as, “ways to escape being killed by the Mafia.” None of the ad hoc categories had a narrowly constraining context. Results indicated that subjects generated more of the same items in common rather than ad hoc categories. Nonetheless, graded structure could still be seen in ad hoc categories in that some exemplars occurred more that others.
Experiment 2b A selection of six exemplars (from those generated in 2a) for each category were judged for typicality by subjects. Results showed that the range of typicality ratings for these items in ad hoc categories was comparable to those in common categories. Thus, typicality gradients form ratings and rankings occur in common and ad hoc categories. This implies that the same process is used to create graded structure for both types of categories and the process is not affected by how well established a category is in memory.
Experiment 3 The hypotheses tested in Experiment 3 were that the (1) concept-to-instance associations and (2) category concepts are better established in memory for common than for ad hoc categories. To test these hypotheses free recall and cued recall were used. One group of subjects received words from common categories. Another group heard words from ad hoc categories, and a third group heard unrelated words. In the two category conditions, the items were grouped by category and preceded by their category label. More common category items were recalled in both the free recall and cued recall conditions. In addition, more categories were accessed in the common category condition. There was also more intrusion in the common category condition, which is consistent with the conclusion that common categories are well established in memory. Thus, the concept-to-instance associations are better established in memory for common rather than for the ad hoc categories. Ad hoc categories do, however, provide a means for the organization of information. This is evidenced by the fact that there was greater recall for ad hoc category items than for random items.
Experiment 4 Subjects in this experiment were given exemplars from one common category, or one ad hoc category, or random items, and asked generate a category label. Half of the subjects received context information. This experiment showed that the instance-to-concept associations are better established in common than in ad hoc categories. Ad hoc categories were more difficult to label in the absence of context, but context is not necessary to identify common category labels. It is much easier to label common categories in the absence of a priming context.
Conclusions
Both common and ad hoc categories posses graded structure, and some ad hoc categories possess typicality gradients similar to those of common categories. The process of building graded structure in common and ad hoc categories is the same, and it is independent of how well established a category is in memory. Although ad hoc and common categories possess similar structure, they are represented differently. The concept-to-instance association for common categories is stronger than for ad hoc categories resulting in fast access during exemplary production for common categories and a more difficult categorization process for ad hoc categories. Common categories are better represented in memory than ad hoc categories. Ad hoc categories can be used so frequently that they become well established in memory and function like common categories for some people.
Points for Discussion
Do ad hoc categories serve more of a function than provide a means to organize information? For example, can they ever be used to make predictions?
Do we ever really think of ad hoc categories in terms of bottom up processing? That is, if the purpose of ad hoc categories is to achieve goals, does it make sense to work backward from the items to the category? For example, how would you describe an apple? Are there actually a small number of common categories? For example, what is the category label for hook, earth, and flat?
If it is to be expected that intrusions occur for recall of items in common categories, what is the implication for this for false memories?
|
|
This page was last updated: |