The Representation of Concepts: An Evaluation of the Abstractive and Episodic Perspectives

Abstract

<p>The field of concept formation has been dominated until recently by the abstraction perspective, which holds that categories are mentally represented by abstract summaries of their members. Two variants of this view are the prototype models, which employ singular, central representation via an abstracted central tendency, and strength models, which represent categories through abstracted counts of the frequency of their members feature compounds. In conflict with these notions are instance models, which reject summary representation in favour of separate encodings of individual experiences of category members. The three types of models make similar generalization predictions in stimulus domains whose density is greatest near the central tendency, but make importantly different predictions in other domains.</p> <p>The assumptions of abstraction models regarding the representation of variability and contingency relationships of stimulus features were formalized, and a variety of models differing in the complexity of their assumptions were tested, employing perceptual identification, recognition and categorization tasks. Models based on traditional assumptions of the prototype perspective could not account for the variety of generalization patterns obtained, while the assumptions of models which were successful in accounting for the data were argued to violate the cardinal prototype values of economical and summary representation.</p> <p>A new instance model, the "episode model", was proposed. This model was found to account successfully for a wide variety of patterns of generalization in a variety of domains of differing density, through parallel processing of multiple prior encodings. An important aspect of the model is its emphasis on the the degree of integration of prior encodings, which is held by the model to determine the breadth of generalization of performance supported by prior episodes. This aspect of the model reflects its concern with the effects of processing differences on performance.</p> <p>One class of feature-frequency models was also found to be capable of accounting for the patterns of results. However, the instance account was argued to be preferable on grounds of economy and simplicity of representation, sensitivity to processing context and differences in processing, and heuristic utility in directing attention to important adaptive abilities of the organism.</p>