A model for schematic concept formation: further development and assessment

Abstract

The purpose of this investigation was the development and assessment of an extended model for schematic concept formation. The original model, CODER, was developed as a higher order functional unit having the capability to form representations of different kinds of patterns, this theoretical machine thus describes a process by which objects can be categorized on the basis of information provided by the environment itself. The basic assumption made in developing such a system was that potentially useful categories are indicated in the environment by association with a large number of covarying characteristics. In this context, the extension of CODER in the present research involved the specification of an additional pattern representation system for representing the characteristics which prove most useful for the categorization process. The extended version of CODER, specified in the form of a computer program, was used to make¿ quantitative predict ions of human schematic: concept formation in two mixed schemata discrimination tasks. The first task required the subjects to discriminate among VARGUS 7 stimuli sampled from two different schema families, the second task was identical to the first except that three different schema families were represented in the task. The parameter Values used in the operating the model were the same for both of the above discrimination tasks. Furthermore, this same set of parameters had been previously used in a simulation of CODER in a free-sorting task. The generality of the basic model's suppositions was thus evaluated in a manner seldom reported in other simulation experiments. A secondary objective of the research was the development of suitable procedures for assessing the degree of correspondence between the group performance of the human subjects and the predictions of the model. These assessment procedures provided quantitative measures of both general and detailed correspondences between human performance and the model's predictions. These comparison methods also provided objective criteria for deciding what degree of correspondence is sufficient for concluding that the model is a useful predictor of human performance. The simulations of the extended version of CODER successfully predicted both the gross and detailed characteristics of human schematic concept formation. Increasing the number of schema families represented in the task had little effect on the degree of correspondence between human performance and the model's predictions. These results thus support the suggestion that CODER' s basic components may be useful in constructing models for performance in a variety of tasks related to schema theory.