Schematic concept formation: an extension of previous models

Abstract

Much of the contemporary research in schema theory involves the applicability of the CODER model to a variety of pattern recognition tasks. The most explicit statements of schema theory have been presented as computer simulations of CODER. CODER is a theoretical model that describes a procedure whereby patterns or objects can be categorized into classes based on the information provided by the object or pattern. Earlier computer simulations of CODER have demonstrated the applicability of the model in classification tasks using sequential VARGUS 7 histoform patterns. The purpose of the present research was to extend the explicit statements of schema theory, as embodied in computer simulation programs, to VARGUS 9 patterns and to judgments of pattern similarity. The VARGUS 9 patterns used were represented by histoform and schematic aircraft format transforms of numerical patterns. The Ss' task was to classify pairs of patterns as representing "same" patterns or "different" patterns and to judge the similarity of the patterns. Two levels of pattern difficulty, the actual pattern classification and the two pattern formats (Histoform or Aircraft), represented the independent variables used in the study. Analysis of the human data indicated that the pattern difficulty and pattern format variables interacted for both classification and similarity responses. The computer simulation was generally equivalent to the human performance data as evidenced by ANOVA and trial-by-trial correlations. Approximately 16 percent of the 45 percent of the variance accounted for by the independent variables was attributed to differences between the performance of the model and the human Ss when the number of correct responses was used as the dependent variable. When similarity judgments were used as the dependent variable, only approximately 4 percent of the 92 percent of the variance accountable for by the independent variables was attributable to differences between the performance of the model and the performance of the human Ss. The analyses indicated specific deficiencies in the simulation; however, the majority of these deficiencies were related to the pattern difficulty by pattern format interaction evidenced by the human 2s and not replicated by the simulation. The present research indicates that an extension of the previous simulations of the CODER model to VARGUS 9 patterns and to judgments of pattern similarity is not only feasible but is, to a large extent, accomplished. The specific deficiencies noted in the simulation do not present theoretical difficulties for the CODER model. A number of techniques for reducing the discrepancies between the simulation and human performance were discussed.