THE APPLICATION OF THE ADAPTIVE RESONANCE THEORY TO THE ACQUISITION AND THE REPRESENTATION OF CONCEPTUAL KNOWLEDGE

In the present paper simulations with the ART1, a neural net of the ad aptive resonance theory, are applied to aspects of concept formation. The first simulations will show that different types of categories are produced depending on the time course of learning. Both classical wel l-defined categories and prototypical categories can arise. The simula tions especially reveal the transition between these types of categori es. It will be clear that the different models of categorization can b e realized with the neural ART1 dynamics by changing a time-parameter. In this context we will discuss experimental findings indicating that the kind of categorization depends on time factors. A second set of s imulations deals with both the interaction of the parameter vigilance with the typicality of examples and its consequences for the categoriz ation at different levels of abstraction. In particular, the simulatio ns result in different expectations for the children's behaviour regar ding familiar general categories just after the acquisition of a speci fic category. These predictions have already been verified by us, and as a whole are not compatible with other explanations described in the literature. Finally, the time course of competition is studied in the ART1. The duration of competition substantially takes up part of the time of categorization in the net. It is shown that this part of the n et-internal time for choosing a category depends on the same factors w hich determine the reaction time in experiments with human subjects. T hus, the adaptive resonance theory offers a possible explanation for t he differences in the reaction time with a neural dynamics.