ABSTRACT
In this paper we describe three connectionist simulations which explore the formation of hierarchical organization of categories in the framework of an action-based theory of categorization. An organism with a visual system and a 2-segment arm has to reach different points in space depending on the object seen and on context. The context indicates whether to categorize visually perceived objects, which are all equally similar to each other, at basic or superordinate level. The results indicate that learning of hierarchical organization is easier when the actions to perform with members of basic and superordinate level categories are similar. Furthermore, learning Goal Derived categories - i.e., categories that violate categorical boundaries - is more difficult than learning Common Taxonomic categories. Thus, the results replicate empirical results with action-based categories.
