THE FORGETTING ALGORITHM - HOW FRAGMENTARY KNOWLEDGE OF EXEMPLARS CAN ABSTRACT KNOWLEDGE

The position of Perruchet and Pacteau (1990, 1991) concerning what is retained about past exemplars when subjects implicitly learn an artificial grammar oscillates between two extremes (bigram information versus intact exemplars). On the other hand, THIYOS, the model proposed by Mathews, Druhan, and Roussel (1989), on the basis of classifier systems, is capable of finding optimal sets of features to retain through rule competition. It is argued that people implicitly generate abstract rules that cannot be explained in terms of retention of bigrams or larger chunks without retention of spatial information and that THIYOS integrates findings associated with prototype and exemplar models of induction.