Learning a Hierarchical Organization of Categories

Although exemplar models of category learning have been successfully applied to a wide range of classification problems, such models have only rarely been tested on their ability to deal with vertical category learning, that is, cases where the same stimuli may be classified at multiple levels of abstraction. We report an experiment in which participants learned to classify artificial stimuli at both levels of a nested hierarchy and displayed more accurate classification of theseion. We report an experiment in which participants learned to classify artificial stimuli at both levels of a nested hierarchy and displayed more accurate classification of these items at the lower level of the hierarchy than at the more general level. Some authors have suggested that exemplar models would have great difficulty accounting for this phenomenon, but we show that the ALCOVE exemplar model effectively captures the behavioral pattern arising in the experiment. Despite suggestions to the contrary, superior performance at the lower level of a nested hierarchy does not necessarily invalidate the class of exemplar models.