Attribute Verification in Dementia of Alzheimer Type: Evidence for the Preservation of Distributed Concept Knowledge

Abstract

This study reports on a group of patients with dementia of Alzheimer's type DAT who were selected on the basis that they all made similar semantic paraphasias in a word-to-picture matching test. Using knowledge probe questions, it was shown that these patients were more accurate at verifying properties shared by many exemplars in a category i.e. superordinate knowledge, than properties specific to few exemplars in the category i.e. item-specific knowledge. Differences in level of cognitive effort could not explain this effect. Furthermore, it was shown that although concept familiarity correlated with performance on item-specific probes it did not correlate with performance on superordinate probes. It is argued that these results support the predictions of neural network models of semantic memory in which a concept is represented by a distributed pattern of activity across a set of features. In such models, there is noexplicitdistinctionbetween basic level, subordinatelevel, andsuperordinate level features, but rather, features vary in the number of times they occur within the training set.