The Structure and Deterioration of Semantic Memory: A Neuropsychological and Computational Investigation

Wernicke suggested that semantic knowledge arises from the interaction of modality-specific perceptual representations of objects, and of the words used to describe them. We present a computational implementation of this theory, in which semantic representations arise from learning mechanisms that acquire the mappings between visual representations of objects and their verbal descriptions. To test the theory, we first consider data from attribute-listing and drawing experiments designed to measure the similarities in visual appearances and verbal descriptions of objects. The model provides an intuitive account of semantic task performance when its inputs and outputs conform to these data. Next we use the model to understand the structure of impaired performance in semantic dementia (a disorder in which patients present with a selective and progressive loss of conceptual knowledge). Group data from four well-known semantic tasks reveal consistent patterns that find a ready explanation in the model. Finally, predictions of the model are tested in a fifth semantic task.