DEEP DYSLEXIA - A CASE-STUDY OF CONNECTIONIST NEUROPSYCHOLOGY

Deep dyslexia is an acquired reading disorder marked by the occurrence of semantic errors (e.g. reading RIVER as ''ocean''). In addition, pa tients exhibit a number of other symptoms, including visual and morpho logical effects in their errors, a part-of-speech effect, and an advan tage for concrete over abstract words. Deep dyslexia poses a distinct challenge for cognitive neuropsychology because there is little unders tanding of why such a variety of symptoms should co-occur in virtually all known patients. Hinton and Shallice (1991) replicated the co-occu rrence of visual and semantic errors by lesioning a recurrent connecti onist network trained to map from orthography to semantics. Although t he success of their simulations is encouraging, there is little unders tanding of what underlying principles are responsible for them. In thi s paper we evaluate and, where possible, improve on the most important design decisions made by Hinton and Shallice, relating to the task, t he network architecture, the training procedure, and the testing proce dure. We identify four properties of networks that underly their abili ty to reproduce the deep dyslexic symptom-complex: distributed orthogr aphic and semantic representations, gradient descent learning, attract ors for word meanings, and greater richness of concrete vs. abstract s emantics. The first three of these are general connectionist principle s and the last is based on earlier theorising. Taken together, the res ults demonstrate the usefulness of a connectionist approach to underst anding deep dyslexia in particular, and the viability of connectionist neuropsychology in general.