Connectionist modeling of the recovery of language functions following brain damage

This paper reviews the contribution of connectionism to our understanding of behavioral changes in language functions after brain damage. Connectionism is founded upon a neural metaphor in that connectionist networks are made up of many simple, neuron-like units. It is possible to lesion these networks and explore the effects of that damage. It is widely held that damaging connectionist networks informs our understanding of neuropsychology and cognitive psychology. To what extent then does it currently tell us, or is likely to tell us, anything about behavioral change following brain damage? Current connectionist models simulate either spontaneous recovery or the effects of retraining, and I discuss both approaches. Which is taken at present partly depends upon the connectionist framework used as the starting point. Most simulation work involving back-propagation has focused upon retraining lesioned networks, while work involving interactive activation has focused upon making inferences about the time course of spontaneous recovery. I discuss research in modeling deep dyslexia, aphasia, and dementia. I argue that further research on modeling spontaneous recovery is limited by the fixed architecture of most current connectionist networks.