Using a speech perception neural network computer simulation to contrast neuroanatomic versus neuromodulatory models of auditory hallucinations.

A number of studies suggest that schizophrenia may arise from overzealous pruning of synapses that are an extension of normal developmental pruning during adolescence. Moreover, there has been a long history of studies suggesting that this disorder arises from alterations in the dopaminergic neuromodulatory systems. In order to further assess and compare these two hypotheses, a computer simulation of some aspects of speech perception was developed utilizing a recurrent, backpropagation model of working memory previously reported by Elman . This system was found to produce spontaneous percepts simulating hallucinated speech when the working memory component either was excessively pruned or when neuronal responses were modulated to simulatea hyperdopaminergic system. These hallucinogenic systems also demonstrated disruptions in processing input information when "phonetic information" was degraded. The perceptual performance of these systems were compared to that of actual hallucinating patients and normal controls while tracking (repeating while simultaneously listening to) speech that was phonetically degraded. We found that the neural network simulation producing the best match to speech tracking performance of human hallucinators was an overpruned system with compensatory hypodopaminergic adjustments. These data suggest that the primary pathophysiology of schizophrenia arises from curtailed connectivity in working memory systems and that dopaminergic alterations reflect secondary compensatory adjustments.