A Neural Network Model of Semantic Processing Errors in Parkinson's Disease

摘要

Recent research has indicated that degradation of dopaminergic pathways in the mesolimbic forebrain and neocortex are related to the language processing deficits observed in Parkinson's disease (PD) patients, with experimental investigations indicating that the semantic system suffers the greatest impairment. PD patients make slower responses and more errors in semantic judgement tasks compared to controls, even though nonsemantic performance on word-search tasks is unimpaired. These findings are used in this study to predict the performance of the semantic system of the dual-route model of skilled reading (DRC), where the connection strengths between word and word sense levels in the system are systematically manipulated to produce PD-like reductions in dopamine transmission. Low excitation between layers is predicted to produce PD errors in semantic judgements of polysemous words which have a low sense-frequency delta (s.f.d.), that is, a small difference in sense frequency between at least two word senses. Judgements made on high s.f.d words resulted in 100% correct responses over all excitation values, with a stable solution reached at 39–48 epochs. However, for low s.f.d words, incorrect responses were made for low to medium levels of excitation (30–50% errors), only providing the correct solution for medium to high levels of excitation. An interaction was also observed for low s.f.d words between high lateral inhibition in the word sense level, and low-medium excitation, to produce the results most characteristic of PD semantic errors. The results of this study confirm that competitive and inhibitory mechanisms related to dopaminergic gain are responsible for impaired PD performance on semantic processing tasks.