Very little is known about the underlying causes of schizophrenia. Currently, the most reliable means for diagnosing schizophrenia is to observe certain disturbances of language, such as (1) positive thought disorder (i.e. derailed conversational language), (2) delusion formation of the idee fixe type, and (3) negative thought disorder (i.e. curtailed language output). We are using DISCERN, a connectionist natural language processing system that learns stories in order to paraphrase them and answer questions about them, to test hypotheses about the neuropathology that underlies schizophrenia: excessive developmental pruning of network connectivity, dopamine instability of dynamic learning, and semantic network hyperpriming. The original DISCERN model was extended to learn a larger corpus of stories that consist of multiple scripts and include emotion-coding and self-reference. It is then lesioned in several ways modeling the possible underlying causes and the degree to which the system displays the characteristic symptoms of schizophrenia is observed in each case. A parallel, pilot study will test normal and schizophrenic human subjects on similar story-processing tasks, helping to refine the model as well as allowing the results from the model to guide clinical research.