A Neural Network Simulation of Hallucinated “Voices” and Associated Speech Perception Impairments in Schizophrenic Patients

Abstract

The mechanism of hallucinated speech, a symptom commonly reported by schizophrenic patients, is unknown. The hypothesis that these hallucinations arise from pathologically altered working memory underlying speech perception was explored. A neural network computer simulation of contextually guided sequential word detection based on Elman (1990a,b) was studied. Pruning anatomic connections or reducing neuronal activation in working memory caused word “percepts” to emerge spontaneously (i.e., in the absence of external “speech inputs”), thereby providing a model of hallucinated speech. These simulations also demonstrated distinct patterns of word detection impairments when inputs were accompanied by varying levels of noise. In a parallel human study, the ability to shadow noisecontaminated, connected speech was assessed. Schizophrenic patients reporting hallucinated speech demonstrated a pattern of speech perception impairments similar to a simulated neural network with reduced anatomic connectivity and enhanced neuronal activation. Schizophrenic patients not reporting this symptom did not demonstrate these speech perception impairments. Neural network simulations and human empirical data, when considered together, suggested that the primary cause of hallucinated “voices” in schizophrenia is reduced neuroanatomic connectivity in verbal working memory.