Quantifying incoherence in speech: an automated methodology and novel application to schizophrenia.

Incoherent discourse, with a disjointed flow of ideas, is a cardinal symptom in several psychiatric and neurological conditions. However, measuring incoherence has often been complex and subjective. We sought to validate an objective, intrinsically reliable, computational approach to quantifying speech incoherence. Patients with schizophrenia and healthy control volunteers were administered a variety of language tasks. The speech generated was transcribed and the coherence computed using Latent Semantic Analysis (LSA). The discourse was also analyzed with a standard clinical measure of thought disorder. In word association and generation tasks LSA derived coherence scores were sensitive to differences between patients and controls, and correlated with clinical measures of thought disorder. In speech samples LSA could be used to localize where in sentence production incoherence occurs, predict levels of incoherence as well as whether discourse "belonged" to a patient or control. In conclusion, LSA can be used to assay disordered language production so as to both complement human clinical ratings as well as experimentally parse this incoherence in a theory-driven manner.