Files
Abstract
Analysis of speech samples is crucial for schizophrenia research, since language is the outer representation of inner thoughts, and speech abnormality effectively reflects brain deterioration. Detailed psycholinguistic analysis, however, often requires substantial linguistic expertise and tremendous amount of time on the part of the analyst, which are not always available. To demonstrate that sophisticated and useful psycholinguistic measures can be automated at various linguistic levels using cutting-edge NLP technologies, this dissertation describes the design of three NLP applications for schizophrenia research. They are Vocabulary Analyzer analyzing vocabulary rarity at the lexical level, D-Level Rater rating syntactic complexity at the syntactic level, and Idea Density Rater computing idea density at the semantic level. Speech samples from a schizophrenia experiment were used as a test bed for the usability of the software. Results show that, lexically, the schizophrenic patients in the experiment tend to use fewer rare words than the normal controls. Structure-wise, the patients speech features lowered syntactic complexity as measured with D-Level Scale. And, semantically, no significant difference in idea density was found between the speech samples of the patients and those of the controls.