This study investigated whether twelve automated Natural Language Processing (NLP) markers could differentiate transcribed speech excerpts from subjects at clinical high risk for psychosis, first episode psychosis patients, and healthy controls. Several measures showed significant differences between groups, including semantic coherence, speech graph connectivity, and an 'on-topic' measure, which outperformed the related measure of tangentiality. Most NLP measures were weakly related, suggesting they provide complementary information. Speech generated from picture descriptions (Thematic Apperception Test) and a story re-telling task outperformed free speech, highlighting the importance of speech generation method.

Sarah E. Morgan, Kelly Diederen, Petra E. Vértes, Samantha H. Y. Ip, Bo Wang, Bethany Thompson, Arsime Demjaha, Andrea De Micheli, Dominic Oliver, Maria Liakata, Paolo Fusar-Poli, Tom J. Spencer, Philip McGuire