Computer-extracted speech features from acoustic, semantic, and psycholinguistic domains can detect mental states after controlled administration of MDMA and intranasal oxytocin. In a double-blind, placebo-controlled study with 31 healthy adults, speech tasks during peak drug effects yielded cross-validated accuracies up to 87% in the training/validation set and 92% in independent datasets for classifying drug conditions. Oxytocin-driven changes were mostly captured by acoustic features related to emotion and prosody, while MDMA-related mental states manifested across multiple speech domains. The experimental task—whether involving interaction with another individual—also affected speech responses. These results suggest speech analysis can provide objective markers of drug-induced mental states.
Disturbances in self-experience—the sense of being the subject of one's own experiences and actions, and of being distinct from others—are central to schizophrenia. Traditionally assessed by manual interview rating, this study used natural language processing to analyze autobiographical narratives from 167 patients with schizophrenia or schizoaffective disorder and 90 healthy controls, totaling 490,000 words. Topics related to self-experience and agency were significantly more expressed in patients than controls and were decoupled from emotional tone, semantic coherence, and burden-related concepts. A classifier trained on these features discriminated patients from controls with an AUC of 0.80. These findings demonstrate that NLP can automatically detect higher-order metacognitive aspects of self-experience without explicit probing.