Automated Quantitative Semantic Analysis of Speech Discriminates Acute Effects of ±3,4-Methylenedioxymethamphetamine and Methamphetamine

BEDI, GILLINDER; GUILLERMO A CECCHI; DIEGO F SLEZAK; FACUNDO CARRILLO; HARRIET DE WIT

Conferencia; BIOLOGICAL PSYCHIATRY CONFERENCE; 2013

Abused drugs can produce profound mental state alterations. Measuring these effects by self-report relies on the capacity to accurately report introspective experiences. Analyzing the content of speech during intoxication may present a more direct ?window? into these effects. We employed computational analysis of speech semantic structure to study mental state alterations after ±3,4-methylenedioxyethamphetamine (MDMA) and methamphetamine. Methods: In a 4-session, double-blind outpatient study 13 healthy ecstasy users (9 M; 4 F) completed a 10-minute speech after oral MDMA (0.75, 1.5 mg/kg), methamphetamine (20 mg) or placebo. Latent Semantic Analyses identified the semantic distance between speech content and concepts relevant to drug effects. Group-level drug effects on semantic distances were assessed. Machine-learning analyses (forced-choice discrimination with leave-one-out cross-validation) assessed individual-level condition prediction using semantic similarity to ?rapport?, ?support? and ?intimacy?, and verbosity (no. words). Results: Speech on MDMA (1.5 mg/kg) was closer to ?friend? than on placebo, MDMA (0.75 mg/kg) and methamphetamine. Speech on MDMA (0.75, 1.5 mg/kg) was closer to ?rapport? than on methamphetamine. Speech on MDMA (0.75 mg/kg) was closer to ?empathy? than on placebo and MDMA (1.5 mg/kg). Classifiers discriminated between MDMA (1.5 mg/kg) and placebo with 83% accuracy, and MDMA (1.5 mg/kg) and methamphetamine with 95% accuracy. The only conditions not classified above chance were the two MDMA doses. Conclusions: Automated semantic speech analyses capture mental state alterations due to acute drug effects. Moreover, such approaches can identify similarities and differences in drug-induced mental state alterations, discriminating between drug types with a high degree of accuracy.