A window into the intoxicated mind? Speech as an index of psychoactive drug effects

This study (2014) demonstrated with the example of MDMA that speech analysis can capture subtle differences in mental state in drugged versus sober individuals. The authors found that the speech of individuals dosed with MDMA showed closer proximity to such concepts as intimacy and empathy than usual.

Abstract

“Abused drugs can profoundly alter mental states in ways that may motivate drug use. These effects are usually assessed with self-report, an approach that is vulnerable to biases. Analyzing speech during intoxication may present a more direct, objective measure, offering a unique ‘window’ into the mind. Here, we employed computational analyses of speech semantic and topological structure after ±3,4-methylenedioxymethamphetamine (MDMA; ‘ecstasy’) and methamphetamine in 13 ecstasy users. In 4 sessions, participants completed a 10-min speech task after MDMA (0.75 and 1.5 mg/kg), methamphetamine (20 mg), or placebo. Latent Semantic Analyses identified the semantic proximity between speech content and concepts relevant to drug effects. Graph-based analyses identified topological speech characteristics. Group-level drug effects on semantic distances and topology were assessed. Machine-learning analyses (with leave-one-out cross-validation) assessed whether speech characteristics could predict drug condition in the individual subject. Speech after MDMA (1.5 mg/kg) had greater semantic proximity than placebo to the concepts friend, support, intimacy, and rapport. Speech on MDMA (0.75 mg/kg) had greater proximity to empathy than placebo. Conversely, speech on methamphetamine was further from compassion than placebo. Classifiers discriminated between MDMA (1.5 mg/kg) and placebo with 88% accuracy, and MDMA (1.5 mg/kg) and methamphetamine with 84% accuracy. For the two MDMA doses, the classifier performed at chance. These data suggest that automated semantic speech analyses can capture subtle alterations in mental state, accurately discriminating between drugs. The findings also illustrate the potential for automated speech-based approaches to characterize clinically relevant alterations to mental state, including those occurring in psychiatric illness.”

Authors: Gillinder Bedi, Guillermo A. Cecchi, Diego F. Slezak, Facundo Carrillo, Mariano Sigman & Harriet de Wit

Summary

INTRODUCTION

Abused drugs alter mental states, and these mental state alterations are intimately involved in motivations to use drugs.

People have tried to communicate drug-related alterations to mental state through artistic and literary approaches for as long as drugs have been used. The scientific perspective has focused on two main approaches to studying drug-induced mental states: retrospective descriptive reports and standardized self-report measures.

Several studies have examined the quantity of speech emitted during intoxication, but they did not examine speech content. Here, we investigate the hypothesis that speech content can provide a unique window into thought.

Early studies of ketamine intoxication found disordered speech, which could be characterized with automated methods. Recent advances in computational measurement allow quantitative speech analysis using automated methods, which could be used to characterize the effects of emerging drugs.

We used automated analyses of free speech to measure the effects of two abused drugs, MDMA and methamphetamine, relative to placebo. We focused on several specific mental states that might differentiate the effects produced by the two drugs.

The semantic content of a word is determined by its relationship to other words in a language, and can be measured quantitatively using Latent Semantic Analysis. This study employed multivariate machine-learning methods to assess whether speech characteristics identified could differentiate between drug conditions.

Participants

Healthy volunteers reported ecstasy use X twice and underwent comprehensive medical and psychiatric screening. All participants provided written informed consent and were debriefed at completion.

Design and Protocol

Participants received MDMA, methamphetamine, or placebo for four 5-h sessions. They abstained from food, cannabis, alcohol, and illicit drugs for 48 h before each session.

Participants took cardiovascular and self-report subjective measurements, ingested a size 00 gelatin capsule containing MDMA hydrochloride, and completed behavioral tasks beginning 65 min after the capsule.

Assessment Measures

During the free speech task, participants spoke to a research assistant about a person of importance in their life. The speech was recorded and the research assistant applied active listening skills to minimize their impact on speech content.

Analytic Approach

We manually transcribed audio recordings, preprocessed them using the Natural Language Toolkit (NLTK), and then lemmatized each word using the WordNet lemmatizer. This resulted in a string of N tokens for each interview, without punctuation marks or symbols.

We used a machine learning approach to assess group-level effects of drug condition on semantic proximity values, and a graph-based approach to assess structural components of speech.

Semantic proximity is a notion of similarity between words that can be captured by dictionaries, thesauri, and similar databases. LSA is a high-dimensional associative model that assumes that semantically related words will necessarily cooccur in texts with coherent topics.

We used TASA, a collection of educational materials compiled by Touchstone Applied Science Associates, to lemmatize the corpus and perform SVD on the term-frequency matrix to obtain the decomposition. The SVD matrix may be cropped to reduce dimensionality while conserving the range of the original matrix.

We selected words hypothesized to be affected by MDMA but not by prototypical psychostimulants, and assessed group-level drug effects on the mean semantic proximity values for each concept selected using repeated-measures ANOVA followed by planned comparisons between placebo and active drug conditions.

Prediction of drug condition using pattern classification. We used an off-the-shelf Support Vector Machine (SVM) classifier and implemented leave-subject-out cross-validation on the data set consisting of N 1 4 13 subjects and 4 conditions. We implemented a four-way classifier via an off-the-shelf linear discriminant analysis (LDA), using rapport, support, intimacy, and friend as semantic similarity measures, plus verbosity. We used a leave-subject-out cross-validation scheme.

Graph-based analysis of speech structure can be used to identify psychosis from speech. This method considers individual words to be nodes in a network, whereas edges represent grammatical or semantic relationships linking nodes.

We used a method to analyze the structural features of speech to assess group-level differences between drug conditions, including the number of different tokens, the number of unique transitions between different nodes, and the number of loops.

RESULTS

13 participants provided consent for speech recording. They smoked marijuana 9.5 days/month and drank 7.4 drinks/week.

Drug Effects on Semantic Proximity to Concepts of Interest

In the drug conditions, speech had greater proximity to several concepts compared with placebo, including friend, support, intimacy, rapport, and empathy. Speech on METH had lower proximity to compassion than speech on PBO.

Prediction of Drug Condition Using Pattern Classification

We implemented a SVM classification to assess whether a combination of the proximity values could differentiate drug conditions. The four-way LDA classifier achieved a classification accuracy of 59%.

Graph-Based Analyses of the Structure of Speech

We did not expect to find evidence of disorganized speech structure after MDMA, but we did observe a small but statistically significant difference in the number of 1-loops between METH and PBO.

DISCUSSION

MDMA (0.5 mg/kg) increased the semantic proximity of speech to several concepts, including friend, support, intimacy, rapport, and empathy, and methamphetamine (20 mg) decreased proximity to compassion and increased verbosity.

To our knowledge, this is the first study using semantic and topological speech characteristics to study drug-related mental-state alterations. The results show that most effects of MDMA on speech content are dose dependent, emerging only at the higher MDMA dose.

MDMA produces psychoactive effects via serotonin release, with euphorigenic effects mediated by dopamine type 2 receptors and a potential role for norepinephrine. Oxytocin release is implicated in the prosocial effects of MDMA.

In this study, we showed that a multivariate combination of speech characteristics could predict drug condition in the individual subject, and could also differentiate between manic and schizophrenic patients based on speech structure.

Previous studies show that several drugs alter speech quantity, and our finding that methamphetamine increased verbosity is consistent with these findings. However, the present study did not observe disrupted speech structure on MDMA, which suggests that MDMA does not alter the formal structure of speech.

These findings have several potential implications, including the use of automated speech analysis as an adjunct to other approaches to better characterize drug-related mental-state alterations.

This study had limitations, such as using only two MDMA doses and one methamphetamine dose, and a small sample size. However, the semantic analysis employed was accurate and the study used a data-driven, ‘black-box’ analysis.

We used a graph-based approach to assess speech structure to detect effects of MDMA and psychosis, and it was also sensitive to effects of methamphetamine. However, the best method for psychiatric and psychopharmacological applications of automated speech analysis remains an important empirical question.

Future research will need to examine the effects of the speech task selected, as earlier studies used different tasks or asked subjects to describe a dream or a movie.

A task that taps the prosocial effects of MDMA may have resulted in practice effects or variability between sessions that was unrelated to drug effects. A further focus for future research will be the relationship between automated methods of speech analyses and more traditional, manual approaches to coding.

The present study focused on mental-state alterations after drugs, but the results could be used for other clinically relevant mental-state changes. This study also supports the use of automated semantic speech analyses to provide diagnostic or prognostic information about individual patients.

Automated semantic speech analyses could be a useful addition to existing methods to characterize alterations to mental state after drugs.