Facundo Carrillo

Automated analysis of free speech predicts psychosis onset in high-risk youths.

Background/Objectives:Psychiatry lacks the objective clinical tests routinely used in other specializations. Novel computerized methods to characterize complex behaviors such as speech could be used to identify and predict psychiatric illness in individuals.AIMS:In this proof-of-principle study, our aim was to test automated speech analyses combined with Machine Learning to predict later psychosis onset in youths at clinical high-risk (CHR) for psychosis.Methods:Thirty-four CHR youths (11 females) had baseline interviews and were assessed quarterly for up to 2.5 years; five transitioned to psychosis. Using automated analysis, transcripts of interviews were evaluated for semantic and syntactic features predicting later psychosis onset. Speech features were fed into a convex hull classification algorithm with leave-one-subject-out cross-validation to assess their predictive value for psychosis outcome. The canonical correlation between the speech features and prodromal symptom ratings was computed.Results:Derived speech features included a Latent Semantic Analysis measure of semantic coherence and two syntactic markers of speech complexity: maximum phrase length and use of determiners (e.g., which). These speech features predicted later psychosis development with 100% accuracy, outperforming classification from clinical interviews. Speech features were significantly correlated with prodromal symptoms.Conclusions:Findings support the utility of automated speech analysis to measure subtle, clinically relevant mental state changes in emergent psychosis. Recent developments in computer science, including natural language processing, could provide the foundation for future development of objective clinical tests for psychiatry.

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

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.

Prediction of psychosis across protocols and risk cohorts using automated language analysis

Language and speech are the primary source of data for psychiatrists to diagnose and treat mental disorders. In psychosis, the very structure of language can be disturbed, including semantic coherence (e.g., derailment and tangentiality) and syntactic complexity (e.g., concreteness). Subtle disturbances in language are evident in schizophrenia even prior to first psychosis onset, during prodromal stages. Using computer‐based natural language processing analyses, we previously showed that, among English‐speaking clinical (e.g., ultra) high‐risk youths, baseline reduction in semantic coherence (the flow of meaning in speech) and in syntactic complexity could predict subsequent psychosis onset with high accuracy. Herein, we aimed to cross‐validate these automated linguistic analytic methods in a second larger risk cohort, also English‐speaking, and to discriminate speech in psychosis from normal speech. We identified an automated machine‐learning speech classifier – comprising decreased semantic coherence, greater variance in that coherence, and reduced usage of possessive pronouns – that had an 83% accuracy in predicting psychosis onset (intra‐protocol), a cross‐validated accuracy of 79% of psychosis onset prediction in the original risk cohort (cross‐protocol), and a 72% accuracy in discriminating the speech of recent‐onset psychosis patients from that of healthy individuals. The classifier was highly correlated with previously identified manual linguistic predictors. Our findings support the utility and validity of automated natural language processing methods to characterize disturbances in semantics and syntax across stages of psychotic disorder. The next steps will be to apply these methods in larger risk cohorts to further test reproducibility, also in languages other than English, and identify sources of variability. This technology has the potential to improve prediction of psychosis outcome among at‐risk youths and identify linguistic targets for remediation and preventive intervention. More broadly, automated linguistic analysis can be a powerful tool for diagnosis and treatment across neuropsychiatry.

Fast distributed dynamics of semantic networks via social media

We investigate the dynamics of semantic organization using social media, a collective expression of human thought. We propose a novel, time-dependent semantic similarity measure (TSS), based on the social network Twitter. We show that TSS is consistent with static measures of similarity but provides high temporal resolution for the identification of real-world events and induced changes in the distributed structure of semantic relationships across the entire lexicon. Using TSS, we measured the evolution of a concept and its movement along the semantic neighborhood, driven by specific news/events. Finally, we showed that particular events may trigger a temporary reorganization of elements in the semantic network.

Natural speech algorithm applied to baseline interview data can predict which patients will respond to psilocybin for treatment-resistant depression

BACKGROUND Natural speech analytics has seen some improvements over recent years, and this has opened a window for objective and quantitative diagnosis in psychiatry. Here, we used a machine learning algorithm applied to natural speech to ask whether language properties measured before psilocybin for treatment-resistant can predict for which patients it will be effective and for which it will not. METHODS A baseline autobiographical memory interview was conducted and transcribed. Patients with treatment-resistant depression received 2 doses of psilocybin, 10 mg and 25 mg, 7 days apart. Psychological support was provided before, during and after all dosing sessions. Quantitative speech measures were applied to the interview data from 17 patients and 18 untreated age-matched healthy control subjects. A machine learning algorithm was used to classify between controls and patients and predict treatment response. RESULTS Speech analytics and machine learning successfully differentiated depressed patients from healthy controls and identified treatment responders from non-responders with a significant level of 85% of accuracy (75% precision). CONCLUSIONS Automatic natural language analysis was used to predict effective response to treatment with psilocybin, suggesting that these tools offer a highly cost-effective facility for screening individuals for treatment suitability and sensitivity. LIMITATIONS The sample size was small and replication is required to strengthen inferences on these results.

Automated Speech Analysis for Psychosis Evaluation

Psychosis is a mental syndrome associated to loss of contact with reality which may arise in patients with different diseases, such as schizophrenia or bipolar disorder. Symptoms include hallucinations, confused and disturbed thoughts or lack of self-awareness. Recent studies have found that psychotic patients can be objectively screened using graph-theoretical algorithms for speech analysis. This analysis often relies in manually executed tasks such as syntagma generation, text splitting or manual feature selection for classification. To solve this fundamental limitation, we use three fully-automated text analysis tools graph generation methods. In addition, since aspects of psychosis may be manifested in semantic aspects of speech, we also developed a semantic features index based on speech coherence. We show that using this combined approach, classifications obtained from automatic techniques are higher than 85 % in a database of 20 schizophrenic patients, with similar results to previous works. In summary, here we develop and validate a new tool for automated speech processing which includes semantic and structural aspects. The tool performs similar to manual screening procedures providing a new method to complement standard psychometric scales and fostering automated psychiatric diagnosis.

26.4 LANGUAGE DISTURBANCE AS A PREDICTOR OF PSYCHOSIS ONSET IN YOUTH AT ENHANCED CLINICAL RISK

Abstract Background Language offers a privileged view into the mind; it is the basis by which we infer others’ thoughts. Subtle language disturbance is evident in schizophrenia prior to psychosis onset, including decreases in coherence and complexity, as measured using clinical ratings in familial and clinical high-risk (CHR) cohorts. Bearden et al previously used manual linguistic analysis of baseline speech transcripts in CHR to show that illogical and referential thinking, and poverty of content, predict later psychosis onset. Then, Bedi et al used automated natural language processing (NLP) of CHR transcripts to show that decreased semantic coherence and reduction in syntactic complexity predicted psychosis onset. To determine validity and reproducibility, we have applied automated NLP methods, with machine learning, to Bearden’s original CHR transcripts to identify a language profile predictive of psychosis. Methods Participants in the Bearden UCLA cohort include 59 CHR, of whom 19 developed psychosis (CHR+) within 2 years, whereas 40 did not (CHR-), as well as 16 recent-onset psychosis and 21 healthy individuals, similar in demographics; speech was elicited using Caplan’s “Story Game. Participants in the Bedi NYC cohort include 34 CHR (29 CHR+), with speech elicited using open-ended interview. Speech was audiotaped, transcribed, de-identified and then subjected to latent semantic analysis to determine coherence and part-of-speech tagging to characterize syntactic structure and complexity. A machine-learning speech classifier of psychosis onset was derived from the UCLA CHR cohort, and then applied both to the NYC CHR cohort and to the UCLA psychosis/control comparison, with convex hull (three-dimension depiction of model) and receiver operating characteristics analyses. Correlational analyses with demographics, symptoms and manual linguistic features were also done. Results A four-factor model language classifier derived from the UCLA CHR cohort that comprised three semantic coherence variables and one syntax (usage of possessive pronouns) predicted psychosis t with accuracy of 83% (intra-protocol) for UCLA CHR, 79% (cross-protocol) for NYC CHR, and 72% for discriminating psychosis from normal speech (UCLA psychosis/control). Convex hulls were defined as the smallest space containing all datapoints within a set for CHR- or healthy controls: these convex hulls showed substantial overlap, with CHR+ and psychosis speech datapoints largely outside these convex hulls. Coherence was associated with age, but speech variables did not vary by gender, race, or socioeconomic status in this study. While automated text features were unrelated to prodromal symptom severity, they were highly correlated with manual text features (r = 0.7, p < .000001). Discussion In this small preliminary study, we identified and cross-validated a robust language classifier of psychosis risk that comprised measures of semantic coherence (flow of meaning in language) and syntactic usage (usage of possessive pronouns). This classifier had utility in discriminating speech in individuals with recent-onset psychosis from the norm. It demonstrated concurrent validity in that it was highly correlated with manual linguistic features previously identified by Bearden et al, important as automated methods are fast and inexpensive. Automated language features were unrelated to sex, ethnicity or social class in these small samples, and semantic coherence increased with age, consistent with prior studies of normal language development. Of interest, overlapping convex hulls could be defined for groups of individuals without psychosis (UCLA CHR-, NYC CHR- and UCLA healthy), suggesting a constrained hull of normal language in respect to syntax and semantics, from which pre-psychosis and psychosis speech deviates. The RDoC linguistic corpus-based variables of semantic coherence and syntactic structure hold promise as biomarkers of psychosis risk and expression, with initial validation and reproducibility. Next steps in biomarker development include larger multisite studies with standardization of protocols for speech elicitation, test-retest, and attention to traction/feasibility, acceptability, cost, and utility. Mechanistic studies can also yield neural and physiological correlates of abnormal semantic coherence and syntax.

Characterization of the relationship between semantic and structural language features in psychiatric diagnosis

Psychiatry describes speech symptoms that are indicative of disorganized thought, but measuring them is not easy. With natural language processing tools, it is possible to quantify psychiatric symptoms. Graph representations of word trajectories and semantic incoherence have independently been shown to predict the Schizophrenia diagnosis. Both analyses assess thought organization through speech, but the relationship between them is unknown. To fill this gap, here we characterize the relationship between structural and semantic features of free verbal reports from 60 patients and matched controls. Graph connectedness is inversely correlated to semantic incoherence and both explain 54% of negative symptoms variance.