Interdisciplinary approach to identify language markers for post-traumatic stress disorder using machine learning and deep learning.

Post-traumatic stress disorder (PTSD) lacks clear biomarkers in clinical practice. Language as a potential diagnostic biomarker for PTSD is investigated in this study. We analyze an original cohort of 148 individuals exposed to the November 13, 2015, terrorist attacks in Paris. The interviews, conducted 5-11 months after the event, include individuals from similar socioeconomic backgrounds exposed to the same incident, responding to identical questions and using uniform PTSD measures. Using this dataset to collect nuanced insights that might be clinically relevant, we propose a three-step interdisciplinary methodology that integrates expertise from psychiatry, linguistics, and the Natural Language Processing (NLP) community to examine the relationship between language and PTSD. The first step assesses a clinical psychiatrist's ability to diagnose PTSD using interview transcription alone. The second step uses statistical analysis and machine learning models to create language features based on psycholinguistic hypotheses and evaluate their predictive strength. The third step is the application of a hypothesis-free deep learning approach to the classification of PTSD in our cohort. Results show that the clinical psychiatrist achieved a diagnosis of PTSD with an AUC of 0.72. This is comparable to a gold standard questionnaire (Area Under Curve (AUC) ≈ 0.80). The machine learning model achieved a diagnostic AUC of 0.69. The deep learning approach achieved an AUC of 0.64. An examination of model error informs our discussion. Importantly, the study controls for confounding factors, establishes associations between language and DSM-5 subsymptoms, and integrates automated methods with qualitative analysis. This study provides a direct and methodologically robust description of the relationship between PTSD and language. Our work lays the groundwork for advancing early and accurate diagnosis and using linguistic markers to assess the effectiveness of pharmacological treatments and psychotherapies.

What Does the Sleeping Brain Say? Syntax and Semantics of Sleep Talking in Healthy Subjects and in Parasomnia Patients.

Objectives: Speech is a complex function in humans, but the linguistic characteristics of sleep talking are unknown. We analyzed sleep-associated speech in adults, mostly (92%) during parasomnias. Methods: The utterances recorded during night-time video-polysomnography were analyzed for number of words, propositions and speech episodes, frequency, gaps and pauses (denoting turn-taking in the conversation), lemmatization, verbosity, negative/imperative/interrogative tone, first/second person, politeness, and abuse. Results: Two hundred thirty-two subjects (aged 49.5 ± 20 years old; 41% women; 129 with rapid eye movement [REM] sleep behavior disorder and 87 with sleepwalking/sleep terrors, 15 healthy subjects, and 1 patient with sleep apnea speaking in non-REM sleep) uttered 883 speech episodes, containing 59% nonverbal utterance (mumbles, shouts, whispers, and laughs) and 3349 understandable words. The most frequent word was "No": negations represented 21.4% of clauses (more in non-REM sleep). Interrogations were found in 26% of speech episodes (more in non-REM sleep), and subordinate clauses were found in 12.9% of speech episodes. As many as 9.7% of clauses contained profanities (more in non-REM sleep). Verbal abuse lasted longer in REM sleep and was mostly directed toward insulting or condemning someone, whereas swearing predominated in non-REM sleep. Men sleep-talked more than women and used a higher proportion of profanities. Apparent turn-taking in the conversation respected the usual language gaps. Conclusions: Sleep talking parallels awake talking for syntax, semantics, and turn-taking in conversation, suggesting that the sleeping brain can function at a high level. Language during sleep is mostly a familiar, tensed conversation with inaudible others, suggestive of conflicts.

Differences between girls and boys in emerging language skills: evidence from 10 language communities.

The present study explored gender differences in emerging language skills in 13,783 European children from 10 non-English language communities. It was based on a synthesis of published data assessed with adapted versions of the MacArthur-Bates Communicative Development Inventories (CDIs) from age 0.08 to 2.06. The results showed that girls are slightly ahead of boys in early communicative gestures, in productive vocabulary, and in combining words. The difference increased with age. Boys were not found to be more variable than girls. Despite extensive variation in language skills between language communities, the difference between girls and boys remained. This suggests that the difference is caused by robust factors that do not change between language communities.

Syntactic and lexical context of pauses and hesitations in the discourse of Alzheimer patients and healthy elderly subjects.

Psycholinguistic studies dealing with Alzheimer's disease (AD) commonly consider verbal aspects of language. In this article, we investigated both verbal and non-verbal aspects of speech production in AD. We used pauses and hesitations as markers of planning difficulties and hypothesized that AD patients show different patterns in the process of discourse production. We compared the distribution, the duration and the frequency of speech dysfluencies in the spontaneous discourse of 20 AD patients with 20 age, gender and socio-economically matched healthy peers. We found that patients and controls differ along several lines: patients' discourse displays more frequent silent pauses, which occur more often outside syntactic boundaries and are followed by more frequent words. Overall patients show more lexical retrieval and planning difficulties, but where controls signal their planning difficulties using filled pauses, AD patients do not.

Does structural complexity necessarily imply processing difficulty?

Our goal is to establish a link between the time needed to plan a sentence containing an embedded clause and the structure of this sentence. Contrary to a traditional monolithic conception of subordination, three types of embeddings were considered, depending on their degree of syntactic integration: subcategorized, modifier and pseudo-embedded clauses. We hypothesized that in the case of subcategorization, fewer pauses should occur between the matrix and the subordinate clause since the latter is required by the lexical properties of verbs. By contrast, pseudo-embedded clauses are the less integrated. Hence, they should exhibit planning characteristics similar to the ones of simple sentences, the matrix clause and the subordinate clauses being planned in two steps. Twenty texts produced by French speaking adults were recorded. Pauses were characterized according to their duration and position. Globally, both predictions were confirmed. We conclude that supposedly complex sentences are not necessarily difficult to process.