Roman Kotov

Panagiotis Kaliosis, Adithya V Ganesan, Oscar N. E. Kjell et al.

Large language models (LLMs) are increasingly being used in a zero-shot fashion to assess mental health conditions, yet we have limited knowledge on what factors affect their accuracy. In this study, we utilize a clinical dataset of natural language narratives and self-reported PTSD severity scores from 1,437 individuals to comprehensively evaluate the performance of 11 state-of-the-art LLMs. To understand the factors affecting accuracy, we systematically varied (i) contextual knowledge like subscale definitions, distribution summary, and interview questions, and (ii) modeling strategies including zero-shot vs few shot, amount of reasoning effort, model sizes, structured subscales vs direct scalar prediction, output rescaling and nine ensemble methods. Our findings indicate that (a) LLMs are most accurate when provided with detailed construct definitions and context of the narrative; (b) increased reasoning effort leads to better estimation accuracy; (c) performance of open-weight models (Llama, Deepseek), plateau beyond 70B parameters while closed-weight (o3-mini, gpt-5) models improve with newer generations; and (d) best performance is achieved when ensembling a supervised model with the zero-shot LLMs. Taken together, the results suggest choice of contextual knowledge and modeling strategies is important for deploying LLMs to accurately assess mental health.

Adithya V Ganesan, Vasudha Varadarajan, Oscar NE Kjell et al.

While NLP typically treats documents as independent and unordered samples, in longitudinal studies, this assumption rarely holds: documents are nested within authors and ordered in time, forming person-indexed, time-ordered $\textit{behavioral sequences}$. Here, we demonstrate the need for and propose a longitudinal modeling and evaluation paradigm that consequently updates four parts of the NLP pipeline: (1) evaluation splits aligned to generalization over people ($\textit{cross-sectional}$) and/or time ($\textit{prospective}$); (2) accuracy metrics separating between-person differences from within-person dynamics; (3) sequence inputs to incorporate history by default; and (4) model internals that support different $\textit{coarseness}$ of latent state over histories (pooled summaries, explicit dynamics, or interaction-based models). We demonstrate the issues ensued by traditional pipeline and our proposed improvements on a dataset of 17k daily diary transcripts paired with PTSD symptom severity from 238 participants, finding that traditional document-level evaluation can yield substantially different and sometimes reversed conclusions compared to our ecologically valid modeling and evaluation. We tie our results to a broader discussion motivating a shift from word-sequence evaluation toward $\textit{behavior-sequence}$ paradigms for NLP.

Rajath Rao, Adithya Ganesan, Oscar Kjell et al.

Current speech encoding pipelines often rely on an additional text-based LM to get robust representations of human communication, even though SotA speech-to-text models often have a LM within. This work proposes an approach to improve the LM within an audio model such that the subsequent text-LM is unnecessary. We introduce WhiSPA (Whisper with Semantic and Psychological Alignment), which leverages a novel audio training objective: contrastive loss with a language model embedding as a teacher. Using over 500k speech segments from mental health audio interviews, we evaluate the utility of aligning Whisper's latent space with semantic representations from a text autoencoder (SBERT) and lexically derived embeddings of basic psychological dimensions: emotion and personality. Over self-supervised affective tasks and downstream psychological tasks, WhiSPA surpasses current speech encoders, achieving an average error reduction of 73.4% and 83.8%, respectively. WhiSPA demonstrates that it is not always necessary to run a subsequent text LM on speech-to-text output in order to get a rich psychological representation of human communication.

Youngseo Son, Sean A. P. Clouston, Roman Kotov et al.

Background: Oral histories from 9/11 responders to the World Trade Center (WTC) attacks provide rich narratives about distress and resilience. Artificial Intelligence (AI) models promise to detect psychopathology in natural language, but they have been evaluated primarily in non-clinical settings using social media. This study sought to test the ability of AI-based language assessments to predict PTSD symptom trajectories among responders. Methods: Participants were 124 responders whose health was monitored at the Stony Brook WTC Health and Wellness Program who completed oral history interviews about their initial WTC experiences. PTSD symptom severity was measured longitudinally using the PTSD Checklist (PCL) for up to 7 years post-interview. AI-based indicators were computed for depression, anxiety, neuroticism, and extraversion along with dictionary-based measures of linguistic and interpersonal style. Linear regression and multilevel models estimated associations of AI indicators with concurrent and subsequent PTSD symptom severity (significance adjusted by false discovery rate). Results: Cross-sectionally, greater depressive language (beta=0.32; p=0.043) and first-person singular usage (beta=0.31; p=0.044) were associated with increased symptom severity. Longitudinally, anxious language predicted future worsening in PCL scores (beta=0.31; p=0.031), whereas first-person plural usage (beta=-0.37; p=0.007) and longer words usage (beta=-0.36; p=0.007) predicted improvement. Conclusions: This is the first study to demonstrate the value of AI in understanding PTSD in a vulnerable population. Future studies should extend this application to other trauma exposures and to other demographic groups, especially under-represented minorities.