TALQ: A reproducible language model questionnaire for dense measurement of depressive and anxious states
Depression and anxiety are often assessed weeks or months apart with brief questionnaires covering few symptoms, leaving changes between assessments unobserved. Here, we introduce TALQ, a scoring system in which a large language model rates 52 symptoms from text, and use it to score 491 members of two online mental health communities every two weeks for a year. Ratings were stable across samples, predicted human annotations of interpersonal risk factors, low self-esteem and depression severity, and agreed with inventories completed by the same model. The scores mapped into 12 replicable symptom factors, including fearful preoccupation and social strain. Sleep disturbance and self-harm content consistently preceded overwhelm and emotional pain, respectively, two weeks later. People differed mainly in the states they occupied rather than transitions between them, and 13% entered in a crisis that remitted over months. Overall, our findings suggest that TALQ can measure expressed mental states densely and quantify longitudinal dynamics.
Authors
- Reza Abbasi-Asl (ORCID: https://orcid.org/0000-0001-7824-4628)
- Kira E. Poskanzer (ORCID: https://orcid.org/0000-0003-4830-8891)
- Joshua D. Woolley (ORCID: https://orcid.org/0000-0002-4378-0014)
- Maxine E Levesque
Institutions
- University of California, San Francisco (US)
Publication Details
- Journal
- medRxiv
- Published
- 2026-11-15
- DOI
- https://doi.org/10.64898/2026.11.12.26367797
- Primary Topic
- Mental Health via Writing
- Type
- preprint