Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2502.09487 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911702684008448 |
|---|---|
| author | Onysk, Jakub Huys, Quentin J. M. |
| author_facet | Onysk, Jakub Huys, Quentin J. M. |
| contents | Characterising how we verbalise our feelings is central to psychological assessment and intervention, yet the mapping between narrative and affective state remains poorly understood. Across two large studies (n=1257), we parameterised the structure and dynamics of depressive states by quantifying participants' internal narratives through large-language-model representations and their subspaces. In Study 1, we found verbal descriptions of symptom-specific thoughts captured granular information predictive of standardised, self-reported depression scores. Critically, we show preserving the specific covariance between symptoms is essential for construct validity, suggesting high-dimensional text representations mirror the latent geometry of the disorder. Study 2 probed the temporal dynamics of this relationship as participants engaged with emotional narratives. We found quantified changes in internal narratives led to changes in self-report, while the baseline narrative severity predicted the magnitude of subsequent affective change. By framing affect as a computational state, our results highlight its core, therapeutically pertinent functions: constraining the structure of internal narratives and integrating context to shape self-report. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_09487 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Internal narratives parameterise affective states Onysk, Jakub Huys, Quentin J. M. Computation and Language Artificial Intelligence Machine Learning Characterising how we verbalise our feelings is central to psychological assessment and intervention, yet the mapping between narrative and affective state remains poorly understood. Across two large studies (n=1257), we parameterised the structure and dynamics of depressive states by quantifying participants' internal narratives through large-language-model representations and their subspaces. In Study 1, we found verbal descriptions of symptom-specific thoughts captured granular information predictive of standardised, self-reported depression scores. Critically, we show preserving the specific covariance between symptoms is essential for construct validity, suggesting high-dimensional text representations mirror the latent geometry of the disorder. Study 2 probed the temporal dynamics of this relationship as participants engaged with emotional narratives. We found quantified changes in internal narratives led to changes in self-report, while the baseline narrative severity predicted the magnitude of subsequent affective change. By framing affect as a computational state, our results highlight its core, therapeutically pertinent functions: constraining the structure of internal narratives and integrating context to shape self-report. |
| title | Internal narratives parameterise affective states |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2502.09487 |