Brain states recur during longitudinal viewing and relate to content across narratives and sensory formats
- 1Massachusetts Institute of Technology
Abstract
Models of brain dynamics propose that brain activity revisits a repertoire of configurations, and that current input influences which configuration is expressed. How broadly this holds across a person's extended experience is unknown. We analyzed 38-57 hours of fMRI per participant in six adults watching the television series Friends. Participant-specific hidden Markov models identified 42-47 active brain states, and independent fits to separate portions of viewing recovered similar states with similar tendencies to recur. States matched across participants' separately fitted models tended to occur at the same moments of the episodes. Among states retained after temporal screening, features from pretrained video, audio, and language networks partly predicted state identity, most strongly from intermediate-to-late layers. Predictions learned from Friends generalized without retraining to new films, and language-based predictions extended to reading and audiobook listening. State usage corresponded across stimuli little beyond what each recording's spectra and covariance reproduced.
Related
- rpSTS Stimulation Disrupts Shared Neural Responses to Social Content during Naturalistic Viewing
- Association between viewing more media war content and higher intention to use cannabis in adolescents: the indirect effect of post-traumatic stress
- Ethological variability shapes the structure and content of sensory population codes
- Human medial frontal neurons represent integrated semantic content and sensory evidence to inform decision-making
- GBZ-base and GAF-base: Indexed pangenome file formats
BibTeX
@misc{chen2026brain,
title = {{Brain states recur during longitudinal viewing and relate to content across narratives and sensory formats}},
author = {Y. Chen and M. Ghavami and M. St-Laurent and L. Bellec and S. S. Ghosh},
howpublished = {bioRxiv},
year = {2026},
month = may,
doi = {10.64898/2026.05.31.729141},
url = {https://www.biorxiv.org/content/10.64898/2026.05.31.729141v2},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Brain states recur during longitudinal viewing and relate to content across narratives and sensory formats
AU - Y. Chen
AU - M. Ghavami
AU - M. St-Laurent
AU - L. Bellec
AU - S. S. Ghosh
PY - 2026
DA - 2026/05/31
PB - bioRxiv
DO - 10.64898/2026.05.31.729141
UR - https://www.biorxiv.org/content/10.64898/2026.05.31.729141v2
AB - Models of brain dynamics propose that brain activity revisits a repertoire of configurations, and that current input influences which configuration is expressed. How broadly this holds across a person's extended experience is unknown. We analyzed 38-57 hours of fMRI per participant in six adults watching the television series Friends. Participant-specific hidden Markov models identified 42-47 active brain states, and independent fits to separate portions of viewing recovered similar states with similar tendencies to recur. States matched across participants' separately fitted models tended to occur at the same moments of the episodes. Among states retained after temporal screening, features from pretrained video, audio, and language networks partly predicted state identity, most strongly from intermediate-to-late layers. Predictions learned from Friends generalized without retraining to new films, and language-based predictions extended to reading and audiobook listening. State usage corresponded across stimuli little beyond what each recording's spectra and covariance reproduced.
ER -