The abstract is read at the publisher; the summary is JClub's.
An open-weight LLM-based contextual information gain metric correlates with human ratings of event importance and causal centrality, separating causal from semantic centrality.
- Why it matters: Manual labeling of narrative event features is labor-intensive and subjective, limiting scaling to longer, more varied stimuli and slowing discovery of behaviorally and neurally relevant quantities.
- What they did: The authors compute contextual information gain with open-weight large language models and test whether it matches human judgments of event importance and causal centrality, including comparisons to semantic centrality.
- The result: The metric also captures shared coactivation in the default mode network, validating an automated, empirically grounded tool for identifying behaviorally and neurally relevant computational quantities in naturalistic narratives.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.