The abstract is read at the publisher; the summary is JClub's.
In high-level visual cortex across THINGS, NSD, and BMD, a shared cross-modal representational component increases along the hierarchy and dominates explained variance, especially in FFA/EBA.
- Why it matters: It was unclear how much vision and language predictors reflect shared versus unique aspects of neural responses, how this changes across the visual hierarchy and category-selective regions, and whether convergence depends on stimulus type.
- What they did: Using RSA and variance partitioning on three fMRI datasets (N = 21–35 per dataset), the study related vision DNN features and LLM embeddings to similarity judgements of visual stimuli and their language descriptions.
- The result: Vision-language convergence was higher within behaviour than within models; LLM alignment with vision-based similarity increased from objects to dynamic events, while vision DNN alignment stayed stable, and LLMs showed largest unique contributions in FFA and EBA.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.