arxivcs.CLcs.LG2026-06-25
Heterogeneous Neural Predictivity from Language Models During Naturalistic Comprehension
Language-model representations provide structured, high-dimensional annotations of naturalistic language stimuli and can serve as informative neural predictors during comprehension. We analyzed locked derived data from Brain Treebank, MEG-MASC, and Podcast ECoG with eight frozen…