Fig. 2: Shifts in semantic contexts of the movie are tied to shifts in brain state dynamics.

We used NLP tools from natural language processing to decompose the movie subtitle corpus into 200 unique semantic contexts across thousands of vocabulary entries. The timepoint-wise quantification of semantic context occurrences offered the basis to link meaning with brain states. The contexts’ associated wordclouds help interpret the semantics-brain links. a Middle left: The wordcloud for exemplary semantic context No. 103. The captured movie events centered around the character “Lieutenant Dan”. Rest: This representative semantic context described recurring events for the same character (Lieutenant Dan), the associated movie clips (top row) were linked with the projected embedding peaks by an arrow, on which the text illustrated the brief movie events. b Subject hidden Markov model: each dynamic brain state (of subject 1’s DN&AM model) was correlated with certain semantic contexts (colored according to brain state; see Supplementary Fig. 1 for choice of four states). The collection of remaining 160 contexts is shown in gray. As such, we performed a semantic dissection (via latent semantic analysis) of recurring movie themes into 200 unique semantic contexts, which related to complementary contextual information and distinct brain states. Source data are provided as a Source Data file.