SCRIBE is a neuroscience‑informed semantic framework that uses an ensemble of open‑source large language models to produce interpretable word‑level semantic vectors for any text. In a behavioral study with 921 participants we observed systematic context‑dependent shifts in human semantic feature judgments, and SCRIBE captured these shifts, reaching 75%–81% of the estimated human noise ceiling. When predicting neural responses recorded while participants listened to naturalistic stories, SCRIBE outperformed traditional non‑contextual semantic representations in language‑selective brain regions. Variance partitioning showed that SCRIBE contributed unique predictive information beyond measured syntactic features, indicating it captures aspects of meaning not accounted for by syntax alone. Together, these results establish SCRIBE as a scalable and interpretable foundation for studying natural language semantics in the brain.
Review