Framework models how human brain understands words in context
SCRIBE: Scalable Contextual Representations of Interpretable Brain-Grounded Semantics
Artificial IntelligenceComputation and Language
Summary
Understanding how people grasp the meaning of words depending on their context is very tricky. The authors introduced SCRIBE, a new tool that uses several AI language models informed by brain science to create clear, word-by-word meanings from any text. They tested SCRIBE with almost a thousand people and brain scans while listening to stories, showing it matches how humans think and predicts brain activity related to language better than older methods. This tool helps explain how our brains handle meaning beyond simple grammar.
What this means in practice
- •For brain-computer interface developers: Improve real-time decoding of intended meaning by integrating context-aware semantic representations based on brain activity.
- •For natural language processing engineers: Enhance AI language understanding systems by incorporating brain-grounded, interpretable semantic features that adapt to context.
Authors
Camila Kolling, Omer Moussa, Emin Çelik, Mariya Toneva
Abstract
Understanding how the human brain constructs meaning from real-world language remains a fundamental challenge, as semantics depend on context that varies in natural settings. To enable the study of real-world linguistic meaning in the brain, we introduce SCRIBE, a framework grounded in a neuroscience-informed semantic space that leverages an ensemble of open-source large language models to generate interpretable, word-level semantic representations of arbitrary text. In a behavioral study with 921 participants, we show that human semantic feature judgments vary systematically with context, and that SCRIBE captures these shifts, reaching 75-81% of the estimated human noise ceiling. Using brain activity recorded while participants listen to naturalistic stories, we further demonstrate that SCRIBE predicts neural responses in language-selective brain regions, significantly outperforming non-contextual semantic representations. Variance partitioning reveals that SCRIBE explains unique brain-predictive information beyond syntactic features, suggesting that its representations capture aspects of linguistic meaning not accounted for by the measured syntactic features. Together, these findings establish SCRIBE as a scalable and interpretable foundation for studying natural language semantics in the brain.