How the Brain Dynamically Constructs Sentence-Level Meanings From Word-Level Features.

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics
• Psycholinguistics

Background:

• Word meanings are thought to be embodied in neural systems, as proposed by Concept Attribute Representations (CAR) theory.
• Existing theories face challenges explaining how context dynamically influences the weighting of concept attributes.

Purpose of the Study:

• To address the challenge of context-dependent meaning construction in embodied theories.
• To introduce and evaluate the Context-dEpendent meaning REpresentations in the BRAin (CEREBRA) neural network model.

Main Methods:

• Developed the CEREBRA neural network model to quantify contextual effects on word meanings using brain imaging data.
• Conducted computational experiments to analyze changes in concept attributes and word representations within different sentence contexts.
• Performed behavioral analysis to validate the predictive power of CEREBRA's context-based representations.

Main Results:

• Demonstrated that word representations differ based on sentence context.
• Observed that contextual changes in concept attributes reveal unique conceptual combinations.
• Found that contextually updated word representations become more similar to other words in the sentence.

Conclusions:

• CEREBRA successfully models dynamic, context-dependent Concept Attribute Representations (CARs).
• The model provides actionable insights into how the brain constructs meaning dynamically.
• CEREBRA offers a valuable framework for interdisciplinary research on the mental lexicon and embodied cognition.