Using computational semantics to study meaning in the brain

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics
• Neuroscience

Background:

• Human language comprehension involves understanding individual words and combining them to form novel sentence meanings.
• Cognitive neuroscience studies neural activation patterns for language processing, while computational linguistics uses vector-based models to represent meaning in artificial neural networks.
• Recent advances in large language models have spurred research integrating computational semantics with brain imaging studies of language processing.

Purpose of the Study:

• To review and integrate theoretical and experimental approaches to semantic processing in the brain.
• To provide an overview of vector-based semantics models for word and sentence meaning.
• To analyze studies using computational semantics and fMRI to understand brain-based language processing.

Main Methods:

• Overview of major vector-based semantics models (e.g., word embeddings).
• Review of traditional and modern neuroimaging approaches to studying semantics (lexical, compositional).
• Comparative analysis of 57 functional Magnetic Resonance Imaging (fMRI) studies employing vector-based semantics models.

Main Results:

• Identified common trends in using computational semantics to study brain language processing.
• Highlighted significant methodological inconsistencies across fMRI studies, complicating interpretation.
• Demonstrated the utility of vector-based models for investigating semantic processing.

Conclusions:

• Vector-based semantics models offer valuable insights into neural mechanisms of semantic processing.
• Greater methodological standardization is crucial for advancing research in this cross-disciplinary field.
• Integrating computational approaches with neuroimaging enhances our understanding of human language comprehension.