Predicting Neural Activity Patterns Associated with Sentences Using a Neurobiologically Motivated Model of Semantic...

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Linguistics

Background:

• Previous research primarily focused on predicting neural patterns for isolated words.
• Existing models often lack neurobiological interpretability.
• Understanding sentence-level semantic processing in the brain remains a challenge.

Purpose of the Study:

• To introduce and validate an approach for predicting neural representations of sentence meanings.
• To utilize a neurobiologically grounded semantic model incorporating diverse attributes.
• To bridge the gap between word-level semantics and sentence-level brain activity.

Main Methods:

• Developed a method to decompose sentence-level functional Magnetic Resonance Imaging (fMRI) data into individual word representations.
• Employed a neurobiological semantic model (sensory, motor, social, emotional, cognitive attributes) as a foundation.
• Used multiple regression to estimate activation patterns for semantic attributes and synthesized these to predict new sentence representations.

Main Results:

• Successfully predicted neural representations for new sentences by superposing word-level predictions.
• Region-of-interest analyses indicated highest prediction accuracy in left temporal and inferior parietal cortex.
• Demonstrated that semantic information is widely distributed across various brain regions.

Conclusions:

• A neurobiologically motivated semantic model can effectively decompose sentence-level fMRI data.
• Component word activation features can be recombined to predict activation patterns for novel sentences.
• This approach offers a pathway to understanding complex semantic processing in the brain.