fMRI reveals language-specific predictive coding during naturalistic sentence comprehension.

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Psycholinguistics

Background:

• Prediction is a fundamental cognitive process across various domains.
• Debate exists on whether predictive coding arises from domain-general or domain-specific neural circuits.
• Understanding the information sources for predictions is crucial.

Purpose of the Study:

• To investigate whether predictive coding in language processing is domain-specific or domain-general.
• To determine if predictions are based on local word co-occurrence or hierarchical linguistic structure.
• To explore the neural implementation of predictive coding during naturalistic language comprehension.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was employed during a naturalistic language comprehension task.
• A continuous-time deconvolutional regression technique was utilized to analyze blood-oxygen-level-dependent (BOLD) signals.
• This method allowed for data-driven discovery of hemodynamic response functions from continuous BOLD fluctuations.

Main Results:

• Evidence of predictive coding effects was found within the language network, not the multiple-demand network.
• Predictions were sensitive to both local word co-occurrence patterns and hierarchical linguistic structure.
• Surface-level and structural prediction effects were separable within the language network, explaining over 37% of variance.

Conclusions:

• Language processing involves domain-specific predictive coding mechanisms.
• These mechanisms utilize both local and structural linguistic information for prediction.
• Predictions are generated by specialized language processing circuits, not high-level executive control.