Abstract linguistic structure correlates with temporal activity during naturalistic comprehension.

Area of Science:

• Neuroscience
• Psycholinguistics
• Computational Linguistics

Background:

• Sentence comprehension involves a network of brain regions, but the specific information flow remains unclear.
• The debate exists whether syntactic processing relies on hierarchical grammars or linear (Markov chain) models.

Purpose of the Study:

• To investigate the nature of syntactic information processing in the brain.
• To determine if brain activity reflects hierarchical grammar computations or linear sequence prediction.
• To compare predictions from parsing models with functional magnetic resonance imaging (fMRI) data.

Main Methods:

• Quantified cognitive states implied by hierarchical and linear (Markov chain) parsing models.
• Compared processing-complexity predictions from these models against fMRI timecourses.
• Analyzed brain activity in regions associated with sentence comprehension.

Main Results:

• Hierarchical grammar models independently predicted fMRI timecourses in the left anterior and posterior temporal lobes.
• Linear (Markov chain) models predicted activity in these temporal regions and broadly across the language network, including the inferior frontal gyrus.
• Found distinct roles for brain regions in syntactic processing.

Conclusions:

• Linear processing effects are widespread across the language network.
• Specific regions in the left temporal lobe are involved in processing abstract, hierarchical syntactic representations.
• Suggests a dual system for syntactic information processing in the brain.