Using stochastic language models (SLM) to map lexical, syntactic, and phonological information processing in the...

Area of Science:

• Neuroscience
• Computational Linguistics
• Cognitive Science

Background:

• Language comprehension requires simultaneous processing of phonological, syntactic, and lexical information.
• Computational language models provide probabilistic measures like perplexity to analyze language sequences.
• Perplexity has been used to predict behavioral and neural correlates of sentence processing.

Purpose of the Study:

• To investigate the neural correlates of phoneme, part-of-speech, and word processing.
• To determine if distinct brain areas track different streams of linguistic information.
• To utilize stochastic measures from computational models in an fMRI experiment.

Main Methods:

• Computed perplexity from word sequences, parts of speech, and phonemic transcriptions.
• Used functional magnetic resonance imaging (fMRI) to measure brain activity.
• Regressed brain activity time-locked to each word on three model-derived measures.

Main Results:

• Observed distinct neural correlates for processing phonological, syntactic, and lexical information.
• Demonstrated that the brain tracks the statistical structure of these linguistic streams in separate areas.
• Confirmed the utility of computational language models in identifying neural processing streams.

Conclusions:

• The brain segregates the processing of phonological, syntactic, and lexical information into distinct neural populations.
• Computational linguistic measures are effective tools for uncovering the neural basis of language comprehension.
• This research provides insights into the distributed neural architecture supporting language processing.