Parallel processing in speech perception with local and global representations of linguistic context.

Area of Science:

• Cognitive Neuroscience
• Psycholinguistics
• Speech Processing

Background:

• Human listeners predict upcoming speech using linguistic context.
• Two models exist: a two-stage process with local integration and a unified interpretation integrating all information.
• Previous evidence is contradictory regarding which model or if both are used.

Purpose of the Study:

• To distinguish between local and unified predictive models of speech processing.
• To investigate how predictive context integrates with sensory input during continuous speech.
• To identify neural signatures of different predictive models using magnetoencephalography.

Main Methods:

• Magnetoencephalography (MEG) was used to record brain responses to continuous narrative speech.
• Analysis focused on identifying neural signatures supporting local and unified predictive models.
• Neural source localization was employed to determine the anatomical origins of predictive processing.

Main Results:

• Evidence supports the parallel use of both local and unified predictive models.
• Early neural responses reflect both sublexical phoneme sequences and sentence-level constraints.
• Different predictive models originate from distinct areas in the superior temporal lobes, with right hemisphere preference for local models.

Conclusions:

• Speech processing employs both local and unified predictive models concurrently.
• Parallel models enhance perceptual robustness and may aid in language acquisition.
• Findings reconcile previous disparate evidence on predictive speech processing.