Prediction Errors but Not Sharpened Signals Simulate Multivoxel fMRI Patterns during Speech Perception.

Area of Science:

• Cognitive Neuroscience
• Psycholinguistics
• Computational Neuroscience

Background:

• Perception integrates sensory input with prior knowledge, but the mechanism remains unclear.
• Two models explain expectation's influence: Sharpened Signals (enhancement) and Predictive Coding (suppression of expected, processing of unexpected).
• Distinguishing these models is crucial for understanding perception.

Purpose of the Study:

• To differentiate between Sharpened Signals and Predictive Coding models of speech perception.
• To investigate how sensory detail and prior expectations interact to influence speech representation.
• To establish methods for distinguishing Prediction Error and Sharpened Signal representations.

Main Methods:

• Combined behavioral, univariate, and multivariate functional magnetic resonance imaging (fMRI).
• Utilized computational modeling to simulate Sharpened Signals and Prediction Error accounts.
• Assessed effects of sensory detail and prior expectations on speech perception accuracy and neural activity.

Main Results:

• Behavioral and univariate fMRI showed additive effects of sensory detail and expectations on word report accuracy and reduced BOLD signal.
• Multivariate fMRI revealed an interaction: increased sensory detail enhanced speech information without expectations but reduced it with informative expectations.
• Computational simulations supported Prediction Error but not Sharpened Signals for multivariate fMRI findings.

Conclusions:

• The interaction between prior expectation and sensory detail provides strong evidence for a Predictive Coding framework in speech perception.
• This study demonstrates a novel approach to differentiate neural representations of Prediction Errors and Sharpened Signals.
• Findings advance our understanding of how the brain combines sensory evidence with prior knowledge across perceptual domains.