Sensory sharpening and semantic prediction errors unify competing models of predictive processing in human speech...

Area of Science:

• Neuroscience
• Cognitive Science
• Linguistics

Background:

• Human speech comprehension relies heavily on predictive processing, influenced by conversational context.
• Existing models of predictive processing offer conflicting explanations for integrating predictions with incoming speech, focusing on either information sharpening or prediction error.
• Understanding how the brain balances these mechanisms is crucial for a comprehensive model of speech perception.

Purpose of the Study:

• To reconcile divergent models of predictive processing in speech comprehension.
• To investigate the neural mechanisms underlying the integration of predictions and incoming speech.
• To determine if sharpening and prediction error operate at different hierarchical levels in the brain.

Main Methods:

• Electroencephalography (EEG) was used to record neural activity during speech perception.
• Participants listened to identical ambiguous speech stimuli presented in varied speaker contexts.
• Speech decoding and encoding models utilizing pretrained transformers were employed to analyze neural data.

Main Results:

• Listeners demonstrated the ability to learn speaker-specific semantic priors, leading to sharpened sensory representations.
• Prediction errors were identified at higher linguistic levels of processing.
• Neural evidence indicated that both sharpening and prediction error mechanisms function concurrently but at distinct hierarchical levels.

Conclusions:

• The findings support a unified model of predictive processing in speech comprehension.
• Sharpening and prediction errors coexist at different hierarchical levels, contributing to robust perception.
• This dual mechanism facilitates both efficient sensory processing and the development of adaptive world models.