Predictive uncertainty in auditory sequence processing.

Area of Science:

• Cognitive Science
• Music Cognition
• Auditory Perception

Background:

• Auditory expectation studies often focus on retrospective event perception.
• This research investigates predictive uncertainty in the prospective state of expectation before an event.
• The Statistical Learning Hypothesis posits that schematic expectations arise from implicitly learned probabilistic sensory relationships.

Purpose of the Study:

• To model predictive uncertainty in music cognition using Shannon entropy.
• To investigate how musical context complexity and listener expertise (musicians vs. non-musicians) influence perceived uncertainty.
• To compare computational models of auditory expectation with empirical data.

Main Methods:

• Selected melodic contexts with high and low Shannon entropy from simple and complex musical repertoires.
• Collected explicit uncertainty judgments and inferred uncertainty via a probe-tone paradigm.
• Utilized an unsupervised, variable-order Markov model for probability estimates.
• Simulated listener perception using computational models of auditory expectation.

Main Results:

• Listeners reported greater uncertainty in high-entropy musical contexts compared to low-entropy contexts.
• This effect was more pronounced for inferred uncertainty and stronger in musicians.
• Computational models showed varying fits to the empirical data, aiding model comparison.

Conclusions:

• Predictive uncertainty in music is influenced by the probabilistic structure of the auditory input.
• Musical training enhances the ability to model and perceive probabilistic structures, leading to greater sensitivity to uncertainty.
• Findings support the Statistical Learning Hypothesis regarding implicit learning of auditory regularities.