External speakers: Helen Blank. How expectations shape speech perception: A predictive processing perspective

What: How expectations shape speech perception: A predictive processing perspective

Where: BCBL Auditorium and Auditorium zoom room (If you would like to attend to this meeting reserve at info@bcbl.eu)

Who: Helen Blank. PhD, Professor, Research Center One Health Ruhr of the University Alliance Ruhr, Faculty of Psychology, Ruhr-University Bochum, Germany

When:  Thursday,  Mar  12th at 12:00 PM noon.

Speech perception is not a passive decoding of sound but an active inferential process shaped by prior expectations at multiple levels. In this talk, I present a series of studies investigating how different kinds of priors, from semantic beliefs about speakers to recent sensory history, systematically influence what we hear. The fundamental question stimulating our work is: Do expectations sharpen expected speech features, or suppress them to highlight prediction errors?
First, I extend these ideas to naturalistic conversation. I argue that speech comprehension in dialogue relies on speaker-specific semantic priors, expectations about what a particular person is likely to talk about. I discuss how such high-level priors may interact with lower-level acoustic processing, and how sharpening and prediction-error signals may be coordinated across time and levels of the processing hierarchy.
Next, I revisit serial dependence in speech perception. Although attractive biases toward recent input are often interpreted as stabilizing perception, we show that this view is incomplete. When previous stimuli and responses are modeled jointly, robust repulsive effects emerge at sensory levels, and attractive biases fail to scale with environmental correlations. To reconcile these findings, I introduce the Hierarchical Precision Hypothesis, proposing that prior precision decreases across hierarchical levels: highly precise sensory priors generate repulsion, whereas broader, higher-level priors give rise to attraction. Importantly, prior precision is dynamically shaped by environmental statistics, ensuring adaptive stability.
Together, these studies propose a unified, hierarchical account of speech perception in which expectations shape perception through precision-weighted prediction errors, enabling both stability and flexibility in real-world communication.