Schedule of Events | Symposia

Distinct Neural Representations of Phonological and Semantic Predictions and Prediction Errors in Speech Comprehension

Poster Session E - Monday, March 31, 2025, 2:30 – 4:30 pm EDT, Back Bay Ballroom/Republic Ballroom

Yi Du¹ (duyi@psych.ac.cn), Baihan Lyu², Xiuyi Wang³; ¹Institute of Psychology, Chinese Academy of Sciences, ²Department of Psychology, University of Chinese Academy of Sciences

Introduction: The integration of external linguistic inputs with internal predictions from long-term memory is critical for language comprehension. While neural correlates of phonological and semantic prediction have been studied in natural language processing, how top-down prediction and bottom-up perception are co-represented at these levels remains poorly understood. Methods: This fMRI study investigates the neural representation of phonological and semantic information under congruent, absent, and incongruent conditions. Twenty-nine native Chinese speakers were exposed to idiomatic phrases with manipulated final characters, creating Expected, Missing, and Unexpected conditions. We used multiple regressive representation similarity analysis and partitioned the variances uniquely explained by perceived and predicted phonological and semantic information across conditions. Results: In the Missing condition, stronger phonological representations were observed in production-related precentral gyrus, and stronger semantic representations in default mode network regions (e.g., angular gyrus), suggesting a prediction-by-production mechanism and semantic prediction from long-term memory. In the Unexpected condition, distinct frontal and temporal areas encoded phonological and semantic information for perceived incorrect words versus predicted correct words, indicating separate hierarchical representations for bottom-up and top-down processing when predictions are violated. Additionally, the Unexpected condition elicited increased activations in auditory, attention and control systems, with bilateral superior temporal gyrus encoding the extent of subjective and calculated prediction errors (i.e., implausibility and surprisal). Conclusion: Our findings delineate distinct neural pathways for sound-to-meaning mapping and those representing phonological and semantic predictions and prediction errors. Further analysis will explore regions involved in computing and resolving prediction errors, elucidating the brain's adaptive mechanisms in language comprehension.

Topic Area: LANGUAGE: Semantic