Sensory sharpening and semantic prediction errors unify competing models of predictive processing in communication

The human brain makes abundant predictions in speech comprehension that, in real-world conversations, depend on conversational partners. Yet, models diverge on how such predictions are integrated with incoming speech. Predictive coding proposes that the brain emphasises unexpected information via prediction errors, whereas Bayesian models emphasise expected information through sharpening. We reconcile these views through direct neural evidence from electroencephalography showing that both mechanisms operate at different hierarchical levels during speech perception. Across multiple experiments, participants heard identical ambiguous speech in different speaker contexts. Using speech decoding, we show that listeners learn speaker-specific semantic priors, which sharpen sensory representations by pulling them toward expected acoustic signals. In contrast, encoding models leveraging large language models reveal that prediction errors emerge at higher linguistic levels. These findings support a unified model of predictive processing, wherein sharpening and prediction errors coexist at distinct hierarchical levels to facilitate both robust perception and adaptive world models.