Neural Circuit Evidence for Hierarchical Processing of Prediction Errors: Distinct Cerebello–Frontal Connectivity for Phonological and Semantic Errors

Predictive coding theories posit that the brain processes prediction errors (PEs) across multiple levels, yet explicit neural circuit evidence for these mechanisms remains elusive. This study investigated effective connectivity within the right cerebellum (RC)–left inferior frontal gyrus (LIFG)–left middle frontal gyrus (LMFG) network. We leveraged Mandarin Chinese, specifically tone changes (tone sandhi) and classifier–noun agreement, to generate hierarchical phonological and semantic PEs during language processing. Combining fMRI with dynamic causal modeling, we systematically examined how these hierarchical PEs modulated effective connectivity within this network. Our results revealed distinct connectivity patterns driven by specific error types. Phonological PEs uniquely enhanced the connection between the RC and LIFG. In contrast, semantic PEs strengthened the connection between the LIFG and LMFG and the connection from the LMFG to RC. Simultaneously, semantic PEs attenuated the links from LIFG to RC and from RC to LMFG. These findings provide neural circuit evidence for a hierarchical PE processing system, where lower level phonological errors are processed through a more direct cerebello–LIFG pathway, while higher level semantic errors engage a more complex network involving the middle frontal gyrus. This research advances our understanding of the dynamic neural architecture underlying predictive language comprehension.