Gilmore, Greydon

Abstract Statistical learning (SL) is a powerful mechanism that supports the ability to extract regularities from environmental input. Yet, its neural underpinnings are not well understood. Previous EEG studies of SL have found that the brain tracks regularities by synchronizing its activity with the presented stimuli—a phenomenon known as neural entrainment. However, EEG lacks the spatial resolution to unveil the specific brain regions where this process takes place. In our study, 18 patients with drug-resistant epilepsy who were implanted with intracranial electrodes for presurgical investigation listened to a continuous speech stream containing embedded trisyllabic words. Neural entrainment was measured at the syllable and word frequencies, with the latter providing an online index of learning. SL was further assessed through both explicit and implicit behavioral measures. Behaviorally, we found evidence of learning at the group level in both tasks. At the neural level, our analyses revealed three temporal tuning profiles: 25% of contacts showed entrainment at the syllable frequency, 11% of contacts showed entrainment at both the word and syllable frequencies, and 4% showed entrainment only to the word frequency. Word entrainment, indicating sensitivity to word structures, was most commonly found in auditory and language-related regions, including insula, middle temporal gyrus, superior temporal gyrus, and supramarginal gyrus. In contrast, evidence for neural entrainment in the hippocampus was weak. Overall, these results support the idea that speech-based SL is largely supported by modality-specific brain regions.