Neural tracking of prosodic and statistical rhythms jointly supports artificial language learning

Neural tracking of prosodic and statistical patterns supports speech segmentation and word learning. Yet, acoustic and abstract patterns have proven difficult to dissociate empirically. Our study employs an artificial language learning paradigm generated with a method that orthogonalizes acoustic and statistical patterns in auditory stimuli. MEG recordings from 32 human adults who had been exposed to continuous syllable streams reveal distinct brain sources tracking periodic acoustic, prosodic, and statistical patterns. After encoding, M200 and M400 field responses to isolated trisyllabic chunks show that participants retained both prosodic and statistical patterns for learning pseudowords from the syllable streams. Behavioral data extend our findings, suggesting not only that integrating both acoustic and abstract patterns is critical to learning, but also that prior knowledge of the pseudowords is sufficient for recognition. The findings advance our understanding of the neural sources that encode and integrate acoustic and statistical patterns during and after language learning.