Predictive and feedback signals differently shape the formation of language representations

Abstract

Adults vary greatly in how effectively they learn a new language, but the signals driving the learning processes and individual differences remain unclear. Over seven days, we tracked behavioral learning and collected fMRI data on day 1 from 102 adults as they learned an artificial language with corrective feedback. We trained matched transformer models to predict upcoming linguistic elements, learn grammaticality from corrective supervision or reinforcement-like reward signals in the feedback, and compared their internal representations with human brain activity. Representations derived from the prediction-focused model accounted for the largest share of unique neural variance at the group level, despite the human task being feedback-based. Across model-training stages, alignment between the predictive model and neural representations shifted from sensory to higher-order language and associative networks, indicating a model abstraction process. Conversely, neural patterns related to the feedback model were most useful for predicting individuals' future generalization outcomes on Day 7. Reward signals did not significantly account for learning variance of either type. These findings support a multi-signal model of adult language learning, in which prediction shapes a common neural learning architecture across learners, whereas feedback-related mechanisms better explain individual differences over time.