AI and the World's Hard Problems · Linguistics · arXiv · 2026-10-22
Predictive and feedback signals differently shape the formation of language representations
Adults vary greatly in how effectively they learn a new language, but the signals driving this learning remain unclear. Over seven days, 102 adults learned an artificial language with corrective feedback while fMRI data from day 1 were compared with matched transformer models trained to predict upcoming linguistic elements, learn grammaticality from corrective supervision, or use reinforcement-like reward signals. The prediction-focused model accounted for the largest share of unique neural variance at the group level, its alignment shifting from sensory to higher-order language and associative networks across training, while feedback-related neural patterns best predicted individuals' generalization outcomes on Day 7 and reward signals did not significantly account for learning variance of either type.