The development of the relationship between distributional learning and prediction in language

Prediction-based theories propose that children learn language by generating expectations about upcoming input and updating their representations when those expectations are violated, but direct evidence for this mechanism in children is sparse. In this eye-tracking study, North American English-speaking 8- to 12-year-olds learned new distributional information about familiar verbs and their associated syntactic structures. In this verb bias learning task, children heard sentences such as "feel the frog with the feather" and learned to use "the feather" either as an instrument for "feel", or a modifier for "the frog". Children rapidly updated verb-specific structural preferences, showing stronger instrument interpretations, as well as greater anticipatory looks to instruments, for instrument-trained than modifier-trained verbs. Older children demonstrated greater learning in their final interpretations, but younger children demonstrated more growth in anticipatory looking behavior. Crucially, individual differences analyses suggest that children whose initial verb-bias interpretations diverged more from the trained structure showed greater verb bias malleability in both final interpretation and anticipatory first fixation, consistent with error-based learning accounts. These findings suggest a reciprocal relationship between distributional linguistic learning and predictive processing in children.