Predict-and-revise: A key skill for young language learners?

Note. This is a Stage 1 Registered Report with in-principle acceptance from Developmental Science. Abstract: Actively generating predictions that turn out to be incorrect benefits children's learning in many domains, and prediction errors (PE) are a fundamental component of influential, computationally-specified accounts of learning and memory. Crucially, this includes models of how children acquire linguistic knowledge. Yet, empirical evidence that the larger the PE the greater the learning of new linguistic knowledge (word meanings, sentence structures) is both limited and contradictory. This study provides the most comprehensive test of this hypothesis so far. We operationalize PE by measuring a child’s ability to first generate and then revise incorrect predictions (predict-and-revise, assessed via eye-tracking) and, for the first time, test whether this ability is predictive of later learning (i) across linguistic domains (we test the same children on both word and structure learning) and (ii) over time (we test children’s learning immediately after an exposure phase, 24 hours, and one week later). We also report relations between predict-and-revise and individual differences in linguistic knowledge, non-verbal cognitive control and episodic memory, and the relations between these skills and learning, as a preliminary step towards developing a causal model of individual differences in the speed of language development.