Summary
Learning language well depends on how caregivers give feedback to children, but it's hard to figure out exactly which types help the most by just watching real kids. The authors used child-like AI language models to test four kinds of feedback: feedback about communication, matching sentence structures, meaning connection, and emotional support. They found that feedback focusing on matching sentence structures helped the AI models improve their grammar the most, while communication feedback helped a bit, and emotional or meaning-based feedback did not improve grammar but might help other parts of language learning. This suggests that different kinds of caregiver feedback support different parts of how children learn language.
caregiver feedbackgrammar learningreinforcement learninglanguage modelsCHILDESstructural alignmentsemantic contingencyaffective feedbackminimal-pair evaluationfree generation
Authors
Jing Liu, Marianne Schweitzer, Abdellah Fourtassi
Abstract
Social interaction is central to children's language learning, but the effects of different forms of caregiver feedback are difficult to isolate in naturalistic data. We use child-like language models as controlled learners to test which forms of feedback support grammatical development. Small GPT-2-style models are pretrained on child-directed language from CHILDES, then fine-tuned with reinforcement learning using reward models trained to capture four feedback types: communicative feedback, structural alignment, semantic contingency, and affective feedback. Reward fine-tuning yields limited gains on minimal-pair evaluations, but clearer effects in free generation. Structural alignment produces the strongest improvements in grammaticality, providing a novel, plausible mechanistic account of how this feedback can support grammar learning. Communicative feedback yields more moderate gains. In contrast, semantic contingency and affective feedback do not improve grammaticality, although further analyses suggest that they may support other aspects of language learning beyond grammar. These results suggest that different forms of caregiver feedback make complementary contributions to language learning.