Structural caregiver feedback best supports grammar learning in child models
Which Forms of Caregiver Feedback Support Grammar Learning? A Reinforcement-Learning Study of Child-Like Language Models
Computation and Language
Summary
Children learn language through social interactions with caregivers, but it’s hard to tell exactly which types of feedback help most. The authors used small language models trained to act like children and tested different kinds of caregiver feedback. They found that feedback helping the child notice structural patterns in language most improved grammar, while other feedback types like emotional cues did not improve grammar but could help other language skills. This suggests that different feedback types each play unique roles in language learning.
What this means in practice
- •For language technology developers: Design child-like AI systems that improve grammar learning using structural alignment feedback modeled from caregiver interactions.
- •For speech therapy tool builders: Develop interactive speech systems that incorporate structural feedback to help users improve grammatical skills more effectively.
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.