The first 1000 days: Modeling language learning from individual children's everyday speech.

Child-specific cognitively feasible models trained on longitudinal home speech from eight children across their first 1000 days develop phoneme-like and word-form representations that track each child’s learning pace.

• Why it matters: It remains unclear how everyday input shapes children’s language learning, and existing large language models differ from natural child experience in scale and training conditions.
• What they did: The study trained one incremental model per child on ordered speech input from longitudinal home recordings, learning without prior linguistic knowledge and using dense daily sampling plus replay of recent input.
• The result: Learning depended on dense daily sampling and replay, suggesting experience- and memory-related processes are both necessary, and enabling child-specific modeling to test developmental theories.

AI summary of the abstract · may contain errors · not medical advice · read the paper.