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

Abstract

Abstract A longstanding challenge in developmental science is to understand how children learn language from naturalistic everyday input. To study this process, we leveraged the First 1,000 Days (1kD) dataset, which provides longitudinal, ultra-dense daily audiovisual recordings for individual children in their home environments. This unusually detailed, child-specific record of early experience enabled us to pair each child’s rich language input with a cognitively grounded learning agent, linking naturalistic experience (“nurture”) to internal learning mechanisms (“nature”). Trained incrementally on each child’s input without prior linguistic knowledge, the learning agent discovered speech units corresponding to the English phoneme inventory and acquired thousands of words, closely mirroring individual developmental trajectories. Learning generalized across children while preserving individual differences in rate and timing. Interestingly, learning relied not only on linguistic input but also on the rehearsal of past experiences at the end of each training day. These findings demonstrate that everyday environments provide sufficient structure for language acquisition and establish a unified mechanistic framework for studying development in real-world contexts.

The authors' abstract, as published at the source. Science, 2026 · DOI ↗