Infant language acquisition is commonly studied using so-called daylong audio recordings collected from their everyday environments. Manual analysis of such recordings is not feasible due to the sheer scale of the data, limiting our ability to learn from these realistic datasets. This project develops more accurate and reliable tools for processing child-centered audio recordings, thereby greatly expanding the possibilities to study child language experiences using such data. In addition to better analysis tools, the project will result in new methods to assess reliability of the tools across diverse scenarios. These technical contributions will then be used to study age-dependent linguistic properties of child-directed speech using large amounts of naturalistic audio data from longitudinal corpora on Finnish an English, including comparisons between the languages, as well as studying the connections between properties of caregiver speech and consequent infant language development.

2025

2029