Predicting children's early word learning using egocentric videos of their learning environments
Abstract
How do features of children's environmental input predict their word learning? Prior research has suggested that distributional features of the language children hear (e.g., frequency and syntactic complexity) predict variation in early word learning, but these studies have typically relied on aggregated estimates from disjunct samples to calculate input features and word learning outcomes. In this study, we use child-specific input features derived from 1118h of at-home egocentric video recordings to show that word frequency, length in phonemes, and syntactic complexity predict word knowledge within individual children (N = 29). The predictive power of these within-child features was greater than features aggregated across children, supporting theories positing a direct relationship between input distributions and children's language learning. Exploratory analyses involving object frequency, word–object cooccurrences, and context distinctiveness did not reveal any additional effects of these predictors. The use of child-specific naturalistic data provides an avenue for more precise and comprehensive investigations of input–outcome relationships for word learning, allowing researchers to adjudicate among theories of language learning in young children.