Predicting infant vocabulary from neural connectivity and maternal speech: A machine learning approach

Identifying early predictors of language development is essential for understanding how infants acquire vocabulary during the first years of life. While previous studies have established the importance of infant-directed speech (IDS) and neural speech processing, this longitudinal study introduces a novel approach by combining EEG-based functional connectivity analysis and machine learning to assess the joint contribution of maternal and infant neural factors to language outcomes. Data were collected at birth and nine months, including maternal personality and speech characteristics, alongside infant EEG responses during speech processing. Language comprehension and production were assessed at 18 months using a standardized measure. A machine learning framework was used to model the predictive contribution of these variables, revealing that maternal speech prosody and EEG network properties were among the strongest predictors of later vocabulary. These findings highlight the value of integrating neural and environmental data using data-driven approaches and suggest targeted directions for future research on early language acquisition.