Insights into Early Word Comprehension: Tracking the Neural Representations of Word Semantics in Infants.

Abstract

Infants develop foundational language skills and can understand simple words well before their first birthday. This developmental milestone has been studied primarily using looking time paradigms and Event-Related Potential (ERP) techniques, which provide evidence of word comprehension in the infant brain. While prior research validates the presence of semantic representations of words (word meaning) in infants, it does not tell us about the mental processes involved in retrieving these semantic representations or the content of the representations. To this end, we use a machine learning approach to predict the semantic representations of words using Electroencephalograms of infant brain activity. We explore semantic representations in two groups of infants (9-month-old and 12-month-old) and find significantly above chance decodability of overall word semantics, word animacy, and word phonetics. We compare decodability between the age groups and find significant differences later in the analysis window (around 700-900 ms after word onset), with higher decoding performance in 9-month-olds. Through our analysis, we also observe strong neural representations of word phonetics in the brain data for both age groups, some possibly correlated to word decoding accuracy and others not. Finally, we use the Temporal Generalization Method to show that neural representation of word semantics generalizes across the two infant age groups. Our results on word semantics, phonetics, and animacy provide insights into the evolution of neural representation of single word meaning in infants.