Neural signatures of word learning in toddlers

Learning language is an incredible feat that takes years to master. One of the first steps is to learn the meaning of words. After successful learning, infants can demonstrate their understanding through their actions, including physical reactions and responses, speech, and looking behaviour. These actions have been used to indicate successful word learning in study designs and have helped answer the mysteries of early word-learning processes. However, despite the undeniable progress that these behavioural measures have made in our understanding of early word learning, they can’t answer all our questions. Neuroimaging techniques have transformed our understanding of mental and cognitive processes by allowing us to study the underlying neural activity during a task. Advances in neuroimaging technology and statistical techniques have given rise to a newer type of analysis method called multivariate pattern analysis (MVPA). Rather than studying gross neural activity averaged across many sessions and individuals, as in univariate measures, MVPA uses machine learning algorithms to decode an individual’s pattern of brain activity over time and space into the conceptual categories it’s been given. In the current study, I use a MVPA approach to decode toddlers’ electroencephalographic (EEG) responses to word and nonword auditory stimuli. In two studies, I ask (i) if machine learning techniques can successfully decode word and nonword items in toddlers, (ii) if machine-learning-derived indices capture individual differences in language competence and learning, and (iii) if these decoding techniques can be used to track changes in novel word representations as they are learned. My results demonstrate successful decoding of auditory stimuli in the toddler population, but mixed findings related to the word-learning indices and behaviours. Ultimately, this thesis provides a proof-of-concept that this approach to investigating toddler word-learning processes and outcomes is feasible, but more work is needed to determine the validity and reliability of the individual-level metrics derived from the MVPA decoding approach.