Colunga, E. , Schilling , S. . & Sims , C. .
University of Colorado Boulder
In typical development, word learning goes from slow and laborious to fast and seemingly effortless. Typically developing 2-year-olds are so skilled at learning noun categories that they seem to intuit the meaning of a new word after hearing it only once. This is not the case for children below the 20th percentile on productive vocabulary, so-called late talkers. Late talkers are not a homogenous group in terms of their developmental outcomes: some catch up, a few will be diagnosed with Specific Language Impairment, and for some the source of the delay may be environmental (e.g., Rescorla, 2000; 2002). Importantly, although there is continuity in vocabulary measures at the group level, the outcome for individual children cannot be accurately predicted on the basis of vocabulary production or comprehension (e.g., Desmarais, Meyer, Bairati & Rouleau, 2008). In the work presented here, we use connectionist models to understand the processes underlying word learning in children in the bottom and top 20th percentile in vocabulary measures. We look at the vocabulary composition of age-matched 18-30-month-old late- and early-talking children. Our results suggest that noun vocabularies of late takers and those of early talkers are structured differently. Critically, we also show that neural networks trained on vocabularies structured as those of individual late talkers subsequently learn new nouns in different ways than those neural networks trained on vocabularies of individual early talkers. Finally, we have confirmed predictions made by these models, showing that late- and early-talking toddlers do indeed, in the lab, learn new nouns in the ways predicted (Colunga & Sims, 2011, under rev). Our more recent work looks at the interdependencies in the learning of different sorts of words during development in both neural network simulations and toddlers (Schilling, Sims, & Colunga, in press).