Using Computational Models To Create a Word Learning Intervention

The present investigation examines the role of semantic network structure in the mental lexicon of toddlers during language acquisition. The growth principle preferential attachment can be applied to a lexical network; implying networks grow by adding new words based on the connectivity of words already in the network. Aguilar (unpublished) used preferential attachment to create simulations of children’s lexical network growth. The following paper is an extension of Aguilar (unpublished) and applies the same preferential attachment algorithm developed for semantic-based features to real children’s vocabularies. Using real children’s vocabularies offers further insight on what types of words affect a child’s mental lexicon and overall vocabulary. In this experimental study, children were given a custom-made picture book containing sixteen words that have either a highor low-probability of being learned according to the predicted growth of each child’s existing vocabulary. Overall, children in both conditions were able to successfully learn the words in their book as measured by a receptive vocabulary pointing task. In addition, findings show a significant difference in whether or not words are acquired differently for different word-learning populations; depending on what words they already know. Findings suggest children benefit from learning words in an intervention when presented with words they otherwise wouldn't learn. The work presented makes two main contributions. The first contribution is a measure of how well children under the age of two respond to individualized storybook interventions. The second is a test of whether the Aguilar (unpublished) models capture some aspect of actual growth. MENTAL LEXICON INTERVENTION