Files and links (1)
Free to read and download, Open Access
In order to learn a new word, a young child must extricate the correct object from multiple possible items in front of them, make an initial association between the specific word-form and the particular referent, robustly link the new word and referent and integrate the new word into their lexicon. Recent research suggests processes that focus attention on the most novel objects in a complex environment, as well as the child's own developing vocabulary play critical roles in this process. This thesis aims to understand the influence of novelty and prior vocabulary knowledge on referent selection and how the interaction of novelty and knowledge can lead to word learning.
A series of empirical studies first probed the use of children's endogenous novelty bias in a referent selection task, and then explored how the use of novelty was related to retention of newly mapped word-referent pairs. A second set of studies explored children's use of vocabulary knowledge in ambiguous learning situations by varying the strength of knowledge for competing items present during novel word learning. Finally, a Hebbian Normalized Recurrent Network model was used to explore the underlying associative process of referent selection and retention in novelty- or knowledge-based word learning tasks.
Counter to prior work, results here suggest that novelty can override knowledge and in fact, be a detriment to word learning. Children demonstrate a novelty bias across multiple contexts and tasks, but the dominant use of novelty does not translate to retention and does not appear to implicate the use of the child's lexicon. As novelty diminishes and vocabulary knowledge increases, some children can overcome this bias and demonstrate retention for new word-referent pairs. Moreover, the results also suggest that when disambiguation requires the use of weak prior knowledge, more cognitive processing is necessary. The increases in processing subsequently translate to retention for new word-referent pairs. The empirical and computational results together suggest potential limitations of these findings to word learning and suggest future directions exploring variability in object and word representations during learning.