Computational Modeling of Word Learning: The Role of Cognitive Processes

Young children, with no prior knowledge, learn word meanings from a highly noisy and ambiguous input. Moreover, child word learning depends on other cognitive processes such as memory, attention, and categorization. Much research has focused on investigating how children acquire word meanings. A promising approach to study word learning (or any aspect of language acquisition) is computational modeling since it enables a precise implementation of psycholinguistic theories. In this thesis, I investigate the mechanisms involved in word learning through developing a computational model. Previous computational models often do not examine vocabulary development in the context of other cognitive processes. I argue that, to provide a better account of child behavior, we need to consider these processes when modeling word learning. To demonstrate this, I study three phenomena observed in child word learning.;First, I show that individual differences in word learning can be captured through modeling the variations in attentional development of learners. Understanding these individual differences is important since although most children are successful word learners, some exhibit substantial delay in word learning and may never reach the normal level of language efficacy. Second, I have studied certain phenomena (such as the spacing effect) where the difficulty of learning conditions results in better retention of word meanings. The results suggest that these phenomena can be captured through the interaction of attentional and forgetting mechanisms in the model. Finally, I have investigated how children, as they gradually learn word meanings, acquire the semantic relations among them. I propose an algorithm that uses the similarity of words in semantic categories and the context of words, to grow a semantic network. The resulting semantic network exhibits the structure and connectivity of adult semantic knowledge. The results in these three areas confirm the effectiveness of computational modeling of cognitive processes in replicating behavioral data in word learning.