Computational Modeling of Word Learning Biases by using Known Words Meanings

K. Kurosaki and T. Omori (Japan)


connectionist model, cognitive process, word learning bias, novel noun generalization task, vocabulary spurt


In the acquisition of their early nouns, it is well-known that young children have a tendency to understand the mean ing of novel nouns based on the similarity of shape. This phenomenon is called ”shape bias.” Though this bias is re markable in solid objects, it is reported that children over generalize and misapply the bias to non-solid objects. For this phenomenon, learning models using distributed repre sentation are proposed. But the computational mechanism behind such children’s behavior has not been clarified. In this paper we aim to clarify the more detailed computa tional mechanisms of these biases. Therefore, we explicitly define word meanings by a ”word category neuron model” and propose a ”nearest neighbor hypothesis” that represents a plausible mechanism for children’s cognitive processes. Then, from a computer simulation based on the novel noun generalization task of developmental psychology, we show that the proposed hypotheses can better explain the emer gence of word learning bias and deflection in children’s word learning.

Important Links:

Go Back