Early word learning through communicative inference
Speaker: Michael Frank
Institution: Massachusetts Institute of Technology
Abstract:
How do children learn their first words? While they are able to make use of distributional information about the co-occurrence of words and objects, even very young children also seem to take into account information about speakers' communicative intentions. Rather than being though of as purely statistical or purely social, I argue that children's early word learning is best modeled as a process of statistical inference about speakers' communicative intentions. Using a communicative inference framework allows our model to learn words accurately from natural corpus data, to explain a large range of developmental results, and to make novel developmental predictions. In addition, this framework offers insight into how the rich variety of non-linguistic information about speaker's intentions can be used in service of word learning. Joint work with Noah D. Goodman and Joshua B. Tenenbaum.