Presenter: Daniel Yurovsky
Date & Hour: 8 July (Fri) 15:30--
Place: JAIST Collaboration-Room 2
Title: Learning Words in the Lab and in the World: The Role of Cross-Situational Statistics

Abstract:
Learning a language involves solving a number of difficult problems -
understanding when others are trying to communicate, segmenting individual
words from continuous speech, and mapping these words onto their referents
(among others). Modern computational models universally assume that these
problems can be solved, over time, by tracking statistical patterns in the
world. In this talk, I will address three outstanding issues facing this
approach. First, I will present empirical evidence that validates a strong,
untested assumption of these models: that words are learned by the
accumulation of partial knowledge. Second, I will show that humans make use
of partial knowledge of some words to quickly learn new words, and argue
that the competition is the key mechanism responsible for this
bootstrapping. Finally, I will present evidence that statistical word
learning is not only a lab phenomenon, but can scale up to the kind of input
from which children must learn language.