harlan@paradox.psych.columbia.edu
An Interdisciplinary Survey of Word Learning Research
Time:Thursday, March 4, 12:00 - 1:00
PLEASE NOTE THE CHANGE OF TIME:
Place:CS Conference Room in MUDD
Abstract:
The learning of new words is an essential skill for any intelligent entity, from human infants to language-capable artificial agents. Psychologists have long been interested in the complex process of language acquisition, and research has made significant progress recently towards understanding the heuristics and mechanisms that people use to learn new words, both as infants and as adults.
This talk presents an overview of the current word learning
literature, with very brief summaries of current approaches in NLP and
AI, and a more extensive review of the psycholinguistic evidence. The
review will focus on several topics most relevant to researchers in
AI, such as the meta-learning of biases from experience, the role of
statistical processing in word learning, and the relationship between
word learning and category learning. I will briefly note two
psychological word-learning experiments currently in progress here at
Columbia, and will conclude with a wildly speculative discussion of
how practical AI systems could make use of the results of this
interdisciplinary common interest.
About the speaker: Harlan D. Harris is a Post-Doctoral Research Scientist in the Language and Cognition Lab at Columbia University's Department of Psychology. He recently completed his PhD in Computer Science at the University of Illinois at Urbana-Champaign, where his thesis was entitled "New Algorithms for Attribute-Efficient Linear Learning." As a graduate student, he also worked at the Language Production Lab at the Beckman Institute for Advanced Science and Technology, studying computational cognitive modeling and psycholinguistics. His interests are in language and learning, human and artificial.