Seminars & Colloquia
Kristy Boyer
Computer Science, NCSU
"Learning Statistical Models for Natural Language Tutorial Dialogue Systems "
Wednesday December 08, 2010 09:30 AM
Location: 3211, EB2 NCSU Centennial Campus
(Visitor parking instructions)
Creating intelligent systems that engage users in rich natural language dialogue is a central focus for artificial intelligence and computational linguistics research. While great strides have been made toward human-like dialogue systems, today's systems face challenges interpreting complex, noisy user input and have limited flexibility in selecting their own responses. A promising approach to addressing these limitations involves applying statistical machine learning techniques to corpora of human dialogues and utilizing the learned models to automatically author new dialogue system behavior. This talk presents a novel dialogue modeling framework based on hidden Markov models. The framework is applied within a tutorial dialogue context in which a tutor and student work together to solve a computer programming problem. Results suggest that hidden Markov models are a principled choice for dialogue modeling, and moreover, that modeling the hidden dialogue state can significantly improve a learned dialogue policy, as evidenced by increased prediction accuracy for tutor dialogue moves. The findings hold implications for engineering the next generation of natural language dialogue systems.
Kristy Elizabeth Boyer is a Postdoctoral Researcher in the Department of Computer Science at North Carolina State University. Her research focuses on natural language dialogue systems with an emphasis on the cognitive, affective, and social mechanisms by which humans learn through dialogue. She holds an M.S. in Applied Statistics from the Georgia Institute of Technology. She has been recognized with a National Science Foundation Graduate Research Fellowship and is the recipient of the Best Student Paper award at the International Conference on Artificial Intelligence in Education (AIED) 2007.
Host: James Lester, Computer Science, NCSU