Challenges in modern-day natural dialogue systems

Date of Presentation: 
Wednesday, January 11, 2017
2017 Winter
Research Focus: 

Abstract - Natural dialogue systems embody the goals of the Turing test: a computer system that can hold a conversation in a manner indistinguishable from a human. Building such a system is still a non-trivial open problem in the field of natural language processing. Today’s most successful systems are rule-based and work well within a controlled conversational domain. However, these systems lack the ability to reason within the broader, unconstrained environment that is typical of human conversations. Within the past few years, deep neural networks have emerged as a potential replacement for rule-based systems. The long short-term memory (LSTM), in particular, has been successful at replicating and improving performance over rule-based systems. This talk will introduce the LSTM model, highlight some of the challenges that remain with this approach – e.g., consistency, relevance, initiation, and elaboration – and propose ideas for addressing these challenges.
Bio - America Chambers is an assistant professor in the Math and CS Department at the University of Puget Sound where she teaches Introduction to Computer Science, Algorithms, Artificial Intelligence, and Natural Language Processing. She earned her B.A. in Mathematics and Computer Science from Swarthmore College and her Masters and Ph.D. in Computer Science from the University of California, Irvine. She taught as a visiting professor for two years at Pomona College before joining the University of Puget Sound in 2015. Her research interests include artificial intelligence, machine learning, and natural language processing. Currently, she is very interested in creating dialogue systems.

America Chambers
Speaker affiliation: 
University of Puget Sound, Mathematics and Computer Science