Continual Language Learning through Collaborative Interaction with Users

Location:

Zoom (Zoom link; Password if needed: MichiganAI)

ABSTRACT:
Interaction with human users provides ample opportunities for language learning. I will describe how collaborative scenarios provide a strong foundation for learning from interaction with users, by combining strong incentives for teaching and flexibility for adaptation. I will show how this applies to instruction generation learning, where a system learns by comparing human execution of generated instructions and their original system intents to infer the system’s communicative success. We use this signal to formulate a contextual bandit learning problem, where the system continually improves over time through interaction with users. In interaction with real users, we demonstrate dramatic improvements in the system’s ability to generate language over time.

BIO:
Yoav Artzi is an Associate Professor in the Department of Computer Science and Cornell Tech at Cornell University. His research focuses on developing learning methods for natural language understanding and generation in automated interactive systems. He received an NSF CAREER award, and his work was acknowledged by awards and honorable mentions at ACL, EMNLP, NAACL, and IROS. Yoav holds a B.Sc. from Tel Aviv University and a Ph.D. from the University of Washington.