Language Use and Learning in Collaborative Situated Interactions
Abstract: Systems that use language in situated collaborative interactions with human users must reason about language as it is grounded in context. This includes grounding to visual perception and action, but also to the dynamics that arise in multi-turn interactions with human users, wherein users adapt their language and behavior to most effectively collaborate with an agent. While this interactive setting poses a significant challenge, it also opens up new learning opportunities, where a system can continually learn from its interactions with users as they mutually adapt to one another. In this talk, I will discuss a collaborative situated environment that supports studying human-agent language-based interactions, and approaches to continually improve language using agents through these interactions by taking advantage of feedback that is implicitly and explicitly available from these interactions. Finally, I will talk about work in progress on characterizing the wide diversity of phenomena that arise in language-based collaborative interactions between human language users, through the lens of uncertainty.
Bio: Alane Suhr recently joined EECS and BAIR at UC Berkeley as an Assistant Professor. Alane's work focuses on building language-using systems that communicate with and learn from human users in collaborative, situated interactions. Prior to joining Berkeley, Alane completed a PhD in Computer Science at Cornell University / Cornell Tech and spent a year afterwards as a Young Investigator at the Allen Institute for AI.
Accommodation statement: If you require an accommodation such as live captioning or interpretation to participate in this event, please contact Dan Parker (parker.1758@osu.edu). In general, requests made two weeks before the event will allow us to provide seamless access, but the university will make every effort to meet requests made after this date.
