Guest Speaker: Eunsol Choi
Dreese Labs 480
2015 Neil Ave, Columbus, Ohio 43210
Contextual Machine Reading: ultra-fine entity typing and interactive question answering
Machine reading systems that automatically answer questions from provided text must both (1) understand the implicit information about entities being discussed and (2) grasp the intent of the questions to find the best answer. In this talk, we present novel formulation for each of these challenges. First, I will cover ultra-fine-grained entity typing: a new formalism for recognizing what types of entities are being mentioned, in context-dependent ways. While previous work relied on small inventories of entity types, our new task covers thousands of fine-grained types, comes with a natural source of weak-supervision, and improves performance on existing benchmarks. In the second half of the talk, I will introduce a new QA task — Question Answering in Context (QuAC) where we simulate information seeking dialogs composed of a sequence of questions that build on each other. This formalism allows for more open-ended, contextual questions that could not be asked in existing single-turn QA settings, greatly reducing the prevalence of simple factoid questions. Together, these works expand the scope of questions that can be answered with the next generation of machine reading systems.
Bio: Eunsol Choi is a PhD candidate in the Paul G. Allen School of Computer Science and Engineering at the University of Washington, advised by Yejin Choi and Luke Zettlemoyer. Her research focuses on natural language processing, specifically applying machine learning to recover semantics from text. She develops techniques for extracting information about entities from text, and answering natural language questions automatically using large-scale databases or unconstructed text. Prior to UW, she did her undergraduate study at Cornell University. Currently she is supported by Facebook Fellowship.
Host: Wei Xu