Event Details

Human-level natural language understanding requires implicit reasoning, enabling us to anticipate subsequent phrases or make assumptions about unstated facts. Typical event patterns and knowledge about everyday concepts play a critical role in this process. In this talk, I will describe our efforts to use neuro-symbolic approaches to induce event patterns from text narratives and to explore the inferential power of such patterns for downstream natural language understanding tasks. Our work leverages advances in information extraction (entity/event/relation detection, co-reference resolution), semantic parsing, summarization, and generalization. As a starting point, we focus on domain-specific complex event patterns because, in our view, this offers the most practical approach to real world use cases, such as domain-specific decision support.

Speaker Bio

Maria Chang is a Research Scientist at IBM Research AI, specializing in knowledge representation its application to natural language understanding tasks (such as question answering and natural language inference), spatial reasoning, analogical reasoning, and reasoning about events. She has also worked at the intersection of AI, education, and cognitive science, investigating sketch- and dialogue-based intelligent tutoring systems as well as neural characteristics of developmental disabilities in children. She has a PhD in computer science from Northwestern University and a BA in Cognitive Science from UC Berkeley.