We build AI systems using structured knowledge, language, and reasoning - with a focus on scientific knowledge, dialogue, structured data, and trustworthy evaluation.
We study how to build, align, and reason over knowledge graphs drawn from text, tables, and scientific data. Recent work explores scientific and temporal knowledge graphs, knowledge graph construction, and AI methods for analyzing the scientific record.
We investigate how language models reason in conversation, use common ground, compose reasoning steps, and handle long-term interaction. Our work spans commonsense-grounded dialogue, persona generation, theory-of-mind style evaluation, and multi-turn reasoning.
A major part of our work focuses on extracting meaning from tables and semi-structured data. We build methods for table understanding, semantic modeling, fact verification, and mapping diverse data sources into knowledge graph-friendly representations.
We study how to evaluate and improve AI systems in ways that are robust, interpretable, and socially responsible. This includes benchmarking reasoning, understanding representational harms, analyzing toxicity and safety, and measuring scientific and social impacts at scale.