Learning new knowledge by reading texts has long been a dream of Artificial Intelligence. With the astronomical explosion of textual content on the web, the impact of machines that can deeply understand the texts on our daily lives cannot be overestimated. They might help a doctor conduct a quick diagnosis, suggest stocks to buy to an investor, or help an intelligence analyst piece together evidence of a crime.

Past attempts at deep language understanding by machines have been hampered by the need for common sense knowledge to interpret text on the one hand, and the lack of learning algorithms that can work on rich representations of text on the other. Our project seeks to bridge this gap by making novel advances in machine learning for structured inputs and outputs, and by exploiting models of pragmatics of communication to interpret and learn new knowledge from text. Our research spans multiple areas in AI including natural language understanding, structured prediction, relational data mining, and statistical relational learning.