Open Question Answering

Thesis (Ph.D.)--University of Washington, 2014For the past fifteen years, search engines like Google have been the dominant way of finding information online. However, search engines break down when presented with complex information needs expressed as natural language questions. Further, as more people access the web from mobile devices with limited input/output capabilities, the need for software that can interpret and answer questions becomes more pressing. This dissertation studies the design of Open Question Answering (Open QA) systems that answer questions by reasoning over large, open-domain knowledge bases. Open QA systems are faced with two challenges. The first challenge is knowledge acquisition: How does the system acquire and represent the knowledge needed to answer questions? I describe a simple and scalable information extraction technique that automatically constructs an open-domain knowledge base from web text. The second challenge that Open QA systems face is question interpretation: How does the system robustly map questions to queries over its knowledge? I describe algorithms that learn to interpret questions by leveraging massive amounts of data from community QA sites like WikiAnswers. This dissertation shows that combining information extraction with community-QA data can enable Open QA at a much larger scale than what was previously possible