Deducing answers to english questions from structured data.

ABSTRACTWe describe ongoing research using natural English text queries as an intelligent interface for inferring answers from structured data in a specific domain. Users can express queries whose answers need to be deduced from data in different databases, without knowing the structures of those databases nor even the existence of the sources used. Users can pose queries incrementally, elaborating on an initial query, and ask follow-up questions based on answers to earlier queries. Inference in an axiomatic theory of the subject domain links the natural form in which the question is posed to the way relevant data is represented in a database, and composes information obtained from several databases into an answer to a complex question. We describe the status of a prototype system, called Quadri, for answering questions about HIV treatment, using the Stanford HIV Drug Resistance Database [8] and European resources. We discuss some of the problems that need to be solved to make this approach work, and some of our solutions.