Clarification Dialogs: a tool for Question Answering disambiguation,

Abstract

We present a Question Answering system that provides a natural lan- guage interface to Databases. Our system uses clarification dialogues in order to disambiguate the possible meaning of user questions in natural language. The system will only query the user when commitment to some meaning is relevant for the final answer. When there are multiple different answers to a user query, the system initiates a clarification dialogue trying to ascertain the user’s intentions. The clarification question is always about a relevant property of a (user) discourse entity. The answer to the clarification question (its semantic representation) can be incorporated into the initial user question enabling its reinterpretation under the new constraints. This is developed in an integrated logic programming frame- work, based on constraint logic programming using the GnuProlog(-cx) language [3, 8] and the ISCO framework [2]. The use of this LP framework allows the in- tegration of Prolog-like inference mechanisms with classes and inheritance, con- straint solving algorithms and provides the connection with relational databases, such as PostgreSQL.