What Can Be Learned from Previously Answered Questions? A Corpus-Based Approach to Question Answering

Abstract

We present corpus-based approach to question answering, which enables our system to classify a question category, generate a query and verify answer candidates. The system uses the Internet to find information required to provide an answer. The role of the corpus-based modules is to support the system with knowledge that provides the means to effectively use previously answered questions as the system experience base. We present the Query Pattern Generation method, which demonstrates that the system can automatically acquire knowledge on how to optimally generate a query for a given question category and question syntax. The corpus-based answer candidate verification is an effective tool to exclude answer candidates that do not belong to a question category, and is able to provide a proper length for the answer.