Model-driven knowledge-based development of expected answer type taxonomies for restricted domain question answering

Abstract

A Question Answering (QA) system must provide concise answers from large collections of documents to questions stated by the user in natural language. Importantly, a question should be correctly classified by means of a predefined taxonomy in order to determine which is the Expected Answer Type (EAT), thus reducing the searching space over documents, while a right answer is obtained. Designing a proper EAT taxonomy is even more crucial in restricted domain QA, since domain experts use specific terminology, thus asking more precise questions and expecting more precise answers. This paper presents a novel approach in order to ameliorate the task of designing restricted-domain EAT taxonomies by using heterogeneous knowledge resources and collection of documents. To show the applicability of our approach, a set of experiments has been carried out by defining a new EAT taxonomy for being able to answer questions about the agricultural domain. © 2010 Springer-Verlag Berlin Heidelberg.