Linguistically Motivated Question Classification

In this paper we describe a question interpretation module designed as a part of a Question Answering Dialogue System (QADS) which is used for an interactive quiz application. Question interpretation is achieved in applying a sequence of classification, information extraction, query formalization and query expansion tasks. The process of a question classification is performed based on a domain-specific taxonomy of semantic roles and relations. Our taxonomy was designed in accordance with the real spoken dialogue data. The SVM-based classifier is trained to predict the Expected Answer Type (EAT) with the precision of 82%. In order to retrieve a correct answer, focus word(-s) are extracted to augment the EAT identified by the system. Our hybrid algorithm for the extraction of focus words demonstrates the accuracy of 94.6%. EAT together with focus words are formalized in a query, which is further expanded with the synonyms from WordNet. The expanded query facilitates the search and retrieval of the information that is necessary to generate the system’s responses.