Instance-Based Generation for Interactive Restricted Domain Question Answering Systems

Instance-Based Generation for Interactive Restricted Domain Question Answering Systems Matthias Denecke and Hajime Tsukada NTT Communication Science Laboratories, 2-4 Hikaridai, Seika-Cho, Soraku-gun, Kyoto {denecke, tsukada}@cslab.kecl.ntt.co.jp Abstract. One important component of interactive systems is the gen- eration component. While template-based generation is appropriate in many cases (for example, task oriented spoken dialogue systems), inter- active question answering systems require a more sophisticated approach. In this paper, we propose and compare two example-based methods for generation of information seeking questions. 1 Introduction Question answering is the task of providing natural language answers to natural language questions using an information retrieval engine. Due to the unrestricted nature of the problem, shallow and statistical methods are paramount. Spoken dialogue systems address the problem of accessing information from a structured database (such as time table information) or controlling appliances by voice. Due to the fact that the scope of the application defined by the back-end, the domain of the system is well-defined. Therefore, in the presence of vague, ill-defined or misrecognized input from the user, dialogue management, relying on the domain restrictions as given by the application, can interactively request more information from the user until the users’ intent has been determined. In this paper, we are interested in generation of information seeking questions in interactive question-answering systems. 1.1 Our System We implemented a system that combines features of question answering systems with those of spoken dialogue systems. We integrated the following two features in an interactive restricted domain question answering system: (1) As in question answering systems, the system draws its knowledge from a database of unstruc- tured text. (2) As in spoken dialogue systems, the system can interactively query for more information in the case of vague or ill-defined user queries. 1.2 Problem Addressed in This Paper Restricted domain question answering systems can be deployed in interactive prob- lem solving solutions, for example, software trouble shooting. In these scenarios, interactivity becomes a necessity. This is because it is highly unlikely that all facts relevant to retrieving the appropriate response are stated in the query. For exam- ple, in the software trouble shooting task described in [5], a frequent system gen- erated information seeking question is for the version of the software. Therefore, R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 486–497, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 Instance-Based Generation 487 there is a need to inquire additional problem relevant information from the user, depending on the interaction history and the problem to be solved. In this paper, we specifically address the problem of how to generate in- formation seeking questions in the case of ambiguous, vague or ill-defined user questions. We assume that the decision of whether an information seeking ques- tion is needed is made outside of the module described here. More formally, the problem we address can be described as follows: Given 1. A representation of the previous interaction history, consisting of user and system utterances, and retrieval results from the IR subsystem, 2. A decision for a …

Content truncated to 500 words