Classification of Multiple-Sentence Questions

Classification of Multiple-Sentence Questions Akihiro Tamura, Hiroya Takamura, and Manabu Okumura Precision and Intelligence Laboratory, Tokyo Institute of Technology, Japan aki@lr.pi.titech.ac.jp {takamura, oku}@pi.titech.ac.jp Abstract. Conventional QA systems cannot answer to the questions composed of two or more sentences. Therefore, we aim to construct a QA system that can answer such multiple-sentence questions. As the first stage, we propose a method for classifying multiple-sentence questions into question types. Specifically, we first extract the core sentence from a given question text. We use the core sentence and its question focus in question classification. The result of experiments shows that the proposed method improves F-measure by 8.8% and accuracy by 4.4%. 1 Introduction Question-Answering (QA) systems are useful in that QA systems return the answer itself, while most information retrieval systems return documents that may contain the answer. QA systems have been evaluated at TREC QA-Track1 in U.S. and QAC (Question & Answering Challenge)2 in Japan. In these workshops, the inputs to systems are only single-sentence questions, which are defined as the ques- tions composed of one sentence. On the other hand, on the web there are a lot of multiple-sentence questions (e.g., answer bank3, AskAnOwner4), which are defined as the questions composed of two or more sentences: For example, “My computer reboots as soon as it gets started. OS is Windows XP. Is there any homepage that tells why it happens?”. For conventional QA systems, these questions are not expected and existing techniques are not applicable or work poorly to these questions. Therefore, constructing QA systems that can handle multiple-sentence questions is desirable. An usual QA system is composed of three components: question process- ing, document retrieval, and answer extraction. In question processing, a given question is analyzed, and its question type is determined. This process is called “question classification”. Depending on the question type, the process in the an- swer extraction component usually changes. Consequently, the accuracy and the efficiency of answer extraction depend on the accuracy of question classification. 1 http://trec.nist.gov/tracks.htm 2 http://www.nlp.is.ritsumei.ac.jp/qac/ 3 http://www.theanswerbank.co.uk/ 4 http://www.askanowner.com/ R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 426–437, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 Classification of Multiple-Sentence Questions 427 Therefore, as a first step towards developing a QA system that can han- dle multiple-sentence questions, we propose a method for classifying multiple- sentence questions. Specifically, in this work, we treat only questions which re- quire one answer. For example, if the question “The icon to return to desktop has been deleted. Please tell me how to recover it.” is given, we would like “WAY” to be selected as the question type. We thus introduce core sentence extraction component, which extracts the most important sentence for question classifica- tion. This is because there are unnecessary sentences for question classification in a multiple-sentence question, and we hope noisy features should be eliminated before question classification with the component. If a multiple-sentence question is given, we first extract the most important sentence for question classification and then classify the question using the …

Content truncated to 500 words