Question Classification using Head Words and their Hypernyms

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 927–936, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Question Classification using Head Words and their Hypernyms Zhiheng Huang EECS Department University of California at Berkeley CA 94720-1776, USA zhiheng@cs.berkeley.edu Marcus Thint Intelligent Systems Research Center British Telecom Group Chief Technology Office marcus.2.thint@bt.com Zengchang Qin EECS Department University of California at Berkeley CA 94720-1776, USA zqin@cs.berkeley.edu Abstract Question classification plays an important role in question answering. Features are the key to obtain an accurate question classifier. In con- trast to Li and Roth (2002)’s approach which makes use of very rich feature space, we pro- pose a compact yet effective feature set. In particular, we propose head word feature and present two approaches to augment semantic features of such head words using WordNet. In addition, Lesk’s word sense disambigua- tion (WSD) algorithm is adapted and the depth of hypernym feature is optimized. With fur- ther augment of other standard features such as unigrams, our linear SVM and Maximum Entropy (ME) models reach the accuracy of 89.2% and 89.0% respectively over a standard benchmark dataset, which outperform the best previously reported accuracy of 86.2%. 1 Introduction An important step in question answering (QA) and other dialog systems is to classify the question to the anticipated type of the answer. For example, the question of Who discovered x-rays should be classi- fied into the type of human (individual). This infor- mation would narrow down the search space to iden- tify the correct answer string. In addition, this infor- mation can suggest different strategies to search and verify a candidate answer. For instance, the classifi- cation of question What is autism to a definition type question would trigger the search strategy specific for definition type (e.g., using predefined templates like: Autism is ... or Autism is defined as...). In fact, the combination of QA and the named entity recog- nition is a key approach in modern question answer- ing systems (Voorhees and Dang, 2005). The question classification is by no means trivial: Simply using question wh-words can not achieve satisfactory results. The difficulty lies in classify- ing the what and which type questions. Considering the example What is the capital of Yugoslavia, it is of location (city) type, while What is the pH scale is of definition type. Considering also examples (Li and Roth, 2006) What tourist attractions are there in Reims, What are the names of the tourist attractions in Reims, What do most tourists visit in Reims, What attracts tourists to Reims, and What is worth seeing in Reims, all these reformulations are of the same answer type of location. Different wording and syn- tactic structures make it difficult for classification. Many QA systems used manually constructed sets of rules to map a question to a type, which is not effi- cient in maintain and upgrading. With the increasing popularity of statistical approaches, machine learn- ing plays a more and more important role in this task. A salient …

Content truncated to 500 words