A Language Independent Method for Question Classification

A Language Independent Method for Question Classification Thamar Solorio1, Manuel P´erez-Couti˜no1, Manuel Montes-y-G´omez1,2 Luis Villase˜nor-Pineda1 and Aurelio L´opez-L´opez1 1Language Technologies Group, Computer Science Department National Institute of Astrophysics, Optics and Electronics 72840 Tonantzintla, Puebla, Mexico 2Departamento de Sistemas Inform´aticos y Computaci´on Universidad Polit´ecnica de Valencia Espa˜na {thamy,mapco,mmontesg,villasen,allopez}@inaoep.mx Abstract Previous works on question classification are based on complex natural language process- ing techniques: named entity extractors, parsers, chunkers, etc. While these ap- proaches have proven to be effective they have the disadvantage of being targeted to a particular language. We present here a sim- ple approach that exploits lexical features and the Internet to train a classifier, namely a Support Vector Machine. The main fea- ture of this method is that it can be applied to different languages without requiring ma- jor modifications. Experimental results of this method on English, Italian and Span- ish show that this approach can be a prac- tical tool for question answering systems, reaching a classification accuracy as high as 88.92%. 1 Introduction Open-domain Question Answering (QA) sys- tems are concerned with the problem of trying to answer questions from users posed in nat- ural language. What makes these systems a very complex and interesting research area is that the answers they retrieve must be concise, as opposed to traditional search engines that in response to a user query retrieve a list of docu- ments believed to contain the answer. More- over, current evaluation environments of QA systems, such as TREC QA track (Voorhees, 2001) and CLEF (Peters et al., 2003), restrict the size of the answers to a maximum of 50 bytes. Given the complexity involved in this problem, traditional approaches to QA take a divide-and-conquer strategy, where the prob- lem is divided into several less complex subtasks that combined lead to the resolution of the ques- tions. An important subtask of a QA system is question analysis, since it can provide useful clues for identifying potential answers in a large collection of texts. For instance, Question Clas- sification is concerned with assigning semantic classes to questions. This semantic classifica- tion can be used to reduce the search space of possible answers, i.e. if we can determine that the question Who is the Italian Prime Minis- ter? belongs to the semantic category PER- SON, then we only need to look for instances of type PERSON as possible answers. Clearly, the advantage of such classification relies on hav- ing the ability of extracting from the documents such instances. In other words, a good question classification module may be useless if we lack an accurate named entity extractor for the doc- ument collection. Results of the error analysis of an open- domain QA system showed that 36.4% of the errors were generated by the question classifi- cation module (Moldovan et al., 2003). Thus it is not surprising that an increasing interest has arisen aimed at developing accurate ques- tion classifiers (Zhang and Lee, 2003; Li and Roth, 2002; Suzuki et al., 2003). However, most of these approaches …

Content truncated to 500 words