Extracting Exact Answers using a Meta Question Answering System

Proceedings of the Australasian Language Technology Workshop 2005, pages 105–112, Sydney, Australia, December 2005. Extracting Exact Answers using a Meta Question Answering System Luiz Augusto Sangoi Pizzato and Diego Moll´a-Aliod Centre for Language Technology Macquarie University 2109 Sydney, Australia {pizzato, diego}@ics.mq.edu.au http://www.clt.mq.edu.au/ Abstract This work concerns a question answering tool that uses multiple Web search engines and Web question answering systems to retrieve snippets of text that may contain an exact answer for a natural language question. The method de- scribed here treats each Web information re- trieval system in a unique manner in order to ex- tract the best results they can provide. The re- sults obtained suggest that our method is com- parable with some of today’s state-of-the-art systems. 1 Introduction Text-based Question Answering (QA) focuses on finding answers for natural language ques- tions by searching collections of textual docu- ments. This area of research has become espe- cially active after the introduction of a question answering task in TREC-8 (Voorhees, 1999), which was based on open-domain question an- swering. The result of this research is a number of systems and QA methodologies not only for generic domains (Moldovan et al., 2003), but also for restricted domains (Moll´a et al., 2003) and Web-based systems (Zheng, 2002). Each type of QA system has specific issues and methodologies. Thus, open-domain QA can rely on generic tools and resources such as parsers, named-entity recognisers, and lex- ical resources like WordNet (Fellbaum, 1998). This can be seen in recent TREC confer- ences (Voorhees, 2004b) where some of the par- ticipants used readily available third-party re- sources to quickly build systems that obtained satisfactory results for the amount of effort in- vested. On the other hand, restricted domain QA can take advantage of deep knowledge of the cov- ered area by using resources that are specific to the domain such as terminology lists and ontolo- gies, for example in the domain of biomedicine (Zweigenbaum, 2003). Finally, web-based QA can take advantage of the enormous amount of data available on the World Wide Web and use data-intensive ap- proaches that exploit the inherent redundancy to find answers (Brill et al., 2001). Our system belongs to this category. The satisfaction of the user with a certain answer will depend on various factors. For in- stance, someone who wants to find some spe- cific fact would be satisfied with a short and brief answer while someone else may require a more detailed answer. These kind of differences between casual users of generic domain QA sys- tems make the establishment of personalized an- swer models difficult. One way that may satisfy both types of users is by providing an exact answer while at the same time showing a snippet of the original text from where the answer was extracted. This kind of response was required from the partici- pant systems of the main task of the QA-track of TREC-2003 (Voorhees, 2004a). We have adopted this approach by providing the exact answer, a summary, and a link to …

Content truncated to 500 words