<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0507">
  <Title>A Practical QA System in Restricted Domains</Title>
  <Section position="7" start_page="1" end_page="1" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> This paper describes the practical QA system for restricted domains. To be practically used, our system tries to achieve high precision at the sacrifice of question coverage.</Paragraph>
    <Paragraph position="1"> To achieve high accuracy, we pre-designate semi-structured information resource webpages and extracted domain-specific information from them. We also prepare a domain-specific ontology and query frames for the question analysis. The user's request in natural language is converted into SQL expression to generate an answer for the question. Testing with a small set of queries on weather domain, the QA system showed 90.9% of precision and 75.0% of recall. By restricting the coverage of questions, our system could achieve relatively high precision.</Paragraph>
    <Paragraph position="2"> However, the figures are not enough for a real practical system.</Paragraph>
    <Paragraph position="3">  stricted domain QA engine Much work is left for our future work. First, we are expanding the domain for the system. A domain classifier will be added to the QA system to process multiple-domain questions, as represented in Figure 6. We will separate domain dependent resources (query frames, ontology containing domain-dependent information, and etc.) and domain independent resources (linguistic resources, and ontology for domain-independent information) to allow easier domain expansion.</Paragraph>
    <Paragraph position="4"> Second, the information extractor has to be upgraded. Currently, the QA system is using hand-coded wrappers, and the wrappers cannot extract necessary information robustly when the webpages are modified. We are developing an information extractor that can recognize the modification of the webpages and modify the wrappers automatically.</Paragraph>
    <Paragraph position="5"> The upgraded information extractor will improve the robustness of our system.</Paragraph>
    <Paragraph position="6"> Finally, we will increase the size of ontology to cover more question types. From the experimentation, we realize that a larger ontology for weather terms is necessary to classify a question correctly. It seems more query frames are necessary for more proper answers to the users' requests.</Paragraph>
  </Section>
class="xml-element"></Paper>