<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-1023">
  <Title>A Question Answering System Supported by Information Extraction*</Title>
  <Section position="6" start_page="168" end_page="170" type="evalu">
    <SectionTitle>
3 Limitation
</SectionTitle>
    <Paragraph position="0"> The first limitation comes from the types of questions. Currently only wh-questions are handled although it is planned that yes-no questions will be handled once we introduce CE  and GE templates to support QA. Among the wh-questions, the why-question and how-question t are more challenging because the asking point cannot be simply mapped to the NE types/sub-types.</Paragraph>
    <Paragraph position="1"> The second limitation is from the nature of the questions. Questions like Where can l find the homepage for Oscar winners or Where can I find info on Shakespeare's works might be answerable easily by a system based on a well-maintained data base of home pages. Since our system is based on the processing of the underlying documents, no correct answer can be provided if there is no such an answer (explicitly expressed in English) in the processed documents. In TREC-8 QA, this is not a problem since every question is guaranteed to have at least one answer in the given document pool.</Paragraph>
    <Paragraph position="2"> However, in the real world scenario such as a QA portal, it is conceived that the IE results based on the processing of the documents should be complemented by other knowledge sources such as e-copy of yellow pages or other manually maintained and updated data bases.</Paragraph>
    <Paragraph position="3"> The third limitation is the lack of linguistic processing such as sentence-level parsing and cross-sentential co-reference (CO). This problem will be gradually solved when high-level IE technology is introduced into the system.</Paragraph>
  </Section>
class="xml-element"></Paper>