<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1042">
  <Title>Using Knowledge to Facilitate Factoid Answer Pinpointing</Title>
  <Section position="6" start_page="2" end_page="2" type="concl">
    <SectionTitle>
5. Conclusions
</SectionTitle>
    <Paragraph position="0"> It is tempting to search for a single technique that will solve the whole problem (for example, Ittycheriah et al. (2001) focus on the subset of factoid questions answerable by NPs, and train a statistical model to perform NP-oriented answer pinpointing). Our experience, however, is that even factoid QA is varied enough to require various special-purpose techniques and knowledge. The theoretical limits of the various techniques are not known, though Light et al.'s (2001) interesting work begins to study this.</Paragraph>
    <Paragraph position="1">  Semantic relation scores measured only on questions in which they could logically apply.</Paragraph>
    <Paragraph position="2"> We conclude that factoid QA performance can be significantly improved by the use of knowledge attuned to specific question types and specific information characteristics. Most of the techniques for exploiting this knowledge require learning to ensure robustness. To improve performance beyond this, we believe a combination of going to the web and turning to deeper world knowledge and automated inference (Harabagiu et al., 2001) to be the answer. It remains an open question how much work these techniques would require, and what their payoff limits are.</Paragraph>
  </Section>
class="xml-element"></Paper>