<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1188">
  <Title>Information Extraction for Question Answering: Improving Recall Through Syntactic Patterns</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We described a set of experiments aimed at comparing different information extraction methods in the context of off-line corpus-based Question Answering. Our main finding is that a linguistically deeper method, based on dependency parsing and a small number of simple syntactic patterns, allows an off-line QA system to correctly answer substantially more questions than a traditional method based on surface text patterns. Although the syntactic method showed lower precision of the extracted facts (61% vs. 68%), in spite of parsing errors the recall was higher than that of the surface-based method, judging by the number of correctly answered questions (31% vs. 23%). Thus, the syntactic analysis can in fact be considered as another, intensive way of improving the recall of information extraction, in addition to successfully used extensive ways, such as developing larger numbers of surface patterns or increasing the size of the collection.</Paragraph>
    <Paragraph position="1"> Moreover, we confirmed the claim that for a complex off-line QA system, with statistical as well as knowledge-intensive sanity checking answer selection modules, recall of the information extraction module is more important than precision, and a simple WordNet-based method for improving precision does not help QA. In our future work we plan to investigate the effect of more sophisticated and, probably, more accurate filtering methods (Fleischman et al., 2003) on the QA results.</Paragraph>
  </Section>
class="xml-element"></Paper>