<?xml version="1.0" standalone="yes"?>
<Paper uid="H94-1033">
  <Title>Pattern Matching in a Linguistically- Motivated Text Understanding System</Title>
  <Section position="6" start_page="185" end_page="185" type="concl">
    <SectionTitle>
6. Conclusion
</SectionTitle>
    <Paragraph position="0"> Finite-state pattern-matching has already shown to be useful and valuable in data-extraction applications. Its full possible impact is still being investigated. For example, several groups are trying to find automatic ways to derive FS patterns in order to surmount the porting: problem they pose in systems that heavily depend on them.</Paragraph>
    <Paragraph position="1"> However, maintaining a wide-coverage linguistic core can result in excellent data-extraction capability as has been evidenced by PLUM's performance in the government-sponsored MUC evaluations. null Perhaps the most interesting result was that domain-specific patterns, though in principle very powerful, added relatively little to the performance of the linguistically motivated components. Error rate was improved by at most 3 percentage points. Nevertheless, PLUM data extraction system's performance was among the highest of all systems participating in MUC-5.</Paragraph>
    <Paragraph position="2"> While this one case study does not prove the relative efficacy of domain-specific patterns versus domain-independent, linguistically motivated processing, it does suggest that more research and development in linguistically motivated syntactic and semantic processing is promising even in the short term, not just in long range research.</Paragraph>
  </Section>
class="xml-element"></Paper>