<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0202">
  <Title>Comparing Information Extraction Pattern Models</Title>
  <Section position="8" start_page="17" end_page="17" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper compares four IE pattern models: SVO, chains, linked chains and subtrees. Using texts from the management succession and biomedical domains it was found that the linked chains model can represent around 95% of the possible relations contained in the text, given a dependency parse. Subtrees can represent all the relations contained within dependency trees but their use is less practical because enumerating all possible subtrees is a more complex problem and the large number of resulting patterns could limit the learning algorithms that can be applied. This result should be borne in mind during the design of IE systems.</Paragraph>
  </Section>
class="xml-element"></Paper>