<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1045">
  <Title>Less is more: Eliminating index terms from subordinate clauses</Title>
  <Section position="9" start_page="354" end_page="354" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have demonstrated that, as implicitly predicted by RST, index terms may be eliminated from certain kinds of subordinate clauses without substantially affecting precision or recall. Rather than using NLP to generate more index terms, we have found tremendous gains from systematically eliminating terms. The exact severity of the impact on precision and recall that results from eliminating terms varies by genre. In all cases, however, the systematic elimination of subordinate clause material is substantially better than arbitrary deletion of index terms or the deletion of index terms that occur only in main clauses.</Paragraph>
    <Paragraph position="1"> Future research shall attempt to refine the analysis of the kinds of subordinate clauses from which index terms can be omitted, and to integrate these findings with conventional statistical IR algorithms.</Paragraph>
  </Section>
class="xml-element"></Paper>