<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1517">
  <Title>Efficient extraction of grammatical relations</Title>
  <Section position="8" start_page="168" end_page="168" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have described a dynamic programming approach based on the Inside Outside Algorithm for producing weighted grammatical relation output directly from a unification-based parse forest. In an evaluation on a standard test suite the approach achieves substantial improvements in accuracy and parser throughput over a previous implementation.</Paragraph>
    <Paragraph position="1"> The approach is novel as it allows multiple heads (and inside probabilities) per parse forest node instead of manipulating the parse forest so that each node represents only a single head.</Paragraph>
    <Paragraph position="2"> We intend to extend this work to develop more sophisticated parse selection schemes based on weighted GR output. Re-ranking the n-best GR sets results in a consistent but not necessarily a coherent set of GRs. Given the increased upper bound on precision for the high precision GR output, we hope to boost the corresponding recall measure by determining a consistent and coherent set of GRs from the weighted GR set.</Paragraph>
  </Section>
class="xml-element"></Paper>