<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1041">
  <Title>Long-Distance Dependency Resolution in Automatically Acquired Wide-Coverage PCFG-Based LFG Approximations</Title>
  <Section position="11" start_page="0" end_page="0" type="concl">
    <SectionTitle>
10 Conclusions
</SectionTitle>
    <Paragraph position="0"> We presented and extensively evaluated a finite approximation of LDD resolution in automatically constructed, wide-coverage, robust, PCFG-based LFG approximations, effectively turning the &amp;quot;half&amp;quot;(or &amp;quot;shallow&amp;quot;)-grammars presented in (Cahill et al., 2002) into &amp;quot;full&amp;quot; or &amp;quot;deep&amp;quot; grammars. In our approach, LDDs are resolved in f-structure, not trees. The method achieves a preds-only f-score of 80.97% for f-structures with the PA-PCFG in the integrated architecture against the DCU 105 and 78.4% against the 2,416 automatically generated f-structures for the original Penn-II treebank trees. Evaluating against the PARC 700 Dependency Bank, the P-PCFG achieves an f-score of 80.24%, an overall improvement of approximately 0.6% on the result reported for the best hand-crafted grammars in (Kaplan et al., 2004).</Paragraph>
  </Section>
class="xml-element"></Paper>