<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1504">
  <Title>Parsing with Soft and Hard Constraints on Dependency Length[?]</Title>
  <Section position="9" start_page="40" end_page="40" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have described a novel reason for identifying headword-to-headword dependencies while parsing: to consider their length. We have demonstrated that simple bilexical parsers of English, Chinese, and German can exploit a &amp;quot;short-dependency preference.&amp;quot; Notably, soft constraints on dependency length can improve both speed and accuracy, and hard constraints allow improved precision and speed with some loss in recall (on English and Chinese, remarkably little loss). Further, for the hard constraint &amp;quot;length[?]k,&amp;quot; we have given an O(nk2) partial parsing algorithm for split bilexical grammars; the grammar constant is no worse than for state-of-the-art O(n3) algorithms. This algorithm strings together the partial trees' roots along a &amp;quot;vine.&amp;quot; 28The obvious reduction for unsplit head automaton grammars, say, is only O(n4) - O(n3k), following (Eisner and Satta, 1999). Alternatively, one can convert the unsplit HAG to a split one that preserves the set of feasible (length [?] k) parses, but then g becomes prohibitively large in the worst case.</Paragraph>
    <Paragraph position="1"> Our approach might be adapted to richer parsing formalisms, including synchronous ones, and should be helpful as an approximation to full parsing when fast, high-precision recovery of syntactic information is needed.</Paragraph>
  </Section>
class="xml-element"></Paper>