<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1030">
  <Title>Head-Driven Parsing for Word Lattices</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Previous Work
</SectionTitle>
    <Paragraph position="0"> The largest improvements in word error rate (WER) have been seen with n-best list rescoring. The best n hypotheses of a simple speech recognizer are processed by a more sophisticated language model and re-ranked. This method is algorithmically simpler than parsing lattices, as one can use a model developed for strings, which need not operate strictly left to right. However, we confirm the observation of (Ravishankar, 1997; Hall and Johnson, 2003) that parsing word lattices saves computation time by only parsing common substrings once.</Paragraph>
    <Paragraph position="1"> Chelba (2000) reports WER reduction by rescoring word lattices with scores of a structured language model (Chelba and Jelinek, 2000), interpolated with trigram scores. Word predictions of the structured language model are conditioned on the two previous phrasal heads not yet contained in a bigger constituent. This is a computationally intensive process, as the dependencies considered can be of arbitrarily long distances. All possible sentence prefixes are considered at each extension step.</Paragraph>
    <Paragraph position="2"> Roark (2001) reports on the use of a lexicalized probabilistic top-down parser for word lattices, evaluated both on parse accuracy and WER. Our work is different from Roark (2001) in that we use a bottom-up parsing algorithm with dynamic programming based on the parsing model II of Collins (1999).</Paragraph>
    <Paragraph position="3"> Bottom-up chart parsing, through various forms of extensions to the CKY algorithm, has been applied to word lattices for speech recognition (Hall and Johnson, 2003; Chappelier and Rajman, 1998; Chelba and Jelinek, 2000). Full acoustic and n-best lattices filtered by trigram scores have been parsed.</Paragraph>
    <Paragraph position="4"> Hall and Johnson (2003) use a best-first probabilistic context free grammar (PCFG) to parse the input lattice, pruning to a set of local trees (candidate partial parse trees), which are then passed to a version of the parser of Charniak (2001) for more refined parsing. Unlike (Roark, 2001; Chelba, 2000), Hall and Johnson (2003) achieve improvement in WER over the trigram model without interpolating its lattice parser probabilities directly with trigram probabilities. null</Paragraph>
  </Section>
class="xml-element"></Paper>