<?xml version="1.0" standalone="yes"?>
<Paper uid="C96-1033">
  <Title>FeasPar - A Feature Structure Parser Learning to Parse Spoken Language</Title>
  <Section position="8" start_page="191" end_page="192" type="evalu">
    <SectionTitle>
6 Results
</SectionTitle>
    <Paragraph position="0"> FeasPar is compared with a handmodeled LRparser. The handmodeling effort for FeasPar is 2 weeks. The handmodeling effort tbr the LR-parser was 4 months.</Paragraph>
    <Paragraph position="1"> The evaluation environment is the JANUS speech translation system for the Spontaneous Scheduling Task. The system have one parser and one generator per language. All parsers and generators are written using CMU's GLR/GLR* system(Lavie and Tomita, 1993). They all share the same interlingua, ILT, which is a special case of LFG or feature structures.</Paragraph>
    <Paragraph position="2"> All Performance measures are run with transcribed (T) sentences and with speech (S) sentences containing speech recognition errors. Performance measure 1 is the feature accuracy, where all features of a parser-nmde feature structure are compared with feature of the correct handmodeled feature structure. Performance measure 2 is the end-to-end translation ratio for acceptable non-trivial sentences achieved when LR-generators are used as back-ends of the parsers. Performance measure 2 uses an English LR-generator (handmodeled for 2 years), providing results for Englishto-English translation, whereas performance measure 3 uses a German LR-generator (handmodeled  for 6 months), hence providing results for English-to-German translations. Results for an unseen, independent evaluation set are shown in Figure 7.</Paragraph>
    <Paragraph position="3"> As we see, FeasPar is better than the LR-parser in all six comparison perforInance measures made.</Paragraph>
  </Section>
class="xml-element"></Paper>