<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0401">
  <Title>A model of syntactic disambiguation based on lexicalized grammars</Title>
  <Section position="5" start_page="3" end_page="3" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> This paper described a novel model for syntactic disambiguation based on lexicalized grammars, where the model selects the most probable parsing result from the candidates allowed by a lexicalized grammar. Since lexicalized grammars can restrict the possible structure of parsing results, the probabilistic model cannot simply be decomposed into independent events as in the existing disambiguation models for parsing. By applying a maximum entropy model for feature forests, we achieved probabilistic modeling without decomposition.</Paragraph>
    <Paragraph position="1"> Through experiments, we proved the syntax-only model could record with high level of accuracy with a lexicalized grammar, and maximum entropy estimation for feature forests could attain competitive accuracy compared to the traditional model. We see this work as the first step in the application of linguistically motivated grammars to the parsing of real-world texts as well as the evaluation of linguistic theories by consulting extensive corpora.</Paragraph>
    <Paragraph position="2"> Future work should include the application of our model to other lexicalized grammars including HPSG.</Paragraph>
    <Paragraph position="3"> The development of sophisticated parsing strategies is also required to improve the accuracy and efficiency of parsing. Since parsing results of lexicalized grammars such as HPSG and CCG can include non-local dependencies, we cannot simply apply well-known parsing strategies, such as beam thresholding, which assume the local computation of preference scores. Further investigations must be left for future research.</Paragraph>
  </Section>
class="xml-element"></Paper>