<?xml version="1.0" standalone="yes"?>
<Paper uid="P00-1060">
  <Title>An Information-Theory-Based Feature Type Analysis for the Modelling of Statistical Parsing SUI Zhifang +++ , ZHAO Jun + , Dekai WU + +</Title>
  <Section position="19" start_page="121" end_page="121" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> The paper proposes an information-theory-based feature type analysis method, which not only presents a series of heuristic conclusion on the predictive power of the different feature types and feature type combination for syntactic parsing, but also provides a guide for the modeling of syntactic parsing in the view of methodology, that is, we can quantitatively analyse the different contextual feature types or feature types combination's effect for syntactic structure prediction in advance. Based on these analysis, we can select the feature type or feature types combination that has the optimal predictive information summation to build the probabilistic parsing model.</Paragraph>
    <Paragraph position="1"> However, there are still some questions to be answered in this paper. For example, what is the beneficial improvement in the performance after using this method in a real parser? Whether the improvements in PIQ will lead to the improvement of parsing accuracy or not? In the following research, we will incorporate these conclusions into a real parser to see whether the parsing accuracy can be improved or not.</Paragraph>
    <Paragraph position="2"> Another work we will do is to do some experimental analysis to find the impact of data sparseness on feature type analysis, which is critical to the performance of real systems.</Paragraph>
    <Paragraph position="3"> The proposed feature type analysis method can be used in not only the probabilistic modelling for statistical syntactic parsing, but also language modelling in more general fields [WU, 1999a] [WU, 1999b].</Paragraph>
  </Section>
class="xml-element"></Paper>