<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1513">
  <Title>Vancouver, October 2005. c(c)2005 Association for Computational Linguistics A Classifier-Based Parser with Linear Run-Time Complexity</Title>
  <Section position="8" start_page="130" end_page="131" type="concl">
    <SectionTitle>
5 Conclusion and Future Work
</SectionTitle>
    <Paragraph position="0"> We have presented a simple shift-reduce parser that uses a classifier to determine its parsing actions and runs in linear time. Using SVMs for classification, the parser has labeled constituent precision and recall higher than 87% when using the correct part-of-speech tags, and slightly higher than 86% when using automatically assigned part-of-speech tags. Although its accuracy is not as high as those of state-of-the-art statistical parsers, our classifier-based parser is considerably faster than several well-known parsers that employ search or dynamic programming approaches. At the same time, it is significantly more accurate than previously proposed deterministic parsers for constituent structures.</Paragraph>
    <Paragraph position="1"> We have also shown that much of the success of a classifier-based parser depends on what classifier is used. While this may seem obvious, the differences observed here are much greater than what would be expected from looking, for example, at results from chunking/shallow parsing (Zhang et al., 2001; Kudo and Matsumoto, 2001; Veenstra and van den Bosch, 2000).</Paragraph>
    <Paragraph position="2"> Future work includes the investigation of the effects of individual features, the use of additional classification features, and the use of different classifiers. In particular, the use of tree features seems appealing. This may be accomplished with SVMs  using a tree kernel, or the tree boosting classifier BACT described in (Kudo and Matsumoto, 2004).</Paragraph>
    <Paragraph position="3"> Additionally, we plan to investigate the use of the beam strategy of Ratnaparkhi (1997) to pursue multiple parses while keeping the run-time linear.</Paragraph>
  </Section>
class="xml-element"></Paper>