<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-1063">
  <Title>Text Chunking by Combining Hand-Crafted Rules and Memory-Based Learning</Title>
  <Section position="8" start_page="3" end_page="3" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we have proposed a new method to learn chunking Korean by combining the hand-crafted rules and a memory-based learning. Our method is based on the rules, and the estimates on chunks by the rules are verified by a memory-based learning. Since the memory-based learning is an efficient method to handle exceptional cases of the rules, it supports the rules by making decisions only for the exceptions of the rules. That is, the memory-based learning enhances the rules by efficiently handling the exceptional cases of the rules.</Paragraph>
    <Paragraph position="1"> The experiments on STEP 2000 dataset showed that the proposed method improves the F-score of the rules by 2.34 and of the memory-based learning by 2.83. Even compared with support vector machines, the best machine learning algorithm in text chunking, it achieved the improvement of 1.67.</Paragraph>
    <Paragraph position="2"> The improvement was made mainly in noun phrases among four kinds of phrases in Korean. This is because the errors of the rules are mostly related with noun phrases. With relatively many instances for noun phrases, the memory-based learning could compensate for the errors of the rules. We also empirically found the threshold value t used to determine when to apply the rules and when to apply memory-based learning.</Paragraph>
    <Paragraph position="3"> We also discussed some issues in combining a rule-based method and a memory-based learning.</Paragraph>
    <Paragraph position="4"> These issues will help to understand how the method works and to apply the proposed method to other problems in natural language processing. Since the method is general enough, it can be applied to other problems such as POS tagging and PP attachment.</Paragraph>
    <Paragraph position="5"> The memory-based learning showed good performance in these problems, but did not reach the stateof-the-art. We expect that the performance will be improved by the proposed method.</Paragraph>
  </Section>
class="xml-element"></Paper>