<?xml version="1.0" standalone="yes"?>
<Paper uid="I05-3023">
  <Title>Perceptron Learning for Chinese Word Segmentation</Title>
  <Section position="8" start_page="156" end_page="156" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> Weapplied theunevenmarginsPerceptron toChinese word segmentation. The learning algorithm is simple, fast and effective. The results obtained Table5: Theofficial results ontest set: F-measure (%) for close and open tests, respectively.</Paragraph>
    <Paragraph position="1"> as cityu msr pku close 94.4 93.6 95.6 92.7 open 94.8 93.6 95.4 93.8 are encouraging.</Paragraph>
    <Paragraph position="2"> The performance of Perceptron was close to that of the SVM on Chinese word segmentation for large training data. On the other hand, the Perceptron took much less computation time than SVM.Weimplemented the Perceptron withsemiquadratic kernel in primal form. Our implementation was both effective and efficient. Oursystem performed well forthe three of four corpora, as, cityu and msr corpora. But it was significantly worse than the best result on the pku corpora, which needs further investigation.</Paragraph>
  </Section>
class="xml-element"></Paper>