<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0429">
  <Title>Named Entity Recognition using Hundreds of Thousands of Features</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> The SVM-Lattice approach appears to give good results without language-specific tuning; it handily outperforms the CoNLL-2003 Shared Task baseline, and beats a basic HMM tagger as well. Use of SVMs allows the introduction of a large number of features. These features can be introduced with little concern for dependency among features, and without significant knowledge of the target language. It is likely that our results reflect some degree of overfitting, given the large number of parameters we use; however, we suspect this effect is not large. Thus, the SVM-Lattice technique is particularly well suited to language-neutral entity recognition. We expect it will also perform well on other tasks that can be cast as tagging problems, such as part-of-speech tagging and syntactic chunking.</Paragraph>
  </Section>
class="xml-element"></Paper>