<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-1218">
  <Title>Adapting an NER-System for German to the Biomedical Domain</Title>
  <Section position="6" start_page="94" end_page="94" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have demonstrated the adaptation of an NE tagger originally developed for German to the biomedical domain. We believe that the process of adaptation is able to sketch out some interesting aspects of the new domain.</Paragraph>
    <Paragraph position="1"> The names of the biomedical domain have morphological features that can be covered by the subword-form representation with positional character n-grams.</Paragraph>
    <Paragraph position="2"> The failure of the techniques based on the three-level model indicate that the polysemic and homonymic items and the complexity of biological names hamper or even inhibit a further optimization of models based on simple n-grams of words. We believe that the consideration of more  Instead of the positional character n-grams the system is trained on surface words.</Paragraph>
    <Paragraph position="3"> complex units and longer distant phenomena will lead to further progress in NE-tagging. For the biomedical domain, the work of Takeuchi and Collier (2003) demonstrates the successful incorporation of shallow parsing.</Paragraph>
    <Paragraph position="4"> For future research, we plan to address these issues by focusing on learning external evidence, i.e. triggers and longer-distant phenomena from unlabeled texts.</Paragraph>
  </Section>
class="xml-element"></Paper>