<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-3002">
  <Title>Unsupervised Part-of-Speech Tagging Employing Efficient Graph Clustering</Title>
  <Section position="9" start_page="11" end_page="11" type="concl">
    <SectionTitle>
5 Conclusion and further work
</SectionTitle>
    <Paragraph position="0"> We presented a graph-based approach to unsupervised POS tagging. To our knowledge, this is the first attempt to leave the decision on tag granularity to the tagger. We supported the claim of language-independence by validating the output of our system against supervised systems in three languages.</Paragraph>
    <Paragraph position="1"> The system is not very sensitive to parameter changes: the number of feature words, the frequency cutoffs, the log-likelihood threshold and all other parameters did not change overall performance considerably when altered in reasonable limits. In this way it was possbile to arrive at a one-size-fits-all configuration that allows the parameter-free unsupervised tagging of large corpora.</Paragraph>
    <Paragraph position="2"> To really judge the benefit of an unsupervised tagging system, it should be evaluated in an application-based way. Ideally, the application should tell us the granularity of our tagger: e.g. semantic class learners could greatly benefit from the high-granular word sets arising in both of our partitionings, which we endeavoured to lump into a coarser tagset here.</Paragraph>
  </Section>
class="xml-element"></Paper>