<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1088">
  <Title>Wolfson Building Parks Road</Title>
  <Section position="10" start_page="702" end_page="703" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> The NLP community may consider POS tagging to be a solved problem. In this paper, we have suggested two reasons why this is not the case. First, tagging for lexicalized-grammar formalisms, such as CCG and TAG, is far from solved. Second, even modest improvements in POS tagging accuracycanhavealargeimpactontheperformanceof null downstream components in a language processing pipeline.</Paragraph>
    <Paragraph position="1">  We have developed a novel approach to maintaining tag ambiguity in language processing pipelines which avoids premature ambiguity resolution. The tag ambiguity is maintained by using the forward-backward algorithm to calculate individual tag probabilities. These probabilities can then be used to select multiple tags and can also be encoded as real-valued features in subsequent statistical models.</Paragraph>
    <Paragraph position="2"> With this new approach we have increased POS  taggingaccuracysignificantlywithonlyatinyambiguity penalty and also significantly improved on previous CCG supertagging results. Finally, using POS tag probabilities as real-valued features in the supertagging model, we demonstrated performance close to that obtained with gold-standard POS tags. This will significantly improve the robustness of the parser on unseen text.</Paragraph>
    <Paragraph position="3"> In future work we will investigate maintaining tag ambiguity further down the language processing pipeline and exploiting the uncertainty from previous stages. In particular, we will incorporate real-valued POS tag and lexical category features in the statistical parsing model. Another possibility is to investigate whether similar techniques can improveothertaggingtasks,suchasNamedEntity Recognition.</Paragraph>
    <Paragraph position="4"> This work can be seen as part of the larger goal of maintaining ambiguity and exploiting un- null certainty throughout language processing systems (Roth and Yih, 2004), which is important for coping with the compounding of errors that is a significant problem in language processing pipelines.</Paragraph>
  </Section>
class="xml-element"></Paper>