<?xml version="1.0" standalone="yes"?>
<Paper uid="J01-3001">
  <Title>The Interaction of Knowledge Sources in Word Sense Disambiguation</Title>
  <Section position="9" start_page="344" end_page="345" type="concl">
    <SectionTitle>
7. Conclusion
</SectionTitle>
    <Paragraph position="0"> Previously reported WSD systems that enjoyed a high level of accuracy have often operated on restricted vocabularies and employed a single WSD methodology. These methods have often been pursued for sound reasons to do with evaluation, but have been limited in their applicability and also in their persuasiveness regarding the scal- null Computational Linguistics Volume 27, Number 3 ability and interaction of the various WSD partial methods. This paper reported a system which disambiguated all content words in a text, as defined by a standard machine readable dictionary, with a high degree of accuracy.</Paragraph>
    <Paragraph position="1"> Our evaluation shows that disambiguation can be carried out with more accurate results when several knowledge sources are combined. It remains unclear exactly what it means to optimize the combination of modules within a learning system like T+-MBL: we could, in further work, treat the part-of-speech tagger as a partial tagger and not a filter, and we could allow the system to learn some &amp;quot;optimal&amp;quot; weighting of all the partial taggers. It also remains an interesting question whether, because of the undoubted existence of novel senses in text, a sense tagger can ever reach the level that part-of-speech tagging has. However, we believe we have shown that interesting combinations of WSD methods on a substantial training corpus are possible, and that this can show, among other things, the relative independence of the types of semantic information expressed by the various forms of lexical input.</Paragraph>
  </Section>
class="xml-element"></Paper>