<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0804">
  <Title>SENSEVAL-3 TASK Word-Sense Disambiguation of WordNet Glosses</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
4 Conclusions
</SectionTitle>
    <Paragraph position="0"> The underlying guidance for this SENSEVAL-3 task that, in the absence of significant context, participants make use of WordNet relations for disambiguating glosses has led to some significant insights about the use and importance of wordnets.</Paragraph>
    <Paragraph position="1"> These insights emerge from the tension between the reliance on WordNet relations and the imprecision of the tagging process.</Paragraph>
    <Paragraph position="2"> Many investigators, including several of the participants in this task, are attempting to exploit the kinds of relations between lexical entries that are embodied in WordNet. The use of wordnets in NLP applications has become an important basic construct and increasingly valuable. However, the construction of wordnets is expensive and time-consuming, and without any significant prospects for commercial support. While some dictionary publishers are increasingly incorporating wordnet principles into their lexical resources, this process is slow. At present, the publicly available WordNet remains the wordnet of choice.</Paragraph>
    <Paragraph position="3"> The annotation process followed by the XWN project, with the taggings used in this task, has again indicated difficulties with the WordNet sense inventory. The fact remains that WordNet has not had the benefit of sufficient lexicographic resources in the construction of its glosses and in the acquisition of other lexicographic information in its entries. The WordNet project continues its efforts to add information, but with limited resources.</Paragraph>
    <Paragraph position="4"> With the diverse set of approaches represented by the participants in this task, it is possible to envision sets of steps that might be employed to improve the details of the WordNet sense inventory. One step would include continued hand-tagging of WordNet glosses without consideration of WordNet relations.</Paragraph>
    <Paragraph position="5"> Another step would be the use of automated disambiguation routines to act as checks on consistency. Such systems would include those that rely on WordNet relations as well as those that do not, acting as checks on one another.</Paragraph>
  </Section>
class="xml-element"></Paper>