<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0804">
  <Title>SENSEVAL-3 TASK Word-Sense Disambiguation of WordNet Glosses</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 The Senseval-3 Task
</SectionTitle>
    <Paragraph position="0"> Participants were provided with all glosses from WordNet in which at least one open-class word was given a &amp;quot;Gold&amp;quot; quality assignment. These glosses were provided in an XML file, each with its WordNet synset number, its part of speech, and the gloss itself.</Paragraph>
    <Paragraph position="1"> Glosses frequently include sample uses. The samples uses were not parsed in the XWN project and were not to be included in the submissions.</Paragraph>
    <Paragraph position="2"> The task was configured as essentially identical to the SENSEVAL-2 and SENSEVAL-3 &amp;quot;all-words&amp;quot; tasks, except without any context and with the gloss not constituting a complete sentence. Unlike the all-words task, individual tokens to be disambiguated were not identified, so that participants were required to perform their own tokenization and identification of multiword units. The number of words in a gloss is quite small, but a few glosses do contain the same word more than once. Participants were encouraged to consider a synset's placement within WordNet (its hypernyms, hyponyms, and other relations) to assist in disambiguation. The XWN data contains part of speech tags for each word in the glosses, as well as parses and logical forms, which participants were allowed to use. Most of the glosses in the test set have hand-tagged words as well as words tagged by the automatic XWN systems. The senses assigned to other open-class words have a tag of &amp;quot;silver&amp;quot; or &amp;quot;normal&amp;quot;. In submitting test runs, participants did not know which of the words had been assigned a &amp;quot;gold&amp;quot; quality, but were only scored for the &amp;quot;gold&amp;quot; quality words.2 No training data was available for this task since the number of items in the test set was so small.</Paragraph>
    <Paragraph position="3"> Participants were encouraged to become familiar with the XWN dataset and to make use of it in ways that would not compromise their performance of the task.</Paragraph>
  </Section>
class="xml-element"></Paper>