<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1702">
  <Title>Class Based Sense Definition Model for Word Sense Tagging and Disambiguation</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4. Experiments and evaluation
</SectionTitle>
    <Paragraph position="0"> In order to assess the feasibility of the proposed approach, we carried out experiments and evaluation on an implementation of MARS and SWAT based on LDOCE E-C, LLOCE, and Sinorama.</Paragraph>
    <Paragraph position="1"> First experiment was involved with the trainability of CBSDM and CBSTM via MARS. The second experiment was involved with the effectiveness of using SWAT and CBSTM to annotate a parallel corpus with sense information. Evaluation was done on a set of 14 nouns, verbs, adjectives, and adverbs studies in previous work. The set includes the nouns &amp;quot;bass,&amp;quot; &amp;quot;bow,&amp;quot; &amp;quot;cone,&amp;quot; &amp;quot;duty,&amp;quot; &amp;quot;gallery,&amp;quot; &amp;quot;mole,&amp;quot; &amp;quot;sentence,&amp;quot; &amp;quot;slug,&amp;quot; &amp;quot;taste,&amp;quot; &amp;quot;star,&amp;quot; &amp;quot;interest,&amp;quot; &amp;quot;issue,&amp;quot; the adjective &amp;quot;hard,&amp;quot; and the verb &amp;quot;serve.&amp;quot;</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
4.1 Experiment 1: Training CBSDM
</SectionTitle>
      <Paragraph position="0"> We applied MARS to assign LDOCE senses to word classes in LLOCE. Some results related to the test set are shown in Tables 4. The evaluation in Tables indicates that MARS assigns LDOCE senses to an LLOCE class with a high average precision rate of 90%.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
4.2 Experiment 2: Sense Tagging
</SectionTitle>
      <Paragraph position="0"> We applied SWAT to sense tag English words in some 50,000 reliably aligned sentence pairs in Sinorama parallel Corpus based on LDOCE sense inventory. The results are shown in Tables 6.</Paragraph>
      <Paragraph position="1"> Evaluation indicates an average precision rate of around 90%.</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
5. Discussion
</SectionTitle>
    <Paragraph position="0"> The proposed approach offers a new method for automatic learning for the task of word sense disambiguation. The class based approach attacks the problem of tagging and data sparseness in a way similar to the Yarowsky approach (1992) based on thesaurus categories. We differ from the Yarowsky's approach, in the following ways: i. The WSD problem is solved for two languages instead of one within a single sense inventory. Furthermore, an explicit sense tagged corpus is produced in the process.</Paragraph>
    <Paragraph position="1"> ii. It is possible to work with any number of sense inventories. null iii. The method is applicable not only to nouns but also to adjectives and verbs, since it does not rely on topical context, which is effective only for nouns as pointed out by Towell and Voorhees (1998).</Paragraph>
    <Paragraph position="2"> The approach is very general and modular and can work in conjunction with a number of learning strategies for word sense disambiguation (Yarowsky, 1995; Li and Li, 2002).</Paragraph>
  </Section>
class="xml-element"></Paper>