<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0814">
  <Title>Evaluating the results of a memory-based word-expert approach to unrestricted word sense disambiguation.</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The task of word sense disambiguation (WSD) is to assign a sense label to a word in context. Both knowledge-based and statistical methods have been applied to the problem. See (Ide and V'eronis, 1998) for an introduction to the area. Recently (both SENSEVAL competitions), various machine learning (ML) approaches have been demonstrated to produce relatively successful WSD systems, e.g.</Paragraph>
    <Paragraph position="1"> memory-based learning (Ng and Lee, 1996; Veenstra et al., 2000), decision lists (Yarowsky, 2000), boosting (Escudero et al., 2000).</Paragraph>
    <Paragraph position="2"> In this paper, we evaluate the results of a memory-based learning approach to WSD. We ask ourselves whether we can learn lessons from the errors made in the SENSEVAL-2 competition. More particularly, we are interested whether there are words or categories of words which are more difficult to predict than other words. If so, do these words have certain characteristic features? We furthermore investigate the interaction between the use of different information sources and the part-of-speech categories of the ambiguous words. We also study the relation between the accuracy of the word-experts and their number of training items, number of senses and sense distribution. For these experiments, we performed all SENSEVAL-2 experiments all over again.</Paragraph>
    <Paragraph position="3"> In the following Section, we briefly outline the WSD architecture used in the experiments, and discuss the word-expert approach and the optimization procedure. Furthermore, a brief overview is given of the results of the different components of the word-experts on the train set and the SENSEVAL-2 test material. In Section 3, we evaluate the results of the different classifiers per part-of-speech category. In the July 2002, pp. 95-101. Association for Computational Linguistics. Disambiguation: Recent Successes and Future Directions, Philadelphia, Proceedings of the SIGLEX/SENSEVAL Workshop on Word Sense same Section, these results are further analysed in relation to the number of training items, the number of senses and the sense distribution. Section 4 gives a detailed analysis of the results of our approach on the SENSEVAL-2 test material. We end with some concluding remarks in Section 5.</Paragraph>
  </Section>
class="xml-element"></Paper>