<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2404">
  <Title>Combining Lexical and Syntactic Features for Supervised Word Sense Disambiguation</Title>
  <Section position="9" start_page="2" end_page="2" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We conducted an extensive array of word sense disambiguation experiments using a rich set of lexical and syntactic features. We use the SENSEVAL-2, SENSEVAL-1, line, hard, serve and interest data which together have more than 50,000 sense tagged instances. We show that both lexical and syntactic features achieve reasonably good accuracies when used individually, and that the part of speech of the word immediately following the target word is particularly useful in disambiguation as compared to other individual part of speech features. A combination of part of speech features attains even better accuracies and we identify (P  most potent combinations. We show that the head word of a phrase is particularly useful in disambiguating adjectives and nouns. We identify the head and parent as the most potent parse feature combination.</Paragraph>
    <Paragraph position="1"> We introduce the measures Baseline Ensemble and Optimal Ensemble which quantify the redundancy among two feature sets and the maximum accuracy attainable by an ensemble technique using the two feature sets. We show that even though lexical and syntactic features are redundant to a certain extent, there is a significant amount of complementarity. In particular, we showed that simple lexical features (unigrams and bigrams) used in conjunction with part of speech features have the potential to achieve state of the art results.</Paragraph>
  </Section>
class="xml-element"></Paper>