<?xml version="1.0" standalone="yes"?>
<Paper uid="J03-4004">
  <Title>Adjectives Using Automatically Acquired Selectional Preferences</Title>
  <Section position="8" start_page="652" end_page="653" type="concl">
    <SectionTitle>
7. Conclusion
</SectionTitle>
    <Paragraph position="0"> We quantified coverage and accuracy of sense disambiguation of verbs, adjectives, and nouns in the SENSEVAL-2 English all-words test corpus, using automatically acquired selectional preferences. We improved coverage and recall by applying the one-sense-per-discourse heuristic. The results show that disambiguation models using only selectional preferences can perform with accuracy well above the random baseline, although accuracy would not be high enough for applications in the absence of  McCarthy and Carroll Disambiguating Using Selectional Preferences other knowledge sources (Stevenson and Wilks 2001). The results compare well with those for other systems that do not use sense-tagged training data.</Paragraph>
    <Paragraph position="1"> Selectional preferences work well for some word combinations and grammatical relationships, but not well for others. We hope in future work to identify the situations in which selectional preferences have high precision and to focus on these at the expense of coverage, on the assumption that other knowledge sources can be used where there is not strong evidence from the preferences. The first-sense heuristic, based on sense-tagged data such as that available in SemCor, seems to beat unsupervised models such as ours. For many words, however, the predominant sense varies across domains, and so we contend that it is worth concentrating on detecting when the first sense is not relevant, and where the selectional-preference models provide a high probability for a secondary sense. In these cases evidence for a sense can be taken from multiple occurrences of the word in the document, using the one-sense-per-discourse heuristic.</Paragraph>
  </Section>
class="xml-element"></Paper>