<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-0813">
  <Title>Inferring Semantic Similarity from Distributional Evidence: an Analogy-based Approach to Word Sense Disambiguation*</Title>
  <Section position="6" start_page="95" end_page="95" type="concl">
    <SectionTitle>
5 Concluding remarks
</SectionTitle>
    <Paragraph position="0"> In this paper we described a WSD system which uses a notion of semantic similarity based on distributional evidence. Prehminary results look promising.</Paragraph>
    <Paragraph position="1"> The described measure of semantic similarity offers significant advantages compared with methods where word similarity is evaluated either in statistical terms, ultimately based on token frequency, or through reference to a hierarchically structured thesaurus. First, good results are achieved with small quantities of data, part of which are not even semantically disambiguated. Second, the suggested measure is sensitive to similarities'which are relevant to the context being disambiguated, thus overcoming one of the major drawbacks of fixed decontextualised semantic hierarchies.</Paragraph>
    <Paragraph position="2"> On a more practical front, this measure was evaluated as an integral part of the disambiguation strategy of SENSE, whose main advantages over other WSD systems can be summarised as follows: * SENSE does not take attested evidence at face value but always entertains other hypotheses; * SENSE's inferences are not restricted to contexts which exhibit symmetric syntactic dependencies, but also exploit alternations in argument surface realisation with semantically related verbs; * SENSE is sensitive to the semantic generality/specificity of supporting evidence.</Paragraph>
  </Section>
class="xml-element"></Paper>