<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1019">
  <Title>Modelling the substitutability of discourse connectives</Title>
  <Section position="8" start_page="154" end_page="155" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> The concepts of lexical similarity and substitutability are of central importance to psychology, artificial intelligence and computational linguistics.</Paragraph>
    <Paragraph position="1">  To our knowledge this is the first modelling study of how these concepts relate to lexical items involved in discourse-level phenomena. We found a three way correspondence between data sources of quite distinct types: distributional similarity scores obtained from lexical co-occurrence data, substitutability judgements made by linguists, and the similarity ratings of naive subjects.</Paragraph>
    <Paragraph position="2"> The substitutability of lexical items is important for applications such as text simplification, where it can be desirable to paraphrase one discourse connective using another. Ultimately we would like to automatically predict substitutability for individual tokens. However predicting whether one connective can either a) always, b) sometimes or c) never be substituted for another is a step towards this goal.</Paragraph>
    <Paragraph position="3"> Our results demonstrate that these general substitutability relationships have empirical correlates. We have introduced a novel variance-based function of two distributions which complements distributional similarity. We demonstrated the new function's utility in helping to predict the substitutability of connectives, and it can be expected to have wider applicability to lexical acquisition tasks. In particular, it is expected to be useful for learning relationships which cannot be characterised purely in terms of similarity, such as hyponymy. In future work we will analyse further the empirical properties of the new function, and investigate its applicability to learning relationships between other classes of lexical items such as nouns.</Paragraph>
  </Section>
class="xml-element"></Paper>