<?xml version="1.0" standalone="yes"?>
<Paper uid="I05-5010">
  <Title>Automatic generation of large-scale paraphrases</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Much work on paraphrase generation has focussed on lexical variation and syntactic transformation within individual sentences (Barzilay and McKeown, 2001; Carroll et al., 1999; Dras, 1999; Inui and Nogami, 2001; Kozlowski et al., 2003; Langkilde and Knight, 1998; Takahashi et al., 2001; Stede, 1999). Our interest in this paper lies instead with variations at the level of text structuring -- the way in which propositions are grouped into units like paragraphs, sections, and bulletted lists, and linked rhetorically by discourse connectives such as 'since', 'nevertheless', and 'however'. Elsewhere, we have described a text-structuring method in which the options for organising propositions in a text are laid out as a set of constraints, so that acceptable solutions can be enumerated using constraint satisfaction and evaluated using a cost metric (Power et al., 2003).</Paragraph>
    <Paragraph position="1"> In this paper we show how this method, when harnessed to a system for recognising rhetorical structure in an input text, can be employed in order to produce large-scale paraphrases fulfilling purposes like improving coherence and achieving a desired style of layout.</Paragraph>
  </Section>
class="xml-element"></Paper>