<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1632">
  <Title>Using Linguistically Motivated Features for Paragraph Boundary Identification</Title>
  <Section position="4" start_page="267" end_page="267" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Compared to other text segmentation tasks, e.g.</Paragraph>
    <Paragraph position="1"> topic segmentation, PBI has received relatively little attention. We are aware of three studies which approach the problem from different perspectives.</Paragraph>
    <Paragraph position="2"> Bolshakov &amp; Gelbukh (2001) assume that splitting text into paragraphs is determined by text cohesion: The link between a paragraph initial sentence and the preceding context is weaker than the links between sentences within a paragraph. They evaluate text cohesion using a database of collocations and semantic links and insert paragraph boundaries where the cohesion is low.</Paragraph>
    <Paragraph position="3"> The algorithm of Sporleder &amp; Lapata (2004, 2006) uses surface, syntactic and language model features and is applied to three different languages and three domains (fiction, news, parliament).</Paragraph>
    <Paragraph position="4"> This study is of particular interest to us since one of the languages the algorithm is tested on is German. They investigate the impact of different features and data size, and report results significantly better than a simple baseline. However, their results vary considerably between the languages and the domains. Also, the features determined important is different for each setting. So, it may be possible that Sporleder &amp; Lapata do not provide conclusive results.</Paragraph>
    <Paragraph position="5"> Genzel (2005) considers lexical and syntactic features and reports accuracy obtained from English fiction data as well as from the WSJ corpus.</Paragraph>
    <Paragraph position="6"> He points out that lexical coherence and structural features turn out to be the most useful for his algorithm. Unfortunately, the only evaluation measure he provides is accuracy which, for the PBI task,  does not describe the performance of a system sufficiently. null In comparison to the mentioned studies, our goal is to examine the influence of cohesive features on the choice of paragraph boundary insertion. Unlike Bolshakov &amp; Gelbukh (2001), who have similar motivation but measure cohesion by collocations, we explore the role of discourse cues, pronominalization and information structure.</Paragraph>
    <Paragraph position="7"> The task of topic segmentation is closely related to the task of paragraph segmentation. If there is a topic boundary, it is very likely that it coincides with a paragraph boundary. However, the reverse is not true and one topic can extend over several paragraphs. So, if determined reliably, topic boundaries could be used as high precision, low recall predictors for paragraph boundaries.</Paragraph>
    <Paragraph position="8"> Still, there is an important difference: While work on topic segmentation mainly depends on content words (Hearst, 1997) and relations between them which are computed using lexical chains (Galley et al., 2003), paragraph segmentation as a stylistic phenomenon may depend equally likely on function words. Hence, paragraph segmentation is a task which encompasses the traditional borders between content and style.</Paragraph>
  </Section>
class="xml-element"></Paper>