<?xml version="1.0" standalone="yes"?>
<Paper uid="J04-2001">
  <Title>c(c) 2004 Association for Computational Linguistics Inferable Centers, Centering Transitions, and the Notion of Coherence</Title>
  <Section position="2" start_page="0" end_page="120" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> Centering has been proposed as a model of the local attentional states of speakers and hearers involved in the mutual construction of conversation (Brennan, Friedman, and Pollard 1987; Grosz and Sidner 1986, 1998; Walker 1998). Centering mechanisms are designed to model the coherence of discourse by characterizing transitions between utterances in terms of their inferential load and hence their naturalness. These characterizations are intended to capture intuitions about the &amp;quot;flow&amp;quot; (Chafe 1979) or the &amp;quot;ongoing process of meaning&amp;quot; (Halliday 1994) in discourse.</Paragraph>
    <Paragraph position="1"> In this work, we examine a corpus of Japanese e-mail to investigate the mechanisms by which coherence is achieved. Because this corpus contains a high number of discourse elements that are inferable from the discourse context, we have an opportunity to examine the interplay between standard centering transition definitions and the presence of inferable discourse entities. We claim on the basis of intuitions of native speakers that the actual level of coherence in the corpus is much higher than the centering account implies, primarily by virtue of the fact that transitions involving inferable entities are often difficult to specify. We conclude that the standard centering account cannot accurately model the coherence in this corpus. Detailed analysis reveals that one major problem lies in the requirement of identity of discourse elements in adjacent utterances in order for those elements to contribute to coherence. We describe this problem and propose two additions to the usual repertoire of transitions that enable a more authentic account of coherence in this corpus, while remaining within a centering framework.</Paragraph>
    <Paragraph position="2"> The article is organized as follows. In Section 2, we briefly describe centering mechanisms and their role in modeling coherence. We go on to outline the features of the corpus in Section 3 and illustrate how standard centering mechanisms characterize transitions and coherence in this corpus, suggesting that these mechanisms are not adequate for the task. In Section 4, we describe more general problems with the [?] Infant Studies Centre, University of British Columbia, Room 1401, 2136 West Mall, Vancouver, British Columbia, V6T 1Z4 Canada. E-mail: jwlab@psych.ubc.ca.</Paragraph>
    <Paragraph position="3">  Computational Linguistics Volume 30, Number 2 inclusion of inferable discourse entities in centering theory and propose a revision to the standard set of transitions that more accurately describes the corpus. In this section as well, we explore the implications of this proposal for other areas of discourse analysis. In Section 5, we outline some possibilities for improvement and extension of the proposal, and we conclude in the final section.</Paragraph>
  </Section>
class="xml-element"></Paper>