<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0305">
  <Title>Annotating Attribution in the Penn Discourse TreeBank</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> News articles typically contain a mixture of information presented from several different perspectives, and often in complex ways. Writers may present information as known to them, or from some other individual's perspective, while further distinguishing between, for example, whether that perspective involves an assertion or a belief. Recent work has shown the importance of recognizing such perspectivization of information for several NLP applications, such as information extraction, summarization, question answering (Wiebe et al., 2004; Stoyanov et al., 2005; Riloff et al., 2005) and generation (Prasad et al., 2005). Part of the goal of such applications is to distinguish between factual and non-factual information, and to identify the source of the information. Annotation schemes (Wiebe et al., 2005; Wilson and Wiebe, 2005; PDTB-Group, 2006) encode such distinctions to facilitate accurate recognition and representation of such perspectivization of information.</Paragraph>
    <Paragraph position="1"> This paper describes an extended annotation scheme for marking the attribution of discourse relations and their arguments annotated in the Penn Discourse TreeBank (PDTB) (Miltsakaki et al., 2004; Prasad et al., 2004; Webber et al., 2005), the primary goal being to capture the source and degrees of factuality of abstract objects. The scheme captures four salient properties of attribution: (a) source, distinguishing between different types of agents to whom AOs are attributed, (b) type, reflecting the degree of factuality of the AO, (c) scopal polarity of attribution, indicating polarity reversals of attributed AOs due to surface negated attributions, and (d) determinacy of attribution, indicating the presence of contexts canceling the entailment of attribution. The scheme also describes annotation of the text spans signaling the attribution. The proposed scheme is an extension of the core scheme used for annotating attribution in the first release of the PDTB (Dinesh et al., 2005; PDTB-Group, 2006). Section 2 gives an overview of the PDTB, Section 3 presents the extended annotation scheme for attribution, and Section 4 presents the summary.</Paragraph>
  </Section>
class="xml-element"></Paper>