<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0710">
  <Title>Reference Resolution over a Restricted Domain: References to Documents</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Challenges of Reference Resolution over
a Restricted Domain
</SectionTitle>
    <Paragraph position="0"> From a cognitive point of view, the role of referring expressions in discourse is to specify the entities about which the speaker talks. It has long been observed that a more accurate view is that REs rather specify representations of entities in the speaker's or hearer's mind, an abstraction called discourse entities or DEs (Sidner, 1983; Grosz et al., 1995).</Paragraph>
    <Paragraph position="1"> Reference resolution can be defined as the construction of the discourse entities specified by referring expressions, or rather, the construction of computational representations of DEs. This difficult but important task in discourse understanding by computers appears to be more tractable when enough knowledge about a domain is available to a system (Gaizauskas and Humphreys, 1997), or when the representations are considerably simplified (Popescu-Belis et al., 1998).</Paragraph>
    <Paragraph position="2"> The coreference and anaphoric links, that is, links between REs only, are somewhat different aspects of the phenomenon of reference (Devitt and Sterelny, 1999; Lycan, 2000). Coreference is the relation between two REs that specify the same DE.</Paragraph>
    <Paragraph position="3"> Anaphora is a relation between two REs, called antecedent RE and anaphoric RE, where the DE specified by the latter is determined by knowledge of the DE specified by the former. In other terms, the DE specified by the anaphoric RE cannot be fully determined without knowledge of the antecedent RE.</Paragraph>
    <Paragraph position="4"> Depending on how the referent of the second RE is determined by the referent of the first one, the two REs may be coreferent, as in example (1) below, or they can be related by other referring relations, e.g.</Paragraph>
    <Paragraph position="5"> whole/part, function/value, etc., as in (2).</Paragraph>
    <Paragraph position="6">  1. The first articlei is particularly relevant to our company. Iti discusses . . .</Paragraph>
    <Paragraph position="7"> 2. The first articlei is particularly relevant to our  company. The titlej suggests that we. . .</Paragraph>
    <Paragraph position="8"> In the present case, reference resolution over a restricted domain differs significantly both from anaphora resolution (Mitkov, 2002) and from coreference resolution (Hirschman, 1997; van Deemter and Kibble, 2000). The REs available in the dialog transcript must be matched against the set of potential referents or DEs, which can be derived from the document structure. Therefore a computational representation of the referents is here available to serve as DEs. This advantage results directly from our present research goal and could be later extended to DEs derived computationally from document content, such as the persons mentioned in an article. Reference resolution in a restricted domain presents similarities with problems in natural language generation (NLG) and in command dialogs, that is, when the sets of referents are known a priori to the system. In NLG, the problem is to generate REs from existing computational descriptions of entities--see Paraboni and van Deemter (2002) for an application to intra-document references. In command dialogs, the problem is to match the REs produced by the user against the objects managed by the interface, again known formally to the system (Huls et al., 1995; Skantze, 2002).</Paragraph>
  </Section>
class="xml-element"></Paper>