<?xml version="1.0" standalone="yes"?>
<Paper uid="J03-2002">
  <Title>c(c) 2003 Association for Computational Linguistics Implementing the Binding and Accommodation Theory for Anaphora Resolution and Presupposition Projection</Title>
  <Section position="8" start_page="208" end_page="209" type="concl">
    <SectionTitle>
6. Conclusion
</SectionTitle>
    <Paragraph position="0"> The implementation of BAT presented in this article covers a wide spectrum of referential expressions, ranging from simple pronouns to rich presupposition triggers.</Paragraph>
    <Paragraph position="1"> Compared to Van der Sandt's (1992) original formulation of BAT, the implementation discussed here offers improvements on both the representational and inferential levels. Representational aspects of my reformulation of BAT include a new syntax for unresolved DRSs, which allows for selective binding and moreover provides a means of defining free and bound variables on these structures and hence enables us to implement the free-variable constraint on partially resolved DRSs. The inferential aspects involve a formulation of the acceptability constraints concerning consistency and informativeness, in particular the local versions of these, and an implementation of these constraints using general-purpose theorem provers for first-order logic.</Paragraph>
    <Paragraph position="2"> The core of the algorithm is realized as a PROLOG agent within the OAA environment. Off-the-shelf first-order theorem provers are used as agents to perform the required reasoning tasks. Results of tests conducted on a corpus of spoken route instructions suggest that the core of the algorithm performs reasonably well, certainly when one considers the enormous search space that is involved in presupposition resolution. The importance of reducing the search space is demonstrated in an experiment in which the free-variable constraint, imposed on partially resolved representations, shows a substantial increase in performance compared to when it is imposed on completely resolved representations. Furthermore, to distinguish among the different  Bos Implementing Binding and Accommodation Theory anaphoric natures of various presuppositional expressions, a classification of a-types for presuppositions triggered by noun phrases is proposed and used to increase efficiency in the implementation of the resolution algorithm.</Paragraph>
    <Paragraph position="3"> Although the algorithm presented in this article already includes a number of heuristics to state preferences among a number of candidates of resolved discourse representations, it could further benefit from suggestions made in centering theory (Grosz, Joshi, and Weinstein 1995) to express salience among discourse referents. In the terminology of centering, discourse referents can naturally be seen as centers, with forward-looking centers being a subset of the accessible discourse referents and the unique backward-looking center being a designated discourse referent resembling the focus of discourse. These aspects of centering theory will enable us to include a notion of coherence in BAT, but how to establish the ranking of forward-looking centers (to reflect relative prominence) in a DRT-based formalism remains a subject of future research.</Paragraph>
  </Section>
class="xml-element"></Paper>