<?xml version="1.0" standalone="yes"?>
<Paper uid="J03-1003">
  <Title>c(c) 2003 Association for Computational Linguistics Graph-Based Generation of Referring Expressions</Title>
  <Section position="2" start_page="0" end_page="54" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> The generation of referring expressions is one of the most common tasks in natural language generation and has been addressed by many researchers in the past two decades, including Appelt (1985), Reiter (1990), Dale and Haddock (1991), Dale (1992), Dale and Reiter (1995), Horacek (1997), Stone and Webber (1998), Krahmer and Theune (1998, 2002), Bateman (1999), and van Deemter (2000, 2002). In this article, we present a general, graph-theoretic approach to the generation of referring expressions. We propose to formalize a scene (i.e., a domain of objects and their properties and relations) as a labeled directed graph and describe the content selection problem (which properties and relations to include in a description for an object?) as a subgraph construction problem. The graph perspective has four main advantages. (1) There are many attractive and well-understood algorithms for dealing with graph structures (see, e.g., Gibbons [1985], Cormen, Leiserson, and Rivest [1990], or Chartrand and Oellermann [1993]). In this article, we describe a straightforward branch and bound algorithm for finding the relevant subgraphs in which cost functions are used to guide the search process. (2) By defining different cost functions for the graph perspective, we can simulate (and improve) some of the well-known algorithms for the generation of referring [?] Communication and Cognition/Computational Linguistics, Faculty of Arts, Tilburg University, Tilburg, The Netherlands. E-mail: E.J.Krahmer@uvt.nl.</Paragraph>
    <Paragraph position="1"> + Tijgerstraat 2, NL-5645 CK, Eindhoven, The Netherlands. E-mail: sebster@sebster.com.</Paragraph>
    <Paragraph position="2"> ++ Ranonkelstraat 67, NL-5644 LB, Eindhoven, The Netherlands. E-mail: andre@astygian.nl.</Paragraph>
    <Paragraph position="3">  Computational Linguistics Volume 29, Number 1 expressions mentioned above. This facilitates the formal comparison of these algorithms and makes it easier to transfer results from one algorithm to another. (3) The graph perspective provides a clean solution for some problems that have plagued earlier algorithms. For instance, the generation of relational expressions (i.e., referring expressions that include references to other objects) is enhanced by the fact that both properties and relations are formalized in the same way, namely, as edges in a graph.</Paragraph>
    <Paragraph position="4"> (4) The combined use of graphs and cost functions paves the way for a natural integration of traditional rule-based approaches to generating referring expressions and more recent statistical approaches, such as Langkilde and Knight (1998) and Malouf (2000), in a single algorithm.</Paragraph>
    <Paragraph position="5"> The outline of this article is as follows. In Section 2 the content selection problem for generating referring expressions is explained, and some well-known solutions to the problem are discussed. In Section 3, we describe how scenes can be modeled as labeled directed graphs and show how content selection can be formalized as a subgraph construction problem. Section 4 contains a sketch of the basic generation algorithm, which is illustrated with a worked example. In Section 5 various ways to formalize cost functions are discussed and compared. We end with some concluding remarks and a discussion of future research directions in Section 6.</Paragraph>
  </Section>
class="xml-element"></Paper>