<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0903">
  <Title>Less is More: Using a Single Knowledge Representation in Dialogue Systems</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The ways in which knowledge has been represented in multi-modal dialogue systems (MMDS) show that individual representations with different semantics and heterogeneously structured content can be found in various formats within single natural language processing (NLP) systems and applications. For example, a typical NLP system, such as TRAINS (Allen et al., 1996), employs different knowledge representations for parsing, action planning and generation, despite the fact that what is being represented is common to all those representations, e. g., the parser representation for going from A to B has no similarity to the action planner's representation thereof (Ferguson et al., 1996). Also central concepts, for example city, are represented in multiple ways throughout the system.</Paragraph>
    <Paragraph position="1"> The origin for this state of affairs is that the respective knowledge stores are hand-crafted individually for each task. Sometimes they are compiled into code and cease to be externally available. Where an explicit knowledge representation is used, we find a multitude of formats and inference engines, which often cause both performance and tractability problems. In this paper we introduce the results of an effort to employ a single knowledge representation, i. e., an ontology, throughout a complete multi-modal dialogue system. Therefore, we will describe the underlying modeling principles and the benefits of such a rigorously crafted knowledge store for the actual and future MMDS.</Paragraph>
    <Paragraph position="2"> In Section 2 we will introduce the representational formats pertinent to our ontology, followed by a description of our dialogue system in Section 3. In Section 4 we discuss the modeling principles underlying the ontology.</Paragraph>
    <Paragraph position="3"> Section 5 presents some examples of the various ways in which the common ontology is employed throughout the system. Concluding remarks are given in Section 6.</Paragraph>
  </Section>
class="xml-element"></Paper>