<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0302">
  <Title>GoDiS - An Accommodating Dialogue System</Title>
  <Section position="7" start_page="8" end_page="10" type="evalu">
    <SectionTitle>
5 Limitations
</SectionTitle>
    <Paragraph position="0"> GoDiS is a small-scale prototype and as such it suffers from the familiar drawbacks of many experimental systems: its lexicons and databases are very small, and the domain knowledge is limited. The dialogue plans cover only a small (but important) part of the kind of dialogues that can take place in the travel agency and autoroute domains. We hope to be able to extend both the number of domains and the scale of each domain in the future.</Paragraph>
    <Paragraph position="1"> Also, the focus in the development of GoDiS has been on the dialogue management, i.e. the dialogue move engine and the information state. Less effort has been put into the interpretation and generation modules, which currently use (a slightly extended version of) keyword spotting and partially canned text, respectively. We hope to replace these by plugging in existing interpretation and generation technology to GoDiS.</Paragraph>
    <Paragraph position="2"> For the tasks that GoDiS currently is able to handle, full-blown semantics is not needed. We use a very limited semantics where propositions are essentially feature-value pairs. As a consequence, GoDiS  is e.g. not able to handle dialogue with more than one referent; for this, the information state would have to be amended with a set of referents, and propositions would have to include referent information. This is an area where we hope to improve GoDiS in the near future.</Paragraph>
    <Paragraph position="3"> Speech recognition and synthesis is currently being added to GoDiS, but at the time of writing only written input and output is available.</Paragraph>
  </Section>
class="xml-element"></Paper>