<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-0802">
  <Title>A Two-stage Model for Content Determination</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> This paper addresses the problem of content determination in data summarisation. Content determination as the name indicates is the process responsible for determining the content of the texts generated by an NLG system (Reiter and Dale 2000). Although content-determination is probably the most important part of an NLG system from the end-user's perspective, there is little agreement in the NLG community as to how content-determination should be done, with different systems adapting widely varying approaches. Also, algorithms and architectures for content-determination seem to often be based on the intuitions of system developers, instead of on empirical observations, although detailed content determination rules are often based on corpus analysis and interaction with experts.</Paragraph>
    <Paragraph position="1"> In this paper we propose a general architecture for content determination in data summarisation systems which assumes that content determination happens in two stages: first a qualitative overview of the data is formed, and second the content of the actual summaries is decided upon. This model is based on extensive knowledge acquisition (KA) activies that we have carried out in the SUMTIME project (Sripada, 2001), and also matches observations made during KA activities carried out in the STOP project (Reiter et al 2000). We have not yet implemented this model, and indeed one of the issues that we need to think about is to what degree a content-determination strategy used by human experts is also an appropriate one for a computer NLG system.</Paragraph>
  </Section>
class="xml-element"></Paper>