<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-1418">
  <Title>Optimising text quality in generation from relational databases</Title>
  <Section position="7" start_page="138" end_page="138" type="concl">
    <SectionTitle>
6 Summary
</SectionTitle>
    <Paragraph position="0"> While observing people trying to convert an earlier ILEX system to a new domain, we noted the difficulty they had. To avoid these problems, we undertook to re-implement the domain specification aspects of ILEX to simplify the task.</Paragraph>
    <Paragraph position="1"> Towards this end, we have followed a number of steps. Firstly, we reconstructed ILEX to be domain  declarative resource files. This means that domain developers do not have to deal with code.</Paragraph>
    <Paragraph position="2"> Secondly, we built into ILEX the ability to import entity definitions directly from a relational database (although with some restrictions as to its form).</Paragraph>
    <Paragraph position="3"> A database by itself does not provide enough information to produce text. Domain semantics is required. We have provided a system of incremental specification of this semantics which allows a domain developer to hook up adynamic hypertext interface to a relational database quickly, although producing poor quality text. Minimally, the system requires a domain taxonomy, information on lexification of types, and specification of the data type of each record field.</Paragraph>
    <Paragraph position="4"> Additional effort can then improve the quality of text up to a quite reasonable level. The additional information can include: specification of predicate expression, and specifications supporting comparisons, user adaption, and generalisations.</Paragraph>
    <Paragraph position="5"> Figure 13 summarises the obligatory and optional steps in domain specification in ILEX.</Paragraph>
    <Paragraph position="6"> Simplifying the domain specification task is a necessity as text generation systems move outside of research labs and into the real world, where the domain developer may not be a computational linguist, but a museum curator, personnel officer or wine salesman. ~ have tried to take a step towards making their task easier.</Paragraph>
  </Section>
class="xml-element"></Paper>