<?xml version="1.0" standalone="yes"?>
<Paper uid="A88-1006">
  <Title>FROM WATER TO WINE: GENERATING NATURAL LANGUAGE TEXT FROM TODAY'S APPLICATIONS PROGRAMS 1</Title>
  <Section position="9" start_page="47" end_page="47" type="concl">
    <SectionTitle>
CONCLUSION
</SectionTitle>
    <Paragraph position="0"> The specification language has been completely implemented and used in-house since the fall of 1986.</Paragraph>
    <Paragraph position="1"> The templates and the specifics of how objects and expressions in underlying applications and planners are to be linked to Mumble-86 have evolved over that time and may continue to evolve somewhat as we get more experience with other applications. Mumble-86 itself is currently being used both for applications and as a research tool at a variety of sites including the University of Massachusetts, BBN Labs, RADC, and University of Pennsylvania.</Paragraph>
    <Paragraph position="2"> As part of an excercise in learning how to use Mumble-86, two researchers from RADC, Sharon Walter and Doug White, recently extended the program to generate in KRS's domain. It took them only two days to learn the specification language, build input specifications, and make the necessary grammatical and lexical extensions to generate several sentences in their domain, including those shown in Figure 7. Neither had used Mumble-86 before.</Paragraph>
    <Paragraph position="3"> In conclusion we would like to emphasize two main points. The first is the importance of modularity in design and portability of the modules so that research can concentrate on new and hard problems without having to waste effort reinventing the wheel. Mumble-86 has been developed to be just such a portable module. It has the responsibility for all syntactic decisions without making presumptions about the semantic model of the application program that uses it.</Paragraph>
    <Paragraph position="4"> Our second point is that a designer should not compromise the integrity of a well developed module to accomodate one which is less well developed when the two are brought together in the same system.</Paragraph>
    <Paragraph position="5"> This is the purpose of the input specification language we have introduced in this paper. In developing this language, we have clarified what decisions have to be made outside Mumble-86 and which decisions are its responsibility, thus circumscribing its sphere of influence and making it more useful as a domain independent linguistic component and as a tool for research in text planning and discourse structure.</Paragraph>
  </Section>
class="xml-element"></Paper>