<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1105">
  <Title>LANGUAGE GENERATION FROM CONCEPTUAL STRUCTURE: SYNTHESIS OF GERMAN IN A JAPANESE/GERMAN MT PROJECT</Title>
  <Section position="8" start_page="493" end_page="493" type="concl">
    <SectionTitle>
EXPERIMENTS WITH OTHER GENERATION MODULES
</SectionTitle>
    <Paragraph position="0"> We recently studied three generation modules (running in Lisp on our SYMBOLICS 3600) with the objective to find out, whether they could serve as a generation front end for SEMSYN: SUTRA (Busemann, 1983), the German version of IPG (Kempen &amp; Hoenkamp, 1982), and MUMBLE (McDonald, 1983).</Paragraph>
    <Paragraph position="1"> Our IRS is a functional grammar description. The input of SUTRA, the &amp;quot;preterminal structure&amp;quot;, already makes assumptions about word order within the noun group. To use SUTRA, additional transformation rules would have to be written.</Paragraph>
    <Paragraph position="2"> IPG's input is a conceptual structure.</Paragraph>
    <Paragraph position="3"> Parts of it are fully realized before others are considered. The motivation for IPG's incremental control structure is psychological. In contrast, the derivation of our IRS and its subsequent rendering is not committed to such a control structure. Nevertheless, the procedural grarmnar of IPG could be used to produce surface strings from IKBS by providing it with additional syntactic features (which are contained in IRS).</Paragraph>
    <Paragraph position="4"> Both MUMBLE and IPG are conceptually oriented and incremental. MUMBLE's input is on the level of our IKBS. MUMBLE produces functional descriptions of sentences &amp;quot;on the fly&amp;quot;. These descriptions are contained in a constituent structure tree, which is traversed to produce surface text. Our approach is to make the functional description explicit.</Paragraph>
    <Paragraph position="5"> ACKNOWLEDG~4ENTS We have to thank many colleagues in the generation field that helped SEMSYN with their experience. We are especially thankful to Dave McDonald (Amherst), and Eduard Hoenkamp (Nijmegen) whose support - personally and through their software - is still going on. We also thank the members of the ATLAS/II research group (Fujitsu Laboratories) for their support.</Paragraph>
  </Section>
class="xml-element"></Paper>