<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1503">
  <Title>The Parallel Grammar Project</Title>
  <Section position="13" start_page="0" end_page="0" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> The experiences of the ParGram grammar writers has shown that the parallelism of analysis and implementation in the ParGram project aids further grammar development efforts. Many of the basic decisions about analyses and formalism have already been made in the project. Thus, the grammar writer for a new language can use existing technology to bootstrap a grammar for the new language and can parse equivalent constructions in the existing languages to see how to analyze a construction. This allows the grammar writer to focus on more difficult constructions not yet encountered in the existing grammars.</Paragraph>
    <Paragraph position="1"> Consider first the Japanese grammar which was started in the beginning of 2001. At the initial stage, the work of grammar development involved implementing the basic constructions already analyzed in the other grammars. It was found that the grammar writing techniques and guidelines to maintain parallelism shared in the ParGram project could be efficiently applied to the Japanese grammar. During the next stage, LFG rules needed for grammatical issues specific to Japanese have been gradually incorporated, and at the same time, the biannual ParGram meetings have helped significantly to keep the grammars parallel. Given this system, in a year and a half, using two grammar writers, the Japanese grammar has attained coverage of 99% for 500 sentences of a copier manual and 95% for 10,000 sentences of an eCRM (Voice-of-Customer) corpus.</Paragraph>
    <Paragraph position="2"> Next consider the Norwegian grammar which joined the ParGram group in 1999 and also emphasized slightly different goals from the other groups. Rather than prioritizing large textual coverage from the outset, the Norwegian group gave priority to the development of a core grammar covering all major construction types in a principled way based on the proposals in (Bresnan, 2001) and the inclusion of a semantic projection in addition to the f-structure. In addition, time was spent on improving existing lexical resources ( 80,000 lemmas) and adapting them to the XLE format. Roughly two man-years has been spent on the grammar itself. The ParGram cooperation on parallelism has ensured that the derived f-structures are interesting in a multilingual context, and the grammar will now serve as a basis for grammar development in other closely related Scandinavian languages.</Paragraph>
    <Paragraph position="3"> Thus, the ParGram project has shown that it is possible to use a single grammar development platform and a unified methodology of grammar writing to develop large-scale grammars for typologically different languages. The grammars' analyses show a large degree of parallelism, despite being developed at different sites. This is achieved by intensive meetings twice a year. The parallelism can be exploited in applications using the grammars: the fewer the differences, the simpler a multilingual application can be (see (Frank, 1999) on a machine-translation prototype using ParGram).</Paragraph>
  </Section>
class="xml-element"></Paper>