<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1503">
  <Title>The Parallel Grammar Project</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Large-scale grammar development platforms are expensive and time consuming to produce. As such, a desideratum for the platforms is a broad utilization scope. A grammar development platform should be able to be used to write grammars for a wide variety of languages and a broad range of purposes. In this paper, we report on the Parallel Grammar (ParGram) project (Butt et al., 1999) which uses the XLE parser and grammar development platform (Maxwell and Kaplan, 1993) for six languages: English, French, German, Japanese, Norwegian, and Urdu. All of the grammars use the Lexical-Functional Grammar (LFG) formalism which produces c(onstituent)structures (trees) and f(unctional)-structures (AVMs) as the syntactic analysis.</Paragraph>
    <Paragraph position="1"> LFG assumes a version of Chomsky's Universal Grammar hypothesis, namely that all languages are structured by similar underlying principles. Within LFG, f-structures are meant to encode a language universal level of analysis, allowing for cross-linguistic parallelism at this level of abstraction. Although the construction of c-structures is governed 1We would like to thank Emily Bender, Mary Dalrymple, and Ron Kaplan for help with this paper. In addition, we would like to acknowledge the other grammar writers in the ParGram project, both current: Stefanie Dipper, Jean-Philippe Marcotte, Tomoko Ohkuma, and Victoria Ros'en; and past: Caroline Brun, Christian Fortmann, Anette Frank, Jonas Kuhn, Veronica Lux, Yukiko Morimoto, Mar'ia-Eugenia Ni~no, and Fr'ed'erique Segond.</Paragraph>
    <Paragraph position="2"> by general wellformedness principles, this level of analysis encodes language particular differences in linear word order, surface morphological vs. syntactic structures, and constituency.</Paragraph>
    <Paragraph position="3"> The ParGram project aims to test the LFG formalism for its universality and coverage limitations and to see how far parallelism can be maintained across languages. Where possible, the analyses produced by the grammars for similar constructions in each language are parallel. This has the computational advantage that the grammars can be used in similar applications and that machine translation (Frank, 1999) can be simplified.</Paragraph>
    <Paragraph position="4"> The results of the project to date are encouraging.</Paragraph>
    <Paragraph position="5"> Despite differences between the languages involved and the aims and backgrounds of the project groups, the ParGram grammars achieve a high level of parallelism. This parallelism applies to the syntactic analyses produced, as well as to grammar development itself: the sharing of templates and feature declarations, the utilization of common techniques, and the transfer of knowledge and technology from one grammar to another. The ability to bundle grammar writing techniques, such as templates, into transferable technology means that new grammars can be bootstrapped in a relatively short amount of time.</Paragraph>
    <Paragraph position="6"> There are a number of other large-scale grammar projects in existence which we mention briefly here. The LS-GRAM project (Schmidt et al., 1996), funded by the EU-Commission under LRE (Linguistic Research and Engineering), was concerned with the development of grammatical resources for nine European languages: Danish, Dutch, English, French, German, Greek, Italian, Portuguese, and Spanish. The project started in January 1994 and ended in July 1996. Development of grammatical resources was carried out in the framework of the Advanced Language Engineering Platform (ALEP).</Paragraph>
    <Paragraph position="7"> The coverage of the grammars implemented in LS-GRAM was, however, much smaller than the coverage of the English (Riezler et al., 2002) or German grammar in ParGram. An effort which is closer in spirit to ParGram is the implemention of grammar development platforms for HPSG. In the Verbmobil project (Wahlster, 2000), HPSG grammars for English, German, and Japanese were developed on two platforms: LKB (Copestake, 2002) and PAGE.</Paragraph>
    <Paragraph position="8"> The PAGE system, developed and maintained in the Language Technology Lab of the German National Research Center on Artificial Intelligence DFKI GmbH, is an advanced NLP core engine that facilitates the development of grammatical and lexical resources, building on typed feature logics. To evaluate the HPSG platforms and to compare their merits with those of XLE and the ParGram projects, one would have to organize a special workshop, particularly as the HPSG grammars in Verbmobil were written for spoken language, characterized by short utterances, whereas the LFG grammars were developed for parsing technical manuals and/or newspaper texts. There are some indications that the German and English grammars in ParGram exceed the HPSG grammars in coverage (see (Crysmann et al., 2002) on the German HPSG grammar).</Paragraph>
    <Paragraph position="9"> This paper is organized as follows. We first provide a history of the project. Then, we discuss how parallelism is maintained in the project. Finally, we provide a summary and discussion.</Paragraph>
  </Section>
class="xml-element"></Paper>