<?xml version="1.0" standalone="yes"?>
<Paper uid="C82-1035">
  <Title>ARBUS, A TOOL FOR DEVELOPING APPLICATION GRAMMARS</Title>
  <Section position="3" start_page="0" end_page="221" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Parsing is one of the main problems in natural language processing. It is generally recognized that understanding written text requires some kind of structural analysis, even if semantic comprehension would also be needed. In speech recognition research, syntactic constraints are frequently used to help acoustic recognition by reducing the number of possibilities to be dealt with.</Paragraph>
    <Paragraph position="1"> Grammatical analysis becomes even more important when one considers it to i ~lude not only syntax itself, but also any formal constraint. One can therefore d :ne semantic or pragmatic grammars, dialog grammars as well as phonetic or phonological rules. The formalism of syntactic rules is powerful enough to describe many areas of natural language be~ide syntax itself, and the use of gran~mar has been extended accordingly in many systems.</Paragraph>
    <Paragraph position="2"> But parsing is a difficult problem. The design of a parser involves fairly sophisticated programming techniques. And grammar rules are usually numerous, and their interaction may prove quite complex, so that it is not easy to define a grammar. Rules often have to be modified repeatedly for the development of the grammar, and will have to be modified again if'one wants to change the application domain.</Paragraph>
    <Paragraph position="3"> To avoid tinkering constantly with the program, rules should be kept separate from the control mechanism of the parser. The grammar is then considered as data for the parsing program, and the rules can be given in a clear declarative formulation. If this basic precaution is not observed, modifying the rules will require repeated and tiresome reprogramming, and at some point the program may become too complicated for any further extension of the grammar.</Paragraph>
    <Paragraph position="4"> It is also very interesting to help the user to develop his own grammar, by allowing him to define, test and modify the rules easily thanks to a specialized interactive interface. For example the LIFER system (Hendrix, \]977) was specially designed to help build application grammars without extensive prograrmning. It has been successfully used to build complex natural language front-ends such as the  222 D. MEMMI and J, MARIANI LADDER system (Sacerdoti, 1977; Hendrix and al., 1978) in order to access and query databases in natural English.</Paragraph>
    <Paragraph position="5"> Similarly we have designed the ARBUS system as an aid to the development of grammars. With ARBUS one can build, display, test and modify a grammar interactively in a very convenient way. The user never has to deal directly with the underlying programs and no programming is necessary. This was achieved by packaging a parser and a granmmr editor with an elaborate interface, which shields the user from implementation details and guides him as much as possible.</Paragraph>
  </Section>
class="xml-element"></Paper>