<?xml version="1.0" standalone="yes"?>
<Paper uid="C82-1035">
  <Title>ARBUS, A TOOL FOR DEVELOPING APPLICATION GRAMMARS</Title>
  <Section position="10" start_page="221" end_page="221" type="concl">
    <SectionTitle>
CONCLUSION
</SectionTitle>
    <Paragraph position="0"> Th e ARBUS system is thus a useful, pleasant and practical tool for the development of grarmnars. A first version was implemented in PL/I on IBM 370/168; ARBUS was then completely rewritten in INTERLISP/370, a language better suited to the manipulation of symbolic structures. Both versions are operational, but the PL/I version is directly compatible with speech processing programs written in the same language, while grarmmars built in INTERLISP are available through files.</Paragraph>
    <Paragraph position="1"> 226 D. MEMMI and J. MARIANI We have used ARBUS to develop application grammars for speech recognition and to experiment with dialog grammars in man-machine communication. For instance it took less than half an hour ~o define the syntax of a spoken command language for piloting planes by voice, with about I00 words and 250 different states. This grammar was then successfully used in speech recognition. In other similar experiments we have found ARBUS pleasant to use and quite helpful as a development aid.</Paragraph>
    <Paragraph position="2"> But it should be mentioned that this system is more appropriate for application grammars of a limited size. The deliberate choice of a tree representation for syntax and of interactive construction would make it tedious to define very big grammars in this way. For a huge syntax it would be quicker to enter it directly as a file of rules to be compiled, though the development of such a grammar would prove difficult anyway.</Paragraph>
    <Paragraph position="3"> ARBUS might indeed be modified so as to accept rewrite rules directly. Also one could describe grammars as transition networks rather than trees. But the system would become less interactive and more cumbersome to use, while ARBUS was designed to be as interactive and as easy to use as possible. Such changes would then go against the basic purpose of the system.</Paragraph>
    <Paragraph position="4"> Other extensions are more interesting to contemplate. When building the grammar the system could evaluate automatically the complexity of the language, according to some combination of criteria (size of the vocabulary, number of rules, branching factor, etc...). It would thus be possible to obtain meaningful comparisons between grammars to evaluate speech recognition or parsing systems. One might also better adapt ARBUS to tile description of man-machine dialogs by spucifying the respective roles of the user and the system in these dialogs.</Paragraph>
    <Paragraph position="5"> In short, ARBUS is a good example of an interactive development tool, specially designed from the start to ease the user's task. Such a system is thus part of the evolution towards human engineering and graceful interaction which is becoming more and more apparent in many areas of man-machine communication.</Paragraph>
  </Section>
class="xml-element"></Paper>