<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-2128">
  <Title>A Uniform Architecture for Parsing and Generation</Title>
  <Section position="8" start_page="617" end_page="617" type="evalu">
    <SectionTitle>
6 Precursors
</SectionTitle>
    <Paragraph position="0"> Perhaps the clearest espousal of the notion of grammar reversability was made by Kay \[1975\], whose research into functional grammar has been motivated by the desire to &amp;quot;make it possible to generate and analyze sentences with the same grammar.&amp;quot; Other researchers have also put such ideas into effect. Jacobs's PHRED system \[Jacobs, 1985\] &amp;quot;operates from a declarative knowledge base of linguistic knowledge, common to that used by PHRAN&amp;quot;, an analyzer for so-called phrasal grammars. Jacobs notes that other systems ~ have shared at least part of the linguistic information for parsing and generation; for instance, the HAM-ANS \[Wahlster et al., 1983\]'a;nd VII~-LANG \[Steinacker and Buchberger, 1983\] systems utilize the same lexical information for both tasks. Kasper has used a system for parsing grammars in a unification-based formalism (SItI's Z-PATR system) to parse sentences with respect to the large ISI NIGEL grammar, which had been previously used for generation alone.</Paragraph>
    <Paragraph position="1"> Nonetheless, all of these systems rely on often radically different architectures for the two processes. Precedent for using a single architecture for both tasks is much more difficult to find. The germ of the idea can be found in the General Syntactic Processor (GSP) designed for the MIND system at Rand. Kaplan and K~y proposed use of the GSP for parsing with respect to augmented transition networks and generation by traiisformational grammars \[Kaplan, 1973\]. However, detailed implementation was apparently never carried out. In any case, although the PrOposal involved using the same arehitecture~ different formalisms (and hence grammars) were presupposed for the two tasks, ttunning a definite-clanse grammar (DCG) &amp;quot;backwards&amp;quot; has been proposed previously, although the normal Prolog execution mechanism renders such a technique unusable in practice. However, =.- alternative execution models might make the practice feasible. As mentioned above, the technique described here is just such an exe: cution model, and is directly related to the Earley deduction model of Pereira and Warren \[1983\].. Hasida and Isizaki \[1987\] present another method for generating and analyzing using a DCG-like formalism, which they call dependency propagation. The technique seems to entail using dataflow dependencies implicit in the grammar to control processing in a coroutining manner. The implementation status of their method and its practical utility are as yet unclear.</Paragraph>
    <Paragraph position="2"> The use of an agenda and scheduling schemes to allow varying the control structure of a parser also finds precedent in the work of Kaplan \[1973\] and Kay \[1967\]. Kay's &amp;quot;powerful parser&amp;quot; and the GSP both employed an agenda mechanism to control additions to the chart. However, the &amp;quot;tasks&amp;quot; placed on the agenda were at the same time more powerful (corresponding to unconstrained rewrite rules) and more procedural (allowing register operations and other procedural constructs).</Paragraph>
    <Paragraph position="3"> This work merely applies the notion in the context of the simple declarative formalisms presupposed, and provides it with a logical foundation on which a proof of correctness can be developed. TM Because the formalisms are simpler, the agenda need only keep track of one type of task: addition of a chart item.</Paragraph>
  </Section>
class="xml-element"></Paper>