<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-1017">
  <Title>A Constructive View of GPSG or How to Make It Work</Title>
  <Section position="6" start_page="80" end_page="80" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> It has been shown how our constructive version of GPSG avoids the: problem of combinatorial explosion that would have arisen if we had tried to implement the GPSG formalism in its axiomatic version \[GKPS\] in a straightforward way. Our A stem-form lexicon complemented with lemmatization and inflection procedures is better suited to NL processing anyway, at least if sa'ongly inflecting languages such as German are involved.</Paragraph>
    <Paragraph position="1"> change in perspective also leads to an impoitant simplification of the HFC because it is no longer necessary to build all the projections of an ID rule for the determination of the free feature specification sets.</Paragraph>
    <Paragraph position="2"> The dilemma over the ordering of the CAP and the HFC has been removed too, which is crucial for any implementation of the formalism. But, for this to be achieved, we had to sacrifice part of the generality that characterizes the treatment of control in \[GKPS\]; ke. although the qnestion of which constituents have to agree with one another is ~Lot answered ir~ a purely idiosyncratic way by the ID rules (because most of the cases can be accounted for by FCRs, which are, as it were, language-specific generalizations), the fact that agreement depends on functor-argument structures is no longer integrated iuto the formalism.</Paragraph>
    <Paragraph position="3"> This loss, however, is compensated for by the fact that we can treat agreement in cases which the original CAP could not account for (as in the case where a functor is cont,oiled by one of several arguments).</Paragraph>
    <Paragraph position="4"> Although we have had to concentrate our presentation on just a few aspects of the eonsu-uctive view of GPSG, we hope to have made plausible that our modified formalism is, in contrast to the original one, suitable for parsing and generation within an NL processing system.</Paragraph>
  </Section>
class="xml-element"></Paper>