<?xml version="1.0" standalone="yes"?>
<Paper uid="C94-1075">
  <Title>A MODULAR ARCHITECTURE FOR CONSTRAINT-BASED PARSING</Title>
  <Section position="7" start_page="458" end_page="458" type="concl">
    <SectionTitle>
6 CONCLUSION
</SectionTitle>
    <Paragraph position="0"> APOC-II has several advantages. First of all, it provides comparable implementations of the most comnmn parsing Mgorithms. Their efficiency can be abstractly measured, for instance by counting the number of eomlmtation step (EPDA transition applicatiol 0 performed to eomlmte a tree or a complete forest of parse trees. We call this kind of measm'ements abstract \])ecallse it does not rely neither on the implementlttion nor on the machine that runs the parser. Other comparisons could be done statically, on the automaton or on the pstrse forest (e..g. number of transitions, alllOllllt ,)f determi~lisnl, size of the forest, alllOllllt of structure slurring).</Paragraph>
    <Paragraph position="1"> ()therwise, APOC-II cstn be. used as a to(~lkit that provides :t library of modules usefld to implelllent quickly ll(!W parse.r generators. For instance, one has only to write a solver to obtain up to 22 parsing a.lgorithms (perhaps less if tit(', solw!r provides only basic operations). The library contains tools to deal with some constraints, look-ahead, lexing, tabulation, etc. Reusing these tools whenever it is possible saves a lot of work.</Paragraph>
    <Paragraph position="2"> The limitations of APOC-II are that it is mainly convenient for parsing strategies that stre somehow static, i.e. statically determined at compih! time. Also, al)stractloll (full independence between coral&gt;tiers and i,~terpreters) cannot Im achieved for some optimized algorithms. For in-Sl,&amp;llCe, Nederhof presents in \[Ned93\] a parsing strategy called ELI{ for which tsdmlar execution can be optimized. To implement this a.lgorithm tit ollr system, one would have to write a Ilow interpreter dedicated to ELR-EPDAs.</Paragraph>
    <Paragraph position="3"> \Y=e think that our experiment shows the int(~rest of a tlexible modul;trity for studies abollt parsing. We believe that the same technique can fiuitfully apply on other domains of Ns~tural Language Processing.</Paragraph>
  </Section>
class="xml-element"></Paper>