<?xml version="1.0" standalone="yes"?>
<Paper uid="E91-1049">
  <Title>A PREFERENCE MECHANISM BASED ON MULTIPLE CRITERIA RESOLUTION</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
CONCLUSION
</SectionTitle>
    <Paragraph position="0"> It seems to us that two basic tendencies can be observed in the literature with respect to the treatment of preference. On the one hand, preference is conceived of as an essentially lingutstic or psycholinguistic principle or sum of principles (of. the LFG approach m \[Ford et at. 1982\]); although it has important consequences for the parser, preference ts not directly connected to a specific parsing method, on the other nano, preference has been studied in the context of parsing: in such Ireatments (el. \[Pereira 1985\]), preference amounts to a deterministic procedure, which is not necessarily motivated by linguistic evidence. In our approach preference is established on the basis of rules defined by the user and applied by a post-processor. We have in fact focussed on a method to express linguistically meaningful, preference statements rather than on a particular parsing strategy. . we are aware of the fact that, in a system where parsing is seen as a constraint satisfaction problem, preference criteria of the type we are interested in can be treated on the same level as other linguistic constraints and used to resolve ambiguity at parse time (of. \[Van Henteryck 1989\]). However, such an approach would have meant too radical a change to the underlying Eurotra formalism.</Paragraph>
    <Paragraph position="1"> In accordance with the general practice in Eurotra, our preference mechanism does not plead allegiance to any specific linguistic theory. We have, however, been influenced by a theoretical framework, namely the theory of preference rule systems described in \[Jackendoff 1985\].</Paragraph>
    <Paragraph position="2"> According to this framework, preference can only been decided on the basis of a number of criteria, and a preference mechanism is not based on a dichotomy between correct and wrong results, but on a scale of degrees of acceptability. One of_our main concerns m designing the system, in tact, has been allowing various and even contradictory criteria to be combined in a declarative fashion.</Paragraph>
    <Paragraph position="3"> The use of scoring is in this sense crucial.</Paragraph>
    <Paragraph position="4"> The system has been implemented and successfully tested on real input which showed overgeneration due to PP and adverbial attachment, coordination, pronominal resolution and lexical ambiguity. Some testing results are given in the appendix.</Paragraph>
  </Section>
class="xml-element"></Paper>