<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-1058">
  <Title>Constraints over Lambda-Structures in Semantic Underspecification</Title>
  <Section position="5" start_page="357" end_page="358" type="evalu">
    <SectionTitle>
4 Related Work
</SectionTitle>
    <Paragraph position="0"> CLLS allows a uniform and yet internally structured approach to semantic ambiguity. We use a single constraint formalism in which to describe different kinds of information about the meaning of an utterance. This avoids the problems of order dependence of processing that for example Shieber et al. (1996) get by interleaving two formalisms (for scope and for ellipsis resolution). Our approach follows Crouch (1995) in this respect, who also includes parallelism constraints in the form of substitution expressions directly into an underspecified semantic formalism (in his case the formalism of Quasi Logical Forms QLF). We believe that the two approaches are roughly equivalent empirically. But in contrast to CLLS, QLF is not formalised as a general constraint language over tree-like representations of meaning. QLF has the advantage of giving a more direct handle on meanings themselves - at the price of its relatively complicated model theoretic semantics.</Paragraph>
    <Paragraph position="1"> It seems harder though to come up with solutions within QLF that have an easy portability across different semantic frameworks.</Paragraph>
    <Paragraph position="2"> We believe that the ideas from CLLS tie in quite easily with various other semantic formalisms, such as UDRT (Reyle, 1993) and MRS (Copestake et al., 1997), which use dominance relations similar to ours, and also with theories of Logical Form associated with GB style grammars, such as (May, 1977). In all these frameworks one tends to use variable-coordination (or coindexing) rather than the explicit binding and linking relations we have presented here. We hope that these approaches can potentially benefit from the presented idea of rubber bands for binding and linking, without having to make any dramatic changes.</Paragraph>
    <Paragraph position="3"> Our definition of parallelism implements some ideas from Hobbs and Kehler (1997) on the behavior of anaphoric links. In contrast to their proposal, our definition of parallelism is not based on an abstract notion of similarity. Furthermore, CLLS is not integrated into a general theory of abduction. We pursue a more modest aim at this stage, as CLLS needs to be connected to &amp;quot;material&amp;quot; deduction calculi for reasoning with such underspecified semantic representation in order to make progress on this front. We hope that some of the more ad hoc features of our definition of parallelism (e.g. axiom 5) may receive a justification or improvement in the light of such a deeper understanding. null Context Unification. CLLS extends the expressiveness of context unification (CU) (Niehren et al., 1997a), but it leads to a more direct and more structured encoding of semantic constraints than CU could offer. There are three main differences between CU and CLLS.</Paragraph>
    <Paragraph position="4"> 1) In CLLS variables are interpreted over nodes rather than whole trees. This gives us a direct handle on occurrences of semantic material, where CU could handle occurrences only indirectly and less efficiently. 2) CLLS avoids the capturing problem. 3) CLLS provides explicit anaphoric links, which could not be adequately modeled in CU.</Paragraph>
    <Paragraph position="5"> The insights of the CU-analysis in (Niehren et al., 1997b) carry over to CLLS, but the awkward second-order equations for expressing dominance in CU can be omitted (Niehren and Koller, 1998). This omission yields an enormous simplification and efficiency gain for processing. Tractability. The distinguishing feature of our approach is that we aim to develop efficiently treatable constraint languages rather than to apply maximally general but intractable formalisms. We are confident that CLLS can be implemented in a simple and efficient manner.</Paragraph>
    <Paragraph position="6"> First experiments which are based on high-level concurrent constraint programming have shown promising results.</Paragraph>
  </Section>
class="xml-element"></Paper>