<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1026">
  <Title>A Relational Syntax-Semantics Interface Based on Dependency Grammar</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper, we have shown how to build a fully relational syntax-semantics interface based on XDG.</Paragraph>
    <Paragraph position="1"> This new grammar formalism offers the grammar developer the possibility to represent different kinds of linguistic information on separate dimensions that can be represented as graphs. Any two dimensions can be linked by multi-dimensional principles, which mutually constrain the graphs on the two dimensions. We have shown that a parser based on concurrent constraint programming is capable of inferences that restrict ambiguity on one dimension based on newly available information on another.</Paragraph>
    <Paragraph position="2"> Because the interface we have presented makes no assumption that any dimension is more &amp;quot;basic&amp;quot; than another, there is no conceptual difference between parsing and generation. If the input is the surface sentence, the solver will use this information to compute the semantic dimensions; if the input is the semantics, the solver will compute the syntactic dimensions, and therefore a surface sentence. This means that we get bidirectional grammars for free.</Paragraph>
    <Paragraph position="3"> While the solver is reasonably efficient for many (hand-crafted) grammars, it is an important goal for the future to ensure that it can handle large-scale grammars imported from e.g. XTAG (XTAG Research Group, 2001) or induced from treebanks.</Paragraph>
    <Paragraph position="4"> One way in which we hope to achieve this is to identify fragments of XDG with provably polynomial parsing algorithms, and which contain most useful grammars. Such grammars would probably have to specify word orders that are not completely free, and we would have to control the combinatorics of the different dimensions (Maxwell and Kaplan, 1993). One interesting question is also whether different dimensions can be compiled into a single dimension, which might improve efficiency in some cases, and also sidestep the monostratal vs. multi-stratal distinction.</Paragraph>
    <Paragraph position="5"> The crucial ingredient of XDG that make relational syntax-semantics processing possible are the declaratively specified principles. So far, we have only given some examples for principle specifications; while they could all be written as Horn clauses, we have not committed to any particular representation formalism. The development of such a representation formalism will of course be extremely important once we have experimented with more powerful grammars and have a stable intuition about what principles are needed.</Paragraph>
    <Paragraph position="6"> At that point, it would also be highly interesting to define a (logic) formalism that generalises both XDG and dominance constraints, a fragment of CLLS. Such a formalism would make it possible to take over the interface presented here, but use dominance constraints directly on the semantics dimensions, rather than via the encoding into PA and SC dimensions. The extraction process of Section 4.2 could then be recast as a principle.</Paragraph>
  </Section>
class="xml-element"></Paper>