<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2029">
  <Title>Constraining Tree Adjoining Grammars by Unification</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
i1 Introduction
</SectionTitle>
    <Paragraph position="0"> iln the field of natural language analysis, Unification Grammars are a main research topic. Presently, Unification is defined as extension of context-flee grmnm~s.</Paragraph>
    <Paragraph position="1"> Knowing th, e formalism of Tree Adjoining Grammars (in the following called TAGs in short), which is closely related to context-free gr,'tmmars (in the I611owing abbreviated CFG), the idea arises to replace the context-free grammar in a Unification Grammar by ;t TAG. The advantage of TAGs is that complete context-fl'ee derivation trees or parts of them build the rqles of that grammar type (e.=.,, with the intention of ~epresenting a whole linguistic phenomenon). The ~ecursion operation for TAGs allows the replacement of nodes by a tree (defined by a TAG-rule), so that larger t;tructure u'ees are processed.</Paragraph>
    <Paragraph position="2"> In the literature, a first definition for combining these two formalisms was proposed, where the main idea is Io sepwate the two recursion processes - adjoining and unification - to preserve all properties of both Here a different approach is chosen, where both recursion processes are integrated. The main point to emphasize here is that the different approaches not only represent a .';witch between two modes of interpreting the same definition, but a change in the properties of the resulting formalism.</Paragraph>
    <Paragraph position="3"> This can be sirnply demonstrated by the property of monotonicity of the unification. Associated with each lree of a TAG all spC/ification rules for the unification are interpreted at once (e.g., represented as links between the DAG representation of the specification rules at each node in the tree). If now the recursion process of TAGs, the adjoining operation, combines two trees, which both have DAGs, a strategy for reinterpretation of specification information must be defined, because an adjoining modifies inner nodes, where links still are installed.</Paragraph>
    <Paragraph position="4"> la the following, the two formalisms are briefly revisited to have a common terminological basis with rite reader, before the existing definition of Tree Adjoining Grammars with Unification is presented. In contrast to this approach, the new definition is motivated and its properties are discussed. Finally, our experience with an implementation of that definition in the application domain of syntax analysis (e.g., in a natural language dialogue system) is summarized.</Paragraph>
  </Section>
class="xml-element"></Paper>