<?xml version="1.0" standalone="yes"?>
<Paper uid="E91-1050">
  <Title>A Language for the Statement of Binary Relations over Feature Structures</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> Benefits arising from the adoption of unification as a tool in computational linguistics are well known: a declarative, monotonic method of combining partial information expressed in data structures convenient for linguistic applications permits the writing of sensible grammars that can be made independent from processing mechanisms, and a growing familiarity, in both theoretical and computational circles, with the techniques of unification fosters fruitful interchange of ideas and experiences. There are, however, occasions when unification alone is not an appropriate tool. In essence, unification is a ternary relation in which two structures, when merged, form a third; it is less attractive in circumstances where the relation to be expressed is binary - when one would like to manipulate a single feature structure (FS), perhaps simulating the direct transformation of one FS into another. 1 The present paper introduces a declarative formalism intended for the expression of such relations, and shows how it may be applied to some areas of current interest.</Paragraph>
    <Paragraph position="1"> The formalism in question is based upon a notion of 'transfer rule'; informally, a set of such rules may be considered as characterizing We are indebted to Jacques Jayez for comments on an earlier draft of this paper.</Paragraph>
    <Paragraph position="2"> 1 Clearly there is a sense in which such relations can be viewed as ternary: T(FI, R, F2), where 171 and 172 are * 17Ss, and R is the rule set which relates them.</Paragraph>
    <Paragraph position="3"> a binary relation over a set of feature structures, the properties of that relation depending on the content of the particular rule set in use.</Paragraph>
    <Paragraph position="4"> Transfer rules associate the analysis of one FS with the synthesis of another; they may be thought of as a specialized variety of pattern-matching rule. They are local in nature, and permit the recursive analysis and synthesis of complex structures according to patterns specified in a format closely related to that widely employed in unification-based computational linguistics. Indeed, the interpretation of transfer rules involves unification, albeit in a context which restricts it to the role of a structure-building operation. 2 In the remainder of this paper we provide a brief specification of the transfer rule formalism, discuss its interpretation, outline two alternative rule application regimes, and illustrate the use of the formalism in the areas of machine translation and reduction of FSs to canonical form. We conclude with an overview of continuing strands of research.</Paragraph>
    <Paragraph position="5">  2. Rule Format and Interpretation 2.1. General Remarks A transfer rule consists of four parts: (i) a role name; 3 (ii) a set of constraint equations describing a FS; (iii) a set of constraint equations describing a FS; 4 2 The rule formalism is thus monotonic, being unable  to effect changes in the input representation, and constmcting the output by means of unification.</Paragraph>
  </Section>
class="xml-element"></Paper>