<?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="5" start_page="0" end_page="0" type="concl">
    <SectionTitle>
4. Conclusion
</SectionTitle>
    <Paragraph position="0"> We have presented what is to our knowledge the first formalization and implementation of a type of rule and control regime intended for use in situations where it is desired to produce the effect of transforming one feature structure into another. 9 The formalism described above has been implemented as part of ISSCO's ELU ldeg, an enhanced PATR-II style (Shieber, 1986) unification grammar environment, based on the UD system presented by Johnson and Rosner (1989). ELU incorporates a parser and generatot, and is primarily intended for use as a tool for research in machine translation. Use of transfer rules in translation has not so far brought to light instances where the serial rule invocation regime described in section 3.2 proves necessary. ELU grammars permit the use of typed feature structures (cf. Johnson and Rosner, op. cit., Moens et al., 1989) in grammars; although the present transfer rule format does not, they are clearly a desirable addition, since they would provide a means of exerting control over rule interactions.</Paragraph>
    <Paragraph position="1"> A third area in which the transfer rule forrealism might be applied concerns the manipulation of re-entrant structures. While re-entrancy is in general a useful property of FSs, the complexity entailed by its presence is in some cases unwelcome; the method of genera: 9 Van Noord (1990) describes the use of a standard unification grammar to successively instantiate a single feature structure embodying meaning representations for both source and target language expressions in a machine translation application. Similarly, the transfer rules of Zajac (1990) express a relation between subparts of a single complex structure. Such an approach does not appear suitable for the appl/cation discussed in section 3.2 above.</Paragraph>
    <Paragraph position="2">  tion proposed by Wedekind (1988), for example, requires that the LFG-style f-structures which form the input to the generation process be 'unfolded' into unordered trees. This may be done with a suitably formulated rule set of the kind introduced here. The present rule format is unable to preserve the information that distinct sub-FSs in a destination FS arise from the duplication of a single, re-entrant, sub-FS in the source. Ways of incorporating this ability into the rule formalism are under consideration, one possibility being the addition of an indexing mechanism that would flag sub-FSs as originating in a re-entrancy.</Paragraph>
    <Paragraph position="3"> A companion paper describes an interpretation of transfer rule sets in terms of a partial ordering with respect to the specificity of rules, and discusses linguistic and computational motivations for this view; it also comments in greater detail on the rule interaction problems referred to in fn. 3, and on issues of termination, completeness and coherence in transfer. Here, we simply note that, in the current implementation, it is possible to declare to the system the path set of a source FS that is to be subject to transfer, so as to provide rim-time notification ff inadequacies in the rule set result in a specified sub-FS being neglected.</Paragraph>
    <Paragraph position="4"> With respect to a given rule set and source FS, however, correctness of the transfer process is Assured.</Paragraph>
  </Section>
class="xml-element"></Paper>