File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/concl/89/e89-1032_concl.xml
Size: 2,035 bytes
Last Modified: 2025-10-06 13:56:22
<?xml version="1.0" standalone="yes"?> <Paper uid="E89-1032"> <Title>AN ALGORITHM FOR GENERATION IN UNIFICATION CATEGORIAL GRAMMAR</Title> <Section position="5" start_page="0" end_page="0" type="concl"> <SectionTitle> 7. Implementation </SectionTitle> <Paragraph position="0"> The algorithm discussed in this paper has been implemented in C-Prolog. Recent work has looked at generation from semantic representations which are not in canonical format but which are equivalent, under the axioms of associativity and commutativity to the canonical semantics of sentences recognised by the grammar. Our effort is directed at formulating appropriate notions of &quot;semicanonicality&quot;, which lessen the strict (and in many cases unobtainable) requirement that the representation to be generated from is identical to that obtained as the result of parsing. Such notions would increase the utility of generators such as we have presented while avoiding the dangers of factorial complexity.</Paragraph> <Paragraph position="1"> A further source of inefficiency is the naive lexical indexing strategy used by the predicate lexical. We have presented the algorithm as if iexical simply enumerates the lexicon. This is clearly inefficient and some form of indexing strategy seems essential. The simplest is to choose the principal functor of the semantic representation to use as the index for lexical entries which have the same principal functor in their semantics. Much of the time however, the principal functor is simply the conjunction operator. A more sophisticated indexing strategy involves calculating the best (set of) key(s) to identify candidate lexical entries. This necessarily involves considerable complexity itself.</Paragraph> <Paragraph position="2"> Furthermore, if such indexing is to be automatic, very sophisticated compilation techniques and metaknowledge about the possible structure of semantic representations are required. We are also investigating these possibilities.</Paragraph> </Section> class="xml-element"></Paper>