<?xml version="1.0" standalone="yes"?>
<Paper uid="W96-0508">
  <Title>On Lexical Aggregation and Ordering</Title>
  <Section position="4" start_page="30" end_page="31" type="concl">
    <SectionTitle>
4. Conclusions
</SectionTitle>
    <Paragraph position="0"> While Unbounded and Bounded lexical aggregation are related to one another, UL-aggregation operates over an open set and loses some aggregated information, and BL-aggregation operates over a closed set and the aggregated information is retrievable. To select an appropriate UL-aggregation one may employ a hearer model. In both types of lexical aggregation one must check that the aggregands follow each other consecutively in time.</Paragraph>
    <Paragraph position="1"> From the permutation experiment for obtaining the optimally aggregated text we conclude that one should not always select  the shortest text, but the one with the best discourse organization (which we model by the best RST structure). We found certain optimal orderings of text plan clauses before aggregation, each ordering associated with aggregation rule. Regarding the order of applying the aggregation rules, we propose first to use the most powerful rule (namely UL aggregation), then the predicate and direct object (PDO) grouping rule, then the subject and predicate (SP) grouping rule, and finally BL aggregation.</Paragraph>
    <Paragraph position="2"> This paper is an extract of a longer work described in \[Dalianis96b\]. A great deal more work is required on the various aspects of lexical aggregation. Thoroughly studies of text corpora are necessary, as well as more fine-grained definitions of the various phenomena of lexical aggregation. The implementation of the finding will also be conditional upon the specific choices of knowledge representation system and inference support. This study is just a beginning of lot of exciting research !</Paragraph>
  </Section>
class="xml-element"></Paper>