<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-1048">
  <Title>hnproving Search Strategies An Experiment in BesbFirst Parsing Hans ItAUGENEI)ER</Title>
  <Section position="3" start_page="238" end_page="239" type="intro">
    <SectionTitle>
5. Results
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="238" end_page="238" type="sub_section">
      <SectionTitle>
5.1 Scenario of the Test
</SectionTitle>
      <Paragraph position="0"> For the interpretation of the presented results it seems necessary to specify the experimental conditions under which the tests have been performed. The grammar we have been using covers the tbllowing subset of' English: declarative sentences, imperative sentences, questions (direct and indirect y/n-questions, direct and indirect wh-questions for NPs, PPs, APs), sentential complements for verbs and nouns, complete and reduced relative clauses, infinitive complements, clausal conjunction, and subordinate clauses.</Paragraph>
      <Paragraph position="1"> The test sample consisted of a set of 40 sentences and phrases that range from very simple phrases like &amp;quot;the man&amp;quot; to more complex constructions like &amp;quot;John gives the girl Bill admires a book which he does not expect her to read&amp;quot;. The medium sentence length of the sample is 6.5. The homographic ambiguity factor is 1.3, i.e. each word processed is 1.3 times ambiguous on the average.</Paragraph>
    </Section>
    <Section position="2" start_page="238" end_page="239" type="sub_section">
      <SectionTitle>
5.2 Discussion of the Results
</SectionTitle>
      <Paragraph position="0"> When processing tile test sample under the various strategies, it turned out that there were many strategies that showed approximately the same overall behaviour, i.e. demanded almost the identical search effort.</Paragraph>
      <Paragraph position="1"> Especially the variation of the weight for the single factors in general only shows effects when one contrasts the extreme values for the weighting factor (i.e. 0 and 5).The quantitive measures of some selected heuristic functions that have been used in a one-path analysis mode is shown in figure (IV).</Paragraph>
      <Paragraph position="2">  The strategies we used are defined as explicated in (IiI), where AE means overall weight of the incoming active edge, GR weight of the grammar arc, SIE span of the inactive edge, SP span of the active edge to be continued with the inactive edge, IL items left. Besides the heuristic criteria, another important impact on the parsing strategy is the method o.r insertion of tasks into the agenda. It can take place in a local or in a global mode. While in the latter case (SortAll insertion mode) a .general and costly reordering of all tasks in the agenda is performed, in the first case (SortNewToFront insertion mode) only the ordered set of newly generated tasks is put onto the agenda in a stack-like fashion.</Paragraph>
      <Paragraph position="3"> insertion mode heuristic function stratl</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>