<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1802">
  <Title>Conceptual Structuring through Term Variations</Title>
  <Section position="4" start_page="0" end_page="343" type="evalu">
    <SectionTitle>
5 Results and Evaluation
</SectionTitle>
    <Paragraph position="0"> We apply our program on a technical corpus in the field of agriculture which consists of 2,702 scientific abstracts for a total of 427,482 tokens and an average size of a record of 316 tokens.</Paragraph>
    <Paragraph position="1"> Table 3 gives the results of the collecting phase and Table 4 shows the percentages of the different types of variations for candidate terms appearing at least two times and the number of synonymic conflations.</Paragraph>
    <Paragraph position="2"> This conflating has a linguistic precision of 99 %.</Paragraph>
    <Paragraph position="3"> Number of occurrences 1 a0 2 Total base structures  matically identified conceptual relationships. Concerning morphological links, we note that two non-standard functions that have not been presented yet obtain a consequent representativity: MICRO induces from the suffixe micro(-): MICRO(film perfor'e &amp;quot;perforated film&amp;quot;) = film micro-perfor'e &amp;quot;pinhole film&amp;quot;; INTER infers from the suffixe inter expressing a reprocivity relationship: INTER('echelle nationale &amp;quot;national scale&amp;quot;) = 'echelle internationale &amp;quot;international scale&amp;quot;. The average precision of the morphological links is 95 %. The wrong links are 75 % due to the prefixes re, r'e refering to the function AGAIN. Examples of false drop are: action/r'eaction &amp;quot;reaction&amp;quot; in several candidate base-terms: action/r'eaction enzimatique &amp;quot;enzimatic action/reaction&amp;quot;, action/r'eaction acide &amp;quot;acide action/reaction&amp;quot;, a0a1a0a1a0 , production/reproduction, solution/r'esolution &amp;quot;resolution&amp;quot;, etc.</Paragraph>
  </Section>
class="xml-element"></Paper>