<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0207">
  <Title>LoLo: A System based on Terminology for Multilingual Extraction</Title>
  <Section position="9" start_page="63" end_page="63" type="concl">
    <SectionTitle>
5 Afterword
</SectionTitle>
    <Paragraph position="0"> The patterns extracted from the English (and Arabic) corpora confirm to an extent the view of the proponents of local grammar, of a special language, that there are certain words (in our case percent, shares, index) that appear to have a specific grammatical category in the sense that the neighbourhood of these words is occupied by a small number of other words (up, down, fall, rise, &lt;no&gt; for instance). If we were to apply the grammars typically used in part-of-speech taggers and syntactic parsing in general, the idiosyncratic behaviour of the pivotal keywords in specialist language does not become apparent: the pivotal keywords are regarded as noun phrases and the association of these phrases is with other general categories of verb phrase, adjectival phrase and adverbial phrase.</Paragraph>
    <Paragraph position="1"> The patterns we have extracted could have been extracted with the help of a thesaurus. And, this is the question which is critical to us: how to create and maintain a thesaurus within a domain.</Paragraph>
    <Paragraph position="2"> This is illustrated in a small way by our experiment on the Training1 corpus where the term index was not statistically significant for it to appear in the trigrams that populate the local grammar. However, in Training2, the larger corpus did contain significant frequency of the term index for it to make into a pattern of its own.</Paragraph>
    <Paragraph position="3"> Furthermore, many of the patterns in Training1 persisted in Training2. Smaller N-grams persist as well in the various Training and Test corpora - these patterns in themselves act like units around which other trigrams nucleate.</Paragraph>
    <Paragraph position="4"> The evaluation of our Algorithm is still continuing and we are in the process of setting up experiments with human volunteers, especially those with some knowledge of financial matters to evaluate the output of LoLo. We intend to use information retrieval metrics of recall and the various Fb measures.</Paragraph>
    <Paragraph position="5"> The local grammar movement has made erratic progress since its inception in the 1960's.</Paragraph>
    <Paragraph position="6"> Now, with the advent of accessible computers with substantive memories, with the advent of the Internet and the concomitant treasure of multi-lingual text deposits and text streams, one can explore the use of such grammars in addressing the major challenges in information extraction. null</Paragraph>
  </Section>
class="xml-element"></Paper>