<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-3010">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A Hybrid Relational Approach for WSD - First Results</Title>
  <Section position="8" start_page="58" end_page="59" type="concl">
    <SectionTitle>
5 Conclusion and future work
</SectionTitle>
    <Paragraph position="0"> We presented a hybrid relational approach for WSD designed for MT. One important characteristic of our approach is that all the KSs were sense(sent_id,translation).</Paragraph>
    <Paragraph position="1"> sense(sent1,voltar).</Paragraph>
    <Paragraph position="2"> sense(sent2,ir).</Paragraph>
    <Paragraph position="3"> :- modeh(1,sense(sent,translation)).</Paragraph>
    <Paragraph position="4"> :- modeb(11,has_colloc(sent,colloc_id,colloc)).  :- modeb(10,has_bag(sent,word)). ... 1. sense(A, sair) :has_collocation(A, preposition_right, out). 2. sense(A, chegar) :satisfy_restrictions(A, [animal,human],[concrete]); has_expression(A, 'come at').</Paragraph>
    <Paragraph position="5"> 3. sense(A, vir) : null satisfy_restriction(A, [human],[abstract]), has_collocation(A, word_right_1, from). set(evalfn, posonly): learns from positive examples. set(search, heuristic): turns the search strategy heuristic. set(minpos, 2): establishes as 2 the minimum number of positive examples covered by each rule in the theory. set(gsamplesize, 1000): defines the number of randomly generated negative examples to prune the search space.  automatically extracted, either from the corpus or machine-readable lexical resources. Therefore, the work could be easily extended to other words and languages.</Paragraph>
    <Paragraph position="6"> In future work we intend to carry out experiments with different settings: (a) combinations of certain KSs; (b) other sample corpora, of different sizes, genres / domains; and (c) different parameters in Aleph regarding search strategies, evaluation functions, etc. We also intend to compare our approach with other machine learning algorithms using all the KSs employed in Aleph, by pre-processing the KSs in order to extract binary features that can be represented by means of attribute-value vectors. After that, we intend to adapt our approach to evaluate it with standard WSD data sets, such as the ones used in Senseval2.</Paragraph>
  </Section>
class="xml-element"></Paper>