<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2207">
  <Title>Identifying correspondences between words: an approach based on a bilingual syntactic analysis of French/English parallel corpora</Title>
  <Section position="4" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 State of the art
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.1 Term alignment
</SectionTitle>
      <Paragraph position="0"> Two kinds of methods have been basically proposed in order to address the problem of bilingual lexicon extraction. On the one hand, terms are recognized in both source and target language and then they are mapped to each other (Daille, Gaussier and Lange, 1994). On the other hand, only source terms are extracted and the target ones are discovered through the alignment process (Gaussier, 1998; Hull, 2001). The alignment between terms is obtained either by computing association probabilities (Gaussier, 1998 ; Daille, Gaussier and Lange, 1994) or by identifying, for a given source term, a sequence of words in the target language which is likely to contain or to correspond to its translation (Hull, 2001). In so far as the precision rate may be affected by the number of alignments obtained (Daille, Gaussier and Lange, 1994; Gaussier, 1998), the results achieved basically range between 80% and 90%, for the first 500 alignments. As for the method described in (Hull, 2001), the precison reported is 56%.</Paragraph>
      <Paragraph position="1"> It should be noticed that the use of linguistic knowledge is most of the time restricted to the term recognition stage. This kind of knowledge is quite rarely taken into account within the very alignment process, except for the approach implemented by Daille, Gaussier and Lange (1994), which try to take advantage of correspondences between the syntactic patterns defined for each language.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.2 Word alignment
</SectionTitle>
      <Paragraph position="0"> Quite recently attempts have been made in order to incorporate different types of linguistic information sources into word alignment systems and to combine them with statistical knowledge.</Paragraph>
      <Paragraph position="1"> Various and more or less complex sources of linguistic knowledge are exploited: morphological, lexical (Arhenberg, Andersson and Merkel, 2000) and syntactic knowledge (Wu, 2000; Lin and Cherry, 2003). The contribution of these information sources to the alignment process with respect to the statistical data varies according to the considered system. However, as pointed out by Arhenberg, Andersson and Merkel (2000) as well as Lin and Cherry (2003), the introduction of linguistic knowledge leads to a significant improvement in alignment quality. In the first case, the accuracy goes from 91% for a baseline configuration up to 96.7% for a linguistic knowledge based one. In the second, the precision rate is increased from 82.7% up to 89.2% and the improvement noticed have been confirmed within the framework of an evaluation task (Mihalcea and Pedersen, 20003).</Paragraph>
      <Paragraph position="2"> For our part, we propose a method in which the syntactic information plays a major role in the alignment process, since syntactic relations are used to find out new correspondences between words or to confirm the existent ones. We chose this approach in order to achieve a high accuracy alignment both at word and phrase level. Indeed, we aim at capturing frequent alignments between words and phrases as well as those involving sparse or corpus specific ones. Moreover, as stressed in previous works, using syntactic dependencies seems to be particularly well suited to solve n-to-1 or n-to-m alignments (Fluhr, Bisson and Elkateb, 2000) and to cope with the problem of linguistic variation and non correspondence across languages, for instance when aligning terms (Gaussier, 2001).</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>