<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1013">
  <Title>From Words to Corpora: Recognizing Translation</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> I have proposed a language-independent approach to the detection of translational equivalence in texts of any size that works at various bilingual resource levels. Fast, e ective approximations have also been described, suggesting scalability to very large corpora. Notably, tsim is adaptable to any probabilistic model of translational equivalence, because it is an instance of a model-independent de nition of similarity.</Paragraph>
    <Paragraph position="1"> The core of the technique is the computation of optimal matchings at two levels: between words, to generate the tsim score, and between texts, to detect translation pairs.</Paragraph>
    <Paragraph position="2"> I have demonstrated the performance of this technique on English-Chinese and English-French.13 It is capable of pulling parallel texts out of a large multilingual collection, and it rivals the performance of structure-based approaches to pair classi cation (Resnik, 1999), having better agreement with human judges.</Paragraph>
  </Section>
class="xml-element"></Paper>