<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1064">
  <Title>Aligning words using matrix factorisation</Title>
  <Section position="9" start_page="0" end_page="0" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper, we view word alignment as 1/ estimating the association between source and target words, and 2/ factorising the resulting association measure into orthogonal, non-negative factors. For solving the latter problem, we propose an algorithm for ONMF, which guarantees both proper alignments and good coverage. Experiments carried out on the Hansard give encouraging results, in the sense that we improve in several ways over state-of-the-art results, despite a clear bias in the reference alignments. Further investigations are required to apply this technique on different association measures, and to measure the influence that ONMF may have, eg on a phrase-based Machine Translation system.</Paragraph>
  </Section>
class="xml-element"></Paper>