<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1154">
  <Title>Robust Sub-Sentential Alignment of Phrase-Structure Trees</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and future work
</SectionTitle>
    <Paragraph position="0"> We have presented an automatic algorithm which aligns bilingual context-free phrase-structure trees at sub-structural level and applied this algorithm to a subset of the English-French section of the HomeCentre corpus. We have outlined detailed evaluations of our algorithm. They show that while translation coverage was 10% lower using the automatically aligned data, the quality of the translations produced is comparable to the quality of those produced using manual alignments. While DOT systems produce very high quality translations in reasonable time, resource acquisition remains an issue. Manual sub-structural alignment is time-consuming, error-prone and requires considerable linguistic expertise. Our alignment method, on the other hand, is efficient, consistent and languageindependent, constituting a viable alternative to manual sub-structural alignment; thus solving the data acquisition problem.</Paragraph>
    <Paragraph position="1"> We intend to apply our automatic alignment methodology to the full English-French section of the HomeCentre corpus, as well as the English-German and French-German sections, and perform experiments to further validate the the language-independent nature of both our alignment algorithm and the data-oriented approach to translation. We also plan to automatically parse existing bitexts, thus creating further resources for use with our DOT system and, together with our aligner, enabling much larger-scale DOT-based translation experiments than have been performed to date.</Paragraph>
  </Section>
class="xml-element"></Paper>