<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0811">
  <Title>Models for Inuktitut-English Word Alignment</Title>
  <Section position="7" start_page="81" end_page="81" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have presented several individual and combined systems for word alignment of Inuktitut-English bitext. The most successful individual systems were those targeted to the speci c characteristics of the language pair. The combined systems generally outperformed the individual systems, and different combination methods were able to optimize for performance under different evaluation metrics. In particular, per-English-word voting performed well on F-measure, while per-link voting performed well on AER.</Paragraph>
    <Paragraph position="1"> Acknowledgements: Many thanks to Eric Goldlust, David Smith, and Noah Smith for help in using the Dyna language.</Paragraph>
  </Section>
class="xml-element"></Paper>