<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2051">
  <Title>Bridging the Inflection Morphology Gap for Arabic Statistical Machine Translation</Title>
  <Section position="8" start_page="202" end_page="202" type="concl">
    <SectionTitle>
5 Conclusion and Future Work
</SectionTitle>
    <Paragraph position="0"> This work shows the potential for significant improvements in machine translation quality by directly bridging the inflectional gap across language pairs. Our method takes advantage of source and  The IWSLT evaluation did not allow systems to train separately for evaluation on BLEU or NIST, but results from the proceedings indicate that top performers in each metric optimized towards the respective metric.</Paragraph>
    <Paragraph position="1"> target language context when conducting morphological analysis of each surface word form, while avoiding complex parsing engines or refinements to the alignment training process. Our results are presented on moderately sized corpora rather than the scarce resource domain that has been traditionally employed to highlight the impact of detailed morphological analysis.</Paragraph>
    <Paragraph position="2"> By showing the impact of simple processing steps we encourage the creation of simple word and gloss level analysis tools for new languages and show that small investments in this direction (compared to high octane context sensitive parsing tools) can yield dramatic improvements, especially when rapid development of machine translation tools becomes increasingly relevant to the research community.</Paragraph>
    <Paragraph position="3"> While our work focused on processing the morphologically rich language and then translating &amp;quot;down&amp;quot; into the morphologically poor language, we plan to use the analysis tools developed here to model the reverse translation process as well, the harder task of translating &amp;quot;up&amp;quot; into a highly inflected space.</Paragraph>
  </Section>
class="xml-element"></Paper>