<?xml version="1.0" standalone="yes"?>
<Paper uid="H05-1012">
  <Title>Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 89-96, Vancouver, October 2005. c(c)2005 Association for Computational Linguistics A Maximum Entropy Word Aligner for Arabic-English Machine Translation</Title>
  <Section position="12" start_page="94" end_page="95" type="concl">
    <SectionTitle>
11 Conclusion and Future Work
</SectionTitle>
    <Paragraph position="0"> This paper presented a word aligner trained on annotated data. While the performance of the aligner is shown to be significantly better than other unsupervised algorithms, the utility of these alignments in machine translation is still an open subject although gains are shown in two of the test sets. Since features are extracted from a parallel corpus, most of the information relating to the specific sentence alignment is lost in the aggregation of features across sentences.</Paragraph>
    <Paragraph position="1"> Improvements in capturing sentence context could allow the machine translation system to use a rare but correct link appropriately.</Paragraph>
    <Paragraph position="2"> Another significant result is that a small amount (5K sentences) of word-aligned data is sufficient for this algorithm since a provision is made to handle  unknown words appropriately.</Paragraph>
  </Section>
class="xml-element"></Paper>