<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1626">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Distributed Language Modeling for N-best List Re-ranking</Title>
  <Section position="8" start_page="221" end_page="221" type="concl">
    <SectionTitle>
7 Conclusion and future work
</SectionTitle>
    <Paragraph position="0"> In this paper, we presented a novel distributed language modeling solution. The distributed LM is capable of using an arbitrarily large corpus to estimate the n-gram probability for arbitrarily long histories. We applied the distributed language model to N-best re-ranking and improved the translation quality by 4.8% when evaluated by the BLEU metric. The distributed LM provides a flexible architecture for relevance selection, which makes it possible to select data for each individual test sentence. Our experiments have shown that relevant data has better discriminative power than using all the data.</Paragraph>
    <Paragraph position="1"> We will investigate different relevance weighting schemes to better combine n-gram statistics from different data sources. We are planning to integrate the distributed LM in the statistical machine translation decoder in the near future.</Paragraph>
  </Section>
class="xml-element"></Paper>