<?xml version="1.0" standalone="yes"?>
<Paper uid="H01-1033">
  <Title>Improved Cross-Language Retrieval using Backoff Translation</Title>
  <Section position="6" start_page="3" end_page="3" type="concl">
    <SectionTitle>
5. CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> There are many ways of combining evidence from multiple translation lexicons. We use tralexes similar to those usedby Nie et al. [4], but our work differs in our use of balanced translation and a back-off translation strategy (which produces a stronger baseline for our WebDict tralex), and in our comparisonof merging and backoff translation strategies for combining resources. In future work we plan to explore other combinations of merging and backoff and other merging strategies, including post-retrieval merging of the ranked lists.</Paragraph>
    <Paragraph position="1"> In addition, parallel corpora can be exploited for more than just the extraction of a non-contextualized translation lexicon. We are currently engagedin work on lexical selection methods that take advantage of contextual information, in the context of our research on machine translation, and we expect that CLIR results will be improved by contextually-informed scoring of term translations.</Paragraph>
  </Section>
class="xml-element"></Paper>