<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1103">
  <Title>Using a Probabilistic Translation Model for Cross-Language Information Retrieval</Title>
  <Section position="6" start_page="26" end_page="26" type="concl">
    <SectionTitle>
4. Conclusions
</SectionTitle>
    <Paragraph position="0"> MT systems are considered by many as appropriate tools for CLIR. In this paper, we showed that there are better tools for CLIR than MT. We investigated the possibility of using a probabilistic translation model built automatically from a parallel corpus. In comparison with MT, this approach is more flexible. It may be used for any pair of languages for which an appropriate parallel corpus is available.</Paragraph>
    <Paragraph position="1"> When applied to CLIR, MT systems (LOGOS and SYSTRAN) can give a relatively good performance. Simpler approaches based only on bilingual dictionaries or terminology databases like BTQ lead to much poorer performance. Our probabilistic translation model almost rivals the performance of the MT systems, despite the fact that our training corpus is not closely related to the test corpus. In our experiments, we observed different advantages and disadvantages for different approaches to translate queries from a language to another. They often have complementary properties, and may be successfully combined. In this study, we combined our probabilistic translation model with a bilingual dictionary. This combination outperformed the MT systems, leading us to the conclusion that there are better approaches to CLIR than MT.</Paragraph>
    <Paragraph position="2"> In all cases, the performance of CLIR remains substantially lower than that of monolingual IR. Thus there is still a lot of room for further improvement. There may not be any single translation method that will fill the bill. We believe that progress is likely to come from combining various sources of translation knowledge and we intend to continue testing such methods in our future research.</Paragraph>
  </Section>
class="xml-element"></Paper>