<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0806">
  <Title>Augmenting a Small Parallel Text with Morpho-syntactic Language Resources for Serbian-English Statistical Machine Translation</Title>
  <Section position="6" start_page="46" end_page="46" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this work, we have examined the possibilities for building a statistical machine translation system with a small bilingual Serbian-English parallel text.</Paragraph>
    <Paragraph position="1"> Our experiments showed that the translation results for this language pair are comparable with results for other language pairs, especially if the small size of the corpus, unrestricted domain and rich inflectional morphology of Serbian language are taken into account. With the baseline system, we obtained about 45% WER for translation into English and about 53% for translation into Serbian.</Paragraph>
    <Paragraph position="2"> We have systematically investigated the impact of the corpus size on translation quality, as well as the importance of additional bilingual knowledge in the form of short phrases. In addition, we have shown that morpho-syntactic information is a valuable language resource for translation of this language pair. Depending on the availability of resources and tools, we plan to examine parallel texts with other languages, and also to do further investigations on this language pair. We believe that more refined use of the morpho-syntactic information can yield better results (for example the hierarchical lexicon model proposed in (Niessen and Ney, 2001)). We also believe that the use of the conventional dictionaries could improve the Serbian-English translation.</Paragraph>
  </Section>
class="xml-element"></Paper>