<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0823">
  <Title>Statistical Machine Translation of Euparl Data by using Bilingual N-grams</Title>
  <Section position="6" start_page="135" end_page="135" type="concl">
    <SectionTitle>
5 Conclusions and Further Work
</SectionTitle>
    <Paragraph position="0"> As can be concluded from the presented results, performance of the translation system used is much better for French and Spanish than for German and Finnish. As some results suggest, reordering and vocabulary size are the most important problems related to the low translation quality achieved for German and Finnish.</Paragraph>
    <Paragraph position="1"> It is also evident that the bilingual n-gram model used requires the additional feature functions to produce better translations. However, more experimentation is required in order to fully understand each individual feature's influence on the overall log-linear model performance.</Paragraph>
  </Section>
class="xml-element"></Paper>