<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1168">
  <Title>A Uni ed Approach in Speech-to-Speech Translation: Integrating Features of Speech Recognition and Machine Translation</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we presented our approach of incorporating both speech recognition and machine translation features into a log-linear speech translation model to improve speech translation.</Paragraph>
    <Paragraph position="1"> Under this new approach, translation performance was signi cantly improved. The performance improvement was con rmed by consistent experimental results and measured by using various objective translation metrics. In particular, BLEU score was improved by 7.9% absolute.</Paragraph>
    <Paragraph position="2"> We show that features derived from speech recognition: likelihood of acoustic and language models, helped improve speech translation. The N-best recognition hypotheses are better than the single-best ones when they are used in translation. We also show that N-best recognition hypothesis translation can improve speech recognition accuracy of incorrectly recognized sentences.</Paragraph>
    <Paragraph position="3"> The success of the experiments owes to the use of statistical machine translation and log-linear models so that various of e ective features can be jointed and balanced to output the optimal translation results.</Paragraph>
  </Section>
class="xml-element"></Paper>