<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1609">
  <Title>Statistical Machine Reordering</Title>
  <Section position="7" start_page="74" end_page="75" type="concl">
    <SectionTitle>
5 Conclusions and Further Research
</SectionTitle>
    <Paragraph position="0"> In this paper we have mainly dealt with the re-ordering problem for an n-gram-based SMT system. However, our approach could be used similarly for a phrase-based system. We have addressed the reordering problem as a translation from the source sentence to a monotonized source sentence. The proposed SMR system is applied before a standard SMT system. The SMR and SMT systems are based on the same principles and share the same type of decoder.</Paragraph>
    <Paragraph position="1"> In extracting bilingual units, the change of order performed in the source sentence has allowed the modeling of the translation units to be improved (shorter units mean a reduction in data sparseness). Also, note that the SMR approach allows the coherence between the change of order in the training and test source corpora to be maintained.</Paragraph>
    <Paragraph position="2">  Performing reordering as a preprocessing step and independently from the SMT system allows for a more ef cient nal system implementation and a quicker translation. Additionally, using word classes helps to infer unseen reorderings.</Paragraph>
    <Paragraph position="3"> These preliminary results show consistent and signi cant improvements in translation quality.</Paragraph>
    <Paragraph position="4"> As further research, we would like to add extra features to the SMR system, and study new types of classes for the reordering task.</Paragraph>
  </Section>
class="xml-element"></Paper>