<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0903">
  <Title>Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 17-24, Ann Arbor, June 2005. c(c)2005 Association for Computational Linguistics Preprocessing and Normalization for Automatic Evaluation of Machine Translation</Title>
  <Section position="11" start_page="23" end_page="23" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have described several MT evaluation measures.</Paragraph>
    <Paragraph position="1"> We have pointed out common preprocessing steps and auxiliary methods which have not been studied in detail so far in spite of their importance for the MT evaluation process. Particularly, we have introduced a novel method for determining the reference length of an evaluation candidate sentence, and a simple method to incorporate sentence boundary information to m-gram based evaluation measures.</Paragraph>
    <Paragraph position="2"> We then have performed several experiments on these methods on three evaluation corpora.</Paragraph>
    <Paragraph position="3"> The results indicate that both our new reference length algorithm and the use of sentence boundaries improve the correlation of the studied automatic evaluation measures with human evaluation. Furthermore, we have learned that case information should be removed when performing automatic sentence evaluation. On sentence level, evaluator normalization can improve the correlation between automatic and human evaluation.</Paragraph>
  </Section>
class="xml-element"></Paper>