Post-editing English to Japanese statistical machine translation : effort and productivity A comparative analysis of three different Moses configurations
Stéphanie Di Rosa · Archive ouverte UNIGE (University of Geneva) · 2012
This thesis explores the correlations between statistical machine translation quality and post-editing activity of the English to Japanese language pair. Building on some previous research done at Autodesk Development Sàrl., and using a qualitative post-editing (PE) analysis method based on logical edits, we conducted a separate short research to determine the effort involved to post-edit machine translation (MT) stemming from different Moses configurations. We specially wanted to see whether it was establish correlations between MT quality, PE effort and productivity with the designed metric. Although, due to data scarcity, not all results are conclusive, this preliminary study has shows that carrying out a qualitative, logical-edits-based analysis of post-editing is informative as to areas of potential improvement in the raw machine translation output and post-editing activity in general. We suggested future work, in particular to determine the differences with post-editing analyses conducted with automated metrics.