Translate
I wonder whether anybody ever tried to use MT and something like BLEU to improve alignment results?
I envision a source and a target text (already translated, to be aligned), then Source gets machine translated, so we have three texts: SourceHT, TargetHT, TargetMT
- Calculate BLEU scores for all TargetHT and TargetMT segments. (The relationship from SourceHT and TargetMT (as in matching segments) is known, of course.)
- Align TargetHT to SourceHT based on BLEU scores.
This would be especially useful if you could use your trained custom MT or an MT with glossary.
Daniel
