This software generates word-level/phrase-level QE tags for a given translated text and its post-edited version. Following the WMT shared task's notation, correct words/phrases are tagged as OK, and errors are tagged as BAD.
The phrase-level tagger makes use of mgiza.
QET.py -t translation_file -pe post-edited_file {-p source_file mgiza_path}
Where:
- translation_file is the translated text.
- post-edited_file is the post-edited version of the translated text.
- -p switches to phrase-level generation.
- source_file is the original text.
- mgiza_path is the path to mgiza's bin folder (e.g., /opt/mgiza/mgizapp/bin/).
Source: El profundo mar azul .
Translation: The deep sea blue .
Post-edited: The deep blue sea .
Tags: OK OK BAD OK OK
Source: El profundo mar azul .
Translation: The deep sea blue .
Post-edited: The deep blue sea .
Tags: OK BAD BAD OK OK