Understanding KantanBuildAnalytics Scores
This video will show you how to analyse the scores from your
engine.
You should always consult the automatic scores to determine if the engine is suitable to proceed to the next step. The following scores are good indicators of a high quality, production-ready engine:
F-Measure is an automated calculation of recall and precision. The more words the engine correctly selects, the better the vocabulary of your engine.
Aim high, a score of around 60% usually indicates a production ready engine.
BLEU is an automated calculation of fluency. It measures phrase selection capabilities of your KantanMT engine.
Aim high, generally a score of around 50% usually indicates a production ready engine
TER is an automated calculation of post-editing effort. It estimates the amount of post-editing required to bring a generated translation up to high native fluency.
Aim low, generally a score of around 40% or lower usually indicates a production ready engine.
If you encounter any problems or need help, contact us at [email protected]