Unbabel / OpenKiwi

Open-Source Machine Translation Quality Estimation in PyTorch
https://unbabel.github.io/OpenKiwi/
GNU Affero General Public License v3.0
229 stars 48 forks source link

Fix requirements for training sentence-level estimators #19

Closed captainvera closed 5 years ago

captainvera commented 5 years ago

This is a response to #18. The possibility of training an estimator solely for sentence-level hadn't been considered.

The estimator part of the predictor-estimator model erroneously required one of the predict-{target, source, gaps} flag to be active.

Furthermore the estimator always required train-target-tags flag to be passed. This doesn't make sense when training solely for sentence-level.