Currently, we bubble up alignments from machine.py to Serval and store them in the pretranslations. We could easily add the alignment scores as well. This, like confidence, could give us insights into the quality of the drafts we are producing / flag incorrectly set-up draft runs. Of course, it's also possible that these scores could be used in other applications in the future, but the primary motivation for doing this sooner than later would be that this is data that we are already calculating, it's easy to store, and it could be valuable to us both now and retrospectively.
Currently, we bubble up alignments from machine.py to Serval and store them in the pretranslations. We could easily add the alignment scores as well. This, like confidence, could give us insights into the quality of the drafts we are producing / flag incorrectly set-up draft runs. Of course, it's also possible that these scores could be used in other applications in the future, but the primary motivation for doing this sooner than later would be that this is data that we are already calculating, it's easy to store, and it could be valuable to us both now and retrospectively.