In the coming months we’ll begin rolling out the Pretranslation feature in Pontoon.
Pretranslation is the process of using translation memory and machine translation systems to translate content before it is edited by translators. It is intended to speed up the translation process and ease the work of translators.
How it works
Pretranslation will be off by default and only enabled for a selected list of locales within selected projects by the L10n Program Managers.
When enabled, each string without any translations will be automatically pretranslated using the 100% match from translation memory or (should that not be available) using the machine translation engine. If there are multiple 100% matches in TM, the one which has been used the most times will be used.
Pretranslations will be assigned a new translation status – Pretranslated. That will allow translators to distinguish them from community submitted translations and suggestions, and make them easier to review and postedit.
The important bit is that pretranslations will be stored to version control systems immediately, which means the postediting step will take place after translations might have already shipped in the product.
Machine translation engines
We have evaluated several different options and came to conclusion that the following scenario works best for our use case:
- If your locale is supported by the Google AutoML Translation, we’ll use that service and train the engine using the Mozilla translation data sets, which will result in better translation quality than what’s currently suggested in the Machinery tab by generic MT engines.
- For other locales we’ll use Google Translation API or similar engines.
We are in the early stages of the feature rollout. We’re looking for teams that would like to test the feature within a pilot project. If you’re a locale manager and want to opt in, please send an email to firstname.lastname@example.org and we’ll add your locale to our list of early adopters.