It’s important to note that these are not strict criteria: members of staff will evaluate each request to opt in individually, based on their knowledge of the project and direct experience with the locale.
Criteria for enabling pretranslation for a new locale
- Request needs to come from translators or managers active within the last month (translating or reviewing).
- There is an active manager for the locale (last activity within 2 months).
Criteria for enabling pretranslation for a new project
- Less than 400 missing strings, except for projects or locales where existing pretranslation statistics provide high-confidence.
- Average review time for pretranslations in existing projects is faster than 3 weeks.
Criteria for disabling the feature for a locale or a project
- Approval rate drops below 40%.
- Average review time for pretranslations is slower than 6 weeks.
Note that disabling a project would always involve a conversation with reviewers for the locale.
Access Pontoon’s admin console, and select the project: at the bottom of the page there is a section dedicated to Pretranslation.
IMPORTANT: if this is the first project for a locale, the first step is to train and set up the custom machine translation model in Google AutoML Translation.
Use the checkbox
PRETRANSLATION ENABLED to enable the feature for the project, then move the requested locales from the
Available list to
Chosen. Clicking the
PRETRANSLATE button will pretranslate immediately all missing strings in enabled locales, otherwise pretranslation will run automatically as soon as new strings are added to the project.
To improve performance of the machine translation engine powering the pretranslation feature, custom machine translation models are trained for each locale using Pontoon’s translation memory. That results in better translation quality than what’s provided by the generic machine translation engine.
To create a custom translation model, first go to the team page of the locale you are creating a custom translation model for and download its translation memory file. Next, go to the Google Cloud console (requires permission) and follow these instructions — in case of doubt, consult the official instructions.
The first step is to create a translation dataset. In the
Datasets panel, select
- For the
Dataset name, follow the pattern used by existing datasets:
- Select the
Translate from…language (
English (EN)) and the
Translate to…language (e.g.
This operation will take a few seconds. At the end, an empty dataset with the selected name will be available in the list, with
0 in the
Total pairs column. It’s now time to import Pontoon’s translation memory and train the model:
- Click the dataset, then navigate to the
SELECT FILESto select the downloaded TMX file from your device.
Destination on Cloud Storagefield and select
CONTINUEto start the import process. The import process will take a few minutes (it’s possible to close this window and return later to the list of datasets, when completed the
Statuscolumn will say
- Once the import is completed, navigate to the
TRAINtab and click
Note that creating the model is a background job which takes a few hours (when completed the
Status column will say
Success: CreateModel), and models for at most 4 locales can be trained concurrently. When the model is created, store its name (usually starting with
NM, followed by a series of alphanumeric characters) under Google automl model in the Django’s admin interface of the locale.
From that point on, Machinery will start using the custom machine translation model instead of the generic one and you’ll be set to enable pretranslation for the locale.