Globalese 1.6 is out, bringing faster and better processes, changes in the UI, and an extended API.
Under the hood
- The unknown word report has been rewritten so instead of taking almost the same time as translating a file, it only takes a couple of seconds per file. Note: this requires retraining of phrase tables.
- Faster and better learning and translation for Japanese.
- Improved quality estimation algorithm.
- It is now possible to translate Wordbee XLIFF files.
- Instead of parallel text files, bilingual XML-based files can now be used as engine assets (for tuning and quality estimation model training).
- Engine assets can now be created from any bilingual resource in the Repository. When retrieving an asset from the Repository, it is now possible to crop it to any user-defined number of random segments.
- The “ApprovedTranslation” status is considered in SDLXLIFF files.
- When training a quality estimation model, users can now choose which assets to include instead of automatically using all available assets.
- Added “Delete all evaluation files” option to the Files menu on the project page.
- Added a “Stop QE model training button” on engine pages.
- Project files can now be evaluated directly from the project page.
- The status of quality estimation models is displayed on the Engines page.
- A new tab, “Logs” now displays the most important events for models.
- When deleting an engine, it is now possible to delete underlying components that are not in use anywhere else.
- The list of possible API calls has been extended.