Akorbi perform a comprehensive rollout process including a personalized discovery phase for your content, language and quality needs. This may result in proposing Engine Training before kicking off MT in production. MT engine training will help customize the base model to industry and client – specific language and style. This in turn reduces the amount of edits necessary to be made by the post-editor, and thereby time and cost involved in the process, compared to that of an untrained engine.
This approach is particularly valuable for specialized industries, where standardized terminology is essential for effective communication. For engine training, it is important to have large client data in the form of translation memories and glossaries, however even more critical is the quality of this data. The MT engine will directly reflect the data it is trained on, in its output, so benefits will be reaped by maintaining clean data for this process.
It is important to note that training an engine does not re-train the MT base model engine, but acts as an overriding mechanism that applies predetermined translations for specific terms and general style preferences across the text. This ensures that custom terminology and style is prioritized while still benefiting from the extensive training of the underlying language models.