| nav_order | grand_parent | parent | title | description |
|---|---|---|---|---|
2 |
Application areas |
Translation and localisation |
Hybrid translation |
Workflow with both human translations and pure machine translations |
In a hybrid translation workflow, some raw machine translations are never seen or touched by a human translator.
Hybrid translation can be faster and cheaper than full human post-editing.
Hybrid translation requires good machine translation. A signification portion of the machine translated segments should be usable as-is.
The hybrid translation workflow was first presented by Microsoft, VMWare and Unbabel.
Slide from a ModelFront presentation
A risk threshold is set. Each new machine translation is automatically classified as high-risk or low-risk.
A low-risk machine translation is machine-approved as-is.
A high-risk translation is still sent to a human translator for post-editing.
In the translation management system (TMS) and computer-aided translation (CAT) tool, the machine-approved machine translations are handled like translation memory matches. They are either locked or marked as approved.
The key technology for a hybrid translation workflow is translation risk prediction, which is known as quality estimation in the research world.
The first companies to adopt the hybrid translation workflow built quality estimation internally.
- Microsoft
- VMWare
- Wayfair