Link Search Menu Expand Document

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.

Workflow diagram

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.

Technology

The key technology for a hybrid translation workflow is translation risk prediction, which is known as quality estimation in the research world.

Adoption

The first companies to adopt the hybrid translation workflow built quality estimation internally.

  • Microsoft
  • VMWare
  • Wayfair

See also


Machine Translate is created and edited by contributors like you!

Learn more about contributing. Licensed under CC-BY-SA-4.0.