AI post-editing improves translations that already exist. Where AI translation drafts from nothing, post-editing reads each segment's current translation and rewrites it to publishable quality, guided by your translation memory, your terminology, and your instructions. It runs as a supplier on a job, so it fits into your workflows exactly where a human post-editor would.
What AI Post-Editing Is For
-
Cleaning up machine translation output. Chain machine translation and post-editing in one workflow: the first job drafts, the second polishes.
-
Raising fuzzy matches. Memory pre-translation fills segments with partial matches; post-editing rewrites them into correct target text without a linguist touching every one.
-
A quality pass before review. Reduce what human reviewers need to correct.
By default, post-editing works on exactly the content that needs it: fuzzy memory matches and machine translation output. Segments a person edited, exact and context memory matches, pre-translations from a previous document version, and locked segments are left untouched. All of this is adjustable per configuration (see Configuring AI Workflow Parameters).
How to Run It
AI post-editing is assigned like any supplier. The service is defined once, then used on any job:
-
Set up a supplier service with an AI agent, selecting a post-editing workflow, for example Post-Edit workflow: 1 Agent (see AI-powered suppliers for the full service setup).
-
Assign a job to that supplier and service.
-
When the job status moves to In Progress, the workflow starts on its own. Nobody accepts or starts the work manually.
-
When the workflow finishes, the content is updated and the job is completed automatically, triggering whatever comes next in your workflow.
While it runs, the job's AI agent status shows live progress, and the job messages record what happened (see Tracking AI Work).
Choosing a Workflow Variant
|
Variant |
How it works |
When to choose it |
|---|---|---|
|
1 Agent |
One agent post-edits each batch in a single pass. |
Bulk improvement where later steps catch what remains. |
|
3 Agents |
The first agent post-edits, a second validates the result, and a third corrects what was flagged. |
Content going out with little or no human review afterwards. Each extra agent adds processing time and AI usage. |
Learn More
-
AI-powered suppliers: Defining the AI agent service in a supplier pricelist.
-
Configuring AI Workflow Parameters: Reference filters, segment scope, and custom instructions.
-
Tracking AI Work: Progress, errors, and cancellation.