Post-editing
Post-editing is the human correction of machine translation output. ISO 18587 defines two levels: light post-editing fixes only errors that affect meaning and leaves the rest, while full post-editing produces output comparable to a human translation. Anything published should be full post-edited.
How it works
Post-editing takes raw machine translation and brings it up to a target quality level. ISO 18587 sets two: light, which corrects only errors that distort meaning, and full, which corrects everything a reader would notice, matching human-translation quality. The editor decides what to keep from the machine and what to rewrite.
The skill is knowing when to leave good machine output alone. Over-editing wastes the speed advantage; under-editing ships errors. Trained post-editors work to the agreed level, not to personal preference.
How SourceTarget uses it
SourceTarget's post-editors work to ISO 18587 levels, with the level fixed before the job starts. Output metrics and confidence scores flag the segments most likely to need attention, so effort goes where it counts.
Post-editing compared with Translating from scratch
| Post-editing | Translating from scratch | |
|---|---|---|
| Starts from | Machine translation output | A blank page |
| Editor decides | What to keep and what to rewrite | Every choice from nothing |
| Faster when | Machine output is good | Output is poor or content is creative |