Human-in-the-loop
Human-in-the-loop describes a workflow where AI produces the draft and a qualified person reviews, corrects or approves it before it is used. In translation it is the standard responsible pattern: machine speed at the front, human judgement at the quality gate.
How it works
In a human-in-the-loop workflow, automation does the heavy, repetitive part, and a person is placed at the decision points that need judgement or carry risk. For translation that means the machine drafts and a qualified human reviews, corrects and signs off before anything is used.
The design question is where the human sits. Reviewing everything is safe but slow; reviewing nothing is fast but reckless. Good systems use signals like confidence and quality scores to put human attention where it changes the outcome.
How SourceTarget uses it
Human-in-the-loop is the backbone of how SourceTarget works: machine speed for the draft, human judgement at the quality gate. Confidence and quality scores decide how much review a job needs and which segments to check first.
Nothing published leaves without a qualified human having reviewed it to the agreed standard.
Human-in-the-loop compared with Full automation
| Human-in-the-loop | Full automation | |
|---|---|---|
| Final check by | A qualified human | No one |
| Speed | Fast, with review targeted by risk | Fastest |
| Risk | Errors caught before publication | Confident errors ship unnoticed |
| Right for | Anything published or high-stakes | Low-stakes, gist-only use |