AI Quality Evaluation & Supervision
Supervise the machine, don't just translate: grade AI output with MQM, use quality estimation, and design review workflows that scale.
As machines take over first drafts, the human role shifts from translating everything to judging translations at scale, and that supervisor position is one of the most durable in the field. Learn to grade output with the industry-standard MQM error typology, classify errors by severity consistently, build a defensible scorecard, use quality estimation to review smartly instead of reading everything, and govern terminology so quality holds across translators, models, and releases.
What you'll learn
- Grade translation output with the MQM error typology
- Classify errors by type and severity consistently
- Turn a rubric into a repeatable, defensible scorecard
- Use quality-estimation signals to triage what to review
- Design a lightweight, risk-based review workflow
- Govern terminology across an AI translation pipeline
Course content
- 1. From translator to supervisor: the role that survives (14 min)
- 2. The MQM error typology (16 min)
- 3. Severity: minor, major, critical (15 min)
- 4. Building a scorecard you can reuse (15 min)
- 5. Quality estimation: reviewing smartly at scale (15 min)
- 6. Governing terminology and the pipeline (15 min)
Related courses
- AI for Translators: Your First LLM Workflow — Drive ChatGPT, Claude & Gemini as a translation copilot: a repeatable, confidentiality-safe workflow that keeps you the editor in charge.
- Prompt Engineering for Translators — Turn vague AI output into on-brief translations: glossaries, style rules, and reusable templates that make ChatGPT, Claude and Gemini obey.
- Machine Translation Post-Editing (MTPE), Done Right — Post-edit to the right standard, light vs full per ISO 18587, spot the errors machines make, and price MTPE so it actually pays.