A translation competence framework needs observable AI assessment
Europe’s updated translator framework makes LLM use, critical judgment and ethics more explicit, but curriculum teams still need performances that demonstrate those competences under real constraints.
What happened
The European Master’s in Translation network published an updated competence framework on 26 October 2026, retaining five connected areas and 36 skills while expanding expectations around AI, large language models and language technology.
Why it matters
A framework can align curriculum language across universities, but it changes graduate capability only when programmes translate each competence into assessed work, failure conditions and evidence that remains visible after AI assistance is removed.
What changed
The European Commission announcement says the European Master’s in Translation network has updated its professional competence framework. It retains five connected competence areas and 36 skills from the 2022 version while making transfer, thematic and information research, AI, large language models, critical thinking and ethics more explicit. Slator independently describes the shift from machine-translation basics toward LLM literacy, workflow automation and evaluation of generated content.
The framework matters because it is not an isolated course checklist. The EMT network covers 81 programmes at 79 universities in 24 countries, and the framework informs programme design and quality criteria for membership. Its language can therefore shape what curriculum committees buy, teach and assess before the next membership call at the end of 2028.
The implementation gap
A competence statement is not evidence that a student can perform. “Evaluate AI output” may mean spotting obvious errors in a prepared exercise, diagnosing a subtle cultural failure in an unfamiliar domain, defending a revision to a client, or deciding not to use the tool. Those performances have different consequences and require different evidence.
The same distinction applies to critical thinking and ethics. Adding those words across a framework can improve coverage, but programmes still need observable behaviours: identifying hidden assumptions, documenting provenance, escalating a confidentiality problem, preserving authorship decisions and explaining uncertainty to a service recipient. Without an assessment map, technology content can expand while judgment remains implicit.
Build an assessment map
For each AI-related competence, write one performance statement using a verb and a boundary. Name the input, time limit, permitted tools, quality standard and failure condition. Then specify the evidence artefact: a revision trace, source log, risk note, client explanation, bilingual comparison or rejected-output rationale. A separate assessor should be able to see why the decision was made, not only whether the final text reads well.
Add an unaided check where retained human capability matters. A student might first complete a bounded translation or verification task without an LLM, then repeat a related task with assistance and compare errors, time and reasoning. The point is not to ban tools. It is to distinguish tool operation from transferable language judgment and to detect when polished output masks weak understanding.
Assessment should also test transfer. Rotate subject matter, language direction, data sensitivity and client constraint so that success cannot be memorised from one workflow. Record which competences were directly observed and which were inferred. Because the Commission announcement is not an outcome study, programmes should publish neither placement nor proficiency claims without their own denominators and follow-up evidence.
Govern the map like a professional standard, not a one-off syllabus exercise. Version the task, rubric and allowed tools; train assessors on borderline examples; and retain a sample of decisions for moderation. Invite employers and practising translators to challenge whether the task represents current work, while keeping academic owners responsible for validity and student protection. Review the map when tools, privacy requirements or service models change, but preserve comparable anchor tasks so the programme can tell whether performance changed rather than merely the test.
Document accommodations and language-pair limits as part of validity, so accessibility support is not mistaken for weaker competence and one language direction is not generalised to all professional work.
Decision rule
Do not approve a curriculum revision that maps the new framework only to module titles. Require an evidence matrix linking each material competence to a supervised performance, artefact, unaided check, assessor and remediation route. Pilot it on a small set of tasks, inspect disagreements between assessors and revise the rubric before high-stakes use. This applies the same principle as Skills Intelligence’s analysis of AI accountability beyond prompt skills: professional competence must remain visible in work.