The EU Publishes the 2026 EMT Framework: Check Judgement Before You Buy Translation

On October 26, the European Commission’s Directorate-General for Translation announced that the European Master’s in Translation (EMT) network had published an updated competence framework for translators. The PDF metadata shows a production date of October 20, the text was completed in September 2026, and the document states that it takes effect as the EMT competence framework from 2026. Translation competence is defined as five interdependent areas: linguistic and cultural, transfer, digital and technological, thematic and information research, and (inter)personal and service provision.

The first thing to pin down is who the document binds. It sets learning outcomes for master’s programmes and is a reference for universities preparing to join the EMT network. The Directorate-General’s news item places the next membership call at the end of 2028 and expects translations into all official EU languages in February 2027. It is not a regulation that requires companies to rewrite this quarter’s procurement contracts. The same news item says the update keeps the five competence areas and 36 skills of the 2022 version, and also calls transfer and thematic and information research two new areas. The framework text itself still organises competence into five areas, with skills numbered from 1 to 36.

The “two new areas” in the news are still five in the text

The Directorate-General’s news item summarises the change in four points: translation is no longer described as a single skill; the technology section now covers artificial intelligence, large language models and other language technologies; critical thinking and ethics run through the framework; and a glossary, including AI terms, has been added.

The framework and the EMT programme page name the five areas as linguistic and cultural competence, transfer competence, digital and technological competence, thematic and information research competence, and (inter)personal and service provision competence. “Two new areas” describes a renaming inside a five-area model. It does not mean two areas were added on top of the old five, making seven. The skills that can be checked still run from item 1 to item 36.

The technology items ask for judgement, not a record of clicks

Digital and technological competence is items 11 to 19. Graduates are expected to understand the basic principles of machine translation and its effect on the translation process, and to show AI literacy; and to understand the sources of large-language-model training data, training cut-off dates, how representative the data are, and what fine-tuning implies for accuracy, reliability and domain performance. Item 13 asks them to understand the risks and benefits of workflow automation and to bring machine translation into a workflow according to the task. Item 14 asks them to choose a technology, and to explain that choice, against task complexity, data sensitivity, ethics and the client’s requirements.

Item 17 asks them, when working with a specific model, to evaluate automatically generated answers and to do further research. Item 18 includes data protection, bias mitigation, transparency, accessibility, responsible sourcing of language data, environmental impact and effects on professional practice among the principles to consider when using machine translation and AI. These are learning outcomes on a training programme, not measured scores for working translators.

Post-editing and checking sit inside transfer and research

Transfer competence covers meaning transfer across languages, within one language, and across sign systems, and it includes analysis before translation and final quality control. Item 8 requires post-editing of machine translation, reducing priming and interference in the target text, and using a style guide and a glossary, where they exist, to hold quality. Item 6 requires a translation that is fit for purpose.

Item 21, under thematic and information research, requires graduates to recognise and remove inappropriate additions, omissions, bias, hallucinations or fabrications in machine-translation and AI output. Item 20 lists AI-generated content among the sources whose relevance and reliability have to be checked. The premise under linguistic and cultural competence is that quality translator education starts from at least two working languages at CEFR C1, or an equivalent level. That is a premise for training, not a new employment rule for companies.

It governs training programmes, not this quarter’s contract

The foreword says the framework is addressed first to teachers who design curricula and to universities preparing to apply to the EMT network in the next round. Working translators can use it as a reference for lifelong learning. The Directorate-General’s news item says the EMT network currently includes 81 master’s programmes at 79 universities in 24 countries. Translations into all official EU languages are expected in February 2027. Until then, the text that can be checked line by line is the English original.

The publication carries an authors’ statement: the views in the document are the authors’ responsibility and do not necessarily represent the official position of the European Commission. The news item was issued by the Directorate-General for Translation. The foreword is signed by Christos Ellinides, Director-General for Translation and chair of the EMT Board, and the text is presented as the EMT competence framework effective from 2026. A company can use it as the reference that European translator training is adopting. It should not be read as a mandatory standard already written into procurement contracts.

Four gates before rewriting a translator requirement

• Document gate: Use the five names and the numbered skills in the framework text. Do not read “two new areas” in the news item as two areas added on top of five.
• Task gate: When buying post-editing, terminology or quality checks, ask whether the person can evaluate a model’s answers, explain data sensitivity and point out hallucinations. Being able to open a tool is not the judgement described in items 14, 17 and 21.
• Language gate: Keep language level, transfer competence and subject-matter checking separate. C1 is the starting point for training. It does not replace domain fact-checking or post-editing.
• Timing gate: The next EMT membership call is at the end of 2028, and full translations are expected in February 2027. A current contract does not have to pretend that an untranslated standard already applies. A role description can still be written against the judgement the English text already asks for.

The 2026 EMT framework writes the use of machine translation and large language models into the standard actions of translator training, and it leaves selection, checking and responsibility on the human side. When a company going abroad buys language services, the useful question is whether the provider can judge, for the task at hand, whether a technology applies, and can catch fabrication and mistranslation. It is not whether the provider claims to have connected a particular model. Ongoing translation work still needs terminology, a quality target and a named owner of the final text, agreed inside the professional translation workflow.

Sources: the European Commission Directorate-General for Translation news item of 26 October 2026; the European Master’s in Translation Competence Framework 2026 PDF (catalogue number HC-01-26-016-EN-N); the European Commission’s European Master’s in Translation programme page. The framework lists learning outcomes for training programmes, not a uniform examination of working translators. Translations into all official EU languages have not yet been published.