The industry-side impact of the EU AI Act and our company-side implementation are covered in "EU AI Act Impact on ITWA Exhibitors" and "EU AI Act Compliance in Practice". This article is for translators — specifically those of you doing MTPE (machine translation post-editing). The workflow changed in August, and you need to know what shifted.
MTPE is no longer "machine draft, quick polish"
Six months ago the standard MTPE workflow was straightforward: machine translates first draft → human reviews and polishes → deliver. After the EU AI Act took effect, a critical judgement sits at the end of that workflow: did your revision reach "substantial modification"? If yes, the client may avoid marking the translation as AI-assisted. If no, marking may be required.
For translators this means you no longer deliver "one polished translation". You deliver "one polished translation + a revision ratio + a breakdown of what you changed". The added work is modest but it shifts the mindset — MTPE used to be about being fast and accurate; now it also needs to be about being sufficiently thorough.
Three new requirements we set for our MTPE translators
First: revision-ratio tracking is mandatory. We use Word's native revision tracking. Translators paste the raw machine output into Word, turn on "track changes", and work directly on it. Our internal system auto-computes the revision ratio when the file is submitted, excluding formatting-only edits. Translators fill in a final confirmation on the delivery form. You do not calculate the ratio manually, but you are responsible for making genuine substantive revisions — no padding the number with trivial changes just to reach the threshold.
Second: compliance awareness is required. You need to understand that the MTPE files you produce may be submitted to EU regulatory bodies (CE certification) or to EU buyer procurement teams. If your revision ratio is low and the client later faces marking requirements, that is not your fault — but if you falsely confirm a ratio above 50% when the actual ratio is clearly below, you share responsibility. Be honest about your revision ratio.
Third: translation provenance records. For fully human-translated files (not MTPE), we now generate provenance records: translator ID, work timestamps, term library version used. This is not extra work — most translators already use these resources, you just record them in a form field. But the record needs to be specific: "confirmed customer term library v3.2 for automotive electronics", not "used customer terminology".
Three practical techniques to raise revision ratio without wasting effort
Some translators worry: machine translation keeps improving. Sometimes the output is almost good enough — how do you push it above 50% without inventing changes?
Technique one: restructure phrasing for natural flow. Machine translation's most visible weakness is unnatural phrasing — "in light of the fact that" instead of "given that", "with regard to" instead of "on". Industrial clients prefer concise, direct language. Rewriting these improves readability and pushes up revision ratio. It is not padding — it is making the translation actually usable.
Technique two: unify terminology across the full file. Machine translation often renders the same term differently in different contexts — "wafer" as both "wafer" and "crystal" in a semiconductor specification. Go through the file and standardise every instance to the client's confirmed term. A 30-page file can easily contain 100+ ununified terms — fixing all of them pushes the ratio above 50% fast.
Technique three: add translator notes on ambiguous terms. For industry-specific terms that are ambiguous or where multiple renderings exist, add a brief translator's note in brackets: "(translator note: XX renders as XX per customer term library v3.2)". This adds revision-count value while also making the translation more useful to the client.
The benefit of all three: your revisions are genuine improvements, not padding. You raise the ratio without making work for work's sake.
One practical recommendation
The EU AI Act is the first of what will be several waves of AI regulation — the United States, United Kingdom and ASEAN are drafting their own frameworks. Build compliance awareness as a permanent capability, not a temporary burden. MTPE translators' core competitive edge now combines speed, accuracy and depth of revision. If you want to stay competitive in MTPE, treat revision-ratio tracking as a habit, not an extra task.
To join our MTPE translator pool, contact us with your target sector (semiconductors/robotics/automotive electronics/other) and working language pairs.