Skip to content
a Cup of Words

Opinion

Do you still need a translator when AI can translate?

I have been a freelance translator since 2013 and a sworn translator registered with the Court of Appeal of Dijon. Here I answer the question I have been asked most often over the past three years, drawing on what I see from my desk and on what the available research actually supports.

Portrait of Silvia Vannier-Feranec

Silvia Vannier-FeranecTranslator, English and German into French

The question people ask me

Do you still have work as a freelance translator, now that AI is here?

People ask me this regularly, and I understand why: anyone can now translate an entire text in a few minutes, and the result reads well enough. Claiming that nothing has changed would be dishonest, and nobody would believe me.

What has changed is not the amount of work but its nature. Professional translation has always involved two people, one who translates and one who revises. In some projects, machine translation has taken over the first role, and the agencies that use it still entrust the second to a person. That is the role I fill more and more often, and it has a name: post-editing.

Two people, two roles

Traditionally

The translator

The reviser

In some projects

Machine translation

Post-editing

What I do when an agency sends me a machine translation

I take machine translation output and bring it up to the standard expected of a human translation: meaning checked against the source, terminology aligned with your glossaries, style made to read naturally, false friends and mistranslations removed. Light or full post-editing, depending on what the text is for and what the budget allows.

Many translators turn this work down, and the figures make the point better than I can: in the consultation the Société française des traducteurs, the French national translators’ association, ran among its members in March and April 2026, to which 374 of them responded, 93% said they would rather translate than post-edit, and only 6% of those who post-edit chose to do so. I take it on, provided it is recognised for what it is and paid what it is worth.

would rather translate than post-edit
93%
of translators who post-edit do so by choice
6%

Société française des traducteurs, members’ consultation of March and April 2026, report of June 2026, 374 respondents

What I correct is not what you might think

What I fix is hardly ever a spelling or grammar mistake; those errors have all but disappeared. It is the subtle things: a word that sounds better than another in its context, a repetition that weighs the text down, phrasing that needs unpicking, a shift in register that would not go down well with your client.

This is not just my own impression. Two researchers at Ghent University had texts revised by 16 professional translators from eight agencies, without telling them whether they were looking at a human or a machine translation. For the same quality requirement, the machine output needed roughly twice as much of its text changed. And in a 2024 doctoral thesis covering 90 million words of real-world production, more than 40% of post-editors’ interventions changed the meaning of the machine output: these were not merely cosmetic adjustments.

text changed in the machine output, for the same quality requirement
about twice as much
of post-editors’ interventions changed the meaning of the machine output
more than 40%

What the law requires, and no software can provide

When it comes to translations intended for government offices, courts, notaries or consulates, the issue is not quality, but accountability. According to the French Code of Criminal Procedure, if no expert on the official list is available, the judicial authority may appoint another adult to assist the court on their honour and conscience, provided they swear an oath to do so.

So what the law requires is not a particular skill but someone who gives their word and is accountable for their translation. As a court-appointed translator registered with the Court of Appeal of Dijon since 2021, I certify my translations and I am accountable for them.

More about sworn translation

Where I stop correcting and start suggesting

On an advertising tagline, slogan or brand signature, correcting gets you nowhere. What needs to be transferred from one language to another is the effect that the text has on the reader, and a literal translation often misses this. So I suggest several versions, usually three, that no longer resemble the original but work in the target market. This is transcreation: a separate assignment agreed in advance and charged by the time spent rather than by the word count.

It is the one part of my work where a comparison with a machine makes no sense, because the job is no longer to rework a text but to write a new one from an intention.

What this means if you buy translation

  • You are an agency

    If you are an agency, you already know all this, and what matters to you is whether I work on your terms: I work in whichever CAT tool you use, I take tests, I follow your style guides and deadlines, and I invoice on the usual industry terms.

  • You are a company

    If you are a company, the useful question is not whether to use machine translation but what you publish under your name. A text that comes out of an engine is not a finished text, and the gap between the two rarely shows in a language you do not speak.

Tell me what you need and attach the document if you have it: I will reply within one working day with a firm quote, per word or by time spent, depending on the project.

Sources cited on this page

  1. Société française des traducteurs, Consultation des membres sur les usages, les perceptions et l’impact de l’intelligence artificielle générative et de la traduction automatique, results report, June 2026 (data collected from 18 March to 5 April 2026, 374 member respondents). sft.fr (new window)
  2. Joke Daems and Lieve Macken, “Post-editing human translations and revising machine translations: impact on efficiency and quality”, in Koponen, Mossop, Robert and Scocchera (eds), Translation Revision and Post-editing, Routledge, 2020, pp. 50-70. doi.org (new window)
  3. Silvia Terribile, Productivity in the Post-Editing of Neural Machine Translation: A Mixed-Methods Analysis of Speed and Edits at Toppan Digital Language, doctoral thesis, University of Manchester, 2024. research.manchester.ac.uk (new window)
  4. French Code of Criminal Procedure, Article D594-16, version in force since 15 April 2022. legifrance.gouv.fr (new window)