Will AI replace translators?

Machine translation into Arabic is now good enough to fool you, which is the most dangerous stage there is. An obvious error gets fixed; an elegant one gets published.

AA Abdelilah Arahal
4 min read Updated 21 September 2026

Translation was the first thing anyone pointed a computer at, and the promise has been repeated since the 1950s. What is different now is that the promise got close enough to be dangerous.

Machine translation into Arabic is now good enough to fool you. That is the most dangerous stage there is, because an obvious error gets fixed and an elegant one gets published.

What happened to me a few weeks ago

I needed some papers translated. Official ones, the kind nobody is allowed to get wrong.

So I hired a professional translator. The correct choice, surely.

And when the work came back I understood what had actually happened: she had used AI. Then she signed it.

I was not annoyed. Because thinking it through, I realised I had never been buying the translation. I was buying the signature.

Translation is free in my pocket. What I cannot do myself, at any level of tooling, is make the document legally acceptable, with a licensed person carrying responsibility for what is in it.

The round-trip test

I use this in training and it reveals more than people expect.

Take an Arabic text, machine translate it to English, then translate the result back to Arabic with a different tool, and compare against the original.

What survives is literal meaning. What falls out almost every time: tone, register, cultural reference, and wordplay. The text comes back correct and flattened, as though it lost its face.

3the first things to fall, tone, context and wordplay
2tools needed for the test, no more
10minutes it takes

Where it fails specifically in Arabic

Dialects. Models are trained far more heavily on formal Arabic. Egyptian is passable; Gulf and Maghrebi are weaker.

Register. Arabic distinguishes fine gradations of formality; machine translation picks one level and holds it, usually the wrong one for the context.

Ambiguity without diacritics. Many words share a spelling and differ in meaning, with only context to separate them.

Direction. A Latin technical term, a numeral, or a punctuation mark inside an Arabic sentence frequently breaks the ordering in machine output.

Names. Transliterating an Arabic name without a fixed standard produces three versions of the same person in one document.

What remains

Post-editing. Now a profession in its own right, and growing faster than translation from scratch.

Localisation rather than translation. Making a campaign work in the Gulf market is a different job from moving its words.

Texts that carry liability. Legal, medical, official.

Literature. Where the requirement is the effect, not the meaning.

What this means for you

If you translate: price post-editing separately and explicitly. Selling "translation" by the word invites comparison with a free tool; selling "responsibility for correct Arabic" invites no comparison at all.

If you buy translation: ask specifically about the human review stage. Its absence is the difference between text that passes and text that embarrasses.

If you publish in Arabic: run the round-trip test before publishing. Ten minutes prevents what is hard to withdraw later.

In closing

Literal translation became a cheap commodity. Guaranteeing that Arabic text is correct, appropriate, and answerable for is a market that is widening rather than narrowing.

Try this today: take a paragraph from your own site and run the round trip. If it comes back flattened, that is what your visitors are reading.

Common questions

Will AI replace translators?
It replaces literal translation of general text. It does not replace localisation, texts carrying legal or medical liability, or literature. Post-editing is growing faster than translation from scratch.
Where does machine translation fail in Arabic?
Non-standard dialects, register, ambiguity between identically spelled words, ordering when Latin and Arabic mix, and consistent transliteration of names.
How do I check translation quality quickly?
The round-trip test: translate out, translate back with a different tool, compare with the original. Tone, context and wordplay are what usually disappear.
Is machine translation acceptable for contracts and official documents?
Not on its own. These carry legal consequences and many authorities require a certified translator's stamp, which no tool carries, along with the liability behind it.
Share

No comments yet

Leave a comment

Never published. Used only if I reply to you directly.

Comments are read before they appear.

Related reading

12 min read

The best framework for building AI agents in 2026

Seventeen agent frameworks, ranked by what survives production rather than by GitHub stars. Some of the most popular names sit in the bottom tier, and one of them is a security incident waiting to happen.

Start with a short call

Fifteen minutes to understand what your team does and what you want to change. If training is not the right answer, I will say so.