|
The Ghost Has Taken a Seat at the Table
By Volkan Güvenç, Founder — Alafranga Language Solutions  Not long ago, the idea that AI would replace translators felt like something you could safely dismiss — a recurring anxiety in a profession that had survived every previous disruption. I was wrong about the timeline. I did not expect it to arrive this fast, or feel this concrete.
|
Early 2026 looked, briefly, like it might bring a recovery. Then the Iran war changed the mood. Projects froze mid-pipeline. Clients who had been close to signing went quiet. What made this harder to read than a typical downturn is that two things are happening at once: geopolitical disruption compressing demand, and AI capability expanding the alternatives. When uncertainty rises and a cheaper substitute is sitting right there, clients do not wait — they decide.
|
The Tsunami Hit Translators First
That AI represents a permanent shift — not a hype cycle — became the prevailing view somewhere around the time Claude arrived. Before that, you could still find serious people arguing it was overstated. That argument is harder to make now.
|
Translators were among the first knowledge workers to feel this directly, and the damage has not been evenly distributed. So it is worth asking precisely: which kinds of professional translation has AI actually reduced or eliminated?
The clearest case is into-native-language work. A subject-matter expert can now draft in their own language, run it through AI, and review the output themselves — without involving a translation agency at all. This works until the volume becomes unmanageable, at which point they come back. But when they do, they are likely to ask for post-editing rather than full translation. The margin on that work is thinner. The skill required is different.
For translation out of one's native language the situation is starker. You cannot verify what you cannot read. If the output is wrong, who carries the liability? Not the AI. Not the client's internal team. The question of accountability has not gone away — it has just become easier to ignore, right up until something goes wrong.
|
The Expertise Tax
There is a persistent misconception that using AI for translation is a neutral act — that you feed text in and quality comes out, and the only variable is which model you choose. In practice, it is nothing like that.
Getting good output from AI in a professional translation context requires knowing what good looks like. It requires the ability to catch errors that are linguistically fluent but factually or culturally wrong — the kind of error that a non-specialist reviewer will miss entirely. Without that expertise, AI does not reduce the risk of a bad translation. It just makes bad translations faster and cheaper to produce.
|
The consequences can be mundane or serious, sometimes both. An AI-translated video recently rendered a reference to Atatürk in a way that anyone who knows Turkish history would find offensive — the kind of mistake that a human translator with basic cultural literacy would never make. It is easy to laugh at examples like this. It is less easy when the same category of error appears in a medical device manual, a legal contract, or a regulatory submission.
Language is precise. Culture is precise. AI, at present, is not.
|
An Industry That Cannot Quite Commit
There is a persistent misconception that using AI for translation is a neutral act — that you feed text in and quality comes out, and the only variable is which model you choose. In practice, it is nothing like that.
Getting good output from AI in a professional translation context requires knowing what good looks like. It requires the ability to catch errors that are linguistically fluent but factually or culturally wrong — the kind of error that a non-specialist reviewer will miss entirely. Without that expertise, AI does not reduce the risk of a bad translation. It just makes bad translations faster and cheaper to produce.
|
One of the more interesting tensions in the industry right now is the relationship between AI and translation memory — TM. These two technologies do not sit easily together, and the sector has not fully decided what to do about it.
TM was the foundation of professional translation workflows for decades: approved terminology stored at the segment level, carried forward across projects, ensuring consistency across hundreds of thousands of words and multiple linguists. That accumulated value does not disappear just because AI has arrived. Clients have invested in it. Agencies have built workflows around it. Linguists have contributed to it.
AI, by contrast, works holistically. It reads context across paragraphs and sections, which is part of what makes it capable of producing fluent, coherent output. But that is precisely where it conflicts with TM, which operates segment by segment. The two systems are not just different tools — they have a fundamental architectural mismatch. Early attempts to bridge them, LILT being the most discussed example, are promising but not yet seamless. The friction is real and the industry is still working through it.
|
What Has Not Changed
It would be convenient if the uncertainty of this moment pointed neatly toward a simple conclusion. It does not. What I can say, based on two decades in this work, is this: the space for professional translation has not closed. It has narrowed in some areas and deepened in others.
|
AI struggles most where the cost of error is highest and the knowledge required to catch that error is most specialised. Regulated industries — medical, legal, industrial, energy — require documentation that is not only linguistically accurate but terminologically consistent, jurisdictionally aware, and defensible under scrutiny. Building and maintaining that standard requires expertise that AI does not have and cannot independently acquire. It can assist. It cannot own the outcome.
The translators who are finding this period most difficult are, in many cases, those who competed primarily on speed and price in content categories where AI is now genuinely competitive. That market has contracted, and it is unlikely to return. The translators building for the long term are those who went deep in a field — who spent years developing knowledge that cannot be replicated by a model trained on general text.
|
Where We Stand
It would be convenient if the uncertainty of this moment pointed neatly toward a simple conclusion. It does not. What I can say, based on two decades in this work, is this: the space for professional translation has not closed. It has narrowed in some areas and deepened in others.
|
We have been doing this since 2002. Turkish technical translation, founded in Istanbul, now operating from London across more than forty languages. We built AI into our workflows because the evidence for doing so was compelling. We kept human review at every critical step because the evidence for that was equally compelling.
This is not a defence of the status quo. The status quo has changed and will keep changing. It is a statement about where we think value actually sits: in judgment, in accountability, in the kind of knowledge that takes years to develop and cannot be shortcut.
Whatever the tools become, someone has to read the output and stand behind it. That has always been true. It remains true.
The ghost has taken a seat at the table. But the table is still ours to work at.