Why Professional Translation Still Matters in the Age of AI

Machine translation has gotten surprisingly good. DeepL handles most European pairs well, and LLMs can pick up on context that would have tripped up older systems completely. So does anyone still need a human translator?

It depends what’s on the line.

When “good enough” isn’t good enough

For reading a menu abroad or skimming a news article, MT is fine. Nobody’s hiring a professional for that. But regulatory filings for the Kazakh government aren’t a menu. Neither is drug safety documentation going to CIS registration authorities. Get those wrong and you’re looking at compliance failures, project delays, or legal exposure. Not a typo on a restaurant sign.

The Central Asian problem

Central Asian languages make this worse. Kazakh, Kyrgyz, Uzbek, and Tajik have limited digital corpora, so MT models trained on them are weaker than what you’d get for, say, French or German. Kazakhstan is also mid-transition from Cyrillic to Latin script, which automated systems handle inconsistently at best. Government and legal documents require specific formal registers that MT misses more often than it gets right.

Where AI actually helps

We use MT ourselves. We built our own TMS, and for the right content types, machine translation is a reasonable starting point. Our post-editors (ISO 18587-certified) refine the output. But “for the right content types” is doing a lot of work in that sentence. Knowing when MT output is usable and when you need to start over requires judgment about context, audience, and consequences. That’s not something you can automate.

So what?

If your content matters to regulators or partners, get a human translator. It’s that simple.