Featured image: Two hands are visible in the foreground, one of which is holding a pen and writing on a document. AI-generated image; not a photograph of the reported event.
AI dominates the translation business. But for high-stakes work, humans are still needed. Source: bbc.co.uk.
AI Translates Trillions of Words Monthly
Google Translate processes one trillion words per month.
Other tools—like DeepL, Microsoft, Claude, ChatCGT, and Gemini—also contribute heavily.
Direct sales of machine translation, outside general AI subscriptions, were significant last year.
AI Works Well for Routine Tasks
For many everyday uses, AI translation is now sufficient.
Translating internal documents or product listings doesn’t need human review. Fixing errors would cost far more than the AI output.
John Zimmerman, founder of OKA, a Swiss non-profit translating Wikipedia articles, says skipping human checks makes sense in these cases.
AI Drafts Are Common—But Not Perfect
OKA uses AI to generate translation drafts.
Human translators then review every sentence to ensure accuracy.
Once, translators rewrote large portions of AI drafts. Today, many sentences require no editing at all.
Legal Errors Can Have Serious Consequences
AI can misread legal terms. This can lead to serious problems.
Ariella Baltié, a legal translator, found AI translated ‘Le demandeur’ (the plaintiff) as ‘the requester’—a confusing term in court documents.
It also misread ‘Madame X, épouse Y’ (Ms X, wife of Y) as ‘Ms X marries Y’. That turned a noun into a verb and lost the original meaning.
Medical Translations Are Especially Risky
Medical texts often use Greek letters. A single mistake can be life-threatening.
Carol Bereuter, a medical translation specialist, found AI replaced the Greek letter μ with ‘milli’ instead of ‘micro’.
That’s a serious error. Giving someone the wrong amount could be fatal.
AI Text Sounds Natural—But Errors Are Harder to Spot
Older AI tools had grammatical errors that stood out.
Modern AI output sounds fluent and natural. Errors blend in. They’re harder to detect.
As François Yvon of CNRS notes, AI relies on probability. What seems likely isn’t the same as what is correct.
Context Is Key—AI Often Misses It
AI struggles with context.
If it doesn’t know a sentence refers to a football match, it may mis-translate ‘taking a corner’.
Research Shows Human Review Still Matters
OKA conducted a study, funded by Wikimedia.
It analyzed 119 articles, 10 language pairs, and more than 1,000 hours of work.
- 27% of AI-generated text needed correction before publication.
- Three-quarters of those errors were introduced by AI.
- 5 to 6% of errors changed the original meaning.
What’s Next for AI Translation?
Future systems may self-correct using multiple AI tools.
One could include a translator, a critic, and a rewriter working in tandem.
The real breakthrough may be when AI shows when it’s uncertain.
As Yvon says, the key advance won’t be perfect translation—but showing doubt.
For now, humans remain essential in legal, medical, and high-accuracy fields.
AI is powerful. But it’s not infallible.
More on AI and language: Original source
