How long will humans be needed for translation?
AI dominates the translation business, but for high-stakes work, humans are still needed.
Key points
- Google says people now translate one trillion words a month with Google Translate.
- OKA found 27% of AI-generated text was corrected before publication.
- Medical mistranslation of the Greek letter mu caused a thousandfold dosage error.
Machine translation is now widely used, with Google saying people translate one trillion words a month using Google Translate. Direct sales of machine translation outside general AI subscriptions were worth an estimated $1.4bn last year.
But the article asks how long humans will still be needed, citing errors in legal and medical texts. François Yvon, a research director at the French National Centre for Scientific Research CNRS, said a system may handle an idiom in one language but fail to find an equivalent in another.
Errors in legal and medical texts
Ariella Baltié, a translator and legal expert, was asked to check court documents already translated by AI. She found two major errors at the beginning: the plaintiff became the requester, and Ms X, wife of Y became Ms X marries Y.
Carol Bereuter, a medical translation specialist, found a text where the Greek letter mu was translated as milli instead of micro. She said giving a milligram instead of a microgram means giving a thousand times the dose. She added that AI errors are now harder to spot because the output sounds fluid and natural.
The real breakthrough won't come until these systems are not merely capable of translating but also of telling us when they're not sure
— François Yvon
What a study found
Last year OKA ran a study funded by Wikimedia, covering 119 articles, 10 language pairs and more than 1,000 hours of translator work. John Zimmerman, founder of OKA, said 27% of the AI-generated text was still corrected before publication, and 5 to 6% of errors changed the meaning of the original text.
Zimmerman said OKA has published more than 10,000 pieces, mostly into Spanish, Portuguese and English, and could not have done so without AI. He said the question is no longer whether AI translation is good, but whether it is good enough to remove the human.
