AI translation is fluent now, but human checkers remain essential
BBC reports that Google says people now translate one trillion words a month with Google Translate, while direct sales of machine translation outside general AI subscriptions were estimated at $1.4bn last year. OKA (Open Knowledge Association), a Swiss non-profit, ran a Wikimedia-funded study covering 119 articles, 10 language pairs and more than 1,000 hours of translator work, finding that 27% of AI-generated text was still corrected before publication, three-quarters of those corrections were errors the AI introduced, and 5 to 6% of errors changed the meaning of the original. Legal and medical translators cite AI rendering "Le demandeur" (the plaintiff) as "the requester" and translating the Greek letter μ as milli rather than micro, a thousandfold dosage error; CNRS researcher François Yvon says the way these systems function makes 100% reliability hard to imagine.
In legal and medical settings translation errors carry serious consequences: the report notes that errors in machine-translated US asylum applications have led to delays in the release of detainees.
ByJohn Laurenson
Business reporter, Reporting fromParis
"Il pleut des cordes," say the French when it's raining hard.
"It's raining ropes," is the direct translation. It's the kind of idiom which can trip up an AI translator.
"You can imagine the dilemma this throws AI into," says François Yvon, a research director specialised in automatic translation with the French National Centre for Scientific Research CNRS.
#Google Translate#Google#DeepL#ChatGPT#Claude#Gemini#Open Knowledge Association#CNRS#Wikimedia