Google Translate processes one trillion words monthly, but machine translation still stumbles on idioms, legal precision, medical dosages and contextual ambiguity. The question facing the industry is no longer whether AI translation works, but whether it works well enough to replace humans entirely.
François Yvon, a research director at France's National Centre for Scientific Research, notes that AI systems struggle with expressions like the French "Il pleut des cordes" (it's raining ropes). A trained system might substitute "it's raining cats and dogs" for English, but struggles to find equivalents in languages like Indonesian. Context proves especially difficult. When an AI encounters "translate into Polish," it may misinterpret this as shoe cream. A sentence about "taking a corner" can confuse systems unaware they're reading about a football match.
Google Translate handles one trillion words per month, but DeepL, Microsoft, Claude, ChatGPT, Gemini and corporate systems also process enormous volumes. Direct machine translation sales reached an estimated $1.4 billion last year.
John Zimmerman, founder of Open Knowledge Association, a Swiss non-profit that translates Wikipedia articles, says AI excels at low-stakes work. "For an encyclopedia it's different. Wikipedia runs on trust. A wrong fact can sit there for years and get cited by journalists and AI systems," he says. OKA's own study of 119 articles across 10 language pairs found that 27% of AI-generated text required correction before publication. Three-quarters of those corrections involved errors introduced by the machine. Five to six percent of errors altered the meaning of the source text.
High-stakes translation demands human expertise. Ariella Baltié, a legal translator specializing in court documents, was asked to verify an AI translation that rendered "Le demandeur" (the plaintiff) as "the requester" and mistook a noun phrase for a verb, writing "Ms X marries Y" instead of "Ms X, wife of Y."
Medical translation carries similar risks. Carol Bereuter, a medical translation specialist, found an AI system that translated the Greek letter μ as "milli" rather than "micro," a mistake that could result in giving a patient 1,000 times the intended dose. Bereuter notes that modern AI output sounds natural and fluent, making errors harder to detect than with older systems.
François Yvon cautions that fundamental aspects of how these systems work may prevent them from becoming completely reliable. "As things stand, I think the fundamental way these systems function makes it quite hard to imagine how they could become 100% reliable."