AI translation tools are increasingly used across various industries, but they still struggle with idioms and nuanced language. For instance, the French expression 'Il pleut des cordes'—meaning 'it's raining hard'—is better translated as 'it's raining cats and dogs' in English rather than literally 'ropes.' However, finding an equivalent in other languages, such as Indonesian, remains a challenge for AI systems. Experts like François Yvon from CNRS highlight that while AI has improved significantly, it still faces hurdles in understanding context and idiomatic expressions.
The growing reliance on AI translation is evident in the scale of usage. Google Translate alone processes over one trillion words monthly, with other platforms like DeepL, Microsoft, and Gemini also contributing significantly. Direct sales of machine translation services reached $1.4 billion last year, indicating the widespread adoption of these technologies. John Zimmerman of OKA, a Swiss non-profit translating Wikipedia, notes that AI has made translation more efficient and cost-effective, particularly for internal documents and product listings.
Despite these advancements, human oversight remains crucial, especially in high-stakes scenarios. Ariella Baltié, a legal translator, encountered major errors in AI-translated court documents, such as misinterpreting legal terms and altering the meaning of phrases. Similarly, in the medical field, mistranslations can have life-threatening consequences. Carol Bereuter, a medical translator, pointed out an error where the Greek letter μ was incorrectly translated as 'milli' instead of 'micro,' leading to a potentially dangerous dosage error. These examples underscore the risks associated with relying solely on AI in critical contexts.
Zimmerman of OKA emphasizes that while AI has made translation more efficient, the quality of translations is still a concern. His organization's study found that 27% of AI-generated text required corrections before publication, with most errors stemming from the AI itself. Yvon adds that the probabilistic nature of AI translation makes achieving 100% accuracy challenging, especially in complex and context-dependent language. As AI continues to evolve, the balance between automation and human oversight remains a key consideration for industries relying on translation.