--- title: "Machine Translation Struggles with Idiomatic Expressions" url: https://noti.group/machine-translation-struggles-with-idiomatic-expressions/ language: en publisher: "Noti Group" section: "Technology" published: 2026-10-13T23:05:29.000Z updated: 2026-10-14T07:47:17.360Z id: ece76c81-9e82-411f-834a-36eb5909b4d8 source: "BBC News Technology https://www.bbc.co.uk/news/articles/c1wxdjx27ypo?at_medium=RSS&at_campaign=rss" attribution: "Link to https://noti.group/machine-translation-struggles-with-idiomatic-expressions/ and name Noti Group when you quote or summarize this story." --- # Machine Translation Struggles with Idiomatic Expressions The reliance on machine translation raises concerns about its limitations and potential consequences due to struggles with idiomatic expressions. The increasing reliance on machine translation has sparked concerns about its limitations and potential consequences. While AI-powered translators have made significant strides in recent years, they still struggle to accurately convey idiomatic expressions and nuances of language. A French idiom, "il pleut des cordes, which literally translates to it's raining ropes, is a classic example of the challenges faced by machine translation. The phrase is often used to describe heavy rain, but its literal meaning can be confusing for automated translators. According to François Yvon, a research director at the French National Centre for Scientific Research CNRS, well-trained AI systems may recognize that ropes are not falling from the sky and opt for a more idiomatic translation like it's raining cats and dogs. However, this still raises questions about the ability of machine translators to adapt to different languages and cultures. Yvon notes that even with advanced training, AI systems may struggle to find equivalent expressions in other languages, such as Indonesian. The demand for machine translation is growing rapidly, with Google announcing that its Translate service handles over one trillion words per month. Other companies like DeepL, Microsoft, Claude, ChatCGT, and Gemini are also investing heavily in this technology. The market for direct sales of machine translation services alone was estimated to be worth $1.4 billion last year. Industry experts say that while AI-powered translators have improved significantly, there are still situations where human intervention is necessary. John Zimmerman, founder of the Swiss non-profit OKA, argues that multiplying your costs by 100 to catch the last few percent of errors may not be rational for certain use cases, such as translating internal documents or product listings. The efficiency of artificial intelligence (AI) in translation has led to a significant reduction in human input for tasks like checking and editing translated content. This is evident in the work done by OKA, a Swiss non-profit organization that has contributed over 10,000 articles to Wikipedia, mostly in Spanish, Portuguese, and English. The sheer volume of translations made possible by AI would have been unaffordable using traditional methods, with costs running into millions of dollars. Instead, the AI produces a draft, which is then reviewed by human translators for accuracy and quality. This hybrid approach has enabled OKA to achieve its impressive output. However, the reliance on human input is decreasing as machine-generated translations improve in quality. According to John Zimmerman, founder of OKA, the output from AI tools like DeepL is now good enough that skipping human review altogether is becoming a tempting option for many translation companies. For instance, Ariella Baltié, a translator and legal expert specializing in French-language court documents, was recently asked by someone who had used an AI tool to translate important documents. Upon reviewing the output, she found significant errors at the beginning of the document. The implications of this trend are far-reaching, raising questions about the future role of human translators in the industry. As Zimmerman notes, the question is no longer whether AI translation is accurate, but rather if it's good enough to replace human input altogether. A critical case of mistranslation occurred when Le demandeur" was translated as "the requester, rendering it awkward for legal documents. The AI's interpretation of Madame X, épouse Y resulted in a translation that read Ms X marries Y, completely misrepresenting the original phrase by confusing a noun with a verb. Mistranslations in the legal sphere can have severe consequences. In the US, errors in machine-translated asylum applications have led to delays in releasing detainees from custody. The stakes are even higher in medical contexts where accuracy is a matter of life and death. Specialist Carol Bereuter notes that many medical terms include Greek letters, such as those used for dosages and blood tests. Bereuter recalls an instance where an automatically-translated text misinterpreted the Greek letter μ as milli" instead of "micro, which could result in a patient receiving a thousand times their intended dose. The reliability of machine translation systems is a pressing concern for those who rely on them for critical decisions. According to experts, even straightforward sentences can be misinterpreted by AI services. This highlights the limitations of machine learning algorithms in understanding context and nuances of language. In particular, words with multiple meanings can cause problems for AI translators. They may struggle to determine which meaning is relevant, leading to errors that can have significant consequences. For instance, a sentence like translate into Polish might be misinterpreted as referring to shoe cream rather than the actual translation task. Context plays a crucial role in determining the accuracy of machine translations. If an AI translator fails to grasp the context of a sentence or conversation, it may produce incorrect results. This is evident in phrases like taking a corner," which can refer to either a football maneuver or a geographical location. Without proper contextual understanding, AI translators are prone to misinterpretation. A recent study conducted by OKA and funded by Wikimedia sheds light on the extent of machine translation errors. The research involved analyzing 119 articles, 10 language pairs, and over 1,000 hours of translator work. The results showed that a significant proportion of AI-generated text required correction before publication, with many errors being introduced by the AI itself. The challenge of eradicating these errors is substantial, according to experts. As things stand, the fundamental design of machine translation systems makes it difficult to envision how they could become 100% reliable. This raises important questions about the role of humans in translation and whether we will ever be able to rely solely on AI services for critical tasks. The current limitations in AI translation stem from its reliance on probability rather than absolute certainty. In language, nuance and subtlety can be difficult to capture, making it challenging for AI systems to always produce accurate results. A new generation of technology is expected to improve upon this by incorporating multiple AI tools that work simultaneously. One tool would handle the translation itself, while a second would critique the output and a third would generate a revised version, allowing for more precise and reliable translations. Ultimately, the biggest breakthrough will come when AI systems can indicate their own uncertainty or doubt, enabling humans to rely on them with greater confidence. --- Source: [BBC News Technology](https://www.bbc.co.uk/news/articles/c1wxdjx27ypo?at_medium=RSS&at_campaign=rss) Published by Noti Group: https://noti.group/machine-translation-struggles-with-idiomatic-expressions/