Email)">
AI translation tools have limitations in accuracy, especially for less common languages, so there's a risk of misinterpretations.
While AI can be helpful, professional translators are crucial for ensuring accuracy, cultural sensitivity, and understanding nuances.
This trend means that the public sector must be putting language accessibility at the top of its communications priorities. Translating material takes time and adds cost to already strapped budgets. Many are turning to Artificial Intelligence (AI) tools to get it done, with mixed results.
Raftelis has been offering clients translation services for several years now, by subcontracting with a professional translation company, UNO Translations and Communications, LLC (UNO). UNO Founder and CEO, Brigitta Toruño, started doing a deep dive into the use of AI in language translation in early 2024. She was skeptical about it, unsure whether it would be a good idea to use as part of her services, and frankly fearful that it could replace the very personal and specialized approach she offers with real people, certified to do translations. Her skepticism of AI translation tools is warranted, given the many ways AI translations can go wrong.
For one thing, the accuracy is not quite there yet. Though Google Translate continues to improve, research has shown the app is inaccurate in about 10-20% of translations for common languages such as Spanish, Korean, and Chinese. Less common languages, such as Farsi (67%) and Armenian (55%) have lower accuracy levels, likely because the AI models are built in English. In addition to accuracy issues, this means AI might struggle with nuances, grammatical structures, idioms, and humor.
In 2017, Facebook learned this shortcoming the hard way, with an infamous translation that got a man arrested. The man posted a picture of himself next to a bulldozer at a construction site. His post said “yusbihuhum” which is Arabic for “good morning,” but Facebook’s AI translation tool translated the message as “hurt them” in English and “attack them” in Hebrew. Thankfully, the man was cleared, and Facebook issued an apology.
AI will also struggle with language when words have multiple meanings. This can render translations that are far from the intention of the original text, which can have very serious consequences. For example, in the same Google Translate study, it was learned that Coumadin, a common drug used for blood clots, was translated as soybeans in Chinese. The sentence, “Do not take any more soybean until you see your doctor,” doesn’t adequately warn the patient about their Coumadin intake.
Toruño has spent much of 2024 building her understanding of AI translation tools and how to best use them. While she’s optimistic about AI’s potential, she believes that today, “No one should rely on AI alone for translation. You still need a person who speaks that language to review the translation.”
Toruño believes that human oversight and consultants with strong cultural competency are vital to the successful use of AI translation tools. Though Toruño’s team uses a three-pronged quality control process that involves the translator, the proofreader, and the project manager, she sees value in using AI in some cases. Going forward, her team will use AI translation tools to conduct reviews of their work to help identify gross errors and omissions. Her team will use AI with lots of human oversight and be testing the efficacy of the AI tools in the coming months to see whether they agree or not with the suggested changes.
In terms of cultural competency, Toruño is concerned about AI translation tools’ ability to deliver. Cultural competency, also called cultural humility, is the notion of being aware and sensitive to the audience that will be taking in the information – and delivering a translation that will be understood, not only from a technical language perspective but also from a cultural nuance perspective. This is an integral part of the work a firm like Toruño’s does. It’s what sets them apart from the products that AI can deliver. Human translators deliver translations that get to the heart of the reader, while AI only provides a literal translation, which may also have errors in it.
Toruño says, yes, but with caution. If it’s simple messaging in the more common languages with higher accuracy levels, it may be okay. For larger documents, more complicated messaging, or less common languages, it’s best to use a professional translator or a combination of the two, so you can check the accuracy and cultural sensitivity of the tool’s suggested translation.
“I can see the value AI will bring to translations, but I trust my own linguists more,” says Toruño.
If you have questions about translations or need some translation or interpretation assistance, contact Sam Villegas at svillegas@raftelis.com or Brigitta Toruño at btoruno@unotranslations.com.
