New AI tools promise real-time translation so you don't have to. But is that a good thing?
There's a suite of new and upcoming tools designed to make translation between languages easier and faster than ever before — some, with the help of artificial intelligence.
At their I/O 2025 event, for example, Google revealed a live translation service it's added to its Google Meet videoconferencing tool. A demonstration showed two people speaking to each other — one in English, one in Spanish — with their speech translated into the other language with a short, seconds-long delay. The computer-generated voice mimicked the original speaker's voice and intonation.
Roger J. Kreuz, a professor at the University of Memphis who specializes in the psychology of language, said Google's live translation demonstration was "a pretty amazing technological achievement," but its staged nature left questions about how it will work in a real meeting.
"Conversations are rarely as clean as the conversation that we saw in the demo," he said. "They typically will overlap or interrupt, and I can only imagine the cacophony that would occur if people were kind of excitedly talking back and forth ... and voices cutting in and then cutting back out again. How is that controlled?"
Experts caution tools like this raise big questions about what might get lost in translation. Because while tech companies often tout these tools as scientific and objective, language doesn't really work that way in the real world.
In March, Bloomberg reported that Apple is planning to update their AirPod earbuds to allow them to translate languages from speech it hears on the fly. (Google's rival product, the Pixel Buds, have had this feature for years, the report said.)
Apple's reported foray into the live translation game is notable, says WIRED journalist and senior business editor Louise Mataskis, because the company typically doesn't introduce new tech features as quickly as others.
"They tend to hold back until that technology is really mature and that there is a good sense that it's gonna be reliable. So I think that this shows that this technology is really starting to mature," she told The Sunday Magazine's Piya Chattopadhyay.
Google's speech translation currently only features translation between English and Spanish, and it's available only in the U.S. to anyone paying for their Google AI Pro premium services. The company says it doesn't save users' audio, no AI models are trained using your voice, and the feature is opt-in only.
A representative from Google told CBC the service will add more languages "in the next few weeks." They said the feature uses an AI large language model called AudioLM, developed by Google DeepMind.
'Do you have a toilet in your house?'Mataskis says language tools can help people practice learning languages, but cautions that while the tools or apps often present themselves as neutral — i.e. there's only one right way to translate a word or phrase — it might miss important contextual or cultural variations.
"In Mandarin, we don't give people possession of things at their job. So you would never say, in Mandarin, 'do you have a bathroom?' You would say, 'where is the bathroom in this place?'"
Mataskis, who used Google translate when she first started learning Mandarin abroad in Taiwan, got quizzical looks when asking the former in coffee shops. "Often these baristas would look at me funny and I didn't realize that basically I was saying 'do you got a toilet in your house?'"
What's more, the kind of translations you get can inform how a tool's language database was trained. Mataskis says that as she's learned more Mandarin, her "hunch" is that translation tools use Chinese state media texts.
"It's sort of using these like, honorifics to refer to the Chinese Communist Party. Or, like it can be sort of stifled in the way that state media and or government documents are often — you know, sort of very dry and use a lot of formal language," she said.
Kreuz notes that, historically, translation apps have had trouble detecting and properly translating sarcasm or homophones. He ran into the latter when the Turkish translation for one of his books apparently missed the mark on the title.
"I put the title into Google translate. This is 2018. Apparently, literally, it meant: How to Achieve Fluency in a Foreign Language. And what it gave me was: How To Earn Fluency on Foreign Dildos, which was just bizarre," he said.
Muhammad Abdul-Mageed, Canada Research Chair in natural language processing and machine learning, says that companies should take extra care when building AI translation tools for international languages that may have little in common with European ones.
Certain sounds an English speaker makes, for example, may have no equivalent in Arabic, which could present challenges for tools expected to make instant translations.
"We cannot really paint all these languages with the same brush, in a sense," he said.
Abdul-Mageed has been doing work with African languages of late, in the hopes of helping develop sophisticated tools to translate between them as easily as a Google or Apple might focus on English and other European languages.
Doing the work to preserve languages can be seen as another way of preserving biodiversity, he argues — and advances in machine learning and other technologies could be powerful tools to do so.
"We want to preserve the animal kingdom, the plant kingdom, and so on. Language is part of us, right? And if we let certain languages go, we are letting parts of us go," he said.
As convenient as live translation can be, it's no substitute for learning and eventually becoming fluent in a second or third language on your own.
Mataskis says she's spoken to researchers who have found that learning more languages can improve your brain's neuroplasticity. "So there's quite literally health benefits to learning a second language," she said.
She wants to encourage people to use any of the new language tools, including those powered by AI, as a potential learning aid rather than a crutch.
Using it that way can help set you up for the next, possibly best nonacademic setting to learn: going to the bar with a friend fluent in that language and just hanging out and talking together.
"Think about these tools as a way to facilitate that connection, rather than to be an intermediary between you and this other person," she said.
cbc.ca
