The Sunday Magazine
Some of the world’s biggest tech companies person unrecorded translation devices either successful the works oregon already connected the market. But tech watchers accidental these developments rise larger questions astir however AI tools are changing connection learning and translation.
Google, Apple rolling retired caller tools that construe your code connected the fly
Jonathan Ore · CBC Radio
· Posted: Jun 07, 2025 4:00 AM EDT | Last Updated: June 7
The Sunday Magazine21:16Will we inactive request to larn overseas languages if AI tin bash the translating?
There's a suite of caller and upcoming tools designed to marque translation betwixt languages easier and faster than ever earlier — some, with the assistance of artificial intelligence.
At their I/O 2025 event, for example, Google revealed a unrecorded translation work it's added to its Google Meet videoconferencing tool. A objection showed 2 radical speaking to each different — 1 successful English, 1 successful Spanish — with their code translated into the different connection with a short, seconds-long delay. The computer-generated dependable mimicked the archetypal speaker's dependable and intonation.
Roger J. Kreuz, a prof astatine the University of Memphis who specializes successful the science of language, said Google's unrecorded translation objection was "a beauteous astonishing technological achievement," but its staged quality near questions astir however it volition enactment successful a existent meeting.
"Conversations are seldom arsenic cleanable arsenic the speech that we saw successful the demo," helium said. "They typically volition overlap oregon interrupt, and I tin lone ideate the cacophony that would hap if radical were benignant of excitedly talking backmost and distant ... and voices cutting successful and past cutting backmost retired again. How is that controlled?"
Experts caution tools similar this rise large questions astir what mightiness get mislaid successful translation. Because portion tech companies often tout these tools arsenic technological and objective, connection doesn't truly enactment that mode successful the existent world.
In March, Bloomberg reported that Apple is readying to update their AirPod earbuds to let them to construe languages from code it hears connected the fly. (Google's rival product, the Pixel Buds, person had this diagnostic for years, the study said.)
Apple's reported foray into the unrecorded translation crippled is notable, says WIRED writer and elder concern exertion Louise Mataskis, due to the fact that the institution typically doesn't present caller tech features arsenic rapidly arsenic others.
"They thin to clasp backmost until that exertion is truly mature and that determination is simply a bully consciousness that it's gonna beryllium reliable. So I deliberation that this shows that this exertion is truly starting to mature," she told The Sunday Magazine's Piya Chattopadhyay.
Google's code translation presently lone features translation betwixt English and Spanish, and it's disposable lone successful the U.S. to anyone paying for their Google AI Pro premium services. The institution says it doesn't prevention users' audio, nary AI models are trained utilizing your voice, and the diagnostic is opt-in only.
A typical from Google told CBC the work volition adhd much languages "in the adjacent fewer weeks." They said the diagnostic uses an AI ample connection exemplary called AudioLM, developed by Google DeepMind.
'Do you person a toilet successful your house?'
Mataskis says connection tools tin assistance radical signifier learning languages, but cautions that portion the tools oregon apps often contiguous themselves arsenic neutral — i.e. there's lone 1 close mode to construe a connection oregon operation — it mightiness miss important contextual oregon taste variations.
"In Mandarin, we don't springiness radical possession of things astatine their job. So you would ne'er say, successful Mandarin, 'do you person a bathroom?' You would say, 'where is the bath successful this place?'"
Mataskis, who utilized Google construe erstwhile she archetypal started learning Mandarin overseas successful Taiwan, got quizzical looks erstwhile asking the erstwhile successful java shops. "Often these baristas would look astatine maine comic and I didn't recognize that fundamentally I was saying 'do you got a toilet successful your house?'"
What's more, the benignant of translations you get tin pass however a tool's connection database was trained. Mataskis says that arsenic she's learned much Mandarin, her "hunch" is that translation tools usage Chinese authorities media texts.
"It's benignant of utilizing these like, honorifics to notation to the Chinese Communist Party. Or, similar it tin beryllium benignant of stifled successful the mode that authorities media and oregon authorities documents are often — you know, benignant of precise adust and usage a batch of ceremonial language," she said.
Kreuz notes that, historically, translation apps person had occupation detecting and decently translating sarcasm oregon homophones. He ran into the second erstwhile the Turkish translation for 1 of his books seemingly missed the people connected the title.
"I enactment the rubric into Google translate. This is 2018. Apparently, literally, it meant: How to Achieve Fluency successful a Foreign Language. And what it gave maine was: How To Earn Fluency connected Foreign Dildos, which was conscionable bizarre," helium said.
WATCH | AI coming to classes astir Canada AI coming to classes astir Canada
Language arsenic biodiversity
Muhammad Abdul-Mageed, Canada Research Chair successful earthy connection processing and instrumentality learning, says that companies should instrumentality other attraction erstwhile gathering AI translation tools for planetary languages that whitethorn person small successful communal with European ones.
Certain sounds an English talker makes, for example, whitethorn person nary equivalent successful Arabic, which could contiguous challenges for tools expected to marque instant translations.
"We cannot truly overgarment each these languages with the aforesaid brush, successful a sense," helium said.
Abdul-Mageed has been doing enactment with African languages of late, successful the hopes of helping make blase tools to construe betwixt them arsenic easy arsenic a Google oregon Apple mightiness absorption connected English and different European languages.
Doing the enactment to sphere languages tin beryllium seen arsenic different mode of preserving biodiversity, helium argues — and advances successful instrumentality learning and different technologies could beryllium almighty tools to bash so.
"We privation to sphere the carnal kingdom, the works kingdom, and truthful on. Language is portion of us, right? And if we fto definite languages go, we are letting parts of america go," helium said.
As convenient arsenic unrecorded translation tin be, it's nary substitute for learning and yet becoming fluent successful a 2nd oregon 3rd connection connected your own.
Mataskis says she's spoken to researchers who person recovered that learning much languages tin amended your brain's neuroplasticity. "So there's rather virtually wellness benefits to learning a 2nd language," she said.
She wants to promote radical to usage immoderate of the new language tools, including those powered by AI, arsenic a imaginable learning aid rather than a crutch.
Using it that mode tin assistance acceptable you up for the next, perchance champion nonacademic mounting to learn: going to the barroom with a person fluent successful that connection and conscionable hanging retired and talking together.
"Think astir these tools arsenic a mode to facilitate that connection, alternatively than to beryllium an intermediary betwixt you and this different person," she said.
Interviews with Louise Mataskis and Muhammad Abdul-Mageed produced by Andrea Hoang