Picture a street in Tokyo. You do not speak a word of Japanese, but a woman asks you for directions and you understand her, because the earbuds you already own translate her into English a moment after she speaks. You answer, and your phone speaks for you.

This is not science fiction any more. It is a feature you can switch on today. It is not perfect, but it is good enough to make a question that language teachers have been dreading sound perfectly reasonable: why would anyone still spend six years learning French?

The question deserves an honest answer, and the honest answer is not "nothing has changed".

Yes, the machine has already taken part of the job

It is tempting to wave the question away. The numbers do not allow it.

Researchers at Oxford looked at hundreds of regions across the United States. Where machine translation caught on faster, jobs for translators grew more slowly. So did job ads asking for Spanish, Chinese or German. The effect is still moderate, and the researchers expect it to grow.

So one reason to learn a language, the purely practical one of getting things done, is shrinking. Ordering food, reading a letter from the tax office, answering an email from a supplier: a machine can already do most of that for you.

Anyone who tells you otherwise is selling something. The more interesting question is what is left once that part is gone.

School was never going to make you fluent anyway

Before we mourn the old way of learning languages, it is worth remembering how well it worked. Not very.

The US government's own language school for diplomats estimates that an English speaker needs 600 to 750 hours of full-time class to reach a professional level in Spanish or French. For Japanese, Arabic or Chinese it is about 2,200 hours. A school career offers a few hundred hours at best, shared with thirty other people.

Ireland shows what happens when school carries a language on its own. Irish has been a compulsory subject for about a century. In the last census, around 1.9 million people said they could speak it. Only about 72,000 actually use it every day outside school.

Most people who "failed" at languages never had a real chance. They simply never got the hours. We looked at why school lessons have barely changed in a hundred years in a separate article.

So the machine is not replacing a system that worked. It is replacing a system that mostly left people with a few phrases and the private belief that they have no talent for languages.

AI makes practice easier, and that can be the problem

If AI can translate, surely it can also teach? Partly. But one experiment should make everyone pause.

At a high school in Turkey, about a thousand students practised maths with the help of a chatbot. Those who could use it freely solved far more practice exercises than their classmates. Then came a test without the chatbot. They did worse than the students who had never used it.

A second group got a more careful version that only gave hints. They also did much better during practice. On the test, they did no better than anyone else.

The lesson carries straight over to languages. Learning happens in the moment of effort: the pause before you find the word, the sentence you have to build yourself. A tool that removes that effort makes practice feel smooth and leaves little behind. The struggle was not a flaw in the process. It was the process.

AI is useful where it saves you effort that teaches nothing, such as deciding which word to review today or finding an example sentence. It becomes risky when it takes over the part that should stay hard.

A translation carries the words and drops the person

Here is what the earbuds do not tell you. A translation gets the basic message across and loses almost everything around it.

Jokes rarely survive. Puns almost never do. Many languages have several ways of saying "you", depending on how close you are to someone, and machines regularly pick the wrong one. Irony, warmth, hesitation, the polite no that really means no: all of that gets flattened.

Every time that happens, a system is deciding on your behalf how formal you are being, whether you are joking and who you are talking to. That is not just translation. It is someone else editing what you say.

There is a famous quote about this, usually attributed to Nelson Mandela: "If you talk to a man in a language he understands, that goes to his head. If you talk to him in his language, that goes to his heart."

Mandela never wrote it. The line first appeared in the foreword of an American book in 1996. What he actually said, in a conversation around 1992, was more specific: when you speak English, many people understand you, including Afrikaners, "but when you speak Afrikaans, you go straight to their hearts".

The real version is more interesting. Afrikaans was the language of the regime that imprisoned him. Mandela learned it on purpose, to understand the people on the other side of the table. For him, a language was not only about warmth. It was about understanding, and about power.

Being understood is not the same as being spoken to

If that sounds too romantic, here is a hard number.

In Ontario, Canada, researchers followed almost 190,000 older patients who were admitted to hospital. Among patients whose first language was neither English nor French, those treated by a doctor who spoke their language had roughly half the odds of dying in hospital.

Nobody is suggesting the other doctors did not care. But a patient who can describe exactly where it hurts, in their own words, to someone who understands every nuance, is a different patient.

In 2050 there will almost certainly be a translation app in every consulting room. The question is whether that will ever be the same thing.

Languages can disappear quietly

There is a larger story behind all this. When everyone can get by through one big language and a machine, fewer people have a reason to keep a small one alive.

A study of more than 6,500 languages found that over a third are already under threat. Without deliberate effort, at least 1,500 could stop being spoken by the end of this century. One finding surprised the researchers: more years of schooling went hand in hand with more endangerment, possibly because children are taught in a dominant language and stop using the one they grew up with at home.

That also means school is a lever. If schooling can push a language out, it can help keep one alive.

In 2050, speaking a language becomes a choice, and choices say more

So what does language learning look like in 2050? Probably two things at once.

Everyday communication will be almost free. Menus, forms, directions, a first draft of a contract in another language: machines will handle that, first for the big languages and then for the rest. Nobody will need to learn Italian just to order a coffee in Rome.

Actually speaking a language will be worth more. When translation is the default, choosing to speak someone's language becomes a signal. It says: I made the effort, I want to understand you properly, not just roughly. Signals are worth more when they are rare.

Think of the doctor who speaks your language. The negotiator who laughs at the joke before the earbud has translated it. The colleague who understands what was left unsaid, because in that language it is always left unsaid.

There is also a risk. The English-speaking world could stop learning languages altogether, because the machine does it for them, while everyone else keeps doing the work of meeting them halfway. That would not be a technological outcome. It would be a choice, made by default.

Which leaves a question worth asking yourself. If nobody forced you, if no exam and no job required it, which language would you still want to speak? And what would you want to be able to say in it?

Where Vokabulo fits

Most of what this article describes is far bigger than any app. No app will fix school hours, save endangered languages or replace a year abroad.

What Vokabulo does is smaller: it keeps the words that belong to your own life, the ones from your job, your conversations and the series you are watching, together with the sentence you found them in. And it brings them back at the right moment so they stay with you. That is the part of speaking a language that a translation earbud cannot do for you. Vokabulo is free to start.

More on the hours problem in why so many people fail with language apps, on what the levels mean in understanding CEFR language levels, and on starting again decades later in learning a language in later life.

Sources

  • Frey, C. B., & Llanos-Paredes, P. (2025). Lost in translation: Artificial intelligence and the demand for foreign language skills. Oxford Martin School. oxfordmartin.ox.ac.uk
  • US Department of State, Foreign Service Institute. Foreign language training: language learning time estimates. state.gov
  • Central Statistics Office Ireland (2023). Census 2022, Profile 8: The Irish language and education. cso.ie
  • Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2024). Generative AI can harm learning. Working paper. papers.ssrn.com
  • de Galbert, P. (n.d.). On the Mandela quote about language and the heart. Harvard University. scholar.harvard.edu
  • Seale, E., Reaume, M., Batista, R., et al. (2022). Patient–physician language concordance and quality and safety outcomes among frail home care recipients admitted to hospital in Ontario, Canada. CMAJ, 194(26), E899-E908. cmaj.ca
  • Bromham, L., Dinnage, R., Skirgård, H., et al. (2022). Global predictors of language endangerment and the future of linguistic diversity. Nature Ecology & Evolution, 6, 163-173. nature.com