Why Traditional Language Learning Methods Don’t Work

By the British Council’s estimates, well over a billion people are learning English around the world, and many more are studying other languages. The overwhelming majority will never reach comfortable, conversational fluency. I don’t believe that’s because they lack talent, discipline or motivation. I think it’s because the method almost all of them are using was never designed to work.

I want to explain what that method is, where it came from, and why it fails — because once you see it, you can’t unsee it.

The “code model”

Nearly all language teaching rests on an assumption linguists call the code model: a language is a code. Words are tokens, grammar is the rule set, and if you memorise enough tokens and learn the rules, you can encode your thoughts into the language and decode other people’s back into your own.

This is the logic underneath almost every textbook, course and app on the market. Learn the vocabulary. Study the grammar. Practise encoding and decoding. Take a test. It sounds completely reasonable. It is also wrong.

Where it actually came from

The grammar-translation method wasn’t designed by people studying how humans learn to speak. It grew out of how Latin and Greek were taught in 18th- and 19th-century European schools — dead languages, taught for reading and mental discipline, to large classes, by teachers who often couldn’t speak them either.

It survived for one reason: it was convenient to administer. It could be delivered to big groups, examined with written tests, and standardised across schools. It was built for institutional efficiency, not for the human brain. And yet this is still, in its bones, how most of the world teaches living languages today — even when it’s dressed up with streaks, points and an app icon.

Why it fails

Your brain doesn’t decode language. When you hear your own language, you don’t identify words, look up meanings, parse the grammar and assemble the message. You just understand — instantly, effortlessly, whole. That’s because your brain runs on patterns of meaning, not rules. You don’t know that “I goed to the shop” is wrong because you memorised an irregular verb; you know because it sounds wrong. That’s pattern recognition, not rule-following.

Comprehension comes before production. Every human being understood their first language before they could speak it, and spoke it before they knew a single rule. Meaning leads; structure follows. The code model inverts the natural order and then wonders why it’s so slow.

Translation is a bottleneck. The code model trains you to translate in both directions — think in your language, encode, speak; hear, decode, understand. It’s slow, effortful and fragile, and it’s the opposite of fluency. Fluent speakers don’t translate; they think in the language.

Tests measure the wrong thing. Grammar-translation is very good at producing people who can conjugate a verb on paper, and not good at producing people who can follow a film or hold their nerve in a meeting. Test performance and real communication are different skills, built by different processes — and the method optimises for the one that doesn’t matter in real life.

The scale of it

Step back and the waste is staggering. Enormous sums of money, and years of human effort by some of the most motivated people on earth, poured into a process that mostly doesn’t deliver what it promises. Hundreds of millions of people have a language app on their phone. Most of them are learning about a language rather than actually acquiring it — and many can feel the gap, even if they can’t name it.

What replaces it

The answer isn’t a tidier version of the code model. It’s a different approach altogether — one built on how the brain actually acquires language: through comprehensible input, pattern recognition, physical training, emotional safety and real, meaningful use.

That’s what I laid out in my original TEDx talk — now watched more than 37 million times — and it’s what the five principles and seven actions describe in practice.

I take the comparison much further in my newest TEDx talk, recorded in Hong Kong, which directly contrasts the two models — the old code model versus a pattern-recognition, meaning-based approach — and shows what changes when you switch from one to the other. It’s due to be published by TED later in 2026. Read more about it here.

I don’t think 37 million people watched that first talk because it was clever. I think it’s because it put words to something a billion frustrated learners already suspected: it shouldn’t be this hard.

It shouldn’t. And it doesn’t have to be.