Almost everyone has met this person, and in several large European countries this person is the majority. They studied English for eight years, or French for five. They sat an exam and passed it. Ten years later they can read a menu, recognise perhaps a third of a news broadcast, and cannot hold five minutes of conversation without the whole thing collapsing into apology.

The standard explanation is that school language teaching is stuck in the nineteenth century. Rows of desks, verb tables, a set text, a vocabulary test on Friday. Nothing has changed since our grandparents, so of course nothing works.

That explanation is almost exactly backwards, and it is worth taking apart, because the real answer is more interesting and points at things that could actually be fixed.

What actually changed, four times

The idea that language teaching never modernised would have surprised the people who spent the last century modernising it, usually while insisting the previous lot had ruined everything.

Grammar-translation, the method everyone pictures when they complain, was inherited from the teaching of Latin and applied to living languages in the nineteenth century. It had already been attacked as a scandal by 1882, when the philologist Wilhelm Viëtor published a pamphlet whose title was its whole argument: Der Sprachunterricht muss umkehren!, language teaching must turn around. Viëtor and the Reform Movement wanted the spoken language first, phonetics taken seriously, and connected texts instead of disconnected sentences about the gardener's aunt.

What followed was the direct method, which banned the mother tongue from the classroom entirely and was commercialised with great success by Berlitz. Then, out of American army language programmes in the 1940s, audiolingualism: pattern drills, habit formation, and the language laboratory, which arrived in the 1950s and 1960s with the same air of technological inevitability that software carries now. Then, from the 1970s, communicative language teaching, built on Hymes's argument that knowing a language means knowing how to use it appropriately, and on Wilkins's notional-functional syllabus, which organised courses around what a learner needs to do rather than around the grammar of the language. Then task-based learning. Then, from 2001, the Common European Framework, which quietly did something more radical than any of them by redefining a level as a list of things a person can do. Then content and language integrated learning, which teaches other subjects through the foreign language.

That is four or five complete overhauls in a hundred and forty years, roughly one per teaching generation, each sold as the end of the old way.

And the aggregate result, measured properly for the first time in 2011, was this. The European Survey on Language Competences tested almost 54,000 pupils across 16 education systems. In their first foreign language, 42% reached independent user level, B1 or B2. An identical 42% were at A1 or below, which is to say functionally at the beginning. In the second foreign language, only 25% reached independent user level and 58% were at basic level or lower (First European Survey on Language Competences, executive summary, European Commission, 2012).

The spread between systems was enormous and instructive. Malta and Sweden had 82% of pupils at independent user level in English. France had 14% in English. England had 9% in French.

So the method changed and the outcome did not. Which means the method was probably never the binding constraint.

The arithmetic nobody puts on the syllabus

Start with how much time it takes to learn a language, using the least controversial numbers available. Cambridge English publishes guided learning hours for its own exams, cumulative from a standing start: roughly 180 to 200 hours to A2, 350 to 400 to B1, 500 to 600 to B2, 700 to 800 to C1 (Cambridge English, guided learning hours). About 200 hours per level, in other words.

Now look at what school gives. Across Europe, foreign languages take between 5% and 10% of instruction time in primary school, which works out at roughly 30 to 69 hours a year in most systems, rising to between 10% and 19% in secondary (Key Data on Teaching Languages at School in Europe, Eurydice, 2023).

Take a reasonably generous secondary school: three lessons a week of 45 minutes, 38 weeks a year, so about 85 hours a year, six years of it, plus a few hundred hours in primary. Call it 700 hours. On Cambridge's scale, that lands somewhere around C1.

Almost nobody arrives anywhere near C1. So the hours are not real hours, and it is worth being precise about where they go.

Some of it is the ordinary friction of a school: settling the class, taking the register, handing back work, explaining the homework. Some of it is the mother tongue, in which instructions are given, grammar is explained and discipline is maintained. Some of it is writing, which is silent. And the part that is left, the part where a pupil actually produces the language out loud, has to be divided by the number of people in the room.

Suppose a very good teacher gets 20 minutes of pupil speech into a 45-minute lesson. In a class of 28 that is 43 seconds each. Three lessons a week is a little over two minutes. A school year is about 80 minutes of speaking. Six years is roughly eight hours.

Eight hours. Then we are surprised that people cannot speak.

There is one honourable escape from this arithmetic and communicative teaching found it decades ago: pair and group work. Twenty minutes in pairs is ten minutes of speaking each, not 43 seconds, and it changes the numbers by a factor of fourteen. This is exactly why the reform happened. It is also exactly what large classes, noise, room layout, mixed ability, an anxious inspection culture and an exam that does not measure speech all push back against. The method knew the answer. The building did not cooperate.

The survey that skipped speaking

There is a detail in that European survey which is worth sitting with. It tested reading, listening and writing. It did not test speaking, which was, in the report's own words, considered logistically difficult for this first round.

The largest and most careful measurement of school language competence Europe has ever attempted left out the one skill that the entire public complaint is about, because it was hard to organise.

That is not a scandal. It is a symptom, and it is the single most useful fact in this article, because assessment is the strongest lever any education ministry actually has. Teachers teach towards the exam. Not out of cynicism, but because that is what their pupils need from them, and because a school's reputation is built on results. If the exam consists of reading comprehension, a grammar section and a written text, then the exam has defined the subject, and the definition does not include talking.

Japan tried to fix precisely this and the attempt collapsed. The government planned to fold external commercial proficiency tests into the Common Test for university admissions, specifically to introduce a speaking component that the national exam lacked. It ran into what Butler, Lee and Peng call a mountain of criticism, largely on equity grounds: pupils differed in their access to the test centres and the expensive preparation industry around them, and the plan was shelved (Butler, Lee & Peng, 2022, English Today 38(4)).

The lesson there is not that speaking exams are impossible. It is that outsourcing them to a commercial market makes them politically fatal.

Excellent at maths, poor at languages

This is where the comparison people reach for instinctively turns out to be genuinely informative.

Japan came fifth in the world in PISA 2022 mathematics, with 536 points against an OECD average of 472, and it sits near the top of the table in reading and science as well. On any measure of whether a school system can teach a demanding subject to a whole population, Japan is at the top of the table.

In the EF English Proficiency Index for 2025, Japan ranked 96th out of 123 countries and territories, in the lowest of the five bands, below China at 86th and well below South Korea at 48th (Nippon.com, 2025). That index deserves a caveat, since it is built from people who choose to take a free online test rather than from a representative sample, and Japan's position in it has fallen partly as more countries joined. But nobody in Japan disputes the direction, and the country has been running reform after reform on exactly this basis for thirty years.

So a system that can teach mathematics to the top of the world cannot teach English to the middle of it. Aptitude is not the explanation, and neither is effort. Four things are.

Distance. English and Japanese are about as far apart as two major languages get, in writing system, sound inventory, word order and the entire apparatus of politeness. The US Foreign Service Institute's own estimates put Japanese in its hardest category for English speakers, at roughly three to four times the hours of French, and the difficulty runs in both directions. A Dutch teenager learning English is doing a structurally easier task than a Japanese teenager learning English, and no amount of pedagogy erases that.

Exams. Japanese entrance exams have historically rewarded reading, grammar and translation, which are markable at scale by people who are not themselves fluent speakers. The washback from that reaches down into every classroom in the country, and the attempt to change it failed for the reasons above.

Exposure. Japan dubs, and English has very little presence in the ordinary day. A Dutch or Swedish teenager meets English on television undubbed, in games, in music, in half the internet they use, and eventually in a university lecture or a work email.

Need. You can live an entire professional life in Japan without English. In the Netherlands you cannot, and everyone knows it at fourteen.

The Netherlands does not out-teach Japan by a factor of five. It out-exposes it, and it gives its pupils a reason. France, at 14% in that European survey, is a useful check on any smug reading of this: France dubs, has a large domestic culture in its own language, and does not need English to run daily life, and its results look far more like Japan's than like Sweden's. This is not about national character. It is about how many hours of a language a country's ordinary life hands out for free.

The subtitle experiment Europe ran by accident

Which countries dub and which subtitle was decided in the 1930s, for reasons that had nothing whatever to do with education: market size, the cost of dubbing relative to the size of the audience, and, in several countries, governments that preferred foreign films to arrive in the national language. That historical accident makes a rather good natural experiment, and economists have used it.

Comparing countries by translation mode, Rupérez Micola, Aparicio Fenoll, Banal-Estañol and Bris found a large positive effect of subtitled original-version broadcasting, as against dubbing, on national English proficiency, with the effect strongest in listening comprehension. Their conclusion was blunt: governments interested in foreign language skills should consider promoting subtitling (Rupérez Micola et al., 2019, Journal of Economic Behavior & Organization 158, 487-499).

It is a cross-country comparison and should be read as such rather than as a precise causal coefficient. But the mechanism is not mysterious, and it is the same mechanism as everything else in this article. Subtitling countries give their teenagers several hundred additional hours of comprehensible listening a year, at no cost to any ministry, with no timetable slot and no teacher.

"Why not simply hire native speakers?"

This is the question everyone asks, usually in a tone that suggests the answer is obvious and the education system is being obtuse. It deserves a serious answer, because it has four parts and only the last one is really about pedagogy.

There are not enough of them, by an order of magnitude. Germany alone employs tens of thousands of English teachers. Add France, Poland, Italy, Spain, Turkey, Brazil, Indonesia, China. The number of native English speakers who are also trained teachers, and are willing to spend a career in a small town on a local salary teaching thirteen-year-olds, is not a shortfall to be closed by better recruitment. It is a category error. And the argument reverses instantly: Germany could not staff the world's German lessons either, nor France its French, which is why almost every language on earth is taught, everywhere, mostly by people who learned it.

Being a native speaker is not a qualification. Speaking a language and being able to teach it are different skills, and the second one is the rare one. Robert Phillipson named the assumption that the ideal teacher is a native speaker the native speaker fallacy, and Péter Medgyes set out the compensating case in The Non-Native Teacher in 1994: the teacher who learned the language is a living model that success is possible, can teach learning strategies because they used them, can say more about how the language works, can anticipate where their pupils will get stuck, is generally more sympathetic when they do, and shares the pupils' first language. A French teacher of German who once fought the case system knows exactly which mistake is coming next week. A native speaker often has never consciously known the rule at all.

The big state experiments did not produce the jump. Japan's JET Programme began in 1987 with 848 assistants from four countries. It now runs at 5,734 participants from 53 countries and has placed 82,628 people in Japanese schools over nearly four decades (JET Programme history). Hong Kong has run its Native-speaking English Teacher scheme in primary and secondary schools for decades. Korea's EPIK scheme was expanded and then cut back by several provincial education offices from 2011 onwards on cost-effectiveness grounds. None of this is a controlled trial and the counterfactual is genuinely unknown, so the honest statement is narrow: forty years of a large, well-funded, well-run native-speaker programme did not visibly move a national proficiency figure. That is at least evidence that the person at the front of the room was not the binding constraint.

What native speakers are actually good for is real, and it is not what the argument assumes. They bring input that is not the teacher's input, a reason to speak to someone with whom the mother tongue is genuinely not an option, and a cultural context that no textbook carries. In team teaching, where the local teacher owns the pedagogy and the sequence and the visitor owns the interaction, that combination is worth having. As a replacement for a trained local teacher it is worth much less, and in hiring practice the preference for native speakers frequently ends up standing in for nationality and appearance rather than for teaching ability, which the applied linguistics literature has been documenting uncomfortably for thirty years.

The deeper problem with the question is that it locates the failure in the accent at the front of the room, when the failure is in the number of minutes each pupil speaks. Replace the teacher and the arithmetic is unchanged.

What the better systems actually do

They buy hours. That is the entire pattern, and it survives every case.

The Netherlands teaches other subjects in the language. As of October 2021, 134 of the country's 648 secondary schools ran an accredited Dutch-English bilingual programme, in which a substantial part of the curriculum is taught in English (Mearns, van Kampen & Admiraal, 2023). Pupils in those streams reach higher English proficiency, faster, than their mainstream peers, and the gains are largest at the start of the programme (Verspoor, de Bot & Xu, 2015, Journal of Immersion and Content-Based Language Education 3(1), 4-27). Just as importantly, the two fears people raise about this have not materialised: exam results in subjects like geography and history are not significantly different, and pupils do not score lower in Dutch.

There is a selection caveat that honest advocates state themselves. Bilingual streams tend to attract motivated pupils from supportive families, so the raw gap overstates what the programme itself causes. The effect is real; it is smaller than the headline.

The Nordic countries and Malta stack exposure on top of school. Subtitles, an early start, English-medium university teaching, English at work. By the time a Swedish pupil sits the test that put Sweden at 82%, school was one of several sources and possibly not the largest.

Luxembourg is the case that keeps everyone honest. It is the most deliberately multilingual school system in Europe: German as the main language of instruction, Luxembourgish alongside it, French from the second year. It is also a warning. Fewer than a third of primary pupils now speak Luxembourgish as a first language. Pupils from French-speaking or Portuguese-speaking homes consistently underperform in German and Luxembourgish listening regardless of socioeconomic background, and they do worse in mathematics as well, largely because mathematics arrives through German. Of the pupils who dropped out in 2023/2024, 84.1% had repeated a school year at least once. In July 2025 the government legislated to let families choose French-language literacy instead, from 2026/2027 (Education and Training Monitor 2025, Luxembourg).

The lesson is not that multilingual schooling fails. It is that teaching a subject through a language sorts children by which language they arrived with, and a system that does not plan for that will convert a language policy into an inequality.

And the reform that everybody adopted did the least. Starting earlier is now nearly universal: in most European systems every pupil begins a foreign language between the ages of 6 and 8, and six systems start at 3, 4 or 5. Yet between 2013 and 2020 the share of lower secondary pupils learning two or more foreign languages moved by 0.8 percentage points. Starting earlier without raising intensity mostly buys more years of low-intensity contact, which is the one thing the arithmetic above says will not work.

Three things that would actually change the outcome

1. Examine the thing you want people to be able to do

Speaking has to be in the school-leaving exam, weighted heavily enough that no department can quietly skip it.

This is the cheapest structural lever in education, because it reaches every classroom in a country at once without anyone having to be persuaded of anything. Teachers respond to assessment because assessment is what their pupils are judged on. As long as the exam is reading, grammar and a written text, every pressure in the system points away from the thing everyone says they want.

The objections are real. Speaking assessment is expensive, slow and awkward. The European survey could not manage it once, across sixteen systems, with a research budget. Marking it consistently is genuinely hard, and it is the format most vulnerable to accusations of unfairness, since a nervous pupil in front of an examiner is being tested on something other than language.

But it is done, in plenty of places, and the ingredients are known: internally conducted, externally moderated orals, recorded so that a sample can be re-marked, with double marking on borderline cases. Recording and remote marking have become cheap in a way that changes the cost picture entirely. The failure mode is Japan's: hand the speaking test to commercial providers and it becomes a question of who can afford the fee and who lives near a test centre, and it dies. Keep it inside the school system, or do not attempt it.

The second-order effect matters more than the exam itself. Once speaking is examined, pair work stops being a thing good teachers do when they can afford it and becomes the core of the lesson, because it is now the thing being trained. The 43 seconds becomes ten minutes. That is the whole game.

2. Buy hours, from other subjects and from the evenings

No minister can find three hundred extra language hours in a timetable that is already full, and every attempt to do it turns into a fight with mathematics, history and sport that the language department loses.

So take the hours from a subject you are already teaching, by teaching it in the language. That is what the Dutch bilingual streams do, and the Dutch evidence answers the two questions everyone asks first: subject knowledge does not suffer measurably, and the first language does not suffer either.

The honest constraint here is teacher supply, and it is severe. A geography teacher who can teach geography in fluent English is rarer than either a geographer or an English teacher, and cannot be conjured by policy. This is the real reason CLIL is a stream in 134 schools rather than the design of all 648. Which suggests the sensible version is not a national mandate but a national qualification path: pay for serving subject teachers to reach the level and be certified, and treat it as a career track rather than a favour.

Then there is the free half, which costs no teacher at all. Subtitling has a measurable effect on national listening ability and requires a broadcasting rule rather than a budget line. The modern caveat is that a policy about broadcast television is increasingly a policy about a medium teenagers do not use. The current equivalent is defaults: what a streaming service does when a household selects a country, whether the original audio is one tap away or four, and whether anyone has ever shown a fourteen-year-old how to change it. That last one is not a policy at all. It is a thing a teacher can do on a Tuesday afternoon, and it is worth more than the lesson it interrupts.

3. Change what school promises to deliver, and hand over the tools

Nothing above closes the gap to fluency, and pretending otherwise is how this subject keeps failing publicly.

Put the numbers side by side. Reading a novel without constant help takes something like 8,000 to 9,000 word families, and following ordinary spoken language takes around 6,000 to 7,000 (Nation, 2006, Canadian Modern Language Review 63(1), 59-82). Vocabulary size at B2 sits at roughly 3,250 to 3,750 of the most frequent 5,000 words on Meara and Milton's scale, and most school leavers are well below that. School realistically delivers a third of the requirement, at the exact moment its pupils stop having lessons.

Calling that failure is unfair to teachers who did what forty-five minutes allowed. Calling it completion is dishonest to pupils. The useful move is to redefine the deliverable: a language education produces a solid core plus a working method for continuing, and the method is examinable content, not a motivational poster in week one.

That means teaching, explicitly: how to keep a record of the words you personally meet, how spaced repetition works and why cramming does not survive the summer, how to mine a film or a podcast or a contract for language instead of merely consuming it, how to find and keep a conversation partner, and how to read your own plateau when it arrives. None of this is soft. It is the difference between a subject that ends at eighteen and one that compounds for forty years.

It is also the part of the problem we work on, and it is deliberately a small part: Vokabulo exists for the several thousand words that belong to one person's life rather than to any syllabus, kept with the context they were met in and returned on a spaced repetition schedule. It cannot teach the grammar skeleton, it cannot run a conversation class, and it is no substitute for a teacher. This article is largely an argument that the teacher's constraints were never the teacher's fault.

So, has it changed?

Both halves of the question turn out to be true, which is why the argument never resolves.

The doctrine has been rebuilt four or five times since 1882, each time by people who were right about what was wrong with the last version. And the lesson still looks much as it did, because what a language lesson looks like is set by forty-five minutes, twenty-eight chairs, one adult and an exam that measures reading. No method survives contact with those four things, and every method has been asked to.

The countries that do better did not find a better method. They found more hours: in other subjects, on television, in the street outside, in the plain fact that a Dutch teenager will need English and knows it. And where those hours are not lying around to be picked up, they have to be bought deliberately, or the outcome will be what it has been for a century, which is a nation of people who studied a language for eight years and apologise for not speaking it.

More on the hours problem in why so many people fail with language apps, on what the levels in this article actually mean in understanding CEFR language levels, on the point where school leavers get stuck in breaking the intermediate plateau, and on picking this up decades later in learning a language in later life.

Sources: First European Survey on Language Competences, European Commission, 2012. Key Data on Teaching Languages at School in Europe, Eurydice, 2023. Cambridge English guided learning hours. OECD, PISA 2022. EF English Proficiency Index 2025, via Nippon.com. Rupérez Micola, Aparicio Fenoll, Banal-Estañol and Bris, Journal of Economic Behavior & Organization, 2019. Butler, Lee and Peng, English Today, 2022. Verspoor, de Bot and Xu, Journal of Immersion and Content-Based Language Education, 2015. Mearns, van Kampen and Admiraal, CLIL in the Netherlands, 2023. Education and Training Monitor 2025, Luxembourg. JET Programme history. Péter Medgyes, The Non-Native Teacher, 1994. Robert Phillipson, Linguistic Imperialism, 1992. I.S.P. Nation, 2006. Meara and Milton, vocabulary size and the CEFR.