By the third session of my German course in Berlin, I was still trying to keep apart what sounded to me like several versions of “sh” and to understand what the dots had done to the vowels. Then the teacher put up a table with every present-tense form of a dozen common verbs. It even included some irregular ones. I was supposed to practice it at home, which I did a bit, but very little of it stuck. A middle-aged dude learning his fourth language? What do you expect?
It might not have been only a me-problem, though, as the rest of the class was not faring that much better. A bunch of supposedly motivated adults: a Brazilian software engineer, an Indian data scientist, a French hipster, and a Korean opera singer in training. Also, a retired Iranian lady, thanks to whom I did not win the distinction of being the oldest in the class. I have not told her yet that I was born in Iran, and somehow she has not put it together from my name yet. My name is a simplified form of a long Iranian name I settled on after becoming a US citizen and being tired of people mispronouncing my name. There is another American who always sits next to me and had also fled US tech to land a German tech job at a neobank. And amusing detail: English is enough for her neobank job, but if she wants to move to a legacy German bank, she needs C1-level German. (Neobanks are the kind of software company you try if you want to penetrate an archaic tech market like Germany’s, and you do not want to try too hard. Incumbent German banks are ancient and easy to beat, and rival American companies do not know how to navigate the scary German bureaucracy.)
The problem was not the teacher either. The teacher is a nice older German dude who maintains that heat does not bother him. Nonetheless, on a 33-degree day he looked at us soft foreigners, left the room, and came back with the maintenance dude, who turned on the air conditioning. Why does it take a maintenance dude to turn on the air conditioning in a Berlin language school? I wish my German listening comprehension were good enough that I could tell you.
The class was not one of the free state-funded ones. The reviews were scary and there was too much paperwork in German. The supposed modern book this class uses looks to me like an extended version of the little travel books tourists buy to learn a bit of a language. A classic language course starts by establishing the phonemes; this one is skill-based, organized by situation. My name is X. What do you do for a living? This is my family. The textbook’s index reads like a tour itinerary: introductions, hobbies, housing, health, festivals, getting around. The exercises in the book also kept using things we had not covered. Apparently we were supposed to work them out; this is called discovery and is considered good for us.
I had learned English very differently, in Tehran at the Iran Language Institute, the nationalized descendant of the Iran-America Society. We did not do much discovery. Every lesson there had the same architecture: a dialog, a text, and exercises. The dialog trained your ear and mouth; the text trained your eyes and vocabulary; the exercises used what you had already been given. I did not recognize the rule until decades later: nothing you had not seen before, with complexity rising in small steps. Grammar was a ladder you climbed once. A text that followed the rule was called a “graded text”. As far as I remember, the goal was not to be able to order coffee but to actually “learn” the language. In fact, when I ordered my first coffee in English, I struggled a bit, despite my relative competence in GRE essays. The coffee ended up not being to my liking. But hey, it didn’t kill me. Plus, the place where I ordered it probably could not have produced a coffee to my liking anyway. Turns out there was a coffee snob in me waiting to leave Iran too.
There was a piece of lore attached to the place. Our teachers said it was the method Americans used to train intelligence officers. It sounded like exactly the sort of story an Iranian teacher would make up to impress children who were just making fun of his belt behind his back. I remember the belt. It was less than a centimeter wide and, to us children, funny.
While researching this post, I remembered the claim and went looking. Turns out the poor guy had told the approximate truth. The trail led back to the Army Specialized Training Program in 1943, where linguists helped rapidly train thousands of soldiers for the war effort in languages including Japanese, German, and Russian. After the war, related methods continued at the Foreign Service Institute and the Defense Language Institute, and eventually the method traveled abroad through American binational centers of the kind that preceded the institute where I studied. In that setting, you could not fake competence for long. A wrong translation based on guesswork might cost lives. It made sense to me that a method built under those stakes would lean toward graded steps, nothing unseen, mastery pacing, and enormous contact hours.
So what happened to it? The old method saw language as habit formation; that was its theory. It also offered a bargain: the institution would make the sequence explicit, track what the student had learned, organize progress around mastery rather than the calendar, and accept some responsibility for whether the student arrived. But then Chomsky came along and helped discredit that theory by arguing that language was too creative to be explained as a set of habits learned through reward. When the theory died, the practice that had grown out of it was executed alongside it. The replacement, the communicative approach, won on the strength of a new theory without ever winning a head-to-head comparison against the old method.
The comparisons, where they happened, kept going the wrong way. Benjamin Bloom reported one-to-one tutoring with mastery feedback beating conventional classrooms by about two standard deviations, and classroom mastery learning alone beating them by about one. A later meta-analysis put mastery learning nearer half a standard deviation. Project Follow Through, at the time the largest and most expensive evaluation in education or the social sciences, found scripted, graded Direct Instruction to be the strongest on basic skills. Neither result remade the mainstream classroom. Paul Nation later put two numbers near my school’s house rule: unfamiliar words become easier to infer around 95 or 96 percent coverage; around 98 percent supports reliable unassisted comprehension. This was the coverage that graded texts tried to create and “authentic” discovery materials destroy. The method didn’t really lose in practice; it was simply refuted in theory and abandoned.
The method was even vindicated inside language teaching itself, under a new name, and was ignored there too. Canada’s French-immersion schools came close to the communicative ideal: children taught in the language for years, maximal authentic input, minimal explicit grammar. The result, documented across two decades of studies, was graduates who were fluent, confident, and persistently inaccurate, their grammar stuck at a level thousands of additional hours did not repair. By the nineties the field was quietly bolting explicit instruction back on under the deliberately modest label “focus on form”. Finally, the discipline audited itself when a meta-analysis by Norris and Ortega in 2000 showed explicit teaching outperformed implicit across the literature, including on delayed post-tests. That did not change much in terms of teaching and stayed an academic result; the coursebooks were in too deep.
A year or two before Berlin, I had heard the whole argument without really understanding it. My son did kindergarten and first grade at P.S. 321 in Park Slope, Brooklyn, and the school talked about reading constantly: phonemes, phonics, guided reading. One term was always in the past tense, always something now disavowed with an innocent-looking name: “balanced literacy”. I did not know what balanced literacy was. To me, it was parent-noise from a war I had not watched. What I also did not realize was that I had arrived at the war’s very end: I was hearing the vocabulary of the correction without ever having seen the thing corrected. I was just there because it was considered one of the better public schools in NYC, and because I, the progressive leftist idealist parent, naively believed in public education. And here is the part I only learned later. P.S. 321 was not just a good school that used balanced literacy. It sat unusually close to the movement’s center.
More than twenty-five years ago, Lucy Calkins of Teachers College brought the writing-process approach to the school’s teachers, and P.S. 321 became a longtime model site for her Reading and Writing Project. The school became the “model literacy school” that hosted visiting educators from everywhere, its teachers attending TC summer institutes and working alongside TC staff developers in their classrooms. Its longtime principal, Liz Phillips, after converting from book publishing to teaching, went on to mentor a network of Brooklyn principals herself, so the school exported both the method and the people who carried it. When Bloomberg mandated balanced literacy across most of the city in 2003, with Calkins as its leading advocate, P.S. 321’s stellar results made it a natural showcase. Never mind that an affluent Park Slope school where the Parent and Teachers Association can raise about a million dollars a year is exactly the population where almost any method would look like it worked.
What does it mean that the method worked? They meant the children did well on the state reading test. Those tests use unseen passages and ask children to make inferences and support them from the text. They could not tell whether a child had decoded a difficult word or used the rest of the sentence to get past it. The score still evaluated the child, not the method: it could not separate the workshop model from everything else the child had been taught at school or at home. This is the same problem as my language course. A certificate can show that I performed the defined tasks at level X. It cannot tell how much the course taught me, how much I brought in, or how much I repaired at home.
Then it all collapsed. The 2022 podcast Sold a Story exposed the damage done by the cueing practices embedded in Calkins’s curriculum; Teachers College transitioned the project into Advancing Literacy in 2023; and the school chancellor mandated that every New York City elementary school adopt one of three science-of-reading-aligned curricula starting in 2023–24, removing Calkins’s program as a core option in the city that made it famous. Phillips openly criticized the new mandate at parent meetings, and the school kept as much of the workshop model as the rules allowed. I had been nodding at newsletters from one of the movement’s model schools without knowing it.
My experience in the Berlin class is what finally made it click. Ken Goodman had described reading as a “psycholinguistic guessing game”, built on sampling cues and predicting meaning. Running on this idea, balanced literacy told children to guess unknown words from pictures and context, often pulling them away from decoding. When the sentence and the word on the page disagreed, poor readers had more trouble suppressing the guess and reading the word that was actually there. That is the child’s version of what my German course was doing to me. Guessing the word from the picture is discovery learning for seven-year-olds; the verb dump and the “authentic” exercises are three-cueing for adults. Both were the same idea: language is acquired, not taught; thus, we don’t teach it. The difference was that as an adult with a level of metacognition, I gave myself permission to be confused, then to feel insulted, and to decide that the book had failed to teach a prerequisite. An eight-year-old has much less authority to return the judgment.
What followed when this approach became policy at scale in the American school system was not pretty. California’s 1987 literature-based framework was widely implemented as whole language; by 1994 its fourth-graders were reading near the bottom of the national tables, and the state reversed course within a decade. Mississippi adopted the correction statewide and with unusual thoroughness. In raw fourth-grade reading scores, it went from second-worst in 2013 to 21st in 2022. Nationally, the standing audit repeated the verdict year after year: in the 2019 federal assessment, roughly a third of fourth-graders fell below the assessment’s “Basic” threshold, failing even to demonstrate partial prerequisite skills for solid reading.
Decades into this method, America has a mass literacy problem. The scale is not marginal and cannot be waved off as the cost of doing business. In the 2023 Survey of Adult Skills, 28 percent of adults scored at Level 1 or below: able to manage short texts, explicit cues, and mostly single-step tasks. More than half failed to reach Level 3, where readers can reliably integrate and evaluate information across longer texts. This is functional illiteracy at population scale. And, ironically, the approach’s undoing was not the accumulation of contrary findings; it was a podcast. Not because the podcast did new reading science; it just went viral and turned a specialist argument into a public one. New York City dropped Calkins’s curriculum in 2023, just a year or two before my son started first grade.
The children with the best chance of recovering are those whose families can buy the missing education: phonics tutoring, extra time, maybe a private school. And even then, a private school can use the same bad method, just as paying for my German class did not buy me an escape from communicative teaching. Zooming out, though, what you need to survive is just enough choices and compensating resources to route around whatever the school leaves out. Basil Bernstein named this in 1975: “invisible pedagogies”, diffuse and implicit learning opportunities, workable for children whose homes pre-supply the hidden knowledge, locked away for children whose homes do not. Lisa Delpit re-derived it from the other side: a process pedagogy can incidentally withhold the explicit codes of the culture of power from the children least likely to inherit those codes at home. Translation: the damage the method did was a function of your social class. Park Slope could afford the bad method for years because most of its families could route around it. Many families cannot.
And when a child struggles in the balanced-literacy classroom, they get a diagnosis: not a reader. The adult in the communicative classroom gets one too: not a language person. In both cases, the failure is returned to sender, which is what protects the method from scrutiny. Because the system was never really aiming at the child. It was firing into the air to hit its own requirements — material to cover, can-do lists, pacing, the exam date — and the children who got hit by the rounds coming down are recorded as the ones who couldn’t learn.
What selected this method, if not evidence? As usual, look at what it economizes. A syllabus of “authentic materials” outsources the expensive work of creating a graded corpus. “Discovery” outsources the sequencing to the learner and rebrands the omission as autonomy. A teacher demoted to “facilitator” is easier to treat as interchangeable. Calendar pacing — A1 in 8 weeks, in our case — fits cohorts and revenue planning in a way mastery pacing never can. And the CEFR’s can-do checklists convert the whole arrangement into clean deliverables: certificates, the one output verifiable at the point of sale. Even better, instruction is a credence good: the buyer cannot assess quality at purchase, the outcome arrives years later, the failure is attributed to the buyer, so the market competes on experience and credentials, never on learning. The sad part is that the pedagogy was not born neoliberal. It arose from a romantic ideal: the learner discovers, acquisition happens naturally, and instruction interferes. Then the neoliberal market came along and selected its institutional form. The teacher became a facilitator; sequence became coverage; mastery became calendar throughput; learning became a credential legible at the point of sale. The learner kept the obligation to succeed while the provider surrendered responsibility for teaching. The market did not care that the method was progressive. It cared that the institution’s output was legible and, more importantly, that the learner’s failure was private.
Germany’s contribution was to give the American creation legal force. Here, a language level like B1 is not merely a description of language ability; it is an administrative object wired into citizenship and residence law. The integration course is a state-procured market in which providers are reimbursed per participant and teaching unit; the certificates are administrative facts. People are not learning much? Do it harder! After the PISA shock at the beginning of the century, Germany expanded Kompetenzorientierung, codifying it in national educational standards. In other words, it doubled down on measurable competencies.
This is an odd turn for the country that gave us the Ausbildung, the apprenticeship system the whole world admires. Unlike the language class, the apprenticeship system is structured and externally examined. Its chamber exam committees have equal numbers of employer and employee representatives, the latter nominated by unions, plus at least one vocational-school teacher. The failures land on people who hire graduates or work beside them. If Ausbildung doesn’t work, it is not only the student who gets screwed; employers lose money too. The system can teach. It does teach, for the reproduction of its own workforce.
But we cannot afford all that for immigrants, can we? So we do educational standards and a more detailed list of can-dos. Let’s ask them for B2/C1: 57 percent of German establishments now require at least B2 when hiring. C1 is kind of hard to achieve, and a software engineer might not need it, but what can you do? We need to make sure the person is integrated. Let’s attach a moral judgment to it while we are at it. The certificate means integration; difficulty reaching C1 means insufficient will. Plus it is an efficiency gain. It is not cheap to interview people.
Any teacher will tell you that caring about your students and respecting them is an essential ingredient. It might sound like sentiment, but it is really just engineering. This is why the old rules read the way they do: “never show the student anything not built from what they’ve mastered”; “check mastery before advancing”; “build responsibility for failure into the design”. Put it all together and squint, and you are looking at what a caring teacher does one-on-one, industrialized. You want to encode this care as architecture, so the system doesn’t rely on every teacher to be a saint.
My current German teacher seems to care; he is the one who turned on the air conditioning even though he is supposedly immune to the heat, as Germans are or pretend to be. But a caring teacher inside a coverage system can only soften the blow; he must hit the exam date, use the book, deliver the verb dump. Germany’s public integration system fills many of the same teaching roles with freelancers paid by the teaching unit, most of them at the minimum possible rate. Its framework leaves lesson sequencing to teachers but fixes the modules, the hours, and the exam target. Care without authority over the calendar is admirable, but not very effective.
And who survives the careless method? People like me: privileged, with a previous language-learning success to steal structure from, plus money and time. And of course, there are ways to approximate the Tehran house rules while using the in-person classes as supplements. Just the other day I realized I can use a paid AI app called ElevenReader to read the graded texts that do still exist, at a pace I can follow, and I run my own loop through them.
Which brings me back to the joke of the day. A derelict, frozen-in-time language institute in Iran has taught me a foreign language arguably better than a modern private language school in Berlin is teaching me one now. And it was an accident. The revolution nationalized the Iran-America Society into the Iran Language Institute. Foreign institutions and trainers left; sanctions narrowed the commercial traffic in global coursebooks; the old graded sequence stayed, frozen like a Chevrolet in Havana. An anti-American revolution had preserved a piece of American state pedagogy by luck.
Maybe it is the nostalgia of a middle-aged immigrant. But I felt something in that Tehran classroom that I could not have named at fifteen and can name now: the method respected me. For all its defects, it cared whether I arrived. If I did the work and still fell short, I did not have to take that failure as proof that I was the problem; it was possible the design had failed too. I could work around the method and try again.
What I feel in the Berlin classroom is the opposite, and I think now I can tell you what it is. A structural distance and shallowness. A system that does not see me, one that treats the language levels and requirements as its clients rather than me. I was even pissed off enough to write a rather long post about it. Who doesn’t want to be seen?