For three months, you can feel yourself improving. For the next three, you cannot. You are still studying. You are still showing up. You open a book in your target language, read a page, understand most of it, and put it down feeling exactly where you were half a year ago.
If this has happened to you, you are not unmotivated, and you are not bad at this. The wall you have hit has a name, and a date, and a research paper that predicted you would hit it. Almost no one outside of academic linguistics has heard of any of them.
The level the CEFR literally calls 'Threshold'
The Common European Framework of Reference for Languages — the six-level scale, A1 through C2, that most schools and apps use to describe your level — gives the levels official labels almost no one quotes. A1 is Breakthrough. A2 is Waystage. B1 is Threshold. B2 is Vantage. C1 is Effective Operational Proficiency. C2 is Mastery.
The label for B1 was chosen carefully. At B1 you have been let into the language. You can hold a conversation about something concrete. You can follow the main points of clear speech on familiar topics. You can write a connected, simple text about something you know. You cannot do any of this well yet, and you cannot do it at speed, but you can do it. That is real, and it is more than nothing.
What B1 also is, by the CEFR's own internal logic, is the edge of a wall. The level above it is called Vantage — the level from which, for the first time, you can see across the language. Threshold and Vantage are not synonyms. They are the inside and outside of the same boundary, and you can stand on Threshold a long time before you find a way over.
By Cambridge English's published guided-learning-hours estimates, reaching B1 takes roughly 350 to 400 hours of study. Reaching B2 takes another 150 to 200.
| Level | Cumulative hours |
|---|---|
| A2 | 180–200 |
| B1 | 350–400 |
| B2 | 500–600 |
| C1 | 700–800 |
| C2 | 1,000–1,200 |
On paper, the step from B1 to B2 is no bigger than the step from A2 to B1. In practice, almost every adult learner reports it takes longer — often two or three times as long as the table says. The hours are not the bottleneck. What the hours are doing is.
In 1982, two researchers gave the stuck learner a name
In 1982, two American linguists named Theodore Higgs and Ray Clifford published a paper called The Push Toward Communication. The paper made an argument that was unpopular at the time and is still mostly unread outside applied linguistics. It said that the way most adults were being taught second languages — with a strong emphasis on getting them speaking as soon as possible, even if what they said was full of grammatical errors — was producing a kind of learner who could communicate but who would never, beyond a certain point, get any better.
Higgs and Clifford were not theorising. Their evidence came from US government language schools — the Foreign Service Institute, which trains American diplomats, and the Defense Language Institute, which trains military and intelligence personnel. These schools use a six-point proficiency scale called the ILR (the Interagency Language Roundtable scale, developed in the 1950s), with Level 5 being native-speaker fluency. Higgs and Clifford had observed, across a population of learners under the most intensive instruction available anywhere in the world, that a substantial number stalled out at ILR Level 2 — Limited Working Proficiency — and could not be pushed further.
They called these learners terminal 2s.
Their argument was that the methods that got the learners to Level 2 — letting them talk, letting them make errors, prioritising fluency over accuracy — had built into the learners' interlanguage a stable set of systematic mistakes that no later teaching could remove. The learner's grammar had, in effect, set.
That sounds harsh, and the argument has critics. The empirical observation it rests on, though, is the cleanest version of what you may be experiencing four decades later after a year of an evening class or two years of an app. The level Higgs and Clifford were describing maps almost directly to today's B1.
A decade earlier, this already had a different name
Ten years before Higgs and Clifford published their paper, in 1972, a linguist named Larry Selinker had given the underlying mechanism a name. He called it fossilisation, and the term has stuck.
Selinker's argument was that when an adult tries to learn a second language, they do not learn the target language directly. They build, instead, a third system — their own internal version of the target language, which Selinker called the interlanguage. The interlanguage has its own grammar. It changes over time as the learner is exposed to more input. And sometimes — often, in Selinker's claim — parts of it stop changing.
A learner whose interlanguage has stopped changing has fossilised. They will continue producing the same systematic errors no matter how much more they are exposed to the language, how much more they study, how many more years they spend in the country.
Selinker proposed five sources for fossilised forms. The learner's first language leaks through into the second. The way they were taught creates lasting patterns — including patterns of error. The strategies they developed to learn the language become entrenched. The strategies they developed to communicate around things they did not yet know — paraphrasing, avoiding hard structures, working with what they had — become permanent workarounds. And rules they were taught get over-applied beyond where the rule actually holds.
The five processes are, between them, a near-complete description of what an intermediate learner is doing all day. Selinker's claim was that those same processes, which let the learner climb to where they are, are the processes that hold them there.
Why the strategies that got you here are what keep you here
The trap is structural. Higgs and Clifford saw it in 1982. Selinker had described the mechanism in 1972. A more recent treatment — Moving Beyond the Plateau, published by the linguist Jack Richards in 2008 — names three specific features that nearly every plateaued intermediate learner shares.
The learner gets faster at saying the same things they could already say. Sentences become smoother. The toolkit stays the same size.
Over-use of a small set of comfortable lower-level words; failure to acquire the higher-level vocabulary the next level demands. Even the words that are picked up tend to come as dictionary entries rather than collocations.
The learner can communicate, but at the cost of consolidating grammatical errors that were never corrected — Higgs and Clifford's mechanism, described in different language.
Richards puts the first of these plainly: the development of fluency may mean greater ease of known language forms but does not necessarily imply development in the complexity of the learner's language. The second comes with a sharper edge. Even the new vocabulary the plateaued learner does pick up tends to come in as headwords — dictionary entries — rather than as collocations, the patterns the word lives in for real speakers. Knowing the word make is not the same as knowing make a decision, make a fool of yourself, make do with. The third is the diverging line between fluency and accuracy: communication succeeds, errors consolidate, and the learner gets quietly fluent in a version of the language that no native speaker actually uses.
Behind all three is the same machinery. At B1, you can mostly get your point across using a comfortable, recycled set of structures. Those structures are working. You can buy a train ticket, ask for directions, hold a basic argument, write a passable email. The reward your brain is getting for using them is being understood, which is the strongest possible signal that what you are doing is right. Reaching for a harder structure — the subjunctive, a clausal subordinator you only half remember, a word you have seen but never produced — interrupts that reward. The harder structure goes nowhere. The easier one does the job. Over and over, hour after hour, your interlanguage learns that the easier choice is the correct choice. And then it stops.
There is, on top of all of that, a structural vocabulary problem. Paul Nation, the New Zealand researcher who has spent his career mapping the size of vocabulary that different tasks require, calculates that an adult reading unassisted in a novel needs roughly 9,000 word families to cover 98% of the running words on the page. A B1 learner's productive vocabulary — the words they can actually call up to use — is on the order of 2,000 to 3,000.
The receptive-to-productive asymmetry makes it worse. The Israeli researcher Batia Laufer, in a 2004 study of 435 learners, tested vocabulary knowledge across four levels: passive recognition, active recognition, passive recall, active recall. The finding, consistent with forty years of similar work: receptive knowledge develops faster than productive knowledge, and the productive side rarely catches up. The I know that word when I see it but I can never produce it when I need it feeling at B1 is not a personal failing. It is the most consistently documented pattern in second-language vocabulary research, and one of the reasons reading in a new language tends to outpace speaking long before either becomes fluent.
The case that the wall isn't permanent
The story so far is the story Higgs and Clifford, Selinker, and Richards tell. There is a serious argument in the field that this story overclaims.
In 2003, the second-language-acquisition researcher Michael Long published a chapter-length argument that most of the published evidence for fossilisation does not actually support the claim that interlanguage forms stop changing forever. The studies are too short. They observe a stretch of months or a year in which a form does not change, and call that fossilisation, when the data only justifies calling it stabilisation — a temporary halt that, given the right input, can move again.
The 2004 work of Zhaohong Han, the leading researcher on the question since, sits in the middle: long-term stabilisation may be fossilisation's harbinger, but the line between them is fuzzy, and proving the form is permanently fixed requires more longitudinal data than most studies have collected.
This matters, because if the wall is stabilisation rather than fossilisation, the wall can be moved. The question becomes what moves it.
Long and others — Catherine Doughty among them — have argued that what moves stabilised forms is something specific: brief, well-timed attention to form during meaningful communication. Not grammar drills in isolation. Not more input. A short corrective recast in the middle of an actual exchange, on a structure the learner has half-acquired but cannot yet control under speaking pressure. Doughty describes the moment of the recast as opening a "cognitive window" — the few seconds in which the learner's mind, having just been pushed to produce something, is also open to reorganising what it just tried to produce.
The empirical case for this kind of intervention is much more developed than the bare "more input" case, and it is the closest thing the field has to an evidence-based answer to what gets a learner past B1.
What actually moves the wall
There is no single answer. There are three things the research consensus supports, and a fourth thing it has stopped supporting.
The first is productive practice on the structures you can recognise but cannot yet produce. Recognising a subjunctive is not the same as using one; recognising a conditional clause is not the same as forming one under speaking pressure. The gap between recognition and production is exactly where stabilisation lives, and the work of producing the form — slowly, with a chance to notice the mistake and try again — is what unsticks it.
The second is brief attention to form during meaningful communication. This is the central finding of Long, Doughty, and the focus-on-form school. Not classroom grammar drills, and not more input — corrective feedback on a real attempt to say something real, delivered in the moment the learner has just tried to say it. In a tutoring setting this is a recast from a teacher. In self-study it can be a passage you write and then revise against a model; a piece you record and then compare against a native version; a sentence you produced that a language partner gently reformulates while you are still inside the thought that produced it.
The third is vocabulary work on collocation and register, not just headwords. The B2 vocabulary jump is not 4,000 new dictionary entries. It is 4,000 entries you also know how to use — what goes with what, what a writer uses but a speaker does not, what level of formality each word carries. This is the slow, deliberate work of attention to actual native usage, ideally on texts you have read closely and are returning to with a sharper question.
The fourth thing — the one that has stopped being supported — is more of what got you here. More vocabulary lists at your current level. More comprehensible input below the edge of your ability. More conversation in the comfortable register. These continue working in the sense that they keep your interlanguage warm. They do not continue working in the sense that they move it forward, because the strategies that power them are the strategies the research identified as the source of the plateau in the first place.
You are not lazy. You are not unmotivated. The standard advice has been doing what it always did, and it is hitting the limit the research described forty years ago.
This is part of why we built ToTo the way we did. Every day, you get a short story in your target language at a level where you understand around 95% of the words — close to the comfortable edge, but with that missing 5% present in every paragraph as something your brain is actively trying to resolve. That is the input substrate. What you do with it — whether you let the comfortable structures keep doing the work, or whether you reach for the harder forms that are showing up on the page — decides whether the wall moves. Six languages so far: Spanish, Italian, Romanian, Dutch, Swedish, and English. Three minutes. Daily. Free on Android.
