Is the intermediate plateau real?

Yes, and it has two parts. The first is a real flattening. In learner corpora, accuracy improves most on the way into upper intermediate and then stabilises, and each additional thousand words of vocabulary covers much less of a page than the thousand before it. The second part is a measurement failure. Progress continues in dimensions you were never taught to watch, such as the complexity of your sentences. One measure of idiomatic speech even falls as learners cross into upper intermediate and stop leaning on stock phrases. The feeling of being stuck is accurate about the measure and wrong about the learning.

How many hours does it take to get from B1 to B2?

Nobody has measured this. There is no study of how long an adult takes to cross from intermediate to advanced, so any figure in hours or months is an estimate rather than a measurement of learners. The only multi-year plateaus in the research literature were reported in children in Canadian immersion programmes, from group data rather than the same children tracked over time — and in those cases substantial progress resumed later.

Are my mistakes permanent now — is this fossilisation?

Probably not, and the honest answer is that nobody has shown fossilisation exists. Michael Long's review of the research found the evidence "largely impressionistic", with most studies assuming permanence rather than demonstrating it, though he adds that it would be premature to dismiss the idea. ZhaoHong Han, who argues fossilisation is real, describes it as selective and idiosyncratic: particular features of one learner's language stalling, not the learner stopping. The more defensible word for what you are experiencing is stabilisation — observable, and temporary.

Why does my speaking feel clumsier than it did a year ago?

It can be a side effect of improving. In a corpus of spoken language tests, learners crossing from lower to upper intermediate used a wider variety of word combinations and less idiomatic ones, because they repeated fewer stock phrases and reached for more specific vocabulary. The researchers counted that as development rather than decline. Reaching for the precise word and getting a clumsy sentence is a harder thing to do than producing a smooth phrase you drilled.

Will more listening and reading get me past the plateau?

It helps, and on its own it is unlikely to be enough. Krashen's position was that comprehension drives acquisition and that productive ability follows from it; Richards argues this is often not confirmed by learners' experience, and that two further things are needed — noticing, and output that focuses attention on form. The disagreement is live rather than settled. The practical reading is that input is half the job and cannot do the other half.

The intermediate plateau in language learning is real

You haven't stopped. The measure you were using stopped, and nobody told you it was going to.

Yes, the intermediate plateau is real. Two things are happening, and they pull in opposite directions.

The first is a genuine flattening. Your accuracy improves fast on the way into upper intermediate, then levels off. Your vocabulary keeps growing, but each new thousand words buys you less of the page than the thousand before.

The second is that you lost sight of your progress, not the progress itself. It carries on. It just moves somewhere you were never taught to look — into the complexity of your sentences, and the precision of your word choice. Some of it even looks like going backwards: one measure of how idiomatic learners sound drops as they improve.

Both things are true, and you need both. Hear only the first and someone has just told you your learning stopped. Hear only the second and someone has just told you the thing you can feel isn't there.

What it isn't is laziness.

What is the intermediate plateau?

The intermediate plateau is the stage where you can function in the language, keep putting the hours in, and cannot perceive any further progress. The term comes from Jack Richards' Moving Beyond the Plateau, a 2008 Cambridge booklet for teachers. Richards calls it an apparent plateau: learners "appear to have reached a plateau in their language learning and do not perceive that they are making further progress." Not perceiving progress and not making it are different claims, and he picks the first deliberately.

Early on, he notes, progress is "marked and noticeable": after their first 200 or so hours, beginners break through into being real users of the language, however simply. Nothing that visible happens again.

The booklet splits the plateau into five separate problems:

You understand far more than you can say; productive skill stays "relatively static".
Old errors survive, and reappear under pressure.
Fluency ran ahead of complexity, and the harder patterns never arrive.
Your vocabulary range is narrow, so you over-use the words you have.
It's correct and it doesn't sound like a person — grammatical, fluent, bookish.

The first has a post of its own: why you understand a language but can't speak it. This page is about the rest.

Why is intermediate harder than beginner?

Because beginner progress adds abilities you didn't have, and intermediate progress refines abilities you already have. That sentence is our reasoning, not anybody's finding. The arithmetic underneath it is published.

Paul Nation counted how much of a text the commonest 1,000 words cover, then the commonest 2,000, and so on, in How Large a Vocabulary Is Needed For Reading and Listening? His figures for Lady Chatterley's Lover, counting word families — a word plus its relatives — names excluded:

Words you know	Share of the novel you recognise	What that thousand added
first 1,000	80.88%	—
2,000	88.09%	+7.21
3,000	91.23%	+3.14
4,000	93.01%	+1.78
5,000	94.08%	+1.07
6,000	94.77%	+0.69
7,000	95.38%	+0.61
8,000	95.85%	+0.47
9,000	96.17%	+0.32

The right-hand column is subtraction on Nation's own figures. His summary: "from about the sixth 1,000 onwards, each additional 1,000 word-family provides only a small increase in coverage but still involves a reasonable number of word-families." Same thousand words, same months of work, a tenth of the visible return — and most words in that sixth thousand turned up once in the whole novel, too rarely to learn by accident.

The teaching literature puts intermediate learners in the flat part of that curve. Richards reports about 3,000 words as the lower-intermediate target, and quotes 5,000 to 6,000 as readiness for an advanced programme. Those are thresholds people proposed, not boundaries anyone measured.

Nation's table is arithmetic over English text, not a study of learners: it says how much of a page a vocabulary covers, and nothing about speaking.

How do I know if I'm still improving?

Stop measuring with mistakes. Counting them was the right thing to do while you were climbing into upper intermediate, because that is exactly where accuracy improves fastest. You have made that climb. The measure retired and nobody sent word.

One study tagged the errors in 223 learner essays, from lower intermediate up to near-native — B1 to C2 on the European scale. Across more than forty kinds of error, the picture was "dominated by progress and stabilization trends". And the progress was "often located between B1 and B2": on the way into upper intermediate.

A later study of the same range tracked accuracy and sentence complexity together. Crossing into upper intermediate, accuracy is what visibly improves. At the advanced end, the complexity of the sentences moves most, while accuracy carries on improving quietly underneath.

So the dimension that improves changes. If "am I getting better?" means "am I making fewer mistakes?", the answer goes honestly flat while something else is still going up.

Then the finding worth the whole article. Paquot and colleagues measured word combinations — the pairings that make language sound idiomatic — in a corpus of speaking tests. Crossing from lower to upper intermediate, the variety of combinations went up and how idiomatic they were went down. Their reading: learners at upper intermediate and above "repeat fewer combinations and use more specific vocabulary, which sometimes leads to less idiomatic combinations that should nevertheless be considered as traces of qualitative development."

You stop leaning on drilled stock phrases, you start reaching for the exact word, and you score lower on sounding idiomatic — because you got better. If your speaking feels clumsier than it did a year ago, you may be right about the clumsiness and wrong about what it means.

Why has everyone stopped correcting me?

Because your mistakes stopped getting in the way. Richards' explanation: the errors that persist at this stage don't stop the listener understanding you. They may be irritating, they may be stigmatised, but they don't break the sentence, so they never prompt the "sorry, what?" that used to flag them. And because nobody flags them, you stop noticing them yourself — which is where Richards says the chain breaks: "unless the speaker notices such errors, it is unlikely that he or she will correct them."

The silence means you crossed a threshold, not that nobody cares.

For me it is the elephant-in-the-room feeling. Deep down you know you got it wrong, and the person in front of you just nods. Communication worked: the message arrived. The impression you gave of yourself is not the one you wanted.

How long does the intermediate plateau last?

Nobody has measured it in adults. Not "it varies" — measured at all. The durations on page one of Google have no sources behind them: Migaku's page has plateaus arriving "typically after 6-12 months" and lifting "within a few weeks", with no source for either.

What the research literature has is this. Michael Long reports plateaus as long as four years — in English-speaking children in Canadian French immersion, grades 1 to 4 — "but noted that this was followed by renewed, often substantial, progress by grade 10." Those are group averages at different grades rather than the same children tracked through, as Long says himself. Plateaus lasting years have been reported, in children, from imperfect data, and in every reported case development resumed.

Permanent is the real fear, and the word for it is fossilisation. Long's review of that literature found the evidence "has been largely impressionistic", with most studies assuming fossilisation rather than demonstrating it. He is fair the other way too. He calls it "premature to dismiss fossilization as an empty construct" and "quite unwarranted to assume its reality."

ZhaoHong Han has spent a career on fossilisation and argues it is real. Even that account describes it as "selective, idiosyncratic, and contingent" — parts of one learner's language stalling while other parts carry on.

Two senior researchers, disagreeing about whether the thing you are most afraid of has ever been shown to exist. Long's suggestion is to say stabilisation instead — observable, temporary, testable.

How do I get from intermediate to advanced?

Five things, none of which needs us.

1. Change what you count. Pick a measure still moving at your stage, because "how many mistakes did I make" has stopped being one. Longer sentences that hold together; a word you reached for that nobody showed you; the second sentence of an idea instead of stopping at the first. Complexity is where the movement goes once accuracy flattens — which of those you count is our suggestion, not a finding.

2. Let yourself sound worse. When the stock phrase is right there and the precise word isn't, take the precise word and the clumsy sentence. That exchange — fewer stock phrases, more specific words — is what learners crossing into upper intermediate were measured doing while their idiomaticity score fell. Whether doing it deliberately speeds anything up has not been tested.

3. Ask for the correction, because it isn't coming on its own. Once people understand you, your errors stop prompting the question that used to flag them. Ask someone to tell you the thing they would normally let pass — the error that makes you sound foreign rather than the one that makes you unclear. That follows from Richards' argument, and no trial backs it.

4. Add production to your input rather than swapping one for the other. Stephen Krashen, whose theory of comprehensible input shaped how most of us were taught, held that understanding drives acquisition and that the ability to produce the language follows from it. Richards replies that learners' experience often doesn't confirm this, and names two things to add: noticing, and output that puts your attention on form. The disagreement is live, and the fix on offer sits on top of comprehension. Three things to do with no partner are in the post on understanding versus speaking.

5. Put the rare words where you'll use them. A word from the sixth thousand appears once in a novel, so meeting it again is not a plan. If you want to keep a rare word, put it somewhere you will have to produce it.

Where Pezmu fits

Pezmu was built for people at exactly this stage, so take this as disclosure rather than advice.

It's typed chat with a tutor that remembers your life. The tutor pitches its own language a notch above yours, so the input stays ahead of you instead of settling where you are comfortable. It corrects you — the corrections that stopped arriving once people could understand you. It doesn't get bored, and it's there at eleven at night.

Underneath the conversation is a record of what you actually produced. A word moves from Saved to Hinted when you say it after a nudge. It reaches On your own only when you produce it with nothing showing it to you. Recognising it moves nothing. A word you are still learning stays in front of you until you use it, and comes back when it fits what you are talking about.

Ask how you are doing and you get an answer about your language rather than a score: which words you used unprompted, which mistakes repeat, how far your sentences reach. That measure keeps moving long after "fewer mistakes" has gone flat.