Stuck at intermediate for years: what the English plateau really is

You've spent four years calling yourself "a B1, almost B2." You watch shows without subtitles, you follow the meetings, you read in English without a dictionary. And you still make exactly the same mistakes you made four years ago. What's happening to you has a technical name, and it isn't "fossilization": it's stabilization. The distinction matters, because fossilization is permanent by definition and is almost never actually demonstrated, while stabilization is a temporary state you can come out of. You're not stuck for lack of hours or lack of talent. You're stuck because people already understand you, and once people understand you, nothing corrects your errors. What reverses that isn't more input. It's attention to form, plus correction.

You are not fossilized. You are stabilized.#

The word you use for your problem decides whether the problem has a solution. "Fossilized" means permanent, and that's a far stronger claim than the evidence supports. In his chapter of the Handbook of Second Language Acquisition, Michael Long shows that most published fossilization studies fail the basic methodological bar: they select informants badly, collect too little data, and analyze it inadequately. Proving that something has fossilized would mean following the same speaker for years. Almost nobody has done it.

The term specialists actually defend is stabilization: a state in which fluctuation has temporarily stopped. The nuance isn't cosmetic. ZhaoHong Han, the field's leading researcher on the topic, revisited her own hypothesis forty years after Selinker proposed it and arrived at a selective version of the phenomenon: what hardens is specific structures, not your whole English. Your conditionals sit frozen solid while your technical vocabulary grows every month. You don't have a level. You have dozens of sub-systems moving at different speeds.

The plateau has an author, a date, and a description#

The intermediate plateau has been documented in English-teaching literature for more than fifteen years. Jack C. Richards devoted an entire essay to it, Moving Beyond the Plateau, where he defines it as the point at which a learner has developed a functional but limited version of the language: they communicate, they get understood, they get things done. And precisely because they get things done, nothing forces them to improve accuracy, range, or naturalness. The plateau is the price of communicative success, not proof of personal failure.

The plateau also has instructional and motivational causes, and they deserve the same honesty. When Mirzaei and colleagues asked 31 upper-intermediate learners directly why they had stalled, in a qualitative study that also brought in ten teachers, the causes that surfaced weren't purely linguistic. What came up was the same syllabus repeated year after year, motivation that had dropped, no awareness of their own process, and an inability to automate what they already knew. If you've spent three courses recycling the same intermediate material, the plateau isn't a cognitive mystery. It's the syllabus.

You understand more and more, and speak the same. That is not a contradiction.#

Understanding and producing are different operations, which is why they come apart. Merrill Swain observed the extreme case in Canadian immersion programs: after years of extremely rich input, students understood almost like natives and still produced like foreigners. The mechanism is simple. To understand, partial analysis is enough: context, four key words, and the topic fill in the rest. To produce accurately, you need the complete analysis, because nobody is going to fill in your gaps.

The gap shows up in national numbers. In EF's 2025 English proficiency index, Spain scores 558 on reading and 462 on speaking: nearly a hundred points of difference inside the same average speaker. The index is built from voluntary online tests, so read it as an indication rather than a census. The pattern still matches what Swain documented in Canada, and what you notice every time you understand a question perfectly and answer it with half the words you wanted.

Swain ended up putting the weight on speaking out loud with someone in front of you. As she puts it: "In dialogue we can see language learning in progress." Producing isn't a demonstration of what you already know. Producing is the moment you find out what you don't know.

The hours myth: B1 to B2 is not the longer step#

B1 to B2 costs no more than A2 to B1. People say every level is more expensive than the last, and that this is why you stopped at B1. Cambridge English's own guided learning hours say otherwise: 350–400 cumulative hours for B1 and 500–600 for B2. Those are totals from zero, not per-step costs. The jump from B1 to B2 works out at roughly 150–200 hours. The genuinely steep step sits much higher up, between C1 and C2.

What does grow brutally is the vocabulary denominator. Paul Nation's work on how much vocabulary you need puts numbers on it: covering 95% of a text takes 4,000–5,000 word families, and reaching 98%, the threshold for understanding without help, takes 8,000–9,000 in writing and 6,000–7,000 in speech. You double your vocabulary to gain three percentage points. Your first two thousand words changed your life in six months; the next six thousand stay invisible for years. You're still climbing. What changed is the resolution of your progress, not its existence.

What actually breaks it: noticing form, and being corrected#

More listening does not repair an error nobody has ever pointed out. Richard Schmidt established in 1990, in The Role of Consciousness in Second Language Learning, that subliminal language learning does not exist: input becomes learning when the learner notices the form. If you've spent ten years saying I have seen him yesterday and nobody has ever flagged it, you can stack ten more years of podcasts on top without that sentence ever changing.

Correction works, and there are numbers behind it. Shaofeng Li's meta-analysis of 33 studies in Language Learning found a medium effect for corrective feedback that held up over time, with one counterintuitive finding: implicit feedback, recasting what you said instead of marking it in red, was retained better than explicit correction. It also found larger effects in foreign-language settings than in second-language ones, which is exactly the situation of anyone studying English from Madrid, Bogotá, or Mexico City.

The most useful testimony comes from inside the profession. Scott Thornbury, who trains English teachers, documented his own stabilized Spanish in a series of posts: he used demonstratives at random for years because he pointed with his finger, and nobody at the bakery ever noticed. "There are even sets of like-sounding words that I also deploy in free variation, the verbs planear, plantear, plantar, and planificar being a case in point," he writes. Nothing broke, so nothing got repaired. In that same piece he quotes Diane Larsen-Freeman, for whom that mess is "the instability that precedes a phase shift" and the moment when "a pedagogical intervention might be optimal." Translated: the chaos is the signal that you're right before the jump, not after the end.

Breaking a plateau means changing the kind of practice, not the amount. Producing out loud, against a clock, on specific structures, with something telling you whether it came out right before the wrong version settles in again. That loop — say it, hear yourself, correct, come back to it in a few days — is what Tabla automates with its pronunciation exercises: you speak, speech recognition scores you, and the system brings back the word you missed right when you were about to forget it. How many of your lifelong mistakes would survive a month of somebody flagging them every single day?

Pick one structure this week that you know you get wrong (conditionals, the present perfect, time prepositions) and chase it until it feels uncomfortable. The plateau doesn't break by turning up the volume on input. It breaks by pointing the flashlight somewhere very small.

Try this today: record yourself talking for 60 seconds about what you did yesterday, transcribe the audio word for word, and underline the three mistakes you would never have caught while speaking.