Language Learning vs Language Practicing: The Distinction That Explains the Intermediate Plateau

This one's a sponsored post from Geoff Bloom, the founder of Dioma. I gave Dioma 4/5 in my review, and this is the problem it's built around. Take it away, Geoff!

You've seen the grammar before. You've read the explanation, and you could probably teach it to a friend.

You've done real work on the language too. You've heard the structure in context 100 times: podcasts, an Anki deck you open daily, fifty or sixty hours with a tutor.

Then you're in a conversation, and you reach the point in the sentence where the mood has to change, and you freeze. You produce the wrong form, or you steer around it and say something easier.

Nothing is wrong with your knowledge. You just haven't practiced enough.

What Canadian schoolchildren showed us in the 1960s

Starting in the 1960s, Canada ran one of the largest natural experiments language learning has ever had.

English-speaking children were enrolled in French immersion programs. They didn't study French so much as do school in it: math, science, history, all in French, for years.

They received the thing every learner is told to chase, in quantities almost nobody can arrange for themselves. Thousands of hours of rich, contextual, comprehensible input.

Merrill Swain studied their progress and published her findings in 1985. Their comprehension was superb. In listening and reading, immersion students performed at or near the level of native-speaking peers.

Their production was not. Years into full immersion, those same students were still immediately identifiable as non-native speakers and writers.

They had weaker command of grammar, less precision in vocabulary and morphology, and lower accuracy than same-age native speakers. Thousands of hours of perfect input had built enormous understanding. It had not built the ability to produce the language accurately.

Swain's conclusion is the one sentence I'd hand every intermediate learner: comprehensible input is necessary, and it is not sufficient.

What was missing she called pushed output. Being required to produce language precisely and coherently, slightly beyond what you can comfortably manage, and then finding out where you were wrong.

Why input alone won't finish the job

You already know that tapping the right word out of a word bank isn't speaking.

The tools you graduated to after your beginner app mostly aren't finishing the job either.

Podcasts, shows, graded readers and YouTube all build the understanding half of learning a language. None of them ask you to produce anything, and none tell you when you were wrong.

Anki is genuine retrieval, but for most decks it's retrieval of a specific item you've learned to recognize.

Even tutoring splits down the middle, depending on how you use the hour. An hour of comfortable conversation about your weekend, in structures you already know well, doesn't move you forward.

What moves you forward is an hour where you're pushed past what you can already do, and corrected on the specific structures you keep missing.

The math that used to make this impossible

Swain named the fix in 1985, and then everyone ran into the same wall. Pushed output requires somebody competent listening to you and telling you what was wrong.

That has always meant a teacher. It still mostly does.

A good tutor on italki or Preply remains the gold standard, as long as you go in and ask to be corrected rather than kept in flowing conversation.

The trouble is volume. The number of times you have to reach for the subjunctive under real pressure before it becomes automatic isn't five, and it isn't twenty.

At tutoring rates, almost nobody can buy that many repetitions. So people buy an hour or two a week, spend most of it on comfortable conversation, and wonder why progress stalled.

There's a free path, and it's a real one. Language exchange apps, correction communities like LangCorrect, and the correction channels on the bigger language Discord servers will all get you corrected at no cost.

They're uneven, they run on reciprocity, and they work if you put in the time and effort to contribute to the community.

For everyone else there's a gap between "an hour a week of tutoring" and "the volume of corrected production this actually takes." Almost nothing on the market sits in it.

Why I built Dioma

I didn't set out to start a company.

I learned Spanish from scratch as an adult and got to B2 over about three years. I'm working through Hebrew now, which I started in country at nineteen and then walked away from for fifteen years.

Both times I hit exactly the wall this article is about. Both times what moved me was producing the language badly in front of someone who told me precisely what was wrong with it.

I wanted more practice than my calendar or wallet could support. So I built it.

Dioma is a practice engine. It isn't a course, and it isn't trying to be. Here's what a session actually is.

You don't browse a library. Dioma picks today's exercises from where you sit on your level map. That's the deliberate-practice principle applied honestly: you work on your demonstrated gaps, not on whatever you feel like doing tonight.

You produce, from nothing. You speak and write your own responses. No word bank, no multiple choice, no sentence to complete. This is Swain's pushed output, and it's the whole reason the product exists.

A correction comes back in about fifteen seconds, and it names the structure. Not "good job," and not a silent rephrase. It tells you what was wrong and why, which is what the corrective-feedback research says separates feedback that moves a learner from feedback that just happens at them.

Every correction is anchored to a topic a human educator wrote. If a correction can't be tied to a specific topic in our curriculum, we don't show it to you. That curriculum runs more than 700 topics across Spanish, French and Hebrew, written by professional educators who had built CEFR-aligned curricula and taught adults one on one before they came to us. None of it is generated.

You get re-levelled every three sessions, on fixed accuracy thresholds, measured from what you actually produced. Not from a quiz, and not from how many days in a row you logged in.

There are no streaks, no XP, and no characters. Plenty of this industry runs on exactly those things.

On the question this audience always asks, and deserves a straight answer to: generative AI does exactly three jobs inside Dioma. It corrects what you write and say against the teacher-written curriculum, it adds variety to our exercises, and it drafts your weekly topic plan.

Placement, level moves, progress scoring and session structure all run on fixed rules, with no model involved. The curriculum is human all the way down.

Where Dioma won't work for you

Dioma won't suit everybody.

It's web only right now. There's no iOS or Android app today in the app store, though it is mobile optimized and can be attached to a shortcut on your phone. We're working on a native app.

It covers Spanish, French and Hebrew, and nothing else. Going deep meant going narrow. If your target language isn't one of those three, this isn't your tool today.

There's no beginner track. Dioma starts at A2 to B1 depending on the language, because a practice engine needs something to correct. If you can't yet build sentences, get a foundation course first.

There's no free tier. That's the honest consequence of paying educators to write every topic by hand, and of running real-time AI corrections.

And the last one isn't about the product at all.

Producing language badly, on purpose, in front of something that will tell you it was bad, is the least pleasant twenty minutes available in your target language.

Reading a well-written explanation of the subjunctive feels like progress immediately. Practice feels like work, and for the first few weeks it mainly produces evidence of what you can't yet do.

That's not a flaw in the method. That's the method.

Try it for 21 days

If any of this describes your last two years, the fastest way to test the diagnosis is to produce something and see what comes back.

Dioma is free for 21 days, no credit card required. You get full access for the whole trial: the real curriculum, the real corrections, the real level map.

If you decide to continue, plans start at $13 a month billed annually, with a 30-day money-back guarantee once your plan begins.

Start your 21-day free trial at dioma.com

This is a sponsored article written by Geoff Bloom, founder of Dioma. Actual Fluency was paid to publish it. The views and product claims above are the author's own.