Why you understand a language but can't speak it

Understanding a language and speaking it are two different skills, and you get better at whichever one you practise. You can follow a film, read the news and still stall halfway through a sentence of your own, because recognising a word and retrieving it are separate jobs, and nearly everything in an ordinary week of learning is the first job. Your ear is fine. Your memory is fine.

The standard prescription is more input: more films, more podcasts, another six months of flashcards. It isn't wrong, it's just more of what you already have. An hour spent understanding the language buys you a better hour of understanding the language.

Why can I understand a language but not speak it?

Recognising a word and producing one are different operations. Recognition is a match: something arrives, you check it against what you've stored, and context does half the work. Production is retrieval from nothing: a meaning you want, no prompt, no options, and you supply the form.

I'll say it for myself — this physically hurts. Producing makes my brain feel like it's moving through gravel.

Recognition and production come apart in a predictable order, clearest in people losing a language rather than gaining one. A 2019 paper on foreign-language attrition makes the point that how much someone seems to have forgotten depends on how you test them: ask for a half-forgotten word and they often can't produce it, show them the same word and they still recognise it. Their summary of the literature: "productive recall failure appears to precede recognition inability."

Production is the first thing to go and the last thing to arrive.

Forgetting is only the clearest case. The same gap shows up in people still learning: Webb tested learners on both and found receptive vocabulary larger, with the gap widest for the rarer words. How much larger is not a fact. Mark the same answers generously and there was "little difference" between the two, so any page telling you your passive vocabulary is three to five times your active one is quoting nobody.

Why practice is so narrow has a name, the only jargon in this post: transfer-appropriate processing — memory serves you best when what you do at the moment of use matches what you did while learning. A 2021 review in Psychological Bulletin lists it as one of three explanations for why being quizzed beats re-reading. You get good at the exact move you rehearsed, in the format you rehearsed it. When DeKeyser and Sokalski taught 82 adults in their first year of Spanish, some by comprehension practice and some by production practice, each group came out better at what it had practised, though the authors note the pattern blurs with which structure you test and when.

Does watching films and listening to podcasts help you speak?

It helps, and it isn't the same exercise. To get better at producing you have to practise producing, but input is doing real work and shouldn't be cut.

A 2013 review pooling comparisons of comprehension-based and production-based grammar teaching found "large effects on both receptive and productive knowledge" from both kinds: practice built on understanding really does build language you can produce, and by a lot. Where it gives way is time. On the delayed tests, run between a week and about ten weeks after the teaching, production practice was the one ahead on production.

The argument is live rather than settled: VanPatten has long held that comprehension-only practice builds accuracy in production as well as comprehension, and a 2025 review in the TESL Canada Journal notes "this claim has been challenged by other researchers."

Input isn't the enemy — it's half the job. It just can't do the other half for you.

How do I practise speaking when there's nobody to talk to?

Three things that need no partner, strongest first.

1. Tell it, then tell it again. Describe your weekend — out loud, typed to a friend, dictated into a notes app — then describe it again. Across 31 studies of task repetition, saying the same thing again improved accuracy more than any other cheap technique anyone has measured. The second telling is where the language moves: the ideas are settled, so more of your attention is free for form. What those studies measured was writing, not speech.

2. Reach for the word before you look it up. Spend five seconds failing before you open the dictionary: the failing is the exercise, the retrieval attempt you're short of. No trial backs this particular habit; it's the mechanism applied.

For me, those few seconds are the part that feels like doing the work rather than cheating. I can feel the connection to the word getting stronger.

3. Build a whole sentence — then make it one you'd actually say. Reviewing the word shop does less than writing "the shop on my corner closes before I finish work", and aiming the sentence at your own life adds a little on top. In a 2025 study, learners who built a sentence about themselves remembered noticeably more of the new words a week later than learners who drilled the translation. What the study tested was whether learners could still say what the words meant, not whether they could produce them.

You don't have to tell the truth, incidentally. Swap the names, invent the street, give yourself a dog. The sentence only has to be one you had a reason to write.

The sorting rule, for anything else in your week:

What you're doing	What it rehearses
Watching with subtitles, reading an article	Recognising words in context
A card showing the word, asking for the English	Recognition
A card showing the English, asking for the word	Retrieval — harder, and the one you're short of
Re-telling something you already told, better	Retrieving the same forms with the thinking done
Writing a real message to a real person	Retrieval, plus building the sentence

If your week is all top half, the bottom half is what you're waiting on.

Where Pezmu fits

Pezmu was built on the idea that you only get better at what you produce, so read this as disclosure rather than advice.

It's typed chat with a tutor that remembers you, with a record underneath of everything you've produced. A word moves from Saved to Hinted when you say it after a nudge, and reaches On your own only when you produce it with nothing showing it to you. Recognising it moves nothing. Ask any time for feedback on where you're at, and every line of it points back at something you wrote. Corrections sit under the reply instead of inside it, so the conversation never stops.

The understanding half is in there too: the tutor's own language is pitched at your level and a notch past it, so you're reading something slightly harder than you'd have written yourself, every turn. Input isn't what we're arguing against: we built ours to arrive at the right level. What we won't do is count it as practice you've done.

The concession we'd rather make than have made for us: typing is not speaking. Ziegler's 2016 review of typed versus face-to-face interaction found typed conversation produced real progress, roughly on a par with talking in person, with a small edge for typing on written and productive measures — and a small edge for face-to-face on the spoken ones. Pezmu reads its replies aloud, and talking back instead of typing is not ready yet; nothing listens to you speak, and nothing grades your accent.

Seven languages today: Spanish, French, German, Italian, Portuguese, Russian and Greek. Greek is the one with real hours on it — the rest opened in October 2026. And we don't nudge you to repeat a task, so the strongest item above is yours to use anywhere, including nowhere near us.