You understand everything but cannot speak

You can follow a podcast. You read the news without a dictionary. You understand your teacher completely.

Then someone asks you a question in a shop and you produce four words, two of them wrong, at half speed, while a perfectly good sentence assembles itself in your head about six seconds too late.

This is the most common complaint in adult language learning, and it is not a sign that something has gone wrong. It is the predictable result of how nearly everyone studies.

Two different operations

Comprehension and production feel like two directions of the same skill. They are not.

Recognition works from a rich cue. The word arrives, in a sentence, in a context, on a topic you already know about, spoken by someone whose next three words further constrain the meaning. You do not need to retrieve the word — you need to confirm it against a stream of evidence. A partial match is enough. If you catch 70% of the words, context supplies the rest, and you experience that as understanding.

Production has no cue. You need a specific word, from nothing, right now, plus its gender, plus the right preposition, plus a conjugation, assembled in the correct order, while simultaneously planning the next clause and monitoring the listener’s face — in about 400 milliseconds, because that is how long a conversational gap can be before it becomes awkward.

These are not the same task with the arrow reversed. One is a lookup with hints; the other is unassisted retrieval under real-time pressure.

COMPREHENSION                    PRODUCTION
─────────────────                ─────────────────
cue provided          →          no cue
partial match OK      →          exact form required
context fills gaps    →          you supply everything
listener sets pace    →          you set the pace, fast
one operation         →          retrieve + inflect + order + plan

The gap between what you can recognise and what you can produce is normal in every language, including your first. In a foreign language, the gap is simply much wider — and everything about conventional study widens it further.

Why normal study makes it worse

Look at what a typical week contains: reading, listening, watching, flashcards showing the target word and asking for the meaning, exercises where the answer is on the page.

Every one of those is recognition. They are pleasant, they feel productive, and they grow the passive store — which is the side that was already ahead.

Two specific culprits:

Recognition-direction flashcards. A card showing hablar and asking “what does this mean?” trains the easy direction. Reverse it and the same card becomes genuinely hard, which is the point. See spaced repetition on why difficulty is the mechanism rather than an obstacle.

Comprehensible input as a complete method. Input is necessary. It is not sufficient. The strongest input-only advocates predict that production emerges naturally given enough comprehension, and the observed pattern in adults who follow this is enormous comprehension alongside slow, halting speech that improves markedly the moment they start actually speaking. Input builds the store. It does not build the access path.

The four things that work

1. Forced output with constraints

The core problem with free conversation practice is that you can avoid everything you find hard. Learners develop a comfortable subset — a few hundred words and three constructions — and then have the same conversation for two years, fluently, without progress.

The fix is constraint. Not “tell me about your weekend” but:

Tell me about your weekend. Use obwohl, deshalb and trotzdem at least once each.

Or: describe the same event three times — past, future, hypothetical. Or retell a story you just read, without looking, using at least five words from it.

The constraint forces retrieval of things not yet in the comfortable subset. That is the entire trick, and it is the thing a good teacher does that self-study rarely replicates.

2. Production-direction retrieval

Flip the flashcards. Front: the meaning or the situation. Back: the target sentence. Say it out loud, fully, before flipping.

This is uncomfortable in a specific way — cards you “knew” in the recognition direction turn out to be unavailable in the production direction, which is exactly the diagnosis. Those cards were never active vocabulary; they were recognisable strings.

Better still: sentence cards rather than word cards. A word retrieved in isolation still has to be inflected and placed. A sentence retrieved whole comes with its grammar attached, and after enough repetitions it becomes a motor pattern rather than a construction problem.

3. Recorded monologue

Two minutes a day, alone, out loud, recorded. No listener, so no social pressure; recorded, so it counts.

Pick a prompt, speak for two minutes, then listen back. Listening back is the part people skip and the part that works — you will hear the same three errors and the same four crutch words, and you will hear the places where you slowed down to search.

Write down what you could not say. That list is the highest-value study material you will generate all week, because every item on it is a gap you were personally motivated to fill.

4. Shadowing, for the speed problem

Some of the gap is not knowledge but speed. You know the sentence; you cannot assemble it fast enough for the conversation to survive.

Shadowing attacks this directly by removing your control over the pace. It also trains prosody, which is the other half of sounding like you can speak — a learner with correct words and wrong rhythm reads as much less fluent than the reverse.

What to expect

The reassuring part: if you understand a lot, you are not starting from nothing. The store is built. What is missing is access, and access responds fast.

Weeks 1–2   Speaking feels worse than before. You are noticing
            errors you previously spoke straight past.
Weeks 3–6   Crutch phrases decrease. Sentences finish more often.
Months 2–4  Noticeable speed increase. Less translation from L1.
Months 4–6  Constructions you only recognised start appearing
            in your own speech unprompted.

Three to six months of deliberate output practice produces a large change for most learners, and it is a change in retrieval, not in knowledge. This is why the plateau feels so strange from the inside: nothing new is being learned, and yet everything is improving.

A weekly minimum

For someone with a lesson a week and twenty minutes a day:

Daily	10 min production-direction cards, out loud
Daily	2 min recorded monologue, listened back
3× week	10 min shadowing, one clip
Weekly	One lesson with an explicit constraint on what you must use
Weekly	Turn the “could not say” list into next week’s cards

The loop that matters is the last one. Every week you generate a list of things you tried to say and could not; every week those become the material you retrieve next week. That closes the circle between what you need and what you study, and it is the single fastest route out of the gap.

The mechanical half of that loop — capturing the sentence you failed to produce, getting it translated, transcribed and turned into a card — is what Teachee automates during the lesson itself. The failure gets recorded at the moment it happens, which is the only moment you actually know it mattered.