Why we don't correct everything

There is a tempting fantasy in language-learning software: catch every mistake, mark every wrong article, flag every dropped verb ending. The user gets crisp, complete feedback. The product looks rigorous. Everyone goes home happy.

In practice it produces learners who don't speak.

Anyone who has been on the receiving end of relentless correction — from a parent, a strict teacher, a partner with strong feelings about your accent — knows the feeling. After a while, you stop offering sentences you're not sure about. You retreat to safe ones. You speak less. The correction worked, in a narrow sense: you make fewer errors, because you're producing less to be wrong with.

This is exactly the trap Jaimy is built to avoid.

Six kinds of correction, not one

In 1997, two researchers named Roy Lyster and Leila Ranta sat through nearly nineteen thousand minutes of language classrooms and catalogued every time a teacher reacted to a student error. They ended up with a taxonomy of six feedback types. We use a version of it every time Jaimy decides what to do.

Recasts — The teacher quietly says it back correctly. Student: "I went to the cinema yesterday and watch a film." Teacher: "Ah, you went to the cinema and watched a film — what film was it?" The student's ear catches the fix; the conversation never stops.

Elicitations — The teacher leaves the gap and lets the student fill it. "I went to the cinema and..." with a little look. Students who self-correct remember better than students who get told.

Clarification requests — "I'm sorry, could you say that again?" This works for errors the student already knows how to fix; it gives them a chance to try.

Metalinguistic feedback — A tiny grammar hint, without giving the answer. "Watch out — past tense there."

Explicit correction — The full stop-and-explain. "You said 'watch' — it should be 'watched', because it's past tense." The clearest. Also the most disruptive.

Repetition — Repeating the student's error back with rising intonation. "You watch?" A softer way to point at it.

Lyster and Ranta measured how often each one produced uptake — the student actually using the corrected form themselves afterwards. The results were not what you would expect.

Recasts, the gentlest and most common, had the lowest uptake — around 31%. The more disruptive feedback types — elicitations, clarifications, metalinguistic prompts — produced uptake more than half the time. The lesson was clear: the correction style that feels nicest is not always the one that teaches best.

But — and this is the important part — the disruptive feedback only works if the learner is still engaged. Use it constantly and the learner shuts down. The correction system has to know when to push and when to let things slide.

Jaimy's rules of the road

We boil this research into three working principles.

Always correct things that block communication. If you said something and Jaimy genuinely couldn't tell what you meant, that gets explicit feedback every time. Communication is the whole point.

Periodically and gently address patterns. If you have dropped the same verb ending four sessions in a row, Jaimy will, at some point, slow down and explain it once. Not every time it happens. Once.

Let small slips go. A wobbly article, a slightly wrong preposition, a missed gender agreement that didn't affect the meaning — Jaimy notices, but does not stop the conversation. The flow is more valuable than the fix in that moment.

The cost of overcorrection is silence. The cost of letting a few prepositions wobble is approximately nothing. It's not a hard trade.

A1 looks different from C1

The other piece of this is that the right correction style changes with level.

For a brand-new beginner, almost everything is a recast. The learner's job at A1 is to be brave enough to speak at all; piling on explicit corrections would be cruel and counterproductive. The errors will largely self-correct with more exposure.

By C1, the learner has the meta-language and the confidence to handle direct feedback. They want the upgrade — "this works, but a native speaker would say it like this." Recasts at C1 actually feel patronising; the learner already knows the standard form. They want the next level up.

Jaimy's correction style slides across this spectrum as you progress. At A1, you'll barely notice corrections happening. At C2, you'll be getting style notes on register and idiom. Same principle: choose the kind of feedback that this learner, at this level, can actually use.

The thing about being corrected by a friend

There is a difference between the way a friend corrects your language and the way a teacher does. A friend recasts. A friend asks you to say it again because they didn't catch it. A friend laughs at the funny mistake and moves on. A friend doesn't track your errors in a notebook.

The result, oddly, is that you learn faster with the friend. Because you keep talking.

Jaimy is trying to be the friend.