A language you have stopped using fades in a predictable order, and knowing the order tells you what to maintain. Speed goes first: the words are still there and they take two seconds instead of a fifth of one. Then rare words stop arriving. Then production generally becomes effortful. Understanding a written page is last and it survives for years.
That order is why maintenance by reading is the commonest mistake here. It is pleasant, it is easy to keep up, and it exercises the one thing that was not going to go.
Eight facts about losing it and getting it back
- attrition is slow after B2. A language taken past the intermediate stage is remarkably durable, and one abandoned at A2 is not.
- production goes before comprehension. You will still understand a film years after you can no longer hold a conversation.
- relearning is much faster than learning. What took a year to build often comes back in a few weeks of contact.
- a language is rarely lost entirely. What feels like loss is usually retrieval failure, and the material is still there.
- childhood languages fade fastest when contact stops early, which is the one genuine exception to the durability above.
- twenty minutes a day maintains what a thousand hours built, which is an extraordinarily good rate of exchange.
- reading is the cheapest maintenance, requiring no partner, no schedule and no equipment.
- schedule maintenance like review. A fixed weekly slot for a language you are not actively learning is what stops the slide.
Coming back is much faster than arriving
The encouraging half is solid. Relearning something you once knew is dramatically quicker than learning it the first time, which Ebbinghaus measured on himself in 1885 and which everybody who has picked up an old language has experienced: three weeks of work restores something that took two years to build.
That is consistent with everything else on the subject. What faded was mostly the route rather than the item, so what a return does is rebuild access to material that is still in place, and rebuilding access is a much smaller job than storing something new.
A gap in a language is not a catastrophe, and treating it as one is the reason people decide not to start again. What it is is expensive to leave open, because the effort of the return grows with the length of the gap even though it never approaches the original cost.
The maintenance dose is small and it is the right kind
Holding a language costs a fraction of building one, and the fraction is smaller than most people assume. What it does require is that the small amount is spent on the thing that fades rather than on the thing that lasts.
That means production, and it means retrieval under some kind of time pressure. A monthly conversation does more to hold a language than daily reading, and it is a smaller commitment. A weekly voice note to somebody does more than a podcast every morning.
Where it stops: maintaining with input
The stall is a well-meant habit aimed at the wrong target. Somebody who does not want to lose a language keeps reading in it, or keeps a series going, and feels that the language is being kept up.
Comprehension stays excellent, which is the evidence they are looking at, so the habit is confirmed every time. Meanwhile production has been decaying for two years, and the discovery happens at the moment somebody addresses them in it and nothing arrives.
The correction is not more time, it is a different kind of time. Ten minutes a month of speaking beats an hour a week of reading for this specific purpose, and it is much less pleasant to arrange, which is why nobody does it.
One call a month, with one person
A recurring monthly call with a single person is the whole maintenance programme, and it is worth setting up while the language is still working rather than after it has started to go.
It replaced promising yourself that you will keep it up, which names no dose, no date and no other person, and which therefore never happens.
Its limitation is the obvious fragility: it depends on one person continuing to be available, and when they stop, the arrangement stops with them. Having two is not twice the work, and it is the difference between a habit and a coincidence.
What it looks like when it goes wrong
The recognisable failure is understanding a conversation completely and being unable to join it. That is not the language having gone; it is the half that goes first having gone, in a person who has been maintaining the half that never does.
The second is the assumption of total loss. Somebody who has not used a language for a decade will usually say they have forgotten it, and a week in the country contradicts that within days, which is worth knowing before deciding not to bother trying.
What makes a plateau a fact about a method rather than about a person, and the four places it happens while you are still working, is the stalls every learner hits, and the way through each, and the reason a faded word is a routing problem rather than a missing one is the curve everyone is fighting, and what bends it.
The open question is whether to hold two languages at once, because a maintenance dose for each is affordable and the two interfere with each other in ways that a single language never does: when it works, and the pairs that interfere.