Most of what you have forgotten is still stored. The evidence is ordinary: the word arrives an hour later while you are doing something else, or the first letter is enough to produce it, or you recognise it instantly when somebody says it. Nothing that had genuinely gone would do any of that.
So forgetting is mostly a failure to find the thing rather than a loss of it, and that changes what the work is. You are not refilling a leaking container. You are building more routes to something already in there.
Eight facts about losing what you learned
- forgetting is the default. Memory decays unless something interrupts the decay, and the interruption is retrieval.
- the curve is steep in the first day. Most of what is lost is lost within twenty four hours, which is why the same day review matters so much.
- recognition fades slower than recall, so a word you can no longer say is often still one you would recognise.
- interference between similar words is a real effect: learning two similar items together makes both weaker.
- learn opposites separately. Left and right, or the words for buy and sell, are much harder learned as a pair than a week apart.
- context dependent recall. A word learned only at a desk is harder to retrieve in a shop, which is an argument for varying where you study.
- relearning is faster than learning. A language lost over ten years comes back in weeks rather than years, and the saving is measurable.
- forgetting is not failure. It is the mechanism the schedule is built around, and expecting to remember without review is what makes people quit.
More contexts beats more repetitions
If the problem is routing, then what helps is meeting a word in circumstances that are not identical to the last time. The same word in five different sentences builds five ways in. The same sentence five times deepens one, and the one it deepens is the one where the word sat in a deck rather than in a conversation.
That is the practical difference between a word you can produce in a review and a word that arrives when you need it. Both are stored. Only the second has been reached from somewhere other than the front of a card.
It also explains the experience everybody has of understanding a word perfectly in a book and failing to recognise it when somebody says it. Reading and hearing are two different routes, and building one does very little for the other.
What actually bends the curve
Three things do, and they can be stated in a line each. Being made to retrieve something, rather than being shown it again. Leaving a gap before the retrieval, long enough that it takes effort. And connecting the item to something you already hold, which is what makes a hook rather than an isolated fact.
One thing does not, and it is what most study consists of: reading the material again. It produces a strong feeling of familiarity, which is precisely the signal that misleads, because familiarity is what recognition feels like and recognition was never the difficulty.
Sleep is the other half of it and it is not a metaphor. Consolidation happens overnight, so a review done in the evening and a review done at noon are not equivalent, and a night's sleep between two study sessions is doing work that neither session did.
Where it stops: counting repetitions
The stall is treating the number of exposures as the variable. A word that will not stick gets reviewed more often, in the same form, from the same card, and it goes on not sticking, and the learner concludes that they have a bad memory for words.
Nothing about that conclusion is available to be corrected from inside the deck, because the deck reports exposures and cannot report contexts. Ten reviews of one card look identical in the statistics to ten meetings in ten sentences, and only one of those does what you wanted.
The repair for a stubborn item is not another review. It is to go and find the word somewhere else: in a sentence somebody wrote, in something spoken, in a use you had not seen.
Finding the same word in thirty other sentences
Searching an exact phrase in Google Books returns the word inside dozens of published sentences, free and with no account, which is the cheapest way to manufacture the contexts this page is about. For a mid-frequency word, waiting to meet it again by chance takes months; this takes ten seconds and produces a dozen meetings.
It replaced hoping to run into the word again, which is a fine strategy for common words and no strategy at all for the ones that actually fail.
Its limitation is the corpus. The sources skew old and literary, so the contexts are real English and not necessarily the English anybody is speaking this year, which makes it excellent for meaning and unreliable for register.
What it looks like when it goes wrong
The recognisable failure is the word on the tip of your tongue: you can produce its first letter, its length, sometimes a rhyme, and not the word. That is a routing failure in its purest form, and it is proof of storage rather than evidence of loss.
The second is a word that is solid in a deck and absent in speech. The route from a card is built and the route from a situation is not, and no additional card reviews will build the second one.
How far apart the reviews should be, and why the intake rate matters more than the schedule, is reviewing a word exactly as you are about to lose it, and what a card has to contain for a review to be a retrieval at all is making cards that teach rather than cards that pile up.
The unresolved question is where the extra contexts are supposed to come from in a week that is already full, since manufacturing them one word at a time does not scale past a handful: filling your day with the language without moving country.