Why you forget vocabulary you were sure you had learned

Forgetting is the default, not the failure

The experience is universal: you look a word up, you understand the sentence, you feel you have learned it, and four days later it is gone. People treat this as evidence of a bad memory. It is evidence of a normal one.

Ebbinghaus measured the shape of this in the 1880s and it has held up: retention of something learned once falls steeply in the first day or two, then flattens into a long slow decline. The curve is not a personal failing to be overcome by trying harder. It is the baseline behaviour of memory, and the only thing that changes it is meeting the item again.

Recognition is not retrieval, and only one of them counts

Here is the trap. When you read a word with its translation underneath, you recognise it. Recognition is easy, feels like knowing, and is a poor predictor of whether you could produce the word tomorrow.

Retrieval — being shown the word and pulling the meaning out of your own head, or being shown the meaning and producing the word — is harder, feels worse, and is what actually strengthens the memory. The difficulty is not incidental to the benefit; it is the benefit. Psychologists call this desirable difficulty, and it is why highlighting a textbook does so little.

This is the single reason passive watching with dual subtitles teaches less than it feels like it should. Every line is recognised and none is retrieved. The subtitles are not the problem — they are what makes the material comprehensible at all — but they have to be paired with something that makes you produce.

The spacing effect, in one paragraph

Reviewing the same word five times in one sitting is worth far less than reviewing it five times across three weeks, even though the total effort is identical. Spreading the reviews out forces a partial forgetting between them, and recovering from partial forgetting is what makes the memory durable.

The optimal moment to review is just before you would have forgotten — late enough that recall is effortful, early enough that it succeeds. Nobody can judge that by feel, which is the entire reason scheduling software exists. A system that widens the interval on a correct answer and shortens it on a wrong one is approximating that moment for every card independently.

Why the sentence matters as much as the word

A word saved as a bare pair — arbitrage / refereeing — is a translation exercise. A word saved with the sentence it appeared in, the source it came from and the moment it was said is a memory with handles on it.

This matters more than it sounds. Most words do not have one meaning; they have a meaning in a context. Saving the context means that when the card comes back in nine days you are recalling something that actually happened to you, and that recall is far more likely to succeed.

It is worth doing this by hand if your tool does not do it for you — paste the sentence into the card. NoaLingua keeps the sentence, its translation, the source and the timestamp automatically because the alternative is a deck of orphaned pairs, but the principle holds whatever you use.

What to actually do

Three changes, in order of how much they return for the effort.

1. Stop re-reading. If you are not producing an answer from memory, you are not studying, however long you spend.
2. Capture the sentence, not the word. Any word worth saving appeared somewhere; save where.
3. Let something else decide when to review. The interval that feels right is almost always too short — see <a href="/features/vocabulary-review/">how the schedule works</a> for one implementation of this.