Learning vocabulary: what really helps against forgetting

What does the forgetting curve really show?

The famous curve comes from Hermann Ebbinghaus, who around 1880 memorised lists of nonsense syllables himself. After a break he learned the same list again and measured how much time he saved doing so. This “saving” is the number in the curve.⁠^[3]

In 2015, researchers at the University of Amsterdam repeated the experiment: one person, 70 hours of learning, the same intervals as Ebbinghaus.⁠^[3] The result is similar to the original:

Two things stand out. First, the curve drops steeply at the start and then flattens – most is lost in the first hours. Second, in the replication the savings after one day were even slightly higher than after nine hours; some researchers explain such jumps with the sleep in between.⁠^[3]

What the curve does not show is that you have “forgotten 70%” after a day. What was measured was the time saved when relearning nonsense syllables, not how many real words you still know.

Why testing yourself beats rereading

The best-known experiment on this was published in Science. Students learned 40 Swahili–English word pairs, such as mashua – boat. Until every word had been recalled once, all groups learned equally fast. After that they were treated differently: some kept being tested, others only got to reread words they had learned, or dropped them altogether.⁠^[1]

A week later the difference was clear. Those who had kept being tested knew about 80% of the words. Those who stopped being tested after the first correct answer recalled 36% and 33% – including the group that reread every word in every round. Extra reading added practically nothing; extra retrieval added a great deal.⁠^[1]

Remarkably, the students themselves had no idea. In all four groups they predicted they would know about half the words after a week.⁠^[1] That fits a large review of study techniques: testing yourself and spaced practice got the best rating there, rereading and highlighting the worst.⁠^[4]

Why spacing helps

That spaced learning beats cramming is one of the best-established findings in learning research. A 2006 meta-analysis already covered 839 assessments from 317 experiments.⁠^[5] For second language learning, a 2022 meta-analysis pooled 48 experiments with 3,411 participants: spaced practice had a medium-to-large effect.⁠^[2]

How far apart should reviews be? That depends on how long you want to remember something. In an experiment with more than 1,350 people, the best gap before a review grew the later the final test came. For a test after one week it was about 20 to 40% of that week; for a test after one year, about 5 to 10% of the year – a few weeks, in other words.⁠^[6] The material there was facts, though, not vocabulary.

The second language meta-analysis shows the same: short gaps work as well as long ones on a test right afterwards, but worse on a later test.⁠^[2] So short gaps are enough for tomorrow's quiz – for words you still need in a few weeks, longer ones are better.

Growing or equal intervals?

Many apps and the Leitner flashcard box let the intervals grow: first after a day, then after three, then after a week. That saves reviews. In research on second language learning, however, growing and equal intervals were statistically equivalent.⁠^[2]

For you this means: the exact algorithm matters less than is often claimed. What matters is that you test yourself at all – spread over days and weeks, not all in one evening.

Isn't it enough to read and listen a lot?

Reading and listening do bring new words, but slowly. In studies of incidental learning, learners on average kept only 9 to 18% of the new words they had met while reading, listening or watching, on a test right afterwards; for reading it was 17%, for listening 15%.⁠^[7] The two approaches complement each other: stories and conversations show you words in context, deliberate testing makes them stick.