Spaced repetition is the most-cited idea in language learning and one of the least-read. It gets passed around as folklore: review at increasing intervals, trust the algorithm, done. The underlying research turns up as a sentence, if at all.
Which is a shame, because the literature is more interesting than the slogan, and a lot clearer about where the method stops working.
Every claim below is linked to its source. Check them.
The starting point: Ebbinghaus and his curve
In the 1880s, Hermann Ebbinghaus ran memory experiments on a single subject: himself. He memorized lists of nonsense syllables, consonant-vowel-consonant strings with no meaning, and measured how much effort it took to relearn them after various delays. The resulting "forgetting curve" showed memory dropping steeply at first, then flattening.
You have seen that curve. It appears in more or less every article written on this subject, usually without mentioning that it came from one man memorizing gibberish in the 1880s. So: does it hold up?
It does. In 2015, Jaap Murre and Joeri Dros published a careful replication in PLOS ONE. Their subject put in roughly 70 hours, learning and relearning lists of 104 syllables at delays running from 20 minutes to 31 days. The results tracked Ebbinghaus closely.
So the curve survives. What it never did is tell you what to do about it. It describes forgetting, it doesn’t prescribe a schedule. For that you need a different body of work.
The big one: Cepeda et al., 2006
The strongest evidence for spacing is a meta-analysis by Nicholas Cepeda, Harold Pashler, Edward Vul, John Wixted, and Doug Rohrer, published in Psychological Bulletin in 2006 (volume 132, pages 354–380). They pulled together 839 assessments from 317 experiments across 184 articles. For a psychology finding, that is an enormous evidence base.
Spacing beats cramming. You knew that. The finding worth your attention is the one the summaries skip:
The optimal gap between reviews grows as the delay before your test grows. There is no single correct interval. The right spacing depends entirely on how long you need to remember.
Which is why the interval schedules people copy off forums disappoint them. Cramming for a conversation next week? Short gaps are fine. Want the word in ten years? The gaps have to stretch much further than feels reasonable.
The finding that matters more than spacing
Here is the part that should change how you study. In 2008, Jeffrey Karpicke and Henry Roediger published "The Critical Importance of Retrieval for Learning" in Science. Handily for us, they ran it on foreign-language vocabulary.
Students learned Swahili-English word pairs. Once an item had been recalled correctly, it was either studied again, tested again, or dropped entirely. Then everyone came back a week later. Repeated studying of already-learned items did essentially nothing for recall. Repeated testing of the same items produced a large one. A week later the students who kept retrieving recalled around 80% of the pairs. The ones who only kept re-reading managed about a third.
Re-reading feels productive. It mostly isn’t. The work of dragging something back out of memory is what makes it stick, and spacing matters largely because it makes that dragging harder.
There’s a second result in that paper I find more unsettling than the first. The students’ predictions of their own performance were uncorrelated with how they actually did. They could not tell which method was working.
Desirable difficulties
Robert Bjork gave this family of effects a name in the 1990s: desirable difficulties. Conditions that slow you down during practice and improve what you retain afterwards. Spacing qualifies. So does retrieval practice, and interleaving topics rather than blocking them.
The consequence is awkward. Your sense of whether a session went well is close to worthless, and the sessions where everything flowed are often the ones that taught you least.
Where the evidence gets thin
Now the part that gets left out of the confident blog posts. Several things here are much less settled than the consensus suggests.
- Most studies use simple materials. Nonsense syllables and word pairs. Whether that extends cleanly to grammar intuition or spontaneous speech is much less established.
- Expanding intervals are not clearly superior. The intuitive idea that gaps should grow after each success is not decisively supported. Cepeda and colleagues found the evidence for expanding schedules over fixed ones weaker than commonly assumed.
- Recognition is not production. Recognising a word on a card is not the same skill as producing it mid-sentence while someone waits. Your deck stats flatter you.
- Lab conditions are not your Tuesday evening. Participants in these experiments show up and comply. You will skip four days, return to 300 due cards, and consider deleting the deck. Adherence swamps interval optimisation, and almost nobody studies it.
What this means for your actual routine
Stripped down to what the evidence actually supports:
- Test, do not re-read. Cover the answer and make yourself produce it. A card you never fail is a card that taught you nothing.
- Let the intervals get long. Months, for anything you want permanently. Reviewing while recall is still easy feels responsible and does very little.
- Accept the discomfort as the signal. Difficulty during practice predicts later retention. Your sense of how it’s going does not.
- Keep the daily load survivable. Twenty cards a day forever beats two hundred a day for a fortnight. The research mostly assumes this problem away; you can’t.
- Practice production separately. Cards build recognition. Writing your own sentences and saying them out loud builds the thing you actually wanted.
If you want the scheduling handled for you, our vocabulary tools implement these principles directly, with retrieval-first reviews with intervals that adapt to how well you actually recall each word, rather than a fixed ladder.
Short version: spacing is real and well evidenced. Retrieval is probably the bigger lever. Expanding intervals are oversold. And the imperfect schedule you keep will beat the optimal one you abandon.
