The forgetting curve
In 1885, Hermann Ebbinghaus published a study on his own memory in which he showed that information is forgotten on a roughly exponential curve: most of the loss happens in the first hour or two, then tapers. The curve flattens — but only if the material is revisited before it's fully gone.
Each successful review pushes the curve out: the second time you recall something, the interval before you forget again is longer. The third time, longer still. This is the spacing effect.
What spaced repetition systems do
A spaced repetition system (SRS) automates the timing. After each review, it asks how confidently you recalled the material, and it schedules the next review accordingly. Items you almost forgot come back sooner. Items you know solidly come back later.
The dominant scheduling algorithm is SM-2, designed by Piotr Woźniak in 1987 for SuperMemo, and inherited by Anki, Mnemosyne, and many descendants. More recent algorithms (FSRS, the one used in Anki 23.10+, is a notable one) apply per-card difficulty and stability modeling for better predictions.
The SM-2 algorithm assigns each card an interval and an ease factor. Successful recalls increase the interval and may increase the ease factor; failures reset the interval and decrease the ease factor.
KanjiKanji Modern's three deviations from SM-2
- Per-recall-path scheduling. Meaning, on'yomi, and kun'yomi for the same kanji are scheduled independently. Getting the meaning right doesn't reset the reading interval.
- Interleaving against visual confusables. Similar kanji (休/体, 末/未, 待/持) are never scheduled back-to-back in a review session, even when their intervals coincide.
- Leech routing. Kanji that fail N consecutive reviews are pulled from the daily pile and routed into Smart Review's weak-point queue with handwriting practice and mnemonic re-encoding.
Why it works
Three effects combine. Spacing: distributed practice beats massed practice for long-term retention. Testing: actively retrieving information is more effective than re-reading it. Desirable difficulty: some struggle during retrieval is a feature, not a bug — easy recall doesn't strengthen the memory trace.
SRS arranges your practice so all three effects are working at once. That's why a 15-minute SRS session can produce more retention than an hour of re-reading.
Why kanji needs a tuned variant
Generic SRS treats every card the same. A kanji card is not the same as a country-capital card — there are multiple recall paths (meaning, on'yomi, kun'yomi, stroke order), interlocking radicals, and rich failure modes. Naïve SRS will tell you that you know a kanji when you only know one of its readings.
KanjiKanji Modern tracks recall per kanji per path, applies interleaving to keep similar kanji apart, and surfaces leeches into weak-point sessions rather than burying them in the daily pile. Deeper dive in the blog.
What SRS is not
SRS is not a learning technique — it's a memory maintenance technique. You still have to encode the kanji in the first place: meaning, sound, shape, context. SRS keeps the encoding alive. If you're using SRS as a substitute for active encoding, you'll feel productive while making little progress.
Related: Adaptive SRS feature · The Heisig method · Spaced repetition for kanji (blog)
Last updated