Spaced repetition is the most validated learning technique in cognitive psychology. The forgetting curve, the spacing effect, the testing effect — all of it is reproducible, robust, and decades old. So you'd think dropping kanji into a generic SRS tool would Just Work.
It mostly doesn't. Here's why, and what to do about it.
What naïve SRS gets right
Take SuperMemo's SM-2 algorithm (and its Anki / Mnemosyne descendants). For each card, you rate recall on a scale, and the algorithm picks the next interval. Items you almost forgot come back sooner. Items you knew solidly come back later. The math works.
For a single, atomic fact — "the capital of Honduras is Tegucigalpa" — this is exactly what you want. One question, one answer, one recall attempt, one interval adjustment.
What naïve SRS gets wrong about kanji
A kanji is not an atomic fact. Take 学 (study). To "know" it means knowing:
- The meaning(s): study, learning, science.
- The on'yomi reading: ガク (gaku).
- The kun'yomi readings: まな(ぶ) (mana-bu).
- The stroke order (8 strokes).
- How it composes with other kanji: 学校 (school), 学生 (student).
- How it's used in context: 学ぶ (to learn), 大学 (university).
Naïve SRS picks one question per card and grades you on that. If your card asks "what does 学 mean?" and you get it right, you've reset the interval — even though you might still blank on the kun'yomi reading. The schedule says you know it. You don't.
Four failure modes you'll hit
1. The blind spot
You drill meanings and never test readings. Three months in, you can recognize 500 kanji but can't pronounce any of them. The SRS gave you the wrong feedback because you asked it the wrong question.
2. The interference cascade
Similar kanji (休 / 体, 末 / 未, 待 / 持) keep getting confused. Each time you mis-recall one, the schedule resets it but doesn't address the actual problem: the two characters are interfering with each other. They need to be reviewed apart, with disambiguation cues, not more frequently.
3. The leech that never dies
Some kanji just don't stick. Every review is a fresh failure. Naïve SRS keeps them in your daily pile forever, eroding morale. The algorithm needs leech detection — at some point, the cost of keeping a leech in rotation exceeds the value of learning it, and it should be set aside for a different approach (a custom mnemonic, a hand- written practice session, or simply a parking lot).
4. The "Easy" trap
You hit Easy on a card because you knew it cold. The schedule expands the interval aggressively. You don't see the card for two months. Two months later you blank. The Easy button is a self-trap if used too liberally — most learners should use Good for the majority of their responses.
What KanjiKanji Modern does differently
Adaptive scheduling per kanji per recall path. Meaning, on'yomi, and kun'yomi are tracked separately for the same kanji. The schedule responds to your weakest path, not your strongest.
Interleaving prevents similar kanji from being reviewed back-to-back. If you mis-recall 休 as 体, the next review of either is separated by at least a few unrelated cards.
Leech detection routes consistently-failed kanji into focused weak-point practice sessions. You stop seeing them in the daily review pile until they've been re-encoded.
Smart Review surfaces kanji you're ready to recall — not just kanji whose interval has expired. The result is a smaller daily pile with the same retention outcomes.
What you can do today, regardless of app
- Track readings separately from meanings. If your tool can't, supplement with a separate vocabulary deck that drills readings in context.
- Use Good, not Easy. Save Easy for the trivial cases. Default to Good even when you knew the card cold.
- Set a leech threshold. Any kanji you've failed 5+ times in a row gets pulled out of the schedule for a focused re-learning session — handwriting it three times, writing a sentence with it, drawing the radicals.
- Don't chase a 100% retention rate. The optimal recall rate for long-term retention is around 85–90% — meaning you should be failing roughly 1 in 10 cards. If you're hitting 100%, your intervals are too short.
One more thing
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 learning, you'll feel productive while making little progress. Encode well, then maintain. In that order.
Last updated