I Forgot All My Kanji: Why Kanji Decay Happens and How to Recover

If "I forgot all my kanji" is the panic that hit you after finishing or pausing an isolated kanji deck, the loss is a known, mechanical process. It is not a verdict on your ability.¹ It is one of the skill-specific stalls covered by the diagnostic guide to a stalled Japanese plateau. Characters learned in isolation decay faster than characters anchored in vocabulary, and the same mechanics that explain the loss also point straight at the recovery.

Overview

Forgetting is the default path of any memory that is not re-encountered.¹ An isolated kanji card, tied to a single English keyword, has few retrieval hooks. That puts it near the worst case on the forgetting curve.¹²

This article diagnoses what actually decayed, gives a relearning sprint that uses what you still half-know, and explains why vocabulary-driven exposure, not a revived isolated deck, is the durable fix.

Why You Forgot the Kanji

Decay is the default, not a personal failure

Forgetting is the normal fate of a memory that is not re-encountered, not a sign of a defective learner. Hermann Ebbinghaus established the first experimental forgetting curve in 1885 using nonsense syllables, chosen specifically to strip away prior meaning and isolate raw retention.¹

The curve is steep at first and then flattens. Most loss happens soon after learning, and the rate of further loss slows over time.¹ A 2015 replication confirmed a sharp initial drop, with the curve leveling after roughly the first day.³ This forgetting curve is what scheduled review on a spaced interval is designed to counteract.

An isolated kanji card is closer to one of Ebbinghaus's nonsense syllables than to a meaningful word. A bare character tied to one English keyword has few retrieval hooks, so it sits near the worst case on the curve.¹²

The savings effect is the reassuring part

Ebbinghaus measured retention with the savings method: how much less time, or how many fewer repetitions, relearning takes compared with first learning.¹ His results ran from about 58% saved at 20 minutes down to about 21% saved at 31 days.³ Even material you cannot consciously recall relearns faster than it was first learned.

The isolated-deck trap

Levels-of-processing research shows that how deeply material is encoded, not just how long it is rehearsed, predicts retention. Shallow encodings (a surface form, a single label) leave fragile traces prone to rapid decay. Deep, meaning-based encodings leave more durable traces.²

Craik and Tulving demonstrated this directly: words processed for meaning and fitted into a sentence context were remembered far better than words processed only for surface features.⁴ An isolated kanji-with-keyword card is the shallow-end case. The same character met inside a real word the learner understands is the deep-end case.

A bare English keyword is, by Heisig's own framing, a means to an end rather than the target knowledge. Remembering the Kanji 1, the basis of the Heisig method, deliberately teaches only meaning and writing. It defers readings entirely.⁵

Once a learner starts reading, the keyword has no role in comprehension and is among the first things shed.⁶ The isolated deck supplies only that thin, single-link encoding and no recurring meaningful context. Once review stops, its characters sit closest to the steep part of the forgetting curve.¹²

Diagnosing What Actually Decayed

Keyword vs. reading vs. production

"I forgot the kanji" mixes together three separable kinds of knowledge. Each decays at its own rate and needs a different fix. A blanket "restart everything" wastes effort on the layer that may not be worth saving and under-treats the layer that matters.

What decayed	What it is	How fragile
Meaning / English keyword	The label a deck attaches to a character	Thinnest and most disposable; designed to be discarded once reading proficiency rises56
Reading in context	The on'yomi or kun'yomi, or reading, a character takes inside a specific word	The knowledge that actually supports reading; an isolated meaning-only deck never built it5
Production / handwriting	Reproducing the character from memory	The most demanding retrieval; decays fastest7

The reading layer supports comprehension, and an isolated deck never trained it. That is where relearning should concentrate.

You know more than the panic suggests

Recognition memory is supported by familiarity, a faster and more durable process than the conscious recollection that recall requires. The two processes can separate, and recognition typically survives when free recall has already failed.⁷

In practice, most "forgotten" kanji are recognized instantly when met inside a real word, even when the learner could not produce or define them cold. That half-known recognition signal is the lever the relearning sprint uses.

This fits the savings effect: even material that yields zero conscious recall still relearns faster than it was first learned, because a sub-threshold trace persists.¹³

The Relearning Sprint

Don't restart the deck from card 1

Relearning is reliably faster than first learning. This is the savings effect: the second pass needs fewer repetitions than the first, because a residual trace persists even below the threshold of recall.¹³

The practical move is to mark genuinely known cards as known and let the SRS resurface only the true leeches, rather than grinding the whole deck from the start. The algorithm plus the savings effect does the triage for you.

Let the algorithm find the leeches

Distributed (spaced) practice compounds the savings effect: spacing slows the acquisition pass but increases later retention and accelerates relearning on test, a robustly replicated finding.⁸ Resetting a deck to "known" wherever you genuinely recognize a card lets spaced review concentrate on the genuine leeches.

Sprint structure and review-load math

A sustainable relearning block caps the daily new and relearn count so the workload does not spike. Spaced practice is more effective than massed cramming for durable retention, so a steady, time-boxed daily block beats a marathon session.⁸

Put recognition before production. Recognition relies on the durable familiarity process and recovers first. Production (handwriting) relies on effortful recall and should be deferred or treated as optional.⁷

The specific card-load numbers and burnout thresholds belong to the dedicated discussions on computing a sustainable new-card ceiling and the exit signs of SRS burnout. This sprint points there rather than re-deriving them.

Anchor every kanji to a real word

The operative move is to relearn each character through a vocabulary item rather than a bare keyword. That gives it a second, meaning-based retrieval hook from the first relearn. This is the kanji-in-context approach applied to recovery. It follows the levels-of-processing prediction directly: semantic, in-context encoding produces a more durable trace than a shallow surface-label encoding.²⁴

Nation's account of vocabulary learning supports the same move. Words are learned and retained through repeated, meaningful encounters across varied contexts rather than through isolated drills. Learning is cumulative and must be recycled.⁹ In practice, this means building the relearn deck from words you actually meet while reading, using the sentence-mining workflow, rather than from a pre-made keyword list.

Why Vocabulary-Driven Exposure Is the Durable Fix

Insufficient encountering is the root cause

Kanji decay because they are not met often enough in meaningful input. Retention of a lexical item, such as a word, depends on multiple meaningful encounters over time. A stable representation requires repeated spaced exposure in context, not a one-pass drill.⁹

An isolated deck supplies no natural re-encountering once review stops; its only repetitions are the artificial card reviews. Reading supplies meaningful re-encounters continuously and without extra drill. That is what moves a character off the steep part of the forgetting curve.¹⁹

Levels-of-processing explains why at the encoding level: each meaningful in-context encounter is a deep, semantic processing event. Deep processing leaves the durable trace that shallow keyword review cannot.²⁴

Jukugo multiply your retrieval hooks

A single kanji typically appears across many compound words (jukugo, 熟語). Every time any one of those words is read, the shared character is re-encountered and reinforced. One kanji therefore accumulates many independent retrieval hooks instead of the single keyword link an isolated deck gives it.⁹

This is the multiple-meaningful-encounters principle operating structurally. The in-context approach converts one character into as many memory anchors as it has high-frequency compounds. That is precisely the variety-of-contexts condition Nation identifies as driving retention.⁹

The deck route gives the character one fragile hook. The vocabulary route gives it one hook per compound the learner reads, each renewed on every encounter.

To relearn kanji through the words that use them rather than in isolation, J-Compass recommends Amenokori. It surfaces each kanji inside its vocabulary entries, showing on'yomi, kun'yomi, and meanings across a frequency-sorted treatment of the 2,136 jōyō characters, scheduled with the FSRS spaced-repetition algorithm.¹⁰¹¹ This puts the relearn-via-vocabulary move at the center of this article into practice.

Read the per-level word counts as vendor leveling

The per-level word counts Amenokori shows on its collection cards are vendor leveling, not official JLPT counts. The JLPT has published no official vocabulary list since the 2010 redesign.¹¹ Treat Amenokori as one tool in a small field, J-Compass's recommended fit for this exact problem, rather than as an authority on what belongs at each level.

When (and whether) to revive an isolated deck

A brief keyword-refresh sprint can be worth it for component literacy: the ability to recognize shared shapes and radicals across characters. RTK-style decks are built to teach the writing-and-meaning layer that in-context reading does not directly drill.⁵

A permanent second isolated deck, however, repeats the original mistake. It rebuilds the thin, single-link encoding that decayed in the first place, instead of the multiple-meaningful-encounter exposure that produces durable retention.¹²⁹

Three variables decide the trade-off: your goal (reading-only vs. handwriting), how much component and radical literacy you still have, and whether you have a reading habit that supplies natural re-encounters. The default is a short, time-boxed keyword refresh at most. Then shift the durable load to vocabulary-driven reading.⁹

Good to know

Handwriting decays first and that may be fine

Production (handwriting from memory) depends on effortful recall and is the most fragile of the three knowledge types. Recognition, supported by the durable familiarity process, persists long after handwriting has faded.⁷ Losing the ability to write a character by hand does not mean the reading or meaning is gone.

Whether to fight handwriting decay depends on your goals. That debate has its own home, so this article does not re-argue it here.

Recognition without reading is a mirage

Recognizing a character and recalling its English keyword feels like knowledge, but it does not supply the in-word reading that comprehension needs. RTK explicitly teaches meaning while deferring readings.⁵ The relearning sprint must test the reading in context, not just keyword recognition. Otherwise, it certifies a layer that does not transfer to reading.

Don't over-mnemonic the relearn

Recognition memory means most "forgotten" characters retain a sub-threshold familiarity trace that relearns quickly through the savings effect.¹³⁷ Building elaborate new stories for those characters wastes the recognition signal you already have.

Mnemonics are deep, elaborative encoding and stay useful for genuine leeches that resist relearning.⁴ The pitfall is applying them indiscriminately to characters you already half-know.

See also

• Why You've Hit a Japanese Plateau: A Diagnostic Guide to Stalled Progress
• Learning Kanji Through Vocabulary: The Kanji-in-Context Approach
• How to Learn Kanji: A Strategic Overview of Heisig, WaniKani, and Kanji-in-Context
• A Daily Kanji Study Routine: How Many Kanji per Day, Review-Load Math, and the Three-Block Schedule
• Why Anki Has Become Painful (and How to Fix It): A Diagnostic Triage
• When SRS Becomes Counterproductive: Anki Burnout, Leeches, and the Exit Signs

References

Footnotes

1. Ebbinghaus, Hermann. Über das Gedächtnis: Untersuchungen zur experimentellen Psychologie. Duncker & Humblot, 1885. (English: Memory: A Contribution to Experimental Psychology, trans. Henry A. Ruger and Clara E. Bussenius, Teachers College, Columbia University, 1913.) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³

2. Craik, Fergus I. M., and Robert S. Lockhart. "Levels of Processing: A Framework for Memory Research." Journal of Verbal Learning and Verbal Behavior, vol. 11, no. 6, 1972, pp. 671–684. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

3. Murre, Jaap M. J., and Joeri Dros. "Replication and Analysis of Ebbinghaus' Forgetting Curve." PLOS ONE, vol. 10, no. 7, 2015, e0120644. https://doi.org/10.1371/journal.pone.0120644 ↩ ↩² ↩³ ↩⁴ ↩⁵

4. Craik, Fergus I. M., and Endel Tulving. "Depth of Processing and the Retention of Words in Episodic Memory." Journal of Experimental Psychology: General, vol. 104, no. 3, 1975, pp. 268–294. ↩ ↩² ↩³ ↩⁴

5. Heisig, James W. Remembering the Kanji 1: A Complete Course on How Not to Forget the Meaning and Writing of Japanese Characters. 6th ed., University of Hawai'i Press, 2011. ↩ ↩² ↩³ ↩⁴ ↩⁵

6. Tatsumoto Ren. "Learning Kanji." Refold / AJATT-lineage immersion guide, community blog. https://tatsumoto-ren.github.io/blog/learning-kanji.html (limitation: community immersion blog; used only for the illustrative keyword-decay observation, which is corroborated by ²⁴⁵) ↩ ↩²

7. Mandler, George. "Recognizing: The Judgment of Previous Occurrence." Psychological Review, vol. 87, no. 3, 1980, pp. 252–271. ↩ ↩² ↩³ ↩⁴ ↩⁵

8. Cepeda, Nicholas J., Harold Pashler, Edward Vul, John T. Wixted, and Doug Rohrer. "Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis." Psychological Bulletin, vol. 132, no. 3, 2006, pp. 354–380. ↩ ↩²

9. Nation, I. S. P. Learning Vocabulary in Another Language. 2nd ed., Cambridge University Press, 2013. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

10. 文化庁 (Agency for Cultural Affairs). 「常用漢字表」(Jōyō Kanji Table), 2010 revision. 2,136 characters. https://www.bunka.go.jp/kokugo_nihongo/sisaku/joho/joho/kijun/naikaku/kanji/ ↩

11. Amenokori. Product landing page. https://amenokori.com ↩ ↩²