There are 2,136 characters in the list a Japanese education is built around, fixed at that number since the 2010 revision, and the first thousand or so account for the large majority of what turns up in ordinary text. So the job is bounded, which is unusual and worth saying, and the difficulty is not the count. It is that a character does not have a pronunciation until a word selects one.
Which decides the method. Learn characters inside words from the first day and the readings arrive attached to something that chose them; learn a character with a list of its readings and you have memorised a menu with no way to order from it.
Six facts that make kanji tractable
- 2136 jōyō kanji, the official list for everyday use, learned across twelve years of Japanese schooling. That is the target for reading a newspaper and it is a finite, published set.
- on and kun readings. Most characters have at least two: one borrowed from Chinese, used in compounds, and one native Japanese, used when the character stands alone. This is the single hardest thing about kanji.
- 山 is san or yama, mountain, and which one you say depends on the word it is in. Fujisan and yama are the same character.
- radicals organise the dictionary and also hint at meaning: characters containing the water radical tend to concern liquids, which makes a guess possible where a reading is not.
- furigana gives the reading, small kana printed above a character, used in material for children and learners and for unusual names. It is the built in answer key.
- stroke order is how you look it up by hand and how a phone recognises your finger, so it is practical rather than traditional.
Two families of reading, and the word decides
Readings divide into those that came in from Chinese and those that are native Japanese, and they surface in different places. The Chinese readings tend to appear when characters are stacked together into compounds, which is most of the formal and technical vocabulary. The native readings tend to appear when a character stands alone with hiragana attached to it, which is most of the everyday verbs and adjectives.
That is a tendency rather than a rule and it is strong enough to be useful. A compound of two kanji with no hiragana between them is probably taking Chinese readings; a kanji followed by hiragana is probably taking a native one. A learner who has that expectation guesses right often enough to read past unfamiliar words, which is the difference between reading and stopping.
The components underneath are the other half of the economy. Characters are built from a few hundred recurring parts, and a character met as three known components in an arrangement is cheap where the same character met as a picture is not. Ordering the study by component rather than by frequency means learning some characters before they are useful and every later one costing less.
The thousand kanji plateau
Here is where most people stop, and it is arithmetic rather than motivation. Up to roughly a thousand characters, every one learned is one that turns up constantly, so each hour of study is repaid within days and progress is obvious. Past that point the remaining characters are individually rarer, each hour buys something that appears once a fortnight, and the visible return collapses while the effort stays the same.
Nothing has gone wrong and the curve has not flattened; the reward per hour has. The exit is a change of method rather than more of the same. Past a thousand, characters arrive faster from reading than from a deck, because a rare character in a text comes with a context that makes it stick and the same character on a card comes with nothing. That means switching the balance from study to reading at exactly the point where reading first becomes possible, which most learners postpone because the deck still feels like the responsible option. The same shape of stall, and the same kind of exit, is set out in the stalls every learner hits, and the way through each.
The order, with no dates
How the character work actually gets built
There are deliberately no timescales on this. The honest range is one year to four and almost all of the variance sits at the fourth stop.
- startThe componentsA few hundred recurring parts, learned before the characters01
- 02Characters as shapesBuilt from parts already known, with the readings left out
- Words that contain themWhere the readings arrive, attached to something that chose them03
- 04Graded reading, in volumeThe switch most learners postpone, and the one that ends the plateau
- Rare characters from contextCheaper here than on a card, because the context does the work05
The fourth stop is the one that changes the rate, and it is the only stop that looks like less work rather than more. A learner who keeps adding cards past a thousand characters is choosing the slower of two roads because it is the one with a number on it.
Two things that order the work
Remembering the Kanji is the book that made the component-first argument, and it is still the clearest statement of it: characters are ordered so that every one is built from parts already learned, and each is attached to a story rather than to a reading. It replaced the school order, which follows what a Japanese child needs at each age and is therefore in the wrong order for an adult who cannot yet read. It is paid, it deliberately omits the readings, and that omission is the thing people argue about: it buys speed on the shapes and leaves the readings to be built afterwards, from words.
Tadoku's graded readers are the other half, free, written inside stated vocabulary and kanji limits so a whole book can be read without stopping. They replaced reading native material too early, which fails badly in Japanese because an unknown kanji cannot be sounded out and looked up the way an unknown word can in an alphabetic language. They are the material that makes the plateau exit possible, since a learner who cannot read anything cannot learn characters from reading.
The mistake: drilling readings as lists
A frequent character carries half a dozen readings and a few carry more than that. Learning the list is a very large amount of memorisation and it produces nothing usable, because in a real sentence you do not need to recall six possibilities, you need to know which one this word takes, and the list does not contain that information.
Learn a few hundred words that happen to contain the character and the readings arrive already attached to the words that select them. That is the same amount of memory doing a job you can actually use, and it has a side effect worth having: the vocabulary is what you needed anyway. The rule that follows is simple and almost nobody follows it at the start: no card ever has a bare character on the front, and every card has a word.
What the characters connect to
Reading a kanji compound depends on knowing which of the three vocabularies it belongs to, since the readings track the origin of the word, and that is native words, Chinese readings, and English borrowings. The hiragana attached to a character is what tells you it is taking a native reading, and it is also carrying the grammar, in particles, and a verb that waits at the end. None of this is legible until the kana themselves are invisible, which is ninety two symbols, and the fortnight it takes. And where the hours for two thousand characters come from in an ordinary week is turning a vague intention into hours on a calendar.