Why one kanji has several readings, and how to pick one

A kanji has several readings because Japan borrowed the characters without borrowing the language they were written for, so each character ended up carrying two things at once: an imitation of its Chinese pronunciation, and whichever native Japanese word already meant that thing. You choose between them by looking at the word the character sits in, never at the character alone. Kana trailing behind it point to the native reading; a run of kanji with no kana between them points to the borrowed one.

That rule is right most of the time, and the rest of this is where it comes from, what breaks it, and why the reading turns out to be a property of the word rather than of the character.

Japan borrowed the characters, not the language

Chinese writing reached Japan by way of the Korean peninsula, into a country that already had a spoken language and no way to write it. Japanese and Chinese are unrelated: different grammar, different sound system, different word shapes. Nothing about the characters fit, and they were used anyway.

Two things happened at the same time, and both stuck.

Scribes approximated the Chinese pronunciation using Japanese sounds. That is the on'yomi (音読み), the sound reading. And they attached each character to the Japanese word that already meant the same thing, which is the kun'yomi (訓読み), the gloss reading.

水 is the clean case. Its on reading is すい and its kun reading is みず. Nobody stopped saying みず when the character arrived; they were simply given a way to write it. Meanwhile すい came along with a whole cargo of imported vocabulary that still uses it: 水曜日, 水分, 洪水.

So the two readings are not variants of each other, and neither is the real one. They are two vocabularies sharing a sign — a native layer and a borrowed layer, roughly the way English keeps ask alongside interrogate and uses them in different rooms.

One written verb can also hide two unrelated native words: 認める read みとめる or したためる means “acknowledge” in one context and “compose a letter” in another. The object and the register settle the reading.

A whole compound can do the same thing: 仮名 is かな for the kana scripts but かめい for an assumed name. Here the reading selects a different word and sense even though the written form is identical. 明白 hides the same split: it is めいはく for "obvious", while the separate word あからさま, "blatant", is filed under those characters too and normally left in kana — one of the N1 words for truth, clarity and their opposites.

The split between 博士（はかせ）for an expert and 博士（はくし）for a doctoral degree follows the same principle: the phrase around the spelling tells you which word is present.

Single-character nouns can do the same thing. In the N3 fabric and fashion set, 柄 is え for an umbrella handle and がら for a fabric pattern, while 綿 is めん on a cotton label and わた for loose padding. The surrounding phrase carries the choice.

The N3 map-and-route lesson puts that rule to work twice: 角 is かく for an angle and すみ for the assigned inner-corner record, while 表 is ひょう for a table even though the front of a sheet is おもて.

At N1, 値 is ね for price but あたい for numerical value, while 包む can be くるむ or つつむ.

The same principle reaches a whole word in the N1 textile set: 流行 is はやり when it names the fashion itself, but りゅうこう in 流行する. A visible 流行り or the following する settles the reading before any general kanji rule does.

The mirror case exists too. Sometimes two readings of one word carry the same sense, so the reading choice is free and a different signal has to do the work. 尊い reads both とうとい and たっとい with no shift in meaning, and it is the kanji that decides: 尊い for what is revered, 貴い for what is merely precious. That is the axis the N1 courtesy set turns on.

The borrowing happened more than once

If it had been a single event, every character would have exactly two readings and this would be a short article. Several characters have five or six, and the reason is that the borrowing came in waves, centuries apart, from different parts of China.

• 呉音 (go-on) — the oldest layer, arriving early and indirectly, and concentrated in Buddhist vocabulary.
• 漢音 (kan-on) — brought back from the Tang capital by envoys and students, and promoted by the court as the correct pronunciation. Most on readings you meet are this layer.
• 唐音 (tō-on) — later still, arriving with Zen and trade. Rare, and it clings to a small set of specific words.

行 collected all three:

• ぎょう — 行列, 修行
• こう — 銀行, 旅行
• あん — 行灯, 行脚

Add the native readings いく, ゆく and おこなう and one character is carrying six pronunciations. Not one of them is an irregularity. Each is a fossil of a particular century, sitting in the vocabulary that arrived with it.

This is more useful than it sounds. When a character has several on readings, the reading generally travels with its vocabulary layer, so 修行 and 銀行 are not a coin toss once you know the words themselves. The same split reaches into formal vocabulary: 文書 is ぶんしょ for a modern document but もんじょ in 古文書, one of the N1 words for putting language on the record.

The shape of the written word tells you which reading to apply

Here is the part that resolves most real cases. Do not look at the character. Look at what is around it.

Kanji followed by kana that change takes the kun reading. 生きる, 生まれる, 読まない. Those trailing kana — okurigana — are the inflecting tail of a native Japanese verb or adjective, and a borrowed Chinese syllable never had one. The presence of okurigana is close to proof.

Two or more kanji locked together with no kana takes the on reading. 学生, 生活, 一生, 銀行, 説明. These are compounds built on the Chinese pattern, and they use the imported pronunciations.

To practise that prediction with a bounded beginner set, split ten N4 compounds into their on-reading sound blocks, then compare the clean joins with 出発（しゅっぱつ）.

A suffix can narrow the reading further. 消極的 and the ending ～的 show 的 read てき inside a descriptive word, with な or に revealing how that word fits the sentence. The complete phrase supplies more information than a loose kanji reading.

A single kanji standing alone as a noun is usually kun. The four-word N5 nature set keeps 山（やま）and 川（かわ）in that standalone form; 人（ひと）and 空（そら）follow the same broad pattern.

That third rule leaks, and it is worth knowing exactly how. Plenty of single-character nouns take the on reading — 本 (ほん), 駅 (えき), 茶 (ちゃ), 肉 (にく), 点 (てん) — because the word was borrowed whole, not just the character. The test was never the character's appearance. It is whether the word came from Chinese, which is the same principle stated once more.

Run 生 through all of it and the system shows itself:

written	read	why
生きる	いきる	okurigana, so a native verb
生まれる	うまれる	a different native verb, same character
学生	がくせい	kanji compound, so on reading
一生	いっしょう	compound, the other on reading
生ビール	なまビール	native reading, no okurigana
芝生	しばふ	neither rule; this one you look up

Food labels add another split: 生（き） for undiluted, beside 生（なま） for raw or unheated, and 生醤油 can take either reading when the product distinction changes.

Five of those six are predictable from the shape of the word. The sixth is the honest reason the rule is a rule of thumb.

Two mixed patterns look like the rule failing, and both have names

Some compounds take one reading from each layer, which feels like a violation until you learn that Japanese noticed it centuries ago and named both cases after words that demonstrate themselves.

重箱読み (jūbako-yomi) is on followed by kun. The word 重箱 is itself じゅう (on) plus ばこ (kun). Others: 台所 (だいどころ), 番組 (ばんぐみ), 本屋 (ほんや).

湯桶読み (yutō-yomi) is the reverse, kun followed by on. 湯桶 is ゆ (kun) plus とう (on). Others: 場所 (ばしょ), 手本 (てほん), 夕刊 (ゆうかん).

A mixed reading in a practical word is 歯医者（はいしゃ）: native 歯（は） joins 医者（いしゃ）. The health-word lesson places it in a dental appointment and compares it with 入院（にゅういん） and 退院（たいいん）, where the shared 院 keeps いん.

Then there is 熟字訓 (jukujikun), which is a different animal. Here a native reading is assigned to a compound as a whole and cannot be divided among its characters. 今日 is きょう. There is no sense in which 今 is きょ and 日 is う. The characters carry the meaning, and the pronunciation is a separate native word laid over the top of them.

The common ones are extremely common: 昨日 (きのう), 大人 (おとな), 一人 (ひとり), 果物 (くだもの), 眼鏡 (めがね), 時計 (とけい). 明日 is the same trick and shows the layers stacked: あした in speech, and みょうにち when a train announcement wants to be formal. For 昨日, 今日 and 明日 in sentences, the N5 time-word lesson pairs those readings with their usual particle pattern.

Menus are unusually thick with them — 海老 (えび), 小豆 (あずき), 蕎麦 (そば) — which is one reason a menu can defeat you even when you recognise every character on it. Knowing the characters and knowing the word are separate achievements, and jukujikun is where the gap between them is widest. The N3 travel and lodging lesson applies that point to everyday 土産（みやげ）, then separates it from the rare どさん reading. At N1 the same problem reappears inside ordinary-looking words such as 玄人 and 土産, where the characters are familiar and the reading is anything but.

Readings change shape where words join

The other reason a reading you genuinely know can arrive unrecognisable: it gets deformed at the seam.

Rendaku (連濁) voices the first consonant of the second element. 手 plus 紙 gives てがみ. 花 plus 火 gives はなび. 人 doubled gives ひとびと. It is regular enough to expect, and it has a well-known brake: rendaku is blocked when the second element already contains a voiced obstruent, which is why 山風 stays やまかぜ and never becomes やまがぜ.

Gemination does the same job to on readings. Those ending in く, ち or つ tend to collapse into a small っ before the next consonant. 学 (がく) plus 校 (こう) gives がっこう. 出 (しゅつ) plus 発 (はつ) gives しゅっぱつ. 一 (いち) plus 回 gives いっかい.

The counter 匹 shows the whole business in one line: 一匹 (いっぴき), 三匹 (さんびき), 六匹 (ろっぴき). One character, three surface shapes, and no dictionary lists them as three readings, because they are not.

This matters more than it looks. Without it, a word that sounds nothing like the reading you learned is experienced as I learned this and it was wrong. Usually the reading is intact and something predictable happened to it on contact.

Names run on a separate system, and you are not meant to guess

Personal and place names use 名乗り (nanori) readings, a set that exists in names and effectively nowhere else.

The surname 羽生 is read はにゅう by some families and はぶ by others. Identical characters, and nothing in the writing distinguishes them. Places do it too: 日本橋 is にほんばし in Tokyo and にっぽんばし in Osaka, same three characters both times.

The proof that this is not a gap in your study: Japanese official forms print a ふりがな field above the line where you write your name. The institution assumes that a native speaker, holding the characters, cannot reliably produce the sounds.

So when a name defeats you, nothing has gone wrong. Ask, or read the furigana, and expect none of it to generalise.

Most characters carry far less than 生 does

生 gets quoted constantly because it is the extreme case, and quoting it makes the language look worse than it is. The ordinary character is unremarkable:

• 校 — こう. One reading.
• 駅 — えき. One.
• 犬 — けん and いぬ. Two, one from each layer.

The 常用漢字 list, 2,136 characters since its 2010 revision, sets out the readings taught and expected in general use, and the characters hoarding five or six of them are a small and famous group: 生, 行, 上, 下, 日, 人. Those are also the characters you meet every single day, which means the words teach you their readings whether or not you ever sit down with a table.

The impression that Japanese has thousands of unpredictable readings comes from meeting that handful early and generalising from them.

Which is why the unit you learn is the word

If a reading only resolves inside a word, then drilling a character on its own is teaching yourself a set of sounds with no rule for choosing between them. You can recite that 生 is セイ・ショウ・いきる・うまれる・なま・き and still stop dead at 芝生, because the recitation never contained the information you needed.

Turned around, it gets much easier. Learn the word and let it carry the character: meet 生 in 学生, 生ビール, 生まれる and 一生, and you have four readings, each already attached to a meaning and a situation that tells you when it fires. A character you have met in four words is known in a way that forty repetitions in isolation cannot produce.

Which puts the weight on the card. A Japanese card without a reading field is not a card — the characters do not tell you the sound, and that is the part you are actually learning. Kikusho writes the reading, the romaji, the meaning, a register note and an example sentence onto every card by default, and lets you select any word inside that example sentence to make it a card of its own. That last part matters here specifically: it is how one character arrives six times as six words, rather than once as a table you will not remember.

That confusion is also the failure mode readings are most prone to, because recognition decays differently from recall. You see 生活, think I know that one, and never test whether せいかつ was actually there. It was not, and the deck told you nothing, because you never asked it the question.

The character is the sign. The word is the thing you are learning. Japanese only looks arbitrary when you mistake one for the other.