Reading patterns

MC -k → -ku, -ki (Mandarin: lost)

Middle Chinese syllables could end in a stop: -k, -t or -p (the 'entering tone'). Japanese could not end a syllable in a consonant, so it added a vowel: -k became -ku or, after front vowels, often -ki. Mandarin dropped these stops entirely, which is why 学 is xué but gaku.

Where it breaks: Before a voiceless consonant in a compound the -ku/-ki often turns into a doubled consonant (学校 gakkō, not gakukō). Which of -ku and -ki appears is not fully predictable.

MC -t → -tsu, -chi (Mandarin: lost)

Final -t was given a vowel too: -tsu or -chi. Both endings are old; -chi is frequent in readings of the go-on layer (一 ichi, 日 nichi) and -tsu in kan-on readings (一 itsu, 日 jitsu), but many characters have only one.

Where it breaks: Like -ku, -tsu doubles the next consonant in compounds (出発 shuppatsu). The -chi / -tsu split by layer is a tendency, not a rule.

MC -p → -fu → -u (long vowel); sometimes -tsu

Final -p became -pu, later -fu, and that -fu then merged into a long vowel. Old kana spelling still shows it: 十 was written ジフ (jifu), now jū. A few characters took -tsu instead (立 ritsu, 雑 zatsu), especially before voiceless consonants.

Where it breaks: Because -p merged into long vowels, these readings look exactly like readings from -ng (next pattern): 十 jū and 中 chū rhyme in Japanese but not in Chinese.

MC -ng → -u (ō, ū, yō) or -i (ei)

Japanese had no -ng, so it was heard as a nasal glide and written ウ or イ. Today those are long vowels: 東 tō, 中 chū, 生 sei. Mandarin kept -ng, so an on'yomi with a long vowel very often matches a Mandarin syllable in -ng.

Where it breaks: Many -ng characters have two readings, -yō (go-on) and -ei (kan-on): 明 myō / mei, 生 shō / sei. The -ei reading is more common in modern words.

MC -m, -n → -n (Mandarin: -n)

Middle Chinese kept -m and -n apart (三 sam, 山 srean). In modern on'yomi both are -n, exactly as in Mandarin. So a Mandarin syllable in -n almost always has an on'yomi in -n, and vice versa.

Where it breaks: Cantonese and Hokkien still keep -m (三 saam, 心 sam), so their speakers can tell the two apart; Mandarin and Japanese cannot. This is one of the most reliable patterns here.

MC m- → go m- / kan b- ; MC n-, ny- → go n- / kan d-, z-, j-

In the Chang'an dialect that kan-on copied, nasal initials had become partly oral (m → mb, n → nd). So kan-on readings start with b-, d-, z- or j- where older go-on readings keep m- or n-. Mandarin went its own way again: m- stayed (or became w-), ny- became r-.

Where it breaks: Which reading a word uses is lexical: Buddhist and older everyday words lean go-on, learned vocabulary leans kan-on, but you learn each compound as it comes.

MC b-, d-, g-, dz-, h- → go b, d, g, z, j / kan h, t, k, s

Middle Chinese had voiced initials (b-, d-, dz-, h-…) that later lost their voicing in the north. Go-on, borrowed earlier, keeps them as Japanese voiced consonants; kan-on, borrowed from Tang Chang'an where they were devoicing, uses voiceless ones. Mandarin also devoiced them, and splits them by tone into aspirated (平 píng) and unaspirated (地 dì).

Where it breaks: Japanese also voices consonants inside compounds for its own reasons (rendaku: 火山 kazan), so a voiced reading in a compound is not always go-on.

MC k-, kh-, g-, x-, h- → ja k-, g- ; Mandarin h-, or j-, q-, x- before i/ü

Old Japanese had no h-sound like Chinese x- or h-, so Chinese velar and glottal initials were all borrowed as k- or g-. Mandarin later turned velars before i and ü into j-, q-, x-, and did the same to the old sibilants s-, ts-. So a Mandarin j-, q- or x- has two possible sources, and Japanese still shows which: k/g for old velars, s/sh/z for old sibilants.

Where it breaks: A few h- characters came into Japanese as w- (和 wa, 話 wa) or with no consonant (王 ō). The k/s split for j/q/x is reliable; it is the old 尖团 (sharp/round) distinction.

MC p-, ph-, b- → ja h- (once p-, then f-) ; go-on b- for voiced b-

Japanese h- was once p-, then f-; it reached h- only in the last few centuries. So Chinese p- and ph- were borrowed as the sound that is now h-, and so was Chinese f- (which itself came from p- before certain vowels). The voiced b- shows up as b- in go-on and h- in kan-on.

Where it breaks: Inside compounds h- often turns into b- or p- (rendaku and gemination: 三本 sanbon, 一本 ippon, 出発 shuppatsu), so look at the reading of the character on its own.

go-on (5th–6th c.) · kan-on (7th–9th c.) · tō-on (12th c. →, few words)

Japan borrowed Chinese readings more than once. Go-on came early, roughly the 5th–6th centuries, through Korea and the southern dynasties, and lives on in Buddhist and old everyday words. Kan-on was deliberately learned from Tang Chang'an in the 7th–9th centuries and became the standard for learned vocabulary. Tō-on (sometimes sō-on), from the Song period onward, survives in a handful of words. One character can therefore have two or three on'yomi.

Where it breaks: The dates are approximate and the go-on layer is not one uniform system. 東京 Tōkyō uses the go-on kyō although it is a 19th-century name; layers describe readings, not when a word was coined.