Rendaku in Kanji Compounds: Why 紙 Becomes -gami
Rendaku in kanji compounds is the sequential voicing of a second element's initial consonant when two morphemes join. That is why 紙 (kami, "paper") surfaces as -gami in 折り紙 (origami).1 The kanji shape never changes; only the spoken reading shifts. This is why a learner who has memorised on'yomi and kun'yomi still meets compounds whose readings do not follow the table cleanly.1
Overview
What this article covers, and what it defers
The phonological theory of rendaku lives in the prerequisite article on rendaku in compound words. That includes the obstruent-voicing rule, its origin in Old Japanese prenasalisation, and the cross-stratum statistics in the abstract.23 This article applies that theory to kanji compounds specifically.
The scope here is narrower and more practical: how rendaku surfaces in kanji writing, why it correlates with the kun+kun reading pattern, why it usually skips on+on compounds, and how to predict it while reading.14
Audience and JLPT level
The target reader is an N3+ self-studier who already understands on'yomi versus kun'yomi and has met unpredictable voicing in compound vocabulary. The wago versus kango split that drives rendaku is essentially the kun+kun versus on+on distinction, so the on'yomi versus kun'yomi prerequisite article is the natural companion read.14
The 折り紙 case in one paragraph
折り紙 breaks down as 折り (ori, the kun'yomi stem of 折る "to fold") plus 紙 (kami, kun'yomi "paper").5 Both elements are wago (native), so the kun+kun compound triggers rendaku by default: kami voices to gami, giving origami.15
折り紙を折るのが好きです。5
"I like folding origami."
The first thing to notice is the writing-system asymmetry. The kanji 紙 stays 紙 inside 折り紙; the voicing appears only in kana (おがみ with が, not か).6
How rendaku surfaces in kanji compounds
The kanji stays, the reading voices
Rendaku is a sound process at morpheme boundaries, not a spelling process. The kanji 紙 is the same character in 紙 (kami, "paper"), 折り紙 (origami), and 手紙 (tegami); only the spoken reading shifts from /k/ to /ɡ/.1
In kana writing, the voicing is visible: 折り紙 is spelled おりがみ, with the k-row kana か replaced by its g-row counterpart が via dakuten.16 Furigana track the spoken reading, not the kanji-isolated reading. So a dictionary entry for 折り紙 prints おりがみ above the kanji block, with が above 紙, even though the standalone reading of 紙 is かみ.5
The kanji shape is stratum-agnostic: it does not show whether the word is wago or kango. The spoken voicing is stratum-conditioned. 紙 has one kun'yomi (kami) that voices to gami at the compound's right edge. The character does not carry a second reading; the compound conditions a sound shift on the same reading.1
The four voicing shifts in kanji-compound form
The four target rows are the four obstruent rows of the gojūon that already take dakuten in the writing system: k→g, s→z, t→d, h→b.12 Rendaku has no effect on m, n, y, w, or r initials. Those rows are already voiced or are sonorants, and they have no voiced-obstruent counterpart in the kana inventory.12
| Voicing | Kana shift | Compound example | Composition |
|---|---|---|---|
| k → g | か → が | 折り紙 origami | ori + kami → ori-gami5 |
| k → g | か → が | 手紙 tegami "letter" | te + kami → te-gami7 |
| s → z | さ → ざ | 山桜 yamazakura "wild cherry" | yama + sakura → yama-zakura1 |
| s → z | す → ず | 巻き寿司 maki-zushi "rolled sushi" | maki + sushi → maki-zushi1 |
| t → d | ち → ぢ／じ | 鼻血 hana-ji "nosebleed" | hana + chi → hana-ji67 |
| t → d | つ → づ | 三日月 mikazuki "crescent moon" | mika + tsuki → mika-zuki6 |
| h → b | は → ば | 石橋 ishi-bashi "stone bridge" | ishi + hashi → ishi-bashi7 |
| h → b | ひ → び | 花火 hana-bi "fireworks" | hana + hi → hana-bi1 |
The t-row case has a spelling wrinkle. The kana ち historically voices to ぢ, and in most modern contexts it is written じ. The 1986 Cabinet notification on modern kana usage (内閣告示第一号) preserves the ぢ spelling in two narrow situations: same-morpheme reduplication of the chi/tsu series (ちぢむ "to shrink") and two-word compounding where the rendaku boundary is transparent (はなぢ "nosebleed", みかづき "crescent moon").6 The full obstruent-voicing table and its phonological motivation are in the parent rendaku article.2
母に手紙を書いた。7
"I wrote a letter to my mother."
Reduplication and 々
The iteration mark 々 doubles the preceding kanji. It is the most reliable visual cue that rendaku will fire: 人々 hito-bito "people", 時々 toki-doki "sometimes", 国々 kuni-guni "countries", 神々 kami-gami "gods", 木々 ki-gi "trees".18
時々本を読みます。8
"I read books sometimes."
Reduplicated compounds are reliable for a structural reason. Every reduplication is wago by definition, and the morpheme boundary is unambiguous. The second element is the same morpheme as the first, so the writer already knows its stratum. Lyman's Law also has only one chance to fire because there is only one obstruent in the second element.12 The result is rendaku in roughly every reduplicated wago compound whose initial is k, s, t, or h.12
山々 yamayama is wago and reduplicated, but it still does not voice. The reason is structural rather than mimetic. Rendaku targets only the four voiceless obstruent rows /k s t h/. yama starts with /j/ (the y-row glide), which has no voiced-obstruent counterpart in the kana inventory and therefore no rendaku target. The same applies to any m, n, y, w, or r-initial reduplication, including 海々 (vowel-initial, also no target).12
Native compounds (kun+kun): rendaku is the default
Why kun+kun compounds voice
Kun'yomi readings are native Japanese morphemes (wago). Rendaku is a wago-internal sound process. It applies to roughly 90% of eligible wago compounds, a figure standardly cited from Vance 1996 and reproduced in Irwin 2005 and the NINJAL Rendaku Database.4910
Historically, rendaku traces back to a Late Old Japanese reanalysis: an enclitic genitive /no/ followed by prenasalisation of the following stop. Over time, this was absorbed into the second-element initial as a voiced obstruent.3 The pattern therefore lives natively in the wago stratum and was already entrenched before kango entered the lexicon in volume.32
"Eligible" matters here. A wago second element must start with /k s t h/ and must not contain an internal voiced obstruent (Lyman's Law). The compound must also be head-modifier rather than dvandva.1
Worked patterns
A short tour of standard kun+kun voicings shows the rule's reach.
- 折り紙 ori + kami → ori-gami (k→g, both wago).5
- 手紙 te + kami → te-gami (k→g, both wago).7
- 石橋 ishi + hashi → ishi-bashi (h→b, both wago); attested in the proverb 石橋を叩いて渡る "tap the stone bridge before crossing".7
- 花火 hana + hi → hana-bi (h→b, both wago).1
- 山道 yama + michi → yama-michi "mountain path" (no rendaku because michi is m-initial, no target).2
- 中島 naka + shima → naka-jima (s→z surface, written じ; common surname).7
夏祭りで花火を見た。7
"We saw fireworks at the summer festival."
Surnames and place names
Surnames and place names are where the wago default becomes especially important. The same kanji string can split into two readings with two meanings.
山川 distinguishes yamakawa "mountains and rivers" (dvandva, parallel and coordinate, no rendaku) from yamagawa "mountain river" (head-modifier, rendaku fires).111 中田 splits regionally between nakata (no rendaku) and nakada (rendaku) without any meaning contrast. The choice is lexicalised family by family.1 中島 carries the parallel split nakajima versus nakashima, again family-specific.7
山川さんは京都出身です。7
"Ms. Yamakawa is from Kyoto."
Sino-Japanese compounds (on+on): rendaku usually skips
The on+on default: no voicing
Kango compounds undergo rendaku in roughly 10% of binoms. Mononoms inside hybrid wago plus kango compounds voice in about 20% of cases, per Vance 1996 and Irwin 2005.4910 That is about an order of magnitude less than wago.
Standard "no rendaku" examples fill everyday vocabulary: 学校 gakkō, 図書館 toshokan, 大学 daigaku, 新聞 shinbun, 計算 keisan, 公園 kōen, 電車 densha.112
毎朝学校に行きます。12
"I go to school every morning."
図書館で本を借りた。12
"I borrowed a book at the library."
Why kango resists rendaku
Three explanations point to the same prediction.
The homophony-avoidance account observes that Sino-Japanese morphemes often begin with voiced obstruents on their own, including 元 gen, 大 dai, 図 zu, and 病 byō. Voicing the initial of a kango second element would erase a contrast the lexicon actively uses. By contrast, wago morphemes very rarely begin with voiced obstruents, so wago-internal voicing collides with nothing.12
The prosodic and structural account treats kango compounds as right-branching at the morpheme level, with each on'yomi bound and tightly connected. Otsu's branching constraint predicts that rendaku will not apply in such structures.11
The historical account is the simplest. Rendaku is older than most of the kango stratum. The prenasalisation source died out before kango entered the lexicon in volume, so kango never had the chance to participate in rendaku productively.3 The deeper theoretical motivation is in the parent rendaku article.2
When kango does voice: the lexicalised exceptions
A short, memorisable list captures most of the kango compounds that do voice. Treat these as lexicalised exceptions to memorise, not productive cases to derive.49
| Compound | Reading | Voicing |
|---|---|---|
| 株式会社 | kabushiki-gaisha | kaisha → gaisha112 |
| 黒砂糖 | kuro-zatō | satō → zatō12 |
| 和菓子 | wa-gashi | kashi → gashi12 |
| 福袋 | fuku-bukuro | fukuro → bukuro13 |
| 紙袋 | kami-bukuro | fukuro → bukuro1 |
| 単行本 | tankō-bon | hon → bon12 |
| 青写真 | ao-jashin | shashin → jashin12 |
| 無事 | bu-ji | (lexicalised voicing inside a kango binom)12 |
| 湯豆腐 | yu-dōfu | tōfu → dōfu12 |
The common thread is high-frequency, everyday use. The second element has been "naturalised" into wago-stratum behaviour through centuries of daily collocation.49
福袋 is a useful diagnostic case. 福 (fuku) is kango, but 袋 (fukuro) is wago. This makes 福袋 technically a hybrid (kango + wago) compound, so rendaku on the wago second element is expected rather than exceptional.413 株式会社 is the showcase case of genuine kango plus kango rendaku: both 株式 and 会社 are kango, yet daily-business frequency has driven 会社 into wago-stratum behaviour in this collocation.112
父は株式会社の社長です。12
"My father is the president of a corporation."
Mixed compounds: jūbako and yutō
Mixed-reading compounds are named for two standard examples: 重箱 jū-bako (on + kun, the jūbako pattern) and 湯桶 yu-tō (kun + on, the yutō pattern).12
Rendaku behaviour in mixed compounds follows the stratum of the second element. A wago second element voices as in a wago compound: 重箱 jū-bako (箱 hako is wago, voices to bako).14 A kango second element behaves like kango: 湯桶 yu-tō, where 桶 tō is the on'yomi of 桶, takes no rendaku.1 Contrast 湯豆腐 yu-dōfu, where the second element 豆腐 (tōfu, fully naturalised kango) voices to dōfu in line with the same lexicalised-exception pattern as 株式会社.112
Lyman's Law in kanji compounds
The rule
Rendaku is blocked when the second element of the compound already contains a voiced obstruent (/ɡ/, /z/, /d/, /b/) anywhere in its sound shape.114 The constraint is named for Benjamin Smith Lyman, who formulated the modern statement in 1894. The same generalisation had already been observed by the Edo-period kokugakusha (national-learning scholars) 賀茂真淵 (Kamo no Mabuchi, 1765) and 本居宣長 (Motoori Norinaga, late 1700s).14 The historical and phonological backstory, including the OCP-voice account and the prenasalisation residue, is in the parent rendaku article.215
Kanji-compound examples that block
The standard blockers are easy to verify: read the second element aloud and listen for an internal voiced obstruent.
- 神風 kami + kaze → kami-kaze, not *kami-gaze; blocked because kaze contains /z/.114
- 山風 yama + kaze → yama-kaze "mountain wind"; same /z/.14
- 春風 haru + kaze → haru-kaze "spring breeze"; same blocker.1412
- 紙屑 kami + kuzu → kami-kuzu "wastepaper"; blocked because kuzu contains /z/.12
- 一人旅 hitori + tabi → hitori-tabi "solo travel"; blocked because tabi contains /b/.114
- 山火事 yama + kaji → yama-kaji "mountain fire"; blocked because kaji contains a voiced obstruent at /dʑ/.14
- 山門 yama + kado → yama-kado; blocked because kado contains /d/.114
神風は歴史の言葉です。1
"Kamikaze is a word from history."
Kanji-compound examples that fire
The same shape, with no internal voiced obstruent in the second element, yields rendaku.
- 折り紙 ori + kami → ori-gami.5
- 手紙 te + kami → te-gami.7
- 石橋 ishi + hashi → ishi-bashi.7
- 紙袋 kami + fukuro → kami-bukuro; the /k/ in -kuro is voiceless, so it is not a Lyman's-Law blocker.113
- 紙箱 kami + hako → kami-bako "paper box".7
- 紙皿 kami + sara → kami-zara "paper plate".7
The kami-X series shows Lyman's Law as a clean asymmetric filter. As long as the second element has no internal /ɡ z d b/, k→g (or s→z, h→b) is free to fire in the kami compound the learner encounters.1
Exceptions and dialectal noise
A compound that seems to flout Lyman's Law usually fits one of three distinct mechanisms. Each is its own story, not a softening of the law.
The first is the lexicalised-exception class typified by 縄梯子 nawa-bashigo "rope ladder", where rendaku fires despite hashigo containing /ɡ/. The NINJAL Rendaku Database catalogues a small handful of such forms. They are entries in the lexicon, not derivations from rule.149
The second is jukujikun, where the reading is assigned to the kanji compound as a whole rather than derived morpheme by morpheme. 大蛇 has the regular kango reading ō-ja (on + on, no rendaku) and the irregular gloss reading orochi, a native morpheme matched to the whole compound. The orochi reading is not a rendaku product at all, so Lyman's Law has no foothold.7
The third is parallel-meaning dvandva structure. The 山川 yamakawa / yamagawa pair is rendaku-blocked when read yamakawa because the compound is parallel-meaning ("mountains and rivers"). It is not blocked because either element contains a voiced obstruent. Reading the same kanji as yamagawa ("mountain river") makes it head-modifier, and rendaku fires.111
Predicting rendaku in a compound you have not seen
A three-question check
Run the decision flow in order. The first "no" stops the chain.
Stated in prose, the three questions are these.
- Is the second element kun'yomi (wago)? If on'yomi (kango), expect no rendaku unless the word is on the lexicalised-exceptions list above (株式会社, 黒砂糖, 和菓子, 福袋, 単行本, 青写真, 無事, 湯豆腐).1412
- Does the second element already contain /ɡ z d b/? If yes, Lyman's Law blocks.14
- Is the compound a parallel-meaning dvandva (山川 "mountains and rivers", 父母 chichi-haha "father and mother", 好き嫌い suki-kirai "likes and dislikes")? If yes, no rendaku.112
If all three checks pass, expect rendaku and confirm it in a dictionary. The flow predicts about 90% of wago compounds correctly and converges on "no rendaku" for the vast majority of kango compounds.49
Why this is a guess, not a rule
Even with all three blockers cleared, wago compounds still fail to voice in roughly one case in ten. The 90% rate is empirical, not categorical.4910 Surnames and place names are the most treacherous environment. 中田 nakata versus nakada and 中島 nakajima versus nakashima split family by family, without any rule the learner can derive.17
Use rendaku as a default to predict the reading of an unfamiliar compound, then verify it in a dictionary. Do not treat it as a rule for deriving every reading from scratch.4
Where this fits in the larger reading-prediction flow
The full reading-prediction flow for a new kanji compound has four steps: recognise on+on versus kun+kun versus mixed (jūbako or yutō); apply rendaku based on the stratum check above; check for phonetic-component clues on unfamiliar on'yomi; fall back to jukujikun lookup for irregular gloss readings. Rendaku is step two of the four.
Good to know
The 々 iteration mark almost always voices
人々 hito-bito, 時々 toki-doki, 国々 kuni-guni, 神々 kami-gami, and 木々 ki-gi all follow the same default.18 Every reduplicated compound is wago, and the morpheme boundary is unambiguous. Lyman's Law has only one chance to fire, and the dvandva blocker does not apply because a thing repeated is not "X and Y" in the relevant sense.1
山々 yamayama is the famous exception, and the reason is mechanical. Rendaku targets only the four voiceless obstruent rows /k s t h/. yama starts with /j/ (y-row glide), which has no voiced-obstruent counterpart in the kana inventory.12 The same explanation covers any reduplication of an m, n, y, w, or r-initial morpheme.12
Dakuten you will and will not see in writing
Rendaku is invisible in pure-kanji writing (折り紙 prints with the same 紙 that reads kami in isolation) and visible in kana writing (おりがみ, with が).65 Furigana follow the spoken reading, not the kanji-isolated reading. So a dictionary's furigana over 折り紙 reads おりがみ, with が above 紙, even though 紙 alone reads かみ.65 This asymmetry is what makes beginners stumble when they read aloud straight from a dictionary entry.
Why 株式会社 voices but 図書館 does not
Both are kango. The difference is register and frequency, not etymology. 株式会社 has been so domesticated by daily business use that 会社 behaves as a quasi-wago morpheme in this collocation. The same dynamic applies to 黒砂糖, 和菓子, 単行本, and 青写真.1412 図書館 belongs to an institutional and academic register that preserves Sino-Japanese phonology. Its constituents 図書 and 館 do not collocate broadly with wago in everyday use, so neither half has been naturalised.412 Rendaku in kango is conditioned on frequency and lexical stratum drift, not on a phonological rule the learner can run from outside.49
Kun+kun voices, on+on does not, and the reason is stratum
The on+on default is the "Chinese loanword" stratum. Loanwords historically resist phonological assimilation across many languages.14 The kun+kun default is the "native words" stratum, and native words follow native rules. Rendaku is one of those native rules.12 Once the learner sees rendaku as a wago-stratum process that kango opts out of, the asymmetry stops feeling arbitrary.1
鼻血 is hana-ji, not hana-chi: the 1986 cabinet-notification precision
A learner who reads 鼻血 as *hana-chi has missed the rendaku. The correct reading is hana-ji.67 The corrected form belongs in a two-line ruby block.
鼻血が出た。6
"I got a nosebleed."
The kana ち (chi, /tɕi/) voices to its dakuon counterpart, which is historically ぢ and in most modern contexts じ. The 1986 Cabinet notification on modern kana usage (内閣告示第一号, 現代仮名遣い 本文 第2) preserves the ぢ and づ spellings in two narrow situations: same-morpheme reduplication of the chi/tsu series (ちぢむ "to shrink", つづく "to continue") and two-word compounding where the rendaku boundary is transparent (はなぢ "nosebleed", みかづき "crescent moon", そこぢから "innate strength").6
This is a spelling exception, not a sound exception. It is the only correct way to write はなぢ when the rendaku origin matters.6 Standard publishing accepts はなじ as well.67
Etymology of the word "rendaku" itself
連 ren means "consecutive" and 濁 daku means "muddy" or "voiced", so 連濁 reads as "consecutive voicing".17 The same 濁 is the kanji in 濁点 (dakuten, "voicing mark") and 濁音 (dakuon, "voiced sound"). Japanese metalinguistic vocabulary treats voicing as a quality of "muddiness" rather than "loudness".17 The term 連濁 names what the writing system already names with dakuten.17
See also
- Go-on, Kan-on, Tō-on: The Historical Layers Behind a Kanji's Multiple On'yomi
- Ateji (当て字): Kanji Chosen for Sound, Not Meaning
- Reading 生: The Kanji With Over 150 Attested Readings
- Jukugo (熟語): How Kanji Combine to Form Japanese Words
- Japanese Compound-Word Pitch Accent: How Two Words Combine into One Accent Pattern