Rendaku: When K Becomes G in Compound Words

Rendaku is Japanese sequential voicing. In compounds, it turns the first voiceless obstruent (a stop or fricative such as /k/, /s/, /t/, or /h/) of the second element into its voiced partner.^{1 2} It is why 紙 kami becomes gami inside 折り紙 origami, why 手紙 tegami writes its second half with が rather than か, and why 人々 reads hitobito instead of hitohito.

Overview

What rendaku is

Rendaku (連濁) is the regular voicing of a compound's right-hand initial consonant from /k s t h/ to /g z d b~p/.¹ In writing, the shift appears as the dakuten (゛) on the か, さ, た rows and as the dakuten or handakuten (゜) on the は row. In speech, it is a one-step voicing of the obstruent.¹

The term comes from Edo-period kokugaku scholarship. Kamo no Mabuchi (1765) and Motoori Norinaga (1767–1798) independently identified the voicing-blocking constraint now called Lyman's Law.³ Benjamin Smith Lyman's 1894 paper The Change from Surd to Sonant in Japanese Compounds stated the rule in English and gave the constraint its Western name.⁴

The literature usually describes the structure this way: in an [A + B] compound, the initial consonant of B voices under lexical, phonotactic, and structural conditions.^{1 2}

Voiced obstruent means /g z d b/

Lyman's Law and the rendaku rules below all depend on the term "voiced obstruent." This class covers exactly /g/, /z/, /d/, and /b/, the dakuon series. The sonorants /m n r w y/ are not obstruents, so they do not block rendaku.^{4 5}

Where you meet it as a learner

Rendaku is not a JLPT grammar point. It is a phonological process whose results appear in the standard N5 through N1 vocabulary lists.¹ A learner first meets it in a small set of high-frequency words whose second half has unexpected dakuten: 手紙 tegami "letter," 折り紙 origami "paper-folding," 人々 hitobito "people," 鼻血 hanaji "nosebleed," 黒砂糖 kurozatō "brown sugar," 花火 hanabi "fireworks."^{6 7 8 9 10}

Counter compounds are the next major place to notice rendaku. The forms 一本 ippon, 三本 sanbon, 六本 roppon sit at the intersection of rendaku and gemination. The underlying counter is 本 hon, and the surface shape depends on the preceding mora.¹

Surname and place-name components use the same process: -gawa (川), -jima (島), -bashi (橋), -saki/-zaki (崎). These are lexicalized choices, discussed under "Semantic factors" below.²

The four voicing shifts

Rendaku targets the four voiceless obstruent series at the start of the second element. The pairings are fixed by the writing system: each voiceless onset has exactly one voiced reflex.

Voiceless onset	Voiced reflex	Kana shift
/k/	/g/	か→が, き→ぎ, く→ぐ, け→げ, こ→ご
/s/, /ɕ/	/z/, /(d)ʑ/	さ→ざ, し→じ, す→ず, せ→ぜ, そ→ぞ
/t/, /tɕ/, /ts/	/d/, /(d)ʑ/, /dz/	た→だ, ち→ぢ, つ→づ, て→で, と→ど
/h/, /ç/, /ɸ/	/b/ (default) or /p/ (handakuten)	は→ばぱ, ひ→びぴ, ふ→ぶぷ, へ→べぺ, ほ→ぼ~ぽ

k goes to g

The /k/ initial of the second element voices to /g/.¹ This is the clearest of the four shifts because the underlying /k/ has no allophonic complications, meaning it does not change pronunciation by environment before rendaku. Wago compounds with a second-element /k/ show rendaku at the wago baseline rate, around 87% in Rosen 2001 and 84% in Irwin 2005.^{11 12}

手紙てがみ⁸
"letter (the kind you write and post)"

折おり紙がみ⁶
"origami; paper-folding"

縁えん組ぐみ¹³
"matrimonial alliance; adoption arrangement"

s goes to z

The /s/ initial of the second element voices to /z/.¹ The /sh/ ~ /ɕ/ series (し, しゃ, しゅ, しょ) voices to /j/ ~ /(d)ʑ/, written じ, じゃ, じゅ, じょ. Phonetically, this is [dʑ] in onset position.^{1 2}

黒砂糖くろざとう⁷
"unrefined brown sugar"

青空あおぞら²
"blue sky"

夜桜よざくら¹³
"cherry blossoms viewed at night"

t goes to d or dʒ

The /t/ initial of the second element voices to /d/.¹ Before /i/ and /u/, the underlying /t/ surfaces as the allophones [tɕ] and [ts]. The voiced reflexes are ぢ /dʑi/ and づ /dzu/ in historical orthography.³

Modern 現代仮名遣い (the 1946 spelling reform) writes both ぢ and づ as じ and ず in almost every context. The exceptions are rendaku compounds and etymologically transparent reduplications like ちぢむ, where ぢ and づ are retained as visible markers of the rendaku origin.³ The textbook example is 鼻血 hanaji "nosebleed."⁹

鼻血はなぢ⁹
"nosebleed"

三日月みかづき¹³
"crescent moon"

火種ひだね¹
"spark; kindling; the seed of a fire (literal or figurative)"

h goes to b (and sometimes p)

The /h/ initial of the second element voices to /b/ by default, or to /p/ in the handakuten reflex.¹ The /h/-to-/b/ correspondence is regular wherever the structural rules allow rendaku, even though /b/ is phonetically farther from /h/ than /g/ is from /k/.

The reason is historical. The h-row originated as Proto-Japanese _/p/. It shifted to the bilabial fricative [ɸ] before the end of the 16th century, then split into [h] before /a e o/, [ç] before /i/, and remained [ɸ] before /u/.³ The voiced counterpart of _/p/ has been /b/ throughout, and rendaku preserves that ancient pairing.

The /h/-to-/p/ (handakuten) reflex appears predominantly when the first element ends in the moraic obstruent っ or the moraic nasal ん.¹ The counter forms 一本 ippon, 三本 sanbon, 六本 roppon are the highest-frequency examples; Sino-Japanese 出発 shuppatsu shows the same conditioning inside a kango compound.¹

人々ひとびと¹⁰
"people; various people"

花火はなび¹
"fireworks"

川舟かわぶね¹
"river boat"

三本さんぼん¹
"three (long thin objects)"

一本いっぽん¹
"one (long thin object)"

When rendaku fires and when it does not

The conditioning rules are presented below in order of reliability. Position is absolute. Lyman's Law is hard. The strata generalizations are statistical. Branching and dvandva are structural, with attested exceptions.

Position: only the second element

Rendaku targets the initial consonant of the right-hand element of a binary compound. The first element keeps its lexical voicing.^{1 2}

In three-element compounds, only the rightmost morpheme of a right-branching sub-constituent can voice. That constraint is the subject of "Branching" below.¹⁴

Lyman's Law: no double voicing

Vance's modern formulation is simple: rendaku is blocked when the second element of a compound already contains a voiced obstruent phoneme (/g/, /z/, /d/, or /b/).^{1 2} Lyman's 1894 original wording named the same constraint: "the general rule does not apply ... when b, d, g, j, p, or z already occurs anywhere in the second part of the compound."⁴

Modern theoretical work restates Lyman's Law as the surface effect of a morpheme-internal OCP (Obligatory Contour Principle) constraint. In plain terms, that constraint bans two voiced obstruents inside a single morpheme.⁵

The positive control: 折り紙 origami above. Kami contains no voiced obstruent, so rendaku fires.⁶

The canonical negative example:

山風やまかぜ¹
"mountain wind (the wind blowing down off a mountain)"

Lyman blocks because kaze already contains the voiced obstruent /z/. The form is yamakaze, not yamagaze.^{1 4}

A second blocked compound:

山火事やまかじ¹
"wildfire; mountain fire"

Lyman blocks because kaji contains the voiced [dʑ] (phonemically /z/). The form is yamakaji, not yamagaji.¹

Lyman has lexicalized exceptions

The hard constraint has a small set of attested counterexamples in wago compounds. The most cited is 縄梯子 nawa-bashigo "rope ladder," from hashigo, which already contains /g/. Rendaku still fires.¹ Vance treats these as lexicalized exceptions rather than evidence against the constraint.²

Lexical strata: wago vs kango vs gairaigo

Japanese vocabulary divides into three strata, or historical layers: wago (和語, native Japanese), kango (漢語, Sino-Japanese loans from Middle Chinese), and gairaigo (外来語, later loans, predominantly from European languages and now written in katakana).¹⁵

Rendaku rates fall sharply by stratum. The figures below come from Rosen's 2001 University of British Columbia dissertation, Irwin's 2005 Journal of East Asian Linguistics article, and Irwin's 2011 Loanwords in Japanese monograph.

Stratum	Rendaku rate as second element	Source
Wago	~87% of items show rendaku in at least one compound	Rosen 200111
Wago	~84% (alternative count, restricted to compound-initial second-element position)	Irwin 200512
Kango (mononoms)	~20%	Irwin 200512
Kango (binoms)	~10%	Irwin 200512
Gairaigo	near zero	Irwin 201115; Vance 20152

Vance explains the stratum split this way: rendaku is a wago-internal historical process. Kango compounds entered Japanese after that process had largely stopped applying productively, and they were borrowed with fixed on'yomi readings that resist further alternation. Gairaigo arrived later still and behave as orthographically frozen units in katakana.^{1 2}

For learners, a practical stratum check is this: a morpheme written in katakana is gairaigo by default; a morpheme whose kanji are read with on'yomi inside a compound is kango by default; a morpheme whose kanji are read with kun'yomi (the native gloss) is wago by default.¹⁵ The heuristic breaks for the small set of hybrid compounds (重箱読み and 湯桶読み), so it is a guide rather than a rule.

Branching in three-element compounds

Otsu's Right-Branch Condition (1980) states that rendaku applies only when a potential rendaku segment sits in a right-branching constituent, that is, a grouped unit on the right side of the compound.¹⁴ In a three-element compound A+B+C, the structure determines whether the middle element B can voice.

• Left-branching [[A + B] + C]: rendaku fires on C, because C is in the right branch of the top node.
• Right-branching [A + [B + C]]: rendaku is blocked on B; rendaku may still fire on C inside the [B + C] sub-constituent.

The canonical contrast comes from Otsu 1980 and is revisited in every later treatment.^{14 16}

Status of the constraint: psycholinguistic experiments (Kumagai 2014) show only weak speaker sensitivity to the Right-Branch Condition, with average d′ around 0.14. Vance 2015 also questions whether the "loose" versus "strict" compound distinction can be applied reliably.^{16 2} The constraint remains the standard description in the literature. Treat it as a strong tendency in three-element compounds rather than a hard law.

Semantic factors: dvandva and proper names

Dvandva compounds are coordinate compounds, meaning "X and Y." They tend to block rendaku.^{1 2} The classic minimal pair uses a single written form:

山川やまかわ¹⁷
"mountains and rivers" (dvandva: no rendaku)

山川やまがわ¹⁷
"mountain river; a river flowing through a mountain" (modificational: rendaku fires)

The kanji and morphemes are the same, but the compounds differ in meaning. Rendaku marks that difference by being present or absent.

Proper-noun compounds (surnames and place names) are lexicalized.² The same morpheme can take either form in different names: 中島 reads as both Nakajima and Nakashima depending on the family; 大川 reads as both Ōgawa and Ōkawa.² For learners, the rule is simple: look up the specific name, do not predict.

Working through a new compound

A learner who knows the four shifts and Lyman's Law still has to decide whether the second element voices in an unfamiliar compound. The check below turns the conditioning rules into a fixed order of questions.

A four-question check

The questions, in order:

1. Stratum check. Is the second element wago, kango, or gairaigo? If kango, rendaku is unlikely (~10–20%). If gairaigo, rendaku is near zero. If wago, continue.^{12 15}
2. Lyman check. Does the second element already contain a voiced obstruent (/g z d b/, including the [dʑ] allophone written じ or ぢ)? If yes, rendaku is blocked.^{1 4}
3. Branching check. In a three-element compound, is the structure right-branching? If yes, rendaku is blocked on the middle element.¹⁴
4. Semantics check. Is the compound a dvandva (coordinate, "X and Y")? If yes, rendaku is blocked.^{1 2}

If all four checks point to "fire," expect rendaku in a wago compound at roughly the wago baseline rate (≈85% per Rosen 2001 and Irwin 2005). Treat the result as a strong prediction, not a guarantee. A dictionary entry overrides the prediction whenever the two disagree.^{11 12}

Worked examples

The three examples below run the four-question check on a predictable wago hit, a Lyman block, and a kango block.

Predictable wago hit.

折おり紙がみ⁶
"origami; the art of paper-folding"

Stratum: wago + wago. Lyman: kami contains no voiced obstruent. Branching: binary compound, so the three-element question does not arise. Semantics: modificational (the paper is the object being folded), not coordinate. All four checks pass. Rendaku fires: kami → gami.

Lyman block.

山風やまかぜ¹
"mountain wind; the wind that blows down from a mountain"

Stratum: wago + wago. Lyman: kaze contains /z/; blocked. Stop. The form is yamakaze, not yamagaze. This is the canonical Lyman's Law example used in every Japanese phonology paper since Lyman 1894.^{4 1}

Kango (strata) block.

説明せつめい¹
"explanation"

Stratum: kango + kango. Kango compounds resist rendaku (~10% rate per Irwin 2005). The on'yomi of 明 is mei, not bei. Rendaku is blocked at the strata check, so there is no need to run the Lyman check.¹² The form is setsumei.

Good to know

The dakuten you see in the spelling is the rendaku you hear

Rendaku writes itself with the same dakuten (゛) and handakuten (゜) marks that learners already know from the basic kana chart. The spelling side of the story belongs to the J-Compass article on dakuten and handakuten as voicing marks on hiragana. There is no separate "rendaku diacritic."¹ The regular voicing marks do the work.

Why h goes to b, not v

The h-row of modern Japanese descends from Proto-Japanese _/p/. Frellesvig 2010 dates the shift: _/p/ became the bilabial fricative [ɸ] before the end of the 16th century, then split during the 17th–18th centuries into [h] before /a e o/, [ç] before /i/, and remained [ɸ] before /u/.³

The voiced counterpart of */p/ has always been /b/, so rendaku in the h-row preserves the original /p/–/b/ pairing. The modern alternation hito → bito is the surface form of the ancient *pitö → *bitö voicing.³ The same explanation accounts for the handakuten /p/ reflex after moraic っ or ん in compounds such as ippon and sanbon.¹

Mountain wind is the mnemonic for Lyman's Law

The compound 山風 yamakaze is the canonical example used in every introduction to rendaku from Lyman 1894 onward.^{4 1} Memorizing the pair (yama + kaze, blocked because kaze already has a /z/) gives you the rule.

Rendaku is a strong tendency, not a law

Even wago compounds show roughly 13–16% non-rendaku in second-element position (the complement of the 84–87% wago rate cited above), and Lyman has attested exceptions like nawa-bashigo 縄梯子.^{11 12 1} The four-question check produces a strong prediction. A dictionary entry overrides it whenever the two disagree.^{13 18}

How rendaku probably arose

The dominant historical hypothesis is the genitive-particle fusion account. In Old Japanese, the genitive postposition の no could appear between the two elements of an N+N compound. For example, yama-no-tori "mountain-GEN-bird" gave modern yamadori 山鳥 "copper pheasant," via an intermediate stage where no reduced to a prenasalized voiced segment that fused with the following obstruent.^{3 19}

The reconstruction sits in the Old Japanese period (roughly the 7th–8th centuries CE) and earlier. Frellesvig 2010 is the standard reference for the timing.³ The hypothesis explains why rendaku is partial rather than universal: only the subset of compounds that historically passed through the no-fusion path show the voicing reflex.

A second hypothesis treats rendaku as a juncture marker that originally signalled the boundary between elements of a compound. Both accounts coexist in the literature. Frellesvig 2010 reviews them.³

See also

• Rendaku in Kanji Compounds: Why 紙 Becomes -gami
• Japanese Compound-Word Pitch Accent: How Two Words Combine into One Accent Pattern
• How to Predict the Reading of an Unknown Kanji Compound: The On+On Default, Jūbako, Yutō, and the Look-It-Up Bucket
• The Japanese Vowel Inventory: Five Vowels, Done Right
• Japanese Vowel Devoicing: Why です Sounds Like "Des"
• Stress vs. Pitch: Does Japanese Have Stress?

References

Footnotes

1. Vance, Timothy J. The Sounds of Japanese. Cambridge University Press, 2008. Chapter 8 "Sequential voicing (rendaku)", pp. 166–183. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷ ↩¹⁸ ↩¹⁹ ↩²⁰ ↩²¹ ↩²² ↩²³ ↩²⁴ ↩²⁵ ↩²⁶ ↩²⁷ ↩²⁸ ↩²⁹ ↩³⁰ ↩³¹ ↩³² ↩³³ ↩³⁴ ↩³⁵ ↩³⁶

2. Vance, Timothy J. "Rendaku." In Haruo Kubozono (ed.), Handbook of Japanese Phonetics and Phonology. De Gruyter Mouton, 2015, pp. 397–441. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵

3. Frellesvig, Bjarke. A History of the Japanese Language. Cambridge University Press, 2010. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

4. Lyman, Benjamin Smith. The Change from Surd to Sonant in Japanese Compounds. Oriental Club of Philadelphia, 1894. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

5. Kawahara, Shigeto. "The phonology of Japanese accent." In Haruo Kubozono (ed.), Handbook of Japanese Phonetics and Phonology. De Gruyter Mouton, 2015. (Cited for the modern restatement of Lyman's Law as a morpheme-internal OCP constraint on voiced obstruents.) ↩ ↩²

6. Wiktionary (English). "折り紙." https://en.wiktionary.org/wiki/%E6%8A%98%E3%82%8A%E7%B4%99 (etymology entry, cross-referenced to 広辞苑). ↩ ↩² ↩³ ↩⁴

7. Wiktionary (English). "黒砂糖." https://en.wiktionary.org/wiki/%E9%BB%92%E7%A0%82%E7%B3%96 (etymology entry, cross-referenced to 大辞林). ↩ ↩²

8. Wiktionary (English). "手紙." https://en.wiktionary.org/wiki/%E6%89%8B%E7%B4%99 ↩ ↩²

9. Wiktionary (English). "鼻血." https://en.wiktionary.org/wiki/%E9%BC%BB%E8%A1%80 ↩ ↩² ↩³

10. Wiktionary (English). "人々." https://en.wiktionary.org/wiki/%E4%BA%BA%E3%80%85 ↩ ↩²

11. Rosen, Eric Robert. Phonological Processes Interacting with the Lexicon: Variable and Non-Regular Effects in Japanese Phonology. PhD dissertation, University of British Columbia, 2001. ↩ ↩² ↩³ ↩⁴

12. Irwin, Mark. "Rendaku-based lexical hierarchies in Japanese: The behaviour of Sino-Japanese mononoms in hybrid noun compounds." Journal of East Asian Linguistics 14 (2005), pp. 121–153. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

13. 新村 出 (編). 『広辞苑』第七版. 岩波書店, 2018. (Standard reference for compound headword entries and reading attestations.) ↩ ↩² ↩³ ↩⁴

14. Otsu, Yukio. "Some aspects of rendaku in Japanese and related problems." In Yukio Otsu and Ann Farmer (eds.), MIT Working Papers in Linguistics 2: Theoretical Issues in Japanese Linguistics. MIT, 1980, pp. 207–227. ↩ ↩² ↩³ ↩⁴

15. Irwin, Mark. Loanwords in Japanese. John Benjamins, 2011. ↩ ↩² ↩³ ↩⁴

16. Kumagai, Gakuji. "The psychological status of the right-branch condition on rendaku." Studies in Language Sciences 13 (2014), pp. 124–145. ↩ ↩²

17. Wiktionary (English). "山川." https://en.wiktionary.org/wiki/%E5%B1%B1%E5%B7%9D (entry distinguishes yamakawa "mountains and rivers" from yamagawa "mountain river"). ↩ ↩²

18. NHK放送文化研究所 (編). 『NHK日本語発音アクセント新辞典』. NHK出版, 2016. (Standard reference for the standard-Tokyo reading of compound headwords including rendaku forms.) ↩

19. Kindaichi, Haruhiko. The Japanese Language (transl. Umeyo Hirano). Tuttle, 1978. (Cited for the genitive-particle fusion hypothesis as one of two main reconstructions; the other is the "voicing-as-juncture-marker" hypothesis.) ↩