Say 手 (て, te, "hand") and 紙 (かみ, kami, "paper"). Now stick them together to make the word for "letter." You might expect te-kami — but it is てがみ (tegami). The か has become が. This is rendaku (連濁, "sequential voicing"): when a word is glued on as the second element of a compound, its initial voiceless consonant frequently turns voiced. It is one of the most pervasive sound patterns in Japanese, and it explains a whole category of "wait, why did that change?" moments — origami, hanabi, the -boshi in place names.
Rendaku is only partly predictable, so ultimately you learn compounds one at a time. But there is real structure here: a set of regular sound changes, and one dependable law that tells you when voicing is blocked. Learn those and the chaos shrinks to something manageable.
What voices into what
Rendaku swaps a voiceless kana at the start of the second element for its voiced (dakuten) counterpart. The pairs:
| Voiceless | → Voiced | Example |
|---|---|---|
| か・き・く・け・こ (k) | が・ぎ・ぐ・げ・ご (g) | 手 + 紙 → 手紙 てがみ |
| さ・し・す・せ・そ (s) | ざ・じ・ず・ぜ・ぞ (z) | 青 + 空 → 青空 あおぞら |
| た・ち・つ・て・と (t) | だ・ぢ・づ・で・ど (d) | 本 + 棚 → 本棚 ほんだな |
| は・ひ・ふ・へ・ほ (h) | ば・び・ぶ・べ・ぼ (b) | 花 + 火 → 花火 はなび |
The one that surprises learners is the last row: は-row voices to ば (b), not to anything with an "h" sound. That is a fossil of history — this row was pronounced with /p/–/b/ long ago, so its voiced partner is /b/. The voicing is always written with the two-stroke dakuten mark (か → が); see Dakuten and Handakuten for how it looks on the page.
祖母から久しぶりに手紙が届いた。
sobo kara hisashiburi ni tegami ga todoita
A letter arrived from my grandmother for the first time in a while.
夏になったら、みんなで花火を見に行こう。
natsu ni nattara, minna de hanabi o mi ni ikō
When summer comes, let's all go watch the fireworks together.
子供が折り紙できれいな鶴を折った。
kodomo ga origami de kirei na tsuru o otta
The kid folded a beautiful crane out of origami paper.
Notice 折り紙 (おりがみ): 折り (おり) + 紙 (かみ) → origami, the same 紙 → がみ voicing as in 手紙. Once you see 紙 rendaku-ing to がみ in one word, you recognize it in the next.
Repetition and more examples
Rendaku is also what voices the second half of doubled words. 人 (ひと, hito, "person") doubled for "people" is 人々 (ひとびと, hitobito) — the second ひと becomes びと.
お祭りには町の人々が集まってきた。
o-matsuri ni wa machi no hitobito ga atsumatte kita
The townspeople gathered for the festival.
おばあちゃんが寝る前に昔話をしてくれた。
obāchan ga neru mae ni mukashibanashi o shite kureta
Grandma told me an old folk tale before bed.
今夜は空に三日月がきれいに出ている。
kon'ya wa sora ni mikazuki ga kirei ni dete iru
Tonight a lovely crescent moon is out in the sky.
昔話 (むかしばなし) voices 話 (はなし) → ばなし; 三日月 (みかづき) voices 月 (つき) → づき. These are ordinary, everyday words — rendaku is not a rare textbook curiosity, it is baked into the core vocabulary.
Lyman's Law: the reliable brake
If rendaku were automatic, 山 (やま) + 風 (かぜ, "wind") would give yamagaze. It doesn't — it stays 山風 (やまかぜ), no voicing. Compare 山 + 川 (かわ) → 山川 (やまがわ), which does voice. Same 山, same structure — so what blocks the wind one?
The answer is Lyman's Law: rendaku is suppressed when the second element already contains a voiced obstruent (a b, d, g, or z sound) somewhere inside it. 風 (かぜ) already has ぜ (a /z/), so it refuses a second voiced obstruent and か stays か. 川 (かわ) has no such sound, so it voices freely.
| Compound | Second element | Rendaku? | Why |
|---|---|---|---|
| 山川 やまがわ | 川 かわ | ✅ voices (か→が) | no voiced obstruent inside 川 |
| 山風 やまかぜ | 風 かぜ | ❌ blocked | かぜ already has ぜ (/z/) |
| 竹とんぼ たけとんぼ | とんぼ | ❌ blocked | とんぼ has ぼ (/b/) |
| 大とかげ おおとかげ | とかげ | ❌ blocked | とかげ has が (/g/) |
This is a genuinely old and robust observation — the Edo-period scholar Motoori Norinaga noted it, and it is named in English after Benjamin Smith Lyman. One subtlety keeps learners accurate: the blocker must be a voiced obstruent (b, d, g, z). Nasal sounds (m, n) and the moraic ん do not block. In 竹とんぼ it is the ぼ that blocks, not the ん — a word whose only extra sound was a nasal could still rendaku.
春風が気持ちよくて、つい昼寝してしまった。
harukaze ga kimochi yokute, tsui hirune shite shimatta
The spring breeze felt so nice I ended up dozing off. (春 + 風 → harukaze, not *harugaze — Lyman's Law)
昔はよく竹とんぼを作って飛ばして遊んだ。
mukashi wa yoku taketombo o tsukutte tobashite asonda
We used to make bamboo-copters and fly them a lot back then. (blocked by the ぼ in とんぼ)
Why it stays only "partly predictable"
Lyman's Law tells you when rendaku can't happen, but not reliably when it will. Several other factors tilt the odds, and none is absolute:
- Word origin. Native Japanese words (和語, wago) rendaku readily. Sino-Japanese on-yomi compounds (漢語) and foreign loanwords (外来語) usually don't. You will not hear a coffee compound voice アイスコーヒー into aisugōhī; loanwords sit outside the pattern.
- "A and B" compounds resist it. When the two halves are coordinate — A and B, not B of A — rendaku tends to stay away. 山川 read やまかわ means "mountains and rivers" (coordinate) and does not voice, whereas やまがわ, "a river in the mountains" (B modified by A), does. The same two kanji, two readings, split by whether the parts are partners or one modifies the other. 白黒 (しろくろ, "black and white") likewise stays voiceless.
- Plain lexical stubbornness. Some words simply refuse rendaku, or allow both forms, for no rule you can derive. These you memorize.
Because of all this, the honest bottom line is: you acquire rendaku word by word. But that is far from hopeless — the sound changes are regular, Lyman's Law is dependable, and the origin and coordinate tendencies catch most of the rest. Knowing the tendency turns a hundred unexplained "sound changes" into one understood process with a handful of brakes.
How this compares to English
English has no rendaku, but it is not as alien as it first looks. English also voices a consonant in certain derived forms, lexically and unpredictably: leaf → leaves, knife → knives, wolf → wolves — yet chief → chiefs and roof → roofs stubbornly don't. That "the base consonant voices in the derived word, but only for some words, and you must learn which" instinct is exactly the mindset rendaku asks of you. So you already tolerate this kind of thing in your own language; Japanese just does it at the front of compound second-elements instead of at the ends of plurals.
Common Mistakes
祖母に手紙を書いた。
sobo ni tegami o kaita
❌ if pronounced 'te-kami.' Never-voicing is the #1 error — the compound is てがみ (tegami). English speakers under-apply rendaku.
山風が冷たい。
yamakaze ga tsumetai
❌ if 'voiced up' to *yamagaze. Blocked by Lyman's Law — 風 (かぜ) already has ぜ. Over-generalizing rendaku is the mirror-image error.
この空の色、本当にきれいな青空だね。
kono sora no iro, hontō ni kirei na aozora da ne
✅ 青 + 空 → あおぞら voices correctly (そら→ぞら); 空 has no voiced obstruent, so nothing blocks it.
白黒の写真が好きだ。
shirokuro no shashin ga suki da
❌ if voiced to *shiroguro. It's an 'A and B' (black-and-white) coordinate compound — those resist rendaku. Stays しろくろ.
昔話をもう一度聞かせて。
mukashibanashi o mō ichido kikasete
✅ 話 (はなし) → ばなし voices, because はなし has no b/d/g/z to trigger Lyman's Law.
Key Takeaways
- Rendaku voices the initial consonant of a compound's second element: k→g, s→z, t→d, h→b (手紙 てがみ, 青空 あおぞら, 本棚 ほんだな, 花火 はなび).
- The は-row voices to ば (a /b/ sound), a leftover from its old /p/ pronunciation; all voicing is written with dakuten.
- Lyman's Law blocks rendaku when the second element already holds a voiced obstruent (b/d/g/z): 山川 やまがわ voices, but 山風 やまかぜ, 竹とんぼ, 大とかげ don't. Nasals and ん don't block.
- Native words voice readily; on-compounds and loanwords usually don't, and coordinate "A and B" compounds resist (山川 = やまかわ, not がわ).
- Rendaku is acquired word by word, but the regular sound changes plus Lyman's Law explain most of it — like English leaf → leaves, it's lexical but learnable.
Now practice Japanese
Reading grammar gets you part of the way. The exercises are where it sticks — free, no signup needed.
Start learning Japanese→Related Topics
- Rendaku in Spelling: Voiced Compound ElementsN4 — Why a familiar word 'changes sound' inside a compound — sequential voicing (rendaku) adds a dakuten to the second element, turning te + kami into tegami and hana + hi into hanabi.
- Dakuten and Handakuten: Voicing MarksN5 — The two small diacritics that expand hiragana — the dakuten (゛) that voices k→g, s→z, t→d, h→b, and the handakuten (゜) that turns the h-row into p — plus the じ/ぢ, ず/づ tangle.
- Compound NounsN4 — Japanese builds compound nouns by stacking noun on noun — Sino-Japanese jukugo read with on'yomi, native compounds that often rendaku, and mixed forms — always head-final, so long compounds decode right-to-left.