If you’ve studied Japanese for at least a little while, you’ve probably run across a phenomenon known as rendaku. Rendaku means “sequential voicing” and it’s when the reading of a word changes when coupling multiple parts of the word together. I think it’s probably easier to show you some examples of common words that are affected by rendaku:
人々 ＝ ひとびと（not ひとひと)
手紙 ＝ てがみ（not てかみ)
When the two parts of these words are separated they are different from when they are combined. ひと changes to びと and かみ to がみ. To most Japanese learners, this behavior probably feels random. But, there’s a method behind the madness, and you’re going to learn about it via this guide.
Examples Of Rendaku Words
The best way to understand rendaku is to look at some examples.
さくら turns to ざくら
かっぱ turns to がっぱ
はな turns to ばな
ひと turns to びと
とき turns to どき
かいしゃ changes to がいしゃ
You can see how in all these words the first kana in the second part of the word changes to its dakuten version (dakuten is the little quotation mark / circle that modifies a kana). So, these are examples of words that do get converted via rendaku. What about the ones that don’t?
What Doesn’t Cause Rendaku?
I think an easier way to go about understanding rendaku would be to figure out what doesn’t cause a word to change. Looking at it this way produces a more solid set of rules that you can work off of.
1. When A Word Already Has a Dakuten/Dakuon
If the second word in the sequence has a dakuten there’s nothing you can do to make it go rendaku. There’s no anti-rendaku that removes a dakuten/dakuon.
2. Lyman’s Law
Lyman’s Law is perhaps the most famous way to figure out whether or not a word will get the rendaku treatment. Lyman’s law states that:
Rendaku does not occur when the second element of the compound contains a voiced obstruent in any position.
Okay. What the hell is a voiced obstruent (seriously, wtf linguistic people?). A voiced obstruent, put simply, is a consonant sound (so, not a vowel) which is formed by “obstructing airflow in your throat.” Try saying the following sounds and see if you can figure out what I mean:
b, d, g, v, j, z
If you do a lot of these sounds slowly you’ll find your throat has to close a bit (obstruct airflow) in order to be made. Of course, not all of these sounds are applicable. That being said, there are exceptions to this “law” but it will keep you to stay clear most of the time.
Anyways, the idea is basically this: Rendaku won’t occur as much when the second word/section has a voiced obstruent in it. Let’s take a look at some examples:
The second section (originally ひと) has no voiced obstruent in it. Because it didn’t have a voiced obstruent, it does do the rendaku treatment.
This is a weird one. In words that are duplicators (same thing twice), the second part is usually rendaku’d. Same goes with 人々.
山火事（やまかじ）＝ Forest Fire
The second word/section (かじ) does contain a voiced obstruent. Thus, the rendaku does not occur (it is やまかじ not やまがじ).
As I’ve mentioned, there are plenty of exceptions out there, but Lyman’s law will help a lot if you’re ever having trouble figuring out the reading of a word.
3. Foreign Words
Japanese has a lot of loan words, where non-Japanese words have been incorporated into the Japanese language. These are most often written in katakana. In cases like this, you’ll hardly ever see rendaku being used.
アイスコーヒー (Iced Coffee)
You would not change this to アイスゴーヒ, it is a foreign word.
Besides foreign words like these, though, there are also words that were imported via the kanji (from China). The on’yomi reading of a kanji is considered the Chinese reading. The kun’yomi is the Japanese reading, which later got associated with the kanji. When working with the on’yomi reading of kanji, you’ll often combine these together to create a single word. These types of jukugo words rarely get the jukugo treatment, though it does happen from time to time.
The kun’yomi reading of kanji, however, gets the rendaku treatment quite a bit. If you see some kana attached to a kanji or two this is a good indication that the word you’re trying to read uses kun’yomi readings and therefore may be a rendaku word.
Groups Of Rendaku Behavior
Lyman’s law and other “behavior rules” aren’t always reliable. There are some other things you can do to help figure out when rendaku takes place or doesn’t take place. There are different “groups” of words that have different kinds of rules. If you know these groups (especially this first one) you’ll be able to figure out rendaku words more effectively.
Group #1: Immune To Rendaku
This set of Japanese (“Yamato”) nouns (versus nouns imported from China … we’re talking kun’yomi here) never undergo the rendaku treatment. Out of allllll the Japanese word nouns out there, this is a tiny fraction. It’s the exception so to speak, but it’s also solid, without any exceptions of its own.
- 浜（はま）＝ Beach
- 下（した）＝ Below
- 土（つち）＝ Earth
- はし（はし）＝ Edge
- かまち（かまち）＝ Framework
- 滓（かす）＝ Garbage, Scum
- 艶（つや）＝ Gloss
- 枷（かせ）＝ Handcuffs
- 暇（ひま）＝ Leisure
- 北（きた）＝ North
- 姫（ひめ）＝ Princess
- 形（かたち）＝ Shape
- 煙（けむり）＝ Smoke
- 紐（ひも）＝ String
So, with these examples, you can conclude that none of them will appear in the rendaku form (ひま will never be びま, けむり will never be げむり, and so on). Want some examples of this? Here you go:
横浜（よこはま）＝ Yokohama (the city)
よこはま doesn’t turn to よこばま
As far as I can tell, this doesn’t have a voiced obstruent in the はま, but because it’s one of the exception words it doesn’t change to ばま.
顔形（かおかたち）＝ Facial Features
かたち stays at かたち. It is immune to rendaku.
血煙（ちけむり）＝ Squirt Of Blood
けむり stays at けむり, just like it should.
You can also do a search on jisho.org where you take the word (above) and put a * before it. That will show you all the words with something before the word you put in. For example, if you put in *暇 you’ll get all the words with stuff before 暇 in them. You’ll have to look for the ones that are pronounced ひま though, rather than the on’yomi or other pronunciations.
Example: Look up *暇
If you learn this group of words (and you will eventually, though it’ll happen automagically over time as you gain more experience), you’ll at least know a list of words that never get the rendaku treatment. There are others, as well, though, and they’re not quite as friendly.
Then again, just like with everything rendaku, there’s exceptions with the “never go rendaku” words too. Not so “never go rendaku” are you, rendaku? Just shows why this topic is so difficult to pin down.
Group #2: Rendaku Resistors
The above Group never gets all rendaku up in your face. This group of nouns just resists the rendaku treatment, where it is the exception when it happens.
According to a study done by Eric Rosen (“Systematic Irregularity in Japanese Rendaku: How the grammar mediates patterned lexical exceptions” … seriously, a mouthful of a title), these 8 nouns consist of 50% of all the cases of rendaku resisting words. That means if study these eight words and the combo-words they’re involved in, you can make educated guesses on some of the words you’re not sure about when you see them written in kanji.
- 草（くさ）＝ Grass → Resists rendaku 84% of the time
- 原（はら）＝ Field → Resists 57%
- 癖（くせ）＝ Habit→ Resists 75%
- 皮（かわ）＝ Skin → Resists 42%
- 先（さき）＝ Tip → Resists 100%
- 木（き）＝ Tree → Resists 61%
- 子（こ）＝ Child → Resists 38%
- 手（て）＝ Hand → Resists 75%
Now, there’s something interesting about the rendaku resisters. Their resistance occurs only in “short-short” compounds. That is, compounds that are two or less kana long (on both sides). When there is a long compound, these “rendaku resistant” words can no longer resist. For example:
常盤木（ときわぎ）＝ ときわ ＋ ぎ
You can see the first compound is three kana long. Thus, it is a “long-short” compound. Anything that’s not a short-short compound will not resist rendaku (if it is a rendaku resisting word).
But, when 木 is used in a short-short compound, things are totally different. Even though it doesn’t resist all the time, the percentage is much better (i.e. not 100% rate of rendaku like long compounds).
This is a short-short compound
本木（もとき）＝ Original Tree Trunk
もとき is a short-short compound (both sides are two or less kana long), and it is more resistnat.
生木（なまき）＝ Live Tree
That being said, there are still short-short compounds for the rendaku resisting words that still can’t resist. It’s just that they don’t occur as often (though they still do occur).
山木（やまぎ）＝ Mountain Trees
Even though it’s a short-short compound, it can’t resist the rendaku (き becomes ぎ). There are exceptions, since it’s only rendaku resistant and not immune.
I’d say the most tricky ones are the resistant words. They tend to not go rendaku (if you had to bet, you’d at least have better than 50% odds most of the time, I suppose?) but they sometimes do as well. It’s a tricky business, but not something that’s too hard once you’ve studied Japanese for a while and you know a decent number of words.
Group #3: Rendaku Lovers
Then, there’s a group of nouns that love rendaku. They never resist, and you’ll almost always see them in the rendaku form if they are the second part of the word. This is Rosen’s short list of of words that make up 39% of all the occurrences of these rendaku loving nouns.
- 風呂（ふろ）＝ Bath → 100% Rendaku’d
- 腹（はら）＝ Belly → 100% Rendaku’d
- 船（ふね）＝ Boat → 100% Rendaku’d
- 骨（ほね）＝ Bone → 100% Rendaku’d
- 花（はな）＝ Flower → 100% Rendaku’d
- 笛（ふえ）＝ Flute → 100% Rendaku’d
- 金（かね）＝ Gold → 100% Rendaku’d
- 口（くち）＝ Mouth → 100% Rendaku’d
- 底（そこ）＝ Bottom/Sole → 100% Rendaku’d
- 箱（はこ）＝ Box → 100% Rendaku’d
- 紙（かみ）＝ Paper → 100% Rendaku’d
- 人（ひと）＝ Person → 100% Rendaku’d
- 形（かた）＝ Shape → 100% Rendaku’d
Then, these ones are still Rendaku Lovers, but they don’t love rendaku quite as much as the previous list.
- 鳥（とり）＝ Bird → 80% Rendaku’d
- 雲（くも）＝ Cloud → 80% Rendaku’d
- 川（かわ）＝ River → 61% Rendaku’d
- 玉（たま）＝ Ball → 80% Rendaku’d
These words tend to be a little bit more reliable (at least compared to the rendaku resistors). Most of them are 100%, and the ones that aren’t 100% tend to be a lot closer (80% for all but one). If you take a look at these words, they should voice the rendaku every (or almost every) time.
That being said, I found some exceptions to these rules (as in, some of the 100% ones aren’t 100%). There’s 悪口 (わるくち, which to be fair can also be written わるぐち) and there’s 仕口 (しくち, which isn’t a super common word, but you get my drift). There are others as well, but I think Rosen’s 100% list is pretty close, though I’d bring some of them down to around 90% (still close enough to make smart bets in my book).
Then Again, You Could Just Learn The Words
This is a huge amount of information. I never knew any of this until just recently and somehow got by just fine. I also doubt that anyone else learns about this either, and they somehow get by okay too. I think the above information is helpful, don’t get me wrong, but I think it’s more helpful for making guesses when you aren’t sure how to read a word written out in kanji.
Most likely, it’s just going to come down to experience. The more words you see the more natural rendaku will become. You’ll do it (or not do it) naturally without even thinking. Sometimes you’ll get caught up somewhere, but that’s why you have these rules to help you to guess more effectively.
Good luck with your rendaku endeavors. It’s overwhelming. It’s difficult. But it’s interesting as well, I think.