Three separate systems are stacking on top of each other. Gemination: the numbers 1, 6, 8, 10 and 100 historically ended in a stop consonant, so they collapse into a small っ before an unvoiced consonant. The h to p shift: the modern は row was originally p, so after that っ it reverts, giving 一本. Rendaku: voicing after a moraic ん or inside a native compound, giving 三本 and 手紙. Nothing here is irregular. Learn the three rules and roughly ninety percent of the table stops needing memorisation.
Every learner meets the same wall around the third month. You look up how to count long thin objects, and the textbook hands you this:
一本 二本 三本 四本 五本 六本
Same character. Three different readings. And then, almost always, the same piece of advice: do not try to find a rule, just get exposure.
That advice is wrong, and it is expensive. There is a rule. In fact there are three, they were discovered a long time ago, and once you see them the counter tables stop being a memory problem and become a two second calculation. Better still, the same three processes explain 手紙, 花火, 人々 and a large slice of the vocabulary you will meet for the rest of your studies.
Rule 1: the numbers that break
The Sino-Japanese readings 一, 六, 八, 十 and 百 all end in a consonant in the Chinese they were borrowed from. Japanese has no room for a bare final consonant, so those syllables were padded out with い or く. When another consonant follows immediately, the padding collapses back into a geminate, written っ.
The numbers 二, 五, 七 and 九 never had a final consonant, so they never trigger anything. That is why they look "normal" in every table you have ever seen.
The refinement that most guides skip
Gemination does not fire in front of every counter. It needs an unvoiced consonant to lean against, which in practice means the か, さ, た and は rows. Counters beginning with m, n, r, w, y or a vowel are untouched, which is why 一枚 stays exactly as you would expect and never becomes いっまい.
And the numbers themselves are not identical. 一, 八 and 十 geminate in front of か, さ, た and は. But 六 and 百, which end in く rather than ち, only geminate in front of か and は. This is why 六冊 stays quiet while 八冊 does not.
| Number | before か | before さ | before た | before は |
|---|---|---|---|---|
| 一 (いち) | いっかい | いっさつ | いったい | いっぽん |
| 六 (ろく) | ろっかい | ろくさつ | ろくたい | ろっぽん |
| 八 (はち) | はっかい | はっさつ | はったい | はっぽん |
| 十 (じゅう) | じゅっかい | じゅっさつ | じゅったい | じゅっぽん |
| 二, 五, 七, 九 | no change | no change | no change | no change |
Rule 2: why は turns into ぱ
This is the piece of history that makes the whole thing click. In Old Japanese, the sounds now written は ひ ふ へ ほ were pronounced pa pi pu pe po. Over centuries they softened to an f-like sound and then to today's h. But the original p never fully disappeared: it survives in exactly the environments where a hard consonant is needed, which means directly after a geminate っ.
So the reading is not an exception. 本 is simply reverting to what it always was.
いち loses its vowel and becomes っ, then the old p inside は resurfaces because a geminate now precedes it.
The same shift is visible everywhere once you know to look for it. 日本 beside 日本. 出発 from 発. 切符 from 符. 一杯 from 杯. It is one of the most productive patterns in the language and almost nobody tells beginners it exists.
Rule 3: rendaku, and the law that stops it
When two native Japanese words join into a compound, the first consonant of the second word often becomes voiced. The mapping is exactly the dakuten you already know:
か → が さ → ざ た → だ は → ば
Lyman's Law: the brake
Rendaku does not fire indiscriminately, and the condition that stops it is remarkably clean. It was described for Japanese in the nineteenth century by Benjamin Smith Lyman, and it says:
This single constraint explains a pile of words that otherwise look like arbitrary exceptions:
Put 目薬 and 山風 side by side and the mechanism is obvious. 薬 contains no voiced obstruent, so く happily becomes ぐ. 風 already carries ぜ, so か is left alone. Two words, same structure, opposite outcome, one reason.
Two more things that block rendaku
Coordinate compounds. When the two halves are equal partners rather than modifier plus head, rendaku stays away. 好き嫌い is すききらい, likes and dislikes, not すきぎらい. Same with 飲み食い, のみくい.
Sino-Japanese vocabulary. Compounds built from on'yomi normally do not take rendaku, which is why 学校 and 会社 keep their unvoiced consonants. The famous exception is 会社 inside a longer word: 株式会社, where it has been absorbed deeply enough to behave like a native second element.
The freebie: 々 almost always voices
Repeated-word plurals are the highest hit rate rendaku in the language. If the word begins with か, さ, た or は, expect voicing.
And when the word starts with something else, nothing happens, exactly as the rule predicts: 山々, 色々, 我々.
Putting it together: the 本 table, derived rather than memorised
Now the table that started this article writes itself. Take 本, an は row counter, and run the three rules.
| Reading | Which rule fired | |
|---|---|---|
| 1本 | いっぽん | gemination + h to p |
| 2本 | にほん | nothing |
| 3本 | さんぼん | voicing after ん |
| 4本 | よんほん | nothing (see below) |
| 5本 | ごほん | nothing |
| 6本 | ろっぽん | gemination + h to p |
| 7本 | ななほん | nothing |
| 8本 | はっぽん | gemination + h to p |
| 9本 | きゅうほん | nothing |
| 10本 | じゅっぽん | gemination + h to p |
| 何本 | なんぼん | voicing after ん |
Why 三本 voices but 四本 does not
This is the detail that makes people give up on finding a system, and it has a clean historical answer. The ん in 三, 何, 千 and 万 comes from a nasal ending in the Chinese these words were borrowed with, and that nasal voiced whatever followed it.
四 is a completely different animal. The Sino-Japanese reading of 四 is し, and よん is a native Japanese form that was pressed into service later, largely because し collides with 死, death. Its ん never went through the Chinese sound history, so it triggers nothing. Same with 七 replacing しち.
So the shape of the rule is: a moraic ん of Chinese origin voices the counter; a native ん does not. That is not an exception list, it is etymology doing exactly what you would expect.
Where the system gets counter specific
Honesty matters more than a tidy story here. Gemination and the h to p shift are close to automatic. Voicing after ん is likely but decided by the individual counter, and there are two patterns worth knowing.
| Counter | 1 | 3 | 6 | 8 | 10 | 何 |
|---|---|---|---|---|---|---|
| 本 long objects | いっぽん | さんぼん | ろっぽん | はっぽん | じゅっぽん | なんぼん |
| 匹 small animals | いっぴき | さんびき | ろっぴき | はっぴき | じゅっぴき | なんびき |
| 杯 cupfuls | いっぱい | さんばい | ろっぱい | はっぱい | じゅっぱい | なんばい |
| 分 minutes | いっぷん | さんぷん | ろっぷん | はっぷん | じゅっぷん | なんぷん |
| 軒 houses | いっけん | さんげん | ろっけん | はっけん | じゅっけん | なんげん |
| 階 floors | いっかい | さんがい | ろっかい | はっかい | じゅっかい | なんがい |
| 回 times | いっかい | さんかい | ろっかい | はっかい | じゅっかい | なんかい |
Two things to take from this table. First, 分 is the odd one out: after ん it takes the handakuten and gives 三分 rather than さんぶん, so it behaves like the geminated forms rather than the voiced ones. Second, and far more useful in daily life, compare the last two rows. 三階 is the third floor. 三回 is three times. Both are written with かい in isolation, and the voicing is what keeps them apart in speech.
Want to know which JLPT band this material sits in for you? The free level test runs 25 questions across N5 to N1, five per level, and shows exactly where your accuracy falls off.
Take the free JLPT level testWhy this quietly wrecks your listening
Sound changes are not a written-language curiosity. They are a listening problem, and they are one of the specific reasons that a passage you could read comfortably becomes unrecognisable when someone says it out loud.
Your brain searches its vocabulary by sound. If you stored 本 as hon and only as hon, then sanbon arrives as a word you have never met, and the half second you spend deciding what it was is the half second the speaker used to say the next two phrases. This is the mechanism behind the very common complaint that Japanese sounds too fast. It usually is not too fast. The words simply are not arriving in the shape you filed them under.
One ordinary office sentence, three sound changes, none of them optional. Read it and it is easy. Hear it cold and it is three unfamiliar sound shapes in a row.
The four step method
How to work out any counter in about two seconds
- Look at the number. Is it 1, 6, 8, 10 or 100? If yes, expect a small っ. If it is 2, 5, 7 or 9, expect nothing at all and stop here.
- Look at the counter's first sound. Gemination only fires before か, さ, た and は rows. Anything else, and the number stays whole.
- Check for an h. If the counter starts with an は row sound and a っ now sits in front of it, the h becomes p. 本 becomes ぽん, 匹 becomes ぴき, 杯 becomes ぱい.
- Handle the ん numbers separately. With 3, 何, 千 and 万, expect voicing on most counters. With 四 and 七, expect nothing, because those readings are native.
Run that on a counter you have never met and you will be right far more often than you are wrong. And crucially, when you are wrong, you will be wrong in a specific, correctable way rather than blank, which is what turns a mistake into a memory.
How to actually get this into your ear
Rules make the system make sense. They do not, on their own, make sanbon feel like a word. That takes the sound arriving in your ear enough times, spaced correctly, attached to a real sentence rather than a table cell.
This is where Kanjijo does the part a blog post cannot. Every vocabulary and kanji item carries an exclusive mnemonic written to be remembered rather than translated, and every example sentence comes with furigana over the kanji and a full English translation underneath, so the changed reading is right there in context instead of buried in a chart. The SRS engine then schedules each item for the day you are about to lose it, which is exactly the treatment sound-change items need, because they decay faster than ordinary vocabulary.
Between sessions the material keeps working. A home screen widget and a lock screen widget put today's items where your eyes already go, and the test widget quizzes you from the home screen without opening anything. When you meet a counter in the wild on a price tag, a floor directory or a menu, the OCR scanner reads it off the image and turns it into a study item on the spot.
For the listening half of the problem, the dictation track makes you write down what you actually heard, which is the fastest way to discover that you have been storing a word under the wrong sound. Shadowing puts your own mouth through the same shifts, JLPT listening sets test them at exam speed, and full mock exams tell you whether it has stuck.
Reference summary
- Gemination
- 1, 6, 8, 10, 100 produce っ before か, さ, た, は rows. 6 and 100 only before か and は.
- h to p
- は row reverts to ぱ row after っ, because it was originally p.
- Rendaku
- か→が, さ→ざ, た→だ, は→ば on the second element of a native compound.
- Lyman's Law
- Rendaku is blocked if the second element already contains a voiced obstruent.
- Other blockers
- Coordinate compounds and most Sino-Japanese compounds.
- ん voicing
- Chinese-origin ん (三, 何, 千, 万) voices most counters. Native ん (四, 七) does not.
- Notable outlier
- 分 takes handakuten after ん: 三分 is さんぷん.
Go deeper
- On'yomi and kun'yomi: the four historical layers of Japanese readings
- The 15 counters you actually need, and what each one counts
- Long vowels and the small tsu, and why they change meaning
- Why you can read it but cannot hear it
- Same reading, different kanji: 聞く vs 聴く and the rest
- Free JLPT level test: find your real level in 25 questions
Stop memorising tables that have a rule underneath them
Kanjijo gives you the full kanji, vocabulary and grammar bank for every JLPT level, exclusive mnemonics on every item, furigana and English on every example sentence, an SRS engine, lock screen and home screen widgets, a test widget, OCR scanning, JLPT reading and listening practice, mock exams, conversation, shadowing and dictation.
Download Kanjijo FreeFrequently Asked Questions
Three rules are stacking. 一 (いち) geminates into っ, and the h in 本 reverts to its original p, giving いっぽん. 二 (に) triggers nothing, so 本 stays ほん. 三 (さん) ends in a Chinese-origin moraic ん, which voices the counter into ぼん. The counter is not irregular; the numbers are doing different things to it.
Rendaku (連濁), or sequential voicing, is when the first consonant of the second element in a compound becomes voiced: か→が, さ→ざ, た→だ, は→ば. It is why 手 plus 紙 gives てがみ and 花 plus 火 gives はなび.
Rendaku is blocked when the second element already contains a voiced obstruent, meaning a g, z, d or b sound. That is why 山 plus 風 stays やまかぜ rather than becoming やまがぜ, since かぜ already contains ぜ. The same explains 春風 and 紙屑.
1, 6, 8, 10 and 100 cause gemination because they historically ended in a stop consonant. 3, 何, 千 and 万 end in a Chinese-origin ん which voices most counters. The numbers 2, 5, 7 and 9 trigger nothing at all.
Voicing after ん is decided counter by counter rather than automatically. 階 voices to がい while 回 stays かい. 分 goes further still and takes handakuten, giving さんぷん. Gemination and the h to p shift are near automatic; voicing after ん is only highly likely.
Exposure alone works eventually, but it is slow and it leaves you unable to guess anything you have not already heard. The rules take about fifteen minutes to learn and immediately let you predict counters you have never met, which is what converts exposure into learning rather than repetition.