Japanese Number Irregulars

The Hard Part

The five readings that shift

Here's where it gets a little fiddly. Five place combinations refuse to follow the rule you just learned: 300 should be さんひゃく, but it isn't. Frustrating? A bit. The reassuring part is that there are only five of them and the changes aren't random. The sounds shift for ease of speech — a process called euphonic assimilation (音便, onbin) — and two mechanisms account for every case, so once you can name them they stop feeling like exceptions at all.

Rendaku — the voicing of 三

Rendaku (連濁, "sequential voicing") turns a crisp consonant into its voiced partner when it follows certain sounds. After さん, the ひ of 百 voices to び (300 → さんびゃく) and the せ of 千 voices to ぜ (3,000 → さんぜん). The nasal ん is doing the pulling — voiced sounds flow more smoothly after it than voiceless ones.

Gemination — the doubled stop in 六・八

Gemination (the small っ, sokuon) clips a syllable into a held consonant before the next one. ろく + ひゃく collapses to ろっ + ぴゃく; はち + せん collapses to はっ + せん. Before 百, that clipped stop also pushes ひ all the way to a popped ぴ (the handakuon), giving ろっぴゃく and はっぴゃく.

How They Stack

The combinations follow for free

The five base cases are the whole story — every larger number that contains them simply inherits the shift. You never memorize the combinations separately: lock in the five bases and the rest assemble themselves. Here is the complete set of shifts within 1–9,999, which covers every combination you will encounter below 万:

Every row is predictable from the five base rules. A number like 6,300 changes only in the hundred slot; 3,800 and 8,800 each fire in both slots independently. From 万 upward, each four-digit block operates on its own, so the irregulars keep firing inside each block exactly as they would for the standalone number — さんびゃく and はっせん work identically at any scale.

The Other Kind of Irregular

4, 7, 9 — the unlucky readings

The five above are phonetic. Three more readings shift for a cultural reason: their Sino-Japanese sound collides with an unlucky word, so speakers reach for the native reading instead. This is why よん, なな, and きゅう dominate everyday counting.

The taboo is real enough to shape the built world: many hospitals and hotels skip floor and room numbers 4 and 9, and gift sets avoid quantities of four. 42 attracts particular attention — し + に can be parsed as しに, echoing 死に (dying) — so building directories and product counts often quietly skip it alongside 4. 七 (しち) is dodged less for superstition than for clarity — over a phone it blurs into いち, so なな wins.

Common Questions

Quick answers

From Knowledge to Fluency

Knowing the reason isn't the same as the reflex

Understanding that さんびゃく is rendaku makes the rule stick, but in conversation you don't have time to think "rendaku." You need 600 to surface as ろっぴゃく before you've finished hearing it. The five irregulars are exactly where learners freeze under pressure, because they're rare enough to feel uncertain and common enough to matter.

The pattern of mistakes is consistent: learners nail the regular hundreds and thousands under no pressure, then revert to the default forms the moment a price, an address, or a year flashes by at conversational speed. Knowing the mechanism (rendaku / gemination) is the first step; drilling the five cells to automaticity is the second — and the second step is the one that requires reps, not re-reading.

Retrieval practice closes that gap: see the number, commit to the reading out loud, then check. The misses are the lesson — and there are only five of these to get wrong, so a few rounds and they really do click. Twenty focused reps is worth more than re-reading this page five times.