69.1% of Japanese verbs are accented; frequently used verbs tend to be flat.

The Data

All verbs from my JLPT deck (N5 through N1), excluding:

  • サ変動詞 (noun + する verbs like 勉強する)
  • Polyphonic entries (words with multiple pitch patterns like 致す (②, ⓪), 差し上げる (⓪, ④))

The pitch accent is notated using circled numbers where:

  • ⓪ (Heiban/Flat): Pitch rises after first mora and stays high
  • ① (Atamadaka): First mora high, then drops
  • ②③④… (Nakadaka): Pitch drops after the indicated mora
Data Filter
This filter applies to all statistics below

Overall Results

Level Total Heiban (⓪) Atamadaka (①) Nakadaka (②+)

We see a very interesting trend: lower level verbs tend to be heiban. It’s safe to say that frequently used verbs are more likely to be heiban, reaching roughly 50% at the N5 level.

Testing the [-2] Hypothesis

A commonly cited rule: verbs tend to have their pitch drop on the second-to-last mora (the [-2] position).

For a 3-mora verb like かくす, this would be ②. For a 5-mora verb like あたためる, this would be ④.

Results:

Loading...

So yeah, the hypothesis holds some water.

Detailed Breakdown by Pitch Number

Level Total

Following Up

Would be interesting to weigh the verbs by frequency to produce an “effective” pitch accent distribution.

Raw Data

Non-Compound Verbs (1183)

Loading...

Compound Verbs (349)

Loading...

*Compoundness is determined with LLM, mostly accurate with minor mis-classifications.