Why are the pitch accents in Mac dictionary, OJAD, and KaniWani different from actual audio examples

Why is the pitch in the Mac OS dictionary different from actual audio examples for many words? I read on a WaniKani Dogen article that you can check pitch accent for words in the Mac OS dictionary and on OJAD (https://www.gavo.t.u-tokyo.ac.jp/ojad/eng/pages/home). Also, KaniWani support told me that they get their pitch accent images from these sources - the Mac dictionary and OJAD. (I’m referring to the lines with circles on them showing the pitch for a word to be Heiban, Atamadaka, Nakadaka, or Odaka pitch).

However, the actual pitch from audio examples (on WaniKani, Forvo, Jisho) sound different to me.
For example, for the word Handicraft, 工作, the pitch accent shown in the Mac dictionary as well as in KaniWani is Heiban (low to high)

…but on WaniKani, Jisho, and Forvo, the audio example sounds a bit to me like Atamadaka (like the last “u” sound goes down)
WaniKani: https://www.wanikani.com/vocabulary/工作
Jisho: https://jisho.org/search/工作
Forvo: https://forvo.com/word/工作/#ja

(I’m assuming all be the standard Japanese pitch accent.)

Why is the pitch accents in the Mac dictionary and OJAD different for many words than what actual audio examples sound like on WaniKani, Forvo, Jisho? (I asked KaniWani this and he said he couldn’t say why there are discrepancies)

Everyone saying Kyoko’s pronunciation on WaniKani of 工作 is heiban made me listen to several times over - and I think I see why now y’all are saying it is heiban. I thought the last “u” sound was lower but it is _— with the “k” being low, and the rest being higher.
And thank you for reminding me (as Dogen mentioned as well) that going up in pitch is more subtle than going down in pitch. In the handicraft example, it was very subtle for my beginner ears indeed.

1 Like

Just a note on the sources, I’m pretty sure the Jisho audio comes from the old WaniKani audio files (from before the current Kyoko and Kenichi recordings were done), so for all intents and purposes, “Jisho” and “WaniKani” are the same source in this discussion. WK let Jisho use the data. One would expect them to be consistent with each other, and if there were then discrepancies with another source, that would be a separate discussion.

One thing I would check is when the Mac Dictionary and OJAD data is from. Pitch accent isn’t 100% stable over time. That is, it’s possible for resources to be published in the early 2000s that don’t include pitch accents that are now widely used by younger people. My electronic NHK pitch accent dictionary is the 1999 version and I sometimes find it isn’t completely accurate to modern pronunciation, and more recently published versions are better.

It’s possible that there are multiple acceptable pitch accents for a word and the pitch resources are just behind.


In Wanikani it sounds like heiban to me.

Contrast it with an actual nakadaka word like 食べる

Sounds very different.


How does a Canadian pronounce “about”? Depends on the Canadian.

Merriam Webster: " all dictionaries may be classified as descriptive or prescriptive, and some seek to be both types. A descriptive dictionary is one that attempts to describe how a word is used, while a prescriptive dictionary is one that prescribes how a word should be used"

I wouldn’t get too hung up on figuring out the definitive, crystallized pitch pattern, as language is not a crystallized thing. I’d learn the prescribed way but I’d also want to be aware of how it’s actually used, whether it’s dialect differences or local/individual idiosyncrasies.

Language is messy.

Another thing is that the high/low binary pattern for japanese pitch is a simplification. I don’t mean to bash on it, it’s certainly useful enough as a frame work to understand undulation in pronunciation. If you wanna see an example of what I mean, turn on the “pitch contour” function of the OJAD and look up an atamadaka word like 修理 or 荷物. They start high, go higher within the same mora, and then go lower than they started.


I agree with @alo. The WaniKani audio for 工作 is definitely heiban.

It’s possible that occasionally there’s an actual discrepancy between audio and dictionary pitch data, but mostly you probably just need more practice. Also, note that while pitch diagrams show heiban going up after the first mora, this shift is more subtle than down steps in non-heiban words.


WaniKani sounds heiban to me and so does what you linked on Forvo.
Jisho on the other hand is atamadaka.

I have the pitch accent extension for WaniKani and yea, some of the pitches are “wrong” in the sense that the speaker uses a different accent. Sometimes the female and male voice even differ between themselves, but its not super often. Maybe you’re sometimes misshearing? I know I had trouble telling the difference when I started learning.

1 Like

Kyoko is definitely Heiban, Kenichi is maybe not as clear, but possibly a lazy heiban

Edit: I’d add, in general I find Kenichi’s pitch to be less consistent with the apple dictionary (which I check on most words I learn). Occasionally he gets it “right” though when Kyoko gets it wrong. A clear example from memory would be 居間, where Kenichi and Kyoko definitely have opposite pitch patterns, Kyoko’s matching the apple dictionary if memory serves correctly.

Second edit:
I’ve changed my mind, listened to it one more time and I think Kenichi is a pretty standard heiban. Part of the problem is I’ve listened to Kenichi’s voice so much, if he says a word over all just a little different than one of his usual heiban recordings I think, “huh, usually that’s not how he sounds when he says a heiban word, maybe I’m missing something.” In reality, I think he just said the whole word in a slightly lower register than he usually speaks.

P.S. my personal favorite Kenichi recording is 従える, where it sounds like his voice cracks or something.


I appreciate the responses and insight! I am new and will likely understand it more over time.

Everyone saying Kyoko’s pronunciation on WaniKani of 工作 is heiban made me listen to several times over - and I think I see why now y’all are saying it is heiban. I thought the last “u” sound was lower but it is _— with the “k” being low, and the rest being higher.
And thank you for reminding me (as Dogen mentioned as well) that going up in pitch is more subtle than going down in pitch. In the handicraft example, it was very subtle for my beginner ears indeed.

…compared to another 3 mora/syllable Kanji - the vocab for sheep: 羊 WaniKani / Vocabulary / 羊 I can definitely hear the last two mora/syllables つじ are higher than the first. Is a bit more clear than the 工作 heiban. But I can see how 工作 is heiban now.

(If I go by the pitch graphic alone - and not hear the audio examples - my English stress accent self adds way to much pitch to the ups and downs - I need to become more subtle Japanese lol)

Thank you also for reminding me how language is messy and isn’t going to be 100% accurate with all audio examples and the sources. I was hoping they would match perfectly to make it easier ><

If I could ask a couple other questions:

The example @kokopelli121123 mentioned about 居間, the Mac dictionary shows the pitch is #2 (nakadaka; low-high-low); however, the word is just 2 mora/syllable “i” and “ma”, so shouldn’t the Mac dictionary say the pitch is #0 (heiban; low to high). Isn’t a 2 mora/syllable kanji, like 居間 “いま” either only atamadaka (high to low) or heiban (low to high) - because there are only 2 sounds?

Also, in regards to pitch accent, how much does the pitch accent change for a kanji when that kanji is in a sentence. I have not yet looked at any kanji in sentences or even looked at full sentences yet - so I’m just curious if, for example, the heiban pitch 工作 would be different in a sentence. So if I learn the pitch accent for each kanji as I go, I am essentially just learning the pitch accent for when saying that word/kanji by itself, e.g. just when saying “handicraft” by itself. But when a kanji is part of a sentence, that pitch accent that I memorized may be totally different?, e.g. “her hobbies include reading and handicraft” - in this sentence, I assume the heiban pitch I memorize for handicraft would not be the same in a sentence like this. So is memorizing pitch accent of each kanji mostly just for when saying that kanji by itself (and not in a sentence)?

I was just doing WaniKani, but I am now going to do WaniKani and KaniWani (or KameSame). And when I do KaniWani or KameSame, I’m going to just test myself on the pitch accent and also the drawing. Do you think this is wise? Is this what anyone else does / have done? I know will rarely draw the kanji but might need to and can help remember them, and thought I might as well learn as I go (and am not overwhelmed).

1 Like

It means that the pitch drops after the last mora, meaning if the word is followed by a particle the particle is low. By contrast, a heiban word stays high even on the particle that follows.

I assume you mean “word” and not “kanji” (because kanji don’t have pitch accents). If so, there is sentence level pitch and overall sentence flow. I don’t think it makes learning pitch or pitch patterns for individual words useless though. I haven’t studied sentence level pitch patterns yet, so I can’t say any more than that.


That one is tricky because both of the beginning mora are devoiced. So in a sentence, like @seanblue mentioned with particles following after a word, you’re really listening to what comes after it to be able to tell for sure.

For that one, a particle after a word with the accent at the end will drop in pitch.

The Wikipedia article has a lot of good info: Japanese pitch accent - Wikipedia

It’s not wildly different and there are patterns that you can pick up.

I honestly never learned word level pitch accent and only really did a ton of shadowing. Having to mimic the sentence flow helped with not just pitch accent, but also places where natural pauses occur.

1 Like

Pitch patterns in sentences either stay more or less the same (sometimes less dramatic towards the end of the sentence)
Sometimes they change based on rules that require you to know the pitch accent of the original word. For example nominalizing の does different things depending on the pitch pattern of the preceding verb. Also conjugated words change pitch (I.e. 食べる is Nakadaka, and 食べた is atamadaka, 読む is atamadaka and 読まない is nakadaka). These rules will apply to all similar verbs. That is, other Ichidan verbs that are nakadaka (I.e. 起きる or 閉める) will become atamadaka (in 起きた and 閉めた), and other godan atamadaka words (飲む or 富む) will become nakadaka (as in 飲まない and 富まない) where the accent is on the syllable before ない (with the pitch being dropped for the な)

Also I guess…
There are also some words that always mess with pitch like みたい (that is さる in the phrase 猿みたい will act heiban because all words befor みたい are heiban).

All that said, if you don’t know the pitch pattern of the original word, you still won’t know how to use it conjugated or in a sentence. I’d recommend learning pitch for words as you learn them (or at least listening for it). Also, I’d recommended early on getting some of the sentence basics down so you can listen for it in the wild and practice using it as you learn it. I think OJAD works relatively well for short and simple sentences. It makes weird mistakes every now and then, but if you regularly check how words in sentences and contexts change and behave via ojad, my guess is overtime, when ojad makes weird mistakes, you’ll notice and catch it yourself. Idk


oh super interesting thank you for mentioning that

oh that I also didn’t know. That is really good to know. Thank you! I will continue to just learn the pitch of each word and keep in mind that those rules. Thank you so much!

forvo and both of the WK examples sound very clearly heiban to me

rather than ignoring them, I’d pay attention to them and try to hear them when you listen to audio samples.

I think they’re pretty much always correct with standard japanese (tokyo dialect). it’s worth bearing in mind though, that other dialects differ wildly, and some don’t even have pitch accent

1 Like

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.