English Rhythm for Japanese Speakers: Stress Is Volume, Not Pitch

チョコレート is five morae. Chocolate is two syllables. CHOC-lit, and not because anyone is being lazy: two is the main entry in Cambridge's American dictionary, and the three-syllable version is the listed variant. Three of your five beats were never in the English word at all.

English rhythm is the last thing a Japanese speaker fixes, and usually the wrong thing they get told to fix. If you have already put in the hours on R and L, pulled your vowels back out of the katakana set, and people still say "sorry, one more time," the problem has moved somewhere your textbook probably filed under advanced. Search this and you will be told to work on your timing. The measurements say your timing is the part that already works.

Your timing is already fine. Your volume is flat.

39.1% against 39.4%. The first number is how much native English speakers rely on duration to mark a syllable as stressed. The second is how much advanced Japanese learners of English rely on it. Konishi, Yun and Kondo measured all four acoustic cues across 72 Japanese learners and 25 native English speakers, and on duration the two groups are essentially identical.

Now loudness. Natives get 21.7% of the job done with intensity. Advanced Japanese learners, after years of study, get 10.0%. The cue they reach for instead is pitch: 35.0% against the natives' 24.6%.

Say banana out loud, with your hand flat on your chest. Three syllables, and English marks the middle one. Did your voice climb on -na-, or did you feel a push against your hand? Both make a syllable stand out. English uses both, and the push is the one you are leaving out.

That substitution makes sense, because pitch is the only tool Japanese ever gave you for this job. 箸 and 橋 are both hashi. What separates them is melody and nothing else: 箸 falls (HA-shi), 橋 rises (ha-SHI), and a third hashi, 端 (edge), is flat: when a particle follows, 橋 drops on the が while 端 stays high. There is no loudness anywhere in that system. So when an English teacher says "stress this syllable," the instruction arrives in a language where prominence means melody, and you do the sensible thing with it.

Hold on to those numbers, because the next section says your vowel durations are too even, which sounds like the opposite claim. It is not. Konishi measured which cue you reach for; the recordings below measure how far you actually move it. You reach for duration as readily as a native speaker does, and you still do not move it far enough. Only the second of those is a timing problem, and it stays small because the cue built to carry the rest of the load is the one you are barely touching.

Konishi's own description of the fix is that the move from beginner to advanced is more intensity and less pitch.

What a native ear listens for is the spread

"Every syllable comes out the same length" is the standard description of a Japanese accent in English, and it is not what the recordings show. Kawase, Davis and Kim recorded ten Japanese and ten Australian speakers producing English and compared the proportion of speaking time spent in vowels. The difference between native English and Japanese-accented English was not significant, p = .821. Japanese English does not contain too much vowel.

What differed was the spread. On VarcoV, which measures how much vowel durations vary across an utterance, native English scored 50.0 and Japanese-accented English 39.95, p < .001. Your long syllables are long and your short ones are short. They are simply not far enough apart.

In native English, vowel intervals varied significantly more than consonant intervals. In Japanese-accented English the two did not separate, p = .101. English uses its vowels as the elastic part of a sentence and lets its consonants vary less than its vowels do. Japanese English stretches both about equally, which leaves a listener nothing to grab.

This is also the honest answer to why English sounds fast. It is not that Americans talk quickly. It is that the words carrying the information are propped up and everything between them is collapsed, so a listener tuned to even beats hears a blur where a native hears three clear landing points and some connective tissue.

This does not describe every Japanese speaker, and the honest version says so. Kawase's group note that Grenon and White found no significant difference on the same measures in their own sample, and put the discrepancy down to those participants having spent around two years living abroad. The pattern fades with immersion. If you have been living in an English-speaking country for years and people have stopped asking you to repeat, you may already have done this work without naming it.

You already have a silent vowel, but Japanese devoicing is a different machine

Say 靴下 out loud. Kutsushita, socks. Four morae, and three of them start with a voiceless consonant and carry a high vowel, which is the exact environment for Japanese devoicing. Only two go silent. The く devoices, the し devoices, and the つ in the middle survives, because the accent sits on it. That alternating pattern is not a quirk of this word: in an NHK dictionary survey, 84.7% of three-in-a-row devoicing environments come out silent-voiced-silent.

So you have been silencing vowels since you were three. した, 好き, 学生: every one of those traps a high vowel between two voiceless consonants, which is where the rule is near-obligatory. です is the weaker, word-final version, and it devoices because it is everywhere rather than because that environment is strong. Every Japanese speaker learning English gets told to reduce their unstressed vowels, hears that as "do the です thing," and finds it does not work. Here is why.

Funatsu and Fujimoto put sensors on the tongue and watched what it does during a devoiced vowel. The finding, reported in Fujimoto's handbook chapter, is that tongue movements are identical for /kide/ with a voiced /i/ and /kite/ with a devoiced /i/. Devoicing is "accomplished solely by laryngeal articulation." Your tongue makes a perfect /i/. Your vocal folds simply do not switch on.

English reduction is the opposite operation. The tongue never gets where it was going. Flemming measures medial schwas at around 64 ms against roughly 150 ms for a tense vowel in fluent speech, and the shortness comes first: "the short duration of non-final unstressed syllables motivates the neutralization of vowel quality contrasts in these contexts." The gesture runs out of time and lands somewhere in the middle.

You can watch the same machinery from the other side in what happens to the i and u in desu and suki.

The schwa is where English keeps its function words

The, uh, a, to, of, was. Mines, Hanson and Shoup went through 103,887 phonemes of recorded American conversation and found that 54% of every schwa in the corpus belongs to six of the ten most frequent words in the language. Schwa is not scattered evenly through English vocabulary. It is mostly the grammar. That is also why it is the most common vowel sound in English, at 7.30% of all phonemes in conversation.

Function words carry two pronunciations, and dictionaries label them. Cambridge prints a strong form and a weak form for each:

The weak form is the default. A, the and and reduce essentially always. To, of and at go strong at the end of a sentence, and can and have go strong there and when they contract with not. None of that is often. Hearing them coming at you is the other half of this problem, and it is what connected speech does to a sentence. Say all eight at full value in one sentence and you have handed a listener eight loud syllables that carry no information, on top of the ones that do.

Why "I can send it" sounds like "I can't send it"

"I can send it Tuesday." Your manager writes back asking what the blocker is. You said you could. They heard that you couldn't.

You have probably been told the difference is the /t/, or that it is the vowel. In American English it is mostly neither. Takahashi and Ooigawa put it flatly: "American English does not have any phonemic-level vowel distinction" between can and can't. Both are /æ/. And the /t/ frequently does not survive at all.

What happens in fast American speech is that can't becomes [kæ̃ʔ]. The /n/ drops, leaving nasalization on the vowel, and the /t/ becomes a catch in the throat. Broeders and Gussenhoven give this as a general rule of American English and name can't as their example. The negation has dissolved into the color of a vowel. Meanwhile the positive can has shrunk in the other direction, to /kən/ and sometimes all the way to a hummed kn with no vowel left in it.

The contrast is real, then, but it lives in size and nasality rather than in any segment you can point at. The one thing reliably present when the sentence is positive is the [n], which runs backwards from every other modal, where the [n] in shouldn't and wouldn't is what marks the negative. That is a hard listening task, and the numbers say so. Takahashi and Ooigawa played the pair to 30 Japanese listeners and asked them to tell the two apart. On American English: 62.5% correct, where guessing gets you 50%. On Australian English, where can't has the long broad vowel it shares with RP, 77.5%. The same study notes that ten out of eleven professional interpreters reported running into this on the job. If it has been catching you out, you are in company that gets paid for listening.

Now compare your own language. できる, できない. Negation in Japanese is ない, a 助動詞 that attaches to the stem and conjugates. Segments. Always there, always audible. Japanese never asks a listener to recover a negation from the nasal quality of a vowel.

So the practical fix runs opposite to the instinct. Shrink your positive can. If every can you produce is a full loud /kæn/, you have removed the only thing distinguishing it from the negative, and your listener is down to guessing.

Count the beats before you fix the sounds

Back to チョコレート. Five morae, two English syllables, and the three extra beats come from three different places at once.

チョ is two characters and one mora. The ー in レー is a mora that is not a syllable. And the コ and the ト carry vowels English never had, added so the word fits Japanese syllable structure. Japanese Wikipedia's article on the mora uses this exact word as its worked example: syllables チョ｜コ｜レー｜ト, morae チョ｜コ｜レ｜ー｜ト. The inserted vowels are their own problem, and they belong to the katakana trap.

Strengths is the extreme case. One syllable, nine letters, four consonant sounds stacked after the vowel, and one single beat.

This is the one place katakana earns its keep. As a pronunciation guide it does real damage. As a beat counter it works in reverse: whatever number the katakana gives you, the English word has fewer, and you can usually find the target by asking how many vowels survive when a native says it at speed. Stress shift inside a single word is a separate puzzle with the same moral, and it is why photograph and photographer do not share a vowel. The rules are laid out in a post for French speakers.

Neither language is a metronome, and the English rhythm drill still works

Three stressed words, three syllables or nine, and the line is supposed to take about as long either way. Here is the ladder, straight out of an MIT pronunciation handout:

CATS CHASE MICE. The CATS CHASE MICE. The CATS have CHASED MICE. The CATS have CHASED the MICE. The CATS have been CHASING the MICE. The CATS might have been CHASING the MICE.

Everything unstressed gets squeezed into the gaps.

The claim underneath the exercise is false, and you should know that before you spend a month on it. Mark Liberman measured the gaps between stresses directly and watched them stretch from 278.8 ms to 535.4 ms as he added syllables in between. Japanese does not survive its half of the story either: Beckman tested constant mora duration and reported that "neither of these predictions was borne out." What actually separates the two languages is structural, and you can count it. English permits more than fifteen syllable types and Japanese four, and English reduces its unstressed vowels while Japanese does not.

Use the ladder anyway. It is a training input rather than a description of English, and for this exact population there is evidence it moves something: Sugiura and Hori primed Japanese adolescent learners with a beat before they spoke and the ratio between their stressed and unstressed syllable durations shifted significantly toward the native pattern, measured immediately afterwards with no follow-up test.

Make the small words half the size

In the native recordings Sugiura and Hori used as their model, stressed syllables averaged 394 ms and unstressed syllables 173 ms, a little over two to one. Duration is what they measured, and two to one is the size difference to aim for. Duration is also the biggest single cue in English, and it is the one you are already working like a native. Intensity is the one you run at half a native's rate, so that is where the room is.

Record yourself reading the CATS ladder and listen for whether the small words are shrinking or merely speeding up. If the and have are arriving at full volume, faster, you are doing the Japanese version of the exercise.

Two limits worth knowing before you start. Sugiura's earlier repetition study found the effect mostly in duration rather than vowel quality, and a week later it had held only for schwa at the start of a word. And Derwing, Munro and Wiebe's twelve-week comparison, run on mixed-L1 ESL classes rather than Japanese speakers specifically, found that prosody training and segment training both improved read sentences, while only the prosody group got more comprehensible in spontaneous speech. That is the reason to do this work, stated accurately: it is what transfers when you are not reading off a page.

Pick one sentence you say often. "I can send it this afternoon" works, since it carries the can problem and four function words. Say it once with the small words swallowed, once carefully, and compare the recordings. If the careful version sounds clearer to you and the swallowed version sounds sloppy, that instinct is the thing to argue with. To an American ear the swallowed one is the one that sounds like a sentence.