Above the word
Pitch accent is usually taught one word at a time, with each word given a pattern: where the pitch is high, and where it drops. That is accurate for a word said alone and incomplete for a word said in a sentence, which is the only place words normally occur.
In connected speech, words group into phrases, and each phrase has its own overall pitch shape: generally starting higher and drifting down, then resetting at the start of the next phrase. Individual word accents sit inside that larger movement and are modified by it.
This is why a learner who has memorised word patterns can still sound wrong. The words are right and the shape they sit in was never learned.
The gradual descent
Across a phrase, pitch tends to drift downwards. Later words in a phrase start lower than earlier ones, so the same word has a different absolute pitch depending on where it falls.
The practical consequence is that trying to hit a fixed high pitch for every accented word produces speech that sounds bumpy. The high of a later word is lower than the high of an earlier one, and following the drift is what sounds natural.
| Japanese | Romaji | Meaning |
|---|---|---|
| きょうは、あめがふっています。 | Kyō wa, ame ga futte imasu. | It is raining today. |
| わたしのともだちがきました。 | Watashi no tomodachi ga kimashita. | My friend came. |
| あしたのかいぎはなんじからですか。 | Ashita no kaigi wa nanji kara desu ka. | What time is tomorrow's meeting? |
| これはとてもおいしいです。 | Kore wa totemo oishii desu. | This is very tasty. |
| きのうえいがをみました。 | Kinō eiga o mimashita. | I watched a film yesterday. |
| せんせいがきょうしつにいます。 | Sensei ga kyōshitsu ni imasu. | The teacher is in the classroom. |
Say each of these as one falling arc rather than as a series of separately accented words. The arc is the thing to copy.
Where the phrase resets
The descent does not continue indefinitely. At certain boundaries the pitch resets upwards and a new descent begins, and those boundaries usually correspond to meaningful units: a topic, then a comment; a subordinate clause, then a main clause.
Getting the resets right does more for naturalness than getting individual word accents right, because it is what tells the listener how the sentence is organised.
| Boundary | Reset? | Why |
|---|---|---|
| After the topic particle | Usually | Topic, then comment |
| After a clause connector | Usually | New clause |
| Between listed items | Usually | Parallel units |
| Between a noun and its particle | No | One unit |
| Inside a compound word | No | One unit |
| Before the final verb | Sometimes | Depends on emphasis |
The fourth and fifth rows matter most for learners. Resetting pitch between a noun and its particle is a common error and it makes speech sound chopped.
The particle carries the drop
For many words the accent drop is not audible within the word itself. It appears on the particle that follows, which means practising bare nouns removes exactly the information you were trying to learn.
Always drill with a particle attached. It is a small change to how you practise and it makes the contrast audible.
| Japanese | Romaji | Meaning |
|---|---|---|
| はしが | hashi ga | Chopsticks, drop before the particle |
| はしが | hashi ga | Bridge, drop on the particle |
| あめが | ame ga | Rain, drop early |
| あめが | ame ga | Sweets, flat through the particle |
| やまが | yama ga | Mountain |
| ゆきが | yuki ga | Snow |
What is safe to ignore
Pitch accent has a large amount of detail, including regional variation, compound word rules and verb conjugation effects. Very little of it is worth systematic study for a learner whose goal is conversation.
The prioritisation below is worth following, because time spent on the bottom rows is time not spent on length and rhythm, which matter considerably more.
- Worth learning: the phrase-level descent and where phrases reset.
- Worth learning: the accent on the handful of common pairs you actually use.
- Worth noticing: that particles carry the drop.
- Safe to postpone: compound word accent rules.
- Safe to postpone: regional variation, unless you live with it.
- Safe to ignore: memorising dictionary patterns word by word.
Practising something you cannot hear yourself doing
Phrase-level pitch is the hardest thing in this guide to self-assess, because you hear your own intention rather than your output, and because the target is a shape rather than a discrete right answer.
Shadowing gives you the model and scoring gives you the check. Unihongo's pronunciation lab has a pitch accent focus mode and scores your attempt against a native line, which turns a vague sense that a sentence sounded off into something specific enough to repeat and fix.
It then has to survive real speech, where you are also choosing words. That gap is normal, and it closes with conversation volume: sessions in the immersive 3D classroom put the phrase shapes under load, and the report shows where the sentence came apart.
If you are choosing where to spend limited pronunciation time, length and rhythm come first. Pitch is what to work on once those are stable.