What you will be able to do
- Give a neutral syllable the pitch its neighbour hands it, instead of a tone of its own.
- Produce 儿化 as one syllable, and apply the changes -i, -n and -ng undergo before it.
- Apply third-tone sandhi across a phrase by chunking it, not by counting syllables.
- Gain speed by shortening unstressed syllables rather than by flattening tones.
Chinese words in this text can be activated to show a dictionary gloss.
What happens when syllables run together
The neutral tone has a pitch — it is just not its own
A neutral syllable is short, light and unstressed, and its height is decided by the syllable in front of it.
- After a first tone it sits a little below the high line: māma.
- After a second tone it lands in the middle: péngyou.
- After a third tone it comes out relatively high, because the voice is rising off the bottom: jiějie, wǒmen.
- After a fourth tone it stays low, where the fall left it: xièxie, bàba.
This is why with two full fourth tones sounds so wrong: the second syllable is not merely too heavy, it is at the wrong height.
Neutral tones come in two kinds. Grammatical ones are always neutral: , , , , , , the aspect , the plural , the suffixes and . Lexical ones belong to particular words and simply have to be learned: péngyou, shíhou, dōngxi, piányi, yīfu, rènshi, xuésheng, míngzi.
And the weight of that second syllable can carry meaning on its own:
- dōngxi thing / dōngxī east and west
- dìdao authentic, the real thing / dìdào a tunnel
: a colour, not a syllable
yìdiǎnr is two syllables, not three. The is not pronounced as er; it curls the end of the rime it attaches to, and the whole thing stays one beat.
What happens to the rime depends on how it ended:
- Ends in a vowel — just add the curl: huār, gēr, huàr.
- Ends in -i or -n — the -i or -n disappears: xiǎoháir, yìdiǎnr, wánr, wèir. Say wán, then say it again with the tongue curling and the n never arriving.
- Ends in -ng — the ng disappears and the vowel is nasalised instead: kòngr, diànyǐngr.
- The empty rime after z, c, s, zh, ch, sh, r is replaced outright: shìr, guāzǐr.
A short core list is genuinely standard and worth producing: zhèr, nàr, nǎr, yìdiǎnr, yíhuìr, wánr, xiǎoháir.
Beyond that list, usage is regional. It is heavy in Beijing and across the north, much lighter in the south, and in Taiwan zhèlǐ, nàlǐ and nǎlǐ are the everyday forms. A learner should master the core set and resist adding more by ear — over-applied is one of the few foreign accents that sounds like an imitation rather than a mistake.
Third-tone sandhi across a phrase
The rule is unchanged — a third tone before another third tone is said as a second tone — but in a long string you have to know where the phrase divides, because the sandhi applies inside each chunk.
- wǒ hěn hǎo is one chunk of three: work from the right, and it comes out wó hén hǎo.
- wǒ yě hěn hǎo, the same way: wó yé hén hǎo.
- wǒ xiǎng mǎi divides as + , and comes out wó xiáng mǎi.
- wǒ mǎi hǎo jiǔ divides as + . Each half changes on its own, so it comes out wó mǎi háo jiǔ — and the third tone in the middle survives, which the syllable count alone would never have predicted.
The lesson is short: the sandhi follows the sense, not the arithmetic. If you know where the phrase breaks, you know where the tones change.
Two other changes ride along in the same strings. A third tone before any non-third tone loses its rise altogether and simply stays at the bottom — lǎoshī, hǎochī, wǒmen — which is the shape you will use far more often than the full dip. And goes light and neutral inside hǎo bu hǎo and inside the potential tīng bu dǒng.
Rhythm, stress, and speed
Mandarin gives every full-toned syllable roughly the same length. What shortens are the neutral and unstressed ones, and that is where the sense of speed comes from.
- Within a phrase, the last full-toned syllable is the most prominent.
- Function words compress: , , , , and every particle.
- Do not pause between the syllables of a two-syllable word. is one unit; a gap in the middle is more foreign-sounding than a wrong tone.
The commonest way learners try to sound faster is to flatten the tones, which makes them harder to understand rather than easier. Speed comes from the opposite move: keep the tones fully shaped on the stressed syllables and let the light ones almost vanish.
A shadowing protocol
Take thirty seconds of natural audio — a dialogue you already understand — and work it for a week.
- Listen five times without the text.
- Read the text while listening, twice.
- Speak along with the recording, overlapping, five times. Do not try to be intelligible; try to arrive at the same moment as the model.
- Record yourself saying it alone.
- Compare, listening only for where the two recordings diverge in timing. Timing errors reveal chunking errors, and chunking errors are what cause the sandhi to go wrong.
Thirty seconds, ten minutes a day, one week. Then change the clip.
How to work through it
Where the light syllable sitsread
Read the first section, then say 妈妈, 朋友, 姐姐 and 谢谢 in a row, listening only to the second syllable of each. It is a different height every time, and none of those heights is a tone you chose.
Two words, one spellingaudio
Listen to 东西 said both ways — dōngxi and dōngxī — and to 地道 as dìdao and dìdào. Nothing separates the members of each pair except the weight of the second syllable. Decide which you heard before you look.
Cut the phrase, then change the tonesspeaking
Take 我买好酒 and draw the division before you speak: 我买 / 好酒. Say it, then say 我很好 as a single chunk of three and hear the difference. Then do the same with 我想买两把小雨伞 — divide first, speak second.
The core 儿化 wordsvocabulary
Seven 儿化 words plus 东西 for its neutral tone. Count the beats as you say each one — 一点儿 is two, 一会儿 is three, 小孩儿 is three. If you produce an extra syllable, the curl has become a word again.
Thirty seconds, ten timesspeaking
Run the shadowing protocol on one short dialogue you already understand. On the final comparison, ignore the consonants and the tones and listen only for timing — where you and the model fall out of step is where your phrasing is wrong.