Why Your Tones Collapse in Sentences

You can say 买 mǎi and 卖 mài perfectly. You have done the minimal pairs, you pass the listening drills, and on a flashcard your tones are clean. Then you open your mouth to say a whole sentence and something you can hear but not control flattens all of it into a grey line.

This is not a sign that your tone work was wasted. It is a sign that you have finished the first skill and not yet started the second. Tones in isolation and tones in running speech are related the way scales are related to playing a piece: necessary, not sufficient, and taught in the wrong proportion almost everywhere.

Here is what is actually happening, and what to do about each part.

1. Your attention budget runs out

Producing a Mandarin sentence requires you to choose the words, order them, decide whether a marker is needed, remember a measure word, and shape four or five pitch contours — all in real time. Attention is finite. When it runs short, the first thing dropped is whatever is least automatic, and at your stage the pitch contour is the least automatic thing you own.

The fix is not to concentrate harder. It is to make everything else cheaper. Practise tones on sentences you have already mastered grammatically — a dialogue you understand completely, a paragraph you have read four times. If you are still assembling the sentence, you are not practising tones; you are practising assembly.

2. The pitch range shrinks across the phrase

In connected Mandarin, the overall pitch register drifts downwards across a long phrase, and everything is squeezed into that narrowing band. A first tone near the end of a sentence is genuinely lower in absolute pitch than a first tone at the start. This is normal and native — it is called downdrift, and it is why the "high level" tone can sound not very high at all by the end of a long clause.

What this means for you: stop measuring your tones against the pitch of an isolated syllable. The contour has to stay the right shape relative to its neighbours, not hit a fixed note. Learners who try to hold every first tone at the same height end up sounding as though they are chanting.

3. Sandhi is not optional and it does not appear in the writing

Three changes happen every time and are written nowhere.

Third tone sandhi. Two third tones in a row: the first becomes rising. 你好 is written nǐ hǎo and said ní hǎo. 很好 → hén hǎo. 我也想 → wó yé xiǎng. In a longer run the grouping decides which syllable keeps the low tone: 我很好 is said wó hén hǎo, and 展览馆 (zhǎnlǎnguǎn, exhibition hall) comes out as zhánlánguǎn.

一 (yī). 一个 yí ge, 一下 yíxià before a fourth tone; 一杯 yì bēi, 一天 yì tiān, 一直 yìzhí elsewhere; 第一 dì-yī when counting.

不 (bù). 不是 bú shì, 不客气 bú kèqi before a fourth tone; 不好 bù hǎo, 不难 bù nán elsewhere.

If you rehearse the written form, you are rehearsing an error. Say what is said.

4. Neutral tones are not weak tones — they are short tones

谢谢 xièxie, 名字 míngzi, 我们 wǒmen, 桌子 zhuōzi, 起来 qǐlai, 的 de, 了 le. The second syllable has no contour of its own; its pitch is decided by what came before, and its real signature is that it is quick and light.

Two failure modes. Learners who ignore neutral tones give every syllable a full contour, which sounds mechanical. Learners who over-apply the idea start swallowing syllables that should carry a tone — 中国 is Zhōngguó, both syllables full.

5. Your native prosody keeps leaking in

French pushes prominence to the end of a rhythmic group and rises on continuation; the result is a sentence-final tone that goes up regardless of what it should be doing. Arabic places stress by syllable weight and lengthens the stressed syllable; the result is one syllable in every phrase given a size that Mandarin does not give it. English shrinks unstressed syllables to a schwa, which flattens their tones outright.

None of this is a defect. It is your first language doing its job in the wrong room. The only cure is hearing yourself do it, which brings us to the method.

The method: build the sentence backwards

Back-chaining is the single most effective technique for sentence-level tones, because it keeps the end of the sentence — the part that decays first — under practice the whole time.

Take 我明天想去图书馆看书。Wǒ míngtiān xiǎng qù túshūguǎn kàn shū.

1. 看书 kàn shū
2. 图书馆看书 túshūguǎn kàn shū
3. 去图书馆看书 qù túshūguǎn kàn shū
4. 想去图书馆看书 xiǎng qù túshūguǎn kàn shū
5. 明天想去图书馆看书 míngtiān xiǎng qù túshūguǎn kàn shū
6. 我明天想去图书馆看书。

Say each stage three times before adding the next. Because the tail is always the part you have said most, it never gets the chance to collapse.

Do the same with 我们下午两点在学校门口见。Wǒmen xiàwǔ liǎng diǎn zài xuéxiào ménkǒu jiàn. — note 两点 liǎng diǎn, a third tone followed by a third tone, so the first rises.

Three more things that work

Hum the sentence before you say it. No consonants, no vowels, just the melody on a single syllable. It isolates pitch from articulation, and you will immediately hear which contour you cannot produce.

Chunk, then join. Speak in the units the language actually groups: 我明天 / 想去图书馆 / 看书. Pausing at chunk boundaries is normal and native. Pausing in the middle of a chunk is what makes you sound broken.

Slow down without stretching. Slowing a sentence is fine; stretching each syllable is not, because a lengthened syllable invites you to add a rise or a dip that should not be there. Keep the syllables short and put the extra time in the gaps.

What to record

Once a week, record three sentences you actually use, and compare each against a native model — not for accent, only for the pitch line. You are checking four specific things: does a second tone actually rise, or does it start too high and stay flat? Does a fourth tone start high enough to have somewhere to fall? Does a non-final third tone stay low instead of dipping and rising? And does the very last syllable of the sentence go up when it should not?

That last one catches more learners than the other three combined.

What good looks like

You will not arrive at a state where every tone in every sentence is textbook-correct; native speakers do not have that either. What you are aiming for is a sentence whose melodic outline is right: the peaks and troughs land where a listener expects them, the sandhi happens automatically, and nothing rises at the end unless you meant it to.

When that clicks, people stop asking you to repeat things — not because your individual tones improved, but because your sentences finally have a shape.