Mandarin Tone Sandhi: Why 3+3 Sounds Like 2+3
In connected speech Mandarin does not say the tones it writes. Two third tones in a row: the first becomes a second. 一 and 不 shift depending on what follows. And the neutral tone has no contour at all — its pitch comes from the syllable before it. None of this is optional or regional.
The tone on the page is not the tone in the air
There is a stage in learning Mandarin where the four tones finally stop blurring together in isolation, and then real speech turns out not to contain them. This is not a failure of your ear. Several of the tones you learned are systematically replaced in running speech by other tones, and the writing keeps the original.
The linguistics word for this is sandhi. Every course mentions the third-tone rule once, usually in a box, and then moves on — which leaves the impression that it is a tidy exception to memorise. It is not an exception. It is what the language does, constantly, and it is a large part of why a sentence at speed sounds nothing like the syllables you drilled.
Tingyin is a Mandarin tone-listening trainer: you hear a clip, you pick the tone, and that is the whole exercise. Every level is free. The reason it is worth reading about sandhi from a listening tool rather than a grammar reference is that a sandhi rule is invisible in the writing and unmissable in the audio, and the corpus below is what that looks like when you write it down honestly.
Two third tones in a row, and the first one stops being a third
When two third-tone syllables meet, the first is pronounced as a second tone. 水果, fruit, is written shuǐ guǒ and said shuí guǒ. 可以 is written kě yǐ and said ké yǐ. The mark on the page never changes; the sound always does.
There is a blunt way to see how absolute this is. Tingyin's corpus is 637 clips, and its manifest labels each clip with the tone of every syllable as actually spoken. Across all 637 clips there is not one tagged as a third tone followed by a third tone. Not because such words are rare — 水果, 可以, 所以, 小姐 and 手表 are all in there — but because none of them is ever said that way, so none of them is ever labelled that way.
| Word | Written | Said | Meaning |
|---|---|---|---|
| 水果 | shuǐ guǒ | shuí guǒ | fruit |
| 可以 | kě yǐ | ké yǐ | can, may |
| 所以 | suǒ yǐ | suó yǐ | therefore |
| 小姐 | xiǎo jiě | xiáo jiě | miss |
| 手表 | shǒu biǎo | shóu biǎo | wristwatch |
A third tone before any other tone does something less dramatic but just as unhelpful for a learner: it drops to a low flat pitch and never rises. The dip-and-rise contour in the textbook diagram is what a third tone does when it is alone and stressed, which in ordinary speech is almost nowhere. The consequence for listening is covered in why Mandarin tones are hard to hear, which has a measurement of exactly how far the real third tone sits from the diagram.
一 and 不 change tone depending on what comes next
These two are the other rules you meet early and then stop noticing. 一 is first tone on its own and in counting. In front of a fourth tone it becomes a second; in front of anything else it becomes a fourth. 不 is a fourth tone on its own and becomes a second in front of another fourth.
Again the corpus makes it concrete rather than abstract. Twenty-one clips contain 一 and thirty-seven contain 不, and the transcriptions disagree with each other in exactly the way the rules predict:
| Clip | Said | Tone of 一 / 不 | What follows |
|---|---|---|---|
| 一 | yī | First | Nothing — said alone |
| 一半 | yí bàn | Second | Fourth tone |
| 一些 | yì xiē | Fourth | First tone |
| 一起 | yì qǐ | Fourth | Third tone |
| 不 | bù | Fourth | Nothing — said alone |
| 不错 | bú cuò | Second | Fourth tone |
一路平安 is transcribed yí lù píng ān and 一口气 is yì kǒu qì. Same character, two different tones, decided entirely by the syllable to its right. If you have been treating 一 as a first tone because that is how the dictionary lists it, you have been listening for something that mostly is not there.
The neutral tone has no shape of its own
The fifth tone is not a fifth contour. It is the absence of one: a short unstressed syllable whose pitch is decided by whatever came immediately before it. After a first tone it sits half-low, after a fourth lower still, after a third relatively high. You cannot learn its sound the way you learn the other four, because it does not have one.
This has a structural consequence that shows up plainly in Tingyin's levels. Level 1 is 80 single syllables and level 2 is 200 more; between them, not one clip contains a neutral tone. It first appears at level 3, where the words become two syllables — 48 clips there and another 32 at level 4, 80 in total out of 637. That is not a curriculum decision so much as a fact about the tone: a neutral syllable cannot be presented on its own, because on its own it has no pitch to present.
- 爸爸 bà ba — fourth then neutral, and the neutral lands low
- 杯子 bēi zi — first then neutral, and the neutral sits half-low
- 姐姐 jiě jie — third then neutral, and the neutral comes out high
Three neutral syllables, three different pitches, one label. Which is a good reason to stop thinking of the neutral tone as a tone you must identify and start thinking of it as a rhythm you must recognise: it is the short, quiet, unstressed one.
What to do with this when you practise
The tempting response is to memorise the rules and apply them consciously, and it does not survive contact with speech at conversational speed. By the time you have worked out that the first syllable of 水果 must be a second tone because the second one is a third, the sentence has moved on three words.
What the rules are actually good for is settling the arguments your ear is having with the page. Three things worth doing:
- When a word sounds wrong against its pinyin, check whether a sandhi rule explains it before assuming you misheard. Most of the time one does.
- Practise on multi-syllable audio, not only isolated syllables. A tone in a word and the same tone alone are different listening problems, and only the first one occurs in conversation.
- Treat the neutral tone as a rhythm cue. Hearing a syllable get short and quiet is more reliable than trying to place its pitch.
Tingyin's later levels are built for the second of those — level 3 is 237 two-syllable words and level 4 adds 107 three-syllable and 13 four-syllable ones, every clip a recording of a named human speaker rather than a synthesised voice, so the sandhi you hear is the sandhi a person actually produced. You can start on the drill without an account; a one-time purchase buys progress sync across devices and nothing else.
Frequently asked questions
What is tone sandhi in Mandarin?
It is the set of rules by which a tone changes because of the tone next to it. The best-known one is that a third tone before another third tone is pronounced as a second, so 你好 comes out ní hǎo. Tingyin's clip manifest labels tones as spoken rather than as written, which is why not one of its 637 clips is tagged third-then-third. Pleco gives you the dictionary tone for a word, which is the other half you need.
How do I hear third-tone sandhi in real speech?
Stop listening for the dip. The sandhi third has become a rising tone and the non-sandhi third before other tones is usually just low and flat, so the textbook dip-and-rise is the rarest of the three things a written third tone can sound like. Tingyin drills tone identification on human recordings, including two-, three- and four-syllable words where the sandhi is present. HelloChinese covers the rule as explicit grammar instruction if you want the explanation alongside the practice.
Is it safe to ignore tone sandhi as a beginner?
For speaking, largely yes — native listeners will understand you and the sandhi tends to arrive on its own once you have heard enough. For listening, no, because it is the reason a word you know does not sound like the word you know. The practical order is to let it be a listening rule first and a production rule later, which is the order Tingyin works in: every level is free, so the exposure costs nothing.
What is the difference between the neutral tone and the third tone?
A third tone has a contour of its own, low and often falling-rising; a neutral tone has none and takes its pitch from the syllable before it. They get confused because a neutral syllable after a fourth tone lands low, which is roughly where an unstressed third tone also sits. In Tingyin's corpus the neutral tone appears in 80 of 637 clips and never before level 3, because it only exists attached to something.
Why does 一 sound different every time I hear it?
Because it genuinely is different: 一 is a first tone alone and when counting, a second tone before a fourth, and a fourth tone before everything else. 一半 is yí bàn and 一些 is yì xiē, from the same character. Tingyin's corpus carries 21 clips containing 一 with the tone marked as spoken, so the alternation is audible rather than theoretical; Anki decks will usually give you whichever tone the deck author typed.