Mandarin Tone Pairs: Why Two Syllables Are the Wall
Four tones plus a neutral do not give sixteen two-syllable combinations. They give twenty, because a neutral tone can end a word but never start one. Nineteen of those twenty turn up in ordinary speech; the twentieth, third-then-third, is never heard, because Mandarin rewrites it into second-then-third before it reaches your ear. That is why a learner who can name all four tones in isolation still loses words at speaking speed: the four contours were never the hard part.
Twenty pairs, not sixteen
The arithmetic people expect is four times four. The arithmetic that matches speech is four times five: the first syllable of a two-syllable word carries one of the four tones, and the second carries one of the four or the neutral tone. Twenty combinations.
The asymmetry is the interesting part. A neutral syllable has no contour of its own - it borrows its pitch from whatever came immediately before it. A word cannot open with a borrowed pitch when there is nothing yet to borrow from, so the neutral tone never appears in first position. Four tones can start a word; five can finish one.
This is not a tidy way of putting it. It is what a corpus of recordings looks like when you sort it. Tingyin's two-syllable material is 237 clips, and every one of them has a first syllable in the range one to four. Not a single clip in the whole 637 opens on a neutral tone, because no such word exists to record.
What arrives with the second syllable
Three things happen at the join, and none of them is visible when you drill syllables one at a time.
- Tones get rewritten. Mandarin substitutes tones at boundaries. The famous case is a third tone before another third tone, which comes out as a second, but 不 and 一 both shift as well depending on what follows them.
- The neutral tone turns up. It has no shape of its own, so you cannot learn it as a fifth contour; you can only learn it as four different landings.
- You have to find the seam. Two syllables run together, and deciding where the first ends is a separate act from identifying either tone.
Tingyin is a Mandarin tone-listening trainer, and only that: you hear a clip and pick the tone, with no character writing, no grammar and no vocabulary lists. Every clip it plays is a recording of a person rather than a synthesised voice, and each one carries its source, speaker, licence and a content hash - 637 clips from eight named speakers, 622 of them under CC BY-SA 4.0 and 15 under CC0. Level 1 runs without an account at all, and level 2 is free once you sign in.
The grid, counted from a real corpus
Below is Tingyin's two-syllable set laid out as a grid: rows are the tone of the first syllable, columns the tone of the second. The numbers are clips.
| First ↓ / second → | 1st | 2nd | 3rd | 4th | neutral |
|---|---|---|---|---|---|
| 1st | 12 | 12 | 12 | 12 | 12 |
| 2nd | 12 | 12 | 20 | 13 | 12 |
| 3rd | 12 | 12 | 0 | 12 | 12 |
| 4th | 12 | 12 | 12 | 12 | 12 |
Seventeen of the nineteen occupied cells hold exactly twelve clips. The three cells that break the pattern - one empty, two overfull - are the same phenomenon seen from different angles.
The empty cell, and where its words went
Third-then-third is zero. Not rare - zero, across all 637 clips at every length. The reason is that Mandarin does not permit the sequence: when a third tone is followed by another third tone, the first is spoken as a second. 你好 is written third-third and said second-third, and the same is true of 水果, 可以, 所以 and every other word of that shape.
Tingyin labels each syllable with the tone as spoken rather than as written, so those words are not missing from the corpus - they are filed one row up. That is what the 20 in the second-then-third cell is: all twenty are words written third-then-third, which is also why that cell is the fullest in the grid. The 13 in second-then-fourth is the smaller version of the same story, and every one of those thirteen begins with 不 or 一, the two words that shift upward before a fourth tone.
For a learner this is better news than it sounds. It means there are nineteen pairs to train, not twenty, and it means the pair that looks most frightening on paper is one you will never have to identify. What you do have to identify is second-then-third, which is now carrying two jobs at once - the full set of substitution rules covers which words end up there and why.
The neutral tone is four sounds, not one
Because a neutral syllable takes its pitch from the syllable in front of it, its actual landing point is different after each of the four tones. Treating it as a single thing you learn once is the reason it keeps slipping.
| Pair | Where the neutral lands | Example |
|---|---|---|
| 1st + neutral | Steps down from a high level start | 杯子 bēi zi, cup |
| 2nd + neutral | Lands mid, after the rise has finished | 房子 fáng zi, house |
| 3rd + neutral | Comes off the bottom of the dip, so it sounds higher | 姐姐 jiě jie, older sister |
| 4th + neutral | Lowest of the four, following the fall down | 爸爸 bà ba, father |
Eighty of Tingyin's clips contain a neutral tone and all of them are at least two syllables long, which is the only way a neutral tone can be recorded at all. If you have been drilling single syllables and feel you have never really met the neutral tone, that is because you have not. What the neutral tone actually is goes through it properly.
Which pairs to drill first
Not all nineteen deserve equal time, and the ones worth starting on are not the ones that look hardest.
- Second-then-third, because it is doing double duty. Every word written third-third arrives here, so it is both the most frequent pair and the one whose spelling will not match what you hear.
- The four neutral pairs, as a set rather than one at a time. The point is the contrast between them; drilled separately they teach you four unrelated facts instead of one rule.
- Third-then-first and third-then-fourth, where the third tone is not phrase-final and so never completes its dip. It is spoken low and short, which is not what a third tone sounds like when you learned it in isolation.
What does not work is drilling the pair you find hardest until it yields. Tone perception moves on contrast - hearing two things that differ in one dimension, close together - so a session that alternates second-then-third against third-then-third-as-written will do more than an hour of the same cell. Why some people cannot hear the difference at first covers the mechanism.
Where the pairs actually live
In Tingyin the split is clean and it is worth being straight about: levels 1 and 2 are single syllables, 280 clips of them, and they are the free part. Every two-syllable word is level 3, and the longer material is level 4, so the entire tone-pair grid sits behind the paywall. On the web, the only place to pay while the native apps are unpublished, that is $2.49 a week, $5.99 a month or $15.99 once for lifetime access; paying also adds progress sync and offline downloads.
If you would rather not pay, the pairs can be drilled elsewhere with more setup. An Anki deck of two-syllable words with audio will do it, provided you tag each card with the spoken tones rather than the written ones - otherwise you will train yourself on a third-third that does not exist. Pleco will pronounce any word you look up, which makes it a decent way to check a pair you are unsure of, though it is a dictionary rather than a drill. The ladder from syllable to sentence sets out what comes before and after this rung.
Frequently asked questions
What is the best way to practise Mandarin tone pairs?
Drill them as contrasts rather than one cell at a time, because tone perception improves when two similar things are heard close together and barely improves from repetition of one. Tingyin's two-syllable level holds 237 clips spread across the nineteen combinations that occur, all of them human recordings rather than synthesis. Anki with a two-syllable deck works too and costs nothing, but you have to source the audio and tag the tones as spoken yourself, which is where most home-made decks go wrong.
How do I stop confusing the second and third tone inside a word?
Listen to where the syllable ends rather than where it starts, because the two differ most at the finish: a second tone is still climbing, a third has already bottomed out. Inside a pair this is harder than in isolation, since a third tone that is not the last syllable is spoken low and short without completing its dip. Practising the two against each other in the same session is what separates them; practising them on different days does not.
Is it safe to learn tone pairs from a text-to-speech app?
For the shape of a pair, mostly; for the neutral tone, no. Synthesis tends to give a neutral syllable a tidy fixed pitch, when the whole point is that it lands somewhere different after each of the four tones, so a learner trained on it hears something real speakers do not produce. Tingyin ships only recordings of people, from eight named speakers with the licence and a content hash on every file. If you do use a synthetic voice, use it for the four full tones and find human audio for anything with a neutral syllable in it.
What is the difference between tone pairs and tone sandhi?
A tone pair is a combination you might hear; tone sandhi is the rule that decides which combinations are possible in the first place. Sandhi is why one of the twenty pairs is empty - third followed by third is rewritten to second followed by third before anyone says it - so the pairs are the surface and sandhi is the machinery underneath. You can train the pairs without knowing the rules, but you cannot make sense of why a word looks one way and sounds another.
Why do I keep mishearing words I am sure I know?
Usually because you learned them one syllable at a time and the word does not behave like the sum of its syllables. The tones shift at the join, the neutral syllable moves depending on what came before it, and the seam between the two is something you have to find in real time. It is not a memory problem and it does not mean the earlier practice was wasted - it means the two-syllable rung is a genuinely different listening task, and almost nobody is told that before they hit it.