Getting a tone wrong almost never causes a misunderstanding, and that is precisely why tones do not get fixed. Context repairs the error before anybody notices it: nobody at a fruit stall thinks you are asking to buy a horse, so the sentence goes through, no correction arrives, and the wrong tone is quietly reinforced by having worked.
Which means the feedback a learner needs has to be arranged deliberately, because the language will not supply it. Ordinary conversation is the one place tones cannot be diagnosed.
The four, and the three rules that change them
- mā, má, mǎ, mà, mother, hemp, horse, scold. High and level, rising, dipping, falling sharply.
- the third tone dips and rises in isolation, but in ordinary connected speech it is usually just low and flat. Learners who practise the full dip sound slow and unnatural.
- two third tones become second then third. Nǐ hǎo is actually said ní hǎo. The rule is automatic and it applies wherever two meet.
- bù changes before a fourth tone, becoming second: bù plus duì is bú duì.
- yī changes with what follows, taking a fourth tone before first, second and third, and a second tone before a fourth.
- the neutral tone has no mark and is short and light. It appears on grammatical particles and on the second half of many two syllable words, and getting it wrong makes speech sound heavy.
A tone is a shape in your own range, not a note
The four tones are contours rather than pitches, so they move with the speaker, with the sentence and with how loudly you are talking. That is why a learner who has drilled them against a recording can score perfectly and produce nothing usable: they have learned four absolute shapes and the language wants four relative ones.
Three things then act on those contours in connected speech. Tones change each other, so two third tones in a row turn the first into a second, and the words bu and yi shift according to what follows them. Many second syllables lose their tone entirely and become neutral. And at speed the contours compress towards each other, so the difference that was obvious at dictation speed becomes a matter of a few tens of milliseconds.
The consequence is a rule about practice rather than about theory. Tone lives in the phrase, so a tone practised on a single syllable has been practised in the one environment where it does not have to survive anything.
Being understood is what makes them permanent
Here is the stall and it is caused by success. Every wrong tone that gets understood is a repetition of the wrong tone, and in a normal conversation almost all of them are understood, because the topic, the sentence and the other person's willingness are doing the disambiguation.
The damage shows up later and somewhere else. On a telephone, with a stranger, on a subject the listener did not expect, the context is gone and comprehension collapses in both directions, which learners experience as the other person speaking badly. The exit is a judge with no interest in being kind, and the cheapest one is speech recognition: dictate a sentence and read the characters that come back, because a tone error produces a visibly different word rather than a polite nod. Setting up a source of feedback that the situation does not naturally provide is the general move described in the stalls every learner hits, and the way through each.
What a wrong tone actually costs
Where the error is repaired for you, and where it is not
The first three rows are why tones fossilise. The last three are where the bill for that arrives, usually a year later.
Will the context rescue this tone
- YesA shop, a restaurant, a taxiThe topic is known and the sentence is expectedYes
- YesA conversation with your teacherTrained to understand learners, and paid to be encouragingYes
- YesAnything where you can point at itThe strongest repair of all, and the least like a testYes
- NoA telephone callNo face, no context, no repairNo
- NoA name, a number, an addressNothing to predict from, which is why these fail firstNo
- Not checkedWhether the listener noticed and said nothingNobody has counted this and everybody has done itNot checked
The last row is the honest one. A learner cannot tell a sentence that worked from a sentence somebody decoded politely, which is exactly the information that would have been useful, and it is not available from inside a conversation.
Two ways to get told
Chinese dictation on a phone is the free judge. Say a sentence and look at what appears: the recogniser is choosing characters, so a wrong tone shows up as the wrong word on the screen rather than as an unnoticed error. It replaced asking a native speaker whether that sounded right, which returns encouragement, because they understood you and are not going to stop the conversation. Free and already installed, and worth reading in the same direction as everywhere else: a wrong transcription is a real finding and a right one is weak evidence.
Tone pair drills are the other half and they fix what single syllable drills cannot. There are sixteen possible combinations of two tones and a handful of them are genuinely hard for English speakers; practising the pairs rather than the four tones alone puts the sandhi and the compression into the exercise. It replaced drilling four tones on one syllable, which is the standard first week and which trains the only environment tones never occur in.
The mistake: treating the tone as part of the pronunciation
A learner who thinks of tone as an accent feature will treat a wrong tone as a small blemish, in the way an English speaker's Spanish r is a blemish. It is not that. The tone is part of the word, in the same way a consonant is, and a syllable with the wrong tone is a different word rather than the same word said imperfectly.
The practical consequence is about storage rather than about speech. A card carrying a character and a meaning and no tone is an incomplete card, and a learner who has stored several hundred words that way has a vocabulary they cannot say. Tone goes on the card at the moment the word is first met, in the same movement as the meaning, and that is a habit rather than a technique. The same decision, and the same expensive triage when it is made too late, is described in making cards that teach rather than cards that pile up.
What tones need from the rest
Producing a tone correctly and hearing one at speed are separate skills that fail separately, and the second usually lags the first by about a year, which is hearing tone at conversational speed. The Roman spelling records the tone with a mark that is frequently dropped in the wild, and it misleads in several other ways, in the Roman spelling, and where it misleads English readers. The characters carry no tone information at all, so nothing in the writing will remind you, in radicals, components, and how to remember a few thousand. And a wrong tone matters more in Chinese than an accent does elsewhere because the syllable inventory is small, which is two character words, and guessing the meaning of a new one.