Tone Change Rules: Sandhi, the Neutral Tone and Erhua

Why tones change at all

The first thing most courses teach is that Mandarin has four tones, and that each syllable carries one of them. That is true of syllables in isolation. It is not true of connected speech. When tones are produced in a stream, one tone regularly modifies the shape of the next: a dipping tone becomes a rising one before another dipping tone, a high tone loses its height after certain neighbours, and some syllables give up their tone altogether and become short and light. These adjustments are called tone sandhi (变调).

The reason is physical. Producing a full dip-and-rise on every third-tone syllable in a sentence would require the speaker to restart the pitch contour constantly, which is slow and tiring. Languages solve this the way they solve most articulatory problems: they simplify, as long as the simplification does not create ambiguity. The rules below are those simplifications, and they are not optional flourishes — a learner who produces every tone exactly as the dictionary marks it will sound unnatural and, at speed, will be harder to follow, not easier.

For the sound inventory these tones attach to — initials, finals and whole syllables — see the pinyin chart. For characters whose tone or vowel changes when the meaning changes, rather than because of a neighbour, see common polyphonic characters.

The four tones, and how to actually produce them

Tone is described with a five-point pitch scale, where 5 is the top of your comfortable speaking range and 1 the bottom. Every tone is a contour traced on that scale. The descriptions below give both the contour and a physical way to find it.

• First tone (阴平), 55 — high and level. Hold a steady high pitch for the whole syllable, as if humming a single sustained note. It is not a stressed English syllable; the pitch is level from beginning to end. Example: mā (妈, mother). A useful trick for English speakers is to say the syllable on the pitch you would use for the first note of a song, and to keep it there.
• Second tone (阳平), 35 — rising. Start in the middle of your range and rise to the top, like the English question intonation on "What?" Example: má (麻, hemp). Learners who come from languages without contour tones tend to make this too weak; it needs a real upward movement, comparable to a surprised "hm?".
• Third tone (上声), 214 — low, dipping, then rising. Begin at mid-low, drop to the bottom of your range, and come back up part of the way. Example: mǎ (马, horse). This is the tone learners over-produce. In connected speech the full dip-and-rise is often reduced to just the low part, sometimes called a "half third tone". Producing a long, theatrical dip on every third tone makes speech sound stilted. The low point is what carries the meaning; the rise is frequently dropped.
• Fourth tone (去声), 51 — falling sharply. Start high and drop steeply to the bottom, like a firm English command such as "Stop!". Example: mà (骂, to scold). English speakers usually find this the easiest tone, because the contour resembles emphatic English intonation.

The four-way contrast on a single syllable is worth practising as a set: mā má mǎ mà. Then move to real minimal sets that differ only in tone, such as shū (书, book), shú (熟, ripe), shǔ (属, to belong), shù (树, tree). If you can distinguish and produce those four reliably, the tone system is under control; everything that follows is about how tones behave in combination.

The neutral tone (轻声)

A neutral-tone syllable is short, unstressed, and has no fixed pitch of its own — its pitch is inherited from the syllable in front of it. It is written in pinyin without any tone mark. Learners often try to give it a tone anyway, which is one of the most recognisable foreign accents. The neutral tone is not a fifth tone; it is the absence of tone, comparable to the weak syllables in English "button" or "sofa".

The question that matters in practice is which syllables go neutral. There is no complete rule, but the categories cover the large majority of cases:

• Grammatical particles. The sentence-final and aspect particles are essentially always neutral: 的 de, 了 le, 着 zhe, 过 guo, 吗 ma, 呢 ne, 吧 ba, 啊 a. These are among the highest-frequency characters in the language, so getting them right matters far more than getting a rare noun right.
• Suffixes. The noun suffix 子 zi and the word-forming 头 tou are neutral when they follow a stem: 桌子 zhuōzi (table), 房子 fángzi (house), 石头 shítou (stone), 木头 mùtou (wood). But note that 子 and 头 keep their full tones when they are meaningful words in their own right — zǐ as in 子女 (children), tóu as in 头发 (hair). The distinction is grammatical: suffix versus content word.
• The second half of reduplicated words. In kinship terms and in reduplicated verbs and adjectives, the repeated syllable is neutral: 妈妈 māma (mother), 爸爸 bàba (father), 看看 kànkan (to take a look), 慢慢 mànmàn — though in this last case the whole word is often written with a neutral second syllable in speech.
• Positional and directional words used as endings. 上 shang, 里 li, 面 mian and similar words are neutral when they attach to a noun as a locative ending: 桌上 zhuō shang (on the table), 家里 jiā li (at home).
• Common everyday words with an established neutral reading. A large set of high-frequency words simply have a neutral second syllable as standard: 朋友 péngyou (friend), 东西 dōngxi (thing), 明白 míngbai (to understand), 谢谢 xièxie (thank you), 什么 shénme (what).

Where the neutral tone is genuinely difficult is that it can be meaning-bearing. 东西 dōngxi means "thing"; 东西 dōngxī means "east and west". 大意 dàyì (the gist) versus 大意 dàyi (careless). Learners should treat these meaning-bearing cases as vocabulary items to learn individually, not as a rule to apply.

Third-tone sandhi: two and three in a row

The best-known tone rule is this: when two third tones occur in a row, the first one changes to a second tone. The underlying tones are still 3 + 3; what you say is 2 + 3. The classic example is the greeting nǐ hǎo (你好, hello), written with two third tones and pronounced ní hǎo. The same applies across a great many compounds: hěn hǎo (很好, very good) is said hén hǎo, and shuǐ guǒ (水果, fruit) is said shuí guǒ.

Three third tones in a row are where the rule gets interesting, and where the practical answer depends on how the words are grouped. Chinese has a strong tendency to build words in groups of two, and sandhi applies inside each group before it applies between groups. So a three-syllable sequence is normally resolved as if it were 1 + (2 + 3) or (2 + 3) + 1 depending on the structure:

• wǒ hěn hǎo (我很好, I am fine) groups as wǒ + hěn hǎo. The inner pair becomes second + third, then the remaining third follows: wó hén hǎo.
• A sequence like zhǎn lǎn guǎn (展览馆, exhibition hall) groups as zhǎn lǎn + guǎn: zhán lán guǎn.

What learners should take from this is not a decision procedure for every three-syllable phrase. It is that sandhi is applied to word groups, not across the whole sentence, and that in rapid speech a final third tone before a pause is often reduced to a low tone without the rise anyway. Practising nǐ hǎo until it is automatic, and then listening for grouping in recordings, is more productive than memorising exceptions.

The changing tones of 一 and 不

Two extremely common characters have their own sandhi rules, and because they are so frequent, the rules are worth learning properly rather than absorbing by accident.

一 (yī, one)

The dictionary tone of 一 is first tone. In connected speech it changes according to the tone that follows:

• Before a fourth tone, 一 becomes second tone: yí yàng (一样, the same), yí gè (一个, one item), yí cì (一次, once).
• Before first, second or third tone, 一 becomes fourth tone: yì tiān (一天, one day), yì nián (一年, one year), yì qǐ (一起, together).
• When it stands alone, is used as an ordinal, or falls at the end of a phrase, it keeps its first tone: counting aloud yī, èr, sān; dì yī (第一, first); shí yī (十一, eleven).

不 (bù, not)

不 is normally fourth tone, and changes in exactly one environment:

• Before another fourth tone, 不 becomes second tone: bú shì (不是, is not), bú duì (不对, incorrect), bú yào (不要, do not want).
• Before first, second or third tone, it stays fourth tone: bù hǎo (不好, not good), bù lái (不来, is not coming), bù shuō (不说, does not say).
• Before a neutral-tone syllable it also stays fourth tone in practice: bù shì spoken quickly often flattens, but the standard description is fourth tone before neutral.

Note how similar the two rules are: both characters switch to second tone before a fourth tone, and both behave normally elsewhere. That parallel is the reason these two are usually taught together. Note too that the underlying tone is what dictionaries record and what you should write when making study materials — the changed tone is a pronunciation effect, not a spelling change.

Erhua: the r-colouring

In northern Mandarin, and especially in Beijing speech, a large number of syllables take an -r ending that fuses with the final. This is erhua (儿化), written with the character 儿 or as a plain r on the syllable. It is not simply "add an r": when erhua attaches, the final itself changes shape. a becomes a more open r-coloured vowel; i and n are lost and the vowel is nasalised and then rhotacised, so wánr (玩儿, to play around) is not "wan-r" but a single rhotacised syllable.

Erhua does real work. It forms nouns from verbs and adjectives — gài (盖, to cover) becomes gàir (盖儿, a lid) — it adds a sense of smallness or affection, and it distinguishes meanings: tóu (头, head) versus tóur (头儿, boss).

For learners, the honest advice is that erhua is regionally variable and largely optional for a beginner. Southern varieties of Mandarin use far less of it, many textbooks in mainland China mark it but do not require it in production, and it is not needed for comprehension of standard written text. Where it matters is listening: if you are preparing for a proficiency exam that uses northern-standard audio, or watching Beijing-set television, recognising erhua is essential. Producing it is a later refinement. Do learn the meanings of the most common erhua pairs, because they are meaning-bearing (nǎr, 哪儿, where; yìdiǎnr, 一点儿, a little; zhèr, 这儿, here), but do not spend weeks drilling the phonetics.

What to practise, and what you can safely ignore

Tone rules are easy to over-study. A learner who tries to master every sandhi pattern in the first month will spend their time on effects that are largely automatic once the individual tones are solid. The following split is the one that pays off.

Practise from the start. The four tones as isolated contours, with minimal pairs, until the difference between mā má mǎ mà is obvious to you. The third-tone rule inside two-syllable words, because it applies constantly and the pairs are easy to find. The neutral tone on grammatical particles — 的 了 着 过 吗 呢 吧 — because these characters appear in almost every sentence you will ever read, and getting them wrong is conspicuous. The 不 rule, which has exactly one condition and therefore costs almost nothing to learn.

Practise next, once the basics are stable. The 一 rule, which has three cases but is mechanical. The neutral tone on suffixes and reduplicated words, which is mostly vocabulary learning. The third tone before a pause, which in careful speech is often realised as a low tone without a rise.

Deprioritise at beginner level. The full three-syllable sandhi calculus, which resolves itself once your two-syllable groupings are secure. The precise pitch inherited by a neutral tone after each of the four tones — real, but a detail of advanced accuracy rather than communication. And the production of erhua, which is a stylistic feature of one regional standard, not a requirement of the language.

Two habits matter more than any of this. First, learn every new word with its tone from the beginning; retrofitting tones onto a vocabulary you learned without them is far harder than learning them correctly once. Second, read aloud from annotated text. That is exactly what the online pinyin tool produces: paste any Chinese passage and it returns the text with pinyin and tone marks above each character, including the changed neutral readings of particles, with polyphonic characters flagged so you can select the right reading for the context. Because the underlying tone is what the tool marks, it doubles as a way to check your own pronunciation against the standard rather than against the sandhi form.

For the full sound inventory behind these tones, go back to the pinyin chart; for the characters whose reading changes with meaning rather than with a neighbour, see common polyphonic characters. If your tones are reasonably secure and the persistent problem is the consonants around them — the retroflex series, the nasal finals, ü after j, q, x — those contrasts are worked through one by one in commonly mispronounced sounds. And for the characters you will actually meet most often, with their default readings, see 500 essential characters. The complete series is listed in the Tutorials. To hear all of this in practice rather than in the abstract, paste a passage into the online pinyin tool and read the annotated result aloud: the tones are marked above every character, the particles come back with their neutral readings, and the sandhi patterns you have just read about are the ones you will actually be producing.