Every learner has this moment: you confidently read 银行 as , and someone gently says — bank. But you learned that 行 was ! You did learn it right; the character simply has more than one reading. These are called polyphonic characters, 多音字 (), literally "many-sound characters", and Mandarin has a few hundred of them — fortunately, a few dozen account for most of what you will actually meet.
What polyphones are and why they exist
A polyphone is a single written character with two or more standard pronunciations, each tied to a different meaning or use. They are not errors or dialect leftovers; they are fossilized history. Written Chinese is several thousand years old, and the same character was often borrowed to write different words with related or shifted meanings. Over centuries the words kept their different sounds while sharing one written form — the opposite of what happened in English spelling, where sounds shifted but letters froze.
Sometimes the two readings split along grammatical lines: one reading for the noun, another for the verb, as in 长, which is as an adjective (long) but as a verb (to grow, or elder). Sometimes they reflect different source words entirely, as with 行: (to walk, to do) and (row, profession, and — via "row of shops" — bank). And in a handful of cases the difference survives mostly in fixed words, like 石, which is in 石头 (stone) but keeps an older reading as the measure word for stone weight. For a learner the message is the same in every case: the reading belongs to the word, not to the character.
The most common polyphones, with word pairs
Here are the polyphones worth knowing cold, each with two words that force the two readings. Cover the pinyin and test yourself — these pairs are chosen because both words are common, so you cannot avoid either reading.
行 — 行走 (to walk) / 银行 (bank)
长 — 很长 (very long) / 长大 (to grow up)
乐 — 快乐 (happy) / 音乐 (music)
重 — 很重 (heavy) / 重说 (to say again)
发 — 出发 (to set off) / 头发 (hair)
干 — 干净 (clean) / 干活 (to work)
给 — 给我 (give me) / 补给 (to supply)
还 — 还有 (still, also) / 还书 (to return a book)
觉 — 觉得 (to feel) / 睡觉 (to sleep)
角 — 牛角 (horn) / 角色 (role)
都 — 都是 (all are) / 首都 (capital city)
教 — 教书 (to teach) / 教师 (teacher)
没 — 没有 (not have) / 淹没 (to submerge)
中 — 中间 (middle) / 中奖 (to win a prize)
好 — 好人 (good person) / 爱好 (hobby)
难 — 很难 (difficult) / 灾难 (disaster)
地 — 慢慢地 (slowly) / 土地 (land)
得 — 得到 (to obtain) / 跑得快 (runs fast) / 我得走 (I must go)
Notice how often the second reading lives in just one or two fixed words: for hair, for sleep. That is typical, and it is good news — you do not need a general rule for when 觉 is ; you need to know the word 睡觉. The dictionary page for each character, like the page for 行, lists every reading with its example words, which is exactly the list you should memorize instead of the character alone.
How to guess the right reading from context
When you meet a polyphone in the wild, three signals usually decide the reading. First, the partner characters: 行 before 走, 人, or 为 means the walking verb and reads ; 行 after 银 (银行, bank) or 同 (同行, same profession) belongs to the profession word and reads . The point is that the word around it, not your preference, decides. Second, the grammar slot: if the character is doing verb work in one slot and noun or adjective work in another, the readings often split the same way ( grows, is long). Third, frequency asymmetry: in real text, one reading usually dominates — covers most uses of 行, and covers most of 还 — so when truly stuck, the common reading is the better bet, and the rare reading is the one worth checking in a dictionary.
Polyphones hiding in names and places
Surnames are a polyphone trap with real social stakes, because a small set of family-name characters keep archaic readings that survive nowhere else. The surname 单 is read , not (single); 解 is , not (to untie); 曾 is , not (once); and 查 is , not (to check). Place names do the same: the Anhui city 六安 is , with the number six reading everywhere else. The practical rule: when a character appears in a personal or place name and your reading produces a word that means nothing, suspect a name reading and check — dictionaries list them, and getting a new colleague's surname right on the first try is worth the thirty seconds.
Why character-by-character lookup fails
The natural beginner workflow — look up each character's pronunciation individually, then glue the results together — produces wrong pinyin for polyphones almost immediately, and not just occasionally: some of the highest frequency words in the language are polyphone pairs. Worse, the errors are self-reinforcing, because the wrong reading is what your dictionary page for 行 alone seems to license. The correct workflow reads words: identify the word, then read the word. This is exactly how ToPinyin's converter works — it matches your text against the dictionary word by word and only falls back to single characters for unknown stretches, which is why its output can be trusted for real sentences in a way a per-character table cannot. For a single stubborn case, search the character in the hanzi dictionary and compare the readings side by side with their words.
Polyphones interact with tones too, of course — the two readings of a character usually differ in tone as well as letters, as and show. Learn the whole syllable, tone included, as one package per word, and polyphones stop being traps and become interesting etymology.
A final encouragement for the discouraged: the polyphone problem shrinks steadily as your vocabulary grows. Every new word you learn carries its characters with their readings already resolved, so the polyphone set you have actually memorized grows faster than the list of tricky characters you have merely seen. Beginners meet polyphones at every page because their vocabulary is small; intermediate learners meet them mainly in names, rare words, and classical set phrases — and by then the habit of reading words instead of characters is automatic. The converter demonstrates the same trajectory: the longer and more real the sentence, the more context it has, and the more reliably every ambiguous character settles into its correct reading.