A polyphone, 多音字 (duōyīnzì), is a Chinese character with more than one reading, and mainland China and Taiwan rule on those readings in separate documents. We compared two lists across the 3,360 level-one characters, the commonest tier of the mainland's official list, that map to one Taiwan form. The 2013 dictionary keyed to that list makes 392 of them polyphones. Taiwan's 2012 draft table, which is not the table in force, makes 419. In all, 273 are polyphones in both, 69.6% of the first list and 65.2% of the second, and 218 of those have identical readings. Taiwan's 1999 table, in force, keeps 957 polyphones of 4,253 listed; we sampled it and did not cross-tabulate it.
Across published counts the polyphone rate moves more than fivefold
The highest rate below is 5.6 times the lowest, on our arithmetic, because each row counts a different base.
| Source | Base | Polyphones | Share |
|---|---|---|---|
| 2013 dictionary, our count | 3,500 level-one characters | 430 | 12.3% |
| Park and Lee, 2020 | 17,720 characters in Chinese Wikipedia | 762 | 4.30% |
| Taiwan 2012 draft, before review | 5,266 reviewed | 1,263 | 24.0% |
| Taiwan 2012 draft, after review | 5,266 reviewed | 580 | 11.0% |
| Taiwan 1999 table, in force | 4,253 listed | 957 | No rate |
The 1999 row has no rate because its 4,253 are candidates, never all characters. The dictionary is a published book whose readings Unicode transcribes in the Unihan database; our guide to homophones per Mandarin syllable also uses the 430.
Inside Taiwan's table in force 957 characters stay polyphones
Taiwan's Ministry of Education announced the 《國語一字多音審訂表》 (審訂表 means review table) in March 1999. Its status page, from 2013, calls that version 「現行規範」, the current norm, and says textbook review and the language contests run by Taiwan's Ministry of Education follow it until a revision is formally announced. That ministry's old file address is gone, so this link and the next are archived copies.
The 2012 draft reviewed a different list and merged harder, so 卡 has kǎ (ㄎㄚˇ) and qiǎ (ㄑㄧㄚˇ) in the 1999 table and one reading in the draft. The 119 characters that are polyphones only on the mainland side, and the 146 only in the draft, therefore describe the draft and not the rule. One gap does hold in the 1999 table: the dictionary gives 法 only fǎ, and the table adds ㄈㄚˊ (fá), restricted to 法子.
Under the mainland's 1985 table only words with a variant reading get a ruling
The mainland's 《普通话异读词审音表》 is another kind of document. Three bodies issued it in 1985; Peking University's Sun Yuwen, in a September 2026 interview, calls it 「异读裁定的国家标准」, the national standard for ruling on variant readings. We read its text in the left column of the attachment to the 2016 revision notice from the mainland's Ministry of Education; both links are archived copies, because the ministry's own pages returned 404 to us on 1 November 2026.
Its subject is the 异读词, a word with variant readings; the 2016 notes say a word without one is not audited, and neutral tone is not ruled on. Its 847 headed rows, on our count, are therefore no list of polyphones and cannot be set beside Taiwan's counts. Of those rows, 586 carry 统读, one reading for every word using the character. The revision itself is covered in our guide to pronunciation changes.
Against Taiwan's draft 143 single-reading rulings from the mainland read differently
A 统读 ruling can be checked character by character, so this is the cleanest cross-strait test we could run. Of the 586, 84 fall outside the draft's 5,266; among the remaining 502, 143 (28.5%) do not match the 2012 draft. Of those, 109 have a different single reading, 74 of them by tone alone; 31 keep the mainland reading plus another; 3 have several readings without it.
Mainland 危 is wēi, and the 1999 table lists ㄨㄟˊ (wéi) alone. Mainland 期 is qī, and the 1999 table has ㄑㄧˊ (qí) and ㄐㄧ (jī). In the 2012 draft alone, unchecked against 1999, 垃圾 is ㄌㄜˋ ㄙㄜˋ (lè sè) against lā jī.
The 143 is a draft figure and we give no in-force one. The mainland rules 骑 qí, the 1999 table has ㄑㄧˊ alone, and only the draft adds ㄐㄧˋ (jì), so that character counts here without being a difference in force. A dictionary set against a table that merges on purpose also measures editorial policy, and neither says how people talk.
Within a frequency corpus the commonest reading holds over 96% of the time
Polyphones look scarce by character and common by appearance. Park and Lee's g2pM paper of 2020 made that contrast before us: 762 of 17,720 characters in Chinese Wikipedia, 4.30%, were polyphones in the CC-CEDICT dictionary, yet they made up 18.49% of 363 million tokens, or occurrences in running text.
We tested the commonest reading. Unihan's kHanyuPinlu field carries reading counts based in part on a 1986 frequency dictionary; with Traditional duplicates removed we have 1,722,455 tokens on 2,848 characters. That describes the corpus, and cannot stand in for modern text.
| Counted as a polyphone | Share of all tokens | All tokens off commonest reading |
|---|---|---|
| Any two readings | 38.9% | 3.68% |
| Neutral tone merged | 21.0% | 1.51% |
| Tones ignored | 13.5% | 0.79% |
On the broadest definition the commonest reading holds for 96.32% of all tokens, by our arithmetic; on tokens of polyphones alone it holds for about 90.5%. The 88 level-one characters with no reading in common between the dictionary and the 2012 draft, 究 and 击 among them, carry 0.4% of tokens, a count that misses words where both sides hold a reading and use it differently.
For a pinyin converter each reading comes from a word list
A converter resolves polyphones by matching words, so its answer depends on the list it holds; our guide to Zhuyin readings and Taiwan's dictionary measured that from the Taiwan side.
Until 1 November 2026 our polyphone lookup showed the limit itself: its Sample read 行长 as háng cháng, 的 returned "not found", and only 50 characters were covered. The rebuilt version names its source and covers 597 characters from Unihan. Scan 行长 and it gives háng zhǎng; look up 的 and it gives de, dī, dí and dì. Its readings are mainland-sourced: scan 星期 and 期 comes out qī, which the 1999 table's entry does not have.
The 88 counts characters with no shared reading, the 143 counts rulings, and our Zhuyin guide's "about 200" counts characters where one software library's reading is absent from a school dictionary.
Beyond the tables in force we found no finished revision on either side
On the mainland, a national-standard plan of the same name as the 1985 table, 20180782-T-360, was issued on 18 July 2018 as a recommended standard with a 12-month term. On 1 November 2026 the standards portal listed it as 正在批准, being approved. The page does not say the plan is the 2016 draft, and we found no promulgation.
Taiwan's Ministry of Education passed the 2012 draft's results to the National Academy for Educational Research in 2019. A county government letter of March 2020 says the revision was finished and under the academy's review; we have not read it. The academy's survey report of January 2025 still describes the 1999 table as the basis for textbooks and teaching. We found no final replacement by 1 November 2026, and neither record gives a date for one.
