A Chinese word puzzle hands you a few characters and asks which real words they can build. Its pieces are whole syllables, each usually carrying meaning, and that changes the odds sharply. We drew 10,000 random bags of five characters from a list of 104,396 two- to four-character words built from CC-CEDICT, an open Chinese-English dictionary, in its release of 24 May 2026. A little over half of the bags, 54.9%, made no word at all. Not one of the 10,000 made a four-character word.
Random bags of up to five characters form no word in most draws
The draw was weighted by frequency, so a character found in many words was more likely to be picked, which flatters the result. Drawn evenly from the list's 6,361 distinct characters, 96.8% of five-character bags made nothing. The table comes from 1,000 weighted bags per size on a fixed seed; the share making any word passes half only at six characters.
| Characters in the bag | Bags making at least one word | Median words per bag |
|---|---|---|
| 2 | 7.6% | 0 |
| 3 | 16.9% | 0 |
| 4 | 33.2% | 0 |
| 5 | 47.2% | 0 |
| 6 | 58.4% | 1 |
| 8 | 76.8% | 2 |
| 10 | 89.2% | 3 |
| 20 | 99.9% | 12 |
Four-character words stay out of reach. In runs of 10,000 bags, no bag of ten characters made one; at twenty characters, 21 bags did, and none of those words was an idiom.
The search behind these counts is the 重组 (chóngzǔ, rearrange) mode of our Chinese word unscrambler, which returns every word that can be built from the characters typed, using each no more often than it was typed. Enter 天地人和气 and it returns eight two-character words and nothing longer; among them are 天气 tiānqì, which CC-CEDICT glosses as "weather", and the reversed pair 人气 rénqì, "popularity", and 气人 qìrén, "to anger". A setter chooses characters on purpose, so a real puzzle is built to beat a random bag.
About 1,500 character pairs form a word in either order
Where two characters do combine, some combine both ways. Among the list's 60,865 two-character words, 1,775 pairs read as a word in either order. Set aside pairs where either side is a proper noun, counting a word as a name when every one of its CC-CEDICT readings is capitalised, and 1,547 remain. Their 3,094 words are about 5% of all two-character words, on our arithmetic. In 55 of them one side is only a cross-reference to another entry.
The meaning may or may not survive the swap. CC-CEDICT glosses 蜜蜂 mìfēng as "bee" and 蜂蜜 fēngmì as "honey"; 牛奶 niúnǎi is "cow's milk" and 奶牛 nǎiniú a "milk cow".
Linguists have studied these pairs, under names such as 同素逆序词 (tóngsù nìxù cí, words of the same morphemes in reverse order). Yuliya Korbashova and Olga Te, writing in Theoretical and Applied Linguistics in 2022, worked through the 2016 edition of the 《现代汉语词典》 (Xiàndài Hànyǔ Cídiǎn, the Dictionary of Modern Chinese). They found 260 pairs, and on each feature they compared, meaning among them, pairs that matched outnumbered pairs that differed.
Our 1,547 is about six times their 260. The gap comes from the dictionaries and the rules: CC-CEDICT holds more than 120,000 entries, rare and classical words included, and we admit any pair whose two orders are both headwords. Neither figure is the number of reversible words in Chinese.
Shuffled idioms make another idiom in 76 groups
In the list's 19,149 four-character words, 230 groups of two or more share exactly the same characters, 469 words in all. In 76 of those groups at least two members carry CC-CEDICT's idiom tag, 156 idioms between them.
Some swaps keep the sense. CC-CEDICT glosses 一诺千金 yī nuò qiān jīn as "a promise worth a thousand pieces of gold" and 千金一诺 qiān jīn yī nuò as "a promise worth one thousand in gold". Others turn it inside out. 事半功倍 shì bàn gōng bèi is glossed "to achieve twice the result with half the effort", and 事倍功半 shì bèi gōng bàn "to expend twice the effort for half the result". The largest groups have four members, such as 扶危济困, 扶困济危, 济危扶困 and 济困扶危, two of which the dictionary glosses as "to help those in distress". A puzzle that scrambles an idiom and expects one answer can therefore have two, or four.
Each named list yields its own idiom total
Whether a phrase is an idiom depends on whose list you ask. Taiwan's Ministry of Education, in the 2021 revision of its 《重編國語辭典修訂本》, defines 成語 chéngyǔ as 「漢語中有典源，具多層表義功能的固定語。可作為句子的成分，形式不一，以四言為主。」 In our rendering: a fixed expression with a literary source and layered meaning, varied in form but mostly of four characters. The proverbs guide uses the same entry.
CC-CEDICT publishes no definition. Its editors tag entries "(idiom)", and on the project's labels page the meaning cell beside that label is empty. In the 24 May 2026 release, 5,066 two- to four-character headwords carry the tag, all but 21 of them four characters long.
The two part company on the ministry's own examples. Of 矛盾, 綿裡針, 勢如破竹 and 篳路藍縷, CC-CEDICT tags the last two. It holds 矛盾 máodùn untagged, as "contradiction", and does not hold 绵里针 at all.
Totals move with the edition, too: CC-CEDICT's release of 3 November 2026 has 125,360 entries, against 124,945 in May. The chengyu jielong guide works with a list of 29,502.
A single headword can make two different words, as 便宜 does
CC-CEDICT gives 便宜 as biànyí, "convenient", and as piányi, "cheap; inexpensive", in separate entries. 大夫 is dàfū, a "senior official (in imperial China)", or dàifu, a "doctor; physician". 朝阳 has three entries: cháoyáng, "to be exposed to the sun", zhāoyáng, "the morning sun", and a capitalised reading that points to places of that name.
On our count, 766 of the list's headwords carry more than one reading once entries sharing a pinyin are merged. Until a revision dated 4 November 2026, our word finder kept only the first entry for each headword, so 便宜 showed "convenient" alone, and it read the idiom tag from glosses cut at 60 characters, which left it counting 4,602 four-character idioms. The polyphones guide follows the same problem through single characters.
Place names produce most of the words built on 县, 区 and 市
Rank characters by how many words contain them and the top looks odd. 县 xiàn, "county", appears in 1,487 words, third after 不 and 子. 区 qū, "district", and 市 shì, "city", also make the top ten.
Proper nouns account for it. Of the words containing 县, 1,467 are names. The share is 990 of 1,167 for 区 and 764 of 926 for 市, on our count. 不, the most productive character of all, sits in 1,684 words, and only 13 of them are names.
Across the whole list, 16,617 entries, about 16%, are names by CC-CEDICT's capital-letter mark. At the other end of the ranking, the median character appears in 9 words, and 1,123 characters appear in exactly one. Strip the names away and 县 makes 20 words, 县城 xiànchéng, "county seat", among them.
View author profile → · Editorial policy
