A Chinese compound joins two or more characters into one word, and grammar textbooks sort the joins by how the parts relate: one character modifies the other, the two are equals, a verb takes an object or a result, or a subject takes a predicate. On the HSK 3.0 exam syllabus, 78.5% of the 10,896 distinct words are two characters long, on our count. We drew 100 two-character words from the first three levels and sorted them by hand. On our reading, the first character modified the second in 40. The two were equals in 25. Only one joined a subject to a predicate, the greeting 你好.
Paired characters join up in most entries on the HSK list
The syllabus is 《中文水平考试 HSK 考试大纲》 from 中外语言交流合作中心 (CTI), released in November 2025 for use from July 2026. Its 11,000 rows come to 10,896 distinct forms once we strip sense numbers and keep each word at its lowest level. Of those, 8,548 are two characters (78.5%) and 1,503 are one (13.8%).
The two-character share rises from 48.0% at level 1 to 81.3% at level 5. These are shares of list entries, a different measure from running text. Our count of the same syllabus by level has the totals.
Modifiers attach to their heads in the largest group of our sample
A 2018 paper by Cui and colleagues in Frontiers in Psychology sets out the five-way scheme: "two-constituent Chinese compounds are divided into five groups". We added bins for affixed, doubled and opaque words. Python's random.seed(208) drew 100 from the 588 two-character forms at levels 1–3, and one reader classed each by the structure a learner can still see in its present meaning. A star marks a word we think could fairly go elsewhere.
| Type | Count | Sample words |
|---|---|---|
| Modifier-head | 40 | 大人 饮料 常用 节日 平时 花园 雨衣 同意* 邻居* 电脑 书店 难过 后天 不用 不同 不要 好多 常见 衬衫 校园 小时 球场 秋天 铅笔 小学 那边 牛奶 迟到 数学 外国 网球 这边 水果 牙刷 电梯 奶茶 后年 邮件 路口 相信 |
| Coordinate | 25 | 回答 季节 班级 应该 奇怪 姓名 已经 语言 东北 关注 历史* 号码 解决 喜欢 快乐 结束 舒服* 喜爱 世界 凉快 发生* 身体 安全 事情 决定* |
| Verb-object | 10 | 着急 拍照 放假 洗澡 认真 照相 下课 住院 发烧 对话* |
| Verb-complement | 5 | 遇到 变成 遇见 回去 起来 |
| Subject-predicate | 1 | 你好* |
| Affixed | 4 | 本子 她们 咱们 突然 |
| Reduplicated | 1 | 爸爸 |
| No visible structure | 14 | 怎样 从小 马上 漂亮 可是 所以 比如 其实 容易 有的 方便 这些* 后来 一定 |
In a modifier-head word the second character names the thing and the first narrows it: 雨衣 is a coat for rain, 电脑 an electric brain. A second reading of the list disagreed with ours on two words, moving 历史 and 对话 here, which would make 42. With 100 words, the two largest counts could each sit about ten points either way.
Equal partners pair up side by side in about a quarter of the sample
Coordinate words set two related characters together, and neither modifies the other. 身体 joins two words for the body; 东北 joins two directions. This is also where the judgement calls cluster. 决定 and 发生 can be read as a verb with its result, and 舒服 is hard to split at all.
Verbs couple with an object or a result in fifteen sample words
Ten words on our reading are a verb and its object. 洗澡 xǐzǎo is wash plus bath. 认真, earnest, now works as an adjective, but its parts still read as verb and object. Five more are a verb and its result, as 变成, change and become.
Schemes disagree on real words. The Cui paper glosses its examples in English. Its supplement example is glossed "manage and machine, meaning driver", which fits 司机; our rules would make that verb-object. Its subject-predicate example, glossed as age plus light, meaning young, fits 年轻, a level-3 word in our pool that the draw missed. Ours caught only 你好, which could as fairly be called a phrase.
Dictionary file order detached everyday words from a capped list
A list of words containing a character answers a different question from a list of common words. CC-CEDICT, which MDBG publishes, held 124,945 entries in its 24 May 2026 release, stored in the code-point order of each Traditional headword. Until 3 November 2026 our Chinese word formation tool said it showed the most common words, but it took those entries in file order, shortest first, and stopped at 36. We rebuilt all 5,993 of its lists from that rule and every one matched. Under 学 it did not show 学生 or 学校, and it listed 52 of the 163 multi-character HSK level-1 words under every character they contain.
The lists rebuilt that day order each length band by wordfreq, Robyn Speer's frequency table, which includes SUBTLEX-CH (Cai and Brysbaert, 2010) and, by its README, reflects usage "through about 2021". Measured by us against the syllabus, they put 147 of those 163 words under every character, and three under none: 不客气, 面条儿 and 外边.
Fourteen sample words have fused past any inner grammar
马上 is literally on horseback and now means at once; 所以, so, behaves as a single function word. They have their own bin because we classed by present meaning. Class by origin and 马上 其实 一定 后来 怎样 这些 become modifier-head, while 比如 and 方便 become coordinate. On our arithmetic, modifier-head then rises to 46, coordinate to 27, and the bin shrinks to six. Modifier-head leads under either rule.
A neutral tone flags where a suffix has stuck on
Tone is a good signal for when a written ending is a suffix versus a character with its own meaning. CC-CEDICT records a neutral zi in 105 of the 121 HSK words ending in 子, as in 桌子 zhuōzi. Twelve keep a full zǐ, and four carry both, among them 妻子; the syllabus writes all four with the neutral tone.
The ending 头 runs the other way. CC-CEDICT gives a neutral tou to 14 of the 42 HSK words ending in it, among them 木头, wood. The other 28 keep tóu, head or end, as in 开头, the beginning. The syllabus writes 码头, 拳头 and 势头 with different tones from the dictionary, and on its spellings the count is 13. Of the 36 ending in 儿, CC-CEDICT joins 30 to the syllable before, as in 哪儿 nǎr; five keep a full ér, as in 女儿 nǚ'ér, and 口哨儿 is not in CC-CEDICT.
Tone is evidence and falls short of a definition. 老实 lǎoshi and 东西 dōngxi end in a neutral syllable and carry no suffix. The prefix 老 gives no tone signal at all, so we checked its 19 HSK words by hand. Six keep the meaning old, as 老人 does. Eight read as a prefix, among them 老师, teacher, and 老虎, tiger. The other five, such as 老家, could go either way.
