Chiang Kai-shek's name in his family's genealogy was 蔣周泰. His son was 蔣經國. His grandson was 蔣孝勇. Read the three marked characters in order — 周, 國, 孝 — and they are three consecutive words of a twenty-character poem the Chiang family of Xikou have been naming children from for generations: 祁斯肇周國，孝友得成章，秀明啟賢達，奕世慶吉昌.
That is the whole mechanism, and the first trap, in one family. A family fixes a poem, and each generation takes the next character. In 蔣周泰 the character is first; in 蔣經國, his son, it is last. If you have been looking for a shared character in the middle of every name, you have probably been looking in the wrong place.
This guide is about finding yours. Our guide to how Chinese given names are built explains what a generation character is and where it sits in the name; this one assumes you already have some names in front of you and want to know which poem they came from, and what to do when nobody in the family has the book.
A poem belongs to a branch, not to a surname
The Chinese term is 字辈 (zìbèi), also written 派语 or 昭穆. It is a character shared by brothers and by paternal cousins of the same generation, and Chinese Wikipedia's article on the practice puts the position plainly: it is usually one character of the given name, 多為第一個字 — most often the first. Singapore's National Library says the same of the jiapu (族譜, a family lineage book) held here: everyone born in the same generation carries the same character as the first character of their names. FamilySearch, which maintains a primer on generation poems for genealogists, hedges it correctly — most often first, in some families second.
The unit that owns a poem is not the surname. It is the branch. Zhu Yuanzhang, the founding Ming emperor, wrote twenty-four poems in one sitting — one for each of his sons' households, twenty characters each — precisely so that cousins in different houses could not collide. The text of the order survives in the History of Ming: he had it done because his descendants were numerous and he feared repeated names. Every one of those twenty-four houses is a Zhu. None of them shares a poem.
He went further. The second character of every two-character given name in the imperial line had to carry one of the five-phase radicals, cycling fire, earth, metal, water, wood. We checked that rule against every name in the line that inherited the throne, reading the radical out of the Unicode character database rather than from a list we had written ourselves, so the check could not quietly agree with us. It holds for 14 of 14 names in the imperial line, from 朱高熾 — 熾, fire — to 朱由檢, whose 檢 is wood, across more than two centuries. (Fourteen names, not fourteen emperors: the last of them was heir apparent and never reigned.) Eleven of the twenty characters were ever used. The dynasty ended before the other nine.
Three names is usually enough to identify a branch
The useful thing about generation poems is that they barely overlap. We took thirty-three poems we could trace to a published text — the twenty-four Ming princely poems, the Confucius fifty, the Yan thirty, three cyclic Song imperial sets, the Qing imperial twenty-six, the two Mao poems from Shaoshan and the Chiang poem above — and looked for any four-character line that appeared in more than one of them.
There were none. Not one shared line across four dynasties and nine centuries of independent families. Inside Zhu Yuanzhang's twenty-four poems, written by one man to one brief, 477 of the 480 character slots are unique; exactly three characters repeat, each in two houses, and only three of the 276 possible pairs of houses share so much as a single character.
Individual characters are a different story. Across the non-Ming poems 世 appears five times, 承 four, 永 four, 公, 德, 昌, 顯, 光 and 啟 three each. These are the auspicious words any family might reach for. So the rule for reading your own family back is: a shared character means nothing, and a shared line means the two poems are not independent. If you can put three or four consecutive names in generation order — your grandfather, your father, you, a cousin — you have a line, and a line is close to unique. One name is not.
Our generation poem lookup will match a run of characters against the poems it carries. Two warnings about it, both ours to own. It indexes Mandarin pinyin only, so if your family name is written Tan, Teo, Ng, Ooi or Khoo you will need the Chinese character or the Mandarin reading to find anything — and typing Tan will return 譚, which is not the surname a Hokkien Tan carries. And most of the poems it ships came from surname aggregator sites rather than from a named genealogy. Treat a match there as a lead, not an answer, until you can put the poem next to a printed source.
Two websites, two poems, and how to tell which is yours
This is the question the open web answers worst. Search 字辈查询 and you will find a whole industry of surname sites, each with a long list of poems and almost none with a citation. They contradict each other, and the contradictions have a shape you can use.
Run the overlap test on them. In the thirty poems our own tool shipped, the line 世澤綿長 appears in eight supposedly unrelated clans, 文章華國 in seven, 詩禮傳家 in seven. Five pairs of unrelated surnames carry character-for-character identical sets. Against thirty-three sourced poems with zero shared lines, that is not what real genealogies look like — it is what copying looks like. A poem that shares a whole line with another surname's poem is either evidence of a documented relationship or evidence that somebody pasted.
The genuine shared cases are instructive because they are messier than the story. It is often said that the four sage clans—Kong, Meng, Yan, and Zeng—all name from the same poem. The sources tell a more complicated story.
- Meng does align with Kong exactly. 孟繁驥 was 74th generation, 孟祥協 75th, 孟令繼 76th — 繁, 祥, 令, the same characters at the same generation numbers as the Kong line.
- Zeng uses the same characters two generations out of step. The Zeng blocks begin 宏聞貞尚衍 at the 63rd generation; Kong's 宏 is the 61st. So a Kong of the 75th generation is 祥 and a Zeng of the 75th is 慶. They are not peers.
- Yan does not share it at all. Chinese Wikipedia states the exception twice and gives the reason: Confucius's mother was a Yan, so the families are related by marriage rather than descent. The Yan poem runs from the 61st generation and begins 公重從嗣胤，伯光紹懋崇 — thirty characters, nothing in common.
If a site tells you your Yan ancestors named from the Confucius poem, you have caught it copying.
Where to look when nobody has the book
The formal answer is the jiapu, and in the region that answer often does not exist in the form people imagine. Singapore's National Library, writing about the genealogies actually held here, notes that a modern Singapore jiapu is commonly written in English, records only three to five generations because the migrant family's history here is short, and includes women — a departure from the older norm. That is a real document and a useful one. It is not a thousand-year scroll.
In rough order of how often they work, the practical routes are: names on a grave or ancestral tablet, which are often the formal genealogy names; the clan association for your surname and county; older relatives who were given Chinese names before the family stopped; and the ancestral village itself, where a branch that stayed may still hold the book. Our guide to tracing Chinese ancestry from Southeast Asia goes through those records in detail.
One route we could not check for you. The Shanghai Library runs the largest genealogy collection in the world, and its online platform refused our requests, so we have no count of how many poems survive there and are not going to print one we could not verify.
Why the poem stopped, and why that is not a failure
The most common version of this question in Singapore and Malaysia is not "what is my poem" but "why did we stop". Usually the answer is that the family stopped giving Chinese given names at all, and a generation character cannot survive that. Singapore's romanisation history is the other half: there has never been one policy, dialect spellings persist across a single family tree, and the Chinese requirement for pinyin names introduced for pre-primary and Primary 1 in 1981 was reversed in 1991. A branch can go from 陳 to Tan to Tang to Chen inside three generations of paperwork, none of it anyone's mistake.
Poems also run out. Every one of them is finite: fifty characters for the Confucius line, twenty for a Ming house, thirty for the Yan. When a family reached the end it petitioned for more, and the grants are dated. The Confucius sequence is the best-documented case in the world. Its grant dates are worth having because they show a living document, not a static heirloom:
- 1400 — eight characters, 公彥承弘聞貞尚胤, granted by the Jianwen Emperor. Not, despite what a great many pages say, by the founding Hongwu emperor; 希 and 言 were placed in front afterwards, because the 56th and 57th dukes had already been titled.
- 1628 — ten more, 興毓傳繼廣，昭憲慶繁祥, petitioned by the 65th duke.
- Under the Qing — ten more, 令德維垂佑，欽紹念顯揚. The date is disputed and we will not pick for you: the best-cited sources say 1863, petitioned by the 75th duke on his succession, while many Chinese sources say 1744. The block before it ran from the 66th to the 75th generation, which is suggestive of 1863 and is reasoning rather than evidence.
- 1919–1920 — the last twenty, drafted by the 76th duke, who died before submitting them; his widow convened the clan and the government of the day approved them.
Two characters in that sequence changed shape rather than meaning. 弘 became 宏 to avoid the Qianlong emperor's personal name, and 胤 became 衍 to avoid the Yongzheng emperor's. A family document that swerves around an emperor's name is behaving normally, not badly, and it is one reason two printed versions of the same poem can differ.
What a generation character cannot tell you
It cannot date anyone. We took every generation of the senior Confucius line with a recorded birth year — thirty-four generations spanning 1,008 years — which averages 29.6 years to a generation. But the twenty-five directly observed father-to-son steps run from 16 years to 48, with a median of 24. So one generation places a birth inside a window about 32 years wide; three generations, roughly a century; ten, three centuries. A generation character is an excellent ordering device but a poor dating device. Any site that converts a character straight into a birth decade is overselling it.
What it does is place you exactly in a line. The poem is still running: 孔德成 (1920–2008) carried 德, his son 孔維益 carried 維, his grandson 孔垂長 carries 垂, and his great-grandson 孔佑仁, born in 2006, carries 佑—the fifth character of a block granted in the nineteenth century.
One last piece of evidence for how seriously families took this. The Confucius clan ran a second poem at the same time, for courtesy names, recorded in the 闕里志: 起鍾振體京，顯法澤羽瑞. As far as we can find, nobody has checked it. We aligned it against the dukes' recorded courtesy names and it holds for exactly six generations from 1628 — 起, 鍾, 振, 體, 京, 顯 — and then stops dead, four consecutive generations with no match. A family adopted a second naming scheme alongside the famous one, ran it faithfully for six generations, and quietly let it go. Which is, in the end, the ordinary fate of these poems, and no reason not to go looking for yours.
The Ming order and the twenty-four princely poems are quoted from 《明史》卷100 as printed on Chinese Wikisource; the Confucius and Yan sequences, the grant dates and the taboo substitutions from Chinese Wikipedia's 孔子世家 and 行辈 articles, which cite the clan's own published genealogy; the Singapore material from the National Library's BiblioAsia issue on the jiapu and from the published account of Singapore's Chinese romanisation policy. The overlap counts, the five-phase check and the generation-length figures are ours, computed from those texts on 24 September 2026; the radical readings come from the Unicode character database. Where two sources disagree — the Qing grant date, the text at generations 94 to 96 — we have said so rather than chosen.
View author profile → · Editorial policy
