The Chinese Bible Follows Zipf's Law (α 0.993)
Answering: “does the Chinese Bible follow Zipf's law”
Updated 2026-09-22 · plain text · JSON
Zipf's law says the second most common word occurs about half as often as the first, the third about a third as often, and so on — frequency proportional to 1/rank. It holds for most natural language, and this page reports how well it holds for the Chinese text of the Bible specifically. The fit is ordinary least squares of log(occurrences) against log(rank) over the 1,000 most frequent segmented words: exponent 0.993, R² 0.994.
The checkpoints below show what that means in plain counts: the most frequent word occurs 46,164 times, the 100th 876 times, the 1,000th 83. Rank ten-thousand occurs once. That is the same curve behind every coverage figure on this site — it is why 825 words cover 80% of the text while the full vocabulary runs to 11,168.
Fitted over all 11,168 ranks instead of the head, the exponent comes out at 1.656 (R² 0.939). That is not a better or worse answer, it is a different question: the tail is dominated by the 2,579 words that occur exactly once, and a straight line through them describes the tail, not the law. Reporting the head fit is the convention, and the whole-range figure is published next to it so nobody has to guess which was used.
| Rank | Occurrences | Predicted by the fit |
|---|---|---|
| 1 | 46,164 | 46,164 |
| 10 | 7,028 | 4,691 |
| 100 | 876 | 477 |
| 1,000 | 83 | 48 |
| 5,000 | 7 | 10 |
| 10,000 | 1 | 5 |
Prediction is the fitted curve anchored on the most frequent word, shown so the fit can be judged rather than taken on trust.
Questions
- Which words are at the top?
- Function words: 的, 了, 是 and the pronouns. They are excluded from every "words to learn" list on this site but counted in full here, because a frequency distribution with the commonest words removed is not a frequency distribution.
- Does the character distribution behave the same way?
- Similarly, but over a much smaller inventory: 2,996 distinct characters against 11,168 distinct words, with the 73 commonest characters covering half the text.
- How was this computed?
- From the segmented text: forward maximum matching against CC-CEDICT (this site's own segmentation). The fit is least squares on the log-log ranks, and the dataset with every count is at /raw/corpus-stats.json.
Related reference pages
Take it further
- 97 ready-made vocabulary decks — by HSK level, by book or by topic, exportable to Anki, Pleco or Quizlet.
- Printable 田字格 worksheets — passages with pinyin, meanings and tracing rows.
- The parallel reader — Chinese, pinyin and your own language side by side, every word tappable.
- The dataset behind these numbers — plain JSON, free to reuse with attribution.
How these numbers were produced
- Text: 和合本 (Chinese Union Version), simplified script — public domain — 66 books, 1,189 chapters, 31,021 verses.
- Words: forward maximum matching against CC-CEDICT (this site's own segmentation). A different segmenter gives slightly different word counts; character counts are unaffected.
- HSK levels: HSK 3.0 levels from the complete-hsk-vocabulary list; band 7 covers HSK 7-9.
- Reading times: 260 characters a minute — an assumption about a fluent adult reader, not a measurement.
- Computed: 2026-09-22, from the text on this site.
These are counts anyone can reproduce from the published dataset. They are not a peer-reviewed linguistic study, and the difficulty rankings are one stated formula rather than a validated readability score.
Definitions: CC-CEDICT, licensed CC BY-SA (https://creativecommons.org/licenses/by-sa/4.0/). If you share this file onward, keep this notice and share alike. Character data: Make Me a Hanzi (Arphic Public License / LGPL).