关于「字频的长尾:通用规范汉字表、齐普夫模型与一份示意字频等级表」,《中文可读性与字频句长》给出的定量结论是什么?
结论:程序先核对三级字数相加:3500 加 3000 加 1605,等于 8105。满足目标覆盖率 0.99 的指数是 s = 1.3840。在这个模型里,前 1000 个字覆盖 0.9676,前 2500 个字覆盖 0.9850,一级字表覆盖 0.9900,即每百字有 1.00 个字落在一级字表之外;一、二级合计 6500 字覆盖 0.9977。
数据口径:作者自拟的四篇小样本中,四字短句堆僻字的文本平均句长最短,按本书的示意难度公式却最难,得分 19.15;同样短句、全用常用字的文本只有 0.45。 书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议。
对照数据表
下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.9676, 0.9850, 0.99, 0.9900, 0.9977, 1.00
| levels 3500 + 3000 + 1605 = 8105 | ||
| Zipf exponent giving level-one coverage 0.99: s = 1.3840 | ||
| top k | coverage | outside per 100 hanzi |
| 100 | 0.8844 | 11.56 |
| 500 | 0.9497 | 5.03 |
| 1000 | 0.9676 | 3.24 |
| 2500 | 0.9850 | 1.50 |
| 3500 | 0.9900 | 1.00 |
| 6500 | 0.9977 | 0.23 |
| 8105 | 1.0000 | 0.00 |
| chance a sentence holds >= 1 hanzi outside the common set | ||
| L | c=0.99 | c=0.95 |
| 4 | 0.0394 | 0.1855 |
| 8 | 0.0773 | 0.3366 |
| 16 | 0.1485 | 0.5599 |
| 25 | 0.2222 | 0.7226 |
| 37 | 0.3106 | 0.8501 |
| 72 | 0.5150 | 0.9751 |
查看完整验算程序
# Chapter 2. p21: how much text the 3500 level-one hanzi cover, in a toy
# Zipf model. Facts: the Table of General Standard Chinese Characters
# (State Council, 2013) lists 8105 hanzi in three levels of 3500, 3000
# and 1605. Model (illustrative): token share of rank r is proportional
# to r^-s over the closed set of 8105; s is solved so that level one
# covers a target share C1 = 0.99 (an assumed round figure, not a
# measured one). A sentence of L independent hanzi then contains at
# least one hanzi outside level one with chance 1 - c^L.
import math
N, LEVELS, C1 = 8105, (3500, 3000, 1605), 0.99
def harmonic(k, s):
return math.fsum(r ** -s for r in range(1, k + 1))
def coverage(k, s):
return harmonic(k, s) / harmonic(N, s)
def solve_s(target, k):
lo, hi = 0.5, 3.0
for _ in range(60):
mid = 0.5 * (lo + hi)
if coverage(k, mid) < target:
lo = mid
else:
hi = mid
return 0.5 * (lo + hi)
if __name__ == "__main__":
print("levels {} + {} + {} = {}".format(*LEVELS, N))
s = solve_s(C1, LEVELS[0])
print("Zipf exponent giving level-one coverage {:.2f}: s = {:.4f}"
.format(C1, s))
print(" top k coverage outside per 100 hanzi")
for k in (100, 500, 1000, 2500, 3500, 6500, 8105):
c = coverage(k, s)
print(" {:5d} {:.4f} {:6.2f}".format(k, c, 100 * (1 - c)))
print("chance a sentence holds >= 1 hanzi outside the common set")
print(" L c=0.99 c=0.95")
for n in (4, 8, 16, 25, 37, 72):
print(" {:3d} {:.4f} {:.4f}".format(
n, 1 - 0.99 ** n, 1 - 0.95 ** n))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《中文可读性公式与字频句长、落地页与专著段落的阅读难度母典》第 2 章「字频的长尾:通用规范汉字表、齐普夫模型与一份示意字频等级表」(版本 1.0.0). 5D Library. DOI 待发布
DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-323 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026323,
author = {Fan, Yafu},
title = {Chinese Readability from Character Frequency and Sentence Length: Reading Difficulty of Landing Pages and Monograph Paragraphs},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}