关于「「句子越短越好读」的两种误判:句长与常用字比例的二维平面」,《中文可读性与字频句长》给出的定量结论是什么?

结论:第一章推导单看句长排错文本对的概率:句长与少见字比例的相关系数为 0.8 时错误率 0.109,为负 0.6 时升到 0.390。

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。

English version

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: 0.109, 0.390, 0.6, 0.8

sx = 8, sy = 3, D = 0.10 ASL + 0.30 RP
rhocorr(ASL,D)misorder_ASLcorr(RP,D)misorder_RP
+0.90.9720.0760.9780.067
+0.80.9420.1090.9550.096
+0.60.8810.1570.9070.138
+0.40.8150.1970.8570.172
+0.20.7430.2330.8040.203
+0.00.6640.2690.7470.231
-0.20.5750.3050.6860.259
-0.40.4710.3440.6200.287
-0.60.3400.3900.5490.315
a coin flip misorders 0.500 of pairs
查看完整验算程序
# Chapter 1. p13: how often ranking by sentence length alone misorders
# two texts. Model: across a corpus ASL and RP are jointly normal with
# SDs sx = 8 hanzi and sy = 3 per 100 (illustrative) and correlation
# rho. For D = a X + b Y and a proxy P, the chance that a random pair is
# ordered differently by P and by D is arccos(corr(P, D)) / pi (exact for
# jointly normal differences). a = 0.10, b = 0.30 as in the book.
import math

A, B, SX, SY = 0.10, 0.30, 8.0, 3.0


def corr_with_d(rho, proxy):
    vd = math.fsum([(A * SX) ** 2, (B * SY) ** 2,
                    2 * A * B * rho * SX * SY])
    if proxy == "asl":
        cov = math.fsum([A * SX * SX, B * rho * SX * SY])
        return cov / (SX * math.sqrt(vd))
    cov = math.fsum([B * SY * SY, A * rho * SX * SY])
    return cov / (SY * math.sqrt(vd))


def discord(rho, proxy):
    return math.acos(corr_with_d(rho, proxy)) / math.pi


if __name__ == "__main__":
    print("sx = {:.0f}, sy = {:.0f}, D = 0.10 ASL + 0.30 RP".format(SX, SY))
    print("   rho   corr(ASL,D)  misorder_ASL  corr(RP,D)  misorder_RP")
    for rho in (0.9, 0.8, 0.6, 0.4, 0.2, 0.0, -0.2, -0.4, -0.6):
        print("  {:+4.1f}     {:6.3f}       {:6.3f}       {:6.3f}"
              "      {:6.3f}".format(rho, corr_with_d(rho, "asl"),
                                     discord(rho, "asl"),
                                     corr_with_d(rho, "rp"),
                                     discord(rho, "rp")))
    print("a coin flip misorders 0.500 of pairs")

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 1 章「「句子越短越好读」的两种误判:句长与常用字比例的二维平面」(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}
}

本书其他章节结论