关于「齐普夫曲线:累积覆盖率的解析式与词表大小的反解」,《词汇覆盖率与速成边界》给出的定量结论是什么?

结论:第二章推导调和数的闭式近似与幂律尾部的比值公式,并用两个已发表锚点拟合齐普夫—曼德博参考曲线,两千词族约覆盖 0.8618。

数据口径:书中参数为示意,全部计算由书中程序生成;本书是非个人化的经营与教学参考,不是针对任何个人或企业的经营建议,也不承诺任何课程能在某个期限内带来流利。 书中以标注框区分已发表的证据、可运行的 Python 计算(19 段程序)、原典与作者的假说。

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: 0.8618

anchors95 % at 4500 families, 98 % at 8000 families
V s q C(1000) C(2000) n90 n99
150001.884 256.0 0.7755 0.8774 2456 10515
200002.389 664.5 0.7266 0.8618 2647 11158
300002.698 963.9 0.7031 0.8537 2726 11637
500002.836 1106.6 0.6939 0.8504 2756 11882
sensitivity at V20000: move the anchors inside their ranges
95 % at 4000, 98 % at 8000: s1.915 C(2000) = 0.8973
95 % at 5000, 98 % at 8000: s3.224 C(2000) = 0.8087
95 % at 4500, 98 % at 9000: s1.769 C(2000) = 0.8902
查看完整验算程序
# Chapter 2. p23: a Zipf-Mandelbrot curve through the two published anchors
# Model: f(r) proportional to (r + q)^(-s) for word families r = 1..V.
# Coverage C(n) = S(n) / S(V), with S(n) = sum_{r<=n} (r + q)^(-s)
# computed exactly for r <= 20 and by Euler-Maclaurin beyond. For each
# illustrative family universe V we solve for (s, q) so that the curve
# passes through Laufer and Ravenhorst-Kalovski's anchors, 95 % at the
# midpoint 4500 families and 98 % at 8000 families, then read off what
# the curve implies for 1000 and 2000 families and for 99 %.
import math


def partial(n, s, q):
    m = min(n, 20)
    head = [(r + q) ** (-s) for r in range(1, m + 1)]
    if n <= 20:
        return math.fsum(head)
    a, b = 20 + q, n + q
    if abs(s - 1.0) < 1e-12:
        integ = math.log(b / a)
    else:
        integ = (a ** (1 - s) - b ** (1 - s)) / (s - 1)
    ends = 0.5 * (b ** (-s) - a ** (-s))
    slope = -s * (b ** (-s - 1) - a ** (-s - 1)) / 12.0
    return math.fsum(head + [integ, ends, slope])


def cover(n, s, q, v):
    return partial(n, s, q) / partial(v, s, q)


def q_for(s, v, n2, c2):
    lo, hi = 0.0, 20000.0
    for _ in range(90):
        mid = 0.5 * (lo + hi)
        if cover(n2, s, mid, v) > c2:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)


def fit(v, n1=4500, c1=0.95, n2=8000, c2=0.98):
    lo, hi = 1.0, 8.0
    for _ in range(70):
        s = 0.5 * (lo + hi)
        q = q_for(s, v, n2, c2)
        if cover(n1, s, q, v) > c1:
            lo = s
        else:
            hi = s
    s = 0.5 * (lo + hi)
    return s, q_for(s, v, n2, c2)


def words_for(c, s, q, v):
    lo, hi = 1, v
    while hi - lo > 1:
        mid = (lo + hi) // 2
        if cover(mid, s, q, v) >= c:
            hi = mid
        else:
            lo = mid
    return hi


if __name__ == "__main__":
    print("anchors: 95 % at 4500 families, 98 % at 8000 families")
    print("      V      s        q    C(1000)  C(2000)   n90    n99")
    for v in (15000, 20000, 30000, 50000):
        s, q = fit(v)
        print("  {:5d}  {:5.3f}  {:7.1f}   {:.4f}   {:.4f}  {:5d}  {:5d}"
              .format(v, s, q, cover(1000, s, q, v),
                      cover(2000, s, q, v), words_for(0.90, s, q, v),
                      words_for(0.99, s, q, v)))
    print("sensitivity at V = 20000: move the anchors inside their ranges")
    for n1, n2 in ((4000, 8000), (5000, 8000), (4500, 9000)):
        s, q = fit(20000, n1=n1, n2=n2)
        print("  95 % at {}, 98 % at {}: s = {:.3f}  C(2000) = {:.4f}"
              .format(n1, n2, s, cover(2000, s, q, 20000)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《二语词汇覆盖率九十八阈值、齐普夫频率曲线与马来语闽南语速成边界母典》第 2 章「齐普夫曲线:累积覆盖率的解析式与词表大小的反解」(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-228 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026228,
  author    = {Fan, Yafu},
  title     = {The 98% Lexical-Coverage Threshold, Zipfian Frequency Curves and Fast-Track Limits for Malay and Hokkien Learners},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论