关于「实战工具与检查器规格:文气节律台账、预注册检验与推翻条件」,《机器文气与句长节律》给出的定量结论是什么?

结论:第八章给出蒸馏小工具「文气节律与套话密度检查器」、按四件套分组的检查规格表、文气节律台账与预注册检验,配对盲测在选择率 0.60 时需要 153 次判断。本书是写作工艺书:5D Library 书库各卷如实标注 AI 辅助,本书工具用于提高读者体验,不得用于在须披露的场合隐去机器参与;具身细节必须真实。书中以标注框区分已发表的证据、可运行的 Python 计算(21 段程序)、原典与作者的假说。

数据口径:书中示例均为作者自写的合成文字或公有领域文本,参数为示意,全部计算由书中程序生成。 第一至六章依次讲句长指纹与对数正态、突发度与顺序量、检测器的贝叶斯账与偏差、套话密度的泊松检验与群体超额词汇、三元连发与压缩冗余、只动句读的动态规划与真实细节账本:合成草稿只改 5 处句读,变异系数就从 0.123 升到 0.546,而字词一个不动。

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: 0.60, 153

(a) paired preference, one-sided 5%, power 0.8
p10.55: 617 paired judgments
p10.60: 153 paired judgments
p10.65: 67 paired judgments
p10.70: 37 paired judgments
(b) n 200 p_o 0.800 p_e 0.516 kappa 0.587 se 0.058
passages for half-width 0.10263
(c) drafts per type for +/-0.05 on mean CV16
(c) drafts per type for +/-0.03 on mean CV43
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered tests.
# (a) Blind paired preference: readers see the original and the
#     cadence-revised version of one passage in random order and pick
#     the one they would rather keep reading. H0: p = 0.5; one-sided
#     5 per cent, power 0.8, normal approximation.
# (b) Two raters label 200 passages "flat" or "not flat"; Cohen's kappa
#     from an ILLUSTRATIVE table, with its large-sample standard error,
#     and the passages needed for a 95 per cent half-width of 0.10.
# (c) Ledger: drafts per type needed before a CV threshold may move,
#     so that the mean CV is known within 0.05 (sd of CV 0.10, read off
#     the n = 40 band of p22).
import math
from statistics import NormalDist

Z = NormalDist()
za, zb = Z.inv_cdf(0.95), Z.inv_cdf(0.80)
print("(a) paired preference, one-sided 5%, power 0.8")
for p1 in (0.55, 0.60, 0.65, 0.70):
    n = ((za * 0.5 + zb * math.sqrt(p1 * (1 - p1))) / (p1 - 0.5)) ** 2
    print("    p1 = {:.2f}: {:>4} paired judgments".format(p1, math.ceil(n)))

a, b, c, d = 62, 18, 22, 98            # rater1 x rater2 counts, n = 200
n = a + b + c + d
po = (a + d) / n
pe = ((a + b) * (a + c) + (c + d) * (b + d)) / (n * n)
kappa = (po - pe) / (1 - pe)
se = math.sqrt(po * (1 - po) / (n * (1 - pe) ** 2))
print("(b) n {}  p_o {:.3f}  p_e {:.3f}  kappa {:.3f}  se {:.3f}".format(
    n, po, pe, kappa, se))
need = po * (1 - po) / ((1 - pe) ** 2) * (1.959964 / 0.10) ** 2
print("    passages for half-width 0.10: {}".format(math.ceil(need)))

sd_cv = 0.10
for half in (0.05, 0.03):
    k = (1.959964 * sd_cv / half) ** 2
    print("(c) drafts per type for +/-{:.2f} on mean CV: {}".format(
        half, math.ceil(k)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《机器生成文本统计指纹剥离、句长突发律动重构与宗师级文气呼吸感母典》第 8 章「实战工具与检查器规格:文气节律台账、预注册检验与推翻条件」(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-394 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026394,
  author    = {Fan, Yafu},
  title     = {Stripping Statistical Fingerprints of Machine Prose, Sentence-Length Burstiness Reconstruction and Master-Level Stylistic Cadence},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论