关于「章节信息熵:惊讶度预算、换角度重述与修辞动作的语法」,《专著可读性的可证伪称量》给出的定量结论是什么?

结论:第一章用香农熵称量三份草稿,故事草稿的词熵落在套话到术语区间的 0.930 处,好读来自解码成本低而不是信息少;原句照抄三遍留下的漏读是换角度重述的 5.22 倍。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律意见,也不是个人化的商业或财务建议。 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。

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: 0.930

draftwordsdistinctH bits/wordzlib ratiolong words
boilerplate52143.5390.3580.000
jargon54445.3340.6790.074
story54415.2080.6240.000
entropy gap jargon minus boilerplate: 1.795 bits/word
story sits at 0.930 of the boilerplate-to-jargon range
查看完整验算程序
# Chapter 1. p11: word entropy and compressibility of three drafts.
# Model: three English drafts explain the same theorem (the source
# coding bound). For each we count words, distinct words, the unigram
# word entropy H = -sum p log2 p in bits per word, and the zlib ratio
# (compressed bytes over raw bytes) as a crude redundancy gauge, and
# the share of long words (ten letters or more) as a crude gauge of
# vocabulary a lay reader must decode.
# Texts are written for this book; the measures are exact.
import collections
import math
import re
import zlib

DRAFTS = {
    "boilerplate": (
        "It is important to note that entropy is important. It is "
        "important to note that the code length is important. It is "
        "important to note that the bound is important. It is important "
        "to note that the result is important for the code and the "
        "bound and the entropy and the result."),
    "jargon": (
        "For a discrete memoryless source with entropy H, every uniquely "
        "decodable code satisfies expected length L at least H, by the "
        "Kraft inequality and Gibbs; Huffman construction attains L below "
        "H plus one, and block coding over n symbols drives the per symbol "
        "overhead under one over n, so H is the asymptotic infimum."),
    "story": (
        "Picture a game of twenty questions played against a stingy "
        "oracle. You may only ask yes or no. If the oracle favours some "
        "answers, clever questions split the likely ones first, and on "
        "average you finish sooner. Entropy is the fewest questions you "
        "will ever need on average, no matter how clever you are."),
}


def measures(text):
    words = re.findall(r"[a-z]+", text.lower())
    n = len(words)
    counts = collections.Counter(words)
    h = -math.fsum(c / n * math.log2(c / n) for c in counts.values())
    raw = text.encode("ascii")
    ratio = len(zlib.compress(raw, 9)) / len(raw)
    long_share = len([w for w in words if len(w) >= 10]) / n
    return n, len(counts), h, ratio, long_share


print("draft        words  distinct  H bits/word  zlib ratio  long words")
rows = {}
for name in ("boilerplate", "jargon", "story"):
    n, d, h, r, lw = measures(DRAFTS[name])
    rows[name] = (n, d, h, r, lw)
    print("{:<11s}  {:5d}  {:8d}  {:11.3f}  {:10.3f}  {:10.3f}".format(
        name, n, d, h, r, lw))
gap = rows["jargon"][2] - rows["boilerplate"][2]
print("entropy gap jargon minus boilerplate: {:.3f} bits/word".format(gap))
print("story sits at {:.3f} of the boilerplate-to-jargon range".format(
    (rows["story"][2] - rows["boilerplate"][2]) / gap))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《畅销专著修辞信息论、硬核公式具象隐喻映射与章末悬念母机》第 1 章「章节信息熵:惊讶度预算、换角度重述与修辞动作的语法」(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-149 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026149,
  author    = {Fan, Yafu},
  title     = {Information-Theoretic Rhetoric of Bestselling Monographs, Concrete Metaphor Mapping of Hard Formulas and the Chapter-End Suspense Engine},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论