关于「句长是旧指纹:从尤尔、威廉斯到 Delta」,《机器文气与句长节律》给出的定量结论是什么?

结论: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。

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

text       n   mean     sd     CV       B      r1  short
zh steady  8  28.88   1.05  0.036  -0.930   0.097  0.000
zh breath 10  14.10  11.23  0.796  -0.113  -0.356  0.400
en steady  8  13.38   0.99  0.074  -0.862  -0.367  0.000
en breath 10   9.90   5.50  0.556  -0.285  -0.187  0.300
lengths zh steady: [30, 29, 28, 27, 30, 30, 29, 28]
lengths zh breath: [3, 29, 3, 10, 22, 21, 10, 35, 6, 2]
lengths en steady: [13, 14, 13, 12, 15, 14, 12, 14]
lengths en breath: [5, 17, 7, 9, 16, 13, 8, 18, 5, 1]
查看完整验算程序
# Chapter 1. p11: split sentences, then length mean, sd, CV, B, r1.
# Four SYNTHETIC passages written by the author for this book: a flat
# "steady" passage and a varied "breathing" passage in each language.
# None is a real order, customer or event. Chinese length = hanzi per
# sentence; English length = words per sentence. sd is the population
# standard deviation; B = (sd - mean) / (sd + mean) as in Goh and
# Barabasi; r1 = lag-1 autocorrelation; short = <= 6 hanzi / <= 5 words.
import math
import re

ZH_STEADY = (
    "在当今快速发展的时代,写作能力已经成为每个人不可或缺的重要素养。"
    "首先,清晰的结构能够帮助读者快速理解文章的核心观点与主要内容。"
    "其次,准确的用词可以有效提升表达的专业性、可信度和整体感染力。"
    "此外,合理的段落安排有助于增强文章的逻辑性、连贯性与层次感。"
    "值得注意的是,不同类型的文本对于语言风格有着各自不同的具体要求。"
    "因此,写作者需要根据目标读者的需求灵活调整自己的表达方式与策略。"
    "与此同时,持续的阅读积累也是提升写作水平的重要途径和有效方法。"
    "总之,只有不断练习、认真反思和虚心学习,才能真正写出优秀的文章。"
)
ZH_BREATH = (
    "货到了。"
    "三只纸箱,两只完好,第三只的一角压出一道斜折痕,像被人用膝盖顶过。"
    "先不拆。"
    "拍照,记下运单尾号,再拆。"
    "里面的书脊没有伤,只有塑封起了皱,用指甲一刮就平。"
    "这一单原本可以写成一句客套话:货品完好,请放心。"
    "可客人问的是那道折痕。"
    "所以回信只写看见的东西:哪一只箱子,压在哪个角,书有没有受损,"
    "照片在附件第几张。"
    "她回了两个字。"
    "收到。"
)
EN_STEADY = (
    "In today's fast-paced world, clear writing has become an essential "
    "skill for everyone. Firstly, a clear structure helps readers "
    "understand the main ideas of the text quickly. Moreover, precise "
    "word choice can enhance the clarity, credibility, and impact of "
    "writing. Additionally, well-organized paragraphs improve the logic, "
    "coherence, and flow of the argument. It is worth noting that "
    "different types of texts have their own specific stylistic "
    "requirements. Therefore, writers need to adjust their approach "
    "according to the needs of their audience. Furthermore, consistent "
    "reading is an important way to improve overall writing ability. In "
    "conclusion, only through practice, reflection, and learning can we "
    "write truly excellent texts."
)
EN_BREATH = (
    "The parcel came at noon. Three boxes: two intact, the third creased "
    "across one corner as if someone had knelt on it. We did not open it "
    "at once. Photo first, then the waybill number, then the knife. The "
    "spines inside were fine; only the shrink-wrap had wrinkled, and it "
    "flattened under a thumbnail. A form letter would have said that the "
    "goods arrived in good condition. But the customer had asked about "
    "the crease. So the reply said which box, which corner, whether any "
    "book was marked, and which attachment showed it. She answered in "
    "one word. Thanks."
)

HAN = re.compile(r"[一-鿿]")
WORD = re.compile(r"[A-Za-z][A-Za-z'\-]*")


def split_zh(text):
    parts = re.split(r"(?<=[。!?])", text)
    return [len(HAN.findall(p)) for p in parts if HAN.search(p)]


def split_en(text):
    parts = re.split(r"(?<=[.!?])\s+", text.strip())
    return [len(WORD.findall(p)) for p in parts if WORD.search(p)]


def stats(xs, short_cut):
    n = len(xs)
    m = math.fsum(xs) / n
    sd = math.sqrt(math.fsum((x - m) ** 2 for x in xs) / n)
    cv = sd / m
    b = (sd - m) / (sd + m)
    num = math.fsum((xs[i] - m) * (xs[i + 1] - m) for i in range(n - 1))
    den = math.fsum((x - m) ** 2 for x in xs)
    r1 = num / den if den > 0 else 0.0
    short = len([x for x in xs if x <= short_cut]) / n
    return n, m, sd, cv, b, r1, short


rows = [("zh steady", split_zh(ZH_STEADY), 6),
        ("zh breath", split_zh(ZH_BREATH), 6),
        ("en steady", split_en(EN_STEADY), 5),
        ("en breath", split_en(EN_BREATH), 5)]
print("text       n   mean     sd     CV       B      r1  short")
for name, xs, cut in rows:
    n, m, sd, cv, b, r1, sh = stats(xs, cut)
    print("{:<9} {:>2} {:>6.2f} {:>6.2f} {:>6.3f} {:>7.3f} {:>7.3f} {:>6.3f}"
          .format(name, n, m, sd, cv, b, r1, sh))
print("lengths zh steady:", split_zh(ZH_STEADY))
print("lengths zh breath:", split_zh(ZH_BREATH))
print("lengths en steady:", split_en(EN_STEADY))
print("lengths en breath:", split_en(EN_BREATH))

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 1 章「句长是旧指纹:从尤尔、威廉斯到 Delta」(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