关于「「的」字链与定语堆叠:左分支负担的精确计算」,《去翻译腔的中文文案计算》给出的定量结论是什么?

结论:第二章证明定语堆叠的依存长度近似平方增长,四层时全部关系保持住的示意概率只剩 0.098。

数据口径:书中的例句与参数均为示意;全部计算由书中程序生成;本书不构成个别化的法律、广告合规或营销建议,只讨论守规矩的说服。 第一章用手工标注的示意依存树说明,西化稿的平均依存距离是 3.429,母语改写是 1.515;与保持树结构的随机语序相比,西化稿的比值为 0.950,几乎等于随机,改写稿为 0.640。

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: 0.098

m = 3 words per modifier, r = 0.93
k | stacked: DL max MDD keep | topic-first: DL max MDD keep
1 |62 1.500 0.865 |63 2.00 0.804
2 |166 2.000 0.560 |123 2.00 0.647
3 |3010 2.500 0.271 |183 2.00 0.520
4 |4814 3.000 0.098 |243 2.00 0.419
5 |7018 3.500 0.027 |303 2.00 0.337
6 |9622 4.000 0.005 |363 2.00 0.271
at k = 4: stacked keeps 0.098, topic-first keeps 0.419
查看完整验算程序
# Chapter 2. p22: Why stacked pre-nominal modifiers hurt. A noun phrase
# with k modifiers, each of m words plus 的, all hanging on one head noun
# ("[w w M 的] [w w M 的] N"), against a topic-first rewrite in which the
# noun comes first and the k modifiers become k short clauses chained one
# after another ("N, w w M, w w M").
# Both trees are built explicitly and measured with the same function.
# Retention: each intervening word keeps the pending link with
# probability r = 0.93 (an illustrative decay, not a measured constant);
# P(all links kept) is the product over arcs of r ** (distance - 1).
import math

R = 0.93
M = 3


def stacked(k, m):
    heads, pos = {}, 0
    noun = k * (m + 1) + 1
    for _ in range(k):
        first = pos + 1
        mhead = first + m - 1
        for w in range(first, mhead):
            heads[w] = mhead
        heads[mhead] = noun
        heads[mhead + 1] = mhead          # 的 attaches to its modifier
        pos = mhead + 1
    heads[noun] = 0
    return heads


def topic_first(k, m):
    heads = {1: 0}
    prev, pos = 1, 1
    for _ in range(k):
        first = pos + 1
        mhead = first + m - 1
        for w in range(first, mhead):
            heads[w] = mhead
        heads[mhead] = prev
        prev, pos = mhead, mhead
    return heads


def measure(heads):
    d = [abs(a - b) for a, b in heads.items() if b]
    keep = math.exp(math.fsum((x - 1) * math.log(R) for x in d))
    return math.fsum(d), max(d), math.fsum(d) / len(d), keep


print("m = {} words per modifier, r = {}".format(M, R))
print(" k | stacked: DL  max   MDD   keep | topic-first: DL max  MDD  keep")
for k in range(1, 7):
    a = measure(stacked(k, M))
    b = measure(topic_first(k, M))
    print("{:>2} | {:>11.0f} {:>4} {:>5.3f} {:>5.3f} | {:>15.0f} {:>3}"
          " {:>4.2f} {:>5.3f}".format(k, a[0], a[1], a[2], a[3],
                                      b[0], b[1], b[2], b[3]))
k = 4
a, b = measure(stacked(k, M)), measure(topic_first(k, M))
print("at k = {}: stacked keeps {:.3f}, topic-first keeps {:.3f}"
      .format(k, a[3], b[3]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《跨语境母语修辞韵律、去翻译腔句法树重构与马新台华语转化语料学》第 2 章「「的」字链与定语堆叠:左分支负担的精确计算」(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-144 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026144,
  author    = {Fan, Yafu},
  title     = {Native Cross-Context Prosody, De-Translationese Syntax-Tree Rewriting and a Conversion Corpus of Malaysian, Singaporean and Taiwanese Chinese},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论