关于「原典考辨:严复三难、孔子正名、哲罗姆意译与施莱尔马赫两条路」,《嵌词阈值与单语等效》给出的定量结论是什么?
结论:七种译法里,帕累托意义上不被别人全面压过的有四种:直译、借义复合、全覆盖构式与直接嵌入。逐词硬译(word-for-word)在三项上都不占优:「信」0.800,「达」只有 0.587,「雅」只有 0.40,被语用标记移位的构式全面压过。直接嵌入的「达」是 0.767,「信」却只有 0.820,因为一成多读者不认得这个词。按严复规则,「达」过了 0.6 线的译法里,「信」最高的是全覆盖构式 hyper+,「信」1.000、「达」0.670、「雅」0.95。改用三项等权平均,结果仍是它,平均分 0.873。
数据口径:嵌词收益只在读者共享双语词库时成立:把单语表达、嵌入与嵌入加注释写成覆盖率的三条成本线,示意参数下 kiasu 在书面渠道的覆盖率阈值为 0.844;读者不发问时,失败会变成沉默的误读;各地区分页面值得嵌的词,合成混合页面后可能全部跌破阈值。 书中参数为示意,全部计算由书中程序生成,不对应任何真实客户、对话或成交;马来语、闽南语、粤语、英语、华语一律平等对待。
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: 0.40, 0.587, 0.6, 0.670, 0.767, 0.800
| rendering | xin | da | ya | pareto |
|---|---|---|---|---|
| literal | 0.400 | 0.936 | 1.00 | yes |
| compound | 0.600 | 0.875 | 1.00 | yes |
| marker | 0.800 | 0.717 | 1.00 | no |
| hyper+ | 1.000 | 0.670 | 0.95 | yes |
| word-for-word | 0.800 | 0.587 | 0.40 | no |
| insert | 0.820 | 0.767 | 1.00 | yes |
| insert+gloss | 0.964 | 0.648 | 0.90 | no |
| Yan Fu rule (da >= 0.6, then xin, then ya): hyper+ | ||||
| equal weights instead: hyper+ (mean 0.873) |
查看完整验算程序
# Chapter 7. p71: Yan Fu's three difficulties as three scores.
# Renderings of "kiasu" for a reader group with coverage p = 0.82.
# xin (fidelity) = expected feature coverage; da (getting through) =
# decode probability * exp(-length / 30); ya (elegance) = 1 minus a
# translationese penalty. Coverage and lengths come from p61; the
# word-for-word line is a hard literal rendering of an English gloss.
# Yan Fu rule: keep renderings with da >= 0.6 ("a translation that does
# not get through is no translation"), then maximise xin, then ya.
import math
P = 0.82
CANDS = [
# name, coverage if understood, length, decode prob, ya
("literal", 0.40, 2, 1.0, 1.00),
("compound", 0.60, 4, 1.0, 1.00),
("marker", 0.80, 10, 1.0, 1.00),
("hyper+", 1.00, 12, 1.0, 0.95),
("word-for-word", 0.80, 16, 1.0, 0.40),
("insert", 1.00, 2, P, 1.00),
("insert+gloss", 1.00, 13, 1.0, 0.90),
]
def scores(c):
name, cov, ln, dec, ya = c
if name == "insert":
xin = dec * cov
elif name == "insert+gloss":
xin = P * cov + (1 - P) * 0.80
else:
xin = cov
da = dec * math.exp(-ln / 30.0)
return xin, da, ya
def dominated(a, others):
for b in others:
if b != a and all(x >= y for x, y in zip(b, a)) and b != a:
if any(x > y for x, y in zip(b, a)):
return True
return False
if __name__ == "__main__":
s = {c[0]: scores(c) for c in CANDS}
vals = list(s.values())
print("rendering xin da ya pareto")
for name, v in s.items():
print("{:<14} {:>5.3f} {:>6.3f} {:>6.2f} {}".format(
name, v[0], v[1], v[2], "no" if dominated(v, vals) else "yes"))
ok = [(v[0], v[2], n) for n, v in s.items() if v[1] >= 0.6]
pick = max(ok)
print("Yan Fu rule (da >= 0.6, then xin, then ya): {}".format(pick[2]))
eq = max((math.fsum(v) / 3, n) for n, v in s.items())
print("equal weights instead: {} (mean {:.3f})".format(eq[1], eq[0]))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《多语参杂语码转换信息论、词汇空缺补偿与单语种等效精准表达母典》第 7 章「原典考辨:严复三难、孔子正名、哲罗姆意译与施莱尔马赫两条路」(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-381 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026381,
author = {Fan, Yafu},
title = {Information Theory of Code-Switching, Lexical-Gap Compensation and Monolingual Precision Equivalence in Multilingual Markets},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}