What does Compliant Rewriting of Absolute Terms conclude, in numbers, about “Keeping the Meaning: Why Surface Similarity Rewards the Violation”?
Answer: Chapter 5 shows that surface similarity rewards the violation; a score built on proposition slots has a rank correlation of minus 0.949 with edit-distance similarity.
Scope of the numbers: Its lexicons, copy, facts and parameters are illustrative; all calculations are generated by its programs; it is not individual legal advice and supplies no way of slipping past platform review. Marked boxes separate published evidence, runnable Python calculations (18 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it.
original: ben dian zhen pi bao zui jia xuan ze rewrite len levenshtein dice jaccard R1 delete 5 0.556 0.667 0.556 R2 comparative 10 0.500 0.471 0.357 R3 rating fact 13 0.385 0.400 0.400 R4 repurchase fact 13 0.385 0.400 0.294 R5 absolute synonym 11 0.636 0.556 0.538 highest Levenshtein similarity: R5 absolute synonym highest bigram Dice: R1 delete the only non-compliant rewrite is R5
Show the full program
# Chapter 5. p51: surface similarity rewards keeping the violation.
# Model: three surface measures between an original line and five
# rewrites, all on syllables (one syllable = one Chinese character):
# normalised Levenshtein similarity 1 - d / max(len), Dice coefficient
# of syllable bigrams, and Jaccard index of syllable sets. Lines and
# figures in them (ratings, review counts) are illustrative.
ORIG = "ben dian zhen pi bao zui jia xuan ze"
REWRITES = [("R1 delete", "ben dian zhen pi bao"),
("R2 comparative", "ben dian zhen pi bao geng shi he tong qin"),
("R3 rating fact",
"ben dian zhen pi bao ping fen 4.8 gong 1236 tiao ping jia"),
("R4 repurchase fact",
"ben dian zhen pi bao jin jiu shi tian fu gou lv 0.18"),
("R5 absolute synonym",
"ben dian zhen pi bao wu ren neng bi xuan ze")]
def lev(a, b):
prev = list(range(len(b) + 1))
for i, x in enumerate(a, 1):
cur = [i]
for j, y in enumerate(b, 1):
cur.append(min(prev[j] + 1, cur[j - 1] + 1,
prev[j - 1] + (x != y)))
prev = cur
return prev[-1]
def bigrams(t):
return [tuple(t[i:i + 2]) for i in range(len(t) - 1)]
def dice(a, b):
ba, bb = bigrams(a), bigrams(b)
pool = list(bb)
common = 0
for g in ba:
if g in pool:
pool.remove(g)
common += 1
return 2.0 * common / (len(ba) + len(bb))
def jaccard(a, b):
sa, sb = sorted(set(a)), sorted(set(b))
inter = [x for x in sa if x in sb]
union = sorted(set(sa) | set(sb))
return len(inter) / len(union)
o = ORIG.split()
print("original: " + ORIG)
print("rewrite len levenshtein dice jaccard")
rows = []
for name, line in REWRITES:
t = line.split()
lv = 1.0 - lev(o, t) / max(len(o), len(t))
rows.append((name, len(t), lv, dice(o, t), jaccard(o, t)))
print("%-20s %4d %.3f %.3f %.3f" % rows[-1])
top = max(rows, key=lambda r: (r[2], r[0]))
print("highest Levenshtein similarity: %s" % top[0])
top = max(rows, key=lambda r: (r[3], r[0]))
print("highest bigram Dice: %s" % top[0])
print("the only non-compliant rewrite is R5")Single source and citation
Fan, Y. (2026). Compliant Rewriting for Platform-Sensitive Terms: Absolute-Term Automata under the Advertising Law, Meaning-Preserving Rewrites and Conversion Elasticity (Version 1.0.0), Chapter 5: Keeping the Meaning: Why Surface Similarity Rewards the Violation. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-274 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026274,
author = {Fan, Yafu},
title = {Compliant Rewriting for Platform-Sensitive Terms: Absolute-Term Automata under the Advertising Law, Meaning-Preserving Rewrites and Conversion Elasticity},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · The Risk Ledger of an Absolute Term: How Much Conversion Must One "Best" Buy?
- Chapter 2 · One Pass for Every Term: the Aho-Corasick Multi-Pattern Automaton
- Chapter 3 · False Kills and Misses: Exempt Contexts, Audit Additions and Audit Sample Sizes
- Chapter 4 · Choosing Rewrites under Constraints: Evidence, Re-scanning and Title Length
- Chapter 6 · Conversion Elasticity: How Much Does a Rewrite Move Conversion, and Can a Test Tell?
- Chapter 7 · Eastern and Western Sources: "If Names Are Not Correct" and Orwell's Rules for Writing
- Chapter 8 · Closing the Loop: a Compliant-Rewrite Library for the Persuasive-Copy Checker, and a Non-Inferiority Test