Repository path: workshop/translations/maigan/R40-v1/translation.md · rendered 2026-09-09

劉基《賣柑者言》 — The Words of the Orange Seller, whole, under R40

Source. 劉基 Liu Ji (1311–1375), 《賣柑者言》, Ming, written before the founding of the dynasty he helped to found. 312 Han characters, 22 sentences as numbered in ../copy-text-adopted.txt. Copy-text and its nine variant readings: ../collation.md. Public domain; both witnesses freely reachable, both read whole.

Regime. R40 (workshop/regimes/R40-matched-shape-cjk.md), which is R39 moved to a Chinese source: the source-side class is redefined for a language that matches by measure rather than by derivational pattern, and the English resource table (A)–(D) is imported verbatim and unchanged, so that a verdict here and a verdict in T-kalila-fanza-R39-v1 are about the same English question.

Inventory frozen at 771a2315, before the rendering file existed — inventory-frozen.md, 12 loci. The exact strength of that freeze, which is weaker than R39's, is stated on the inventory page and is not restated as stronger here.

1. The rendering

[1] In Hangchow there was a seller of fruit who had the art of keeping mandarins: they passed through cold and heat and did not rot. [2] He would bring them out gleaming, of jade in substance and of gold in colour. [3] Set out in the market, they fetched ten times the price, and people fought to buy them. [4] I bought one, cut it open, and something like smoke struck at my mouth and nose. [5] I looked inside: dry, like rotted floss. [6] I thought this strange and asked him: "These that you sell to people — are they to fill the dish, to serve the rite, to feast the guest? [7] Or are they to make a show of the outside and gull the ignorant and the blind? [8] It is too much, this cheating!"

[9] The seller laughed and said: "I have followed this trade for years now; I depend on it to feed this body of mine. [10] I sell them, they take them, and there has never been a word about it — and are you the one man it does not satisfy? [11] The world is not short of cheats: am I the only one? [12] You have not thought about it, sir. [13] Consider those who wear the tiger tally, who sit on the tiger skin: so bold in the look of them, the tools of shield and wall — can they truly pass on the arts of Sun and Wu? [14] Those who rear the great cap, who trail the long sash: so grand in the look of them, the tools of court and hall — can they truly set up the works of Yi and Gao? [15] Thieves rise and they know not how to check them; folk starve and they know not how to save them; clerks cheat and they know not how to curb them; law rots and they know not how to mend it. They sit there wasting the grain of the public granaries, and are not ashamed. [16] Look at those who sit in the high hall, ride on the great horse, swill down the rich wine, gorge on the fat meat: which of them is not of the towering sort, that a man must fear; of the blazing sort, that a man must copy! [17] And where will you go and not find gold and jade without, rotted floss within? [18] Now this you do not look into, and you look into my mandarins!"

[19] I was silent and had no answer. [20] I withdrew and thought over what he had said: it was of a piece with the drollery of Master Dongfang. [21] Was he a man hot against the age, hard against the wrong? [22] And had he put it on the mandarins, to make his satire?

Four proper names, glossed here and not in the text (R40 rule 2 forbids adding to the sense, and a gloss inside the sentence would be an addition): 孫、吳 are Sun Wu, author of the Art of War, and Wu Qi, the Warring States general — bywords for military competence. 伊、皋 are Yi Yin and Gao Yao, ministers to Tang and to Shun — bywords for administrative competence. 東方生 is Dongfang Shuo (c. 160–93 BC), the Han courtier whose 滑稽 jesting carried remonstrance inside a joke; 「類東方生滑稽之流」 places the orange seller in that line, and is the sentence on which the whole essay turns from anecdote to satire.

2. The census — R40 rule 3, all twelve

Columns: the English written · the plain wording refused at the same span (rule 5, written at the same sitting) · the resource under R39 §2's table · whether a resource other than (C) was available, and why it was not taken.

2.1 The measures, so the match is checkable and not merely asserted

R39 §2's (C) requires a matched syntactic frame with matched member length. Member length is counted here in English syllables, and the count is printed so a reader can disagree with it.

Every locus is exactly matched in measure. That is a fact about this rendering, not a claim about its quality, and the syllable counts are the thing to check first if the rendering is doubted.

2.2 The four numbers this census exists to print

IMPOSSIBLE was never written, and the honest report is that the verdict was available and was not reached — as at T-kalila-fanza-R39-v1 §4.2. Unlike that chapter, no locus here came out free: at all twelve the plain wording actually refused is unmatched, so all twelve are usable as paired material, where four of the Arabic chapter's twenty-four were not.

2.3 The resource, and the fact that it is one resource

All twelve loci were answered with resource (C), the matched syntactic frame. In the Arabic chapter the same hand, working from the same table, used (A) six times, (B) six times, (C) twice and (D) ten times.

This was noticed while rendering. It was not predicted beforehand, and it is not reported here as a prediction that held. The reading that suggests itself — that a source which matches by measure and construction pulls an English hand toward English's measure-and-construction resource, while a source which matches by derivational pattern pulls him toward English's suffixes — is a conjecture formed after the fact, on one text, one language pair and one hand. It is written down so that a later session can register it before rendering a third source, which is the only way it could become evidence.

Where a non-(C) resource was genuinely available and was refused, with the reason:

• M1. (D), matched inflection at member-end, was available as jade-fleshed and gold-skinned. Refused under rule 2: 質 is substance and 色 is colour, and fleshed / skinned both state where in the fruit the property sits, which the Chinese does not. The only addition-free (D) here would need jade-substanced, which is not English.
• M6. (D) was available as world-hating and vice-loathing, and it is a real match. Refused on craft: 憤 is heat rather than hatred and 邪 is the crooked rather than vice, and the compound-adjective mannerism is worse English than the frame. The option is named here so that a reader can weigh the refusal instead of taking it on trust.
• M12. (A) was reachable as gilded without, rotted within. Refused under rule 2: it drops 玉, the jade.

At the other nine loci the members are phrases rather than words, and (A) and (D) — which are suffix resources and want word-sized members — had nothing to attach to.

2.4 What the regime cost in length, measured

There is no unmatched control rendering of this text, so no cost-of-forcing figure is available here and none is claimed. The comparable figure exists only for the Arabic chapter, where the forced rendering ran at 1.829 English words per Arabic word against 1.849 for the same hand without the clause — i.e. no cost. Whether the same holds for Chinese is untested, and saying so is cheaper than implying it.

2.5 Where the rendering deliberately breaks

Sentence 15 sets four members of six characters marching and ends with one of eight — 坐糜廩粟而不知恥. The inventory declines it, and R40 rule 4 forbids a matched figure outside the inventory, so the English breaks measure there too: after four clauses of exactly ten syllables, They sit there wasting the grain of the public granaries, and are not ashamed. The break is the sentence's argument — the four failures are things not done, and the fifth is a thing done — and a hand who tidied it into the pattern would be improving Liu Ji.

3. Translator's log — decisions

Frozen with the rendering, before any evaluation of it was designed (charter §3, rule 8). What was hard, what was live, what was chosen, what was lost. No quality claims; the one evaluative sentence in this log is flagged in §3.4.

3.1 The two hardest loci

M8, 洸洸乎干城之具也 ／ 昂昂乎廟堂之器也. The difficulty is not the metaphors — 干城 shield and wall is from the Odes and 廟堂 temple and hall is the seat of government, and both survive literally in English. It is 洸洸 and 昂昂: two reduplicatives whose whole work is to sound like swagger while meaning almost nothing. Three options were live. Martial and majestic gave the sense and killed the measure (2 syllables against 3). Bold, bold and high, high reproduced the reduplication and read as verse. What was chosen puts the emptiness where the Chinese puts it — so bold in the look of them says that the boldness is in how they look — at the cost of six syllables per member that the Chinese does not spend.

M5, the four-member series. The trap is that English will take three of these easily and balk at the fourth: sit in the high hall, ride on the great horse, drink the good wine — and then 飫肥鮮, sated on fat and fresh, has no English of the same measure that is not either clinical (eat one's fill of rich food) or comic. Swill down and gorge on were chosen together, as a pair, because the contempt in this sentence is Liu Ji's and not an English addition: 醉 and 飫 are both words for taking more than is needed. What is lost is 鮮, the freshness; fat meat keeps the fat and drops the fresh.

3.2 Two words that recur, and were made to recur

敗絮 — rotted silk wadding — appears at [5] and again at [17], and the second occurrence is the essay's most quoted phrase. The English repeats rotted floss in both places. Floss rather than wadding was chosen at [5] because of [17]: the famous line wants a monosyllable to sit against gold and jade, and a hand who wrote wadding at [5] and floss at [17] would be hiding the repetition that the Chinese makes.

察 likewise recurs inside [18], 不察 … 察, and the English repeats look into. This is repetition, not matched shape; the inventory declines it under criterion (ii), and repeating it in English is not a device placed outside the inventory but the ordinary business of rendering a sentence that repeats a word.

3.3 What the sentence-final particles cost

The Chinese ends five sentences in this text with 耶, 哉, 也, 乎 — particles that mark a question as rhetorical, or an assertion as exclamatory, and that English has to buy with word order or with punctuation. [8] 甚矣哉，為欺也 is the extreme case: an exclamatory 甚矣哉 fronted before the thing it exclaims about. It is too much, this cheating! buys the fronting with a right-dislocation, which is the nearest English construction and is more colloquial than the Chinese.

3.4 One evaluative sentence, flagged

internal-judgment-only: true — the uniformity of resource (C) across all twelve loci makes me suspect that I reached for the frame first and tested the other resources afterwards, rather than weighing four resources at each locus. §2.3 names the three loci where another resource was genuinely reachable; it cannot show what a hand who had begun from (A) would have found.

4. Contamination

contamination: none — measured, not asserted.

tools/dependence_check.py, run after the rendering and the inventory were committed at 771a2315, against the only complete English rendering of this essay this project can freely reach: an anonymous 2013 translation, Discourse with the Mandarin Vendor, at vacantmountain.wordpress.com. 0 shared 7-grams, 0 shared 12-grams, 0 shared 15-grams, longest common run 5 tokens — people fought to buy them — on 449 tokens against 416. dependence.json in this directory.

The rendering was written from the Chinese alone. No English version of this text was opened until after the freeze commit; the search that located the comparator was run before translating and returned titles and URLs only, and the comparator's text was not fetched until the rendering existed.

What is not claimed. The Words of the Orange Seller is an anthology piece — it stands in 《古文觀止》, the standard classical reader — and 「金玉其外，敗絮其中」 is a proverb this hand has certainly met before. This measurement is against one comparator, and other English renderings exist in print that could not be reached. none here means measured clean against the reachable comparator, and nothing wider.

Why the rendering would be usable either way. The design it feeds does not require independence from any published version — its items are matched and unmatched wordings of this English, and no comparator enters the experiment. That is what CLAUDE.md's standing rule turns on: the constraint binds where a design's validity turns on independence, and this one's does not.