Repository path: framework/v0.3/entries/HB-sound-in-prose.md · rendered 2026-09-09

Sound figures and ornament in prose: answering a figure, inventing one, compensating — and what each tells a reader about the original

Standing. Written at the project's close (2026-09-09, S257) as a consolidation of the record without the translation limb the procedure's step 6 requires — no fresh passage was translated under this entry, so it stays status: draft and its §5 Application reads "not applied". It rests on one Tier 1 precedent anchor (Knatchbull 1819, AR→EN; the lead's census, re-coded in part by blind model seats), a two-hand published panel on the Nights (Lane 1839, Burton 1885), and eight model-seat experiments — six on the lead's own renderings, one on the Arabic alone, one re-coding the anchor's English — over six Kalīla wa-Dimna chapters (AR→EN) and two classical Chinese texts (ZH→EN). Evidence classes X1a, X1b, X2 only; every count internal-judgment-only. Tier D is NOT PASSED and is EXHAUSTED (config/models.md, RS-20260906-tierD-verdict-v3): the seats were asked what they believed about the original or what relation they saw, never which rendering is better. Consolidated from v0.2 §7.12, §7.14, §7.15, §7.16, §7.17, §7.18, §7.20, §7.22, §7.23.

1. The problem

Arabic literary prose rhymes its clause-ends (saj‘), echoes one root in two shapes (jinās) and matches its members in morphological pattern (muwāzana); classical Chinese matches members by syllable count and word class; both set phrases where English has only a proposition. English has no derivational pattern to match with and no enclitic to rhyme on, so the sound does not cross, and the handbook's oldest positive instruction — compensate: put a device of your own at that place — has stood in for a decision. The evidence splits it into four: which places are the source's figures (the target set, under whose rule); whether to answer in place, invent elsewhere, or refuse, and what each tells a reader who cannot see the source; at a matched shape, what counts as an English match and how a refusal is recorded; and whether a device defended by naming its job does that job.

2. What published translators do

The regularities, as what the hands did:

1. The one published English hand from the Arabic answers about one figure in seven and no matched shape at all, and its 1818 preface said so in advance — impossible to express the sententious brevity of the Arabic with strict fidelity; muwāzana is the sententious brevity, at 0 of 18, found at 0 of 38 by a second coding hand that corrected the lead twice toward less answered — not confirming a zero (upper bound 0.193) but showing it is not the coder's (RS-20260816j §5, RS-20260821 §2–§3).
2. What crosses in a published hand is the colon structure, not the sound. In span A neither hand drops a colon (RS-20260814 §5); in span C, at the one locus where Burton lacks the second member, the source-inference dies and Lane, who keeps it, carries it (RS-20260816 §8).
3. Where a hand answers, it answers repetition first (REPEAT 25%, SHAPE 0%), and answers the enclitic chain where English supplies the answer for free. Cite 3 of 11, never 2 of 2 (anchor §3.3).

3. What this project's own practice found

1. A regime that refuses English sound can be executed (RS-20260816, «السائح والصائغ», 998 Arabic → 1,862 English under R30, three seats, 372 bodies): heard as sound-patterned at 0 of 26 cells. One substituted device at each of 13 figure loci (R31) was heard at 4 of 13; the same classes at 13 plain loci at constant length (R32) at 0 of 13 (P = 0.0478). Asked was the original doing something with sound here? the seats fired at 7 of 13 compensated figure loci (base 2 of 13) and 1 of 13 plain ones. The primary was withheld; the post-hoc account (the surviving members carry the inference) went to §7.12 as untested.
2. That account failed its first test (RS-20260816b, 402 bodies): two arms token-identical except for one word at 11 of 13 plain loci (+52 and +52 words), chiming (R34) or not (R35). Chiming: the original was doing something at 9 of 13; de-sounded twin 4 of 13 (Δ +0.3846, exact score 0.0312); base 0. The seats saw the members in both and wrote "only ordinary parallel phrasing" on the twin: the sound does the work, and it is read as the author's. At figure loci a genuine compensation scored 8 of 13; at plain loci the invention scored 9 of 13 (gap −0.0769): a source-blind reader cannot tell a compensation from an invention.
3. The target set is a count under a rule (RS-20260816c, «الناسك والضيف», 21 sites): shown the Arabic alone, three seats separated the inventory's figured from unfigured sites by 0.1886 against a 0.40 gate, agreeing on 5 of its 10 figures. Source-blind, the supplied device at plain sites produced belief in an original figure at 0.7500 against 0.3750 for the translator's own plain wording (+0.3889); readers of the Arabic endorse those sites at 0.2941; at figure sites removing the device moved belief −0.0167, the plain wording still repeating what the Arabic repeats. Limb T-kalila-ibn-urs-R36-v1 (570 → 1,075 words): 19 figures enumerated, 19 answered, 0 devices elsewhere, +35 words (3.3%); the ornamentalist T-kalila-nasik-R34-v1 spent +34 words (6.1%) on 10 answers and 12 inventions.
4. The readers' set is determinate; what was rule-dependent was one clause of the project's own rule (RS-20260816h, «اللبؤة والإسوار», 599 → 1,078 under R37, 28 loci): asked would a translator be right to treat this as a sound effect his English ought to answer? the seats owed 1.0000 where both rules admit and 0.0000 where neither does (24 of 24 bodies); by their own labels ROOT 1.00, SHAPE 1.00, RHYME 0.97, REPEAT 0.67, PARALLEL (grammar or sense, not sound) 0.00. Rhyme on an enclitic pronoun — excluded by name — was owed 9 of 9; the exclusion is struck. The primary (readers side with the sound rule by ≥ +0.40) failed at +0.2778. The union of form rule and sound rule holds all 19 owed places plus 4 — over-answering about one place in six, about 4.5% in length.
5. Answering under the repaired rule costs about 4% (T-kalila-qird-R38-v1, 1,201 → 2,220, RS-20260816j): 54 loci, 9 existing only because the exclusion was struck; all answered for +98 words (4.4%); the restored enclitic loci the cheapest class (nine loci, fourteen words).
6. Matched shape is unattempted in English, not inexpressible (T-kalila-fanza-R39-v1, 1,284 → 2,348, RS-20260821): under a clause forbidding a silent reduction — MATCHED or IMPOSSIBLE at every shape locus, English quoted, no third verdict — 24 MATCHED, 0 IMPOSSIBLE, 16 by English morphology, 4 free; ratio 1.829 against 1.849 for the same hand's previous chapter without the clause.
7. A matched shape has no plain version to be checked against (RS-20260821b, 劉基《賣柑者言》 under R40, 312 characters → 449 words, 12 loci 12 MATCHED, all by frame; plus the Arabic chapter's refusals): at 22 loci the wording recorded as the plain wording refused still echoed in form at 16 of 22 (blind screen) and said something different at 10 of 22 (parity screen, 4 of 4 planted errors caught); 2 survived against a floor of 10, primary withheld before dispatch. Prompted whole-passage readers recovered the built matches at 0.879 and a figure at the same places in the "plain" version at 0.845. Method note (bqr).
8. A translator's account of a device's job held at step 1 for manner and ornament (withheld at step 2, §7) and failed for attitude (RS-20260816g, 蒲松齡〈種梨〉, T-zhongli-R04-v1, 686 English words, one hand writing translation, labels and plain alternatives): declared-probe movement 0.8642 at nine job loci against 0.0741 at three loci declared inert (P = 0.00455). MANNER 3 of 3, ORNAMENT 3 of 3, STANCE 0 of 3: the attitude survives in the plain version at 0.67–1.00 because it is in the proposition, and the device was instead telling the reader the Chinese had a set phrase (set-phrase probe +0.333 to +0.778; the Chinese stretches are set forms).
9. "Write the passage without the device" is not a well-defined instruction (RS-20260820b, same text, an independent hand writing the replacements blind to the labels): in great mouthfuls → greedily; Δmanner 0.000 at all three manner loci; at the rhetorical question 18 bodies differed on nothing. The plainer version read as more judging (Δstance −0.444, −0.222: "greedily tells how and judges"). The attitude exception held at a set idiom (怫然: set-phrase belief fell +0.889, attitude stayed) and not at a form of address (your worship → you: attitude fell 0.889 → 0.667). In this run's own data the inert loci moved less (0.1111 vs 0.4603; descriptive — cross-run comparison is barred by the design).

4. The options

The evidence prices five moves and ranks none; the device-function findings (§3 items 8–9) go straight to §5 items 6–7:

• Refuse sound everywhere (R30). Executable and measurable as executed. The source's matched members still cross where English cannot render the parallel without them (2 of 13 figure loci read as the original's with no device), and readers infer the figure from surviving colon structure in published hands (§2 item 2).
• Answer in place (R31, R36, R38). Read as the original's at 7–8 of 13 figure loci against 2 of 13 with all sound refused (RS-20260816 §7, RS-20260816b §7); costs 3.3–4.5% in length when every figure is answered. On another chapter, removing the device at a figure site moved belief −0.017, because the plain wording there still repeated what the Arabic repeats (RS-20260816c §7.1). Published practice does this at 15% of loci.
• Invent elsewhere (R34). At the words an ornamentalist takes, read as the original's at 9 of 13 places the source is plain — indistinguishable from compensation (RS-20260816b §7) — and, on another chapter, at 0.75 against the 0.29 readers of the Arabic endorse (RS-20260816c §7). Not measured as better or worse English; measured as telling the reader something about the author the translator made up — the unlicensed-mark diagnostic (HB-emphasis-and-typography's count) in its sharpest form, the mark read as a licence.
• Declare the rule that fixes the target set, then answer the union (R37). Misses nothing the readers owed, over-answers about one place in six; the sound rule alone misses 8 of 19, the form rule alone over-claims 4. Readers of the Arabic decline outright a parallel in grammar or sense but not in sound.
• At matched shape, declare the English resources and forbid a silent reduction (R39, R40): suffix, measure, frame, inflection; MATCHED or IMPOSSIBLE with the English quoted. No length cost on the one chapter measured. Its audit record — the plain wording refused — is not plain and cannot serve as a control.

The set depends on whether the reader can see the source (untested — §6), which rule the reader of the source applies (item 4), and whether the figure is sound (removable) or position (not, item 7).

5. Guidance

For a translator

1. Enumerate the source's figures from the source before any English exists, under a written rule, and say which — rhyme however produced (enclitic included), root in two shapes, matched pattern, repetition; not a parallel in grammar or sense without sound. Expect a reader of the source to owe the union of a form rule and a sound rule, and over-answering at about one place in six if you answer it. — evidenced (AR→EN, two chapters' readers; three chapters' inventories under the repaired rule, R37–R39).
2. Do not put a sound device where the source has none unless you accept that a reader who cannot see the source will read it as the author's. The inference carries no information about where the source's figures are, and your own plain wording lands the reader roughly where readers of the source sit. — evidenced (AR→EN, two chapters, three model seats).
3. Where you answer in place, expect it to cost words; what it adds to a source-blind reader's belief depends on the plain alternative — +0.38 against a rendering that refuses all sound (RS-20260816 §7), −0.017 against a plain wording that still repeats what the Arabic repeats (RS-20260816c §7.1). Budget about 4% for answering every figure. — evidenced (AR→EN, three chapters for cost, two for belief).
4. Keep the count and adjacency of the source's members even when you refuse the sound. Hands that drop no colon carry the inference; the one dropped member killed it. Craft, not the withdrawn §7.12 explanation. — evidenced (AR→EN, Lane/Burton, 8 loci), weakly.
5. At a matched shape, write down before translating what you will accept as an English match (suffix, measure, frame, inflection), and at every such locus record either the match or the reason you could not — never "English does not afford it" decided locus by locus. Separate the free matches. The frame is the cheap half and most of what a prompted reader registers, the form the expensive half; only the second is optional. — evidenced (AR→EN one chapter, ZH→EN one text; one hand).
6. Do not test a figure by writing the passage plainly and trusting your own plain version. Write the replacement down and show it to someone; where two versions of "plain" disagree, that is the finding. At a matched shape there is no plain version at all. — evidenced (ZH→EN, two texts).
7. Trust your sense that a choice is doing nothing; distrust your sense of what a stance device does. Here a set idiom or a form of address carried the reader's belief in a source set phrase, not the attitude, and the plainer adverb judged where the figure showed. — evidenced (ZH→EN, one tale, one translator).
8. When answering figures from Arabic, say that the available published precedent does not — 15% of loci, no matched shape — and that its translator called the loss taste. A fact to issue with the instruction, not an argument against it. — evidenced (AR→EN, one hand, 1819).

For a pipeline

1. Enumerate figure loci from the source under a declared rule (union of form and sound; enclitic rhyme admitted; parallelism without sound excluded), frozen before rendering. — evidenced (AR→EN, R37/R38/R39) as executed by runs; reliability in a second hand untested.
2. Classify each locus by ground (REPEAT / ROOT / RHYME / SHAPE) and, for SHAPE, by English resource available. — evidenced (AR→EN, R38, R39).
3. Set the policy parameter: refuse (R30), answer in place, nowhere else (R36/R38), or ornamentalist (R34), with each policy's reader-belief consequence declared as a cost. — executed as regimes; selection as a parameterised choice untested.
4. Render, emitting MATCHED or IMPOSSIBLE with the English quoted at every SHAPE locus, no third verdict. — evidenced (AR→EN R39, ZH→EN R40, one hand).
5. Check the device count against the enumerated loci: every device at a non-locus is a supplied mark, flagged read-as-the-author's; every IMPOSSIBLE is a visible refusal. Do not build a plain-version contrast as the check. — untested as an automated step.
6. Do not use a source-blind was the original doing something here? as a fidelity check: it tracks the translator's hand. — evidenced (AR→EN), negative.

Human entry points. Step 1's rule and step 3's policy are the decisions a person may take over; step 3 is a declared-purpose decision (wiki/goodness-senses.md). Without one, default to answer in place, nowhere else under the union rule (cost measured, about 4% length; reader-side effect least misleading) and MATCHED/IMPOSSIBLE at every shape locus.

Application. Not applied: written at close-out without a translation limb.

6. Not evidenced, and open

• One work, one hand, one recension, 1819 English for the anchor; one pair for the Nights panel; internal experiments mostly on the lead's renderings, three fixed model seats, n = 9 to 28 loci per arm. Nothing reaches human readers, a contemporary hand from Arabic, or other pairs. Four of Knatchbull's 18 matched-shape loci and all 54 non-shape calls are unchecked by a second hand; the relation menu is unpiloted; its loose rung (same part of speech, same length) must not be reported as a match.
• Whether a reader who can see the source tells an invented ornament from a compensation, and counts it against the translation, was never measured — withheld behind a failed gate and never re-registered (ARM-invented-ornament's constituting question).
• Detection is prompted. No page says an unprompted reader notices a matched shape, a chime, or a supplied device (RS-20260821b §6).
• The R39 resource list has one hand's use only: frame at 12 of 12 on Chinese, mixed on Arabic — an observation, not a prediction.
• Open: the distance at which an unchanged repetition stops being owed (RS-20260816h §5). Open: the threshold gap — a device too slight to be called ornament still licensing a belief about the author (RS-20260816b §8). Open: a contrast varying which resource carries a match with position held fixed (RS-20260821b §8; ARM-radif built it for Persian line-ends, family C). Open: a second hand applying the two written inventory rules blind (ARM-invented-ornament). Open, inherited: whether adverbial reduplication surviving once and failing once is a property of the words or of one ear (HB-register §6, Verga IT→EN); whether a figure refused for economy against overuse rather than for cost to sense is a fourth refusal-currency for the Gulistan tajnīs grading (ES-20260905-craft-report-gulistan D78, FA→EN — one instance; watch for a second).

7. Sources consumed

• v0.2 §7.12 (S191, RS-20260816-answering-figure): the refusing regime; relocation vs constant-length invention; the published panel — item 1, §2.
• v0.2 §7.14 (S196, RS-20260816b-invented-figure): the sound does the work; compensation and invention indistinguishable blind — item 2.
• v0.2 §7.15 (S197, RS-20260816c-checked-ornament): the target set under two rules; the device manufactures belief at plain sites; R36 — item 3.
• v0.2 §7.17 (S202, RS-20260816h-target-set): the readers' set determinate; exclusion (a) struck; answer the union — item 4.
• v0.2 §7.18 (S203, RS-20260816j-published-figure, A-knatchbull-kalila): census, preface, R38 — §2, item 5.
• v0.2 §7.22 (S208, RS-20260821-matched-shape): the zero re-coded; R39 — §2, item 6.
• v0.2 §7.23 (S209, RS-20260821b-matched-heard): no plain version of a matched shape; R40 — item 7.
• v0.2 §7.16 (S201, RS-20260816g-device-function) and §7.20 (S206, RS-20260820b-device-function-2): the subtractive test and its failure — items 8–9.
• Not in the row, consumed by headline only: RS-20260814-saj-carriage, RS-20260814g-supplied-sound, RS-20260815-supplied-sound-confirm (the Nights Lane/Burton panel) — §7.12 cites them; no v0.2 section or v0.3 index row owns them, and HB-rhymed-prose tags the first as family L. The close-out should assign them in one place.
• Withdrawn along the way: §7.12 item 3 (the members carry the inference; the sound you add is heard as yours) and its heading's sound is the wrong channel — withdrawn by §7.14 (RS-20260816b). §7.15's claim that the source's figures are not a determinate set — narrowed to say which rule fixes your target set by §7.17 (RS-20260816h), which also withdrew two of RS-20260816c §5's three "rules" as single-locus readings. The standing inventory rule's exclusion (a) — struck by §7.17. §7.16 item 1 (the subtractive test as an instruction) — withdrawn by §7.20 (RS-20260820b); §7.16 item 3 (manner and ornament three for three) — withheld, two loci gated out; §7.16 item 4's attitude exception — restated as locus-specific. The lead's two credited Knatchbull matched-shape answers in the third chapter — overturned by RS-20260821 §3; the "2 of 2" enclitic corroboration — scoped to 3 of 11 by RS-20260816j §6. §7.22's implied audit (compare the match against the plain wording refused) — withdrawn by §7.23 (RS-20260821b). The relation menu's loose rung as a match — retired by RS-20260821 §6. RS-20260814g's headline (Burton supplies sound at 4 of 8) — not replicated at magnitude by RS-20260815-supplied-sound-confirm.