Repository path: framework/v0.3/entries/HB-radif-and-refrain.md · rendered 2026-09-09

When every rhyming line of the source closes on the same word (the Persian radif), and English can end a clause only on some words

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'. Evidence: two published hands read whole against the Persian (Leaf 1898, 28 odes; Payne 1901, 199 odes — X1a/X1b, carriage coded mechanically), a 495-ghazal census of the source (X2), eight lead renderings under frozen logs (X2 craft, internal-judgment-only). Two model-seat measurements — a preference level, a well-formedness rating — are panel-scored (X3), inadmissible, and appear only as the instrument's behaviour. Tier D is NOT PASSED and EXHAUSTED (config/models.md, RS-20260906-tierD-verdict-v3): nothing here says one rendering is better than another on a jury's word. Consolidated from v0.2 §7.36, §7.40, §7.41, §7.42, §7.46, plus §7.32's struck sentence, §7.43.1/§7.43.3, §7.47.2 and §7.49.1's radif stratum — sections the row's titles hid (§7).

1. The problem

A Persian rhyming position is […qāfiya][radif]: a word or phrase repeated identically at every rhyming position, the rhyme on the word immediately before it, nothing between. Persian puts the verb last, so a clause can end wherever the poet likes and the radif can be any word: 305 of Hafez's 495 ghazals carry one (0.616), the commonest being finite verbs, copulas and the object marker — است, بود, دارد, کرد, را (RS-20260829-radif-hands §3). English ends a clause on an object, a complement or an adverbial, and its genitive puts a preposition and an article where Persian puts the bearer. A translator must decide, before the first line, whether one English word can stand last at every position and serve every collocation the poet gives it, and — where it cannot — which half to keep: the repeated tail or the chime before it. The same device arises in rhymed prose (Sa'di's Gulistan), where no rhyme is hunted and the repetition alone is at issue. Refrain occurs in the record only as Leaf's and Bell's word for the radif; a refrain proper — a repeated line or stanza — was never measured (§6).

2. What published translators do

The regularities:

1. Every cell the record printed or conjectured impossible was carried by a published hand, except one. Transitive finite verb: Payne's highest cell. Pronoun in ezāfe: Leaf XXV. Negated finite verb: Leaf XVI. Bare copula: Payne. The case particle را is refused by every hand that had the chance — five by Payne, one by Leaf, one enumerated refusal by the lead: seven refusals, zero carriages (RS-20260831b-radif-hands §4).
2. The carrying move is visible on the page, and it is one move: inversion. Object, complement or adverbial before the verb so the radif can stand last ("Of us many a year made"). The plainer hand uses it too (Leaf XVI). Where the radif governs its bearer, Leaf re-attaches (thy → for thee), paying in the relation; where one English verb will not serve every collocation, Payne nominalises and fronts (made good actions her practice, ode CXVI) (RS-20260829-radif-hands §4; RS-20260831b-radif-hands §6). Why Payne carries more than Leaf is not settled: register withdrawn (§7), room-in-the-line dead (median hemistich 15 syllables against 14, RS-20260901-inversion-habit), habit withheld by its gate.
3. The radif's grammatical class explains little of whether a hand carries it. Payne's three "unreachable" cells pooled 33 of 46 (0.717) against 27 of 34 (0.794) elsewhere, difference 0.077, interval [−0.119, 0.253] — underpowered, declared in advance (RS-20260831b §3).
4. English manufactures a radif the Persian metrics do not record, for one reason. A language that marks person, tense or the copula on the word forces English to externalise the mark as a free word: nine of nine in Payne, two of two in Leaf (RS-20260831b §5; RS-20260830 §7).
5. A carried radif does not take the sense a rhyme displaces. On the 14 radif renderings the trailing words are stock lexicon (is, of the friend); the displaced sense never lands there, and the chime sits before them (RS-20260908-chime-slot §2).

3. What this project's own practice found

Eight renderings under frozen logs, all internal-judgment-only.

1. Six Sa'di ghazals whole under R51 (39 bayts, 45 rhyming positions; T-saadi-ghazals-radif-R51-v1, suspected: 0 shared 7-grams against King 1926 whole, the same poems unlocated). A prediction table from the radif's part of speech, written before any line, came out 6 of 6: آنجاست is there HELD 8 of 8; برخاست arose HELD 8 of 8 with two sense-strains; افتاده‌ست SHORT (radif 7 of 7, chime 3, 4 refused); انداخت cast UNREACHABLE 0 of 8; مرا radif 7 of 7, chime 0 of 7; ترست SHORT 4 of 7 (RS-20260826b-radif §2). The rule read off it was later withdrawn for the transitive cell (§7); the craft survives: a Persian radif repeats a form and lets the sense move under it (برخاست rose up at six positions, departed at two) while English repetition drags the sense along, so the hand buys one sense and pays at the others; a comparative radif supplies an English chime unbidden (every comparative ends on the same unstressed syllable; the no-chime arm graded NEAR 3 of 3); the rhyme family binds first — ۴۹ came out at seven distinct bearers in eight positions, Sa'di's own ratio; Sa'di repeats his own bearers (ایطا) in four of the six.
2. Three Sa'di ghazals twice each under R54 (21 bayts, 24 positions; T-saadi-ghazals-forced-R54-v1, suspected; ۹۴'s arms share a 19-token run in the a-hemistichs). All three radifs pass the part-of-speech test and fail: دوست and تو نیست stand in a genitive to their bearer, and the English of-genitive puts of the in the rhyme slot; اوست leaves the slot free and the eight bearers share no English rhyme family (best 3 of 8). REP (tail kept) graded IDENTICAL at 28 of 28 pairs per poem; CHI (chime kept) 68 STRICT + 16 NEAR of 84. The costs are opposite: REP forces an unstressed line-ending at every position in two of three poems and cost no logged strain; CHI ends every line on a stressed rhyme and cost three logged strains (metonyms, a named emotion) plus a fourth the equivalence gate found unlogged — an added assertion; REP alone reproduces Sa'di's ایطا, CHI silently corrects the poet. The family that worked held a word for the hardest position (RS-20260828-forced-half §2, §4). Seat record, X3, inadmissible: three model seats took CHI 170 to 100 across five disclosure conditions, no disclosure moving it past +0.111 of 2.000; the page calls it a policy comparison confounded by rhythm and says nothing about a reader.
3. Hafez ۱۳۱ and ۴ under R55 (T-hafez-ghazals-radif-R55-v1, none). ۱۳۱, radif کرد with the qāfiya as its object at every position (غارت کرد made a plundering): carried 9 of 9, chime 9 of 9, no filler, no inversion. ۴, radif را: UNREACHABLE with four options enumerated before any English (no object-marking word of any class; a preposition adds a relation; a vocative cannot stand last; a discourse tag installs a speaker); chime HELD 5 of 9, SHORT 4 (RS-20260829-radif-hands §9).
4. Hafez ۱۳۰ under R55 in a declared plain register (T-hafez-sahar-bolbol-R55-v1, none): the same radif کرد, the object relation at 4 of 11; graded UNREACHABLE, the log naming a lexical wall (neither do nor make collocates with all eleven nouns); chime HELD 7 of 11. Payne's CXVI is this poem and carries made at 9 of 9, including two of the three positions the log named as blocked. The obstacle was the declared register, not the lexicon; the frozen log stands unamended (RS-20260831b-radif-hands §6).
5. Hafez sh124 and sh118 under R56 (T-hafez-darad-R56-v1, none): 18 bayts × four arms (direct/inverted × plain/archaic), the archaic arms generated by script so only the named tokens differ. Seat record, X3: within-idiom well-formedness fell by 2.407 of ten under inversion in plain lexis and 1.926 in archaic; 101 of 106 forced choices rejected the inverted order; the inverted archaic line sat below the plain direct one. Reported only for what it withdrew (§7); no better/worse claim. The log's end-weight conjecture (D8) is registered and untested (RS-20260829b-inversion-price §7–§8).
6. Hafez ۷۷ under R58 in declared plain modern English (T-hafez-daasht-R58-v1, none): the finite transitive radif داشت carried as had at 9 of 9, inversion LOCAL 6 · SPREAD 3 · REBUILT 0 — only the verb moved at six positions, the predicate was reshaped at none; §7.46.5's test (one English collocate for all nine complements) was applied in advance and passed; chime HELD 3 of 9, REFUSED 6. An existence result about a register, not a rate (RS-20260901-inversion-habit; v0.2 §7.47.2).
7. Hafez ۱۸۷ under R57, Leaf's contract (T-hafez-bekonad-R57-v1, none): بکند → will do at 8 of 8 at fifteen syllables a line — the radif survives the metre; the losses are nouns and numerals, none caused by rhyme or radif (RS-20260830-leaf-contract; v0.2 §7.43.3).
8. Rhymed prose: Gulistan bāb 2 under R05 (T-gulistan-bab2-R05-v1, contamination: high). Rule V18 (gulistan-bab2/register.md): a radif of lexical material is carried as the same repeated English phrase; a radif of grammatical material is not, and the loss is declared — گو مباش say, let there be none whole and free at b271–b272 and b294; را (b308–b310) and کن (b316–b317) not, their qāfiya going with them. Nothing was hunted: the English says the phrase twice because the Persian does (D48). Span E's D62: a single lexical word, گیر take, closes both hemistichs of b384 after qāfiya on رضا/خدا and is held at both English line-ends by inverting the clause — V18 is about lexicality, not phrase length.

4. The options

• Carry whole (tail and chime). Available when one English word or phrase can stand clause-final at every position and serves every collocation the poet makes with it in the register declared (the §7.46.5 test, passed in advance at R58). Cheapest where the radif is a light verb whose object is the qāfiya at every position (R55 ۱۳۱: no inversion at all); otherwise the usual price is an inversion at each line.
• Carry by inversion — object, complement or adverbial before the verb. Payne's move in the matched odes; Leaf XVI; the lead at 9 of 9 in plain register. Register does not decide availability. Its cost has no admissible number: the only pricing is seat-rated (X3), and there it varied from nothing to six of ten across eighteen bayts.
• Re-attach. Break the genitive and hang the pronoun on the clause (thy → for thee): keeps the shape, changes the relation (Leaf XXV).
• Nominalise and front (made good actions her practice): what Payne's register reached where the lead's declared plain one hit a wall (CXVI).
• Keep the tail, drop the chime (REP). No sense strain, every line free to be literal, the poet's ایطا kept; with a genitive radif, an unstressed line-ending at every position.
• Keep the chime, drop the tail (CHI). A stressed rhyme at every line; bought with sense (three logged strains plus one unlogged across 24 positions); the poet's own repetitions corrected; the family chosen from the hardest position.
• Refuse and declare. The one clean case is the object marker را (seven refusals, zero carriages). In prose, V18 declares the grammatical loss.
• Externalised bound morphology is not a choice. Where the Persian repeats a bound copula or person suffix, the English carries a radif whether or not one is wanted — rendering the source's morphology, not adding a refrain.
• Prose against verse. In rhymed prose the lexical radif crosses free (V18); in verse the tail and the chime compete for one slot.

5. Guidance

For a translator

1. Before the first line, list every rhyming position's radif and bearer and ask three things: can an English clause end on the radif; does the radif govern its bearer (the English of-genitive puts a function word in the rhyme slot); does one English word serve every collocation the poet makes with it, in the register you have chosen. The third question decides, and it is answerable from the source. — evidenced (FA→EN), craft.
2. Do not treat the radif's part of speech as a verdict. Finite transitive verbs, negated verbs, copulas and pronouns in ezāfe were all carried by published hands; the one cell no hand carries is the object marker را. — evidenced (FA→EN).
3. Expect to invert. The claim that it needs an archaic register was withdrawn (§7); nothing admissible prices it in either register. — evidenced (FA→EN, one poem, lead) for availability in a plain register; the price untested (X3 only).
4. Where the radif verb's object is the rhyme word, carry both halves as one constituent (made a plundering) — but check the relation holds at every position; where it lapses, item 1's collocation test governs. — evidenced (FA→EN, two poems).
5. Where you cannot carry both halves, choose the half by the poem's cost, priced in advance: keeping the tail costs the line-endings (unstressed, if the radif is a genitive) and keeps the poet's repetitions; keeping the chime costs literalness at some positions and corrects the poet. Choose the rhyme family from the least tractable position. — evidenced (FA→EN, three poems), craft; nothing says which half a reader wants.
6. A Persian radif's sense moves under the repeated form; yours will not. Decide which sense to buy where they diverge, and log the others as paid. — evidenced (FA→EN).
7. When the Persian repeats a bound morpheme or the qāfiya word itself, expect your English to carry a radif the metrics do not record; neither suppress it as an addition nor count it as carriage of a radif. — evidenced (FA→EN, two hands).
8. Do not plan to park a displaced sense on the radif words; the chime slot before them is where a rhyme's cost is spent, and it is the slot to protect rather than "the last word". — evidenced (FA→EN, 14 renderings), descriptive.
9. In rhymed prose, carry a lexical radif verbatim and declare a grammatical one lost; hunt no rhyme for it. — evidenced (FA→EN, one chapter, one hand), craft.

For a pipeline

1. Enumerate the radif mechanically (longest common trailing word sequence over the rhyming positions, minus one word for the qāfiya) and every bearer; classify the radif blind on the frozen checklist — part of speech; governs its bearer; case particle; light verb with the qāfiya as object; bound morphology. — the census and classification ran as study stages (RS-20260829, RS-20260831b); untested as a pipeline step.
2. Test collocation before rendering: for each candidate English radif, check one lexeme against every position's complement in the declared register; grade the poem CARRY / FORCED / REFUSE. — untested.
3. Set the policy parameter for a FORCED poem: REP, CHI, or both renderings of record (R54); for CHI, select the rhyme family from the hardest position. — untested.
4. Render under R55 (carry, or refuse with the refusal enumerated) or R54; log each inversion as LOCAL / SPREAD / REBUILT (R58). — untested.
5. Check: mechanical grade of tail (IDENTICAL) and chime (tools/rhyme_pairs.py); a sense-equivalence gate on every position where the chime was bought; flag manufactured radifs where the Persian repeats a bound form. — untested.

Human entry points. Step 3 is a purpose decision (declared-purpose): which half to keep, and whether a reader who sees no Persian is to be told the form (no disclosure tested moved a seat — X3). Step 2's register declaration is a person's: a plain register shut one poem the archaic hand carried. Without a person, default to both renderings of record for a FORCED poem and REFUSE for را.

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

6. Not evidenced, and open

• Persian→English only; two poets; two published hands in verse, one register each, Brockhaus text not Qazvini–Ghani, OCR unaudited, Payne's accepted set enriched for tail-carrying poems (0.688 against 0.577); no published hand in prose — the Gulistan hands were coded for rhyme, never for radif.
• Refrain proper — a repeated line or stanza, in any language — was never measured.
• What repetition is worth to a reader: no number, and none can come (Tier D exhausted); style-correspondence records repetition as still not evidenced — a comparative radif supplies a chime and a genitive radif a weak ending whether or not wanted, so the halves were never separated as a device.
• The price of the inversion exists only as seat ratings; the end-weight conjecture (D8) and the object/adverbial distinction (n = 11 / 4) are untested; the habit account (does Payne invert radif or no radif) was withheld; no inversion rate may be cited.
• Whether the radif's class predicts carriage is underpowered ([−0.119, 0.253]).
• §7.49.1's by-radif robustness claim is unresolved at n = 8 renderings ([−0.128, +0.364]).
• Open: the manufacture rule (bound morphology → English radif) in a third hand or a second source language that marks person or copula on the word. Open: the end-weight account of the inversion's price, codable off the page. Open: whether R58's plain-register carriage generalises past one poem. Open: whether the radif survives a fixed metre beyond one poem (§7.43.3). Open: the conjecture RS-20260828-forced-half §7 named — a disclosure that favours neither arm has nothing to move a choice with — which belongs with HB-telling-the-reader.

7. Sources consumed

• v0.2 §7.36 (S224, RS-20260826b-radif): the prediction table and three craft facts; its measured half unestablished by its own order gate — §3.1.
• v0.2 §7.40 (S229, RS-20260828-forced-half): the genitive case, the two policies and their costs; the disclosure null and the seat level — §3.2, §4.
• v0.2 §7.41 (S231, RS-20260829-radif-hands): nine matched ghazals, the refutations, the surviving را prohibition, the census — §2, §3.3.
• v0.2 §7.42 (S232, RS-20260829b-inversion-price): the register clause withdrawn — §3.5, §7.
• v0.2 §7.46 (S236, RS-20260831b-radif-hands): Payne whole, the manufacture rule, the whole-poem collocation test — §2, §3.4.
• RS-20260908-chime-slot (S254), with v0.2 §7.49.1's radif stratum: the chime, not the line-end, is the spent slot — §2 row 5, regularity 5.
• Sections the row's titles hid, consumed for their radif content only: v0.2 §7.32's struck sentence (the target of §7.36's correction); §7.43.1 and §7.43.3 (RS-20260830-leaf-contract, A-leaf-hafez — Leaf XVI, the 7 of 28 count, the radif surviving the metre); §7.47.2 (RS-20260901-inversion-habit — the plain-register carriage at 9 of 9). §7.47.1/§7.47.3 and §7.51 belong to family D.
• gulistan-bab2/register.md V18 and R05-v1/translation.md D48, D62; goodness-senses.md §style-correspondence (the two repetition entries); regimes R51, R54, R55, R56.
• Withdrawn along the way, one line each: §7.32's "English has no slot before a repeated word to rhyme in" — withdrawn by §7.36 (RS-20260826b). §7.36.1's transitive-verb cell "unreachable" — refuted by §7.41 (RS-20260829), withdrawn entirely by §7.46 (RS-20260831b). §7.36.1's oblique pronoun and §7.40's "a radif standing in a genitive to its bearer cannot be carried, whatever its part of speech" — refuted on Leaf XXV by §7.41, reported as 1 of 3 by §7.46. §7.41.3's "you can, at this price, in this register" — withdrawn by §7.42 (RS-20260829b). §7.41.3's "Leaf, writing plainer, never uses it and drops every one of those radifs" — withdrawn by §7.43.1 (RS-20260830-leaf-contract). §7.41.5's "where the radif is a light verb whose object is the qāfiya, the shape crosses whole … no inversion" — corrected to the whole-poem collocation test by §7.46.5 (RS-20260831b). S230's conjecture that English cannot end a clause on a bare copula — refuted by §7.41.4 (Payne's is). The T-hafez-sahar-bolbol-R55-v1 log's "the obstacle is lexical, not syntactic" — corrected by RS-20260831b §6, the frozen log left unamended. §7.49.1's "factor of eleven" and its equal-in-both-strata claim — corrected to about 5.5 and to unresolved by RS-20260908-chime-slot.