Repository path: workshop/experiments/E-20260831-arabic-in-persian/loci-frozen.md · rendered 2026-09-09

Sa'di's Arabic and what four English hands do with it — loci and predictions, frozen

Frozen 2026-08-31, before any published English of باب دوم was opened, and after T-gulistan-bab2-R05-v1 span C was committed at 651d64cc. Nothing below was written with an English text in view.

1. The question

«گلستان» is a bilingual book. Sa'di writes Persian and quotes Arabic — Qur'an, hadith, Arabic verse, Arabic proverb — and a Persian reader sees the language change on the page. English cannot change language at those places, because the whole book has already become English.

register.md V5, made at span B, says the English marks the Arabic only where Sa'di marks it: where he frames the quotation the frame survives translation and does the marking by itself; where he does not, the language change is the only signal there is, it cannot survive, and the loss is declared on the artifact rather than repaired with a device Sa'di did not use. Span C had to apply that rule four times in eleven tales, and the sharpest case (b236) is a Qur'anic half-line that rhymes with the Persian half-line before it — the code-switch is what makes the rhyme.

The question this study asks is not whether V5 is right. It is what other translators did. Four published English hands rendered this book between 1806 and 1899. At every place where Sa'di changes language, each of them made the same decision, in print, and the decisions are countable.

Subject-rule sentence (wiki/tracks.md): this unit teaches what translators do when the source is bilingual and the target cannot be — whether the foreign voice is marked, absorbed, or dropped — which is a standing problem of literary translation and one the translation limb met four times this session.

2. Materials

Source side. «گلستان» باب دوم whole, 390 blocks, single-witness copy-text of record (collation-chapter.md). Loci enumerated from the Persian by two independent detectors — a diacritic-density screen over whole blocks and a maximal-run screen over tokens — with every candidate then read and classified by hand. Both detectors and their raw output are in enumerate.py.

Target side. The four hands of A-gulistan-hands, from the Internet Archive, public domain (1806–1899): Gladwin 1806 (gulistan00unkngoog), Ross 1823 (gulistanorflowe00rossgoog), Eastwick 1852 (gulistanorrosega00sadiuoft), Arnold 1899 (frompersianguli00arnogoog). Provenance and OCR quality as recorded at E-20260822c-persian-hands/materials/hands.md.

One instrument change from that experiment, and it is a gate, not a unit. _djvu.txt loses italic setting, and italic is one of the four signals this study codes. The _abbyy.gz derivative carries <formatting italic="true"> and exists for all four books; it is used here, and where a book's OCR gives no italic runs at all that fact is reported rather than read as "this hand used no italics".

3. The loci — 21 Arabic quotations in باب دوم

Excluded by definition, and the exclusions are part of the frozen design: pious formulae (رَحْمَةُ اللهِ عَلَیْهِ, عَزَّوَجَلَّ, سُبْحانَ اللهِ, رَضِیَ اللهُ عَنْهُ, عَلَیْهِ السَّلامُ) — they are Arabic but they are formulae, not quotations, and V9 governs them; naturalised Arabic connectives (فی‌الجمله at b207 and b356); Arabic divine vocatives inside Persian speech (یا غَفور، یا رحیم at b005); and the very large Arabic lexical stratum of ordinary Persian.

17 unframed, 4 framed (one of the four also glossed). V5 treats the framed case as the one its rule is built around; in this chapter it is the minority by more than four to one.

4. Coding, fixed before the books are opened

Each locus × each hand is one cell. Two codes per cell, independent of each other.

Presence. PRESENT — the content of the Arabic appears in the English. DROPPED — it does not. UNREACHABLE — the hand does not translate this part of the book, or the scan is illegible at the place. (Arnold 1899 covers the first four bábs, so باب دوم should be within all four.)

Marking. MARKED if the English carries any of four signals at the locus, and the signal is recorded: IT italic setting · QU quotation marks · TR the Arabic transliterated or printed in Arabic type · SAY an explicit phrase naming the language or the source (in Arabic, the Koran says, the Prophet said) or a footnote doing so. PLAIN if none of the four is present. SAY counts only where the English introduces the naming and the Persian did not — at the four framed loci the Persian frame is already there, so SAY at those loci is translation, not marking, and is coded SAY-SOURCE and kept apart in the analysis.

Verse form is not coded here; A-gulistan-hands §2 already records each hand's verse policy and RS-20260822c measured it.

5. Predictions, registered

1. The framed loci will be marked far more often than the unframed ones. Concretely: marking at framed loci ≥ 0.75 of reachable cells, marking at unframed loci ≤ 0.35. This is V5's premise, and if it fails V5 is resting on an intuition the tradition does not share.
2. Some hand will repair. At least one of the four will mark at least three unframed loci, because the impulse to tell the reader this was Arabic is strong and a footnote is cheap. If no hand repairs anywhere, V5 is the tradition's policy as well as this arm's.
3. Dropping will be rare and unevenly spread. ≤ 0.15 of cells DROPPED overall, and the dropping concentrated in one or two hands rather than spread evenly — on the pattern of RS-20260822c, where the hands' policies differed by hand and not by locus.
4. The rhyme-bearing locus (b236) will be PLAIN in every hand, because carrying it needs both a rhyme and a language change and no hand has produced a rhyme inside Sa'di's prose at any of 128 opportunities (A-gulistan-hands §3).

Failure criteria. The study reports a null if (a) fewer than three hands are reachable for باب دوم, or (b) locus alignment fails for more than a quarter of cells — that is, if the lead cannot say with confidence where in the English a given Arabic locus falls. Alignment is by tale number and by the surrounding sentence, both of which every hand prints.

Known limitation, declared in advance. The marking codes are the lead's reading of nineteenth-century OCR, unanchored, and IT in particular depends on the ABBYY layer being right about italics. Every MARKED cell is quoted verbatim on the result page so the reading is checkable; a hand whose ABBYY layer yields no italics anywhere is reported as such.