Repository path: workshop/experiments/E-20260814-saj-carriage/design.md · rendered 2026-09-09

E-20260814 — does Arabic rhymed prose reach English? A census of two published hands at eight frozen loci

Frozen 2026-08-14 before either comparator was opened. The translation it hangs off (T-alf-layla-R05-v1, commit 2c5b612) was frozen first, in a separate commit, with its eight saj' loci already tabulated. Neither Lane's nor Burton's rendering of span A has been read at the time of this freeze. That ordering is the whole design: the loci cannot be chosen to suit the answer if they were fixed before the answer was visible.

1. Question

Saj' — rhymed, cadenced prose — is the signature of literary Arabic prose and is the thing translators of it say cannot be brought over. Does it reach English at all, and does the answer differ between two published hands with opposite declared programmes?

The translating raised it. Eight loci in eighty words of preamble consumed the largest share of the sitting of anything in span A, and the lead's own answer (five carried, one partial, two lost, by moving the rhyme to an -ing chain that falls nowhere near where the Arabic rhyme falls) is a craft record and not evidence: contamination: high, declared with two named priming events. The finding this experiment can support is about Lane and Burton, not about the lead.

2. What this teaches about translating literature (subject rule)

It measures whether a specific formal device of a source literature survives translation into English, on the published record, at loci fixed in advance — the same question shape the project has run on Gogol's ellipses, Poe's italics, Bengali honorifics and Hungarian tag-verbs, now on the device that Arabic prose is most defined by. It is not about the project's instruments.

3. Materials

• Loci: the eight saj' loci of T-alf-layla-R05-v1's log table, D2, frozen at 2c5b612.
• Copy-text: Hindawi 2022, Būlāq/Calcutta II family (collation.md §2).
• Hand A — Edward William Lane, 1839–41, Project Gutenberg #34206, vol. I. Worked from Būlāq. Declared programme: an English Nights for the Victorian family, with an ethnographic apparatus.
• Hand B — Richard F. Burton, 1885, Project Gutenberg #3435, vol. 1. Worked mainly from Macnaghten (Calcutta II). Declared programme: literalism, with an English built to be strange.
• Payne 1882 was searched for and is not freely reachable (collation.md §5). Two hands.

4. Procedure

1. Locate, in each hand, the English corresponding to each of the eight loci. Record the located text verbatim.
2. Code each locus in each hand on a fixed four-value scale: - carried — the English cadence carries a rhyme or a true assonance across two or more members at the corresponding join; - partial — a rhyme across some members of a chain of three or more, but not all; - lost — the passage is translated and carries no rhyme at the join; - absent — the passage is not in that hand at all (recension difference, or omission).
3. absent is reported separately and never counted as lost. A locus a translator never saw is not a locus a translator dropped.
4. Record the rhyming English word-pair at every carried and partial, so a later session can audit the coding without re-locating the text.
5. Separately, code the one non-saj' locus that span A's collation turned up: ونيك (D4), on which the copy-text's own second Arabic witness performs a deletion. Scale: rendered / euphemised / omitted / absent.
6. Run tools/dependence_check.py on the three renderings of span A — lead, Lane, Burton — and report the longest common run and the shared 7-gram count, per the standing rule.

5. Registered predictions

P1 — Burton codes carried at strictly more loci than Lane. His declared programme is literalist and he is the hand famous for trying to reproduce Arabic form in English.

P2 — neither hand reaches carried at more than four of the eight loci.

P3 — at locus 4 (أعلم · أحكم · أعز · أكرم, a four-member epithet chain) at least one hand drops members rather than rhyming them.

P4 — ونيك is omitted in Lane and not rendered in Burton. Lane's expurgation is a documented fact of his edition; Burton's reputation for frankness was made on his notes and his Terminal Essay as much as on his text.

P5 — the lead's overlap against both hands is low (no 12-gram shared with either), despite contamination: high, because the register decision D1 points away from both.

6. Failure criteria, registered

• Under-power: if fewer than six of the eight loci can be located in both hands, the census is reported as under-powered and no comparative claim is made from it.
• Coder: the coding is done by the lead, single-handed, having already translated the same span. This is a real threat and it is not remediable within this session. It is bounded by rule 4 above — every verdict must exhibit its English word-pair — so the coding is auditable even though it is not independent. Any claim from this census carries internal-judgment-only.
• Recension: the copy-text is not the text either hand worked from. absent verdicts are therefore ambiguous between omitted by the translator and not in his Arabic, and will be reported as ambiguous rather than resolved.