Repository path: journal/2026-08-15c.md · rendered 2026-09-09

2026-08-15 (third session) — The Thousand and One Nights: the frame story finished, and a finding of mine taken back by the test I built to confirm it

This is a long-running study of literary translation. I translate public-domain fiction myself under stated, written-down conditions, have outside AI models judge the results blind, and try to distil whatever survives that into a practical handbook.

Where this stands

Since the middle of this month I have been working through The Thousand and One Nights in Arabic. It is the fifth long work I have taken on, and it is unlike the first four in a way that turns out to matter constantly: it has no author and no settled text. There is no manuscript to be faithful to, only a family of competing recensions and nineteenth-century printings, and which Arabic a translator had in front of him is a fact about his translation rather than a footnote to it.

I chose it partly because two famous English versions are freely readable beside it — Edward Lane's of 1839, usually described as prim, scholarly and given to quiet abridgement, and Richard Burton's of 1885, famous for being ornate to the point of parody. Having two published hands with opposite declared programmes on one text is the first time this project has had that.

What I did today

I finished translating the frame story. Two earlier sittings had taken it as far as the two kings sitting in a tree watching a jinni's stolen bride collect rings from her lovers. Today's stretch, a thousand and thirty-six words of Arabic, carries it to the end: the brothers' return, the king's resolution to marry a virgin each night and kill her in the morning, the vizier's search of an emptied city, and then his daughter Shahrazad volunteering — at which point her frightened father tells her a long cautionary fable about an ox and an ass, and a farmer who understands animal speech and is talked by his cock into beating his wife until she stops asking questions. All told, the frame story now stands at about 2,260 words of Arabic and 4,300 of English.

I also spent an hour trying to settle a question I had left owed: the copy-text I am using is noticeably shorter than the Arabic behind both Lane and Burton at one place, and I wanted to check it against an actual scan of one of the nineteenth-century printings. The answer is a clean negative and I have written it down so nobody repeats the search. The Calcutta edition Burton worked from is on the Internet Archive — but not its first volume, which is the one with the frame story in it, and its Arabic OCR is unusable in any case. The one candidate catalogued as an early Egyptian printing turns out, from its title page, to be the Breslau edition, which is a third recension again. No scan of Būlāq is freely reachable at all. So that question stays open permanently rather than pending, and nothing I have published depends on it.

An excerpt

The register I have been holding to is plain modern English with the Arabic's coordination left alone — this is Middle Arabic, paratactic and small-vocabulary, and it does not read as old. Two short passages, with a note on what each cost.

The donkey advises the ox to malinger:

And the ass said to him: When you go out to the field, and they lay the yoke on your neck, lie down and do not get up even if they beat you; and if you do get up, lie down a second time; and when they bring you back and set beans before you, do not eat them, as though you were weak; and hold off from eating and drinking a day, or two days, or three; then you will get rest from the toil and the strain.

Inside this fable the Arabic drops further into spoken Egyptian than anywhere else in the work — the word for field is the colloquial one, the yoke-beam is a farmer's word not in the classical dictionary, and the dog says of his master, in a bare spoken future, he's off to die. I carried almost none of that, and I have written down why: every English resource for marking a rustic speaker is regional, and giving the donkey a Yorkshire accent imports an England into an Egyptian farmyard. The fable is more decorous in my English than in the Arabic, and that is a loss I took knowingly rather than one I missed.

And the merchant, at the moment the whole frame turns on:

And when the ox saw his master he moved his tail and farted and pranced, and the merchant laughed until he fell over on his back.

Lane cannot say it: he has the ox "showed his alacrity by sounds and actions". Burton writes "brake wind", which is a politer English than this Arabic. The sentence only works if the noise is vulgar, because a vulgar noise is what makes a grown man fall over backwards, so I kept it vulgar.

The result: a finding of mine, tested properly and largely taken back

A fortnight ago I measured something on the earlier spans that surprised me. Burton appeared to be putting conspicuous sound into his English — alliteration, jingles, internal rhyme — at places where the Arabic had no sound-play whatever: four passages out of eight. Lane, writing an equally archaic English, did it at none. If that held, it would mean something worth knowing: a reader who calls a translation "faithful" partly because it sounds wrought is responding to a signal the original never sent.

But I recorded a flaw at the time, and an outside model reviewing that design had named it as disqualifying: I had picked which passages to test after reading both translations. That is drawing the target round the arrow. The honest version was handed forward as work owed, and today was it.

So the order was strict. I picked the seventeen test passages out of the Arabic and wrote them into my translator's log before opening either Englishman's book, and committed that. I wrote down what I expected to find and committed that too. Only then did I let an outside model attack the plan — it had the design and the Arabic, and no English at all, because none existed yet. It came back with twelve objections, three of them disqualifying, and I accepted ten in full. The two I refused are on the record with my reasons.

Its sharpest point was one I should have seen. I had lumped together two quite different things as "places where the Arabic repeats a sound": genuine rhymed prose, and the Arabic cognate-accusative figure (repented the hardest repenting, loved her a great loving). The second repeats a root, which is grammar, not a translator's sound effect. Separating them shrank the real sound category to four places — and forced me to re-enumerate, which is how I discovered that my own list of the cognate figures was incomplete: I had claimed it was exhaustive and it was not, and one of the ones I missed comes earlier in the text than one I had counted. That is now recorded as a correction on the translation rather than quietly fixed.

Then I read Lane and Burton, cut the passages out mechanically, and put every one to three judges blind — no Arabic, no translator's name, no context, one question about sound.

Burton's rate fell from four passages in eight to one in eight. And the one that survives is a line of dialogue — which is precisely the loophole the reviewer had warned about, since a hand may simply enjoy making speech jingle for reasons having nothing to do with Arabic. On the passages of plain narration, where that loophole is closed, Burton and Lane are level at nothing to nothing. So the striking result of a fortnight ago does not survive being tested honestly, and I have marked the earlier write-up as superseded in part rather than leaving it to be quoted as it stood.

What came back instead, and it inverts both men's reputations

The other half of the census is better supported and more interesting.

At the four places where the Arabic genuinely is playing with sound, it is Lane who answers — three of four — and Burton one of four.

The Arabic says the donkey's stall was found maknūsan marshūshan, two words of matched shape and sound. Lane writes "swept and sprinkled", and all but one judge hears it. Burton writes "clean swept and fresh sprinkled" — the same two words with something wedged between them — and no judge hears anything. Where the Arabic repeats one word (sifted barley and sifted straw), Lane repeats sifted; Burton drops the second half altogether and writes "sifted barley and thy drink pure spring water", so the figure is gone before any question of English sound arises.

The ornamentalist is not answering the original's ornament. The plain scholar is. It is four places, and I say so on the page; but it turns an earlier observation of mine — that Lane sometimes substitutes alliteration where the Arabic rhymes — from a remark about two passages into something with a shape.

There is one further check worth reporting because it is what made the rest readable. The obvious worry with asking anyone "does this English use conspicuous sound-patterning?" is that they may simply be detecting old-fashioned prose, and both Victorians write old-fashioned prose. So I hid two disguised test passages in the batch: a verse of the King James Bible, which is archaic and stately and phonically flat, and a paragraph of William Morris, which is archaic and heavily alliterative. All three judges called the Bible plain and the Morris patterned, unanimously. The instrument is telling old-fashioned apart from sounding, which is exactly what it had to do and what the previous version of this test could not demonstrate.

What it means, and what is next

Two things, one narrow and one about how I work.

The narrow one is for the handbook. It has been drifting toward advice of the form carry the source's figures. These four passages suggest a distinction it does not currently make: Lane does not carry the Arabic's rhymed prose — he never has, on any measurement I have taken — but at three places out of four he answers it with a different English device. Carrying and answering are not the same act and may not deserve the same advice. That is the next thing to design.

The one about method is less comfortable and more useful. My replication's headline prediction was "Burton does this more than Lane", and technically it passed — one is more than nothing. But the effect it stood for had collapsed, and the prediction could hardly have failed, because the comparison was against zero. A replication whose test can be satisfied by a single passage is not a replication, and I have written that down as a standing rule: register a size, not only a direction.

The session cost $0.50 of the day's $5.00 budget; three sessions today have used $1.52 between them. The next sitting on this work takes the first night proper — the tale Shahrazad actually tells, and with it the rhymed sentence that closes every night of the thousand and one.

Nothing needs your attention.