Repository path: workshop/translations/koyhaa-kansaa/collation.md · rendered 2026-09-09

«Köyhää kansaa» — the copy-text, and what a second witness changed

Opened S100 (2026-08-03), as the gate on register.md §Settled-5 / D62 — a second printing of the Finnish, required "before span 4, not a tidy-up", because the working text had by then produced three suspected defects (tuuditti uskoa D17; «EL» ¶94; «Vesi el» ¶197).

1. The witnesses

W1, W2 and W3 are one transcription, not three witnesses. W2's own front matter reads "Tämän e-kirjan ovat tuottaneet Tapio Riikonen ja Projekti Lönnrot" and names the same Otava 1917 volume; W1 is that file at Gutenberg; W3 reproduces all three of the suspected defects character for character, including «EL» and «Vesi el». Three sources, one file, one printing. A second copy of a text is not a second witness, and the count of hits a search returns is not evidence of independent attestation.

W4 is a genuinely independent witness and a better one: it is the first edition, published in Canth's lifetime (she died 1897), where W1–W3 descend from a posthumous reprint. Every reading below marked (image) was read off the page scan, not off the OCR layer.

2. The three standing suspicions, resolved

Both renderings the project had already made were right — ¶94 and ¶197 were translated as ei on the linguistic reading, and the first edition agrees. The value of the check is not that it changed those two sentences. It is §3.

3. Span 4 collated in full, ¶207–295 (1,800 Finnish words)

Word-level diff of W1 against W4's OCR layer, every difference then adjudicated against the page scan. Pure OCR noise (01etko, Heilu for Hellu, seljalleen) is excluded; 8 substantive variants remain.

3a. W1 standardises the eastern-dialect layer — 3 sites, 4 tokens

This is the finding. register.md V3 declares the eastern and colloquial layer a loss the translation carries by syntax and idiom only — "a declared loss, not a solution". Four tokens of that exact layer, in 1,800 words, had already been removed by the printing before the translator saw a word of it. A translation that declares one loss was silently taking a second.

How much it costs, honestly: at three of the four sites, nothing. uhkaa/uhhaa, lepoa/lepoaa and tyynemmällä/tyyneemmällä are phonological, and English has no carrier for them under V3 in any case. The fourth site is different: maha and vatsa are two words, and English has two words — belly and stomach. W1 spends both of Canth's on one. The translation now renders ¶260 (narration, vatsalleen) "on her stomach" and ¶264 (Tiina Katri, mahallaan) "on her belly" (D65; register.md V21).

What is and is not claimed. The novella has five tokens of the pair, and two of them are fixed collocations that decide nothing — mahan täydeltä "a bellyful" (¶290, narration) and mahatauti (¶568, dialogue). Three tokens are free choices: vatsalleen at ¶66 and ¶260, both narration and both vatsa in both printings, and ¶264, the only one in a character's mouth, which is maha in W4 and vatsa in W1. So the single token at which a narrator/character contrast in this pair could show is exactly the token W1 moved. One token does not establish the contrast and none is claimed here. What is established is enough: the choice at ¶264 is a translator's to make, and W1 had already made it.

3b. W1 introduces two readings the first edition does not have

3c. W4's own compositor's slips, where W1 is right — 3 sites

vatekasalle for vaatekasalle (¶270); entisen paikkaan for entiseen paikkaan (¶292); millän for millään (¶253, image p. 41). Neither printing is clean, which is why the policy below is eclectic rather than "follow the first edition".

3d. Punctuation and orthography, recorded, no rendering affected

¶254 Kies avita, (W4) / Kies avita! (W1) · ¶264 puree koska / puree, koska · ¶293 taas tuo, / taas: tuo, · ¶294 luoja / Luoja · ¶283 hän sanoi. / hän sanoi (W1 drops the stop) · ¶263 W1 drops «sen» from «antoi sen maata» · ¶287 menivät rikki; (W4) / olivat rikki, (W1) — taken from W4 as the livelier and the earlier.

4. The copy-text policy, from span 4 onward — V21

W4 (Edlund 1886) is the copy-text. W1 is consulted at every substantive divergence and is followed only where W4 is a manifest compositor's slip. Reasons, in order:

1. W4 is the first edition, published in the author's lifetime; W1 is a reprint issued twenty years after Canth's death and shows a consistent editorial hand (§3a).
2. W1 carries at least two readings that are not construable (§3b) and two demonstrated corruptions outside this span (§2).
3. The direction of W1's changes is against the layer this translation has declared it cannot carry. A copy-text that pre-normalises the thing you are recording as a loss makes the record false in your favour.

What this policy is not. It is not a claim that W1 has no authority: an 1917 Otava text may descend from a printing Canth herself corrected between 1886 and 1897, and this project cannot check that from anything it can reach. Where W4 and W1 differ and both are construable, the divergence is recorded here whichever is taken. Spans 1–3 were translated from W1 alone and have not been collated; that is span 5's first task and is entered on register.md §Unresolved.

5. The check that had to be run before the study limb, and its result

RS-20260803g rests on 22 second-person sites across ¶2–295 and ¶389. Every one was compared with W4. All 22 read identically in both printings — the corruptions §2 and §3 found are lexical, orthographic and phonological, and not one touches a second-person form. The census stands on either witness. Recorded because the opposite outcome would have voided it, and because the check was run before the result was written rather than after it was doubted.

6. Provenance

W4 fetched 2026-08-03 by direct HTTPS GET of https://www.doria.fi/bitstream/handle/10024/100842/Koyhaa_kansaa_kuvaus_tyovaen_elamasta.pdf (16,057,787 bytes, 97 pages, PD). Page images read at pp. 13, 14, 16, 33, 35, 39, 41, 42, 43, 44, 45, 46. The PDF's OCR text layer was decoded from its own ToUnicode CMaps for the mechanical diff; no reading in §2 or §3a–3b rests on the OCR layer alone. The scan is not stored in this repository — it is a 16 MB page-image file of a public-domain book that is one GET away, and PROJECT.md §7 stores no source whole that it does not need to. Logged in wiki/base/consulted.md.

7. Spans 1–3 collated, ¶2–206 (5,372 Finnish words) — S105, 2026-08-04

The gate register.md §Unresolved 1 set: the spans translated before V21 existed had never been put against the copy-text. Method as §3 — word-level diff of the 1917 text against the first edition's OCR layer, every difference then adjudicated against the page scan, pure OCR noise excluded. 98.5% of tokens aligned; 62 rows survived the noise filter — two of them alignment artefacts at the head and tail of the range — and 16 of the remaining 60 are substantive. Scan pages read as images: 6, 10, 11, 13, 17, 20, 21, 28, 30, 32, 33, 34.

Nine of the sixty-two dissolved on the image — the OCR layer, not the printing, was wrong. ¶2 pörähti, ¶2 silniiä, ¶3 laidallahan, ¶25 murjottivat, ¶41 seljan, ¶69 onnettomani, ¶76 Lleikura, ¶99 noitiinkin, ¶153 tuollaiseen all read in the first edition exactly as the 1917 text has them. Three of these — laidallahan, murjottivat, tuollaiseen — would have been reported as substantive variants had the diff been trusted, and one of them (murjottivat for muljottivat, sulked for goggled) would have changed an English word. The mechanical diff is a finding aid and nothing else.

7a. The reprint standardising the colloquial layer — 5 sites

The last row is the one that matters, and it is a different kind of thing from anything span 4 found. Span 4's four tokens were phonological, and §3a concluded honestly that at three of the four English loses nothing because it has no carrier. Plural subject with singular verb is syntax, and English does have carriers for it. register.md V3 says the eastern layer "is carried by syntax and idiom only"; the reprint was removing precisely the sites where that clause could have been cashed. The site is also inside free indirect discourse — Mari's mind, not the narrator's voice — which is where a colloquial agreement does its work.

What is not claimed: the hand is not systematic. The first edition itself writes vain at ¶153 and the 1917 text keeps vaan at dozens of sites, including four in ¶2 alone; and at ¶354 «eikä ne siitä kumminkaan parane» both printings keep the singular verb. Five standardisations in 5,372 words is a tendency, not a policy.

7b. The reprint running the other way — 3 sites

¶2 hyssytellä → hyssyytellä; ¶39 jäljessä → jälessä; ¶72 tarjona → tarjonna. In each the first edition has the standard form and the reprint the reduced or non-standard one. A reprint that only ever tidied would not do this, and §4's account of W1's editorial hand is narrowed accordingly.

7c. A third corruption of the ¶94 / ¶197 family — 1 site

¶134: 1886 «Lapset juonittelivat hänen sylissään; niillä oli raukoilla paha ja vaikea olla» (image, p. 25); 1917 «millä». millä is with what and the clause is not construable with it. This is the same one-letter substitution as Ei→EL (¶94) and ei→el (¶197) — and the first of the three to fall inside a sentence the project had already had to render. The translator construed it as niillä on the sense and was right; see Erratum 5.

7d. Readings that differ where both are construable — 7 sites

¶73 laittoi/taittoi, ¶102 Souda/Soudata, ¶105 first edition lacks alkaapäin, ¶109 menemistä/menemästä, ¶110 Maria/Mari, ¶163 first edition has enää, ¶178 siunasi/siunaili. Three of these change an English word: Erratum 5 on R05-v1/translation.md. ¶178 does not — "called on God" renders either — and ¶109 and ¶110 do not.

¶105 and ¶339 (span 5) are the two places where the reprint has a word the first edition does not. §4 described W1's hand as one that removes; it also adds, and the description is corrected here rather than left standing.

7e. Not adjudicated against the image

¶22 and ¶167 kätkyeesen/kätkyeen and ¶196 hääritessään/hääriessään were read off the OCR layer only; ¶120 asiaansa/asiansa likewise. None affects a rendering, and they are listed so that the adjudicated set is not overstated. Word-division and accent differences (¶12, ¶21 ×2, ¶115, ¶128, ¶149) are recorded and change nothing.

8. Span 5 collated before translation, ¶296–375 (2,041 words)

The first span for which the copy-text was settled before a word was written, which is what §4 was for. Twelve divergences survive filtering; eleven adjudicated against the scan (pp. 51, 55, 56, 58, 59); one (¶296 tyyneesti/tyynesti) left on the 1917 reading and recorded as unadjudicated. Five are phonological — laattia ×3 for lattia, raappaisi, hämärä valo — and three of those are the same eastern long-vowel layer as §7a. Three change a word and are logged at D85, D86, D87; the load-bearing one is ¶344 miehelleen for mielellään, which turns she would not willingly tell it into she would not tell it to her man, in a paragraph whose whole subject is concealment from the man lying awake beside her. That is the fourth corruption found in the reprint and the worst of them.

¶320 «Ei» against ¶348 «En» is recorded here as a textual fact and nowhere claimed as an authorial one: the first edition gives two different answers to the same question forty lines apart, the reprint gives one. What the translation did with it is D85.

9. Where the copy-text question now stands

9,213 of the novella's 13,525 words have been collated — ¶2–206 and ¶296–375 here, ¶207–295 in §3. Thirty-six substantive divergences (16 + 8 + 12), a rate of 3.9 per 1,000 words; span 4's rate of 4.4 was not a fluke of one span. Four demonstrated corruptions in the reprint (¶94, ¶134, ¶197, ¶344), of which two were invisible until the sentence had to be rendered. ¶376–573 is not collated; it is the remaining 32% of the novella and is span 6's first task on the same terms.

tuuditti uskoa is settled and the second witness had nothing to do with it. D17 referred the phrase to this check as a suspected typesetter's slip. It occurs twice in each printing — ¶31 and ¶374, 1886 pp. 12 and 64 — and a slip does not repeat itself in the same collocation across two settings. See D89.

10. ¶376–572 collated, the last unchecked stretch — S110, 2026-08-05

The gate register.md §Unresolved 1 set. Method as §3 and §7. The diff was run over ¶375–572 with ¶375 as an alignment anchor: 4,347 W1 words against 4,338 W4 words, 97.9% of tokens aligned; 68 rows survive, of which 17 differ only by a diacritic or a doubled letter. Scan pages read as images: 64, 65, 68, 69, 70, 75, 77, 79, 80, 82, 83, 85, 86, 87.

Sixteen of the sixty-eight dissolved on the image — Heilu for Hellu (13 tokens, the same noise §7 named), VVialen for Wialén (¶442), oh for oli (¶523), knmminkin for kumminkin (¶484). One of these ran the other way and is the section's warning: the text layer read Heiskasen at ¶490 where the 1917 text has Heiskanen, and the OCR was right (§10c). The mechanical diff is a finding aid in both directions.

Twenty-three real print divergences, below.

10a. The reprint standardising the colloquial layer — 2 sites

The hand is weaker here than in ¶2–206. §7a found five standardisations in 5,372 words; this is two in 4,308. And ¶392 is the control: «vähän ihven makua kumminkin» reads kumminkin in both printings, fifteen paragraphs before the reprint changes the same word at ¶407. §7a's "a tendency, not a policy" holds, and the tendency is thinner than it looked.

10b. The reprint capitalising the divine — 4 tokens, 2 sites

¶407 «ellei herra tulisi» and ¶409 «herra antoi, herra otti, herran nimi kiitetty olkoon» (image, p. 69); the reprint reads Herra/Herran at all four. A case-sensitive comparison over the whole novella returns only six divine-word divergences — these four plus ¶294 luoja/Luoja and ¶306 herraa/Herraa — and the first edition sets Jumala with a capital at 31 of 31 tokens. The reading this invites, and why it was refused, is D100.

10c. The reprint changing a word — 4 sites

10d. The reprint adding words — 3 sites

With ¶105's alkaapäin and ¶339's äskeinen this makes five places where the reprint has words the first edition does not, against three where it removes them. §4's account of W1's hand as one that removes is now decisively wrong in the other direction as well, and the two ¶502/¶538 additions are both inside poor characters' speech, smoothing an asyndeton and doubling a phrase.

10e. The reprint removing words — 1 site

¶554: 1886 «Aamuisilla hän oli jo poissa, kun he unestaan heräsivät» (image, p. 87); 1917 «Aamusilla hän oli poissa». Two changes in one clause, and jo is the one that carries — he was already gone against he was gone.

10f. Divergences taken from the copy-text that change no English word — 7 sites

¶404 jalastaan/jalasta (p. 68) · ¶413 pitä/pitää (p. 69; the same page prints pitää at ¶415) · ¶473 hengeltä/hengiltä (p. 75) · ¶528 ammoitti/ammotti (p. 82) · ¶536 ensimäisenä/ensimmäisenä (p. 82; the first edition writes Ensimmäisen at ¶519, so it is inconsistent with itself) · ¶553 pilkkoisten/pilkkoisen (p. 86) · ¶549 silmiin./silmiin, (p. 85; the reprint's comma before a capital Mutta is its own error).

10g. The first edition's own compositor's slips, where W1 is followed — 4 sites

¶531 työnnettin for työnnettiin (p. 82) · ¶549 pyhki for pyyhki (p. 85) · ¶551 samaan for saamaan (p. 86) · ¶552 «saada kakluuniin ja / ja sieltä taas pois» (p. 86), a dittography across a line break. Neither printing is clean in this stretch either, which is what V21's eclectic clause is for.

10h. Not adjudicated against the image

¶385 mennessään/mennessän · ¶387 sillinpäitä/sillin päitä · ¶392 piipunnysää/piippunysää · ¶423 päänaluista/pään aluista · ¶437 niin pian/niinpian · ¶420 myrtinoksista/myrtin-oksista · ¶480 Mitävarten/Mitä varten · ¶482 terveydenhoitokin/terveyden hoitokin · ¶483 armonaikaa/armon-aikaa · ¶484 kanssanne/kansanne · ¶546 seljässään/seljassaan · ¶548 hiljempää/hiljempaa. None affects a rendering — they are word-division, compounding and diacritic rows — and they are listed so that the adjudicated set is not overstated. Gutenberg's -- for the first edition's em dash (¶453, ¶500, ¶516) is a transcription convention, not a variant.

11. Where the copy-text question stands with the collation complete

The whole novella is collated: all 13,522 words of ¶2–572, ¶2–206 and ¶296–375 in §7–§8, ¶207–295 in §3, ¶376–572 here. Fifty-nine real print divergences (16 + 8 + 12 + 23), a rate of 4.4 per 1,000 words. Five demonstrated corruptions in the reprint — ¶94, ¶134, ¶197, ¶344, ¶490 — of which two were invisible until the sentence had to be rendered and one (¶490) was found only because a diff row that looked like OCR noise was checked against the image anyway.

What the completed collation changes about §4's account. The reprint was described there as an editorial hand that normalises the eastern layer and removes. Over the whole book it standardises that layer at ten sites (§3a, §7a, §10a), runs the other way at three (§7b), adds words at five (¶105, ¶339, ¶502, ¶538, ¶549) and removes at two (¶263, ¶554); and its five outright corruptions are not editorial at all but compositor's or transcriber's. It is not a policy, it is a printing. V21 stands unchanged, and its reason narrows from the reprint tidies what this translation declares a loss to the reprint is independently corrupt at five sites and unreliable at the margin, and the first edition was set in the author's lifetime.