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

«Köyhää kansaa» — the binding register

R05 rule 4. What later spans are bound by, each citing the log decision that created it, plus the questions the next span is required to settle. Revisions require a numbered erratum on R05-v1/translation.md, not a silent edit here.

Bound

Terminology fixed so far

Settled by span 2 — the five questions, answered

1. Jumal' antakoon — V10 did NOT survive. Erratum 1; V10 → V10a. The formula does not "recur": ¶79 is its only occurrence in the novella, and D11 wrote a rule for it in span 1 without reading it in place. In place, the exchange is a secular greeting from above answered by a pious formula from below, and cutting the formula deletes the work's first class marking. Standing consequence: V17.
2. V1's period English held the rent conversation, and did not turn Dickensian. The reason is worth recording: the scene's own register is transactional, not pathetic — Heikura is polite, reasonable, and takes the bed — and V1's plain end ("right enough", "what's to be done about it", "not a stick of furniture") sits on that without strain. No erratum. V1 stands, now tested on four cross-class exchanges.
3. The count was taken, and the question was pointed the wrong way. hän-resolution forced 0 added nouns in span 2 (span 1: 1), because Canth disambiguates for herself at the one dangerous site (¶130 supplies «rouva» mid-sentence). What forces additions is the zero person and the passive: 8 sites (¶86, ¶95, ¶107, ¶115 ×3, ¶126, ¶131), all enumerated in D45 and all class-marked in English (V15). The running count is the zero-person count, not the hän count; span 3 continues it.
4. ~~raahtia / hennoa: the collapse costs one more site in the whole novella, and stands.~~ CLOSED TOO EARLY — CORRECTED, span 4, D74. There is a third verb: raskita, at ¶287, the novella's only occurrence, rendered with the same English idiom ("had not the heart to"). The collapse therefore costs four tokens and three verbs, not three and two. raahtia ¶8, ¶164; hennoa ¶25; raskita ¶287. The conclusion is unchanged — four tokens across 13,525 words is not worth manufacturing an English distinction for — and the family is now closed by whole-text search, not by prediction from a span.
5. ~~Escalated, not settled.~~ DISCHARGED, span 4, D65–D66 and ../collation.md. The 1886 Edlund first edition is free, reachable and independent; the three e-texts the project could find are one transcription of one 1917 reprint. The first edition reads Ei at ¶94 and ei at ¶197 (both read off the page scan): the working text is corrupt at both, and both renderings were already right. ¶228's unbalanced quotation is not a corruption — it is in the first edition too (D72). V21 makes the first edition the copy-text.

Settled by span 3 — the four questions, answered

1. The honorific at ¶175 was marked, measured, and un-marked, all in one session. Canth supplies «Äiti» four times in the three utterances around the site, so the kin term is already spent — a compensation competes for the same slot as the source's own marking, and this is the first time in three spans that the slot was occupied. "Please to get up!" was taken at this site only, put blind to three seats with one string varying, and it worked: +1.167 on the girl's standing rating, exact permutation p = 0.0368, three seats of three moving the same way, five of six quoting the device by name, and the mother's direction unmoved. It was reverted anyway, on the naturalness tripwire frozen before dispatch (0.833 against 0.50). V18 → V18a; Erratum 4. Full record: wiki/findings/results/RS-20260803b-honorific-carry.md. The loss at this site is now this translation's choice, made on a measurement, and not a limit of English.
2. The direction reversal does not reverse the address forms; it abolishes them. In the whole visit (¶172–194) the two ladies speak three times and all three to a four-year-old boy; they never address Mari. Mari speaks to them twice and uses no second-person form of any kind. V13 gets no second clause — its site did not recur, because the scene contains no address across the class line in either direction. English carries this for free: it is a fact about who speaks to whom, not about grammar. D58.
3. The zero-person count continues: span 1 = 2, span 2 = 8, span 3 = 7, running total 17, all seven enumerated at D61, with one deliberately uncounted borderline (¶198) listed there too.
4. The comparator priced two further things, and one of them corrected the span. Erratum 3: «toista vuotta» was rendered "this two years and more" — a disjoint interval — and is now "going on two years now"; Hertzberg reads «fem år» at the same site, which is wrong on any reading, so the witness that caught the error is worse at it. And V15's over-specification is a fact about the target, not a translator's failing: seven English sites in this span need a class-marked generic and Swedish needs none, because man is neutral and Hertzberg uses it six times in the road scene alone.
5. A second printing of the Finnish, before span 4 — still outstanding, and now sharper. ¶197 reads «Vesi el kumminkaan mitään maksa». The i/l substitution now appears twice (with ¶94's «EL»), which makes a typesetting or OCR fault much likelier than deliberate emphasis. The question for a second witness is now specific: does it read ei at ¶94 and ¶197. D62.

Settled by span 4 — the four questions, answered

1. arvon — the gloss sits, and D53's reason for it was wrong. D73. There are four occurrences, not three: ¶153, ¶194, ¶221 (narration, unknown to D53) and ¶232, which is dialogue with a lausui tag and not the narration D53 predicted. "Sure enough" sits at both span-4 sites. No erratum. What is recorded is that a decision can be right and one of its stated reasons false, which is V17's lesson holding even where the call came out well.
2. ~~V18a needs no re-test; it needs a count. Both known honorific-to-a-parent sites…~~ ANSWERED, span 4, D78 — and the count in this entry was wrong by half. There are four sites, not two, all Hellu to her mother: ¶67 «Elkää, hyvä äiti, elkää» (span 1, already translated plain, and behind the span-by-span counter), ¶103 «Viivyttekö kauan?», ¶175 «Nouskaa ylös!», ¶389 «Äiti, kuulkaa…» (chapter II). All four confirmed in the 1886 first edition. Span 4 has none. No erratum — V18a mandates plain and all three translated sites read plain. The loss is worth four sites in 13,525 words. The count was run over the whole text the moment the question was asked, rather than accumulated span by span, and that is the only reason the span-1 site was found; V17 forbids a rule for an unread site and does not forbid a count.
3. The prediction held at 15 sites out of 15, and it changed what te is called. D77; RS-20260803g-distance-not-deference. Every second-person form between acquainted adults of the poor class in span 4 is T-singular — not one te. The six te events are plural by number (four beyond argument; ¶222 and ¶240 contested by an adversarial seat and carried in the result's limits) and one is V-plural upward to officialdom in reported speech (¶248), which is the cell D60 said the book had not supplied. The whole-text distribution then broke the project's own vocabulary: te is symmetric across the landlord/tenant gap and is therefore not an honorific. V23.
4. ohimovika (D50) is still unresolved and was not re-opened this session — the Swedish remains the only reachable witness, and whatever it says is a reading, not an authority. Two further words join it as declared readings, D81: kuoninki (¶248, rendered "the hem") and vantis (¶293, rendered "by its bow"). Both read identically in the 1886 first edition, so neither is a corruption to be argued away.

Settled by span 5 — the three questions, answered

1. Spans 1–3 collated. DONE, and it cost three sentences. ¶2–206, 5,372 words: 16 substantive divergences, three of which change an English word (Erratum 5: ¶73, ¶105, ¶163) and two of which vindicate a rendering already made (¶102, ¶134). A third corruption of the ¶94/¶197 letter-substitution family was found at ¶134, inside a clause the translator had had to guess at. And the reprint's hand is not the one-directional tidying ../collation.md §4 described: it standardises the colloquial layer at five sites, runs the other way at three, and adds a word at two (¶105, ¶339). §§7–9 of ../collation.md.
2. Chapter II is not yet the falsifier, and V23 gains a cell. DONE for this span. Span 5 has 17 utterances carrying second-person forms and not one te — every form T-singular, sixteen of them between husband and wife and one of them Mari to God (¶301), a cell the novella had not previously supplied. V23 stands. ¶376–573 is still unread and the exposure is still open. D97.
3. ¶240's addressee — untouched, as span 4 predicted it would be. Span 5 supplies nothing on it. It goes to the craft report as a declared hole unless span 6 or 7 supplies a parallel.

Settled by span 6 — the five questions, answered

1. ¶376–572 collated. DONE, and the whole novella is now on the copy-text. 4,308 words, 68 surviving rows, 23 real print divergences, of which thirteen fall inside span 6 and are applied in ../span-6-source.txt before a word was translated. A fifth demonstrated corruption in the reprint, ¶490 Heiskanen for Heiskasen, found only because a row that looked like OCR noise was checked against the image anyway. ../collation.md §§10–11. No new gate is owed on the text: ¶442–572 is collated and waiting for span 7, and §11 restates what V21's reason now is.
2. V23's exposure closes, and it closes by being sharpened rather than confirmed. V25. ¶376–572 contains five single-addressee te events — four of them mutual between Tiina Katri and Holpainen, one of them the last child→mother honorific (¶389, D101) — and not one te between people inside a family. But the same stretch supplies the within-speaker control the earlier spans could not: Tiina Katri says sinä to Mari and te to Holpainen, and both are returned. te is therefore non-kinship, not non-intimacy. D103.
3. V24 got no test, and no rule was stretched to give it one. §Unresolved 3 asked what happens if a span supplies a zero person that is neither a command nor propositional. Span 6 supplied no zero-person command at all and only two zero-person/passive sites in 1,733 words — the lowest count of any span (D111). The question is carried to span 7 unanswered, which is the honest state of it.
4. D96's generic got no deciding site either. ¶376's «eihän niitä mielellään tahtoisi antaa» is a plain woman speaking about mothers and inside her own statement, so V20 governs it cleanly and the ¶365 case — speaker outside, referent inside — did not recur. Carried.
5. ohimovika (D50), kuoninki and vantis (D81) remain declared readings. All three now sit inside a fully collated text and all three read identically in both printings, so none of them is a corruption to be argued away; they are hard words, and nothing this project can reach settles them.

Discharged — what span 6 was required to settle (all five answered above)

1. ¶376–573 has never been collated — 4,309 words, 32% of the novella, and the largest unchecked stretch left. The four corruptions found so far were found at a rate of about one per 2,300 collated words, and two of the four were invisible until the sentence had to be rendered. Collating ¶376–573 is span 6's first task, before any new prose, on §8's terms.
2. V23's exposure closes or fails in ¶376–573. A te between intimates there falsifies it. Every second-person site is recorded as it is translated, against the grid.
3. V24 has one span's evidence and one construction's. The bare imperative carried the zero-person command; the propositional zero person still takes V15's generic. If span 6 supplies a zero person that is neither — a question, or a conditional — the rule is under-specified and must say so rather than be stretched.
4. D96's generic is unresolved on purpose. ¶365 needed a generic where the speaker is outside it and the referent inside; V15 and V20 both fail there and no rule was written, per V17. The next such site decides it.
5. ohimovika (D50), kuoninki (D81) and vantis (D81) remain declared readings, not authorities. Nothing this session touched them.

Settled by span 7 — the four questions, answered, and the register closes

1. ¶449's Swedish stays Swedish. V26, D115. Three options were live — keep it bare, English it, or keep it and add in Swedish to the tag — and the bare one is taken, unitalicised, in the same double quotes as everything else. Englishing it deletes the book's sharpest class device at its most loaded moment: two gentlemen deciding over a bound woman's body, in a language her household does not have, whether she is worth curing. The cost is a change of the reader's seat, not a loss of information, and the study limb that went looking for its price could not measure it: every one of four seats shown the bare Swedish reported its content, two of them by translating it. RS-20260805g.
2. V1 reaches educated speech, and the rule says which of its clauses did the work. V27, D117. The pastor–doctor argument is the novella's thesis and V1's plain-end examples reach none of it. What was applied is V1's prohibition — nothing that reads as contemporary literary fiction — and "purge your religion; that doctrine won't do any longer for the people of the present day" is period and is not that. V1 is under-specified, not too narrow, which is the third rule in this register found under-specified by a referent it had never been tested against. And V16's clitic diagnostic does not survive these two speakers at all (D118): the doctor and the pastor carry the colloquial clitics at Heikura's density, so they take contractions like everyone else, and what separates them is subordination and abstract vocabulary, which English carries without a rule. V16 stands as an instruction and is withdrawn as a description.
3. V24's carried question is answered, and the zero person has a third class. D120. §Settled-6 item 3 asked what happens if a span supplies a zero person that is neither a command to oneself nor propositional. ¶484 supplies four in one paragraph — «julistetaan», «saa tappaa», «Juljetaanpa», «varoitetaan», «Ei ajatella» — and they are agent suppression, not generics: the doctor's indictment of war never once names who declares it, who is permitted to kill, or who prays for the blessing. Rendered with English passives and dummy subjects, never with V15's generic, which would put a man where the source has deliberately put nobody. V15 and V24 both stand and V15's domain narrows to the genuinely propositional case. The highest zero-person density in the novella is the educated speaker's, doing the opposite work from everyone else's. Whole-work count: 2, 8, 7, 16, 8, 2, 10 — 53 sites.
4. D96's deciding site arrived, and the answer was in the previous sentence. D121. §Settled-6 item 4 carried an unresolved zero person where the speaker stands outside the implied agent. ¶541: Tiina Katri's «Antaisi poloisille edes puhdasta ilmaa», reproaching an unnamed responsible party she is plainly not part of. Rendered "They might give the poor creatures clean air at least" — and the English they is not invented, because the source supplies ne in the immediately preceding sentence, so the zero person is anaphoric to an indefinite agent already on the page. A reproaching zero person takes indefinite "they", never the generic. No rule is written: there is no further span to bind, and this belongs to the craft report.
5. The three loaded copy-text sites, and one of them cost nothing. D114. ¶502's undoubled «Ei saa sitoa» makes Mari less mad than the reprint does — she asks once and then obeys, and the obedience is the point of the paragraph that follows; ¶549's absent niin, keeps a child's thought running on flat where the reprint puts a breath into it; ¶554's jo is the whole content of its sentence, and without it his leaving and their waking become simultaneous. ¶538's asyndeton is the one the reprint's smoothing does not cost anything, because English has the asyndetic pair and "that did his good, his bad, under him" carries it.

The novella is translated whole and this register is closed. There is no §Unresolved for span 8, because span 8 is the craft report and binds no prose. Three declared readings stand unresolved and are not going to be resolved by anything this project can reach: ohimovika (D50), kuoninki and vantis (D81), joined at the last paragraph by konttuori (D126).

Discharged — what span 7 was required to settle (all four answered above)

1. ¶449 «Kan hon botas?» is Swedish inside the Finnish, and Canth does not gloss it. At the deathbed the pastor asks the doctor, in Swedish, whether the woman can be cured, and gets no answer; at ¶457 the two of them «vaihtoivat muutaman sanan ruotsiksi». This is the device RS-20260802d found English better placed to carry than Swedish is — «mutta ruotsiksi, ettei Mari ymmärtänyt» (¶130) depends on Swedish being a switch away from the language of the book. Here the switch is not narrated, it is simply printed, which is the harder case and the one the register has no rule for. Decide before translating: does the Swedish stay Swedish on the page.
2. The pastor–doctor dialogue (¶463–488) is the novella's argument, and V1 has never been tested on educated speech at length. V16 gives the merchant class uncontracted verbs and everyone else contractions; a socialist doctor and a Lutheran pastor arguing about euthanasia and the causes of poverty are neither. V1's plain end may not reach it, and if it does not, the rule must say so rather than be quietly widened.
3. V24 and D96 carry over from span 6 undecided (§Settled-6 items 3 and 4).
4. ¶502 and ¶538 are the loaded copy-text sites in span 7's range — the first edition says «Ei saa sitoa» once where the reprint says it twice, and «joka hyvänsä pahansa» without the reprint's ja. Both are inside poor characters' speech and both are places where the reprint smoothed. Translate from the copy-text and record what the smoothing would have cost.