Repository path: workshop/translations/jeli-il-pastore/R05-v1/span2-fid-prereg.md · rendered 2026-09-09

Span 2 — free indirect discourse: site inventory, policy, and predicted costs

Frozen before any English of span 2 existed. This page was written and committed in its own commit; the translation commit follows it. That ordering is the whole point: a policy written after the prose is a rationalisation, and a "predicted" cost written after the fact is not a prediction.

What it is for. Register row U1 — how to mark free indirect discourse — was left open by span 1 with the reason that two instances cannot set a policy binding on 9,000 further words. Span 2 is the first stretch dense in it. The policy below is derived from the author's own statement of method (S-verga-farina), not from the translator's ear, and it is applied to a site set enumerated from the Italian before translating so that it cannot be quietly fitted to whatever the English turned out to want.

Scope. Span 2 = paragraphs 28–61 of ../source-it-full.txt, 2,013 source words, ending at «Dio ne scampi.» — a narrative break: paragraph 62 opens the fair-day movement.

(Paragraph indices are into the repaired source file — see the translation page's erratum E2. The repair merged a false break and does not touch span 1's range.)

1. Operational definition of a site

A community-voice site is a stretch inside narration — not inside dash-set speech — carrying at least one of:

• (i) guillemets — Verga's own mark. Mechanical.
• (ii) a tense outside the narrative sequence: a gnomic present indicative, or a conditional, inside past narration. Found by reading, not by a parser; every instance is quoted in full below so the set is auditable.
• (iii) an idiom or evaluative epithet attributable to the community rather than to a neutral narrator.* This criterion is a judgment**, and is flagged as such wherever it is the only one met.

Criterion (iii) is the soft one and it is declared soft in advance. The three classes are reported separately in the result so that a reader can discount (iii) entirely and still see what happened to (i) and (ii).

2. The site inventory — 21 sites, enumerated before translating

Every Italian string below was machine-verified present in the stored source at the stated paragraph (note (g)).

Class A — Verga's own guillemets (criterion i), n = 3

Noted in advance, because it is the difference from span 1: span 1's two guillemet sites were whole clauses of proverb («Era piovuto dal cielo, e la terra l'aveva raccolto»; «dove la malaria si poteva mietere»). All three of span 2's are bare noun phrases — the community's word for a thing. The span-1 policy could not see this class.

Class B — tense outside the narrative sequence (criterion ii), n = 9

Class C — unmarked community idiom or epithet (criterion iii — judgment), n = 9

3. The policy, derived from S-verga-farina

Verga's letter to Farina states the method the device serves: the story is retold «colle medesime parole semplici e pittoresche della narrazione popolare», the reader meets the fact «senza … la lente dello scrittore», and «la mano dell'artista rimarrà assolutamente invisibile». Four rules follow.

• P1 — No added frame. Where the Italian gives community voice no attributive frame, the English adds none: no as they said, no people thought, no quotation marks, no italics. A frame restores the lens the method removes.
• P2 — Verga's own marks are kept, and kept as one mark. Guillemets → single inverted commas, no gloss, no expansion. The marked/unmarked contrast is authorial — 8 guillemet sites in 11,474 words — and flattening it in either direction destroys a distinction the author drew.
• P3 — Dash-speech takes English quotation marks, and the attributive tag stays where Verga puts it (after the first clause, inside the sentence), not fronted. Typographic convention, not framing; confirms span 1's practice.
• P4 — The gnomic present stays present. Where the Italian steps out of the narrative past into a gnomic present or a conditional, the English keeps the step rather than harmonising to the past. This is the one carrier of P1 that English shares grammatically, and it survives even when the lexis is neutralised.

4. Predicted costs — three predictions, falsifiable, made now

Prediction 1 — unattributability. Under P1, at least one Class C site will read in English as the narrator's own assertion where the Italian reads as reported community opinion, because English narration defaults to narratorial authority more strongly than Verga's does. The cost falls on accuracy (who is speaking), the gain on voice and style-correspondence. Falsified if the log records no site where P1 produced an attribution English does not carry.

Prediction 2 — P4 will be strained and I will break it at least once. Keeping an Italian gnomic present as an English present inside past narration will at some site read as a tense slip rather than a shift, and I predict I will harmonise there. Falsified if all nine Class B sites keep the shift, or if none of them reads as a slip.

Prediction 3 — P2 fails specifically on the bare noun phrases. Span 1's guillemets were proverbs, where inverted commas read as quoting. A bare NP in inverted commas in English reads as scare-quoting — ironic distance — which is the reverse of what Verga is doing (solidarity with the community's word). I predict P2 costs something at A1 «nel fatto suo» in particular, where the phrase is also idiomatically opaque. Falsified if all three Class A sites take inverted commas without strain.

And a fourth thing, predicted as a conflict rather than a cost. P1 collides with register row V4. V4 (adopted by drift at D19) says contractions in speech, none in narration. An unmarked community-voice site is narration by punctuation and speech by voice; under V4 it must go uncontracted, which pulls directly against P1's purpose. I predict the collision is real at Class C, and that resolving it requires a third register category — the narrator speaking in the community's voice — which no short unit has ever needed because no short unit has both sustained dialogue and unmarked FID. Falsified if V4 as written handles every Class C site without amendment.

5. What settling U2 requires

Span 2 contains 28 dash-set speech turns across paragraphs 29–59, including two sustained exchanges (the death-bed reckoning, ¶46–48; the leave-taking, ¶52–59). That is the "scene of sustained dialogue" U2 named as the test. U2 is settled by whether V4 survives it — separately from the P1 collision above, which is about narration, not speech.

Post-freeze correction, 2026-07-27, marked rather than silently applied. The figure 28 in §5 is the count of em-dashes, not of speech turns: four dashes close a turn rather than open one (¶44, ¶57, and the — oppure — pair at ¶43). The turn count is 24. Caught by note (g) at hand-off, after the freeze. The error does not touch the site inventory, the policy or the predictions — §5 is scene-setting for U2 — and it is corrected here in place rather than in the text above so that the frozen page still shows what was actually frozen.

6. Things span 2 forces that are register business, not U1

Pre-listed so that the log cannot be credited for noticing them afterwards:

• campaio (¶31) against campiere (¶55) — the source uses two words for one job; register row N5 has only campiere.
• ecalibbiso (¶42, ¶43) against eucaliptus (¶42) — the peasants' pronunciation and the correct form, two sentences apart, the first inside Massaro Agrippino's speech and the second in the narrator's own. A marking problem in miniature.
• tumoli (¶46) — a Sicilian dry measure; N20 keeps onze untranslated and this must follow or be argued.
• compare (¶41 ff.), zia Agata (¶43), zio Cosimo (¶51) — three more address terms for D4's fork.
• La Mara (¶28) — the article before a personal name; N11 dropped it for la Bianca.
• 2 onze e quindici (¶47) — the source sets the numeral as a digit inside speech.