Seven Renderings of Effingers Chapter 25 — A Comparative Analysis

A study of Chapter 25 (“Frühling”) of Gabriele Tergit’s Effingers (1951) rendered into English by A, B, C, C-let, D-let, D-aim (six AI arms differing on a persona-source axis and a brief-nudge axis), and H (Sophie Duvernoy’s published English version, NYRB 2025). All seven render the same German source; differences are the unit of analysis.

The analysis is in three layers, addressed in order:

1. The two controlled probes: C ↔︎ C-let (no-persona license effect) and D-let ↔︎ D-aim (persona-anchored license vs school instruction).
2. The seven-way comparison: where the arms diverge, where they cluster, what the personas say and how that shows up, and where H sits in this geometry.
3. Patterns that emerge from a free read — and a specific test of the domesticated Tergit as a region, not a point hypothesis.

The source-chapter is a Berlin Saturday in March 1887, told in seven scenes time-stamped by a recurrent refrain — “Was für ein Frühlingstag, dieser Sonnabend im März des Jahres 1887! Was für eine Süße…” — at 10am, 11am, 1pm, 5pm, 6pm, 8pm, and 3am. Each scene shifts class, register, and dramatic mode: the wealthy Eugenie with her seamstress, the schoolgirl Sofie writing a forbidden love-letter, the Privatdozent Waldemar arguing with an old historian and then bedding the singer Susanna, the working- class Effingers and the ruined banker Mayer on the Chausseestraße, the Friedrichstraße street tableau, and Theodor’s jealous despair ending in a wine-tavern with a 17-year-old prostitute named Wanda. Berlin dialect, embedded songs (Schumann, Wagner, Heine), period honorifics, costume vocabulary, and free indirect speech are all load-bearing.

Layer 1 — The two controlled probes

Probe A — C ↔︎ C-let (no-persona base; system-prompt permission)

Both C and C-let are control arms without a persona. They share model, novel-read, pass budget, and isolation. C’s brief tail is “No style guidance. No epoch guidance. No usage guidance. No format guidance. The interpretation is yours, from the text alone.” C-let replaces that tail with a “Style permission” block that explicitly authorizes (but does not require) anglicization of foreign vocabulary, plain English for dialect, paragraph fusion, English rhyme for embedded lyric, and domestication of culturally specific references. The only experimental difference is that brief tail.

Where the permission moved C-let. The clearest movements are: a structural cut, a dialect-and-idiom shift, a chiastic-line naturalization, and a refrain compression.

The structural cut is the most striking. C-let deleted Paul’s exclamation “»Das Leben!« sagte Paul” and Mayer’s sarcasm preamble “Wie recht Sie haben. Wenn ich noch den Sarkasmus meiner Jugend hätte…” entirely. Where C, A, B, D-let, D-aim, and H all preserve “Such is life!” / “How right you are. If I still had the sarcasm of my youth…”, C-let merged the speeches and dropped the line. C-let’s pass-1 log openly cites the permission as authority for cutting: “the system prompt grants that permission”. No other arm cut at this scale.

The dialect-and-idiom shift. For the carter’s tirade — “Sie doofe Ziege, Ihnen müßte man die Hammelbeine langziehen, damit Sie sehen, det hier’n Ferd steht” — C gives “You daft goat, somebody ought to stretch your mutton-legs for you, so you’d see there’s a horse standing here.” C-let gives “You stupid cow, somebody ought to give you what-for, then you’d see there’s a horse standing here!” Permission moved C-let off the literal Berlin calque (“Hammelbeine” = mutton-legs, a fixed idiom meaning “give a piece of one’s mind”). C kept the visual livestock image because the pass-1 log explicitly noted that “Fresh Mutton” (the pub name) “chimes with the mutton-leg insult & the wool/mutton market” — C heard a within-chapter motif and preserved it. C-let dropped the motif for natural English.

The chiastic-line naturalization. Waldemar’s “Schäm dich, daß du dich schämst!” — a tight reflexive paradox the German lets echo. C: “Be ashamed that you are ashamed!” (preserved exactly). C-let: “Be ashamed of being ashamed!” (idiomatic English).

The refrain compression. C’s refrain runs “What a spring day, this Saturday in March of the year 1887! What sweetness, at ten o’clock in the morning!” C-let drops “of the year” and “o’clock”: “What a spring day, this Saturday in March of 1887! What sweetness, at ten in the morning!” Both elisions move toward the published human translation H, which also drops “of the year” and “the morning” only (keeps “o’clock”). C-let’s pass-1 log notes the variant: “kept the German’s three deliberate VARIANTS rather than regularizing” — meaning it preserved the German’s own internal variation (the 6pm refrain in German drops “des Jahres”; C-let extended this compression to the regular refrain too).

Where the permission did NOT move C-let — and where it moved AWAY from H. Permission did not produce a wholesale anglicization. C-let kept Fräulein and Herr throughout, with the pass-2 log spelling out the choice: “Fräulein / Frau / Herr kept throughout. Decided against Anglicizing in this pass; they’re functioning as period markers and class-form markers that English’s Miss / Mrs. / Mr. would flatten.” It kept Privatdozent: “no English equivalent.” It kept Frischer Hammel in German italics, where C calqued to “Fresh Mutton.” And on three lyrical-cultural titles, C-let moved toward the German where C had translated: the Schumann cycle “Frauenliebe und -leben” (C-let; “A Woman’s Love and Life” C), the Wagner “Feuerzauber” (C-let; “Magic Fire Music” C), and the pub name. C-let used italics as a cosmopolitan marker — preserving the German word as a recognized foreign object rather than translating it.

Most surprising: in pass 3, C-let actively reversed a London- domesticating move. Pass-1 had “in the east end of Berlin”; the pass-3 log corrects to “in the east of Berlin” with the reason: “East end carries London (the East End) in English ears. German Berliner Osten is just ‘the east of Berlin’ — generic compass direction, not a specific district name.” Permission did not make C-let reach for substitution-domestication; it sometimes pushed it to avoid it.

Pattern at the chapter level. C-let did not walk to a different conception of the chapter from C. It read the chapter the same way — the same scene structure, the same character voicing, the same refrain spine — and applied the permission selectively at the sentence level, scoring genuine domesticating moves on dialect and chiastic shape and structural compression while foreignizing on musical/pub titles via italics. The text is 115 words shorter (-2.6%) — consistent with the cut and the tightened dialogue. The permission worked as a license, not as a school: read as “you can move here where it serves,” not as “move all the way over here.”

Probe B — D-let ↔︎ D-aim (shared persona; permission vs school)

D-let and D-aim share the same persona built from the expanded corpus (B’s German materials plus Wharton, Powell, Mitford, Isherwood), the same step-1/2/3 notes, and the same persona- revision-after-novel. The conversation forked at Step 4. D-let ran with the standard kickoff (relying on the permission already in the system prompt — the same paragraph C-let had). D-aim’s Step-4 kickoff replaced the permission paragraph with a categorical school instruction:

“For this translation, the operating principle is domesticating translation, in Schleiermacher’s sense — bring the author to the reader, not the reader to the author. … Faithfulness operates at the level of meaning, scene, character, voice, and effect; not at the surface level of structure, syntax, or vocabulary.”

The persona is shared. Only the Step-4 kickoff differs.

Where the school instruction moved D-aim. The movement is heaviest on honorifics, academic ranks, social-historical lexicon, song titles, and period vocabulary.

Honorifics. D-let writes “Fräulein Winkel,” “Fräulein Mayer,” “Fräulein Kelchner,” “Herr Mayer,” “Herr Effinger,” “Herr Kollege.” D-aim writes “Miss Winkel,” “Miss Mayer,” “Miss Kelchner,” “Mr Mayer,” “Mr Effinger,” “My dear colleague.” This is the only arm in the seven that anglicizes Fräulein and Herr to Miss and Mr. Even H — the published human translation — keeps “Fräulein Winkel” and “Herr Mayer.”

Academic rank. D-let: “Privatdozent Waldemar Goldschmidt.” D-aim: “Waldemar Goldschmidt, the young lecturer.” The two pass logs explicitly contradict each other on this word, citing different authorities. D-let’s pass-1 log: > “Kept Privatdozent (no English equivalent), Herr Kollege, > Kaiserreich, Monumenta, Junker, and the Latin tag in verba > magistri”

D-aim’s pass-1 log on the same word: > “Privatdozent → the young lecturer (rejected: keeping the > German loanword in italics; chosen: anglicising fully — the > contemporary anglophone reader does not parse ‘Privatdozent’ > without a footnote, and footnotes are forbidden by the > domesticating principle)”

The persona was the same; the cited authority differs (persona- derived “no English equivalent” rule vs. school-derived “footnotes forbidden” rule). The school instruction overrode a persona- specified default.

Social-historical lexicon. D-let kept Kaiserreich, Kaiser, Hessen, Schloß, Stadtrat, Stadtwald. D-aim translated all of these — to Empire, Emperor, Hesse, the Palace, the Councillor, the woods outside the town. The persona’s stated principle (“I will keep Geheimrat, Kommerzienrat, Maultaschen, Stollen, Verlobung, where they carry irreplaceable cultural weight; I will translate them where they can land cleanly”) is applied differently because the school instruction shifted what counts as “landing cleanly.”

Song titles. D-let kept the Schumann cycle as “Frauenliebe und -leben” and the Heine song verbatim as “Die Lotosblume ängstigt …”. D-aim translated both: “A Woman’s Love and Life” and “The lotus flower trembles…”. The shift on “Die Lotosblume” is particularly stark — D-let preserves the German lyric line entirely intact; D-aim renders to English with the dramatic verb “trembles” (coinciding, independently, with H’s exact rendering).

Period vocabulary. D-let “peignoir” (French period); D-aim “wrapper” (American period). D-let “tabouret” (period French); D-aim “low stool” (modern English).

Locale. D-let kept “in the east of Berlin.” D-aim wrote “in the East End of Berlin,” with the pass-1 log defending the move: > “Berliner Osten → the East End of Berlin (the English idiom — > London-specific in its origin but instantly resonant for the > anglophone reader and exactly the right social register; perfect > domestication)”

This is the strongest substitution-domestication in any of the seven arms. C-let, given the same permission as D-let, explicitly rejected this move in pass 3 (see Probe A). D-aim, given the school, embraced it. And — important for Layer 2 — H also does NOT go to “East End”: H writes “in the east of Berlin.”

Refrain compression. D-let kept “of the year” in the refrain; D-aim dropped it. Both adjustments are independently applied (D-aim added the line “Tomorrow they were to leave for the Riviera, for Nice, for the Hôtel Barblan.” with a French circumflex on Hotel — a quietly foreignizing move while domesticating elsewhere).

Dialect address. D-let “Paule, must you?” (kept “Paule”). D-aim “Paul, love, must you really?” — added the British working-class endearment “love” (a Mitford or Wharton-novelistic move) where the German has no such word. The lodger and wife dialect-spats are otherwise close to identical across D-let and D-aim — both arms applied the persona’s “plain working-class English with no eye-dialect” rule.

Where D-aim did NOT move from D-let. The voice-models from the persona held across both. D-aim’s pass log explicitly: “The four anglophone voice-models settled in Step 2 — Isherwood-clear, Wharton-periodic, Mitford-quick, Powell-deadpan — are doing the per-scene register work.” Berlin dialect treatment held: no eye-dialect in either. Free indirect speech handled the same. Specific cultural items both kept: Frischer Hammel in German italics (both), in verba magistri in Latin (both), Adieu as period salon-farewell (both), Junkers (both), Annettchen handled similarly (D-let kept; D-aim “my little Annette” — both preserve the diminutive’s pinpoint of cruelty for Eugenie’s recognition), Unter den Linden named explicitly in both (D-let pass 3; D-aim pass 2 — independent rediscovery of the same disambiguation).

Length. D-aim is longer than D-let (4,418 vs 4,326 words — +2.1%), opposite to C-let’s contraction relative to C. The domesticating direction unpacked German compounds and titles into English multi-word descriptions (“Privatdozent” → “the young lecturer”; “Stadtwald” → “woods outside the town”), adding length.

Pattern at the chapter level. D-aim did not walk to a wholly different conception of the chapter from D-let. It kept the persona’s voice models, the scene-by-scene register mapping, the dialect- without-eye-dialect rule, the kept-German treatment of Frischer Hammel and street names and Latin tags. What changed was a band of sociolinguistically-marked vocabulary — honorifics, academic ranks, social-historical terms, song titles, locale-naming — that the school instruction specifically targeted by overriding the persona’s “cultural-weight” exemption clause. The reasoning-language in the pass log shifted accordingly: D-let cites the persona; D-aim cites the “domesticating principle.” Same persona, different cited authority, different specific picks on roughly fifteen items.

The interaction between the two probes

This is the analytically interesting question. The two probes together test: does the same nudge-direction (license to domesticate, in two formats and at two intensities) produce the same effect on a no-persona base vs. a persona-anchored base?

The answer is no, and the asymmetry is illuminating.

Finding 1 — The persona dampens the permission. D-let with the permission kept MORE German than C-let. D-let kept Stadtrat, Kaiserreich, Herr Kollege (C-let either rewrote or didn’t have to encounter these). The expanded-corpus persona — with its anglophone reference writers — provided strong defaults that mostly preserved the German on lexicon, and the permission landed on already-formed preferences. The same brief tail had different effects depending on whether there was a persona for it to filter through.

Finding 2 — The school instruction overrides the persona on specific bands. D-aim’s persona is the same as D-let’s, including the explicit “I will keep [items] where they carry irreplaceable cultural weight” rule. The school instruction overrode this rule for honorifics, academic rank, social-historical lexicon, two of three song titles, and Berliner Osten. The override is visible in the pass log’s switch in cited authority — from persona-derived defaults to “the domesticating principle … footnotes are forbidden.”

Finding 3 — The format matters as much as the content. C-let’s permission and D-let’s permission are the same paragraph, but because C-let has no persona to act as a filter, the permission has to be the operating principle for any case where the source-text alone doesn’t decide. D-let has the persona-as-filter, so the permission is at most a tie-breaker. D-aim’s school instruction in the Step-4 kickoff sits chronologically after the persona reading and is framed as an operating principle (“Every choice — scene by scene, word by word — follows from this principle”), so it acts as a re-anchoring — not a tie-breaker but a new categorical default.

Finding 4 — Both nudge-formats move in the same direction but pick different features to operate on. C-let domesticated dialect- idiom and chiastic-line shape; D-aim domesticated honorifics and academic titles. Neither moved all the same features. The same direction-name (domestication) covers operations in different operational regions of the translation. This is the seed of the region-vs-point finding (Layer 3).

Finding 5 — Neither arm landed at H’s specific picks. C-let overshot H on the “stupid cow” cluster (matching H exactly on the substituted-noun choice) but undershot on the Schumann title (C-let kept German; H also kept German — coincidental convergence on a feature both saw as cosmopolitan, not as foreignizing). D-aim overshot H on Fräulein/Herr (translated where H keeps), overshot on Berliner Osten (D-aim wrote “East End” where H stays generic), undershot on Frischer Hammel (D-aim kept German where H translates to “Young Ram”), and missed H’s structural moves entirely (paragraph fusion, Schumann rhyme, etc.).

Layer 2 — The seven-way comparison

Where the seven diverge and converge

I’ll work through the chapter’s load-bearing loci in order, with the German source and each English rendering, and read off the clustering pattern.

The opening refrain

German: “Was für ein Frühlingstag, dieser Sonnabend im März des Jahres 1887! Was für eine Süße, morgens um zehn Uhr!”

Three clusters. A/B/D-let exactly converge (one verbatim formula). C adds “o’clock.” C-let and D-aim drop “of the year.” H is alone in three ways: (1) adds the adjective beautiful to spring day — where the German has just Frühlingstag; (2) restructures Was für eine Süße from elliptical noun-exclamation to a declarative past-tense clause “How sweet the air was” — inventing a subject (“the air”) and a verb; (3) drops “of the year” and keeps “o’clock.” H’s reading converts the rhetorical form completely.

All five AI arms with personas (A, B, D-let, D-aim) and one without (C) preserve the elliptical noun-exclamation form. C-let shortens within the form. H alone leaves the form. The structural choice on the refrain is the boldest reading-difference in the chapter, and it belongs to H.

“Of the year” and the refrain’s mutations

The German refrain has natural variants at 6pm (“im März 1887, abends um sechs Uhr! Welch ein Gewimmel auf der Chausseestraße!” — drops des Jahres, fuses time into first sentence, replaces Was für eine Süße with Welch ein Gewimmel) and 8pm (single sentence, no second exclamation). All seven preserve the 6pm tail (C: “throng”, B: “swarming”, A: “throng”, C-let: “swarm of people”, D-let: “swarm”, D-aim: “teeming throng”, H: “crowd”). A/B/D-let also preserve “of the year” only in the regular instances and drop it at 6pm (matching the German variation). C-let, D-aim, and H globalize the elision — dropping “of the year” everywhere. This is the simplest evidence of a domesticating direction: drop incantatory specificity that doesn’t add reader-comprehension.

The final 3am refrain has no comma between Süße and morgens in the German — a small one-line typographic irregularity. C-let, D-let, D-aim, A, B, and C all preserve the missing comma in the final refrain (each pass log notes the deliberate match to source). H drops the rhetorical form entirely so the question doesn’t arise. This is one of the clearest tells that the AI arms are all reading the German typography with close attention.

Honorifics — the cleanest probe of the domesticating axis

D-aim is the only arm that anglicizes Fräulein and Herr to Miss and Mr — and it does this despite sharing a persona with D-let that has the same “keep where it carries weight” rule. D-let is the only arm that keeps “Herr Kollege” and “Stadtrat” in German.

Two interesting non-trivial converges: - A’s persona explicitly states “honorifics simplified — the Prussian Rat titles all to ‘Councilor,’ ‘Frau Kommerzienrat’ to ‘Mrs. Oppner’ — because rendered literally they go stilted and lofty-wrong.” A applied this to Stadtrat → “the Councilor” but did NOT apply it to Fräulein → Miss. The persona’s principle was inherited from H’s writing, and H also does not anglicize Fräulein/Herr. Two arms (A and H) converge on NOT anglicizing Fräulein/Herr despite both having the simplification principle available. - C-let’s pass-2 log explicitly considered Anglicizing: “Fräulein / Frau / Herr kept throughout. Decided against Anglicizing in this pass; they’re functioning as period markers and class-form markers that English’s Miss / Mrs. / Mr. would flatten.” The permission was used to think the question through, not to trigger the move.

D-aim alone is at the “all the way to Miss/Mr” end. Its pass-1 log: “Privatdozent → the young lecturer (rejected: keeping the German loanword in italics; chosen: anglicising fully — the contemporary anglophone reader does not parse ‘Privatdozent’ without a footnote, and footnotes are forbidden by the domesticating principle).” The school instruction operated as the authority for the override.

The Schumann lyric (Sofie’s letter)

German: “›Seit ich Dich gesehen, glaub’ ich, blind zu sein, immer wie im Traume seh’ ich Dich allein.‹” (Sofie misquotes Schumann/ Chamisso — original is “Seit ich ihn gesehen, glaub’ ich blind zu sein, wo ich hin nur blicke, seh’ ich ihn allein” — Sofie personalizes with Dich for ihn and rewrites the second clause.)

The rhyme. Only H invents an English rhyme (blind/mind), which the German source does not have at all (Chamisso’s poem has a rhyme scheme but Sofie’s misquotation breaks it). H’s pick is a poetic-translation craft move: replace the song-line’s rhythmic shape with an English equivalent that feels sung. D-let attempted a different kind of period-marking by using archaic “Thee” — a King James register signal. D-aim added a line break (/) to suggest couplet shape but no rhyme.

The Du-respect convention. B and D-let alone capitalize “You” in the love-letter, preserving the German Du-letter typographic convention. A, C, C-let, D-aim, and H all use lowercase “you” in the letter. The convention is a period-letter formal feature; D-let explicitly notes it: “Period English love-letter register. Capitalised You / Your per the German Du convention, which English period letters also observed.” This is the persona’s operational reasoning at work.

Cultural song/work titles

Two convergences worth flagging. First, C-let and H both keep the German titles for Frauenliebe and Feuerzauber. H (the domesticated human) preserves these German cultural names; C-let (the no-persona permission-armed arm) also preserves them. The permission armed C-let to use italics as a cosmopolitan flag rather than as a calque trigger. The arm independently invented a solution that matches H’s published choice.

Second, on “trembles” for “ängstigt” — D-aim and H converge on the identical verb. The German ängstigt (causes anxiety / makes fearful) is rendered literally by A (“is afraid”), B/C (“fears”), C-let (“fears”). D-let preserves the German entirely. D-aim’s pass-1 log: > “Die Lotosblume ängstigt → The lotus flower trembles > (italicised as a song-quotation; ‘trembles’ carries the > ängstigt-emotional-shrinking better than ‘fears’ or ‘is afraid’).”

H also chose “trembles” — independent convergence on the better English verb. This is one of the few exact matches between D-aim and H. It comes from each translator independently reading the German and judging which English verb captures the affect.

“Frischer Hammel” — the pub name

H is alone in inventing an English pub name — “Young Ram” reads as a believable English pub (compare “Bull and Bush,” “King’s Head,” “Coach and Horses”). B and C calque (“Fresh Mutton” — a direct translation that doesn’t read as an English pub name). A, C-let, D-let, and D-aim keep the German. The three nudged AI arms (C-let, D-let, D-aim) cluster on German-italic preservation. H’s move requires translator-pragmatic invention — knowing what English pubs sound like and being willing to invent rather than translate. No nudge-direction conveys this; it is a craft pick.

The Berlin dialect — the carter’s tirade

German: “»Sie doofe Ziege, Ihnen müßte man die Hammelbeine langziehen, damit Sie sehen, det hier’n Ferd steht.« — »Mann, schimpfen Sie doch nicht so!« — »Reden Sie nich so mit die frisierte Schnauze!«”

Three clusters. - A, B, C stay on the German Hammelbeine metaphor with goat/ goose substitutions for Ziege. - C-let, D-let, D-aim, H abandon the metaphor and substitute “cow” for Ziege — but each picks a different English paraphrase: what-for / hiding / thrashing / good slap. Four arms, four picks. - H goes further: “Can’tcha see” eye-dialect, and rewrites the “frisierte Schnauze” line entirely as “I’ll have none of yer fancy talk” — abandoning the snout/mug imagery for a register comment on speech itself.

This is one of the cleanest pieces of evidence for the region-vs-point hypothesis: the four domesticating arms all share the direction (abandon literal mutton-legs) but each picks a different English destination.

Free indirect speech

The chapter’s load-bearing FIS moments — Volksbegeisterung, dachte Waldemar; Guter, ahnungsloser Papa! dachte Amalie; Wie es stank im Hausflur!; Welche Wohltat … ehrlich sinnlich sein zu dürfen; Der Dämel! dachte sie — are handled differently:

For Amalie’s Guter, ahnungsloser Papa! dachte Amalie all seven preserve the form essentially literally: “Good/Dear, unsuspecting/ innocent Papa! thought Amalie.” This is unanimously preserved as exclamation-with-thought-tag. H’s “Dear, innocent Father!” is the only one to use “Father” instead of “Papa” — domesticating the address-form to match how an English bourgeois daughter would address her father.

For Susanna’s Welche Wohltat … ehrlich sinnlich sein zu dürfen! the renderings are: - A, B, C: “What a blessing, thought Susanna, … to be allowed to be honestly sensual!” - C-let: “What a blessing, Susanna thought, as she lay beside him — to be allowed, at last, to be honestly sensual.” (C-let adds “at last” — interpretive intensification.) - D-let, D-aim: “What a relief, thought Susanna … to be allowed to be honestly sensual.” - H: “How good it feels, thought Susanna as she lay next to him, to be honest about my desires!”

H alone breaks into first-person — my desires — making the FIS into direct interior monologue. And H rephrases Wohltat / to be allowed to be honestly sensual into a more colloquial How good it feels … to be honest about my desires. The German construction (impersonal Wohltat + dürfen = “be permitted”) is naturalized to a direct first-person sensation. This is the most interpretively-active free-indirect rendering in the seven.

For Wanda’s Der Dämel!: A and C-let “The chump!”, B “The great fool!”, C “The ninny!”, D-let “Idiot!”, D-aim “The idiot!”, H “What a sop!” — six different English picks for one German word. None matches another exactly. The German Dämel is Berlin slang for soft-headed fool; H’s “sop” is the most regionally-marked British/American period word; D-let and D-aim cluster on “idiot”; A, B, C, C-let on the period-British (chump/fool/ninny) cluster.

Length

H is the shortest. A is second shortest (A’s persona explicitly: “the lightest possible hand — no additions, no rewriting, no explaining”). The within-probe comparison is informative: C-let SHORTER than C (-115 w; cut + tighter dialogue), D-aim LONGER than D-let (+92 w; unpacking compounds and titles). Same nudge family, opposite length effect — because the no-persona arm contracted (cut a sentence, naturalized a chiasmus) while the persona-armed arm expanded (unpacked German compounds into descriptive English).

Clustering across the seven

No two arms have the same feature profile. But three approximate directions are visible:

1. Layered foreignizing — B, D-let, partly A. Preserve German place-name suffixes (ß spelling); keep period-French and German vocabulary (tabouret, peignoir, Privatdozent); keep song titles in German italic; preserve the Schäm-dich chiasmus; preserve the Du-respect capitalized “You” in love-letter (B and D-let). B preserves the literal Berlin idiom “mutton-legs”; D-let uses archaic “Thee” in the Schumann line.

2. Middle / hybrid — C, A, C-let. Most kept-German vocabulary intact but lighter on dialect-literalism; A and C-let keep Frischer Hammel German while C calques. C-let domesticates dialect-idiom and chiasmus, foreignizes song titles via italics — an articulated middle position.

3. Strongly domesticating — D-aim, H. Both anglicize Privatdozent to “lecturer” (sole convergence on this); D-aim also anglicizes Fräulein/Herr (H does not); H invents “Young Ram” for the pub (D-aim keeps German); H fuses paragraphs and invents Schumann rhyme (D-aim does neither). The two arms share direction but diverge sharply on specific features — operating in different operational regions of “domestication.”

What the personas say, and how that shows up

The three personas (A, B, D) differ markedly in how explicit they are about translation craft principles versus literary stance.

A’s persona is the most operationally specific. It was built from Sophie Duvernoy’s writing about Tergit — a working translator’s craft commitments. The persona lists concrete defaults:

“foreign society-words naturalized — a coupé becomes a carriage, a cul becomes the rear … honorifics simplified — the Prussian Rat titles all to ‘Councilor,’ ‘Frau Kommerzienrat’ to ‘Mrs. Oppner’ … Berlinisch with a light hand — period Anglo slang and dropped consonants, not mapped onto Cockney or Brooklyn”

A’s translation applies most of these: Coupé → “carriage,” Taburett → “stool,” Stadtrat → “the Councilor,” Berlin dialect without eye-dialect or location-marking. But A’s translation does NOT apply the rule to Fräulein/Herr — it explicitly flags this in pass-1 (“Revisit: does the established Duvernoy voice anglicize to ‘Mr./Mrs./Miss’?”) and chose against. The inherited principles were applied, but case-by-case, not categorically.

B’s persona is the most literary/stance-oriented. Long, prose- poetic, full of biography, politics, and history. Operating principles are stated as literary attitude rather than as craft-rules: “the chorus,” “the object,” “the terse line,” “the flat death-sentence,” “laughter in the misery of going-under,” “it is not written from the end.” The pass log cites these stances directly: “the law of the book: it is not written from the end” / “I will not turn Berlinisch into Cockney or Franconian into a Scots — that is the translator’s vanity.” B’s translation preserves more German vocabulary than A’s (tabouret, capitalized “You,” literal “mutton-legs,” capitalized “Federalist” as period- specific noun, Sofie’s love-letter as a Du letter). B’s pass-2/3/4 work is mostly fidelity-correction (vergessen=forget; Recht=law; Gotthardtunnel-as-subject) — sense-fixes, not style-shifts.

B’s after-novel persona revision adds a single paragraph: “the true hero of the book is no person but the Time.” This is literary-stance not craft. It does not change any specific translation choice in a visible way.

D’s persona is operationally specific too — and explicitly cites its anglophone reference writers. The persona names four English voice models by scene assignment: Isherwood-clear for street- vignettes, Wharton-periodic for Tiergarten interiors, Mitford-quick for dialogue, Powell-deadpan for mercantile/modern. The persona also gives an explicit lexicon rule: “I will keep Geheimrat, Kommerzienrat, Maultaschen, Stollen, Verlobung, where they carry irreplaceable cultural weight; I will translate them where they can land cleanly.” And: “I will let the Hauptverben land at the end of the clause where English wants them; I will not preserve the German word-order.”

The after-novel addendum (+419 words, the largest revision in the experiment) adds four points: (1) the refrain is the chapter’s formal signature and must work four times; (2) “The keyword is Süße. Sweetness. Use the word”; (3) letters are the connective tissue; (4) “Lightness is the law.” It also names Wharton as the closest model for Ch.25 specifically, and explicitly states the no-eye-dialect rule for the Berlin scenes. Both D-let and D-aim applied this rule — neither uses eye-dialect, even D-aim, the strongest-nudged arm.

The interesting test: D-let and D-aim share this persona, yet diverge on roughly 15 lexical items (honorifics, academic ranks, song titles, social-historical terms, Berliner Osten). The persona provided common defaults; the brief overrode them for these specific items in D-aim. D-aim’s pass log makes this explicit at the moment of override (“rejected: keeping the German loanword in italics; chosen: anglicising fully … footnotes are forbidden by the domesticating principle”). D-let’s pass log does not perform this override — it cites the persona’s “no English equivalent” rule and keeps the German.

Comparing the three personas as readings of the author. - A’s persona: a domesticating professional translator’s mid- Atlantic Wharton-Powell-Mitford settled-voice, with Berlin grounding via place-name preservation and light-hand dialect. - B’s persona: a Berliner Vertriebene chronicler’s voice — the laconic, terse, anti-pathetic stance of New Objectivity, with the silent-suffering moral weight of Wolff’s Berliner Tageblatt. - D’s persona: an aged London-Berlin chronicler reading her own prose through four anglophone novelist-companions, with explicit craft rules about what to keep and what to translate.

Each persona is internally coherent and historically plausible. Each produces a recognizable English voice. None of the three predicts the others — A’s principles (honorifics simplified) overlap somewhat with what D’s persona names (Wharton-periodic), but B’s “laughter in the misery of going-under” is its own stance. The personas converge on stance (Tergit’s anti-pathos, flat-line-deaths, dialogue-driven scenes) but diverge on the operational layer (how to handle dialect, how foreign to keep the lexicon, whether to anglicize titles).

Where the human sits

H is at its own point in the geometry. Most distinctive features:

1. Paragraph fusion at refrain transitions. H folds the refrains into the preceding paragraph at lines 73, 175, 225, 283 — only the first two and the final are independent paragraphs. NO AI arm does this. The visual rhythm of H’s chapter is denser and more prose-block than any AI’s.

2. English rhyme invented for Schumann lyric (“blind/mind”). NO AI arm does this. The German source has Sofie’s misquotation that breaks Chamisso’s original rhyme; H invents a new English rhyme that creates the song-feel.

3. Heavy eye-dialect on Berlin scenes (“d’you haff to,” “Yer old man,” “Lettit go,” “Can’tcha,” “leggo,” “piss away your week’s pay”). NO AI arm does this. Every AI persona/pass log explicitly rejects eye-dialect (B: “translator’s vanity”; D’s persona: “plain working-class English with no eye-dialect”; A’s persona: “not mapped onto Cockney or Brooklyn”). H goes the other way.

4. Refrain restructured (“How sweet the air was at ten o’clock”). H adds an English subject and verb where the German is elliptical, and adds the adjective “beautiful” to Frühlingstag. NO AI arm does this.

5. Naturalized address-form for Amalie’s interior (“Dear, innocent Father!”). H translates Papa to Father here, where all six AI arms keep “Papa.”

6. Pub name invented in English (“Young Ram”). No other arm.

7. First-person interior monologue for Susanna’s FIS (“to be honest about my desires”). All six AI arms keep the German dürfen/permission frame.

8. “Sop” for Dämel — H’s most regionally-marked English word. All six AI arms used more standard period-British words (chump/ninny/great fool/idiot).

9. Modernized opaque period reference (“Matches! Matches!” drops “wax” from Wachsstreichhölzer). H is the only arm to drop the period specificity.

10. Selective foreignizing in unexpected places. “Casino toilette” for Kasinotoilette — H is the MOST literal of all seven on this period-fashion word. A says “dinner gown,” B and C “casino gown,” D-aim “evening gown.” H’s literalism here is striking: a translator who fuses paragraphs and invents pub names also leaves an obscure period-fashion term untranslated.

H is also the shortest translation (≈4,285 w). The combination of compression (paragraph fusion, dropped articles, sometimes-tighter dialogue) and modest expansion in the refrain (“How sweet the air was” vs “What sweetness”) balances out.

Does A resemble H most among the AI arms? Some — but selectively. A and H share: keep Fräulein/Herr; light-hand Berlin dialect (no eye-dialect); period-French vocabulary moderated. A and H diverge: A keeps Frischer Hammel, Privatdozent, Sofie’s “Frauenliebe und -leben” → translated to “A Woman’s Love and Life” while H keeps German. A does NOT fuse paragraphs, does NOT invent Schumann rhyme, does NOT use eye-dialect, does NOT invent “Young Ram.” A’s persona was built from H’s writing-about-Tergit, but A’s specific picks differ from H’s published practice at most loci.

The reason is structurally informative. H’s writing-about-Tergit gave A general principles (light hand; mid-Atlantic register; Wharton/Powell/Mitford as peer-group; “lightest possible hand”). But H’s published practice contains many craft-pragmatic picks — “Young Ram” specifically, paragraph fusion specifically, eye- dialect specifically, “blind/mind” rhyme specifically — that are not derivable from the general principles. A’s persona inherited the orientation but not the case-by-case picks. The principles A inherited are real and operative (the persona is doing work — A is shorter, lighter, more naturalized than B), but they don’t get A to H’s specific destinations.

Does C-let resemble H more than C does? On a few specific features yes: - The “stupid cow” cluster + abandonment of mutton-legs idiom (C-let matches H exactly on noun-substitute) - Drops “of the year” from refrain (matches H) - Idiomatic “of being ashamed” (closer to H than the chiasmus C preserves) - Keeps Schumann title in German (matches H — coincidence of cosmopolitan flagging)

On other features C-let does NOT resemble H: - Keeps Frischer Hammel German (H translates) - Keeps Fräulein/Herr (matches H — both keep, so neither moves) - Cuts Mayer’s “Such is life!” sentence (H keeps; this is a C-let move H does not make)

So C-let resembles H more than C does on dialect and idiom, but not on song titles (where both keep) or on the structural cut (C-let alone).

Does D-aim resemble H more than D-let does? Yes on overall direction; specifically on: - Anglicized Privatdozent → “lecturer” (D-aim matches H; D-let kept German) - “Trembles” for ängstigt (D-aim matches H; D-let kept German) - Dropped “of the year” (D-aim matches H; D-let kept)

D-aim resembles H LESS than D-let does on: - Frauenliebe title (D-aim translates; H and D-let both keep German) - Feuerzauber (D-aim “Magic Fire Music”; H keeps German; D-let “Magic Fire”) - Berliner Osten (D-aim “East End”; H “east of Berlin”; D-let “east of Berlin”) - Fräulein/Herr (D-aim translates; H and D-let both keep) - Casino toilette (D-aim “evening gown”; H “casino toilette”; D-let “ball-gown”)

The verdict is split. D-aim moves closer to H on some features and farther from H on others. The school-instruction direction is real, but the picks within that direction don’t align with H’s picks. D-aim has its own internally-coherent domesticating profile — and it’s not H’s.

The deeper question — what does the D-aim ↔︎ H gap tell us?

The gap is informative. D-aim went heaviest on the dimensions a school-instruction can directly target: a categorical class of items (honorifics, academic ranks, social-historical lexicon, locale-naming). H went heaviest on dimensions that require case-by-case craft pragmatic invention: paragraph fusion at scene transitions, English-rhyme creation for a song lyric, believable English pub-name invention, register-specific eye-dialect, modernized opaque period words. These two sets of moves are quite different in kind.

A school instruction can shift defaults across a category (“anglicize honorifics”). It cannot generate the right pub name (“Young Ram”) because the right name depends on knowledge of English pub-naming conventions and a feel for what sounds plausible. It cannot decide to fuse paragraphs at refrain transitions because that requires reading the chapter’s visual rhythm and judging that the refrain-as-standalone-paragraph reads as too telegraphic in English prose. It cannot invent “blind/mind” because that requires hearing what sounds like an English song-line.

H’s translation contains many such picks. Each is locally-grounded and not derivable from any principle. They are translator- pragmatic decisions: case-by-case craft calls that draw on specific knowledge the brief doesn’t carry.

The school instruction format moved D-aim in the right general direction. What it didn’t and couldn’t do is supply the specific moves. This is a structural limit of nudge-based translation: direction is shiftable; picks are not.

Layer 3 — Patterns and the region-vs-point hypothesis

Patterns that name themselves

Reading the seven freely, several recurrent patterns emerge that cross the categorical lines above.

Italics as cosmopolitan flag. C-let and D-let invented this pattern independently: preserve a German cultural-recognized item in its original spelling, but italicize it — so the German word reads as a recognized foreign work rather than as a foreign friction. Frauenliebe und -leben, Feuerzauber, Frischer Hammel, Die Lotosblume ängstigt. The italic is doing register work: this is a thing English speakers know in its German form, not something we have to translate. A keeps the German without italics (treating it as part of the world); B and C translate or calque. D-aim partially abandons the pattern (translates Frauenliebe and Lotosblume; keeps Frischer Hammel). H uses some German titles italicized (Frauenliebe und Leben, Feuerzauber) but translates the pub name. The italics-as-cosmopolitan-flag is a craft device the permission unlocked.

Voice-model layering by scene. Both D arms cite Wharton/ Mitford/Powell/Isherwood by name and scene-assignment in their pass logs (“Wharton-periodic in the Tiergarten salon and the Hiller lunch, Mitford-quick in the conversation, Powell-deadpan where the prose turns mercantile”). The persona’s voice-model mapping holds across the two forks. A’s pass log cites “Wharton/Powell/Mitford register” as one voice; B’s cites Fontane and the Weltbühne chorus. None of the no-persona arms (C, C-let) explicitly cite literary voice-models; they reason from register-by-character.

Independent fidelity-correction cluster. Several arms found and fixed the same meaning errors independently. Eugenie’s selber lieben misread as “loves oneself” was fixed by C-let (pass 2), D-let (pass 2), and D-aim (pass 4) — three independent rediscoveries of the same model-tendency error. vergessen (= forget) misread as “forgive” was fixed by B (pass 2) and C-let (pass 4). These corrections are visible in each pass log as substantive meaning-fixes (“This is the most consequential change of the pass”). The model has shared tendencies on certain German constructions; the arms each detected them on re-reading.

Structural conservatism by all AI arms. No AI arm fuses paragraphs at refrains. No AI arm restructures the refrain’s “What sweetness” into “How sweet the air was.” No AI arm invents English rhyme for the Schumann lyric. No AI arm uses eye-dialect on Berlin scenes. No AI arm invents “Young Ram” for the pub. All five structural/poetic-craft moves belong to H alone. The AI arms — all of them, including the strongest-nudged D-aim — operate at the clause and word level. They preserve the source’s paragraph structure, sentence-form, and rhetorical shape. H operates also at the structural level, and the differences are read off the page.

Different axes of domestication. Each of the four “domesticating- end” arms (C-let, D-let, D-aim, H) operates on a different axis:

• C-let domesticates structure (cuts a sentence!), idiom (“what-for”), chiastic-line shape (“of being ashamed”). Foreignizes via italics on song titles.
• D-let domesticates syntax/rhythm (English word-order, no German front-loading), dialect (no eye-dialect). Foreignizes on lexicon (kept Kaiserreich, Stadtrat, etc.) and on the Schumann lyric (archaic Thee).
• D-aim domesticates lexicon and honorifics (Fräulein → Miss, Privatdozent → lecturer), locale-naming (East End), song titles (mostly). Stays loyal to source paragraph structure.
• H domesticates structure (paragraph fusion), rhythm (drops articles, fuses sentences), embedded-song-form (invents rhyme), dialect (eye-dialect), period reference (modernizes). Foreignizes selectively on lexicon (keeps Fräulein, Herr, casino toilette, Frauenliebe).

These are four different things being called “domestication.” The arms are not arrayed along a single axis. They are at different points in a multi-axis space.

Testing the region-vs-point hypothesis

The hypothesis: domesticated Tergit is a region in choice-space, not a single point. Multiple domesticating arms (C-let, D-let, D-aim, H) sit toward the domesticating end of various axes but at distinct locations.

The evidence is strongly for the region hypothesis.

First, on every locus where the domesticating-end arms agree on direction, they disagree on destination. For Hammelbeine langziehen, the four “abandon-the-mutton-legs” arms produce four different English idioms: “what-for” (C-let), “hiding” (D-let), “thrashing” (D-aim), “good slap” (H). The direction is shared; the specific English pick is independent.

Second, the same persona produces different points depending on the brief. D-let and D-aim share the persona including its explicit “keep where it carries weight” rule. The persona produces two different translations because the brief alters what counts as “weight” worth keeping. The same persona is consistent with at least two distinct points in the choice-space. The persona constrains; it does not determine.

Third, the strongest-nudged AI arm (D-aim) does not converge on H’s picks. The point hypothesis predicts that stronger nudges should move the AI toward H’s specific picks. The evidence is mixed — D-aim moves toward H on Privatdozent and “trembles” and the refrain compression, but moves AWAY from H on Fräulein/Herr (D-aim translates where H keeps), Frischer Hammel (D-aim keeps where H translates), Berliner Osten (D-aim “East End” where H “east of Berlin”), Frauenliebe (D-aim English where H German), and the structural/poetic moves H makes (paragraph fusion, rhyme). Stronger nudge does not equal “closer to H.”

Fourth, H is itself at a particular point with distinctive features. H keeps casino toilette (the most literal of the seven on this word) while fusing paragraphs and inventing a pub name. H goes furthest on dialect eye-dialect while keeping all Fräulein and Herr titles. H’s domestication profile is anti-categorical: case by case, with specific knowledge of which features warrant naturalization and which warrant preservation. The profile is coherent within itself (H’s craft has a logic) but not reachable by direction-nudge (you cannot get there by telling an AI to domesticate harder).

Fifth, the AI-arms-that-cluster-with-H on direction do not match H’s particular picks. The two arms most like H on direction are D-aim and (less so) C-let. Neither matches H pick-by-pick. C-let and H both keep Schumann’s German title — but the rest of C-let’s profile (keeping Frischer Hammel, the structural cut, the chiastic naturalization) does not align with H’s profile (fusing paragraphs, inventing rhyme, using eye-dialect, keeping Fräulein).

The region has structure. It is bounded — no arm goes infinitely far in the domesticating direction; each arm’s direction is bounded by what its persona, brief, or close-reading of the source recommends preserving. And it is multi-dimensional — arms can be “domesticating on dialect but not on lexicon,” or “domesticating on lexicon but not on structure,” or any other combination.

The point hypothesis is supported only on a few features (refrain compression; “trembles” verb-choice). On those, multiple domesticating arms converge with H. But on most features, the domesticating arms each pick differently from H, and from each other. The region hypothesis is the better fit for the seven-way geometry.

What the gap between D-aim and H tells us

This is, ultimately, a finding about what categorical instructions can and can’t do for translation. The school instruction reliably shifted D-aim’s defaults across whole categories of lexical substitution (honorifics, academic ranks, social-historical lexicon, song titles). What it did not and could not do is invent the specific English picks H made.

H’s most distinctive moves — “Young Ram” for the pub, “blind/mind” for the Schumann rhyme, “Can’tcha see” for eye-dialect, paragraph fusion at refrains, “the air was” for the refrain’s elliptical form, “casino toilette” preserved as period-fashion vocabulary, “sop” for Dämel, “my desires” for Susanna’s FIS — are case-by- case craft pragmatics. Each pick draws on a specific kind of target-language knowledge (English pub-naming inventories, English-song idiom, English working-class register conventions, English prose-paragraph rhythm, English period-fashion vocabulary, English casual-fool diction, English first-person sensual-discourse). None is derivable from “domesticate” as a categorical instruction.

A categorical instruction can move an AI across the macroregional axis. It cannot specify the local picks within the region. The local picks require either case-by-case human craft, or a much more granular and operationally specific kind of instruction than “domesticate in the Schleiermacher sense.”

The experimental finding: even the strongest plausible domesticating nudge (school instruction at the translation step) moves the AI to its own coherent domesticating point — different from the un-nudged AI’s point, different from the permission-armed AI’s point, and different from H’s point. The domesticating direction is a real and shiftable orientation. The destination is craft.

Closing note on the experimental geometry

Six AI arms differ on two axes (persona-source: none / inherited / self-built-German / self-built-expanded; nudge: none / system- prompt-permission / step-4-kickoff-school) and produce six distinct translations of the same chapter. The published human translation H is a seventh. No two of the seven are at the same point. They cluster directionally but not by destination.

The clearest single finding: the persona controls the defaults, the brief controls the deviations, and H’s published practice contains many specific craft picks that neither persona nor brief can produce. The personas A, B, D each produce internally coherent voices that recognizably read Tergit; the brief-nudges shift specific features without changing the voice; the published human translation makes structural and rhythmic moves no AI arm makes, while preserving some foreign vocabulary even the most-foreignizing AI arms translate.

This is a description of the experimental space, not a ranking. Six of these renderings are made under controlled conditions for the purpose of the comparison; one is a published human translation read in its own light. None of the seven is bad; each is a coherent and intelligent reading of the German source. They differ, and the differences are where translation craft lives.