The Persona, the Brief, and the Human Hand

Stance. The seven renderings are treated as co-equal columns. This is description and synthesis — not a ranking. Which rendering is the better or more faithful literary translation is a judgment reserved for human reviewers and is not made here. Every chart is built from the study's verified quantitative layer; every quoted rendering is from the finished translations.

1The design & the two probes

Chapter 25 of Effingers is a Berlin Saturday — 16 March 1887 — narrated hour by hour and stitched together by a recurrent, time-stamped spring-refrain:

Was für ein Frühlingstag, dieser Sonnabend im März des Jahres 1887! Was für eine Süße, [hour]!

The chapter cross-cuts the whole society — a fitting at the wealthy Eugenie's; the schoolgirl Sofie's love-letter; the Privatdozent Waldemar arguing Roman law and then bedding the singer Susanna; the working-class Effingers and the ruined banker Mayer; a street tableau; Theodor's despair in a wine-tavern. Its load-bearing features make translation choices visible: the incantatory refrain and its variants, montage paragraphing, Berlin working-class dialect, embedded lyric (Schumann, Heine), period honorifics and academic ranks, costume and society vocabulary, and free indirect speech.

Six AI arms render the chapter, varying on two axes; the human translation is folded in as a seventh, co-equal column. Two of the pairings isolate a single varied input and are the cleanest probes in the study.

Limitations, up front. n = 1 chapter, one instance per AI arm (a case study, not a statistically powered result). The six AI arms share a common base model — a plausible driver of their mutual closeness; the agreement figures measure how alike the outputs are, not independent convergence. This is an informed, not blind, comparison (arm identities known). All numbers below are from the study's independently verified quantitative layer.

2The composite gradient — Chart 1

Scoring each rendering on a 0–100 foreignization↔domestication composite — the coverage-weighted mean of six tests, each measure weighted by the number of loci it covers (proper nouns, loanwords/titles, dialect, refrain/structure, idioms, honorifics) — places the seven on a single line. H stands far out at the domesticating end; the six AI arms sit in a narrow band, roughly a third of the way along.

3The 7×7 agreement heatmap — Chart 2

Two arms "agree" at a locus when they make the same salient pick (out of 45). The matrix is symmetric; darker green = more agreement. The six AI arms form a dark green block; the H row and column are a pale band — H sits apart from all of them.

4The 2-D cluster scatter — Chart 3

A second axis separates lexical domestication from editorial boldness — how far a rendering departs from the German's form (reshaping the refrain, fusing paragraphs, inventing rhyme, eye-dialect, dropping locale), scored independently of vocabulary. The two axes are correlated for H but separable for the AI: the school pushes D-aim rightward without lifting it upward.

5Region, not point — Chart 4

Is there a single "domesticated Tergit" target toward which stronger nudges converge? The data say no: where the domesticating arms abandon a vivid German idiom, they reliably share the direction but scatter to different destinations. The clearest case is the carter's Berlin tirade "die Hammelbeine langziehen" — four domesticating arms, four distinct English landings.

The verified layer quantifies this across the region loci — those where all four domesticating-end arms move off the literal:

The two loci the brief pre-flagged — Hammelbeine and Bärenhunger — are maximally dispersed (four movers, four destinations each). The honest counter-example is doofe Ziege, where the four converge on one image ("cow") and differ only in the modifier — a near-point. The conditional pattern is clean: when the domesticating arms abandon a vivid German idiom they scatter; when they abandon a flat insult they converge.

6Where the human sits — Chart 5

Domestication is not one dial. Comparing each arm's domesticating-mean to H's on six sub-axes (MORE / at / less than H) separates the dials the composite nets together. The single red cell tells the story: the strongest-nudged arm overshoots H on exactly one narrow dial.

H's structural and craft signature

What defines H is not more of what the AI arms do but a different kind of move. Across the 45 loci the "six AI agree, H alone differs" signature is the largest single partition pattern (14 loci); H is a singleton at 23. Five moves belong to H alone and to no AI arm — not even D-aim:

To these the layer adds first-person interior monologue for Susanna's free indirect speech ("to be honest about my desires" — all six AI arms keep the German permission-frame), "Father" for Papa in Amalie's interior, "sop" for Dämel (the most regionally marked word in the field), and a modernized opaque period reference ("Matches!" dropping the wax of Wachsstreichhölzer).

And yet H foreignizes selectively — the anti-categorical signature. The same translator who fuses paragraphs and invents a pub name keeps the French "casino toilette" (the most literal of all seven on this period word), keeps Fräulein and Herr (which D-aim anglicizes), and keeps Frauenliebe und Leben and Feuerzauber in German. H domesticates the references an English reader cannot catch and preserves the cultural artifacts an English reader is likelier to recognize. Its profile is coherent within itself but not reachable by direction-nudge: D-aim, told to domesticate harder, landed on a different coherent profile — anglicizing the honorifics H keeps, keeping the pub name H invents.

7How the arms reasoned

The six AI arms left pass logs; H did not. An inductive typology of what each arm cites as its authority turns the output differences into a picture of why the arms diverged. The cleanest result sits inside Probe 2, where only the brief's modality varies.

The modality asymmetry — a permission is quiet, a school is loud

D-let's permission is background: present in its prompt, almost never named, exercised silently while each move is justified by persona or period-usage. D-aim's school is foreground: cited at nearly every domesticating choice and reasserted in every pass header — and it overrides the very English-survival facts D-let uses to keep the German. On Taburett, both logs state the same fact ("tabouret survives in English period vocabulary") and reach opposite verdicts. Same content in the two briefs; opposite citation-behaviour by modality.

Two patterns that map onto the geometry

• Voice-models are corpus-gated. The named anglophone authorities (Wharton, Mitford, Powell, Isherwood) appear only where the corpus supplied them — A and the D pair. B cannot cite them: its German-only corpus contains zero mentions. This maps onto the gradient, where B's German-only persona sits near the foreignizing floor (19.7, second-lowest of the six). A cites craft defaults where B cites literary stance.
• Independent fidelity-corrections reveal shared-base-model tendencies. Three or four arms caught and fixed the same meaning-errors on re-reading (e.g. selber lieben misread as "loves oneself," fixed independently by C-let, D-let, and D-aim). That the arms rediscover the same error separately is the clearest in-log sign of the shared base model the agreement figures must be read against.

8Discussion

The seven-way geometry and the two probes converge on one clean law:

The persona controls the defaults. The composite ordering of the AI arms tracks the persona-source axis more than anything else within the AI band: the two German-leaning persona arms on the lighter brief (D-let 17.1, B 19.7) sit lowest; the persona dampens a permission (D-let kept more German than C-let under the identical paragraph); and the corpus that built a persona gates which authorities it can cite. The persona does real work — but on the baseline disposition, not on individual destinations.

The brief controls the deviations. Holding the persona constant, the school moved D-aim's composite by +29.0 and the permission moved C-let's by +1.3, both in the predicted direction, both on the features the nudge could target — and the format set the reach: a quiet permission is a tie-breaker a strong persona can absorb, while a loud school re-anchors the operating principle and overrides persona defaults on whole bands of vocabulary. The deviation is real and measurable, but it is a deviation from a persona-set baseline, not a relocation to a fixed point.

The human's craft picks are produced by neither. H's defining moves — restructuring the refrain, fusing paragraphs, inventing the Schumann rhyme, writing eye-dialect, inventing "Young Ram," breaking into first-person free indirect speech — are made by no AI arm, including the strongest-nudged. They are case-by-case craft pragmatics, each drawing on a specific kind of target-language knowledge no categorical direction carries. And H's foreignizing exceptions are equally case-by-case.

The practical upshot is a statement of what categorical instructions can and cannot do for translation. A categorical instruction can move an AI across the macro-regional axis — it reliably shifts defaults across a whole category of lexical substitution. It cannot specify the local picks within the region — the right pub name, the decision to fuse paragraphs, the rhyme that makes a misquoted lyric sound sung. The domesticating direction is a real and shiftable orientation. The destination is craft.

This is a description of an experimental space, not a verdict on it. Six of the renderings were made under controlled conditions for the purpose of comparison; one is a published human translation read in its own light. None of the seven is bad; each is a coherent reading of the German source. They differ — and the differences are where translation craft lives. Limitations: n = 1 chapter, one instance per AI arm; the six AI arms share a common base model (a live confound for every similarity among them); this is an informed, not blind, analysis. The companion paper (paper.md), the verified quantitative layer (quant/quant_layer.md), and the reasoning-typology exhibit (authority/authority.md) carry the underlying detail; a separate persona-readings study compares the arms' readings of Tergit.