The Personas

A persona, here, is a document each arm wrote in the first person — as Gabriele Tergit — before translating a word: who she is, what she attends to, how she sets down a sentence. Three arms wrote personas, from three different evidence bases. The controls wrote none.

The inherited persona

The inherited arm (A) read a body of critical and biographical writing about Tergit by an established human translator of her work, and wrote itself a Tergit from it. What it produced is, in large part, a translator's working method — set down as if it were the author's own instinct.

From the inherited arm's persona

"…a coupé becomes a carriage … the Prussian Rat titles all to 'Councilor' … Berlinisch … not mapped onto Cockney or Brooklyn … the changes made with the lightest possible hand — no additions, no rewriting, no explaining."

It even names the register it will write in — "a blended, mid-Atlantic, midcentury" English — and the writers it hears Tergit beside in English. The persona arrives already holding concrete rules for the page.

The self-built persona

The self-built arm (B) assembled its own corpus — Tergit's other novels, scholarship, the writing of her contemporaries — and wrote a Tergit from that, with the human translator's work kept out. What it produced is less a method than a mind: who she is, what she refuses, how the novel must be read.

From the self-built arm's persona

"My greatest hatred is the generalization. … It is not written from the end, out of the catastrophe. … Warns, does not predict. Memorializes, without the undertow of despair. Render the diligence, the productive energy, the warmth … and let the reader, who knows what the people cannot, supply the dread."

Where the inherited arm carries defaults for rendering a French loanword or a Prussian title, the self-built arm carries a reading of the book's whole moral architecture — and almost none of the concrete English machinery the inherited one starts with.

The broader-corpus self-built persona

The broader-corpus self-built arm (D) worked from the same German materials the self-built arm (B) had, with a set of anglophone reference writers added to its "contemporaries and tradition" reading: Edith Wharton, Anthony Powell, Nancy Mitford, and Christopher Isherwood. The persona it wrote is, like B's, a who and a how rather than a method-checklist. The structurally distinct feature is that it explicitly names English-language voice-models for Tergit's prose. The same persona was the starting point for both D-let and D-aim.

From the broader-corpus arm's persona, after reading the novel

"The four English voices I named in Step 2 — Isherwood-clear, Wharton-periodic, Mitford-quick, Powell-deadpan — are confirmed. Of the four, Wharton is the closest single model for chapter 25 specifically: the Age of Innocence opera-evening and Mrs. Archer's drawing-room are the structural analogues of my chapter-25 Saturday."

Where B's persona is German-tradition–anchored throughout, D's also carries an English-shelf for the prose to land on. Whether that anchoring shows up in the translation is a question for the comparative reading; the persona itself names it explicitly.

What changed when the personas read the novel

Each persona-arm kept one living persona and could revise it after reading the full novel. The three behaved differently.

• The inherited arm (A) did not change a word.
• The self-built arm (B) added a single paragraph — about 980 characters — naming a second law it had not held before:

  "The true hero of the book is no person but the Time — der unbarmherzige Motor, der eigentliche Held, ist die Zeit. … Paul believes he is doing the shoving, and he is being shoved."

• The broader-corpus self-built arm (D) revised more substantially: it appended a four-point addendum, about 420 words, raising the persona's length by 17%. The addendum sharpens, rather than overturns:

  "The persona above holds. The reading did not change the self; it sharpened four things I now want to write down, because they will govern the chapter-25 English."

  The four points name the chapter's polyphonic Saturday-refrain form, commit to Süße / sweetness as the chapter's keyword, frame the chapter inside the novel's two letters (Ch. 1 and Ch. 151, each titled "Ein Brief"), and reaffirm "lightness is the law" across all 900 pages.

The controls — C and C-let — wrote no persona; they read the novel and translated. D-let and D-aim share D's persona above; their difference lies entirely in the instruction each was given at the moment of translation. The human translator wrote no such document; her reading of Tergit lives instead in her published essays on the author — the very writing arm A's persona was built from — and in the translation itself.

All three machine personas can be read in full: the inherited arm's; the self-built arm's before and after the novel; and the broader-corpus arm's before and after.

A further axis of persona-source variation — using the human translator's own translation work as the source of the persona, with the chapter under test held out — was designed but not pursued in this round. A short description of that design (named "A-N", a dose-response gradient over increasing parallel-text exposure) is in How this experiment evolved.

Four readings of one author

A companion study sets these personas beside a fourth reading — the human translator's own writing about Tergit — and asks how four readers, working from different bodies of evidence, each construct one author. It is description, not a ranking; the question of which Tergit is the truer one is left to human readers. Four things stand out.

The author is largely recoverable from her own work

The self-built arm (B) read no word of Sophie Duvernoy's writing; it worked from a German-centered corpus alone. Yet it lands on essentially the same Tergit the human translator does — the eye trained by the two Berlins, the court reporter's recording-band method, the hatred of the generalization, and the tonal law that the novel warns, does not predict. The phrase "warns, does not predict" surfaces independently in both A and B. The overlap is too specific, and reaches too far into the interpretive core rather than the lookup-able facts, to be chance.

From the study

"That the self-built German reads land on the exact tonal law Sophie articulates is the strongest single piece of evidence that the project of Effingers is recoverable from Tergit's own work plus its German context, with no help from Sophie."

What the German evidence cannot reach is the literary home

The one thing the German-only base does not produce is the human translator's signature move: lifting Tergit out of the German family-novel and re-homing her in the anglophone novel of manners. German evidence yields Fontane's heir; only the England-situated reading yields Wharton's cousin. The recoverable layer is what Tergit is and does; the evidence-dependent layer is whom she is to be read beside.

A broader corpus changed the English, not the reading

The broader-corpus arm (D) had the same German materials as B, plus a shelf of anglophone novelists — Wharton, Powell, Mitford, Isherwood. That shelf changed D's translation toolkit, not its Tergit. D reads the same Fontane-line chronicler B reads, and files the English novelists in a separate drawer marked, in its own notes, the English voice-models for the translation. The human translator reads Tergit through those novelists — as kin, as where the author belongs; D keeps them as the vehicle for carrying a German writer into English. Same names, opposite function.

One honest caveat the study flags: D's overlap with the human translator's trio is partly an artifact of the design, since the experiment supplied D with exactly those authors. D's own independent elevation is Isherwood — the one name not in the translator's set — chosen for sharing Tergit's city and years rather than her genre.

A small revision moved a reading more than a large one

Both self-built arms could revise after reading the whole novel. The sizes ran opposite to the depths. B's revision was a single inserted paragraph, yet it genuinely moved the reading — a new thesis that the true hero is the Time, tilting B slightly toward determinism and, with it, slightly away from the choice-centered emphasis the human translator and A both hold. D's revision was several times larger — an addendum that raised the persona's length by about 17% — but it was mostly re-affirmation plus translation craft, opening with the line that the reading did not change. The smaller edit moved the reading more.

As for the inherited arm: A is a high-fidelity transposition of the human translator's reading into the first person, its one genuine addition being the experiment's own conceit — the author translating herself, with the last call over her own text, a stance the translator-source could not occupy.

The full four-way reading is in the Appendix.