After Babel, The Machine?

The library

What each instance read — the complete reading list, arm by arm

The personas in this experiment were never given style rules. They were given reading. This page names every text each instance was given to read before it translated, so that the source of each translator’s “voice” is fully transparent. The works themselves are under copyright and are not reproduced here — they are named, not copied. Where a work is in the public domain it is freely findable elsewhere; where it is not, this is a citation, not a reprint.

Six of the seven arms are AI; the seventh, H, is the published human translation. The arms differ along a persona-source axis. Two of them — the controls C and C-let — read nothing but the novel itself. Three — A, B, and D — each read a corpus first and built a translator-persona from it. Below is a side-by-side map of who read what, followed by the full annotated lists.

The map — who read what

Reading group A B C C-let D-let / D-aim
Tergit’s own works (5) ● ● — — ●
Scholarship & reference on Tergit — ● — — ●
German peers & contemporaries — ● — — ●
German literary tradition — ● — — ●
Period & cultural context — ● — — ●
Anglophone reference novelists — — — — ●
Inherited frame — Sophie Duvernoy’s writing ● — — — —
Effingers — the full novel (Step 3) ● ● ● ● ●

A inherited its interpretive frame from an established human translator’s writing on Tergit and did not perform the self-built German research. B built its own frame from the German corpus, walled off from any writing by that translator. D read everything B read plus a set of anglophone novelists — its one distinguishing addition; D-let and D-aim share a single identical corpus and reading, and diverge only at the translation step. C and C-let read the novel and nothing else — no corpus, no persona. Not shown in the table is H, the novel’s published translator (Sophie Duvernoy, New York Review Books, 2025), who brought a lifetime of reading no document can enumerate; no constructed corpus is listed for her.

1. Tergit’s own works — the primary corpus

Read in full by A, B, and D (Step 1) — five works, in roughly compositional order:

2. Scholarship & reference on Tergit

Read by B and D (Step 2):

3. German peers & contemporaries

Read by B and D — the writers around her:

4. German literary tradition

Read by B and D — the realist–bourgeois line Tergit writes within:

5. Period & cultural context

Read by B and D — encyclopedic briefings on the world of the novel: the German Empire (Kaiserreich) and the Gründerzeit; the Weimar Republic and the Golden Twenties; the inflation of 1914–1923; antisemitism and the history and emancipation of Germany’s Jews; the Kurfürstendamm and the Grunewald; the feuilleton and Die Weltbühne; New Objectivity (Neue Sachlichkeit); and exile literature.

6. Anglophone reference novelists

Read by D only — the expanded corpus, D’s one distinguishing addition:

7. The inherited frame — Sophie Duvernoy’s writing

Read by A only. A did not assemble its own research; it built its persona from an established human translator’s own writing about Tergit:

A inherited the interpretive frame in these pages, but the translator’s name was withheld from A itself — so the arm never “knew” it was tracking the same translator who supplies the human column, H.

8. The novel

Read in full at Step 3 by every AI arm — A, B, C, C-let, and D:

The controls C and C-let read this and nothing else. H is the novel’s published translator and produced the human column from it.

All source materials were converted once to plain text and held in a single archive, so that no instance ever loaded a binary file into its context. The reading was real: full-read protocol at every corpus and novel step, no sampling.