Ernst Klugmann Is Us: Translating Four Untranslated Latin Political Texts With LLMs

Under the name “Ernst Klugmann,” we are self-publishing German translations of four early-modern Latin political treatises that, as far as we can determine, have never been translated into German before. Two things about that sentence need to be stated plainly, right here, rather than saved for later: Ernst Klugmann is a pseudonym Westwood Works uses for this publication, not a real historian. And the translations were produced with LLM assistance, not by a human classicist working alone. Neither fact is a reveal. Both are just true, and this post is about how the project actually went.

This is not a product. It is a side project that grew out of a simple question: what happens if you point a modern LLM at a genre of text — dense, syntactically brutal, 400-year-old Latin — that sits well outside anything most people would consider a safe use case, and check the results carefully instead of taking them on faith?

Finding texts nobody had translated

The genre is Tacitism — early-modern political writing built on ratio status (“reason of state”) and the reading of Tacitus as a guide to how power actually works, as opposed to how it should. It is a real scholarly niche, and it already has two active German translation series covering it: the Bibliothek des deutschen Staatsdenkens and the Editionen zur Frühen Neuzeit. Before touching any text, we checked both series and searched the KVK and K10plus union catalogs — the standard way to confirm whether a German edition of a given Latin work already exists. A web search cannot prove a negative here; only a library catalog check comes close.

That process ruled out several strong candidates (Lipsius’s Politica, Althusius, Clapmar — all already translated) and left four texts that, as best we can establish, have not:

• Justus Lipsius, Monita et Exempla Politica (1605) — a mirror-for-princes built from ancient Roman and Greek exempla.
• Cardinal Mazarin (attrib.), Breviarium Politicorum (1684) — blunt, practical maxims on staying in power, later translated to French with a preface by Umberto Eco, but never to German.
• Kaspar Schoppe, Paedia politices (1623) and Gabriel Naudé, Bibliographia politica (1633) — a two-part primer on how to study politics at all, bound together in the 17th century by Hermann Conring.
• Henning Arnisaeus, De jure majestatis (1610) — a systematic treatise on the legal foundations of sovereignty.

The translation pipeline, and where it broke

Each text got translated twice: once as a literal, text-faithful rendering that preserves the syntax and aphoristic rhythm of the original, and once as an idiomatic modern German version aimed at a general reader. The two run side by side on facing pages so a reader can check one against the other.

The first pass worked from the OCR text layer of existing digital scans. That was a mistake we caught, not avoided. Historical type — long ſ, ligatures, inconsistent spacing — produces OCR output with real errors, and translating a corrupted source text just launders the corruption into fluent-sounding German. We rebuilt the pipeline to feed the LLM the actual page images instead, letting a vision-capable model read the historical typeface directly and produce its own transcription alongside the translation. That vision pass then served as a cross-check against the original OCR-based translations, not a replacement generated once and trusted.

What the verification actually showed

We are not going to claim these are scholarly, peer-reviewed translations, because they are not. No critical apparatus, no institutional review, no classicist checking every line against the manuscript tradition. What we did do: for the Schoppe text, we checked our German version against the 2022 Italian critical edition (A. Panichi) that prints the Latin facing an Italian translation — the closest thing to a scholarly reference any of these four texts has. For Mazarin, we ran a systematic comparison against the one existing modern translation, the published French edition, sentence by sentence.

That comparison found four real divergences in meaning across roughly 100 pages — three moderate, one minor, none changing the substance of any chapter. We then adjudicated each one against our own vision-based Latin transcription. In three of the four cases, our German version matched the Latin and the French reference translation was the one in error — a compass direction reversed, a rank mistranslated as a physical stature, a cash payment turned into a laurel wreath. That is a genuinely useful result, but it is a spot check against one comparison text, not a certification. We are not overselling it.

Why the pseudonym, and why say so

“Ernst Klugmann” exists because these translations aren’t a Westwood Works software product and we didn’t want them to read like one. But an invented byline that quietly obscures who actually made something is a different thing from a pen name that’s openly acknowledged, and we’d rather be the second kind. So: Ernst Klugmann is us, the translations involved LLMs at every stage, and both facts are here in the first paragraph, not the last.

Where this stands

The Mazarin volume is furthest along — we already have print-ready paperback and Kindle layouts assembled. The other three are behind it. We’re planning to sell physical copies at cost — a price that covers print production and nothing more, no markup, no ISBN or storefront to link to yet because neither exists at the time of writing. If you’re interested in a copy or in the project generally, reach us at hi@westwoodworks.de.

This is the kind of LLM use case we find genuinely interesting: not a chatbot bolted onto a workflow, but a task with a real, checkable source text, a way to verify the output against reference material, and an honest accounting of what the verification did and didn’t cover.