MIGNE.APP

Editorial method

How these translations are made

What the rule is, what it forbids, and how to check us against the Latin.

What Migne's text is

The Latin and Greek on these pages is Migne's printing, and Migne was a publisher, not a critical editor. The Patrologia was by design a reprint: its texts came from the great early-modern collections — the line that runs from la Bigne's Bibliotheca of 1575 through the Maxima of Lyon (1677) and Galland — and above all from the Maurist editions of the seventeenth and eighteenth centuries, the Benedictine scholarship of Saint-Germain-des-Prés, reproduced entire. Where those failed, the shop took whatever the nineteenth century had printed. Migne's workshop reset the type at speed — five stages of proof and an error bounty against the new mistakes, but no fresh collation of manuscripts behind them. What you are reading, then, is the received text of the fathers as the age of print transmitted it: the manuscripts as the Maurists and their predecessors edited them, as Migne's compositors reset them, errors of the lineage included.

Two practical consequences. First, where a modern critical edition exists (Vienna's CSEL, Berlin's GCS, Turnhout's CCSL, Sources Chrétiennes), it supersedes Migne for scholarly citation of the text itself — what this site gives you is Migne, complete and addressable, which no critical corpus begun since has managed to be. Second, column numbers here follow the original Migne plates; the Garnier reprints made after the 1868 fire were reset with different column breaks, so citations against a Garnier printing may drift.

A note on the word Englished

This site says that a work has been Englished, lists its authors as the Englished, and tells you of a work it has not reached that it is not yet Englished. On a page that also discloses that the drafting is done by machine, the word can look like something a machine coined. It is a good deal older than that. To English is a medieval English verb — the Middle English Dictionary gives englishen, "to put into English: to paraphrase or translate a foreign text," with citations from the fifteenth century — and it stayed in ordinary use for the five hundred years after. Ronald Knox, who translated the whole Vulgate single-handed at the commission of the bishops of England and Wales (the New Testament published in 1945, the Old in 1949), called his account of that work On Englishing the Bible (London: Burns, Oates & Washbourne, 1949).

It is used here for the reason it has always been used: it names the act without claiming the honours of authorship. A text that has been Englished has been carried over into the language and can now be read by someone who has no Latin. It has not thereby been given its definitive English version, and this edition does not pretend to have made one.

How the translation is made

The English on this site is drafted by large language models working under a fixed rulebook, and then verified — partly by machine, against the Latin, and partly by a human reading the result. We say so plainly here because the alternative is worse: a reader who works it out for himself has every reason to wonder what else was not said, and a method is easier to defend when it is described than when it is discovered.

The objection this raises is the right one to raise. A machine asked for a translation will produce fluent English about the right subject whether or not it has understood the sentence in front of it, and the fluency is exactly what makes the failure hard to see. Everything below exists to close that gap. The short version: the model is never asked to translate a work. It is asked to render a numbered chunk of Latin under a written rule that forbids nearly every liberty a translator might want to take, and the result is then checked against the Latin by a program that does not care how good the English sounds.

Three things do the work.

A frozen rulebook. Every translation is made against a register rulebook that is versioned in the project's repository and grows only by explicit ruling. It fixes the register — literal, never paraphrase; every clause rendered, nothing summarized or compressed — and it fixes what happens at each recurring construction that has caused trouble before, with a worked example named for each. When a reading of a finished work turns up a bad habit, what gets written down is not a fix for that sentence but a rule for the class of sentence, and the existing English is then swept for other members of the class. There are a million quoniams in eighty-five million words; the sentence is not worth fixing, the pattern is.

A style anchor. Alongside the rulebook, every draft is made with a fixed pair of files in view: one chunk of Latin and its approved English. The conventions that are easier to show than to state — how a heading is set, how a marker is carried, how dense a sentence is allowed to get — travel by that example rather than by description.

Chunks, not works. Texts are cut into numbered chunks at safe boundaries and translated in those units, with the column anchors of the printed page preserved inside them. This is what makes the checking below possible: the Latin and the English are partitioned identically, so any segment of English can be laid against the exact Latin it renders rather than against a general impression of the work.

What happens when the page is defective

This is where a translation is most tempted to improve on its original, and it is the rule this project is strictest about. The English renders what Migne's plate prints, not what it ought to have printed. That holds when the printed reading is odd, when it is ungrammatical, and when it says the opposite of what the argument plainly requires.

The reason is not pedantry. Migne's corruptions are frequently the most interesting thing on the page, and every silent correction destroys evidence a later scholar cannot recover from our English. A dropped negative that makes Richard of St Victor assert the contrary of his own argument — there are two such places in the works translated so far, at PL 196 col. 1054A and col. 1069D — is exactly the kind of thing an edition exists to expose. Emend it and you have produced smooth prose and a worthless witness.

The rule is not this project's invention, and it is not an artifact of translating by machine. Ronald Knox met exactly this case while Englishing the Vulgate. At Acts xvii. 6 the Clementine text reads urbem, the city, because a copyist had written it for orbem, the world; Knox knew which reading was right and said so in print — and then translated the corrupt one anyway, "who turn the state upside down," on the ground that "it is no part of the translator's business to alter, on however good grounds, his original." The true reading went to the foot of the page and the text kept what the page printed. That is the division of labour here, with the cruces doing the work Knox's footnotes did.

In practice:

Two classes that look like corruption and are not, and are therefore left alone: spellings attested in medieval Latin, which are correct Latin for their period rather than mistakes, and Latin grammatical gender, which has no English exponent at all — there is nothing there to get right or wrong.

Why the scripture is translated from the page

Scripture quotations in these works are translated from the Latin as printed — not replaced with any conventional English version. This is intentional. The fathers and their medieval readers quote the Vulgate loosely, from memory, from Old Latin versions older than Jerome, or with the wording deliberately bent to the argument — and then the argument rides on the wording. Abbo of Fleury's Joel runs eruca, locusta, bruchus — caterpillar, locust, grub — where the King James's creatures come in another order entirely; a translator who pastes in the familiar verse has quietly corrected the author's Bible out from under his sermon. Where the quoted Latin does match the standard text, our rendering is free to echo the traditional English ("Thou art Peter, and upon this rock I will build my Church"); where it diverges, the divergence is preserved, because the divergence is evidence — of the version the author read, the memory he quoted from, the point he was making. The parenthetical references in small type stand on Migne's page — editorial identifications, inherited from the edition he reprinted or supplied by his own shop, and reproduced here exactly as printed; they will take you to the verse in whatever Bible you trust. The English on this page tells you what the author actually wrote.

Quotation marks, and telling us from Migne

The two kinds of quotation mark on these pages mean different things, and the difference is provenance rather than taste. « » reproduces a mark that Migne prints, placed exactly where he places it. " " is a mark we have supplied, because English needs one and the plate marks the quotation some other way, or not at all.

The consequence is that a reader can tell the edition from the plate by eye, with no apparatus to consult. It also makes the rule checkable by machine: the guillemets in the English must match the Latin's one for one, while the straight quotes are free. One corollary is worth stating because it looks like an error — where Migne prints an attribution inside the quotation, « I am, he says, the way and the truth », the English keeps it inside rather than breaking the quotation around the speech tag as English convention would. And where Migne opens a quotation and never closes it, neither do we.

Citations

Migne's citations are reproduced as printed, and this occasionally means reproducing one that is wrong. The policy separates the two things a citation does: the text keeps what Migne prints, while the project's index resolves to the reference actually meant and retains the printed form beside it, so that every correction is auditable. Fourteen such corrections have been entered so far, each verified against the quoted Latin before entry — including, in one work, a reference to III Cor., a book that does not exist.

Citations Migne prints inline rather than in notes are kept verbatim in Latin rather than expanded into English titles, on the same grounds: expanding De consid. into a title would put words on the page that Migne does not print.

What is checked, and by what

Nothing ships on the strength of the English reading well. Before a work is built, a verifier compares every English chunk against its Latin twin and fails on any of the following:

  1. Frontmatter identical to the Latin chunk's.
  2. Column-anchor sequence identical — same anchors, same order, same count. These are the citation-resolving unit of the whole site and are treated as sacred.
  3. Migne's [n: …] citation notes: same count, same content, same order.
  4. Section-heading count identical.
  5. English-to-Latin word ratio within bounds. English runs about 1.5× the Latin; a chunk well outside that is either padded or has dropped something, and gets read.
  6. A duplicate scan across the whole work, catching text re-emitted at a chunk boundary — with a deliberate exception where the Latin itself repeats, because canon collections really do cite the same authority twice.
  7. Guillemet parity against the Latin, per the rule above.
  8. Well-formedness of the markers the edition itself adds.

These are mechanical checks, and mechanical checks catch mechanical failures: omission, duplication, a dropped anchor, a quotation silently converted. They do not certify that a sentence has been understood. That is what the cruces and the reading are for.

The cruces

Each translated work carries a list of cruces — the places where the plate is defective or the reading uncertain, given by column, with the printed form, what was done, and why. Sixty-four of the works translated so far have one, running to something on the order of nine hundred entries in all. They are not a disclaimer. They are the part of the edition that lets you check the part you could not otherwise check: every place a judgment was made is listed, with the printed Latin beside it, so that a reader who disagrees can see exactly what he is disagreeing with.

A crux is required even when the decision was to do nothing. That matters more than it sounds. The failure this project has actually caught in its own work was not invention but tidiness — a defect described accurately in a note and then quietly smoothed in the text. Two works had deleted a printed repetition while explaining in their notes that they were avoiding nonsensical English. Both were restored.

What this method refuses to do

Stated flatly, because these are the suspicions a translation of theological texts has to answer:

It does not paraphrase. The register is literal and every clause is rendered. Nothing is summarized, compressed, or dropped for length, and the mechanical ratio check exists partly to catch it if it were.

It does not smooth doctrine. Where an author says something the translator would not, the author says it. Where the plate makes an author say the opposite of his own argument, the plate's reading stands in the text and the conjecture goes in the notes — the two dropped negatives named above are the worked examples, and they are precisely the cases where a well-meaning corrector would have restored the orthodox sense and destroyed the evidence.

It does not substitute a familiar Bible for the author's. See above; this is the most common way a patristic translation quietly rewrites its author's argument.

It does not correct Migne — not his Latin, not his punctuation, not his citations, not his attributions. Corrections live in the notes and in the index, in a form that lets you audit them.

It does not claim more than it can show. Every work carries one of two claims: first English translation or new English translation. The strong claim is the only one that needs evidence, and it is made only where a verified survey found no prior English. Where the evidence is missing, ambiguous, or merely a search engine's confident assertion, the page says "new" — which is always true of our work and cannot mislead. When four works were found carrying a first-translation claim without evidence behind it, the claim was downgraded rather than defended.

It does not fill gaps with invention. Where the source is missing, illegible, or absent from the digitization, the gap is recorded as a gap. Nothing is supplied from a modern translation or from general knowledge to paper over it.

Why translate this way

Everything above defends the procedure. It does not yet answer the prior question, which a reader may fairly still be holding: granting every safeguard, why should this be done by machine at all?

Because nobody was doing it. There is no complete English Migne. There is no English Acta Sanctorum, and no complete English Bonaventure. These are not gaps waiting on a translator who is nearly finished — they are gaps that have not moved since the nineteenth century, because the cost of closing them is measured in human lifetimes. Two hundred and twenty-one volumes of Latin and a hundred and sixty-one of Greek stood no closer to being read in English this year than they would a thousand years from now. A corpus of this size is not translated slowly. It is not translated at all.

The aim is therefore not a perfect translation. It is to unlock the language gate to the tradition.

There will be errors, including serious ones — as there are in Migne, and in the Maurists he reprinted, and in every edition of this kind ever assembled. The model carries a fault forward, a line or a spacing or an artifact of the transcription, much as a clerk copying his exemplar carried forward the mistake set in front of him. Working this way has made me feel nearer to the medieval copyists than to a modern translator.

That is not an excuse offered in advance. It is the reason the rest of this page exists: if the errors are inevitable, then what matters is whether they are findable — which is why the Latin sits beside the English, why every judgment is logged as a crux, why the column anchors resolve, and why the licence lets you publish a correction without asking.

Two companion projects work the same way and under the same rules: the Acta Sanctorum at actasanctorum.org and Bonaventure's Opera Omnia at bonaventure.wrootpress.com. Three corpora, one method, each vouching for the others.

How to check us

The site is built to make this easy rather than to ask for trust.

The right posture toward this edition is the one a careful reader takes toward any translation, and toward Migne himself: use it, and check it where it matters. It is offered as a serious and disclosed piece of work, not as an authority.