Bonaventure

This project presents an English translation of the Opera Omniaof St. Bonaventure of Bagnoregio (1221–1274), the Seraphic Doctor — drawn from the Quaracchi critical edition (Vols. I–X, 1882–1902) prepared by the Fathers of the Collegium S. Bonaventurae.

All four books of the Commentary on the Sentences of Peter Lombard — Volumes I through IV — are complete and published here. Work is now under way on Volume V, the opuscula, beginning with the Breviloquium; the intention is to finish all ten volumes, through the Itinerarium mentis in Deum, the Collationes in Hexaemeron, the disputed questions, and the sermons. No complete English edition of the Opera Omnia has ever been published.

Every unit of text on this site is drafted with the help of a large language model working from the Quaracchi Latin, and is then checked against the printed page by a procedure that has been written down, tested, and revised across some five hundred units of text. The first half of that sentence is what people ask about. The second half is what decides whether the translation is any good, so it is set out here in full.

The Latin is not read off the page image. It is set from the Internet Archive’s OCR of the Quaracchi scan, which is markedly more accurate than reading small nineteenth-century type at any practical resolution — a direct read of the print produces errors the OCR gets right (bonus taken for utens, homo for bonum, a proceditur reversed into procedam). Where the OCR is sound it governs, and the printed page is consulted to settle what it garbles.

From Volume II onward the edition is set in two columns, and there the OCR fails in a particular way: it shatters diagonally across the gutter, a token to a line, through exactly the passages that matter most — the Respondeo, the Solutio, the scholia, and every page footer. In those regions the priority inverts. The page is extracted at 450 dpi, cropped into column bands, and read band by band, top to bottom, left column and then right. The gutter is measured on each page, because it wanders and alternates with the page’s parity; a band cropped at the wrong split truncates a column without any sign that it has. At least one word in the published corpus — transumtum, where the page reads transumtivum— was wrong for precisely that reason until a later pass caught it.

The English is literal and it is parallel: paragraph for paragraph, and footnote marker for footnote marker, each marker standing where it stands in the Latin rather than at the end of the clause. Standard scholastic formulae are rendered the same way every time — Videtur quod as “It seems that,” Sed contraas “On the contrary,” Respondeo. Dicendum quodas “I respond: It must be said that” — and a table of key terms is fixed for the project and applied across all ten volumes.

The scholia and the critical apparatus are translated in full, to the same standard as the body. That is unusual, and it is deliberate. Those notes carry much of the scholarly value of the Quaracchi edition, they have never been put into English, and they are the first thing an abridgment drops.

Quaracchi’s footnotes are the hardest part of the edition to reproduce, and the part a paraphrase would quietly lose. Three properties of the printed page make them dangerous.

• The numbering restarts on every printed page. A passage spanning four pages carries four independent sequences of 1, 2, 3. Footnote labels are therefore qualified by page internally, so that two notes numbered 4 cannot collide and silently overwrite one another; the reader sees the printed number.
• Notes run over.A note reaching the foot of a column continues at the head of the next column’s footer — sometimes on the following page, sometimes in the middle of a word — and the continuation carries no number. The fragment left behind at the column foot usually reads as a complete entry, so the loss is invisible. Every footer’s last entry is treated as incomplete until its continuation is either found or positively excluded from both sides. Of the sixteen units in the Breviloquium prologue and first part, ten carried a runover.
• The OCR loses footers wholesale.In the last distinction of Book IV it dropped three entire footer registers — twenty-six notes across three pages — with nothing to indicate that anything was missing. They were recovered because the apparatus is read off the page bands as a matter of course, not only when something looks wrong. In Volume V the OCR does not render footnote numerals at all; every superscript arrives as a stray punctuation glyph, and the band image is the only source there is.

Before any unit is committed it passes a set of scripted audits: a paraphrase audit, which scores word overlap against the raw Latin and separates literal transcription from summary; a header audit, which counts the semantic divisions in the source against those in the file and catches a whole question gone missing; an apparatus count; and a formatting scan confirming that every footnote definition has a matching marker in both languages. A seam screen examines every boundary that falls inside a printed page, which is where text goes missing, and the site is rebuilt to confirm the file parses and the two languages carry the same anchors.

On a fixed cadence — roughly every hundred printed pages, and unconditionally at the end of each work — three further passes run. Every ambiguity flag is resolved against a 600 dpi read of the page or else formally accepted as illegible, with the reason recorded. The whole corpus is scanned for formatting drift, not merely the new part. And every boundary in the finished section is swept to confirm that no text and no footnote was lost where two units meet. In the closing gate on Book IV, a hundred and eleven mid-page seams were checked and three dropped notes recovered.

Several of these checks exist because an earlier one failed. The formatting scan spent months hardcoded to two volumes, so four volumes’ worth of clean results were false; when that was found the tool was repaired and sixty-one genuine defects surfaced behind it. A register of known open defects is kept with the project, along with the resolution log for every gate, and both are consulted before new work rather than after.

Plain scanning artifacts are repaired silently — a word broken across a line rejoined, an ahquid restored to aliquid. Anything genuinely uncertain is flagged rather than guessed, and the flag stays in the file until someone resolves it against the page.

Quaracchi’s own errors stay. Where the editors number a footnote eleven on a page carrying ten, the page is reproduced as printed and the discrepancy noted; no eleventh note is invented to make the sequence come out. Where a scriptural citation in the apparatus is wrong — Matth.12:14 for what is plainly 12:40 — the printed reading stands. An earlier pass had silently corrected that one, and the correction was reverted. This is an edition of Quaracchi, not an improvement on it.

Beneath the question of human against machine lies a prior question that settles it: nobody was doing this. There is no complete English Bonaventure. There is no English Acta Sanctorum. There is nothing approaching an English Migne. These are not gaps waiting on a translator who is nearly finished. They are gaps that stood no closer to being filled this year than they would a thousand years from now, because the cost is measured in human lifetimes and that cost has not moved since the nineteenth century. A corpus of this size is not translated slowly. It is not translated at all.

The other two are now being attempted the same way, by the same hand and under the same procedure: the Bollandists’ Acta Sanctorum at actasanctorum.org, and Migne’s Patrologia at migne.app. Neither had been attempted whole before.

So the aim is not a perfect translation. The aim is to unlock the language gate to the tradition. Someone with no Latin can now reach the passage, follow what Bonaventure is arguing, and see exactly where to check him. That is worth more than a flawless rendering of a single question arriving in fifty years.

There will be errors in this, including serious ones. There are serious ones in Migne, and in Quaracchi, and in every edition of this kind ever assembled. What is striking is the shape they take. The model will carry a fault forward — a line, a spacing, 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.

The following scholastic terms are rendered consistently: uti → to use; frui → to enjoy; ratio → account/ground; processio → procession; suppositum → supposit; potentia → potency/power; actus → act/actuality; caritas → charity; exemplar → exemplar; vestigium → vestige; similitudo → likeness; honestum → the honorable; complacentia → complacency; actus quietativus→ quietative act.

Everything here is a draft and should be cited as one; passages still under review are marked as such in the text. Corrections are wanted, not merely welcomed. If you read Latin and find something wrong — a mistranslation, a dropped note, a misread page — write to wilson@wrootlabs.com. It will be fixed, and the correction recorded. Inquiries from scholars who would like to review or contribute to the work are welcome on the same address.