One Sentence of Hermes, Copied for Sixteen Hundred Years

The Asclepius is a dialogue in which Hermes Trismegistus teaches his pupils about God, the world and the human soul. It was written in Greek, probably in the third century, and translated into Latin by the early fifth, when Augustine quoted the Latin in the City of God. For most of the dialogue, including the passage followed here, the Greek is lost. Every reader since late antiquity has known it through that one Latin translation.

A little of the Greek does survive. Its title was the Logos teleios, the “Perfect Discourse”. Lactantius, writing about 310, quotes it in Greek and refers to three different parts of it. The prayer that closes the dialogue is preserved in Greek in a magical papyrus now in the Louvre. A Coptic translation of chapters 21 to 29 and of the prayer was found at Nag Hammadi in 1945. None of these reaches chapter 32.

The Latin is not Marsilio Ficino’s, though it is often taken for his. Ficino translated the fourteen Greek treatises of the Pimander in 1463, and their Greek survives. From the early sixteenth century printers bound the old Latin Asclepius in with his Pimander, as the Aldine press did in 1516, and the two have travelled together since. The Asclepius translation is a thousand years older than Ficino. The manuscripts credit it to Apuleius, which Scott and other modern editors reject, and its translator is unknown.

This note follows a single sentence from chapter 32 through the copies of it we hold: a fifteenth-century manuscript, printed editions of 1516, 1532, 1549, 1593 and 1778, a Franciscan commentary of 1590, and Walter Scott’s critical edition of the 1920s. Each copy was read against its scan. The sentence turned out to carry almost every kind of error a text can pick up on its way to us, including some we added ourselves this year.

The sentence

Hermes has been describing three kinds of mind: God’s, the world’s and ours. Then the Latin says:

Intelligentia enim sensus humani qualis aut quanta sit, tota est in memoria praeteritorum. Per eam enim memoriae tenacitatem, gubernator effectus est terrae. Intellectus autem naturae et qualitatis sensus mundi ex omnibus quae in mundo sensibilia sunt, poterit pervideri.

Word for word: “For the intelligence of the human sense, of what kind or how great it is, is all in the memory of things past. Through that tenacity of memory he was made governor of the earth. But the understanding of the nature and quality of the sense of the world can be seen through from all sensible things in the world.” The middle sentence is clear. The first and third are not: “the intelligence of the human sense, of what kind or how great it is” has no work to do in its own sentence, and the third sentence repeats the oddity with the world in place of man. A few lines later it comes round a third time, for God.

What went wrong in the translation

Walter Scott, editing the Hermetica in the 1920s, said plainly that the Latin makes no sense here, and asked why. His argument is short. A copyist garbles a phrase once, at random. This phrase is wrong in the same way three times, so the fault is not in the copying of the Latin but in the translation itself (Scott’s commentary, pp. 214–15).

He then works backwards to the Greek. Suppose Hermes wrote ἡ γνῶσις ἡ κατὰ τὸ ποιόν, “the knowledge that corresponds to the character of” a given mind. The preposition κατά here means “in keeping with”. If the translator took it to mean “of”, he would turn “the knowledge that fits the human mind” into “the knowledge of the quality of the human mind”, and then, treating ποιόν and ποσόν (“of what kind”, “how much”) as questions, into qualis aut quanta sit. Read with κατά restored, the passage sets out three grades of knowledge, each matched to a kind of mind. This is how Scott translates it (p. 357):

The knowledge which corresponds to the character and extent of the human mind is based wholly on man’s memory of the past; it is the retentiveness of his memory that has given him dominion over the earth. The knowledge which corresponds to the nature and character of the cosmic mind is such as can be procured from all the sensible things in the Kosmos.

Scott marks every Greek word he supplies with a question mark, and says that the sense he recovers cannot be considered certain. With the Greek lost, it is a reconstruction. It is also the only reading so far that makes the sentence mean something.

Copied faithfully, for centuries

The broken phrase is in every copy we hold, word for word: the fifteenth-century manuscript now in the Biblioteca Laurenziana in Florence, the Aldine edition of 1516, the Basel edition of 1532, Jean de Tournes’s Lyon edition of 1549, Francesco Patrizi’s Magia philosophica of 1593 and the 1778 edition of Apuleius. None of them flags it or tries to mend it. Faced with a sentence they could not understand in a book ascribed to Hermes, these copyists and editors preserved it exactly.

In 1590 the Calabrian Franciscan Hannibal Rosseli, teaching in Kraków, reached this sentence in his six-volume commentary on Hermes. He sets it as the heading of a chapter (Book VI, Commentary XI) and spends two pages explaining it through Boethius and Augustine. In our translation of his Latin:

Rightly, therefore, was it said by the most wise Mercury that the intelligence of the human sense—that is, to understand, or the act of understanding of the human intellect—proceeds from the memory of the past

His gloss drops qualis aut quanta sit and keeps the part that makes sense. A few lines earlier on the same page he quotes Augustine: “Whoever understands any thing otherwise than it is, is deceived.”

Copyists and printers add their own

The translator’s error is consistent. The errors that came after it are not. Scott’s apparatus records that the medieval manuscripts already disagree over the third sentence. One group reads et qualitate, another ex qualitate, a third qualitate et, and Scott prints the emendation et qualitatis (p. 356). The printed editions inherit the split. The Aldine of 1516 has naturae, & qualitatis, & sensus mundi; Basel in 1532 has naturae qualitate & sensus mundi; the Apuleius of 1778 has naturae et qualitatis sensus mundi.

Sixteen years separate the Aldine from the Basel edition, and in that time a clause went missing. After umbra dignoscitur the Aldine continues ubi enim quid temporum dimensione cognoscitur, ubi sunt mendacia: “for where anything is known by the measure of time, there are falsehoods”. In the Basel edition dignoscitur is followed directly by Vbi sunt mendacia. Six words have dropped out: the eye went from dignoscitur. ubi to cognoscitur. ubi. Patrizi’s edition of 1593 has the same gap, and the Lyon edition of 1549 and the Apuleius of 1778 do not. A shared omission of this kind is how editors work out which copy was made from which: Patrizi’s text belongs to the Basel line, not the Aldine one.

Patrizi’s 1593 edition adds a slip of its own a few lines further on. The 1778 text reads:

Hoc autem differt intellectus a sensu, quod intellectus noster ad qualitatem sensus mundi intelligendam…

Patrizi’s page has:

The words between sensu and the next sensus, “quod intellectus noster ad qualitatem”, are gone. The eye of the compositor, or of whoever made his copy, jumped from one word to the same word further on and carried on from there. Textual critics call this saut du même au même, the jump from same to same, and it is the commonest copying error there is. What is left reads “the intellect differs from the sense of the world, to be understood and discerned”, which is not Latin anyone wrote.

The same mistakes, in 2026

Source Library’s transcriptions are made by AI models reading the page images, and its translations by AI models reading the transcriptions. When we checked our text of this passage against the scans, we found our own copies of it had picked up three of the old kinds of error.

The long s

Just after our sentence comes a phrase that Scott thinks has strayed from earlier in the chapter: aeternitas quae secunda est, “eternity, which is second”. Printers of the period set a long s, ſ, at the start and in the middle of words. It differs from f only by the short bar, which on the ſ stops at the left side of the stem and on the f crosses it. Misreading one as the other is the classic mistake of anyone new to early print.

All five printed editions whose scans we checked print secunda, with the long s. Our OCR read four of them as fecunda, “fruitful”, and our translations follow it. It read the fifth correctly:

Fecunda is a real Latin word, and “fruitful eternity” sounds like something Hermes might say. The wrong reading is easy to accept because it is plausible.

The skipped lines

The Aldine page is a folio set in long lines of small type, and on it our OCR made the same slip as Patrizi’s compositor, in almost the same place. The last four lines of the Aldine crop shown earlier read hoc autem differt intellectus a sensu, quod intellectus noster, ad qualitatem sensus mundi intelligendam … & sic contingit nobis hominibus. Our transcription has hoc autem intellectus a sensu, quod intellectus noster, git nobis hominibus. It stopped at the end of one line, at noster, and resumed two lines further down, in the middle of the word contingit. Two lines of Hermes are missing from our text of that page, and the English built on it reads “our intellect grants us humans that we might see those things that are in heaven as if through a mist”, a sentence assembled from the two ends of the gap.

The easier reading

The Florence manuscript is written in a small, heavily abbreviated cursive hand, and it is much harder for a machine to read than print. Here our OCR did something different. Where it could not read the hand, it wrote fluent Latin that is not on the page.

Each machine reading looks like ordinary Latin and makes some sense, and a reader of our transcription would have no reason to doubt it. Textual critics have a name for this too: the lectio facilior, the easier reading. A copyist who cannot make out or cannot follow the text in front of him writes what he expects it to say. That is why editors give preference to the harder reading. A scholar comparing witnesses from our transcription would record “ex uberioris talis rei” as a variant of the Florence manuscript. It is not one.

A fluent translation of a broken sentence

The last problem is quieter. Scott says the Latin of the sentence “is impossible to make sense of”. Our English of the 1532 text reads:

For the intelligence of the human sense, of whatever sort or size it may be, exists entirely in the memory of past things. Through that tenacity of memory, man was made the governor of the earth. But the intellect of nature and the sense of the world can be perceived from all sensible things that are in the world.

It is a fair rendering of the Latin, and it gives no sign that the Latin is in trouble. “Of whatever sort or size it may be” smooths the broken phrase into an ordinary English aside. Rosseli did the same thing in 1590 by explaining it.

Three kinds of error

Followed through these copies, the errors in this one sentence fall into three groups.

Misunderstanding. The translator misread how a Greek construction worked. Because he misread it the same way every time, the error is regular, and its regularity is what let Scott detect it sixteen centuries later.

Eye and hand. Skips from a word to the same word further on, letters taken for similar letters, and small slips that spread through the manuscripts and then the printed editions. These are random, local and usually easy to correct once two copies are set side by side.

Expectation. A reader who trusts the text explains the error or replaces it with something more plausible. This kind lasts longest, because the result reads well. The 1590 commentary, the manuscript’s easier readings and our fluent translation all belong here. A machine trained to produce likely text is very good at this kind of error.

The safeguard is the same as it has been since the Renaissance editors: keep the witness in view. Every page in Source Library sits beside its scan so that a reader can check the transcription against the leaf, and every edition of a work we hold is listed on its book page so that copies can be compared. This note used nothing else.

Our errors in this passage. In the passage followed here, we checked our transcriptions against the scans of five printed editions and one manuscript. Four of the five printed transcriptions misread secunda (printed with a long s) as fecunda, and the 1516 transcription skips two lines. The manuscript transcription had at least three readings that are not on the leaf. These are logged with the other manuscript and early-print OCR cases on issue #4877. The 1516 Aldine has since been re-read with a stronger model, and its Latin transcription of this passage now matches the page; the English translation beside it had not yet been redone. The other editions and the manuscript had not been corrected when this was published. If you find an error on any page, the feedback button on that page reaches us.

Sources: Walter Scott’s Hermetica, vol. 1 (Oxford, 1924) for the Latin text and translation of the Asclepius, and vol. 3 (1926) for his commentary on chapter 32; the manuscript and the printed editions are linked where they appear. Quoted English is Scott’s own or Source Library’s translation of the page it links to.