Here’s a pretty question, sent in by email. I imagine that more such letters are in my future, but I am really not sure what to do with them.
I thought I would drop you a little note to let you know about a project that I have recently uploaded. This is my first attempt at using ChatGPT to translate a Latin text that I have been wanting to read for many years. It is Marius Mercator’s “Commonitorium de Coelestio”/”Commonitorium super nomine Cælestii”. I, of course, do not know Latin but I am familiar enough with Pelagianism (having studied it intensely for several years) to know what to expect and from that perspective the output text was pretty much what I expected. To my knowledge this has never been available in English before. I have put it up on the Patristics In English website here:
https://www.seanmultimedia.com/Pie_Marius_Mercator_Memo_On_Coelestius_With_Latin_Text.html
I am not sure of the value of AI translations. On one hand I feel like I was able to get a good understanding of a text that I had wanted to read for many years. On the other, one is left to wonder how accurate they are. At any rate, unless people come after me with pitchforks and torches like Frankenstein’s monster after reading it, I would like to do more translations like this on Pelagianism. Maybe Pope Zosimus will be next.
A translation of Marius Mercator would be very valuable to have.
But … this is an AI output. Is it valuable?
Well, in some ways, yes. The Latin text has been put online, and the author has run it through ChatGPT which saves us all from having to do the same in order to get an idea of what it says. He’s given it a read to check that the ideas of Pelagianism are actually there, which is a bonus. Surely this is better, far better, than nothing?
But in other ways, no. AI is not a translator. It’s a search engine with a chatbot on the front, turning the search results into something plausible. It can often produce excellent readable translations of a passage. It can also generate material which isn’t in the source at all. AI is a good tool, to reduce the amount of effort needed to produce a first draft of a translation. But every word and every sentence needs to be checked, compared against the original, in order to produce a reliable translation. The unwary reader will suppose that the AI output is a translation, when it is not.
What on earth do we make of this?
There, he was accused before Aurelius, the bishop of the said city, through a book sent by a certain Deacon Paulinus of holy memory, the bishop of Milan.
ibique de infra scriptis capitulis apud Aurelium episcopum memoratae urbis per libellum a quodam Paulino Diacono sanctae memoriae Ambrosii Mediolanensis episcopi est accusatus,
Just reading the first few lines I notice that the AI has omitted a translation of “de…capitulis” and made a mistake by assigning to Paulinus the bishopric of Milan. Also, a “libellum” is more likely an “indictment” than a “book.” Celestius was condemned for chapters transcribed and discussed later in the Marius Mercator’s account and the person of blessed memory and bishop of Milan would be Ambrose. Paulino was Ambrose’s deacon.
Thank you. I feared as much.
An old friend of mine used to say that if it’s worth doing it’s worth doing badly, and there’s some truth in that, even though it is a problematic statement when it comes to publication. Publishing a translation implies that you have understood a text and have the ability to convey its menaing in another language. It seems a terrible prospect to me that the web will fill up with error-laden AI translations and that it will be difficult to tell which are reliable and which are not. Latin, with its complex syntax and relatively small vocabulary, where the same word can bear many meanings, is still a major challenge for machine translation, as Douglas has shown above: an omission, a mistranslation, and a major misunderstanding in just a single sentence. This effort is no doubt well-intentioned, but we all know how urban legends and the like work: once they become established they can be absolutely uncorrectable. At the very least AI “translations” have to be clearly labelled as AI products.
For a very severe judgement on unskilled translators (too severe, I suppose) see Mark Thakkar’s scathing review of two Wyclif translations: https://www.academia.edu/44398256/Duces_caecorum_On_Two_Recent_Translations_of_Wyclif