I first learned of Old Church Slavonic (OCS) when I encountered the work of Dr. Florin Curta on the Bogomils, about two years ago. I don’t recall what led me to investigate the Bogomils but my present intrigue about OCS and Slavic history is easy to explain. We help run an orphan ministry and have a daughter from Ukraine. We also currently sponsor a young Ukrainian who fled the war.
Regarding OCS and Slavic culture I know just enough from personal experience, my brief investigation of the Bogomils, and the broader Carolingian era to realize they occupy a unique and important place in church history. It appears Slavic Christianity is a somewhat distinct thread that develops within the timeframe of the Holy Roman and Byzantine “Empires.” Not surprisingly, the Slavic world adopted Christianity from the latter. If I know nothing more, I suspect this exceeds most in the modern, Western, English-speaking world. If Eastern Christianity in general tends to be obscure in the Western world, the history of Slavic Christianity is only more so, I assume.
In another post, I shared an experiment using a combination of AI (ChatGPT) and Google Translate to translate a 12th century Latin text. With a completely different alphabet, far fewer resources, and many more obscurities for a neophyte like myself, translating OCS seems to present a much greater challenge than Latin. Sometimes unique challenges present unique opportunities.
I don’t know enough about the corpus of untranslated OCS texts (or even where to locate them) to speculate on the potential here. The entire corpus is apparently relatively small to begin with. Nevertheless, I hope this will contribute something useful and/or generate interest in further exploring the possibilities, even for other languages. I would love get some feedback on this and information on other texts that are good candidates for translation. I hope readers will comment if they know of any.
Some general thoughts on AI
Even with just a little dabbling with Generative artificial intelligence (AI), the possibilities it seems to present amaze me. I’ve worked in technology for around 25 years and my professional niche is in Identity Management/Cybersecurity. It makes for an interesting dilemma: Which little voice in my head do I listen to, the one that encourages me to think through the risks presented in using/training something like ChatGPT or the one that encourages me to dive in and leverage it for something I find useful, hoping to realize it’s benefits? At this point, it doesn’t bother my conscience enough to deter me. I do hope such advancements in technology contribute positively to the study of church history and it seems the translation of texts is an area that is worthy of much exploration.
Why use AI to Translate Old Church Slavonic?
Google Translate has evolved into a rather useful service, over the years. In our home and ministry, going on six years, we’ve often relied on it for both spoken and written communication with native Ukrainian/Russian speakers. I’ve used it when traveling in Ukraine, Poland, and Israel. It works remarkably well. Roger Pearse has written on it’s usefulness in translating Latin here. That’s all great but what do you do with an ancient text written in a language that online translators like Google, Bing, and Yandex don’t support? Well, I suppose you can find a professional, commission the translation, and wait for it. Or, you can use AI. Of course, it’s not appropriate to suppose the results are equivalent. Clearly there are pros and cons each way.
Before detailing my quick experiment, here is my attempt at using Google Translate with OCS, using it’s language detection:
As you can see, Google’s language detection isn’t entirely useless with OCS. Apparently early Cyrillic and it’s modern Bulgarian equivalent, as well as some of the vocabulary, are similar enough that Google “detects” Bulgarian and renders both individual words and some complete sentences. In a pinch, it seems there are OCS dictionaries one could slog through, cobbling together what Google can’t. That sounds rather miserable. ChatGPT, as we’ll see, was able to use my description of the text as Old Church Slavonic, and produce a full translation. What follows is information on the source text, an of outline the results, and a comparison to an actual human translation.
Translating a portion of the Life of Good King Wencelaus
The University of Texas at Austin’s Linguistics Research Center has a website with various language lessons. One of the languages covered is OCS. The page has ten sample-text lessons, seven of them are Biblical passages and the other three are extracts from other writings. For this experiment, I chose Lesson 9, from the Life of Good King Wencelaus. It is a legend about a pious Bohemian king. It’s an interesting read but I’m not concerned with the story itself here, only the translation. Characteristics of the text considered important to linguists and historians, may be of interest. Here is a brief description from the lesson:
“This text is of paramount importance for two major reasons. Firstly, at a cultural level this text shows the emergence of a true Slavonic spiritual identity. No longer is the ecclesiastical corpus relegated to stories from neighboring cultures. With the story of Wencelaus, who is both king and saint, the Slavonic culture asserts its legitimacy in contradistinction to the territorial influence of both the Holy Roman Empire and Byzantium. Second, on a linguistic level the native Slavic subject matter ensures that there is no Greek or Latin original to be sought.”
The text used in the lesson is 17 sentences long. It seems an adequate length for getting a decent idea of how well ChatGPT handles OCS. I won’t reproduce the OCS source text here. They explain their English translations on the website. Reproducing the 17 sentences of English may be incidental but makes the comparison a lot easier. I hope it doesn’t trigger any nastygrams from the copyright police. This is a sentence-by-sentence comparison and and I hope it’s easy to follow. Each sentence, as it appears in the lesson, is here alongside it’s ChatGPT equivalent.
The results
Line 1:
Behold, now the prophetic word, which our Lord Jesus Christ spoke himself, was fulfilled:
ChatGPT:
Here is a prophetic word, which the Lord Jesus Christ Himself spoke.
Line 2:
‘For it will happen’, he said, ‘in the final days, which we believe to be now, brother shall rise against his own brother and son against his father, and a man’s foes shall be they of his own household.’
ChatGPT:
For it will be, He says, in the last days: brothers will rise against brothers, and sons against their fathers, and sons-in-law against fathers-in-law; man will be at enmity with his household.
Line 3:
For men will be unpitying to one another, and God will dispense according to their deeds.
ChatGPT:
People will become heartless, and God will render to them according to their deeds.
Line 4:
But there was a certain prince in the Czech Lands, Bratislav by name, and his wife was called Dragomira.
ChatGPT:
There will be a king named Bratislav, and his wife will be called Dragomira.
Line 5:
And having produced their first-born son, they christened him and gave him the name Venceslav.
ChatGPT:
They will bear a son, their firstborn, and they will christen him and name him Vyacheslav.
Line 6:
And as he grew so big as to tonsure him, his father Bratislav invited to this tonsure a certain bishop, Notar by name, and his clerics with him
ChatGPT:
As he grows, like a shoot, Bratislav, his father, will summon him and send him to be tonsured as a bishop by the name of Notar, and he will become his cleric.
Line 7:
And when they celebrated the mass for him in the Church of the Holy Maria, the bishop took up the boy, placed him on the top step before the altar, and blessed him, saying:
ChatGPT:
Having sung praises to him in the holy church of Saint Mary, the bishop will take the young boy, place him on the steps before the altar, bless him, and say,
Line 8:
‘Lord God, Jesus Christ, bless this boy, as you have blessed all yours who are just.’
ChatGPT:
‘Lord God, Jesus Christ, bless this boy, as You have blessed all Your righteous ones.’
Line 9:
And so with a blessing he was tonsured.
ChatGPT:
Thus, with the blessing of the bishop, he will be tonsured.
Line 10:
And by this we believe that, with the blessing of that just bishop and by his prayers, the boy began to grow, protected by God’s grace.
ChatGPT:
It is believed that through the blessing of that righteous bishop and his prayers, the boy will grow, nurtured by God’s grace.
Line 11:
He learned well the Slavonic and Latin letters.
ChatGPT:
He will be taught in the Slavic and Latin languages.
Line 13:
And Boleslav, his brother, grew alongside him; both were still young.
ChatGPT:
His brother Boleslav will grow up under his protection, and both will prosper.
Line 14:
But their mother Dragomira ruled the land, and adminstered to the people, while Venceslav was growing up; and when he was grown, he began himself to administer to his people.
ChatGPT:
To their mother Dragomira, the land will be entrusted, and she will build structures until Vyacheslav grows up, and as he grows, he will build his own people.
Line 15:
But by the grace of God, prince Venceslav truly learned well not only letters, but was also filled by faith.
ChatGPT:
By God’s grace, Vyacheslav, the prince, will not only excel in worldly knowledge but also be perfected in faith.
Line 16:
He would perform good deeds among all the poor, clothe the naked, nourish the hungry, take in strangers according to the word of the evangel.
ChatGPT:
He will do good for all the poor, clothe the naked, feed the hungry, and welcome strangers according to the Gospel’s voice.
Line 17:
He would not permit a widow to be disgraced — he showed mercy to all people, rich or poor; he served those who served God, and adorned many churches with gold.
ChatGPT:
He will not allow widows to be oppressed. He will show compassion to all people, both the poor and the rich. He will serve those who work for God and adorn churches with much gold.
Conclusion
Not bad, in my estimation. It seems one could learn some basics and, with a little patience, produce a decent amateur translation. Some of the sentences (e.g. lines 3 and 10) actually seem like they may render a bit better. The translation of “languages” vs. “letters” in line 11 seems like it might capture the intent. Perhaps this is an example of literal vs. dynamic translation.
It’s worth noting that I didn’t make any attempt at refining the translation. It may be possible to improve the result with a little work. For example, telling the tool that verb tenses are off and directing it to correct them may yield an improved translation. Likewise, one may even be able to provide contextual information about a text that is useful.
These are fun little experiments. What other candidates for AI translation are out there?
Interesting post. I noticed that the human-made translation seemed to write in the past tense, while the AI translation wrote in the future tense. Do you think there might be a reason why the verb tense is different between the two?
Thanks for the comment and question. I’m not a linguist and don’t know much about the language but I would guess it’s possible “tense” cannot be expressed as it is in English. Perhaps it’s context dependent. Latin apparently doesn’t have definite and indefinite articles and is likewise dependent on context. In this case, one might deduce the tense based the story involving historical narrative. I haven’t tried it but I wouldn’t be surprised if ChatGPT could “correct” the tense of most of the sentences if the text were described as a historical narrative, written mostly in the past tense.