This post at Three Pillars Blog came to my attention yesterday. Scott Cooper is experimenting with Google Translate and ChatGPT AI to see whether we can get anything useful out of Old Church Slavonic.
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.
He’s picked an Old Slavonic text which is online, and indeed already has an English translation online, and run ChatGPT against it.
Obviously we have to ask – is that translation already in the ChatGPT database? If so, the AI translation will not in fact be doing anything much. What we need is some u untranslated Old Slavonic.
But very interesting!