This is fascinating! Thank you so much for sharing this groundbreaking work. I am interested to know more about how much you have tested it with unpublished/untranslated examples, since in some of my experiments with AI translation, on the one hand it was able to produce something just by providing a reference to the papyrus number, while on the other, when I gave it a significant portion of a text that has never been translated into English, it was able to get some of it and then filled in plausible text around it. The propensity to do that will necessitate caution, while on the other hand I am eager to see how this very aspect of LLMs might be harnessed to explore reconstruction of fragmentary manuscripts and inscriptions.
Thanks! My tests were all about the more wonky aspect of structuring texts. We haven’t tried translation yet, but I just heard of someone running these experiments on lemmatization of Greek, which it seems to be doing well. My sense is that the value of this work will not be as much in reconstruction (where I’m not sure it does better than a human, and in any case needs supervision) than in large scale analyses.
This is fascinating! Thank you so much for sharing this groundbreaking work. I am interested to know more about how much you have tested it with unpublished/untranslated examples, since in some of my experiments with AI translation, on the one hand it was able to produce something just by providing a reference to the papyrus number, while on the other, when I gave it a significant portion of a text that has never been translated into English, it was able to get some of it and then filled in plausible text around it. The propensity to do that will necessitate caution, while on the other hand I am eager to see how this very aspect of LLMs might be harnessed to explore reconstruction of fragmentary manuscripts and inscriptions.
Thanks! My tests were all about the more wonky aspect of structuring texts. We haven’t tried translation yet, but I just heard of someone running these experiments on lemmatization of Greek, which it seems to be doing well. My sense is that the value of this work will not be as much in reconstruction (where I’m not sure it does better than a human, and in any case needs supervision) than in large scale analyses.