When I’m stumped by a Greek sentence, I check the LCL English translation. Most of the time the translation ends my perplexity. But when it doesn’t, I might give the sentence or phrase to google ai, and ask for help.
I lack the ability evaluate google ai’s answers, in the same way that I lack the ability to evaluate the answers to queries here. For as I see it, if I could evaluate textkit guru replies, then I wouldn’t post a query, for I wouldn’t be baffled to start with. I use guru replies here as hints for study, which I always need.
For example, I queried google ai about a sentence in Isocrates 1.109, and asked for translation and parse. It translated the phrase “ὡς ἡμᾶς” as “to us”, a new meaning of “ὡς” for me. I tested this meaning against the lexicon, and it was there, the very last meaning given by CGL. I had overlooked it in my earlier dictionary work. Then I read again the LCL translation, and recognized how “ὡς ἡμᾶς” had been Englished, something I had not noticed before.
I find google ai more helpful at translations and parsing that at explaining grammatical precepts.
I also find Gemini great for getting the gist of things - it’s quite capable of giving idiomatic English translations. It’s also great at giving high-level summaries and literary analysis.
I wouldn’t press it too much though. It’s OK with basic morphology, but for anything slightly more tricky (especially in verse) it tends to make up forms and claim they are “dialectal”. But if your goal is to keep reading at a good pace and just need some help every now and then to remove a road block, it’s better than most other free resources online. It’s also very convenient because you can just paste the sentences you don’t understand.
My theory is that all those language models have ingested enough translations and commentary that they basically “remembered” the meaning of each sentence (the entire body of Classical Greek text is not that large anyway), and since they are quite good at phrasing English sentences they’ll often give you something more natural-sounding than their source material. But they haven’t actually learned enough of Greek grammar to tell you exactly why a sentence means that.
I’ll add: they seem to have an enormous database of translations and commentaries. They find the location of the quotes, and match these with their database of commentaries and translations. (AI services are downloading data in huge amounts from Perseus, as I have been told, which explains why we may encounter delays while using Perseus.)
Incidentally, I often query google ai about how it came up with explanations. Yesterday I disparaged its flattering manner in response to my questions —E.g.: it says things like “that is a brilliant insight, for you have pinpointed” something or other. Google ai explained why it talked like that, and printed some boilerplate to put in my next query to make it cut out the BS.
Google AI explained that it has no “self” and no ability to remember past transactions. Each query calls up a new search, tabula rasa. But for the current transaction with a user, it keeps a sort of scratch-pad file, which enables it to follow references to earlier points in that transaction.
Overall, I’d say AI matches the statement of Chief Dan George in the film “Little Big Man”: “Sometimes the magic works, sometimes it doesnt.”