I just realised, the DeepL machine translation tool I have been using since early this year is missing a feature that would be useful: translating from classic Latin to modern languages.
This should be doable I would expect: there’s enough but also a limited corpus of text, meaning what needs translation is clear and bounded. There’s existing translations of most if not all those classical texts, and the language structure has been fixed for ages because it’s no longer in use. All of it is in the public domain. So much so that you might even be better off without algorithmic translation and go with straightforward database searches. Google, Baidu and Yandex machine translation all seem to incorporate Latin, but both Google and Yandex mangle even the simplest of things into unrecognisable output, if not leaving most words simply untranslated.
Some online searching suggest it may be less straightforward than that. This page by Susan Shin, Summer Zhou, and Zoe Zitzewitz from two years ago shares a neural translation model for Latin. They say that there is a lack of accurate translations and accessible resources which I find surprising, making it difficult to build a model for Latin. It also says that Google Translate only uses a statistical method, and has no model for grammar, syntax, idiom, or meaning. The latter is readily apparent from using Google Translate for a few Latin phrases, but it’s again surprising. After all there are machine translators for Klingon. 😉
I mean all this w.r.t. translating from classic Latin to modern languages, for translations from modern languages to Latin all the above doesn’t apply.
The title of this posting is incorrect Latin, but hey, I blame (the absence of) machine translation.
Moving on from Latin, how about older versions of current modern languages?