To be honest, I've never tried it on a completely unaccented text. I use it for generating verbal paradigms and it successfully does that (so to that extent is 100% accurate). It requires a small amount of lexical knowledge even for verbs. Nouns would obviously need rich lexical knowledge.Jonathan Robie wrote:Cool! To what extent can I use this to, say, add accents to Greek from Alan Bunning's text? How accurate is it?jtauber wrote:https://github.com/jtauber/greek-accentuation
It also has general tools for testing the law of limitations, stripping accents, etc, all of which could be used in a tool for helping accent new text.
Because, in Alan's case, we're talking about variants rather than a completely new text, one might get a long way just by building an unaccented->accented mapping from existing accented text. I'm not sure accenting actual transcriptions is useful (or even wise, although I'm not a textual critic) but this approach could be used to jumpstart accentuation of a critical text. I found a few errors in SBLGNT when it first came out with some basic heuristic so as a validation mechanism it might be useful.
I mostly mentioned it, not because I think it would specifically solve Alan's use case, but because a lot of people who might find it useful for other things don't seem to realise the library exists.