On 22/07/2026 11:21, Werner LEMBERG wrote:
This is intended as an experiment that could start with a few
chapters, say, to find out whether such an approach is feasible. Of
course, we would have to discuss the conditions for the LLM to get a
good result.
Given the quality of LLMs nowadays (much as I don't like the things), I
would think a fairly basic LLM, given a music technical dictionary and
the raw English text, should do a very good job of translating.
You've then got to munge it back into the form we want for lilypond ...
In spite of the grammatical quality I expect that a lot
of technical terms will be mistranslated; we thus have to proofread
the results (I volunteer for that).
With a decent dictionary? Nah, it should do a near perfect job. You
could probably feed it a calc spreadsheet with two columns - "English"
and "German". Put all your technical terms in that, and see what it does.
But as the guy who proofread pretty much the entire 2.4 manuals back in
the day, you're volunteering for one heck of a job!
Cheers,
Wol
