Dear Werner, David, Dan and Jonas, reading this thread makes me wonder about a different approach to the question.
Given a cultural context where readers would commonly face a high language barrier to understand the Lilypond documentation (and I think Kunpeng's case is a possible example of such a situation), it seems to me at the moment the project can provide: 1- no documentation 2- unvetted, machine generated translations (that would be clearly marked as such) 3- outdated and/or incomplete documentation, though the little that is available was quite accurate when it was new 4- properly authored complete documentation I am an observer of state 3 for the Italian set, for example: at least in the past (several years back) where I had casually browsed it, the Italian set of docs was off sync with the English, so I stopped using it completely. Of course, because I am effectively bilingual, this poses no problem to me personally, but that's neither here nor there, except to indicate that in a way one could see that some people might prefer 2 to 3 in practice (likely on a case-by-case basis, arguably). I am very much wondering if providing 2 as a base for everything is in all cases better than 1, and moving to a model where all languages make available a complete doc set, where the varying degrees of documentation qualities are not measured to completeness or how recent they are, but rather to how much time humans have been able to spend editing the translated pages. And my reasoning here is that if a user does not know English, it seems to me they are better served by a fresh, potentially "meh" translation, than they would be by no translation at all, or a translation that has fallen out of sync with the program. Per David's point, this observation is based on the fact that the quality of the LLM-based automatic translations is actually quite high, it's miles ahead of what Google translate did a couple years back. In fact I have found GPT to have considerable competence in music theory matters, much deeper than I would have expected at first. If we were to adopt this view, the guidance for folks like Kunpeng would be more like: - start by machine-translating everything you like/can take on - mark it all very clearly as automated translation - then take all you can take on for manual review, review it and unmark it as you go along Wouldn't this be an overall better service to lilypond users? Ciao, L On Thu, May 14, 2026 at 10:56 AM Kunpeng He via Discussions on LilyPond development <[email protected]> wrote: > Thanks for your kindness criticism and guidance. I completely agree that > unreviewed AI-generated content could do more harm than good, and I > really appreciate you raising that concern. > > The LilyPond documentation is so comprehensive that a fully manual > translation would take an enormous amount of time, so I wanted to > explore whether an LLM could help speed up the process. For now, I've > only taken a quick look at the output, and the quality seems adequate > for my own learning. > > As for contributing translations back to LilyPond, my current plan is to > review each chapter carefully together with friends who have a > background in music, so that the overall quality is at a level the > project can rely on before anything is submitted. > > Thanks for your concern and Werner's guidance. > > Best, > > Kunpeng He > > 在 2026/5/13 20:14, Dan Eble 写道: > > On 2026-05-12 19:10, Dan Eble wrote: > >> On 2026-05-12 09:17, Kunpeng He via Discussions on LilyPond > >> development wrote: > >> > >>> As an amateur music lover, I cannot guarantee the overall quality of > >>> the translation. > >> > >> I am not a lawyer, I don't set policy around here, and I understand > >> that none of us really guarantees anything; however, I don't think > >> it's reasonable to accept content from a submitter unless he is > >> willing to put his own reputation on the line for it. > >> > >> If in your own eyes you are not competent to judge the quality of the > >> work, and if there is no mentor to guide you and "co-sign," it would > >> be best if you did not submit it. > >> > >> Nobody needs another hel-arabic.ly. > >> > >> Having said that, I do think that machine translations could help a > >> lot of people as a separate project with clearly presented goals and > >> caveats. > > Someone thought my reply was rude. I'm sorry if it stirred you in > > ways that I didn't intend. > > > > I see a tsunami of poorly reviewed AI slop rising up into open-source > > software. Guard your good name, friends. > > -- Luca Fascione
