Idea: Expanding multilingual support for WebODM Documentation

Agree that it would be very useful to have documentation translated in other languages. The biggest hurdle is keeping the docs in sync once a translation is started. I've also had this challenge with another open source project of mine (LibreTranslate).

The most scalable approach I've found is to use AI to write a rough draft of the translation in different languages, and then have people review the translation. AI is not perfect, but keeps the structure homogeneous and gets the translation about 90% correct for common languages like French, Italian, German. At every change in the English translation, the AI can sync up other languages (but still needs a reviewer in the loop).

If this approach sounds interesting, I can open a PR for Italian and someone needs to review it / edit it afterwards. Then, occasionally there will be a need to check the accuracy of the synchronization when the English docs change.

Just as a rough idea and a starting point for discussion, I’ve sketched out a potential automated workflow to show how we could set this up using standard, free open-source tools.

The concept relies on GitHub Actions (which are 100% free for open-source projects) and a lightweight Python helper script:

• Trigger (GitHub Actions): A workflow monitors the docs/en/ folder. Every time a .md or .rst file is created, modified, or deleted on the main branch, the Action wakes up.
• Translation Engine (Python + LibreTranslate): A Python script reads the changes, splits the text into smart paragraphs (to avoid hitting API character limits and protect Markdown layout), and translates them. Since Piero mentioned LibreTranslate, the script should use LibreTrasnlate.
• Pull Request Automation : The workflow automatically commits the translated Italian files into a separate branch (like l10n-it-update) and opens a Pull Request.
• Notification : The workflow automatically opens a GitHub Issue, tags the registered Italian reviewers, and links the PR so we can quickly review, edit, and merge.

btw:
" I can open a PR for Italian and someone needs to review it / edit it afterwards."
i'll do, maybe in some days, but i'll do!

@pierotofy @role What about this PoC tool?
https://github.com/lucabonuccelli/sampleGitHubAction4translate

Is the PoC AI generated?

https://github.com/lucabonuccelli/sampleGitHubAction4translate/tree/main/docs/it/test-webodm-docs1

most of it is AI generated.
I started testing it few minute ago.
Is AI generated a problem?

Only if it takes me longer to review it than it took AI to generate it 🙂

let me test more please....

it is just a POC not a prototype.....

Try and see how it handles difficult formatting cases, e.g. how does it handle markdown tables? Or image tags?

all the webod docs...

Or how does it handle code blocks

references where the original paper's title is in English

I need to add markdown recognition and a smart way for formatting case.

yep is a must to add in every page.

But importantly, LibreTranslate/Google Translate might not be the right tool for the job

these systems don't have sufficient context awareness and get confused

LLMs might be better suited for these kind of tasks

Yes, probably, but we'd need a freely accessible and free installation.
As a model, I like Mistral for European languages; it works quite well.

Personally what has worked for me, is a well written AGENTS.md (I will post it soon) and Gemini Pro, no need for ad-hoc rules like regexes and line splitting, just tell it how to translate and it generates a decent draft.

https://github.com/WebODM/Docs/pull/6 rough draft, needs to be carefully reviewed. If you notice a lot of problems it might be worth to start over rather than fixing all the problems. The AGENTS.md is hand written by me.

Wow.
I'll start.

My goal is minum one file per day.
I'll try to schedule 30' every day for it