Context: I maintain multilingual automation workflows across several Microsoft OSS repositories and started reaching out to DevTool maintainers to understand how they approach documentation localization.
Hi from Argentina! We use a lot of loaned words while programming and related stuff. I usually prefer to read the English version, because in the Spanish version I have to untranslate. It's more common in es-es localized docs, they like to translate every single word.
A translated version may be useful for first year students, but somewhere in the middle of the degree every book is in English, the variables must be named in English, the API must use English names, and when you talk you use 95%Spanish+5%English.
The only exception I make is for some real world objects. I work in the first year of the University of Buenos Aires. I made a custom program that must handle the 10 "sedes" where the classrooms are. How do I translate "sede"? It's usually a "building", but sometimes a "sede" has two "buildings", or half a "building". Do I call them "campus", we used to have to "sedes" that were nearby "buildings" inside the same big park that everyone would call "campus". Now each "sede" is like 5 miles away form the nearest one, so they are pretty discrete. So I made an exception and the variable names in the code say "sede".
Hi HN,I help maintain translations for several Microsoft open-source education repositories. The hardest part wasn’t translation quality. It was keeping documentation in sync whenever the English version changed.So I built Localizeflow, a GitHub App that detects documentation changes and automatically opens translation pull requests. No config files, no Api keys.
I've been running it across 14 Microsoft OSS repositories for the past few months. It turned a weekly maintenance task into something that just runs in the background.
Would love feedback from maintainers who've dealt with documentation localization at scale.
This is a great project. My father is an English teacher, and I recently introduced him to generating exam papers using AI. He was impressed, but the biggest hurdle was the fragmented workflow, jumping between LLM chats, document editors, and sharing tools.
These were some patterns that came up repeatedly.