A common challenge for our project is when starting a new language, translators have to translate 15000 entries with very little context.
This is especially a challenge for rich languages such as Arabic or Chinese which have several closeley related syonomyns which should be used depending on the context.
The common suggestion is to provide a screenshot in Transifex and code comments e.g. Translators: read is for reading from hard drive. Both require a lot of effort and screenshots gets dated quickly.
I think with the current LLM technology it would be possible to provide a reasonable summary about the button or translation text by having the LLM read the code repository and provides contextual information that’s more useful that a path and a line number (search/views.py:52) → (This is the placeholder for the searchbar, it appears when the text is empty).