You could still use text for raw identifiers (which tends to be very convenient for debugging) but define language-specific alternate mnemonics as part of your

You could still use text for raw identifiers (which tends to be very convenient for debugging) but define language-specific alternate mnemonics as part of your language's equivalent of header files. These alternate mnemonics could even just be some specially formatted "doc comments", with the actual text substitution being done by a LSP-like system.