With Catalan, we now have another special case, similar to Turkish, of specific discretionary handling in core/hyphenator-liang.lua

I would suggest we discuss moving those in some dedicated language-specific classes for handling post-segmentation exceptions. Other noteworthy points

• We should likely cache those discretionary nodes (here below and for Turkish as well), instead of rebuilding them each time they are needed...
• Specific files for these hacks would need to use a language-based filename rather that merely comparing the language string. When supporting qualified BCP47 languages, we may want to distinguish ca (canonical Catalan = ca-ES) from ca-FR (as used in the French Pyrenees)...

Originally posted by @Omikhleia in #1964 (comment)