It uses plain ASCII characters, letter meaning [A-Za-z]. SGML, being 30 years old, obviously predates Unicode, and was contemporary with 7-bit character sets and bank-swapping concepts that grew into the horrible mess known as ISO-2022.

I don't know if XML extends this, allowing other characters in "identifier" -like tokens. C and C++ lists all legal letters and idiographs in all character sets permitted in identifiers; I don't know if Perl references a formal standard, but it also allows Unicode characters in symbol names. I good first appoximation is to use the Unicode "properties" to pick letters, letter-like symbols, etc.

—John