I do agree with you in the general case. But honestly, guv, there are extenuating circumstances here. Its a very low user admin interface, dealing with content that is basically english. The existing interface groups by english alphabetic ranges and would have silently dropped items that did not fit. I did suggest alternatives to the UI team, but they prefered to keep the existing interface. Regardless...

I would like to hear better approaches, and minimize surprise to users. The problem splits into (at least) two parts: a) sorting a list that may contain non-english words. And b) grouping said list into groups that map somewhat to english character ranges. We can assume that the target audience is English speaking. Both of these may be basically impossible to do correctly for all cases. So whats the least surprising behaviour?