Non-ASCII characters in "browse by metadata" lists are sorted according to UTF-8 code (I think?), not the equivalent/normalised corresponding ASCII value. This means a term like Āna is sorted after Aziz, instead of before Amanda.
The bi_dis_*_filter fields used in Discovery browse are segmented like facets: "value|||Value" - where the first segment is lowercased. Since this is the value used for searching and sorting (but not displaying the value), we should normalise the first segment.