I believe that the key to getting these characters encoded is establishing that there is a vital semantic importance to the character that is lost if it is stripped away. This is the grounds for the Mathematical Alphanumeric Symbols block.
Unfortunately, figures 1 and 2 from JTC1/SC2/WG2 N3915 actually provide a reason -against- encoding. The meaning of the diacritic in these two examples is that the transliterated letters were ligated in the original text. In this usage, the mark can span any arbitrary number of letters; indeed, figure 2 shows the mark in question spanning four letters. This makes it a much better candidate for use in higher-level markup than a set of combining marks. Figures 3 and 4 present a better case and show a stronger need for some combining triple diacritic. I notice that all seven examples between the two figures represent what would normally be two letters with a double diacritic, but some modifier symbol intervenes and stretches the tie to span three. However, proposing the triple diacritics used this way would require proof that the sequence of letters with the diacritic has some important difference from the same sequence of letters without, which N3915 fails to establish. In any event, I happen to know that there is in some phonetic transcription system an "sch" with breve below. It is used to represent [ʒ], which contrasts with the unmarked sch used to represent [ʃ]. This is a clear semantic distinction, and so the sch with breve below should be encoded in some fashion, either as a sequence of characters or some fully composed one. --Ben Scarborough
