I feel like there should have been a better solution to RTL/LTR annotation. Stuff like this makes me wish that there would be some well-defined subset of Unicode which would be less pita to handle.