Parallel interpolation, splitting, and relevance in belief change