On representational abstraction