If you're going to put a lot of energy into binding syntax to sound to semantics, you only want to do it once (per language, at least.)