Positional Grammars: a Formalism for LR-like Parsing of Visual Languages