Basically, when you send a message, you have to specify a symbol, that corresponds to a method name, and parameters. The protocol specifies the "routing", like if there is no method that corresponds to this name, then send (forward) the very same message to your superior (exactly like they do in army, which is an implementation of the actor model).
If supers doesn't know what to do with the message, then there is panic - an exception.
The protocol also specifies the rules of shadowing of symbols, such that a method could be redefined (shadowed) within a closure, so it "overrides" a method with the same name from its ancestors.
Object methods (inner closures) could send messages to classes (invoking class methods) and then doing something with a returned value. Usually, there is some syntactic sugar for that.
In short, inheritance is a set of rules, and these rules could be implemented using various mechanism, like explicit message-passing (a-la Erlang) or implicitly, using procedures calls.
Brian Harvey in classic Berkeley CS61A explained these ideas much better than me.