Neural representation of action sequences: how far can a simple snippet-matching model take us?