Why a propositional single-process model of associative learning deserves to be defended