Absolute Ungrammaticality

Abstract Given the principles of optimality theory, there should be a grammatical output for every input. After all, of the possible outputs corresponding to a particular input, the one that is the most harmonic with respect to a set of ranked universal constraints is grammatical by definition. Nonetheless, it appears to be a rather obvious fact that certain constructions have no realization at all in some, or even all, languages. How is this possible, then? Given the structure of the grammar in optimality theory, several types of solu tions have been proposed for this problem. An Optimality-theoretic grammar consists of two components: a device GEN that generates all possible structural realizations of a particular input and a function EVAL that selects out of this candidate set the structure that is optimal. One possibility therefore is to argue that there are restrictions on GEN such that certain constructions never enter into competition. It is also possible that two candidates score equally on all constraints but one. In that case, the candidate that wins out on this particular constraint will block the other candidate under any constraint ranking. In the terminology of Prince and Smolensky 1993, this is expressed by saying that the latter is harmonically bound by the former.