This talk describes a revision to Optimality Theory (Prince and Smolensky 1993/2004) and offers evidence in support of it. In the revised theory, candidates consist of chains of forms that somewhat resemble the derivations of rule-based phonology. For example, the mapping /ktub/ -> [ʔuktub] in Arabic is accomplished via the candidate chain <ktub, uktub, ʔuktub>.
 



Two main conditions on chain well-formedness are posited: gradualness (one unfaithful mapping is added at each link in the chain) and harmonic improvement (each link in the chain is more harmonic than its predecessor, according to the constraint hierarchy of the language in question). Under the gradualness requirement, for example, <ktub, ʔuktub> is not a valid chain, and under the harmonic improvement requirement, <ktub, ʔktub, ʔuktub> is not a valid chain.
 



Three main results of this approach will be described in this talk: candidate generation and evaluation involve only finite sets; multiple unfaithful mappings cannot be recruited in support of some local markedness goal; and phonological opacity is an effect of constraints on the order of unfaithful mappings in a chain.