Since all four of these candidates only violate constraints from below the cut-off, it is predicted that all four are possible outputs.
Each group, such as (a)-(d), contains a set of candidates that compete with one another, holding main-stress location and foot type constant.
The scales in (4) are not constraints - they cannot interact with other constraints in evaluating candidates.
The correctness of the constraint ranking in (26) is further supported when more elaborate candidates are taken into account.
The appropriate competition is among the stressed syllables in different candidates.
The candidates for paradigm-uniformity constraints are full paradigms.
However, this approach does not extend to other cases such as [bblbntorel], where losing candidates such as *[bblnbtorel] have the same number of closed syllables.
Let us discuss some individual candidates to support this conclusion.
