If you've looked over my reviews you'll know by now that I am suspicious of optimality theory as a general rule, but this is probably the least foolish version. Boersma and Hayes's Gradual Learning Algorithm allows the ranking of constraints according to strictness and probabilistic data, on a continuous ranking scale: continuous (progressive) scale of constraint strictness. Continuous scale becomes more meaningful when differences in distance have observable consequences—when in:
A_________________B_____C
A is more important than B by a greater margin than B is more important than C by. B/H suggest that at “EVAL time”, the position of each constraint is temporarily perturbed by a “selection point”, a random positive or negative value, and then goes back to its “ranking value”. Predictive power, but not as a model for what actually goes on in our heads, thank'ee.
( )