Quadratic Alignment Constraints and Finite State Optimality Theory
T. Biró · Repository of the Academy's Library (Library of the Hungarian Academy of Sciences) · 2003
Previous approaches to Finite State Optimality Theory have supposed that one can build transducers modelling the behaviour of the constraints, e.g. that would assign the correct number of violation marks to the candidates.A constraint is called linear if there is a linear function of the input string's length that is an upper bound on the number of violation marks assigned.Quadratic constraints can assign a number of marks quadratic in the input's length.We shall prove that only linear constraints can be realized as a finite state transducer.Some widely used alignment constraints, e.g. for stress assignment, are not linear.Interestingly, these constraints have also been criticized in recent phonological literature.