Practical Finite State Optimality Theory
Dale Gerdemann, Mans Hulden · Finite-State Methods and Natural Language Processing · 2012
Previous work for encoding Optimality Theory grammars as finite-state transducers has included two prominent approaches: the socalled ‘counting’ method where constraint violations are counted and filtered out to some set limit of approximability in a finite-state system, and the ‘matching’ method, where constraint violations in alternative strings are matched through violation alignment in order to remove suboptimal candidates. In this paper we extend the matching approach to show how not only markedness constraints, but also faithfulness constraints and the interaction of the two types of constraints can be captured by the matching method. This often produces exact and small FST representations for OT grammars which we illustrate with two practical example grammars. We also provide a new proof of nonregularity of simple OT grammars.