Constructive optimality theoretic syntax

This paper documents efforts at developing a constructive optimality theory. Several years ago, two 3rd year AI-students, Chris Regenboog and Jacco van Willegen, came up with an idea to implement optimality theory, in a special case —a system partially describing Dutch word order—. Me and some students have been trying to work out their ideas further in the period since, but many issues still remain open. This paper reports on a reconstruction of Bresnan's Optimal Syntax and a new integrated approach to word order, in three Germanic languages: Dutch, English and German, as a direct illustration of the concept. The idea is the following: interpret every constraint as a procedure that tries to add more structure to an underspecified specification and call the procedures in the order of the strength of the constraints they implement. The main consequence is that the notion of error disappears altogether from optimality theory. Errors are failed attempts to overwrite structure that is already filled in. The process is monotone increase of information and fully deterministic. There should be enough constraints to make sure that the process delivers a fully instantiated ground structure. Two other things that disappear from sight are GEN and various economy constraints: what can be constructed gives an upper bound on what should be in GEN, economy violations are just not generated.