On the Problem of Infinity and Gen-Eval in Optimality Theory
Paul Hagstrom, MIT
November 1993

In the framework of Optimality Theory proposed in Prince & Smolensky
(1993), an infinitely large set of candidates is filtered by ranked
constraints, simultaneously, to determine the optimal candidate.
Although this appears to be computationally impossible given current
notions of computability, I suggest here looking at the problem
"backwards."  Rather than viewing the constraints as ordered filters,
we might instead be able to use them as weighted constructors of
the optimal candidate.  Thus, the finite number of "constructors"
(the analog of OT constraints from this vantage point) becomes the
upper limit of the computation and not the possibly infinite number
of candidate representations.  I discuss examples of "constructor
analogs" to Onset, HNuc, Fill, Parse, No-Coda, and Coda-Cond as
set forth in the OT literature, and show how they might achieve the
correct results given a method of "finalizing" most optimal
structures to the exclusion of incompatible structures.