Is OT deterministic?

[Wataru Nakamura]

Hello again!
I remember (I was in his talk) he contrasted Variable Rule Model with
a version (since some OTers do not hesitate to appeal to graded
constraints) of OT with the following two crucial properties,
graded constraints and strict dominance (i.e. constraints should not
tie). (If I remember correctly) Guy argued that non-graded constraints
are special cases of graded constraints, but not the other way around.
I believe this is the reason he argued that OT is part of Variable Rule
Model (I think we really should ask him directly. I might have misquoted
him).
        As far as I know, there seem two ways to incorporate optionality
into OT:

 (1) Imcomplete ranking (I am not sure if this is the right term. If I
     am wrong, someboy correct me) (Hammond 1994), in which a constraint
     set is such that it produces more than one output.

 (2) Tied constraints: make constraints unordered (Pesetsky 1995, Brohier
     1995).

I am not sure how far (1) can go, but (2) obviously runs counter to
strict dominance. If OT may be characterized by the two properties,
strict dominance and graded constraints and (1) and (2) are the only
ways to handle optionality in OT, it seems fair to say that OT is not
non-deterministic.
        One possible argument against introducing graded constraints is
that if you introduce graded constraints, it is difficult to capture
typological variation, which is supposed to be derived from re-ranking
of universal constraints. Notice, however, that this arguments does not
have any force unless you check all possible rankings of the constraint set
you have and find at least one real or 'plausible' (I do not know how to
judge plausibility, though) language for each possible ranking. And I
do not know any OT paper which go through the whole process, i.e, come
up with a set of constraints for describing a particular case, check
all possible ranking of the constraints, and find languages for each
ranking.
        If I said something wrong, stupid, or whatever, please correct
me.


Wataru Nakamura


On Tue, 13 Feb 1996, Bill Turkel wrote:

> From: "Ellen F. Prince" <ellen at CENTRAL.CIS.UPENN.EDU>
>
> > my understanding is that, given a set of constraints and
> > an ordering/weighting, there's only one possible outcome for a particular case
> > in ot. and that's how i understand 'deterministic' and that's what i would
> > guess guy had in mind.
>
> So, three observations.
>
> 1) The outcome can be a (possibly empty) *set* of optimal candidates.  This
>         seems to leave room for nondeterminism and optionality.
>
> 2) A particular constraint might be nondeterministic in action.  This relates
>         to Dan's point about compatibility with formal/functional models.
>
> 3) The way that the outcome is computed may involve nondeterminism.
>
> Thanks to everyone for clearing up Guy's claim.
>
> Bill
>