Determinism/non-determinism

Bill Turkel bill at HIVNET.UBC.CA
Tue Feb 13 19:01:33 UTC 1996


I am glad that I asked for clarification, because it seems that we
have a cluster of related ideas.

In terms of programming, 'non-determinism' can mean that the same
question can yield more than one answer, or that several elements
can be in a particular relation to a given element, or that a
number of possible steps are available at a choice point.

If we look at OT in terms of the 'answer', I don't think that there
is anything in the theory that says you have to have one and only
one optimal answer.  Think of the discussion of harmonic serialism
in Prince and Smolensky (1993), where particular iterations of
evaluation will return a collection of optimal candidates.  To say
that OT forces you to have exactly one optimal output is wrong.

If we look at OT in terms of the 'path' (to use Ellen Prince's term,
i.e., in terms of possible steps at a choice point) then OT does
not commit to any particular *way* of computing things.  Constraints
could be evaluated always in the same order, or in a random order, or
in parallel, or whatever.  It sounds like Guy was refering to the
'answer' interpretation, however.

Bill



More information about the Funknet mailing list