|Title:||What Constraints Should OT Allow?|
|Comment:||20 pp. Talk handout from LSA 1997. (9 Jan 1997)|
|Abstract:|| What Constraints Should OT Allow?
Jason Eisner - University of Pennsylvania
January 9, 1997 (corrected version)
Handout for talk at LSA Annual Meeting, Chicago, 1/4/97.
Optimality Theory (OT) has shown itself to be an elegant framework for
phonological description. Two important questions remain to be
settled, however: What constraints are allowed? And what kind of
representations do they constrain? Formalizing what OT can and cannot
say is part of stating UG.
This talk proposes an approach to constraining OT, called "primitive
Optimality Theory" (OTP). Most constraints given in the literature
can be reformulated (not always obviously) as coming from one of two
simple, local families of ``primitive constraints'':
Alignment (licensing): Each a temporally overlaps some b.
(If not, it incurs one violation mark.)
Disalignment (clash): Each a temporally overlaps no b.
(Each overlap incurs one violation mark.)
Here, a and b may be constituents, edges of constituents, or
restricted kinds of conjunctive or disjunctive configurations.
We formalize these families and the representations that they
constrain. As in Optimal Domains Theory, neither the constraints nor
the representations use association lines. The constraints control
only the relative timing of articulatory gestures, and other
phonological or morphological constituents, along a continuous
A list of hundreds of constraints drawn from the literature is
presented, showing how every degree of freedom of OTP is exploited in
each of several areas: features, prosody, feature-prosody interaction,
input-output relationships, and morphophonology. To show that the
primitive constraints are not merely necessary, but also close to
sufficient, we also discuss how to handle a few apparently difficult
cases of non-local phenomena.