Chapter 1: Optimality Theory: Constraint Interaction in Generative Grammar
Rules
There appear to be rules in phonology that operate thus:A → B / C _ D
i.e. feature A of a segment changes to features B when it appears between segments with features C and D
For example, in American English, /t/ → /ɾ/ between two vowels, the second of which is unstressed (think data - it sounds a bit like a [d] to the English ear.)
Optimality Theory: the basic components
In Optimality Theory, rather than changing the input with a specific rule, in a specific environment, there are instead constraints on the possible output:*CAD : the cluster of segments with features C, A and D in that order is forbidden
Multiple outputs are generated from the input by a function Gen (a fixed part of the Universal Grammar), and the output which is optimal is chosen (see below) to be pronounced - is the 'real' output.
"Gen contains information about representational primitives and their universally irrevocable relations". For exampple, the notions of syllable and onset, and their hierarchical relations: onsets are part of syllables, and not vice versa.
H-Eval is the function which evaluates harmony - decides which output is most harmonious, and therefore optimal - and it is the ways in which this function can vary which produce differences between languages.
Constraints are ranked, and outputs which violate higher ranked constraints are discarded first, until there is only one candidate remaining.
(If you are left with multiple candidates that all violate the same constraint, then the one with fewest violations 'wins'.)
Iterative or parallel?
Either Gen generates all candidates at the same time, and they are evaluated in parallel; or it generates some, which are evaluated, and the results fed back into Gen for another round.Prince and Smolensky - the instigators of OT - base their theory on parallel analyisis.
Difference from rule-based phonology
In rule-based phonology, e.g. phonotactic constraints are "language-particular statements of phonotactic truth".
In OT, constraints are universal, and an individual grammar consists of ranking these constraints, and choosing the highest-ranked constraint in event of conflict. Because they are ranked, constraints are regularly violated.