Because this rule includes RTR/Hi&Fr as a “target condition”, rightward spread is blocked whenever it encounters a high front segment (i.e., iyšj). But RTR/Hi&Fr is not included in the formulation of the leftward spread rule, which is not subject to any target condition. Thus, the constraint RTR/Hi&Fr is process-specific because it is built directly into the one process it controls. This is the nature of process-specificity in the rule-based parametric framework.
Elaborating on his argument against OT, Davis offers the following question about Southern Palestinian (p. 495): “how can Optimality Theory account for the fact that rightward emphasis spread, but not leftward emphasis spread, is subject to the grounded conditions RTR/Hi and RTR/Fr, under the view that these conditions are general constraints of the language and not process specific?” Undoubtedly, OT does not countenance complex language-particular rule schemata like (), with a parametric slot in which the process-specific constraints can reside. The challenge, then, is to see whether OT can obtain the process-specificity effect by other means.
In fact, this situation presents no challenge whatsoever to OT. Constraint ranking, the essential element of OT, straightforwardly leads to situations in which some processes are blocked by a constraint and others are not. To prove this point, I will outline the main elements of an OT analysis of the southern Palestinian material in ().
OT does not characterize processes in terms of operations like (). Rather, in OT the operational notion of a process roughly translates into a constraint ranking in which some structural constraint M crucially dominates some faithfulness constraint F: M >> F. The constraint M prohibits some kind of output structure, such as a coda, a non-branching foot, or a front round vowel; taken together, the various constraints like M constitute the universal theory of markedness. On the other hand, F is one of the family of constraints that demand faithfulness of surface forms to underlying forms. When F is obeyed, surface and underlying forms are identical in the F-relevant characteristic; when F is violated, they differ in that characteristic. The ranking M >> Fasserts that obedience to M is required even at the expense of violation of F, so unless some higher-ranking constraint vitiates the force of M, any potential M-violating form will be altered to an M-conforming one, even if this means unfaithfulness with respect to F. This unfaithfulness in the underlying surface map, required by constraint interaction under M >> F, is the approximate OT analogue to a “process” in operational theories.2
Turning from these abstract considerations to more concrete ones, we can apply the M >> F schema to RTR harmony in Palestinian Arabic. The relevant faithfulness constraint is Stay-ATR — assuming full specification, a segment which is underlyingly ATR must remain so.3(i) Stay-ATRorIdent(ATR)
If S1, S2, , and is ATR, then is ATR.
As defined, this constraint presupposes full specification of S1 and S2. An appropriate modification for underspecificational assumptions can be readily accommodated: replace “is ATR” with “is not RTR”, in both instances. It is crucially dominated by structural constraints that demand harmony of the feature RTR, whenever that feature is present in a form. Following proposals by Akinlabi (1994, 1995, to appear), Beckman (1994), Cole and Kisseberth (1994), Kirchner (1993), Pulleyblank (1993, 1994), and Ringen and Vago (1995, to appear), I will formulate these harmony constraints in terms of the Generalized Alignment model of McCarthy and Prince (1993b).4 The responsible high-ranking constraints in Southern Palestinian are these:
() Constraints on RTR Alignment
Align([RTR], Left, Word, Left)
“Any instance of [RTR] is aligned initially in Word.”
Align([RTR], Right, Word, Right)
“Any instance of [RTR] is aligned finally in Word.”
The M >> F ranking is proven by the constraint tableaux 1 and 2. (In these tableaux and others, I have reckoned violations of alignment and faithfulness in terms of feature-geometric root nodes, so aa and a both count as one, but nothing hinges on this detail, since only the degree of violation matters.) In tableau 1, satisfaction of RTR-Left is bought at the price of unfaithfulness to the ATR specifications of b, a, l, and a; the alternative candidate, with RTR-in-situ, fails because the alignment constraint is top-ranked. Tableau 2 shows the same thing, mutatis mutandis, for the constraint RTR-Right. Tableau 1
RTR-Left >> Stay-ATR
* ! ***
RTR-Right >> Stay-ATR
* ! ***
As these ranking arguments make clear, alignment takes precedence over faithfulness to the input, securely establishing the constraint ranking given in ():