John J. McCarthy Similar phonological processes can be governed by different constraints. Davis (1995) claims that the effect of such process-specific constraints cannot be obtained in Optimality Theory, exemplifying this point with material from harmony in Palestinian Arabic. On the contrary, I show that process-specific constraints are a natural and expected result of constraint ranking, which is the fundamental idea of Optimality Theory. Furthermore, Optimality Theory makes a novel and restrictive prediction, the subset criterion, about co-existent process-specific constraints within a single grammar. This prediction derives empirical support from a reanalysis of the Palestinian harmony systems.
Davis also presents evidence that epenthetic segments have featural specifications, claiming that Optimality Theory says they are featureless. This is incorrect; Optimality Theory is a model of constraint interaction, not of the representation of epenthetic segments.
Keywords: Arabic dialects, constraints, epenthesis, harmony, Optimality Theory
Research in phonology during the last two decades has securely established that otherwise general processes may be blocked by output constraints. For example, the Obligatory Contour Principle (OCP — Leben 1973, Goldsmith 1976), which prohibits adjacent identical elements, is known to block vowel deletion between identical consonants (McCarthy 1986) or to block high-tone spreading onto a syllable that is adjacent to a high tone (Myers 1987).
In the context of his analysis of tongue-root harmony in Palestinian Arabic, Davis (1995) presents evidence that these blocking effects are process-specific: the same basic process in different grammars or similar processes in the same grammar can be governed by different constraints or by no constraint at all. Concretely, leftward tone spreading might be blocked by the OCP and rightward tone spreading might ignore it, in the same grammar, on a language-particular basis. To express this observation, Davis proposes, following Archangeli and Pulleyblank 1994, that phonological processes are formulated as language-particular rules with a slot available to record any constraints that limit their applicability. That is to say, a constraint that blocks a process is included as a language-particular codicil in the process’s formulation (see () below for an example).
At the conclusion of his article, Davis goes on to say that process-specific constraints, in the sense just described, present a challenge to Optimality Theory (OT — Prince and Smolensky 1991, 1993). OT does not recognize parametric rules with constraints embedded in them. Rather, OT constructs grammars from language-particular rankings of universal constraints. This difference leads Davis to raise the following rhetorical question (p. 495): “how can the effect of process-specific constraints be produced within Optimality Theory just by constraint ranking, without processes or derivations?” That is, how does OT obtain the effect of process-specific constraints without rules in which to embed the constraints?
The answer to this question is basic to OT, and it is the topic of this article. Reanalysis of the Palestinian Arabic material (in sections 2 and 4) illustrates some important Optimality-Theoretic modes of explanation. I will show that the conception of constraint interaction inherent to OT is superior to the parametric rule-based theory Davis advocates, yielding a more explanatory account of the very material he discusses. This discussion leads to a novel and restrictive claim about the nature of process-specificity under OT (section 3), adopting a suggestion by Alan Prince. Finally, in section 5, I address an unrelated point Davis raises against OT, involving supposed representational commitments in the case of epenthesis.
In the Southern Palestinian dialect described by Davis, there is bidirectional harmony of the phonological property known traditionally as “emphasis,” phonetically a kind of uvularization. He identifies this property with the distinctive feature RTR (retracted tongue-root), which is present underlyingly on certain consonants.1 Leftward harmony of RTR is unlimited within the word (a), while rightward harmony is blocked by any of the high front segments iyšj (b, c). (As a transcriptional convenience, I will use underscoring to indicate the extent of the surface RTR span and I will capitalize the consonant that underlyingly bears the RTR feature. The vowel written as a is phonetically [æ], except in an RTR span, when it is approximately [ ]. The data are taken directly from Davis’s article, which may be consulted for the glosses.)
The forms in (a) show that leftward harmony of RTR is unimpeded even by high, front segments like i (tamšiiTa). The forms in (b) show that there is rightward harmony as well, but the examples in (c) prove that high front segments block rightward harmony. The blockers are i (Tiinak), y (Sayyaad), š( aTšaan), and j ( ajjaat).
According to Davis, the high front segments block rightward harmony because a process-specific constraint prevents them from linking to RTR; since they cannot be skipped over either, they put a stop to harmony. The responsible process-specific constraint can be called RTR/Hi&Fr; it is the conjunction of the constraints RTR/Hi ( *[high, RTR]) and RTR/Fr ( *[front, RTR]) (after Archangeli and Pulleyblank 1994). The combined constraint RTR/Hi&Fr asserts that [RTR] segments cannot be both high and front: