Consonant cluster simplification in Basque depends on the prosodic domain. Simplification across morpheme boundaries is obligatory while simplification across word boundaries is optional. Moreover, strident-affricate clusters inside words are simplified by deaffrication, while phrase-internally such a mapping is not observed. Depending on the variety, clusters may be avoided by simplification or vowel epenthesis. Previous OT-based analyses of Basque clusters posited that the aforementioned phenomena are triggered by an output requirement that stops must be followed by vowels (Côté 1999, 2000). The empirical argument in favor of this view is that stops and affricates before a pause are also avoided. Importantly, such avoidance is less likely phrase-finally than phrase-internally. In order to account for the aforementioned asymmetries, positional markedness constraints relativized to prosodic boundaries were invoked. The goal of this paper is to construct a uniform OT grammar that accounts for stop deletion, affricate spirantization, deaffrication, spirant coalescence, voiced stop lenition as well as voice assimilation. Two conspiracies are identified: against non-prevocalic stops and against adjacent stridents. It is argued that prosodic anchoring as the locus of the asymmetries in the application of the listed processes is superior to the positional markedness approach. In a probabilistic grammar, the former yields satisfactory results while the latter generates unattested outputs. From the theoretical side, the anchoring model posits one driver for each conspiracy. The positional markedness model, on the other hand, has no formal device of expressing the conspiracies since a different markedness constraint triggers simplification word-internally, phrase-internally and phrase-finally.
J. DUNIN-BORKOWSKI (Fri,) studied this question.