Speaker: Ezer Rasin (MIT) Title: The stress-encapsulation universal and phonological modularity Date/Time: Monday, September 12, 5:00-6:30pm Location: 32-D831
Cross-linguistically, the distribution of segmental features is often conditioned on the position of stress. In American English, for example, [t] is flapped between a preceding stressed vowel and a following unstressed vowel (políDical vs. politícian), voiceless stops are aspirated at the onset of a stressed syllable (opphóse vs. opposítion), stressless vowels undergo reduction (át@m vs. @tó mic), and [h] is deleted before an unstressed, non-initial vowel (vé[h]icle vs. vehícular). As noted by Blumenfeld (2006), stress-segmental interactions in the other direction are almost non-existent: stress is sensitive to supra-segmental features such as syllable structure and tone, but – to the exclusion of sonority – it is never sensitive to any segmental features (such as voicing, continuancy, place of articulation, and so on). My main claim is that reported sonority-sensitive stress patterns do not require direct reference to sonority. The claim will be based on a review of the sonority-driven stress literature and a re-evaluation of some of the reported cases. The result is that Blumenfeld’s list of universal asymmetries between stress and segmental features becomes a generalization over all features: the distribution of stress is never conditioned on segmental features. I refer to this result as the “Stress-encapsulation Universal”. The stress-encapsulation universal is surprising under existing theories of phonology: rule-based theories of stress (e.g., Halle and Vergnaud, 1987) have used rules that make direct reference to segment quality, and even if reference to segment quality is avoided, the fact that stress rules would consistently ignore the same information in their input (segmental features) would be left as an accident. In Optimality Theory (Prince and Smolensky, 1993), stress and segmental processes are computed in parallel: output markedness constraints that trigger stress-sensitive segmental processes are symmetric and may be used to (undesirably) trigger quality-sensitive stress. Given such constraints, standard Optimality Theory has no general way of banning quality-sensitive stress processes. I will show that the stress-encapsulation universal can be derived in a modular architecture of phonology where a stress-computation module is encapsulated from the rest of the system. In this modular architecture, the input to the stress module excludes representations of segmental features, and outside of the stress module, stress representations cannot be changed. I will propose a concrete theory of the interface to the stress module within a serial rule-based framework and discuss its predictions regarding indirect effects of segmental features on the position of stress, including vowel invisibility to stress and effects on stress through syllable structure. Finally, I will evaluate alternative explanations for the universal, including non-modular phonological explanations (such as fixed constraint rankings within OT) and explanations that attribute the universal to extra-phonological factors.