The current study explores whether natural constraints are sufficient for predicting typology in the case of Post-Nasal Devoicing (PND). Previous proposals have suggested that PND is a counterexample to theories that require constraints to be natural (Hyman 2001), since it's attested and appears to require the unnatural constraint *ND. However, I show that a Duke-of-York derivation (Pullum 1975, McCarthy 2003) in Stratal OT (Booij 1996, Kiparsky 2000) can represent PND using only natural constraints (following multiple, similar diachronic proposals: Dickens 1984, Hyman 2001:163, Beguš submitted). I then test the learnability of this representation using the Maximum Entropy Stratal Learner from Nazarov and Pater (to appear). While this representation of PND is reliably learnable, giving the grammar the ability to use Duke-of-York derivations also predicts unattested patterns. An example of one of these is Word-Final Voicing, which can also be represented using Stratal OT and which is also shown to be learnable. This seems to present a problem for natural constraints, since allowing the representation of an unnatural, attested pattern also allows an unnatural, unattested one.
Proceedings of the 35th West Coast Conference on Formal Linguistics
edited by Wm. G. Bennett, Lindsay Hracs, and Dennis Ryan Storoshenko
Table of contents