On the configuration of height features

Abstract

While it has long been known that certain groups of features pattern together in rules, this observation has only recently been directly encoded in segment structure. In Sagey (1986), features are grouped according to the articulator with which they are executed. Due to the success of this model, much subsequent work on feature geometry has further pursued the approach. In spite of this, I argue in this dissertation that height features are located in a component separate from the remaining place features. Part I focusses on the configuration of height features and on related issues in place geometry. Part II deals with certain learnability issues which arise in connection with the proposals set forth in Part I. In Part I, I motivate a geometry in which: (a) all features are monovalent; (b) height features define their own constituent 'Vocalic' which is, along with Place, directly dominated by Supralaryngeal; (c) height features ( (open) and (low)) are in a subset relation under Vocalic; (d) (atr) patterns as a height feature and occupies the same position in the geometry as (low); (e) (atr) and (rtr) are formally independent of one another and are located in different parts of the geometry, under Vocalic and Place respectively; and (f) a new feature (front), doubly dependent on both Coronal and Dorsal, replaces the traditional (-anterior) and (-back). A reappearing theme throughout Part I is that a language's inventory drives its feature selection which in turn drives its phonological processes. This is the basic premise of Contrastive Specification, and in direct contrast to Radical Underspecification and Combinatorial Specification where feature selection is instead process-driven. In Part II, this approach is argued to have positive consequences for a theory of acquisition. Specifically, if feature selection is inventory-driven, determined on the basis of minimal contrasts among segments in a language, the child need initially know nothing about the phonological processes operative in his/her language. The result is a simple learning algorithm: segment classes are compared pairwise to determine feature selection. The child then uses the inventory to 'bootstrap' into the phonology (cf. Pinker 1984). (Copies available exclusively from Micrographics Department, Doheny Library, USC, Los Angeles, CA 90089-0182.)