Tiers in articulatory phonology, with some implications for casual speech

Abstract

Introduction We have recently begun a research program with the goal of providing explicit, formal representations of articulatory organization appropriate for use as phonological representations (Browman and Goldstein 1986; Goldstein and Browman 1986). The basic assumption underlying this research program is that much phonological organization arises from constraints imposed by physical systems. This is of course a common assumption with respect to the elements – features – used in phonological description; it is not such a common assumption, at least in recent years, with respect to the organization of phonological structures. In our view, phonological structure is an interaction of acoustic, articulatory, and other (e.g. psychological and/or purely linguistic) organizations. We are focusing on articulatory organization because we believe that the inherently multidimensional nature of articulation can explain a number of phonological phenomena, particularly those that involve overlapping articulatory gestures. Thus, we represent linguistic structures in terms of coordinated articulatory movements, called gestures , that are themselves organized into a gestural score that resembles an autosegmental representation. In order to provide an explicit and testable formulation of these structures, we are developing a computational model in conjunction with our colleagues Elliot Saltzman and Phil Rubin at Haskins Laboratories (Browman, Goldstein, Kelso, Rubin and Saltzman 1984; Browman, Goldstein, Saltzman, and Smith 1986). Figure 19.1 displays a schematic outline of this model, which generates speech from symbolic input. As can be seen from the number of submodels in the figure, gestures are relatively abstract. Even articulatory trajectories are one step more abstract than the output speech signal – they serve as input to the vocal tract model (Rubin, Baer, and Mermelstein 1981), which generates an acoustic signal.