Towards a Computational Articulatory Model of Spanish Phonology

Many aspects of Spanish phonology remain poorly described within feature-based frameworks, in part because it is not well understood to what extent phonological features are grounded in the phonetic domain. In this paper, we introduce a computational model of Spanish phonology based on articulatory primitives (Browman & Goldstein 1992) that is currently being developed as an extension of the Task Dynamics Application (TaDA: Saltzman & Munhall 1989; Nam et al. 2004). Under this model, lexical items are represented as molecules of coordinated gestures. Hypotheses about phonological representations are tested by comparing the results of articulatory and acoustic simulations against phonetic data acquired from Spanish speakers producing the same utterances. Two phenomena of particular interest in Spanish phonological theory that have been examined using this methodology are intervocalic stop lenition and liquid neutralization. A wide range of realizations are observed in the production of the Spanish stop series /b-d-g / and /p-t-k/. This allophony is influenced by factors including speech rate, stress, context and individual speaker variation (Cole et al. 1999; Soler & Romero 1999; Ortega-Llebaria 2004), and has proven difficult to describe in terms of features. Under the articulatory framework propose