Investigating speech articulatory control through dynamical model fitting

Dynamical models of speech articulation link the continuous trajectories of articulators with stable parameters of phonological knowledge (e.g., Browman & Goldstein, 1989). We examined the fit of a recently proposed dynamical model (Stern & Shaw, 2024) to electromagnetic articulography (EMA) recordings of twelve English speakers and twelve Mandarin speakers. While fits to (labial) consonant constrictions were excellent (mean R 2 = .97), fits to consonant release movements and vowel constrictions were slightly worse (mean R 2 = .91 and.89, respectively). We explore two implications of this result. (1) Consonant release and vowel constriction target achievement may be coordinated in time, unlike consonant constriction target achievement (Kramer et al., 2023). The addition of a coupling term may improve model fit, constituting evidence for target-based gestural coordination in speech (Turk & Shattuck-Hufnagel, 2020). (2) Vowel constriction movements may be governed by two separate tasks, one controlling constriction location and one controlling constriction degree (Saltzman & Munhall, 1989). Model fit may be improved by separating vowel constriction dynamics into two systems, shedding light on the dimensionality of articulatory control.