Articulatory control parameters underlying speech rate variation

Speech rate variation in articulation was observed using electromagnetic articulographic recordings of American English speakers. Each participant produced the sentences “I will be pop/popped on Monday” 16 times at 11 distinct rates cued by a visual meter. Constriction and release gestures for each /p/ in /pɑp(t)/ were parsed using the lip aperture signal, and /ɑ/ gestures from a sensor on the tongue body. We found that some kinematic measures (duration) scaled monotonically with rate, while others (peak velocity) showed non-monotonic (U-shaped) patterns. A nonlinear dynamical model of articulation (Stern & Shaw, 2025) was fit to each movement. We found that each of the two control parameters of the model scaled monotonically with rate: r (controlling rapidity) increased at faster rates, and T (controlling spatial target) became less extreme (larger for constrictions, smaller for releases) at faster rates. The presence of the second coda consonant (in /pɑpt/) also increased T, though in speaker-idiosyncratic ways. Importantly, movements simulated from the model using the fitted parameters reproduced both the monotonic and non-monotonic kinematic patterns, providing evidence that the complex articulatory timing relations underlying speech rate variation can arise from monotonic parameter scaling of a nonlinear dynamical control system. [Work supported by NSF-BCS2141275.]