Thai speakers time lexical tones to supralaryngeal articulatory events

• Acoustic and EMA study of tone production in Thai. • Onset-to-onset tonal lags more stable and less variable than target-to-target lags. • Articulatory lags more stable, less variable, and more informative than acoustic lags. • Tones are timed onset-to-onset to vocalic gestures. • A stochastic model of tonal timing is developed. What do speakers do when they produce tones? Do they aim for the synchronization of f0 targets with segmental acoustic events, or do they execute a routine in which f0 changes and oral articulations are precisely coordinated? This paper explores these questions in Thai using acoustic and electromagnetic articulography data from eight speakers. Drawing on analyses of variability, stability, and informativity, our findings indicate that the timing of the onsets of tonal and oral articulatory gestures is generally more stable than the timing of tonal and oral targets, both in articulatory and acoustic measurements. When comparing the two modalities directly, we found that the lag between tonal onset and vocalic gesture onset exhibits the lowest variability and the highest mutual information among a large set of timing measures. Additionally, only articulatory lags remain stable under rate and context perturbations. To explain these findings, we propose that Thai tones are timed onset-to-onset with vocalic gestures and develop a model that formally implements this proposal. This model also accounts for otherwise puzzling acoustic patterns, such as a negative lag between tonal onset and acoustic syllable boundaries at slower speech rates. Further temporal patterns, such as surface non-zero time-locking rather than perfect synchrony of events, are also clarified. In sum, this work advances our understanding of tonal timing in Thai and outlines its implications for more general theories of phonology and speech production.