Time Group types in Mandarin syllable annotations

We describe an inductive procedure for automatically parsing syllable sequences in speech annotations into Time Groups (TGs) based on deceleration and acceleration measures, rather than predefined units from the prosodic hierarchy. The effect of different minimal duration difference thresholds (MDDTs) on the size and relation of these TGs to different grammatical units, e.g. `word', `phrase', `sentence', is examined. The relative syllable isochrony in TGs is also determined with the nPVI measure. We found several non-trivial effects, demonstrating the plausibility of the methodology: (a) a stepwise, non-continuous relation between MDDT increases and number of induced TGs; (b) correspondence of induced TGs to sizes and ranks of grammatical units; (c) better correspondence with shorter phrasal units in the acceleration condition and longer discourse units in the deceleration condition; (d) comparability of relative isochrony using syllabic nPVI with the vocalic nPVI results from previous studies.