On the target of phonetic convergence: Acoustic and linguistic aspects of pitch accent imitation

Studies in phonetic convergence have shown that speakers imitate phonetic patterns to which they are exposed (e.g., Pardo et al., 2013; Nielsen, 2011). However, it is less clear what aspects of the speech signal speakers are responding to when they change their speech behavior and to what extent: specific acoustic values, linguistically-interpreted targets, or some combination. To explore this issue, we conducted two imitation experiments, one in which participants were exposed to a linguistically-salient manipulated pitch accent realization, and one in which participants were exposed to a linguistically-less-salient overall pitch difference. Phonetic convergence might target acoustic values of the speech signal, leading participants to converge acoustically toward the low f0 and fast speech rate of the model taker. On the other hand, phonetic imitation might target linguistic patterns in the stimuli, leading participants to imitate the enhanced pitch accent in f0 and duration, even if it would result in acoustic divergence.

Our results show that it is both: participants converged to the model talker in linguistically-interpreted targets (pitch accent type, relative rise of the peak f0 on L + H*, and relative duration of the target phrase) and acoustic values (carrier phrase f0 and duration). However, our data indicated that higher-level linguistically-interpreted convergence may be manifested more robustly than low-level convergence. We also observed an asymmetry in convergence robustness between f0 and duration, as well as individual difference in the patterns of convergence, which suggest that there are constraints on convergence related to structural properties in the stimuli as well as speakers’ characteristics in perception and production.