“Look at the shark”: Evaluation of student- and actress-produced standardised sentences of infant- and foreigner-directed speech

Standardised sentence production is routinely used in speech research to avoid content variability typical of natural speech production. However, the validity of such standardised material is not well understood. Here, we evaluated the use of standardised sentences by comparing them to two existing, non-standardised datasets of simulated free and natural speech (the latter produced by mothers in real interactions). Standardised sentences and simulated free speech were produced by students and actresses without an interaction partner. Each dataset comprised recordings of infant- (IDS), foreigner- (FDS) and adult-directed (ADS) speech, which were analysed for mean F0, vowel duration and hyperarticulation. Whilst students’ mean F0 pattern in standardised speech was closer to the natural speech than their previous ‘simulated free speech’, no difference in vowel hyperarticulation and duration patterns was found for students’ standardised sentences between the three speech styles. Actresses’ F0, vowel duration and hyperarticulation patterns in standardised speech were similar to the natural speech, and a part improvement on their ‘simulated free speech’. These results suggest that successful reproduction of some acoustic measures (e.g., F0) can be achieved with standardised content regardless of the type of speaker, whereas the production of other acoustic measures (e.g., hyperarticulation) are context- and speaker-dependent.

Research highlights► Successful imitation of natural speech modifications depends on research paradigm. ► Standardised sentences yield more natural results than free speech in simulated tasks. ► Successful imitation of natural speech modifications also depends on type of speaker. ► Untrained and uninstructed speakers can approximate some natural speech modifications. ► Trained and uninstructed speakers can approximate most natural speech modifications.