Presentation

Cosimo Iaia^1,2, Mirko Grimaldi², Alessandro Tavano¹; ¹Max Planck Institute for Empirical Aesthetics, ²University of Salento

Speech rhythms across languages appear to converge on similar low frequency (LF) modulations of the speech carrier, suggesting a remarkable degree of regularity. It is unclear the extent to which such regularity reflects only low-level speech units, such as syllabic series, or projects to higher-order syntactic chunking. We have mapped variability in unit duration across four speech levels of increasing complexity - phonemes, syllables, words and sentences -, as well as four types of syntactic constituents - Noun phrases, Verb phrases, Adverbial phrases, and Clauses. Twenty-three subjects (5 male, mean age = 23.3, std ± 3.5) took part in a behavioral and EEG study. They were asked to listen to the first chapter of two audiobooks in the Italian language, read by a voice actor. Stimuli were presented through two loudspeakers at a comfortable volume (70 dB). Each chapter was roughly 9 min long, segmented into 10 trials. Linguistic levels in the acoustic signal were annotated manually at the phoneme, syllable, word, syntactic phrase, and sentence time scales using PRAAT software. Syntactic constituency analysis was performed with Stanza, an nlp library in Python. The duration of each temporal unit was extracted, and variance was compared across levels using the coefficient of variation and individual nonlinear fits. Preliminary analyses show significant differences in variance between the phonemic/syllabic levels and the word/sentence levels with larger variance for the latter. While durations at phonemic and syllabic levels were highly correlated, no other significant correlation was found. Importantly, the larger variance for word/sentence levels was due to a bimodal distribution of duration estimates, relative to the unimodal profile of phonemic/syllabic levels. Similarly, NPs showed the largest variance and a bimodal distribution, while the remaining syntactic units were unimodally distributed. We then tested whether pairwise phase consistency in neural data reflected behavioral variability, displaying one or two peaks of activity, in correspondence to the temporal profiles.

Topic Areas: Speech Perception, Syntax