Speaking rate, information density, and information rate in first-language and second-language speech

Using a corpus of multilingual recordings of a standard text (the North Wind and the Sun passage, NWS) in 11 languages, speaking rate (SR, syllables/second) and information density (ID, number of syllables for the NWS text) were examined in first-language (L1) and second-language (L2) speech. Replicating prior work, cross-language comparison of L1 speech showed a trade-off between SR and ID such that relatively low density languages (many syllables for the NWS text) tended to be produced at relatively fast rates, and vice versa. Furthermore, L2 English was characterized by both slower rate and lower ID than L1 English. That is, L2 English involved more syllables than L1 English for the same NWS text. A comparison of the number of acoustic syllables (i.e. amplitude peaks) with the number of orthographic syllables (i.e. dictionary-based syllable counts for the NWS text) indicated that L1 speech involved substantial syllable reduction (fewer acoustic than orthographic syllables) while L2 speech involved substantial syllable epenthesis (more acoustic than orthographic syllables). These findings suggest that L2 speech production involves temporal restructuring beyond increased segment, syllable and word durations, and that the resultant information rate (information bits transmitted/second) of L2 speech diverges substantially from that of L1 speech.