Abstract

Although early ideas about the isochronous organisation of speech units have consistently been shown to lack empirical support, the proposition that speech is rhythmically organised remains prevalent. The concept of rhythm class is frequently invoked in typological descriptions, and recent studies of the entrainment of neural oscillators to the amplitude envelope of speech often assume a quasi-periodic signal. Typically, however, speech encodes linguistically important distinctions both through spectral and temporal variation, building robust redundancy into the signal and giving rise to a strong tendency to “antirhythm” (cf., Nolan & Jeon, 2014; Pointon, 1980; White, 2014). Presenting results from studies of language discrimination and artificial language learning, I argue that sensitivity to speech rate is evidence of the importance of prediction for listeners, with deviations from temporal expectation interpreted as linguistically meaningful. Expectation arises not only from the signal, however, but also from linguistic context and listener-specific factors such as social experience. Thus, speakers may modulate temporal regularity to trade off their audience’s current need for predictability against a tendency to maximise the informational efficiency of the signal.