The Use of Domain-initial Strengthening in Segmentation of Continuous English Speech

Prosodic structure in English speech is signalled, in part, by stronger articulation of consonants at the onset of intonational phrases (IPs) than of consonants that are IP-medial. In two cross-modal priming experiments, American English listeners heard sentences and decided whether visual letter strings, presented during the sentences, were real words. We manipulated sentence type (either no IP boundary or an IP boundary in a critical two-word sequence), splicing (whether the onset of the sequence’s second word was spliced from another token of that sentence or cross-spliced from a matched sentence with or without an IP boundary), and relatedness (whether the visual target was the first word in the spoken sequence). There was a relatedness effect on target responses for sentences with no IP boundary only when they were cross-spliced, that is, where splicing provided evidence of domain-initial strengthening. Listeners thus use this evidence when segmenting continuous speech.