Stochastic suprasegmentals: relationships between redundancy, prosodic structure and care of articulation in spontaneous speech

Within spontaneous speech there are wide variations in the articulation of the same word by the same speaker. Some words become extremely reduced while others seem to stand out more strongly in a phrase or sentence. This thesis explores these variations in articulation from two different but, arguably, related perspectives, prosodic structure and redundancy. I argue that the constraint of producing robust communication while efficiently expending articulatory effort leads to: 1. An inverse relationship between language redundancy and care of articulation 2. The need for a strong ’checking’ signal The inverse relationship improves robustness by spreading the information more smoothly across the speech signal leading to a smoother signal redundancy profile. Checking in contrast leads to a more robust signal by ensuring that errors are detected and corrected. I argue that smooth signal redundancy and a checking signal could be implemented by prosodic prominence and prosodic boundaries. Prosodic prominence increases care of articulation and appear to coincide with unpredictable sections of speech. In doing so prosodic prominence leads to a smoother signal redundancy. Prosodic boundaries cause syllabic lengthening and, by bounding self contained chunks of information (such as a word or phrase), signal that a listener should have a meaningful section of speech as well as offering a location for a listener to request clarification or re-transmission. In this way prosodic boundaries could be regarded as a checking signal. The work presented here concentrates on the issue of smoothing redundancy. In order to explore this idea quantitatively, prosodic coding, metrics of language redundancy (word frequency, syllabic trigrams and givenness) and of care of articulation (normalised syllabic duration and vowel quality) are formulated and applied to a large corpus of English spontaneous task-oriented dialogue. Results confirm the strong relationship between prosodic structure and care of articulation as well as an inverse relationship between language redundancy and care of articulation. In addition, when an opportunity for a checking signal is controlled for, in some circumstances language redundancy can predict up to 65% of the variance in raw syllabic duration. This is comparable with 64% predicted by prosodic structure. Moreover most (62%) of this predictive power is shared. This leads to the conclusion that, within English, prosodic structure is the means with which constraints caused by requiring a robust signal are expressed in spontaneous speech. Finally it is argued that, if redundancy is indeed a driving force behind prosodic structure, notions of redundancy and predictability should be more formally included into prosodic theory.