The form and auditory control of downward trends in intonation
Of all the areas of intonational research, study of the tendency of the frequency of vocal fold vibration to decline during the course of an utterance - F0 declination - is likely initially to be the most fruitful in determining the interaction between perceptual and productive processes. A general introduction to the phenomenon is augmented by analysis of different methods of determining declination lines; theoretical treatments are then introduced. One particular local factor contributing to the downward trend, downstep, is discussed, and its pivotal role in the intonational phonology developed by Janet Pierrehumbert critically examined. In the light of the theoretical discussion, two competing hypotheses are presented as to the mediation of the declination effect, which is the effect that of two accented syllables in an utterance, the second has to have a lower peak F0 value than the first for them to be judged to have equal prominence. The Global Declination Hypothesis attributes this to the use by speakers and hearers of one or two abstract reference lines declining through the course of a tone-unit. The Local Declination Hypothesis attributes it to the disposition of F0 excursions surrounding the two accents as well as to the respective peak values. The Global Declination Hypothesis is tested by presenting listeners with pairs of dual-peak accented utterances with the two peaks identical in F0, without any physically present local declination, and asking them to rate the prominence of the second peak of each such utterance. No significant differences are found in the prominence ratings, so the Local Declination Hypothesis appears to be favoured. That hypothesis is itself tested through the development of a model of individual accent prominence, which incorporates terms for surrounding unaccented context. This is then used as the basis of a model of the perceptual constraints on the production of intonation in the scaling of target peaks. The model predicts that local slope between accents and slope of the context after the target accent, as well as other local variables, jointly determine the F0 value of a peak with a particular targetted prominence relationship with its predecessor. If the interaccentual stretch is declining, the declination effect is predicted to occur, ceteris paribus. The model is found to be initially acceptable. In addition, a global interpretation of downstep is made within the model. The mechanisms the model is suggested to represent are auditory feedback control loops of a variety of possible degrees of complexity. An experiment is devised to test for the basic existence of a feedback loop which is used to prevent local slope exceeding an arbitrary threshold value. Auditory feedback In subjects was disrupted by headphone-administration of low-pass filtered masking noise during their utterance of a sustained vowel, and a short and a long dual peak-accented sentence. The disruption was sufficient to alter the apparent mechanism controlling the production of the sustained vowel, but the Lombard effect, whereby subjects automatically raise the level of their voice in ambient noise, was found to be a vitiating factor. General conclusions are drawn on the nature of the declination phenomenon In intonation, and proposals made for future research.