The present study tests a prediction from the prevalent Autosegmental-Metrical (AM) model of American English intonation: the existence of distinct phonological contrasts among nuclear tunes composed of a pitch accent (here H*, L+H*, L*+H), phrase accent (H-, L-) and boundary tone (H%, L%), which in combination yield an inventory of 12 tonally distinct nuclear tunes. Using an imitative speech production paradigm and AX discrimination task with L1 speakers of Mainstream American English (MAE) we test the extent to which each of 12 predicted tunes is distinct from the others in the production and perception of intonation. We tackle this question with a series of analytical methods. We use GAMM modeling of time-series F0 trajectories to test for differences among all of the twelve nuclear tunes, and compare these results to a method that does not rely on pre-defined tune categories, k-means clustering for time-series data, to discover emergent classes of tunes in a "bottom-up" fashion. We complement these time-series analyses with an analysis of the temporal tonal center of gravity (TCoG) over the F0 trajectories of nuclear tunes to assess tonal timing distinctions and their relation to top-down tune classes (defined by the AM model) and bottom-up classes (emergent from clustering). Production results are further compared to perceptual discrimination responses, which together point to a hierarchy of distinctions among nuclear tunes: a set of primary tune distinctions are emergent in clustering and always distinct in perception. Other tune distinctions, although evident in top-down analyses of (labeled) F0 trajectories, are lost in emergent clusters, limited in magnitude and scope, and often confused in perception. Results are discussed in terms of implications for a theory of intonational phonology.
Steffman et al. (Mon,) studied this question.