TY - GEN
T1 - The rise and fall of American English pitch accents
T2 - Evidence from an imitation study of rising nuclear tunes
AU - Steffman, Jeremy
AU - Shattuck-Hufnagel, Stefanie
AU - Cole, Jennifer
PY - 2022/5/26
Y1 - 2022/5/26
N2 - Rising pitch movements associated with pitch accents are frequently described in terms of alignment and scaling; for example, L+H* versus H* accents vary in these parameters. We examine how 12 American English nuclear tunes, created by combining three pitch accents {H*, L+H*, L*+H}, and four edge tone sequences {H-H%, H-L%, L-H%, L-L%}, are distinguished in an imitative speech production paradigm. Bottom-up clustering analyses of unlabeled time-series f0 identify a robust distinction between trajectories that rise throughout (rise-only) and those with rising-falling movements (rise-fall). Additional clustering distinctions between tunes with different pitch accents are observed only in the rise-only cluster, and further reflect variation in holistic nuclear tune shape. For rise-fall movements, further distinctions in clustering are best defined by ending f0, corresponding to a boundary tone distinction {H%, L%}. With only 4 distinct clusters emerging from the imitated tunes, it appears that some tune distinctions are lost. Nevertheless, modeling trajectories with ToBI labels using a GAMM, and testing alignment of f0 turning points, reveals small differences between tunes in f0 scaling and alignment, distinguishing ToBI labels that were grouped together in clustering. We discuss these results in terms of the hierarchy of distinctions they imply, and categories of tune shapes.
AB - Rising pitch movements associated with pitch accents are frequently described in terms of alignment and scaling; for example, L+H* versus H* accents vary in these parameters. We examine how 12 American English nuclear tunes, created by combining three pitch accents {H*, L+H*, L*+H}, and four edge tone sequences {H-H%, H-L%, L-H%, L-L%}, are distinguished in an imitative speech production paradigm. Bottom-up clustering analyses of unlabeled time-series f0 identify a robust distinction between trajectories that rise throughout (rise-only) and those with rising-falling movements (rise-fall). Additional clustering distinctions between tunes with different pitch accents are observed only in the rise-only cluster, and further reflect variation in holistic nuclear tune shape. For rise-fall movements, further distinctions in clustering are best defined by ending f0, corresponding to a boundary tone distinction {H%, L%}. With only 4 distinct clusters emerging from the imitated tunes, it appears that some tune distinctions are lost. Nevertheless, modeling trajectories with ToBI labels using a GAMM, and testing alignment of f0 turning points, reveals small differences between tunes in f0 scaling and alignment, distinguishing ToBI labels that were grouped together in clustering. We discuss these results in terms of the hierarchy of distinctions they imply, and categories of tune shapes.
KW - intonation
KW - nuclear tunes
KW - alignment
KW - scaling
KW - clustering
KW - imitative speech production
U2 - 10.21437/SpeechProsody.2022-174
DO - 10.21437/SpeechProsody.2022-174
M3 - Conference contribution
T3 - Speech prosody
SP - 857
EP - 861
BT - Proceedings of the International Conference on Speech Prosody
ER -