Toward measurement of pitch alignment

This paper discusses theoretical and practical issues underlying the measurement of pitch alignment. We define the alignment concept as the relationship between pitch trajectories and articulatory/acoustic trajectories. This concept is formalized within a general superpositional framework, according to which a pitch curve is viewed as the sum of component curves, such as phrase curves, accent curves, and segmental perturbation curves. According to a special case of the general superpositional concept, the Linear Alignment Model, a given intonational-phonological pitch accent class can be characterized as the combination of an underlying accent template (that represents the basic shape of the pitch excursion, e.g., rise, rise-fall) and an alignment parameter matrix (that specifies how to warp this template to be properly aligned with the segmental stream with which a pitch accent is associated, taking into account the segmental/durational structure of this stream). Traditional measurement of alignment is customarily based on the surface-point-to-surface-point approach (P-P approach). In this approach, the time interval (in ms or as a percentage of some structural unit such as the syllable) is measured between pitch points (i.e., points on the surface pitch contour, such as local F0 minima and maxima) and segmental points (i.e., points in the segmental stream such as boundaries between segments, syllable constituents, or syllable boundaries). A special case of the P-P approach is the search for segmental points that serve as anchors for pitch points. We show how the Linear Alignment Model can account for the systematic dependency of pitch point timing on the segmental/durational structure of the segmental stream. Specifically, we show how apparent changes of alignment as measured by the P-P approach, resulting from some independent variable, are in fact predictable via the model as direct consequences of the effects of the independent variable on the segmental/durational structure, and thus may not be changes in alignment at all. We also show how the model can account for phonological-perceptual changes associated with small changes in alignment in combination with unchanged segmental/ durational structure and pitch accent shape.