Dissecting coarticulation: How locus equations happen

A programmatic series of studies aimed at expanding our understanding of coarticulation in V1·CV2 sequences is presented. The common thread was examining coarticulatory dynamics through the prism of locus equations (LEs). Multiple experimental methodologies (articulatory synthesis, X-ray film, Principal Component Analysis, and extraction of time constants for F2 transitions), guided by a few theoretical assumptions about speech motor planning and control, were used to uncover the articulatory underpinnings responsible for the trademark acoustic form of LE scatterplots. Specific findings were: (1) the concept of a stop consonantal ‘target’ was quantitatively derived as a vowel-neutral, ‘deactivated,’ tongue contour; (2) the linearity of LEs is significantly enhanced by the uniformity of F2 transition time constants, which normalize with respect to F2 transition extents, and an inherent linear bias created by the smaller frequency range of [F2onset−F2vowel] relative to F2vowel frequencies; (3) realistic LE slopes and y-intercepts were derived by modeling different extents of V2 overlap onto stop consonantal target shapes at closure; and (4) a conceptually simple model, viz. interpolation between successive articulatory target shapes, followed by derivation of their formant values expressed as LEs, came surprisingly close to matching actual LEs obtained from our speaker.

► We sought to explain how locus equations index CV coarticulation and why they are linear.
► Stop closure ‘target’ contours for [bdɡ] were reconstructed from PCA analyses.
► V-neutral tongue shapes were then used to simulate locus equations in five vowel contexts.
► Varying degrees of vowel overlap with C-target shapes were captured by locus equation slopes.
► Linearity was significantly linked to near invariant F2 transition time constants irrespective of extent.