Predicting 3D Tongue Surfaces from Midsagittal Contours

M Stone, M Epstein, M Sutton


It would be very useful to be able to predict 3D tongue shape from midsagittal data.  3D data reveal the tongue’s true degrees of freedom, as well as physiological and linguistic constraints, but 3D data are difficult to collect.  We believe, however, that it is possible to predict 3D shapes from midsagittal data, because any single language uses a limited set of tongue gestures.  For example, coronal data show that the tongue’s cross-sectional motions range from a midline arch to a midline groove (cf. Stone and Vatikiotis-Bateson, 1995,  J. Phon, 81-100).  Although this shape range is valid throughout the tongue, different tongue segments may be executing different shapes at any one time. Thus predicting 3D shapes can be complicated.  For American English, however,  we already have 3D surface information, within which is embedded midsagittal and coronal data.  From this, midsagittal heights and coronal shapes can be determined for any specific sound.  Therefore, any 3D surface can be decomposed into a series of coronal arches and grooves from front to back, as well as a single midsagittal shape.  Conversely, a series of coronal shapes from front-to-back can be constructed into a 3D surface, and a midsagittal shape is embedded in these data sets as well.  Three-dimensional tongue surface shapes have been documented for all the sounds of American English (Stone and Lundberg, 1996, JASA, 3728-3737).   Using this data set, midsagittal points will be mapped to 3D surfaces.  The goals are (1) to determine the strength of the midline height/coronal shape correlation at each A-P tongue location and (2) define the 3D surfaces by their composite 2D coronal contours and determine their predictability from the midsagittal shape.