Experimental evaluation of the constant tongue volume hypothesis

The human tongue is considered to be a muscular hydrostat (Kier and Smith, 1985). As such, it is considered to be incompressible. This constant volume hypothesis has been incorporated in various mathematical models of the tongue, which attempt to provide insights into its dynamics (e.g., Levine et al., 2005). However, to the best of our knowledge, this hypothesis has not been experimentally validated for the human tongue during actual speech production. In this work, we attempt an experimental evaluation of the constant tongue volume hypothesis. To this end, volumetric structural Magnetic Resonance Imaging (MRI) was used. A database consisting of 3D MRI images of subjects articulating continuants was considered. The subjects sustained contextualized vowels and fricatives (e.g., IY in “beet,” F in “afa”) for 8 seconds in order for the 3D geometry to be collected. To segment the tongue and estimate its volume, we explored watershed (Meyer and Beucher, 1990) and region growing (Adams and Bischof, 1994) techniques. Tongue volume was estimated for each lingual posture for each subject. Intra-subject tongue volume variation was examined to determine if there is sufficient statistical evidence for the validity of the constant volume hypothesis. [Work supported by NIH and a USC Viterbi Graduate School Ph.D. fellowship.]