Kalveram, K.Th., & Natke, U. (1997) Stuttering and misguided learning of articulation, or why it is extremely difficult to estimate the physical parameters of limbs. In W. Hulstijn, H. Peters & P. H. H. M. Van Lieshout (Eds.): Speech production: Motor Control, Brain Research and Fluency Disorders. Amsterdam: Elsevier Science Publishers, 89-98.

Abstract
The speech tract transforms muscle forces exerted on the articulatory system, larynx included, into actual speech sounds. Speaking then means to invert this transformation neurally. The related inverse model can be acquired by auto-imitation. In order to identify the control type of speaking, we tried to measure inertia, stiffness and damping of the jaw during speech. For this sake, we applied an external force impulse with known shape to the jaw and recorded the related acceleration, velocity and position. This is an auto-imitation-like procedure and should provide the parameters looked for. However it turned out that mathematically the problem is ill posed, whereby small errors in the recorded data will lead to big errors in the parameters to be estimated. Typical regularization methods failed. To get deeper insight into the problem we simulated a three-jointed arm and the acquisition of its control. An exact solution of the inverse dynamics could only be obtained with no noise in the feedback lines. Already under moderate noise, in 5-10 percent of the cases misguided learning or-cured disrupting movement control, which follows from the ill posedness also present in this instance. We concluded that the organism is exposed to the same problem when trying to get control of the articulators, larynx included. This holds especially for children of about four years, when they learn to unstress formerly stressed, but unimportant syllables. This demands a (second) auto-imitative process which sometimes is assumed to be misguided in a manner, that phonation occasionally is prevented. Taking into account models of prosody control and speech planning, it is argued that this could explain iterations of syllables.