Background and Objectives: This chapter presents results since our previous work on machine-learning based Model Order Reduction (MOR) method applied to a complex 3D Finite Element (FE) biomechanical model of the human tongue. This model is important to adress real-time simulation problems essential for future work on speech motor control.Methods: The proposed method is the same “a posteriori” MOR method as the one presented in our previous work but engaging within a more complex framework involving contacts and a reduced amount of training data.Results: The MOR method is evaluated for simulations involving only one muscle but with the complexity of contacts generating discontinuities in the boundary conditions. It is shown to be able to take into account, with sub-millimeter spatial accuracy, the behavior of the model with a reduced number of simulations for learning and extrapolating the prediction.Conclusion: Further evaluations of the MOR method will include tongue movements induced by multiple muscle activations and contact diversity. At this stage, our MOR method offers promising prospects for the use of the tongue model in a speech motor control context.