Here's my first shot at Julius' described model. I think it isn't what's supposed to happen. But I do think this is the basic mechanism. I suppose you can get more pleasant results when you know the function for the lookup table. And when you know how to vary the lookup table. Julius describes that the lookup table changes when you change the applied force.
I've been fiddling around with this lately. This model works pretty well, especially when you use your mouse as a controller. It's based on research by Julius O. Smith. And I ripped some parts of the patch from the Synthesis ToolKit in C++ of Perry R. Cook & Gary P. Scavone.
To get the patch to work make sure you set the signal vector size to 1 sample. Otherwise, the tuning will be off. I could have wrapped a part of the patch into a poly~ to set the vector size to 1, but I didn't do that just yet. I also know gen~ is better for this kind of application. But I don't own gen~.