Tuesday, December 09, 2008

Connectionist modeling for stuttering?


I am currently doing a postgraduate Open University course on Exploring cognition: damaged brains and neural networks. The course explores two methodologies to study the brain: case studies on damaged brains, and connectionist modeling. We learn much more about the brain when it does not work properly as opposed to when it works well!

Think of your car. After any breakdown be it the battery or gears, you know much better how the damn thing really works! Or go back to the pain and handicap from your last toothache or back problems. Now you really know what and how teeth or backs do every single day. Stuttering is no different. A speech scientist should see the disorder as a blessing. Any scientist who claims to know how the brain speaks must be able to explain why the brain does not speak fluenty for some people. In fact, stuttering is key to understanding normal speech production. If we know why we stutter, we also know much more about the normal brain processes underlying speech and related activities.

The second methodology is connectionist modeling which models brain processes by constructing mathematical models that mimic features of networks of neurons and "injures" them to see the effects. I do not know of any group working on this topic in stuttering. I vaguely remember Pete Howell, professor at University College London, trying to get a group in connectional modelling to work on stuttering. The Boston group around Prof Guenther does mathematical modeling, especially his PhD student Oren Civier: see here. However, they use the differential equations approach which is a very different modeling technique. Ludo Max has worked on models together with them, but I think they agreed to disagree on the right approach. Within the research community their work is completely ignored, simply because no-one understand what they have done or are doing! It took me some time to understand what they have done, but I would need to play around with the equations myself to give good feedback on the usefulness of their models. My intuition tells me that the models are too simple to capture the essentials, and that even if a model reproduces the essentials of stuttering, it does not necessarily imply that it is implemented in the brain in this way.

2 comments:

Oren said...

Thanks for mentioning my research, Tom. Your interested readers would like to know that we're now modeling the basal ganglia, which will allow us to account not only for stuttering, but also for spasmodic dysphonia.
For a movie of stuttering produced by our model, click http://speechlab.bu.edu/movies/Good_doggie_stuttering.mpg

Oren

thaddeus said...

this is a really good post
just puts more emphasis on all of my many theories