GVU Technical Report Number: GIT-GVU-09-06

Title: Recognizing Sign Language from Brain Imaging

Authors: Nishant A. Mehta, Thad Starner, Melody Moore Jackson, Karolyn O. Babalola, George Andrew James

Abstract: The problem of classifying complex motor activities from brain imaging is relatively new territory within the fields of neuroscience and brain-computer interfaces. We report positive sign language classification results using a tournament of pairwise support vector machine classifiers for a set of 6 executed signs and also for a set of 6 imagined signs. For a set of 3 contrasted pairs of signs, executed sign and imagined sign classification accuracies were highly significant at 96.7% and 73.3% respectively. Multiclass classification results also were highly significant at 66.7% for executed sign and 50% for imagined sign. These results lay the groundwork for a brain-computer interface based on imagined sign language, with the potential to enable communication in the nearly 200,000 individuals that develop progressive muscular diseases each year.