Highly Efficient Neural Network Language Model Compression Using Soft Binarization Training | IEEE Conference Publication | IEEE Xplore