Multiple parallel hidden layers and other improvements to recurrent neural network language modeling | IEEE Conference Publication | IEEE Xplore