Adaptive Importance Sampling to Accelerate Training of a Neural Probabilistic Language Model