Strategies for Training Large Vocabulary Neural Language Models

Training neural network language models over large vocabularies is computationally costly compared to count-based models such as Kneser-Ney. We present a systematic comparison of neural strategies to represent and train large vocabularies, including softmax, hierarchical softmax, target sampling, noise contrastive estimation and self normalization. We extend self normalization to be a proper estimator of likelihood and introduce an efficient variant of softmax. We evaluate each method on three popular benchmarks, examining performance on rare words, the speed/accuracy trade-off and complementarity to Kneser-Ney.