This paper investigates the scaling properties of Recurrent Neural Network Language Models (RNNLMs). We discuss how to train very large RNNs on GPUs and address the questions of how RNNLMs scale with respect to model size, training-set size, computational costs and memory. Our analysis shows that despite being more costly to train, RNNLMs obtain much lower perplexities on standard benchmarks than n-gram models. We train the largest known RNNs and present relative word error rates gains of 18% on an ASR task. We also present the new lowest perplexities on the recently released billion word language modelling benchmark, 1 BLEU point gain on machine translation and a 17% relative hit rate gain in word prediction.

Our entry to the parallel corpus filtering task uses a two-step strategy. The first step uses a series of pragmatic hard 'rules' to remove the worst example sentences. This first step reduces the effective corpus size down from the initial 1 billion to 160 million tokens. The second step uses four different heuristics weighted to produce a score that is then used for further filtering down to 100 or 10 million tokens. Our final system produces competitive results without requiring excessive fine tuning to the exact task or language pair. The first step in isolation provides a very fast filter that gives most of the gains of the final system.

No abstract