Greedy Search with Probabilistic N-gram Matching for Neural Machine Translation
This addresses training inefficiencies in neural machine translation, offering a more stable alternative to reinforcement learning for researchers and practitioners, though it is incremental as it builds on existing sequence-level training ideas.
The paper tackles the problems of improper translation evaluation and exposure bias in neural machine translation by introducing a differentiable sequence-level training objective based on probabilistic n-gram matching, which avoids reinforcement learning and uses greedy search to align training with inference. Results show it outperforms reinforcement-based methods and improves by 1.5 BLEU points on average over a strong baseline in Chinese-to-English translation tasks.
Neural machine translation (NMT) models are usually trained with the word-level loss using the teacher forcing algorithm, which not only evaluates the translation improperly but also suffers from exposure bias. Sequence-level training under the reinforcement framework can mitigate the problems of the word-level loss, but its performance is unstable due to the high variance of the gradient estimation. On these grounds, we present a method with a differentiable sequence-level training objective based on probabilistic n-gram matching which can avoid the reinforcement framework. In addition, this method performs greedy search in the training which uses the predicted words as context just as at inference to alleviate the problem of exposure bias. Experiment results on the NIST Chinese-to-English translation tasks show that our method significantly outperforms the reinforcement-based algorithms and achieves an improvement of 1.5 BLEU points on average over a strong baseline system.