Checkpoint averaging is a simple and effective method to boost the performance of converged neural machine translation models. The calculation is cheap to perform and the fact that the translation improvement almost comes for free, makes it widely adopted in neural machine translation research. Despite the popularity, the method itself simply takes the mean of the model parameters from several checkpoints, the selection of which is mostly based on empirical recipes without many justifications. In this work, we revisit the concept of checkpoint averaging and consider several extensions. Specifically, we experiment with ideas such as using different checkpoint selection strategies, calculating weighted average instead of simple mean, making use of gradient information and fine-tuning the interpolation weights on development data. Our results confirm the necessity of applying checkpoint averaging for optimal performance, but also suggest that the landscape between the converged checkpoints is rather flat and not much further improvement compared to simple averaging is to be obtained.

Encoder-decoder architecture is widely adopted for sequence-to-sequence modeling tasks. For machine translation, despite the evolution from long short-term memory networks to Transformer networks, plus the introduction and development of attention mechanism, encoder-decoder is still the de facto neural network architecture for state-of-the-art models. While the motivation for decoding information from some hidden space is straightforward, the strict separation of the encoding and decoding steps into an encoder and a decoder in the model architecture is not necessarily a must. Compared to the task of autoregressive language modeling in the target language, machine translation simply has an additional source sentence as context. Given the fact that neural language models nowadays can already handle rather long contexts in the target language, it is natural to ask whether simply concatenating the source and target sentences and training a language model to do translation would work. In this work, we investigate the aforementioned concept for machine translation. Specifically, we experiment with bilingual translation, translation with additional target monolingual data, and multilingual translation. In all cases, this alternative approach performs on par with the baseline encoder-decoder Transformer, suggesting that an encoder-decoder architecture might be redundant for neural machine translation.

The neural hidden Markov model has been proposed as an alternative to attention mechanism in machine translation with recurrent neural networks. However, since the introduction of the transformer models, its performance has been surpassed. This work proposes to introduce the concept of the hidden Markov model to the transformer architecture, which outperforms the transformer baseline. Interestingly, we find that the zero-order model already provides promising performance, giving it an edge compared to a model with first-order dependency, which performs similarly but is significantly slower in training and decoding.

In order to combat overfitting and in pursuit of better generalization, label smoothing is widely applied in modern neural machine translation systems. The core idea is to penalize over-confident outputs and regularize the model so that its outputs do not diverge too much from some prior distribution. While training perplexity generally gets worse, label smoothing is found to consistently improve test performance. In this work, we aim to better understand label smoothing in the context of neural machine translation. Theoretically, we derive and explain exactly what label smoothing is optimizing for. Practically, we conduct extensive experiments by varying which tokens to smooth, tuning the probability mass to be deducted from the true targets and considering different prior distributions. We show that label smoothing is theoretically well-motivated, and by carefully choosing hyperparameters, the practical performance of strong neural machine translation systems can be further improved.

Attention-based encoder-decoder models have achieved great success in neural machine translation tasks. However, the lengths of the target sequences are not explicitly predicted in these models. This work proposes length prediction as an auxiliary task and set up a sub-network to obtain the length information from the encoder. Experimental results show that the length prediction sub-network brings improvements over the strong baseline system and that the predicted length can be used as an alternative to length normalization during decoding.