Neural Machine Translation with Key-Value Memory-Augmented Attention
This addresses translation quality issues in NMT for language processing applications, representing an incremental improvement over existing attention mechanisms.
The paper tackles the problem of repeating and dropping translations in neural machine translation by proposing a key-value memory-augmented attention model, which improves translation adequacy and shows superior results on Chinese-English and WMT17 German-English tasks.
Although attention-based Neural Machine Translation (NMT) has achieved remarkable progress in recent years, it still suffers from issues of repeating and dropping translations. To alleviate these issues, we propose a novel key-value memory-augmented attention model for NMT, called KVMEMATT. Specifically, we maintain a timely updated keymemory to keep track of attention history and a fixed value-memory to store the representation of source sentence throughout the whole translation process. Via nontrivial transformations and iterative interactions between the two memories, the decoder focuses on more appropriate source word(s) for predicting the next target word at each decoding step, therefore can improve the adequacy of translations. Experimental results on Chinese=>English and WMT17 German<=>English translation tasks demonstrate the superiority of the proposed model.