Neural Machine Translation Leveraging Phrase-based Models in a Hybrid Search
This work addresses translation quality for machine translation systems, but it is incremental as it combines existing NMT and SMT approaches rather than introducing a fundamentally new paradigm.
The paper tackles the problem of improving neural machine translation quality by integrating phrase-based statistical models into the NMT beam search, resulting in up to 2.3% absolute BLEU score improvement over a strong NMT baseline on German-English and English-Russian translation tasks.
In this paper, we introduce a hybrid search for attention-based neural machine translation (NMT). A target phrase learned with statistical MT models extends a hypothesis in the NMT beam search when the attention of the NMT model focuses on the source words translated by this phrase. Phrases added in this way are scored with the NMT model, but also with SMT features including phrase-level translation probabilities and a target language model. Experimental results on German->English news domain and English->Russian e-commerce domain translation tasks show that using phrase-based models in NMT search improves MT quality by up to 2.3% BLEU absolute as compared to a strong NMT baseline.