Neural Machine Translation: Recent Topics and Advances

Ensemble and System Combination

• COLING 2016, Ensemble Learning for Multi-Source Neural Machine Translation.
• ACL 2017, Neural System Combination for Machine Translation.

Combining SMT and NMT

• COLING 2016, Pre-Translation for Neural Machine Translation.

Dynamic Vocabulary

• ACL 2016, Vocabulary Manipulation for Neural Machine Translation.
• AAAI 2018, Neural Response Generation with Dynamic Vocabularies.
  • Might work for NMT, another way of distributing uncertainty (probability mass).

Open Vocabulary

• ACL 2016, Achieving Open Vocabulary Neural Machine Translation with Hybrid Word-Character Models.

Data Augmentation

• EMNLP 2016, Exploiting Source-side Monolingual Data in Neural Machine Translation.
  • Reordering objective and self-training.

Paraphrasing with/for NMT

• EACL 2017, Paraphrasing Revisited with Neural Machine Translation.

Translation Memory for Training/Decoding

Dynamic Programming with Structural Bias for NMT

• EMNLP 2016, Online Segment to Segment Neural Transduction.
• ICLR 2018, Towards Neural Phrase-based Machine Translation. github.

Fast Softmax

• AISTAT 2017, Complementary Sum Sampling for Likelihood Approximation in Large Scale Classification.
• NIPS 2017, SVD-Softmax: Fast Softmax Approximation on Large Vocabulary Neural Networks.

---