Adversarial Machine Translation

Experimental Neural Machine Translation model, based on Transformer Networks with Adversarial Target.

Instead of learning translation model by Likelihood maximization, this model relies on minimization Wasserstein-1 distance between real translations and parametrized with this model distribution.