Translation Reinvented
Our Adversarial-NMT agent uses a 2D convolutional neural network as an adversary to challenge and refine the output of a neural machine translation (NMT) model. By co-training our NMT model and the adversary with a policy gradient method we can produce high quality translations that are indistinguishable from human ones.
​
Our Solution is a single system that can be trained directly on source and target text, without the need for any intermediate steps or specialized tools.
​
Multilingual
Data that requires optimized features handling and guided attention is trained using a Machine Translation Task and translated into multiple languages.
Natural Language Flexibility
Given a sequence of text in a source language, there is not one single best chat of that text to another language. This is because of the natural ambiguity and flexibility of human language.
Efficient Translation
The Agent can expertly and efficiently convert text from the source language to the target language, taking into account a vast number of rules and exceptions.