MetaMT,a MetaLearning Method Leveraging Multiple Domain Data for Low Resource Machine Translation

This addresses the challenge of limited training data for machine translation, but appears incremental as it builds on existing meta-learning and data manipulation techniques.

The paper tackles the problem of low-resource machine translation by leveraging multiple domain data through meta-learning, resulting in robust neural models.