In plain words: A free, open software package that trains and runs neural translation systems, built so researchers can swap in new model designs, input features, or source types, with clear teaching docs. It keeps translation quality competitive while needing only reasonable training time.

Abstract

We describe an open-source toolkit for neural machine translation (NMT). The toolkit prioritizes efficiency, modularity, and extensibility with the goal of supporting NMT research into model architectures, feature representations, and source modalities, while maintaining competitive performance and reasonable training requirements. The toolkit consists of modeling and translation support, as well as detailed pedagogical documentation about the underlying techniques.

Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, Alexander M. Rush
arXiv:1701.02810 · cs.CL, cs.AI, cs.NE · submitted Jan 10, 2017 · updated Mar 6, 2017
abstract · pdf · html · Report for http://opennmt.net