Despite the promising results achieved in last years by statistical machine translation, and more precisely, by the neural machine translation systems, this technology is still not error-free. The outputs of a machine translation system must be corrected by a human agent in a post-editing phase. Interactive protocols foster a human computer collaboration, in order to increase productivity. In this work, we integrate the neural machine translation into the interactive machine translation framework. Moreover, we propose new interactivity protocols, in order to provide the user an enhanced experience and a higher productivity. Results obtained over a simulated benchmark show that interactive neural systems can significantly improve the classical phrase-based approach in an interactive-predictive machine translation scenario. c 2016 Elsevier Ltd. All rights reserved.

The authors wish to thank the anonymous reviewers for their careful reading and in-depth criticisms and suggestions. This work was partially funded by the project ALMAMATER (PrometeoII/2014/030). We also acknowledge NVIDIA for the donation of the GPU used in this work.