Applied Soft Computing Journal 78 (2019) 230–239 Contents lists available at ScienceDirect Applied Soft Computing Journal journal homepage: www.elsevier.com/locate/asoc Assembling translations from multi-engine machine translation outputs Debajyoty Banik a,b , Asif Ekbal a,b , Pushpak Bhattacharyya a,b , Siddhartha Bhattacharyya c,d,∗ a Department of Computer Science and Engineering, India b Indian Institute of Technology Patna, India c Department of Information Technology, India d RCC Institute of Information Technology, India highlights • A hybrid architecture is proposed to assemble different system combina- tion models. • The architecture is influenced by both statistical and neural network based system combination techniques. • The training datasets are prepared by using different types of MT sys- tems. graphical abstract article info Article history: Received 11 September 2018 Received in revised form 28 January 2019 Accepted 19 February 2019 Available online 28 February 2019 Keywords: Statistical approach Neural network Machine translation Neural Machine Translation (NMT) Statistical Machine Translation (SMT) abstract In this paper, we present a hybrid architecture for developing a system combination model that works in three layers to achieve better translated outputs. In the first layer, we have various machine translation models (i.e. Neural Machine Translation (NMT), Statistical Machine Translation (SMT), etc.). In the second layer, the outputs of these models are combined to leverage the advantages of both the systems (i.e SMT and NMT systems) by using the statistical approach and neural-based approach. But each approach has some advantages and limitations. So, instead of selecting an individual combined system’s output as the final one, we apply these outputs in the final layer to produce the target output by assigning appropriate preferences to SMT based and neural-based combinations. Though there are some techniques for system combination but no such approach exists which uses preferences from various system combination models (statistical and neural) for the purpose of better assembling. Empirical results show improved performance in the terms of translation accuracy. Our experiments on two benchmark datasets of English–Hindi and Hindi–English pairs show that the proposed model performs significantly better than the participating models. Apparently, the efficacy of proposed model is significantly better than the state-of-the art machine translation combination systems (6.10 and 4.69 BLEU points for English-to-Hindi, and Hindi-to-English, respectively). © 2019 Elsevier B.V. All rights reserved. ∗ Corresponding author. E-mail addresses: debajyoty.pcs13@iitp.ac.in (D. Banik), asif@iitp.ac.in (A. Ekbal), pb@iitp.ac.in, swalpa@cse.jgec.ac.in (P. Bhattacharyya), dr.siddhartha.bhattacharyya@gmail.com (S. Bhattacharyya). 1. Introduction The SMT system [1,2] is better at preserving adequacy and handling rare words [3], the NMT system, in contrast, [4,5] is https://doi.org/10.1016/j.asoc.2019.02.031 1568-4946/© 2019 Elsevier B.V. All rights reserved.