FBK’s Participation to the English-to-German News Translation Task of WMT 2017

Proceedings of the Conference on Machine Translation (WMT), Volume 2: Shared Task Papers, pages 271–275 Copenhagen, Denmark, September 711, 2017. c⃝2017 Association for Computational Linguistics FBK’s Participation to the English-to-German News Translation Task of WMT 2017 Mattia A. Di Gangi1,2 and Nicola Bertoldi2 and Marcello Federico2 1ICT International Doctoral School - University of Trento, Italy 2Fondazione Bruno Kessler - Trento, Italy {digangi,bertoldi,federico}@fbk.eu Abstract In this paper we report on FBK’s par- ticipation to the English-to-German news translation task of the Second Conference on Machine Translation (WMT’17). The submitted system is based on Neural Ma- chine Translation using byte-pair encod- ing segmentation on both source and tar- get languages for open-vocabulary trans- lations. Back-translations of news mono- lingual data are used for improving the translations fluency on the in-domain data. With respect to last year’s evaluation, our baseline outperforms the 2016 best sys- tem’s baseline on the test sets 2015 and 2016. However, in our set-up back- translations produced a smaller improve- ment than expected. The final submission is given by the combination of 7 systems, including a system trained only on true parallel data and two right-to-left systems, which improves over our single best sys- tem by 1.5 BLEU points. 1 Introduction FBK’s participation to the news translations shared task in WMT 17 focused this year on the English-German language direction. Our purpose was to explore the state of the art and build a competitive neural machine translation [3] sys- tem in order to gain a practical knowledge of the available tools. With respect to our participa- tion in the IWSLT 2016 evaluation campaign, we switched from the Nematus-Theano framework to the OpenNMT-Torch framework [16]. The rea- sons were twofold: higher baseline performance and significantly faster training. In our primary submission we used back- translations [22], BPE-encoding [23] and sys- tem combination [11]. In this paper, we report about the tools used for the submitted system and the choices we have taken in terms of hyper- parameters and used data. The presentation is structured as follows: in Sec- tion 2 we briefly introduce the theoretical back- ground for NMT. In Section 3 we describe our baseline system. In Sections 4 and 5 we de- scribe the details of the back-translations and sys- tem combination, which have been used for our final submission. Evaluation results are discussed in Section 6, while Section 7 is devoted to discus- sion and conclusions. 2 Neural Machine Translation Neural machine translation [25] represents the state of the art for machine translation since the outstanding results obtained on IWSLT2015 [17] IWSLT2016 [1, 7] and WMT16 [24, 5] where the neural models greatly outperformed phrase-based systems. NMT is based on the encoder-decoder- attention architecture [3] which jointly learns the translation and the alignment model with a sequence-to-sequence learning model. Given a se- quence of words f1, f2, . . . , fm in the source lan- guage, they are used to index an embedding look- up table and retrieve the vectors x1, x2, . . . , …

Content truncated to 500 words