Re-design of the Machine Translation Training Tool (MT3)

Re-design of the Machine Translation Training Tool MT 3 Emiliano Cuenca Paula Estrella Laura Bruno Faculty of Languages & FaMaF University of C´ordoba Argentina Emiliano.Cuenca@unc.edu.ar Paula.Estrella@unc.edu.ar lbruno@unc.edu.ar Jonathan Mutal Sabrina Girletti Lise Volkart Pierrette Bouillon FTI/TIM, University of Geneva Switzerland Jonathan.Mutal@unige.ch Sabrina.Girletti@unige.ch Lise.Volkart@unige.ch Pierrette.Bouillon@unige.ch Abstract We believe that machine translation (MT) must be introduced to translation students as part of their training in preparation for their professional life. In this paper, we present a new version of the tool called MT 3, which builds and extends on a joint effort undertaken by the Faculty of Lan- guages of the University of C´ordoba and the Faculty of Translation and Interpret- ing of the University of Geneva to develop an open-source web platform to teach MT to translation students. We also report on a pilot experiment with the goal of test- ing the viability of using MT 3 in an MT course. The pilot allows us to identify ar- eas for improvement and collect feedback from students on the tool’s usability. 1 Introduction Machine translation (MT) has made enormous progress over the past few years with the devel- opment of neural systems (Koehn and Knowles, 2017), and the translation industry has therefore been increasingly integrating it into daily work- flow processes. This may have a direct impact on translators, who need to learn how to work with MT in a step called post-editing (O’Brien and Moorkens, 2014). It is therefore important to help translation students understand the underlying concepts of MT, including how MT systems are trained from bilingual corpora, what distinguishes systems from one another in terms of the algo- rithms they use, as well as the impact of their inter- nal functioning on translation quality and, hence, c⃝ 2020 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. on the post-editing task. Even if the idea be- hind these systems sounds simple (producing the most likely translation), it can be challenging to explain to non-tech savvy students how these sys- tems produce a final output. Currently, there are many open-source tools available to train statisti- cal and neural MT models (the most widely used are Moses (Koehn et al., 2007) for statistical MT and TensorFlow1 or OpenNMT (Klein et al., 2017) for neural MT) or commercial tools (such as Kan- tanMT2 or Microsoft Custom Translator3). How- ever, it is difficult to integrate them in the class- room because open-source tools are mostly de- signed for IT professionals, who do not need a graphical user interface (GUI), as opposed to trans- lation students; as for commercial tools, the costs are far too steep for some institutions. The Faculty of Languages of the University of C´ordoba (FL-UNC) and the Faculty of Translation and Interpreting of the University of Geneva (FTI- UniGe) have been collaborating since 2017 to de- sign and prototype a tool that will support lecturers in the classroom. In this paper, we present a new version of the tool called MT …

Content truncated to 500 words