MMPE: A Multi-Modal Interface for Post-Editing Machine Translation

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1691–1702 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 1691 MMPE: A Multi-Modal Interface for Post-Editing Machine Translation Nico Herbig1, Tim D¨uwel1, Santanu Pal1,2, Kalliopi Meladaki1, Mahsa Monshizadeh2, Antonio Kr¨uger1, Josef van Genabith1,2 1German Research Center for Artificial Intelligence (DFKI), Saarland Informatics Campus, Germany 2Department of Language Science and Technology, Saarland University, Germany {firstname.lastname}@dfki.de {firstname.lastname}@uni-saarland.de Abstract Current advances in machine translation (MT) increase the need for translators to switch from traditional translation to post-editing (PE) of machine-translated text, a process that saves time and reduces errors. This affects the design of translation interfaces, as the task changes from mainly generating text to cor- recting errors within otherwise helpful trans- lation proposals. Since this paradigm shift of- fers potential for modalities other than mouse and keyboard, we present MMPE, the first prototype to combine traditional input modes with pen, touch, and speech modalities for PE of MT. The results of an evaluation with professional translators suggest that pen and touch interaction are suitable for deletion and reordering tasks, while they are of limited use for longer insertions. On the other hand, speech and multi-modal combinations of se- lect & speech are considered suitable for re- placements and insertions but offer less poten- tial for deletion and reordering. Overall, partic- ipants were enthusiastic about the new modali- ties and saw them as good extensions to mouse & keyboard, but not as a complete substitute. 1 Introduction As machine translation (MT) has been making sub- stantial improvements in recent years1, more and more professional translators are integrating this technology into their translation workflows (Zaret- skaya et al., 2016; Zaretskaya and Seghiri, 2018). The process of using a pre-translated text as a basis and improving it to create the final translation is called post-editing (PE). Older research showed a strong dislike of translators towards PE (Lagoudaki, 2009; Wallis, 2006), and more recent studies agree that translators are still cautious about PE and ques- tion its benefits (Gaspari et al., 2014; Koponen, 1WMT 2019 translation task: http://matrix.statmt.org/, ac- cessed 16/04/2020 2012), partly because they see it as a threat to their profession (Moorkens, 2018). Experienced translators in particular exhibit rather negative atti- tudes (Moorkens and O’Brien, 2015). Conversely, novice translators have been shown to have more positive views on PE (Yamada, 2015). Green et al. (2013) demonstrated that some translators actually strongly prefer PE and argue that “users might have dated perceptions of MT quality”. Apart from translators’ preference, productivity gains of 36% when using modern neural MT for PE (Toral et al., 2018) already result in substan- tial changes in translation workflows (Zaretskaya and Seghiri, 2018) and will probably continue to do so the better MT becomes. Thus, PE requires thorough investigation in terms of interface design, since the task changes from mostly text produc- tion to comparing and adapting MT and translation memory (TM) proposals, or put differently, from control to supervision. Previous elicitation-based research (Herbig et al., …

Content truncated to 500 words