ELITR: European Live Translator

ELITR: European Live Translator Ondˇrej Bojar1 Dominik Mach´aˇcek1 Sangeet Sagar1 Otakar Smrˇz1 Jon´aˇs Kratochv´ıl1 Peter Pol´ak1 Ebrahim Ansari1 Dario Franceschini2 Chiara Canton2 Ivan Simonini2 Thai-Son Nguyen3 Felix Schneider3 Sebastian St¨uker3 Alex Waibel3 Barry Haddow4 Rico Sennrich4 Philip Williams4 1Charles University, 2PerVoice, 3Karlsruhe Institute of Technology, 4University of Edinburgh Coordinator email: bojar@ufal.mff.cuni.cz Abstract ELITR (European Live Translator) project aims to create a speech translation system for simultaneous subtitling of conferences and online meetings targetting up to 43 languages. The technology is tested by the Supreme Audit Office of the Czech Re- public and by alfaview®, a German online conferencing system. Other project goals are to advance document-level and mul- tilingual machine translation, automatic speech recognition, and meeting summa- rization. 1 Description ELITR (European Live Translator, elitr.eu) is a three-year EU H2020 Research and Innovation Programme running from 2019 to 2021. The con- sortium consists of Charles University, University of Edinburgh, Karlsruhe Institute of Technology (research partners), PerVoice (integrator) and alfa- training (user partner). 2 Objectives ELITR objectives are research and innovations in the field of spoken language and text translation and automatic summarization of meetings. 2.1 Simultaneous Subtitling In ELITR, we aim to develop a system for simulta- neous subtitling of conferences and online meet- ings. Our affiliated user partner is the Supreme Audit Office of the Czech Republic. It is hosting a congress of EUROSAI (European Organization © 2020 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. of Supreme Audit Institutions). The congress par- ticipants are natives of 43 languages, and many of them have difficulties in understanding any of the six congress official languages, into which it is in- terpreted by humans, or to understand some non- native accents. For this and other similar cases, we develop a simultaneous speech translation sys- tem from 7 spoken languages (English, German, Russian, Italian, French, Spanish, and experimen- tally Czech) subtitling into 43 languages, including those for which a human interpreter would not be available for capacity reasons. The 43 languages are 24 EU official languages and 19 others, spoken between Morocco and Kazachstan. With our other user partner, alfatraining, we connect our system with an online meeting plat- form, alfaview®. 2.2 Other Research Topics The most visible application goal of live subti- tling is supported by our advancements in the re- lated areas. We research into document-level ma- chine translation to enable conference participants to translate documents between all the 43 lan- guages in high-quality, taking inter-sentential phe- nomena into account (Voita et al., 2019a; Voita et al., 2019b; Vojtˇechov´a et al., 2019; Popel et al., 2019; Rysov´a et al., 2019). We research into multilingual machine transla- tion to reduce the cost of targeting many languages at once, and to leverage multiple language variants of the source for higher quality (Zhang et al., 2019; Zhang and Sennrich, 2019). To face challenges of simultaneous translation, such as robustness to noise, out-of-vocabulary words, domain adaptation, and non-standard ac- cents (Mach´aˇcek et al., 2019), …

Content truncated to 500 words