ISTIC’s Neural Machine Translation System for IWSLT’2020

FINDINGS OF THE IWSLT 2020 EVALUATION CAMPAIGN Ebrahim Ansari Charles U./IASBS Amittai Axelrod DiDi Labs Nguyen Bach Alibaba Ondˇrej Bojar Charles U. Roldano Cattoni FBK Fahim Dalvi QCRI Nadir Durrani QCRI Marcello Federico Amazon AI Christian Federmann Microsoft Research Jiatao Gu Facebook AI Fei Huang Alibaba Kevin Knight DiDi Labs Xutai Ma JHU/Facebook AI Ajay Nagesh DiDi Labs Matteo Negri FBK Jan Niehues Maastricht U. Juan Pino Facebook AI Elizabeth Salesky JHU Xing Shi DiDi Labs Sebastian St¨uker KIT Marco Turchi FBK Alex Waibel CMU/KIT Changhan Wang Facebook AI Abstract The evaluation campaign of the International Conference on Spoken Language Translation (IWSLT 2020) featured this year six chal- lenge tracks: (i) Simultaneous speech transla- tion, (ii) Video speech translation, (iii) Offline speech translation, (iv) Conversational speech translation, (v) Open domain translation, and (vi) Non-native speech translation. A total of 30 teams participated in at least one of the tracks. This paper introduces each track’s goal, data and evaluation metrics, and reports the results of the received submissions. 1 Introduction The International Conference on Spoken Lan- guage Translation (IWSLT) is an annual scien- tific conference (Akiba et al., 2004; Eck and Hori, 2005; Paul, 2006; Fordyce, 2007; Paul, 2008, 2009; Paul et al., 2010; Federico et al., 2011, 2012; Cettolo et al., 2013, 2014, 2015, 2016, 2017; Niehues et al., 2018, 2019) for the study, development and evaluation of spoken lan- guage translation technology, including: speech- to-text, speech-to-speech translation, simultane- ous and consecutive translation, speech dubbing, cross-lingual communication including all multi- modal, emotional, para-linguistic, and stylistic as- pects and their applications in the field. The goal of the conference is to organize evaluations and sessions around challenge areas, and to present scientific work and system descriptions. This pa- per reports on the evaluation campaign organized by IWSLT 2020, which features six challenge tracks: • Simultaneous speech translation, address- ing low latency translation of talks, from En- glish to German, either from a speech file into text, or from a ground-truth transcript into text; • Video speech translation, targeting multi- modal speech translation of video clips into text, either from Chinese into English or from English into Russian • Offline speech translation, proposing speech translation of talks from English into German, using either cascade architectures or end-to-end models, able to directly translate source speech into target text; • Conversational speech translation, target- ing the translation of highly disfluent conver- sations into fluent text, from Spanish to En- glish, starting either from audio or from a ver- batim transcript; • Open domain translation, addressing Japanese-Chinese translation of unknown mixed-genre test data by leveraging hetero- geneous and noisy web training data. • Non-native speech translation, considering speech translation of English-to-Czech and English-to-German speech in a realistic set- ting of non-native spontaneous speech, in somewhat noisy conditions. The challenge tracks were attended by 30 par- ticipants (see Table 1), including both academic and industrial teams. This correspond to a signifi- cant increment with respect to the last year’s eval- uation campaign, which saw the participation …

Content truncated to 500 words