Findings of the IWSLT 2022 Evaluation Campaign

Proceedings of the 19th International Conference on Spoken Language Translation (IWSLT 2022), pages 98 - 158 May 26-27, 2022 c⃝2022 Association for Computational Linguistics FINDINGS OF THE IWSLT 2022 EVALUATION CAMPAIGN Antonios Anastasopoulos George Mason U. Lo¨ıc Barrault Le Mans University Luisa Bentivogli FBK Marcely Zanon Boito U. Avignon Ondˇrej Bojar Charles U. Roldano Cattoni FBK Anna Currey AWS Georgiana Dinu AWS Kevin Duh JHU Maha Elbayad Meta Clara Emmanuel Apple Yannick Est`eve Avignon University Marcello Federico AWS Christian Federmann Microsoft Souhir Gahbiche Airbus Hongyu Gong Meta Roman Grundkiewicz Microsoft Barry Haddow U. of Edinburgh Benjamin Hsu AWS D´avid Javorsk´y Charles U. Vˇera Kloudov´a Charles U. Surafel M. Lakew AWS Xutai Ma JHU/Meta Prashant Mathur AWS Paul McNamee JHU Kenton Murray JHU Maria N˘adejde AWS Satoshi Nakamura NAIST Matteo Negri FBK Jan Niehues KIT Xing Niu AWS John Ortega Le Mans University Juan Pino Meta Elizabeth Salesky JHU Jiatong Shi CMU Matthias Sperber Apple Sebastian St¨uker Zoom Katsuhito Sudoh NAIST Marco Turchi FBK Yogesh Virkar AWS Alex Waibel CMU/KIT Changhan Wang Meta Shinji Watanabe CMU Abstract The evaluation campaign of the 19th Interna- tional Conference on Spoken Language Trans- lation featured eight shared tasks: (i) Simul- taneous speech translation, (ii) Offline speech translation, (iii) Speech to speech transla- tion, (iv) Low-resource speech translation, (v) Multilingual speech translation, (vi) Di- alect speech translation, (vii) Formality con- trol for speech translation, (viii) Isometric speech translation. A total of 27 teams partic- ipated in at least one of the shared tasks. This paper details, for each shared task, the pur- pose of the task, the data that were released, the evaluation metrics that were applied, the submissions that were received and the results that were achieved. 1 Introduction The International Conference on Spoken Lan- guage Translation (IWSLT) is the premier annual scientific conference for all aspects of spoken lan- guage translation. IWSLT is organized by the Spe- cial Interest Group on Spoken Language Trans- lation, which is supported by ACL, ISCA and ELRA. Like in all previous editions (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; Ansari et al., 2020; Anastasopoulos et al., 2021), this year’s conference was preceded by an evaluation campaign featuring shared tasks addressing scien- tific challenges in spoken language translation. This paper reports on the 2022 IWSLT Evalua- tion Campaign, which offered eight shared tasks: • Simultaneous speech translation, addressing low latency speech translation either streamed by a speech recognition (ASR) system or di- rectly from the audio source. The translation directions for both conditions are: English to German, English to Japanese, and English to Mandarin Chinese. • Offline speech translation, proposing speech 98 Team Organization AISP-SJTU Aispeech and Shanghai Jiao Tong University, China (Zhu et al., 2022) ALEXA AI Amazon Alexa AI, USA (Shanbhogue et al., 2022) ALEXA AI Amazon Alexa AI, USA (Zhang et al., 2022a) APPTEK AppTek, …

Content truncated to 500 words