EVALUATION

IWSLT 2020 will feature five challenges that address the following focus areas:

  • End-to-end SLT models (Chair: M. Turchi, FBK) – overcoming the cascaded approach
  • Simultaneous translation (Chair: J. Gu, Facebook) – low latency, real-time interpretation
  • Conversational speech translation (Chair: E. Salesky, JHU) – highly disfluent speech
  • Open domain translation (Chair: A. Nagesh, Didi Labs) – new domains and languages
  • Non-native speech translation (Chair: O. Bojar, Charles U.) – non-native disfluent speech

Training and development data for each challenge will be released. Organizers will provide baseline component systems to participants that do not have access to an individual component. Human and automatic scoring will be performed on each challenge, published and discussed. An opportunity for secondary system submissions will be given in the week following primary submissions.