First WMT shared task on sign language translation


27/07/2022 Upon request we are extending the submission deadline by 1 week. See the updated schedule.

21/07/2022 Submissions are open on OCELoT.

14/07/2022 The source side of our test data is now released. The submission platform will take slightly longer, it is scheduled to open on July 21.

05/07/2022 New development data is now public.

04/07/2022 We now provide an automated Tensorflow datasets loader for our training data.

04/07/2022 A second call for participation is now public.

30/06/2022 We clarified the participation and submission procedure.

12/06/2022 All our training data can now be accessed.

10/06/2022 We released code for baseline systems.

09/06/2022 A description of our data sets is now public.

09/06/2022 One of our training corpora, FocusNews, can now be downloaded from Zenodo! We apologize for the delay.

Shared task in a nutshell

This shared task is concerned with automatic translation between signed and spoken languages. The task is novel in the sense that it requires processing visual information (such as video frames or human pose estimation) beyond the well-known paradigm of text-to-text machine translation.

We provide training data for one language pair: Swiss German Sign Language (DSGS) and German (DE).

Translation from DSGS to German is our primary translation direction: submitted systems are ranked on a leaderboard and we will provide baseline systems. Systems translating from German to DSGS are not ranked on the leaderboard while the task is running, but we still encourage participants to submit such systems.

We aim to provide human evaluation for all submitted systems, regardless of the translation direction.

We will ask participants to submit a system description paper to WMT 2022.