We are not running a separate multilingual task this year, but multilingual models are allowed and encouraged for this year’s offline and low-resource tasks!

Offline task

For the offline task this year, participants may submit for up to three target languages and several pretrained models allowed, many of which are multilingual models. Which submissions use multilingual models will be noted in the overview paper, and analysis will be done to see if there are any systematic differences between those submissions which are multilingual and monolingual. Check out the task page for more information.

Low-Resource task

For the low-resource track, there will be three language pairs and any data except the provided evaluation sets are allowed for training, with multilingual and pretrained models encouraged. What is most important is to see which resources and modeling techniques are needed to improve the target task. Check out the task page for more information.

Please have a look at these tasks, and let us know on the IWSLT google group if you have any questions!

Contact: Elizabeth Salesky