WMT23: General translation shared task

[Tom Kocmi]

Hello,

 

we are happy to announce WMT23 General translation shared task. Please, help us by filling the poll [1] about pretrained models.

The main changes for this year are:

 

• Human evaluation is going to be professional only (no Mechanical Turks)
• Hebrew – English is the main low-resource LP
• German – English is going to be document-level (no sentence breaking)
• Czech – Ukrainian is the main non-English LP
• Submission process will be more transparent to avoid missing submissions
• Redefinition of constrained/unconstrained track

 

With the recent progress with pretrained models and LLMs, we want to redefine the constrained/unconstrained tracks to level the field.

We would like to know the opinion of the community on this topic, please, help us filling this short poll even if you don’t plan to participate [1]

 

Thank you and good luck with building the winning model,

Tom

(in Germany, he/him)

 

[1] https://docs.google.com/forms/d/e/1FAIpQLSdHmQav6ZD0HqLA7ODEq_vJ4YqqNbz5slWlo4Zf_roKEn03EQ/viewform