GERBIL-NLG will be used as a leaderbooard for the WebNLG2020 challenge.

Disclaimer: Human evaluation is the primary evaluation for the WebNLG challenge

Please refer to the videos and text below to know How to use GERBIL-NLG


NLG task

Here you simple upload your hypothesis and upload your reference file for your own experiments.

WebNLG RDF2Text

Here you evaluate your hypothesis on the WebNLG 2020 development set and this evaluation goes to the leaderboard.
You simply upload your txt file.
Important Note: The reference data is a concatenation of all development files.
Basically, the ref data is a merge following the order from 1-N-triples file to 7-N-Triples files.
Thus, your uploaded hypothesis file has to follow the same order.

WebNLG Text2RDF

Here you need to provide your annotation following the format provided by Chris at Parsing script. Afterward, you will upload your results, this experiment is similar to NLG task in the sense of no leaderboard.