The code for our entry on Task 11 SemEval 2023 - Learning With Disagreements
See https://semeval.github.io/ for a description of the tasks.
Our contribution is a comparision between different approaches, to see what worked and what did not. This includes ensemble techniques, multi-task learning, gaussian processes, use of statistical features, different loss functions and additional datasets.
In the end, our best approach was a multi-task learning scheme which included soft loss.