Human Annotation

We Need to Consider Disagreement in Evaluation

Evaluation should model meaningful disagreement instead of reducing every task to a single gold label.