Better than Random: Reliable NLG Human Evaluation with Constrained Active Sampling

Open in new window