Efficient Elicitation of Annotations for Human Evaluation of Machine Translation