Towards Efficient Machine Translation Evaluation by Modelling Annotators