Gravar-mail: Human Evaluation of Models Built for Interpretability