A Survey on Bias and Fairness in Natural Language Processing