Adversarial Training Objectives for Generative Attacks on Text Classifiers