Evaluating Character Attacks