Using Punctuation as an Adversarial Attack on Deep Learning-Based NLP Systems: An Empirical Study F i n d i n g s - E A C L