Lingxiao Kong

Backdooring Rationalization

Rationalization model has recently garnered significant attention for enhancing the interpretability of natural language processing by first using a generator to select the most relevant pieces from the text with respect to the label, before passing the text input to the predictor. However, the robu