Text this: An exploratory study of interpretability for face presentation attack detection