Revisiting the Robustness of Watermarking to Paraphrasing Attacks