Using Transformers on Noisy vs. Clean Data for Paraphrase Identification in Mexican Spanish