Investigating Large Language Models' Linguistic Abilities for Text Preprocessing