Computational linguistics

A text classifier removes every punctuation mark during preprocessing. What is a sound concern about this choice?