Yes. You may notice that texts often include many special characters, abbreviations, numbers, etc. And, for example, numbers can be pronounced differently depends on the context. It is especially actual for languages like Russian, where words can be modified by tense, number, gender, etc. It is not very hard for a human to understand how to pronounce such abbreviations. But for machine learning models it is a hard task. That's why we need to unfold all these abbreviations before doing machine learning. This is called text normalization.