I was going to say the same thing about information extraction (e.g. NER). Not all words need to be assigned any class. So if you fail to extract a "person" name, it would be a false negative for the person class but it would not also be a false positive for any other class.