Sometimes the author of a PDT 2.0 text uses a word incorrectly - e.g. a name of a woman as a man's name etc. In such cases, the real usage should be annotated, not the should-be usage.
The texts can contain errors. It is reasonable to correct some of them (but the original - errorneous - word form should always be preserved in the
origf attribute). However, only low-level errors (spelling and morphology) should be corrected. We do not want to correct Engels' text into Heidegger's. Never replace a colloquial form with an official one (e.g. *zelené města → zelená města, *bez noh → bez nohou), even if the analyzer does not know the form.