The article considers the range of possibilities of pragmatic markers (PM) annotation: from the speaker’s code to the speaker’s commentaries for all difficult cases. The research is based on the material of two corpora of everyday Russian speech – “One Day of Speech” (ORD; dialogues / polylogues) and “Balanced Annotated Text Collection” (SAT; monologues). Two main annotation levels have become the objects of research in this paper: the part of speech of the original lexical unit, from which the basic version of the PM derived (POS), and the model of formation for the PM which consist of more than one word (Model). The research shows the low feasibility of trying to fit PM into the system of traditional parts of speech, and, conversely, the importance and role of defining a model of formation of PM for their systematic description. In any case, the automatic annotation of corpus material turns out to be considerably difficult.