EN
The aim of the article is both to point to the descriptive character of a majority of corpus linguistic analyses and to argue that this character (which is manifested by the classification, sorting or labelling of language data) represents a limit in corpus linguistic research. Further, an experimental approach (in the sense of empirical testing of a hypothesis) is proposed as a possible way of overcoming this limit. Finally, some methodological aspects of current state of corpus linguistics, namely the notion of representativeness and the interpretation of quantification, are critically discussed.