Gramatika v korpusu, korpus v gramatice: příspěvek k diskusi o vyhledávání gramatické informace v korpusech

This article discusses some aspects of searching for grammatical information in corpora. It argues that any search procedure must consist of at least three principally different steps. First, a hypothesis regarding some grammatical property of the language system must be formulated in terms of an available 'tagging' menu. Second, general instructions concerning the sample size, relevant context size, etc. must be stated, and only then can the third step, i.e. the proper search and interpretation of the attested data, be taken. Examples from the Czech National Corpus are offered to show that the boundary between grammaticality and non-grammaticality of a phenomenon or category is represented by a probability scale with more than just two opposing values and that the corpus may serve as an important tool for locating the most probable (favorite) point on the scale. The issue of zero or non-zero occurrence of a phenomenon is discussed in greater detail. It is argued that if no example of a phenomenon is attested in the corpus, it does not necessarily follow that the corpus is too small and that it is necessary or significant to intervene in favor of a larger one.