Abstract
This study explores the interdependence of qualitative and quantitative analysis in articulating empirically plausible and theoretically coherent generalizations about grammatical structure. I will show that the use of large electronic corpora is indispensable to the grammarian''s work, serving as a rich source of semantic and contextual information, which turns out to be crucial in categorizing and explaining grammatical forms.
These general concerns are illustrated by the patterns of use of Czech relative clauses (RC) with the non-declinable relativizer CO, by taking a set of existing claims about these RCs and testing their accuracy on corpus material. The relevant analytic categories revolve around the referential type of the relativized noun, the interaction between relativization and deixis, and the semantic relationship between the relativized noun and the proposition expressed by the RC.
The analysis demonstrates that some of the existing claims are fully invalid in the face of regularly attested semantic distinctions, while others are more or less on the right track but often not comprehensive or precise enough to capture the full richness of the facts.