Abstract
The article tries to advocate the fact that although the field of rule-based syntactic analysis of natural languages has recently been practically taken over by data-driven (mostly stochastic) methods. it is still important both for the theory of formal description of natural languages as well as for certain types of applications, especially those dealing with an ill-formed input. The arguments are based upon the experience gained in the process of development a pilot implementation of a grammar checker of Czech.
Although the methods described in the paper did not lead directly to a commercial application, they definitely increased the level of understanding of certain complicated linguistic phenomena, namely the phenomenon of grammaticality and non-projectivity of Czech sentences.