Grammar Development for Czech Syntactic Parser with Corpus-based Techniques

In the paper, we present the description of the Czech syntactic parser synt developed at FI MU NLP laboratory. The presented system is based on the meta-grammar formalism with a head-driven chart parser. The parsing technique provides fast analysis of the context free backbone with successive evaluation of the contextual constraints using so called “forest of values.” The meta-grammar formalism allows to capture complicated grammatic relations with a maintainable number of rules. Besides the description of the synt system, we display the process of the meta-grammar development. One of the first phases is formed by construction of corpus data for testing. In the paper, we demonstrate the exploitation of the corpus on testing a method for detection of the “best analysis” selection with the results of testing the synt analysis on Czech corpus.