A Method of Accurate Robust Parsing of Czech

Abstract

This paper advocates the claim that the property of a robustness of a certain automatic natural language parser is something different than a simple ability to construct a syntactic structure for each sequence of word forms (sentence) of a given language.

The robustness in our terminology should be more accurate in a sense that it should be able to distinguish between “good” and “bad” ill-formed sentence. We propose to use two measures for this purpose, the node-gap complexity which describes the complexity of the sentence with regard to nonprojective constructions, and the degree of robustness which takes into account the number of syntactic inconsistencies encountered in the process of robust parsing. These measures make it possible to develop a scale of global constraints which allow a kind of gradual parsing of both syntactically well-formed and ill-formed sentences of a natural language.