The prevailing current view of a (symbolic, computational) grammar is basically that of a set of rewriting rules using feature-structured categories. However, whenever such a grammar is aimed at development of a real-world applied project, at least two disadvantages become clear. First, it breaks with the traditional understanding of a grammar as a network of phenomena (such as agreement, subcategorization, etc.), thus impeding the (direct) incorporation of this knowledge into such grammars. Second, a realistic grammar is inevitably huge and simultaneously contains very complex interdependencies among rules. This makes any modularization of grammar engineering (aka division of labour within a team) and above all maintaining and debugging realistic grammars a virtually impossible task. This paper presents an alternative view of formal (computational) grammars of natural language allowing for smooth modularization of the grammar-writing process and hence for meeting the pressing task of distributed grammar-development. The examples of both problems and their solutions are related to grammars in HPSG style, however, the problems discussed are in no way HPSG-specific, just on the contrary, they indeed concern any approach making use of feature-structure categories.
展开▼