Customising grammar and style checker rules

Abstract

Abstract In order to make grammar and style checkers customizable to meet writers' individual or organisational house style needs, complex rules specifying how to recognise and replace undesirable forms must be modified by non-expert users. Attempts in current commercial systems to provide such a facility are unsatisfactory: given the notations used to represent rules in these systems, any system that is powerful enough to perform its basic task of grammar and style checking is too complex to be comprehensible to a rule writer. This paper argues that any system with adequate natural language processing (NLP) resources to perform the basic tasks of a grammar and style checker can be augmented with a rule definition facility which, largely making use of those same resources, would be radically more usable than any existing system. The proposed approach is crucially dependent on the modular representation of system knowledge and incorporates techniques from knowledge representation, human-computer interaction and machine learning.