Natural Language Prototypes for Analyzing Design Standards

CADLINE has abstract only. This paper addresses the use of natural language processing for acquiring, processing, and representing knowledge from design standards. A standard is a set of provisions providing principles, models, rules, limits, and particulars that are established by some authority for some purpose. In their textual form as written documents, design standards cannot directly be used in computer-aided design (CAD) systems. This paper demonstrates how standards can be transformed, using natural language processing techniques, from their textual form to alternative representations that more readily lend themselves to use in computer-aided design systems, supporting a variety of design applications. The language being transformed is the Building Officials and Code Administrators Building Code, one set of requirements that govern the design of buildings. Prototype computer subsystems have been developed that transform natural language sentences to case-grammar format and finally to subject-relationship- object triplets. The three prototypes that achieve these transformations are described: a parser, a semantic analyzer, and a query system. During one processing cycle, the system identifies that data items in a provision and the relationships between the data items. It also interacts with the user to add new data items to its knowledge bases, to verify data items found, and to add to its vocabulary. Alternatively, it responds to natural-language questions about the contents of the standard by identifying the relevant provisions within the standards. Processing formal documents requires knowledge about vocabulary, word-order, time, semantics, reference, and discourse. Despite the relative clarity of formal writing as it occurs in standards, the difficulties of implied responsibility, multiple meanings, and implied data items remain. A long-term research program at North Carolina State University has been defined that builds on these prototypes to further investigate knowledge acquisition and representation for standard