Neural Semantic Parsing of Building Regulations for Compliance Checking

Abstract Computerising building regulations to allow reasoning is one of the main challenges in automated compliance checking built environment. While there has been a long history translating manually, recent years, natural language processing (NLP) used support or automate this task. rule- and ontology-based information extraction transformation approaches have achieved accurate translations for narrow domains specific regulation types, machine learning (ML) promises increased scalability adaptability new styles. Since ML usually requires many annotated examples as training data, we take advantage code computerisation use corpus manually translated train transformer-based encoder-decoder model. Given relatively small corpus, model learns predict logical structure extracts entities relations reasonably well. translation quality not adequate fully process, shows potential serve an auto-completion system identify that need be reviewed.