Automated regulatory compliance checking requires automated extraction of requirements from regulatory textual documents and their formalization in a computer-processable rule representation. Such information extraction (IE) is a challenging task that requires complex analysis and processing of text. Natural language processing (NLP) aims to enable computers to process natural language text in a human-like manner. This paper proposes a semantic, rule-based NLP approach for automated IE from construction regulatory documents. The proposed approach uses a set of pattern-matching-based IE rules and conflict resolution (CR) rules in IE. A variety of syntactic (syntax/grammar-related) and semantic (meaning/context-related) text features are used in the patterns of the IE and CR rules. Phrase structure grammar (PSG)-based phrasal tags and separation and sequencing of semantic information elements are proposed and used to reduce the number of needed patterns. An ontology is used to aid in the recognition of semantic text features (concepts and relations). The proposed IE algorithms were tested in extracting quantitative requirements from the 2009 International Building Code and achieved 0.969 and 0.944 precision and recall, respectively.

To fully automate regulatory compliance checking of construction projects, regulatory requirements need to be automatically extracted from various construction regulatory documents and then transformed into a formalized format that enables automated reasoning. To address this need, the authors propose an approach for automatically extracting information from construction regulatory textual documents and transforming them into logic clauses that could be directly used for automated reasoning. This paper focuses on presenting the proposed information transformation (ITr) methodology and the corresponding algorithms. The proposed ITr methodology utilizes a rule-based, semantic natural language processing (NLP) approach. A set of semantic mapping (SeM) rules and conflict resolution (CoR) rules are used to enable the automation of the transformation process. Several syntactic text features (captured using NLP techniques) and semantic text features (captured using an ontology) are used in the SeM and CoR rules. A bottom-up method is leveraged to handle complex sentence components. A consume and generate mechanism is proposed to implement the bottom-up method and execute the SeM rules. The proposed ITr algorithms were tested in transforming information instances of quantitative requirements, which were automatically extracted from the International Building Code 2009, into logic clauses. The algorithms achieved 98.2 and 99.1% precision and recall, respectively, on the testing data.

Existing automated compliance checking (ACC) efforts are limited in their automation and reasoning capabilities; the state of the art in ACC still uses ad hoc reasoning schema/methods, with lack of support for complete automation in ACC reasoning. First-order logic (FOL) representation and reasoning can provide a generalized reasoning method to facilitate complete automation in ACC reasoning. This paper presents a new FOL-based information representation and compliance reasoning (IRep and CR) schema for representing and reasoning about regulatory information and design information for checking regulatory compliance of building designs. The schema formalizes the representation of regulatory information and design information in the form of semantic-based (ontology-based) logic clauses that could be directly used for automated compliance reasoning. Two alternative subschemas, following a closed-world assumption and an open-world assumption for noncompliance detection, respectively, were proposed and tested. The proposed IRep and CR schema was tested in representing and reasoning about quantitative regulatory requirements in Chapter 19 of the International Building Code 2009 and design information of a two-story duplex apartment test case in two ways, using perfect information and imperfect information. The closed-world assumption subschema was selected based on performance results; it achieved 100% recall and precision in noncompliance detection using perfect information and 98.7% recall and 87.6% precision in noncompliance detection using imperfect information.