Automated Identification and Deconstruction of Penalty Clauses in Regulation

Regulations are often complex and therefore difficult to understand. Violations are known to occur oftentimes because of this lack of understanding. Regulations contain penalty clauses that impose some kind of punitive measures on the defaulting party in case of non-compliance. In this paper, we present a model for extracting and deconstructing penalty clauses. The model allows for capturing the essence of penalty clauses in terms of their critical attributes such as Trigger, Consequent Action, Penalized Entity, Subsequent Offence and Exception. We augment the extracted penalty attributes with relevant Definitions and render the deconstructed and augmented penalty clauses for a better understandability. We discuss the results of automating the identification and deconstruction of penalty clauses using a combination of rule-based techniques and Recurrent Neural Network based sequence-to-sequence model designed around the Penalty Model. The evaluation carried out on 39 publicly available regulation documents from sixteen countries indicate that our method can identify penalty clauses with a F1-score of 95% and extract 85% of the attribute instances correctly. This will help the requirement analysts to identify critical software requirements implied in penalty clauses, prioritize them, and thereby contribute to building compliant software systems.