A rationale-augmented NLP framework to identify unilateral contractual change risk for construction projects

Abstract

Construction projects often see owners who exert their dominance by modifying the standard contracts without contractors’ prior consent. This can lead to undesirable outcomes for contractors, thus referred to as unilateral contractual change risk (UCCR) in this study. As such, identifying UCCR proactively becomes essential for contractors, particularly in engineering-procurement-construction (EPC) projects where the claim scope is limited. Although natural language processing (NLP) has shown promise in the identification of UCCR, it faces difficulties due to the high cognitive demands required. To fill the gap, we propose a rational-augmented NLP framework that emulates human reasoning to identify UCCR in EPC contracts in an explainable and effective manner. The framework (1) leverages NLP techniques to disassemble contract text into features that draw the attention of human readers for obligation-related comprehension, and (2) generates a coherent sequence of intermediate reasoning steps using a customized Microsoft Excel and-in interface to identify predetermined categories of UCCR. Our framework, which achieves an outstanding F1 score of 0.87, is trained on a widely used standard form of EPC contract. It also presents a user-friendly interface for contractors to discern any intentional or malicious acts committed by the owner during the contract stage. Furthermore, our methodology can be adapted to enhance risk management in other sectors.