Enhancing Software Requirements Quality: Ambiguity Detection and Resolution Using Large Language Models

Software Requirements Specification (SRS) is a crucial artifact in the software development lifecycle that bridges the gap between stakeholders (clients, users, developers) by establishing a common understanding of the software’s intended functionality and behavior. Different stakeholder interpretations could lead to misunderstandings about the project’s goals and requirements. Ambiguous requirements result in incorrect implementations and inconsistent designs, causing rework and inefficiencies. Addressing misunderstandings and rework due to ambiguity increases the overall project costs. Finding and properly understanding the ambiguous words can address these problems more efficiently. This paper presents a novel framework that uses large language models (LLMs) to identify, categorize, and resolve ambiguities in software requirements documents. This paper addresses seven types of ambiguity: semantic, syntactic, functional, operational, scope, temporal, and quality. By harnessing LLMs’ natural language understanding capabilities, the framework detects ambiguous statements and proposes unambiguous alternatives, potentially improving the quality and clarity of software requirements specifications. It ensures the final developed product meets the stakeholder’s needs and expectations, leading to higher satisfaction.