Towards Transforming Free Natural Language Text to Conceptual Models with Generative AI

Abstract: The current systems engineering state-of-the-practice is to start the lifecycle of a system by providing a system requirements document written freely in natural language (NL) that serves as a basis for the downstream lifecycle stages of modeling, analysis, and design. Model-based systems engineering (MBSE) advocates using a conceptual model or a set of related models as the single source of truth—the backbone of the ensuing analysis and design of the contemplated system. Converting the NL requirements to a model is a daunting task and is an inescapable stage in the MBSE value chain. Since conceptual modeling requires time and intellectual engineering resources, it is often skipped at a price to be paid later. Large language models (LLMs) can facilitate text-to-model conversion, significantly reducing this resource-consuming human effort. Automating the process of building system models from free text can save precious systems engineers time, improving their ability to spot flaws and mistakes in the textual specification requirements prior to system architecting and design. To enable text-to-model conversion, we describe NL2OPL, a task of converting single NL sentences into semantically equivalent sentences in Object-Process Language (OPL)—a well-defined subset of English that serves as the textual modality of Object-Process Methodology (OPM) ISO 19450:2024. This work is a first step towards automating conceptual model construction from NL text. When fully realized, this process can have a profound positive effect on the way systems engineering will be conducted.