ChatGPT in the Loop: A Natural Language Extension for Domain-Specific Modeling Languages

This paper presents an approach to no-code development based on the interplay of formally defined (graphical) Domain-Specific Languages and informal, intuitive Natural Language which is enriched with contextual information to enable referencing of formally defined entities. The paper focuses on the use and automated integration of these enriched intuitive languages via ChatGPT-based code generation to exploit the best of both language paradigms for domain-specific application development. To compensate for the lack of control over the intuitive languages we apply automated system-level validation via automata learning and subsequent model checking. All this is illustrated using the development of point-and-click adventures as a minimal viable example.