From Scenario to Case Model: Generating CMMN Models from Natural Language with Large Language Models

Translating rich, user-centric narratives into formal process models remains a major challenge in model-driven engineering, particularly for flexible and adaptive workflows. This paper introduces a novel two-step method that leverages GPT4.0 to first enrich natural language scenarios with behavioral design principles and then generate executable CMMN models. We evaluate four prompt strategies across diverse e-commerce cases. Findings show that role-playing prompts effectively guide scenario enrichment, while a combined strategy integrating full guidance, few-shot examples, and role-playing produces the most accurate and semantically aligned CMMN models. This work lays the groundwork for LLM-driven, human-centric modeling and opens new directions for integrating cognitive insights into automated model synthesis.