Towards Taming Large Language Models with Prompt Templates for Legal GRL Modeling

The Legal Goal-oriented Requirements Language (Legal GRL) is a promising conceptual modeling approach for supporting regulatory compliance analysis. Yet, despite early attempts at automation, such Legal GRL models are still manually created, being time consuming and error prone. Recent work has demonstrated how Large Language Models can support the creation of conceptual models. Although showing promise, the application scenarios for conceptual modeling are often limited to well structured, and scoped, scenarios. Dealing with practical, less controlled, regulatory analyses, whereby often a particular actor or topic needs to be pulled into focus, is an open issue. In this paper, we propose using prompt templates to structure the process of using LLMs to create a Legal GRL model from text. The core idea is that prompt templates are created from state of the art prompt patterns, which can restrict LLM output, can manage a LLM conversation context, and can structure a LLM conversation. We report on an initial assessment of prompt templates on multiple law articles from the healthcare and energy community domains. Our initial results are promising for Legal GRL modeling, but at the same time show that caution is warranted.