Imagine navigating the complex web of legal texts governing public services. It's a daunting task for anyone, but what if AI could help? Researchers are exploring how Natural Language Processing (NLP) can automatically analyze these intricate documents, specifically focusing on German legal norms for administrative processes. They've put three different methods to the test: rule-based systems (classic symbolic AI), deep discriminative models (like the popular BERT), and deep generative models (the cutting-edge of AI text generation). The goal? To identify key elements within these texts, such as the main actors, required documents, deadlines, and conditions for service delivery. Surprisingly, the deep discriminative models came out on top, outperforming both the rule-based systems and the more advanced generative models in most categories. This is intriguing because it suggests that for highly structured, yet semantically and syntactically diverse texts like legal norms, the focused learning of discriminative models might be more effective than the broader knowledge base of generative AI or even carefully crafted human-made rules. However, the research also highlights the significant challenge posed by the "data field" category, which proved difficult for all models to accurately identify, suggesting that context and domain-specific knowledge play a crucial role. The future of this research lies in exploring hybrid approaches, perhaps combining the strengths of rule-based systems and deep learning models, or even integrating generative AI into the mix. This could lead to more robust and efficient tools for legal text analysis, ultimately streamlining public service administration and making access to vital services smoother for everyone.