Automated production systems are complex, long-living mechatronic installations designed to meet modern production requirements. Their control software, typically implemented on PLCs according to IEC 61131, directly interacts with hardware components through input and output mappings. Over their lifetime, the control software must adapt continuously to evolving customer requirements and technical constraints. However, regulatory demands and best practices often mandate thorough documentation and verification of these changes, complicating the engineering process. This paper presents a human-machine collaborative approach to bridge the gap between natural language change requests and formalized change specifications for model-based verification. By combining recent advances in Large Language Models (LLMs) with expert knowledge and functional system descriptions, unformalized change requests are semi-automatically translated into verifiable specifications. This improves verification efficiency and communication between customers, suppliers, and technical experts. A human-machine software prototype streamlines this process by guiding users through the parsing of change requests, identifying semantic sub-requests, classifying changes into predefined patterns, and selecting the appropriate software units and variables for verifiable change specifications. Interaction with the software prototype is demonstrated through a real-world scenario, and suitable LLMs are identified in terms of their performance and quality. Expert feedback highlights its practical utility and suggests future improvements.