Text this: Towards a benchmark dataset for large language models in the context of process automation