Generative language models are now capable of producing impressive texts, but they often rely on implicitly learned information that is difficult to trace and may be outdated. Especially in complex knowledge domains, this can lead to factual errors, a lack of transparency, and reduced trust. WG2Text investigates how explicit knowledge from semantic knowledge graphs can be used for text generation. The aim is to combine structured, verifiable, and updatable information from knowledge graphs with language models in such a way that fact-based, understandable, and traceable texts can be generated. The project develops and compares approaches for transforming knowledge graphs into natural language texts and demonstrates the results in a working demonstrator. In this way, WG2Text contributes to more transparent and reliable language models as well as to improved knowledge communication across different application domains.