Text this: Validation of LLM-Generated Object Co-Occurrence Information for Understanding Three-Dimensional Scenes