Do language models have coherent mental models of everyday things?