Cross-Modal Taxonomic Generalization in (Vision-) Language Models