"Yeah Right!" -- Do LLMs Exhibit Multimodal Feature Transfer?