Parts of Speech-Grounded Subspaces in Vision-Language Models