Parts of Speech–Grounded Subspaces in Vision-Language Models | Read Paper on Bytez