Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models | Read Paper on Bytez