Do Visual-Language Maps Capture Latent Semantics?