Shape similarity, better than semantic membership, accounts for the structure of visual object representations in a population of monkey inferotemporal neurons