Perhaps it's because the representations are fractured. The link above is to the transcript of an episode of Machine Learning Street Talk with Kenneth O. Stanleyabout The Fractured Entangled Representation Hypothesis[1]