Yes, however, there is no way to connect this to human-friendly categories of meaning, since (1) the method is unsupervised and (2) it might be difficult to imp

Yes, however, there is no way to connect this to human-friendly categories of meaning, since (1) the method is unsupervised and (2) it might be difficult to impose that sort of supervision on the training, as human semantic meaning is often difficult to encode.