One thing that is problematic, for me at least, is that unlike phonology which is pretty well understood and for which workable models in historical change exist, semantics has never been formalized in a way that allows one to investigate one lnaguage standing still in time, let alone track change over the period of 40K to 100K years.