Why emergentist accounts of cognition are more theoretically constraining than structured probability accounts: comment on Griffiths et al. and McClelland et al.