Learning and Equilibrium Refinements in Signalling Games