In addition to their tendency towards exaggeration, biological signals also tend to be quite distinct from each other. This is not just true when compared across species and contexts, but also between intraspecific displays of very similar function, such as threat displays. General biases in recognition mechanisms may be responsible for this effect. Previous artificial neural network simulations have demonstrated that perceptual biases in the receiver may cause a coevolving signal to polarize away from another, non-evolving, stimulus. In this paper we extend this work by investigating networks which respond differently to several different stimuli. We show that two relevant signals, ideally producing different responses, also evolve towards converse forms despite being processed through the same network. We found no evidence that attractive signals share common attractive properties. Our results can not be accounted for by existing strategic models of communication.