Using a neural network simulation of a series of language training experiments with chimpanzees, the difference between indexical and symbolic interpretation is explored. From the results of the simulation follows a discussion about the systemic requirements for crossing the symbolic threshold and how the primacy of icons applies to computational models.