Abstract
This study investigates to what extent more advanced joint attentional mechanisms, rather than only shared attention between two agents and an object, influence the results of language games played by these agents. The simulations show that adding constructs that mimic follow attention capabilities substantially increases the performance of agents in these language games. Using follow and direct attention mechanisms, but without Theory of Mind-like capabilities, the agents are able to learn a shared lexicon much faster than when using only checking attention or corrective feedback. These results support the hypothesis that language evolution and evolutionary Theory of Mind develop in a co-evolutionary way, and that joint attentional skills are necessary and sufficient prerequisites for both.
