Natural language as well as other commu- nication forms are constrained by cognitive function and evolved through a social process. Here, we examine whether hu- man memory may be uniquely adapted to the social structures prevalent in groups, specifically small-world networks. The emergence of domain languages is simulated using an empirically evaluated ACT-R-based cognitive model of agents in a naming game played within communities. Several community structures are examined (grids, trees, random graphs and small-world networks). We present pre- liminary results from small-scale simulations, showing relative robustness of cognitive models to network structure.