The models of statistical physics used to study collective phenomena in someinterdisciplinary contexts, such as social dynamics and opinion spreading, donot consider the effects of the memory on individual decision processes. On thecontrary, in the Naming Game, a recently proposed model of Language formation,each agent chooses a particular state, or opinion, by means of a memory-basednegotiation process, during which a variable number of states is collected andkept in memory. In this perspective, the statistical features of the number ofstates collected by the agents becomes a relevant quantity to understand thedynamics of the model, and the influence of topological properties onmemory-based models. By means of a master equation approach, we analyze theinternal agent dynamics of Naming Game in populations embedded on networks,finding that it strongly depends on very general topological properties of thesystem (e.g. average and fluctuations of the degree). However, the influence oftopological properties on the microscopic individual dynamics is a generalphenomenon that should characterize all those social interactions that can bemodeled by memory-based negotiation processes.