The question of how referential choice and interpretation are inuenced by production cost remains unresolved in the literature. A recent paper (Rohde et al., 2012) investigates the conditions under which speakers choose to coordinate using low-cost but ambiguous expressions by conducting a number of experiments in which participants played an iterated referential coordination language game. This dissertation takes a novel approach to modeling referential coordination by simulating Rohde et al.'s results using particle swarm optimization (PSO), a general-purpose optimization method for non-differentiable problems with continuous search spaces. Two PSO-based models are presented, one of which is shown to perform well against a baseline model. Predictions from the more favourable of the two models are presented for several variants of the Rohde et al. language games, extrapolating from the results of the original studies. This model is also shown to partially replicate observed findings of persistent entrainment on lexical forms, even when a changing discourse context causes these forms to become overinformative (Brennan & Clark, 1996). The results from the PSO model are taken to demonstrate that dyadic referential coordination can be framed as a constrained optimization problem in which agents do not need to maintain an explicit representation of the common ground or of each other - a finding in keeping with egocentric accounts of communication from the literature (Horton & Keysar, 1996).