Proceedings of the Annual Meeting of the Cognitive Science Society

Communication in social learning is often limited by time, attention, and explicit penalties, yet many models assume free or fixed exchange. We study how two bundled communication regimes shape networked inference in an agent-based model. Groups of agents on a ring network infer a binary hidden state over 10 rounds from noisy private signals and neighbors' probabilistic messages. The regimes jointly vary per-broadcast penalty, broadcast probability, per-round broadcast cap, and social-update weight. Simulations show that the high-cost, communication-constrained regime produces fewer outgoing broadcasts, slower reduction of logged-belief dispersion, and stronger dependence of final accuracy and score on private-signal reliability. Increasing signal strength therefore yields larger gains in the high-cost regime than in the low-cost regime. These results should be interpreted as regime-level differences rather than as the isolated causal effect of message penalty alone, and they generate testable predictions for human experiments.