It is an interesting question to answer how the society groups its way towards efficient equilibrium in an imperfect world when self-interested agents learn from others. In this paper, we focus on mutual learning. Each agent learns the rule of interaction in the negotiation situations formulated as hawk-dove games. It is known the mixed Nash strategy of hawk-dove games will result in an inefficient equilibrium. In this paper we consider both mimicry and crossover as the methodology of individual learning. We show that all agents mutually learn to behave as doves, which result in social efficiency. We also investigate the meta-rules acquired by agents through mutual learning. With mimicry the meta-rules of all agents are categorized into a few meta-rules. On the other hand, with crossover, almost all agents have acquired different meta-rules with one common feature.