Article révisé par les pairs
| Résumé : | The relation between Lewis' signaling game framework and Steels' language game framework is discussed. The problem of pooling equilibria in signaling games is approached from both angles. Previous results about the requirements on learning for escaping pooling equilibria and achieving convergence are refined, and it is discussed why Roth Erev learning with forgetting should be a good learning rule in this context. This is confirmed empirically in simulation. |
