Communications with Learning Errors

Abstract

In the real-life scenario of human communications, both the local communication process of agents and the information propagation on the underlying network affect the achievement and speed of global convergence. For example, if agents can learn fast and correctly from local communications, and thereafter teach their neighbors to effectively learn the same, then the entire population would be able to reach global convergence efficiently; or, if the acquaintanceship of agents is simple, meaning that the underlying network is simple, then the transmitted information would be propagated efficiently over the entire network. Here, a simple underlying network is one with good connectivity (large average degree), but with less local-clustered structures (low clustering coefficient). However, realistically, language acquisition is always error-prone. This problem in human language leads to ambiguities with learning errors in human conversations, thereby degrading the effectiveness of human communications. Interestingly, it was suggested in [1] that learning errors can actually increase diversity of the linguistic system by introducing additional information. Thus, learning errors are able to help prevent the linguistic system from being trapped in sub-optimum states, beneficial for the evolution of a more efficient language. Here, the linguistic system is evaluated by a function of payoff. It was also found in [1] some thresholds of the learning error rate for certain models, where if the error rate is below the threshold then the system gains advantage from learning errors; otherwise, if the error rate is above the threshold then the errors or mistakes will impair the system, e.g., reducing significantly the payoff of the linguistic system. Moreover, noise may lead to recurrently converging states of a Markov chain model, which is considered beneficial for better detecting social interactions [2]. Therefore, errors or noise may affect the language system positively, to some extent, as in the two cases mentioned above.