On the role of population heterogeneity in emergent communication

Populations have always been perceived as a structuring component for language to emerge and evolve: the larger the population, the more systematic the language. While this observation is widespread in the sociolinguistics literature, it has not been reproduced in computer simulations with deep reinforcement learning agents. In this paper, we thus aim to clarify this apparent contradiction. We thoughtfully explore emergent language properties by varying agent population size in the speaker-listener Lewis Game. After reproducing the experimental paradox, we challenge the simulation assumption that the agent community is homogeneous. To do so, we control the training speed or capacity of the agents by either asymetryzing the speaker-listeners properties or modifing each individuals within the population. We show that introducing such heterogeneities naturally sort out the initial paradox for larger communities start developing more systematic and structured languages.

• Machine Intelligence