Automata networks model for alignment and least effort on vocabulary formation
This work addresses the problem of understanding language emergence in artificial systems for researchers in computational linguistics and multi-agent systems, but it is incremental as it builds on existing models like the Naming Game.
The study investigated whether artificial agent communities can develop language with Zipf law scaling by modeling vocabulary formation using Automata Networks under alignment and least effort strategies, finding that minimizing speaker effort on one-dimensional lattices best promotes shared language formation.
Can artificial communities of agents develop language with scaling relations close to the Zipf law? As a preliminary answer to this question, we propose an Automata Networks model of the formation of a vocabulary on a population of individuals, under two in principle opposite strategies: the alignment and the least effort principle. Within the previous account to the emergence of linguistic conventions (specially, the Naming Game), we focus on modeling speaker and hearer efforts as actions over their vocabularies and we study the impact of these actions on the formation of a shared language. The numerical simulations are essentially based on an energy function, that measures the amount of local agreement between the vocabularies. The results suggests that on one dimensional lattices the best strategy to the formation of shared languages is the one that minimizes the efforts of speakers on communicative tasks.