Dynamic Analysis of Metropolis Hastings Naming Game by Experimental Semiotics

Abstract

How people invent and share signs through communication is a central question in the study of symbol emergence systems. A previous study examined human communication through a subject experiment and found that higher acceptance probabilities, generated using the Metropolis-Hastings method in a category formation task, led to higher overall acceptance rates by humans. However, the previous work was unable to statistically determine whether humans accepted names given to them by others with such probabilities. The aim of this paper is to investigate whether humans accept or reject a given name based on the acceptance probabilities generated by the Metropolis-Hastings method. We conducted asked participants to play a naming game, observing their acceptance and rejection judgments, and compared them with various computational models of human judgment. Our findings suggest that among the models tested, the mode using the Metropolis-Hastings method most closely approximates how humans make acceptance/rejection judgments in this task.