Analysis of the occurrence of hierarchical communication in dilemma games

Abstract

Humans typically use communication systems with linguistic ambiguity. Recently, artificial agents interacting with humans have been actively developed, although it is still difficult to engage in human-like communication. We consider that a deeper understanding of human language use is beneficial for developing artificial agents with advanced communication abilities. This study utilized a dilemma environment to create a complex social model and observed the occurrence of novel language based on experimental semiotics. We analyzed lexical ambiguity in the sense of connotation and denotation, which is calculated as the strength of the one-to-one relationship between the message tokens and their meaning. Furthermore, we analyzed syntactic ambiguity by adopting different chunking methods. The results of our analysis suggest the advantage of such quantitative analysis showing connections between the emergence of hierarchical communication and the intentions of the participants who generated the message.