[3L3-GS-8] Robot and real worlds: symbol emergence / knowledge extraction

[3L3-GS-8-04] Dynamics verification based on the Metropolis-Hastings method of symbol emergence experiments in multiple humans

〇Ryota Okumura¹, Yoshinobu Hagiwara¹, Akira Taniguti¹, Tadahiro Taniguti¹ (1. Ritsumeikan University)

Keywords:Symbol emergence in robotics, Multi-agent, Human robot interaction

Symbol emergence is computationally modeled as an inter-personal multimodal categorization in a previous study. Inter-personal categorization consists of a language game played between two agents and is shown to lead the agents to an agreement about the usage of signs and categorization of objects. Although the model suggests that agents in symbol emergence systems should accept or reject other agents' signs based on the probability calculated by the Metropolis-Hastings algorithm, it is unclear whether people accept other people's proposals based on this acceptance probability.
This study examines whether actual people, i.e., human participants, follow the acceptance probability or not.
We designed a language game in which participants were asked to classify color stimuli generated from three multidimensional Gaussian distributions and to name the presented color stimuli to each other.
The results show that people often accept other person's opinions even when the inferred acceptance probability is close to zero. At the same time, we observed that the percentage of human acceptance is higher when the inferred acceptance probability is generally higher.