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

[3L3-GS-8-01] Multimodal Symbol Emergence with Lexical Synthesis

〇Kazuma Furukawa¹, Yoshinobu Hagiwara¹, Akira Taniguchi¹, Tadahiro Taniguchi¹ (1. Ritsumeikan University)

Keywords:symbol emergence in robotics, intelligent robotics, artificial intelligence

Several computational models have been proposed for symbol emergent systems, which organize symbolic systems used by multi-agents for communication and cognition. However, these models only deal with the emergence of a word among agents, and cannot deal with compositionality, which is a characteristic of human language. The compositionality is the property that the meaning of a sentence depends on its constituent words and their order. In this study, we propose a computational model of a multi-agent system that emerges a symbol system with compositionality. The proposed model was developed by combining the model of cross-situational learning, which learns word sequences by assuming categories for each modality, with the model of symbol emergent systems. The experimental results of interpersonal cross-modal inference demonstrated that the proposed model emerges a symbol system with compositionality and has higher accuracy than baseline in the task of predicting unknown patterns from a sentence.