DIALOGUES BETWEEN ADAM AND EVE: EXPLORATION OF UNKNOWN CIVILIZATION LANGUAGE BY LLM

WangXu, Fengzhou Wang, YiquanWang

Keywords: Language Emergence, Reinforcement Learning, Language Acquisition, Agent-based Simulation, Semantic Similarity

TL;DR: This study simulates language emergence in LLMs through dialogue between a rule-based generator (Adam) and a learner (Eve), revealing that reinforcement learning strategies enhance efficiency and mirror human language acquisition patterns.

Abstract: Language serves as an irreplaceable bridge for cultural communication, yet its origins and mechanisms remain largely uncharted territory. The emergence of intelligent agents empowered by large language model and NLP technology offers fresh approaches to investigate language understanding and generation across civilizations.This study constructs "Adam" and "Eve" agents that evolve through multi-scenario dialogues and iterative Q&A strategy learning, thereby elucidating novel language acquisition processes.Our framework reveals fundamental mechanisms of linguistic emergence, offering novel insights into intelligent interaction patterns during language development.

Submission Number: 10