MORE | Spring 2023

Emergent Communication in Context-Dependent Scenarios: A Novel Approach to Language Creation

Data icon, disabled. Four grey bars arranged like a vertical bar chart.

This project investigates the role of context-dependent scenarios in shaping emergent communication among agents that learn to interact without any pre-established linguistic system. A novel method to model context-dependent scenarios using multi-agent reinforcement learning is proposed, and the influence of various factors on the emergence of compositional, coherent, and adaptive communication is examined. The implications of these results for natural language evolution and co-creation are also discussed.

Student researcher

Ethan Clark

Computer science

Hometown: Tempe, Arizona, United States

Graduation date: Fall 2023