Drawing with Strangers: Population Scaling Drives Zero-Shot Mutual Intelligibility in Emergent Sketching

2026-06-09

AI Summary

This work investigates how populations of independent agents without prior interaction can achieve zero-shot mutual intelligibility (ZMI), proposing ZMI as a novel generalization dimension in emergent communication. Employing a vision-based sketch communication paradigm, the study systematically examines the impact of population size on ZMI through large-scale multi-agent training, perceptual similarity analysis, and cross-population variability metrics. Experimental results demonstrate that scaling up the training population significantly enhances ZMI performance: intra-population communication diversity increases while inter-population divergence diminishes, with emergent symbols becoming more tightly anchored to objective visual features of target images. These findings reveal a scale-driven structural convergence mechanism underlying the emergence of mutually intelligible communication systems.

Abstract

Generalization in emergent communication has largely focused on novel inputs or linguistic structures, yet the capacity for agents to communicate with strangers from strictly disjoint communities remains relatively unexplored. In this work, we formalize this capability as \textit{zero-shot mutual intelligibility (ZMI)}: successful communication between independently trained populations without prior exposure. Leveraging emergent sketching -- in which agents communicate through sets of drawn strokes -- as a visually grounded modality, we find that scaling the training population substantially improves ZMI across independent groups. Crucially, as we scale the population size, in-group communicative variation increases, preventing co-adaptation into homogeneity. Simultaneously, cross-group variation decreases, indicating a structural convergence toward a certain type of universality. Further analysis reveals that this universality is achieved through perceptual grounding: scaled populations increasingly anchor their emergent sketches on the objective visual resemblance of the target images. Together, these results position ZMI as a distinct axis of generalization in emergent communication and suggest a route toward socially interoperable artificial agents.

Problem

• zero-shot mutual intelligibility
• emergent communication
• population scaling
• sketching
• generalization

Innovation

• zero-shot mutual intelligibility
• emergent sketching
• population scaling
• perceptual grounding
• structural convergence