Signal Soup: Sharedness vs. code alignment

This paper introduces Signal Soup, a computational model exploring how communication can emerge among multiple agents without assuming they share identical codes. Unlike traditional models that focus on pairs of agents exchanging signals with explicit feedback, Signal Soup models stigmergic coordination: agents interact indirectly by modifying and responding to signals in a shared environment. Each agent learns associations between perceived signals and produced responses through simple pattern matching. Simulations reveal two remarkable emergent regularities: agents stabilize at approximately 5.6 sign functions (well below their memory capacity), and the type-token ratio converges to 1/3 regardless of population size. While sign functions become partially shared across agents, the majority remain idiosyncratic. The model suggests that sharedness—partial rather than complete code alignment—is sufficient for stable communicative interaction, indicating that loose, distributed alignment may be relevant to robust communication systems.