Affective Marking as a Bifurcation Point Between Irony and Deception in Evolved Communicators

How does non-literal language evolve, and what determines whether populations develop cooperative (ironic) versus exploitative (deceptive) communication strategies? We present an artificial life simulation in which populations of neural-network-controlled agents evolve communication strategies in a foraging environment with information asymmetry. Both irony and deception involve the same surface mechanism — signal inversion, where an agent uses a signal whose established meaning mismatches the true state of the world. We show that the presence of an affective channel (tone) acts as a bifurcation point: when agents can mark signals with tone, populations evolve intentional irony (tone-marked inversion that listeners interpret correctly); when the affective channel is absent, populations evolve deception instead (unmarked inversion that exploits listener credulity). Across 7 ablation conditions with 20 independent evolutionary runs each (500 generations, 50 agents), we find that: (1) conditions without an affective channel produce zero intentional irony and significantly elevated deception rates (p < 0.001); (2) above-baseline intentional irony emerges most reliably in populations with shared experience (p < 0.001 across all tone thresholds tested), consistent with Relevance Theory’s emphasis on common ground; and (3) irony displaces deception over evolutionary time, with irony events increasing from ~26 to ~70 per generation while deception decreases from ~43 to ~37 over 500 generations. These findings suggest that the capacity for affective expression is a prerequisite for the evolution of cooperative non-literal communication, and that irony and deception represent competing attractors in the space of non-literal signaling strategies.