Cooperation among selfish individuals provides the fundamentals for social organization among animals and humans. Cooperation games capture this behavior at an abstract level and provide the tools for the analysis of the evolution of cooperation. Here we use the Rock-Paper-Scissors (RPS) game with positive and negative draw outcomes (i.e. when the draw outcome has a positive or negative impact on the players) to study the evolution of cooperative behavior in communities of simulated selfish agents. The agents communicate to each other using a probabilistic language and the cooperation game is set in an uncertain resource generation context. The offspring of the agents may clump together or may spread out, simulating the easy and difficult identification of possible cooperation partners. The results show that more uncertainty leads to more cooperation both in positive and negative draw games. Surprisingly we found that in negative draw games the level of cooperation is statistically significantly higher, although close to, the level that would be expected from random choice of RPS decisions. We also analyzed language complexity correlates of cooperation. The agent-based simulations and the results described here are applicable to social institutions or ecological systems with more than two, non-transitively comparable, decision states that can be described abstractly as RPS games.