Starting from language origins theories, which connect prelinguistic primate abilities such as deixis, to modern linguistic systems, we seek to let emerge universals of human languages such as dispersion principles and the quantal aspect of speech. For this aim, we model a society of Bayesian sensory-motor agents, which interact and evolve in an environment filled with objects they can identify. We show how these agents may converge on a common phonological code for communication. This enables us to validate our assumptions and suggest probabilistic models of language origins theories.ISSP