3 présentations

• 
  16h10 - 16h35

  Further evolution of natural categorization systems: A new approach to evolving color concepts

  • Maryam Gooyabadi, prés., University of California Irvine

  A dynamic model of language evolution is applied to the naturally occurring color categorizations of 110 linguistic communities taken from the World Color Survey (Berlin & Kay 1969) in order to assess the stability of their respective color categorizations. The evolutionary dynamics is modeled after human communication and specified by Komorova et al.’s Discrimination-Similarity and 2-player teacher games
  (Komarova & Jameson 2007), where color-naming systems are evolved to stable equilibria through agent interactions. This approach remedies the sparseness of empirical, time-series data – in some cases impossible to attain – that would be ideal for studying natural evolution trends in real human communities by broadly approximating such evolutionary processes. Results suggest that our simulations uphold the original integrity of each community’s categorizations – did not impose an external structure – but are still able to evolve each system to a stable equilibrium. Further, our simulation-based approach offers insights in the relative maturity of color concepts within each linguistic population that was previously lacking from analyses of static data. The stability of color categorization can be used as a criterion for grouping different language communities together for the purposes of cross-population analysis. More broadly, we demonstrate that dynamic models initialized with static data can provide valuable insight and have real implications for research across a large variety of fields.

• 
  16h35 - 17h00

  Hierarchical Models for the Evolution of Compositional Language

  • Calvin Cochran, prés.,
  • Jeffrey Barrett, UC-Irvine
  • Brian Skyrms, UC-Irvine

  We present three hierarchical models for the evolution of compositional language. Each has the basic structure of a two-sender/one receiver Lewis signaling game augmented with executive agents who can learn to influence the behavior of the basic senders and receiver. With each game, we move from stronger to weaker modeling assumptions. The first game shows how the basic senders and receiver might evolve a compositional language when the two senders have pre-established representational roles. The second shows how the two senders might coevolve representational roles as they evolve a reliable compositional language. Both of these games impose an efficiency demand on the agents. The third game shows how costly signaling alone might lead role-free agents to evolve a compositional language.

• 
  17h00 - 17h25

  Evolutionary branching of forgiveness in the iterated Prisoner's Dilemma

  • Taneli Pusa, prés., Inria / University of Lyon
  • Tadeas Priklopil, University of Lausanne

  We study the evolution of forgiveness in a population of individuals playing the iterated Prisoner’s Dilemma game.
  We focus on a probabilistic TFT-like strategy where mistakes happen in implementing one’s moves, in which case cooperative interactions are followed by harmful cooperation-defection cycles. Interacting players may, however, forgive the defective move of their opponent and recover cooperative interactions.
  We study the adaptive dynamics of forgiveness and give analytical conditions when forgiveness is selected for.
  We find that both the likehood of mistakes and the cost of cooperation influence the level of forgiveness the population will evolve to.
  Most interestingly, we find conditions, in terms of payoffs, where forgiveness undergoes evolutionary branching thus leading to more than one level of forgiveness in the population.