Abstract For many adaptive complex systems information about the environment is not simply recorded in a look-up table, but is rather encoded in a theory, schema, or model, which compresses information. The grammar of a language can be viewed as such a schema or theory. In a prior study [Teal et al., 1999] we proposed several conjectures about the learning and evolution of language that should follow from these observations: (C1) compression aids in generalization; (C2) compression occurs more easily in a “smooth”, as opposed to a “rugged”, problem space; and (C3) constraints from compression make it likely that natural languages evolve towards smooth string spaces. This previous work found general, if not complete support for these three conjectures. Here we build on that study to clarify the relationship between Minimum Description Length (MDL) and error in our model and examine evolution of certain languages in more detail. Our results suggest a fourth conjecture: that all else being equal, (C4) more complex languages change more rapidly during evolution.

We have evolved artificial neural networks to control the wandering behavior of small robots. The task and environment were very simple—to touch as many squares in a grid as possible during a fixed period of time. A number of the simulated robots were embodied in a small Lego™ robot, controlled by a Motorola™ 6811 processor; and their performance was compared to the simulations. We observed that: (a) evolution was an effective means to program the robot's behavior; (b) progress was characterized by sharply stepped periods of improvement, separated by periods of stasis that corresponded to levels of behavioral/computational complexity; and (c) the simulated and realized robots behaved quite similarly, the realized robots in some cases outperforming the simulated ones. Introducing random noise to the simulations improved the fit somewhat (from r = 0.73 to 0.79). Hybrid simulated/embodied selection regimes for evolutionary robots are discussed.