The Evolution of Behavioural and Linguistic Skills to Execute and Generate Two-word Instructions in Agents Controlled by Dynamical Neural Networks

This paper illustrates an agent-based simulation model focused on the acquisition of linguistic skills. Populations of simulated agents controlled by dynamical neural networks are trained by artificial evolution to perfom two tasks: the behaviour-production task which consists in accessing and executing linguistic instructions; and the behaviourrecognition task which consists in linguistically recognising behaviours. During training the agent experiences only a subset of all linguistic instructions/behaviours. Trained agents successfully acquire an ability to perform both tasks. Moreover some of the successfull agents proved to be able to access and execute also linguistic instructions not experienced during training. However, none of the successfull agents manage to linguistically recognise behaviours corresponding to the execution of linguistic instructions not experienced during training. We conclude by speculating on potential factors that may have inhibited the agents from developing fully compositional semantics structures.