Evolve a language. It has been repeatedly shown in computer simulations and robotic experiments that aspects of language can evolve through self-organisation by a group of agents who interact with each other, and who invent and adopt aspects of the language such as words, meanings and grammatical rules, see, e.g., (Xxxxxxx 2002; Xxxxxxxxx & Xxxxxx 2001, Xxxxx 2002, Steels 1997) for overviews. Following the models outlined by Xxxxxx (2003) and Xxxx (2003a), meanings will be grounded from the interaction of the agents with their environment by (1) detecting features about the events they are experiencing, (2) categorising these, and (3) associating the categories with existing or newly invented words. The features agents can detect relate to the features of objects and actions. These features are categorised into categories stored in the agents’ memories, or into newly constructed categories in case of failure, thus expanding their memories. This way, agents develop their own private repertoire of categories used to form the meanings of the language. The categories are to be associated with words and become meaningful in the way they are used by the population. If the categories are novel to agents who wish to communicate, the agents can invent new words. Sharing is done by communication and learning from each other. It has been shown in Xxxx and Xxxxxxx (2003) that learning is most efficient if learners learn a language from experienced language users, as modelled by the iterated learning model (Xxxxx & Xxxxxxx 2001). Therefore our population will consist of adult agents and learners. Largely unexplored in language grounding is the emergence of grammatical structures. The core innovation of the language evolution model we intend to implement relates to the emergence of compositional structures in a grammar that is grounded. This part will be largely based on studies investigating the origins of syntactic structures with predefined semantics (e.g., Brighton 2003, Xxxxx 2001) and a recent prototype on grounding compositional structures (Xxxx 2003b). One of the main issues will be scaling the existing models. It is expected that this will be partly solved by the distributed nature of the environment and the spatial organisation of the agents themselves. This will cause the agents to communicate locally, thus evolving local languages or dialects. The more agents will spread out, the more the languages will distribute over the population, and possibly put pressure on the need to c...
