The Cultural Self-organisation of Cognitive Grammar


This is an international, multi-disciplinary research project on cultural selforganisation in the origins of grammar. The project brings together members from key disciplines relevant to modelling the evolution of language: linguistics, neuroscience, psychology, artificial intelligence, and complex systems. The main hypothesis of the project is that language has emerged and continues to evolve as a cultural selectionist system. This means that generic cognitive abilities, such as associative memory, structural processing, categorisation, etc. have been recruited for language, and that the language system complexifies and changes in the process of cultural transmission. Coherence of a language in a community and the spreading of new features is hypothesised to be due to self-organisation.

This functionalist and cultural evolution hypothesis is clearly in radical opposition to the popular hypothesis of a genetic encoding and origin of the "language instinct". However it is standard scientific practice to examine opponent hypotheses and to gather evidence for each. It is only by examining the cultural evolution hypothesis seriously that we can see the necessity for a strong genetic determination of language, or the degree thereof. For the project we have chosen a number of phenomena for which a large amount of cross-linguistic data exists and for which grammaticalisation processes have been studied. They will concern case, aspect, and determiner systems. The language processing, cognitive architecture, and sensori-motor components required for these phenomena will be studied and simulations will be set up that show how language constructs can emerge. We intend to show that the investigated linguistic phenomena can be explained in terms of cognitive constraints, task constraints, constraints on human brain architecture, and embodiment, rather than through innate genetic determinism.
The primary aim of the Brussels AI-Laboratory in this project is to set up a number of concrete experiments in which artificial, robotic agents develop autonomously a shared communication system that has characteristics of human natural language. The robots will do this by engaging in a series of language games about situations in the real world, which they perceive, categorise, and conceptualise. The robots will be given the necessary sensori-motoric abilities, as well as components for the use, acquisition and invention of language. Each experiment should show a qualitative correspondence with established, empirical observations in natural language, and thus show the sufficiency of the components introduced for getting the desired language behaviour. We realise that this aim is very ambitious but recent advances in all sub-fields concerned, particularly in the domain of humanoid robotics, give grounds for optimism. Moreover, a series of experiments in the domain of lexical evolution have already shown the way. This project will carry further these prior results and the used methodology, involving both computer modelling and robotic modelling, into the domain of grammar. One important spin-off of the project will be a set of tools for carrying out experiments in language evolution that will be made accessible to the research community at large. These tools include computational frameworks for modelling evolving grammars, a "teleportation" infrastructure for carrying out experiments at multiple sites, and various kinds of monitoring tools that quantify language evolution, such as communicative success, coherence, spreading or cumulative change.


Collaborating research groups

  • VUB AI Lab Brussels, Prof. Luc Steels
    The VUB AI Lab will be responsible for the main robotic set-up on which other groups will be able to execute experiments. They will also work on the computational framework for the grammar and the execution of various experiments. Particularly, the experiments for the emergence of a determiner and aspect system will be studied.
  • Institut des Sciences Cognitives (Lyon, France), Dr. Peter F. Dominey
    This group will be responsible for introducing neural realism into the agent's cognitive architectures in order to examine precisely which architectural characteristics might be responsible for certain universal tendencies of grammar. The group will also examine conceptual categorisation in some specific domains, particular the categorisation of action.
  • Department of Communication Science (Sienna, Italy), Prof. Castelfranchi
    This group will be responsible on the one side for developing the theoretical foundations of evolutionary language games by linking to models developed in economics and biology; on the other side for a model of the evolution of tacit conventions, implicit communication and agreements as a basis for "negotiating" linguistic rules.
  • Institut für Afrikanistik (Cologne, Germany) / Anglistic III (Düsseldorf, Germany), Prof. Heine, Prof. Kouteva
    This group's role in the project will be to catalog universal properties of case systems, aspect systems, and determiners and to provide information on grammaticalisation processes as observed in the world's languages. These results will be a target for the computer simulations and robotic experiments carried out by the other partners.
  • Cognitive Science (Lund, Sweden), Dr Balkenius The Lund team will be carrying out experiments in grounded language evolution on robotic agents, in strong collaboration with the Brussels team. Some of the components they have already developed are crucial for developing complete agent systems.

Associated research groups

  • Max Planck Institute for Evolutionary Anthropology (Leipzig, Germany), Prof. Tomasello
  • Departamento de Filología Inglesa (Murcia, Spain), Dr. Javier Valenzuela
  • Max Planck Institute for Psycholinguistics (Nijmegen, Netherlands), Prof. Bowerman


Involved members: