Computational Models of Language Evolution in Multi-agent Systems

EPSRC Grant GR/N01118

 

 

 

Summary     Objectives & Workplan       Background         Results        Publications        Software

 

 

Investigators

Angelo Cangelosi (Principal investigator)

Huck Turner (PhD student)

Davide Marocco (visiting PhD student)

Collaborators

Stevan Harnad (External advisor, University of Southampton - UQAM Montreal)

Domenico Parisi (CNR Rome)

Alberto Greco (Universita' di Genova, Italy)

Thomas Riga (now at University of Plymouth)

Barbara Giolito (Universita' Piemonte Orientale, Italy)

Summary

The aim of this research is to investigate and develop computational models of the origins and evolution of language and communication. The modelling approach mainly uses artificial neural networks, genetic algorithms and evolutionary robotics to simulate evolving populations of communicating autonomous agents. Some simulations also use an English parser model to study the emergence of syntactic universals under cognitive contraints. Various simulation programs have been developed and used for numerous experimental investigations. These were designed to investigate some of the biological, neural, and adaptive mechanisms that lead to the evolution of language. This research studies have significantly contributed to (a) develop new neural network models for the acquisition of categories through grounded symbols, (b) identify the sensorimotor basis of syntactic categories such as nouns and verbs, (c) investigate the evolution of optimal word orders under working menory constraints, (d) test Deacon's hypothesis on the co-evolution of language and brain structures and (e) identify different Baldwinian phenomena in language evolution. The research has potential applications in Robotics and Artificial intelligence (e.g. evolutionary robotics) for the development of adaptive and self-organising algorithms for communication in multi-agent and multi-robot systems. It is also of interest to the scientific community of Computational Neuroscience (e.g. new applications of synthetic brain imaging) and Cognitive Science (e.g. relationship between sensorimotor knowledge and linguistic categories, grounding of symbols in cognitive models). The simulation programs and the publications are available on the world wide web.