A MULTI-AGENT SIMULATION OF THE EVOLUTION OF LANGUAGE Dimitar Kazakov, Mark Bartlett Department of Computer Science University of York Heslington, York YO10 5DD, UK Tel: +44 1904 43 4775/3378; fax: +44 1904 432767 e-mail: kazakov,bartlett@cs.york.ac.uk ABSTRACT This paper discusses the evolution of language as an emerging phenomenon with both genetic and social components that are shaped under evolutionary pressure. Communication between relatives is seen as an act of kinship-driven altruism and the chances of survival of such behavour discussed from a Neo- Darwinist point of view. The paper provides motivation for the use of multi-agent systems in the simulation of the evolution of language and describes one setup taking into account the above-mentioned issues. 1 INTRODUCTION In recent years, there has been much research carried out in attempting to model the evolution of language through computer simulation. Within this field, there are two somewhat disjoint problems being tackled. One set of researchers are investigating communication systems that are stored and passed between entities genetically [2, 10], while others are approaching the problem of learned communication systems [4, 7, 8]. A few are attempting to study the interaction between the two [1]. Our research is based within the domain of simulating learned communication systems and focusses on why any creature should develop or choose to use such a system to speak. More specifically, we present a framework in which the urge to use language is seen as an inherited feature selected by evolution, while language itself is a social phenomenon that is passed on through interaction rather than genetically inherited. Clearly an entity that is able to use such a communication system to understand the meanings of others' speach is at an advantage, as it can gain information through the work of other entities rather than its own toil. However, it is less apparent why a creature should choose to speak, when this will clearly give other creatures an advantage which this one has had to work hard to gain. 2 ALTRUISM AND NEO-DARWINISM In Darwinian terms, by helping other creatures with no obvious benefit to itself, the creature has acted altruistically decreasing its own fitness relative to that of others, and therefore we would expect such behaviour to be selected against by nature. However, the existence of human language clearly shows that in at least one case natural selection has acted opposite to this expectation. Researchers studying learned languages have not studied this question, but several researchers [6, 11] have looked at similar problems in the domain of innate communication systems and we look to this work for possible approaches. They have found that, in their most abstract models, communication does indeed seem to be selected against if an agent can choose not to speak without penalty. However, there are possible modifications to these systems that seem to encourage communication to occur. A spatial distribution is one such modification that can be applied, with agents interacting more with those spatially adjacent to them. This promotes reciprocal altruism, in which both entities benefit by cooperating rather than competing. Another possible explanation is to look at the issue of altruism from a Neo-Darwinian perspective. Hamilton [3] shows us that if we view the basic element of evolution not as the individual, but as the gene, we find that natural selection may actually favour selfless acts in the form of kinship-driven altruism. This form of altruism involves helping relatives proportionally to their degree of kinship to the altruistic entity. For instance, should such an entity die saving the lives of three of its children, there will probably be more copies of its genes remaining alive than if the creature had preserved its own life. Through this mechanism, a hypothetical gene promoting altruism would be able to spread itself. We propose to investigate whether either this form of altruism or reciprocal altruism can be used to explain the existence of learned language.