LIP6 2000/022
- Reports
L'émergence d'un lexique dans une population d'agents autonomes - F. Kaplan
- 295 pages - 06/26/2000- document en - http://www.lip6.fr/lip6/reports/2000/lip6.2000.022.pdf - 4,696 Ko
- Contact : Frederic.Kaplan (at) nulllip6.fr
- Ancien Thème : OASIS
- Keywords : multi-agent simulation, language evolution, self-organisation
- Publisher : Valerie.Mangin
The goal of this research is to understand how to bootstrap a cultural phenomenon, in our case the emergence of a lexicon, in a population of artificial agents undergoing some strong constraints of realism. The agents can only use words and gestures to communicate, and initially they know neither words nor categories. Given such assumptions, how can they collectively build a lexicon to name real world objects ? How can the words of this lexicon be associated to shared meanings when the agents are neither perceiving nor categorizing the world in the same way ?
Our approach consists in studying experimentally a succession of increasingly complex models. In the most simple ones, the artificial agents are only composed of an associative memory and the environment is just a set of discrete symbols. In the most complex ones, the agents are grounded into physical robots with visual perception and interact about their perception of the real world. At each step, we identify the collective dynamics enabling lexicon self-organisation. We show how this lexicon manages to be culturally transmitted from one generation to the next one, how it gets modified, more accurate and adapted to describe the real or virtual environment that the agents perceive. We show how the structure of the lexicon becomes more regular, easier to learn, easier to transmit.
Results arising from this research may prove to be of importance in two different domains. From a linguistic point of view, studying the collective dynamics that account for lexicon emergence in a population of artificial agents can lead to relevant hypotheses to understand natural languages evolution and human language origin. From a technological point of view, studying the same dynamics can lead to new techniques for human-machine communication and interactions between robots.
Our approach consists in studying experimentally a succession of increasingly complex models. In the most simple ones, the artificial agents are only composed of an associative memory and the environment is just a set of discrete symbols. In the most complex ones, the agents are grounded into physical robots with visual perception and interact about their perception of the real world. At each step, we identify the collective dynamics enabling lexicon self-organisation. We show how this lexicon manages to be culturally transmitted from one generation to the next one, how it gets modified, more accurate and adapted to describe the real or virtual environment that the agents perceive. We show how the structure of the lexicon becomes more regular, easier to learn, easier to transmit.
Results arising from this research may prove to be of importance in two different domains. From a linguistic point of view, studying the collective dynamics that account for lexicon emergence in a population of artificial agents can lead to relevant hypotheses to understand natural languages evolution and human language origin. From a technological point of view, studying the same dynamics can lead to new techniques for human-machine communication and interactions between robots.