Emergence in a population of agents of a lexicon based on an individual conceptualization - HAL-SHS - Sciences de l'Homme et de la Société Accéder directement au contenu
Poster De Conférence Année : 2002

Emergence in a population of agents of a lexicon based on an individual conceptualization

Résumé

This work deals with the emergence of a lexicon based on an individual conceptualization of the world in a population of agents. The multi-agents paradigm seems well suited for the investigation of the emergence of language. Previous works have shown that, language games, i.e. simple well-defined, communicative interactions, permit populations of agents to develop shared realistic phonological systems, elements of syntax or lexicon. In most previous attempts (Steels, 2000), words used by agents were to name single object of the world, single agent or single feature. The conceptual structure presented here is richer; we can name an object a “hammer”, a “tool” or an “artifact” according to our purpose. In this study, agents construct an individual representation of world that they use to communicate not only about the objects of the world, but also about categories of objects. Conceptual spaces have been proposed by Gärdenfors (2000) to be a framework for modeling the formation and the evolution of concepts. They are multi-dimensional geometric spaces. The different dimensions, named quality dimension, correspond to the different ways objects can be judged similar. Within the conceptual space, particular points are the center of spheres that represent prototype of categories. Classical realistic semantics states that meaning is a mapping from a word into an object or a relation between objects of the world. In contrast, cognitive semantics argue that words are mapped into mental entities. Gärdenfors propose that concepts as modeled in conceptual spaces are good candidates for these mental entities. On the basis of these considerations, we constructed a world composed of objects and agents endowed with conceptual spaces, initially empty, and a lexicon that associates each prototype with one or more words (associations are weighted with a score). Agents play a modified version of a language game known as the naming game. In the game they play, an agent, the initiator, chooses a set of objects, and communicates his choice to another agent, the recipient, in a non linguistic way. Then the recipient categorizes the objects according to his representation of the world by determining which prototype is the closest of the gravity center of the projections of the objects in his conceptual space. The concept represented by the prototype is more or less abstract, depending on the similarity of the objects. The recipient indicates to the initiator how he calls the concept, and the initiator, determines whether he agrees with the word user by the recipient. Depending of the issue of the game, both agents modify their lexicon and/or their conceptual space: they may modify the score of the associations, acquire a new word or a new concept, or modify the structure of their conceptual space. The results of the simulations show that although agents develop their own representation of the world in their conceptual space, a shared lexicon emerge from the population and allows the agents to communicate not only about the objects, but also about categories that regroup objects and/or other categories.

Mots clés

sma
Fichier non déposé

Dates et versions

halshs-01174447 , version 1 (09-07-2015)

Identifiants

  • HAL Id : halshs-01174447 , version 1

Citer

Jean-Philippe Magué. Emergence in a population of agents of a lexicon based on an individual conceptualization. Meeting of the European Society for Philosophy and Psychology, Jul 2002, Lyon, France. 2002. ⟨halshs-01174447⟩
73 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More