Collection Mémoires et thèses électroniques
Accueil À propos Nous joindre

Dans cette thèse, nous proposons un cadre unifié pour la pragmatique et la sémantique de la communication entre agents logiciels. La pragmatique traite la façon dont les agents utilisent les actes communicatifs lorsqu’ils participent aux conversations. Elle est liée à la dynamique des interactions entre agents et à la manière avec laquelle les actes individuels sont reliés pour construire des conversations complètes. La sémantique, quant à elle, est intéressée par la signification de ces actes. Elle établit la base pour une signification concise et non ambiguë des messages échangés entre les agents. Ce cadre unifié vise à résoudre trois problèmes majeurs dans le domaine de communication entre agents :

1- L’absence d’un lien entre la pragmatique et la sémantique.

2- L’inflexibilité des protocoles actuels de communication entre agents.

3- La vérification des mécanismes de communication entre agents.

Les contributions principales de cette thèse sont :

1- Une approche pragmatique formelle basée sur les engagements sociaux et les arguments.

2- Un nouveau formalisme pour la communication entre agents appelé Réseau d’Engagements et d’Arguments.

3- Un modèle logique définissant la sémantique des éléments utilisés dans l’approche pragmatique.

4- Une technique de vérification de modèles basée sur une sémantique à tableaux pour vérifier une famille de protocoles flexibles de communication entre agents appelée protocoles à base de jeux de dialogue.

5- Un nouveau protocole de persuasion à base de jeux de dialogue.

L'idée principale de notre approche pragmatique est que la communication entre agents est modélisée comme des actions que les agents accomplissent sur des engagements sociaux et des arguments. La dynamique de la conversation entre agents est représentée par cette notion d’actions et par l’évolution de ces engagements et arguments. Notre formalisme (Réseau d’Engagements et d’Arguments) basé sur cette approche fournit une représentation externe de la dynamique de communication entre agents. Ce formalisme peut être utilisé par les agents comme moyen pour participer à des conversations d’une manière flexible parce qu’ils peuvent raisonner sur leurs actes communicatifs en utilisant leurs systèmes d’argumentation et l’état actuel de la conversation.

Notre modèle logique est une sémantique, à base d’un modèle théorique, pour l’approche pragmatique. Il définit la signification des différents actes de communication que nous utilisons dans notre approche pragmatique. Il exprime également la signification de quelques actes de discours importants dans le contexte de communication multi-agents et il capture la sémantique des arguments annulables. Ce modèle logique permet d’établir le lien entre la sémantique et la pragmatique de communication entre agents.

Nous traitons le problème de vérification des protocoles à base de jeux de dialogue en utilisant une technique de vérification de modèles basée sur une sémantique à tableaux. Ces protocoles sont spécifiés sur la base de notre modèle logique. Nous montrons que notre algorithme de vérification offre une technique, non seulement pour vérifier si le protocole à base de jeux de dialogue (le modèle) satisfait une propriété donnée, mais également si ce protocole respecte la sémantique des actes communicatifs.

Notre protocole de persuasion à base de jeux de dialogue est spécifié dans le contexte de notre cadre unifié en utilisant un langage logique. Il est implémenté en utilisant une programmation logique et un paradigme orienté-agent. Dans ce protocole, le processus décisionnel des agents est basé sur les systèmes d’argumentation et sur la notion de crédibilité des agents.

In this thesis, we propose a unified framework for the pragmatics and the semantics of agent communication. Pragmatics deals with the way agents use communicative acts when conversing. It is related to the dynamics of agent interactions and to the way of connecting individual acts while building complete conversations. Semantics is interested in the meaning of these acts. It lays down the foundation for a concise and unambiguous meaning of agent messages. This framework aims at solving three main problems of agent communication:

1- The absence of a link between the pragmatics and the semantics.

2- The inflexibility of current agent communication protocols.

3- The verification of agent communication mechanisms.

The main contributions of this thesis are:

1- A formal pragmatic approach based on social commitments and arguments.

2- A new agent communication formalism called Commitment and Argument Network.

3- A logical model defining the semantics of the elements used in the pragmatic approach.

4- A tableau-based model checking technique for the verification of a kind of flexible protocols called dialogue game protocols.

5- A new persuasion dialogue game protocol.

The main idea of our pragmatic approach is that agent communication is considered as actions that agents perform on social commitments and arguments. The dynamics of agent conversation is represented by this notion of actions and by the evolution of these commitments and arguments. Our Commitment and Argument Network formalism based on this approach provides an external representation of agent communication dynamics. We argue that this formalism helps agents to participate in conversations in a flexible way because they can reason about their communicative acts using their argumentation systems and the current state of the conversation.

Our logical model is a model-theoretic semantics for the pragmatic approach. It defines the meaning of the different communicative acts that we use in our pragmatic approach. It also expresses the meaning of some important speech acts and it captures the semantics of defeasible arguments. This logical model allows us to establish the link between the semantics and the pragmatics of agent communication.

We address the problem of verifying dialogue game protocols using a tableau-based model checking technique. These protocols are specified in terms of our logical model. We argue that our model checking algorithm provides a technique, not only to verify if the dialogue game protocol satisfies a given property, but also if this protocol respects the underlying semantics of the communicative acts.

Our persuasion dialogue game protocol is specified in our framework using a logical language, and implemented using a logic programming and agent-oriented programming paradigm. In this protocol, the agents’ decision making process is based on the agents’ argumentation systems and the notion of agents’ trustworthiness.

First, I would like to warmly thank my thesis supervisor, Bernard Moulin, for his invaluable support and patience, for his precious and never lacking enthusiasm for research and for his continuous assistance in preparing and writing this thesis. I also want to thank him for having so strongly believed in me and provided me with insights which helped me solve many of the problems I encountered in my research. I also acknowledge him for assisting me financially during my Ph.D and for his generosity which gave me opportunities to attend several international conferences and to visit Utrecht University in order to validate my work with the members of Intelligent Systems Group.

I would like to express my gratitude to my co-supervisor, Brahim Chaib-draa, whose expertise, understanding and patience, added considerably to my graduate experience. I want to thank him for highly stimulating discussions (which generally took place in the department corridor) that have led to the discovery of many results in this thesis and for having given me the opportunity to collaborate with many researchers from Utrecht University, Université Paris 9 Dauphine and University of Melbourne.

I want to thank Yves Lespérance from York University (Canada) for having accepted to evaluate the first version of this thesis. I would like to thank him for his very interesting comments and suggestions that allowed me to improve the quality of this thesis. I appreciated the many discussions that I have had with him about the different parts of this dissertation.

I also wish to gratefully thank Frank Dignum from Utrecht University (the Netherlands) and Josée Desharnais from Laval University (Canada) for having accepted to evaluate my thesis and participate in the jury.

A very special thanks goes out to John-Jules Ch. Meyer from Utrecht University (the Netherlands) for his extremely appreciated collaboration that allowed me to improve the logical model and the model checking technique proposed in this thesis. I have very much enjoyed working with him. I also would like to thank him for having received me in his Intelligent Systems Group for 2 weeks during the last month of my thesis and for having given me the opportunity to discuss my work with Frank Dignum, Mehdi Dastani, Henry Prakken, Broersen Jan, Martin Caminada, Van Riemsdijk Birna and Davide Grossi. Particular thanks goes out to Frank Dignum for his helpful comments and interesting suggestions.

I am particularly appreciative of the help I did receive from: Leila Amgoud from Institut de Recherche en Informatique de Toulouse (France), Nicolas Maudet from Université Paris 9 Dauphine (France), Marco Colombetti from Politecnico di Milano (Italy), Iyad Rahwan from University of Melbourne (Australia) and British University in Dubai (United Arab Emirates), Rance Cleaveland from State University of New York at Stony Brook (USA), Girish Bhat from Cosine Communication Inc. (USA), Peter McBurney from University of Liverpool (UK), Simon Parsons from City University of New York (USA), Claude Bélisle from Université Laval, Department of Mathematics and Statistics (Canada), Josée Desharnais from Université Laval (Canada), Daniel Vanderveken from Université de Québec à Trois-Rivières, Department of Philosophy (Canada), Sylvain Delisle from Université de Québec à Trois-Rivières (Canada) and my colleague Philippe Pasquier. I also would like to thank Johanne Savard for her friendship and linguistic help.

My years as a Ph.D student would not have been as much fun without my friends, particularly my colleagues: Nafaa Jabeur, Nabil Sahli, Walid Ali, Hedi Haddad, Mondher Bouden, Tarek Sbeoui, and Boulekrouche Boubaker with whom I shared the laboratory and unforgettable moments. I would like to thank him for being the surrogate family and for their continued moral support. From the staff, Linda Goulet and Violaine Pellerin are especially thanked for their care and attention.

Finally, I would like to thank my mother and all my family (Fatima, Mohammed, Faisal, Hassan, Fadila, Bouchra, and Wiam) for the support they provided me through my entire life and for their endless patience, without whose love, encouragement and editing assistance, I would not have finished this thesis.

© Jamal Bentahar, 2005