Computational models of argument

Argumentation is a generic approach for solving different AI problems including defeasible reasoning, decision making, classification, etc. It is based on the justification of claims by arguments.

In this topic, our research can be partitioned into 3 distinct fields.

Evaluation of arguments

A key step of the approach based on argumentation is the evaluation of arguments. This notion of evaluation can be studied following two points of view: we can consider either an objective evaluation of the arguments (objective in the sense that the impact on the “target of the argumentation task” is not taken into account) or a cognitive one.

Considering the objective point of view, since 1993, several evaluation methods, called semantics, were proposed. However, their foundations were largely unexplored. We laid the foundations of semantics; we proposed axioms or principles that should be satisfied by reasonable semantics and investigated four types of argumentation graphs: flat attack graphs, weighted attack graphs, weighted support graphs, weighted bipolar graphs; for each type of graphs, we analyzed and compared existing semantics, and proposed novel ones that satisfy the axioms. Also, we have conducted an analysis (subsumption, incompatibility,..) of principles, adapted or dedicated, to ranking-based semantics. We also proposed new semantics in order to complete the panoramic view of the extensions of the classical framework defined by Dung (argumentation graphs with recursive/high-order attack/support). Beside this core, some other interesting points have been studied: for instance a structural analysis of some semantics using a matrix approach.

Considering the cognitive evaluation of arguments, we have proposed in a formal analysis of the persuasive impact that an argument can produce on a human agent. This analysis allows the definition of a formal model based on the dual process theory proposed by Kahneman. Then this model has been applied to the automatic evaluation of arguments. This work is the first step towards a best understanding of some crucial aspects of the collective decision using argumentation strategies and persuasion process.

Languages and algorithms

It is of course important to define and characterize the evaluation of arguments. Nevertheless, in practice, it remains useless if we are not able to compute them in a realistic time. That leads us to consider both languages and algorithms for evaluation of arguments.

We laid the foundations of a logic of argumentation in which arguments, as well as attacks and supports among arguments are all defined in a unifying formalism. We introduced a series of inference rules relating arguments and showed how the resulting logic captures important features of argumentation that hitherto have not been captured by existing formalisms. We have also proposed a translation of abstract argumentation graphs (flat, bipolar and recursive) into logical formalisms. Another approach was the definition of a logic able to support argumentation and logic programming.

The study of these different logical approaches for argumentation has allowed us to implement some tools and to open the door to algorithms based on efficient SAT solvers for computing argument evaluation (the more efficient the translation and the SAT solver, the larger the argumentation graphs we can evaluate). Note that, in parallel, a study of the links between abstract argumentation and logic is the subject of a CIMI project.

Argument mining  

Argument mining is a highly complex information retrieval task. Given a claim, the aim is to find, in a large variety of texts, propositions that support or attack it, how, with what strength and on the basis of what knowledge and reasoning schemes. In the past years, from linguistic and conceptual evidence, we developed an original approach: knowledge-based argument mining which offers a high analysis accuracy and a good explanation power. This approach enables a better understanding of argumentative processes than, e.g., systems based on learning from annotations.

References

  • Leila Amgoud, Jonathan Ben-Naim. Axiomatic foundations of acceptability semantics. Dans / In : International Conference on Principles of Knowledge Representation and Reasoning (KR 2016), Cap Town, Afrique du Sud, 25/04/2016-29/04/2016, James Delgrande, Frank Wolter (Eds.), AAAI Press, p. 2-11, avril / april 2016.
  • Leila Amgoud, Jonathan Ben-Naim, Dragan Doder, Srdjan Vesic. Acceptability semantics for weighted argumentation frameworks. Dans / In : International Joint Conference on Artificial Intelligence (IJCAI 2017), Melbourne, Australie, 19/08/2017-25/08/2017, AAAI Press, p. 56-62, août / august 2017.
  • Leila Amgoud, Jonathan Ben-Naim. Evaluation of arguments from support relations: Axioms and Semantics. Dans / In : International Joint Conference on Artificial Intelligence (IJCAI 2016), New York, USA, 09/07/2016-15/07/2016, Kambhampati Subbarao (Eds.), AAAI Press, p. 900-906, juillet / july 2016.
  • Leila Amgoud, Jonathan Ben-Naim. Weighted bipolar argumentation graphs: Axioms and Semantics. Dans / In : International Joint Conference on Artificial Intelligence (IJCAI 2018), Stockholm, Suède, 13/07/2018-19/07/2018, Jérôme Lang (Eds.), AAAI Press, p. 5194-5198, juillet / july 2018.
  • Claudette Cayrol, Jorge Fandinno, Luis Fariñas del Cerro, Marie-Christine Lagasquie-Schiex. Structure-Based Semantics of Argumentation Frameworks with Higher-Order Attacks and Supports. Dans / In : International Conference on Computational Models of Argument (COMMA 2018), Warsaw, Poland, 12/09/2018-14/09/2018, Sanjay Modgil, Katarzyna Budzynska, John Lawrence (Eds.), IOS Press, p. 29-36, septembre / september 2018.
  • Claudette Cayrol, Jorge Fandinno, Luis Fariñas del Cerro, Marie-Christine Lagasquie-Schiex. Argumentation Frameworks with Recursive Attacks and Evidence-Based Supports. Dans / In : International Symposium on Foundations of Information and Knowledge Systems (FoIKS) (FolKS 2018), Budapest, Hungary, 14/05/2018-18/05/2018, Flavio Ferrarotti, Stefan Woltran (Eds.), Springer, LNCS 10833, p. 150-169, mai / may 2018.
  • Yuming Xu, Claudette Cayrol. Initial sets in abstract argumentation frameworks. Dans / In : Journal of Applied Non-Classical Logics, Taylor & Francis Group, Vol. 28 N. 2-3, p. 260-279, avril / april 2018.
  • Pierre Bisquert, Madalina Croitoru, Florence Dupin De Saint Cyr, Abdelraouf Hecham. Formalizing Cognitive Acceptance of Arguments: Durum Wheat Selection Interdisciplinary Study. Dans / In : Minds & Machines, Springer, Vol. 27 N. 1, p. 233-252, avril / april 2017.
  • Leila Amgoud, Philippe Besnard, Anthony Hunter. Foundations for a Logic of Arguments. Dans / In : Journal of Applied Non-Classical Logics, Taylor & Francis Group, Vol. 27 N. 3-4, p. 178-195, février / february 2018.
  • Florence Dupin De Saint Cyr, Pierre Bisquert, Claudette Cayrol, Marie-Christine Lagasquie-Schiex. Argumentation update in YALLA (Yet Another Logic Language for Argumentation). Dans / In : International Journal of Approximate Reasoning, Elsevier, Vol. 75, p. 57-92, 2016.
  • Claudette Cayrol, Marie-Christine Lagasquie-Schiex. Logical Encoding of Argumentation Frameworks with Higher-order Attacks. Dans / In : International Conference on Tools with Artificial Intelligence (ICTAI 2018), Volos, Grèce, 05/11/2018-07/11/2018, Miltos Alamaniotis (Eds.), IEEE : Institute of Electrical and Electronics Engineers, p. 667-674, décembre / december 2018.
  • Jorge Fandinno, Luis Fariñas del Cerro. Constructive Logic Covers Argumentation and Logic Programming. Dans / In : International Conference on Principles of Knowledge Representation and Reasoning (KR 2018), Tempe, Arizona, 27/10/2018-02/11/2018, AAAI Press, p. 128-137, 2018.
  • Patrick Saint-Dizier. Challenges of Discourse Processing: the case of technical texts, Cambridge University Press, février / february 2014.
  • Patrick Saint-Dizier. A Two-Level Approach to Generate Synthetic Argumentation Reports. Dans / In : Argument and Computation Journal, Taylor & Francis Group, Vol. 9(1), janvier / january 2018.
  • Patrick Saint-Dizier. Mining Incoherent Requirements in Technical Specifications: Analysis and Implementation. Dans / In : Data and Knowledge Engineering, Elsevier, Vol. 116, avril / april 2018.
  • Patrick Saint-Dizier. Knowledge-Driven Argument Mining Based on the Qualia Structure. Dans / In : Argumentation and Computation, Taylor & Francis Group, Vol. 6(3), p. 132-160, octobre / october 2017.