Ebook: Computational Models of Argument

Models of argumentation, that take pros and cons for some conclusion into account, have been extensively studied over a number of years, and some basic principles have now been clearly established. More recently, there has been interest in computational models of argumentation where the aim is to develop software tools to assist users in constructing and evaluating arguments and counterarguments and/or to develop automated systems for constructing and evaluating arguments and counterarguments.

After the very successful First Conference on Computational Models of Argument that was instigated by the members of the EU-funded ASPIC Project, and hosted by the University of Liverpool in September 2006, the papers in this volume form the programme for the Second Conference on Computational Models of Argument hosted by the Institut de Recherche en Informatique de Toulouse (IRIT) in May 2008.

The range of papers in the volume provides a valuable snapshot of the leading research questions in the area of computational models of argument. This volume includes papers drawing on the wealth of research on the philosophical questions surrounding the notions arising in argumentation, papers that are addressing knowledge representation and reasoning issues emerging from modelling argumentation, and papers that are considering appropriate models of wider rhetorical issues arising in argumentation. This volume also includes papers that are proposing and evaluating algorithmic solutions associated with generating and judging constellations of arguments, and papers that are proposing standards for exchanging information associated with argumentation so that different systems can work together. We are also pleased to include some papers that report practical working tools for computational argumentation.

We would like to thank the programme committee who did an excellent job in selecting the 37 papers in this volume from the 60 papers that were originally submitted. We would also like to thank the Local Organization Committee (i.e. Leila Amgoud, Claudette Cayrol, Véronique Debats, Marie-Christine Lagasquie-Schiex, Jérôme Mengin, Laurent Perrussel, and Henri Prade) for all the hard work they put into making the conference a success.

March 2008

Philippe Besnard (Conference Chair)

Sylvie Doutre (Local Organization Chair)

Anthony Hunter (Programme Chair)

Possibilistic Defeasible Logic Programming (P-DeLP) is an argumentation framework based on logic programming which incorporates a treatment of possibilistic uncertainty at object-language level. In P-DeLP, the closure of justified conclusions is not always consistent, which has been detected to be an anomaly in the context of so-called rationality postulates for rule-based argumentation systems. In this paper we present a novel level-based approach to computing warranted arguments in P-DeLP which ensures the above rationality postulate. We also show that our solution presents some advantages in comparison with the use of a transposition operator applied on strict rules.

Persuasion is one of the main types of dialogs encountered in everyday life. The basic idea behind a persuasion is that two (or more) agents disagree on a state of affairs, and each one tries to persuade the other to change his mind. For that purpose, agents exchange arguments of different strengths.

Several systems, grounded on argumentation theory, have been proposed in the literature for modeling persuasion dialogs. These systems have studied more or less deeply the different protocols required for this kind of dialogs, and have investigated different termination criteria. However, nothing is said about the properties of the generated dialogs, nor on the behavior of the interacting agents. Besides, analyzing dialogs is a usual task in everyday life. For instance, political debates are generally deeply dissected.

In this paper we define measures for analyzing dialogs from the point of view of an external agent. In particular, three kinds of measures are proposed: i) measures of the quality of the exchanged arguments in terms of their strengths, ii) measures of the behavior of each participating agent in terms of its coherence, its aggressiveness in the dialog, and finally in terms of the novelty of its arguments, iii) measures of the quality of the dialog itself in terms of the relevance and usefulness of its moves.

In a recent work we have proposed a comprehensive set of evaluation criteria for argumentation semantics and we have shown that none of a set of semantics including both traditional and recent proposals is able to meet all criteria. This naturally raises the question whether such criteria are actually satisfiable altogether: this paper provides a positive answer to this question by introducing a new family of argumentation semantics, called resolution-based and showing that all the desirable criteria are met by the resolution-based version of grounded semantics.

The issue of characterizing classes of argumentation frameworks where different semantics agree has been considered in the literature with main focus on the relationships between agreement and topological properties. This paper contributes to this kind of investigation from a complementary perspective, by introducing a systematic classification of agreement classes concerning a comprehensive set of argumentation semantics on the basis of properties of their sets of extensions only. In particular, it is shown that 14 distinct classes out of 120 nominal ones exist, and a complete analysis of their set-theoretical relationships is carried out.

We introduce a new semantics for value-based argumentation frameworks (VAFs) – the uncontested semantics – whose principal motivation is as a mechanism with which to refine the nature of objective acceptance with respect to a given audience. The objectively accepted arguments of a VAF w.r.t. an audience [Rscr ], are those considered justified by all subscribing to the audience, [Rscr ], regardless of the specific value orderings that individuals may hold. In particular we examine how the concept of uncontested acceptance may be used in examination dialogues. The proposed semantics bear some aspects in common with the recently proposed ideal semantics for standard – i.e. value–free – argumentation frameworks. In this paper we consider applications of the new semantics to a specific “real” example and examine its relationship to the ideal semantics as well as analysing some basic complexity-theoretic issues.

Mapping scholarly debates is an important genre of what can be called Knowledge Domain Analytics (KDA) technology – i.e. technology which combines both quantitative and qualitative methods of analysing specialist knowledge domains. However, current KDA technology research has emerged from diverse traditions and thus lacks a common conceptual foundation. This paper reports on the design of a KDA ontology that aims to provide this foundation. The paper then describes the argumentation extensions to the ontology for supporting scholarly debate mapping as a special form of KDA and demonstrates its expressive capabilities using a case study debate.

In this paper we propose a formal dialogue game in which two players aim to determine the best explanation for a set of observations. By assuming an adversarial setting, we force the players to advance and improve their own explanations as well as criticize their opponent's explanations, thus hopefully preventing the well-known problem of ‘tunnel vision’. A main novelty of our approach is that the game supports the combination of argumentation with abductive inference to the best explanation.

The purpose of this paper is to provide a formal model of a persuasion process in which a persuader tries to influence audience's beliefs. We focus on various interactions amongst agents rather than only exchange of arguments by verbal means. Moreover, in our approach the impact of the parties of the dispute on the final result of the persuasion is emphasized. Next, we present how to formalize this model using the modal system inspired by Logic of Graded Modalities and Algorithmic and Dynamic Logics. We also show how to investigate local and global properties of multi-agent systems which can be expressed in the language of the proposed logic.

Students, researchers and professional analysts lack effective tools to make personal and collective sense of problems while working in distributed teams. Central to this work is the process of sharing—and contesting—interpretations via different forms of argument. How does the “Web 2.0” paradigm challenge us to deliver useful, usable tools for online argumentation? This paper reviews the current state of the art in Web Argumentation, describes key features of the Web 2.0 orientation, and identifies some of the tensions that must be negotiated in bringing these worlds together. It then describes how these design principles are interpreted in Cohere, a web tool for social bookmarking, idea-linking, and argument visualization.

The past ten years have shown a great variety of approaches for formal argumentation. An interesting question is to which extent these various formalisms correspond to the different application domains. That is, does the appropriate argumentation formalism depend on the particular domain of application, or does “one size fits all”. In this paper, we study this question from the perspective of one relatively simple design consideration: should or should there not be contrapostion of (or modus tollens) on defeasible rules. We aim to show that the answer depends on whether one is considering epistemical or constitutive reasoning, and that hence different domains require fundamentally different forms of defeasible reasoning.

In this paper we discuss the development of tools to support a system for e-democracy that is based upon and makes use of existing theories of argument representation and evaluation. The system is designed to gather public opinions on political issues from which conclusions can be drawn concerning how government policies are presented, justified and viewed by the users of the system. We describe how the original prototype has been augmented by the addition of well motivated tools to enable it to handle multiple debates and to provide analyses of the opinions submitted, from which it is possible to pinpoint specific grounds for disagreement on an issue. The tool set now supports both argumentation schemes and argumentation frameworks to provide representation and evaluation facilities. We contrast our system with existing fielded approaches designed to facilitate public consultation on political issues and show the particular benefits that our approach can bring in attempting to improve the quality of such engagement online.

Human beings share a common competence for generating relevant arguments. We therefore hypothesize the existence of a cognitive procedure that enables them to determine the content of their arguments. The originality of the present approach is to analyse spontaneous argument generation as a process in which arguments either signal problems or aim at solving previously acknowledged problems.

We present an argumentation-based approach to contract negotiation amongst agents. Contracts are simply viewed as abstract transactions of items between a buyer agent and a seller agent, characterised by a number of features. Agents are equipped with beliefs, goals, and preferences. Goals are classified as either structural or contractual. In order to agree on a contract, agents engage in a two-phase negotiation process: in the first phase, the buyer agent decides on (a selection of) items fulfilling its structural goals and preferences; in the second phase, the buyer agent decides on a subset of the items identified in the first phase fulfilling its contractual goals and preferences. The first phase is supported by argumentation-based decision making taking preferences into account.

We analyse the computational complexity of the recently proposed ideal semantics within abstract argumentation frameworks. It is shown that while typically less tractable than credulous admissibility semantics, the natural decision problems arising with this extension-based model can, perhaps surprisingly, be decided more efficiently than sceptical admissibility semantics. In particular the task of finding the unique maximal ideal extension is easier than that of deciding if a given argument is accepted under the sceptical semantics. We provide efficient algorithmic approaches for the class of bipartite argumentation frameworks. Finally we present a number of technical results which offer strong indications that typical problems in ideal argumentation are complete for the class P^NP_‖ : languages decidable by polynomial time algorithms allowed to make non-adaptive queries to an np oracle.

Classical propositional logic is an appealing option for modelling argumentation but the computational viability of generating an argument is an issue. Here we propose ameliorating this problem by harnessing the notion of a connection graph to reduce the search space when seeking all the arguments for a claim from a knowledgebase. For a set of clauses, a connection graph is a graph where each node is a clause and each arc denotes that there exist complementary disjuncts in the pair of nodes. For a set of formulae in conjunctive normal form, we use the notion of the connection graph for the set of clauses obtained from the conjuncts in the formulae. When seeking arguments for a claim, we can focus our search on a particular subgraph of the connection graph that we call the focal graph. Locating this subgraph is relatively inexpensive in terms of computational cost. In addition, using (as the search space) the formulae of the initial knowledgebase, whose conjuncts relate to this subgraph, can substantially reduce the cost of looking for arguments. We provide a theoretical framework and algorithms for this proposal, together with some theoretical results and some preliminary experimental results to indicate the potential of the approach.

In this paper we present a model for defeasible decision making that combines decision rules and arguments. In this decision framework we can change the agent's decision policy in a flexible way, with minor changes in the criteria that influence the agent's preferences and the comparison of arguments. Our approach includes a simple methodology for developing the decision components of the agent. A decision framework designed with this methodology exhibits some interesting properties. If the agent (decision maker) has available all the relevant knowledge about its preferences among the different alternatives that could be conceivably posed to it, then our proposal implements a rational preference relation. In opposition, if the agent has partial knowledge about its preferences, the decisions made by the agent still exhibits a behavior consistent with the weak axiom of revealed preference of the choice-based approach, a more flexible approach to Individual Decision Making than the preference-based approach. The principles stated in this work are exemplified in a robotic domain, where a robot should make decisions about which box must be transported next.

We present a variant of AB-dispute derivations for assumption-based argumentation (ABA), that can be used for determining the admissibility of claims. ABA reduces the problem of computing arguments to the problem of computing assumptions supporting these arguments. Whereas the original AB-dispute derivations only manipulate sets of assumptions, our variant also renders explicit the underlying dialectical structure of arguments (by a proponent) and counter-arguments (by an opponent), and thus supports a hybrid of ABA and abstract argumentation beneficial to developing applications of argumentation where explicit justifications of claims in terms of full dialectical structures are required. We prove that the proposed variant of AB-dispute derivations is correct.

This paper formulates in the first part some requirements for a certain sort of computational argumentation systems, namely those which are designed for a very specific purpose: to motivate reflection on one's own thinking, and to induce cognitive change. This function of argumentation systems is important for argument-based conflict negotiations, deliberation processes, intercultural communication, text analysis, and learning through argument visualization. In all these situations success is only possible when people are able to change their mind, learn something, or start to reframe well-established ways of perceiving and interpreting things. Based on these requirements, I defend and explain in the second part my decision to use for Logical Argument Mapping—a method specifically designed for supporting reflective argumentation—only argument schemes that are deductively valid.

Proposals for logic-based argumentation have the potential to be adapted for handling diverse kinds of knowledge. In this paper, a calculus for representing temporal knowledge is proposed, and defined in terms of propositional logic. This calculus is then considered with respect to argumentation, where an argument is pair 〈Φ,α〉 such that Φ a minimally consistent subset of a database entailing α. Two alternative definitions of an argument are considered and contrasted.

Extended abstract frameworks separate conflicts and preference between arguments. These elements are combined to induce argument defeat relations. A proper defeat is consequence of preferring an argument in a conflicting pair, while blocking defeat is consequence of incomparable or equivalent-in-strength conflicting arguments. As arguments interact with different strengths, the quality of several argument extensions may be measured in a particular semantics. In this paper we analyze the strength of defenses in extended argumentation frameworks, under admissibility semantics. A more flexible form of acceptability is defined leading to a credulous position of acceptance.

This paper is concerned with the general problem of constructing decision tables and more specifically, with the identification of all possible outcomes of decisions. We introduce and propose basic influence diagrams as a simple way of describing problems of decision making under strict uncertainty. We then establish a correspondence between basic influence diagrams and symmetric generalised assumption-based argumentation frameworks and adopt an argumentation-based approach to identify the possible outcomes. We show that the intended solutions are best characterised using a new semantics that we call liberal stability. We finally present a number of theoretical results concerning the relationships between liberal stability and existing semantics for argumentation.

A recent extension to Dung's argumentation framework allows for arguments to express preferences between other arguments. Value based argumentation can be formalised in this extended framework, enabling meta-level argumentation about the values that arguments promote, and the orderings on these values. In this paper, we show how extended frameworks integrating meta-level reasoning about values can be rewritten as Dung frameworks, and show a soundness and completeness result with respect to the rewrites. We then describe how value orderings can emerge, or be ‘formed’, as a result of dialogue games based on the rewritten frameworks, and illustrate the advantages of this approach over existing dialogue games for value based argumentation frameworks.

An abstract framework for formalising persuasion dialogues has recently been proposed. The framework provides for a range of speech acts, and protocols of varying levels of flexibility. However, the framework assumes the availability of preference information relevant to determining whether arguments moved in a dialogue defeat each other. However, preference information may only become available after the dialogue has terminated. Hence, in this paper, we describe dialogues conducted under the assumption of an attack relation that does not account for preferences. We then describe how the resultant dialogue graph can be pruned by a preference relation in order to determine whether the winner of the dialogue is still the winner given the newly available preference information. We also describe a class of protocols that account for subsequent pruning by a preference relation, and show that under a restriction on the pruning, if the player defending the dialogue's main topic is winning the dialogue, then (s)he remains the winner irrespective of the preference relation applied.

This paper describes a method for decision making using argumentation. The method is intended to produce the decision considered most likely to promote the agent's ams ans aspirations. First, the problem situation is formulated in terms of an Action-Based Alternating Transition System, representing the actions available to the agents relevant to the situation, and their consequences, taking into account the possible effects of the choices of the other relevant agents. Next, arguments are constructed by instantiating an argumentation scheme designed to justify actions in terms of the values they promote and subjecting these instantiations to a series of critical questions to identify possible counter arguments. The resulting arguments are then organized into a Value-Based Argumentation Framework (VAF), so that a set of arguments acceptable to the agent can be identified. Finally the agent must select one of the acceptable actions to execute. The methodology is illustrated through the use of a detailed case study.