Ebook: Computational Models of Argument

The papers in this volume formed the programme of the 1^st International Conference on Computational Models of Argument (COMMA), which was hosted by the Dept. of Computer Science of The University of Liverpool from Sept. 11^th–12^th, 2006. This conference originated from the ASPIC project

European Commission Project, IST-FP6-002307

The field of argumentation, once the preserve of linguistic and philosophical investigation, is now rightly seen as providing a core approach of great significance to many aspects of Artificial Intelligence. A central challenge for A.I. researchers, however, concerns how best to develop the long established body of work from more speculative disciplines, such as philosophical treatments of argument and reasoning, into effective and practical computational paradigms: one aim of COMMA, well reflected in the papers contributing to this volume, has been to engage with the issues raised by this challenge. Thus the topics addressed range from formal questions involving properties of algorithms and semantic models, through proposals for robust implementation of argumentation based systems, to reports of applications built on argumentation technology.

It is, of course, the case that the success of any conference depends not only on the quality of the research presented but also on the contributions of many other individuals. The organisers are grateful to the members of the Programme Committee and additional reviewers whose detailed reports and subsequent discussions considerably eased the diffult task of forming the final selection of papers. It is also a pleasure to thank Ken Chan, Phil Jimmieson and Dave Shield for their work in providing technical support throughout the period from the initial announcement to the conference itself, together with Thelma Williams who kept track of assorted budget and financial matters. In addition the editors appreciate the efforts of Carry Koolbergen, Maarten Fröhlich, and Paul Weij of IOS Press in promptly and efficiently handling the many questions that arose during the preparation of this volume. Finally, and by no means least, we thank Catherine Atherton who maintained the conference web pages as well as dealing with general queries.

June 2006, Paul E. Dunne, Trevor Bench-Capon, Michael Wooldridge

This paper demonstrates the potential of the Semantic Web as a platform for representing, navigating and processing arguments on a global scale. We use the RDF Schema (RDFS) ontology language to specify the ontology of the recently proposed Argument Interchange Format (AIF) and an extension thereof to Toulmin's argument scheme. We build a prototype Web-based system for demonstrating basic querying for argument structures expressed in the Resource Description Framework (RDF). An RDF repository is created using the Sesame open source RDF server, and can be accessed via a user interface that implements various user-defined queries.

In order for one agent to meet its goals, it will often need to influence another to act on its behalf, particularly in a society in which agents have heterogenous sets of abilities. To effect such influence, it is necessary to consider both the social context and the dialogical context in which influence is exerted, typically through utterance. Both of these facets, the social and the dialogical, are affected by, and in turn affect, the plan that the influencing agent maintains, and the plans that the influenced agents may be constructing. The i-Xchange project seeks to bring together three closely related areas of research: in distributed planning, in agent-based social reasoning, and in inter-agent argumentation, in order to solve some of the problems of exerting influence using socially-aware argument.

Argumentation is becoming increasingly important in the design and implementation of autonomous software agents. We believe that agents engaged in decision-making and reasoning should have access to a general purpose argumentation engine that can be configured to conform to one of a range of semantics. In this paper we discuss our current work on a prototype light-weight Java-based argumentation engine that can be used to implement a non-monotonic reasoning component in Internet or agent-based applications.

We consider agents in a multi-agent system, each equipped with a Bayesian network model (BN) of its environment. We want the agents to reach consensus on one compromise network, which may not be identical to a single one of the BNs initially held by the agents, but rather a combination of aspects from each BN. The task can be characterized as the need for agents to agree on a specific state (a BN) of a variable with an enormous state space (all possible BNs). The grandness of the task is reduced by the fact that BNs are composed of local relationships, and it should therefore be possible to reach the compromise by gradually agreeing on parts of it. In the metaphor of the variable, the agents should be able to agree on successively smaller subsets of the enormous state space. However, these same local relationship can interact, and understanding the extent to which partial agreements affect the possible final compromise is a highly complex task. In this work we suggest using formal argumentation as the reasoning mechanism for agents solving this task, and suggest an open-ended agora approach that ensures agents high quality compromises in an anytime fashion.

In this paper we discuss the integration of two systems that are based on a specific theory of argumentation: the first, an existing web-based discussion forum; the second, a method to enable autonomous software agents to perform practical reasoning based upon their subscription to social values. Weshow how the output from the first of these systems can be used as input to the second and how the information gathered can be reasoned about through computer support. The purpose of the approach is to demonstrate how current theories of argumentation can be used to assist with the analysis of public attitude in a particular debate, with the specific example domain used being that of eDemocracy. We also provide some discussion and comparison of these current tools with similar, earlier systems.

Argument Based Machine Learning (ABML) is a new approach to machine learning in which the learning examples can be accompanied by arguments. The arguments for specific examples are a special form of expert's knowledge which the expert uses to substantiate the class value for the chosen example. Možina et al. developed the ABCN2 algorithm - an extension of the well known rule learning algorithm CN2 - that can use argumented examples in the learning process. In this work we present an application of ABCN2 in the medical domain which deals with severe bacterial infections in geriatric population. The elderly population, people over 65 years of age, is rapidly growing as well as the costs of treating this population. In our study, we compare ABCN2 to CN2 and show that using arguments we improve the characteristics of the model. We also report the results that C4.5, Naïve Bayes and Logistic Regression achieve in this domain.

In this paper we present a novel approach for combining Case-Based Reasoning (CBR)Argumentation. This approach involves 1) the use of CBR for evaluating the arguments submitted by agents in collaborative decision making dialogs, and 2) the use of Argument Schemes and Critical Questions to organize the CBR memory space. The former involves use of past cases to resolve conflicts among newly submitted arguments by assigning them a strength, and possibly submitting additional arguments deemed relevant in similar past deliberations. The latter enables use of agents' submitted arguments instantiating Argument Schemes and Critical Questions, to assess the similarity among cases. This use of CBR and argumentation is formulated with the ProCLAIM model, which features a Mediator Agent that directs proponent agents in their deliberation and subsequently evaluates their submitted arguments so as to conclude whether a proposed decision is valid. To motivate and substantiate the practical value of this approach, we illustrate its application in the human organ transplantation field.

One difficulty that arises in abstract argument systems is that many natural questions regarding argument acceptability are, in general, computationally intractable having been classified as complete for classes such as NP, CO-NP, and Π^p₂. In consequence, a number of researchers have considered methods for specialising the structure of such systems so as to identify classes for which efficient decision processes exist. In this paper the effect of a number of graph-theoretic restrictions is considered. For the class of bipartite graphs, it is shown that determining the acceptability status of a specific argument can be accomplished in polynomial time under both credulous and sceptical semantics. In contrast to these positive results, however, deciding whether an arbitrary set of arguments is “collectively acceptable” remains NP–complete in bipartite systems. In addition, a construction is presented by means of which questions posed of arguments in any given finite argument system may be expressed as questions within a related system in which every argument attacks and is attacked by at most two arguments. It follows that bounding the number of attacks on individual arguments is unlikely to produce a computationally more tractable environment.

The hitherto most abstract, and hence general, argumentation system, is the one described by Dung in a paper from 1995. This framework does not allow for joint attacks on arguments, but in a recent paper we adapted it to support such attacks, and proved that this adapted framework enjoyed the same formal properties as that of Dung. One problem posed by Dung's original framework, which was neglected for some time, is how to compute preferred extensions of the argumentation systems. However, in 2001, in a paper by Doutre and Mengin, a procedure was given for enumerating preferred extensions for these systems. In this paper we propose a method for enumerating preferred extensions of the potentially more complex systems, where joint attacks are allowed. The method is inspired by the one given by Doutre and Mengin.

This paper presents a query-answering algorithm to compute minimal lines of defence around an individual argument. The algorithm returns all such defence sets together with an indication whether the defence is grounded or admissible. For every argument encountered in the search process the algorithm further indicates whether that argument is IN, OUT, or UNDEC (undecided) according to the grounded semantics. The presentation of the algorithm is followed by a correctness proof and a complexity analysis of other than worst cases. The algorithm is already functional in argument analysis and visualization tools.

In this paper, we examine an argument-based semantics called semi-stable semantics. Semi-stable semantics is quite close to traditional stable semantics in the sense that every stable extension is also a semi-stable extension. One of the advantages of semi-stable semantics is that there exists at least one semi-stable extension. Furthermore, if there also exists at least one stable extension, then the semi-stable extensions coincide with the stable extensions. This, and other properties, make semi-stable semantics an attractive alternative for the more traditional stable semantics, which until now has been widely used in fields such as logic programming and answer set programming.

This paper describes a generic approach to implement propositional argumentation frameworks by means of quantified Boolean formulas (QBFs). The motivation to this work is based on the following observations: Firstly, depending on the underlying deductive system and the chosen semantics (i.e., the kind of extension under consideration), reasoning in argumentation frameworks can become computationally involving up to the fourth level of the polynomial hierarchy. This makes the language of QBFs a suitable target formalism since decision problems from the polynomial hierarchy can be efficiently represented in terms of QBFs. Secondly, several practicably efficient solvers for QBFs are currently available, and thus can be used as black-box engines in potential implementations of argumentation frameworks. Finally, the definition of suitable QBF modules provides us with a tool box in order to capture a broad range of reasoning tasks associated to formal argumentation.

We present a procedure for computing the sceptical “deal semantics” for argumentation in assumption-based frameworks. This semantics was first proposed for logic programming in [1], extending the well-founded semantics. The proof procedure is defined by means of a form of dispute derivations, obtained by modifying the dispute derivations given in [2] for computing credulous admissible argumentation. The new dispute derivations are sound for the “ideal semantics” in all cases where the dispute derivations of [2] are complete for admissible argumentation. We prove that this is the case for the special kind of assumption-based frameworks with a finite underlying language and with the property of being “p-acyclic”.

In the context of Dung's theory of abstract argumentation frameworks, the comparison between different semantics is often carried out by resorting to some specific examples considered particularly meaningful. This kind of comparison needs to be complemented by more general evaluation criteria based on “example-independent” basic principles. We review several principles for argumentation semantics, identify their formal counterpart in terms of extensions, and analyze their relationships with the notion of argument justification state. Then, we evaluate and compare several semantics on the basis of the introduced principles.

In the early 20th century, J.H. Wigmore described a new method for analysing and laying out arguments in legal cases. His proposal was the first system of argument diagramming, and it is still in use in jurisprudence today. Wigmore diagrams offer a rich ontology of argumentation concepts which in some respects are close to ideas in other, more modern systems of argument analysis and argument diagramming – whilst in other areas, is much richer and more specific than alternatives. The features of Wigmore analyses might reasonably be expected to contribute to modern, computational approaches to argument, both in the legal domain and more broadly. This paper explores some of the key issues in representing Wigmore analyses and translating between them and other systems of analysis such as those founded upon Toulmin models and scheme-based models.

This paper presents our efforts to create argument structures from meeting transcripts automatically. We show that unit labels of argument diagrams can be learnt and predicted by a computer with an accuracy of 78,52% and 51,43% on an unbalanced and balanced set respectively. We used a corpus of over 250 argument diagrams that was manually created by applying the Twente Argument Schema. In this paper we also elaborate on this schema and we discuss applications and the role we foresee the diagrams to play.

We present a formal, mathematical model of argument structure and evaluation, called the Carneades Argumentation Framework, which applies proof standards [1] to determine the defensibility of arguments and the acceptability of statements on an issue-by-issue basis. Carneades uses three kinds of premises (ordinary premises, presumptions and exceptions) and information about the dialectical status of statements (undisputed, at issue, accepted or rejected) to model critical questions in such a way as to allow the burden of proof to be allocated to the proponent or the respondent, as appropriate.

The Pierson vs. Post case [1] has become an important benchmark in the field of AI and Law for computational models of argumentation. In [2], Bench-Capon used Pierson vs. Post to motivate the use of values and value preferences in his theory-construction account of legal argument. And in a more a recent paper by Atkinson, Bench-Capon and McBurney [3], it was used to illustrate a formalization of an argumentation scheme for practical reasoning. Here we offer yet another reconstruction of Pierson vs. Post, using our Carneades Argumentation Framework, a formal mathematical model of argument structure and evaluation based on Walton's theory of argumentation [4], and compare it to this prior work. Carneades, named in honor of the Greek skeptic philosopher who emphasized the importance of plausible reasoning, applies proof standards [5] to determine the defensibility of arguments and the acceptability of statements on an issue-by-issue basis.