Ebook: Computational Models of Argument
In its classical form, the study of argumentation focuses on human-oriented uses of argument, such as whether an argument is legitimate or flawed, engagement in debate, or the rhetorical aspects of argumentation. In recent decades, however, the study of logic and computational models of argumentation has emerged as a growing sub-area of AI.
This book presents the Seventh International Conference on Computational Models of Argument (COMMA’18), held in Warsaw, Poland, from 12 to 14 September 2018. Since its inception in 2006, the conference and its related activities have developed alongside the steady growth of interest in computational argumentation worldwide, and the selection of 25 full papers and 17 short papers, out of a total of 70 submissions, and 15 demonstration abstracts included here reflect the broad multidisciplinary nature of argumentation and the increasing body of work which establishes the relevance of computational models to various disciplines and real world applications.
Subjects covered include: algorithm development; innovative applications; argument mining, argumentation-based models of dialogue; abstract argument frameworks; and structured argumentation.
Representing an overview of current developments in the field, this book will appeal to all those with an interest in computational models of argument.
In its classical treatment within philosophy and the social sciences, from Aristotle to the present day, the study of argumentation has focused on ‘human orientated’ uses of argument, such as when an argument can be considered legitimate or flawed, the processes by which participants engage in debate, rhetorical aspects of argumentation etc. However, in the last couple of decades, the study of logic, and more broadly ‘computational’ models of argumentation, has emerged as a growing sub-area of AI. While researchers from a number of fields contributed to this growth, it was those with a background in logics for non-monotonic and uncertain reasoning, and logic based models of argumentation for epistemic reasoning, decision making and dialogue, that fuelled this surge of interest in argumentation. In particular, the first biennial International Conference on Computational Models of Argument (COMMA), was supported by the EU 6th Framework Programme project ASPIC, and was hosted by the University of Liverpool in 2006. Since then a steering committee promoting the continuation of the conference was established, and the subsequent steady growth of worldwide interest in computational argumentation has gone hand in hand with the development of the conference itself, and of related activities. Specifically, plenary invited talks by world-leading researchers, and a software demonstration session, were introduced to the programme in the second edition (COMMA'08) hosted by IRIT in Toulouse. COMMA'10, hosted by the University of Brescia in Desenzano del Garda, saw the addition of a best student paper award, and the same year saw inauguration of a new journal – Argument and Computation – closely related to the COMMA community. The fourth edition, organized by the Vienna University of Technology in 2012, introduced an Innovative Application Track and the proceedings now included a section for Demonstration abstracts. COMMA'14, hosted by the Universities of Aberdeen and Dundee in Pitlochry, was preceded by the first Summer School on Argumentation: Computational and Linguistic Perspectives, while the same year also saw the launch of the first International Competition on Computational Models of Argumentation (ICCMA). The sixth edition of COMMA, organized by the University of Potsdam, included two additional satellite workshops: Systems and algorithms for formal argumentation and Foundations of the language of argumentation.
This year COMMA is hosted by the Institute of Philosophy and Sociology of the Polish National Academy of Sciences in Warsaw, Poland. It is a testament to both the broad multidisciplinary interest in argumentation and the increasing body of work establishing the relevance of computational models to these various disciplines, and in real world applications, that COMMA'18 is one among a number of events constituting the Warsaw Argumentation Week (WAW). Indeed, WAW also includes the Third Summer School on Argumentation: Computational and Linguistic Perspectives, as well as the 16th ArgDiaP Conference on Argumentation and Corpus Linguistics and two colocated workshops on methodologies for research on rhetoric and methodologies for research on legal argumentation.
The COMMA'18 programme reflects the broad reach and burgeoning interdisciplinary focus and application uses of computational models of argument. COMMA'18 includes the second edition of the workshop Systems and algorithms for formal argumentation, as well as the additional workshops: Argumentation and Society and Argumentation and Philosophy. For the first time, the programme also includes an industry afternoon bringing together businesses, NGOs, academics and students interested in practical applications of argument technologies in industry. The topic of argument mining has substantial representation, both in the accepted papers and demos, and in the invited talks. Dr. Noam Slonim, head of a team at the IBM Haifa Research Lab, presents his team's ground breaking work on debating technologies, and a stalwart of the computational argumentation community – Professor Francesca Toni from Imperial College London – gives an invited talk that traverses the pipeline from argument mining to argumentation semantics. The programme also includes papers reporting on algorithm development, innovative applications, argumentation-based models of dialogue, and abstract argument frameworks whose various types of argument relations reflect the diversity of human uses of argument. A notable recent trend has been the extent to which development of computational models of argument are being informed by the use of argument and proof in mathematical, scientific and philosophical contexts. This is reflected in papers in these proceedings, and in the invited talk given by Marcello D'Agostino, professor of philosophy at the University of Milan. Finally, we point out the presence of papers on structured argumentation, many of which adhere to the Dung paradigm wherein arguments in some formal language are related by conflict based relations in Dung argument frameworks, and variants thereof. It is a testament to the profound and widespread influence of P.M. Dung's work, that his seminal paper On the Acceptability of Arguments and its Fundamental Role in Nonmonotonic Reasoning, Logic Programming and N-person Games, first published in 1995, was awarded the 2018 AIJ Classic Paper Award, which recognizes a paper published at least 15 years ago that has shown to be exceptional in its significance and impact. Indeed, Dung would we believe look favourably on many of the developments reflected in the COMMA'18 programme, and the co-located events in WAW. His paper's dialectical characterisations of non-monotonic inference explicitly aimed at relating nonmonotonic logics to real world reasoning and debate.
We conclude by acknowledging that the success of a conference and its co-located events, depends on the contributions of many people. We would like to thank IOS press for publishing these proceedings and continuing to make them Open Access. We thank the invited speakers, Francesca Toni, Noam Slonim and Marcello D'Agostino for their insightful and inspiring talks. We acknowledge steady support and encouragement by the COMMA Steering Committee, and are very grateful to the Programme Committee and additional reviewers whose invaluable expertise and efforts have led to the selection of 25 full papers and 17 short papers, out of a total of 70 submissions, and 15 demonstration abstracts. The submission and reviewing process has been managed through the Easychair conference system, which we acknowledge for supporting COMMA since the first edition. We would like to express our gratitude for the considerable efforts of the local organizing committee Marcin Koszowy and Maria Załęska, and Brian Plüss for managing the COMMA'18 website. Our thanks also to Barbara Konat and Jakub Zygucki, coordinators of the industry afternoon, the COMMA'18 workshop chairs Matthias Thimm, Katie Atkinson and Jacky Visser, and the summer school chairs Bartłomiej Skowron and Magdalena Kacprzak. Last but not least, we thank all the authors for contributing to the success of the conference with their hard work and commitment.
Sanjay Modgil (Programme chair)
Katarzyna Budzynska (Conference chair)
John Lawrence (Demonstrations chair)
London/Warsaw/Dundee July 2018
Argumentation frameworks have been widely studied both in terms of formal properties they exhibit under different semantics and in terms of applications they can support. But where are argumentation frameworks coming from, and how can argumentation, a model-based approach to AI, beneficially integrate with the nowadays-much-widespread data-centric AI perspective? In this talk I will overview applications empowered by a variety of (extension-based and gradual) semantics for abstract and bipolar argumentation frameworks automatically obtained from data (including but not limited to text) and from logical rules. Some of these applications require the integration of argumentation and machine learning, and result in a mixed model-based and data-centric pipeline. For some applications, the semantics informs the definition of the frameworks rather than, as is conventionally the case, being enforced on frameworks a posteriori.
Project Debater is the first AI system that was shown to debate humans in a meaningful manner in a full live debate. Developing this system started in 2012, as the next AI Grand Challenge pursued by IBM Research, following the demonstration of Deep Blue in Chess in 1997, and Watson in Jeopardy! In 2011. The Project Debater system was demonstrated for the first time in San Francisco in June 2018, in two full live debates vs. expert human debaters, and correspondingly received massive media attention. This talk will present the challenges in developing this system, its current capabilities and present limitations, as well as how we envision its future.
The argumentation turn in logic paves the way for importing problems and ideas from general philosophy of science into logical formalisms. The main problems are: when faced with an inconsistency, how do we choose the hypotheses/beliefs that are the most plausible candidates for revision? Do we really need to react to each single inconsistency or should we focus on predictive success and temporarily ignore “minor” inconsistencies? How do we distinguish ad hoc adjustments from heuristically fruitful revisions? These questions call for a sophisticated notion of “epistemic entrenchment”. The basic idea has played an important role both in the area of belief revision (Gärdenfors and Makinson) and, under different denominations, in that of general philosophy of science (Quine, Kuhn, Lakatos), with little or no interaction between them. The main aim of this talk is to bring the philosophical debate on this notion to the attention of the formal argumentation community and suggest how this interaction could lead to substantial advances in the field.
We investigate Dung's semantics for assumption-based frameworks (ABFs) that are induced by contrapositive logics. We show that unless the falsity propositional constant is part of the defeasible assumptions, the grounded semantics lacks most of the desirable properties it has in abstract argumentation frameworks (AAFs), and that for simple definitions of the contrariness operator and the attacks relations, preferred and stable semantics are reduced to naive semantics. We also show the tight relations of this framework to reasoning with maximally consistent sets, and consider some properties of the induced entailments, such as being cumulative or preferential relations that satisfy non-interference.
In Dung's abstract argumentation theory, an extension can be represented by subsets of it in the sense that from each of these subsets, the extension can be obtained again by iteratively applying the characteristic function. Such so-called regular representations can be used to differentiate argumentation frameworks having the same extensions. In this paper we provide a full characterization of relations between seven different types of representation equivalence.
In this paper, we propose a generalisation of Dung's abstract argumentation framework that allows representing higher-order attacks and supports, that is attacks or supports whose targets are other attacks or supports. We follow the necessary interpretation of the support, based on the intuition that the acceptance of an argument requires the acceptance of each supporter. We propose semantics accounting for acceptability of arguments and validity of interactions, where the standard notion of extension is replaced by a triple of a set of arguments, a set of attacks and a set of supports. Our framework is a conservative generalisation of Argumentation Frameworks with Necessities (AFN). When supports are ignored, Argumentation Frameworks with Recursive Attacks are recovered.
We present a probabilistic interpretation of the plausibility of attacks in abstract argumentation frameworks by extending the epistemic approach to probabilistic argumentation with probabilities on attacks. By doing so we also generalise the previously proposed attack semantics by Villata et al. to the probabilistic setting and provide a fine-grained assessment of the plausibility of attacks. We also consider the setting where partial probabilistic information on arguments and/or attacks is given and missing probabilities have to be derived.
In this paper, we consider SETAFs due to Nielsen and Parsons, an extension of Dung's abstract argumentation frameworks that allow for collective attacks. We first provide a comprehensive analysis of the expressiveness of SETAFs under conflict-free, naive, stable, complete, admissible and preferred semantics. Our analysis shows that SETAFs are strictly more expressive than Dung AFs. Towards a uniform characterization of SETAFs and Dung AFs we provide general results on expressiveness which take the maximum degree of the collective attacks into account. Our results show that, for each k>0, SETAFs that allow for collective attacks of k+1 arguments are more expressive than SETAFs that only allow for collective attacks of at most k arguments.
Abstract dialectical frameworks (ADFs) are generalizations of Dung argumentation frameworks where arbitrary relationships among arguments can be formalized. This additional expressibility comes with the price of higher computational complexity, thus an understanding of potentially easier subclasses is essential. Compared to Dung argumentation frameworks, where several subclasses such as acyclic and symmetric frameworks are well understood, there has been no indepth analysis for ADFs in such direction yet (with the notable exception of bipolar ADFs). In this work, we introduce certain subclasses of ADFs and investigate their properties. In particular, we show that for acyclic ADFs, the different semantics coincide. On the other hand, we show that the concept of symmetry is less powerful for ADFs and further restrictions are required to achieve results that are similar to the known ones for Dung's frameworks. We also provide experiments to analyse the performance of solvers when applied to particular subclasses of ADFs.
Many recent studies of dynamics of formal argumentation in AI focus on the well-known formalism of argumentation frameworks (AFs). Despite their use-fulness in many areas of argumentation, their abstract notion of arguments creates a barrier for operators that modify a given AF, namely in the case that dependencies between arguments have been abstracted away that might be subsequently missed. In this paper we aim to support development of dynamic operators on formal models in abstract argumentation by providing constraints imposed on the modification of the structure that can be used to incorporate information that has been abstracted away. Towards a broad reach, we base our results on the general formalism of abstract dialectical frameworks (ADFs) in abstract argumentation, and study the complexity of the proposed structural constraints. To show applicability, we adapt an extension enforcement operator on AFs to ADFs that is allowed to only add support relations between arguments. We show feasibility of our approach by an experimental evaluation of an implementation of this operator.
While work on abstract argumentation frameworks has greatly advanced the study of argumentation in AI, its use is not without danger. One danger is that the direct modelling of examples in abstract frameworks instead of through a theory of the structure of arguments and the nature of attacks leads to ad-hoc modellings. Another danger is that it may be overlooked that abstract accounts of argumentation can implicitly make assumptions that are not shared by many of their instantiations. A variant of this is where assumptions valid for specific argumentation contexts are incorrectly generalised by abstracting away from the context. This paper gives examples of both dangers. A lesson drawn from this is that abstraction in AI research, although necessary for understanding the essentials of the object of study, can oversimplify in ways that are not easily noticed without an explicit account of the structure of arguments and the nature of attack.
We directly instantiate metalevel argumentation frameworks (MAFs) to enable argumentation-based reasoning about information relevant to various applications. The advantage of this is that information that typically cannot be incorporated via the instantiation of object-level argumentation frameworks can now be incorporated, in particular information referencing (1) preferences over arguments, (2) the rationale for attacks, and (3) the dialectical effect of critical questions that shifts the burden of proof when posed. We achieve this by using a variant of ASPIC+ and a higher-order typed language that can reference object-level formulae and arguments. We illustrate these representational advantages with a running example from clinical decision support.
We relate the ANGELIC methodology for acquiring and encapsulating domain knowledge to the ASPIC+ framework for structured argumentation. In so doing we hope to facilitate the building of applications in concrete domains by linking a successful methodology to a proven theoretical framework. We use an example from the ASPIC+ literature to illustrate the relationship.
This discussion paper describes the “Laboratory of Dilemmas”, a paradigmatic decision problem presented as a video installation at La Biennale of Venice, and discusses the challenges it poses for argument-based models of decision making. It then sketches and compares two investigation directions to address these challenges and provides some relevant preliminary technical observations.
In many expert and everyday reasoning contexts it is very useful to reason on the basis of defeasible assumptions. For instance, if the information at hand is incomplete we often use plausible assumptions, or if the information is conflicting we interpret it as consistent as possible. In this paper sequent-based argumentation, a form of logical argumentation in which arguments are represented by a sequent, is extended to incorporate assumptions. The resulting assumptive framework is general, in that some other approaches to reasoning with assumptions can adequately be represented in it. To exemplify this, we show that assumption-based argumentation can be expressed in assumptive sequent-based argumentation.
Argument-based decision making has been employed to support a variety of reasoning tasks over medical knowledge. These include evidence-based justifications of the effects of treatments, the detection of conflicts in the knowledge base, and the enabling of uncertain and defeasible reasoning in the health-care sector. However, a common limitation of these approaches is that they rely on structured input information. Recent advances in argument mining have shown increasingly accurate results in detecting argument components and predicting their relations from unstructured, natural language texts. In this study, we discuss evidence and claim detection from Randomized Clinical Trials. To this end, we create a new annotated dataset about four different diseases (glaucoma, diabetes, hepatitis B, and hypertension), containing 976 argument components (697 containing evidence, 279 claims). Empirical results are promising, and show the portability of the proposed approach over different branches of medicine.
Much research in computational argumentation assumes that arguments can be obtained in some way. Yet, to improve and apply models of argument, we need methods for acquiring them. Current approaches include argument mining from text, hand coding of arguments by researchers, or generating arguments from knowledge bases. In this paper, we propose a new approach, which we call argument harvesting, that uses a chatbot to enter into a dialogue with a participant to get arguments and counterarguments from him or her. Because it is automated, the chatbot can be used repeatedly in many dialogues, and thereby it can generate a large corpus. We describe the architecture of the chatbot, provide methods for clustering arguments by their similarity and value, and an evaluation of our approach in a case study concerning attitudes of women to participation in sport.
Endorsing the character of allies and destroying credibility of opponents is a powerful tactic for persuading others, impacting how we see politicians and how we vote in elections, for example. Our previous work demonstrated that ethos supports and attacks use different language, we hypothesise that further distinctions should be made in order to better understand and implement ethotic strategies which people use in real-life communication. In this paper, we use the Aristotelian concept of elements of ethos: practical wisdom, moral virtue and goodwill, to determine specific grounds on which speakers can be endorsed and criticised. We propose a classification of types of ethos supports and attacks which is empirically derived from our corpus. The manual classification obtains a reliable Cohen's kappa κ=0.52 and weighted κ=0.7. Finally, we develop a pipeline to classify ethos supports and attacks into their types depending on whether endorsement or criticism is grounded in wisdom, virtue or goodwill. The automatic classification obtains a solid improvement of macro-averaged F1-score over the baseline of 10%, 25%, 9% for one vs all classification, and 16%, 18%, 10% for pairwise classification.
We present a family of stochastic local search algorithms for finding a single stable extension in an abstract argumentation framework. These incomplete algorithms work on random labellings for arguments and iteratively select a random mislabeled argument and flip its label. We present a general version of this approach and an optimisation that allows for greedy selections of arguments. We conduct an empirical evaluation with benchmark graphs from the previous two ICCMA competitions and further random instances. Our results show that our approach is competitive in general and significantly outperforms previous direct approaches and reduction-based approaches for the Barabási-Albert graph model.
We propose natural generalizations of the credulous and skeptical acceptance problems in abstract argumentation for incomplete argumentation frameworks . This continues earlier work on a similar generalization of the verification problem. We provide a full analysis of the computational complexity of the generalized problems for all original semantics, showing that, in almost all cases, acceptance problems for incomplete argumentation frameworks are significantly harder than the respective problems for argumentation frameworks without uncertainty. All our hardness results for the classes NP, coNP, Πp2, and Σp2 are derived from one generic reduction.
We present a computational study of effectiveness of declarative approaches for three optimization problems in the realm of abstract argumentation. In the largest extension problem, the task is to compute a σ-extension of largest cardinality (rather than, e.g., a subset-maximal extension) among the σ-extensions of a given argumentation framework (AF). The two other problems considered deal with a form of dynamics in AFs: given a subset S of arguments of an AF, the task is to compute a closest σ-extension within a distance-based setting, either by repairing S into a σ-extension of the AF, or by adjusting S to be a σ-extension containing (or not containing) a given argument. For each of the problems, we consider both iterative Boolean satisfiability (SAT) based approaches as well as directly solving the problems via Boolean optimization using maximum satisfiability (MaxSAT) solvers. We present results from an extensive empirical evaluation under several AF semantics σ using the ICCMA 2017 competition instances and several state-of-the-art solvers. The results indicate that the choice of the approach can play a significant role in the ability to solve these problems, and that a specific MaxSAT approach yields quite generally good results. Furthermore, with impact on SAT-based AF reasoning systems more generally, we demonstrate that, especially on dense AFs, taking into account the local structure of AFs can have a significant positive effect on the overall solving efficiency.
To address the needs of the EU NoAW project, in this paper we solve the problem of efficiently generating the argumentation graphs from knowledge bases expressed using existential rules. For the knowledge bases without rules, we provide a methodology that allows to optimise the generation of argumentation graphs. For knowledge bases with rules, we show how to filter out a large number of arguments and reduce the number of attacks.