by K Atkinson · Cited by 152 — Abstract argumentation theory is probably the sub- field of computational models of argument that has attracted most research attention in the last two decades,

122 KB – 12 Pages

PAGE – 1 ============
Humans argue. 1This distinctive feature is at the same time an important cognitive capacity and a powerful social phenomenon. It has attracted attention and careful analysis since the dawn of civilization, being inti- mately related to the origin of any form of social organiza- tion, from political debates to law, and of structured think- ing, from philosophy to science and arts.As a cognitive capacity, argumentation is important for handling con !icting beliefs, assumptions, viewpoints, opin- ions, goals, and many other kinds of mental attitudes. When we are faced with a situation where we “nd that our infor- mation is incomplete or inconsistent, we often resort to the use of arguments in favor and against a given position in order to make sense of the situation. When we interact with other people we often exchange arguments in a cooperative or competitive fashion to reach a “nal agreement or to defend and promote an individual position.Articles FALL 2017 25 Copyright © 2017, Association for the Advancement of Arti “cial Intelligence. All rights reserved. ISSN 0738-4602 Toward Arti ! cial Argumentation Katie Atkinson, Pietro Baroni, Massimiliano Giacomin, Anthony Hunter, Henry Prakken, Chris Reed, Guillermo Simari, Matthias Thimm, Serena Villata !The !eld of computational models of argument is emerging as an important aspect of arti !cial intelligence research. The reason for this is based on the recognition that if we are to develop robust intelligent systems, then it is imperative that they can handle incom- plete and inconsistent information in a way that somehow emulates the way humans tackle such a complex task. And one of the key ways that humans do this is to use argumentation either internally, by evaluating arguments and counterargumentsâ or externally, by for instance entering into a discussion or debate where arguments are exchanged. As we report in this review, recent devel- opments in the !eld are leading to tech- nology for arti !cial argumentation, in the legal, medical, and e-government domains, and interesting tools for argu- ment mining, for debating technologies, and for argumentation solvers are emerging.

PAGE – 2 ============
Occurring continuously both in our mind and in the social arena, argumentation pervades our intelli- gent behavior and the challenge of developing arti “-cial argumentation systems appears to be as diverse and exciting as the challenge of arti “cial intelligence itself.Indeed, this rich and important phenomenon offers an opportunity to develop models and tools for argumentation and to conceive autonomous arti “cialagents that can exploit these models and tools in the cognitive tasks they are required to carry out. To this purpose, a number of interesting lines of research are being investigated within arti “cial intelligence and several neighbor “elds, leading to the establishment of computational models of argument as a promising interdisciplinary research area. Progress in this area is expected to contribute to signi “cant advances in the understanding and modeling of various aspects of human intelligence.In this article, we review formalisms for capturing various aspects of argumentation, and we present advances in their applications, with the aim to com- municate how research is making progress toward the goal of making arti “cial argumentation tech- nologies and systems a mature and widespread reali- ty. In this brief review, we are unable to discuss or cite all the relevant literature, and we suggest that the interested reader seek more detailed coverage of the foundations from Rahwan and Simari (2009), of applications from Modgil et al. (2013), and of recent developments from the proceedings of the Interna- tional Conference on Computational Models of Argument series, 2and the Argument and Computation journal.3Models of ArgumentComputational models of argument are being devel- oped to re !ect aspects of how humans build, exchange, analyze, and use arguments in their daily lives to deal with a world where the information may be controversial, incomplete, or inconsistent (Bench- Capon and Dunne 2007, Rahwan and Simari 2009). The diversity of the manifestations of arguments in real life implies diversity in the relevant models too and the impossibility to reduce the vast available lit- erature to a single reference scheme. It is possible however to identify some layers that can be regarded as basic building blocks for the construction of an argumentation model. Speci “c modeling approaches may differ in the selection of which layers they actu- ally use, in the way the selected layers are combined, and in the formalization adopted within each layer. We consider the following “ve main layers: struc- tural, relational, dialogical, assessment, and rhetori- cal. They are described in the following and also sum- marized in Figure 1. Note that while each layer has its own nature and distinctive traits, the boundaries between layers may not be so neat in some contexts, and speci “c formalisms may inextricably merge together aspects relevant to different layers.Structural LayerThis layer concerns the structure of the arguments and how they are built: essentially it speci “es, in a given context, what an argument looks like, in terms of its internal structure, and which are the ingredi- ents for its construction. To exemplify, in contexts where arguments are built from a logical knowledge base the ingredients are the logical formulae includ- ed in the knowledge base. Then one way to build arguments is by simply applying the logic of the lan- guage in which the knowledge base is stated to derive conclusions. An argument here can be seen as a pair (!, “) where !is a subset of the knowledge base (a set of formulae) that logically entails “(a formula). Here, !is called the support, and “is the claim, of the argu- ment. Other approaches consider argument con- struction from knowledge bases as applying rules to the formulas from the knowledge base, where the rules may be defeasible. In these rule-based approach- es an argument is typically seen as a tree whose root is the claim or conclusion, whose leaves are the premises on which the argument is based, and whose structure corresponds to the application of the rules from the premises to the conclusion. Investigations into the structural layer were initiated by Pollock (1992). Prominent examples of rule-based formalisms are ASPIC+, assumption-based argumentation (ABA), and defeasible logic programming (DeLP). For a tuto- rial introduction to formalisms for structured argu- mentation, see Besnard et al. (2014).Arguments are not built from knowledge bases only, however. For instance, interactive systems that acquire arguments from users may adopt the approach of argumentation schemes (Walton, Reed, and Macagno 2008), namely stereotypical reasoning patterns, where in addition to the premises and the claim, a set of critical questions is considered. Criti- cal questions provide a sort of checklist of issues that can be raised to challenge arguments built on the basis of a given scheme. Argumentation schemes have also been used as a source of defeasible infer- ence rules in rule-based approaches to argument con- struction from knowledge bases. In addition, argu- mentation schemes are often considered in the context of argument mining (see the Argument Min- ing section) where the goal is to identify and extract the argumentative structures embedded in a natural language source, providing a machine-processable representation of them.The variety of existing argument models raises the issue of exchanging or sharing arguments among dif- ferent systems. This problem is addressed by the argument interchange format initiative (ChesŒevar et al. 2006), aimed at providing an interlingua between various more concrete argumentation lan- guages, on the basis of a generic abstract ontology. Articles 26AI MAGAZINE

PAGE – 3 ============
Articles FALL 2017 27 Relational LayerArguments do not live in isolation and are linked to each other by various types of relations: the relation- al layer deals with identifying and formally repre- senting them, in view of their use in other layers or even for descriptive and presentation purposes, since they are essential for an understanding of what is actually going on in an argumentation process. Examples of important relationships are (1) the sub- argument and superargument relationships, indicat- ing how an argument is built incrementally on top of other arguments; (2) the attack relationship, indicat- ing that an argument is incompatible with another argument in some sense, for example, because they have contradictory claims, or one claim contradicts some premise or assumption on which the other is based; (3) the support relationship, intuitively mean- ing that an argument provides some backing to another, and admitting several, even rather dissimi- lar, interpretations, depending on the actual nature of this backing; (4) a preference relationship, ranking arguments according to some criterion, and admit- ting again a variety of instantiations ranging from strength to credibility to value-based evaluations.What relationships are signi “cant and how to identify them are highly context-dependent matters. Note in particular that identifying argument rela- tions may be an easy mechanical procedure in set- tings where arguments are formally built from a knowledge base, while in an argument mining sce- nario it is a task as challenging as the identi “cationof the arguments themselves.Dialogical LayerThis layer deals with the exchange of arguments among different agents (or even between an agent and itself, in a scenario where argumentative reason- ing is conceived as a monological activity) according to formal dialogue rules. Agents may engage in the exchange of arguments for a variety of purposes with several dialogue types having been identi “ed in the literature, like inquiry, negotiation, information- seeking, deliberation, and persuasion. In all cases the exchange can be formalized as a dialogue game, which is normally made up of a set of communica- tive acts called moves, and a protocol specifying Figure 1. Key Aspects of Argumentation. Structural layer: How are arguments constructed?Relational layer: What are the relationships between arguments?Dialogical layer: How can argumentation beundertaken in dialogues?Assessment layer: How can a constellation of interacting argumentsbe evaluated and conclusions drawn?Rhetorical layer: How can argumentation be tailored for an audience so that it is persuasive?

PAGE – 4 ============
which moves can be made at each step of the dia- logue. It concerns representing and managing the locutions exchanged between the agents involved, as well as specifying the contents of these locutions in terms of entire arguments or components of argu- ments. Moreover, the dialogue protocol may establish the allowed moves on the basis of argument relation- ships. For instance, a protocol may specify that a move is legal only if it presents an argument attack- ing an argument presented in a previous move. For these reasons the dialogical layer requires strict con- nections with the structural and relational layers. Moreover some dialogue protocols are de “ned so as to embed an argument assessment method: in these cases the dialogical layer is intertwined with the assessment layer, described next. Assessment LayerThis layer concerns the assessment of a set of argu- ments and of their conclusions in order to establish their justi “cation status. The need for this layer arises from the presence of attacks among arguments, pre- venting them so as to be accepted altogether and call- ing for a formal method to solve the con !ict. This problem is addressed in a principled and highly styl- ized form in the context of the theory of abstract argu- mentation frameworks (Dung 1995), where argu- ments are treated as abstract entities, deprived of any structural property and of all their relations but attack. We give an example of an argumentation framework, based on textual arguments, in “gure 2. Given its abstract nature, an argumentation framework is often referred to as argument graph, and this term is also used to refer to similar representations where addi- tional relations, like support, are considered.An abstract argumentation semantics is a formal cri- terion to determine which sets of arguments, called extensions, are able to survive the con !ict together and can be regarded as collectively acceptable. Abstract argumentation theory is probably the sub- “eld of computational models of argument that has attracted most research attention in the last two decades, due to its generality and theoretical clean- ness. In particular Dung (1995) has shown the ability of the formalism to capture as instances several other approaches, especially in the area of nonmonotonic reasoning. DungÕs approach abstracts from the origin and nature of the attack relation. A natural idea is to de”ne this relation in terms of a more basic notion of con!ict between arguments (for example, two argu- ments having contradictory conclusions) and a notion of relative argument strength or preference. In the literature, there are two ways to connect these ideas to DungÕs frameworks. The “rst approach leaves DungÕs frameworks as they are but connects them with models at the structural layer of argument to de”ne attack in terms of preferences or argument strength while taking the structure of arguments into account. The second approach instead extends DungÕs frameworks with some abstract notion of argument strength or preference, while possibly also adding an abstract support relation between argu- ments. Moreover, while most approaches consider a qualitative notion of acceptance, quantitative assess- ments methods are being investigated too.Further, it must be noted that the evaluation of argument acceptability is only a part, actually the most basic one, of the assessment tasks required in an argumentative process. In particular the “nal goal of an agent is usually the assessment of the justi “cationstatus of the statements supported by arguments, which, in the end, amounts to determining what to believe or what to do. Since many arguments may have the same conclusion, assessing the status of a statement involves a synthesis of the statuses of the arguments supporting it. As in real life, the task of deciding what to believe may be carried out adopting different attitudes, ranging from extremely skeptical to extremely credulous, corresponding to different formal methods for statement justi “cation synthesis. Articles 28AI MAGAZINE Figure 2. An Example of Argumentation Framework Consisting of Three Arguments in the Medical Domain and Their Attacks. Arguments A1and A2are two alternatives for treating a patient with hypertension, and A3provides a reason against one of the options. Here, we assume that A1and A2attack each other because giving one treatment precludes the other, and we assume that A3attacks A2because it provides a counterargument to A2.A1 = Patient hashypertension soprescribe diureticsA2 = Patient hashypertension so prescribe betablockers A3 = Patient has emphysema, whichis a contraindicationfor betablockers

PAGE – 5 ============
Rhetorical LayerNormally argumentation is undertaken in some wider context of goals for the agents involved, and so individual arguments are presented with some wider aim and according to some strategical consid- erations. For instance, if an agent is trying to per- suade another agent to do something, then it is like- ly that some rhetorical device is harnessed and this will re !ect the nature of the arguments used (for example, a politician may refer to investing in the future of the nationÕs children as a way of persuading colleagues to vote for an increase in taxation). With the roots of the study of rhetoric going back to Aris- totle,4recent studies into aspects of the rhetorical lev- el include believability and impact of arguments from the perspective of the audience, use of threats and rewards, appropriateness of advocates, and val- ues of the audience. The rhetorical layer may be absent in some contexts, for example, when neu- trally building arguments from a knowledge base, but can permeate all the other layers in other contexts since goal-oriented considerations may drive the decisions of which arguments to build, taking into account their relations with other arguments, of whether, how, and when to use the arguments in a dialogue, and of which assessment method (for example, whether a more skeptically or more credu- lously oriented one) to apply. The following sections review several prominent domains that exploit computational models of argu- ment for the development of actual applications and, at the same time, stimulate the relevant theoretical development by providing case studies and impor- tant modeling challenges.Legal ArgumentationThe law is an obvious application domain for argu- mentation research, since legal reasoning is essen- tially argumentative and to a large extent recorded in documents. This has led to highly stylized forms of argumentation, which makes it easier to formulate and validate formal and computational models of argument than in many other domains. In this sec- tion, we brie !y discuss work and research themes in this area. A more detailed survey can be found in the paper by Prakken and Sartor (2015).In legal cases, “rst the facts have to be determined. Because of the diverse nature of the evidence in most cases and the need for explanation to statistical laypeople, legal evidential reasoning is an excellent test bed for combined qualitative and quantitative models of defeasible reasoning. At the practical side, so-called sense-making systems have been proposed, with which crime investigators or triers of fact can structure their arguments and scenarios to make sense of a large body of evidence.After the facts of a case have been established, they must be classi “ed under the conditions of legal rules, which involves interpreting these rules. Two in !uen-tial AI and law models of this are the HYPO system by Kevin Ashley and its successor CATO by Vincent Aleven, which model how lawyers in common-law jurisdictions make use of past decisions when argu- ing a case. Their underlying argumentation model is for factor- or dimension-based reasoning, where cas- es are collections of abstract fact patterns that favor or oppose a conclusion, either in an all-or-nothing fashion (factors) or to varying degrees (dimensions). This work inspired subsequent formal work using the tools of formal argumentation, resulting in formal- ized versions of traditional legal argument forms such as appeal to precedent and policy and the bal- ancing of goals, values, and interests (Horty and Bench-Capon 2012).Finally, when the facts have been classi “ed, the legal rules must be applied to them. Legal rules can have exceptions or con !ict on other grounds. Rule- based argumentation logics with preferences have proved useful here.Legal reasoning usually takes place in the context of a dispute between adversaries, within a prescribed legal procedure. This makes the setting inherently dynamic and multiparty, and raises issues of strategy and choice. For example, there is work on optimal strategies for adversaries in debates with an adjudica- tor, given their preferences over the possible out- comes of a debate and their estimates of what the adjudicator will likely accept.While the theoretical advances on models of legal argument have been impressive and a number of valuable prototype systems have been developed, no systems have been deployed in everyday practice yet. One reason is the conservative attitude to technology in the legal world and its billing-by-the-hour culture, which does not stimulate innovation. Another reason is the fact that building realistic systems of legal argu- ment requires vast amounts of commonsense knowl- edge. However, recently things have changed. First, clients of law “rms increasingly demand the use of modern technology. Moreover, the recent advances in natural language processing, machine learning, and data science combined with the massive digital avail- ability of legal data and information have created the prospects for combining AI models of legal argument with argument mining techniques. In fact, two of the “rst argument mining projects were on legal argu- ment (Palau and Moens 2011). If this technology is combined with AI and lawÕs computational models of argument, then practical applications of these models could be well within reach.Medical ArgumentationHealth care is a potentially important domain for developing and applying computational models of argument. It is common for health-care information to be complex, heterogeneous, incomplete, and Articles FALL 2017 29

PAGE – 6 ============
inconsistent, and therefore argumentation is appeal- ing for those involved as it allows for important con- !icts to be highlighted and analyzed and unimpor- tant con!icts to be suppressed.One of the pioneers of argumentation technology, John Fox, developed a number of prototype systems for medical decision support such as the Capsule sys- tem (Walton et al. 1997). Capsule supports a family practitioner when she or he is about to prescribe a speci”c drug for a patient. The system uses a standard database of equivalent treatments that is routinely used by clinicians, and the patient records, to provide arguments pro and con each of the alternatives. The arguments are based on whether the patient has pre- viously expressed a preference for or against the alter- native, whether the patient has previously exhibited a negative reaction to the alternative, whether there is possible negative interaction with other drugs being taken by the patient, and the relative cost of the alternative. In a formal trial of the Capsule system, with 42 clinicians using 36 simulated cases, the sys- tem was shown to help clinicians improve the quali- ty of their prescribing and to improve their compli- ance with medical guidelines.Over recent years, there has been substantial shift in health care to evidence-based practice. This means that health-care professionals need to use the best available evidence to inform their decision making. For deciding on interventions, this normally calls for evidence from randomized clinical trials. The prob- lem with this is that there are many such trials pub- lished each year, and it is dif “cult for clinicians to keep abreast of this literature. To help them, there are medical guidelines and systematic reviews that aggre- gate this evidence by providing recommendations. Unfortunately, these recommendations can rapidly become out of date, they do not take local circum- stances into account, and they normally do not con- sider patients with comorbidities. To address these problems, an argument-based approach to aggregat- ing clinical evidence has been proposed by Hunter and Williams (2012). The framework is a formal approach to synthesizing knowledge from clinical tri- als involving multiple outcome indicators (where an outcome indicator is either positive such as the num- ber of patients who survive the disease after 1 year, or 2 years, and so on, or negative such as the proportion of those treated who have a particular side-effect). Based on the available evidence, evidence-based argu- ments are generated for claiming that one treatment is superior to another for a given patient.Preference criteria over evidence-based arguments are speci “ed in terms of the outcome indicators, and the magnitude of those outcome indicators, in the evidence. Various kinds of counter-arguments attack the evidence-based arguments depending on the quality of evidence used (for example, evidence could be attacked because a trial was not conducted correctly). The arguments and counter-arguments constitute an argument graph, and using abstract argumentation semantics, the winning arguments are identi”ed, and thereby argument-based recommen- dations for which treatments are superior can be identi”ed. The approach has been evaluated by com- parison with recommendations made in published health-care guidelines (Hunter and Williams 2012) and it has been used to publish, in the medical liter- ature on lung cancer, a more re “ned systematic review of the evidence. An ongoing study is using this technique in a systematic review on brain cancer for publication by Cochrane.These examples are just two of a number of appli- cations of argumentation being developed for sup- porting health-care professionals and patients. Fur- ther applications include dealing with the con !ictsthat can occur when using multiple clinical guide- lines, supporting multidisciplinary teams of health- care professionals when dealing with dif “cult clinical cases, and supporting medical image interpretation.e-GovernmentAn important feature of democracies is that citizens can engage their governments in dialogues about policies. Traditionally this was done by writing letters and holding town hall debates, but over the past two decades new methods of interaction have been devel- oped to exploit the bene “ts of current technology. Citizens may wish to respond in several ways to pol- icy proposals made by their governments. They may simply seek a justi “cation of the proposed policy; they may wish to object to the proposed policy; or they may want to propose policies of their own. Such dialogues can be facilitated through tools to support e-democracy, and computational models of argument can be put to effective use in such tools.For example, consider a local government authori- ty that is deciding what to build on a plot of waste- land in a community. One proposal by the local authority may be to permit the building of a new supermarket on the grounds that this will provide jobs for the local community and shopping facilities for local residents. Citizens may be consulted on this proposal and critique this policy as well as put for- ward their own proposals. For example, the local authorityÕs proposal could be critiqued by stating that the action of building a new supermarket will not have the intended effect of creating jobs as there will be job losses from local shop owners being put out of business by the supermarket. An alternative proposal could be that instead of building a new supermarket, the site should be used to build a new play center for local residentsÕ children. Such opinions can be formed into arguments by distinguishing the premis- es (for example, there is little unemployment in the community and play centers promote social interac- tion) and conclusion (we should build a new chil- drenÕs play center). Argumentation-based tools can Articles 30AI MAGAZINE

PAGE – 8 ============
Random breath tests to public vehicle drivers can hard- ly be called an invasion of privacy or an investigation without due cause, because public safety is at stake. Random tests are routinely carried out by many train and bus companies and are being introduced on air- lines as well. The same applies for other drivers, who are a major liability to the safety and lives of other drivers.Randomly testing employees cannot be considered an invasion of privacy. People who have to take random breath tests to drive trucks or fly planes as part of their jobs are taking the test as part of their job. They are being paid and must do what their employer wants them to do in order to keep their job. Searching ran- dom people outside of the context of employment with no suspicion of a crime is very different as it erodes civil liberties and sets a dangerous precedent.The “rst goal of the argument mining pipeline con- sists in extracting the arguments from this text. In the previous example, we highlight the four arguments that can be identi”ed:Random breath tests to public vehicle drivers can hard- ly be called an invasion of privacy or an investigation without due cause, because public safety is at stake [A1].Random tests are routinely carried out by many train and bus companies and are being introduced on air- lines as well. The same applies for other drivers, who are a major liability to the safety and lives of other drivers [A2].Randomly testing employees cannot be considered an invasion of privacy [A3]. People who have to take ran- dom breath tests to drive trucks or !y planes as part of their jobs are taking the test as part of their job. They are being paid and must do what their employ- er wants them to do in order to keep their job. Search- ing random people outside of the context of employment with no suspicion of a crime is very different as it erodes civil liberties and sets a dangerous precedent [A4].Given these four arguments (that is, A1, A2, A3, and A4), the relations among them have to be identi “ed.Let us consider for the explanatory purpose of this example that the two relations we aim at identifying are the attack and the support relations only. In this case, we have that, taking into account the temporal dimension of the debate to decide the direction of the relations, argument A3supports argument A1, and argument A4attacks argument A2.11It is worth noticing that the identi “cation of the arguments and their relations is much more subtle and ambiguous than what emerges from this explanatory example, and may often be a matter of interpretation that current state-of-the-art argument mining systems cannot tackle yet. For instance, argu- ment A1can be considered as a subargument of argu- ment A2as ÒThe same applies ÉÓ refers to A1, and argument A3can be interpreted as a kind of persua- sive statement meant to strengthen argument A4.To address this kind of issue and build more capa- ble applications, it is necessary to enhance the exist- ing tools used to analyze, aggregate, synthesize, struc- ture, summarize, and reason about arguments in texts, with more sophisticated natural language pro- cessing (NLP) methods. However, and considering the complexity of the task, to do so it is still necessary to reach a deeper level of understanding of the inner workings of natural language, and evolve new meth- ods expanding the ones currently found in natural language processing.Moreover, to tackle these challenging tasks, high- quality annotated corpora are needed for use as a training set for any kind of aforementioned identi “-cation. These corpora are mainly composed by three different elements: an annotated data set that repre- sents the gold standard whose annotation has been checked and validated by expert annotators and is used to train the system for the required task (that is, arguments or relations extraction), a set of guidelines to explain in a detailed way how the data has been annotated, and “nally, the unlabelled raw corpus that can be used to test the system after the training phase. The reliability of a corpus is ensured by the cal- culation of the interannotator agreement that meas- ures the degree of agreement in performing the anno- tation task among the involved annotators. 12Currentprototypes of argument mining systems require to be trained against the data the task is addressed to, and the construction of such annotated corpora remains among the most time-consuming activities in this pipeline.For an exhaustive state of the art review on argu- ment mining techniques and applications, we refer the reader to the paper by Lippi and Torroni (2016). Debating Technologies There is a long tradition of computer-aided debate systems with roots in e-democracy, decision support, and so on. These systems have two things in com- mon: “rst, they implement idiosyncratic and new dialogical structures, or games, with little reuse or incremental development. The second is that little or no contribution to the debate itself is offered by the machine. The system role has been one of support and facilitation only. With a variety of techniques for automatically mining argument structures from both monological and dialogical resources, an exciting new possibility is opened up for not just supporting and enhancing new human-human arguments but also developing new human-machine arguments: this is the space of debate technology, a speci “c sub- “eld of argument technology in general.Several systems have demonstrated stand-alone applications of debate technology focusing on domains of use such as pedagogy (Pinkwart and McLaren 2012), in which both responsibility for the structuring of a debate and its automatic furthering are taken on by the machine. The key bottleneck in such systems, however, is the availability of data. Articles 32AI MAGAZINE

122 KB – 12 Pages