A Review of Argumentation for the Social Semantic Web

Semantic Web 0 (0) 1 1 IOS Press

Editor(s): Harith Alani, The Open University, UK Solicited review(s): Fouad Zablith, American University of Beirut, Lebanon; Simon Buckingham Shum, The Open University, UK; Iyad Rahwan, Masdar Institute, Abu Dhabi Open review(s):

Jodi Schneider a, Tudor Groza a,b, and tention of scholars from fields such as artificial in- Alexandre Passant a telligence [137], communication theory [5], business a Digital Enterprise Research Institute, National management [88] and e-government [107]. At the University of Ireland, Galway, same time, argumentation researchers began establish- fi[email protected] ing the foundations for a World Wide Argument Web b School of ITEE, The University of Queensland, (WWAW) as “a large-scale Web of interconnected ar- Australia, [email protected] guments posted by individuals to express their opinions in a structured manner” [138]. Arguments on the Web can be used in decision- making contexts. Decision-making often requires discussion not just of agreement and disagreement, but Abstract. Argumentation represents the study of views and also the principles, reasons, and explanations driving opinions that humans express with the goal of reaching a the choices between particular options. Furthermore, conclusion through logical reasoning. Since the 1950’s, sev- arguments expressed online for one audience may be eral models have been proposed to capture the essence of informal argumentation in different settings. With the emer- of interest to other (sometimes far-flung) audiences. gence of the Web, and then the Semantic Web, this model- It can be difficult to re-find the crucial turning points ing shifted towards ontologies, while from the development of an argumentative discussion, even one in which we perspective, we witnessed an important increase in Web 2.0 have participated. Yet on the Web, we cannot subscribe human-centered collaborative deliberation tools. Through a to arguments or issues, nor are there tools that support review of more than 150 scholarly papers, this article pro- searching for arguments. Nor can we summarize the vides a comprehensive and comparative overview of ap- rationale behind a group’s decision, even when the dis- proaches to modeling argumentation for the Social Seman- cussion took place entirely in public venues such as tic Web. We start from theoretical foundational models and investigate how they have influenced Social Web tools. We mailing lists, blogs, IRC channels, and Web forums. also look into Semantic Web argumentation models. Finally By providing common languages and principles to we end with Social Web tools for argumentation, including model and query information on the Web (such as online applications combining Web 2.0 and Semantic Web RDF [90], RDFS [1], OWL [3], SPARQL [2], Linked technologies, following the path to a global World Wide Ar- Data principles [18], etc.), the Semantic Web [20] is gument Web. an appropriate means to represent arguments and argumentation uniformly on the Web, and to enable, for Keywords: Argumentation, Semantic Web, Social Web, Se- mantic Web, Ontologies instance, browsing distributed argumentation patterns that appear in various places on the Web. Indeed, researchers have shown that the Semantic Web can be 1. Introduction used for visualization and comparison in decision rationale [102]. In recent years, the problem of representing large- In this context, this paper discusses research in mod- scale argumentation on the Web has attracted the at- eling argumentation as it relates to the Social Seman-

1570-0844/0-1900/$27.50 c 0 – IOS Press and the authors. All rights reserved 2 Schneider et al. / A Review of Argumentation for the Social Semantic Web tic Web [10,70,27], focusing on foundational models mentation (Section 4) from a variety of fields, compare of argumentation, their applications in the Social Web, them (Section 5), and present applications of these the- and on ontologies (as in computer science [69]). oretical models (Section 6). Subsequently we present In particular, our purpose is to investigate ontologies (Section 7) and compare (Section 8) Semantic Web and tools which may be useful for argumentation on models of argumentation. Then we move on to review- the Social Semantic Web, a field where the aforemen- ing tools: in Section 9 we highlight thirteen notewor- tioned Semantic Web technologies support Social Web thy features of Social Web argumentation tools, based [123] applications, while at the same time Social Web on a comprehensive analysis of thirty-seven relevant paradigms are used to generate Semantic Web data col- tools (see the Appendix for full details). Finally we laboratively and at large scale. This convergence aims conclude the paper in Section 10. at providing new and improved ways to integrate and discover data, following the vision of Social Machines provided by Berners-Lee [19], both on the Web and 2. Background in the enterprise [127]. In the context of argumentation, this could help to aggregate arguments from var- 2.1. Argumentation ious websites — for instance a discussion starting on Twitter and followed up on a mailing list, later frozen Argumentation theory is the study of agreement, on a wiki once consensus is reached — thus providing disagreement, and of the dialogues and writing through new means to follow argumentative discussions on the which we convince ourselves and others of our points Web. This would enable an argument-centric view of of view [65]. Informal argumentation occurs through- the Web. out conversations, online and offline, often in conjunc- Moreover, the Social Web does not yet have widely- tion with persuasion or with joint decision-making. used argumentative ontologies, though this problem has been noted [67], along with the need for federation Even logically sound decisions may involve choices infrastructures [125]. Thus, in order to identify how based on values and preference judgements: people different argumentation models and tools can be used may agree on the facts of a situation yet disagree on for the Social Semantic Web, this paper offers a review the preferred outcome or decision to be taken. That is of more than 150 research papers on the topic, from vitally different from disagreeing on the facts of a sit- 1945 to 2011, from which we compare: uation (in which case more information is called for). We are concerned with argumentative discussions, – 14 theoretical models of argumentation which we take to be online, mainly textual messages – 14 Semantic Web models for argumentation (i.e. and discussions, in which subjective perspectives or ontologies) differences of opinion are important and relevant. – 37 tools for representing argumentation on the Groups may use online conversations and social media Web. to coordinate and support decision-making; such argumentative discussions can be found in many online dis- As the focus is on human-centered argumentation cussion fora such as standardization bodies’ listservs, [87], with the goal of improving access and providing Wikipedia editors’ wiki pages, and open source com- overviews and visualizations, this article will briefly munities’ IRC channels, bug reports, and listservs. In- mention, but not analyze, the agent-based argumenta- dividuals may also draw on online conversations and tion domain1 social media in order to form personal opinions and Following the introduction, we provide brief overviews clarify their own preferences, based on sensemaking of argumentation (Section 2.1) and of the Social Web and analysis of others’ experiences; such argumenta- (Section 2.2), then discuss requirements for support- tive discussions can be found, for instance, on product ing argumentation on the Social Semantic Web (Sec- reviews websites, political blogs, in patient advocacy tion 3). We next present theoretical models of argu- and support group discussions, and in brief anecdotes shared via microblogs. 1 See for instance the Argumentation in Multi-Agent Systems There are a variety of common argument structures (ArgMAS) Workshop series, in its ninth year in 2012. In particular, in the social context, Heras has presented argumentation work from [134]. A single premise may directly support a conclu- a social perspective using case-based reasoning and ontologies e.g. sion (as in (i) of Figure 1 on the facing page), but more the ArgCBROntology 2. [76,73]. commonly, they are combined to come to a conclusion. Schneider et al. / A Review of Argumentation for the Social Semantic Web 3

Yet the internal structure of these discussions – such as whether the participants agree or disagree, are con- tributing diverse ideas, or debating in circles – is still not represented in SIOC. Capturing such underlying arguments would be valuable, and research is begin- ning to address this for instance by identifying argument schemes used in Amazon reviews [75,208] and by modeling the speech acts in Twitter conversations [148]. Yet infrastructure for argumentation on the So- Fig. 1. Common argument patterns, from [134]. cial Semantic Web is still needed. Premises and conclusions may also be chained (as in (iv) of Figure 1 on the next page). Further, as we will see in this review, there are differ- 3. Requirements ent ways of thinking about and modeling arguments. Argumentation theorists have variously modeled indi- What are the requirements for supporting argu- vidual argument structure (e.g. Toulmin, discussed in mentation on the Social Semantic Web? Arguments Section 4.1, page 4), argument chaining (e.g. Arau- must be identified, resolved, represented and stored, caria3 [150,142,144]), and groups of arguments (e.g. queried, and presented to users. Identification involves Dung, discussed in Section 4.5, page 6). There is a sig- mining arguments, in the form of claims, from text nificant difference between what is possible to analyze (Section 6.12.2, page 17), eliciting them from users, when looking at these different levels of argumenta- or some combination of these approaches. Resolving tion: they show micro- and macro- structures which involves indicating the relationships between the in- are not commensurate, so it is important to have clarity dividual claims that make up arguments: are they on about what kind of analysis suits a situation. the same topic? Do they agree or disagree? Represent- We believe that for social media, the basic units ing and Storing arguments requires a suitable ontol- of argumentation are claims and justifications. By ogy to represent claims and the relationships between a claim, we mean an assertion of fact or opinion4. them. This supports Querying and enables Presenting Justifications–reasons for believing the claim–are of- the Social Semantic Argument Web, i.e. using these ten elicited when a claim is questioned. Some justifi- ontologies to facilitate access to conversations, sum- cations take the form of explanations; opinions may be marizing the contentious and agreed-upon points of a elaborated upon and explained even when no disagree- discussion. ment is expected. The representations chosen are key to this process, since they determine what stored information can be 2.2. Social Semantic Web retrieved, and what information needs to be mined and resolved. Existing representations will need to be aug- The interaction of users around the Web has been mented, since the information we can retrieve depends shifting from individual siloed Web systems, towards 5 on what information we store. The desired ontologies more open and interlinked social applications . In dis- should encompass not only the structural features of cussion environments, such interlinkage is particu- posts (such as the date and author of a post) and of larly important: the same community may discuss top- conversations (such as the reply structure of multi- ics across multiple sites, and use multiple types of ple posts), but also additional argumentative features, sites, such as blogs and microblogs, discussion forums, for instance to mark claims and to indicate the rela- and wikis. Crosslinking the discussions of these sys- tionships between them. The simplest relationships for tems is a first step, which has been taken by SIOC representing argumentation indicate whether pairs of – Semantically-Interlinked Online Communities [26]. claims support or challenge each other. Yet in general, these relationships do not just pertain to pairs; in 3http://araucaria.computing.dundee.ac.uk/ general, entire groups of argumentative messages may 4 A propositional commitment, in Walton and Krabbe’s terminol- need to be considered together. The meaning of a dia- ogy [198] 5http://oreilly.com/web2/archive/ logue may be lost by chunking messages and treating what-is-web-20.html them individually, out of the original context. 4 Schneider et al. / A Review of Argumentation for the Social Semantic Web

Even simple scenarios may give rise to complex Another factor is how–by what process and agent– argumentation involving chains of statements (e.g. arguments are translated into the formal semantics. [178]), and context-dependent relationships in which This may be the responsibility of a human or machine. the conclusion of one argument is premise of another If argumentation is annotated by humans–either the [209]: this makes the graph structures of the Seman- person posting a comment or other individuals–they tic Web a natural fit. Wyner et al. suggest that be- will need sufficient understanding of the model as well sides agreement and disagreement, the semantic types as a suitable incentive or motivation for annotating. of arguments should at least include introduction of Meanwhile, machine-based (algorithmic) annotation is a premise or exception, refinement, and pronomial limited by our current understanding of informal ar- anaphora and call for a modular architecture “where gumentation, and by the multilayered meaning of con- different relationships or debate components may be versations. Further, argumentation can be treated as added systematically” [209]. static (for completed discussions) or dynamic (for ongoing conversations). If participants annotate the dis- 3.1. Example Applications & Requirements cussion, the requirement to annotate can distract from the discussion; yet non-participants and machines may We envision two main approaches to studying argu- be limited by lack of awareness of the context and of mentation on the Social Semantic Web: subtle language cues. Models supporting human navigation (e.g. to sup- 1. Focusing on the real-time, dialogical nature of port our second example above) should make psycho- the Social Web, i.e. by soliciting arguments from logical sense; this is not a factor for internal represen- humans through conversation and real-time ex- tations for machines. In either case, to ensure feasibil- change. ity (for either human or algorithmic entry), argumenta- 2. Focusing on the Social Web as a source of ar- tion classifications need to be clearcut. Thus, the gran- tifacts, i.e. by using existing natural language ularity of the model must be limited. conversations and reconfiguring the traces and archives of these conversations. Examples of the first case would be a chatbot or 4. Theoretical Models of Argumentation an interactive webform; these could help populate a knowledge base or enable argumentative interaction This section discusses fourteen theoretical models. between humans and intelligent agents. Examples of Seven are designed for capturing argument structure: the second case would be discussion summaries or in- Toulmin [180], Issue-Based Information Systems [96], teractive conversation browsers; a discussion summary Walton’s argumentation schemes and critical ques- could highlight the agreement and disagreement about tions [197], Walton and Krabbe’s dialogue types [198], a topic expressed in a number of Social Web sources, Dung’s Argumentation Frameworks [51], Value-based or a review browser could enable faceted navigation Argumentation Frameworks [15], and Factor Analysis through reviews based on the factors they mention, and and Dimension Analysis [16]. An additional seven lin- the polarity and strength of the reviewer’s perspective guistic approaches deal with issues relevant to argu- on each such factor. ment structure or detection: Speech Act Theory [158], Formal semantics will be needed in both cases, Language/Action Perspective [205], Pragma-dialectic but for different functions. In the first case, formal [186], Metadiscourse and Structural Elements of Text semantics translate from natural language to agent- [79], Rhetorical Structure Theory [110], Coherence appropriate vocabularies, potentially enabling reason- [91], and Cognitive Coherence Relations [152]. ing over human input. Argumentation has long been used for planning between agents: agent-based ap- 4.1. Toulmin proaches to the Semantic Web are common [179], and there has been some work in mediating between hu- The study of informal argumentation originated mans and agents [169,202]. In the second case, the se- in philosophy in 1958 with Toulmin [180]. Toulmin mantics will mainly be useful for presenting humans sought to find a common underlying basis for argu- with visualizations and overviews, therefore the ease ments in every field of human activity. His model of mining and presenting representations, and the suit- applies, for instance, to legal, scientific, and infor- ability for human understanding, should be preferred. mal conversational arguments. In Toulmin’s theory, Schneider et al. / A Review of Argumentation for the Social Semantic Web 5 evidence and rules called Warrants support Claims. makers–each of whom need to communicate with each Claims may also be qualified (i.e. with constraints or other and who must also get information from existing to indicate uncertainty); Rebuttals may be used to ar- records and documentation. gue against an argument. Toulmin’s argument pattern IBIS, as originally designed, is a documentation sys- is shown in Figure 2: Data is supported by Warrants tem, meant to organize discussion and allow subse- which have Backings, showing that a Claim holds with quent understanding of the decision taken; this ex- Qualifiers regarding the situation, unless there is a Re- plains the use of “Information System” in its acronym. buttal . Figure 3 shows Toulmin’s now-famous argu- The context of the discussion is a discourse about a ment, presented according to this structure. topic. Issues may bring up questions of fact and be discussed in arguments. Here, “Arguments are constructed in defense of or against the different positions until the issue is settled by convincing the opponents or decided by a formal decision procedure,” [96]. IBIS also recognizes model problems, such as cost-benefit models, that deal with whole classes of problems. Several kinds of relationships exist between issues: direct successor, generalization, relevant anal- ogy, compatible, consistent, or inconsistent. The method Fig. 2. An interpretation of Toulmin’s argument pattern, from [29]. also distinguishes issue content, as factual, deonic (“Shall X become the case?”), explanatory, or instru- mental (“Shall we take approach X to accomplish Y?"). Originally implemented as a paper-based system, IBIS influenced several ontologies and numerous tools (see Section 6.2, page 12) as well as procedures such as dialogue mapping [47].

4.3. Walton’s Argumentation Schemes and Critical Questions

Fig. 3. Toulmin’s example argument from page 105 of [180]. The Canadian philosopher Walton has written extensively on argumentation for more than thirty years (e.g. [190,193,194]); a 2010 festschrift honoring his 4.2. Issue-Based Information System (IBIS) contributions [143] shows how his work has influenced and been applied to computer argumentation. Informal IBIS, Issue-Based Information System, is a problem- argumentation is one of Walton’s specialties [195], and solving structure first published in 1970 [96]. As the in this section, we discuss two of his key theories, start- name suggests, IBIS centers around controversial is- ing with argumentation schemes. sues which take the form of questions. Specialists from According to Rahwan [134], while many taxonomies different fields may use the same words with differ- of argumentation have been proposed [129,63,185,85], ent assumptions and intentions6, hampering commu- Walton’s taxonomy [191] provides the point of depar- nication. IBIS is especially intended to support com- ture for computational models of argumentation. In munity and political decision-making. In this scenario, there may be three separate groups–the participants in his detailed classification from 1995 [191], Walton de- the discussion, the relevant experts, and the decision scribes each scheme with a name, a conclusion, a set of premises, and a set of critical questions. Critical questions address the points where this argument scheme 6“Many central terms used are proper names for long stories specific of the particular situation, with their meaning depending very may break down, and suggest attacks against the argu- sensitively on the context in which they are used." [96] ment. For example, the following six critical questions 6 Schneider et al. / A Review of Argumentation for the Social Semantic Web are associated with the Argument from Expert Opinion large, middling, or minor role. Inquiry, Discovery, [62]7: and Information-seeking dialogues are almost entirely knowledge-based, while knowledge plays only a mi- 1. How credible is E as an expert source? nor role in Negotiation (aiming at a harmonious set- 2. Is E an expert in the field that A is in? tlement) and Eristic (quarrels, beneficial mainly for 3. Does E’s testimony imply A? venting emotions). Knowledge plays some role in the 4. Is E reliable? remaining two types: in Persuasion and Deliberation, 5. Is A consistent with the testimony of other ex- opinion and belief also have a large role. Further com- perts? plexity arises because dialogue types may shift in an 6. Is A supported by evidence? actual discussion, and argument schemes may be em- Walton’s 2008 book [197], coauthored with computa- bedded in one another [196]. tional argumentation researchers, presents 65 general argumentation schemes, presumably updating [191]. 4.5. Dung’s Argumentation Frameworks

4.4. Walton and Krabbe’s Dialogue Types Dung provides a powerful graphical model of argumentation frameworks in [51], which has been Discussion types, first developed by Walton and widely used in computational argumentation. Argu- Krabbe [198], have also been influential. 8 mentation frameworks are defined as sets of arguments Seven types of dialogue are shown in Figure 4 on and attacks between them. Formally, an argumentation the next page. These types are Persuasion, Inquiry, framework is a pair AF = hAR, attacksi where AR Discovery, Negotiation, Information-Seeking, Deliber- is a set of arguments, and attacks is a binary relation ation, and Eristic. They are distinguished by the ini- on AR, i.e. attacks ⊂ AR × AR. tial situation, the individual goals of the participants, and the overall goal of the dialogue. For instance, an information-seeking dialogue and an inquiry have similar goals, but differ in the initial situation: one person is believed to have the answer in an information- seeking dialogue, while in an inquiry, no one has the answer. Persuasion and deliberation are distinguished by whose preferences are used: in a persuasion dialogue, the outcome depends on the preferences of the individual to be persuaded, while in a deliberation, the group preferences are used. Understanding the goal of a conversation is important for determining the outcome, and for determining what conversational moves are relevant. We may be able to access to pragmatics [92] of a conversation by understanding its goals. This is also how we evaluate a conversation: What was A trying to achieve? What Fig. 5. Example of an argumentation framework was B trying to achieve? Did they achieve it? In our own view, these types of dialogue can Then the questions of interest are to find maximal be classified based on whether knowledge plays a sets of arguments that do not attack each other (these are called conflict-free), and to find arguments that are not defeated by a given set of arguments (these are 7[62] attributes this to page 49, D. Walton, Appeal to Expert Opin- ion, Penn State Press, University Park, 1997. called acceptable). A conflict-free set of arguments is 8Walton has revised this taxonomy several times. ‘Discovery’ was then considered to be admissible if each argument is not in several earlier formulations, such as [193](p. 183); it is moti- acceptable with respect to the set. vated by choosing the best hypothesis for testing. Debate and Peda- Dung then finds maximal admissible sets, known as gogical appeared in an earlier formulation [192] which provides de- preferred extensions grounded scriptions of the goals of each dialogue. Other scholars have also . Also important are the suggested extensions and modifications, for instance Dunne et al. extensions, which represent the least fixed point of the have proposed adding examination dialogues [52]. function mapping an argumentation framework to the Schneider et al. / A Review of Argumentation for the Social Semantic Web 7

Fig. 4. Walton’s seven types of dialogue, from [196]. set of acceptable arguments of that framework. A sta- draws from Perelman’s notion of audience [129]: ar- ble extension is a conflict-free set of arguments that at- guments are often addressed to particular audiences, tacks each argument that does not belong to the set. and persuasive arguments are those aligned with the A simple argumentation framework is shown in Fig- audience’s values and preferences. Value-based Argu- ure 5. In this example, A and B attack each other; A mentation Frameworks provide a method for logically also attacks C; and D is not attacked. Thus A,D and calculating consistent approaches, distinguishing be- B,C,D are preferred extensions. tween the facts of a situation and community mem- In Dung’s theory, there is a notion of ‘defend’ – to bers’ values. defeat the attackers – but there is no direct notion of ‘support’. Arguments ‘support’ another one by not be- 4.7. Factor and Dimension Analysis ing defeated, and by not attacking, a given argument. There is a large business market in legal information 4.6. Value-based Argumentation Frameworks retrieval, and one method for classifying and indexing legal cases has been the key factors of a case, for in- Value-based Argumentation Frameworks [15], based on Dung’s argumentation frameworks, address persua- stance for the early legal argumentation system CATO sion and preferences. It is not just differences about [7]. Factor analysis can be helpful in supporting com- the facts, or failures in logic that can cause reason- munity decision-making or in summarizing reviews. able people to disagree: differences in values can also Factors are simplifications that are either present or ab- be to blame. For instance: “Despite the fact that the sent; when present, a factor “always strengthens the weather is beautiful, I choose to stay inside, because case for the same disputant" [16]. I have something important to do". In practical rea- Further development along the lines of CATO soning, two people can come to different, consistently yielded Ashley’s later system, HYPO [11], which uses logical opinions, based on a difference of values: “A dimensions rather than factors. Dimensions “capture key element in persuasion is identifying the value con- the legal relevance of a cluster of facts to the merits flict at the root of the disagreement so that preference of a claim". Dimensions such as “Obligation to aid the between values can explicitly inform the acceptance victim" or “Failure to heed traffic signs" contribute to or rejection of the competing arguments." The the- determinations of culpability, and have been recorded ory of Value-based Argumentation Frameworks thus in manually constructed databases [12]. Dimensions 8 Schneider et al. / A Review of Argumentation for the Social Semantic Web can be present to a greater or lesser degree and it may Perspective, it uses speech acts, further developing unclear which side they favor. Searle’s theory in order to model argumentation. Rather than focusing on the logical forms and patterns 4.8. Speech Act Theory of reasoning, as Walton does, van Eemeren and Groo- tendorst’s pragma-dialectic theory views argumenta- Several approaches to conversation and argumentation as a social process, used to settle “a difference of tion have been derived from speech act theory. Searle’s opinion by verbal means” [187](pp. ix-xii). Speech Act Theory [158] describes five categories of speech acts: assertives, directives, commissives, expressives, and declaratives. Speech acts are about the force of a statement: what effect they seek to have on the hearer or the world. Assertives (‘The sky is blue’) assert that something is true. Directives (‘Clean your room’) order, permit, or request something. Commis- sives are vows or pledges (‘I swear to tell the truth’). Expressives offer thanks or congratulations, or express feelings (‘Great work!’). Declarations (‘I now pro- nounce you man and wife’) enact what they say, ef- fectively changing reality.9 Speech act theory is not a complete model of argumentation, yet it is a relevant theory that has been widely influential. For example, earlier (Section 4.4, page 6) we discussed that understanding the goals of a conversation could be important for interpreting it; the same speech acts can be used in different ways, depending on the goals of a dialogue 10.

4.9. Language/Action Perspective

Fig. 6. Distribution of speech acts in a critical distribution from The Language/Action Perspective (LAP) [205] em- [186]. beds Speech Act Theory in a task-based framework. Argumentation is found in each of the three types Depending on the context, the same speech acts can of conversations which accomplish goals in the Lan- function as an explanation, a piece of information, or guage/Action Perspective, according to de Moor and as an argumentation: for argumentation, the context Aakhus: Conversations for action involve making must include a difference of opinion. Depending on commitments; conversations for possibility create a their order and position in the discussion, speech acts context for action; and conversations for disclosure al- take on different meanings, as we will see in Figure 6. low participants to share their views and concerns [49]. Further, Searle’s illocutionary speech acts, which function at the sentence-level, combine in argumentation 4.10. Pragma-dialectic into a higher-order textual element: the argumentation is itself seen as a complex speech act. The pragma-dialectic approach is a complete argu- The main speech acts within an argumentation are mentative theory, which has been developed over a assertives, commissives, and directives. Expressives– number of years in numerous scholarly works (espe- which express emotions–do not help resolve the differ- cially [184,185,186]) and popularized in the authors’ ence of opinion, but may affect how or whether the dis- textbooks (e.g. [187]). Like the Language/Action cussion proceeds. Declaratives–which bring a state of affairs into being–are relevant for definitions, specifications, amplifications, and explanations; van Eemeren 9As with all speech acts, sincerity is a criterion, and social criteria, e.g. ceremony, may also hold. and Grootendorst call these “usage declaratives”. 10For a pertinent example see the persuasion and deliberation sce- The pragma-dialectic approach also stresses the narios discussed in Atkinson et al. [13]. principles of clarity, honesty, efficiency, and relevance, Schneider et al. / A Review of Argumentation for the Social Semantic Web 9 updating Grice’s Cooperation Principle [64]–which fo- Similarly, set phrases often indicate the topics of cuses on the intention of language–with the Searlean a debate–helping us detect a participant’s standpoint, focus on the communicative aspects of language use. which “expresses a certain positive or negative posi- Relevance, for example, can be global, local, subject tion with respect to a proposition" [186](p. 3). Stand- matter-specific, or probative. An argument may be rel- points may often be obliquely stated, yet they can evant at one phase, but irrelevant at another; for exam- sometimes be recognized by the appearance of partic- ple an argument related to selecting the topic of dis- ular phrases, which can be baldly stated (“my stand- cussion is not relevant once the topic has been agreed point is that”, “we are of the opinion that”) or explicit upon. (“I think that”, “if you ask me”, “therefore”). Some To understand the force of a speech act–whether phrases that may be used to indicate a standpoint also an assertive, commissive, or directive–, we must iden- permit alternate interpretations (“the way I see it”, “in tify where we are in the argumentation. van Eemeren other words”, “all things considered”). Other patterns, and Grootendorst identify four dialectical stages of ar- like “shouldn’t”, “you must never”, “that...is”, “ought gumentation: Confrontation, Opening, Argumentation, to be”, commonly co-occur when a standpoint is ex- and Closing [186]. In the confrontation stage, the issue pressed [187](pp. 10-12). at hand is announced, agreed upon, or clarified. In the Such phrases, like the metadiscourse markers we opening stages, the rules are agreed to (perhaps implic- discuss next, can potentially be applied in detecting itly). In the main stage (Argumentation), each party is and mining argumentation (which we later discuss in expected to make a serious effort to support his point Section 6.12.2, page 17). However, phrase-detection of view, while also allowing the other party to make alone does not suffice, and significant challenges re- his case. Finally, the argument closes when the goal is main. Language is multivalent and context is needed fulfilled or the parties agree to end the debate. for properly reconstructing argumentation. “In lan- The pragma-dialectic approach is far more exten- guage use there is often the case that there is more than sive in attending not only to linguistic patterns, but also one purpose at the same time, and if language is used to the social context in which they are embedded (i.e. argumentatively, the argumentative function need not the linguistic pragmatics, see e.g. [92]). Argumenta- always be the most important," [186](p. 23), meaning tive discourse starts with the assumption that the lis- that reconstructing argumentation must extract a single tener does not (necessarily) agree with the speaker’s thread of meaning out of many. position, and aims at “coming to a reasonable agreement” [187], p 4. Speakers may anticipate objections, 4.11. Linguistic Markers of Argumentation: explaining their reasoning to account for expected (im- Metadiscourse and Structural Elements of Text plicit) differences of opinion. Or, they may wait to hear their conversational partners’ standpoints or doubts, We next discuss linguistic markers of argumenta- and then respond. tion; strictly speaking these are not theoretical models The issue, according to the pragma-dialectic the- of argumentation, yet they identify possible argumen- ory, may be single or multiple, and mixed or non- tation, essentially enabling the argumentation struc- mixed. Single disagreement is about just one proposi- tures to be annotated in and abstracted from the text. tion while multiple disagreement is about more than Metadiscourse refers to the “aspects of a text which one proposition. If both a positive and a negative stand- explicitly organize a discourse or the writer’s stance to- point are taken on the issue, the disagreement is mixed, wards either its content or the reader.” [79](p. 14). Ar- otherwise it is simple. This is a particularly useful dis- gumentative words and phrases such as ‘but’ and ‘ac- tinction for conversations in social media. cording to X’ are prominent examples. Metadiscourse Pragma-dialectic is also particularly useful for de- is used not only to structure text but also to influence termining which parts of social media discourse can the reader’s view. be considered argumentative, since it presents phrases Hyland classifies metadiscourse into interactive and that tend to mark argumentation, and since it treats interactional types. Interactive metadiscourse, which speech acts from an argumentative perspective. For ex- organizes text, includes transitions (in addition, but, ample, doubt is often implicit, but certain phrases mark thus, and); frame markers (finally, to conclude, my it more explicitly, such as “I don’t know whether", purpose is); endorphoric markers (noted above, see “I’m not yet convinced that”, “Couldn’t it be that”, and Fig, in section 2); evidentials (according to X, Z “I’ll have to think about whether”. states); and code glosses (namely, e.g., such as, in 10 Schneider et al. / A Review of Argumentation for the Social Semantic Web other words) [79](p. 49). Interactional metadiscourse, Argumentation relies on coherence: Adding ‘be- which makes the author’s views explicit and invites cause’ works in the first example but not in the sec- readers’ response, includes hedges (might, perhaps, ond example. The word ‘because’ stresses the ex- possible, about); boosters (in fact, definitely, it is clear pected causal relationship, making the informal argu- that); attitude markers (unfortunately, I agree, surpris- ment more evident. ingly); self mentions (I, we, my, me, our); and engage- 3. “Tim must love that Belgian beer, because the crate ment markers (consider, note, you can see that). in the hall is already half empty." As markers of the persuasive and rhetorical elements of texts, metadiscourse elements are likely to be In Sentence 3, the reader must still infer some informa- useful signals for identifying claims and arguments in tion, such as that the crate in the hall contains Belgian social media. beer, and that Tim is the main person drinking the con- tents of the crate. Such missing premises are typical in 4.12. Rhetorical Structure Theory informal argumentation. 4.14. Cognitive Coherence Relations Rhetorical Structure Theory (RST) [110], a method for analyzing texts according to their structure and One actionable way of expressing coherence is by rhetorical role, was developed at the University of using specific signaling terms. The causal relationship Southern California’s Information Sciences Institute to (expressed in ‘because’) is one of the Cognitive Co- assist with computer-based text generation. In RST, herence Relations which Sanders uses to explain how structures such as ‘Concession’, ‘Evidence’, and ‘Jus- readers understand text [152]. The four Cognitive Co- tify’, called ‘relations’, describe the relationship of two herence Relations are: Basic Operation (causal or ad- or more spans of text. Generally one span (the most ditive), Source of Coherence (semantic or pragmatic), important) is called the nucleus, while the less impor- Polarity (positive or negative), Order of Segments (for tant spans are known as satellites. In some situations causal relations only: basic or non-basic, depending on (such as sequences and contrasts), both spans are nu- whether or not the antecedent appears before the con- clei of equal weight. Justifications and hedges are more sequent). Based on this, the relationship signaled in be- likely to appear in satellites while the nucleus is more cause in Sentence 3 (above) is causal, pragmatic, pos- likely to contain claims; this has potential application itive, and basic. in detecting arguments and in summarizing social web applications. 5. Comparison of Theoretical Models 4.13. Coherence Any of these models could be expressed in se- Coherence is another important concept in text and mantic formats (e.g. RDF) since they are compatible dialogue. Coherence is not an argumentation theory with a graph-based representation of argumentation. per se, but it is both an essential part of text, and an Yet for modeling argumentation on the Social Seman- essential part of argumentation, since before an argu- tic Web, it is most meaningful to examine the chal- ment can be understood (much less formally evalu- lenges and opportunities that might advantage any one ated), how its parts hold together or interrelate must be model or framework over the others. As previously understood. Knott compares the following two exam- noted, various units have been modelled–individual ples [91]. argument structures (e.g. Toulmin, discussed in Sec- tion 4.1, page 4), argument chaining (e.g. Araucaria11 1. “Tim must love that Belgian beer. The crate [150,142,144]), and groups of arguments (e.g. Dung, in the hall is already half empty." discussed in Section 4.5, page 6). 2. “Tim must love that Belgian beer. He’s six We can make several further distinctions between foot tall." models, for instance based on the community in which While the first example is coherent the second exam- they originated, their purpose or use, the extent to ple is more challenging to make sense of: the reader which they focus on disagreement, the unit of analysis expects (but does not get) a sensible explanation or evidence for why Tim must love that Belgian beer. 11http://araucaria.computing.dundee.ac.uk/ Schneider et al. / A Review of Argumentation for the Social Semantic Web 11 on which they focus, their granularity, and their suit- tion and shared visions; disagreement is analyzed or ability for automation or for aiding human reasoning. understood only to the extent necessary for coming to consensus or providing an overview of viewpoints. 5.1. Community of Origin 5.4. Unit of Analysis Various communities have contributed models, particularly the argumentation and linguistics communi- Different units of analysis have been used. At the ties. The IBIS model comes from management and language layer, models may focus on the relation- was later taken up by design rationale and human- ship between different clauses (Metadiscourse, Coher- computer interaction (HCI) communities. The Lan- ence, Cognitive Coherence, RST) or the communica- guage/Action Perspective originated in artificial intel- tive function of different words, phrases, and sentences ligence and HCI and was later adopted by communi- (Speech Act Theory and the Language/Action Per- cation theorists. In some cases, models bridge commu- spective). The pragma-dialectic approach focuses on nities: the pragma-dialectic approach is an argumenta- the propositional level, while factors analysis looks tion model which has been heavily influenced by lin- at important attributes or dimensions. Other models guistics, and by the theory of pragmatics [92] in par- focus on classifying individual arguments and their ticular. Models have been shaped by their originators relation to a whole (IBIS), or studying the internal and proponents, and the purposes for which they were structure of arguments (Toulmin, Walton, Argumenta- intended. tion Frameworks, Value-based Argumentation Frame- works). 5.2. Purpose or Use 5.5. Granularity The intended purpose for each model depends largely on its origin. Models put forth by the argu- Models of linguistic features are more granular, mentation community are generally designed to sup- but sometimes less focused on the overall structure. port either analysis (e.g. to determine the reasoning Coarse-grained and simple models, such as IBIS, more patterns used and to identify fallacies) or formal rea- common in application. Yet even IBIS is not gener- soning, in order to address questions such as compu- ally applied in its full complexity, but is rather reduced tational decisions of which argument won, what the to focusing identifying issues, and then on identifying deciding factors were, or what values and preferences pros and cons for a particular issue. Fuller versions be- were expressed in the discussion. Models of linguistic come more complex by looking at the relationships be- features may be used in discourse analysis, for summa- tween arguments: what responds to what. rization, and to support natural language generation by both machines and non-native speakers. Models from 5.6. Ease of Automated Application other communities are generally intended to support flow-based process analysis, for instance to organize Mechanistic application is possible for some mod- information in order to avoid information overload, to els but not for others. In particular, classification for speed human decision-making, and to provide a record Walton’s model would be quite difficult due to the of collaborative thought processes. large number (65) of argument categories and the need for detailed reasoning. On the hand, in many cases 5.3. Agreement/Disagreement Focus language technologies can be mechanically analyzed. Identification and classification of argumentation via Disagreement and the process of coming to con- language technologies is still in its infancy, yet offers sensus are the core of argumentation. While disagree- great potential to expand algorithmic understanding of ment and agreement are central in models coming from language. the argumentation community, other models focus on this core to a greater or lesser extent. Most linguistic 5.7. Support for Human Reasoning models are considerably broader and less focused on the argumentation aspects, yet in addressing conversa- To aid human reasoning, however, linguistic models tion, they provide valuable insights as well as analy- that mainly use cue words are probably too granular sis tools. HCI models focus on supporting collabora- since they occur at the sentence level, probably more 12 Schneider et al. / A Review of Argumentation for the Social Semantic Web granular than needed. For this purpose, Walton’s critical questions are very useful, because they can point humans to the questions that need to be addressed, opening the door to checklists for reasoning, which could be applied consistently by groups. Value-based frameworks also address the basic reasoning underlying social decisions: each person has their own reasons, which may be revealed in the course of a discussion. Focusing on what the values are, and being able to articulate them, can help both in developing em- pathy for dissenting viewpoints, and in making clear the rationale for group decisions when consensus is needed. Fig. 7. The core Toulmin argument model ontology [130].

6. Applications of the Models to the Social and 6.2. Applications of Issue-Based Information System (IBIS) Semantic Web Although many tools are described as ‘using the The theoretical models discussed have been quite in- IBIS model’ or ‘IBIS-like’, there is significant varia- fluential, and in many cases we can directly trace So- tion in the underlying structure of these models [82]. cial and Semantic Web applications of these models. In In our view, these models use ‘IBIS-like’ to mean that this section we describing some applications, review- they concern decision-making or design rationale, pro- ing each of the models in turn. vide graphical representations, and use some form of polarity. 6.1. Applications of Toulmin The IBIS model has a long history of use, particularly with early hypertext systems. Early critiques Toulmin is cited frequently and in numerous fields, of IBIS came from the design rationale community. from rhetoric (e.g. [199]) to education (e.g. [37]) to One difficulty was that only deliberated issues were computer argumentation (e.g. [103]). While his model included; Procedural Hierarchy of Issues (PHI) mod- is a useful abstraction, scholars have argued about ifies IBIS to allow inclusion of subissues which are whether people actually think in terms of Toulmin’s not deliberated [56]. PHI was adopted by another early warrants [121]. system, the Author’s Argumentation Assistant [157], One early hypertext system, SEPIA, drew from the which also drew from the authors’ earlier Toulmin- Toulmin system [171]. A modified version of Toul- based system, SEPIA [171]. min’s argumentation schemes have been used to de- Another difficulty, representing the relationships scribe a cooperative dialogue game, implemented in and interdependencies of issues [56], remains chal- Prolog, in which the participants’ goal is to reach a lenging to resolve, though ideas such as nonfunctional claim on which they agree, while also producing a requirements and dependencies (Section 7.6, page 20), supporting argumentation structure for the claim [14]. might be relevant. Such dialogue games could be modified for use in IBIS has also been used outside of design ratio- mixed-initiative systems. nale. For instance, Gerosa et al. [59] discuss an e- In the Semantic Web, the Toulmin Argument Model learning message board system adopting a modification of IBIS, where message types are specified. In has been implemented by an OWL 2 DL ontology addition to the IBIS-analogues, Question, Argumen- that imports CiTO12 [130]. It follows Toulmin’s model tation, and Counter-Argumentation, the system adds closely, as shown in Figure 7. two types: Seminar (a general topic for the week) and Clarification. IBIS has also influenced the design 12http://www.essepuntato.it/2011/02/ of modern ontologies, including the SALT Rhetorical argumentmodel Ontology, SIOC-Argumentation, DILIGENT, and the Schneider et al. / A Review of Argumentation for the Social Semantic Web 13

Change Ontology which we discuss after reviewing Semantically-Interlinked Online Communities [26] — IBIS’ RDF representation. a model that focuses on representing online communities and the content shared within them. 6.2.1. IBIS RDF While SIOC simply focuses on the notion of replies IBIS RDF13 is an RDF representation of the IBIS (sioc:reply_to) to represent connections be- model. refersTo is modelled as a subProperty of tween discussion items, the SIOC-Argumentation mod- dcterms:reference with two subproperties, pro ule goes further and provides finer-grained representa- and con. The larger IBIS vocabulary provides Pub- tion of discussions and argumentations in online com- lished Subject Indicators14 for important terms, in- munities. pro con Idea Question Argument cluding , , , , , So far a modification of SIOC that draws from DILI- Decision Reference , and . GENT and OMDoc has been used in the math wiki 6.2.2. SALT Rhetorical Ontology system SWiM16 [99], a Semantic Wiki for Mathemati- SALT [68] is a rhetorical ontology for scholarly cal Knowledge Management. The SIOC/OMDoc argu- communication. In SALT, opposing arguments can be mentation ontology (Figure 9 on the following page) connected together with the relation is further described in Lange’s dissertation [97]. It in- hasCounterArgument, while a RhetoricalElementcorporates IBIS-style classes from SIOC (Position, can also be connected with what it argues for (Argument Decision, Idea, and Issue), as well as domain- and hasArgument, for instance). SALT’s argu- specific argumentation classes for math (e.g. Wrong, mentation also includes Reason, which contains Keep_as_Bad_Example, Incomprehensible). Argument (further specified to PositiveArgument As opposed to IBIS-RDF, SIOC-Argumentation or NegativeArgument) and CounterArgument. (Figure 10 on page 15) provide the means to easily integrate argumentation modelling patterns with So- 6.2.3. DILIGENT ontology cial Web applications since it relies on SIOC, already DILIGENT provides an argumentative structure used in various applications (Drupal7, etc.). However, for collaborative ontology construction; the acronym SIOC-Argumentation has limitations: it does not rep- comes from the phrase DIstributed, Loosely-controlled resent taxonomic, causal, or similarity relations, which and evolvInG Engineering processes of oNTologies. prevents its use in contexts that require deeper analysis DILIGENT draws from both RST (Section 4.12, page of arguments. 10) and IBIS (Section 4.2, page 5), as shown in Fig- ure 8 on the next page. 6.2.6. SWAN-SIOC SWAN-SIOC [128] harmonizes the argumentation 6.2.4. Change Ontology (ChAO) in Collaborative aspects of two pre-existing ontologies: along with Protégé SIOC, it is based on SWAN — Semantic Web Appli- DILIGENT [36] itself influenced the Change On- cations in Neuromedicine [44] — an ontology which tology in Collaborative Protégé. Castro et al. distin- focuses on scientific communication in neurology. guish between an argument (which is well-focused and SWAN/SIOC uses twelve terms, as shown in Fig- specific) and an elaboration (which provides support ure 11 on page 15. The most general term is relatedTo, for the argument, possibly with file attachments) [36]. which has five direct descendents or subterms. These, Positions become clear through the dispute-resolution in turn, may have subterms, until we reach the base process. With Protégé, various argumentation-related terms in the ontology: disagreesWith, agreesWith, Annotations can be added, including Explanation, and discusses. SWAN/SIOC provides a simple Proposal, and AgreeDisagreeVote [156]. model for the relationships between items. Tools using 6.2.5. SIOC-Argumentation SWAN-SIOC include PDOnline, which is discussed in Section A.31, page xiv. The SIOC-Argumentation15 model [98] expands on the IBIS model with terms such as Decision and 6.3. Applications of Walton Argument. It is provided as an extension of SIOC — Walton’s model has been widely applied in compu- 13http://purl.org/ibis tational argumentation [143], and recent research has 14http://www.topicmaps.org/xtm/index.html# def-published-subject-indicator 15http://rdfs.org/sioc/argument 16http://kwarc.info/projects/swim/demo.html 14 Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 8. An overview of the core DILIGENT ontology from [174].

Fig. 9. The argumentation ontology from SWiM extends SIOC-Argumentation and DILIGENT Argumentation [99]. demonstrated how argument schemes could be used – to achieve new circumstances S, to aid sensemaking in Amazon reviews [75,74,208]. – which will realize some goal G, Avicenna and ArgDF incorporate Walton’s schemes. – which will promote some value V. The only Social Web application we are aware of is The following sixteen critical questions are associ- Parmenides17 [32,33,34], which uses the following ar- ated with Argumentation Scheme AS1: gumentation scheme and value-based argumentation frameworks [15]: CQ1 Are the believed circumstances true? Argumentation Scheme AS1: CQ2 Assuming the circumstances, does the action have the stated consequences? – In the current circumstances R, CQ3 Assuming the circumstances and that the action – we should perform action A, has the stated consequences, will the action bring about the desired goal? 17http://cgi.csc.liv.ac.uk/~parmenides/ CQ4 Does the goal realize the value stated? Schneider et al. / A Review of Argumentation for the Social Semantic Web 15

Fig. 10. An overview of the SIOC-Argumentation module from [98].

Fig. 11. SWAN-SIOC ontology from [42].

CQ5 Are there alternative ways of realizing the same sively studied [132], and the wider area of dialogue consequences? games has been an active area of agent argumentation CQ6 Are there alternative ways of realizing the same [113]. In applications to human argumentation, some goal? work on deliberation (e.g. [189] and the Parmenides CQ7 Are there alternative ways of promoting the system Section A.30, page xiii as a whole) has used same value? dialogue type to focus the discussion. CQ8 Does doing the action have a side effect which demotes the value? 6.5. Applications of Dung’s Argumentation CQ9 Does doing the action have a side effect which Frameworks demotes some other value? CQ10 Does doing the action promote some other value? Dung’s argumentation frameworks have been in- CQ11 Does doing the action preclude some other ac- credibly influential: as of 2011, the original 1995 paper tion which would promote some other value? has over 450 citations in the ACM Digital Library, and CQ12 Are the circumstances as described possible? over 1500 in Google Scholar. A Java reasoner called CQ13 Is the action possible? Dungine implements Dung’s acceptability semantics CQ14 Are the consequences as described possible? [170]. Dungine is part of ArgKit18, a Java 5 develop- CQ15 Can the desired goal be realized? ment toolkit for applications that use argumentation; it CQ16 Is the value indeed a legitimate value? is open source, licened under LGPL. Dung’s approach has driven computational research 6.4. Applications of Walton & Krabbe’s Dialogue in argumentation and provided the basis for a large Types body of theoretical work. Among theoretical exten- Walton & Krabbe’s Dialogue Types’s have been influential. Persuasion, in particular, has been exten- 18http://argkit.org/ 16 Schneider et al. / A Review of Argumentation for the Social Semantic Web sions of Dung’s work, we have focused on Value-based model online conversations to classify them and create Argumentation Frameworks, and the relevant Social visual maps, used for information retrieval: Web application of Dung is in fact an application of “Using current search engines, the searcher could that extension, as we next explain. search for the words Vietnam, war, and critique. 6.6. Applications of Value-based Argumentation However, many critiques of the war might not con- Frameworks tain the word critique, and would thus be lost (or receive a low ranking) in such a search. If the Parmenides uses value-based argumentation frame- searcher was able to issue a query such as Viet- works in addition to the argument scheme and criti- nam war (critique) where critique is the purpose cal questions discussed above. It pinpoints the source of at least one participant in the conversation, of the disagreement, by having participants respond to she would likely get better results. The search for a series of questions in a survey format. The group’s the semantic meaning of the words Vietnam war preferences are revealed in the results, which are dis- using conventional searching techniques would played to administrators as graphical argumentation then be combined with the search for the pragmatic frameworks. force of the word critique, yielding a search result with higher precision than searching on semantic 6.7. Applications of Factor Analysis meaning alone.” [181](bold, underline added). Attending to Speech Acts can also help predict decep- Factor analysis has been applied in Ashley’s legal tion, which uses ‘fewer assertions and more expres- argumentation systems, but we know of no specific So- sives’ [181]. cial Web applications. The factors analysis approach is still used by com- 6.10. Applications of Pragma-dialectics mercial providers of legal information [207]. More recently, automatic text mining has been used to iden- We are not aware of any argumentation tools spe- tify these factors [207]. Generalizing factors, perhaps cific to pragma-dialectics. de Moor, however, has taken using argument schemes and critical questions, could a reconstructive approach to argumentation based on provide another approach to argument mining; see pragma-dialectics (e.g. [49]). for instance Heras’ manual application of argument Pragma-dialectics has also been very influential schemes to Amazon reviews [75] and Schneider’s fac- in the argumentation community, integrated into e.g. tors analysis of Wikipedia deletion discussions [155]. Walton’s textbook descriptions of argumentation [195] 6.8. Applications of Speech Act Theory and discussed in at least one journal special issue [25] and edited collection [183]. Speech acts are used to model the flow of online conversation in several recent works. Jeong et al. 6.11. Applications of Metadiscourse and Structural [83] use semi-supervised machine learning to iden- Elements of Text tify speech acts in email and forum posts. Ritter et al. [148] model Twitter conversations with Speech Act Annotating argumentation in natural language often Theory in combination with topic modelling and show takes advantage of detecting metadiscourse and docu- a Speech Act transition map with probabilities for each ment structure. Two particularly promising approaches state. come from Teufel and Sándor. Teufel’s rhetorical com- One central use is in provenance in the Semantic ponent extraction uses machine learning to extract and Web. For assertions modelled in RDF, Carroll et al. classify text according to its rhetorical status [175]. [30] use the idea of performative warrants, to describe Sándor’s concept-matching framework detects con- assertions made legitimately by the authority signing a trasting ideas linguistically, using metadiscourse and Named Graph. rhetorical markers [6]. Rather than determining the relations between text spans, Sándor uses her concept- 6.9. Applications of the Language/Action Perspective matching framework to infer contrasts, novel information, etc. from the author’s metadiscourse [6]. Using the Language/Action Perspective and draw- Teufel and Moens focus on the document-level con- ing from Speech Act Theory, Twitchell et al. [181] text, rather than the relationship between text spans. In Schneider et al. / A Review of Argumentation for the Social Semantic Web 17 their argument zoning, machine learning is used to ex- pus19, based on Context Free Grammars [207] as well tract and classify text from academic articles accord- as techniques from Teufel and Moens. Earlier Grover ing to its rhetorical status [175]. Sándor and Teufel and et al. [66] adapted Teufel and Moens’ approach to Moens provide contributions in risk assessment, an- determine the argumentative role of sentences drawn notation, and audience- and task-specific summariza- from a corpus of legal judgements. tion. Reuse of their work has included an application In “Automatic Argumentation Detection and its to find rhetorical features of related work sections, first Role in Law and the Semantic Web" [117], Mochales- using classification algorithms, and then applying on- Paulau and Moens suggest that argument mining could tologies [9]. However, these techniques are of particu- contribute to the World Wide Argument Web [138], lar interest because of further work in argument mining by extracting argument structures without human an- drawing on these ideas, which we soon discuss (Sec- notation. As they point out, automatic argument detection 6.12.2, page 17). tion is needed at multiple levels: the inner structure of each argument as well as the overall structure of how multiple arguments are combined to contribute to the 6.12. Applications of Rhetorical Structure Theory argumentative discourse.

RST has been widely used for a variety of pur- 6.13. Applications of Coherence poses and in 2006 a paper summarizing its applications [173] was published. Recently, Mentis et al. [114] Related notions of coherence are used in Thagard’s used RST to analyze group decision rationale, compar- explanatory coherence theory, a computational theory ing new and established groups using relations such “an explanatory hypothesis is accepted if it coheres as ‘Interpretation & Evaluation’, ‘Evidence’, ‘Elabo- better overall than its competitors” [176]. This theory ration’, ‘Concession’, and ‘Antithesis’. Summarization has been applied to analyze scientific reasoning and le- research has frequently drawn upon RST [111,112]. gal trials. Some further applications of RST and related ap- Another theory it influenced–that of cognitive co- proaches are discussed in a recent survey of work herence relations, which we discuss next–has been in- on discourse structure [200]; argumentation might be fluential. found in several genres they discuss, such as in the es- 6.14. Applications of Cognitive Coherence Relations say analysis and scoring and opinion mining applications. Cognitive Coherence Relations contributed to the 6.12.1. DILIGENT development of ScholOnto [109]. In separate work, DILIGENT, briefly discussed above as an applica- Mancini’s cinematic hypertext [108] used Cognitive tion of IBIS (Section 4.2, page 5), also draws from Coherence Relations to develop a visual language for RST (Section 4.12, page 10) as shown in Figure 8 on structuring hypertext links, increasing the coherence page 14. To improve the agreement, clarity, and sat- of argumentation conveyed in non-linear hypertext by isfaction [174] of discussions for ontology creation clearly expressing the rhetorical relationships between and refinement, DILIGENT restricts the argumenta- chunks of text. Meanwhile, agent-based argumentation has used Cognitive Coherence Relations as a theory of tion. Five argumentative relations — alternative, pragmatics [126]. evaluation and justification, counterExample, The ScholOnto [166,17] project, which ran at Open elaboration, and example — were drawn from University’s Knowledge Media Institute from 2001- RST [99], based on the arguments that advanced the 2004, focused on modeling claims and arguments ontology creation process in an experiment [131]. in scholarly communication. ClaiMapper, ClaiMaker, 6.12.2. Argumentation Mining and ClaimSpotter were among the tools21 developed Drawing on rhetorical parsing, argument mining is a new area of study which seeks to detect and ex- 19AraucariaDB20, the Arucaria corpus of arguments, draws in part tract arguments from texts algorithmically. Mochales- from discussion fora (BBC Talking Point, Christian Apologetics & Paulau’s dissertation work [116] focused on mining ar- Research Ministry Discussion Boards, MSNBC discussion forum, NPR discussion boards, and Global Greens) [86]. guments [117,207,115] from the European Court of 21http://projects.kmi.open.ac.uk/scholonto/ Human Rights and from the Araucaria annotated cor- software.html 18 Schneider et al. / A Review of Argumentation for the Social Semantic Web in the project, which was seen as part of sense- ment and scheme nodes (S-nodes) holding the relation- making research. An open source web publishing ships between arguments. Scheme nodes are further di- tool called the Digital Document Discourse Environ- vided into three main types, for representing logical in- 22 3 ment , or D E [164] was also developed in related ference (RA nodes), preferences or values (PA nodes), research. ScholOnto made an RDF Schema avail- and conflicts between I-nodes (CA nodes). More re- able, but database queries with SQL were preferred to cent work on AIF envisions classification of argumen- querying based on this RDF Schema (SPARQL was tation schemes, enabling automated reasoning over the first released as a working draft in 2004). The under- schemes [136]. In Spring 2012, new specifications for lying ScholOnto ontology for these projects is shown AIF were released in OWL and RDF; an SQL database in Figure 6.14 on the facing page. This ontology now definition is also in progress23. underlies one of the applications we later discuss: Co- here is argument mapping software for sensemaking AIF forms the foundation for the World Wide Ar- that integrates annotation and argumentation for the gument Web (WWAW). The WWAW is “a large-scale general public [162]. Web of interconnected arguments posted by individuals to express their opinions in a structured manner” [138], where RDFS and OWL were suggested to be 7. Ontologies incorporating argumentation used for AIF. The foundations of the World Wide Ar- gument Web have been further discussed by Rahwan In addition to the seven ontologies discussed above, and others (e.g. [139,134,135]). which implement or follow from particular theories– AIF has continued to develop, and several published ChAO, DILIGENT, IBIS RDF, SALT, ScholOnto, extensions of AIF exist. Rahwan adds form nodes (F- SIOC-Argumentation, and SWAN/SIOC–we now re- nodes) [138] in order to more fully represent generic view seven ontologies relevant to argumentation. These argument schemes (as opposed to the instantiations include both dedicated argumentation ontologies (the of those schemes). Then Walton’s argument schemes Argument Interchange Format) and ontologies de- can be represented, using ConflictSchemes to capture signed with substantial input from the argumentation 24 community (the Legal Knowledge Interchange For- exceptions/Critical Questions. With AIF-RDF , Rah- mat) as well as ontologies that incorporate small num- wan et al. [138] add RDFS extensions to an AIF im- bers of argumentative elements (the Annotation On- plementation. In this implementation, edges are explic- tology, bio-zen-plus, the Citation Typing Ontology, itly typed. Letia and Groza add a Context Node, used the Non-functional requirements and Design Rationale to evaluate the same argument in different contexts Ontology, and the Semantic Annotation Vocabulary). [100]. Rahwan et al. [136] present a new formalization of AIF in OWL-DL, implemented in Avicenna (Sec- 7.1. Argument Interchange Format (AIF) tion A.9, page iii). Work on this area continues, largely published in the Computational Models of Argument The Argument Interchange Format [120] is a pow- (COMMA) conference series25, with a large body of erful, dedicated ontology for argumentation, originally promising research in the 2012 edition of this biannual designed to ensure interoperability of argumentation conference. software such as ArgDF, ArgKit, Carneades, and On- While AIF was intended to model monological ar- line Visualisation of Argument. AIF would be chal- guments, dialogue has been another area of inter- lenging to apply to the Social Web because it requires est in AIF extensions, with work from Modgil and argumentation schemes to be specified. In fact, even arguments themselves are not necessarily clearly spec- McGinnis [118] and Reed et al. (e.g. [146]). Ear- ified in the informal argumentation found in the So- lier work began the process of extending monological cial Web! Thus, for example, enthymemes make for- AIF for use in representing dialogical argumentation mal specification of arguments challenging [22]. [141,145,147,140]. The original core ontology, shown in Figure 7.1 on page 20 consists of two disjoint sets of nodes: information nodes (I-nodes) holding the content of the argu- 23http://www.arg.dundee.ac.uk/?page_id=197 24http://argdf.org/source/ArgDF_Ontology. rdfs 22http://d3e.sourceforge.net/ 25http://www.comma-conf.org/ Schneider et al. / A Review of Argumentation for the Social Semantic Web 19

Fig. 12. Class structure of the Scholarly Discourse Ontology from [166] .

7.2. Annotation Ontology properties, supported-by and in-conflict-with, augmenting the argumentation-related Argumentation enters into the Annotation Ontol- correlation-concepts, ogy’s26 [43] curation use case. In that use case, a hu- such as Positive correlation (unsigned), man curator reviews an annotation created by a text Positive correlation (signed), mining service, and first rejects it. This curator sub- Negative correlation (unsigned), and sequently changes her mind after a discussion with a Negative correlation (signed), which are second curator, and finally accepts the annotation af- found within the bio-zen ontology30. ter all. The statuses Rejected , Discusses, and Accepted express an argumentative workflow in this 7.4. Citation Typing Ontology (CiTO) situation. The Annotation Ontology has been used by existing CiTO31 [160,161] is an ontology for citation net- tools such as the SWAN Annotation Framework and works in scholarly publications. Its argumentative Utopia PDF reader [177]. It is currently being incorpo- terms include corrects, confirms, rated into a new W3C standard for Open Annotation gives support to, is agreed with by, 27 . is ridiculed by, qualifies, and refutes. Papers can thus be semantically enhanced.32 For ex- 7.3. bio-zen-plus ontology framework ample, an author could indicate in a paper that it updates a previous publication, and critiques a The bio-zen-plus ontology28 [151] is an ontology piece of related work, while using evidence from an- for biology; as the name suggests, it is an extension of other paper (citesAsEvidence). Readers can as- the bio-zen ontology29. It includes two argumentative semble bibliographies using CiTO properties, for instance with the bibliographic management software 26http://code.google.com/p/ annotation-ontology/ 27http://www.openannotation.org/spec/core/ 30http://neuroscientific.net/bio-zen.owl 28http://neuroscientific.net/bio-zen-plus. 31http://purl.org/spar/cito owl 32http://imageweb.zoo.ox.ac.uk/pub/2008/ 29http://neuroscientific.net/index.php?id=43 plospaper/latest/ 20 Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 13. Concepts and relations of the Argument Interchange Format

CiteULike33, showing the possibilities of semantic an- 7.6. The NDR Ontology notation. The Non-functional requirements and Design Ra- 7.5. Legal Knowledge Interchange Format tionale (NDR) Ontology [102] addresses the visual-

34 ization of non-functional requirements as Softgoal In- Legal Knowledge Interchange Format (LKIF) , terdependency Graphs. While some classes (such as developed as part of the ESTRELLA project, is an Softgoal) are specific to this domain, the NDR OWL ontology [3] for the legal domain. Its Rules Ontology introduces useful argumentative labels and & Argumentation Module deals with Exceptions, causal relationships. For example, the label prop- Rules, Arguments, and Assumptions [133]. It erty can be used to indicate the extent to which also imports the LKIF Expression Module, which pro- goals are met (i.e. whether they are adequately ‘satis- vides “a vocabulary for describing, propositions and Denied Weakly denied Undecided propositional attitudes (belief, intention), qualifica- ficed’): , , , tions, statements and media" [133]. It includes terms Weakly satisficed, Satisficed, or Conflict. for various PropositionalAttitudes, as well NDR also has classes for Argumentation, Claim, as Intention and Lie, for instance. Contribution, and Interdependency (including a subclass, Correlation Contribution 33http://www.citeulike.org/ ). The of child goals 34http://www.estrellaproject.org/lkif-core/ to the parent goal can be labelled as Break, Hurt, lkif-rules.owl# Unknown, Help, Make, etc. Schneider et al. / A Review of Argumentation for the Social Semantic Web 21

7.7. Semantic Annotation Vocabulary 9. Features of Social Web Tools

The Semantic Annotation Vocabulary [60] was de- We conducted a thorough review of argumentation veloped for the Trellis system (Section A.35, page tools for the Social Web, attending especially to exist- xvi). They used various dimensions: pertinence, reli- ing Social Web sites, tools using the Semantic Web, and prototypes from the research community in Social ability, credibility, causality (e.g. contribute to, indi- Web and Semantic Web. In this section we describe re- cate), and temporal ordering, as well as structural rela- lated reviews of argumentation tools, the scope of our tionships (such as part/whole, example-of, describes). review, and some highlights concerning thirteen particular features of these systems–visualization, ease of use, collaboration, user engagement, balancing contri- 8. Comparison of Semantic Web Models butions, deliberative polling, distributive and federated systems, annotation, incremental formalization, populating knowledge bases from user input, mixed ini- In Figure 14 on the following page we present a tiative, search, reasoning and querying. These provide comparison of the Semantic Web models discussed. ideas of the aspects that may be need to be consid- Topics addressed include whether each ontology is ered for argumentation systems on the Social Seman- centered on relations or concepts as well as whether tic Web. For further detail about each system we re- it is IBIS-like (i.e. does it contain concepts function- viewed, consult the Appendix, Section A, page i. ally equivalent to IBIS’ ‘Statement’, ‘Issue’, ‘Posi- tion’, and ‘Argument’?). We also cover what types 9.1. Related Reviews of relations it contains, drawing from ScholOnto’s types: causal, similarity, generic, supporting, challeng- Argumentation tools have been reviewed and overviewed ing, taxonomic (e.g. hierarchical categorization), and in various publications, including two contemporary problem related. Further, we describe whether polarity books. Visualizing Argumentation [87] presents eight (e.g. positive vs. negative) and weights are explicit or chapters which cover the history and cognitive foun- implicit and whether the ontology speciﬁes other on- dations of argumentation tools; describe tools for col- tologies to use for content provenance and authorship laborative learning and deliberation; provide insight provenance (such as from FOAF, SIOC, or PAV–the into map-based facilitation of in-person meetings; and Provenance & Authoring and Versioning ontology35) describe mapping scholarly debates. Of particular in- and domain knowledge (such as from DOLCE, SKOS, terest in this exceptional volume is the chapter on “The Roots of Computer Supported Argument Visu- or the PRotein Ontology). We have used a ‘?’ to indi- alization" [165]. Knowledge Cartography: Software cate that we were not able to ﬁnd this information in Tools and Mapping Techniques [122] provides seven- publications, or when information was ambiguous, to teen case studies of using mapping and argumentation reconcile it. tools, primarily in education, but also in science, poli- Some models provide a shallow view of arguments tics, and organizational knowledge transfer. yet are situated within a larger (perhaps social) con- Argumentation tools have also gained attention in text. Yet other models, originating in the argumenta- e-government (e.g. [107], or for a tools review see tion community, focus on representing the arguments ‘Opinion gathering in e-democracy’, Chapter 3 of themselves, often including the internal structure of Cartwright’s dissertation, [35]) and education ([153]). the arguments. The argumentation community’s inter- Crossover interest in politics from the IEEE commu- est in the Semantic Web has been motivated in part by nity is evidenced by a ‘Trends & Controversies’ sec- the idea of The World Wide Argument Web (WWAW) tion “AI, E-government, and Politics 2.0" [39]. [138], while the semantic web community’s interest Scheuer et al. [153] review 45 argumentation sys- has centered on communication structures, rather than tems36 used in Computer Supported Collaborative the details of argumentation or rhetoric. Learning and discuss 13 empirical studies involving

36Six of these systems are also discussed in our review below: 35http://swan.mindinformatics.org/spec/1.2/ Argunet, ConvinceMe, CoPe_it!, Debategraph, Debatepedia, and pav.html SEAS. 22 Schneider et al. / A Review of Argumentation for the Social Semantic Web

(a)

(b)

Fig. 14. A comparison of Semantic Web models for argumentation in terms of (a) structure; and (b) features. Schneider et al. / A Review of Argumentation for the Social Semantic Web 23 the use of argumentation systems in education. Inter- Tools failing the ‘Web-based’ criterion included esting results from their work are that arguments are email tools such as WIT and Zest, SAIC’s SIAM and constructed in learning applications in five main ways: Causeway, and the argumentation tools Carneades, free-form arguments, argumentation based on back- Araucaria, and Convince Me39. WIT40 [19] and Zest ground materials, arguments rephrased (e.g. reworded [210] focused on argumentation in email. SIAM41 and and rekeyed) from a transcript, arguments extracted Causeway42 are Windows-based software for influence (e.g. copied/pasted) from a transcript, and system- net modeling, designed for analyst use and primarily provided units, with combined approaches also used in for collaboration inside the firewall; although HTML some applications. Further, they compare the advan- can be exported, Web-based collaboration is not sup- tages and disadvantages of user-controlled and system- ported. Similarly, Carneades43 maps can be shared in controlled layouts for education. Their discussion of LKIF, but not directly visualized online. Araucaria44 ontologies is limited. [150,142,144] offers a searchable online argument cor- This tools coverage in this section differs from pre- pus, but not online display of its arguments. While vious coverage in its scope of collaborative, Web- Convince Me45 offers a Java applet for display, argu- based tools with argumentation components, and in ments cannot be saved or published via the applet. its attempt at comprehensiveness. A further bias has Tools failing the ‘argumentative discussion’ crite- been software aimed at use by the public, rather than rion included general Web2.0 tools as well as Anek- exclusively for government consultation, enterprise dotz, and Vox Populi. General Web2.0 tools (e.g. Twit- decision-making, or learning argumentation and criti- ter46, Facebook47) and social software (generic mail- cal thinking skills. However, we have deliberately in- ing lists, forums) were excluded since their argumen- cluded several research prototypes which focus on Se- tation support is peripheral. Anekdotz48 failed because mantic Web approaches to argumentation on the Web the sites currently using it focus on the emotional, and on supporting the nascent World Wide Argumen- rather than the rational, aspects of argumentation. For tation Web. example, the breakups section of When You Knew asks commenters to click on either ‘Put their stuff on 9.2. Scope: Collaborative, Web-based tools with the curb’ or ‘Give em another shot’ to solicit feed- argumentation components back, which is marked as positive, negative, or neutral. Vox Populi49 [24] supports documentary filmmakers Tools were considered in-scope if they were collab- in generating argumentative film sequences based on orative (i.e. involved sharing information among mul- annotated interviews. tiple parties who could build upon each others’ work Further, tools were treated differently depending on in some way), Web-based (i.e. allowed display of in- their origin and availability; for instance, it was con- formation on the Web), and had argumentative dis- sidered helpful to include many contemporary research cussion components. By argumentative discussion, we systems even though we were not able to interactively mean discussion around disagreements, explanations, explore Web-based demo versions for some of those and reasons, coming from or including a rational (i.e. systems. We have inevitably missed some relevant sys- reason-based) standpoint. tems, and would appreciate the reader’s assistance in Some prospective tools were excluded due to failing fixing this flaw. one or more of these conditions. Tools failing the ‘collaborative’ criterion included 39Note that we do include the similarly named ‘ConvinceMe’ site the EUProfiler37, and the HealthCentral/Washington in Section A.17, page viii. Post Poligraph 200838. Users of these tools viewed per- 40http://www.w3.org/WIT/ 41 sonalized visualizations, based on their answers to a http://www.inet.saic.com/inet-public/siam. htm questionnaire, however they are not asked (or able) to 42http://www.inet.saic.com/inet-public/ share their comments on others’ views, to interact with causeway.htm other users (adding to a larger debate), or to contribute 43http://carneades.berlios.de/ to sensemaking or analysis of existing argumentation. 44http://araucaria.computing.dundee.ac.uk/ 45http://codeguild.com/convinceme/ 46http://twitter.com 37http://euprofiler.eu/ 47http://www.facebook.com/ 38https://www.washingtonpost.com/wp-srv/ 48http://www.anekdotz.com/ health/interactives/poligraph/ 49http://homepages.cwi.nl/~media/demo/IWA/ 24 Schneider et al. / A Review of Argumentation for the Social Semantic Web

9.3. Classifications of social tools pendix. Argument maps are one classic representation, which continues to be popular with a variety of tools, Aakhus and collaborators [49,4] classify argumen- including Argunet, Cohere, and Climate CoLab. tation software by use: issue networking, funnelling, On Argunet, users have significant control over the or reputation (Figure 15 on the next page). Shum says presentation of arguments, such as colors and descrip- that each tool is ‘tuned’ to a different task: “foraging tions of different argument families. Related maps can for material, classifying and linking it, discussing it in be published in series, as shown in Figure 17(a) on meetings and online, and evaluating specific points in page 27. In the argument map representation, each more depth" [163]. We later use this categorization, as node can be opened up to reveal a matrix listing ‘Functional type’ in our comparison of tools. which other arguments support, attack, are supported Scheuer et al. [153] compare the visualization and by, and are attacked by the given node (Figure 17(b) representation styles of argumentation tools used in on page 27). computer-supported collaborative learning. They summarize the pros and cons of 5 representation styles, as shown in Figure 16 on the facing page. We later use this categorization, as ‘Representation style’ in our comparison of tools. Scheuer’s representation styles are typically used for discussions (linear representation), modeling (container50), or both (threaded, graph). For instance, graph representations are highly expressive, with explicit labelling of relationships, but make it hard to see temporal sequences.

(a) 9.4. List of social and Semantic Web tools considered

In this section we draw from our review of thirty- seven online argumentation tools: AGORA: Partici- pate - Deliberate, ArgDF, Arguehow, Argument Blog- ging, Argumentum, Argumentations.com, Argunet, Avicenna, bCisiveOnline, Belvedere, Cabanac’s annotation system, Climate CoLab, Cohere, Compet- ing Hypotheses, ConsiderIt, ConvinceMe, CoPe_it!, CreateDebate, Debate.org, Debategraph, Debatepe- dia, Debatewise, DiscourseDB, Dispute Finder, Hy- (b) pernews, LivingVote, Opinion Space, Online Visual- isation of Arguments, Parmenides, PDOnline, REA- Fig. 17. Argunet can show an (a) overview of several related argument maps; and (b) in each individual map, nodes can be opened up SON, Riled Up!, SEAS, Trellis, TruthMapping, and to show arguments they support, attack, are supported by, and are Videolyzer. These systems are described and discussed attacked by. individually, in alphabetical order, in the Appendix, Section A, page i. In Argumentum51 arguments are colored to indicate the supporting (green) and opposing (red) argu- 9.5. Visualization ments (Figure 18 on page 27). Comments, but not their replies, are similarly colored to indicate agree- Overviews and visual representations aid under- ment or disagreement. Pro and con arguments are dis- standing. Here we point out visualization features of tinguished by green and red lines to the left of a com- some Social Web argumentation tools, along with ment, posted linearly, rather than in two columns. screenshots; further details are available in the ap- Competing Hypotheses52 supports breaking down information into hypotheses, evidence, and analysis,

50The container approach uses discrete visual areas to group related items. For example in Debatepedia each question is contained 51http://arg.umentum.com/ in a frame with pro and con arguments on that question. 52http://competinghypotheses.org/ Schneider et al. / A Review of Argumentation for the Social Semantic Web 25

Fig. 15. Issue-networking, funnelling, and reputation, from [49].

Fig. 16. Comparison of the visualization and representation styles of CSCL argumentation tools, from [153]. 26 Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 18. In Argumentum, the left-hand color bars indicate the sup- Fig. 21. At CreateDebate, users add and comment on pro and con porting (green) and opposing (red) arguments. arguments. which are entered into a matrix as shown in Fig- ure 36(a) on page vii. The matrix can help visually indicate the most likely and least likely scenarios.53 Mul- tiple analyses can be combined to provide a group view (Figure 36(b) on page vii), or compared pairwise. ConsiderIt54 [94,93] powers the Living Voters’ Guide55. What is unique is the possibility to drill down to understand other voters’ perspectives. In addition to seeing pros and cons on an issue from all voters, re- gardless of their stance, (Figure 20(a) on the following page), the Living Voters’ Guide can show the key points for a particular group of voters (Figure 20(b) on the next page), such as those undecided on the issue or strongly supporting it. This can help users understand Fig. 22. Opinion Space maps comments in a constellation view. what makes an issue controversial. Users indicate how they feel about an issue before and after reading an SEAS [105,104] structures arguments as templates, argument (deliberative polling), which could also be showing a colored tree view (Figure 23 on the facing used to ﬁnd the most convincing arguments. page). SEAS visualization features are also consider- CreateDebate (Figure 21 on page 27) offers nu- able: to visualize multiple dimensions, SEAS uses star- merous statistics for each debate, such as the lan- burst, constellation, and table views. guage grade level, average word lengths, and vocabulary overlap, as well as a wordcloud. Some debates 9.6. Ease of use have more than two sides. Replies can ‘support’, ‘dis- By ‘ease of use’, we mean how easy an interface is pute’, or ‘clarify’ a given point. 56 In Opinion Space [55], opinions are mapped in a to use, based on our own perception. ConvinceMe constellation, using principal component analysis, to lets users add arguments to pro or con columns, or add show a user where they stand compared to other re- a rebuttal by clicking a button. Arguments can be voted up or down. Various other systems (discussed in the spondents, as shown in Figure 22 on page 27. Each appendix) have similar functionality. point in the visualization represents a perspective; Debatepedia57, a wiki-based system, provides an larger points represent more popular perspectives. intuitive editing environment, where users can edit the entire page or just the relevant section, such as the pro 53More sophisticated ACH-based software uses matrices as input or con for a topic. to Bayesian probabilistic reasoning. 54http://engage.cs.washington.edu/ considerit/ 56http://www.convinceme.net/ 55http://www.livingvotersguide.org/ 57http://debatepedia.idebate.org/ Schneider et al. / A Review of Argumentation for the Social Semantic Web 27

(a) (b)

Fig. 19. In Competing Hypotheses, (a) each individual’s analysis is represented in a consistency matrix; (b) multiple analyses can be combined to create a group matrix. In the group view, darker shades of purple indicate more disagreement.

(a) (b)

Fig. 20. The Living Voters’ Guide compiles pro and con lists on each issue. They give (a) an overview of what all voters think about the issue; as well as (b) the key points for undecided voters.

Living Vote58 asks participants to vote on arguments in a tree. “To vote on an argument, you must ﬁrst prove you’ve read it by answering a simple question.” Users can also add arguments to the tree.

9.7. Collaboration

Systems approach collaboration in various ways. bCisive Online is intended for real-time collaboration in conjunction with with audio conferencing. One person edits the map at a time, adding nodes and connections between nodes while others can point with their cursor or request editing control. Climate CoLab uses moderation to help review comments and deal with the learning curve for argument mapping. While any user can add Pros, Cons, and Issues directly to an argument map, moderators are ex- Fig. 23. SEAS visualizes arguments as a tree; colors indicate the credibility level [105]. 58http://www.LivingVote.org/ 28 Schneider et al. / A Review of Argumentation for the Social Semantic Web pected to keep track of the conversation, adding new on page xvii), users can continually modify each con- ideas to the argument map as needed. tribution, but can only post one critique on each node. Competing Hypotheses59 has persistent chat (es- Anyone can leave a rebuttal, but only one user, the sentially a comment thread) for the entire project as original arguer, can modify the map. The system in- well as message boards for each hypothesis, evidence dicates when comments are out of sync with the map, item, and evidence-hypothesis pair. and a wiki-style history is available to better trace the On Debatewise60, everyone can collaborate in cre- conversation. ating the strongest case both for and against a given issue, using teams. 9.10. Deliberative Polling

9.8. User Engagement One measure of arguments is how persuasive they are. Measuring how users feel about an issue before Debatewise also makes it easy to get involved by and after reading an argument–known as deliberative providing suggestions of 5-minute, 20-minute and 1- polling [57]–is used in some systems, such as Con- hour tasks and showing “7 things you should have an siderIt, Living Vote, and OpinionSpace. opinion on" in rotating images on the homepage. Debate.org uses deliberative polling but places Social networking is one part of the engagement at more importance on other factors besides agreement Debate.org, which allows users to search for people when scoring debates, giving the most importance to with particular profile attributes, such as income, lo- using reliable sources and making convincing argu- cation, ideology, gender, president, religion, and who ments. they are interested in and looking for. Along with adding comments, on many sites, users 9.11. Distributed and Federated Systems can vote for arguments that convinced them (e.g. Cre- ate Debate). A user’s reputation is generally based on Systems can be distributed, like Argument Blog- the success of their arguments. Votes may be weighted, ging, in which JavaScript code is placed on blogs to for instance after 10 up or down votes, further votes have less influence on ArgueHow. link back to a server which provides access to a larger Competition helps engage users at other sites, which conversation. can be quite explicit about this aspect; for instance, Federation is an exciting direction: Argunet [154] Riled Up!’s tagline is “Like Raising Cain? So Do We." uses an open source federation system for sharing ar- 62 Other sites, like ConvinceMe treats debates as games; gument maps from a desktop tool . Uses can run their 63 in one such game, the debater whose idea is most pop- own server or use a public server, Argunet.org , which ular is crowned “King of the Hill”. Competitive de- allows authors to make maps public or restrict viewing bating environments may use point schemes and user and/or editing to a specified group. rankings to motivate contributions. These mechanisms also impact the privacy of a system–whether work can be saved privately, used col- 9.9. Balancing Contributions laboratively with a small group, or shared publicly with the world. There are several approaches to balancing contributions. For example, by removing authorship markers, 9.12. Annotation argument maps may increase the neutrality of a conversation. Annotating materials or commenting on existing Another approach to balancing is taken by TruthMap- discussions can be an important way to interact on ping61, which focuses on having a persistent con- the Social Web. Annotation can also be used to clas- versation which can not be drowned out by a single sify messages. For instance, Hypernews64 [21] asks opponent. Users can leave feedback in critiques at- users to indicate what kind of message they are post- tached to each premise and conclusion (Figure 53(b)

62http://www.argunet.org/editor/ 59http://competinghypotheses.org/ 63http://www.argunet.org/debates/ 60http://debatewise.org/ 64http://www.hypernews.org/HyperNews/get/ 61http://www.truthmapping.com/ hypernews/reading.html Schneider et al. / A Review of Argumentation for the Social Semantic Web 29 ing (None, Question, Note, Warning, Feedback, Idea, et al. [206] propose using a controlled natural lan- More, News, Ok, Sad, Angry, Agree, Disagree). guage called Attempto Controlled English [46] for Videolyzer65 [50] provides an integrated discussion high-stakes argumentative discussions, in order to gen- forum for annotating and challenging the claims a erate a first-order-logic representation of the discus- video makes. Segments likely to be of interest are iden- sion. tified ahead of time by processing both the transcript Dispute Finder66 [53,54] provides just-in-time in- and the video. formation, alerting users when information they read Annotation has also been treated in from an argu- is disputed, based on its database of disputed claims. mentation perspective in the research of Cabanac et This relies on a disputes database which was first pop- al., who study social validation of argumentative de- ulated by hand-annotation by activists (interested in in- bates through collective annotations [28]. forming or convincing others) and then extended algorithmically. The algorithm, which can be applied on 9.13. Incremental Formalization the Web at large, uses a set of 54 patterns to identify possible disputed claims. With incremental formalization, representations are useful before they are fully interpretable by the com- 9.15. Mixed Initiative puter. Incremental formalization can be helpful since people find it difficult to make structure, content, or Another possibility is to use Mixed Initiative sys- procedures explicit [159]. As the user’s understanding tems, wherein the actions of both humans and ma- (or goals) change, some systems facilitate systematic chines are important. Online Visualisation of Argu- additions or changes. ment (OVA)67 is part of a pipeline of argumentation With Argunet, arguments can be quickly sketched tools [168] which starts to bridge the gap between or reconstructed as premises and conclusions, support- human-oriented argumentation tools and calculation- ing incremental formalization. based agent argumentation. Mixed initiative discus- CoPe_it! transforms the user-created informal spa- sions are enabled by the argument maps created by tial hypertext view (Figure 39(a) on page ix) into an OVA or any other AIF-based tool. Thus, instead of rep- issue chart (Figure 39(b) on page ix) according to rules resenting one’s point of view countless times in a fo- shown in Figure 39(c) on page ix). Users can also cus- rum or FAQ, it would be possible to delegate these tomize the transformation rules. conversations to a machine agent using an underlying argument map, as prototypes like MAgtALO68 9.14. Populating Argumentative Knowledge Bases [145,202] and the Google Wave discussion bot Arvina From User Input show.

Other approaches are also iterative, similar to incre- 9.16. Search mental formalization but focusing speciﬁcally on the input phase. For instance, user input may be processed, Semantic search focuses not on mere keyword and the output presented to the user, who can then cor- matches but on retrieving structured data, such as rect it. whether an opinion is argued for or against. Seman- Trellis introduced a language processing technique tic search is possible with several tools. ArgDF69 uses called “Annotation Canonicalization through Expres- the AIF-RDF ontology described above [139,138,212] sion synthesis" [23], which applied an ontology to a and Sesame RDF70. In ArgDF, it is possible to display user-supplied sentence, checked the computer’s ontol- all the arguments in which a claim is involved (e.g. ogy application by presenting a paraphrase to the user, where it is used as a conclusion or as a premise) or all and solicited additions to the ontology from unknown the arguments using, say, the Argument from Expert or misunderstood words. Opinion reasoning scheme. Controlled natural languages, which adopt a more restrictive grammar and vocabulary in order to facili- 66http://ennals.org/rob/disputefinder.html tate parsing, have also been used to take in informa- 67http://ova.computing.dundee.ac.uk tion, formalizing it for reasoning. For instance, Wyner 68http://www.arg.dundee.ac.uk/?page_id=61 69http://argdf.org/ 70Various other tools tools export AIF without directly imple- 65http://videolyzer.com/ menting semantic search. 30 Schneider et al. / A Review of Argumentation for the Social Semantic Web

DiscourseDB71 uses Semantic MediaWiki [95] with tive properties, Avicenna can support argument chain- the SemanticForms72 extension. This makes it possi- ing, such as retrieving all arguments that directly or in- ble to list all commentary written by particular person, directly support a given conclusion. Avicenna is also published in a particular venue, and so forth. Further, used to infer the classification hierarchy of argument since items indicate the position they take on a topic, schemes: for example, an appeal to expert opinion is a DiscourseDB can list all commentary for or against specialization of an argument from position to know. a given position. When a topic has multiple positions (e.g. Darfur73), DiscourseDB is especially helpful in summarizing the discussion. Semantic search uses a 10. Discussion & Conclusion simple syntax: for instance, on the Darfur conflict issue, to search for commentary opposing the position We have reviewed argumentation theory, existing that the U.N. should send peacekeepers, this code is ontologies with argumentative aspects, and Web tools used: [[is against::Darfur conflict / United Nations for argumentation. We now discuss three main gaps should send peacekeepers]]. Since the results are al- based on our observations. First, the ontologies given ready displayed on summary pages, most end users need further adaptation to meet the existing variety would not need to create or modify queries, but Dis- of social tools and purpose. In particular, arguing is courseDB’s semantic search is a powerful tool for cre- a social activity. The varieties of argument tools on ating summaries. the Social Web need distinct types of interface support and social engineering. The remaining question is what 9.17. Reasoning and Querying appropriate Social Web ontologies for argumentation would be. Rather than a single ontology, we envision As previously mentioned, Attempto Controlled modular, interoperable components. English can generate first-order-logic representations, which allow inference and consistency-checking, and 10.1. Arguing is a Social Activity can be translated into OWL and RuleML [58]. Mean- while, Open Vocabulary Executable English can be As argumentation scholars have long realized, hu- used for simple reasoning in the Internet Business mans argue for a variety of reasons, not always to solve Logic System74. “wicked problems". Rather, arguing is a social activ- Parmenides75 [32,33,34] is a structured survey tool ity people may use to position and establish them- for gathering public opinion on a proposal. Based on selves. This kind of arguing is important in the Social argument schemes and critical questions from argu- Web, where people play by arguing such as with Con- mentation theory, Parmenides can pinpoint the source vinceMe’s the ‘King of the Hill’ game, or create net- of the disagreement, by having participants respond to works of friends and enemies, such as on Riled Up! a series of questions. At the end of the survey, users and Create Debate. Arguing can also be used to con- are offered the choice of submitting an alternative pro- nect people such as on Debate.org. Ontologies for the posal, and are shown the answers they chose. Admin- Social Semantic Web will need to respect these social istrators can then analyze the group’s responses, which aspects, and may need to incorporate emotive indica- are displayed in graphical argumentation frameworks tors such as the heat of the debate as well as the manner [51]. in which the outcome will affect the participants. More advanced reasoning and querying is enabled The notion of debate, where two parties face off, is by Avicenna, an OWL-based argumentation system also well-represented in existing social tools. Debate on the Web which uses Jena [31], ARQ76, and Pel- may allow individuals to show their expertise, to find let [167]. Since OWL supports inference over transi- the best arguments, or simply to practice their rhetorical skills. Debate topics may be reused, for ongoing 71http://discoursedb.org/ issues with two or more defendable positions, espe- 72http://www.mediawiki.org/wiki/Extension: cially when a topic is controversial. This suggests two Semantic_Forms opportunities. First of all, future Social Semantic Web 73 http://discoursedb.org/wiki/Darfur_ prototype tools for sensemaking and argument map- conflict 74http://www.reengineeringllc.com/ ping could be tested with for argumentation for some 75http://cgi.csc.liv.ac.uk/~parmenides/ common debate topic in order to find a large audience 76http://jena.sourceforge.net/ARQ/ of potential evaluators. Second, providing meaningful Schneider et al. / A Review of Argumentation for the Social Semantic Web 31 ways to discover new debate topics, and potentially can be more complex and involved than those for ca- record and share the outcome of these debates, could sual use by the general public. E-government and de- be helpful. Frequent debaters may also provide an in- liberation tools have the strictest usability needs for teresting class of users since we might expect them to this reason. be more familiar with fallacies and argument diagramming, making them potentially more savvy about ar- 10.3. Bridging the Social Web and the Semantic Web gumentation schemes and similar abstractions. to Manifest the World Wide Argument Web

10.2. Current Use To conclude, we discuss the obstacles to manifesting the Social Semantic Argumentation Web, along with a While argumentation support has become more research agenda. mainstream, it is still a niche. While there is a desire for public discussion systems, especially in areas such 10.3.1. Problems as e-government, social discussion systems and social Despite candidate technical architectures for a World networks are driven by network effects (e.g. you are Wide Argument Web, the WWAW is not yet viable on persuaded to use them by the ability to communicate the Social Web at large. We notice several interrelated with your friends and colleagues) and by ease of use. obstacles: first, the existing ontologies are not meant Argumentative elements in generic social media tools for integrating wide-scale informal discussion; second, are very basic: Facebook and Google Plus use ‘Like’ current approaches to supporting argumentation gen- and ‘+1’ buttons, which imply a semantics of agree- erally require substantial human effort; and third, de- ment; YouTube adds a ‘dislike’ button, and flagging termining the appropriate uses and re-uses for social posts for moderation (e.g. on Craigslist) or downvoting media posts depends on their context (e.g. the type of posts (e.g. on StackOverflow or Reddit) also implies discourse). dislike. First, the very informality of social media can make With existing systems, discussed in the Appendix, discussions more difficult to integrate. Argumentation there is a continuum from those with little use to those is used in many contexts and while formal argumen- with wide use. Some (non-research) sites have few tation can be represented with ontologies such as AIF, users and seem to have been abandoned. Some re- argumentation on the Social Web can be quite infor- search prototypes are not accessible at all (and have mal, with missing premises and unexpressed argument been discussed based on papers and screenshots). schemes. While human analysis can sometimes bridge Other research prototypes are available, and some the gap between AIF and the Social Web, (facilitated seem to have users. Some are widely (or at least some- for instance by tools such as OVA, Section A.29, page what) used – showing (perhaps) what’s needed to build xiii), more scalable solutions are needed. Several ap- a Social Web infrastructure for argumentation. proaches will be needed to more routinely express the Argumentation support has not yet moved firmly existing argumentation on the Social Semantic Web. from the academic lab, into the mainstream. While dis- Second, most current approaches to supporting ar- cussion is widespread, argumentation needs are often gumentation still require substantial human effort; specific to the reasoning schemes used – which vary little automatic processing is available. Issues and by discipline and area. Such constraints simplify the stances can be categorized by hand, mapped and ana- reasoning process for humans as well as for argumen- lyzed, or voted up and down. Tools for certain tasks– tation support. Further, in dialogue, most argumenta- situational analysis, argumentation reconstruction, and tion happens informally: we can count on our conver- argument mapping–are highly developed. These tools sational partners to indicate what is missing and to de- have become, while not mainstream, widely accepted mand that we explain what is unclear to them. It is for certain communities. The value of spatial hypertext difficult to systematically indicate assumptions and to visualization systems cannot be discounted, and some make reasoning explicit; while this is needed for ideal automation does exist: leveraging human-devised ma- reasoning support, it is not feasible or reasonable to ex- trices using Bayesian networks (in more advanced ver- pect in everyday discourse. This leads into a discussion sions of ACH tools like Competing Hypotheses), or of usability. summarizing human-answered surveys with argumen- Usability needs depend on the task at hand and the tation frameworks (Parmenides). Yet since these value- target audience. Tools for in-depth analysis by experts added tools still require substantial human effort to en- 32 Schneider et al. / A Review of Argumentation for the Social Semantic Web ter information in particular formats, their very use is Language (CNL); whereas on the general Social Web, an encouraging sign that some forms of argumentation CNL would restrict input, in analysis tools, CNL might support would be beneficial. open the vocabulary. Third, context is important for integrating conversa- Second, to further address the human effort re- tions and claims. One strength of the Semantic Web is quired, we need to motivate wide-scale participation in bringing conversations together; this has been very and improve automatic argument detection. To encour- powerful for the Social Semantic Web in general. Yet age ongoing human effort, it would be helpful to create the rhetorical effect of an argument depends on certain a virtuous cycle–by which people benefit from the Ar- contextual information, such as the surrounding con- gument Web, thereby motivating increased participa- versation, its participants, and the medium. Extract- tion. Understanding the niches filled by existing tools, ing and summarizing conversations without this con- and whether these needs could be fulfilled better by text has risks (i.e. potentially presenting misleading or a larger Argument Web, would help in this regard. overstated arguments). For instance, while abstract argument schemes may not be well understood by users, Parmenides shows 10.3.2. Research Agenda that stepwise processes based on these schemes can To overcome these obstacles and manifest the So- be powerful. Opening up the analysis tools, so that a cial Semantic Argumentation Web, we see a need for group could view aggregate responses, would take Par- various approaches. First, we need ontologies suitable menides to a new level of collaboration. While Par- for representing informal Social Web arguments, and menides focuses on gathering multiple responses on to map to these ontologies. Second, to address and re- the same set of issues, a different approach would be duce the human effort required we also need to moti- to crowdsource work based on an argument scheme. vate participation, and find ways to infer argumenta- Many groups already do this informally with check- tive relationships. Third, we need to further investigate lists and procedures, for instance in Wikipedia’s article context. promotion process. Providing templates where users First, we need ontologies that map between the could indicate which critical questions they have asked social world and the argumentative world. A modu- and answered, and at what point in time, might help lar approach will be needed, reusing existing work, to distribute and share this work, while making the un- both in domain knowledge and in Social Web mod- derlying process more transparent. eling, for instance by importing existing ontologies Automatic detection of arguments might help fur- (particularly SIOC). Maximal benefit from a Social ther bootstrapping the existing Social Web into the Ar- Web ontology for argumentation will come from align- gument Web. In the scholarly communication and le- ing and crosswalking to popular ontologies (especially gal fields, argument detection relies on rhetorical fea- AIF). Many tools can already output AIF, and analyst- tures. Argumentative markers would also help in mod- oriented tools can be brought onto the Argument Web ifying these argument detection approaches for use on with comparatively little effort. Motivated users and the Social Web. Analyzing existing Social Web cor- defined argumentation schemes ease this process. puses, such as DisputeFinder’s claims database and the Mapping to these ontologies will also leveraging the Discussion Fora from the Aracaria corpus may help in existing human effort already used in argumentation determining such markers. Various corpus-processing tools. SEAS, for example, already uses argument tem- techniques and approaches may be useful for detecting plating. Such templates appear to be specialized ar- argumentation, which shares rhetorical features with gument schemes, which could be expressed in shared other sorts of speech. Linguistic pragmatics dominate repositories and even classified (for instance using in much argumentation, so one form of progress would OWL as Avicenna does). Once the argument schemes be to find unassailable features which mark argumen- can be referenced, SEAS might provide another source tative contexts on the Social Web. Relevant approaches of AIF data, as well as point to further enrichment may come from opinion mining [124], question an- needed. The ACH process underlying Competing Hy- swering and explanation [119], contradiction detection potheses seems to use a narrower set of reasoning; its [149], controversy detection [72], persuasion detection data, similarly, might be encompassed by understand- [211,8], stance detection [188] and automatically typing and expressing the ACH argument scheme. The an- ing links [45]. alyst community is also a good place to start with in- Third, we need to investigate how to preserve the terface interventions such as using Controlled Natural rhetorical effect of an argument even when it is di- Schneider et al. / A Review of Argumentation for the Social Semantic Web 33 vorced from its original context. Some aspects of con- Acknowledgements text are straightforward: for instance, items can be con- fusing or non-sensical when stripped of context. The This work was supported by Science Foundation canonical example from metadata scholars is the “on Ireland under Grant No. SFI/09/CE/I1380 (Líon2). a horse” [204] problem: in the context of a Theodore We are grateful for the extensive comments of the Roosevelt collection, the description “on a horse" ad- reviewers, Iyad Rahwan, Simon Buckingham Shum, equately describes a photo; yet outside of that context and Fouad Zablith. Thanks also to Ágnes Sándor for her comments which have prompted further clarifica- it is unclear and diffuse. Other contextual factors in- tions and furthered our thinking. The re-revision of clude the type of argumentative discourse. The vari- this paper also benefited significantly from conversa- ous types of argumentative discourse, shown in Fig- tions with and feedback from Katie Atkinson, Trevor ure 4 on page 7, vary in the amount of interest and Bench-Capon, and Adam Wyner, thanks to a visit to value they generate, outside the immediate context of the University of Liverpool sponsored by the COST the discussion. Eristic dialogue, of personal conflicts, Action IC0801 on Agreement Technologies, STSM for instance, is generally not worth reusing (though it 1868. could be used to establish, for instance, who started an argument, or that a dyad should avoid further discussions). Discovery dialogue, on the other hand, which References seeks to find and defend a hypothesis, can be useful both for understanding the process undertaken and the [1] RDF Vocabulary Description Language 1.0: RDF Schema. W3C Recommendation 10 February 2004, World Wide outcome. Web Consortium, 2004. http://www.w3.org/TR/ Support may also depend in part by who reuses rdf-schema/. discussions–the participants, outside parties, or both– [2] SPARQL query language for RDF. W3C Recommenda- and how much support they need at various points in tion 15 January 2008, World Wide Web Consortium, 2008. http://www.w3.org/TR/rdf-sparql-query. time. Reviews, for instance, are mainly written for an [3] OWL 2 Web Ontology Language: Document Overview. external audience. Blog and microblog posts may be W3C Recommendation 27 October 2009, World Wide read by others but also searched by the author as a Web Consortium, 2009. http://www.w3.org/TR/ form of externalized memory. There may also be a owl2-overview/. [4] Mark Aakhus. Pragmatic Web as communication design temporal component: standards bodies use their own practice. In PragWeb 2006: Proceedings of the First Inter- discussions in order to make decisions, but after the national Conference on the Pragmatic Web 21-23 September fact, these discussions may be of considerable interest 2006 Stuttgart, Germany. Ges. für Informatik, Bonn, 2006. to non-participants trying to understand why a partic- [5] Mark Aakhus and Miriam Greenfeld Benovitz. Argument reconstruction and socio-technical facilitation of large scale ar- ular decision was taken. The decision-making associ- gumentation. In Proceedings of the 3rd International Confer- ated with discussions may be a particular point where ence on the Pragmatic Web: Innovating the Interactive Soci- support is needed. Decisions can be taken by groups ety, pages 77–81, 2008. (e.g. standards bodies) or by individuals (e.g. from re- [6] Ágnes Sándor. Using the author’s comments for knowledge discovery. In Semaine de la connaissance, Atelier texte et views) and may depend on a centralized discussion or connaissance, Nantes, France, 2006. widely distributed pieces of information. [7] Vincent Aleven and Kevin D. Ashley. Evaluating a learning Overall, there is significant potential for support- environment for case-based argumentation skills. In Proceed- ing argumentation on the Social Semantic Web, but a ings of the 6th international conference on Artificial Intelli- gence and Law, ICAIL 1997, pages 170–179, New York, NY, large amount of work remains to be done, in creat- USA, 1997. ing ontologies, easing human annotation of arguments, [8] Pranav Anand, Joseph King, Jordan Boyd-Graber, Earl Wag- improving techniques for detecting and mining argu- ner, Craig Martell, Doug Oard, and Philip Resnik. Believe mentation, and in marking the context, pragmatics, me–we can do this! Annotating persuasive acts in blog text. In AAAI Workshop on Computational Models of Natural Ar- and provenance of dialogues. Given the vast amount gument at the Twenty-Fifth AAAI Conference on Artificial In- of commentary on the social web – much of it argu- telligence, 2011. mentative, persuasive, or opinionated – there is a great [9] M.A. Angrosh, Stephen Cranefield, and Nigel Stanger. need for further research and technical developments Ontology-based modelling of related work sections in research articles: Using CRFs for developing semantic data to search and organize these discussions with semantic based information retrieval systems. In iSemantics 2010, web technologies. 2010. 34 Schneider et al. / A Review of Argumentation for the Social Semantic Web

[10] Anupriya Ankolekar, Markus Krötzsch, Duc Thanh Tran, and [27] John G. Breslin, Alexandre Passant, and Stefan Decker. The Denny Vrandecic. The two cultures: Mashing up Web 2.0 and Social Semantic Web. Springer, 2009. the Semantic Web. Journal of Web Semantics, 6(1):70–75, [28] Guillaume Cabanac, Max Chevalier, Claude Chrisment, and 2008. Christine Julien. Social validation of collective annotations: [11] Kevin D. Ashley. Modeling legal arguments: Reasoning with Definition and experiment. Journal of the American Soci- cases and hypotheticals. MIT Press, Cambridge, MA, USA, ety for Information Science and Technology, 61(2):271–287, 1991. 2010. [12] Kevin D. Ashley and Edwina L. Rissland. But, see, accord: [29] Daniela V. Carbogim, David Robertson, and John Lee. Generating blue book citations in HYPO. In Proceedings of Argument-based applications to knowledge engineering. The the 1st International Conference on Artificial Intelligence and Knowledge Engineering Review, 15(2):119–149, 2000. law, pages 67–74, 1987. [30] Jeremy J. Carroll, Christian Bizer, Patrick Hayes, and Patrick [13] Katie Atkinson, Trevor J. M. Bench-Capon, and Doug Wal- Stickler. Named graphs, provenance and trust. In Proceed- ton. Distinctive features of persuasion and deliberation dia- ings of the 14th International Conference on World Wide Web, logues. Argument & Computation, 2012. in press. 2004. [14] Trevor J. M. Bench-Capon. Specification and implementa- [31] Jeremy J. Carroll, Ian Dickinson, Chris Dollin, Dave tion of Toulmin dialogue game. In Proceedings of JURIX, Reynolds, Andy Seaborne, and Kevin Wilkinson. Jena. In volume 98, pages 5–20, 1998. Proceedings of the 13th international World Wide Web con- [15] Trevor J. M. Bench-Capon. Persuasion in practical argument ference - Alternate track papers & posters - WWW Alt. ’04, using Value-based Argumentation Frameworks. Journal of page 74, 2004. Logic and Computation, 13(3):429–448, 2003. [32] Dan Cartwright and Katie Atkinson. Political engagement [16] Trevor J. M. Bench-Capon and Edwina L. Rissland. Back to through tools for argumentation. In Computational models the future: Dimensions revisited. In Proceedings of JURIX of argument: Proceedings of COMMA 2008, pages 116–127. 2001, 2001. IOS Press, 2008. [17] Neil Benn, Simon Buckingham Shum, John Domingue, and [33] Dan Cartwright and Katie Atkinson. Using computational ar- Clara Mancini. Ontological foundations for scholarly debate gumentation to support e-participation. IEEE Intelligent Sys- mapping technology. In Second International Conference tems, 24(5):42–52, 2009. on Computational Models of Argument (COMMA ’08), May [34] Dan Cartwright, Katie Atkinson, and Trevor Bench-Capon. 2008. Supporting argument in e-democracy. In Electronic Gov- [18] Tim Berners-Lee. Linked Data. Design issues for the world ernment: Third International Electronic Democracy (EDEM wide web, World Wide Web Consortium, 2006. http:// 2009), pages 151–160, 2009. www.w3.org/DesignIssues/LinkedData.html. [35] Daniel Robert Cartwright. Digital Decision-Making: Using [19] Tim Berners-Lee and Mark Fischetti. Weaving the Web: The Computational Argumentation to Support Democratic Pro- Original Design and Ultimate Destiny of the World Wide Web cesses. PhD thesis, University of Liverpool, 6 2011. by its Inventor. Harper Collins Publishers, New York, 1999. [36] Alexander Garcia Castro, Angela Norena, Andres Betancourt, [20] Tim Berners-Lee, James A. Hendler, and Ora Lassila. The and Mark A. Ragan. Cognitive support for an argumentative Semantic Web. Scientific American, 284(5):34–43, 2001. structure during the ontology development process. In Pro- [21] David Alan Black. Introducing HyperNews. Linux Journal, ceedings of the 9th International Protégé Conference, Stan- (27), July 1996. ford, California, July 2006. [22] Elizabeth Black and Anthony Hunter. Using enthymemes in [37] Andy R. Cavagnetto. Argument to foster scientific literacy. an inquiry dialogue system. In Proceedings of the 7th interna- Review of Educational Research, 80(3):336–371, 2010. tional joint conference on Autonomous agents and multiagent [38] Steve Cayzer. Semantic blogging and decentralized knowl- systems-Volume 1, pages 437–444. International Foundation edge management. Communications of the ACM, 47(12):47– for Autonomous Agents and Multiagent Systems, 2008. 52, 2004. [23] Jim Blythe and Yolanda Gil. Incremental formalization of [39] Hsinchun Chen. AI, e-government, and politics 2.0. IEEE document annotations through ontology-based paraphrasing. Intelligent Systems, 24(5):64–86, 2009. In Proceedings of the 13th International Conference on World [40] Timothy Chklovski, Yolanda Gil, Varun Ratnakar, and John Wide Web, pages 455–461, 2004. Lee. TRELLIS: supporting decision making via argumenta- [24] Stefano Bocconi, Frank Nack, and Lynda Hardman. Sup- tion in the Semantic Web. In Proceedings of the Second In- porting the generation of argument structure within video se- ternational Semantic Web Conference, October 2003. quences. In Proceedings of the 16th ACM conference on Hy- [41] Timothy Chklovski, Varun Ratnakar, and Yolanda Gil. User pertext and hypermedia, pages 75–84, 2005. interfaces with Semi-Formal representations: a study of de- [25] David Braybrooke. Introduction to special issue of argumen- signing argumentation structures. In Proceedings of the tation originating in a conference at The University of Texas 10th International Conference on Intelligent User Interfaces, at Austin. Argumentation, 17(4):361–363, 2003. pages 130–136, 2005. [26] John G. Breslin, Andreas Harth, Uldis Bojars,¯ and Stefan [42] Paolo Ciccarese. Scientific Discourse Relationships Ontol- Decker. Towards Semantically-Interlinked Online Commu- ogy Specification. Technical report, Massachusetts General nities. In The Semantic Web: Research and Applications, Hospital, December 2008. Proceedings of the second European Semantic Web Confer- [43] Paolo Ciccarese, Marco Ocana, Sudeshna Das, and Tim ence (ESWC ’05), number 3532 in LNCS, pages 500–514. Clark. AO: an open Annotation Ontology for science on the Springer, 2005. Web. In Bio-Ontologies 2010, 2010. Schneider et al. / A Review of Argumentation for the Social Semantic Web 35

[44] Paolo Ciccarese, Elizabeth Wu, Gwen Wong, Marco Ocana, [60] Yolanda Gil and Varun Ratnakar. TRELLIS: An Interac- June Kinoshita, Alan Ruttenberg, and Tim Clark. The SWAN tive Tool for Capturing Information Analysis and Decision biomedical discourse ontology. Journal of Biomedical Infor- Making. Internal Project Report, USC Information Sciences matics, 41(5):739–751, October 2008. Institute, 2001. http://www.isi.edu/expect/web/ [45] Chip Cleary and Ray Bareiss. Practical methods for automat- trellis/Trellis-8-01.pdf. ically generating typed links. In Proceedings of the Seventh [61] Yolanda Gil and Varun Ratnakar. Trusting information ACM Conference on Hypertext, pages 31–41, 1996. sources one citizen at a time. In Proceedings of the First Inter- [46] Juri Luca De Coi, Norbert E Fuchs, Kaarel Kaljurand, and national Conference on The Semantic Web, pages 162–176. Tobias Kuhn. Controlled english for reasoning on the Seman- Springer-Verlag, 2002. tic Web. In Semantic Techniques for the Web–The REWERSE [62] Thomas F. Gordon, Henry Prakken, and Douglas Walton. The Perspective, volume 5500 of Lecture Notes in Computer Sci- Carneades model of argument and burden of proof. Artificial ence, pages 276–308. Springer, 2009. Intelligence, 171(10-15):875–896, July 2007. [47] Jeffrey Conklin. Dialogue Mapping: Building Shared Under- [63] Wayne S. Grennan. Informal Logic: Issues and Techniques. standing of Wicked Problems. John Wiley and Sons, Ltd., McGill Queens University Press, Montreal, Quebec, 1997. 2005. [64] H.P. Grice. Logic and conversation. In Syntax and semantics, [48] Sudeshna Das, Michael Rogan, Holli Kawadler, Samuel Brin, Vol.3, pages 41–58, 1975. Stéphane Corlosquet, and Tim Clark. PD Online: A case [65] Leo Groarke. Informal logic. In Edward N. Zalta, editor, The study in scientific collaboration on the Web. In Workshop on Stanford Encyclopedia of Philosophy. Spring 2012 edition, the Future of the Web for Collaborative Science, 19th Inter- 2012. national World Wide Web Conference, 2010. [66] Claire Grover, Ben Hachey, and Chris Korycinski. Summaris- [49] Aldo de Moor and Mark Aakhus. Argumentation support: ing legal texts: Sentential tense and argumentative roles. In From technologies to tools. Communications of the ACM, Proceedings of the Text Summarization Workshop at HLT- 49(3):93–98, 2006. NAACL 2003, Human Language Technology Conference of [50] Nicholas Diakopoulos, Sergio Goldenberg, and Irfan Essa. the North American Chapter of the Association for Computa- Videolyzer: quality analysis of online informational video for tional Linguistics Volume 5, pages 33–40, 2003. bloggers and journalists. In Proceedings of the 27th Interna- [67] Tudor Groza, Siegfried Handschuh, John G. Breslin, and Ste- tional Conference on Human Factors in Computing Systems, fan Decker. An abstract framework for modeling argumenta- pages 799–808, 2009. tion in virtual communities. International Journal of Virtual [51] Phan Minh Dung. On the acceptability of arguments and its Communities and Social Networking, 1:35–47, 2009. fundamental role in nonmonotonic reasoning, logic program- [68] Tudor Groza, Siegfried Handschuh, Knud Möller, and Stefan ming and n-person games. Artificial Intelligence, 77(2):321– Decker. SALT - Semantically Annotated LaTeX for scien- 357, 1995. tific publications. In The Semantic Web: Research and Appli- [52] Paul E. Dunne, Sylvie Doutre, and Trevor J. M. Bench-Capon. cations, Proceedings of ESWC 2007, LNCS, pages 518–532. Discovering inconsistency through examination dialogues. In Springer, 2007. Leslie Pack Kaelbling and Alessandro Saffiotti, editors, IJ- [69] Thomas R. Gruber. Towards principles for the design of on- CAI, pages 1680–1681. Professional Book Center, 2005. tologies used for knowledge sharing. International Journal [53] Rob Ennals, Dan Byler, John Mark Agosta, and Barboara Human-Computer Studies, 43(5–6):907–928, 1995. Rosario. What is disputed on the Web? In WWW 2010, 2010. [70] Thomas R. Gruber. Collective knowledge systems: Where [54] Rob Ennals, Beth Trushkowsky, and John Mark Agosta. the Social Web meets the Semantic Web. Journal of Web Highlighting disputed claims on the Web. In WICOW 2010 at Semantics, 6(1):4–13, 2008. WWW 2010, 2010. [71] Ali Gürkan, Luca Iandoli, Mark Klein, and Giuseppe Zollo. [55] Siamak Faridani, Ephrat Bitton, Kimiko Ryokai, and Ken Mediating debate through on-line large-scale argumenta- Goldberg. Opinion Space: A scalable tool for browsing on- tion: Evidence from the field. Information Sciences, line comments. In Proceedings of the 28th International Con- 180(19):3686–3702, 2010. ference on Human Factors in Computing Systems, CHI ’10, [72] Karl Gyllstrom and Marie-Francine Moens. Clash of the pages 1175–1184, 2010. typings: finding controversies and children’s topics within [56] Gerhard Fischer, Andreas C. Lemke, Raymond McCall, and queries. In Proceedings of the 33rd European conference Anders I. Morch. Making argumentation serve design. on Advances in information retrieval, ECIR’11, pages 80–91. Human-Computer Interaction, 6(3):393–419, 1991. Springer-Verlag, 2011. [57] James S. Fishkin. Democracy and Deliberation: New Direc- [73] Stella Heras. Case-Based Argumentation in Agent Societies. tions for Democratic Reform. Yale University Press, 1991. PhD thesis, Universitat Politècnica de València, 2011. [58] Norbert E. Fuchs, Kaarel Kaljurand, and Gerold Schneider. [74] Stella Heras, Katie Atkinson, Vicente Botti, Floriana Grasso, Attempto controlled english meets the challenges of knowl- Vicente Juliana, and Peter McBurney. Applying argumenta- edge representation, reasoning, interoperability and user in- tion to enhance dialogues in social networks. In Proceedings terfaces. In FLAIRS Conference, pages 664–669, 2006. of the IJCAI Workshop on Computational Models of Natural [59] Marco Aurélio Gerosa, Mariano Gomes Pimentel, Hugo Argument (CMNA), 2010. Fuks, and Carlos Lucena. Analyzing discourse structure to [75] Stella Heras, Katie Atkinson, Vicente Botti, Floriana Grasso, coordinate educational forums. In Intelligent Tutoring Sys- Vicente Juliana, and Peter McBurney. How argumentation tems 2004, number 3220 in LNCS, pages 23–36. Springer- can enhance dialogues in social networks. In Computational Verlag, 2004. Models of Argument - Proceedings of COMMA 2010, number 216 in Frontiers in Artificial Intelligence and Applications, 36 Schneider et al. / A Review of Argumentation for the Social Semantic Web

pages 267–274. IOS Press, 2010. 2011 edition, 2011. [76] Stella Heras, Vicente Botti, and Vicente Juliana. On a Compu- [93] Travis Kriplean, Jonathan Morgan, Deen Freelon, Alan Born- tational Argumentation Framework for Agent Societies, vol- ing, and Lance Bennett. Supporting reflective public thought ume 6614 of LNCS, pages 123–140. Springer-Verlag New with ConsiderIt. In Proceedings of the ACM 2012 conference York Inc, 2011. on Computer Supported Cooperative Work, CSCW ’12, pages [77] Michael Hoffmann. Logical Argument Mapping 265–274, New York, NY, USA, 2012. (LAM) - A Manual. Working paper, Georgia Tech, [94] Travis Kriplean, Jonathan T. Morgan, Deen Freelon, Alan 2011. http://www.spp.gatech.edu/faculty/ Borning, and Lance Bennett. ConsiderIt: improving struc- workingpapers/wp50.pdf. tured public deliberation. In Proceedings of the 2011 annual [78] Michael H. G. Hoffmann. Logical argument mapping: conference extended abstracts on Human factors in comput- A cognitive-change-based method for building common ing systems, CHI EA ’11, page 1831–1836, 2011. ground. In Proceedings of the 2nd international conference [95] Markus Krötzsch, Denny Vrandeciˇ c,´ Max Völkel, Heiko on Pragmatic Web, ICPW ’07, pages 41–47, 2007. Haller, and Rudi Studer. Semantic Wikipedia. Journal of Web [79] Ken Hyland. Metadiscourse. Continuum, 1st edition, 2005. Semantics, 5(4):251–261, 2007. [80] Joshua E. Introne. Adaptive Mediation in Groupware. PhD [96] Werner Kunz and Horst W.J. Rittel. Issues as elements of thesis, Brandeis University, 2008. information systems. Working Paper 131, Center for Plan- [81] Joshua E. Introne. Supporting group decisions by mediat- ning and Development Research, University of California at ing deliberation to improve information pooling. In Proceed- Berkeley, Berkeley, CA, 1970. ings of the ACM 2009 International Conference on Support- [97] Christoph Lange. Semantic Web Collabora- ing group work - GROUP ’09, page 189, 2009. tion on Semiformal Mathematical Knowledge. [82] Alex P. J. Jarczyk, Peter Löffler, and Frank M. Shipman III. Ph.D., Jacobs University, School of Engineering Design rationale for software engineering: a survey. In Pro- and Science, Bremen, Germany, Germany, 2010. ceedings of the Twenty-Fifth Hawaii International Confer- https://svn.kwarc.info/repos/swim/doc/phd/phd.pdf. ence on System Sciences, 1992, volume ii, pages 577–586, [98] Christoph Lange, Uldis Bojars,¯ Tudor Groza, John G. Breslin, 1992. and Siegfried Handschuh. Expressing argumentative discus- [83] Minwoo Jeong, Chin-Yew Lin, and Gary Geunbae Lee. Semi- sions in social media sites. In Proceedings of the ISWC2008 supervised speech act recognition in emails and forums. Workshop on Social Data on the Web (SDoW2008), October In EMNLP: Proceedings of the Conference on Empirical 2008. Methods in Natural Language Processing, pages 1250–1259, [99] Christoph Lange, Tuukka Hastrup, and Stephane Corlosquet. 2009. Arguing on issues with mathematical knowledge items in [84] David R. Karger and Dennis Quan. What would it mean to a semantic wiki. In LWA: Lernen, Wissensentdeckung und blog on the Semantic Web? In The Semantic Web – ISWC Adaptivität, volume 448 of Technical Report, pages 41–48. 2004, pages 214–228. Springer, 2004. Department of Computer Science, University of Würzburg, [85] Joel Katzav and Chris Reed. On argumentation schemes Germany, 2008. and the natural classification of arguments. Argumentation, [100] Ioan Alfred Letia and Adrian Groza. Contextual exten- 18(2):239–259, June 2004. sion with concept maps in the Argument Interchange For- [86] Joel Katzav, Chris Reed, and Glenn Rowe. Argument research mat. In Argumentation in Multi-Agent Systems, volume 5384. corpus. In Proceedings of the 2003 Conference on Practical Springer, 2009. Applications in Language and Computers, pages 229–239. [101] Anna De Liddo and S. Buckingham Shum. Cohere: A proto- Peter Lang, 2003. type for contested collective intelligence. In Collective Intel- [87] Paul A. Kirschner, Simon J. Buckingham Shum, and Chad S. ligence In Organizations - Toward a Research Agenda, 2010. Carr. Visualizing Argumentation: Software tools for collab- [102] Claudia Lópea, Pablo Inostroza, Luiz Marcio Cysneiros, and orative and educational sense-making. Springer, 1 edition, Hernán Astudillo. Visualization and comparison of architec- 2003. ture rationale with Semantic Web technologies. The Journal [88] Mark Klein. Achieving collective intelligence via large-scale of Systems and Software, 82(8):1198–1210, 2009. on-line argumentation. Technical Report Research Paper [103] Rui Pedro Lourenço and Jo ao Paulo Costa. A note on democ- No. 4647-07, CCI Working Paper No. 2007-001, MIT Sloan racy and Group Support Systems concepts. In CISTI: 2010 School of Management, April. 5th Iberian Conference on Information Systems and Tech- [89] Mark Klein. The MIT Collaboratorium: Enabling Effective nologies, pages 1–6, 2010. Large-Scale Deliberation for Complex Problems. SSRN eLi- [104] John Lowrance, Ian Harrison, Andres Rodriguez, Eric Yeh, brary, (Research Paper No. 4679-08), December 2007. Tom Boyce, Janet Murdock, Jerome Thomere, and Ken Mur- [90] Graham Klyne and Jeremy J. Carroll. Resource De- ray. Template-based structured argumentation. In Knowledge scription Framework (RDF): Concepts and abstract syn- Cartography, Advanced Information and Knowledge Process- tax. W3C Recommendation 10 February 2004, World ing, pages 307–334. Springer, 2008. Wide Web Consortium, 2004. http://www.w3.org/ [105] John D. Lowrance. Graphical manipulation of evidence in TR/rdf-concepts/. structured arguments. Law, probability and risk, 6(1-4):225, [91] Alistair Knott and Ted Sanders. The classification of coher- 2007. ence relations and their linguistic markers: An exploration of [106] Ann Macintosh. Moving toward "Intelligent" policy develop- two languages. Journal of Pragmatics, 30:135–175, 1998. ment? IEEE Intelligent Systems, 24(5):79–82, 2011. [92] Kepa Korta and John Perry. Pragmatics. In Edward N. Zalta, [107] Ann Macintosh, Thomas F. Gordon, and Alastair Renton. editor, The Stanford Encyclopedia of Philosophy. Summer Providing argument support for e-participation. Journal of Schneider et al. / A Review of Argumentation for the Social Semantic Web 37

Information Technology & Politics, 6(1):43–59, 2009. [123] Tim O’Reilly. What Is Web 2.0: Design Patterns and Business [108] Clara Mancini and Simon Buckingham shum. Cognitive Co- Models for the Next Generation of Software, 30 September herence Relations and hypertext: from cinematic patterns to 2005. http://www.oreillynet.com/lpt/a/6228. scholarly discourse. In Proceedings of the 12th ACM confer- [124] Bo Pang and Lillian Lee. Opinion mining and sentiment anal- ence on Hypertext and Hypermedia, pages 165–174, Arhus, ysis. Foundations and Trends in Information Retrieval, 2(1- Denmark, 2001. 2):1–135, 2008. [109] Clara Mancini and Simon J. Buckingham Shum. Modelling [125] Jack Park. Boundary infrastructures for IBIS federation: De- discourse in contested domains: A semiotic and cognitive sign rationale, implementation, and evaluation. Ph.D. The- framework. International Journal of Human-Computer Stud- sis Proposal KMI-10-01, The Open University, Milton Keyes, ies, 64(11):1154–1171, November 2006. 2010. [110] William C. Mann and Sandra A. Thompson. Rhetorical Struc- [126] Philippe Pasquier, Iyad Rahwan, Frank Dignum, and Liz ture Theory: Toward a functional theory of text organization. Sonenberg. Argumentation and persuasion in the cogni- Text, 8(3):243–281, 1988. tive coherence theory. In Proceedings of the 2006 confer- [111] Daniel Marcu. Building up rhetorical structure trees. In AAAI- ence on Computational Models of Argument: Proceedings 96 Proceedings, 1996. of COMMA 2006, pages 223–234, Amsterdam, The Nether- [112] Daniel Marcu. The Theory and Practice of Discourse Parsing lands, The Netherlands, 2006. IOS Press. and Summarization. The MIT Press, November 2000. [127] Alexandre Passant. Semantic Web Technologies for Enter- [113] Peter McBurney and Simon Parsons. Dialogue games for prise 2.0. IOS Press, 2010. agent argumentation. Argumentation in Artificial Intelligence, [128] Alexandre Passant, Paolo Ciccarese, John G. Breslin, and Tim pages 261–280, 2009. Clark. SWAN/SIOC: aligning scientific discourse representa- [114] Helena M. Mentis, Paula M. Bach, Blaine Hoffman, tion and social semantics. In Proceedings of the Workshop on Mary Beth Rosson, and John M. Carroll. Development of de- Semantic Web Applications in Scientific Discourse (SWASD cision rationale in complex group decision making. In Pro- 2009), at (ISWC-2009), volume 523. CEUR-WS.org, 2009. ceedings of the 27th International Conference on Human Fac- [129] Chaim Perelman and Lucie Olbrechts-Tyteca. The New tors in Computing Systems, pages 1341–1350, 2009. Rhetoric: A Treatise on Argumentation. University of Notre [115] Raquel Mochales and Marie-Francine F. Moens. Argumen- Dame Press, June 1969. tation mining. Artificial Intelligence and Law, 19(1):1–22, [130] Silvio Peroni and David Shotton. Fabio and cito: Ontologies 2011. for describing bibliographic resources and citations. IEEE [116] Raquel Mochales Palau. Automatic detection and clas- Intelligent Systems, 2011. Preprint. sification of argumentation in a legal case. PhD thesis, [131] H. Sofia Pinto, Christoph Tempich, and Steffen Staab. DILI- K.U.Leuven, Faculty of Engineering Science, July 2011. GENT: towards a fine-grained methodology for DIstributed, Moens, Marie-Francine and De Schreye, Daniel (supervi- loosely-controlled and evolvInG engingeering of oNTologies. sors). In Proceedings of the 16th European Conference on Artificial [117] Raquel Mochales-Palau and Marie-Francine Moens. Auto- Intelligence (ECAI 2004), August 22nd - 27th, pages 393– matic argumentation detection and its role in law and the 397, August 2004. Semantic Web. In Law, Ontologies and the Semantic Web [132] Henry Prakken. Formal systems for persuasion dialogue. The - Channelling the Legal Information Flood, volume 188 of Knowledge Engineering Review, 21(2):163–188, 2006. Frontiers in Artificial Intelligence and Applications, pages [133] ESTRELLA Project. Deliverable 1.4: OWL ontology of basic 115–129. IOS Press, 2009. legal concepts (LKIF-Core). Deliverable IST-2004-027655, [118] Sanjay Modgil and Jarred McGinnis. Towards characterising University of Amsterdam, January 2007. argumentation based dialogue in the Argument Interchange [134] Iyad Rahwan. Mass argumentation and the Semantic Web. Format. In Argumentation in Multi-Agent Systems, volume Web Semantics: Science, Services and Agents on the World 4946. Springer, 2008. Wide Web, 6(1):29–37, February 2008. [119] Bernard Moulin, Hengameh Irandoust, Micheline Bélanger, [135] Iyad Rahwan and Bita Banihashemi. Arguments in OWL: a and Gaëlle Desbordes. Explanation and argumentation capa- progress report. In Computational models of argument: Pro- bilities: Towards the creation of more persuasive agents. Ar- ceedings of COMMA 2008, pages 297–310. IOS Press, 2008. tificial Intelligence Review, 17:169–222, 2002. [136] Iyad Rahwan, Bita Banihashemi, Chris Reed, Douglas Wal- [120] Carlos Ches nevar, Jarred McGinnis, Sanjay Modgil, Iyad ton, and Sherief Abdallah. Representing and classifying ar- Rahwan, Chris Reed, Guillermo Simari, Matthew South, Ger- guments on the semantic web. The Knowledge Engineering ard Vreeswijk, and Steven Willmott. Towards an Argument Review, 26(04):487–511, 12 2011. Interchange Format. The Knowledge Engineering Review, [137] Iyad Rahwan and Guillermo Simari. Argumentation in Artifi- 21(4):293–316, 2006. cial Intelligence. Springer, 2009. [121] Susan E. Newman and Catherine C. Marshall. Pushing [138] Iyad Rahwan, Fouad Zablith, and Chris Reed. Laying the Toulmin too far: Learning from an argument representa- foundations for a World Wide Argument Web. Artificial In- tion scheme. Technical Report Technical Report SSL-92- telligence, 171(10-15):897–921, July 2007. 45, Xerox PARC, 1992. http://www.csdl.tamu.edu/ mar- [139] Iyad Rahwan, Fouad Zablith, and Chris Reed. Towards large shall/toulmin.pdf. scale argumentation support on the Semantic Web. In Pro- [122] Alexandra Okada, Simon J. Buckingham Shum, and Tony ceedings of the 22nd national conference on Artificial Intelli- Sherborne. Knowledge Cartography: Software Tools and gence - Volume 2, pages 1446–1451. AAAI Press, 2007. Mapping Techniques. Springer, 2008. [140] Andrew Ravenscroft, Simon Wells, Musbah Sagar, and Chris Reed. Mapping persuasive dialogue games onto argumen- 38 Schneider et al. / A Review of Argumentation for the Social Semantic Web

tation structures. In Symposium on Persuasive Technology [156] Daniel Schober, James Malone, and Robert Stevens. & Digital Behaviour Intervention at AISB: Artificial Intelli- Practical experiences in concurrent, collaborative ontol- gence and Simulation of Behaviour, April 2009. ogy building using Collaborative Protégé. In Pro- [141] Chris Reed. Representing dialogic argumentation. ceedings of the International Conference on Biomedi- Knowledge-Based Systems, 19(1):22–31, March 2006. cal Ontology, Buffalo, NY, 2009. Nature Precedings. [142] Chris Reed and Glenn Rowe. Araucaria: Software for argu- http://precedings.nature.com/documents/3517/version/2. ment analysis, diagramming and representation. International [157] Wolfgang Schuler and John B. Smith. Author’s argumenta- Journal on Artificial Intelligence Tools, 13(4):961–979, 2004. tion assistant (AAA): a hypertext-based authoring tool for ar- [143] Chris Reed and Christopher W. Tindale. Dialectics, dialogue gumentative texts. In Hypertext: Concepts, systems and appli- and argumentation: An examination of Douglas Walton’s the- cations (Proceedings of the ECHT’90 European Conference ories of reasoning and argument. College Publications, Lon- on Hypertext), pages 137–151. Cambridge University Press, don, 2010. 1992. [144] Chris Reed, Douglas Walton, and Fabrizio Macagno. Argu- [158] John R. Searle. Speech Acts. Cambridge University Press, ment diagramming in logic, law and Artificial Intelligence. New York, 1969. The Knowledge Engineering Review, 22(01):87–109, 2007. [159] Frank M. Shipman and Catherine C. Marshall. Formality con- [145] Chris Reed and Simon Wells. Dialogical argument as an inter- sidered harmful: Experiences, emerging themes, and direc- face to complex debates. IEEE Intelligent Systems, 22(6):60– tions on the use of formal representations in interactive sys- 65, 2007. tems. Computer Supported Cooperative Work, 8(4):333–352, [146] Chris Reed, Simon Wells, Katarzyna Budzynska, and Joseph December 1999. Devereux. Building arguments with argumentation: the role [160] David Shotton. CiTO, the citation typing ontology, and its use of illocutionary force in computation models of argument. for annotation of reference lists and visualization of citation In Computational Models of Argument - Proceedings of networks. In Bio-Ontologies 2009, a Special Interest Group COMMA 2010, number 216 in Frontiers in Artificial Intelli- meeting at ISMB 2009, 2009. gence and Applications, pages 415–426. IOS Press, 2010. [161] David Shotton. CiTO, the citation typing ontology. Journal [147] Chris Reed, Simon Wells, Joseph Devereux, and Glenn Rowe. of Biomedical Semantics, 1(Suppl 1):S6, 2010. AIF+: dialogue in the Argument Interchange Format. In [162] Simon Buckingham Shum. Cohere: Towards web 2.0 argu- Computational models of argument: Proceedings of COMMA mentation. In Computational models of argument: Proceed- 2008, pages 311–323. IOS Press, 2008. ings of COMMA 2008, volume 172 of Frontiers in Artificial [148] Alan Ritter, Colin Cherry, and Bill Dolan. Unsupervised Intelligence and Applications. IOS Press, 2008. modeling of Twitter conversations. In Proceedings of Hu- [163] Simon Buckingham Shum. AIF use case: man Language Technologies: The 11th Annual Conference of Iraq debate. AIF 2.0 meeting contribution, the North American Chapter of the Association for Computa- http://projects.kmi.open.ac.uk/hyperdiscourse/docs/AIF- tional Linguistics, 2010. UseCase-v2.pdf, 2010. [149] Alan Ritter, Doug Downey, Stephen Soderland, and Oren Et- [164] Simon Buckingham Shum and Tamara Sumner. JIME: an zioni. It’s a contradiction—no, it’s not: a case study using interactive journal for interactive media. First Monday, 6(2), functional relations. In EMNLP:Proceedings of the Confer- February 2001. ence on Empirical Methods in Natural Language Process- [165] Simon J. Buckingham Shum. The roots of computer sup- ing, pages 11–20. Association for Computational Linguistics, ported argument visualization. In Visualizing Argumenta- 2008. tion: Software tools for collaborative and educational sense- [150] Glenn W. A. Rowe, Chris A. Reed, and Joel Katzav. Arau- making, pages 3–21. Springer, 2003. caria: Marking up argument. In European Conference on [166] Simon J. Buckingham Shum, Victoria Uren, Gangmin Li, Computing and Philosophy, 2003. Bertrand Sereno, and Clara Mancini. Modeling naturalistic [151] Matthias Samwald and Klaus-Peter Adlassnig. The bio-zen argumentation in research literatures: Representation and in- plus ontology. Applied Ontology, 3(4):213–217, 2008. teraction design issues. International Journal of Intelligent [152] Ted J.M. Sanders, Wilbert P.M. Spooren, and Leo Noordman. Systems, 22(1):17–47, 2007. Towards a taxonomy of coherence relations. Discourse Pro- [167] Evren Sirin, Bijan Parsia, Bernardo Cuenca Grau, Aditya cesses, 15:1–35, 1992. Kalyanpur, and Yarden Katz. Pellet: A practical OWL-DL [153] Oliver Scheuer, Frank Loll, Niels Pinkwart, and Bruce reasoner. Web Semantics: Science, Services and Agents on the McLaren. Computer-supported argumentation: A review of World Wide Web, 5(2):51–53, 2007. the state of the art. International Journal of Computer- [168] Mark Snaith, Joseph Devereux, John Lawrence, and Chris Supported Collaborative Learning, 5(1):43–102, 2010. Reed. Pipelining argumentation technologies. In Computa- [154] David C. Schneider, Christian Voigt, and Gregor Betz. tional Models of Argument - Proceedings of COMMA 2010, ArguNet–a software tool for collaborative argumentation number 216 in Frontiers in Artificial Intelligence and Appli- analysis and research. In Proceedings of the IJCAI Workshop cations. IOS Press, 2010. on Computational Models of Natural Argument (CMNA), [169] Mark Snaith, John Lawrence, and Chris Reed. Mixed ini- 2007. tiative argument in public deliberation. In Proceedings of [155] Jodi Schneider, Alexandre Passant, and Stefan Decker. Dele- the Fourth International Conference on Online Deliberation, tion discussions in Wikipedia: Decision factors and outcomes. 2010. In 8th International Symposium on Wikis and Open Collabo- [170] M. South, G. Vreeswijk, and J. Fox. Dungine: A JAVA Dung ration, WikiSym 2012, 2012. reasoner. In Proceeding of the 2008 conference on Computa- tional Models of Argument (COMMA 2008), pages 360–368. Schneider et al. / A Review of Argumentation for the Social Semantic Web 39

IOS Press, 2008. [188] Marilyn A. Walker, Pranav Ananda, Rob Abbott, Jean E. Fox [171] Norbert A. Streitz, Jörg Hannemann, and Manfred Thüring. Tree, Craig Martell, and Joseph King. That’s your evidence?: From ideas and arguments to hyperdocuments: Travelling Classifying stance in online political debate. Decision Sup- through activity spaces. In Proceedings of the second an- port Systems, in press 2012. nual ACM conference on Hypertext, HYPERTEXT ’89, pages [189] Doug Walton, Katie Atkinson, Trevor Bench-Capon, Adam 343–364, 1989. Wyner, and Daniel Cartwright. Argumentation in the frame- [172] Daniel D. Suthers. Empirical studies of the value of conceptu- work of deliberation dialogue. Arguing Global Governance: ally explicit notations in collaborative learning. In Knowledge Agency, Lifeworld and Shared Reasoning, page 210, 2010. Cartography, Advanced Information and Knowledge Process- [190] Douglas Walton. Plausible argument in everyday conversa- ing, pages 1–24. Springer, 2008. tion. SUNY Press, 1992. [173] Maite Taboada and William C. Mann. Applications of Rhetor- [191] Douglas Walton. Argumentation Schemes for Presumptive ical Structure Theory. Discourse Studies, 8(4):567–588, Au- Reasoning. Routledge, 1995. gust 2006. [192] Douglas Walton. How can logic best be applied to arguments? [174] Christoph Tempich, H. Sofia Pinto, York Sure, and Steffen Logic Journal of the IGPL, 5(4):603–614, July 1997. Staab. An argumentation ontology for DIstributed, loosely- [193] Douglas Walton. Fundamentals of Critical Argumentation. controlled and evolvInG engineering processes of oNTologies Cambridge University Press, October 2005. (DILIGENT). In The Semantic Web: Research and Applica- [194] Douglas Walton. Dialog Theory for Critical Argumentation. tions, pages 241–256. Springer, 2005. John Benjamins, 2007. [175] Simone Teufel and Marc Moens. Summarizing scientific arti- [195] Douglas Walton. Informal Logic: A Pragmatic Approach. cles: Experiments with relevance and rhetorical status. Com- Cambridge University Press, 2nd edition, 2008. putational Linguist., 28(4):409–445, 2002. [196] Douglas Walton. Types of dialogue and burdens of proof. [176] Paul Thagard. Explanatory coherence. Behavioral and Brain In Computational Models of Argument - Proceedings of Sciences, 12(3):435–502, 1989. COMMA 2010, number 216 in Frontiers in Artificial Intelli- [177] Terri Attwood Anita de Waard Tim Clark, Paolo Ciccarese gence and Applications, pages 13–24. IOS Press, 2010. and Steve Pettifer. A round-trip to the annotation store: Open, [197] Douglas Walton, Chris Reed, and Fabrizio Macagno. Argu- transferable semantic annotation of biomedical publications. mentation Schemes. Cambridge University Press, 1st edition, In Beyond the PDF workshop, University of California San August 2008. Diego, 2011. [198] Douglas N. Walton. Commitment in dialogue : basic concepts [178] Francesca Toni and Paolo Torroni. Bottom-up argumentation. of interpersonal reasoning. State University of New York First International Workshop on the Theory and Applications Press, Albany, 1995. of Formal Argumentation, 2011. [199] Barbara Warnick. Rhetoric Online: Persuasion and politics [179] Paolo Torroni, Federico Chesani, and Marco Gavanelli. Ar- on the World Wide Web. Peter Lang, 2007. gumentation in the Semantic Web. IEEE Intelligent Systems, [200] Bonnie Webber, Markus Egg, and Valia Kordoni. Discourse 22(6):66–74, 2007. structure and language technology. Natural Language Engi- [180] Stephen Edelston Toulmin. The Uses of Argument. University neering, pages 1–54, FirstView Article 2011. Press, Cambridge [Eng.], 1958. [201] Simon Wells, Colin Gourlay, and Chris Reed. Argument blog- [181] Douglas P. Twitchell, Mark Adkins, Jay F. Nunamaker Jr., ging. In Proceedings of the IJCAI Workshop on Computa- and Judee K. Burgoon. Using speech act theory to model tional Models of Natural Argument (CMNA), 2009. conversations for automated classification and retrieval. In [202] Simon Wells and Chris Reed. MAgtALO: An agent-based (LAP’04):Proceedings of the 9th International Working Con- system for persuasive online interaction. In AISB 2008: Arti- ference on the Language-Action Perspective on Communica- ficial Intelligence and Simulation of Behaviour, 2008. tion Modelling, pages 121–129, 2004. [203] Simon Wells and Chris Reed. A domain specific language for [182] Manolis Tzagarakis, Nikos Karousos, Nikos Karacapilidis, describing diverse systems of dialogue. Journal of Applied and Dora Nousia. Unleashing argumentation support systems Logic, 2011. to appear. on the Web: The case of CoPe_it! In Web Science 2009, [204] Robin Wendler. The eye of the beholder: Challenges of image Athens, Greece, March 2009. description and access at harvard. In Metadata in Practice, [183] F. H van Eemeren, editor. Advances in Pragma-Dialectics. pages 51–69. ALA Editions, 2004. SicSat/Vale Press, 2002. [205] Terry Winograd. A Language/Action perspective on the de- [184] Frans H. van Eemeren and Rob Grootendorst. Speech Acts sign of cooperative work. Human-Computer Interaction, in Argumentative Discussions: A Theoretical Model for the 3(1):3–30, March 1987. Analysis of Discussions Directed Toward Solving Conflicts of [206] Adam Wyner, Tom Van Engers, and Kiavash Bahreini. From Opinion. de Gruyter Mouton, June 1984. policy-making statements to first-order logic. In Electronic [185] Frans H. van Eemeren and Rob Grootendorst. Argumenta- Government and the Information System Perspective, volume tion, Communication and Fallacies: A Pragma-Dialectical 6267 of LNCS, pages 47–61. Springer, 2010. Perspective. Erlbaum, 1992. [207] Adam Wyner, Raquel Mochales-Palau, Marie-Francine [186] Frans H. van Eemeren and Rob Grootendorst. A systematic Moens, and David Milward. Approaches to text mining ar- theory of argumentation: the pragma-dialectical approach. guments from legal cases. In Semantic Processing of Legal Cambridge University Press, 2004. Texts, number 6036 in Lecture Notes in Computer Science, [187] Frans H. van Eemeren, Rob Grootendorst, and Fran- pages 60–79. Springer, 2010. cisca Snoeck Henkemans. Argumentation: analysis, evalua- [208] Adam Wyner, Jodi Schneider, Katie Atkinson, and Trevor tion, presentation. Erlbaum, 2002. Bench-Capon. Semi-automated argumentative analysis of on- 40 Schneider et al. / A Review of Argumentation for the Social Semantic Web

line product reviews. In Proceedings of the 2012 conference on Computational Models of Argument (COMMA 2012). [209] Adam Wyner and Tom van Engers. A framework for enriched, controlled on-line discussion forums for e- government. In eGOV 2010, pages 1–11, 2010. [210] Ka-Ping Yee. Zest: Discussion mapping for mailing lists. In Proceedings of the 2002 ACM conference on Computer- Supported Cooperative Work, 2002. [211] J. Young, C. Martell, P. Anand, P. Ortiz, and H. T. Gilbert IV. A microtext corpus for persuasion detection in dialog. In Workshops at the Twenty-Fifth AAAI Conference on Artiﬁcial Intelligence, 2011. [212] Fouad Zablith. ArgDF: Arguments on the Seman- tic Web. Master’s, The British University in Dubai Jointly with The University of Edinburgh, February 2007. http://www.argdf.org/source/. Schneider et al. / A Review of Argumentation for the Social Semantic Web i

Appendix of handling reputation: users start with a reputation of 50, which increases or decreases according to the votes A. Tools their points accrue. Votes are weighted: for instance, points with 10 ‘cred’ or ‘crud’ votes change less in A.1. List of social and Semantic Web tools considered response to further votes, and votes on users’ ﬁrst 20 discussion points affect their reputation less than later In this section we draw from our review of thirty- contributions, allowing them to learn the system. seven online argumentation tools: AGORA: Partici- pate - Deliberate, ArgDF, Arguehow, Argument Blog- ging, Argumentum, Argumentations.com, Argunet, Avicenna, bCisiveOnline, Belvedere, Cabanac’s annotation system, Climate CoLab, Cohere, Compet- ing Hypotheses, ConsiderIt, ConvinceMe, CoPe_it!, CreateDebate, Debate.org, Debategraph, Debatepe- dia, Debatewise, DiscourseDB, Dispute Finder, Hyper- news, LivingVote, MAgTALO, Opinion Space, Online Visualisation of Arguments, Parmenides, PDOnline, REASON, Riled Up!, SEAS, Trellis, TruthMapping, and Videolyzer. For further details about our inclusion criteria, see Section 9, page 21.

A.2. AGORA: Participate - Deliberate

Michael Hoffman’s system, AGORA: Participate - Deliberate [78], uses Logical Argument Mapping Fig. 25. Arguhow offers structured discussion. [77](Figure 24 on page ii), providing support for representing deductively valid arguments, using one of seven schemes: modus ponens; modus tollens; disjunc- A.5. Argument Blogging tive syllogism; not-both syllogism; conditional syllogism; equivalence; and constructive dilemma. It relies The idea of argument blogging was proposed by on concept mapping software called CmapTools77. Wells, Gourlay and Reed [201] as a way to bring blogs into the WWAW, based on standard Web technolo- A.3. ArgDF gies, and augmented by argument speciﬁc technologies. In addition to AIF, argument blogging relies on ArgDF78 is a Semantic Web-based argumentation the AIF Database (AIFDB) and Dialog Game Descrip- system using the AIF-RDF ontology described above tion Language (DGDL). AIFDB is a MySQL database [139,138,212]. ArgDF uses Sesame RDF for storage for storing AIF documents which can be serialized as and querying and Phesame for communicating with RDF and accessed via a RESTful Web service. DGDL the Sesame through PHP pages. [140,203] is a grammar for describing the rules of dialogue games. A.4. Arguehow Argument blogging uses text from the current Web as a departure point for the WWAW. When browsing ArgueHow79 (Figure 25) is a argument-based dis- the Web, users select text and click a JavaScript book- cussion board aimed at a general audience. Its purpose marklet, to indicate whether they will attack an infer- is to help ﬁnd the best points supporting a position. ence, support or refute the selected text. This gener- Discussion points are sorted by votes for (‘Creds’) and ates a fragment of embeddable JavaScript the user can against (‘Cruds’) them. ArgueHow has a unique way paste onto his/her blog. Once a blogger opts in to the WWAW by adding JavaScript to a webpage, the page 77http://cmap.ihmc.us/ displays a badge which links back to argument blog- 78http://argdf.org/ ging server, where the distributed dialog can be visual- 79http://arguehow.com/ ized or exported as text. ii Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 24. A sample LAM map, from [78].

Earlier work on semantic blogging predates the WWAW but focused more attention on the visualization of reply graphs of messages from multiple blogs [84] or the possibilities for inference [38].

A.6. Argumentum

Argumentum80 is an argument-based discussion site aimed at airing discussions. Debaters add topics and their arguments are colored to indicate the supporting (green) and opposing (red) arguments (Figure 26). Comments, but not their replies, are similarly colored to indicate agreement or disagreement. Users some- Fig. 26. In Argumentum, users can indicate support for an argument times want to agree or disagree without leaving com- with money. The left-hand color bars indicate the supporting (green) and opposing (red) arguments. ments; currently this leaves a default comment that says “Type the reason why you oppose..." A.7. Argumentations.com Argumentum’s most unique feature is that users can put their “2 cents" in literally: credibility, earned with Argumentations83 serves analysts who want to de- good arguments, is measured in ‘cents’ and can be velop arguments collaboratively. Arguments, which spent to influence a debate result. Users can also con- are classified as either claims or open-ended issues, tribute arguments without starting from the Argumen- can be added or edited; an example is shown in Fig- tum website, using bookmarklets81 or through Gmail ure 27 on page iv. To help suggest topics and build and Facebook82. Further, loggers and publishers can arguments, users can import news stories and extract also contribute using Argumentum buttons or widgets. statements (declarative sentences) from stories. Argumentations offers several unique features. First, arguments–whether claims or open-ended issues–are 80http://arg.umentum.com/ evaluated depending on their type. Claims are evalu- 81http://arg.umentum.com/share 82http://arg.umentum.com/wiki/ more-ways-to-argue 83http://www.argumentations.com/ Schneider et al. / A Review of Argumentation for the Social Semantic Web iii ated with a truth value and confidence. Open-ended issues are evaluated based on Desirability, Importance, Volatility, Likelihood, and Confidence. Second, along with tag clouds, Argumentations uses ‘tag spheres’ (Figure 28). Further, arguments can be opened in Sil- verlight. Finally, they offer some interesting tutorials which display mindmaps84.

(a)

Fig. 28. The global warming ‘tag sphere’ from Argumentations.

A.8. Argunet (b)

Argunet [154] is a desktop tool85 coupled with an Fig. 29. Argunet can show an (a) overview of several related argu- open source federation system for sharing argument ment maps; and (b) in each individual map, nodes can be opened up to show arguments they support, attack, are supported by, and are 86 maps. A public server, Argunet.org , allows authors attacked by. to make maps public or restrict viewing and/or editing to a specified group. Connecting to other servers is A.9. Avicenna also possible; this focus on federation, makes Argunet unique. Rahwan and Banihashemi’s OWL-based argumen- Argunet also has other unique features. Argunet is a tation system Avicenna (Figure 30 on page iv) was multi-lingual environment which records the language demonstrated at COMMA 2008 [135] and recent de- of the map. Maps published at Argunet.org, must be scriptions and screenshots appear in [136]. Extending released under the CC-BY license. An extensive online the work of ArgDF, Avicenna is a Web-based system manual provides instruction, and they promote embed- using Jena[31], ARQ87, and Pellet [167]. Since OWL ding debates. Users also have significant control over supports inference over transitive properties, Avicenna the presentation of arguments, such as colors and de- can support argument chaining, such as retrieving all scriptions of different argument families. Related maps arguments that directly or indirectly support a given can be published in series, as shown in Figure 29(a). conclusion. Avicenna is also used to infer the classi- In the argument map representation, each node can be fication hierarchy of argument schemes: for example, opened up to reveal a matrix listing which other ar- an appeal to expert opinion is a specialization of an guments support, attack, are supported by, and are at- argument from position to know. tacked by the given node (Figure 29(b)). Argunet appears to support incremental formalization since ar- A.10. bCisive Online guments can be quickly sketched or reconstructed as premises and conclusions. bCisive Online88 is an online argument mapping and spatial hypertext environment for real-time collaboration and team problem-solving (Figure 31(a) on 84http://www.argumentations.com/ Argumentations/Help/Tutorials/Tutorials.aspx 85http://www.argunet.org/editor/ 87http://jena.sourceforge.net/ARQ/ 86http://www.argunet.org/debates/ 88http://www.bcisiveonline.com/ iv Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 27. In Argumentions, colored dots indicate the supporting (green) and opposing (red) arguments.

Fig. 30. Avicenna uses Walton’s critical questions and argument schemes [136]. page v). Aimed at the business market and individual viewers to pan, zoom, hide and show parts of the map) decision-makers, bCisive Online is a commercial prod- or exported as PowerPoint. Snapshots can be saved as uct from AusThink, the makers of the Rationale desk- history items, to allow restoring to or reviewing a pre- top tool; the free option allows up to three users to vious map. collaborate, or users can upgrade with a monthly sub- scription fee. bCisive Online is unique in that it is in- A.11. Belvedere tended for real-time use with audio conferencing. One 89 person edits the map at a time, adding nodes and con- Belvedere is open source software for problem- nections between nodes (Figure 31(b) on page v) while based collaborative learning. It provides multiple others can point with their cursor or request editing control. Maps can be embedded in blogs (which allows 89http://belvedere.sourceforge.net/ Schneider et al. / A Review of Argumentation for the Social Semantic Web v

(a) (b)

Fig. 31. (a) Collaborative maps for bCisive Online can be used for decision-making and requirements analysis. (b) bCisive Online’s node types show the kinds of discussions that it facilitates. views, such as tables, graphs, and argument maps, of A.12. Cabanac’s annotation system the same topic (Figure A.11). It has been extensively Cabanac used a Java-based system90 to research so- investigated in studies of computer-supported collabo- cial validation of the arguments in comments [28]. rative learning [172]. Users did not contribute new content to an ongoing public debate, but analyzed the argumentative status of document comments. Uniquely, sliders were used to indicate the extent to which items were refuted, neutral, or conﬁrmed (Figure 33 on page vi). In effect, users were asked to synthesize the discussion. Aggregated information was not viewed by the users, but held by the experimenter. However, in principle, this approach could be used to promote collaborative sensemaking not just of annotations but also of debate.

A.13. Climate CoLab

The Climate CoLab91 is a deliberation platform under development at MIT, building on former projects such as the Deliberatorium and the ClimateCollabato- rium [88,89,71]. The community runs an annual con- test to gather proposals for mitigating global warming from the general public; once proposals are filtered by experts, everyone is invited to discuss the finalists. Users deliberate in the Positions tab, which facilitates constructing an argument map, voting, and commenting on each of five key topics. Moderators are ex- Fig. 32. Belvedere has both argument maps and tables to help orga- pected to review comments and add new ideas to the nize evidence in collaborative learning. argument map; users can also add Pros, Cons, and Is-

90http://www.irit.fr/~Guillaume.Cabanac/ expe/ 91http://climatecolab.org/ vi Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 33. Cabanac had users flag items (refuted, neutral, confirmed) and indicate their types (question, modification, example). sues directly to an argument map. The Climate CoLab A.15. Competing Hypotheses is unique for integrating argument maps into a larger debate, and for its moderator support, which allows Competing Hypotheses94 is open source analysis users to benefit from argument maps without necessar- software based on the CIA methodology “Analysis ily needing to understand how to edit them. of Competing Hypotheses" (ACH). The software supports breaking down information into hypotheses, evi- A.14. Cohere dence, and analysis, which are entered into a matrix as shown in Figure 36(a) on page vii. The matrix can help Cohere is open source software for sensemaking visually indicate the most likely and least likely scenar- 95 which integrates annotation and argumentation for the ios. Multiple analyses can be combined to provide general public [162,101]. At the Cohere website92, a group view (Figure 36(b) on page vii), or compared users can view and create maps, or import them from pairwise. Competing Hypotheses has persistent chat the Compendium desktop software. Maps consist of (essentially a comment thread) for the entire project as ideas, which users can add directly on the site (Fig- well as message boards for each hypothesis, evidence ure 35), draw from Cohere’s global pool of public item, and evidence-hypothesis pair. We excluded earlier ACH implementations such as PARC ACH96. Un- ideas, or clip via a Firefox plugin while browsing. like these systems, Competing Hypotheses has a testing server97 which allows online collaboration. It is unique for its visualization structure and its use of both individual and group information.

A.16. ConsiderIt

ConsiderIt98 [94] is a new open source deliberation platform under development at the University of Wash- ington. It powers the Living Voters’ Guide99, a deliberation and voter-information platform for Washington State voters. What is unique is the possibility to drill down to understand other voters’ perspectives. In addition to see- Fig. 35. Adding an idea to Cohere. ing pros and cons on an issue from all voters, regard-

Cohere is unique for its integration with the Com- 94http://competinghypotheses.org/ 95 pendium desktop software, its incorporation of social More sophisticated ACH-based software uses matrices as input to Bayesian probabilistic reasoning. bookmarking, and the ability to mark information as 96http://www2.parc.com/istl/projects/ach/ private, public, or shared with a group. Cohere also of- ach.html fers an API93. 97http://groups.google.com/group/ach-users/ browse_thread/thread/d87a5ec4df8be6c0 98http://www.livingvotersguide.org/ 92http://cohere.open.ac.uk/ considerit 93http://cohere.open.ac.uk/help/code-doc/ 99http://www.livingvotersguide.org/ Schneider et al. / A Review of Argumentation for the Social Semantic Web vii

(a)

(b)

Fig. 34. At Climate CoLab, (a) the positions tab shows an argument map which users can edit or comment on. (b) argument maps are introduced with contextual help.

(a) (b)

Fig. 36. In Competing Hypotheses, (a) each individual’s analysis is represented in a consistency matrix; (b) multiple analyses can be combined to create a group matrix. In the group view, darker shades of purple indicate more disagreement. viii Schneider et al. / A Review of Argumentation for the Social Semantic Web

(a) (b)

Fig. 37. The Living Voters’ Guide compiles pro and con lists on each issue. They give (a) an overview of what all voters think about the issue; as well as (b) the key points for undecided voters. less of their stance, (Figure 37(a)), the Living Voters’ Guide can show the key points for a particular group of voters (Figure 37(b)), such as those undecided on the issue or strongly supporting it. This can help users understand what makes an issue controversial. Users indicate how they feel about an issue before and after reading an argument (deliberative polling), which could also be used to ﬁnd the most convincing arguments.

A.17. ConvinceMe

ConvinceMe100 is a competitive debating environment which uses a point scheme and user rankings to motivate contributions to several types of debates. In the King of the Hill game, the most popular choice (and the debater who suggested it) wins. Battles are one- on-one debates between two users, who add arguments Fig. 38. In ConvinceMe’s Open Debates, users can vote for an argu- and evidence in hopes of getting readers’ votes; the ment that convinced them debate ends when one side gets a pre-agreed number of votes. Open debates (Figure A.17) are ongoing and government domains. Users can form groups to share accept pro or con arguments from any registered user, maps, but communicate only through email on the site. as well as rebuttals to existing arguments; users con- Maps can be imported from Compendium, and entire vinced by an argument vote for it. These various types discussions from external webforums in phpNuke for- of debate games make ConvinceMe unique. mat can be imported using a URL. One unique aspect of in CoPe_it! is its approach A.18. CoPe_it! to incremental formalization. CoPe_it! transforms the CoPe_it!101 [182] is a spatial hypertext environ- user-created informal spatial hypertext view (Fig- ment for collaboration, aimed at the learning and e- ure 39(a) on page ix) into an issue chart Figure 39(b) on page ix according to rules shown in Figure 39(c) on 100http://www.convinceme.net/ page ix. Users can also customize the transformation 101http://copeit.cti.gr/ rules. Schneider et al. / A Review of Argumentation for the Social Semantic Web ix

(a) (b) (c)

Fig. 39. CoPe_it! has (a) an informal spatial hypertext view; and (b) a formalized view, created by (c) automatically transforming items.

A.19. CreateDebate

CreateDebate102 is a social debate community, aimed at the general public as well as primary and sec- ondary school classes103. The highest-rated arguments are shown at the top, based on user votes (and ignoring the down votes), which are also used to determine a point score for the user. They offer bookmarklets and promote JavaScript buttons to webmasters104. Some unique features are that the debate moderator can add a ‘Topic Research’ section with RSS feeds from other sites, and that, in addition to pro/con debates, Cre- ateDebate has Perspective debates, which generally have more than two sides, are scored based on user- Fig. 40. At CreateDebate, users add and comment on pro and con arguments. applied tags. A wordcloud and various statistics (Fig- ure 40), including the language grade level, average questions: (1) Agreed with before the debate: (worth word lengths, and vocabulary overlap are calculated 0 points) (2) Agreed with after the debate: (worth 0 for each debate. points) (3) Who had better conduct: (worth 1 point) (4) Had better spelling and grammar: (worth 1 point) A.20. Debate.org (5) Made more convincing arguments: (worth 3 points)

105 (6) Used the most reliable sources: (worth 2 points) Debate.org is a social networking site for debate . Points are awarded, with the most importance given lovers. Debates take place between two members and to using reliable sources and making convincing argu- have four cycles: the challenge period, debating pe- ments. riod, voting period, and post voting period. The de- Another unique feature is Debate.org’s focus on user bating period consists of 1-5 time-limited rounds in proﬁles, where various user details are displayed in- which debaters post arguments. While comments can cluding information such as income, location, ideol- be added at any time, votes are only accepted dur- ogy, gender, president, religion, and who they are in- ing the voting period. Voting involves choosing one of terested in and looking for. These can be used to search the debators (or ‘tied’) for each of the following six for people with particular proﬁle attributes, and aggregate user demographics106 are also available. De- 102http://www.createdebate.com/ bate.org also determines the percentage to which other 103http://www.createdebate.com/about/sites/ members agree with you on “the big issues" (cultural, school 104http://www.createdebate.com/share/buttons 105http://debate.org/ 106http://www.debate.org/about/demographics/ x Schneider et al. / A Review of Argumentation for the Social Semantic Web religious, and political hot topics). Individual members A.23. Debatewise are also ranked by their percentile, based on the outcomes of previous debates. On Debatewise110, everyone can collaborate in creating the strongest case both for and against a given issue. As part of a partnership with the International Debate Education Association (iDebate), they provide links to Debatepedia and iDebate’s reference site De- batabase111. Karma, teams, and lists of recent participants and new editors help motivate participation.

Fig. 41. Debate.org is a social networking site promoting debate.

A.21. Debategraph

Debategraph107 [106] is a wiki debate visualization tool which has been adopted for use at the Kyoto climate change summit and is being tested by EU projects Fig. 43. Debatewise offers an executive summary, followed by a de- such as WAVE108. Debategraph offers several visual- tailed pro/con debate. izations, including the Debate Explorer view shown in Figure 42(a) on page xi and a text-based outline shown There are several unique features. The site makes in Figure 42(b) on page xi. Visualizations can be em- it easy to get involved by providing suggestions of 5- bedded in other websites, and Debategraph encourages minute, 20-minute and 1-hour tasks and showing “7 users to add links to related webpages within graphs. things you should have an opinion on" in rotating images on the homepage. Edit histories are available for A.22. Debatepedia each pro and con point. Debates are structured as ad- judicated debates between two teams; other users can Debatepedia109 bills itself as the “the Wikipedia of make comments, vote, and subscribe to debates. pros and cons". Sponsored by the International Debate Education Association, Debatepedia is a collaborative A.24. Discourse DB community effort to summarize arguments. Each argument page provides an overview, then a list of is- DiscourseDB112 is used to collaboratively collect sues, with pros and cons supported by news articles policy-related commentary. Opinion pieces (Figure 44(a) and similar sources. It provides an intuitive editing en- on page xi) are collected from notable sources, news- vironment, where users can edit just the relevant sec- papers and websites with at least 50,000 circula- tion, such as the pro or con for a topic. Debatepedia tion/unique visitors per month. Users categorize these is unique for providing an easily-editable wiki of pros opinion pieces, selecting a quote, indicating the topic and cons. 110http://debatewise.org/ 107http://debategraph.org/ 111http://www.idebate.org/debatabase/intro. 108http://www.wave-project.eu/ php 109http://debatepedia.idebate.org/ 112http://discoursedb.org/ Schneider et al. / A Review of Argumentation for the Social Semantic Web xi

(a) (b)

Fig. 42. Debategraph for CNN’s Amanpour TV shown in (a) Debate Explore view; (b) text view.

(a) (b)

Fig. 44. In DiscourseDB, (a) users catalog opinion pieces; (b) this generates an overview of the positions for, against, and mixed on a topic. and position, along with whether the author’s argument When a topic has multiple positions (e.g. Darfur114), is for, against, or mixed on the position. DiscourseDB is especially helpful in summarizing the DiscourseDB uses Semantic MediaWiki [95] with discussion. the SemanticForms113 extension. This makes it possible to list all commentary written by particular person, A.25. Dispute Finder published in a particular venue, and so forth. 115 Further, since items indicate the position they take Dispute Finder [53,54] is a browser extension on a topic, DiscourseDB can list all commentary for that alerts users when information they read is dis- or against a given position as shown in Figure 44(b). puted, based on a database of disputed claims. This

114http://discoursedb.org/wiki/Darfur_ 113http://www.mediawiki.org/wiki/Extension: conflict Semantic_Forms 115http://ennals.org/rob/disputefinder.html xii Schneider et al. / A Review of Argumentation for the Social Semantic Web database was created by asking activists (who are in- U.S. Department of State is using Opinion Space119 to terested in informing or convincing others) to indi- aggregate opinions about foreign policy and create a cate disputed claims manually, and then extended algo- “virtual town hall" as shown in Figure 47. rithmically. While the Dispute Finder plugin remains available116, it notes that the project has ended; unfortunately, the plugin no longer highlights phrases such as the “abortion reduces crime" phrase used in paper examples.

A.26. Hypernews

Hypernews117 [21] is a general purpose Web forum, inspired by Usenet news. Its use of message types distinguishes HyperNews from other forums. Users are asked to indicate what kind of message they are posting (None, Question, Note, Warning, Feedback, Idea, More, News, Ok, Sad, Angry, Agree, Disagree) as shown in Figure 45(a) on page xiii; the message type Fig. 47. Opinion Space maps comments in a constellation view. is then displayed as an icon in the forum’s thread view (Figure 45(b) on page xiii). Opinion Space is unique in its use of deliberative polling and visualization. With deliberative polling, A.27. LivingVote participants are polled both before and after deliberation, to better understand how public opinion can At Living Vote118, the general public can discuss change based on increased understanding of the issues. pro and con arguments of issues, creating argument Users move sliders to express their opinions on ﬁve is- maps, as shown in Figure 46 on page xiii. A tree view sues. The system then maps the user’s opinion, using provides a coherent view of the argument, which can principal component analysis, to show the user where be drilled down, where arguments and their counter- they stand. Each point in the visualization represents a arguments are presented side-by-side. Users can add perspective; larger points represent more popular per- arguments, and voting colors the nodes according to spectives. Users can also view and rate others’ com- whether you agree (green), disagree (red), or haven’t ments (Figure 48). Ratings can be used to choose the voted (white). most informative comments for display. Living Vote is unique in the way that it handles and uses votes. To vote, users must answer questions designed to test whether they’ve read the arguments. Liv- ing Vote also prunes unhelpful arguments and aims to provide a “complete, persistent, constantly changing and up-to-date record" of everyone’s opinions and the most convincing arguments.

A.28. Opinion Space

Opinion Space is software developed by UC Berke- ley’s Center for New Media “designed to collect and visualize user opinions" on a variety of topics [55]. The Fig. 48. Opinion Space uses sliders to collect and display users’ 116http://addons.mozilla.org/en-US/firefox/ opinions on ﬁve issues. addon/11712/ 117http://www.hypernews.org/HyperNews/get/ hypernews/reading.html 118http://www.LivingVote.org/ 119http://www.state.gov/opinionspace/ Schneider et al. / A Review of Argumentation for the Social Semantic Web xiii

(a) (b)

Fig. 45. (a) Users are asked to specify their message type, using this Hypernews taxonomy; (b) Part of a Hypernews discussion thread.

Fig. 46. At Living Vote, the weight given to a user’s votes increases as they read and vote on more arguments.

A.29. Online Visualisation of Arguments (OVA) oriented argumentation tools and calculation-based agent argumentation. Mixed initiative discussions are Online Visualisation of Arguments120 (OVA) is an enabled by the argument maps created by OVA or online argument analysis and mapping environment any other AIF-based tool. Thus, instead of represent- [169] which exports AIF. In OVA, web pages can be ing one’s point of view countless times in a forum or displayed adjacent to an argument mapping canvas, FAQ, it would be possible to delegate these conversa- helping analysts create a graphical representation of tions to a machine agent using an underlying argument the arguments in online forums or news stories. The re- map, as prototypes like MAgtALO121 [145,202] and sulting argument maps can show the relationships be- the Google Wave discussion bot Arvina [169] show. tween premises (supporting or attacking) as well as the participants responsible for each point of view. In ad- A.30. Parmenides dition to AIF, users can export JPEG and SVG images of the argument. Parmenides122 [32,33,34] is a structured survey tool OVA is part of a pipeline of argumentation tools for gathering public opinion on a proposal. Based on [168] which starts to bridge the gap between human-

121http://www.arg.dundee.ac.uk/?page_id=61 120http://ova.computing.dundee.ac.uk 122http://cgi.csc.liv.ac.uk/~parmenides/ xiv Schneider et al. / A Review of Argumentation for the Social Semantic Web argument schemes and critical questions from argumentation theory, Parmenides can pinpoint the source of the disagreement, by having participants respond to a series of questions. In a Parmenides debates, participants are ﬁrst asked to agree or disagree with a position on a question such as “Should laptops be banned in lecture theatres?" (Figure 49(a) on page xv). Those who disagree are stepped through several screens (such as Figure 49(b) on page xv) of yes/no questions to determine the source of the disagreement. Limited free text boxes allow users to add further information. At the end of the survey, users are offered the choice of submitting an alternative proposal, and are shown the answers they chose. Administrators can then analyze the group’s responses, which are displayed in graphical argumentation frameworks [51]. A greater understanding of the most popular reasons for disagreement could support further discussion and debate about the key issues.

A.31. PDOnline

SWAN/SIOC is itself used in PDOnline123, an on- Fig. 50. Part of an argumentative discussion at PDOnline line community for scientists, funders, and medical professionals working in Parkinson’s disease science, ing the decision-making process; further, in an adap- which is funded by the Michael J. Fox Foundation tive version of REASON, aggregate weights express- [48]. ing the group’s view of each alternative are displayed. Figure 50 shows a PDOnline discussion about a Uniquely, arguments start as threaded discussions in recently-published paper and indicates how the topic REASON, and are colored based on whether they fits into the “PD Guide" taxonomy of research and communication topics. The discussion links both for- agree (blue) or disagree (yellow) with their parent in ward to responses and related contributions and back the thread. to a thread on Papers of the Week (itself contained within a Research Question board). Members’ full A.33. Riled Up! names, credentials, and institutional affiliations are listed, with links to user profiles and institutions. Mem- Riled Up!124 (Figure 51(a) on page xv) is a debate- bers’ profiles link to their publications, and throughout centered site which motivates participation with a the site explicit references to the literature are given. It point-based authority system. Aimed at people who is unique in that it uses scientific argumentation. enjoy debate, Riled Up!’s tagline is “Like Raising Cain? So Do We." Users can add debates, arguments, A.32. REASON and comments, and vote for others’ arguments, as well as add friends and enemies. REASON —Rapid Evidence Aggregation Support- Riled Up! is unique in its comment system–users ing Optimal Negotiation [80,81] — is a Java applet for can respond with positive (green), neutral (grey), or group deliberation, used to arrive at a consensus de- negative (red) comments, as shown in Figure 51(b) on cision. Drawing from decision theory, group-decision support systems, and argumentation, REASON is in- page xv. In addition to a standard layout, a contributor tended to improve information pooling. An argument view gives an overview of the arguments but not the map is used to organize group evidence shared dur- comments.

123http://www.pdonlineresearch.org/ 124http://riledup.com/ Schneider et al. / A Review of Argumentation for the Social Semantic Web xv

(a) (b)

Fig. 49. (a) In Parmenides, participants are asked to agree or disagree with a starting position. (b) Next Parmenides steps participants through a series of yes/no questions to pinpoint the source of their disagreement.

(a) (b)

Fig. 51. RiledUp (a) debates allow structured discussion on a topic; and (b) readers can respond with positive (green), neutral (grey), or negative (red) comments.

A.34. SEAS schemes, allow non-experts to make sound reasoning. SEAS automatically answers some questions based on SRI International’s SEAS125 [105,104] is a template- based structured argumentation tool originally de- earlier responses. The developers report that a threat- signed for collaborative intelligence analysis. It has since been tested in other domains such as by IRS assessment template originally developed by U.S. in- tax auditors and in a simulated public health emer- telligence analysts was successfully applied by non- gency. SEAS’s most unique feature is its emphasis on templating; users can author templates which provide experts in their laboratory. SEAS visualization fea- transferrable notions of how to make an argument, and tures are also considerable: to visualize multiple di- specify authorized coeditors. Figure 52 on page xvi shows one question from such a template. These tem- mensions, SEAS uses starburst, constellation, and ta- plates, which are in essence domain-speciﬁc argument ble views. SRI International runs a SEAS server with

125http://www.ai.sri.com/~seas/ paid accounts and SEAS server software is available. xvi Schneider et al. / A Review of Argumentation for the Social Semantic Web

added rich relationships such as is elaborated by, is supported by, is summarized by, and stands though contradicted by, which the system stored in XML, RDF, and DAML+OIL. In contrast to the detailed argumentation of Rich Trellis, Tree Trellis uses only pro and con, and collaborative discussion is supported, while Table Trellis allows feature and value pairs to be arranged in a matrix, allowing the user to compare and evaluate alter- natives according to their own criteria.

A.36. TruthMapping

TruthMapping127 is an online deliberation tool which seeks to structure the conversation to focus around the “aha!" moment, avoiding digressions and soapboxes, and making hidden assumptions explicit. TruthMap facilitates structured conversations which use argument maps, critiques and rebuttals (Figure 53(a) on page xvii). Users can vote on and rate topics, and watch Fig. 52. SEAS uses a series of questions to structure the argument particular conversations Only one user, the original ar- [105]. guer, modifies the map; feedback comes in critiques attached to each premise and conclusion (Figure 53(b) A.35. Trellis software on page xvii), which can be rebutted. One unique aspect of TruthMapping is that users can continually The argument analysis system Trellis126 [61,40,41] modify each contribution, but can only post one cri- was built on Semantic Web technologies, including the tique on each node. This is designed to make it easier Semantic Annotation Vocabulary Section 7.7, page 20. to contribute a persistent comment to the discussion, Trellis, inspired by intelligence analysis, began as a which can not be drowned out by a single opponent. credibility and analysis system to help structure de- The system indicates when comments are out of sync, cisions, for example to construct a family geneology and a wiki-style history is available. Another unique based on contradictory information [61]. aspect is the use of votes to color the map: as shown in Originally, Trellis was designed to help capture ar- Figure 53(a) on page xvii, each node is colored based gumentation, grounded in documents, whose reliabil- on the percentage of votes agreeing (green) and dis- ity the user rated, and from which the user extracted agreeing (red). statements; although users did not work directly with A.37. Videolyzer the underlying ontology, arguments could be exported into XML, RDF, DAML, and OWL. In addition to Videolyzer128 [50] allows the general public to have the original version, now called Rich Trells, two other sensemaking and argumentative discussions about the modes, Tree and Table Trellis, described in [41], are quality of online videos. It builds on gamelike-creation now supported, for incremental formalization. of video transcripts and on machine tagging of areas In Rich Trellis, statements are given likelihood- of interest in either the transcript (claim verbs, peo- qualifiers such ‘surprise’ (indicating the analyst’s sub- ple, money, and comparison) or the video itself (faces) jective reaction); reliability-qualifiers such as ‘com- (Figure 54(a) on page xvii), to provide an integrated pletely reliable’; and credibility-qualifiers such as discussion forum for annotating and challenging the ‘possibly true’. Statements may also be associated claims a video makes (Figure 54(b) on page xvii). with a document providing evidence. The source for Videolyzer is unique in its focus on integrating argu- each document, including creator, publisher, date, and mentative discussion into a video platform. format, is recorded. Originally, in Rich Trellis, users

127http://www.truthmapping.com/ 126http://www.isi.edu/ikcap/trellis/ 128http://videolyzer.com/ Schneider et al. / A Review of Argumentation for the Social Semantic Web xvii

(a) (b)

Fig. 53. Truthmapping (a) allows users to construct an argument by laying out premises and conclusions. Each node is colored based on the percentage of agreement (green) and disagreement (red). (b) Each premise and conclusion can be critiqued in comments, and critiques can be responded to with rebuttals.

(a) (b)

Fig. 54. Videolyzer (a) allows users to comment on the points made in a video; and (b) algorithmically determines segments of possible interest to help focus the discussion: in the transcript these are claim verbs and comparisons as well as mentions of people and money, and in the video these are peoples’ faces.

B. Matrix Comparison of Tools both tables, we use ‘?’ to indicate that we could not locate a piece of information. First, we record the intended purpose of the tool. We now present comparison charts of the tools re- Next we provide the representation style and func- viewed. Figure 55 on page xxi shows an overall com- tional type. As introduced in Section 9.3, page 23, rep- parison, in which tools are compared according to var- resentation style is drawn from linear, threaded, graph, ious features, which we outline shortly. For the down- container, and matrix (including combinations of these loadable tools, Figure 56 on page xxii provides the li- styles); functional type is drawn from issue network- cense, programming language(s) and data storage. In ing, funnelling, and reputation. Then we indicate what xviii Schneider et al. / A Review of Argumentation for the Social Semantic Web

(a)

(b) Schneider et al. / A Review of Argumentation for the Social Semantic Web xix

(c)

(d) xx Schneider et al. / A Review of Argumentation for the Social Semantic Web

(e)

(f) Schneider et al. / A Review of Argumentation for the Social Semantic Web xxi

(g)

Fig. 55. Overall Comparison of Tools. sort of advanced visualization is offered; ‘-’ indicates We also indicate, in the tags row, whether users that no examples were found (i.e. that the question can provide tags for content. We also indicate which does not apply). The perspective row records whether tools have a bookmarklet for use while browsing, and an individual user has a personal perspective distinct which promote embedding on external sites. The re- from the group view. Next we consider whether a tool maining rows describe features related to the site’s in- has a distributed architecture (allowing multiple copies teraction style, starting with whether it is possible to to synch with one another). attach media in discussions and the input type (such as Then we distinguish downloadable and hosted sys- point and click visual controls or form-based editing). tems (noting that some tools are in both categories or We also indicate which have consistency checking (i.e. use a combined method). To understand their current avoiding obvious contradictions) and credibility met- integration with the Social Web, we record whether rics (usually, but not always, voting) as well as export they use a site-specific login, or allow external creden- capabilities. Tools which export AIF can take advan- tials (such as OpenID, Twitter, or Facebook). tage of an existing infrastructure. We further indicate whether they have any integra- Overall, we can make certain observations regarding tion with third party services; a single row does not these tools: generally they focus either on encourag- do justice to the wide range of integration we found. ing discussion or having a basis in rigorous argumen- For tools with social networking capabilities, we pro- tation models. Significant amounts of innovation has vide an example of the interaction users can have with occurred in the research community, but many ideas each other, or the information they can find out about have not been propagated to the Social Web at large. each other. Stable URLs indicates our success in find- There are certain common mechanisms among many ing reusable bookmarks: in fact these URLs can be systems–basic features such as upvoting, segregating at multiple granularities, such as the entire argument pro and cons, etc. Social Web systems do not have even map, issue, or conversation; each individual comment levels of adoption: some tools are very well-adopted or critique; etc. while others are not. xxii Schneider et al. / A Review of Argumentation for the Social Semantic Web

Fig. 56. Downloadable tools: License, language, and data storage.