1In a 1994 article, Moeschler mischievously pointed out that the connector approach to argumentation, enlightening as it may be to account for “micro” sequences, would struggle to increase the intelligibility of larger argumentative movements: “Asking connectors to function at the macro-structural level”, he wrote, “would be like asking a toothbrush to clean a kitchen efficiently” (1994: 102 ; our translation). This is because argumentation structures the discourses in which it is deployed at different scales. Beyond this question of scale, argumentation can be considered at different levels, notably as a principle structuring relations between propositional contents, or as a (macro)act of language produced in the context of interactions structured by differences of opinion. The aim of this article is to avoid the temptation to discard toothbrushes because they can't clean kitchens – in other words, to pillory devices for visualizing argumentation on the grounds that none of them is able to account for what happens at all scales of argumentative facts, or to capture all levels simultaneously.
2A comparative analysis of five argumentation visualization software packages (MindMup, Rationale, bCisive, OVA and MindManager) will help us to understand the complexity of argumentation. The first step will be to identify the main elements common to most theories of argumentation. This will be based centrally on the triple distinction established by O'Keefe (1982) between argument₁ (propositional content), making-an-argument (the speech act) and argument₂ (argumentative interaction). This distinction will enable us to examine how each software treats (or neglects) these different dimensions of argumentation. Using examples from the play Twelve Angry Men, we’ll illustrate how these software programs generate visual representations of argumentative sequences; we’ll also offer a summary and recall the main argumentative lines that structure the play. Finally, we’ll look at the ontologies of argumentation that underpin the five argument visualization devices considered to show the elements and relationships between elements identified by the different theories of argumentation that they enable to represent, and those they struggle to consider.
3The term “ontology” will be understood here in the sense used in computer science and information science, i.e. in the sense of a data model made up of objects and relationships between these objects. The ontologies mobilized by different approaches to argumentation will be examined; to explore the correspondences, reductions and possible transformations they undergo in visualization software systems designed to translate the structure of argumentation.
4We must first learn from the very important distinctions made by O’Keefe (1977, 1982). The opposition between argument1 and argument2, proposed in O’Keefe (1977), has been retained above all – and sometimes assimilated to the opposition between ‘“argument-as-a-product’“ and ‘“argument-as-a-process’“ (e.g. Reed & Walton 2003); but it is rather the triple distinction between argument1, making-an-argument and argument2 (O’Keefe 1982) that will interest us here.
- 1 This clarification is important: neither “argument” nor “argumentation” in French (nor the verb “ar (...)
5O’Keefe (1982) argues that a number of what he calls “basic” distinctions need to be introduced to give students a better grasp of the phenomena targeted by argumentation studies, and a better understanding of the different perspectives that exist in this field of research. He calls these distinctions “basic” because he bases them in part on the common meanings of the word argument in English1 in everyday speech and suggests that we think in terms of what he calls paradigmatic cases illustrating these different meanings – examples that would be offered to a non-native speaker of English to help them understand what the different meanings of the word prototypically refer to.
- 2 Forgive us for recalling this distinction between argument1 and argument2, which is widely known in (...)
- 3 The pagination shown here corresponds to the republication of the 1977 article in Benoit, Hample & (...)
6In his 1977 article “Two Concepts of Argument”2, he notes that the word argument can refer to two distinct phenomena. He proposes to speak of argument1 when the word refers to a particular type of utterance or communicative act (“a kind of utterance or a sort of communicative act,” 1977: 793), among other communicative acts such as promise, injunction, warning, invitation, etc. This is the meaning intended in sentences such as “He made an argument”, or by the expression “he argued that...”. On the other hand, a sentence like “they had an argument”, or the expression “they argued about...“ refers to argument2, i.e. to a type of interaction involving (at least) two individuals (other types of interaction include the interview, the everyday conversation or the commercial transaction).
- 4 O’Keefe refuses to use the verb arguing to designate this speech act, because of the ambiguity of t (...)
7In 1982, O’Keefe returned to his definition of argument1 as a communicative act. He draws a parallel with other speech acts. A promise, he reminds us, is not a speech act; it’s the act of promising that is (and the promise is “what is conveyed in that speech act”; 1982:12). Similarly, an argument1 is not, he now considers, a speech act: it’s making an argument14 that is (and in performing this speech act, we convey an argument1). Making an argument1 is an act of language performed by an utterance in a communicative situation and presenting a specific linguistic form. A paradigmatic case of making an argument1 “involves the communication of both (1) a linguistically explicable claim and (2) one or more overtly expressed reasons which are linguistically explicit” (O’Keefe 1982: 14). The argument1 constitutes a kind of abstraction built on the basis of the language act performed by making an argument1 and can be analyzed without regard to the particular language form that carried it in a given communicative situation. Let us reproduce here the methodological consequences that O’Keefe draws from this distinction, consequences that will be crucial for the rest of our paper:
I can describe the argument1 that someone made apart from the utterances of the speaker in the act of making the argument1; I can, that is, abstract the argument1 from its communicative vehicle and describe it. Obviously, the speaker’s utterances are related to the argument1 that is made: but in describing the argument1 I am not describing the speaker’s utterances per se but rather the import of those utterances. Given several acts of argument-making, I can abstract the several arguments1 and compare them apart from the acts of argument-making in which they were conveyed; I might, for example, have some notion of the structural elements of arguments1 (perhaps claim, warrant, qualifier, etc.?), and thus could compare arguments1 with respect to those various structural elements. I can, of course, also describe acts of argument-making (to see, perhaps, the various ways in which the act of argument-making is accomplished). (O’Keefe 1982: 18-19)
8Interactions corresponding to argument2 are interactions “in which extended overt disagreement between the interactants occurs” (O’Keefe 1982: 9). He goes on to say that an argument2 (although not a paradigmatic one) is an interaction in which no making-an-argument speech act is produced, and from which the analyst cannot reconstruct any argument1.
9This threefold distinction is of major importance in that it allows us to question, for each theory of argumentation, the phenomenon it seeks to shed light on first and foremost (argument1, making an argument, argument2), the categories it sets up to do so, and eventually, the way in which it conceives the articulation of these phenomena and the concepts used to account for them.
10O’Keefe (1982: 19) points out that argument visualization devices (argument diagrams) are at grips with argument1; for this reason, we shall now focus on the ontology specific to this phenomenon. We’ll come back later to the capacity of the devices examined to restore something of the language act of making an argument1 or arguments2.
11It’s difficult to decide on a term to designate the basic units on which theories of argumentation work when they seek to account for arguments1, so much so that most of the terms considered (statement? sentence? proposition?) entail subsequent theoretical choices and are inseparable from complex conceptual constructs and incommensurable. At this point, we’ll refer to them as “bricks.” This metaphorical term refers to units carrying propositional content and corresponding, at the level of the statement(s) performing the language act of making an argument, to portions of discourse of variable size and structure.
12A brick can enter into an argumentative relationship with another brick. This relationship can be of two types:
-
SUPPORT (one brick is presented as making the other brick more acceptable). When bricks enter into such a relationship, the brick whose acceptability is increased is called “conclusion”; the brick contributing to the increase in acceptability is called “argument“.
-
CONFLICT (the brick is presented as making the other brick less acceptable). When bricks enter into such a relationship, the brick whose acceptability is attacked is called the “conclusion”; the brick contributing to the degradation of its acceptability is called the “counter-argument“.
13A conclusion brick may be supported by a single line of support, or by several independent lines of support (or contestation): we may want to buy a dress [conclusion] because it’s pretty [argument 1], but also because it’s cheap [argument 2]. When the same conclusion is defended by several independent argumentative lines, we speak of convergent argumentation (Govier 2010: 38) or multiple argumentation (Snoeck-Henkemans 1997: 15, 25); the conclusion is supported by free premises.
14It also happens that two (or more) bricks form a block-of-bricks, which can enter into an argumentative relationship (of SUPPORT or CONFLICT) with a brick-conclusion (but each of the bricks in the block, taken independently, can’t). A telling, if spectacularly artificial, example of such argumentative blocks is that of syllogisms containing a disjunctive statement:
Pierre lives in Paris or Marseille.
Pierre doesn’t live in Paris.
So: Pierre lives in Marseille
15From “Pierre lives in Paris or Marseille” we can’t conclude “Pierre lives in Marseille“ – any more than from “Pierre doesn’t live in Paris“. On the other hand, from the block of premises {“Pierre lives in Paris or Marseille“ + “Pierre doesn’t live in Paris“}, we can conclude that Pierre lives in Marseille. The “blocks“ formed by these arguments can be welded together by various relations of interdependence, which Snoeck-Henkemans examines (1997: 25-36). This is coordinative argumentation (Eemeren, Grootendorst & Snoeck Henkemans 2002: 65), or coordinative coumpound argumentation (Snoeck-Henkemans 1997: 15), in which the premises are linked (linked premises, ibid.: 25; Govier, 2010: 25).
- 5 On the logic behind argument classifications, see Plantin 2017, Kienpointner 1992; on specific typo (...)
16The relationships of SUPPORT or CONFLICT between bricks or brick-blocks can be specified using the notion of argument schemes; the inventory of argument schemes and their classification varies from one theoretical model to another.5
- 6 Eemeren, Grootendorst & Snoeck Henkemans refer to substandpoint (2002: 66).
17The term “argumentation” (or, to use Plantin 2016’s term, “argumentative cell”) is used to designate the set formed at least by an argument brick (or counterargument brick) and a conclusion brick. Each argument brick or counterargument brick can in turn be supported or contested and then acquires the status of a6 conclusion regarding the supporting or contending bricks (or blocks-of-bricks). We then have two partially superimposed argumentative cells, with the same brick acting as argument or counterargument in one, and as conclusion in the other. The term “subordinate argumentative structure” is used to designate these phenomena of “chaining” between argumentative cells (Eemeren, Grootendorst & Snoeck-Henkemans 2002: 65). Such constructions can be understood as attempts to anticipate possible criticism (i.e. to justify a proposition in advance, before it is challenged). All argumentative cells that are dependent on the initial argumentation constitute sub-argumentations (ibid.: 66).
18Most theories of argumentation make room for implicit elements in their modeling of arguments1. Perelman & Olbrechts-Tyteca (1969) state that, since “the springs supporting the argumentation are almost never entirely explicitly described”, when we seek to identify the argumentative pattern at work in a fragment of discourse, “we must interpret the words of the speaker, supply the missing links” (Perelman & Olbrechts-Tyteca 1969: 187). As such an operation is “always very risky” (ibid.), many authors strive to frame the reconstruction of elements initially left implicit (see, for example, Lumer 2003). It’s not a question of reconstructing all the implicit in a discourse, but only that which corresponds to missing premises, the restitution of which makes it possible to fill inference gaps (Govier 2018:124).
19Most authors agree that it is the trigger for argumentative activity, and that it involves the articulation of a discourse (understood as a position supported by arguments) with a counter-discourse, the description of which needs to be refined. The first degree of divergence likely to give rise to an argumentation is not, strictly speaking, disagreement, but a questioning, a simple questioning:
Marie: Macron should resign!
Pierre: I’m not sure, in fact I don’t really know what to think.
20In this case, in the words of van Eemeren et al. (2002: 3), we’re in a non-mixed quarrel (only one point of view is under discussion). Pragma-dialectics considers that we move from a non-mixed dispute to a mixed dispute as soon as Pierre no longer simply questions Marie’s position but undertakes to defend the opposite position (“Macron shouldn’t resign”). The distinction between mixed and unmixed argumentation is often very difficult to make, particularly in polemical contexts, where any expression of doubt is immediately interpreted as a sign of support for the opposing position (Amossy 2014) and where, under pressure, a speaker initially inclined to adopt a position of uncertainty comes to take a stand for one side or the other. This is also the case, conversely, in pacified contexts where, to spare the interactants’ faces and not compromise conversational harmony, an opposing point of view will be formulated as a simple question (Eemeren et al. 2002: 8). However delicate the distinction between questioning a position and adopting an opposing one may be, it makes sense for the analyst, insofar as only the second branch of the alternative places an obligation on the speaker to justify himself (if the interlocutor or circumstances so require). In the first case, the speaker has not undertaken to defend any position and therefore has no burden of proof (ibid.: 8).
21Pragma-dialectics considers that, when confronted with complex disputes, the analyst must break them down into as many elementary disputes (simple – single – and non-mixed) as necessary. As a result of this instruction, in every dispute (single and non-mixed) submitted for analysis, disagreement is reduced to a question of doubt. The refutation of a proposition put forward by a speaker L1 (“Macron should resign”) by a speaker L2 is seen as opening a new elementary argument around the position opposed to the one defended by L1 (“Macron should not resign”).
- 7 A similar perspective is developed by Apothéloz, Brandt & Quiroz (1992)
22Rocci (2021) shares the questions raised by pragma-dialecticians regarding the ambivalence of certain interventions in an argumentative exchange: are they to be considered as simply casting doubt on the point of view under discussion? As attacks on that point of view? As arguments in favor of the opposing point of view? For his part, he invites us to consider that any attack (against a conclusion, a premise or an argument) must also be seen as an argument in the service of an opposing conclusion. From this perspective, any critical contribution to an argumentative exchange serves both to reject the argumentative line in progress, and to support the opposite conclusion. But he does not follow the proposal put forward by pragma-dialectics to break down argumentative exchanges to manipulate only argumentative structures reflecting the way in which a protagonist supports his point of view and seeks to integrate this ambivalence into the analysis of critical interventions in argumentative exchanges7.
- 8 Citing Freeman (1991, 2011) reworked by Peldszus & Stede (2013).
23Finally, a critical intervention in an argumentative discussion can be embodied in attacks of various kinds. Rocci (2021)8 distinguishes three cases, depending on whether the strategy adopted consists in
-
rebutting a conclusion. Consider the following argument:
“This building should be demolished; it’s full of asbestos”
the conclusion can be refuted with an anti-orientation argument (e.g. “On the other hand, many people like the view from the roof”).
-
rebutting an argument (e.g. “But so far, no one has done a serious study of its asbestos content’“).
-
undercutting an argument (attacking the inference leading from an argument to the conclusion): this involves weakening the reinforcement of a conclusion by an argument through “undermining” (e.g. “We could have the building decontaminated”) (Rocci 2021: 148).
Each of these attacks can itself be the subject of a support or attack activity.
24We don’t claim to have exhausted all the subtleties of the way in which the various approaches to argumentation propose to model arguments1; but we do now have elements on which to base our understanding and evaluation of the visual representation solutions generated by the software we’ll now be looking at, and which are briefly presented in the following section.
- 9 https://www.mindmup.com
25MindMup is an online tool for creating mind maps and argument maps, accessible via a web browser with no installation required. It is developed by Gojko Adzic and his team at MindMup. Although mainly used for mind mapping, MindMup includes specific functionalities (“argument visualization” module) that enable the generation of complex argumentative structures. This module restricts the nature of the units that can be created (nodes) and the possible relationships between these units (support/opposition); it thus strongly constrains what can or cannot be represented and only allows the integration of elements clearly fulfilling an argumentative function. It is possible to assign different degrees of “strength” to the argumentative lines identified. MindMup automatically generates a hierarchical numbering system for easy, unambiguous reference to the various elements that make up the visual representation of an argument.
- 10 https://www.rationaleonline.com
26Rationale is a software application dedicated to argument mapping, developed by Austhink Software, a company founded by cognitive scientist Tim van Gelder of the University of Melbourne. One of the ways in which Rationale represents argumentative structures is very similar to what MindMup offers in its argument visualization module. Like MindMup, Rationale helps users to structure and analyze complex arguments by providing a visual framework for identifying the functional components of an argument and the relationships between them. It offers the possibility of visually marking the evaluation of an argument in a more refined way than MindMup (it distinguishes between the evaluation of the acceptability of premises and the evaluation of the soundness of inferences). Rationale also offers a range of visualization options (“reasoning”, “analysis”, etc.), pre-defined structuring matrices (Letter to the Editor, Multiple reasons, etc.) and a module for writing argumentative texts. Rationale has been widely used in educational contexts to teach critical thinking, logical analysis and reasoning skills.
- 11 https://www.bCisiveonline.com
27bCisive is based on Rationale and is also developed by Austhink Software. It extends the capabilities of argument mapping to the field of corporate decision-making. It integrates tools for analyzing strategic options, assessing risks and benefits, and facilitating communication between stakeholders. bCisive is designed to enable professionals (consultants, project managers...) to model complex decision-making processes, organize relevant information and present conclusions clearly and persuasively. It offers a very large repertoire of icons representing various types of sources (“publication”, “personal experience”, “web”...) or different functional characterizations, not necessarily argumentative, of content elements (“consequence”, “explanation”, “fact”...). It does not include Rationale’s pedagogical features. The variety of icons made available to users by bCisive makes it visually attractive and adaptable to each user’s objectives; the downside is that the “meaning” associated with these icons is most often under-defined, which is an advantage in terms of flexibility of use in a collective decision-making context, but can be a limitation for those who expect the device to be based on a coherent model of argumentation or debate.
- 12 http://ova.arg-tech.org/
- 13 https://arg.tech/f/IAT_guidelines_and_tutorials-2023-10.pdf
28OVA3 (Online Visualization of Argument version 3) is a web-based tool developed by the Centre for Argument Technology (ARG-tech) at the University of Dundee, headed by Professor Chris Reed. It is the latest avatar of Araucaria. Technically, OVA3 is based on the Argument Interchange Format (AIF), a data exchange format that enables interoperability between different argumentation analysis and representation tools. It offers two distinct modules, depending on whether the user wishes to analyze and visually represent the structure of monological or dialogical arguments. The theoretical framework on which it is based is Inference Anchoring Theory (IAT),13 which seeks to capture dialogical and argumentative structures, and the way in which they are articulated. This theory integrates categories such as argument types, dialogue structures and illocutionary forces, to capture both the logical dimension of argumentative exchanges and their interactional dynamics.
- 14 https://www.mindmanager.com
29MindManager, developed since the early 2000s by the Californian company Mindjet, was originally designed for mind mapping and project management. It was therefore not designed specifically for argumentative analysis, but it can be adapted for this purpose and, like the devices presented above, represent the structure of an argument, connect propositions, identify the relationships between these propositions and make the logic of complex reasoning explicit, even if the result does not always achieve the visual clarity of devices dedicated to the representation of argumentative structures. The addition of icons, labels and comments helps to clarify argumentative relationships and follow the development of a discussion. Any added information (metadata, annotations) can be used to filter, hide or soften elements.
- 15 They do allow some license-free manipulations, but these are so limited as to be negligible (for ex (...)
30All the software programs considered allow you to export visualizations in various formats (pdf, png...). Only OVA allows all data to be retrieved for use in other software, thanks to the AIF (Argument Interchange Format) mentioned above. Rationale, bCisive and MindManager are not free15. The free version of MindMup provides users with its main functionalities, to which the version adds a few “extras” (such as collaborative working possibilities, dedicated storage spaces, the ability to produce trees of unlimited size, etc.). OVA is completely free. All allow online use (without software download).
- 16 Twelve Angry Men (the play as well as the film) is material that has been widely exploited in a var (...)
31To illustrate the types of visual representations made possible by the different devices we’ve just presented, we’ve decided to borrow our examples from a common source, densely argumentative, clear in its argumentative objectives and potentially familiar to the reader: Twelve Angry Men16.
- 17 https://www.sellingyourscreenplay.com/wp-content/uploads/screenplay/scripts/12-Angry-Men.pdf
32Twelve Angry Men is a play written in 1954 for television by American playwright Reginald Rose, after he himself served as a juror at a trial in which a teenager was tried for murder. The film is best known for Sydney Lumet’s 1957 adaptation; however, it’s Rose’s play that we’ll take as our reference text17.
33This play depicts the deliberations of a jury that must decide on the guilt of a young Hispanic man accused of stabbing his father to death. It begins at the end of the closing arguments: the judge tells the jurors that they must reach a unanimous decision of “guilty” or “not guilty”. The judge also specifies that a guilty verdict will necessarily be equivalent to the death penalty. The jurors then retire to a room to deliberate. The first round shows that 11 of the 12 jurors are convinced of the defendant’s guilt; only one expresses uncertainty. The jurors are asked to argue their position to win over the dissenting juror to their point of view.
34The twelve jurors are never referred to by name, but simply by number. We offer a brief characterization:
-
N°1: jury president; high school soccer coach
-
N°2: bank employee; shy
-
N°3: owner of a courier company; hot-tempered, impulsive
-
N°4: stockbroker; methodical, logical
-
N°5: hospital employee; from a poor neighborhood
-
N°6: house painter; realistic, honest, respectful of others
-
N°7: salesman; cynical, extrovert, American soccer fan
-
N°8: architect; scrupulous, patient, persuasive
-
N°9: retired; wise, observant, empathetic
-
N°10: garage owner; categorical, jaded
-
N°11: watchmaker; of European origin, discreet, meticulous
-
N°12: publicist; smooth, charming, unfocused.
35The following section provides a brief overview of the main arguments used by jurors in their deliberations, to facilitate understanding of the examples discussed throughout this article.
36The accused is a 19-year-old Hispanic male. Orphaned from his mother at the age of nine, he has a long criminal record (placement in reform school, arrest for assault, involvement in knife fights).
37At around 8 p.m. on the evening of the murder, the accused and his father argue. The father hit the accused at least twice. Shortly after this violent argument, the accused left the apartment.
38The accusation is based essentially on two testimonies.
39Around midnight, a woman living across the subway tracks from the father’s apartment wakes up. Through the windows of a passing train, she sees the father being stabbed by an assailant she identifies as her son.
40An elderly man who lives in the apartment below the father’s testified that, around the time the woman witnessed the stabbing, he heard the accused shout “I’m going to kill you” and, a second later, heard a body fall to the floor. He says he got up from his bed, went to the door and saw the accused running down the stairs.
41The accused returned to his father’s apartment at around 3am. The police, who had been alerted by the neighbor across the street, immediately questioned him. He claims to have gone to see friends shortly after the argument with his father. At the trial, he stated that he had gone to the cinema alone at around 11pm and had returned at 3am. He has no recollection of the films he saw, and can point to no witnesses who could confirm his statements.
42Finally, the accused admits that, shortly after the fight with his father, he went to a store where he bought a switchblade knife identical to the one found embedded in his father’s chest. He claims to have lost the knife, which would have fallen through a hole in his pocket.
43We will now examine which elements of the ontology outlined above can be supported by argumentation visualization devices, and what solutions they propose.
- 18 Recall that the analytical perspective defended by pragma-dialectics requires recourse only to the (...)
44Devices for visualizing argumentative structures based on a minimal ontology, such as MindMup or Rationale, only distinguish between two types of possible relationships between elements of different levels: a supportive relationship (an element is presented as contributing to reinforcing the acceptability of another element of a higher level) and an oppositional relationship (an element is presented as diminishing the acceptability of another element of a higher level)18.
45Let’s take the intervention of juror N°7 who, faced with the alternative opened up by the jury foreman (do we begin by going round the table so that everyone can state their position for or against the guilt of the accused, or do we launch a general discussion on the elements of the trial?), opts for the first solution in the hope that an immediate consensus will enable the exchanges to be shortened:
N° 7: Let’s vote now. Who knows, maybe we can all go home.
46MindMup invites you to represent this argumentative micro-sequence as follows:
- 19 In the argumentation visualization devices studied here, the green color marks a supportive relatio (...)
47In this representation, the link marked by the green line between 2.1 and 1.1 is a support link19, meaning something like “2.1 makes 1.1 more acceptable”. As mentioned above, many approaches to argumentation allow us to specify such a support relationship by characterizing it in terms of an argument scheme. MindMup or Rationale do not integrate such an indication into the visualization of the argument structure. OVA, on the other hand, does – and even provides the user with categories corresponding to different theoretical frameworks. If we select Walton Presumptive Inference among these frameworks, we can characterize the link between 2.1 and 2.2 as argumentation by positive consequences (one of the variants of pragmatic argument in the terminology of Perelman & Olbrechts-Tyteca 1969: 266).
48This characterization appears on the arrow representing the inference, and the green color indicates, once again, that this is a supportive relationship and not an oppositional one.
49The same type of observation can be made for elements whose function is to weaken the acceptability of an assertion: whereas MindMup, Rationale or bCisive only provide a basic visualization of the relationship these elements have with the conclusion brick (marked by the color red in contrast to the green of the previous example), OVA allows the logic of opposition at work to be specified. Thus, when juror N°8 proposes to examine the possibility that the accused is not guilty, juror N°10 rejects this proposal, discrediting the accused’s version in advance on the grounds of his background, which constitutes an ad hominem refutation:
N°10: You’re not going to tell us that we’re supposed to believe him, knowing what he is. I’ve lived among ‘em all my life. You can’t believe a word they say. You know that.
50All the devices allow us to account for what we mentioned earlier as a structural property of all argumentation – i.e., the fact that any proposition is capable of being argued (justified or criticized) by means of a premise or objection, which in turn, as a proposition, can be the subject of a procedure designed to strengthen or weaken it, and this, ad libitum. In the following example, juror N°7 seeks to support the thesis of the defendant’s guilt by referring to his criminal record, which he believes demonstrates that he is not a “very fine boy” (even if he ironically asserts the contrary):
N° 7: Look at the kid’s record. At fifteen he was in reform school. He stole a car. He’s been arrested for mugging. He was picked up for knife-fighting. I think they said he stabbed somebody in the arm. This is a very fine boy.
51This support can be represented as follows (we use MindMup, but other devices offer similar representations):
52Conclusion 1.1 is not stated in N°7’s intervention reproduced above, but it can unhesitatingly be reinstated insofar as, during the first round, N°7 came out in favor of the thesis of the defendant’s guilt, and the jurors are in the process of justifying their position one after the other during a second round. As we can see, 2.1 has the dual status of an argument in relation to 1.1, and a conclusion in relation to 3.1, 3.2, 3.3 and 3.4, which, although independent of each other, all converge on the ironic statement “This is a very fine boy.”
53As mentioned above, when argumentation theorists address the question of the structure of an argumentative sequence, they emphasize the need to account, in one way or another, for the fact that several elements may contribute independently of each other to reinforcing (or attenuating) the acceptability of a conclusion, whereas they must sometimes be considered together, as a block, which itself fulfils such a function.
54Most devices for visualizing argumentative structures offer a solution for representing these possible links between supporting or contesting elements. Thus, when jurors convinced of the defendant’s guilt review the facts that support their conclusion, they implement (at least) two distinct argumentative lines:
N° 4: The boy’s entire story is flimsy. He claimed he was at the movies. That’s a little ridiculous, isn’t it? He couldn’t even remember what pictures he saw.
N° 3: That’s right. Did you hear that? (To N° 4) You’re absolutely right.
N° 10: Look, what about the woman across the street? If her testimony don’t prove it, then nothing does.
N° 12: That’s right. She saw the killing, didn’t she?
55The first line of argument (2.1) points to the weakness of the alibi put forward by the accused. The second line of argument is based on the testimony of the neighbor across the street (2.2). Each of these lines is independent of the other (as evidenced by the existence of two distinct inferential links).
- 20 A position he expressed during the first round of discussions, and which he sets out to justify her (...)
56On the other hand, when juror N°4, who at this point in the deliberation is leaning towards the guilt of the accused20, mentions the accused’s social background:
N° 4: We’re missing the point here. This boy – let’s say he’s a product of a filthy neighborhood and a broken home. We can’t help that. We’re not here to go into the reasons why slums are breeding grounds for criminals. They are. I know it. So do you. The children who come out of slum backgrounds are potential menaces to society.
57It puts forward two elements which, taken together, make the conclusion that the accused is indeed guilty (1.1) more acceptable: the mention of his social origin (2.1), and a proposal establishing a link between such an origin and the propensity to crime (2.2). The fact that 2.1 and 2.2, taken together, reinforce 1.1 is visually represented by the horizontal brace that groups them together, and the single inferential link that runs from this brace to the conclusion.
58The solution proposed by bCisive differs in that it introduces a visually graspable distinction between the two coupled elements:
- 21 Rigotti & Greco (2019), with the Argumentum Model of Topics, considerably sophisticate the ontology (...)
59In bCisive, a reason is defined as a supporting statement, and the co-premise associated with it by the ampersand is defined as an additional claim that snaps onto a Reason or an Objection to establish relevance. The asymmetry between the two elements echoes Toulminian models (Toulmin 1958) of argumentation, distinguishing data from guarantor; it also recalls, in syllogism logic, the distinction between minor and major premise21.
60An argumentation can be contested in two ways: one concerns the acceptability of the proposition itself, the other the strength of the inference linking the premise or objection to the conclusion to which it points. However, some devices do not allow us to visually distinguish between these two modes of argumentation contestation, while others do.
61This point is illustrated by the discussion, in Twelve Angry Men, of the downstairs neighbor’s testimony that he heard the young man shout “I’m going to kill you” just before he heard a thud, like that of a body falling to the ground:
N° 8: The old man heard the boy say, “I’m going to kill you,” and one second later he heard a body fall. One second. That’s the testimony, right?
N° 2: Right. [...]
N° 8: The old man would have had to hear the boy say, “I’m going to kill you,” while the front of the el was roaring past his nose. It’s not possible that he could have heard it. [...] Now. There’s something else I’d like to point out here. I think we proved that the old man couldn’t have heard the boy say, “I’m going to kill you,” but supposing he really did hear it? This phrase: how many times has each of you used it? Probably hundreds. “If you do that once more, Junior, I’m going to murder you.” “Come on, Rocky, kill him!” We say it every day. This doesn’t mean that we’re going to kill someone.
62One of the main arguments in favor of the young man’s guilt is the testimony of the downstairs neighbor reported in 1A-a. 2A-a directly attacks the propositional content of 1A-a, asserting that it’s impossible for the neighbor to have heard what he claims to have heard, given the circumstances (the aerial metro passed right in front of his open window, producing a deafening din; 3A-a). In 2B-a, N°8 proposes another line of attack: he argues that even if the young man had indeed shouted “I’m going to kill you”, this would not have led to the conclusion that “the young man killed his father”. 2B-a is supported by the evocation of experiences widely shared by all, where, in the throes of exasperation, people have come to call for the death of others, without any murder being committed (3B-a). As we can see, Rationale distinguishes between attacks on the proposition contained in a box (they point to the challenged box) and attacks on the inference that leads from the premise to the conclusion (or from the objection to the proposition it attacks), which point to the term indicating the argumentative orientation of the sequence (here, “supports”).
63Finally, none of the devices considered here allows us to take into account the analytical perspective opened up by Rocci (2021) when he considers that any objection contesting a conclusion C is more or less an argument in favor of an opposite conclusion, for the same reasons that none (with the exception of OVA) allows us to represent divergent arguments (where the same brick serves as an argument for different conclusions) : this would run counter to the tree-like structure adopted by most of them (and the visual solution proposed by Rocci 2021, which is very precise analytically, does not seem to us to be compatible with these devices).
64As we have also seen above, reconstructing the structure of an argument requires us to re-establish certain elements left implicit when their absence generates inference gaps. This is the case in the following exchange, where juror N°8 is reluctant to see the accused as a dangerous killer – as N°3 invites him to do – in view of his age:
N° 3: I never saw a guiltier man in my life. You sat right in court and heard the same thing I did. The man’s a dangerous killer. You could see it.
N° 8: He’s nineteen years old.
- 22 The formulation of the reconstructed implicit element depends, of course, on the analyst, not on th (...)
65In this exchange, N°8’s objection to N°3’s assertion (“The man’s a dangerous killer) only makes sense in association with an implicit element such as “You can’t be a dangerous killer at 19”22. Even if such an element does not appear in the dialogue, it can (and must) be re-established in the analysis, since we can reasonably consider that this is what N°8 has, more or less consciously, in mind when he refuses to see the accused as a dangerous criminal in view of his young age. MindMup makes it possible to distinguish explicit elements, reproducing or reformulating elements present in the source text, from implicit elements re-established by the analyst by hijacking the latter with a broken line, as in 2.2 below:
- 23 In any case, this is hardly surprising for bCisive, insofar as this software is geared less towards (...)
66This marking, which is possible with OVA, is not possible with Rationale or bCisive.23
67Insofar as the argumentation visualization devices considered here were developed in an Anglo-Saxon context strongly marked by normative perspectives, seeking to distinguish good from bad argumentation, most of them provide for a marking of the evaluation of the arguments represented. This marking can be fine-tuned in terms of the evaluation scale taken into consideration, and the objects to which the evaluation can be applied.
68If we take the previous example, the objection made up of the two linked elements 2.1 and 2.2 is not decisive: there are many examples of murderers under the age of 19. MindMup allows you to represent the strength or weakness of an argument or objection by varying the thickness of the link between it and the element it reinforces or challenges. We can visually mark the weakness of 2.1 and 2.2 in this way (“neutral” representation in fig.a; marking the weakness of the argument in fig.b):
69Rationale provides for a double evaluation, distinguishing between the evaluation of the acceptability of the proposition contained in the box and that of the inference between the elements considered. In the case under consideration, the proposition (he’s nineteen) is true (it belongs to the elements validated during the trial); in Fig. d, this is indicated by the green sign✓. On the other hand, the objection (the ability of the assertion to weaken “The man is a dangerous killer”) is weak; this is indicated by the attenuated red color in Fig. d compared to the “neutral” version in Fig. c.
70This finer marking allows us to evaluate all propositional elements separately, whether explicit or not, and at the same time to assign different degrees of strength to the inference linking them to the higher element.
71The following paragraphs will examine the ability of the devices under consideration to render properties that go beyond an argument1 and focus on argumentation as a language activity undertaken by enunciators in a specific context (i.e., in connection with the language act of making-an-argument, or with an argument-type interaction2).
72Visualization systems that focus on rendering arguments1, without considering the linguistic and interactional (or at least, dialogical) activity in which they are involved, do not provide for visual marking of the enunciative instance in charge of a given functional component (argument, objection, conclusion) of the argumentative text.
73Thus, in the following exchange, juror N°10 rejects N°9’s previous intervention (not reproduced here), which refers to an “impression” that he feels and which makes him doubt the defendant’s guilt, to put forward the “facts” which, in his opinion, accuse him beyond all reasonable doubt:
N° 10: What about the facts?
N° 3: You said a mouthful. (To N° 9) Look, the old man heard the kid yell, “I’m gonna kill you.” A second later he heard the father’s body falling, and he saw the boy running out of the house fifteen seconds after that.
N° 12: That’s right. And let’s not forget the woman across the street. She looked into the open window and saw the boy stab his father. She saw it. Now if that’s not enough for you....
74Such an exchange can be represented in MindMup (and a similar representation would result from the use of bCisive or Rationale) as follows:
75Of course, there’s nothing to prevent each piece of content from being preceded (or followed) by an indication of the enunciative source; but the device itself makes no provision for this, which indicates that the ontology on which it is based is not populated by enunciator(s) – contrary to what would be required by approaches aiming to account for the language act of making an argument1 or the interactional dynamics specific to an argument2.
76In addition, as you will have noticed, some of the preceding trees show a considerable distance between the content of the boxes and the letter of the exchange they claim to reflect. Indeed, as O’Keefe (1982) invites us to do, we have allowed ourselves to take liberties with the concrete productions of the characters and to reformulate and synthesize them to make their argumentative orientation more obvious and preserve the readability of the tree. Up to now, we’ve sought to account for the structure of arguments1 – hence our relative indifference to the question of the enunciative support of the different “bricks”, or their sequentiality.
77Yet O’Keefe (1982), as well as Reed and Walton (2003), suggest that to simply consider argumentation as argument1, argument2 or the act of making-an-argument, without examining how these three phenomena articulate with each other, and without thinking about a theory that would enable us to think about this articulation, would be to be satisfied with a perspective that is quite impoverished in view of the richness and complexity of the field.
78It is precisely such bridges that OVA seeks to build, and the Inference Anchoring Theory on which it is based. OVA, in its dialog mode, offers the possibility of combining two forms of restitution. One seeks to give an account of the arguments1 and consists of a reformulation, possibly distanced from an element of the initial text to highlight its argumentative orientation (this is the one shown on the left of the tree below). The other strives to reconstruct what is happening as closely as possible to the interactional dynamics and language choices of the enunciators involved in the argumentative exchange. It looks at how speech acts – including that of making-an-argument – are handled by enunciators whose exchanges are both constrained by the linearity of oral speech, and by the hierarchical structure of conversation (argument2). The visual representation of the previous exchange between N°10, N°3 and N°12 then becomes:
79The elements in the blue boxes on the right-hand side of the tree reproduce fragments of the verbatim, in the order of their enunciation (from top to bottom). These elements (called locution nodes in OVA vocabulary) are associated with a specific enunciator (here, N°10 for the first, N°3 for the next two, N°12 for the last two). Purple ovals can be used to specify the nature of the transition between these verbatim fragments.
80Each locution node is associated with a blue box on the left of the tree (called an information node), to which it is linked by a specific illocutionary force (indicated in the yellow oval box in the middle of the representation). These information nodes contain the propositional content of the locution from which they originate. They may have inferential relations (green ovals), conflict relations (red ovals, not shown in this example) or reformulation relations (orange ovals) which, like the transition links, may or may not be specified.
81Finally, transition relations (purple ovals) connect propositions (green, orange or red ovals) by a link that automatically proposes an argumentative specification of the illocutionary force attached to the locution that follows. For example, the last two lines of the above representation read: “when n°12 (blue box on the right) asserts (yellow oval on the lower line) that the woman on the other side of the street saw through the open window the young man stabbing his father (blue box on the left), he argues (yellow oval on the upper line), through testimony (green oval), in favor of the conclusion that the facts accuse the defendant (blue box at the very top of the tree).”
82Such a representation seeks to articulate a large number of analytical elements:
-
Argumentative units, in their literal form (locution nodes) and in a purified propositional form that better expresses their orientation (information nodes)
-
Mention of the enunciative instance responsible for the verbatim elements integrated into the visual representation
-
The linear nature of the exchanges (since the verbatim fragments to the right of the representation appear, from top to bottom, in the order in which they were uttered)
-
The nature of the transition from one verbatim fragment to another
-
The illocutionary characterization of each fragment, associated with a speech act (in the yellow ovals) and propositional content (in the information nodes)
-
The argumentative relationship, whether supportive or oppositional, between propositional units (between information nodes) derived from verbatim fragments.
83OVA’s ambition to articulate an analysis of argumentation as a linguistic and interactional activity, and an analysis of argumentation as a structure made up of propositional elements linked by relations of support or opposition (in O’Keefe’s terms, of not letting go of the argument2 or the act of making-an-argument for the argument1, or vice versa), is commendable.
84This articulation is particularly valuable in the case of Twelve Angry Men, since the heart of the plot is the successive rallying of the various jurors to the initial position of one of them, juror N°8: we can thus envisage introducing the successive positions taken by the various jurors into the locution nodes and specifying their illocutionary force of agreement/disagreement in relation to those of their co-jurors, on the basis of the arguments put forward by one or other of them.
85This ambition comes at a cost: what is gained in richness of analysis is lost in legibility. While the diagram above only shows a very brief exchange, the linear and constrained nature of the locutions on the right of the diagram (typical of arguments2 and argumentation as a language act) does not sit well with the non-linear relations between the propositional elements on the left (argument1); the whole remains legible in the example we have constructed, but it’s easy to imagine the degree of complexity that can be reached when we try to account for longer sequences.
86At this stage of our examination of the argumentative data that can be returned by existing systems, we can ask ourselves what else seems to be missing.
87First, I’d like to make a point about one of the requirements of these devices, which caused us some problems in the restitution of certain Twelve Angry Men exchanges: they all recommend limiting the bricks to one proposition and one proposition only. Such a requirement is largely linked to the initial aim of most of them, which is to build a representation of the structure of an argument to facilitate its evaluation. However, standard evaluation protocols invite us to question the truth of premises, their relevance to the intended conclusion and the strength of inference; and the truth of premises can only be determined if they are reduced to propositional form. However, the systematic decomposition of certain interventions whose principle of coherence is essentially narrative, such as testimonies, into as many independent propositions (in the previous example, “the old man heard the kid yell, ‘I’m gonna kill you’ “ / “A second later he heard the father’s body falling” / “he saw the boy running out of the house fifteen seconds after that”), as if they were as many linked arguments, is unsatisfactory; and this remark is far from anecdotal when we look at judicial arguments, in which narrative plays a key role (Burnett & Badzinski, 2000 : 390; Vidmar, Beale & Coleman 2008).
- 24 As mentioned above, O'Keefe (1982) insists that the term “argument” can be used to designate an int (...)
88Moreover, as we have seen, only OVA, via its dialectic module, tackles the articulation of elements of analysis of argumentation seen as a propositional structure relating propositional entities through links of support or conflict, and of argumentation seen as an interaction (argument2) including, among other speech acts, that of making-an-argument24.
89The feeling that emerges from examining the representation proposals generated by OVA is mixed – and this is perhaps inevitable for any compromise solution: the argument representations1 have neither the legibility of the structures generated by Rationale or MindMup, nor the great maneuverability offered by these devices when constructing the representation. As for what OVA can represent in terms of exchange dynamics, it remains a very schematic sketch of interactional strategies which, in Twelve Angry Men (as in many non-fictional argumentative exchanges), are sometimes as subtle as they are powerful in terms of generating effects. Juror N°10, for example, considers that the defendant’s version of events should be disregarded because of the lack of credibility of anyone from the underworld. However, the neighbor across the street, whose testimony N°10 relies on to prove the young man’s guilt, comes from the same background. In N°8’s eyes, this seems an inconsistency, which he places before N°10:
N° 8: I’d like to ask you something. How come you believed her? She’s one of “them” too, isn’t she?
90For N°8, it’s a matter of putting N°10 in contradiction, at a distance, with an argumentative line he had put forward several minutes earlier. The analysis that this strategy requires is subtle, insofar as the contradiction brings into play, firstly, the explicit assertion of a principle of credibility (“you can’t believe a word people from that background say, they’re born liars”), and (at least) two elements that underpin another line of argument: if N°10 mentions the testimony of the neighbor across the street as evidence, it’s because he considers it credible; and the neighbor in question comes from much the same background. The first point is not explained; the second is formulated by N°8 as “She’s one of ‘them,’ too”.
91This strategy can’t be described without reference to the enunciators in charge of the different argumentative lines (there’s a contradiction only because it’s the same N°10 who sometimes seems to consider the environment of origin to be relevant in assessing the credibility of an individual’s remarks, and sometimes not); so none of the devices aimed at restoring arguments1 (MindMup, Rationale, bCisive) can shed any satisfactory light on it. But even OVA would struggle to do so, if only because of the time lapse separating the two argumentative moments, which introduces a difficult-to-manage gap between the right-hand side of the representation, dependent on the linearity of the exchanges, and the left-hand side, aimed at restoring their ‘“logical’“ structure independently of their anchoring at a given point in the unfolding of the interaction.
92This same example highlights another shortcoming of the systems examined here. N°8’s contradiction of N°10 triggers a flush of anger in the latter, leading him to stand up and approach his opponent in a way that the other jurors find threatening:
N° 8: I’d like to ask you something. How come you believed her? She’s one of ‘”them” too, isn’t she?
[N° 10 walks over to N° 8:]
N° 10: (sarcastically) You’re a pretty smart fellow, aren’t you?
FOREMAN: (rising) Now take it easy.
[N° 3 gets up and goes to N° 10:]
N° 3: Come on. Sit down. (He leads N° 10 back to his seat.) What’re you letting him get you all upset for?
Relax.
[N° 10 and N° 3 sit down.]
FOREMAN: Let’s calm down now. Number. 5, it’s your turn.
N° 5: I’ll pass it.
93This emotional reaction, the tension it arouses and the collective management of this disturbance by the jurors as a whole phenomenon; should they be accounted for in the analysis of argumentation? The answer, of course, depends on how we conceive of argumentation, and the importance we attach to the various facets of the phenomenon. From the point of view usually adopted by Pierre Pilon, one of the authors of the present article, who uses argumentation visualization devices notably in the context of decision-making meetings, emotions can constitute indications likely to be significant in the construction of an argumentation map (for example, to point out points that we can anticipate will be difficult to get a collective to accept); furthermore, the emotions expressed by participants constitute indications of satisfaction or dissatisfaction that encourage further elaboration of the tree.
- 25 And for us, contrary to Amossy's (2018) proposal to consider all discourse as having at least an ar (...)
94From the point of view of Marianne Doury, the other author, the answer is obviously positive. Her perspective is that of argumentative discourse analysis. In other words, it’s about practicing a form of discourse analysis, which implies considering language productions from a situated perspective, attached to enunciators expressing themselves and interacting in special and significant situations, traversed by stakes that inform the exchanges, the latter allowing, to a certain extent, to re-negotiate them. Moreover, it is an analysis of argued discourse25. Because the analysis aims to account for this argumentative character, it must be supported by specific conceptualizations that enable argumentation to be analyzed:
-
as a logical structure that clarifies the coherence of the discourse,
-
as a language practice with a claim to transformative power (perhaps no more, but no less, than any act of language),
-
and insofar as they engage male and female arguers in exchanges marked by strategies of relative positioning (agreement, disagreement, rapprochement or questioning) likely to have, among other effects, consequences on the relationship that unites them and on their respective faces. And, it should be emphasized, these stakes in argumentative exchanges for the sides of the interactants are not “sidelines” for the analysis of argumentation, but must be integrated and dealt with, insofar as they act on the argumentative component.
95The deliberation process in Twelve Angry Men is strongly determined by relationships of domination/intimidation between the jurors, which are established as soon as they enter the room where the in camera session will take place, and which have as much to do with their psychological determinants (N°2’s almost sickly shyness contrasts with N°7’s extroverted and cynical character) as with their socio-economic capital (N°8, an architect, N°10 a garage mechanic, N°6 a house painter...). These relations of power are renegotiated as the exchanges progress, and while they initially led some of the jurors to seek to avoid speaking out and asserting their point of view, they are now balanced out and they opened a space in which everyone feels legitimate in asserting their alliances or differences with the other jurors. Logics of domination and their possible emotional corollaries are therefore fully relevant to the analysis of argumentative discourse.
- 26 https://rajesh.work/resume
96Does this mean that visualization devices must reflect this? To us, this seems much more uncertain. Perhaps this reservation is due to a lack of imagination on our part: it’s quite possible that the balance of power between interactants engaged in a discussion, or the emotional tone of exchanges, could be the subject of an enlightening visual representation. This, at least, is the inspiration behind Rajesh’s proposal26, reproduced below, which focuses on identifying the main flows of exchange between jurors (chord diagram in the upper part of the visual) and the dynamics of jurors rallying to the position of N°8 (bottom bars):
https://www.behance.net/gallery/66212835/Data-Analysis-with-Movie-Script-12-Angry-Men
97Such a representation, focused on certain aspects of interactional dynamics, seems enlightening to us; but it is only so at the price of eliminating (very many) other possible angles into the data, including, of course, the strictly argumentative ones. It should be noted, however, that even the interactional approach is far from exhausted by this representation.
98In short, what existing devices for the visual representation of argumentative structures can do – and do rather well – is restore the way in which arguments1 are constructed. But just because MindMup, Rationale, bCisive or OVA can’t do everything, that doesn’t mean we should overlook their contribution to the analysis of argumentation. It seems to us, moreover, that proposals aimed at restoring, in a single visual representation, considerations linked to arguments1, to the act of making-an-argument and to arguments2 can only make sense on very limited fragments – and even then, at the cost of considerably impoverishing each of the angles of approach supported. On the other hand, we can dream of a device that allows us to annotate data from every possible angle and can generate a specific type of visual representation, depending on the preferred lighting and thanks to a complex system of filters.
99MindManager opens new avenues in this direction. It allows you to break down an initial text into fragments whose grain can be determined according to the needs of the analyst (in this way, it avoids the requirement of a breakdown into propositions criticized above). It offers the possibility of annotating data from any desired angle (in terms of argumentative function, enunciative support, speech act, emotional tone, etc.). By activating a system of filters, it enables specific insights to be gained into the text under consideration. Since annotations can be exported to Excel tables, MindManager analyses can serve as the basis for visualizations made possible by Excel, or by other devices (e.g. OVA, thanks to AIF).
100The comparative analysis we have carried out reveals a central paradox in the computerized visualization of argumentation: while digital tools offer unprecedented possibilities for representing argumentative structures, they simultaneously confront us with the limits inherent in any attempt to formalize argumentative phenomena. The analytical grid inspired by O’Keefe (1982), by distinguishing argument₁, making-an-argument and argument₂, enabled us to characterize precisely the potentialities and blind spots of each device studied.
101The software reviewed is distributed along a continuum between two poles. At one end, tools like MindMup and Rationale focus on visualizing the logical relationships between propositions (argument₁), offering clear but necessarily reductive representations of argumentative complexity. At the other extreme, OVA attempts to articulate the logical, enunciative and interactional dimensions of argumentation, at the cost of significantly complexifying the schemas produced. Between these two poles, bCisive and MindManager offer different compromises, revealing the trade-offs involved in any endeavor to visualize argumentation.
102Our study corpus, Twelve Angry Men, highlighted crucial aspects of argumentation that systematically elude current devices: relations of domination between speakers, the power of narratives in the construction of argumentative positions, and the emotional dimension of exchanges. These blind spots are not simply technical limitations: they call into question the very possibility of reducing argumentation to a system of formalizable relations. More profoundly, they invite us to re-examine what “visualizing argumentation” can mean.
103This reflection leads to a fundamental rethinking of the epistemological status of visualization tools in argumentative analysis. Rather than devices aimed at descriptive exhaustiveness, these software tools should be seen as heuristic instruments, enabling us to shed light on certain specific dimensions of argumentation, depending on the objectives pursued. From this point of view, the diversity of approaches embodied by the various software packages is less a limitation than a source of richness: each tool, with its theoretical presuppositions and visualization choices, offers a particular insight into the complexity of the argumentative phenomenon.
104These conclusions open several perspectives for the future development of visualization tools. The first would be to explore new representation modes that integrate the currently neglected dimensions of argumentation, notably its socio-relational and narrative aspects. A second avenue would be to develop interfaces that allow dynamic switching between different visualization modes, adapted to different dimensions of analysis. More fundamentally, our results call for further reflection on the impact of digital tools on our analysis practices and our very conception of argumentation.
105At the end of this study, it appears that computerized visualization of argumentation is not simply a technical advance in our methods of analysis: it confronts us with fundamental epistemological questions about the very nature of the argumentative phenomenon. While digital tools undeniably enrich our analytical possibilities, they also remind us that argumentation remains irreducible to its formal representations. It is perhaps in this productive tension between formalization and resistance to formalization that lies one of the major contributions of new technologies to the study of argumentation.